1======================== 2ftrace - Function Tracer 3======================== 4 5Copyright 2008 Red Hat Inc. 6 7:Author: Steven Rostedt <srostedt@redhat.com> 8:License: The GNU Free Documentation License, Version 1.2 9 (dual licensed under the GPL v2) 10:Original Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton, 11 John Kacur, and David Teigland. 12 13- Written for: 2.6.28-rc2 14- Updated for: 3.10 15- Updated for: 4.13 - Copyright 2017 VMware Inc. Steven Rostedt 16- Converted to rst format - Changbin Du <changbin.du@intel.com> 17 18Introduction 19------------ 20 21Ftrace is an internal tracer designed to help out developers and 22designers of systems to find what is going on inside the kernel. 23It can be used for debugging or analyzing latencies and 24performance issues that take place outside of user-space. 25 26Although ftrace is typically considered the function tracer, it 27is really a framework of several assorted tracing utilities. 28There's latency tracing to examine what occurs between interrupts 29disabled and enabled, as well as for preemption and from a time 30a task is woken to the task is actually scheduled in. 31 32One of the most common uses of ftrace is the event tracing. 33Throughout the kernel is hundreds of static event points that 34can be enabled via the tracefs file system to see what is 35going on in certain parts of the kernel. 36 37See events.rst for more information. 38 39 40Implementation Details 41---------------------- 42 43See Documentation/trace/ftrace-design.rst for details for arch porters and such. 44 45 46The File System 47--------------- 48 49Ftrace uses the tracefs file system to hold the control files as 50well as the files to display output. 51 52When tracefs is configured into the kernel (which selecting any ftrace 53option will do) the directory /sys/kernel/tracing will be created. To mount 54this directory, you can add to your /etc/fstab file:: 55 56 tracefs /sys/kernel/tracing tracefs defaults 0 0 57 58Or you can mount it at run time with:: 59 60 mount -t tracefs nodev /sys/kernel/tracing 61 62For quicker access to that directory you may want to make a soft link to 63it:: 64 65 ln -s /sys/kernel/tracing /tracing 66 67.. attention:: 68 69 Before 4.1, all ftrace tracing control files were within the debugfs 70 file system, which is typically located at /sys/kernel/debug/tracing. 71 For backward compatibility, when mounting the debugfs file system, 72 the tracefs file system will be automatically mounted at: 73 74 /sys/kernel/debug/tracing 75 76 All files located in the tracefs file system will be located in that 77 debugfs file system directory as well. 78 79.. attention:: 80 81 Any selected ftrace option will also create the tracefs file system. 82 The rest of the document will assume that you are in the ftrace directory 83 (cd /sys/kernel/tracing) and will only concentrate on the files within that 84 directory and not distract from the content with the extended 85 "/sys/kernel/tracing" path name. 86 87That's it! (assuming that you have ftrace configured into your kernel) 88 89After mounting tracefs you will have access to the control and output files 90of ftrace. Here is a list of some of the key files: 91 92 93 Note: all time values are in microseconds. 94 95 current_tracer: 96 97 This is used to set or display the current tracer 98 that is configured. Changing the current tracer clears 99 the ring buffer content as well as the "snapshot" buffer. 100 101 available_tracers: 102 103 This holds the different types of tracers that 104 have been compiled into the kernel. The 105 tracers listed here can be configured by 106 echoing their name into current_tracer. 107 108 tracing_on: 109 110 This sets or displays whether writing to the trace 111 ring buffer is enabled. Echo 0 into this file to disable 112 the tracer or 1 to enable it. Note, this only disables 113 writing to the ring buffer, the tracing overhead may 114 still be occurring. 115 116 The kernel function tracing_off() can be used within the 117 kernel to disable writing to the ring buffer, which will 118 set this file to "0". User space can re-enable tracing by 119 echoing "1" into the file. 120 121 Note, the function and event trigger "traceoff" will also 122 set this file to zero and stop tracing. Which can also 123 be re-enabled by user space using this file. 124 125 trace: 126 127 This file holds the output of the trace in a human 128 readable format (described below). Opening this file for 129 writing with the O_TRUNC flag clears the ring buffer content. 130 Note, this file is not a consumer. If tracing is off 131 (no tracer running, or tracing_on is zero), it will produce 132 the same output each time it is read. When tracing is on, 133 it may produce inconsistent results as it tries to read 134 the entire buffer without consuming it. 135 136 trace_pipe: 137 138 The output is the same as the "trace" file but this 139 file is meant to be streamed with live tracing. 140 Reads from this file will block until new data is 141 retrieved. Unlike the "trace" file, this file is a 142 consumer. This means reading from this file causes 143 sequential reads to display more current data. Once 144 data is read from this file, it is consumed, and 145 will not be read again with a sequential read. The 146 "trace" file is static, and if the tracer is not 147 adding more data, it will display the same 148 information every time it is read. 149 150 trace_options: 151 152 This file lets the user control the amount of data 153 that is displayed in one of the above output 154 files. Options also exist to modify how a tracer 155 or events work (stack traces, timestamps, etc). 156 157 options: 158 159 This is a directory that has a file for every available 160 trace option (also in trace_options). Options may also be set 161 or cleared by writing a "1" or "0" respectively into the 162 corresponding file with the option name. 163 164 tracing_max_latency: 165 166 Some of the tracers record the max latency. 167 For example, the maximum time that interrupts are disabled. 168 The maximum time is saved in this file. The max trace will also be 169 stored, and displayed by "trace". A new max trace will only be 170 recorded if the latency is greater than the value in this file 171 (in microseconds). 172 173 By echoing in a time into this file, no latency will be recorded 174 unless it is greater than the time in this file. 175 176 tracing_thresh: 177 178 Some latency tracers will record a trace whenever the 179 latency is greater than the number in this file. 180 Only active when the file contains a number greater than 0. 181 (in microseconds) 182 183 buffer_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 dyn_ftrace_total_info: 370 371 This file is for debugging purposes. The number of functions that 372 have been converted to nops and are available to be traced. 373 374 enabled_functions: 375 376 This file is more for debugging ftrace, but can also be useful 377 in seeing if any function has a callback attached to it. 378 Not only does the trace infrastructure use ftrace function 379 trace utility, but other subsystems might too. This file 380 displays all functions that have a callback attached to them 381 as well as the number of callbacks that have been attached. 382 Note, a callback may also call multiple functions which will 383 not be listed in this count. 384 385 If the callback registered to be traced by a function with 386 the "save regs" attribute (thus even more overhead), a 'R' 387 will be displayed on the same line as the function that 388 is returning registers. 389 390 If the callback registered to be traced by a function with 391 the "ip modify" attribute (thus the regs->ip can be changed), 392 an 'I' will be displayed on the same line as the function that 393 can be overridden. 394 395 If a non ftrace trampoline is attached (BPF) a 'D' will be displayed. 396 Note, normal ftrace trampolines can also be attached, but only one 397 "direct" trampoline can be attached to a given function at a time. 398 399 Some architectures can not call direct trampolines, but instead have 400 the ftrace ops function located above the function entry point. In 401 such cases an 'O' will be displayed. 402 403 If a function had either the "ip modify" or a "direct" call attached to 404 it in the past, a 'M' will be shown. This flag is never cleared. It is 405 used to know if a function was every modified by the ftrace infrastructure, 406 and can be used for debugging. 407 408 If the architecture supports it, it will also show what callback 409 is being directly called by the function. If the count is greater 410 than 1 it most likely will be ftrace_ops_list_func(). 411 412 If the callback of a function jumps to a trampoline that is 413 specific to the callback and which is not the standard trampoline, 414 its address will be printed as well as the function that the 415 trampoline calls. 416 417 touched_functions: 418 419 This file contains all the functions that ever had a function callback 420 to it via the ftrace infrastructure. It has the same format as 421 enabled_functions but shows all functions that have every been 422 traced. 423 424 To see any function that has every been modified by "ip modify" or a 425 direct trampoline, one can perform the following command: 426 427 grep ' M ' /sys/kernel/tracing/touched_functions 428 429 function_profile_enabled: 430 431 When set it will enable all functions with either the function 432 tracer, or if configured, the function graph tracer. It will 433 keep a histogram of the number of functions that were called 434 and if the function graph tracer was configured, it will also keep 435 track of the time spent in those functions. The histogram 436 content can be displayed in the files: 437 438 trace_stat/function<cpu> ( function0, function1, etc). 439 440 trace_stat: 441 442 A directory that holds different tracing stats. 443 444 kprobe_events: 445 446 Enable dynamic trace points. See kprobetrace.rst. 447 448 kprobe_profile: 449 450 Dynamic trace points stats. See kprobetrace.rst. 451 452 max_graph_depth: 453 454 Used with the function graph tracer. This is the max depth 455 it will trace into a function. Setting this to a value of 456 one will show only the first kernel function that is called 457 from user space. 458 459 printk_formats: 460 461 This is for tools that read the raw format files. If an event in 462 the ring buffer references a string, only a pointer to the string 463 is recorded into the buffer and not the string itself. This prevents 464 tools from knowing what that string was. This file displays the string 465 and address for the string allowing tools to map the pointers to what 466 the strings were. 467 468 saved_cmdlines: 469 470 Only the pid of the task is recorded in a trace event unless 471 the event specifically saves the task comm as well. Ftrace 472 makes a cache of pid mappings to comms to try to display 473 comms for events. If a pid for a comm is not listed, then 474 "<...>" is displayed in the output. 475 476 If the option "record-cmd" is set to "0", then comms of tasks 477 will not be saved during recording. By default, it is enabled. 478 479 saved_cmdlines_size: 480 481 By default, 128 comms are saved (see "saved_cmdlines" above). To 482 increase or decrease the amount of comms that are cached, echo 483 the number of comms to cache into this file. 484 485 saved_tgids: 486 487 If the option "record-tgid" is set, on each scheduling context switch 488 the Task Group ID of a task is saved in a table mapping the PID of 489 the thread to its TGID. By default, the "record-tgid" option is 490 disabled. 491 492 snapshot: 493 494 This displays the "snapshot" buffer and also lets the user 495 take a snapshot of the current running trace. 496 See the "Snapshot" section below for more details. 497 498 stack_max_size: 499 500 When the stack tracer is activated, this will display the 501 maximum stack size it has encountered. 502 See the "Stack Trace" section below. 503 504 stack_trace: 505 506 This displays the stack back trace of the largest stack 507 that was encountered when the stack tracer is activated. 508 See the "Stack Trace" section below. 509 510 stack_trace_filter: 511 512 This is similar to "set_ftrace_filter" but it limits what 513 functions the stack tracer will check. 514 515 trace_clock: 516 517 Whenever an event is recorded into the ring buffer, a 518 "timestamp" is added. This stamp comes from a specified 519 clock. By default, ftrace uses the "local" clock. This 520 clock is very fast and strictly per cpu, but on some 521 systems it may not be monotonic with respect to other 522 CPUs. In other words, the local clocks may not be in sync 523 with local clocks on other CPUs. 524 525 Usual clocks for tracing:: 526 527 # cat trace_clock 528 [local] global counter x86-tsc 529 530 The clock with the square brackets around it is the one in effect. 531 532 local: 533 Default clock, but may not be in sync across CPUs 534 535 global: 536 This clock is in sync with all CPUs but may 537 be a bit slower than the local clock. 538 539 counter: 540 This is not a clock at all, but literally an atomic 541 counter. It counts up one by one, but is in sync 542 with all CPUs. This is useful when you need to 543 know exactly the order events occurred with respect to 544 each other on different CPUs. 545 546 uptime: 547 This uses the jiffies counter and the time stamp 548 is relative to the time since boot up. 549 550 perf: 551 This makes ftrace use the same clock that perf uses. 552 Eventually perf will be able to read ftrace buffers 553 and this will help out in interleaving the data. 554 555 x86-tsc: 556 Architectures may define their own clocks. For 557 example, x86 uses its own TSC cycle clock here. 558 559 ppc-tb: 560 This uses the powerpc timebase register value. 561 This is in sync across CPUs and can also be used 562 to correlate events across hypervisor/guest if 563 tb_offset is known. 564 565 mono: 566 This uses the fast monotonic clock (CLOCK_MONOTONIC) 567 which is monotonic and is subject to NTP rate adjustments. 568 569 mono_raw: 570 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW) 571 which is monotonic but is not subject to any rate adjustments 572 and ticks at the same rate as the hardware clocksource. 573 574 boot: 575 This is the boot clock (CLOCK_BOOTTIME) and is based on the 576 fast monotonic clock, but also accounts for time spent in 577 suspend. Since the clock access is designed for use in 578 tracing in the suspend path, some side effects are possible 579 if clock is accessed after the suspend time is accounted before 580 the fast mono clock is updated. In this case, the clock update 581 appears to happen slightly sooner than it normally would have. 582 Also on 32-bit systems, it's possible that the 64-bit boot offset 583 sees a partial update. These effects are rare and post 584 processing should be able to handle them. See comments in the 585 ktime_get_boot_fast_ns() function for more information. 586 587 tai: 588 This is the tai clock (CLOCK_TAI) and is derived from the wall- 589 clock time. However, this clock does not experience 590 discontinuities and backwards jumps caused by NTP inserting leap 591 seconds. Since the clock access is designed for use in tracing, 592 side effects are possible. The clock access may yield wrong 593 readouts in case the internal TAI offset is updated e.g., caused 594 by setting the system time or using adjtimex() with an offset. 595 These effects are rare and post processing should be able to 596 handle them. See comments in the ktime_get_tai_fast_ns() 597 function for more information. 598 599 To set a clock, simply echo the clock name into this file:: 600 601 # echo global > trace_clock 602 603 Setting a clock clears the ring buffer content as well as the 604 "snapshot" buffer. 605 606 trace_marker: 607 608 This is a very useful file for synchronizing user space 609 with events happening in the kernel. Writing strings into 610 this file will be written into the ftrace buffer. 611 612 It is useful in applications to open this file at the start 613 of the application and just reference the file descriptor 614 for the file:: 615 616 void trace_write(const char *fmt, ...) 