projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
| Commit | Line | Data |
|---|---|---|
| eb6d42ea SR |
1 | ftrace - Function Tracer |
| 2 | ======================== | |
| 3 | ||
| 4 | Copyright 2008 Red Hat Inc. | |
| a41eebab SR |
5 | Author: Steven Rostedt <srostedt@redhat.com> |
| 6 | License: The GNU Free Documentation License, Version 1.2 | |
| 7 | Reviewers: Elias Oltmanns and Randy Dunlap | |
| eb6d42ea | 8 | |
| a41eebab | 9 | Writen for: 2.6.26-rc8 linux-2.6-tip.git tip/tracing/ftrace branch |
| eb6d42ea SR |
10 | |
| 11 | Introduction | |
| 12 | ------------ | |
| 13 | ||
| 14 | Ftrace is an internal tracer designed to help out developers and | |
| 15 | designers of systems to find what is going on inside the kernel. | |
| 16 | It can be used for debugging or analyzing latencies and performance | |
| 17 | issues that take place outside of user-space. | |
| 18 | ||
| 19 | Although ftrace is the function tracer, it also includes an | |
| 20 | infrastructure that allows for other types of tracing. Some of the | |
| 21 | tracers that are currently in ftrace is a tracer to trace | |
| 22 | context switches, the time it takes for a high priority task to | |
| 23 | run after it was woken up, the time interrupts are disabled, and | |
| 24 | more. | |
| 25 | ||
| 26 | ||
| 27 | The File System | |
| 28 | --------------- | |
| 29 | ||
| 30 | Ftrace uses the debugfs file system to hold the control files as well | |
| 31 | as the files to display output. | |
| 32 | ||
| 33 | To mount the debugfs system: | |
| 34 | ||
| 35 | # mkdir /debug | |
| 36 | # mount -t debugfs nodev /debug | |
| 37 | ||
| 38 | ||
| 39 | That's it! (assuming that you have ftrace configured into your kernel) | |
| 40 | ||
| 41 | After mounting the debugfs, you can see a directory called | |
| 42 | "tracing". This directory contains the control and output files | |
| 43 | of ftrace. Here is a list of some of the key files: | |
| 44 | ||
| 45 | ||
| 46 | Note: all time values are in microseconds. | |
| 47 | ||
| 48 | current_tracer : This is used to set or display the current tracer | |
| 49 | that is configured. | |
| 50 | ||
| 51 | available_tracers : This holds the different types of tracers that | |
| a41eebab | 52 | have been compiled into the kernel. The tracers |
| eb6d42ea SR |
53 | listed here can be configured by echoing in their |
| 54 | name into current_tracer. | |
| 55 | ||
| 56 | tracing_enabled : This sets or displays whether the current_tracer | |
| 57 | is activated and tracing or not. Echo 0 into this | |
| 58 | file to disable the tracer or 1 (or non-zero) to | |
| 59 | enable it. | |
| 60 | ||
| 61 | trace : This file holds the output of the trace in a human readable | |
| 62 | format. | |
| 63 | ||
| 64 | latency_trace : This file shows the same trace but the information | |
| 65 | is organized more to display possible latencies | |
| 66 | in the system. | |
| 67 | ||
| 68 | trace_pipe : The output is the same as the "trace" file but this | |
| 69 | file is meant to be streamed with live tracing. | |
| 70 | Reads from this file will block until new data | |
| 71 | is retrieved. Unlike the "trace" and "latency_trace" | |
| 72 | files, this file is a consumer. This means reading | |
| 73 | from this file causes sequential reads to display | |
| 74 | more current data. Once data is read from this | |
| 75 | file, it is consumed, and will not be read | |
| 76 | again with a sequential read. The "trace" and | |
| 77 | "latency_trace" files are static, and if the | |
| 78 | tracer isn't adding more data, they will display | |
| 79 | the same information every time they are read. | |
| 80 | ||
| 81 | iter_ctrl : This file lets the user control the amount of data | |
| 82 | that is displayed in one of the above output | |
| 83 | files. | |
| 84 | ||
| 85 | trace_max_latency : Some of the tracers record the max latency. | |
| 86 | For example, the time interrupts are disabled. | |
| 87 | This time is saved in this file. The max trace | |
| 88 | will also be stored, and displayed by either | |
| 89 | "trace" or "latency_trace". A new max trace will | |
| 90 | only be recorded if the latency is greater than | |
| 91 | the value in this file. (in microseconds) | |
| 92 | ||
| 93 | trace_entries : This sets or displays the number of trace | |
| 94 | entries each CPU buffer can hold. The tracer buffers | |
| 95 | are the same size for each CPU, so care must be | |
| a41eebab SR |
96 | taken when modifying the trace_entries. The trace |
| 97 | buffers are allocated in pages (blocks of memory that | |
| 98 | the kernel uses for allocation, usually 4 KB in size). | |
| 99 | Since each entry is smaller than a page, if the last | |
| 100 | allocated page has room for more entries than were | |
| 101 | requested, the rest of the page is used to allocate | |
| 102 | entries. | |
| eb6d42ea SR |
103 | |
| 104 | This can only be updated when the current_tracer | |
| 105 | is set to "none". | |
| 106 | ||
| 107 | NOTE: It is planned on changing the allocated buffers | |
| 108 | from being the number of possible CPUS to | |
| 109 | the number of online CPUS. | |
| 110 | ||
| 111 | tracing_cpumask : This is a mask that lets the user only trace | |
| 112 | on specified CPUS. The format is a hex string | |
| 113 | representing the CPUS. | |
| 114 | ||
| 115 | set_ftrace_filter : When dynamic ftrace is configured in, the | |
| 116 | code is dynamically modified to disable calling | |
| 117 | of the function profiler (mcount). This lets | |
| 118 | tracing be configured in with practically no overhead | |
| 119 | in performance. This also has a side effect of | |
| 120 | enabling or disabling specific functions to be | |
| 121 | traced. Echoing in names of functions into this | |
| a41eebab | 122 | file will limit the trace to only these functions. |
| eb6d42ea SR |
123 | |
| 124 | set_ftrace_notrace: This has the opposite effect that | |
| 125 | set_ftrace_filter has. Any function that is added | |
| 126 | here will not be traced. If a function exists | |
| a41eebab SR |
127 | in both set_ftrace_filter and set_ftrace_notrace, |
| 128 | the function will _not_ be traced. | |
| eb6d42ea SR |
129 | |
| 130 | available_filter_functions : When a function is encountered the first | |
| 131 | time by the dynamic tracer, it is recorded and | |
| 132 | later the call is converted into a nop. This file | |
| 133 | lists the functions that have been recorded | |
| 134 | by the dynamic tracer and these functions can | |
| 135 | be used to set the ftrace filter by the above | |
| 136 | "set_ftrace_filter" file. | |
| 137 | ||
| 138 | ||
| 139 | The Tracers | |
| 140 | ----------- | |
| 141 | ||
| 142 | Here are the list of current tracers that can be configured. | |
| 143 | ||
| 144 | ftrace - function tracer that uses mcount to trace all functions. | |
| 145 | It is possible to filter out which functions that are | |
| a41eebab | 146 | to be traced when dynamic ftrace is configured in. |
| eb6d42ea SR |
147 | |
| 148 | sched_switch - traces the context switches between tasks. | |
| 149 | ||
| 150 | irqsoff - traces the areas that disable interrupts and saves off | |
| 151 | the trace with the longest max latency. | |
| 152 | See tracing_max_latency. When a new max is recorded, | |
| 153 | it replaces the old trace. It is best to view this | |
| 154 | trace with the latency_trace file. | |
| 155 | ||
| 156 | preemptoff - Similar to irqsoff but traces and records the time | |
| 157 | preemption is disabled. | |
| 158 | ||
| 159 | preemptirqsoff - Similar to irqsoff and preemptoff, but traces and | |
| 160 | records the largest time irqs and/or preemption is | |
| 161 | disabled. | |
| 162 | ||
| 163 | wakeup - Traces and records the max latency that it takes for | |
| 164 | the highest priority task to get scheduled after | |
| 165 | it has been woken up. | |
| 166 | ||
| 167 | none - This is not a tracer. To remove all tracers from tracing | |
