| | 1 | ftrace - Function Tracer |
| | 2 | ======================== |
| | 3 | |
| | 4 | Copyright 2008 Red Hat Inc. |
| | 5 | Author: Steven Rostedt <srostedt@redhat.com> |
| | 6 | License: The GNU Free Documentation License, Version 1.2 |
| | 7 | (dual licensed under the GPL v2) |
| | 8 | Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton, |
| | 9 | John Kacur, and David Teigland. |
| | 10 | |
| | 11 | Written for: 2.6.28-rc2 |
| | 12 | |
| | 13 | Introduction |
| | 14 | ------------ |
| | 15 | |
| | 16 | Ftrace is an internal tracer designed to help out developers and |
| | 17 | designers of systems to find what is going on inside the kernel. |
| | 18 | It can be used for debugging or analyzing latencies and performance |
| | 19 | issues that take place outside of user-space. |
| | 20 | |
| | 21 | Although ftrace is the function tracer, it also includes an |
| | 22 | infrastructure that allows for other types of tracing. Some of the |
| | 23 | tracers that are currently in ftrace include a tracer to trace |
| | 24 | context switches, the time it takes for a high priority task to |
| | 25 | run after it was woken up, the time interrupts are disabled, and |
| | 26 | more (ftrace allows for tracer plugins, which means that the list of |
| | 27 | tracers can always grow). |
| | 28 | |
| | 29 | |
| | 30 | The File System |
| | 31 | --------------- |
| | 32 | |
| | 33 | Ftrace uses the debugfs file system to hold the control files as well |
| | 34 | as the files to display output. |
| | 35 | |
| | 36 | To mount the debugfs system: |
| | 37 | |
| | 38 | # mkdir /debug |
| | 39 | # mount -t debugfs nodev /debug |
| | 40 | |
| | 41 | (Note: it is more common to mount at /sys/kernel/debug, but for simplicity |
| | 42 | this document will use /debug) |
| | 43 | |
| | 44 | That's it! (assuming that you have ftrace configured into your kernel) |
| | 45 | |
| | 46 | After mounting the debugfs, you can see a directory called |
| | 47 | "tracing". This directory contains the control and output files |
| | 48 | of ftrace. Here is a list of some of the key files: |
| | 49 | |
| | 50 | |
| | 51 | Note: all time values are in microseconds. |
| | 52 | |
| | 53 | current_tracer: This is used to set or display the current tracer |
| | 54 | that is configured. |
| | 55 | |
| | 56 | available_tracers: This holds the different types of tracers that |
| | 57 | have been compiled into the kernel. The tracers |
| | 58 | listed here can be configured by echoing their name |
| | 59 | into current_tracer. |
| | 60 | |
| | 61 | tracing_enabled: This sets or displays whether the current_tracer |
| | 62 | is activated and tracing or not. Echo 0 into this |
| | 63 | file to disable the tracer or 1 to enable it. |
| | 64 | |
| | 65 | trace: This file holds the output of the trace in a human readable |
| | 66 | format (described below). |
| | 67 | |
| | 68 | latency_trace: This file shows the same trace but the information |
| | 69 | is organized more to display possible latencies |
| | 70 | in the system (described below). |
| | 71 | |
| | 72 | trace_pipe: The output is the same as the "trace" file but this |
| | 73 | file is meant to be streamed with live tracing. |
| | 74 | Reads from this file will block until new data |
| | 75 | is retrieved. Unlike the "trace" and "latency_trace" |
| | 76 | files, this file is a consumer. This means reading |
| | 77 | from this file causes sequential reads to display |
| | 78 | more current data. Once data is read from this |
| | 79 | file, it is consumed, and will not be read |
| | 80 | again with a sequential read. The "trace" and |
| | 81 | "latency_trace" files are static, and if the |
| | 82 | tracer is not adding more data, they will display |
| | 83 | the same information every time they are read. |
| | 84 | |
| | 85 | iter_ctrl: This file lets the user control the amount of data |
| | 86 | that is displayed in one of the above output |
| | 87 | files. |
| | 88 | |
| | 89 | trace_max_latency: Some of the tracers record the max latency. |
| | 90 | For example, the time interrupts are disabled. |
| | 91 | This time is saved in this file. The max trace |
| | 92 | will also be stored, and displayed by either |
| | 93 | "trace" or "latency_trace". A new max trace will |
| | 94 | only be recorded if the latency is greater than |
| | 95 | the value in this file. (in microseconds) |
| | 96 | |
| | 97 | buffer_size_kb: This sets or displays the number of bytes each CPU |
| | 98 | buffer can hold. The tracer buffers are the same size |
| | 99 | for each CPU. The displayed number is the size of the |
| | 100 | CPU buffer and not total size of all buffers. The |
| | 101 | trace buffers are allocated in pages (blocks of memory |
| | 102 | that the kernel uses for allocation, usually 4 KB in size). |
| | 103 | If the last page allocated has room for more bytes |
| | 104 | than requested, the rest of the page will be used, |
| | 105 | making the actual allocation bigger than requested. |
| | 106 | (Note, the size may not be a multiple of the page size due |
| | 107 | to buffer managment overhead.) |
| | 108 | |
| | 109 | This can only be updated when the current_tracer |
| | 110 | is set to "nop". |
| | 111 | |
| | 112 | tracing_cpumask: This is a mask that lets the user only trace |
| | 113 | on specified CPUS. The format is a hex string |
| | 114 | representing the CPUS. |
| | 115 | |
| | 116 | set_ftrace_filter: When dynamic ftrace is configured in (see the |
| | 117 | section below "dynamic ftrace"), the code is dynamically |
| | 118 | modified (code text rewrite) to disable calling of the |
| | 119 | function profiler (mcount). This lets tracing be configured |
| | 120 | in with practically no overhead in performance. This also |
| | 121 | has a side effect of enabling or disabling specific functions |
| | 122 | to be traced. Echoing names of functions into this file |
| | 123 | will limit the trace to only those functions. |
| | 124 | |
| | 125 | set_ftrace_notrace: This has an effect opposite to that of |
| | 126 | set_ftrace_filter. Any function that is added here will not |
| | 127 | be traced. If a function exists in both set_ftrace_filter |
| | 128 | and set_ftrace_notrace, the function will _not_ be traced. |
| | 129 | |
| | 130 | available_filter_functions: This lists the functions that ftrace |
| | 131 | has processed and can trace. These are the function |
| | 132 | names that you can pass to "set_ftrace_filter" or |
| | 133 | "set_ftrace_notrace". (See the section "dynamic ftrace" |
| | 134 | below for more details.) |
| | 135 | |
| | 136 | |
| | 137 | The Tracers |
| | 138 | ----------- |
| | 139 | |
| | 140 | Here is the list of current tracers that may be configured. |
| | 141 | |
| | 142 | function - function tracer that uses mcount to trace all functions. |
| | 143 | |
| | 144 | sched_switch - traces the context switches between tasks. |
| | 145 | |
| | 146 | irqsoff - traces the areas that disable interrupts and saves |
| | 147 | the trace with the longest max latency. |
| | 148 | See tracing_max_latency. When a new max is recorded, |
| | 149 | it replaces the old trace. It is best to view this |
| | 150 | trace via the latency_trace file. |
| | 151 | |
| | 152 | preemptoff - Similar to irqsoff but traces and records the amount of |
| | 153 | time for which preemption is disabled. |
| | 154 | |
| | 155 | preemptirqsoff - Similar to irqsoff and preemptoff, but traces and |
| | 156 | records the largest time for which irqs and/or preemption |
| | 157 | is disabled. |
| | 158 | |
| | 159 | wakeup - Traces and records the max latency that it takes for |
| | 160 | the highest priority task to get scheduled after |
| | 161 | it has been woken up. |
| | 162 | |
| | 163 | nop - This is not a tracer. To remove all tracers from tracing |
| | 164 | simply echo "nop" into current_tracer. |
| | 165 | |
| | 166 | |
| | 167 | Examples of using the tracer |
| | 168 | ---------------------------- |
| | 169 | |
| | 170 | Here are typical examples of using the tracers when controlling them only |