617 { 618 va_list ap; 619 char buf[256]; 620 int n; 621 622 if (trace_fd < 0) 623 return; 624 625 va_start(ap, fmt); 626 n = vsnprintf(buf, 256, fmt, ap); 627 va_end(ap); 628 629 write(trace_fd, buf, n); 630 } 631 632 start:: 633 634 trace_fd = open("trace_marker", O_WRONLY); 635 636 Note: Writing into the trace_marker file can also initiate triggers 637 that are written into /sys/kernel/tracing/events/ftrace/print/trigger 638 See "Event triggers" in Documentation/trace/events.rst and an 639 example in Documentation/trace/histogram.rst (Section 3.) 640 641 trace_marker_raw: 642 643 This is similar to trace_marker above, but is meant for binary data 644 to be written to it, where a tool can be used to parse the data 645 from trace_pipe_raw. 646 647 uprobe_events: 648 649 Add dynamic tracepoints in programs. 650 See uprobetracer.rst 651 652 uprobe_profile: 653 654 Uprobe statistics. See uprobetrace.txt 655 656 instances: 657 658 This is a way to make multiple trace buffers where different 659 events can be recorded in different buffers. 660 See "Instances" section below. 661 662 events: 663 664 This is the trace event directory. It holds event tracepoints 665 (also known as static tracepoints) that have been compiled 666 into the kernel. It shows what event tracepoints exist 667 and how they are grouped by system. There are "enable" 668 files at various levels that can enable the tracepoints 669 when a "1" is written to them. 670 671 See events.rst for more information. 672 673 set_event: 674 675 By echoing in the event into this file, will enable that event. 676 677 See events.rst for more information. 678 679 available_events: 680 681 A list of events that can be enabled in tracing. 682 683 See events.rst for more information. 684 685 timestamp_mode: 686 687 Certain tracers may change the timestamp mode used when 688 logging trace events into the event buffer. Events with 689 different modes can coexist within a buffer but the mode in 690 effect when an event is logged determines which timestamp mode 691 is used for that event. The default timestamp mode is 692 'delta'. 693 694 Usual timestamp modes for tracing: 695 696 # cat timestamp_mode 697 [delta] absolute 698 699 The timestamp mode with the square brackets around it is the 700 one in effect. 701 702 delta: Default timestamp mode - timestamp is a delta against 703 a per-buffer timestamp. 704 705 absolute: The timestamp is a full timestamp, not a delta 706 against some other value. As such it takes up more 707 space and is less efficient. 708 709 hwlat_detector: 710 711 Directory for the Hardware Latency Detector. 712 See "Hardware Latency Detector" section below. 713 714 per_cpu: 715 716 This is a directory that contains the trace per_cpu information. 717 718 per_cpu/cpu0/buffer_size_kb: 719 720 The ftrace buffer is defined per_cpu. That is, there's a separate 721 buffer for each CPU to allow writes to be done atomically, 722 and free from cache bouncing. These buffers may have different 723 size buffers. This file is similar to the buffer_size_kb 724 file, but it only displays or sets the buffer size for the 725 specific CPU. (here cpu0). 726 727 per_cpu/cpu0/trace: 728 729 This is similar to the "trace" file, but it will only display 730 the data specific for the CPU. If written to, it only clears 731 the specific CPU buffer. 732 733 per_cpu/cpu0/trace_pipe 734 735 This is similar to the "trace_pipe" file, and is a consuming 736 read, but it will only display (and consume) the data specific 737 for the CPU. 738 739 per_cpu/cpu0/trace_pipe_raw 740 741 For tools that can parse the ftrace ring buffer binary format, 742 the trace_pipe_raw file can be used to extract the data 743 from the ring buffer directly. With the use of the splice() 744 system call, the buffer data can be quickly transferred to 745 a file or to the network where a server is collecting the 746 data. 747 748 Like trace_pipe, this is a consuming reader, where multiple 749 reads will always produce different data. 750 751 per_cpu/cpu0/snapshot: 752 753 This is similar to the main "snapshot" file, but will only 754 snapshot the current CPU (if supported). It only displays 755 the content of the snapshot for a given CPU, and if 756 written to, only clears this CPU buffer. 757 758 per_cpu/cpu0/snapshot_raw: 759 760 Similar to the trace_pipe_raw, but will read the binary format 761 from the snapshot buffer for the given CPU. 762 763 per_cpu/cpu0/stats: 764 765 This displays certain stats about the ring buffer: 766 767 entries: 768 The number of events that are still in the buffer. 769 770 overrun: 771 The number of lost events due to overwriting when 772 the buffer was full. 773 774 commit overrun: 775 Should always be zero. 776 This gets set if so many events happened within a nested 777 event (ring buffer is re-entrant), that it fills the 778 buffer and starts dropping events. 779 780 bytes: 781 Bytes actually read (not overwritten). 782 783 oldest event ts: 784 The oldest timestamp in the buffer 785 786 now ts: 787 The current timestamp 788 789 dropped events: 790 Events lost due to overwrite option being off. 791 792 read events: 793 The number of events read. 794 795The Tracers 796----------- 797 798Here is the list of current tracers that may be configured. 799 800 "function" 801 802 Function call tracer to trace all kernel functions. 803 804 "function_graph" 805 806 Similar to the function tracer except that the 807 function tracer probes the functions on their entry 808 whereas the function graph tracer traces on both entry 809 and exit of the functions. It then provides the ability 810 to draw a graph of function calls similar to C code 811 source. 812 813 "blk" 814 815 The block tracer. The tracer used by the blktrace user 816 application. 817 818 "hwlat" 819 820 The Hardware Latency tracer is used to detect if the hardware 821 produces any latency. See "Hardware Latency Detector" section 822 below. 823 824 "irqsoff" 825 826 Traces the areas that disable interrupts and saves 827 the trace with the longest max latency. 828 See tracing_max_latency. When a new max is recorded, 829 it replaces the old trace. It is best to view this 830 trace with the latency-format option enabled, which 831 happens automatically when the tracer is selected. 832 833 "preemptoff" 834 835 Similar to irqsoff but traces and records the amount of 836 time for which preemption is disabled. 837 838 "preemptirqsoff" 839 840 Similar to irqsoff and preemptoff, but traces and 841 records the largest time for which irqs and/or preemption 842 is disabled. 843 844 "wakeup" 845 846 Traces and records the max latency that it takes for 847 the highest priority task to get scheduled after 848 it has been woken up. 849 Traces all tasks as an average developer would expect. 850 851 "wakeup_rt" 852 853 Traces and records the max latency that it takes for just 854 RT tasks (as the current "wakeup" does). This is useful 855 for those interested in wake up timings of RT tasks. 856 857 "wakeup_dl" 858 859 Traces and records the max latency that it takes for 860 a SCHED_DEADLINE task to be woken (as the "wakeup" and 861 "wakeup_rt" does). 862 863 "mmiotrace" 864 865 A special tracer that is used to trace binary module. 866 It will trace all the calls that a module makes to the 867 hardware. Everything it writes and reads from the I/O 868 as well. 869 870 "branch" 871 872 This tracer can be configured when tracing likely/unlikely 873 calls within the kernel. It will trace when a likely and 874 unlikely branch is hit and if it was correct in its prediction 875 of being correct. 876 877 "nop" 878 879 This is the "trace nothing" tracer. To remove all 880 tracers from tracing simply echo "nop" into 881 current_tracer. 882 883Error conditions 884---------------- 885 886 For most ftrace commands, failure modes are obvious and communicated 887 using standard return codes. 888 889 For other more involved commands, extended error information may be 890 available via the tracing/error_log file. For the commands that 891 support it, reading the tracing/error_log file after an error will 892 display more detailed information about what went wrong, if 893 information is available. The tracing/error_log file is a circular 894 error log displaying a small number (currently, 8) of ftrace errors 895 for the last (8) failed commands. 896 897 The extended error information and usage takes the form shown in 898 this example:: 899 900 # echo xxx > /sys/kernel/tracing/events/sched/sched_wakeup/trigger 901 echo: write error: Invalid argument 902 903 # cat /sys/kernel/tracing/error_log 904 [ 5348.887237] location: error: Couldn't yyy: zzz 905 Command: xxx 906 ^ 907 [ 7517.023364] location: error: Bad rrr: sss 908 Command: ppp qqq 909 ^ 910 911 To clear the error log, echo the empty string into it:: 912 913 # echo > /sys/kernel/tracing/error_log 914 915Examples of using the tracer 916---------------------------- 917 918Here are typical examples of using the tracers when controlling 919them only with the tracefs interface (without using any 920user-land utilities). 921 922Output format: 923-------------- 924 925Here is an example of the output format of the file "trace":: 926 927 # tracer: function 928 # 929 # entries-in-buffer/entries-written: 140080/250280 #P:4 930 # 931 # _-----=> irqs-off 932 # / _----=> need-resched 933 # | / _---=> hardirq/softirq 934 # || / _--=> preempt-depth 935 # ||| / delay 936 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 937 # | | | |||| | | 938 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath 939 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close 940 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd 941 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify 942 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock 943 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd 944 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock 945 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd 946 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close 947 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath 948 .... 949 950A header is printed with the tracer name that is represented by 951the trace. In this case the tracer is "function". Then it shows the 952number of events in the buffer as well as the total number of entries 953that were written. The difference is the number of entries that were 954lost due to the buffer filling up (250280 - 140080 = 110200 events 955lost). 956 957The header explains the content of the events. Task name "bash", the task 958PID "1977", the CPU that it was running on "000", the latency format 959(explained below), the timestamp in <secs>.<usecs> format, the 960function name that was traced "sys_close" and the parent function that 961called this function "system_call_fastpath". The timestamp is the time 962at which the function was entered. 963 964Latency trace format 965-------------------- 966 967When the latency-format option is enabled or when one of the latency 968tracers is set, the trace file gives somewhat more information to see 969why a latency happened. Here is a typical trace:: 970 971 # tracer: irqsoff 972 # 973 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 974 # -------------------------------------------------------------------- 975 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 976 # ----------------- 977 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0) 978 # ----------------- 979 # => started at: __lock_task_sighand 980 # => ended at: _raw_spin_unlock_irqrestore 981 # 982 # 983 # _------=> CPU# 984 # / _-----=> irqs-off 985 # | / _----=> need-resched 986 # || / _---=> hardirq/softirq 987 # ||| / _--=> preempt-depth 988 # |||| / delay 989 # cmd pid ||||| time | caller 990 # \ / ||||| \ | / 991 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand 992 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore 993 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore 994 ps-6143 2d..1 306us : <stack trace> 995 => trace_hardirqs_on_caller 996 => trace_hardirqs_on 997 => _raw_spin_unlock_irqrestore 998 => do_task_stat 999 => proc_tgid_stat 1000 => proc_single_show 1001 => seq_read 1002 => vfs_read 1003 => sys_read 1004 => system_call_fastpath 1005 1006 1007This shows that the current tracer is "irqsoff" tracing the time 1008for which interrupts were disabled. It gives the trace version (which 1009never changes) and the version of the kernel upon which this was executed on 1010(3.8). Then it displays the max latency in microseconds (259 us). The number 1011of trace entries displayed and the total number (both are four: #4/4). 1012VP, KP, SP, and HP are always zero and are reserved for later use. 1013#P is the number of online CPUs (#P:4). 1014 1015The task is the process that was running when the latency 1016occurred. (ps pid: 6143). 1017 1018The start and stop (the functions in which the interrupts were 1019disabled and enabled respectively) that caused the latencies: 1020 1021 - __lock_task_sighand is where the interrupts were disabled. 1022 - _raw_spin_unlock_irqrestore is where they were enabled again. 1023 1024The next lines after the header are the trace itself. The header 1025explains which is which. 1026 1027 cmd: The name of the process in the trace. 1028 1029 pid: The PID of that process. 1030 1031 CPU#: The CPU which the process was running on. 1032 1033 irqs-off: 'd' interrupts are disabled. '.' otherwise. 1034 1035 need-resched: 1036 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set, 1037 - 'n' only TIF_NEED_RESCHED is set, 1038 - 'p' only PREEMPT_NEED_RESCHED is set, 1039 - '.' otherwise. 1040 1041 hardirq/softirq: 1042 - 'Z' - NMI occurred inside a hardirq 1043 - 'z' - NMI is running 1044 - 'H' - hard irq occurred inside a softirq. 1045 - 'h' - hard irq is running 1046 - 's' - soft irq is running 1047 - '.' - normal context. 1048 1049 preempt-depth: The level of preempt_disabled 1050 1051The above is mostly meaningful for kernel developers. 1052 1053 time: 1054 When the latency-format option is enabled, the trace file 1055 output includes a timestamp relative to the start of the 1056 trace. This differs from the output when latency-format 1057 is disabled, which includes an absolute timestamp. 1058 1059 delay: 1060 This is just to help catch your eye a bit better. And 1061 needs to be fixed to be only relative to the same CPU. 1062 The marks are determined by the difference between this 1063 current trace and the next trace. 1064 1065 - '$' - greater than 1 second 1066 - '@' - greater than 100 millisecond 1067 - '*' - greater than 10 millisecond 1068 - '#' - greater than 1000 microsecond 1069 - '!' - greater than 100 microsecond 1070 - '+' - greater than 10 microsecond 1071 - ' ' - less than or equal to 10 microsecond. 1072 1073 The rest is the same as the 'trace' file. 1074 1075 Note, the latency tracers will usually end with a back trace 1076 to easily find where the latency occurred. 1077 1078trace_options 1079------------- 1080 1081The trace_options file (or the options directory) is used to control 1082what gets printed in the trace output, or manipulate the tracers. 1083To see what is available, simply cat the file:: 1084 1085 cat trace_options 1086 print-parent 1087 nosym-offset 1088 nosym-addr 1089 noverbose 1090 noraw 1091 nohex 1092 nobin 1093 noblock 1094 nofields 1095 trace_printk 1096 annotate 1097 nouserstacktrace 1098 nosym-userobj 1099 noprintk-msg-only 1100 context-info 1101 nolatency-format 1102 record-cmd 1103 norecord-tgid 1104 overwrite 1105 nodisable_on_free 1106 irq-info 1107 markers 1108 noevent-fork 1109 function-trace 1110 nofunction-fork 1111 nodisplay-graph 1112 nostacktrace 1113 nobranch 1114 1115To disable one of the options, echo in the option prepended with 1116"no":: 1117 1118 echo noprint-parent > trace_options 1119 1120To enable an option, leave off the "no":: 1121 1122 echo sym-offset > trace_options 1123 1124Here are the available options: 1125 1126 print-parent 1127 On function traces, display the calling (parent) 1128 function as well as the function being traced. 