| 168 | simply echo "none" into current_tracer. | |
| 169 | ||
| 170 | ||
| 171 | Examples of using the tracer | |
| 172 | ---------------------------- | |
| 173 | ||
| 174 | Here are typical examples of using the tracers with only controlling | |
| 175 | them with the debugfs interface (without using any user-land utilities). | |
| 176 | ||
| 177 | Output format: | |
| 178 | -------------- | |
| 179 | ||
| 180 | Here's an example of the output format of the file "trace" | |
| 181 | ||
| 182 | -------- | |
| 183 | # tracer: ftrace | |
| 184 | # | |
| 185 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
| 186 | # | | | | | | |
| 187 | bash-4251 [01] 10152.583854: path_put <-path_walk | |
| 188 | bash-4251 [01] 10152.583855: dput <-path_put | |
| 189 | bash-4251 [01] 10152.583855: _atomic_dec_and_lock <-dput | |
| 190 | -------- | |
| 191 | ||
| 192 | A header is printed with the trace that is represented. In this case | |
| 193 | the tracer is "ftrace". Then a header showing the format. Task name | |
| 194 | "bash", the task PID "4251", the CPU that it was running on | |
| 195 | "01", the timestamp in <secs>.<usecs> format, the function name that was | |
| 196 | traced "path_put" and the parent function that called this function | |
| 197 | "path_walk". | |
| 198 | ||
| 199 | The sched_switch tracer also includes tracing of task wake ups and | |
| 200 | context switches. | |
| 201 | ||
| 202 | ksoftirqd/1-7 [01] 1453.070013: 7:115:R + 2916:115:S | |
| 203 | ksoftirqd/1-7 [01] 1453.070013: 7:115:R + 10:115:S | |
| 204 | ksoftirqd/1-7 [01] 1453.070013: 7:115:R ==> 10:115:R | |
| 205 | events/1-10 [01] 1453.070013: 10:115:S ==> 2916:115:R | |
| 206 | kondemand/1-2916 [01] 1453.070013: 2916:115:S ==> 7:115:R | |
| 207 | ksoftirqd/1-7 [01] 1453.070013: 7:115:S ==> 0:140:R | |
| 208 | ||
| 209 | Wake ups are represented by a "+" and the context switches show | |
| 210 | "==>". The format is: | |
| 211 | ||
| 212 | Context switches: | |
| 213 | ||
| 214 | Previous task Next Task | |
| 215 | ||
| 216 | <pid>:<prio>:<state> ==> <pid>:<prio>:<state> | |
| 217 | ||
| 218 | Wake ups: | |
| 219 | ||
| 220 | Current task Task waking up | |
| 221 | ||
| 222 | <pid>:<prio>:<state> + <pid>:<prio>:<state> | |
| 223 | ||
| 224 | The prio is the internal kernel priority, which is inverse to the | |
| 225 | priority that is usually displayed by user-space tools. Zero represents | |
| 226 | the highest priority (99). Prio 100 starts the "nice" priorities with | |
| 227 | 100 being equal to nice -20 and 139 being nice 19. The prio "140" is | |
| 228 | reserved for the idle task which is the lowest priority thread (pid 0). | |
| 229 | ||
| 230 | ||
| 231 | Latency trace format | |
| 232 | -------------------- | |
| 233 | ||
| 234 | For traces that display latency times, the latency_trace file gives | |
| 235 | a bit more information to see why a latency happened. Here's a typical | |
| 236 | trace. | |
| 237 | ||
| 238 | # tracer: irqsoff | |
| 239 | # | |
| 240 | irqsoff latency trace v1.1.5 on 2.6.26-rc8 | |
| 241 | -------------------------------------------------------------------- | |
| 242 | latency: 97 us, #3/3, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
| 243 | ----------------- | |
| 244 | | task: swapper-0 (uid:0 nice:0 policy:0 rt_prio:0) | |
| 245 | ----------------- | |
| 246 | => started at: apic_timer_interrupt | |
| 247 | => ended at: do_softirq | |
| 248 | ||
| 249 | # _------=> CPU# | |
| 250 | # / _-----=> irqs-off | |
| 251 | # | / _----=> need-resched | |
| 252 | # || / _---=> hardirq/softirq | |
| 253 | # ||| / _--=> preempt-depth | |
| 254 | # |||| / | |
| 255 | # ||||| delay | |
| 256 | # cmd pid ||||| time | caller | |
| 257 | # \ / ||||| \ | / | |
| 258 | <idle>-0 0d..1 0us+: trace_hardirqs_off_thunk (apic_timer_interrupt) | |
| 259 | <idle>-0 0d.s. 97us : __do_softirq (do_softirq) | |
| 260 | <idle>-0 0d.s1 98us : trace_hardirqs_on (do_softirq) | |
| 261 | ||
| 262 | ||
| 263 | vim:ft=help | |
| 264 | ||
| 265 | ||
| 266 | This shows that the current tracer is "irqsoff" tracing the time | |
| 267 | interrupts are disabled. It gives the trace version and the kernel | |
| 268 | this was executed on (2.6.26-rc8). Then it displays the max latency | |
| 269 | in microsecs (97 us). The number of trace entries displayed | |
| 270 | by the total number recorded (both are three: #3/3). The type of | |
| 271 | preemption that was used (PREEMPT). VP, KP, SP, and HP are always zero | |
| 272 | and reserved for later use. #P is the number of online CPUS (#P:2). | |
| 273 | ||
| 274 | The task is the process that was running when the latency happened. | |
| 275 | (swapper pid: 0). | |
| 276 | ||
| 277 | The start and stop that caused the latencies: | |
| 278 | ||
| 279 | apic_timer_interrupt is where the interrupts were disabled. | |
| 280 | do_softirq is where they were enabled again. | |
| 281 | ||
| 282 | The next lines after the header are the trace itself. The header | |
| 283 | explains which is which. | |
| 284 | ||
| 285 | cmd: The name of the process in the trace. | |
| 286 | ||
| 287 | pid: The PID of that process. | |
| 288 | ||
| 289 | CPU#: The CPU that the process was running on. | |
| 290 | ||
| 291 | irqs-off: 'd' interrupts are disabled. '.' otherwise. | |
| 292 | ||
| 293 | need-resched: 'N' task need_resched is set, '.' otherwise. | |
| 294 | ||
| 295 | hardirq/softirq: | |
| 296 | 'H' - hard irq happened inside a softirq. | |
| 297 | 'h' - hard irq is running | |
| 298 | 's' - soft irq is running | |
| 299 | '.' - normal context. | |
| 300 | ||
| 301 | preempt-depth: The level of preempt_disabled | |
| 302 | ||
| 303 | The above is mostly meaningful for kernel developers. | |
| 304 | ||
| a41eebab SR |
305 | time: This differs from the trace file output. The trace file output |
| 306 | included an absolute timestamp. The timestamp used by the | |
| 307 | latency_trace file is relative to the start of the trace. | |
| eb6d42ea SR |
308 | |
| 309 | delay: This is just to help catch your eye a bit better. And | |
| 310 | needs to be fixed to be only relative to the same CPU. | |
| a41eebab | 311 | The marks are determined by the difference between this |
| eb6d42ea SR |
312 | current trace and the next trace. |
| 313 | '!' - greater than preempt_mark_thresh (default 100) | |
| 314 | '+' - greater than 1 microsecond | |
| 315 | ' ' - less than or equal to 1 microsecond. | |
| 316 | ||
| 317 | The rest is the same as the 'trace' file. | |
| 318 | ||
| 319 | ||
| 320 | iter_ctrl | |
| 321 | --------- | |
| 322 | ||
| 323 | The iter_ctrl file is used to control what gets printed in the trace | |
| 324 | output. To see what is available, simply cat the file: | |
| 325 | ||
| 326 | cat /debug/tracing/iter_ctrl | |
| 327 | print-parent nosym-offset nosym-addr noverbose noraw nohex nobin \ | |
| 328 | noblock nostacktrace nosched-tree | |
| 329 | ||
| a41eebab | 330 | To disable one of the options, echo in the option prepended with "no". |
| eb6d42ea SR |
331 | |
| 332 | echo noprint-parent > /debug/tracing/iter_ctrl | |
| 333 | ||
| 334 | To enable an option, leave off the "no". | |
| 335 | ||
| a41eebab | 336 | echo sym-offset > /debug/tracing/iter_ctrl |
| eb6d42ea SR |
337 | |
| 338 | Here are the available options: | |
| 339 | ||
| 340 | print-parent - On function traces, display the calling function | |
| 341 | as well as the function being traced. | |
| 342 | ||
| 343 | print-parent: | |
| 344 | bash-4000 [01] 1477.606694: simple_strtoul <-strict_strtoul | |
| 345 | ||
| 346 | noprint-parent: | |
| 347 | bash-4000 [01] 1477.606694: simple_strtoul | |
| 348 | ||
| 349 | ||
| 350 | sym-offset - Display not only the function name, but also the offset | |
| 351 | in the function. For example, instead of seeing just | |
| a41eebab | 352 | "ktime_get", you will see "ktime_get+0xb/0x20". |
| eb6d42ea SR |
353 | |
| 354 | sym-offset: | |
