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 | |
a97762a7 | 7 | (dual licensed under the GPL v2) |
f2d9c740 SR |
8 | Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton, |
9 | John Kacur, and David Teigland. | |
42ec632e | 10 | Written for: 2.6.28-rc2 |
eb6d42ea SR |
11 | |
12 | Introduction | |
13 | ------------ | |
14 | ||
15 | Ftrace is an internal tracer designed to help out developers and | |
16 | designers of systems to find what is going on inside the kernel. | |
5752674e IM |
17 | It can be used for debugging or analyzing latencies and |
18 | performance issues that take place outside of user-space. | |
eb6d42ea SR |
19 | |
20 | Although ftrace is the function tracer, it also includes an | |
5752674e IM |
21 | infrastructure that allows for other types of tracing. Some of |
22 | the tracers that are currently in ftrace include a tracer to | |
23 | trace context switches, the time it takes for a high priority | |
24 | task to run after it was woken up, the time interrupts are | |
25 | disabled, and more (ftrace allows for tracer plugins, which | |
26 | means that the list of tracers can always grow). | |
eb6d42ea SR |
27 | |
28 | ||
555f386c MF |
29 | Implementation Details |
30 | ---------------------- | |
31 | ||
32 | See ftrace-design.txt for details for arch porters and such. | |
33 | ||
34 | ||
eb6d42ea SR |
35 | The File System |
36 | --------------- | |
37 | ||
5752674e IM |
38 | Ftrace uses the debugfs file system to hold the control files as |
39 | well as the files to display output. | |
eb6d42ea | 40 | |
156f5a78 GL |
41 | When debugfs is configured into the kernel (which selecting any ftrace |
42 | option will do) the directory /sys/kernel/debug will be created. To mount | |
43 | this directory, you can add to your /etc/fstab file: | |
44 | ||
45 | debugfs /sys/kernel/debug debugfs defaults 0 0 | |
46 | ||
47 | Or you can mount it at run time with: | |
48 | ||
49 | mount -t debugfs nodev /sys/kernel/debug | |
eb6d42ea | 50 | |
156f5a78 GL |
51 | For quicker access to that directory you may want to make a soft link to |
52 | it: | |
eb6d42ea | 53 | |
156f5a78 GL |
54 | ln -s /sys/kernel/debug /debug |
55 | ||
56 | Any selected ftrace option will also create a directory called tracing | |
57 | within the debugfs. The rest of the document will assume that you are in | |
58 | the ftrace directory (cd /sys/kernel/debug/tracing) and will only concentrate | |
59 | on the files within that directory and not distract from the content with | |
60 | the extended "/sys/kernel/debug/tracing" path name. | |
eb6d42ea SR |
61 | |
62 | That's it! (assuming that you have ftrace configured into your kernel) | |
63 | ||
64 | After mounting the debugfs, you can see a directory called | |
65 | "tracing". This directory contains the control and output files | |
66 | of ftrace. Here is a list of some of the key files: | |
67 | ||
68 | ||
69 | Note: all time values are in microseconds. | |
70 | ||
5752674e IM |
71 | current_tracer: |
72 | ||
73 | This is used to set or display the current tracer | |
74 | that is configured. | |
75 | ||
76 | available_tracers: | |
77 | ||
78 | This holds the different types of tracers that | |
79 | have been compiled into the kernel. The | |
80 | tracers listed here can be configured by | |
81 | echoing their name into current_tracer. | |
82 | ||
83 | tracing_enabled: | |
84 | ||
85 | This sets or displays whether the current_tracer | |
86 | is activated and tracing or not. Echo 0 into this | |
87 | file to disable the tracer or 1 to enable it. | |
88 | ||
89 | trace: | |
90 | ||
91 | This file holds the output of the trace in a human | |
92 | readable format (described below). | |
93 | ||
5752674e IM |
94 | trace_pipe: |
95 | ||
96 | The output is the same as the "trace" file but this | |
97 | file is meant to be streamed with live tracing. | |
4a88d44a AT |
98 | Reads from this file will block until new data is |
99 | retrieved. Unlike the "trace" file, this file is a | |
100 | consumer. This means reading from this file causes | |
101 | sequential reads to display more current data. Once | |
102 | data is read from this file, it is consumed, and | |
103 | will not be read again with a sequential read. The | |
104 | "trace" file is static, and if the tracer is not | |
105 | adding more data,they will display the same | |
106 | information every time they are read. | |
5752674e IM |
107 | |
108 | trace_options: | |
109 | ||
110 | This file lets the user control the amount of data | |