| | 171 | with the debugfs interface (without using any user-land utilities). |
| | 172 | |
| | 173 | Output format: |
| | 174 | -------------- |
| | 175 | |
| | 176 | Here is an example of the output format of the file "trace" |
| | 177 | |
| | 178 | -------- |
| | 179 | # tracer: function |
| | 180 | # |
| | 181 | # TASK-PID CPU# TIMESTAMP FUNCTION |
| | 182 | # | | | | | |
| | 183 | bash-4251 [01] 10152.583854: path_put <-path_walk |
| | 184 | bash-4251 [01] 10152.583855: dput <-path_put |
| | 185 | bash-4251 [01] 10152.583855: _atomic_dec_and_lock <-dput |
| | 186 | -------- |
| | 187 | |
| | 188 | A header is printed with the tracer name that is represented by the trace. |
| | 189 | In this case the tracer is "function". Then a header showing the format. Task |
| | 190 | name "bash", the task PID "4251", the CPU that it was running on |
| | 191 | "01", the timestamp in <secs>.<usecs> format, the function name that was |
| | 192 | traced "path_put" and the parent function that called this function |
| | 193 | "path_walk". The timestamp is the time at which the function was |
| | 194 | entered. |
| | 195 | |
| | 196 | The sched_switch tracer also includes tracing of task wakeups and |
| | 197 | context switches. |
| | 198 | |
| | 199 | ksoftirqd/1-7 [01] 1453.070013: 7:115:R + 2916:115:S |
| | 200 | ksoftirqd/1-7 [01] 1453.070013: 7:115:R + 10:115:S |
| | 201 | ksoftirqd/1-7 [01] 1453.070013: 7:115:R ==> 10:115:R |
| | 202 | events/1-10 [01] 1453.070013: 10:115:S ==> 2916:115:R |
| | 203 | kondemand/1-2916 [01] 1453.070013: 2916:115:S ==> 7:115:R |
| | 204 | ksoftirqd/1-7 [01] 1453.070013: 7:115:S ==> 0:140:R |
| | 205 | |
| | 206 | Wake ups are represented by a "+" and the context switches are shown as |
| | 207 | "==>". The format is: |
| | 208 | |
| | 209 | Context switches: |
| | 210 | |
| | 211 | Previous task Next Task |
| | 212 | |
| | 213 | <pid>:<prio>:<state> ==> <pid>:<prio>:<state> |
| | 214 | |
| | 215 | Wake ups: |
| | 216 | |
| | 217 | Current task Task waking up |
| | 218 | |
| | 219 | <pid>:<prio>:<state> + <pid>:<prio>:<state> |
| | 220 | |
| | 221 | The prio is the internal kernel priority, which is the inverse of the |
| | 222 | priority that is usually displayed by user-space tools. Zero represents |
| | 223 | the highest priority (99). Prio 100 starts the "nice" priorities with |
| | 224 | 100 being equal to nice -20 and 139 being nice 19. The prio "140" is |
| | 225 | reserved for the idle task which is the lowest priority thread (pid 0). |
| | 226 | |
| | 227 | |
| | 228 | Latency trace format |
| | 229 | -------------------- |
| | 230 | |
| | 231 | For traces that display latency times, the latency_trace file gives |
| | 232 | somewhat more information to see why a latency happened. Here is a typical |
| | 233 | trace. |
| | 234 | |
| | 235 | # tracer: irqsoff |
| | 236 | # |
| | 237 | irqsoff latency trace v1.1.5 on 2.6.26-rc8 |
| | 238 | -------------------------------------------------------------------- |
| | 239 | latency: 97 us, #3/3, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) |
| | 240 | ----------------- |
| | 241 | | task: swapper-0 (uid:0 nice:0 policy:0 rt_prio:0) |
| | 242 | ----------------- |
| | 243 | => started at: apic_timer_interrupt |
| | 244 | => ended at: do_softirq |
| | 245 | |
| | 246 | # _------=> CPU# |
| | 247 | # / _-----=> irqs-off |
| | 248 | # | / _----=> need-resched |
| | 249 | # || / _---=> hardirq/softirq |
| | 250 | # ||| / _--=> preempt-depth |
| | 251 | # |||| / |
| | 252 | # ||||| delay |
| | 253 | # cmd pid ||||| time | caller |
| | 254 | # \ / ||||| \ | / |
| | 255 | <idle>-0 0d..1 0us+: trace_hardirqs_off_thunk (apic_timer_interrupt) |
| | 256 | <idle>-0 0d.s. 97us : __do_softirq (do_softirq) |
| | 257 | <idle>-0 0d.s1 98us : trace_hardirqs_on (do_softirq) |
| | 258 | |
| | 259 | |
| | 260 | |
| | 261 | This shows that the current tracer is "irqsoff" tracing the time for which |
| | 262 | interrupts were disabled. It gives the trace version and the version |
| | 263 | of the kernel upon which this was executed on (2.6.26-rc8). Then it displays |
| | 264 | the max latency in microsecs (97 us). The number of trace entries displayed |
| | 265 | and the total number recorded (both are three: #3/3). The type of |
| | 266 | preemption that was used (PREEMPT). VP, KP, SP, and HP are always zero |
| | 267 | and are reserved for later use. #P is the number of online CPUS (#P:2). |
| | 268 | |
| | 269 | The task is the process that was running when the latency occurred. |
| | 270 | (swapper pid: 0). |
| | 271 | |
| | 272 | The start and stop (the functions in which the interrupts were disabled and |
| | 273 | enabled respectively) that caused the latencies: |
| | 274 | |
| | 275 | apic_timer_interrupt is where the interrupts were disabled. |
| | 276 | do_softirq is where they were enabled again. |
| | 277 | |
| | 278 | The next lines after the header are the trace itself. The header |
| | 279 | explains which is which. |
| | 280 | |
| | 281 | cmd: The name of the process in the trace. |
| | 282 | |
| | 283 | pid: The PID of that process. |
| | 284 | |
| | 285 | CPU#: The CPU which the process was running on. |
| | 286 | |
| | 287 | irqs-off: 'd' interrupts are disabled. '.' otherwise. |
| | 288 | Note: If the architecture does not support a way to |
| | 289 | read the irq flags variable, an 'X' will always |
| | 290 | be printed here. |
| | 291 | |
| | 292 | need-resched: 'N' task need_resched is set, '.' otherwise. |
| | 293 | |
| | 294 | hardirq/softirq: |
| | 295 | 'H' - hard irq occurred inside a softirq. |
| | 296 | 'h' - hard irq is running |
| | 297 | 's' - soft irq is running |
| | 298 | '.' - normal context. |
| | 299 | |
| | 300 | preempt-depth: The level of preempt_disabled |
| | 301 | |
| | 302 | The above is mostly meaningful for kernel developers. |
| | 303 | |
| | 304 | time: This differs from the trace file output. The trace file output |
| | 305 | includes an absolute timestamp. The timestamp used by the |
| | 306 | latency_trace file is relative to the start of the trace. |
| | 307 | |
| | 308 | delay: This is just to help catch your eye a bit better. And |
| | 309 | needs to be fixed to be only relative to the same CPU. |
| | 310 | The marks are determined by the difference between this |
| | 311 | current trace and the next trace. |
| | 312 | '!' - greater than preempt_mark_thresh (default 100) |
| | 313 | '+' - greater than 1 microsecond |
| | 314 | ' ' - less than or equal to 1 microsecond. |
| | 315 | |
| | 316 | The rest is the same as the 'trace' file. |
| | 317 | |
| | 318 | |
| | 319 | iter_ctrl |
| | 320 | --------- |
| | 321 | |
| | 322 | The iter_ctrl file is used to control what gets printed in the trace |
| | 323 | output. To see what is available, simply cat the file: |
| | 324 | |
| | 325 | cat /debug/tracing/iter_ctrl |
| | 326 | print-parent nosym-offset nosym-addr noverbose noraw nohex nobin \ |
| | 327 | noblock nostacktrace nosched-tree |
| | 328 | |
| | 329 | To disable one of the options, echo in the option prepended with "no". |
| | 330 | |
| | 331 | echo noprint-parent > /debug/tracing/iter_ctrl |
| | 332 | |
| | 333 | To enable an option, leave off the "no". |
| | 334 | |
| | 335 | echo sym-offset > /debug/tracing/iter_ctrl |
| | 336 | |
| | 337 | Here are the available options: |
| | 338 | |
| | 339 | print-parent - On function traces, display the calling function |
| | 340 | as well as the function being traced. |
| | 341 | |
| | 342 | print-parent: |
| | 343 | bash-4000 [01] 1477.606694: simple_strtoul <-strict_strtoul |
| | 344 | |
| | 345 | noprint-parent: |
| | 346 | bash-4000 [01] 1477.606694: simple_strtoul |
| | 347 | |
| | 348 | |
| | 349 | sym-offset - Display not only the function name, but also the offset |
| | 350 | in the function. For example, instead of seeing just |
| | 351 | "ktime_get", you will see "ktime_get+0xb/0x20". |
| | 352 | |
| | 353 | sym-offset: |
| | 354 | bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0 |
| | 355 | |