| 355 | bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0 | |
| 356 | ||
| 357 | sym-addr - this will also display the function address as well as | |
| 358 | the function name. | |
| 359 | ||
| 360 | sym-addr: | |
| 361 | bash-4000 [01] 1477.606694: simple_strtoul <c0339346> | |
| 362 | ||
| 363 | verbose - This deals with the latency_trace file. | |
| 364 | ||
| 365 | bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \ | |
| 366 | (+0.000ms): simple_strtoul (strict_strtoul) | |
| 367 | ||
| 368 | raw - This will display raw numbers. This option is best for use with | |
| 369 | user applications that can translate the raw numbers better than | |
| 370 | having it done in the kernel. | |
| 371 | ||
| a41eebab | 372 | hex - Similar to raw, but the numbers will be in a hexadecimal format. |
| eb6d42ea SR |
373 | |
| 374 | bin - This will print out the formats in raw binary. | |
| 375 | ||
| 376 | block - TBD (needs update) | |
| 377 | ||
| 378 | stacktrace - This is one of the options that changes the trace itself. | |
| 379 | When a trace is recorded, so is the stack of functions. | |
| 380 | This allows for back traces of trace sites. | |
| 381 | ||
| 382 | sched-tree - TBD (any users??) | |
| 383 | ||
| 384 | ||
| 385 | sched_switch | |
| 386 | ------------ | |
| 387 | ||
| 388 | This tracer simply records schedule switches. Here's an example | |
| a41eebab | 389 | of how to use it. |
| eb6d42ea SR |
390 | |
| 391 | # echo sched_switch > /debug/tracing/current_tracer | |
| 392 | # echo 1 > /debug/tracing/tracing_enabled | |
| 393 | # sleep 1 | |
| 394 | # echo 0 > /debug/tracing/tracing_enabled | |
| 395 | # cat /debug/tracing/trace | |
| 396 | ||
| 397 | # tracer: sched_switch | |
| 398 | # | |
| 399 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
| 400 | # | | | | | | |
| 401 | bash-3997 [01] 240.132281: 3997:120:R + 4055:120:R | |
| 402 | bash-3997 [01] 240.132284: 3997:120:R ==> 4055:120:R | |
| 403 | sleep-4055 [01] 240.132371: 4055:120:S ==> 3997:120:R | |
| 404 | bash-3997 [01] 240.132454: 3997:120:R + 4055:120:S | |
| 405 | bash-3997 [01] 240.132457: 3997:120:R ==> 4055:120:R | |
| 406 | sleep-4055 [01] 240.132460: 4055:120:D ==> 3997:120:R | |
| 407 | bash-3997 [01] 240.132463: 3997:120:R + 4055:120:D | |
| 408 | bash-3997 [01] 240.132465: 3997:120:R ==> 4055:120:R | |
| 409 | <idle>-0 [00] 240.132589: 0:140:R + 4:115:S | |
| 410 | <idle>-0 [00] 240.132591: 0:140:R ==> 4:115:R | |
| 411 | ksoftirqd/0-4 [00] 240.132595: 4:115:S ==> 0:140:R | |
| 412 | <idle>-0 [00] 240.132598: 0:140:R + 4:115:S | |
| 413 | <idle>-0 [00] 240.132599: 0:140:R ==> 4:115:R | |
| 414 | ksoftirqd/0-4 [00] 240.132603: 4:115:S ==> 0:140:R | |
| 415 | sleep-4055 [01] 240.133058: 4055:120:S ==> 3997:120:R | |
| 416 | [...] | |
| 417 | ||
| 418 | ||
| 419 | As we have discussed previously about this format, the header shows | |
| 420 | the name of the trace and points to the options. The "FUNCTION" | |
| 421 | is a misnomer since here it represents the wake ups and context | |
| 422 | switches. | |
| 423 | ||
| 424 | The sched_switch only lists the wake ups (represented with '+') | |
| 425 | and context switches ('==>') with the previous task or current | |
| 426 | first followed by the next task or task waking up. The format for both | |
| 427 | of these is PID:KERNEL-PRIO:TASK-STATE. Remember that the KERNEL-PRIO | |
| 428 | is the inverse of the actual priority with zero (0) being the highest | |
| 429 | priority and the nice values starting at 100 (nice -20). Below is | |
| 430 | a quick chart to map the kernel priority to user land priorities. | |
| 431 | ||
| 432 | Kernel priority: 0 to 99 ==> user RT priority 99 to 0 | |
| 433 | Kernel priority: 100 to 139 ==> user nice -20 to 19 | |
| 434 | Kernel priority: 140 ==> idle task priority | |
| 435 | ||
| 436 | The task states are: | |
| 437 | ||
| 438 | R - running : wants to run, may not actually be running | |
| 439 | S - sleep : process is waiting to be woken up (handles signals) | |
| 440 | D - deep sleep : process must be woken up (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 give different output depending on whether | |
| 451 | 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 | ||
| a41eebab | 478 | The irqsoff tracer tracks the time interrupts are disabled to the time |
| eb6d42ea SR |
479 | they are re-enabled. When a new maximum latency is hit, it saves off |
| 480 | the trace so that it may be retrieved at a later time. Every time a | |
| 481 | new maximum in reached, the old saved trace is discarded and the new | |
| 482 | trace is saved. | |
| 483 | ||
| 484 | To reset the maximum, echo 0 into tracing_max_latency. Here's an | |
| 485 | example: | |
| 486 | ||
| 487 | # echo irqsoff > /debug/tracing/current_tracer | |
| 488 | # echo 0 > /debug/tracing/tracing_max_latency | |
| 489 | # echo 1 > /debug/tracing/tracing_enabled | |
| 490 | # ls -ltr | |
| 491 | [...] | |
| 492 | # echo 0 > /debug/tracing/tracing_enabled | |
| 493 | # cat /debug/tracing/latency_trace | |
| 494 | # tracer: irqsoff | |
| 495 | # | |
| 496 | irqsoff latency trace v1.1.5 on 2.6.26-rc8 | |
| 497 | -------------------------------------------------------------------- | |
| 498 | latency: 6 us, #3/3, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
| 499 | ----------------- | |
| 500 | | task: bash-4269 (uid:0 nice:0 policy:0 rt_prio:0) | |
| 501 | ----------------- | |
| 502 | => started at: copy_page_range | |
| 503 | => ended at: copy_page_range | |
| 504 | ||
| 505 | # _------=> CPU# | |
| 506 | # / _-----=> irqs-off | |
| 507 | # | / _----=> need-resched | |
| 508 | # || / _---=> hardirq/softirq | |
| 509 | # ||| / _--=> preempt-depth | |
| 510 | # |||| / | |
| 511 | # ||||| delay | |
| 512 | # cmd pid ||||| time | caller | |
| 513 | # \ / ||||| \ | / | |
| 514 | bash-4269 1...1 0us+: _spin_lock (copy_page_range) | |
| 515 | bash-4269 1...1 7us : _spin_unlock (copy_page_range) | |
| 516 | bash-4269 1...2 7us : trace_preempt_on (copy_page_range) | |
| 517 | ||
| 518 | ||
| 519 | vim:ft=help | |
| 520 | ||
| 521 | Here we see that that we had a latency of 6 microsecs (which is | |
| 522 | very good). The spin_lock in copy_page_range disabled interrupts. | |
| 523 | The difference between the 6 and the displayed timestamp 7us is | |
| 524 | because the clock must have incremented between the time of recording | |
| 525 | the max latency and recording the function that had that latency. | |
| 526 | ||
| a41eebab | 527 | Note the above had ftrace_enabled not set. If we set the ftrace_enabled, |
| eb6d42ea SR |
528 | we get a much larger output: |
| 529 | ||
| 530 | # tracer: irqsoff | |
| 531 | # | |
| 532 | irqsoff latency trace v1.1.5 on 2.6.26-rc8 | |
| 533 | -------------------------------------------------------------------- | |
| 534 | latency: 50 us, #101/101, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
| 535 | ----------------- | |
| 536 | | task: ls-4339 (uid:0 nice:0 policy:0 rt_prio:0) | |
| 537 | ----------------- | |
| 538 | => started at: __alloc_pages_internal | |
| 539 | => ended at: __alloc_pages_internal | |
| 540 | ||
| 541 | # _------=> CPU# | |
| 542 | # / _-----=> irqs-off | |
| 543 | # | / _----=> need-resched | |
| 544 | # || / _---=> hardirq/softirq | |
| 545 | # ||| / _--=> preempt-depth | |
| 546 | # |||| / | |
| 547 | # ||||| delay | |
| 548 | # cmd pid ||||| time | caller | |
| 549 | # \ / ||||| \ | / | |
| 550 | ls-4339 0...1 0us+: get_page_from_freelist (__alloc_pages_internal) | |
| 551 | ls-4339 0d..1 3us : rmqueue_bulk (get_page_from_freelist) | |
| 552 | ls-4339 0d..1 3us : _spin_lock (rmqueue_bulk) | |
| 553 | ls-4339 0d..1 4us : add_preempt_count (_spin_lock) | |
| 554 | ls-4339 0d..2 4us : __rmqueue (rmqueue_bulk) | |
| 555 | ls-4339 0d..2 5us : __rmqueue_smallest (__rmqueue) | |
| 556 | ls-4339 0d..2 5us : __mod_zone_page_state (__rmqueue_smallest) | |
| 557 | ls-4339 0d..2 6us : __rmqueue (rmqueue_bulk) | |