111 | that is displayed in one of the above output | |
112 | files. | |
113 | ||
42b40b3d | 114 | tracing_max_latency: |
5752674e IM |
115 | |
116 | Some of the tracers record the max latency. | |
117 | For example, the time interrupts are disabled. | |
118 | This time is saved in this file. The max trace | |
4a88d44a AT |
119 | will also be stored, and displayed by "trace". |
120 | A new max trace will only be recorded if the | |
121 | latency is greater than the value in this | |
122 | file. (in microseconds) | |
5752674e IM |
123 | |
124 | buffer_size_kb: | |
125 | ||
126 | This sets or displays the number of kilobytes each CPU | |
127 | buffer can hold. The tracer buffers are the same size | |
128 | for each CPU. The displayed number is the size of the | |
129 | CPU buffer and not total size of all buffers. The | |
130 | trace buffers are allocated in pages (blocks of memory | |
131 | that the kernel uses for allocation, usually 4 KB in size). | |
132 | If the last page allocated has room for more bytes | |
133 | than requested, the rest of the page will be used, | |
134 | making the actual allocation bigger than requested. | |
135 | ( Note, the size may not be a multiple of the page size | |
3dbda77e | 136 | due to buffer management overhead. ) |
5752674e IM |
137 | |
138 | This can only be updated when the current_tracer | |
139 | is set to "nop". | |
140 | ||
141 | tracing_cpumask: | |
142 | ||
143 | This is a mask that lets the user only trace | |
144 | on specified CPUS. The format is a hex string | |
145 | representing the CPUS. | |
146 | ||
147 | set_ftrace_filter: | |
148 | ||
149 | When dynamic ftrace is configured in (see the | |
150 | section below "dynamic ftrace"), the code is dynamically | |
151 | modified (code text rewrite) to disable calling of the | |
152 | function profiler (mcount). This lets tracing be configured | |
153 | in with practically no overhead in performance. This also | |
154 | has a side effect of enabling or disabling specific functions | |
155 | to be traced. Echoing names of functions into this file | |
156 | will limit the trace to only those functions. | |
157 | ||
07271aa4 CD |
158 | This interface also allows for commands to be used. See the |
159 | "Filter commands" section for more details. | |
160 | ||
5752674e IM |
161 | set_ftrace_notrace: |
162 | ||
163 | This has an effect opposite to that of | |
164 | set_ftrace_filter. Any function that is added here will not | |
165 | be traced. If a function exists in both set_ftrace_filter | |
166 | and set_ftrace_notrace, the function will _not_ be traced. | |
167 | ||
168 | set_ftrace_pid: | |
169 | ||
170 | Have the function tracer only trace a single thread. | |
171 | ||
172 | set_graph_function: | |
173 | ||
174 | Set a "trigger" function where tracing should start | |
175 | with the function graph tracer (See the section | |
176 | "dynamic ftrace" for more details). | |
177 | ||
178 | available_filter_functions: | |
179 | ||
180 | This lists the functions that ftrace | |
181 | has processed and can trace. These are the function | |
182 | names that you can pass to "set_ftrace_filter" or | |
183 | "set_ftrace_notrace". (See the section "dynamic ftrace" | |
184 | below for more details.) | |
eb6d42ea SR |
185 | |
186 | ||
187 | The Tracers | |
188 | ----------- | |
189 | ||
f2d9c740 | 190 | Here is the list of current tracers that may be configured. |
eb6d42ea | 191 | |
5752674e IM |
192 | "function" |
193 | ||
194 | Function call tracer to trace all kernel functions. | |
195 | ||
bc5c6c04 | 196 | "function_graph" |
5752674e IM |
197 | |
198 | Similar to the function tracer except that the | |
199 | function tracer probes the functions on their entry | |
200 | whereas the function graph tracer traces on both entry | |
201 | and exit of the functions. It then provides the ability | |
202 | to draw a graph of function calls similar to C code | |
203 | source. | |
204 | ||
5752674e IM |
205 | "irqsoff" |
206 | ||
207 | Traces the areas that disable interrupts and saves | |
208 | the trace with the longest max latency. | |
209 | See tracing_max_latency. When a new max is recorded, | |
210 | it replaces the old trace. It is best to view this | |
4a88d44a | 211 | trace with the latency-format option enabled. |
eb6d42ea | 212 | |
5752674e | 213 | "preemptoff" |
985ec20a | 214 | |
5752674e IM |
215 | Similar to irqsoff but traces and records the amount of |
216 | time for which preemption is disabled. | |