1129 :: 1130 1131 print-parent: 1132 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul 1133 1134 noprint-parent: 1135 bash-4000 [01] 1477.606694: simple_strtoul 1136 1137 1138 sym-offset 1139 Display not only the function name, but also the 1140 offset in the function. For example, instead of 1141 seeing just "ktime_get", you will see 1142 "ktime_get+0xb/0x20". 1143 :: 1144 1145 sym-offset: 1146 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0 1147 1148 sym-addr 1149 This will also display the function address as well 1150 as the function name. 1151 :: 1152 1153 sym-addr: 1154 bash-4000 [01] 1477.606694: simple_strtoul <c0339346> 1155 1156 verbose 1157 This deals with the trace file when the 1158 latency-format option is enabled. 1159 :: 1160 1161 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \ 1162 (+0.000ms): simple_strtoul (kstrtoul) 1163 1164 raw 1165 This will display raw numbers. This option is best for 1166 use with user applications that can translate the raw 1167 numbers better than having it done in the kernel. 1168 1169 hex 1170 Similar to raw, but the numbers will be in a hexadecimal format. 1171 1172 bin 1173 This will print out the formats in raw binary. 1174 1175 block 1176 When set, reading trace_pipe will not block when polled. 1177 1178 fields 1179 Print the fields as described by their types. This is a better 1180 option than using hex, bin or raw, as it gives a better parsing 1181 of the content of the event. 1182 1183 trace_printk 1184 Can disable trace_printk() from writing into the buffer. 1185 1186 trace_printk_dest 1187 Set to have trace_printk() and similar internal tracing functions 1188 write into this instance. Note, only one trace instance can have 1189 this set. By setting this flag, it clears the trace_printk_dest flag 1190 of the instance that had it set previously. By default, the top 1191 level trace has this set, and will get it set again if another 1192 instance has it set then clears it. 1193 1194 This flag cannot be cleared by the top level instance, as it is the 1195 default instance. The only way the top level instance has this flag 1196 cleared, is by it being set in another instance. 1197 1198 annotate 1199 It is sometimes confusing when the CPU buffers are full 1200 and one CPU buffer had a lot of events recently, thus 1201 a shorter time frame, were another CPU may have only had 1202 a few events, which lets it have older events. When 1203 the trace is reported, it shows the oldest events first, 1204 and it may look like only one CPU ran (the one with the 1205 oldest events). When the annotate option is set, it will 1206 display when a new CPU buffer started:: 1207 1208 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on 1209 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on 1210 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore 1211 ##### CPU 2 buffer started #### 1212 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle 1213 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog 1214 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock 1215 1216 userstacktrace 1217 This option changes the trace. It records a 1218 stacktrace of the current user space thread after 1219 each trace event. 1220 1221 sym-userobj 1222 when user stacktrace are enabled, look up which 1223 object the address belongs to, and print a 1224 relative address. This is especially useful when 1225 ASLR is on, otherwise you don't get a chance to 1226 resolve the address to object/file/line after 1227 the app is no longer running 1228 1229 The lookup is performed when you read 1230 trace,trace_pipe. Example:: 1231 1232 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0 1233 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6] 1234 1235 1236 printk-msg-only 1237 When set, trace_printk()s will only show the format 1238 and not their parameters (if trace_bprintk() or 1239 trace_bputs() was used to save the trace_printk()). 1240 1241 context-info 1242 Show only the event data. Hides the comm, PID, 1243 timestamp, CPU, and other useful data. 1244 1245 latency-format 1246 This option changes the trace output. When it is enabled, 1247 the trace displays additional information about the 1248 latency, as described in "Latency trace format". 1249 1250 pause-on-trace 1251 When set, opening the trace file for read, will pause 1252 writing to the ring buffer (as if tracing_on was set to zero). 1253 This simulates the original behavior of the trace file. 1254 When the file is closed, tracing will be enabled again. 1255 1256 hash-ptr 1257 When set, "%p" in the event printk format displays the 1258 hashed pointer value instead of real address. 1259 This will be useful if you want to find out which hashed 1260 value is corresponding to the real value in trace log. 1261 1262 record-cmd 1263 When any event or tracer is enabled, a hook is enabled 1264 in the sched_switch trace point to fill comm cache 1265 with mapped pids and comms. But this may cause some 1266 overhead, and if you only care about pids, and not the 1267 name of the task, disabling this option can lower the 1268 impact of tracing. See "saved_cmdlines". 1269 1270 record-tgid 1271 When any event or tracer is enabled, a hook is enabled 1272 in the sched_switch trace point to fill the cache of 1273 mapped Thread Group IDs (TGID) mapping to pids. See 1274 "saved_tgids". 1275 1276 overwrite 1277 This controls what happens when the trace buffer is 1278 full. If "1" (default), the oldest events are 1279 discarded and overwritten. If "0", then the newest 1280 events are discarded. 1281 (see per_cpu/cpu0/stats for overrun and dropped) 1282 1283 disable_on_free 1284 When the free_buffer is closed, tracing will 1285 stop (tracing_on set to 0). 1286 1287 irq-info 1288 Shows the interrupt, preempt count, need resched data. 1289 When disabled, the trace looks like:: 1290 1291 # tracer: function 1292 # 1293 # entries-in-buffer/entries-written: 144405/9452052 #P:4 1294 # 1295 # TASK-PID CPU# TIMESTAMP FUNCTION 1296 # | | | | | 1297 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up 1298 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89 1299 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task 1300 1301 1302 markers 1303 When set, the trace_marker is writable (only by root). 1304 When disabled, the trace_marker will error with EINVAL 1305 on write. 1306 1307 event-fork 1308 When set, tasks with PIDs listed in set_event_pid will have 1309 the PIDs of their children added to set_event_pid when those 1310 tasks fork. Also, when tasks with PIDs in set_event_pid exit, 1311 their PIDs will be removed from the file. 1312 1313 This affects PIDs listed in set_event_notrace_pid as well. 1314 1315 function-trace 1316 The latency tracers will enable function tracing 1317 if this option is enabled (default it is). When 1318 it is disabled, the latency tracers do not trace 1319 functions. This keeps the overhead of the tracer down 1320 when performing latency tests. 1321 1322 function-fork 1323 When set, tasks with PIDs listed in set_ftrace_pid will 1324 have the PIDs of their children added to set_ftrace_pid 1325 when those tasks fork. Also, when tasks with PIDs in 1326 set_ftrace_pid exit, their PIDs will be removed from the 1327 file. 1328 1329 This affects PIDs in set_ftrace_notrace_pid as well. 1330 1331 display-graph 1332 When set, the latency tracers (irqsoff, wakeup, etc) will 1333 use function graph tracing instead of function tracing. 1334 1335 stacktrace 1336 When set, a stack trace is recorded after any trace event 1337 is recorded. 1338 1339 branch 1340 Enable branch tracing with the tracer. This enables branch 1341 tracer along with the currently set tracer. Enabling this 1342 with the "nop" tracer is the same as just enabling the 1343 "branch" tracer. 1344 1345.. tip:: Some tracers have their own options. They only appear in this 1346 file when the tracer is active. They always appear in the 1347 options directory. 1348 1349 1350Here are the per tracer options: 1351 1352Options for function tracer: 1353 1354 func_stack_trace 1355 When set, a stack trace is recorded after every 1356 function that is recorded. NOTE! Limit the functions 1357 that are recorded before enabling this, with 1358 "set_ftrace_filter" otherwise the system performance 1359 will be critically degraded. Remember to disable 1360 this option before clearing the function filter. 1361 1362Options for function_graph tracer: 1363 1364 Since the function_graph tracer has a slightly different output 1365 it has its own options to control what is displayed. 1366 1367 funcgraph-overrun 1368 When set, the "overrun" of the graph stack is 1369 displayed after each function traced. The 1370 overrun, is when the stack depth of the calls 1371 is greater than what is reserved for each task. 1372 Each task has a fixed array of functions to 1373 trace in the call graph. If the depth of the 1374 calls exceeds that, the function is not traced. 1375 The overrun is the number of functions missed 1376 due to exceeding this array. 1377 1378 funcgraph-cpu 1379 When set, the CPU number of the CPU where the trace 1380 occurred is displayed. 1381 1382 funcgraph-overhead 1383 When set, if the function takes longer than 1384 A certain amount, then a delay marker is 1385 displayed. See "delay" above, under the 1386 header description. 1387 1388 funcgraph-proc 1389 Unlike other tracers, the process' command line 1390 is not displayed by default, but instead only 1391 when a task is traced in and out during a context 1392 switch. Enabling this options has the command 1393 of each process displayed at every line. 1394 1395 funcgraph-duration 1396 At the end of each function (the return) 1397 the duration of the amount of time in the 1398 function is displayed in microseconds. 1399 1400 funcgraph-abstime 1401 When set, the timestamp is displayed at each line. 1402 1403 funcgraph-irqs 1404 When disabled, functions that happen inside an 1405 interrupt will not be traced. 1406 1407 funcgraph-tail 1408 When set, the return event will include the function 1409 that it represents. By default this is off, and 1410 only a closing curly bracket "}" is displayed for 1411 the return of a function. 1412 1413 funcgraph-retval 1414 When set, the return value of each traced function 1415 will be printed after an equal sign "=". By default 1416 this is off. 1417 1418 funcgraph-retval-hex 1419 When set, the return value will always be printed 1420 in hexadecimal format. If the option is not set and 1421 the return value is an error code, it will be printed 1422 in signed decimal format; otherwise it will also be 1423 printed in hexadecimal format. By default, this option 1424 is off. 1425 1426 sleep-time 1427 When running function graph tracer, to include 1428 the time a task schedules out in its function. 1429 When enabled, it will account time the task has been 1430 scheduled out as part of the function call. 1431 1432 graph-time 1433 When running function profiler with function graph tracer, 1434 to include the time to call nested functions. When this is 1435 not set, the time reported for the function will only 1436 include the time the function itself executed for, not the 1437 time for functions that it called. 1438 1439Options for blk tracer: 1440 1441 blk_classic 1442 Shows a more minimalistic output. 1443 1444 1445irqsoff 1446------- 1447 1448When interrupts are disabled, the CPU can not react to any other 1449external event (besides NMIs and SMIs). This prevents the timer 1450interrupt from triggering or the mouse interrupt from letting 1451the kernel know of a new mouse event. The result is a latency 1452with the reaction time. 1453 1454The irqsoff tracer tracks the time for which interrupts are 1455disabled. When a new maximum latency is hit, the tracer saves 1456the trace leading up to that latency point so that every time a 1457new maximum is reached, the old saved trace is discarded and the 1458new trace is saved. 1459 1460To reset the maximum, echo 0 into tracing_max_latency. Here is 1461an example:: 1462 1463 # echo 0 > options/function-trace 1464 # echo irqsoff > current_tracer 1465 # echo 1 > tracing_on 1466 # echo 0 > tracing_max_latency 1467 # ls -ltr 1468 [...] 1469 # echo 0 > tracing_on 1470 # cat trace 1471 # tracer: irqsoff 1472 # 1473 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1474 # -------------------------------------------------------------------- 1475 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1476 # ----------------- 1477 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0) 1478 # ----------------- 1479 # => started at: run_timer_softirq 1480 # => ended at: run_timer_softirq 1481 # 1482 # 1483 # _------=> CPU# 1484 # / _-----=> irqs-off 1485 # | / _----=> need-resched 1486 # || / _---=> hardirq/softirq 1487 # ||| / _--=> preempt-depth 1488 # |||| / delay 1489 # cmd pid ||||| time | caller 1490 # \ / ||||| \ | / 1491 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq 1492 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq 1493 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq 1494 <idle>-0 0dNs3 25us : <stack trace> 1495 => _raw_spin_unlock_irq 1496 => run_timer_softirq 1497 => __do_softirq 1498 => call_softirq 1499 => do_softirq 1500 => irq_exit 1501 => smp_apic_timer_interrupt 1502 => apic_timer_interrupt 1503 => rcu_idle_exit 1504 => cpu_idle 1505 => rest_init 1506 => start_kernel 1507 => x86_64_start_reservations 1508 => x86_64_start_kernel 1509 1510Here we see that we had a latency of 16 microseconds (which is 1511very good). The _raw_spin_lock_irq in run_timer_softirq disabled 1512interrupts. The difference between the 16 and the displayed 1513timestamp 25us occurred because the clock was incremented 1514between the time of recording the max latency and the time of 1515recording the function that had that latency. 1516 1517Note the above example had function-trace not set. If we set 1518function-trace, we get a much larger output:: 1519 1520 with echo 1 > options/function-trace 1521 1522 # tracer: irqsoff 1523 # 1524 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1525 # -------------------------------------------------------------------- 1526 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1527 # ----------------- 1528 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0) 1529 # ----------------- 1530 # => started at: ata_scsi_queuecmd 1531 # => ended at: ata_scsi_queuecmd 1532 # 1533 # 1534 # _------=> CPU# 1535 # / _-----=> irqs-off 1536 # | / _----=> need-resched 1537 # || / _---=> hardirq/softirq 1538 # ||| / _--=> preempt-depth 1539 # |||| / delay 1540 # cmd pid ||||| time | caller 1541 # \ / ||||| \ | / 1542 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd 1543 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave 1544 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd 1545 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev 1546 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev 1547 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd 1548 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd 1549 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd 1550 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat 1551 [...] 1552 bash-2042 3d..1 67us : delay_tsc <-__delay 1553 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc 1554 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc 1555 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc 1556 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc 1557 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue 1558 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1559 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1560 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd 1561 bash-2042 3d..1 120us : <stack trace> 1562 => _raw_spin_unlock_irqrestore 1563 => ata_scsi_queuecmd 1564 => scsi_dispatch_cmd 1565 => scsi_request_fn 1566 => __blk_run_queue_uncond 1567 => __blk_run_queue 1568 => blk_queue_bio 1569 => submit_bio_noacct 1570 => submit_bio 1571 => submit_bh 1572 => __ext3_get_inode_loc 1573 => ext3_iget 1574 => ext3_lookup 1575 => lookup_real 1576 => __lookup_hash 1577 => walk_component 1578 => lookup_last 1579 => path_lookupat 1580 => filename_lookup 1581 => user_path_at_empty 1582 => user_path_at 1583 => vfs_fstatat 1584 => vfs_stat 1585 => sys_newstat 1586 => system_call_fastpath 1587 1588 1589Here we traced a 71 microsecond latency. But we also see all the 1590functions that were called during that time. Note that by 1591enabling function tracing, we incur an added overhead. This 1592overhead may extend the latency times. But nevertheless, this 1593trace has provided some very helpful debugging information. 