| | 356 | sym-addr - this will also display the function address as well as |
| | 357 | the function name. |
| | 358 | |
| | 359 | sym-addr: |
| | 360 | bash-4000 [01] 1477.606694: simple_strtoul <c0339346> |
| | 361 | |
| | 362 | verbose - This deals with the latency_trace file. |
| | 363 | |
| | 364 | bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \ |
| | 365 | (+0.000ms): simple_strtoul (strict_strtoul) |
| | 366 | |
| | 367 | raw - This will display raw numbers. This option is best for use with |
| | 368 | user applications that can translate the raw numbers better than |
| | 369 | having it done in the kernel. |
| | 370 | |
| | 371 | hex - Similar to raw, but the numbers will be in a hexadecimal format. |
| | 372 | |
| | 373 | bin - This will print out the formats in raw binary. |
| | 374 | |
| | 375 | block - TBD (needs update) |
| | 376 | |
| | 377 | stacktrace - This is one of the options that changes the trace itself. |
| | 378 | When a trace is recorded, so is the stack of functions. |
| | 379 | This allows for back traces of trace sites. |
| | 380 | |
| | 381 | sched-tree - TBD (any users??) |
| | 382 | |
| | 383 | |
| | 384 | sched_switch |
| | 385 | ------------ |
| | 386 | |
| | 387 | This tracer simply records schedule switches. Here is an example |
| | 388 | of how to use it. |
| | 389 | |
| | 390 | # echo sched_switch > /debug/tracing/current_tracer |
| | 391 | # echo 1 > /debug/tracing/tracing_enabled |
| | 392 | # sleep 1 |
| | 393 | # echo 0 > /debug/tracing/tracing_enabled |
| | 394 | # cat /debug/tracing/trace |
| | 395 | |
| | 396 | # tracer: sched_switch |
| | 397 | # |
| | 398 | # TASK-PID CPU# TIMESTAMP FUNCTION |
| | 399 | # | | | | | |
| | 400 | bash-3997 [01] 240.132281: 3997:120:R + 4055:120:R |
| | 401 | bash-3997 [01] 240.132284: 3997:120:R ==> 4055:120:R |
| | 402 | sleep-4055 [01] 240.132371: 4055:120:S ==> 3997:120:R |
| | 403 | bash-3997 [01] 240.132454: 3997:120:R + 4055:120:S |
| | 404 | bash-3997 [01] 240.132457: 3997:120:R ==> 4055:120:R |
| | 405 | sleep-4055 [01] 240.132460: 4055:120:D ==> 3997:120:R |
| | 406 | bash-3997 [01] 240.132463: 3997:120:R + 4055:120:D |
| | 407 | bash-3997 [01] 240.132465: 3997:120:R ==> 4055:120:R |
| | 408 | <idle>-0 [00] 240.132589: 0:140:R + 4:115:S |
| | 409 | <idle>-0 [00] 240.132591: 0:140:R ==> 4:115:R |
| | 410 | ksoftirqd/0-4 [00] 240.132595: 4:115:S ==> 0:140:R |
| | 411 | <idle>-0 [00] 240.132598: 0:140:R + 4:115:S |
| | 412 | <idle>-0 [00] 240.132599: 0:140:R ==> 4:115:R |
| | 413 | ksoftirqd/0-4 [00] 240.132603: 4:115:S ==> 0:140:R |
| | 414 | sleep-4055 [01] 240.133058: 4055:120:S ==> 3997:120:R |
| | 415 | [...] |
| | 416 | |
| | 417 | |
| | 418 | As we have discussed previously about this format, the header shows |
| | 419 | the name of the trace and points to the options. The "FUNCTION" |
| | 420 | is a misnomer since here it represents the wake ups and context |
| | 421 | switches. |
| | 422 | |
| | 423 | The sched_switch file only lists the wake ups (represented with '+') |
| | 424 | and context switches ('==>') with the previous task or current task |
| | 425 | first followed by the next task or task waking up. The format for both |
| | 426 | of these is PID:KERNEL-PRIO:TASK-STATE. Remember that the KERNEL-PRIO |
| | 427 | is the inverse of the actual priority with zero (0) being the highest |
| | 428 | priority and the nice values starting at 100 (nice -20). Below is |
| | 429 | a quick chart to map the kernel priority to user land priorities. |
| | 430 | |
| | 431 | Kernel priority: 0 to 99 ==> user RT priority 99 to 0 |
| | 432 | Kernel priority: 100 to 139 ==> user nice -20 to 19 |
| | 433 | Kernel priority: 140 ==> idle task priority |
| | 434 | |
| | 435 | The task states are: |
| | 436 | |
| | 437 | R - running : wants to run, may not actually be running |
| | 438 | S - sleep : process is waiting to be woken up (handles signals) |
| | 439 | D - disk sleep (uninterruptible sleep) : process must be woken up |
| | 440 | (ignores signals) |
| | 441 | T - stopped : process suspended |
| | 442 | t - traced : process is being traced (with something like gdb) |
| | 443 | Z - zombie : process waiting to be cleaned up |
| | 444 | X - unknown |
| | 445 | |
| | 446 | |
| | 447 | ftrace_enabled |
| | 448 | -------------- |
| | 449 | |
| | 450 | The following tracers (listed below) give different output depending |
| | 451 | on whether or not the sysctl ftrace_enabled is set. To set ftrace_enabled, |
| | 452 | one can either use the sysctl function or set it via the proc |
| | 453 | file system interface. |
| | 454 | |
| | 455 | sysctl kernel.ftrace_enabled=1 |
| | 456 | |
| | 457 | or |
| | 458 | |
| | 459 | echo 1 > /proc/sys/kernel/ftrace_enabled |
| | 460 | |
| | 461 | To disable ftrace_enabled simply replace the '1' with '0' in |
| | 462 | the above commands. |
| | 463 | |
| | 464 | When ftrace_enabled is set the tracers will also record the functions |
| | 465 | that are within the trace. The descriptions of the tracers |
| | 466 | will also show an example with ftrace enabled. |
| | 467 | |
| | 468 | |
| | 469 | irqsoff |
| | 470 | ------- |
| | 471 | |
| | 472 | When interrupts are disabled, the CPU can not react to any other |
| | 473 | external event (besides NMIs and SMIs). This prevents the timer |
| | 474 | interrupt from triggering or the mouse interrupt from letting the |
| | 475 | kernel know of a new mouse event. The result is a latency with the |
| | 476 | reaction time. |
| | 477 | |
| | 478 | The irqsoff tracer tracks the time for which interrupts are disabled. |
| | 479 | When a new maximum latency is hit, the tracer saves the trace leading up |
| | 480 | to that latency point so that every time a new maximum is reached, the old |
| | 481 | saved trace is discarded and the new trace is saved. |
| | 482 | |
| | 483 | To reset the maximum, echo 0 into tracing_max_latency. Here is an |
| | 484 | example: |
| | 485 | |
| | 486 | # echo irqsoff > /debug/tracing/current_tracer |
| | 487 | # echo 0 > /debug/tracing/tracing_max_latency |
| | 488 | # echo 1 > /debug/tracing/tracing_enabled |
| | 489 | # ls -ltr |
| | 490 | [...] |
| | 491 | # echo 0 > /debug/tracing/tracing_enabled |
| | 492 | # cat /debug/tracing/latency_trace |
| | 493 | # tracer: irqsoff |
| | 494 | # |
| | 495 | irqsoff latency trace v1.1.5 on 2.6.26 |
| | 496 | -------------------------------------------------------------------- |
| | 497 | latency: 12 us, #3/3, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) |
| | 498 | ----------------- |
| | 499 | | task: bash-3730 (uid:0 nice:0 policy:0 rt_prio:0) |
| | 500 | ----------------- |
| | 501 | => started at: sys_setpgid |
| | 502 | => ended at: sys_setpgid |
| | 503 | |
| | 504 | # _------=> CPU# |
| | 505 | # / _-----=> irqs-off |
| | 506 | # | / _----=> need-resched |
| | 507 | # || / _---=> hardirq/softirq |
| | 508 | # ||| / _--=> preempt-depth |
| | 509 | # |||| / |
| | 510 | # ||||| delay |
| | 511 | # cmd pid ||||| time | caller |
| | 512 | # \ / ||||| \ | / |
| | 513 | bash-3730 1d... 0us : _write_lock_irq (sys_setpgid) |
| | 514 | bash-3730 1d..1 1us+: _write_unlock_irq (sys_setpgid) |
| | 515 | bash-3730 1d..2 14us : trace_hardirqs_on (sys_setpgid) |
| | 516 | |
| | 517 | |
| | 518 | Here we see that that we had a latency of 12 microsecs (which is |
| | 519 | very good). The _write_lock_irq in sys_setpgid disabled interrupts. |
| | 520 | The difference between the 12 and the displayed timestamp 14us occurred |
| | 521 | because the clock was incremented between the time of recording the max |
| | 522 | latency and the time of recording the function that had that latency. |
| | 523 | |
| | 524 | Note the above example had ftrace_enabled not set. If we set the |
| | 525 | ftrace_enabled, we get a much larger output: |
| | 526 | |
| | 527 | # tracer: irqsoff |
| | 528 | # |
| | 529 | irqsoff latency trace v1.1.5 on 2.6.26-rc8 |
| | 530 | -------------------------------------------------------------------- |