eb6d42ea | 217 | |
5752674e | 218 | "preemptirqsoff" |
eb6d42ea | 219 | |
5752674e IM |
220 | Similar to irqsoff and preemptoff, but traces and |
221 | records the largest time for which irqs and/or preemption | |
222 | is disabled. | |
eb6d42ea | 223 | |
5752674e | 224 | "wakeup" |
eb6d42ea | 225 | |
5752674e IM |
226 | Traces and records the max latency that it takes for |
227 | the highest priority task to get scheduled after | |
228 | it has been woken up. | |
eb6d42ea | 229 | |
5752674e | 230 | "hw-branch-tracer" |
eb6d42ea | 231 | |
5752674e IM |
232 | Uses the BTS CPU feature on x86 CPUs to traces all |
233 | branches executed. | |
234 | ||
235 | "nop" | |
236 | ||
237 | This is the "trace nothing" tracer. To remove all | |
238 | tracers from tracing simply echo "nop" into | |
239 | current_tracer. | |
e2ea5399 | 240 | |
eb6d42ea SR |
241 | |
242 | Examples of using the tracer | |
243 | ---------------------------- | |
244 | ||
5752674e IM |
245 | Here are typical examples of using the tracers when controlling |
246 | them only with the debugfs interface (without using any | |
247 | user-land utilities). | |
eb6d42ea SR |
248 | |
249 | Output format: | |
250 | -------------- | |
251 | ||
f2d9c740 | 252 | Here is an example of the output format of the file "trace" |
eb6d42ea SR |
253 | |
254 | -------- | |
9b803c0f | 255 | # tracer: function |
eb6d42ea SR |
256 | # |
257 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
258 | # | | | | | | |
259 | bash-4251 [01] 10152.583854: path_put <-path_walk | |
260 | bash-4251 [01] 10152.583855: dput <-path_put | |
261 | bash-4251 [01] 10152.583855: _atomic_dec_and_lock <-dput | |
262 | -------- | |
263 | ||
5752674e IM |
264 | A header is printed with the tracer name that is represented by |
265 | the trace. In this case the tracer is "function". Then a header | |
266 | showing the format. Task name "bash", the task PID "4251", the | |
267 | CPU that it was running on "01", the timestamp in <secs>.<usecs> | |
268 | format, the function name that was traced "path_put" and the | |
269 | parent function that called this function "path_walk". The | |
270 | timestamp is the time at which the function was entered. | |
eb6d42ea | 271 | |
eb6d42ea SR |
272 | Latency trace format |
273 | -------------------- | |
274 | ||
4a88d44a AT |
275 | When the latency-format option is enabled, the trace file gives |
276 | somewhat more information to see why a latency happened. | |
5752674e | 277 | Here is a typical trace. |
eb6d42ea SR |
278 | |
279 | # tracer: irqsoff | |
280 | # | |
281 | irqsoff latency trace v1.1.5 on 2.6.26-rc8 | |
282 | -------------------------------------------------------------------- | |
283 | latency: 97 us, #3/3, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
284 | ----------------- | |
285 | | task: swapper-0 (uid:0 nice:0 policy:0 rt_prio:0) | |
286 | ----------------- | |
287 | => started at: apic_timer_interrupt | |
288 | => ended at: do_softirq | |
289 | ||
290 | # _------=> CPU# | |
291 | # / _-----=> irqs-off | |
292 | # | / _----=> need-resched | |
293 | # || / _---=> hardirq/softirq | |
294 | # ||| / _--=> preempt-depth | |
295 | # |||| / | |
296 | # ||||| delay | |
297 | # cmd pid ||||| time | caller | |
298 | # \ / ||||| \ | / | |
299 | <idle>-0 0d..1 0us+: trace_hardirqs_off_thunk (apic_timer_interrupt) | |
300 | <idle>-0 0d.s. 97us : __do_softirq (do_softirq) | |
301 | <idle>-0 0d.s1 98us : trace_hardirqs_on (do_softirq) | |
302 | ||
303 | ||
5752674e IM |
304 | This shows that the current tracer is "irqsoff" tracing the time |
305 | for which interrupts were disabled. It gives the trace version | |
306 | and the version of the kernel upon which this was executed on | |
307 | (2.6.26-rc8). Then it displays the max latency in microsecs (97 | |
308 | us). The number of trace entries displayed and the total number | |
309 | recorded (both are three: #3/3). The type of preemption that was | |
310 | used (PREEMPT). VP, KP, SP, and HP are always zero and are | |
311 | reserved for later use. #P is the number of online CPUS (#P:2). | |
eb6d42ea | 312 | |
5752674e IM |
313 | The task is the process that was running when the latency |
314 | occurred. (swapper pid: 0). | |
eb6d42ea | 315 | |
5752674e IM |
316 | The start and stop (the functions in which the interrupts were |
317 | disabled and enabled respectively) that caused the latencies: | |
eb6d42ea SR |
318 | |
319 | apic_timer_interrupt is where the interrupts were disabled. | |