1594 1595If we prefer function graph output instead of function, we can set 1596display-graph option:: 1597 1598 with echo 1 > options/display-graph 1599 1600 # tracer: irqsoff 1601 # 1602 # irqsoff latency trace v1.1.5 on 4.20.0-rc6+ 1603 # -------------------------------------------------------------------- 1604 # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4) 1605 # ----------------- 1606 # | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0) 1607 # ----------------- 1608 # => started at: free_debug_processing 1609 # => ended at: return_to_handler 1610 # 1611 # 1612 # _-----=> irqs-off 1613 # / _----=> need-resched 1614 # | / _---=> hardirq/softirq 1615 # || / _--=> preempt-depth 1616 # ||| / 1617 # REL TIME CPU TASK/PID |||| DURATION FUNCTION CALLS 1618 # | | | | |||| | | | | | | 1619 0 us | 0) bash-1507 | d... | 0.000 us | _raw_spin_lock_irqsave(); 1620 0 us | 0) bash-1507 | d..1 | 0.378 us | do_raw_spin_trylock(); 1621 1 us | 0) bash-1507 | d..2 | | set_track() { 1622 2 us | 0) bash-1507 | d..2 | | save_stack_trace() { 1623 2 us | 0) bash-1507 | d..2 | | __save_stack_trace() { 1624 3 us | 0) bash-1507 | d..2 | | __unwind_start() { 1625 3 us | 0) bash-1507 | d..2 | | get_stack_info() { 1626 3 us | 0) bash-1507 | d..2 | 0.351 us | in_task_stack(); 1627 4 us | 0) bash-1507 | d..2 | 1.107 us | } 1628 [...] 1629 3750 us | 0) bash-1507 | d..1 | 0.516 us | do_raw_spin_unlock(); 1630 3750 us | 0) bash-1507 | d..1 | 0.000 us | _raw_spin_unlock_irqrestore(); 1631 3764 us | 0) bash-1507 | d..1 | 0.000 us | tracer_hardirqs_on(); 1632 bash-1507 0d..1 3792us : <stack trace> 1633 => free_debug_processing 1634 => __slab_free 1635 => kmem_cache_free 1636 => vm_area_free 1637 => remove_vma 1638 => exit_mmap 1639 => mmput 1640 => begin_new_exec 1641 => load_elf_binary 1642 => search_binary_handler 1643 => __do_execve_file.isra.32 1644 => __x64_sys_execve 1645 => do_syscall_64 1646 => entry_SYSCALL_64_after_hwframe 1647 1648preemptoff 1649---------- 1650 1651When preemption is disabled, we may be able to receive 1652interrupts but the task cannot be preempted and a higher 1653priority task must wait for preemption to be enabled again 1654before it can preempt a lower priority task. 1655 1656The preemptoff tracer traces the places that disable preemption. 1657Like the irqsoff tracer, it records the maximum latency for 1658which preemption was disabled. The control of preemptoff tracer 1659is much like the irqsoff tracer. 1660:: 1661 1662 # echo 0 > options/function-trace 1663 # echo preemptoff > current_tracer 1664 # echo 1 > tracing_on 1665 # echo 0 > tracing_max_latency 1666 # ls -ltr 1667 [...] 1668 # echo 0 > tracing_on 1669 # cat trace 1670 # tracer: preemptoff 1671 # 1672 # preemptoff latency trace v1.1.5 on 3.8.0-test+ 1673 # -------------------------------------------------------------------- 1674 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1675 # ----------------- 1676 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0) 1677 # ----------------- 1678 # => started at: do_IRQ 1679 # => ended at: do_IRQ 1680 # 1681 # 1682 # _------=> CPU# 1683 # / _-----=> irqs-off 1684 # | / _----=> need-resched 1685 # || / _---=> hardirq/softirq 1686 # ||| / _--=> preempt-depth 1687 # |||| / delay 1688 # cmd pid ||||| time | caller 1689 # \ / ||||| \ | / 1690 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ 1691 sshd-1991 1d..1 46us : irq_exit <-do_IRQ 1692 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ 1693 sshd-1991 1d..1 52us : <stack trace> 1694 => sub_preempt_count 1695 => irq_exit 1696 => do_IRQ 1697 => ret_from_intr 1698 1699 1700This has some more changes. Preemption was disabled when an 1701interrupt came in (notice the 'h'), and was enabled on exit. 1702But we also see that interrupts have been disabled when entering 1703the preempt off section and leaving it (the 'd'). We do not know if 1704interrupts were enabled in the mean time or shortly after this 1705was over. 1706:: 1707 1708 # tracer: preemptoff 1709 # 1710 # preemptoff latency trace v1.1.5 on 3.8.0-test+ 1711 # -------------------------------------------------------------------- 1712 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1713 # ----------------- 1714 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0) 1715 # ----------------- 1716 # => started at: wake_up_new_task 1717 # => ended at: task_rq_unlock 1718 # 1719 # 1720 # _------=> CPU# 1721 # / _-----=> irqs-off 1722 # | / _----=> need-resched 1723 # || / _---=> hardirq/softirq 1724 # ||| / _--=> preempt-depth 1725 # |||| / delay 1726 # cmd pid ||||| time | caller 1727 # \ / ||||| \ | / 1728 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task 1729 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq 1730 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair 1731 bash-1994 1d..1 1us : source_load <-select_task_rq_fair 1732 bash-1994 1d..1 1us : source_load <-select_task_rq_fair 1733 [...] 1734 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt 1735 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter 1736 bash-1994 1d..1 13us : add_preempt_count <-irq_enter 1737 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt 1738 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt 1739 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt 1740 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock 1741 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt 1742 [...] 1743 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event 1744 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt 1745 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit 1746 bash-1994 1d..2 36us : do_softirq <-irq_exit 1747 bash-1994 1d..2 36us : __do_softirq <-call_softirq 1748 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq 1749 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq 1750 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq 1751 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock 1752 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq 1753 [...] 1754 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks 1755 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq 1756 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable 1757 bash-1994 1dN.2 82us : idle_cpu <-irq_exit 1758 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit 1759 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit 1760 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock 1761 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock 1762 bash-1994 1.N.1 104us : <stack trace> 1763 => sub_preempt_count 1764 => _raw_spin_unlock_irqrestore 1765 => task_rq_unlock 1766 => wake_up_new_task 1767 => do_fork 1768 => sys_clone 1769 => stub_clone 1770 1771 1772The above is an example of the preemptoff trace with 1773function-trace set. Here we see that interrupts were not disabled 1774the entire time. The irq_enter code lets us know that we entered 1775an interrupt 'h'. Before that, the functions being traced still 1776show that it is not in an interrupt, but we can see from the 1777functions themselves that this is not the case. 1778 1779preemptirqsoff 1780-------------- 1781 1782Knowing the locations that have interrupts disabled or 1783preemption disabled for the longest times is helpful. But 1784sometimes we would like to know when either preemption and/or 1785interrupts are disabled. 1786 1787Consider the following code:: 1788 1789 local_irq_disable(); 1790 call_function_with_irqs_off(); 1791 preempt_disable(); 1792 call_function_with_irqs_and_preemption_off(); 1793 local_irq_enable(); 1794 call_function_with_preemption_off(); 1795 preempt_enable(); 1796 1797The irqsoff tracer will record the total length of 1798call_function_with_irqs_off() and 1799call_function_with_irqs_and_preemption_off(). 1800 1801The preemptoff tracer will record the total length of 1802call_function_with_irqs_and_preemption_off() and 1803call_function_with_preemption_off(). 1804 1805But neither will trace the time that interrupts and/or 1806preemption is disabled. This total time is the time that we can 1807not schedule. To record this time, use the preemptirqsoff 1808tracer. 1809 1810Again, using this trace is much like the irqsoff and preemptoff 1811tracers. 1812:: 1813 1814 # echo 0 > options/function-trace 1815 # echo preemptirqsoff > current_tracer 1816 # echo 1 > tracing_on 1817 # echo 0 > tracing_max_latency 1818 # ls -ltr 1819 [...] 1820 # echo 0 > tracing_on 1821 # cat trace 1822 # tracer: preemptirqsoff 1823 # 1824 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+ 1825 # -------------------------------------------------------------------- 1826 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1827 # ----------------- 1828 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0) 1829 # ----------------- 1830 # => started at: ata_scsi_queuecmd 1831 # => ended at: ata_scsi_queuecmd 1832 # 1833 # 1834 # _------=> CPU# 1835 # / _-----=> irqs-off 1836 # | / _----=> need-resched 1837 # || / _---=> hardirq/softirq 1838 # ||| / _--=> preempt-depth 1839 # |||| / delay 1840 # cmd pid ||||| time | caller 1841 # \ / ||||| \ | / 1842 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd 1843 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1844 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd 1845 ls-2230 3...1 111us : <stack trace> 1846 => sub_preempt_count 1847 => _raw_spin_unlock_irqrestore 1848 => ata_scsi_queuecmd 1849 => scsi_dispatch_cmd 1850 => scsi_request_fn 1851 => __blk_run_queue_uncond 1852 => __blk_run_queue 1853 => blk_queue_bio 1854 => submit_bio_noacct 1855 => submit_bio 1856 => submit_bh 1857 => ext3_bread 1858 => ext3_dir_bread 1859 => htree_dirblock_to_tree 1860 => ext3_htree_fill_tree 1861 => ext3_readdir 1862 => vfs_readdir 1863 => sys_getdents 1864 => system_call_fastpath 1865 1866 1867The trace_hardirqs_off_thunk is called from assembly on x86 when 1868interrupts are disabled in the assembly code. Without the 1869function tracing, we do not know if interrupts were enabled 1870within the preemption points. We do see that it started with 1871preemption enabled. 1872 1873Here is a trace with function-trace set:: 1874 1875 # tracer: preemptirqsoff 1876 # 1877 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+ 1878 # -------------------------------------------------------------------- 1879 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1880 # ----------------- 1881 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0) 1882 # ----------------- 1883 # => started at: schedule 1884 # => ended at: mutex_unlock 1885 # 1886 # 1887 # _------=> CPU# 1888 # / _-----=> irqs-off 1889 # | / _----=> need-resched 1890 # || / _---=> hardirq/softirq 1891 # ||| / _--=> preempt-depth 1892 # |||| / delay 1893 # cmd pid ||||| time | caller 1894 # \ / ||||| \ | / 1895 kworker/-59 3...1 0us : __schedule <-schedule 1896 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch 1897 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq 1898 kworker/-59 3d..2 1us : deactivate_task <-__schedule 1899 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task 1900 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task 1901 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task 1902 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair 1903 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr 1904 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr 1905 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge 1906 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge 1907 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair 1908 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair 1909 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair 1910 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair 1911 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair 1912 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair 1913 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule 1914 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping 1915 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule 1916 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task 1917 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair 1918 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair 1919 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity 1920 ls-2269 3d..2 7us : finish_task_switch <-__schedule 1921 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch 1922 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr 1923 ls-2269 3d..2 8us : irq_enter <-do_IRQ 1924 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter 1925 ls-2269 3d..2 9us : add_preempt_count <-irq_enter 1926 ls-2269 3d.h2 9us : exit_idle <-do_IRQ 1927 [...] 1928 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock 1929 ls-2269 3d.h2 20us : irq_exit <-do_IRQ 1930 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit 1931 ls-2269 3d..3 21us : do_softirq <-irq_exit 1932 ls-2269 3d..3 21us : __do_softirq <-call_softirq 1933 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq 1934 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip 1935 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip 1936 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr 1937 ls-2269 3d.s5 31us : irq_enter <-do_IRQ 1938 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter 1939 [...] 1940 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter 1941 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter 1942 ls-2269 3d.H5 32us : exit_idle <-do_IRQ 1943 ls-2269 3d.H5 32us : handle_irq <-do_IRQ 1944 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq 1945 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq 1946 [...] 1947 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll 1948 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action 1949 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq 1950 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable 1951 ls-2269 3d..3 159us : idle_cpu <-irq_exit 1952 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit 1953 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit 1954 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock 1955 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock 1956 ls-2269 3d... 186us : <stack trace> 1957 => __mutex_unlock_slowpath 1958 => mutex_unlock 1959 => process_output 1960 => n_tty_write 1961 => tty_write 1962 => vfs_write 1963 => sys_write 1964 => system_call_fastpath 1965 1966This is an interesting trace. It started with kworker running and 1967scheduling out and ls taking over. But as soon as ls released the 1968rq lock and enabled interrupts (but not preemption) an interrupt 1969triggered. When the interrupt finished, it started running softirqs. 1970But while the softirq was running, another interrupt triggered. 1971When an interrupt is running inside a softirq, the annotation is 'H'. 1972 1973 1974wakeup 1975------ 1976 1977One common case that people are interested in tracing is the 1978time it takes for a task that is woken to actually wake up. 1979Now for non Real-Time tasks, this can be arbitrary. But tracing 1980it nonetheless can be interesting. 