| 558 | ls-4339 0d..2 6us : __rmqueue_smallest (__rmqueue) | |
| 559 | ls-4339 0d..2 7us : __mod_zone_page_state (__rmqueue_smallest) | |
| 560 | ls-4339 0d..2 7us : __rmqueue (rmqueue_bulk) | |
| 561 | ls-4339 0d..2 8us : __rmqueue_smallest (__rmqueue) | |
| 562 | [...] | |
| 563 | ls-4339 0d..2 46us : __rmqueue_smallest (__rmqueue) | |
| 564 | ls-4339 0d..2 47us : __mod_zone_page_state (__rmqueue_smallest) | |
| 565 | ls-4339 0d..2 47us : __rmqueue (rmqueue_bulk) | |
| 566 | ls-4339 0d..2 48us : __rmqueue_smallest (__rmqueue) | |
| 567 | ls-4339 0d..2 48us : __mod_zone_page_state (__rmqueue_smallest) | |
| 568 | ls-4339 0d..2 49us : _spin_unlock (rmqueue_bulk) | |
| 569 | ls-4339 0d..2 49us : sub_preempt_count (_spin_unlock) | |
| 570 | ls-4339 0d..1 50us : get_page_from_freelist (__alloc_pages_internal) | |
| 571 | ls-4339 0d..2 51us : trace_hardirqs_on (__alloc_pages_internal) | |
| 572 | ||
| 573 | ||
| 574 | vim:ft=help | |
| 575 | ||
| 576 | ||
| 577 | Here we traced a 50 microsecond latency. But we also see all the | |
| a41eebab SR |
578 | functions that were called during that time. Note that by enabling |
| 579 | function tracing, we endure an added overhead. This overhead may | |
| 580 | extend the latency times. But nevertheless, this trace has provided | |
| 581 | some very helpful debugging information. | |
| eb6d42ea SR |
582 | |
| 583 | ||
| 584 | preemptoff | |
| 585 | ---------- | |
| 586 | ||
| a41eebab SR |
587 | When preemption is disabled, we may be able to receive interrupts but |
| 588 | the task cannot be preempted and a higher priority task must wait | |
| eb6d42ea SR |
589 | for preemption to be enabled again before it can preempt a lower |
| 590 | priority task. | |
| 591 | ||
| a41eebab | 592 | The preemptoff tracer traces the places that disable preemption. |
| eb6d42ea SR |
593 | Like the irqsoff, it records the maximum latency that preemption |
| 594 | was disabled. The control of preemptoff is much like the irqsoff. | |
| 595 | ||
| 596 | # echo preemptoff > /debug/tracing/current_tracer | |
| 597 | # echo 0 > /debug/tracing/tracing_max_latency | |
| 598 | # echo 1 > /debug/tracing/tracing_enabled | |
| 599 | # ls -ltr | |
| 600 | [...] | |
| 601 | # echo 0 > /debug/tracing/tracing_enabled | |
| 602 | # cat /debug/tracing/latency_trace | |
| 603 | # tracer: preemptoff | |
| 604 | # | |
| 605 | preemptoff latency trace v1.1.5 on 2.6.26-rc8 | |
| 606 | -------------------------------------------------------------------- | |
| 607 | latency: 29 us, #3/3, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
| 608 | ----------------- | |
| 609 | | task: sshd-4261 (uid:0 nice:0 policy:0 rt_prio:0) | |
| 610 | ----------------- | |
| 611 | => started at: do_IRQ | |
| 612 | => ended at: __do_softirq | |
| 613 | ||
| 614 | # _------=> CPU# | |
| 615 | # / _-----=> irqs-off | |
| 616 | # | / _----=> need-resched | |
| 617 | # || / _---=> hardirq/softirq | |
| 618 | # ||| / _--=> preempt-depth | |
| 619 | # |||| / | |
| 620 | # ||||| delay | |
| 621 | # cmd pid ||||| time | caller | |
| 622 | # \ / ||||| \ | / | |
| 623 | sshd-4261 0d.h. 0us+: irq_enter (do_IRQ) | |
| 624 | sshd-4261 0d.s. 29us : _local_bh_enable (__do_softirq) | |
| 625 | sshd-4261 0d.s1 30us : trace_preempt_on (__do_softirq) | |
| 626 | ||
| 627 | ||
| 628 | vim:ft=help | |
| 629 | ||
| 630 | This has some more changes. Preemption was disabled when an interrupt | |
| 631 | came in (notice the 'h'), and was enabled while doing a softirq. | |
| 632 | (notice the 's'). But we also see that interrupts have been disabled | |
| 633 | when entering the preempt off section and leaving it (the 'd'). | |
| 634 | We do not know if interrupts were enabled in the mean time. | |
| 635 | ||
| 636 | # tracer: preemptoff | |
| 637 | # | |
| 638 | preemptoff latency trace v1.1.5 on 2.6.26-rc8 | |
| 639 | -------------------------------------------------------------------- | |
| 640 | latency: 63 us, #87/87, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
| 641 | ----------------- | |
| 642 | | task: sshd-4261 (uid:0 nice:0 policy:0 rt_prio:0) | |
| 643 | ----------------- | |
| 644 | => started at: remove_wait_queue | |
| 645 | => ended at: __do_softirq | |
| 646 | ||
| 647 | # _------=> CPU# | |
| 648 | # / _-----=> irqs-off | |
| 649 | # | / _----=> need-resched | |
| 650 | # || / _---=> hardirq/softirq | |
| 651 | # ||| / _--=> preempt-depth | |
| 652 | # |||| / | |
| 653 | # ||||| delay | |
| 654 | # cmd pid ||||| time | caller | |
| 655 | # \ / ||||| \ | / | |
| 656 | sshd-4261 0d..1 0us : _spin_lock_irqsave (remove_wait_queue) | |
| 657 | sshd-4261 0d..1 1us : _spin_unlock_irqrestore (remove_wait_queue) | |
| 658 | sshd-4261 0d..1 2us : do_IRQ (common_interrupt) | |
| 659 | sshd-4261 0d..1 2us : irq_enter (do_IRQ) | |
| 660 | sshd-4261 0d..1 2us : idle_cpu (irq_enter) | |
| 661 | sshd-4261 0d..1 3us : add_preempt_count (irq_enter) | |
| 662 | sshd-4261 0d.h1 3us : idle_cpu (irq_enter) | |
| 663 | sshd-4261 0d.h. 4us : handle_fasteoi_irq (do_IRQ) | |
| 664 | [...] | |
| 665 | sshd-4261 0d.h. 12us : add_preempt_count (_spin_lock) | |
| 666 | sshd-4261 0d.h1 12us : ack_ioapic_quirk_irq (handle_fasteoi_irq) | |
| 667 | sshd-4261 0d.h1 13us : move_native_irq (ack_ioapic_quirk_irq) | |
| 668 | sshd-4261 0d.h1 13us : _spin_unlock (handle_fasteoi_irq) | |
| 669 | sshd-4261 0d.h1 14us : sub_preempt_count (_spin_unlock) | |
| 670 | sshd-4261 0d.h1 14us : irq_exit (do_IRQ) | |
| 671 | sshd-4261 0d.h1 15us : sub_preempt_count (irq_exit) | |
| 672 | sshd-4261 0d..2 15us : do_softirq (irq_exit) | |
| 673 | sshd-4261 0d... 15us : __do_softirq (do_softirq) | |
| 674 | sshd-4261 0d... 16us : __local_bh_disable (__do_softirq) | |
| 675 | sshd-4261 0d... 16us+: add_preempt_count (__local_bh_disable) | |
| 676 | sshd-4261 0d.s4 20us : add_preempt_count (__local_bh_disable) | |
| 677 | sshd-4261 0d.s4 21us : sub_preempt_count (local_bh_enable) | |
| 678 | sshd-4261 0d.s5 21us : sub_preempt_count (local_bh_enable) | |
| 679 | [...] | |
| 680 | sshd-4261 0d.s6 41us : add_preempt_count (__local_bh_disable) | |
| 681 | sshd-4261 0d.s6 42us : sub_preempt_count (local_bh_enable) | |
| 682 | sshd-4261 0d.s7 42us : sub_preempt_count (local_bh_enable) | |
| 683 | sshd-4261 0d.s5 43us : add_preempt_count (__local_bh_disable) | |
| 684 | sshd-4261 0d.s5 43us : sub_preempt_count (local_bh_enable_ip) | |
| 685 | sshd-4261 0d.s6 44us : sub_preempt_count (local_bh_enable_ip) | |
| 686 | sshd-4261 0d.s5 44us : add_preempt_count (__local_bh_disable) | |
| 687 | sshd-4261 0d.s5 45us : sub_preempt_count (local_bh_enable) | |
| 688 | [...] | |
| 689 | sshd-4261 0d.s. 63us : _local_bh_enable (__do_softirq) | |
| 690 | sshd-4261 0d.s1 64us : trace_preempt_on (__do_softirq) | |
| 691 | ||
| 692 | ||
| 693 | The above is an example of the preemptoff trace with ftrace_enabled | |
| 694 | set. Here we see that interrupts were disabled the entire time. | |
| 695 | The irq_enter code lets us know that we entered an interrupt 'h'. | |
| 696 | Before that, the functions being traced still show that it is not | |
| 697 | in an interrupt, but we can see by the functions themselves that | |
| 698 | this is not the case. | |
| 699 | ||
| 700 | Notice that the __do_softirq when called doesn't have a preempt_count. | |
| 701 | It may seem that we missed a preempt enabled. What really happened | |
| 702 | is that the preempt count is held on the threads stack and we | |
| 703 | switched to the softirq stack (4K stacks in effect). The code | |
| a41eebab | 704 | does not copy the preempt count, but because interrupts are disabled, |
| eb6d42ea SR |
705 | we don't need to worry about it. Having a tracer like this is good |
| 706 | to let people know what really happens inside the kernel. | |
| 707 | ||
| 708 | ||
| 709 | preemptirqsoff | |
| 710 | -------------- | |
| 711 | ||
| 712 | Knowing the locations that have interrupts disabled or preemption | |