320 | do_softirq is where they were enabled again. | |
321 | ||
322 | The next lines after the header are the trace itself. The header | |
323 | explains which is which. | |
324 | ||
325 | cmd: The name of the process in the trace. | |
326 | ||
327 | pid: The PID of that process. | |
328 | ||
f2d9c740 | 329 | CPU#: The CPU which the process was running on. |
eb6d42ea SR |
330 | |
331 | irqs-off: 'd' interrupts are disabled. '.' otherwise. | |
9244489a SR |
332 | Note: If the architecture does not support a way to |
333 | read the irq flags variable, an 'X' will always | |
334 | be printed here. | |
eb6d42ea SR |
335 | |
336 | need-resched: 'N' task need_resched is set, '.' otherwise. | |
337 | ||
338 | hardirq/softirq: | |
f2d9c740 | 339 | 'H' - hard irq occurred inside a softirq. |
eb6d42ea SR |
340 | 'h' - hard irq is running |
341 | 's' - soft irq is running | |
342 | '.' - normal context. | |
343 | ||
344 | preempt-depth: The level of preempt_disabled | |
345 | ||
346 | The above is mostly meaningful for kernel developers. | |
347 | ||
4a88d44a AT |
348 | time: When the latency-format option is enabled, the trace file |
349 | output includes a timestamp relative to the start of the | |
350 | trace. This differs from the output when latency-format | |
351 | is disabled, which includes an absolute timestamp. | |
eb6d42ea SR |
352 | |
353 | delay: This is just to help catch your eye a bit better. And | |
5752674e IM |
354 | needs to be fixed to be only relative to the same CPU. |
355 | The marks are determined by the difference between this | |
356 | current trace and the next trace. | |
357 | '!' - greater than preempt_mark_thresh (default 100) | |
358 | '+' - greater than 1 microsecond | |
359 | ' ' - less than or equal to 1 microsecond. | |
eb6d42ea SR |
360 | |
361 | The rest is the same as the 'trace' file. | |
362 | ||
363 | ||
ee6bce52 SR |
364 | trace_options |
365 | ------------- | |
eb6d42ea | 366 | |
5752674e IM |
367 | The trace_options file is used to control what gets printed in |
368 | the trace output. To see what is available, simply cat the file: | |
eb6d42ea | 369 | |
156f5a78 | 370 | cat trace_options |
eb6d42ea | 371 | print-parent nosym-offset nosym-addr noverbose noraw nohex nobin \ |
5752674e | 372 | noblock nostacktrace nosched-tree nouserstacktrace nosym-userobj |
eb6d42ea | 373 | |
5752674e IM |
374 | To disable one of the options, echo in the option prepended with |
375 | "no". | |
eb6d42ea | 376 | |
156f5a78 | 377 | echo noprint-parent > trace_options |
eb6d42ea SR |
378 | |
379 | To enable an option, leave off the "no". | |
380 | ||
156f5a78 | 381 | echo sym-offset > trace_options |
eb6d42ea SR |
382 | |
383 | Here are the available options: | |
384 | ||
5752674e IM |
385 | print-parent - On function traces, display the calling (parent) |
386 | function as well as the function being traced. | |
eb6d42ea SR |
387 | |
388 | print-parent: | |
389 | bash-4000 [01] 1477.606694: simple_strtoul <-strict_strtoul | |
390 | ||
391 | noprint-parent: | |
392 | bash-4000 [01] 1477.606694: simple_strtoul | |
393 | ||
394 | ||
5752674e IM |
395 | sym-offset - Display not only the function name, but also the |
396 | offset in the function. For example, instead of | |
397 | seeing just "ktime_get", you will see | |
398 | "ktime_get+0xb/0x20". | |
eb6d42ea SR |
399 | |
400 | sym-offset: | |
401 | bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0 | |
402 | ||
5752674e IM |
403 | sym-addr - this will also display the function address as well |
404 | as the function name. | |
eb6d42ea SR |
405 | |
406 | sym-addr: | |
407 | bash-4000 [01] 1477.606694: simple_strtoul <c0339346> | |
408 | ||
4a88d44a AT |
409 | verbose - This deals with the trace file when the |
410 | latency-format option is enabled. | |
eb6d42ea SR |
411 | |
412 | bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \ | |
413 | (+0.000ms): simple_strtoul (strict_strtoul) | |
414 | ||
5752674e IM |
415 | raw - This will display raw numbers. This option is best for |
416 | use with user applications that can translate the raw | |
417 | numbers better than having it done in the kernel. | |
eb6d42ea | 418 | |
5752674e IM |
419 | hex - Similar to raw, but the numbers will be in a hexadecimal |
420 | format. | |
eb6d42ea SR |
421 | |
422 | bin - This will print out the formats in raw binary. | |
423 | ||
424 | block - TBD (needs update) | |
425 | ||
5752674e IM |