1981 1982Without function tracing:: 1983 1984 # echo 0 > options/function-trace 1985 # echo wakeup > current_tracer 1986 # echo 1 > tracing_on 1987 # echo 0 > tracing_max_latency 1988 # chrt -f 5 sleep 1 1989 # echo 0 > tracing_on 1990 # cat trace 1991 # tracer: wakeup 1992 # 1993 # wakeup latency trace v1.1.5 on 3.8.0-test+ 1994 # -------------------------------------------------------------------- 1995 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1996 # ----------------- 1997 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0) 1998 # ----------------- 1999 # 2000 # _------=> CPU# 2001 # / _-----=> irqs-off 2002 # | / _----=> need-resched 2003 # || / _---=> hardirq/softirq 2004 # ||| / _--=> preempt-depth 2005 # |||| / delay 2006 # cmd pid ||||| time | caller 2007 # \ / ||||| \ | / 2008 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H 2009 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up 2010 <idle>-0 3d..3 15us : __schedule <-schedule 2011 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H 2012 2013The tracer only traces the highest priority task in the system 2014to avoid tracing the normal circumstances. Here we see that 2015the kworker with a nice priority of -20 (not very nice), took 2016just 15 microseconds from the time it woke up, to the time it 2017ran. 2018 2019Non Real-Time tasks are not that interesting. A more interesting 2020trace is to concentrate only on Real-Time tasks. 2021 2022wakeup_rt 2023--------- 2024 2025In a Real-Time environment it is very important to know the 2026wakeup time it takes for the highest priority task that is woken 2027up to the time that it executes. This is also known as "schedule 2028latency". I stress the point that this is about RT tasks. It is 2029also important to know the scheduling latency of non-RT tasks, 2030but the average schedule latency is better for non-RT tasks. 2031Tools like LatencyTop are more appropriate for such 2032measurements. 2033 2034Real-Time environments are interested in the worst case latency. 2035That is the longest latency it takes for something to happen, 2036and not the average. We can have a very fast scheduler that may 2037only have a large latency once in a while, but that would not 2038work well with Real-Time tasks. The wakeup_rt tracer was designed 2039to record the worst case wakeups of RT tasks. Non-RT tasks are 2040not recorded because the tracer only records one worst case and 2041tracing non-RT tasks that are unpredictable will overwrite the 2042worst case latency of RT tasks (just run the normal wakeup 2043tracer for a while to see that effect). 2044 2045Since this tracer only deals with RT tasks, we will run this 2046slightly differently than we did with the previous tracers. 2047Instead of performing an 'ls', we will run 'sleep 1' under 2048'chrt' which changes the priority of the task. 2049:: 2050 2051 # echo 0 > options/function-trace 2052 # echo wakeup_rt > current_tracer 2053 # echo 1 > tracing_on 2054 # echo 0 > tracing_max_latency 2055 # chrt -f 5 sleep 1 2056 # echo 0 > tracing_on 2057 # cat trace 2058 # tracer: wakeup 2059 # 2060 # tracer: wakeup_rt 2061 # 2062 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2063 # -------------------------------------------------------------------- 2064 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2065 # ----------------- 2066 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5) 2067 # ----------------- 2068 # 2069 # _------=> CPU# 2070 # / _-----=> irqs-off 2071 # | / _----=> need-resched 2072 # || / _---=> hardirq/softirq 2073 # ||| / _--=> preempt-depth 2074 # |||| / delay 2075 # cmd pid ||||| time | caller 2076 # \ / ||||| \ | / 2077 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep 2078 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up 2079 <idle>-0 3d..3 5us : __schedule <-schedule 2080 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep 2081 2082 2083Running this on an idle system, we see that it only took 5 microseconds 2084to perform the task switch. Note, since the trace point in the schedule 2085is before the actual "switch", we stop the tracing when the recorded task 2086is about to schedule in. This may change if we add a new marker at the 2087end of the scheduler. 2088 2089Notice that the recorded task is 'sleep' with the PID of 2389 2090and it has an rt_prio of 5. This priority is user-space priority 2091and not the internal kernel priority. The policy is 1 for 2092SCHED_FIFO and 2 for SCHED_RR. 2093 2094Note, that the trace data shows the internal priority (99 - rtprio). 2095:: 2096 2097 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep 2098 2099The 0:120:R means idle was running with a nice priority of 0 (120 - 120) 2100and in the running state 'R'. The sleep task was scheduled in with 21012389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94) 2102and it too is in the running state. 2103 2104Doing the same with chrt -r 5 and function-trace set. 2105:: 2106 2107 echo 1 > options/function-trace 2108 2109 # tracer: wakeup_rt 2110 # 2111 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2112 # -------------------------------------------------------------------- 2113 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2114 # ----------------- 2115 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5) 2116 # ----------------- 2117 # 2118 # _------=> CPU# 2119 # / _-----=> irqs-off 2120 # | / _----=> need-resched 2121 # || / _---=> hardirq/softirq 2122 # ||| / _--=> preempt-depth 2123 # |||| / delay 2124 # cmd pid ||||| time | caller 2125 # \ / ||||| \ | / 2126 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep 2127 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up 2128 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup 2129 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr 2130 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup 2131 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up 2132 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock 2133 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up 2134 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up 2135 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2136 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer 2137 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock 2138 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt 2139 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock 2140 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt 2141 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event 2142 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event 2143 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event 2144 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt 2145 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit 2146 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit 2147 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit 2148 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit 2149 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit 2150 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle 2151 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit 2152 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle 2153 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit 2154 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit 2155 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit 2156 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit 2157 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz 2158 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock 2159 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz 2160 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update 2161 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz 2162 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock 2163 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit 2164 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit 2165 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit 2166 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel 2167 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel 2168 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18 2169 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave 2170 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16 2171 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer 2172 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram 2173 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event 2174 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event 2175 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event 2176 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel 2177 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2178 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit 2179 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward 2180 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward 2181 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11 2182 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns 2183 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns 2184 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18 2185 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave 2186 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns 2187 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns 2188 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns 2189 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event 2190 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event 2191 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event 2192 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns 2193 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2194 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit 2195 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks 2196 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle 2197 <idle>-0 3.N.. 25us : schedule <-cpu_idle 2198 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule 2199 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule 2200 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule 2201 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch 2202 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch 2203 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule 2204 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq 2205 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule 2206 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task 2207 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task 2208 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt 2209 <idle>-0 3d..3 29us : __schedule <-preempt_schedule 2210 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep 2211 2212This isn't that big of a trace, even with function tracing enabled, 2213so I included the entire trace. 2214 2215The interrupt went off while when the system was idle. Somewhere 2216before task_woken_rt() was called, the NEED_RESCHED flag was set, 2217this is indicated by the first occurrence of the 'N' flag. 2218 2219Latency tracing and events 2220-------------------------- 2221As function tracing can induce a much larger latency, but without 2222seeing what happens within the latency it is hard to know what 2223caused it. There is a middle ground, and that is with enabling 2224events. 2225:: 2226 2227 # echo 0 > options/function-trace 2228 # echo wakeup_rt > current_tracer 2229 # echo 1 > events/enable 2230 # echo 1 > tracing_on 2231 # echo 0 > tracing_max_latency 2232 # chrt -f 5 sleep 1 2233 # echo 0 > tracing_on 2234 # cat trace 2235 # tracer: wakeup_rt 2236 # 2237 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2238 # -------------------------------------------------------------------- 2239 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2240 # ----------------- 2241 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5) 2242 # ----------------- 2243 # 2244 # _------=> CPU# 2245 # / _-----=> irqs-off 2246 # | / _----=> need-resched 2247 # || / _---=> hardirq/softirq 2248 # ||| / _--=> preempt-depth 2249 # |||| / delay 2250 # cmd pid ||||| time | caller 2251 # \ / ||||| \ | / 2252 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep 2253 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up 2254 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002 2255 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8 2256 <idle>-0 2.N.2 2us : power_end: cpu_id=2 2257 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2 2258 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0 2259 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000 2260 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch 2261 <idle>-0 2.N.2 5us : rcu_utilization: End context switch 2262 <idle>-0 2d..3 6us : __schedule <-schedule 2263 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep 2264 2265 2266Hardware Latency Detector 2267------------------------- 2268 2269The hardware latency detector is executed by enabling the "hwlat" tracer. 2270 2271NOTE, this tracer will affect the performance of the system as it will 2272periodically make a CPU constantly busy with interrupts disabled. 2273:: 2274 2275 # echo hwlat > current_tracer 2276 # sleep 100 2277 # cat trace 2278 # tracer: hwlat 2279 # 2280 # entries-in-buffer/entries-written: 13/13 #P:8 2281 # 2282 # _-----=> irqs-off 2283 # / _----=> need-resched 2284 # | / _---=> hardirq/softirq 2285 # || / _--=> preempt-depth 2286 # ||| / delay 2287 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2288 # | | | |||| | | 2289 <...>-1729 [001] d... 678.473449: #1 inner/outer(us): 11/12 ts:1581527483.343962693 count:6 2290 <...>-1729 [004] d... 689.556542: #2 inner/outer(us): 16/9 ts:1581527494.889008092 count:1 2291 <...>-1729 [005] d... 714.756290: #3 inner/outer(us): 16/16 ts:1581527519.678961629 count:5 2292 <...>-1729 [001] d... 718.788247: #4 inner/outer(us): 9/17 ts:1581527523.889012713 count:1 2293 <...>-1729 [002] d... 719.796341: #5 inner/outer(us): 13/9 ts:1581527524.912872606 count:1 2294 <...>-1729 [006] d... 844.787091: #6 inner/outer(us): 9/12 ts:1581527649.889048502 count:2 2295 <...>-1729 [003] d... 849.827033: #7 inner/outer(us): 18/9 ts:1581527654.889013793 count:1 2296 <...>-1729 [007] d... 853.859002: #8 inner/outer(us): 9/12 ts:1581527658.889065736 count:1 2297 <...>-1729 [001] d... 855.874978: #9 inner/outer(us): 9/11 ts:1581527660.861991877 count:1 2298 <...>-1729 [001] d... 863.938932: #10 inner/outer(us): 9/11 ts:1581527668.970010500 count:1 nmi-total:7 nmi-count:1 2299 <...>-1729 [007] d... 878.050780: #11 inner/outer(us): 9/12 ts:1581527683.385002600 count:1 nmi-total:5 nmi-count:1 2300 <...>-1729 [007] d... 886.114702: #12 inner/outer(us): 9/12 ts:1581527691.385001600 count:1 2301 2302 2303The above output is somewhat the same in the header. All events will have 2304interrupts disabled 'd'. Under the FUNCTION title there is: 2305 2306 #1 2307 This is the count of events recorded that were greater than the 2308 tracing_threshold (See below). 2309 2310 inner/outer(us): 11/11 2311 2312 This shows two numbers as "inner latency" and "outer latency". The test 2313 runs in a loop checking a timestamp twice. The latency detected within 2314 the two timestamps is the "inner latency" and the latency detected 2315 after the previous timestamp and the next timestamp in the loop is 2316 the "outer latency". 2317 2318 ts:1581527483.343962693 2319 2320 The absolute timestamp that the first latency was recorded in the window. 2321 2322 count:6 2323 2324 The number of times a latency was detected during the window. 2325 2326 nmi-total:7 nmi-count:1 2327 2328 On architectures that support it, if an NMI comes in during the 2329 test, the time spent in NMI is reported in "nmi-total" (in 2330 microseconds). 2331 2332 All architectures that have NMIs will show the "nmi-count" if an 2333 NMI comes in during the test. 2334 2335hwlat files: 2336 2337 tracing_threshold 2338 This gets automatically set to "10" to represent 10 2339 microseconds. This is the threshold of latency that 2340 needs to be detected before the trace will be recorded. 2341 2342 Note, when hwlat tracer is finished (another tracer is 2343 written into "current_tracer"), the original value for 2344 tracing_threshold is placed back into this file. 2345 2346 hwlat_detector/width 2347 The length of time the test runs with interrupts disabled. 2348 2349 hwlat_detector/window 2350 The length of time of the window which the test 2351 runs. That is, the test will run for "width" 2352 microseconds per "window" microseconds 2353 2354 tracing_cpumask 2355 When the test is started. A kernel thread is created that 2356 runs the test. This thread will alternate between CPUs 2357 listed in the tracing_cpumask between each period 2358 (one "window"). To limit the test to specific CPUs 2359 set the mask in this file to only the CPUs that the test 2360 should run on. 2361 2362function 2363-------- 2364 2365This tracer is the function tracer. Enabling the function tracer 2366can be done from the debug file system. Make sure the 2367ftrace_enabled is set; otherwise this tracer is a nop. 2368See the "ftrace_enabled" section below. 