| | 531 | latency: 50 us, #101/101, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) |
| | 532 | ----------------- |
| | 533 | | task: ls-4339 (uid:0 nice:0 policy:0 rt_prio:0) |
| | 534 | ----------------- |
| | 535 | => started at: __alloc_pages_internal |
| | 536 | => ended at: __alloc_pages_internal |
| | 537 | |
| | 538 | # _------=> CPU# |
| | 539 | # / _-----=> irqs-off |
| | 540 | # | / _----=> need-resched |
| | 541 | # || / _---=> hardirq/softirq |
| | 542 | # ||| / _--=> preempt-depth |
| | 543 | # |||| / |
| | 544 | # ||||| delay |
| | 545 | # cmd pid ||||| time | caller |
| | 546 | # \ / ||||| \ | / |
| | 547 | ls-4339 0...1 0us+: get_page_from_freelist (__alloc_pages_internal) |
| | 548 | ls-4339 0d..1 3us : rmqueue_bulk (get_page_from_freelist) |
| | 549 | ls-4339 0d..1 3us : _spin_lock (rmqueue_bulk) |
| | 550 | ls-4339 0d..1 4us : add_preempt_count (_spin_lock) |
| | 551 | ls-4339 0d..2 4us : __rmqueue (rmqueue_bulk) |
| | 552 | ls-4339 0d..2 5us : __rmqueue_smallest (__rmqueue) |
| | 553 | ls-4339 0d..2 5us : __mod_zone_page_state (__rmqueue_smallest) |
| | 554 | ls-4339 0d..2 6us : __rmqueue (rmqueue_bulk) |
| | 555 | ls-4339 0d..2 6us : __rmqueue_smallest (__rmqueue) |
| | 556 | ls-4339 0d..2 7us : __mod_zone_page_state (__rmqueue_smallest) |
| | 557 | ls-4339 0d..2 7us : __rmqueue (rmqueue_bulk) |
| | 558 | ls-4339 0d..2 8us : __rmqueue_smallest (__rmqueue) |
| | 559 | [...] |
| | 560 | ls-4339 0d..2 46us : __rmqueue_smallest (__rmqueue) |
| | 561 | ls-4339 0d..2 47us : __mod_zone_page_state (__rmqueue_smallest) |
| | 562 | ls-4339 0d..2 47us : __rmqueue (rmqueue_bulk) |
| | 563 | ls-4339 0d..2 48us : __rmqueue_smallest (__rmqueue) |
| | 564 | ls-4339 0d..2 48us : __mod_zone_page_state (__rmqueue_smallest) |
| | 565 | ls-4339 0d..2 49us : _spin_unlock (rmqueue_bulk) |
| | 566 | ls-4339 0d..2 49us : sub_preempt_count (_spin_unlock) |
| | 567 | ls-4339 0d..1 50us : get_page_from_freelist (__alloc_pages_internal) |
| | 568 | ls-4339 0d..2 51us : trace_hardirqs_on (__alloc_pages_internal) |
| | 569 | |
| | 570 | |
| | 571 | |
| | 572 | Here we traced a 50 microsecond latency. But we also see all the |
| | 573 | functions that were called during that time. Note that by enabling |
| | 574 | function tracing, we incur an added overhead. This overhead may |
| | 575 | extend the latency times. But nevertheless, this trace has provided |
| | 576 | some very helpful debugging information. |
| | 577 | |
| | 578 | |
| | 579 | preemptoff |
| | 580 | ---------- |
| | 581 | |
| | 582 | When preemption is disabled, we may be able to receive interrupts but |
| | 583 | the task cannot be preempted and a higher priority task must wait |
| | 584 | for preemption to be enabled again before it can preempt a lower |
| | 585 | priority task. |
| | 586 | |
| | 587 | The preemptoff tracer traces the places that disable preemption. |
| | 588 | Like the irqsoff tracer, it records the maximum latency for which preemption |
| | 589 | was disabled. The control of preemptoff tracer is much like the irqsoff |
| | 590 | tracer. |
| | 591 | |
| | 592 | # echo preemptoff > /debug/tracing/current_tracer |
| | 593 | # echo 0 > /debug/tracing/tracing_max_latency |
| | 594 | # echo 1 > /debug/tracing/tracing_enabled |
| | 595 | # ls -ltr |
| | 596 | [...] |
| | 597 | # echo 0 > /debug/tracing/tracing_enabled |
| | 598 | # cat /debug/tracing/latency_trace |
| | 599 | # tracer: preemptoff |
| | 600 | # |
| | 601 | preemptoff latency trace v1.1.5 on 2.6.26-rc8 |
| | 602 | -------------------------------------------------------------------- |
| | 603 | latency: 29 us, #3/3, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) |
| | 604 | ----------------- |
| | 605 | | task: sshd-4261 (uid:0 nice:0 policy:0 rt_prio:0) |
| | 606 | ----------------- |
| | 607 | => started at: do_IRQ |
| | 608 | => ended at: __do_softirq |
| | 609 | |
| | 610 | # _------=> CPU# |
| | 611 | # / _-----=> irqs-off |
| | 612 | # | / _----=> need-resched |
| | 613 | # || / _---=> hardirq/softirq |
| | 614 | # ||| / _--=> preempt-depth |
| | 615 | # |||| / |
| | 616 | # ||||| delay |
| | 617 | # cmd pid ||||| time | caller |
| | 618 | # \ / ||||| \ | / |
| | 619 | sshd-4261 0d.h. 0us+: irq_enter (do_IRQ) |
| | 620 | sshd-4261 0d.s. 29us : _local_bh_enable (__do_softirq) |
| | 621 | sshd-4261 0d.s1 30us : trace_preempt_on (__do_softirq) |
| | 622 | |
| | 623 | |
| | 624 | This has some more changes. Preemption was disabled when an interrupt |
| | 625 | came in (notice the 'h'), and was enabled while doing a softirq. |
| | 626 | (notice the 's'). But we also see that interrupts have been disabled |
| | 627 | when entering the preempt off section and leaving it (the 'd'). |
| | 628 | We do not know if interrupts were enabled in the mean time. |
| | 629 | |
| | 630 | # tracer: preemptoff |
| | 631 | # |
| | 632 | preemptoff latency trace v1.1.5 on 2.6.26-rc8 |
| | 633 | -------------------------------------------------------------------- |
| | 634 | latency: 63 us, #87/87, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) |
| | 635 | ----------------- |
| | 636 | | task: sshd-4261 (uid:0 nice:0 policy:0 rt_prio:0) |
| | 637 | ----------------- |
| | 638 | => started at: remove_wait_queue |
| | 639 | => ended at: __do_softirq |
| | 640 | |
| | 641 | # _------=> CPU# |
| | 642 | # / _-----=> irqs-off |
| | 643 | # | / _----=> need-resched |
| | 644 | # || / _---=> hardirq/softirq |
| | 645 | # ||| / _--=> preempt-depth |
| | 646 | # |||| / |
| | 647 | # ||||| delay |
| | 648 | # cmd pid ||||| time | caller |
| | 649 | # \ / ||||| \ | / |
| | 650 | sshd-4261 0d..1 0us : _spin_lock_irqsave (remove_wait_queue) |
| | 651 | sshd-4261 0d..1 1us : _spin_unlock_irqrestore (remove_wait_queue) |
| | 652 | sshd-4261 0d..1 2us : do_IRQ (common_interrupt) |
| | 653 | sshd-4261 0d..1 2us : irq_enter (do_IRQ) |
| | 654 | sshd-4261 0d..1 2us : idle_cpu (irq_enter) |
| | 655 | sshd-4261 0d..1 3us : add_preempt_count (irq_enter) |
| | 656 | sshd-4261 0d.h1 3us : idle_cpu (irq_enter) |
| | 657 | sshd-4261 0d.h. 4us : handle_fasteoi_irq (do_IRQ) |
| | 658 | [...] |
| | 659 | sshd-4261 0d.h. 12us : add_preempt_count (_spin_lock) |
| | 660 | sshd-4261 0d.h1 12us : ack_ioapic_quirk_irq (handle_fasteoi_irq) |
| | 661 | sshd-4261 0d.h1 13us : move_native_irq (ack_ioapic_quirk_irq) |
| | 662 | sshd-4261 0d.h1 13us : _spin_unlock (handle_fasteoi_irq) |
| | 663 | sshd-4261 0d.h1 14us : sub_preempt_count (_spin_unlock) |
| | 664 | sshd-4261 0d.h1 14us : irq_exit (do_IRQ) |
| | 665 | sshd-4261 0d.h1 15us : sub_preempt_count (irq_exit) |
| | 666 | sshd-4261 0d..2 15us : do_softirq (irq_exit) |
| | 667 | sshd-4261 0d... 15us : __do_softirq (do_softirq) |
| | 668 | sshd-4261 0d... 16us : __local_bh_disable (__do_softirq) |
| | 669 | sshd-4261 0d... 16us+: add_preempt_count (__local_bh_disable) |
| | 670 | sshd-4261 0d.s4 20us : add_preempt_count (__local_bh_disable) |
| | 671 | sshd-4261 0d.s4 21us : sub_preempt_count (local_bh_enable) |
| | 672 | sshd-4261 0d.s5 21us : sub_preempt_count (local_bh_enable) |
| | 673 | [...] |
| | 674 | sshd-4261 0d.s6 41us : add_preempt_count (__local_bh_disable) |
| | 675 | sshd-4261 0d.s6 42us : sub_preempt_count (local_bh_enable) |
| | 676 | sshd-4261 0d.s7 42us : sub_preempt_count (local_bh_enable) |
| | 677 | sshd-4261 0d.s5 43us : add_preempt_count (__local_bh_disable) |
| | 678 | sshd-4261 0d.s5 43us : sub_preempt_count (local_bh_enable_ip) |
| | 679 | sshd-4261 0d.s6 44us : sub_preempt_count (local_bh_enable_ip) |
| | 680 | sshd-4261 0d.s5 44us : add_preempt_count (__local_bh_disable) |
| | 681 | sshd-4261 0d.s5 45us : sub_preempt_count (local_bh_enable) |
| | 682 | [...] |
| | 683 | sshd-4261 0d.s. 63us : _local_bh_enable (__do_softirq) |
| | 684 | sshd-4261 0d.s1 64us : trace_preempt_on (__do_softirq) |
| | 685 | |
| | 686 | |
| | 687 | The above is an example of the preemptoff trace with ftrace_enabled |
| | 688 | set. Here we see that interrupts were disabled the entire time. |
| | 689 | The irq_enter code lets us know that we entered an interrupt 'h'. |