426 | stacktrace - This is one of the options that changes the trace |
427 | itself. When a trace is recorded, so is the stack | |
428 | of functions. This allows for back traces of | |
429 | trace sites. | |
eb6d42ea | 430 | |
5752674e IM |
431 | userstacktrace - This option changes the trace. It records a |
432 | stacktrace of the current userspace thread. | |
02b67518 | 433 | |
5752674e IM |
434 | sym-userobj - when user stacktrace are enabled, look up which |
435 | object the address belongs to, and print a | |
436 | relative address. This is especially useful when | |
437 | ASLR is on, otherwise you don't get a chance to | |
438 | resolve the address to object/file/line after | |
439 | the app is no longer running | |
b54d3de9 | 440 | |
5752674e | 441 | The lookup is performed when you read |
4a88d44a | 442 | trace,trace_pipe. Example: |
b54d3de9 TE |
443 | |
444 | a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0 | |
445 | x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6] | |
446 | ||
5752674e IM |
447 | sched-tree - trace all tasks that are on the runqueue, at |
448 | every scheduling event. Will add overhead if | |
449 | there's a lot of tasks running at once. | |
eb6d42ea | 450 | |
4a88d44a AT |
451 | latency-format - This option changes the trace. When |
452 | it is enabled, the trace displays | |
453 | additional information about the | |
454 | latencies, as described in "Latency | |
455 | trace format". | |
eb6d42ea | 456 | |
eb6d42ea SR |
457 | ftrace_enabled |
458 | -------------- | |
459 | ||
5752674e IM |
460 | The following tracers (listed below) give different output |
461 | depending on whether or not the sysctl ftrace_enabled is set. To | |
462 | set ftrace_enabled, one can either use the sysctl function or | |
463 | set it via the proc file system interface. | |
eb6d42ea SR |
464 | |
465 | sysctl kernel.ftrace_enabled=1 | |
466 | ||
467 | or | |
468 | ||
469 | echo 1 > /proc/sys/kernel/ftrace_enabled | |
470 | ||
5752674e IM |
471 | To disable ftrace_enabled simply replace the '1' with '0' in the |
472 | above commands. | |
eb6d42ea | 473 | |
5752674e IM |
474 | When ftrace_enabled is set the tracers will also record the |
475 | functions that are within the trace. The descriptions of the | |
476 | tracers will also show an example with ftrace enabled. | |
eb6d42ea SR |
477 | |
478 | ||
479 | irqsoff | |
480 | ------- | |
481 | ||
482 | When interrupts are disabled, the CPU can not react to any other | |
483 | external event (besides NMIs and SMIs). This prevents the timer | |
5752674e IM |
484 | interrupt from triggering or the mouse interrupt from letting |
485 | the kernel know of a new mouse event. The result is a latency | |
486 | with the reaction time. | |
eb6d42ea | 487 | |
5752674e IM |
488 | The irqsoff tracer tracks the time for which interrupts are |
489 | disabled. When a new maximum latency is hit, the tracer saves | |
490 | the trace leading up to that latency point so that every time a | |
491 | new maximum is reached, the old saved trace is discarded and the | |
492 | new trace is saved. | |
eb6d42ea | 493 | |
5752674e IM |
494 | To reset the maximum, echo 0 into tracing_max_latency. Here is |
495 | an example: | |
eb6d42ea | 496 | |
156f5a78 | 497 | # echo irqsoff > current_tracer |
4a88d44a | 498 | # echo latency-format > trace_options |
156f5a78 GL |
499 | # echo 0 > tracing_max_latency |
500 | # echo 1 > tracing_enabled | |
eb6d42ea SR |
501 | # ls -ltr |
502 | [...] | |
156f5a78 | 503 | # echo 0 > tracing_enabled |
4a88d44a | 504 | # cat trace |
eb6d42ea SR |
505 | # tracer: irqsoff |
506 | # | |
f2d9c740 | 507 | irqsoff latency trace v1.1.5 on 2.6.26 |
eb6d42ea | 508 | -------------------------------------------------------------------- |
f2d9c740 | 509 | latency: 12 us, #3/3, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) |
eb6d42ea | 510 | ----------------- |
f2d9c740 | 511 | | task: bash-3730 (uid:0 nice:0 policy:0 rt_prio:0) |
eb6d42ea | 512 | ----------------- |
f2d9c740 SR |
513 | => started at: sys_setpgid |
514 | => ended at: sys_setpgid | |
eb6d42ea SR |
515 | |
516 | # _------=> CPU# | |
517 | # / _-----=> irqs-off | |
518 | # | / _----=> need-resched | |
519 | # || / _---=> hardirq/softirq | |
520 | # ||| / _--=> preempt-depth | |
521 | # |||| / | |
522 | # ||||| delay | |
523 | # cmd pid ||||| time | caller | |
524 | # \ / ||||| \ | / | |
f2d9c740 SR |
525 | bash-3730 1d... 0us : _write_lock_irq (sys_setpgid) |
526 | bash-3730 1d..1 1us+: _write_unlock_irq (sys_setpgid) | |