2369:: 2370 2371 # sysctl kernel.ftrace_enabled=1 2372 # echo function > current_tracer 2373 # echo 1 > tracing_on 2374 # usleep 1 2375 # echo 0 > tracing_on 2376 # cat trace 2377 # tracer: function 2378 # 2379 # entries-in-buffer/entries-written: 24799/24799 #P:4 2380 # 2381 # _-----=> irqs-off 2382 # / _----=> need-resched 2383 # | / _---=> hardirq/softirq 2384 # || / _--=> preempt-depth 2385 # ||| / delay 2386 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2387 # | | | |||| | | 2388 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write 2389 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock 2390 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify 2391 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify 2392 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify 2393 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock 2394 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock 2395 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify 2396 [...] 2397 2398 2399Note: function tracer uses ring buffers to store the above 2400entries. The newest data may overwrite the oldest data. 2401Sometimes using echo to stop the trace is not sufficient because 2402the tracing could have overwritten the data that you wanted to 2403record. For this reason, it is sometimes better to disable 2404tracing directly from a program. This allows you to stop the 2405tracing at the point that you hit the part that you are 2406interested in. To disable the tracing directly from a C program, 2407something like following code snippet can be used:: 2408 2409 int trace_fd; 2410 [...] 2411 int main(int argc, char *argv[]) { 2412 [...] 2413 trace_fd = open(tracing_file("tracing_on"), O_WRONLY); 2414 [...] 2415 if (condition_hit()) { 2416 write(trace_fd, "0", 1); 2417 } 2418 [...] 2419 } 2420 2421 2422Single thread tracing 2423--------------------- 2424 2425By writing into set_ftrace_pid you can trace a 2426single thread. For example:: 2427 2428 # cat set_ftrace_pid 2429 no pid 2430 # echo 3111 > set_ftrace_pid 2431 # cat set_ftrace_pid 2432 3111 2433 # echo function > current_tracer 2434 # cat trace | head 2435 # tracer: function 2436 # 2437 # TASK-PID CPU# TIMESTAMP FUNCTION 2438 # | | | | | 2439 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return 2440 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range 2441 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel 2442 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel 2443 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll 2444 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll 2445 # echo > set_ftrace_pid 2446 # cat trace |head 2447 # tracer: function 2448 # 2449 # TASK-PID CPU# TIMESTAMP FUNCTION 2450 # | | | | | 2451 ##### CPU 3 buffer started #### 2452 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait 2453 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry 2454 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry 2455 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit 2456 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit 2457 2458If you want to trace a function when executing, you could use 2459something like this simple program. 2460:: 2461 2462 #include <stdio.h> 2463 #include <stdlib.h> 2464 #include <sys/types.h> 2465 #include <sys/stat.h> 2466 #include <fcntl.h> 2467 #include <unistd.h> 2468 #include <string.h> 2469 2470 #define _STR(x) #x 2471 #define STR(x) _STR(x) 2472 #define MAX_PATH 256 2473 2474 const char *find_tracefs(void) 2475 { 2476 static char tracefs[MAX_PATH+1]; 2477 static int tracefs_found; 2478 char type[100]; 2479 FILE *fp; 2480 2481 if (tracefs_found) 2482 return tracefs; 2483 2484 if ((fp = fopen("/proc/mounts","r")) == NULL) { 2485 perror("/proc/mounts"); 2486 return NULL; 2487 } 2488 2489 while (fscanf(fp, "%*s %" 2490 STR(MAX_PATH) 2491 "s %99s %*s %*d %*d\n", 2492 tracefs, type) == 2) { 2493 if (strcmp(type, "tracefs") == 0) 2494 break; 2495 } 2496 fclose(fp); 2497 2498 if (strcmp(type, "tracefs") != 0) { 2499 fprintf(stderr, "tracefs not mounted"); 2500 return NULL; 2501 } 2502 2503 strcat(tracefs, "/tracing/"); 2504 tracefs_found = 1; 2505 2506 return tracefs; 2507 } 2508 2509 const char *tracing_file(const char *file_name) 2510 { 2511 static char trace_file[MAX_PATH+1]; 2512 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name); 2513 return trace_file; 2514 } 2515 2516 int main (int argc, char **argv) 2517 { 2518 if (argc < 1) 2519 exit(-1); 2520 2521 if (fork() > 0) { 2522 int fd, ffd; 2523 char line[64]; 2524 int s; 2525 2526 ffd = open(tracing_file("current_tracer"), O_WRONLY); 2527 if (ffd < 0) 2528 exit(-1); 2529 write(ffd, "nop", 3); 2530 2531 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY); 2532 s = sprintf(line, "%d\n", getpid()); 2533 write(fd, line, s); 2534 2535 write(ffd, "function", 8); 2536 2537 close(fd); 2538 close(ffd); 2539 2540 execvp(argv[1], argv+1); 2541 } 2542 2543 return 0; 2544 } 2545 2546Or this simple script! 2547:: 2548 2549 #!/bin/bash 2550 2551 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts` 2552 echo 0 > $tracefs/tracing_on 2553 echo $$ > $tracefs/set_ftrace_pid 2554 echo function > $tracefs/current_tracer 2555 echo 1 > $tracefs/tracing_on 2556 exec "$@" 2557 2558 2559function graph tracer 2560--------------------------- 2561 2562This tracer is similar to the function tracer except that it 2563probes a function on its entry and its exit. This is done by 2564using a dynamically allocated stack of return addresses in each 2565task_struct. On function entry the tracer overwrites the return 2566address of each function traced to set a custom probe. Thus the 2567original return address is stored on the stack of return address 2568in the task_struct. 2569 2570Probing on both ends of a function leads to special features 2571such as: 2572 2573- measure of a function's time execution 2574- having a reliable call stack to draw function calls graph 2575 2576This tracer is useful in several situations: 2577 2578- you want to find the reason of a strange kernel behavior and 2579 need to see what happens in detail on any areas (or specific 2580 ones). 2581 2582- you are experiencing weird latencies but it's difficult to 2583 find its origin. 2584 2585- you want to find quickly which path is taken by a specific 2586 function 2587 2588- you just want to peek inside a working kernel and want to see 2589 what happens there. 2590 2591:: 2592 2593 # tracer: function_graph 2594 # 2595 # CPU DURATION FUNCTION CALLS 2596 # | | | | | | | 2597 2598 0) | sys_open() { 2599 0) | do_sys_open() { 2600 0) | getname() { 2601 0) | kmem_cache_alloc() { 2602 0) 1.382 us | __might_sleep(); 2603 0) 2.478 us | } 2604 0) | strncpy_from_user() { 2605 0) | might_fault() { 2606 0) 1.389 us | __might_sleep(); 2607 0) 2.553 us | } 2608 0) 3.807 us | } 2609 0) 7.876 us | } 2610 0) | alloc_fd() { 2611 0) 0.668 us | _spin_lock(); 2612 0) 0.570 us | expand_files(); 2613 0) 0.586 us | _spin_unlock(); 2614 2615 2616There are several columns that can be dynamically 2617enabled/disabled. You can use every combination of options you 2618want, depending on your needs. 2619 2620- The cpu number on which the function executed is default 2621 enabled. It is sometimes better to only trace one cpu (see 2622 tracing_cpumask file) or you might sometimes see unordered 2623 function calls while cpu tracing switch. 2624 2625 - hide: echo nofuncgraph-cpu > trace_options 2626 - show: echo funcgraph-cpu > trace_options 2627 2628- The duration (function's time of execution) is displayed on 2629 the closing bracket line of a function or on the same line 2630 than the current function in case of a leaf one. It is default 2631 enabled. 2632 2633 - hide: echo nofuncgraph-duration > trace_options 2634 - show: echo funcgraph-duration > trace_options 2635 2636- The overhead field precedes the duration field in case of 2637 reached duration thresholds. 2638 2639 - hide: echo nofuncgraph-overhead > trace_options 2640 - show: echo funcgraph-overhead > trace_options 2641 - depends on: funcgraph-duration 2642 2643 ie:: 2644 2645 3) # 1837.709 us | } /* __switch_to */ 2646 3) | finish_task_switch() { 2647 3) 0.313 us | _raw_spin_unlock_irq(); 2648 3) 3.177 us | } 2649 3) # 1889.063 us | } /* __schedule */ 2650 3) ! 140.417 us | } /* __schedule */ 2651 3) # 2034.948 us | } /* schedule */ 2652 3) * 33998.59 us | } /* schedule_preempt_disabled */ 2653 2654 [...] 2655 2656 1) 0.260 us | msecs_to_jiffies(); 2657 1) 0.313 us | __rcu_read_unlock(); 2658 1) + 61.770 us | } 2659 1) + 64.479 us | } 2660 1) 0.313 us | rcu_bh_qs(); 2661 1) 0.313 us | __local_bh_enable(); 2662 1) ! 217.240 us | } 2663 1) 0.365 us | idle_cpu(); 2664 1) | rcu_irq_exit() { 2665 1) 0.417 us | rcu_eqs_enter_common.isra.47(); 2666 1) 3.125 us | } 2667 1) ! 227.812 us | } 2668 1) ! 457.395 us | } 2669 1) @ 119760.2 us | } 2670 2671 [...] 2672 2673 2) | handle_IPI() { 2674 1) 6.979 us | } 2675 2) 0.417 us | scheduler_ipi(); 2676 1) 9.791 us | } 2677 1) + 12.917 us | } 2678 2) 3.490 us | } 2679 1) + 15.729 us | } 2680 1) + 18.542 us | } 2681 2) $ 3594274 us | } 2682 2683Flags:: 2684 2685 + means that the function exceeded 10 usecs. 2686 ! means that the function exceeded 100 usecs. 2687 # means that the function exceeded 1000 usecs. 2688 * means that the function exceeded 10 msecs. 2689 @ means that the function exceeded 100 msecs. 2690 $ means that the function exceeded 1 sec. 2691 2692 2693- The task/pid field displays the thread cmdline and pid which 2694 executed the function. It is default disabled. 2695 2696 - hide: echo nofuncgraph-proc > trace_options 2697 - show: echo funcgraph-proc > trace_options 2698 2699 ie:: 2700 2701 # tracer: function_graph 2702 # 2703 # CPU TASK/PID DURATION FUNCTION CALLS 2704 # | | | | | | | | | 2705 0) sh-4802 | | d_free() { 2706 0) sh-4802 | | call_rcu() { 2707 0) sh-4802 | | __call_rcu() { 2708 0) sh-4802 | 0.616 us | rcu_process_gp_end(); 2709 0) sh-4802 | 0.586 us | check_for_new_grace_period(); 2710 0) sh-4802 | 2.899 us | } 2711 0) sh-4802 | 4.040 us | } 2712 0) sh-4802 | 5.151 us | } 2713 0) sh-4802 | + 49.370 us | } 2714 2715 2716- The absolute time field is an absolute timestamp given by the 2717 system clock since it started. A snapshot of this time is 2718 given on each entry/exit of functions 2719 2720 - hide: echo nofuncgraph-abstime > trace_options 2721 - show: echo funcgraph-abstime > trace_options 2722 2723 ie:: 2724 2725 # 2726 # TIME CPU DURATION FUNCTION CALLS 2727 # | | | | | | | | 2728 360.774522 | 1) 0.541 us | } 2729 360.774522 | 1) 4.663 us | } 2730 360.774523 | 1) 0.541 us | __wake_up_bit(); 2731 360.774524 | 1) 6.796 us | } 2732 360.774524 | 1) 7.952 us | } 2733 360.774525 | 1) 9.063 us | } 2734 360.774525 | 1) 0.615 us | journal_mark_dirty(); 2735 360.774527 | 1) 0.578 us | __brelse(); 2736 360.774528 | 1) | reiserfs_prepare_for_journal() { 2737 360.774528 | 1) | unlock_buffer() { 2738 360.774529 | 1) | wake_up_bit() { 2739 360.774529 | 1) | bit_waitqueue() { 2740 360.774530 | 1) 0.594 us | __phys_addr(); 2741 2742 2743The function name is always displayed after the closing bracket 2744for a function if the start of that function is not in the 2745trace buffer. 2746 2747Display of the function name after the closing bracket may be 2748enabled for functions whose start is in the trace buffer, 2749allowing easier searching with grep for function durations. 2750It is default disabled. 2751 2752 - hide: echo nofuncgraph-tail > trace_options 2753 - show: echo funcgraph-tail > trace_options 2754 2755 Example with nofuncgraph-tail (default):: 2756 2757 0) | putname() { 2758 0) | kmem_cache_free() { 2759 0) 0.518 us | __phys_addr(); 2760 0) 1.757 us | } 2761 0) 2.861 us | } 2762 2763 Example with funcgraph-tail:: 2764 2765 0) | putname() { 2766 0) | kmem_cache_free() { 2767 0) 0.518 us | __phys_addr(); 2768 0) 1.757 us | } /* kmem_cache_free() */ 2769 0) 2.861 us | } /* putname() */ 2770 2771The return value of each traced function can be displayed after 2772an equal sign "=". When encountering system call failures, it 2773can be very helpful to quickly locate the function that first 2774returns an error code. 2775 2776 - hide: echo nofuncgraph-retval > trace_options 2777 - show: echo funcgraph-retval > trace_options 2778 2779 Example with funcgraph-retval:: 2780 2781 1) | cgroup_migrate() { 2782 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */ 2783 1) | cgroup_migrate_execute() { 2784 1) | cpu_cgroup_can_attach() { 2785 1) | cgroup_taskset_first() { 2786 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */ 2787 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */ 2788 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */ 2789 1) 2.335 us | } /* cpu_cgroup_can_attach = -22 */ 2790 1) 4.369 us | } /* cgroup_migrate_execute = -22 */ 2791 1) 7.143 us | } /* cgroup_migrate = -22 */ 2792 2793The above example shows that the function cpu_cgroup_can_attach 2794returned the error code -22 firstly, then we can read the code 2795of this function to get the root cause. 2796 2797When the option funcgraph-retval-hex is not set, the return value can 2798be displayed in a smart way. Specifically, if it is an error code, 2799it will be printed in signed decimal format, otherwise it will 2800printed in hexadecimal format. 2801 2802 - smart: echo nofuncgraph-retval-hex > trace_options 2803 - hexadecimal: echo funcgraph-retval-hex > trace_options 2804 2805 Example with funcgraph-retval-hex:: 2806 2807 1) | cgroup_migrate() { 2808 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */ 2809 1) | cgroup_migrate_execute() { 2810 1) | cpu_cgroup_can_attach() { 2811 1) | cgroup_taskset_first() { 2812 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */ 2813 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */ 2814 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */ 2815 1) 2.335 us | } /* cpu_cgroup_can_attach = 0xffffffea */ 2816 1) 4.369 us | } /* cgroup_migrate_execute = 0xffffffea */ 2817 1) 7.143 us | } /* cgroup_migrate = 0xffffffea */ 2818 2819At present, there are some limitations when using the funcgraph-retval 2820option, and these limitations will be eliminated in the future: 2821 2822- Even if the function return type is void, a return value will still 2823 be printed, and you can just ignore it. 2824 2825- Even if return values are stored in multiple registers, only the 2826 value contained in the first register will be recorded and printed. 2827 To illustrate, in the x86 architecture, eax and edx are used to store 2828 a 64-bit return value, with the lower 32 bits saved in eax and the 2829 upper 32 bits saved in edx. However, only the value stored in eax 2830 will be recorded and printed. 2831 2832- In certain procedure call standards, such as arm64's AAPCS64, when a 2833 type is smaller than a GPR, it is the responsibility of the consumer 2834 to perform the narrowing, and the upper bits may contain UNKNOWN values. 2835 Therefore, it is advisable to check the code for such cases. For instance, 2836 when using a u8 in a 64-bit GPR, bits [63:8] may contain arbitrary values, 2837 especially when larger types are truncated, whether explicitly or implicitly. 2838 Here are some specific cases to illustrate this point: 2839 2840 **Case One**: 2841 2842 The function narrow_to_u8 is defined as follows:: 2843 2844 u8 narrow_to_u8(u64 val) 2845 { 2846 // implicitly truncated 2847 return val; 2848 } 2849 2850 It may be compiled to:: 2851 2852 narrow_to_u8: 2853 < ... ftrace instrumentation ... > 2854 RET 2855 2856 If you pass 0x123456789abcdef to this function and want to narrow it, 2857 it may be recorded as 0x123456789abcdef instead of 0xef. 2858 2859 **Case Two**: 2860 2861 The function error_if_not_4g_aligned is defined as follows:: 2862 2863 int error_if_not_4g_aligned(u64 val) 2864 { 2865 if (val & GENMASK(31, 0)) 2866 return -EINVAL; 2867 2868 return 0; 2869 } 2870 2871 It could be compiled to:: 2872 2873 error_if_not_4g_aligned: 2874 CBNZ w0, .Lnot_aligned 2875 RET // bits [31:0] are zero, bits 2876 // [63:32] are UNKNOWN 2877 .Lnot_aligned: 2878 MOV x0, #-EINVAL 2879 RET 2880 2881 When passing 0x2_0000_0000 to it, the return value may be recorded as 2882 0x2_0000_0000 instead of 0. 