| 713 | disabled for the longest times is helpful. But sometimes we would | |
| 714 | like to know when either preemption and/or interrupts are disabled. | |
| 715 | ||
| 716 | The following code: | |
| 717 | ||
| 718 | local_irq_disable(); | |
| 719 | call_function_with_irqs_off(); | |
| 720 | preempt_disable(); | |
| 721 | call_function_with_irqs_and_preemption_off(); | |
| 722 | local_irq_enable(); | |
| 723 | call_function_with_preemption_off(); | |
| 724 | preempt_enable(); | |
| 725 | ||
| 726 | The irqsoff tracer will record the total length of | |
| 727 | call_function_with_irqs_off() and | |
| 728 | call_function_with_irqs_and_preemption_off(). | |
| 729 | ||
| 730 | The preemptoff tracer will record the total length of | |
| 731 | call_function_with_irqs_and_preemption_off() and | |
| 732 | call_function_with_preemption_off(). | |
| 733 | ||
| 734 | But neither will trace the time that interrupts and/or preemption | |
| 735 | is disabled. This total time is the time that we can not schedule. | |
| 736 | To record this time, use the preemptirqsoff tracer. | |
| 737 | ||
| 738 | Again, using this trace is much like the irqsoff and preemptoff tracers. | |
| 739 | ||
| a41eebab | 740 | # echo preemptirqsoff > /debug/tracing/current_tracer |
| eb6d42ea SR |
741 | # echo 0 > /debug/tracing/tracing_max_latency |
| 742 | # echo 1 > /debug/tracing/tracing_enabled | |
| 743 | # ls -ltr | |
| 744 | [...] | |
| 745 | # echo 0 > /debug/tracing/tracing_enabled | |
| 746 | # cat /debug/tracing/latency_trace | |
| 747 | # tracer: preemptirqsoff | |
| 748 | # | |
| 749 | preemptirqsoff latency trace v1.1.5 on 2.6.26-rc8 | |
| 750 | -------------------------------------------------------------------- | |
| 751 | latency: 293 us, #3/3, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
| 752 | ----------------- | |
| 753 | | task: ls-4860 (uid:0 nice:0 policy:0 rt_prio:0) | |
| 754 | ----------------- | |
| 755 | => started at: apic_timer_interrupt | |
| 756 | => ended at: __do_softirq | |
| 757 | ||
| 758 | # _------=> CPU# | |
| 759 | # / _-----=> irqs-off | |
| 760 | # | / _----=> need-resched | |
| 761 | # || / _---=> hardirq/softirq | |
| 762 | # ||| / _--=> preempt-depth | |
| 763 | # |||| / | |
| 764 | # ||||| delay | |
| 765 | # cmd pid ||||| time | caller | |
| 766 | # \ / ||||| \ | / | |
| 767 | ls-4860 0d... 0us!: trace_hardirqs_off_thunk (apic_timer_interrupt) | |
| 768 | ls-4860 0d.s. 294us : _local_bh_enable (__do_softirq) | |
| 769 | ls-4860 0d.s1 294us : trace_preempt_on (__do_softirq) | |
| 770 | ||
| 771 | ||
| 772 | vim:ft=help | |
| 773 | ||
| 774 | ||
| 775 | The trace_hardirqs_off_thunk is called from assembly on x86 when | |
| 776 | interrupts are disabled in the assembly code. Without the function | |
| 777 | tracing, we don't know if interrupts were enabled within the preemption | |
| 778 | points. We do see that it started with preemption enabled. | |
| 779 | ||
| 780 | Here is a trace with ftrace_enabled set: | |
| 781 | ||
| 782 | ||
| 783 | # tracer: preemptirqsoff | |
| 784 | # | |
| 785 | preemptirqsoff latency trace v1.1.5 on 2.6.26-rc8 | |
| 786 | -------------------------------------------------------------------- | |
| 787 | latency: 105 us, #183/183, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
| 788 | ----------------- | |
| 789 | | task: sshd-4261 (uid:0 nice:0 policy:0 rt_prio:0) | |
| 790 | ----------------- | |
| 791 | => started at: write_chan | |
| 792 | => ended at: __do_softirq | |
| 793 | ||
| 794 | # _------=> CPU# | |
| 795 | # / _-----=> irqs-off | |
| 796 | # | / _----=> need-resched | |
| 797 | # || / _---=> hardirq/softirq | |
| 798 | # ||| / _--=> preempt-depth | |
| 799 | # |||| / | |
| 800 | # ||||| delay | |
| 801 | # cmd pid ||||| time | caller | |
| 802 | # \ / ||||| \ | / | |
| 803 | ls-4473 0.N.. 0us : preempt_schedule (write_chan) | |
| 804 | ls-4473 0dN.1 1us : _spin_lock (schedule) | |
| 805 | ls-4473 0dN.1 2us : add_preempt_count (_spin_lock) | |
| 806 | ls-4473 0d..2 2us : put_prev_task_fair (schedule) | |
| 807 | [...] | |
| 808 | ls-4473 0d..2 13us : set_normalized_timespec (ktime_get_ts) | |
| 809 | ls-4473 0d..2 13us : __switch_to (schedule) | |
| 810 | sshd-4261 0d..2 14us : finish_task_switch (schedule) | |
| 811 | sshd-4261 0d..2 14us : _spin_unlock_irq (finish_task_switch) | |
| 812 | sshd-4261 0d..1 15us : add_preempt_count (_spin_lock_irqsave) | |
| 813 | sshd-4261 0d..2 16us : _spin_unlock_irqrestore (hrtick_set) | |
| 814 | sshd-4261 0d..2 16us : do_IRQ (common_interrupt) | |
| 815 | sshd-4261 0d..2 17us : irq_enter (do_IRQ) | |
| 816 | sshd-4261 0d..2 17us : idle_cpu (irq_enter) | |
| 817 | sshd-4261 0d..2 18us : add_preempt_count (irq_enter) | |
| 818 | sshd-4261 0d.h2 18us : idle_cpu (irq_enter) | |
| 819 | sshd-4261 0d.h. 18us : handle_fasteoi_irq (do_IRQ) | |
| 820 | sshd-4261 0d.h. 19us : _spin_lock (handle_fasteoi_irq) | |
| 821 | sshd-4261 0d.h. 19us : add_preempt_count (_spin_lock) | |
| 822 | sshd-4261 0d.h1 20us : _spin_unlock (handle_fasteoi_irq) | |
| 823 | sshd-4261 0d.h1 20us : sub_preempt_count (_spin_unlock) | |
| 824 | [...] | |
| 825 | sshd-4261 0d.h1 28us : _spin_unlock (handle_fasteoi_irq) | |
| 826 | sshd-4261 0d.h1 29us : sub_preempt_count (_spin_unlock) | |
| 827 | sshd-4261 0d.h2 29us : irq_exit (do_IRQ) | |
| 828 | sshd-4261 0d.h2 29us : sub_preempt_count (irq_exit) | |
| 829 | sshd-4261 0d..3 30us : do_softirq (irq_exit) | |
| 830 | sshd-4261 0d... 30us : __do_softirq (do_softirq) | |
| 831 | sshd-4261 0d... 31us : __local_bh_disable (__do_softirq) | |
| 832 | sshd-4261 0d... 31us+: add_preempt_count (__local_bh_disable) | |
| 833 | sshd-4261 0d.s4 34us : add_preempt_count (__local_bh_disable) | |
| 834 | [...] | |
| 835 | sshd-4261 0d.s3 43us : sub_preempt_count (local_bh_enable_ip) | |
| 836 | sshd-4261 0d.s4 44us : sub_preempt_count (local_bh_enable_ip) | |
| 837 | sshd-4261 0d.s3 44us : smp_apic_timer_interrupt (apic_timer_interrupt) | |
| 838 | sshd-4261 0d.s3 45us : irq_enter (smp_apic_timer_interrupt) | |
| 839 | sshd-4261 0d.s3 45us : idle_cpu (irq_enter) | |
| 840 | sshd-4261 0d.s3 46us : add_preempt_count (irq_enter) | |
| 841 | sshd-4261 0d.H3 46us : idle_cpu (irq_enter) | |
| 842 | sshd-4261 0d.H3 47us : hrtimer_interrupt (smp_apic_timer_interrupt) | |
| 843 | sshd-4261 0d.H3 47us : ktime_get (hrtimer_interrupt) | |
| 844 | [...] | |
| 845 | sshd-4261 0d.H3 81us : tick_program_event (hrtimer_interrupt) | |
| 846 | sshd-4261 0d.H3 82us : ktime_get (tick_program_event) | |
| 847 | sshd-4261 0d.H3 82us : ktime_get_ts (ktime_get) | |
| 848 | sshd-4261 0d.H3 83us : getnstimeofday (ktime_get_ts) | |
| 849 | sshd-4261 0d.H3 83us : set_normalized_timespec (ktime_get_ts) | |
| 850 | sshd-4261 0d.H3 84us : clockevents_program_event (tick_program_event) | |
| 851 | sshd-4261 0d.H3 84us : lapic_next_event (clockevents_program_event) | |
| 852 | sshd-4261 0d.H3 85us : irq_exit (smp_apic_timer_interrupt) | |
| 853 | sshd-4261 0d.H3 85us : sub_preempt_count (irq_exit) | |
| 854 | sshd-4261 0d.s4 86us : sub_preempt_count (irq_exit) | |
| 855 | sshd-4261 0d.s3 86us : add_preempt_count (__local_bh_disable) | |
| 856 | [...] | |
| 857 | sshd-4261 0d.s1 98us : sub_preempt_count (net_rx_action) | |
| 858 | sshd-4261 0d.s. 99us : add_preempt_count (_spin_lock_irq) | |
| 859 | sshd-4261 0d.s1 99us+: _spin_unlock_irq (run_timer_softirq) | |
| 860 | sshd-4261 0d.s. 104us : _local_bh_enable (__do_softirq) | |
| 861 | sshd-4261 0d.s. 104us : sub_preempt_count (_local_bh_enable) | |
| 862 | sshd-4261 0d.s. 105us : _local_bh_enable (__do_softirq) | |
| 863 | sshd-4261 0d.s1 105us : trace_preempt_on (__do_softirq) | |
| 864 | ||
| 865 | ||
| 866 | This is a very interesting trace. It started with the preemption of | |
| 867 | the ls task. We see that the task had the "need_resched" bit set | |
| 868 | with the 'N' in the trace. Interrupts are disabled in the spin_lock | |
| 869 | and the trace started. We see that a schedule took place to run | |