527 | bash-3730 1d..2 14us : trace_hardirqs_on (sys_setpgid) | |
eb6d42ea | 528 | |
eb6d42ea | 529 | |
f2d9c740 | 530 | Here we see that that we had a latency of 12 microsecs (which is |
5752674e IM |
531 | very good). The _write_lock_irq in sys_setpgid disabled |
532 | interrupts. The difference between the 12 and the displayed | |
533 | timestamp 14us occurred because the clock was incremented | |
534 | between the time of recording the max latency and the time of | |
535 | recording the function that had that latency. | |
eb6d42ea | 536 | |
f2d9c740 SR |
537 | Note the above example had ftrace_enabled not set. If we set the |
538 | ftrace_enabled, we get a much larger output: | |
eb6d42ea SR |
539 | |
540 | # tracer: irqsoff | |
541 | # | |
542 | irqsoff latency trace v1.1.5 on 2.6.26-rc8 | |
543 | -------------------------------------------------------------------- | |
544 | latency: 50 us, #101/101, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
545 | ----------------- | |
546 | | task: ls-4339 (uid:0 nice:0 policy:0 rt_prio:0) | |
547 | ----------------- | |
548 | => started at: __alloc_pages_internal | |
549 | => ended at: __alloc_pages_internal | |
550 | ||
551 | # _------=> CPU# | |
552 | # / _-----=> irqs-off | |
553 | # | / _----=> need-resched | |
554 | # || / _---=> hardirq/softirq | |
555 | # ||| / _--=> preempt-depth | |
556 | # |||| / | |
557 | # ||||| delay | |
558 | # cmd pid ||||| time | caller | |
559 | # \ / ||||| \ | / | |
560 | ls-4339 0...1 0us+: get_page_from_freelist (__alloc_pages_internal) | |
561 | ls-4339 0d..1 3us : rmqueue_bulk (get_page_from_freelist) | |
562 | ls-4339 0d..1 3us : _spin_lock (rmqueue_bulk) | |
563 | ls-4339 0d..1 4us : add_preempt_count (_spin_lock) | |
564 | ls-4339 0d..2 4us : __rmqueue (rmqueue_bulk) | |
565 | ls-4339 0d..2 5us : __rmqueue_smallest (__rmqueue) | |
566 | ls-4339 0d..2 5us : __mod_zone_page_state (__rmqueue_smallest) | |
567 | ls-4339 0d..2 6us : __rmqueue (rmqueue_bulk) | |
568 | ls-4339 0d..2 6us : __rmqueue_smallest (__rmqueue) | |
569 | ls-4339 0d..2 7us : __mod_zone_page_state (__rmqueue_smallest) | |
570 | ls-4339 0d..2 7us : __rmqueue (rmqueue_bulk) | |
571 | ls-4339 0d..2 8us : __rmqueue_smallest (__rmqueue) | |
572 | [...] | |
573 | ls-4339 0d..2 46us : __rmqueue_smallest (__rmqueue) | |
574 | ls-4339 0d..2 47us : __mod_zone_page_state (__rmqueue_smallest) | |
575 | ls-4339 0d..2 47us : __rmqueue (rmqueue_bulk) | |
576 | ls-4339 0d..2 48us : __rmqueue_smallest (__rmqueue) | |
577 | ls-4339 0d..2 48us : __mod_zone_page_state (__rmqueue_smallest) | |
578 | ls-4339 0d..2 49us : _spin_unlock (rmqueue_bulk) | |
579 | ls-4339 0d..2 49us : sub_preempt_count (_spin_unlock) | |
580 | ls-4339 0d..1 50us : get_page_from_freelist (__alloc_pages_internal) | |
581 | ls-4339 0d..2 51us : trace_hardirqs_on (__alloc_pages_internal) | |
582 | ||
583 | ||
eb6d42ea SR |
584 | |
585 | Here we traced a 50 microsecond latency. But we also see all the | |
5752674e IM |
586 | functions that were called during that time. Note that by |
587 | enabling function tracing, we incur an added overhead. This | |
588 | overhead may extend the latency times. But nevertheless, this | |
589 | trace has provided some very helpful debugging information. | |
eb6d42ea SR |
590 | |
591 | ||
592 | preemptoff | |
593 | ---------- | |
594 | ||
5752674e IM |
595 | When preemption is disabled, we may be able to receive |
596 | interrupts but the task cannot be preempted and a higher | |
597 | priority task must wait for preemption to be enabled again | |
598 | before it can preempt a lower priority task. | |
eb6d42ea | 599 | |
a41eebab | 600 | The preemptoff tracer traces the places that disable preemption. |
5752674e IM |
601 | Like the irqsoff tracer, it records the maximum latency for |
602 | which preemption was disabled. The control of preemptoff tracer | |
603 | is much like the irqsoff tracer. | |
eb6d42ea | 604 | |
156f5a78 | 605 | # echo preemptoff > current_tracer |
4a88d44a | 606 | # echo latency-format > trace_options |
156f5a78 GL |
607 | # echo 0 > tracing_max_latency |
608 | # echo 1 > tracing_enabled | |
eb6d42ea SR |
609 | # ls -ltr |
610 | [...] | |
156f5a78 | 611 | # echo 0 > tracing_enabled |
4a88d44a | 612 | # cat trace |
eb6d42ea SR |
613 | # tracer: preemptoff |
614 | # | |
615 | preemptoff latency trace v1.1.5 on 2.6.26-rc8 | |
616 | -------------------------------------------------------------------- | |