| | 690 | Before that, the functions being traced still show that it is not |
| | 691 | in an interrupt, but we can see from the functions themselves that |
| | 692 | this is not the case. |
| | 693 | |
| | 694 | Notice that __do_softirq when called does not have a preempt_count. |
| | 695 | It may seem that we missed a preempt enabling. What really happened |
| | 696 | is that the preempt count is held on the thread's stack and we |
| | 697 | switched to the softirq stack (4K stacks in effect). The code |
| | 698 | does not copy the preempt count, but because interrupts are disabled, |
| | 699 | we do not need to worry about it. Having a tracer like this is good |
| | 700 | for letting people know what really happens inside the kernel. |
| | 701 | |
| | 702 | |
| | 703 | preemptirqsoff |
| | 704 | -------------- |
| | 705 | |
| | 706 | Knowing the locations that have interrupts disabled or preemption |
| | 707 | disabled for the longest times is helpful. But sometimes we would |
| | 708 | like to know when either preemption and/or interrupts are disabled. |
| | 709 | |
| | 710 | Consider the following code: |
| | 711 | |
| | 712 | local_irq_disable(); |
| | 713 | call_function_with_irqs_off(); |
| | 714 | preempt_disable(); |
| | 715 | call_function_with_irqs_and_preemption_off(); |
| | 716 | local_irq_enable(); |
| | 717 | call_function_with_preemption_off(); |
| | 718 | preempt_enable(); |
| | 719 | |
| | 720 | The irqsoff tracer will record the total length of |
| | 721 | call_function_with_irqs_off() and |
| | 722 | call_function_with_irqs_and_preemption_off(). |
| | 723 | |
| | 724 | The preemptoff tracer will record the total length of |
| | 725 | call_function_with_irqs_and_preemption_off() and |
| | 726 | call_function_with_preemption_off(). |
| | 727 | |
| | 728 | But neither will trace the time that interrupts and/or preemption |
| | 729 | is disabled. This total time is the time that we can not schedule. |
| | 730 | To record this time, use the preemptirqsoff tracer. |
| | 731 | |
| | 732 | Again, using this trace is much like the irqsoff and preemptoff tracers. |
| | 733 | |
| | 734 | # echo preemptirqsoff > /debug/tracing/current_tracer |
| | 735 | # echo 0 > /debug/tracing/tracing_max_latency |
| | 736 | # echo 1 > /debug/tracing/tracing_enabled |
| | 737 | # ls -ltr |
| | 738 | [...] |
| | 739 | # echo 0 > /debug/tracing/tracing_enabled |
| | 740 | # cat /debug/tracing/latency_trace |
| | 741 | # tracer: preemptirqsoff |
| | 742 | # |
| | 743 | preemptirqsoff latency trace v1.1.5 on 2.6.26-rc8 |
| | 744 | -------------------------------------------------------------------- |
| | 745 | latency: 293 us, #3/3, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) |
| | 746 | ----------------- |
| | 747 | | task: ls-4860 (uid:0 nice:0 policy:0 rt_prio:0) |
| | 748 | ----------------- |
| | 749 | => started at: apic_timer_interrupt |
| | 750 | => ended at: __do_softirq |
| | 751 | |
| | 752 | # _------=> CPU# |
| | 753 | # / _-----=> irqs-off |
| | 754 | # | / _----=> need-resched |
| | 755 | # || / _---=> hardirq/softirq |
| | 756 | # ||| / _--=> preempt-depth |
| | 757 | # |||| / |
| | 758 | # ||||| delay |
| | 759 | # cmd pid ||||| time | caller |
| | 760 | # \ / ||||| \ | / |
| | 761 | ls-4860 0d... 0us!: trace_hardirqs_off_thunk (apic_timer_interrupt) |
| | 762 | ls-4860 0d.s. 294us : _local_bh_enable (__do_softirq) |
| | 763 | ls-4860 0d.s1 294us : trace_preempt_on (__do_softirq) |
| | 764 | |
| | 765 | |
| | 766 | |
| | 767 | The trace_hardirqs_off_thunk is called from assembly on x86 when |
| | 768 | interrupts are disabled in the assembly code. Without the function |
| | 769 | tracing, we do not know if interrupts were enabled within the preemption |
| | 770 | points. We do see that it started with preemption enabled. |
| | 771 | |
| | 772 | Here is a trace with ftrace_enabled set: |
| | 773 | |
| | 774 | |
| | 775 | # tracer: preemptirqsoff |
| | 776 | # |
| | 777 | preemptirqsoff latency trace v1.1.5 on 2.6.26-rc8 |
| | 778 | -------------------------------------------------------------------- |
| | 779 | latency: 105 us, #183/183, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) |
| | 780 | ----------------- |
| | 781 | | task: sshd-4261 (uid:0 nice:0 policy:0 rt_prio:0) |
| | 782 | ----------------- |
| | 783 | => started at: write_chan |
| | 784 | => ended at: __do_softirq |
| | 785 | |
| | 786 | # _------=> CPU# |
| | 787 | # / _-----=> irqs-off |
| | 788 | # | / _----=> need-resched |
| | 789 | # || / _---=> hardirq/softirq |
| | 790 | # ||| / _--=> preempt-depth |
| | 791 | # |||| / |
| | 792 | # ||||| delay |
| | 793 | # cmd pid ||||| time | caller |
| | 794 | # \ / ||||| \ | / |
| | 795 | ls-4473 0.N.. 0us : preempt_schedule (write_chan) |
| | 796 | ls-4473 0dN.1 1us : _spin_lock (schedule) |
| | 797 | ls-4473 0dN.1 2us : add_preempt_count (_spin_lock) |
| | 798 | ls-4473 0d..2 2us : put_prev_task_fair (schedule) |
| | 799 | [...] |
| | 800 | ls-4473 0d..2 13us : set_normalized_timespec (ktime_get_ts) |
| | 801 | ls-4473 0d..2 13us : __switch_to (schedule) |
| | 802 | sshd-4261 0d..2 14us : finish_task_switch (schedule) |
| | 803 | sshd-4261 0d..2 14us : _spin_unlock_irq (finish_task_switch) |
| | 804 | sshd-4261 0d..1 15us : add_preempt_count (_spin_lock_irqsave) |
| | 805 | sshd-4261 0d..2 16us : _spin_unlock_irqrestore (hrtick_set) |
| | 806 | sshd-4261 0d..2 16us : do_IRQ (common_interrupt) |
| | 807 | sshd-4261 0d..2 17us : irq_enter (do_IRQ) |
| | 808 | sshd-4261 0d..2 17us : idle_cpu (irq_enter) |
| | 809 | sshd-4261 0d..2 18us : add_preempt_count (irq_enter) |
| | 810 | sshd-4261 0d.h2 18us : idle_cpu (irq_enter) |
| | 811 | sshd-4261 0d.h. 18us : handle_fasteoi_irq (do_IRQ) |
| | 812 | sshd-4261 0d.h. 19us : _spin_lock (handle_fasteoi_irq) |
| | 813 | sshd-4261 0d.h. 19us : add_preempt_count (_spin_lock) |
| | 814 | sshd-4261 0d.h1 20us : _spin_unlock (handle_fasteoi_irq) |
| | 815 | sshd-4261 0d.h1 20us : sub_preempt_count (_spin_unlock) |
| | 816 | [...] |
| | 817 | sshd-4261 0d.h1 28us : _spin_unlock (handle_fasteoi_irq) |
| | 818 | sshd-4261 0d.h1 29us : sub_preempt_count (_spin_unlock) |
| | 819 | sshd-4261 0d.h2 29us : irq_exit (do_IRQ) |
| | 820 | sshd-4261 0d.h2 29us : sub_preempt_count (irq_exit) |
| | 821 | sshd-4261 0d..3 30us : do_softirq (irq_exit) |
| | 822 | sshd-4261 0d... 30us : __do_softirq (do_softirq) |
| | 823 | sshd-4261 0d... 31us : __local_bh_disable (__do_softirq) |
| | 824 | sshd-4261 0d... 31us+: add_preempt_count (__local_bh_disable) |
| | 825 | sshd-4261 0d.s4 34us : add_preempt_count (__local_bh_disable) |
| | 826 | [...] |
| | 827 | sshd-4261 0d.s3 43us : sub_preempt_count (local_bh_enable_ip) |
| | 828 | sshd-4261 0d.s4 44us : sub_preempt_count (local_bh_enable_ip) |
| | 829 | sshd-4261 0d.s3 44us : smp_apic_timer_interrupt (apic_timer_interrupt) |
| | 830 | sshd-4261 0d.s3 45us : irq_enter (smp_apic_timer_interrupt) |
| | 831 | sshd-4261 0d.s3 45us : idle_cpu (irq_enter) |
| | 832 | sshd-4261 0d.s3 46us : add_preempt_count (irq_enter) |
| | 833 | sshd-4261 0d.H3 46us : idle_cpu (irq_enter) |
| | 834 | sshd-4261 0d.H3 47us : hrtimer_interrupt (smp_apic_timer_interrupt) |
| | 835 | sshd-4261 0d.H3 47us : ktime_get (hrtimer_interrupt) |
| | 836 | [...] |
| | 837 | sshd-4261 0d.H3 81us : tick_program_event (hrtimer_interrupt) |
| | 838 | sshd-4261 0d.H3 82us : ktime_get (tick_program_event) |
| | 839 | sshd-4261 0d.H3 82us : ktime_get_ts (ktime_get) |
| | 840 | sshd-4261 0d.H3 83us : getnstimeofday (ktime_get_ts) |
| | 841 | sshd-4261 0d.H3 83us : set_normalized_timespec (ktime_get_ts) |
| | 842 | sshd-4261 0d.H3 84us : clockevents_program_event (tick_program_event) |
| | 843 | sshd-4261 0d.H3 84us : lapic_next_event (clockevents_program_event) |
| | 844 | sshd-4261 0d.H3 85us : irq_exit (smp_apic_timer_interrupt) |
| | 845 | sshd-4261 0d.H3 85us : sub_preempt_count (irq_exit) |
| | 846 | sshd-4261 0d.s4 86us : sub_preempt_count (irq_exit) |
| | 847 | sshd-4261 0d.s3 86us : add_preempt_count (__local_bh_disable) |