| a41eebab SR |
870 | sshd. When the interrupts were enabled, we took an interrupt. |
| 871 | On return from the interrupt handler, the softirq ran. We took another | |
| 872 | interrupt while running the softirq as we see with the capital 'H'. | |
| eb6d42ea SR |
873 | |
| 874 | ||
| 875 | wakeup | |
| 876 | ------ | |
| 877 | ||
| 878 | In Real-Time environment it is very important to know the wakeup | |
| 879 | time it takes for the highest priority task that wakes up to the | |
| 880 | time it executes. This is also known as "schedule latency". | |
| 881 | I stress the point that this is about RT tasks. It is also important | |
| 882 | to know the scheduling latency of non-RT tasks, but the average | |
| 883 | schedule latency is better for non-RT tasks. Tools like | |
| a41eebab | 884 | LatencyTop are more appropriate for such measurements. |
| eb6d42ea | 885 | |
| a41eebab | 886 | Real-Time environments are interested in the worst case latency. |
| eb6d42ea SR |
887 | That is the longest latency it takes for something to happen, and |
| 888 | not the average. We can have a very fast scheduler that may only | |
| 889 | have a large latency once in a while, but that would not work well | |
| 890 | with Real-Time tasks. The wakeup tracer was designed to record | |
| 891 | the worst case wakeups of RT tasks. Non-RT tasks are not recorded | |
| 892 | because the tracer only records one worst case and tracing non-RT | |
| 893 | tasks that are unpredictable will overwrite the worst case latency | |
| 894 | of RT tasks. | |
| 895 | ||
| 896 | Since this tracer only deals with RT tasks, we will run this slightly | |
| a41eebab SR |
897 | differently than we did with the previous tracers. Instead of performing |
| 898 | an 'ls', we will run 'sleep 1' under 'chrt' which changes the | |
| eb6d42ea SR |
899 | priority of the task. |
| 900 | ||
| 901 | # echo wakeup > /debug/tracing/current_tracer | |
| 902 | # echo 0 > /debug/tracing/tracing_max_latency | |
| 903 | # echo 1 > /debug/tracing/tracing_enabled | |
| 904 | # chrt -f 5 sleep 1 | |
| 905 | # echo 0 > /debug/tracing/tracing_enabled | |
| 906 | # cat /debug/tracing/latency_trace | |
| 907 | # tracer: wakeup | |
| 908 | # | |
| 909 | wakeup latency trace v1.1.5 on 2.6.26-rc8 | |
| 910 | -------------------------------------------------------------------- | |
| 911 | latency: 4 us, #2/2, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
| 912 | ----------------- | |
| 913 | | task: sleep-4901 (uid:0 nice:0 policy:1 rt_prio:5) | |
| 914 | ----------------- | |
| 915 | ||
| 916 | # _------=> CPU# | |
| 917 | # / _-----=> irqs-off | |
| 918 | # | / _----=> need-resched | |
| 919 | # || / _---=> hardirq/softirq | |
| 920 | # ||| / _--=> preempt-depth | |
| 921 | # |||| / | |
| 922 | # ||||| delay | |
| 923 | # cmd pid ||||| time | caller | |
| 924 | # \ / ||||| \ | / | |
| 925 | <idle>-0 1d.h4 0us+: try_to_wake_up (wake_up_process) | |
| 926 | <idle>-0 1d..4 4us : schedule (cpu_idle) | |
| 927 | ||
| 928 | ||
| 929 | vim:ft=help | |
| 930 | ||
| 931 | ||
| a41eebab | 932 | Running this on an idle system, we see that it only took 4 microseconds |
| eb6d42ea | 933 | to perform the task switch. Note, since the trace marker in the |
| a41eebab | 934 | schedule is before the actual "switch", we stop the tracing when |
| eb6d42ea SR |
935 | the recorded task is about to schedule in. This may change if |
| 936 | we add a new marker at the end of the scheduler. | |
| 937 | ||
| 938 | Notice that the recorded task is 'sleep' with the PID of 4901 and it | |
| 939 | has an rt_prio of 5. This priority is user-space priority and not | |
| 940 | the internal kernel priority. The policy is 1 for SCHED_FIFO and 2 | |
| 941 | for SCHED_RR. | |
| 942 | ||
| 943 | Doing the same with chrt -r 5 and ftrace_enabled set. | |
| 944 | ||
| 945 | # tracer: wakeup | |
| 946 | # | |
| 947 | wakeup latency trace v1.1.5 on 2.6.26-rc8 | |
| 948 | -------------------------------------------------------------------- | |
| 949 | latency: 50 us, #60/60, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
| 950 | ----------------- | |
| 951 | | task: sleep-4068 (uid:0 nice:0 policy:2 rt_prio:5) | |
| 952 | ----------------- | |
| 953 | ||
| 954 | # _------=> CPU# | |
| 955 | # / _-----=> irqs-off | |
| 956 | # | / _----=> need-resched | |
| 957 | # || / _---=> hardirq/softirq | |
| 958 | # ||| / _--=> preempt-depth | |
| 959 | # |||| / | |
| 960 | # ||||| delay | |
| 961 | # cmd pid ||||| time | caller | |
| 962 | # \ / ||||| \ | / | |
| 963 | ksoftirq-7 1d.H3 0us : try_to_wake_up (wake_up_process) | |
| 964 | ksoftirq-7 1d.H4 1us : sub_preempt_count (marker_probe_cb) | |
| 965 | ksoftirq-7 1d.H3 2us : check_preempt_wakeup (try_to_wake_up) | |
| 966 | ksoftirq-7 1d.H3 3us : update_curr (check_preempt_wakeup) | |
| 967 | ksoftirq-7 1d.H3 4us : calc_delta_mine (update_curr) | |
| 968 | ksoftirq-7 1d.H3 5us : __resched_task (check_preempt_wakeup) | |
| 969 | ksoftirq-7 1d.H3 6us : task_wake_up_rt (try_to_wake_up) | |
| 970 | ksoftirq-7 1d.H3 7us : _spin_unlock_irqrestore (try_to_wake_up) | |
| 971 | [...] | |
| 972 | ksoftirq-7 1d.H2 17us : irq_exit (smp_apic_timer_interrupt) | |
| 973 | ksoftirq-7 1d.H2 18us : sub_preempt_count (irq_exit) | |
| 974 | ksoftirq-7 1d.s3 19us : sub_preempt_count (irq_exit) | |
| 975 | ksoftirq-7 1..s2 20us : rcu_process_callbacks (__do_softirq) | |
| 976 | [...] | |
| 977 | ksoftirq-7 1..s2 26us : __rcu_process_callbacks (rcu_process_callbacks) | |
| 978 | ksoftirq-7 1d.s2 27us : _local_bh_enable (__do_softirq) | |
| 979 | ksoftirq-7 1d.s2 28us : sub_preempt_count (_local_bh_enable) | |
| 980 | ksoftirq-7 1.N.3 29us : sub_preempt_count (ksoftirqd) | |
| 981 | ksoftirq-7 1.N.2 30us : _cond_resched (ksoftirqd) | |
| 982 | ksoftirq-7 1.N.2 31us : __cond_resched (_cond_resched) | |
| 983 | ksoftirq-7 1.N.2 32us : add_preempt_count (__cond_resched) | |
| 984 | ksoftirq-7 1.N.2 33us : schedule (__cond_resched) | |
| 985 | ksoftirq-7 1.N.2 33us : add_preempt_count (schedule) | |
| 986 | ksoftirq-7 1.N.3 34us : hrtick_clear (schedule) | |
| 987 | ksoftirq-7 1dN.3 35us : _spin_lock (schedule) | |
| 988 | ksoftirq-7 1dN.3 36us : add_preempt_count (_spin_lock) | |
| 989 | ksoftirq-7 1d..4 37us : put_prev_task_fair (schedule) | |
| 990 | ksoftirq-7 1d..4 38us : update_curr (put_prev_task_fair) | |
| 991 | [...] | |
| 992 | ksoftirq-7 1d..5 47us : _spin_trylock (tracing_record_cmdline) | |
| 993 | ksoftirq-7 1d..5 48us : add_preempt_count (_spin_trylock) | |
| 994 | ksoftirq-7 1d..6 49us : _spin_unlock (tracing_record_cmdline) | |
| 995 | ksoftirq-7 1d..6 49us : sub_preempt_count (_spin_unlock) | |
| 996 | ksoftirq-7 1d..4 50us : schedule (__cond_resched) | |
| 997 | ||
| 998 | The interrupt went off while running ksoftirqd. This task runs at | |
| 999 | SCHED_OTHER. Why didn't we see the 'N' set early? This may be | |
| a41eebab SR |
1000 | a harmless bug with x86_32 and 4K stacks. On x86_32 with 4K stacks |
| 1001 | configured, the interrupt and softirq runs with their own stack. | |
| 1002 | Some information is held on the top of the task's stack (need_resched | |
| 1003 | and preempt_count are both stored there). The setting of the NEED_RESCHED | |
| 1004 | bit is done directly to the task's stack, but the reading of the | |
| 1005 | NEED_RESCHED is done by looking at the current stack, which in this case | |
| 1006 | is the stack for the hard interrupt. This hides the fact that NEED_RESCHED | |
| 1007 | has been set. We don't see the 'N' until we switch back to the task's | |
| 1008 | assigned stack. | |
| eb6d42ea SR |
1009 | |
| 1010 | ftrace | |
| 1011 | ------ | |
| 1012 | ||
| 1013 | ftrace is not only the name of the tracing infrastructure, but it | |
| 1014 | is also a name of one of the tracers. The tracer is the function | |
| 1015 | tracer. Enabling the function tracer can be done from the | |