617 | latency: 29 us, #3/3, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
618 | ----------------- | |
619 | | task: sshd-4261 (uid:0 nice:0 policy:0 rt_prio:0) | |
620 | ----------------- | |
621 | => started at: do_IRQ | |
622 | => ended at: __do_softirq | |
623 | ||
624 | # _------=> CPU# | |
625 | # / _-----=> irqs-off | |
626 | # | / _----=> need-resched | |
627 | # || / _---=> hardirq/softirq | |
628 | # ||| / _--=> preempt-depth | |
629 | # |||| / | |
630 | # ||||| delay | |
631 | # cmd pid ||||| time | caller | |
632 | # \ / ||||| \ | / | |
633 | sshd-4261 0d.h. 0us+: irq_enter (do_IRQ) | |
634 | sshd-4261 0d.s. 29us : _local_bh_enable (__do_softirq) | |
635 | sshd-4261 0d.s1 30us : trace_preempt_on (__do_softirq) | |
636 | ||
637 | ||
5752674e IM |
638 | This has some more changes. Preemption was disabled when an |
639 | interrupt came in (notice the 'h'), and was enabled while doing | |
640 | a softirq. (notice the 's'). But we also see that interrupts | |
641 | have been disabled when entering the preempt off section and | |
642 | leaving it (the 'd'). We do not know if interrupts were enabled | |
643 | in the mean time. | |
eb6d42ea SR |
644 | |
645 | # tracer: preemptoff | |
646 | # | |
647 | preemptoff latency trace v1.1.5 on 2.6.26-rc8 | |
648 | -------------------------------------------------------------------- | |
649 | latency: 63 us, #87/87, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
650 | ----------------- | |
651 | | task: sshd-4261 (uid:0 nice:0 policy:0 rt_prio:0) | |
652 | ----------------- | |
653 | => started at: remove_wait_queue | |
654 | => ended at: __do_softirq | |
655 | ||
656 | # _------=> CPU# | |
657 | # / _-----=> irqs-off | |
658 | # | / _----=> need-resched | |
659 | # || / _---=> hardirq/softirq | |
660 | # ||| / _--=> preempt-depth | |
661 | # |||| / | |
662 | # ||||| delay | |
663 | # cmd pid ||||| time | caller | |
664 | # \ / ||||| \ | / | |
665 | sshd-4261 0d..1 0us : _spin_lock_irqsave (remove_wait_queue) | |
666 | sshd-4261 0d..1 1us : _spin_unlock_irqrestore (remove_wait_queue) | |
667 | sshd-4261 0d..1 2us : do_IRQ (common_interrupt) | |
668 | sshd-4261 0d..1 2us : irq_enter (do_IRQ) | |
669 | sshd-4261 0d..1 2us : idle_cpu (irq_enter) | |
670 | sshd-4261 0d..1 3us : add_preempt_count (irq_enter) | |
671 | sshd-4261 0d.h1 3us : idle_cpu (irq_enter) | |
672 | sshd-4261 0d.h. 4us : handle_fasteoi_irq (do_IRQ) | |
673 | [...] | |
674 | sshd-4261 0d.h. 12us : add_preempt_count (_spin_lock) | |
675 | sshd-4261 0d.h1 12us : ack_ioapic_quirk_irq (handle_fasteoi_irq) | |
676 | sshd-4261 0d.h1 13us : move_native_irq (ack_ioapic_quirk_irq) | |
677 | sshd-4261 0d.h1 13us : _spin_unlock (handle_fasteoi_irq) | |
678 | sshd-4261 0d.h1 14us : sub_preempt_count (_spin_unlock) | |
679 | sshd-4261 0d.h1 14us : irq_exit (do_IRQ) | |
680 | sshd-4261 0d.h1 15us : sub_preempt_count (irq_exit) | |
681 | sshd-4261 0d..2 15us : do_softirq (irq_exit) | |
682 | sshd-4261 0d... 15us : __do_softirq (do_softirq) | |
683 | sshd-4261 0d... 16us : __local_bh_disable (__do_softirq) | |
684 | sshd-4261 0d... 16us+: add_preempt_count (__local_bh_disable) | |
685 | sshd-4261 0d.s4 20us : add_preempt_count (__local_bh_disable) | |
686 | sshd-4261 0d.s4 21us : sub_preempt_count (local_bh_enable) | |
687 | sshd-4261 0d.s5 21us : sub_preempt_count (local_bh_enable) | |
688 | [...] | |
689 | sshd-4261 0d.s6 41us : add_preempt_count (__local_bh_disable) | |
690 | sshd-4261 0d.s6 42us : sub_preempt_count (local_bh_enable) | |
691 | sshd-4261 0d.s7 42us : sub_preempt_count (local_bh_enable) | |
692 | sshd-4261 0d.s5 43us : add_preempt_count (__local_bh_disable) | |
693 | sshd-4261 0d.s5 43us : sub_preempt_count (local_bh_enable_ip) | |
694 | sshd-4261 0d.s6 44us : sub_preempt_count (local_bh_enable_ip) | |
695 | sshd-4261 0d.s5 44us : add_preempt_count (__local_bh_disable) | |
696 | sshd-4261 0d.s5 45us : sub_preempt_count (local_bh_enable) | |
697 | [...] | |
698 | sshd-4261 0d.s. 63us : _local_bh_enable (__do_softirq) | |
699 | sshd-4261 0d.s1 64us : trace_preempt_on (__do_softirq) | |
700 | ||
701 | ||
5752674e IM |
702 | The above is an example of the preemptoff trace with |
703 | ftrace_enabled set. Here we see that interrupts were disabled | |
704 | the entire time. The irq_enter code lets us know that we entered | |
705 | an interrupt 'h'. Before that, the functions being traced still | |
706 | show that it is not in an interrupt, but we can see from the | |
707 | functions themselves that this is not the case. | |