| | 848 | [...] |
| | 849 | sshd-4261 0d.s1 98us : sub_preempt_count (net_rx_action) |
| | 850 | sshd-4261 0d.s. 99us : add_preempt_count (_spin_lock_irq) |
| | 851 | sshd-4261 0d.s1 99us+: _spin_unlock_irq (run_timer_softirq) |
| | 852 | sshd-4261 0d.s. 104us : _local_bh_enable (__do_softirq) |
| | 853 | sshd-4261 0d.s. 104us : sub_preempt_count (_local_bh_enable) |
| | 854 | sshd-4261 0d.s. 105us : _local_bh_enable (__do_softirq) |
| | 855 | sshd-4261 0d.s1 105us : trace_preempt_on (__do_softirq) |
| | 856 | |
| | 857 | |
| | 858 | This is a very interesting trace. It started with the preemption of |
| | 859 | the ls task. We see that the task had the "need_resched" bit set |
| | 860 | via the 'N' in the trace. Interrupts were disabled before the spin_lock |
| | 861 | at the beginning of the trace. We see that a schedule took place to run |
| | 862 | sshd. When the interrupts were enabled, we took an interrupt. |
| | 863 | On return from the interrupt handler, the softirq ran. We took another |
| | 864 | interrupt while running the softirq as we see from the capital 'H'. |
| | 865 | |
| | 866 | |
| | 867 | wakeup |
| | 868 | ------ |
| | 869 | |
| | 870 | In a Real-Time environment it is very important to know the wakeup |
| | 871 | time it takes for the highest priority task that is woken up to the |
| | 872 | time that it executes. This is also known as "schedule latency". |
| | 873 | I stress the point that this is about RT tasks. It is also important |
| | 874 | to know the scheduling latency of non-RT tasks, but the average |
| | 875 | schedule latency is better for non-RT tasks. Tools like |
| | 876 | LatencyTop are more appropriate for such measurements. |
| | 877 | |
| | 878 | Real-Time environments are interested in the worst case latency. |
| | 879 | That is the longest latency it takes for something to happen, and |
| | 880 | not the average. We can have a very fast scheduler that may only |
| | 881 | have a large latency once in a while, but that would not work well |
| | 882 | with Real-Time tasks. The wakeup tracer was designed to record |
| | 883 | the worst case wakeups of RT tasks. Non-RT tasks are not recorded |
| | 884 | because the tracer only records one worst case and tracing non-RT |
| | 885 | tasks that are unpredictable will overwrite the worst case latency |
| | 886 | of RT tasks. |
| | 887 | |
| | 888 | Since this tracer only deals with RT tasks, we will run this slightly |
| | 889 | differently than we did with the previous tracers. Instead of performing |
| | 890 | an 'ls', we will run 'sleep 1' under 'chrt' which changes the |
| | 891 | priority of the task. |
| | 892 | |
| | 893 | # echo wakeup > /debug/tracing/current_tracer |
| | 894 | # echo 0 > /debug/tracing/tracing_max_latency |
| | 895 | # echo 1 > /debug/tracing/tracing_enabled |
| | 896 | # chrt -f 5 sleep 1 |
| | 897 | # echo 0 > /debug/tracing/tracing_enabled |
| | 898 | # cat /debug/tracing/latency_trace |
| | 899 | # tracer: wakeup |
| | 900 | # |
| | 901 | wakeup latency trace v1.1.5 on 2.6.26-rc8 |
| | 902 | -------------------------------------------------------------------- |
| | 903 | latency: 4 us, #2/2, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) |
| | 904 | ----------------- |
| | 905 | | task: sleep-4901 (uid:0 nice:0 policy:1 rt_prio:5) |
| | 906 | ----------------- |
| | 907 | |
| | 908 | # _------=> CPU# |
| | 909 | # / _-----=> irqs-off |
| | 910 | # | / _----=> need-resched |
| | 911 | # || / _---=> hardirq/softirq |
| | 912 | # ||| / _--=> preempt-depth |
| | 913 | # |||| / |
| | 914 | # ||||| delay |
| | 915 | # cmd pid ||||| time | caller |
| | 916 | # \ / ||||| \ | / |
| | 917 | <idle>-0 1d.h4 0us+: try_to_wake_up (wake_up_process) |
| | 918 | <idle>-0 1d..4 4us : schedule (cpu_idle) |
| | 919 | |
| | 920 | |
| | 921 | |
| | 922 | Running this on an idle system, we see that it only took 4 microseconds |
| | 923 | to perform the task switch. Note, since the trace marker in the |
| | 924 | schedule is before the actual "switch", we stop the tracing when |
| | 925 | the recorded task is about to schedule in. This may change if |
| | 926 | we add a new marker at the end of the scheduler. |
| | 927 | |
| | 928 | Notice that the recorded task is 'sleep' with the PID of 4901 and it |
| | 929 | has an rt_prio of 5. This priority is user-space priority and not |
| | 930 | the internal kernel priority. The policy is 1 for SCHED_FIFO and 2 |
| | 931 | for SCHED_RR. |
| | 932 | |
| | 933 | Doing the same with chrt -r 5 and ftrace_enabled set. |
| | 934 | |
| | 935 | # tracer: wakeup |
| | 936 | # |
| | 937 | wakeup latency trace v1.1.5 on 2.6.26-rc8 |
| | 938 | -------------------------------------------------------------------- |
| | 939 | latency: 50 us, #60/60, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) |
| | 940 | ----------------- |
| | 941 | | task: sleep-4068 (uid:0 nice:0 policy:2 rt_prio:5) |
| | 942 | ----------------- |
| | 943 | |
| | 944 | # _------=> CPU# |
| | 945 | # / _-----=> irqs-off |
| | 946 | # | / _----=> need-resched |
| | 947 | # || / _---=> hardirq/softirq |
| | 948 | # ||| / _--=> preempt-depth |
| | 949 | # |||| / |
| | 950 | # ||||| delay |
| | 951 | # cmd pid ||||| time | caller |
| | 952 | # \ / ||||| \ | / |
| | 953 | ksoftirq-7 1d.H3 0us : try_to_wake_up (wake_up_process) |
| | 954 | ksoftirq-7 1d.H4 1us : sub_preempt_count (marker_probe_cb) |
| | 955 | ksoftirq-7 1d.H3 2us : check_preempt_wakeup (try_to_wake_up) |
| | 956 | ksoftirq-7 1d.H3 3us : update_curr (check_preempt_wakeup) |
| | 957 | ksoftirq-7 1d.H3 4us : calc_delta_mine (update_curr) |
| | 958 | ksoftirq-7 1d.H3 5us : __resched_task (check_preempt_wakeup) |
| | 959 | ksoftirq-7 1d.H3 6us : task_wake_up_rt (try_to_wake_up) |
| | 960 | ksoftirq-7 1d.H3 7us : _spin_unlock_irqrestore (try_to_wake_up) |
| | 961 | [...] |
| | 962 | ksoftirq-7 1d.H2 17us : irq_exit (smp_apic_timer_interrupt) |
| | 963 | ksoftirq-7 1d.H2 18us : sub_preempt_count (irq_exit) |
| | 964 | ksoftirq-7 1d.s3 19us : sub_preempt_count (irq_exit) |
| | 965 | ksoftirq-7 1..s2 20us : rcu_process_callbacks (__do_softirq) |
| | 966 | [...] |
| | 967 | ksoftirq-7 1..s2 26us : __rcu_process_callbacks (rcu_process_callbacks) |
| | 968 | ksoftirq-7 1d.s2 27us : _local_bh_enable (__do_softirq) |
| | 969 | ksoftirq-7 1d.s2 28us : sub_preempt_count (_local_bh_enable) |
| | 970 | ksoftirq-7 1.N.3 29us : sub_preempt_count (ksoftirqd) |
| | 971 | ksoftirq-7 1.N.2 30us : _cond_resched (ksoftirqd) |
| | 972 | ksoftirq-7 1.N.2 31us : __cond_resched (_cond_resched) |
| | 973 | ksoftirq-7 1.N.2 32us : add_preempt_count (__cond_resched) |
| | 974 | ksoftirq-7 1.N.2 33us : schedule (__cond_resched) |
| | 975 | ksoftirq-7 1.N.2 33us : add_preempt_count (schedule) |
| | 976 | ksoftirq-7 1.N.3 34us : hrtick_clear (schedule) |
| | 977 | ksoftirq-7 1dN.3 35us : _spin_lock (schedule) |
| | 978 | ksoftirq-7 1dN.3 36us : add_preempt_count (_spin_lock) |
| | 979 | ksoftirq-7 1d..4 37us : put_prev_task_fair (schedule) |
| | 980 | ksoftirq-7 1d..4 38us : update_curr (put_prev_task_fair) |
| | 981 | [...] |
| | 982 | ksoftirq-7 1d..5 47us : _spin_trylock (tracing_record_cmdline) |
| | 983 | ksoftirq-7 1d..5 48us : add_preempt_count (_spin_trylock) |
| | 984 | ksoftirq-7 1d..6 49us : _spin_unlock (tracing_record_cmdline) |
| | 985 | ksoftirq-7 1d..6 49us : sub_preempt_count (_spin_unlock) |
| | 986 | ksoftirq-7 1d..4 50us : schedule (__cond_resched) |
| | 987 | |
| | 988 | The interrupt went off while running ksoftirqd. This task runs at |
| | 989 | SCHED_OTHER. Why did not we see the 'N' set early? This may be |
| | 990 | a harmless bug with x86_32 and 4K stacks. On x86_32 with 4K stacks |
| | 991 | configured, the interrupt and softirq run with their own stack. |
| | 992 | Some information is held on the top of the task's stack (need_resched |
| | 993 | and preempt_count are both stored there). The setting of the NEED_RESCHED |
| | 994 | bit is done directly to the task's stack, but the reading of the |