2883 2884You can put some comments on specific functions by using 2885trace_printk() For example, if you want to put a comment inside 2886the __might_sleep() function, you just have to include 2887<linux/ftrace.h> and call trace_printk() inside __might_sleep():: 2888 2889 trace_printk("I'm a comment!\n") 2890 2891will produce:: 2892 2893 1) | __might_sleep() { 2894 1) | /* I'm a comment! */ 2895 1) 1.449 us | } 2896 2897 2898You might find other useful features for this tracer in the 2899following "dynamic ftrace" section such as tracing only specific 2900functions or tasks. 2901 2902dynamic ftrace 2903-------------- 2904 2905If CONFIG_DYNAMIC_FTRACE is set, the system will run with 2906virtually no overhead when function tracing is disabled. The way 2907this works is the mcount function call (placed at the start of 2908every kernel function, produced by the -pg switch in gcc), 2909starts of pointing to a simple return. (Enabling FTRACE will 2910include the -pg switch in the compiling of the kernel.) 2911 2912At compile time every C file object is run through the 2913recordmcount program (located in the scripts directory). This 2914program will parse the ELF headers in the C object to find all 2915the locations in the .text section that call mcount. Starting 2916with gcc version 4.6, the -mfentry has been added for x86, which 2917calls "__fentry__" instead of "mcount". Which is called before 2918the creation of the stack frame. 2919 2920Note, not all sections are traced. They may be prevented by either 2921a notrace, or blocked another way and all inline functions are not 2922traced. Check the "available_filter_functions" file to see what functions 2923can be traced. 2924 2925A section called "__mcount_loc" is created that holds 2926references to all the mcount/fentry call sites in the .text section. 2927The recordmcount program re-links this section back into the 2928original object. The final linking stage of the kernel will add all these 2929references into a single table. 2930 2931On boot up, before SMP is initialized, the dynamic ftrace code 2932scans this table and updates all the locations into nops. It 2933also records the locations, which are added to the 2934available_filter_functions list. Modules are processed as they 2935are loaded and before they are executed. When a module is 2936unloaded, it also removes its functions from the ftrace function 2937list. This is automatic in the module unload code, and the 2938module author does not need to worry about it. 2939 2940When tracing is enabled, the process of modifying the function 2941tracepoints is dependent on architecture. The old method is to use 2942kstop_machine to prevent races with the CPUs executing code being 2943modified (which can cause the CPU to do undesirable things, especially 2944if the modified code crosses cache (or page) boundaries), and the nops are 2945patched back to calls. But this time, they do not call mcount 2946(which is just a function stub). They now call into the ftrace 2947infrastructure. 2948 2949The new method of modifying the function tracepoints is to place 2950a breakpoint at the location to be modified, sync all CPUs, modify 2951the rest of the instruction not covered by the breakpoint. Sync 2952all CPUs again, and then remove the breakpoint with the finished 2953version to the ftrace call site. 2954 2955Some archs do not even need to monkey around with the synchronization, 2956and can just slap the new code on top of the old without any 2957problems with other CPUs executing it at the same time. 2958 2959One special side-effect to the recording of the functions being 2960traced is that we can now selectively choose which functions we 2961wish to trace and which ones we want the mcount calls to remain 2962as nops. 2963 2964Two files are used, one for enabling and one for disabling the 2965tracing of specified functions. They are: 2966 2967 set_ftrace_filter 2968 2969and 2970 2971 set_ftrace_notrace 2972 2973A list of available functions that you can add to these files is 2974listed in: 2975 2976 available_filter_functions 2977 2978:: 2979 2980 # cat available_filter_functions 2981 put_prev_task_idle 2982 kmem_cache_create 2983 pick_next_task_rt 2984 cpus_read_lock 2985 pick_next_task_fair 2986 mutex_lock 2987 [...] 2988 2989If I am only interested in sys_nanosleep and hrtimer_interrupt:: 2990 2991 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter 2992 # echo function > current_tracer 2993 # echo 1 > tracing_on 2994 # usleep 1 2995 # echo 0 > tracing_on 2996 # cat trace 2997 # tracer: function 2998 # 2999 # entries-in-buffer/entries-written: 5/5 #P:4 3000 # 3001 # _-----=> irqs-off 3002 # / _----=> need-resched 3003 # | / _---=> hardirq/softirq 3004 # || / _--=> preempt-depth 3005 # ||| / delay 3006 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3007 # | | | |||| | | 3008 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath 3009 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt 3010 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt 3011 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt 3012 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt 3013 3014To see which functions are being traced, you can cat the file: 3015:: 3016 3017 # cat set_ftrace_filter 3018 hrtimer_interrupt 3019 sys_nanosleep 3020 3021 3022Perhaps this is not enough. The filters also allow glob(7) matching. 3023 3024 ``<match>*`` 3025 will match functions that begin with <match> 3026 ``*<match>`` 3027 will match functions that end with <match> 3028 ``*<match>*`` 3029 will match functions that have <match> in it 3030 ``<match1>*<match2>`` 3031 will match functions that begin with <match1> and end with <match2> 3032 3033.. note:: 3034 It is better to use quotes to enclose the wild cards, 3035 otherwise the shell may expand the parameters into names 3036 of files in the local directory. 3037 3038:: 3039 3040 # echo 'hrtimer_*' > set_ftrace_filter 3041 3042Produces:: 3043 3044 # tracer: function 3045 # 3046 # entries-in-buffer/entries-written: 897/897 #P:4 3047 # 3048 # _-----=> irqs-off 3049 # / _----=> need-resched 3050 # | / _---=> hardirq/softirq 3051 # || / _--=> preempt-depth 3052 # ||| / delay 3053 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3054 # | | | |||| | | 3055 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit 3056 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel 3057 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer 3058 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit 3059 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11 3060 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt 3061 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter 3062 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem 3063 3064Notice that we lost the sys_nanosleep. 3065:: 3066 3067 # cat set_ftrace_filter 3068 hrtimer_run_queues 3069 hrtimer_run_pending 3070 hrtimer_init 3071 hrtimer_cancel 3072 hrtimer_try_to_cancel 3073 hrtimer_forward 3074 hrtimer_start 3075 hrtimer_reprogram 3076 hrtimer_force_reprogram 3077 hrtimer_get_next_event 3078 hrtimer_interrupt 3079 hrtimer_nanosleep 3080 hrtimer_wakeup 3081 hrtimer_get_remaining 3082 hrtimer_get_res 3083 hrtimer_init_sleeper 3084 3085 3086This is because the '>' and '>>' act just like they do in bash. 3087To rewrite the filters, use '>' 3088To append to the filters, use '>>' 3089 3090To clear out a filter so that all functions will be recorded 3091again:: 3092 3093 # echo > set_ftrace_filter 3094 # cat set_ftrace_filter 3095 # 3096 3097Again, now we want to append. 3098 3099:: 3100 3101 # echo sys_nanosleep > set_ftrace_filter 3102 # cat set_ftrace_filter 3103 sys_nanosleep 3104 # echo 'hrtimer_*' >> set_ftrace_filter 3105 # cat set_ftrace_filter 3106 hrtimer_run_queues 3107 hrtimer_run_pending 3108 hrtimer_init 3109 hrtimer_cancel 3110 hrtimer_try_to_cancel 3111 hrtimer_forward 3112 hrtimer_start 3113 hrtimer_reprogram 3114 hrtimer_force_reprogram 3115 hrtimer_get_next_event 3116 hrtimer_interrupt 3117 sys_nanosleep 3118 hrtimer_nanosleep 3119 hrtimer_wakeup 3120 hrtimer_get_remaining 3121 hrtimer_get_res 3122 hrtimer_init_sleeper 3123 3124 3125The set_ftrace_notrace prevents those functions from being 3126traced. 3127:: 3128 3129 # echo '*preempt*' '*lock*' > set_ftrace_notrace 3130 3131Produces:: 3132 3133 # tracer: function 3134 # 3135 # entries-in-buffer/entries-written: 39608/39608 #P:4 3136 # 3137 # _-----=> irqs-off 3138 # / _----=> need-resched 3139 # | / _---=> hardirq/softirq 3140 # || / _--=> preempt-depth 3141 # ||| / delay 3142 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3143 # | | | |||| | | 3144 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open 3145 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last 3146 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last 3147 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check 3148 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement 3149 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action 3150 bash-1994 [000] .... 4342.324899: do_truncate <-do_last 3151 bash-1994 [000] .... 4342.324899: setattr_should_drop_suidgid <-do_truncate 3152 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate 3153 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change 3154 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time 3155 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time 3156 3157We can see that there's no more lock or preempt tracing. 3158 3159Selecting function filters via index 3160------------------------------------ 3161 3162Because processing of strings is expensive (the address of the function 3163needs to be looked up before comparing to the string being passed in), 3164an index can be used as well to enable functions. This is useful in the 3165case of setting thousands of specific functions at a time. By passing 3166in a list of numbers, no string processing will occur. Instead, the function 3167at the specific location in the internal array (which corresponds to the 3168functions in the "available_filter_functions" file), is selected. 3169 3170:: 3171 3172 # echo 1 > set_ftrace_filter 3173 3174Will select the first function listed in "available_filter_functions" 3175 3176:: 3177 3178 # head -1 available_filter_functions 3179 trace_initcall_finish_cb 3180 3181 # cat set_ftrace_filter 3182 trace_initcall_finish_cb 3183 3184 # head -50 available_filter_functions | tail -1 3185 x86_pmu_commit_txn 3186 3187 # echo 1 50 > set_ftrace_filter 3188 # cat set_ftrace_filter 3189 trace_initcall_finish_cb 3190 x86_pmu_commit_txn 3191 3192Dynamic ftrace with the function graph tracer 3193--------------------------------------------- 3194 3195Although what has been explained above concerns both the 3196function tracer and the function-graph-tracer, there are some 3197special features only available in the function-graph tracer. 3198 3199If you want to trace only one function and all of its children, 3200you just have to echo its name into set_graph_function:: 3201 3202 echo __do_fault > set_graph_function 3203 3204will produce the following "expanded" trace of the __do_fault() 3205function:: 3206 3207 0) | __do_fault() { 3208 0) | filemap_fault() { 3209 0) | find_lock_page() { 3210 0) 0.804 us | find_get_page(); 3211 0) | __might_sleep() { 3212 0) 1.329 us | } 3213 0) 3.904 us | } 3214 0) 4.979 us | } 3215 0) 0.653 us | _spin_lock(); 3216 0) 0.578 us | page_add_file_rmap(); 3217 0) 0.525 us | native_set_pte_at(); 3218 0) 0.585 us | _spin_unlock(); 3219 0) | unlock_page() { 3220 0) 0.541 us | page_waitqueue(); 3221 0) 0.639 us | __wake_up_bit(); 3222 0) 2.786 us | } 3223 0) + 14.237 us | } 3224 0) | __do_fault() { 3225 0) | filemap_fault() { 3226 0) | find_lock_page() { 3227 0) 0.698 us | find_get_page(); 3228 0) | __might_sleep() { 3229 0) 1.412 us | } 3230 0) 3.950 us | } 3231 0) 5.098 us | } 3232 0) 0.631 us | _spin_lock(); 3233 0) 0.571 us | page_add_file_rmap(); 3234 0) 0.526 us | native_set_pte_at(); 3235 0) 0.586 us | _spin_unlock(); 3236 0) | unlock_page() { 3237 0) 0.533 us | page_waitqueue(); 3238 0) 0.638 us | __wake_up_bit(); 3239 0) 2.793 us | } 3240 0) + 14.012 us | } 3241 3242You can also expand several functions at once:: 3243 3244 echo sys_open > set_graph_function 3245 echo sys_close >> set_graph_function 3246 3247Now if you want to go back to trace all functions you can clear 3248this special filter via:: 3249 3250 echo > set_graph_function 3251 3252 3253ftrace_enabled 3254-------------- 3255 3256Note, the proc sysctl ftrace_enable is a big on/off switch for the 3257function tracer. By default it is enabled (when function tracing is 3258enabled in the kernel). If it is disabled, all function tracing is 3259disabled. This includes not only the function tracers for ftrace, but 3260also for any other uses (perf, kprobes, stack tracing, profiling, etc). It 3261cannot be disabled if there is a callback with FTRACE_OPS_FL_PERMANENT set 3262registered. 3263 3264Please disable this with care. 3265 3266This can be disable (and enabled) with:: 3267 3268 sysctl kernel.ftrace_enabled=0 3269 sysctl kernel.ftrace_enabled=1 3270 3271 or 3272 3273 echo 0 > /proc/sys/kernel/ftrace_enabled 3274 echo 1 > /proc/sys/kernel/ftrace_enabled 3275 3276 3277Filter commands 3278--------------- 3279 3280A few commands are supported by the set_ftrace_filter interface. 3281Trace commands have the following format:: 3282 3283 <function>:<command>:<parameter> 3284 3285The following commands are supported: 3286 3287- mod: 3288 This command enables function filtering per module. The 3289 parameter defines the module. For example, if only the write* 3290 functions in the ext3 module are desired, run: 3291 3292 echo 'write*:mod:ext3' > set_ftrace_filter 3293 3294 This command interacts with the filter in the same way as 3295 filtering based on function names. Thus, adding more functions 3296 in a different module is accomplished by appending (>>) to the 3297 filter file. Remove specific module functions by prepending 3298 '!':: 3299 3300 echo '!writeback*:mod:ext3' >> set_ftrace_filter 3301 3302 Mod command supports module globbing. Disable tracing for all 3303 functions except a specific module:: 3304 3305 echo '!*:mod:!ext3' >> set_ftrace_filter 3306 3307 Disable tracing for all modules, but still trace kernel:: 3308 3309 echo '!*:mod:*' >> set_ftrace_filter 3310 3311 Enable filter only for kernel:: 3312 3313 echo '*write*:mod:!*' >> set_ftrace_filter 3314 3315 Enable filter for module globbing:: 3316 3317 echo '*write*:mod:*snd*' >> set_ftrace_filter 3318 3319- traceon/traceoff: 3320 These commands turn tracing on and off when the specified 3321 functions are hit. The parameter determines how many times the 3322 tracing system is turned on and off. If unspecified, there is 3323 no limit. For example, to disable tracing when a schedule bug 3324 is hit the first 5 times, run:: 3325 3326 echo '__schedule_bug:traceoff:5' > set_ftrace_filter 3327 3328 To always disable tracing when __schedule_bug is hit:: 3329 3330 echo '__schedule_bug:traceoff' > set_ftrace_filter 3331 3332 These commands are cumulative whether or not they are appended 3333 to set_ftrace_filter. To remove a command, prepend it by '!' 3334 and drop the parameter:: 3335 3336 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter 3337 3338 The above removes the traceoff command for __schedule_bug 3339 that have a counter. To remove commands without counters:: 3340 3341 echo '!