| 1016 | debug file system. Make sure the ftrace_enabled is set otherwise | |
| 1017 | this tracer is a nop. | |
| 1018 | ||
| 1019 | # sysctl kernel.ftrace_enabled=1 | |
| 1020 | # echo ftrace > /debug/tracing/current_tracer | |
| 1021 | # echo 1 > /debug/tracing/tracing_enabled | |
| 1022 | # usleep 1 | |
| 1023 | # echo 0 > /debug/tracing/tracing_enabled | |
| 1024 | # cat /debug/tracing/trace | |
| 1025 | # tracer: ftrace | |
| 1026 | # | |
| 1027 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
| 1028 | # | | | | | | |
| 1029 | bash-4003 [00] 123.638713: finish_task_switch <-schedule | |
| 1030 | bash-4003 [00] 123.638714: _spin_unlock_irq <-finish_task_switch | |
| 1031 | bash-4003 [00] 123.638714: sub_preempt_count <-_spin_unlock_irq | |
| 1032 | bash-4003 [00] 123.638715: hrtick_set <-schedule | |
| 1033 | bash-4003 [00] 123.638715: _spin_lock_irqsave <-hrtick_set | |
| 1034 | bash-4003 [00] 123.638716: add_preempt_count <-_spin_lock_irqsave | |
| 1035 | bash-4003 [00] 123.638716: _spin_unlock_irqrestore <-hrtick_set | |
| 1036 | bash-4003 [00] 123.638717: sub_preempt_count <-_spin_unlock_irqrestore | |
| 1037 | bash-4003 [00] 123.638717: hrtick_clear <-hrtick_set | |
| 1038 | bash-4003 [00] 123.638718: sub_preempt_count <-schedule | |
| 1039 | bash-4003 [00] 123.638718: sub_preempt_count <-preempt_schedule | |
| 1040 | bash-4003 [00] 123.638719: wait_for_completion <-__stop_machine_run | |
| 1041 | bash-4003 [00] 123.638719: wait_for_common <-wait_for_completion | |
| 1042 | bash-4003 [00] 123.638720: _spin_lock_irq <-wait_for_common | |
| 1043 | bash-4003 [00] 123.638720: add_preempt_count <-_spin_lock_irq | |
| 1044 | [...] | |
| 1045 | ||
| 1046 | ||
| 1047 | Note: It is sometimes better to enable or disable tracing directly from | |
| 1048 | a program, because the buffer may be overflowed by the echo commands | |
| 1049 | before you get to the point you want to trace. It is also easier to | |
| 1050 | stop the tracing at the point that you hit the part that you are | |
| 1051 | interested in. Since the ftrace buffer is a ring buffer with the | |
| 1052 | oldest data being overwritten, usually it is sufficient to start the | |
| 1053 | tracer with an echo command but have you code stop it. Something | |
| 1054 | like the following is usually appropriate for this. | |
| 1055 | ||
| 1056 | int trace_fd; | |
| 1057 | [...] | |
| 1058 | int main(int argc, char *argv[]) { | |
| 1059 | [...] | |
| 1060 | trace_fd = open("/debug/tracing/tracing_enabled", O_WRONLY); | |
| 1061 | [...] | |
| 1062 | if (condition_hit()) { | |
| 1063 | write(trace_fd, "0", 1); | |
| 1064 | } | |
| 1065 | [...] | |
| 1066 | } | |
| 1067 | ||
| 1068 | ||
| 1069 | dynamic ftrace | |
| 1070 | -------------- | |
| 1071 | ||
| 1072 | If CONFIG_DYNAMIC_FTRACE is set, then the system will run with | |
| 1073 | virtually no overhead when function tracing is disabled. The way | |
| 1074 | this works is the mcount function call (placed at the start of | |
| 1075 | every kernel function, produced by the -pg switch in gcc), starts | |
| 1076 | of pointing to a simple return. | |
| 1077 | ||
| a41eebab SR |
1078 | When dynamic ftrace is initialized, it calls kstop_machine to make |
| 1079 | the machine act like a uniprocessor so that it can freely modify code | |
| 1080 | without worrying about other processors executing that same code. At | |
| 1081 | initialization, the mcount calls are changed to call a "record_ip" | |
| eb6d42ea SR |
1082 | function. After this, the first time a kernel function is called, |
| 1083 | it has the calling address saved in a hash table. | |
| 1084 | ||
| 1085 | Later on the ftraced kernel thread is awoken and will again call | |
| 1086 | kstop_machine if new functions have been recorded. The ftraced thread | |
| 1087 | will change all calls to mcount to "nop". Just calling mcount | |
| 1088 | and having mcount return has shown a 10% overhead. By converting | |
| 1089 | it to a nop, there is no recordable overhead to the system. | |
| 1090 | ||
| 1091 | One special side-effect to the recording of the functions being | |
| 1092 | traced, is that we can now selectively choose which functions we | |
| 1093 | want to trace and which ones we want the mcount calls to remain as | |
| 1094 | nops. | |
| 1095 | ||
| a41eebab SR |
1096 | Two files are used, one for enabling and one for disabling the tracing |
| 1097 | of recorded functions. They are: | |
| eb6d42ea SR |
1098 | |
| 1099 | set_ftrace_filter | |
| 1100 | ||
| 1101 | and | |
| 1102 | ||
| 1103 | set_ftrace_notrace | |
| 1104 | ||
| a41eebab | 1105 | A list of available functions that you can add to these files is listed |
| eb6d42ea SR |
1106 | in: |
| 1107 | ||
| 1108 | available_filter_functions | |
| 1109 | ||
| 1110 | # cat /debug/tracing/available_filter_functions | |
| 1111 | put_prev_task_idle | |
| 1112 | kmem_cache_create | |
| 1113 | pick_next_task_rt | |
| 1114 | get_online_cpus | |
| 1115 | pick_next_task_fair | |
| 1116 | mutex_lock | |
| 1117 | [...] | |
| 1118 | ||
| 1119 | If I'm only interested in sys_nanosleep and hrtimer_interrupt: | |
| 1120 | ||
| 1121 | # echo sys_nanosleep hrtimer_interrupt \ | |
| 1122 | > /debug/tracing/set_ftrace_filter | |
| 1123 | # echo ftrace > /debug/tracing/current_tracer | |
| 1124 | # echo 1 > /debug/tracing/tracing_enabled | |
| 1125 | # usleep 1 | |
| 1126 | # echo 0 > /debug/tracing/tracing_enabled | |
| 1127 | # cat /debug/tracing/trace | |
| 1128 | # tracer: ftrace | |
| 1129 | # | |
| 1130 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
| 1131 | # | | | | | | |
| 1132 | usleep-4134 [00] 1317.070017: hrtimer_interrupt <-smp_apic_timer_interrupt | |
| 1133 | usleep-4134 [00] 1317.070111: sys_nanosleep <-syscall_call | |
| 1134 | <idle>-0 [00] 1317.070115: hrtimer_interrupt <-smp_apic_timer_interrupt | |
| 1135 | ||
| 1136 | To see what functions are being traced, you can cat the file: | |
| 1137 | ||
| 1138 | # cat /debug/tracing/set_ftrace_filter | |
| 1139 | hrtimer_interrupt | |
| 1140 | sys_nanosleep | |
| 1141 | ||
| 1142 | ||
| 1143 | Perhaps this isn't enough. The filters also allow simple wild cards. | |
| a41eebab | 1144 | Only the following are currently available |
| eb6d42ea | 1145 | |
| a41eebab | 1146 | <match>* - will match functions that begin with <match> |
| eb6d42ea SR |
1147 | *<match> - will match functions that end with <match> |
| 1148 | *<match>* - will match functions that have <match> in it | |
| 1149 | ||
| 1150 | Thats all the wild cards that are allowed. | |
| 1151 | ||
| 1152 | <match>*<match> will not work. | |
| 1153 | ||
| 1154 | # echo hrtimer_* > /debug/tracing/set_ftrace_filter | |
| 1155 | ||
| 1156 | Produces: | |
| 1157 | ||
| 1158 | # tracer: ftrace | |
| 1159 | # | |
| 1160 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
| 1161 | # | | | | | | |
| 1162 | bash-4003 [00] 1480.611794: hrtimer_init <-copy_process | |
| 1163 | bash-4003 [00] 1480.611941: hrtimer_start <-hrtick_set | |
| 1164 | bash-4003 [00] 1480.611956: hrtimer_cancel <-hrtick_clear | |
| 1165 | bash-4003 [00] 1480.611956: hrtimer_try_to_cancel <-hrtimer_cancel | |
| 1166 | <idle>-0 [00] 1480.612019: hrtimer_get_next_event <-get_next_timer_interrupt | |
| 1167 | <idle>-0 [00] 1480.612025: hrtimer_get_next_event <-get_next_timer_interrupt | |
| 1168 | <idle>-0 [00] 1480.612032: hrtimer_get_next_event <-get_next_timer_interrupt | |
| 1169 | <idle>-0 [00] 1480.612037: hrtimer_get_next_event <-get_next_timer_interrupt | |
| 1170 | <idle>-0 [00] 1480.612382: hrtimer_get_next_event <-get_next_timer_interrupt | |
| 1171 | ||
| 1172 | ||
| 1173 | Notice that we lost the sys_nanosleep. | |
| 1174 | ||
| 1175 | # cat /debug/tracing/set_ftrace_filter | |
| 1176 | hrtimer_run_queues | |
| 1177 | hrtimer_run_pending | |
| 1178 | hrtimer_init | |
| 1179 | hrtimer_cancel | |
| 1180 | hrtimer_try_to_cancel | |
| 1181 | hrtimer_forward | |
| 1182 | hrtimer_start | |
| 1183 | hrtimer_reprogram | |