| | 995 | NEED_RESCHED is done by looking at the current stack, which in this case |
| | 996 | is the stack for the hard interrupt. This hides the fact that NEED_RESCHED |
| | 997 | has been set. We do not see the 'N' until we switch back to the task's |
| | 998 | assigned stack. |
| | 999 | |
| | 1000 | function |
| | 1001 | -------- |
| | 1002 | |
| | 1003 | This tracer is the function tracer. Enabling the function tracer |
| | 1004 | can be done from the debug file system. Make sure the ftrace_enabled is |
| | 1005 | set; otherwise this tracer is a nop. |
| | 1006 | |
| | 1007 | # sysctl kernel.ftrace_enabled=1 |
| | 1008 | # echo function > /debug/tracing/current_tracer |
| | 1009 | # echo 1 > /debug/tracing/tracing_enabled |
| | 1010 | # usleep 1 |
| | 1011 | # echo 0 > /debug/tracing/tracing_enabled |
| | 1012 | # cat /debug/tracing/trace |
| | 1013 | # tracer: function |
| | 1014 | # |
| | 1015 | # TASK-PID CPU# TIMESTAMP FUNCTION |
| | 1016 | # | | | | | |
| | 1017 | bash-4003 [00] 123.638713: finish_task_switch <-schedule |
| | 1018 | bash-4003 [00] 123.638714: _spin_unlock_irq <-finish_task_switch |
| | 1019 | bash-4003 [00] 123.638714: sub_preempt_count <-_spin_unlock_irq |
| | 1020 | bash-4003 [00] 123.638715: hrtick_set <-schedule |
| | 1021 | bash-4003 [00] 123.638715: _spin_lock_irqsave <-hrtick_set |
| | 1022 | bash-4003 [00] 123.638716: add_preempt_count <-_spin_lock_irqsave |
| | 1023 | bash-4003 [00] 123.638716: _spin_unlock_irqrestore <-hrtick_set |
| | 1024 | bash-4003 [00] 123.638717: sub_preempt_count <-_spin_unlock_irqrestore |
| | 1025 | bash-4003 [00] 123.638717: hrtick_clear <-hrtick_set |
| | 1026 | bash-4003 [00] 123.638718: sub_preempt_count <-schedule |
| | 1027 | bash-4003 [00] 123.638718: sub_preempt_count <-preempt_schedule |
| | 1028 | bash-4003 [00] 123.638719: wait_for_completion <-__stop_machine_run |
| | 1029 | bash-4003 [00] 123.638719: wait_for_common <-wait_for_completion |
| | 1030 | bash-4003 [00] 123.638720: _spin_lock_irq <-wait_for_common |
| | 1031 | bash-4003 [00] 123.638720: add_preempt_count <-_spin_lock_irq |
| | 1032 | [...] |
| | 1033 | |
| | 1034 | |
| | 1035 | Note: function tracer uses ring buffers to store the above entries. |
| | 1036 | The newest data may overwrite the oldest data. Sometimes using echo to |
| | 1037 | stop the trace is not sufficient because the tracing could have overwritten |
| | 1038 | the data that you wanted to record. For this reason, it is sometimes better to |
| | 1039 | disable tracing directly from a program. This allows you to stop the |
| | 1040 | tracing at the point that you hit the part that you are interested in. |
| | 1041 | To disable the tracing directly from a C program, something like following |
| | 1042 | code snippet can be used: |
| | 1043 | |
| | 1044 | int trace_fd; |
| | 1045 | [...] |
| | 1046 | int main(int argc, char *argv[]) { |
| | 1047 | [...] |
| | 1048 | trace_fd = open("/debug/tracing/tracing_enabled", O_WRONLY); |
| | 1049 | [...] |
| | 1050 | if (condition_hit()) { |
| | 1051 | write(trace_fd, "0", 1); |
| | 1052 | } |
| | 1053 | [...] |
| | 1054 | } |
| | 1055 | |
| | 1056 | Note: Here we hard coded the path name. The debugfs mount is not |
| | 1057 | guaranteed to be at /debug (and is more commonly at /sys/kernel/debug). |
| | 1058 | For simple one time traces, the above is sufficent. For anything else, |
| | 1059 | a search through /proc/mounts may be needed to find where the debugfs |
| | 1060 | file-system is mounted. |
| | 1061 | |
| | 1062 | dynamic ftrace |
| | 1063 | -------------- |
| | 1064 | |
| | 1065 | If CONFIG_DYNAMIC_FTRACE is set, the system will run with |
| | 1066 | virtually no overhead when function tracing is disabled. The way |
| | 1067 | this works is the mcount function call (placed at the start of |
| | 1068 | every kernel function, produced by the -pg switch in gcc), starts |
| | 1069 | of pointing to a simple return. (Enabling FTRACE will include the |
| | 1070 | -pg switch in the compiling of the kernel.) |
| | 1071 | |
| | 1072 | At compile time every C file object is run through the |
| | 1073 | recordmcount.pl script (located in the scripts directory). This |
| | 1074 | script will process the C object using objdump to find all the |
| | 1075 | locations in the .text section that call mcount. (Note, only |
| | 1076 | the .text section is processed, since processing other sections |
| | 1077 | like .init.text may cause races due to those sections being freed). |
| | 1078 | |
| | 1079 | A new section called "__mcount_loc" is created that holds references |
| | 1080 | to all the mcount call sites in the .text section. This section is |
| | 1081 | compiled back into the original object. The final linker will add |
| | 1082 | all these references into a single table. |
| | 1083 | |
| | 1084 | On boot up, before SMP is initialized, the dynamic ftrace code |
| | 1085 | scans this table and updates all the locations into nops. It also |
| | 1086 | records the locations, which are added to the available_filter_functions |
| | 1087 | list. Modules are processed as they are loaded and before they are |
| | 1088 | executed. When a module is unloaded, it also removes its functions from |
| | 1089 | the ftrace function list. This is automatic in the module unload |
| | 1090 | code, and the module author does not need to worry about it. |
| | 1091 | |
| | 1092 | When tracing is enabled, kstop_machine is called to prevent races |
| | 1093 | with the CPUS executing code being modified (which can cause the |
| | 1094 | CPU to do undesireable things), and the nops are patched back |
| | 1095 | to calls. But this time, they do not call mcount (which is just |
| | 1096 | a function stub). They now call into the ftrace infrastructure. |
| | 1097 | |
| | 1098 | One special side-effect to the recording of the functions being |
| | 1099 | traced is that we can now selectively choose which functions we |
| | 1100 | wish to trace and which ones we want the mcount calls to remain as |
| | 1101 | nops. |
| | 1102 | |
| | 1103 | Two files are used, one for enabling and one for disabling the tracing |
| | 1104 | of specified functions. They are: |
| | 1105 | |
| | 1106 | set_ftrace_filter |
| | 1107 | |
| | 1108 | and |
| | 1109 | |
| | 1110 | set_ftrace_notrace |
| | 1111 | |
| | 1112 | A list of available functions that you can add to these files is listed |
| | 1113 | in: |
| | 1114 | |
| | 1115 | available_filter_functions |
| | 1116 | |
| | 1117 | # cat /debug/tracing/available_filter_functions |
| | 1118 | put_prev_task_idle |
| | 1119 | kmem_cache_create |
| | 1120 | pick_next_task_rt |
| | 1121 | get_online_cpus |
| | 1122 | pick_next_task_fair |
| | 1123 | mutex_lock |
| | 1124 | [...] |
| | 1125 | |
| | 1126 | If I am only interested in sys_nanosleep and hrtimer_interrupt: |
| | 1127 | |
| | 1128 | # echo sys_nanosleep hrtimer_interrupt \ |
| | 1129 | > /debug/tracing/set_ftrace_filter |
| | 1130 | # echo ftrace > /debug/tracing/current_tracer |
| | 1131 | # echo 1 > /debug/tracing/tracing_enabled |
| | 1132 | # usleep 1 |
| | 1133 | # echo 0 > /debug/tracing/tracing_enabled |
| | 1134 | # cat /debug/tracing/trace |
| | 1135 | # tracer: ftrace |
| | 1136 | # |
| | 1137 | # TASK-PID CPU# TIMESTAMP FUNCTION |
| | 1138 | # | | | | | |
| | 1139 | usleep-4134 [00] 1317.070017: hrtimer_interrupt <-smp_apic_timer_interrupt |
| | 1140 | usleep-4134 [00] 1317.070111: sys_nanosleep <-syscall_call |
| | 1141 | <idle>-0 [00] 1317.070115: hrtimer_interrupt <-smp_apic_timer_interrupt |
| | 1142 | |
| | 1143 | To see which functions are being traced, you can cat the file: |
| | 1144 | |
| | 1145 | # cat /debug/tracing/set_ftrace_filter |
| | 1146 | hrtimer_interrupt |
| | 1147 | sys_nanosleep |
| | 1148 | |
| | 1149 | |
| | 1150 | Perhaps this is not enough. The filters also allow simple wild cards. |
| | 1151 | Only the following are currently available |
| | 1152 | |
| | 1153 | <match>* - will match functions that begin with <match> |
| | 1154 | *<match> - will match functions that end with <match> |