__schedule_bug:traceoff' > set_ftrace_filter 3342 3343- snapshot: 3344 Will cause a snapshot to be triggered when the function is hit. 3345 :: 3346 3347 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter 3348 3349 To only snapshot once: 3350 :: 3351 3352 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter 3353 3354 To remove the above commands:: 3355 3356 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter 3357 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter 3358 3359- enable_event/disable_event: 3360 These commands can enable or disable a trace event. Note, because 3361 function tracing callbacks are very sensitive, when these commands 3362 are registered, the trace point is activated, but disabled in 3363 a "soft" mode. That is, the tracepoint will be called, but 3364 just will not be traced. The event tracepoint stays in this mode 3365 as long as there's a command that triggers it. 3366 :: 3367 3368 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \ 3369 set_ftrace_filter 3370 3371 The format is:: 3372 3373 <function>:enable_event:<system>:<event>[:count] 3374 <function>:disable_event:<system>:<event>[:count] 3375 3376 To remove the events commands:: 3377 3378 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \ 3379 set_ftrace_filter 3380 echo '!schedule:disable_event:sched:sched_switch' > \ 3381 set_ftrace_filter 3382 3383- dump: 3384 When the function is hit, it will dump the contents of the ftrace 3385 ring buffer to the console. This is useful if you need to debug 3386 something, and want to dump the trace when a certain function 3387 is hit. Perhaps it's a function that is called before a triple 3388 fault happens and does not allow you to get a regular dump. 3389 3390- cpudump: 3391 When the function is hit, it will dump the contents of the ftrace 3392 ring buffer for the current CPU to the console. Unlike the "dump" 3393 command, it only prints out the contents of the ring buffer for the 3394 CPU that executed the function that triggered the dump. 3395 3396- stacktrace: 3397 When the function is hit, a stack trace is recorded. 3398 3399trace_pipe 3400---------- 3401 3402The trace_pipe outputs the same content as the trace file, but 3403the effect on the tracing is different. Every read from 3404trace_pipe is consumed. This means that subsequent reads will be 3405different. The trace is live. 3406:: 3407 3408 # echo function > current_tracer 3409 # cat trace_pipe > /tmp/trace.out & 3410 [1] 4153 3411 # echo 1 > tracing_on 3412 # usleep 1 3413 # echo 0 > tracing_on 3414 # cat trace 3415 # tracer: function 3416 # 3417 # entries-in-buffer/entries-written: 0/0 #P:4 3418 # 3419 # _-----=> irqs-off 3420 # / _----=> need-resched 3421 # | / _---=> hardirq/softirq 3422 # || / _--=> preempt-depth 3423 # ||| / delay 3424 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3425 # | | | |||| | | 3426 3427 # 3428 # cat /tmp/trace.out 3429 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write 3430 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock 3431 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify 3432 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify 3433 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify 3434 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock 3435 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock 3436 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify 3437 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath 3438 3439 3440Note, reading the trace_pipe file will block until more input is 3441added. This is contrary to the trace file. If any process opened 3442the trace file for reading, it will actually disable tracing and 3443prevent new entries from being added. The trace_pipe file does 3444not have this limitation. 3445 3446trace entries 3447------------- 3448 3449Having too much or not enough data can be troublesome in 3450diagnosing an issue in the kernel. The file buffer_size_kb is 3451used to modify the size of the internal trace buffers. The 3452number listed is the number of entries that can be recorded per 3453CPU. To know the full size, multiply the number of possible CPUs 3454with the number of entries. 3455:: 3456 3457 # cat buffer_size_kb 3458 1408 (units kilobytes) 3459 3460Or simply read buffer_total_size_kb 3461:: 3462 3463 # cat buffer_total_size_kb 3464 5632 3465 3466To modify the buffer, simple echo in a number (in 1024 byte segments). 3467:: 3468 3469 # echo 10000 > buffer_size_kb 3470 # cat buffer_size_kb 3471 10000 (units kilobytes) 3472 3473It will try to allocate as much as possible. If you allocate too 3474much, it can cause Out-Of-Memory to trigger. 3475:: 3476 3477 # echo 1000000000000 > buffer_size_kb 3478 -bash: echo: write error: Cannot allocate memory 3479 # cat buffer_size_kb 3480 85 3481 3482The per_cpu buffers can be changed individually as well: 3483:: 3484 3485 # echo 10000 > per_cpu/cpu0/buffer_size_kb 3486 # echo 100 > per_cpu/cpu1/buffer_size_kb 3487 3488When the per_cpu buffers are not the same, the buffer_size_kb 3489at the top level will just show an X 3490:: 3491 3492 # cat buffer_size_kb 3493 X 3494 3495This is where the buffer_total_size_kb is useful: 3496:: 3497 3498 # cat buffer_total_size_kb 3499 12916 3500 3501Writing to the top level buffer_size_kb will reset all the buffers 3502to be the same again. 3503 3504Snapshot 3505-------- 3506CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature 3507available to all non latency tracers. (Latency tracers which 3508record max latency, such as "irqsoff" or "wakeup", can't use 3509this feature, since those are already using the snapshot 3510mechanism internally.) 3511 3512Snapshot preserves a current trace buffer at a particular point 3513in time without stopping tracing. Ftrace swaps the current 3514buffer with a spare buffer, and tracing continues in the new 3515current (=previous spare) buffer. 3516 3517The following tracefs files in "tracing" are related to this 3518feature: 3519 3520 snapshot: 3521 3522 This is used to take a snapshot and to read the output 3523 of the snapshot. Echo 1 into this file to allocate a 3524 spare buffer and to take a snapshot (swap), then read 3525 the snapshot from this file in the same format as 3526 "trace" (described above in the section "The File 3527 System"). Both reads snapshot and tracing are executable 3528 in parallel. When the spare buffer is allocated, echoing 3529 0 frees it, and echoing else (positive) values clear the 3530 snapshot contents. 3531 More details are shown in the table below. 3532 3533 +--------------+------------+------------+------------+ 3534 |status\\input | 0 | 1 | else | 3535 +==============+============+============+============+ 3536 |not allocated |(do nothing)| alloc+swap |(do nothing)| 3537 +--------------+------------+------------+------------+ 3538 |allocated | free | swap | clear | 3539 +--------------+------------+------------+------------+ 3540 3541Here is an example of using the snapshot feature. 3542:: 3543 3544 # echo 1 > events/sched/enable 3545 # echo 1 > snapshot 3546 # cat snapshot 3547 # tracer: nop 3548 # 3549 # entries-in-buffer/entries-written: 71/71 #P:8 3550 # 3551 # _-----=> irqs-off 3552 # / _----=> need-resched 3553 # | / _---=> hardirq/softirq 3554 # || / _--=> preempt-depth 3555 # ||| / delay 3556 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3557 # | | | |||| | | 3558 <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 3559 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 3560 [...] 3561 <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 3562 3563 # cat trace 3564 # tracer: nop 3565 # 3566 # entries-in-buffer/entries-written: 77/77 #P:8 3567 # 3568 # _-----=> irqs-off 3569 # / _----=> need-resched 3570 # | / _---=> hardirq/softirq 3571 # || / _--=> preempt-depth 3572 # ||| / delay 3573 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3574 # | | | |||| | | 3575 <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 3576 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 3577 [...] 3578 3579 3580If you try to use this snapshot feature when current tracer is 3581one of the latency tracers, you will get the following results. 3582:: 3583 3584 # echo wakeup > current_tracer 3585 # echo 1 > snapshot 3586 bash: echo: write error: Device or resource busy 3587 # cat snapshot 3588 cat: snapshot: Device or resource busy 3589 3590 3591Instances 3592--------- 3593In the tracefs tracing directory, there is a directory called "instances". 3594This directory can have new directories created inside of it using 3595mkdir, and removing directories with rmdir. The directory created 3596with mkdir in this directory will already contain files and other 3597directories after it is created. 3598:: 3599 3600 # mkdir instances/foo 3601 # ls instances/foo 3602 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu 3603 set_event snapshot trace trace_clock trace_marker trace_options 3604 trace_pipe tracing_on 3605 3606As you can see, the new directory looks similar to the tracing directory 3607itself. In fact, it is very similar, except that the buffer and 3608events are agnostic from the main directory, or from any other 3609instances that are created. 3610 3611The files in the new directory work just like the files with the 3612same name in the tracing directory except the buffer that is used 3613is a separate and new buffer. The files affect that buffer but do not 3614affect the main buffer with the exception of trace_options. Currently, 3615the trace_options affect all instances and the top level buffer 3616the same, but this may change in future releases. That is, options 3617may become specific to the instance they reside in. 3618 3619Notice that none of the function tracer files are there, nor is 3620current_tracer and available_tracers. This is because the buffers 3621can currently only have events enabled for them. 3622:: 3623 3624 # mkdir instances/foo 3625 # mkdir instances/bar 3626 # mkdir instances/zoot 3627 # echo 100000 > buffer_size_kb 3628 # echo 1000 > instances/foo/buffer_size_kb 3629 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb 3630 # echo function > current_trace 3631 # echo 1 > instances/foo/events/sched/sched_wakeup/enable 3632 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable 3633 # echo 1 > instances/foo/events/sched/sched_switch/enable 3634 # echo 1 > instances/bar/events/irq/enable 3635 # echo 1 > instances/zoot/events/syscalls/enable 3636 # cat trace_pipe 3637 CPU:2 [LOST 11745 EVENTS] 3638 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist 3639 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave 3640 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist 3641 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist 3642 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock 3643 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype 3644 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist 3645 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist 3646 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics 3647 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics 3648 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process 3649 [...] 3650 3651 # cat instances/foo/trace_pipe 3652 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000 3653 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000 3654 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003 3655 <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 3656 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 3657 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000 3658 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000 3659 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 3660 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001 3661 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 3662 [...] 3663 3664 # cat instances/bar/trace_pipe 3665 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX] 3666 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX] 3667 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER] 3668 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU] 3669 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER] 3670 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER] 3671 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU] 3672 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU] 3673 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4 3674 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled 3675 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0 3676 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled 3677 [...] 3678 3679 # cat instances/zoot/trace 3680 # tracer: nop 3681 # 3682 # entries-in-buffer/entries-written: 18996/18996 #P:4 3683 # 3684 # _-----=> irqs-off 3685 # / _----=> need-resched 3686 # | / _---=> hardirq/softirq 3687 # || / _--=> preempt-depth 3688 # ||| / delay 3689 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3690 # | | | |||| | | 3691 bash-1998 [000] d... 140.733501: sys_write -> 0x2 3692 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1) 3693 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1 3694 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0) 3695 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1 3696 bash-1998 [000] d... 140.733510: sys_close(fd: a) 3697 bash-1998 [000] d... 140.733510: sys_close -> 0x0 3698 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8) 3699 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0 3700 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8) 3701 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0 3702 3703You can see that the trace of the top most trace buffer shows only 3704the function tracing. The foo instance displays wakeups and task 3705switches. 3706 3707To remove the instances, simply delete their directories: 3708:: 3709 3710 # rmdir instances/foo 3711 # rmdir instances/bar 3712 # rmdir instances/zoot 3713 3714Note, if a process has a trace file open in one of the instance 3715directories, the rmdir will fail with EBUSY. 3716 3717 3718Stack trace 3719----------- 3720Since the kernel has a fixed sized stack, it is important not to 3721waste it in functions. A kernel developer must be conscious of 3722what they allocate on the stack. If they add too much, the system 3723can be in danger of a stack overflow, and corruption will occur, 3724usually leading to a system panic. 3725 3726There are some tools that check this, usually with interrupts 3727periodically checking usage. But if you can perform a check 3728at every function call that will become very useful. As ftrace provides 3729a function tracer, it makes it convenient to check the stack size 3730at every function call. This is enabled via the stack tracer. 3731 3732CONFIG_STACK_TRACER enables the ftrace stack tracing functionality. 3733To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled. 3734:: 3735 3736 # echo 1 > /proc/sys/kernel/stack_tracer_enabled 3737 3738You can also enable it from the kernel command line to trace 3739the stack size of the kernel during boot up, by adding "stacktrace" 3740to the kernel command line parameter. 3741 3742After running it for a few minutes, the output looks like: 3743:: 3744 3745 # cat stack_max_size 3746 2928 3747 3748 # cat stack_trace 3749 Depth Size Location (18 entries) 3750 ----- ---- -------- 3751 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac 3752 1) 2704 160 find_busiest_group+0x31/0x1f1 3753 2) 2544 256 load_balance+0xd9/0x662 3754 3) 2288 80 idle_balance+0xbb/0x130 3755 4) 2208 128 __schedule+0x26e/0x5b9 3756 5) 2080 16 schedule+0x64/0x66 3757 6) 2064 128 schedule_timeout+0x34/0xe0 3758 7) 1936 112 wait_for_common+0x97/0xf1 3759 8) 1824 16 wait_for_completion+0x1d/0x1f 3760 9) 1808 128 flush_work+0xfe/0x119 3761 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20 3762 11) 1664 48 input_available_p+0x1d/0x5c 3763 12) 1616 48 n_tty_poll+0x6d/0x134 3764 13) 1568 64 tty_poll+0x64/0x7f 3765 14) 1504 880 do_select+0x31e/0x511 3766 15) 624 400 core_sys_select+0x177/0x216 3767 16) 224 96 sys_select+0x91/0xb9 3768 17) 128 128 system_call_fastpath+0x16/0x1b 3769 3770Note, if -mfentry is being used by gcc, functions get traced before 3771they set up the stack frame. This means that leaf level functions 3772are not tested by the stack tracer when -mfentry is used. 3773 3774Currently, -mfentry is used by gcc 4.6.0 and above on x86 only. 3775 3776More 3777---- 3778More details can be found in the source code, in the `kernel/trace/*.c` files. 3779