| 1184 | hrtimer_force_reprogram | |
| 1185 | hrtimer_get_next_event | |
| 1186 | hrtimer_interrupt | |
| 1187 | hrtimer_nanosleep | |
| 1188 | hrtimer_wakeup | |
| 1189 | hrtimer_get_remaining | |
| 1190 | hrtimer_get_res | |
| 1191 | hrtimer_init_sleeper | |
| 1192 | ||
| 1193 | ||
| 1194 | This is because the '>' and '>>' act just like they do in bash. | |
| 1195 | To rewrite the filters, use '>' | |
| 1196 | To append to the filters, use '>>' | |
| 1197 | ||
| a41eebab | 1198 | To clear out a filter so that all functions will be recorded again: |
| eb6d42ea SR |
1199 | |
| 1200 | # echo > /debug/tracing/set_ftrace_filter | |
| 1201 | # cat /debug/tracing/set_ftrace_filter | |
| 1202 | # | |
| 1203 | ||
| 1204 | Again, now we want to append. | |
| 1205 | ||
| 1206 | # echo sys_nanosleep > /debug/tracing/set_ftrace_filter | |
| 1207 | # cat /debug/tracing/set_ftrace_filter | |
| 1208 | sys_nanosleep | |
| 1209 | # echo hrtimer_* >> /debug/tracing/set_ftrace_filter | |
| 1210 | # cat /debug/tracing/set_ftrace_filter | |
| 1211 | hrtimer_run_queues | |
| 1212 | hrtimer_run_pending | |
| 1213 | hrtimer_init | |
| 1214 | hrtimer_cancel | |
| 1215 | hrtimer_try_to_cancel | |
| 1216 | hrtimer_forward | |
| 1217 | hrtimer_start | |
| 1218 | hrtimer_reprogram | |
| 1219 | hrtimer_force_reprogram | |
| 1220 | hrtimer_get_next_event | |
| 1221 | hrtimer_interrupt | |
| 1222 | sys_nanosleep | |
| 1223 | hrtimer_nanosleep | |
| 1224 | hrtimer_wakeup | |
| 1225 | hrtimer_get_remaining | |
| 1226 | hrtimer_get_res | |
| 1227 | hrtimer_init_sleeper | |
| 1228 | ||
| 1229 | ||
| 1230 | The set_ftrace_notrace prevents those functions from being traced. | |
| 1231 | ||
| 1232 | # echo '*preempt*' '*lock*' > /debug/tracing/set_ftrace_notrace | |
| 1233 | ||
| 1234 | Produces: | |
| 1235 | ||
| 1236 | # tracer: ftrace | |
| 1237 | # | |
| 1238 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
| 1239 | # | | | | | | |
| 1240 | bash-4043 [01] 115.281644: finish_task_switch <-schedule | |
| 1241 | bash-4043 [01] 115.281645: hrtick_set <-schedule | |
| 1242 | bash-4043 [01] 115.281645: hrtick_clear <-hrtick_set | |
| 1243 | bash-4043 [01] 115.281646: wait_for_completion <-__stop_machine_run | |
| 1244 | bash-4043 [01] 115.281647: wait_for_common <-wait_for_completion | |
| 1245 | bash-4043 [01] 115.281647: kthread_stop <-stop_machine_run | |
| 1246 | bash-4043 [01] 115.281648: init_waitqueue_head <-kthread_stop | |
| 1247 | bash-4043 [01] 115.281648: wake_up_process <-kthread_stop | |
| 1248 | bash-4043 [01] 115.281649: try_to_wake_up <-wake_up_process | |
| 1249 | ||
| 1250 | We can see that there's no more lock or preempt tracing. | |
| 1251 | ||
| 1252 | ftraced | |
| 1253 | ------- | |
| 1254 | ||
| 1255 | As mentioned above, when dynamic ftrace is configured in, a kernel | |
| 1256 | thread wakes up once a second and checks to see if there are mcount | |
| a41eebab SR |
1257 | calls that need to be converted into nops. If there are not any, then |
| 1258 | it simply goes back to sleep. But if there are some, it will call | |
| eb6d42ea SR |
1259 | kstop_machine to convert the calls to nops. |
| 1260 | ||
| 1261 | There may be a case that you do not want this added latency. | |
| 1262 | Perhaps you are doing some audio recording and this activity might | |
| 1263 | cause skips in the playback. There is an interface to disable | |
| 1264 | and enable the ftraced kernel thread. | |
| 1265 | ||
| 1266 | # echo 0 > /debug/tracing/ftraced_enabled | |
| 1267 | ||
| 1268 | This will disable the calling of the kstop_machine to update the | |
| 1269 | mcount calls to nops. Remember that there's a large overhead | |
| 1270 | to calling mcount. Without this kernel thread, that overhead will | |
| 1271 | exist. | |
| 1272 | ||
| a41eebab SR |
1273 | If there are recorded calls to mcount, any write to the ftraced_enabled |
| 1274 | file will cause the kstop_machine to run. This means that a | |
| eb6d42ea SR |
1275 | user can manually perform the updates when they want to by simply |
| 1276 | echoing a '0' into the ftraced_enabled file. | |
| 1277 | ||
| 1278 | The updates are also done at the beginning of enabling a tracer | |
| 1279 | that uses ftrace function recording. | |
| 1280 | ||
| 1281 | ||
| 1282 | trace_pipe | |
| 1283 | ---------- | |
| 1284 | ||
| 1285 | The trace_pipe outputs the same as trace, but the effect on the | |
| 1286 | tracing is different. Every read from trace_pipe is consumed. | |
| 1287 | This means that subsequent reads will be different. The trace | |
| 1288 | is live. | |
| 1289 | ||
| 1290 | # echo ftrace > /debug/tracing/current_tracer | |
| 1291 | # cat /debug/tracing/trace_pipe > /tmp/trace.out & | |
| 1292 | [1] 4153 | |
| 1293 | # echo 1 > /debug/tracing/tracing_enabled | |
| 1294 | # usleep 1 | |
| 1295 | # echo 0 > /debug/tracing/tracing_enabled | |
| 1296 | # cat /debug/tracing/trace | |
| 1297 | # tracer: ftrace | |
| 1298 | # | |
| 1299 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
| 1300 | # | | | | | | |
| 1301 | ||
| 1302 | # | |
| 1303 | # cat /tmp/trace.out | |
| 1304 | bash-4043 [00] 41.267106: finish_task_switch <-schedule | |
| 1305 | bash-4043 [00] 41.267106: hrtick_set <-schedule | |
| 1306 | bash-4043 [00] 41.267107: hrtick_clear <-hrtick_set | |
| 1307 | bash-4043 [00] 41.267108: wait_for_completion <-__stop_machine_run | |
| 1308 | bash-4043 [00] 41.267108: wait_for_common <-wait_for_completion | |
| 1309 | bash-4043 [00] 41.267109: kthread_stop <-stop_machine_run | |
| 1310 | bash-4043 [00] 41.267109: init_waitqueue_head <-kthread_stop | |
| 1311 | bash-4043 [00] 41.267110: wake_up_process <-kthread_stop | |
| 1312 | bash-4043 [00] 41.267110: try_to_wake_up <-wake_up_process | |
| 1313 | bash-4043 [00] 41.267111: select_task_rq_rt <-try_to_wake_up | |
| 1314 | ||
| 1315 | ||
| 1316 | Note, reading the trace_pipe will block until more input is added. | |
| 1317 | By changing the tracer, trace_pipe will issue an EOF. We needed | |
| 1318 | to set the ftrace tracer _before_ cating the trace_pipe file. | |
| 1319 | ||
| 1320 | ||
| 1321 | trace entries | |
| 1322 | ------------- | |
| 1323 | ||
| 1324 | Having too much or not enough data can be troublesome in diagnosing | |
| 1325 | some issue in the kernel. The file trace_entries is used to modify | |
| a41eebab | 1326 | the size of the internal trace buffers. The number listed |
| eb6d42ea SR |
1327 | is the number of entries that can be recorded per CPU. To know |
| 1328 | the full size, multiply the number of possible CPUS with the | |
| 1329 | number of entries. | |
| 1330 | ||
| 1331 | # cat /debug/tracing/trace_entries | |
| 1332 | 65620 | |
| 1333 | ||
| a41eebab | 1334 | Note, to modify this, you must have tracing completely disabled. To do that, |
| eb6d42ea SR |
1335 | echo "none" into the current_tracer. |
| 1336 | ||
| 1337 | # echo none > /debug/tracing/current_tracer | |
| 1338 | # echo 100000 > /debug/tracing/trace_entries | |
| 1339 | # cat /debug/tracing/trace_entries | |
| 1340 | 100045 | |
| 1341 | ||
| 1342 | ||
| 1343 | Notice that we echoed in 100,000 but the size is 100,045. The entries | |
| 1344 | are held by individual pages. It allocates the number of pages it takes | |
| 1345 | to fulfill the request. If more entries may fit on the last page | |
| 1346 | it will add them. | |
| 1347 | ||
| 1348 | # echo 1 > /debug/tracing/trace_entries | |
| 1349 | # cat /debug/tracing/trace_entries | |
| 1350 | 85 | |
| 1351 | ||
| 1352 | This shows us that 85 entries can fit on a single page. | |
| 1353 | ||
| 1354 | The number of pages that will be allocated is a percentage of available | |
| a41eebab | 1355 | memory. Allocating too much will produce an error. |
| eb6d42ea SR |
1356 | |
| 1357 | # echo 1000000000000 > /debug/tracing/trace_entries | |
| 1358 | -bash: echo: write error: Cannot allocate memory | |
| 1359 | # cat /debug/tracing/trace_entries | |
| 1360 | 85 | |
| 1361 |