| | 1155 | *<match>* - will match functions that have <match> in it |
| | 1156 | |
| | 1157 | These are the only wild cards which are supported. |
| | 1158 | |
| | 1159 | <match>*<match> will not work. |
| | 1160 | |
| | 1161 | # echo hrtimer_* > /debug/tracing/set_ftrace_filter |
| | 1162 | |
| | 1163 | Produces: |
| | 1164 | |
| | 1165 | # tracer: ftrace |
| | 1166 | # |
| | 1167 | # TASK-PID CPU# TIMESTAMP FUNCTION |
| | 1168 | # | | | | | |
| | 1169 | bash-4003 [00] 1480.611794: hrtimer_init <-copy_process |
| | 1170 | bash-4003 [00] 1480.611941: hrtimer_start <-hrtick_set |
| | 1171 | bash-4003 [00] 1480.611956: hrtimer_cancel <-hrtick_clear |
| | 1172 | bash-4003 [00] 1480.611956: hrtimer_try_to_cancel <-hrtimer_cancel |
| | 1173 | <idle>-0 [00] 1480.612019: hrtimer_get_next_event <-get_next_timer_interrupt |
| | 1174 | <idle>-0 [00] 1480.612025: hrtimer_get_next_event <-get_next_timer_interrupt |
| | 1175 | <idle>-0 [00] 1480.612032: hrtimer_get_next_event <-get_next_timer_interrupt |
| | 1176 | <idle>-0 [00] 1480.612037: hrtimer_get_next_event <-get_next_timer_interrupt |
| | 1177 | <idle>-0 [00] 1480.612382: hrtimer_get_next_event <-get_next_timer_interrupt |
| | 1178 | |
| | 1179 | |
| | 1180 | Notice that we lost the sys_nanosleep. |
| | 1181 | |
| | 1182 | # cat /debug/tracing/set_ftrace_filter |
| | 1183 | hrtimer_run_queues |
| | 1184 | hrtimer_run_pending |
| | 1185 | hrtimer_init |
| | 1186 | hrtimer_cancel |
| | 1187 | hrtimer_try_to_cancel |
| | 1188 | hrtimer_forward |
| | 1189 | hrtimer_start |
| | 1190 | hrtimer_reprogram |
| | 1191 | hrtimer_force_reprogram |
| | 1192 | hrtimer_get_next_event |
| | 1193 | hrtimer_interrupt |
| | 1194 | hrtimer_nanosleep |
| | 1195 | hrtimer_wakeup |
| | 1196 | hrtimer_get_remaining |
| | 1197 | hrtimer_get_res |
| | 1198 | hrtimer_init_sleeper |
| | 1199 | |
| | 1200 | |
| | 1201 | This is because the '>' and '>>' act just like they do in bash. |
| | 1202 | To rewrite the filters, use '>' |
| | 1203 | To append to the filters, use '>>' |
| | 1204 | |
| | 1205 | To clear out a filter so that all functions will be recorded again: |
| | 1206 | |
| | 1207 | # echo > /debug/tracing/set_ftrace_filter |
| | 1208 | # cat /debug/tracing/set_ftrace_filter |
| | 1209 | # |
| | 1210 | |
| | 1211 | Again, now we want to append. |
| | 1212 | |
| | 1213 | # echo sys_nanosleep > /debug/tracing/set_ftrace_filter |
| | 1214 | # cat /debug/tracing/set_ftrace_filter |
| | 1215 | sys_nanosleep |
| | 1216 | # echo hrtimer_* >> /debug/tracing/set_ftrace_filter |
| | 1217 | # cat /debug/tracing/set_ftrace_filter |
| | 1218 | hrtimer_run_queues |
| | 1219 | hrtimer_run_pending |
| | 1220 | hrtimer_init |
| | 1221 | hrtimer_cancel |
| | 1222 | hrtimer_try_to_cancel |
| | 1223 | hrtimer_forward |
| | 1224 | hrtimer_start |
| | 1225 | hrtimer_reprogram |
| | 1226 | hrtimer_force_reprogram |
| | 1227 | hrtimer_get_next_event |
| | 1228 | hrtimer_interrupt |
| | 1229 | sys_nanosleep |
| | 1230 | hrtimer_nanosleep |
| | 1231 | hrtimer_wakeup |
| | 1232 | hrtimer_get_remaining |
| | 1233 | hrtimer_get_res |
| | 1234 | hrtimer_init_sleeper |
| | 1235 | |
| | 1236 | |
| | 1237 | The set_ftrace_notrace prevents those functions from being traced. |
| | 1238 | |
| | 1239 | # echo '*preempt*' '*lock*' > /debug/tracing/set_ftrace_notrace |
| | 1240 | |
| | 1241 | Produces: |
| | 1242 | |
| | 1243 | # tracer: ftrace |
| | 1244 | # |
| | 1245 | # TASK-PID CPU# TIMESTAMP FUNCTION |
| | 1246 | # | | | | | |
| | 1247 | bash-4043 [01] 115.281644: finish_task_switch <-schedule |
| | 1248 | bash-4043 [01] 115.281645: hrtick_set <-schedule |
| | 1249 | bash-4043 [01] 115.281645: hrtick_clear <-hrtick_set |
| | 1250 | bash-4043 [01] 115.281646: wait_for_completion <-__stop_machine_run |
| | 1251 | bash-4043 [01] 115.281647: wait_for_common <-wait_for_completion |
| | 1252 | bash-4043 [01] 115.281647: kthread_stop <-stop_machine_run |
| | 1253 | bash-4043 [01] 115.281648: init_waitqueue_head <-kthread_stop |
| | 1254 | bash-4043 [01] 115.281648: wake_up_process <-kthread_stop |
| | 1255 | bash-4043 [01] 115.281649: try_to_wake_up <-wake_up_process |
| | 1256 | |
| | 1257 | We can see that there's no more lock or preempt tracing. |
| | 1258 | |
| | 1259 | trace_pipe |
| | 1260 | ---------- |
| | 1261 | |
| | 1262 | The trace_pipe outputs the same content as the trace file, but the effect |
| | 1263 | on the tracing is different. Every read from trace_pipe is consumed. |
| | 1264 | This means that subsequent reads will be different. The trace |
| | 1265 | is live. |
| | 1266 | |
| | 1267 | # echo function > /debug/tracing/current_tracer |
| | 1268 | # cat /debug/tracing/trace_pipe > /tmp/trace.out & |
| | 1269 | [1] 4153 |
| | 1270 | # echo 1 > /debug/tracing/tracing_enabled |
| | 1271 | # usleep 1 |
| | 1272 | # echo 0 > /debug/tracing/tracing_enabled |
| | 1273 | # cat /debug/tracing/trace |
| | 1274 | # tracer: function |
| | 1275 | # |
| | 1276 | # TASK-PID CPU# TIMESTAMP FUNCTION |
| | 1277 | # | | | | | |
| | 1278 | |
| | 1279 | # |
| | 1280 | # cat /tmp/trace.out |
| | 1281 | bash-4043 [00] 41.267106: finish_task_switch <-schedule |
| | 1282 | bash-4043 [00] 41.267106: hrtick_set <-schedule |
| | 1283 | bash-4043 [00] 41.267107: hrtick_clear <-hrtick_set |
| | 1284 | bash-4043 [00] 41.267108: wait_for_completion <-__stop_machine_run |
| | 1285 | bash-4043 [00] 41.267108: wait_for_common <-wait_for_completion |
| | 1286 | bash-4043 [00] 41.267109: kthread_stop <-stop_machine_run |
| | 1287 | bash-4043 [00] 41.267109: init_waitqueue_head <-kthread_stop |
| | 1288 | bash-4043 [00] 41.267110: wake_up_process <-kthread_stop |
| | 1289 | bash-4043 [00] 41.267110: try_to_wake_up <-wake_up_process |
| | 1290 | bash-4043 [00] 41.267111: select_task_rq_rt <-try_to_wake_up |
| | 1291 | |
| | 1292 | |
| | 1293 | Note, reading the trace_pipe file will block until more input is added. |
| | 1294 | By changing the tracer, trace_pipe will issue an EOF. We needed |
| | 1295 | to set the function tracer _before_ we "cat" the trace_pipe file. |
| | 1296 | |
| | 1297 | |
| | 1298 | trace entries |
| | 1299 | ------------- |
| | 1300 | |
| | 1301 | Having too much or not enough data can be troublesome in diagnosing |
| | 1302 | an issue in the kernel. The file buffer_size_kb is used to modify |
| | 1303 | the size of the internal trace buffers. The number listed |
| | 1304 | is the number of entries that can be recorded per CPU. To know |
| | 1305 | the full size, multiply the number of possible CPUS with the |
| | 1306 | number of entries. |
| | 1307 | |
| | 1308 | # cat /debug/tracing/buffer_size_kb |
| | 1309 | 65620 |
| | 1310 | |
| | 1311 | Note, to modify this, you must have tracing completely disabled. To do that, |
| | 1312 | echo "nop" into the current_tracer. If the current_tracer is not set |
| | 1313 | to "nop", an EINVAL error will be returned. |
| | 1314 | |
| | 1315 | # echo nop > /debug/tracing/current_tracer |
| | 1316 | # echo 100000 > /debug/tracing/buffer_size_kb |
| | 1317 | # cat /debug/tracing/buffer_size_kb |
| | 1318 | 100045 |
| | 1319 | |
| | 1320 | |
| | 1321 | Notice that we echoed in 100,000 but the size is 100,045. The entries |
| | 1322 | are held in individual pages. It allocates the number of pages it takes |
| | 1323 | to fulfill the request. If more entries may fit on the last page |
| | 1324 | then they will be added. |
| | 1325 | |
| | 1326 | # echo 1 > /debug/tracing/buffer_size_kb |
| | 1327 | # cat /debug/tracing/buffer_size_kb |
| | 1328 | 85 |
| | 1329 | |
| | 1330 | This shows us that 85 entries can fit in a single page. |
| | 1331 | |
| | 1332 | The number of pages which will be allocated is limited to a percentage |
| | 1333 | of available memory. Allocating too much will produce an error. |
| | 1334 | |
| | 1335 | # echo 1000000000000 > /debug/tracing/buffer_size_kb |
| | 1336 | -bash: echo: write error: Cannot allocate memory |
| | 1337 | # cat /debug/tracing/buffer_size_kb |
| | 1338 | 85 |
| | 1339 | |
projects / deliverable / linux.git / blame_incremental