eb6d42ea | 708 | |
5752674e IM |
709 | Notice that __do_softirq when called does not have a |
710 | preempt_count. It may seem that we missed a preempt enabling. | |
711 | What really happened is that the preempt count is held on the | |
712 | thread's stack and we switched to the softirq stack (4K stacks | |
713 | in effect). The code does not copy the preempt count, but | |
714 | because interrupts are disabled, we do not need to worry about | |
715 | it. Having a tracer like this is good for letting people know | |
716 | what really happens inside the kernel. | |
eb6d42ea SR |
717 | |
718 | ||
719 | preemptirqsoff | |
720 | -------------- | |
721 | ||
5752674e IM |
722 | Knowing the locations that have interrupts disabled or |
723 | preemption disabled for the longest times is helpful. But | |
724 | sometimes we would like to know when either preemption and/or | |
725 | interrupts are disabled. | |
eb6d42ea | 726 | |
f2d9c740 | 727 | Consider the following code: |
eb6d42ea SR |
728 | |
729 | local_irq_disable(); | |
730 | call_function_with_irqs_off(); | |
731 | preempt_disable(); | |
732 | call_function_with_irqs_and_preemption_off(); | |
733 | local_irq_enable(); | |
734 | call_function_with_preemption_off(); | |
735 | preempt_enable(); | |
736 | ||
737 | The irqsoff tracer will record the total length of | |
738 | call_function_with_irqs_off() and | |
739 | call_function_with_irqs_and_preemption_off(). | |
740 | ||
741 | The preemptoff tracer will record the total length of | |
742 | call_function_with_irqs_and_preemption_off() and | |
743 | call_function_with_preemption_off(). | |
744 | ||
5752674e IM |
745 | But neither will trace the time that interrupts and/or |
746 | preemption is disabled. This total time is the time that we can | |
747 | not schedule. To record this time, use the preemptirqsoff | |
748 | tracer. | |
eb6d42ea | 749 | |
5752674e IM |
750 | Again, using this trace is much like the irqsoff and preemptoff |
751 | tracers. | |
eb6d42ea | 752 | |
156f5a78 | 753 | # echo preemptirqsoff > current_tracer |
4a88d44a | 754 | # echo latency-format > trace_options |
156f5a78 GL |
755 | # echo 0 > tracing_max_latency |
756 | # echo 1 > tracing_enabled | |
eb6d42ea SR |
757 | # ls -ltr |
758 | [...] | |
156f5a78 | 759 | # echo 0 > tracing_enabled |
4a88d44a | 760 | # cat trace |
eb6d42ea SR |
761 | # tracer: preemptirqsoff |
762 | # | |
763 | preemptirqsoff latency trace v1.1.5 on 2.6.26-rc8 | |
764 | -------------------------------------------------------------------- | |
765 | latency: 293 us, #3/3, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
766 | ----------------- | |
767 | | task: ls-4860 (uid:0 nice:0 policy:0 rt_prio:0) | |
768 | ----------------- | |
769 | => started at: apic_timer_interrupt | |
770 | => ended at: __do_softirq | |
771 | ||
772 | # _------=> CPU# | |
773 | # / _-----=> irqs-off | |
774 | # | / _----=> need-resched | |
775 | # || / _---=> hardirq/softirq | |
776 | # ||| / _--=> preempt-depth | |
777 | # |||| / | |
778 | # ||||| delay | |
779 | # cmd pid ||||| time | caller | |
780 | # \ / ||||| \ | / | |
781 | ls-4860 0d... 0us!: trace_hardirqs_off_thunk (apic_timer_interrupt) | |
782 | ls-4860 0d.s. 294us : _local_bh_enable (__do_softirq) | |
783 | ls-4860 0d.s1 294us : trace_preempt_on (__do_softirq) | |
784 | ||
785 | ||
eb6d42ea SR |
786 | |
787 | The trace_hardirqs_off_thunk is called from assembly on x86 when | |
5752674e IM |
788 | interrupts are disabled in the assembly code. Without the |
789 | function tracing, we do not know if interrupts were enabled | |
790 | within the preemption points. We do see that it started with | |
791 | preemption enabled. | |
eb6d42ea SR |
792 | |
793 | Here is a trace with ftrace_enabled set: | |
794 | ||
795 | ||
796 | # tracer: preemptirqsoff | |
797 | # | |
798 | preemptirqsoff latency trace v1.1.5 on 2.6.26-rc8 | |
799 | -------------------------------------------------------------------- | |
800 | latency: 105 us, #183/183, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
801 | ----------------- | |
802 | | task: sshd-4261 (uid:0 nice:0 policy:0 rt_prio:0) | |
803 | ----------------- | |
804 | => started at: write_chan | |
805 | => ended at: __do_softirq | |
806 | ||
807 | # _------=> CPU# | |
808 | # / _-----=> irqs-off | |
809 | # | / _----=> need-resched | |
810 | # || / _---=> hardirq/softirq | |
811 | # ||| / _--=> preempt-depth | |
812 | # |||| / | |
813 | # ||||| delay | |
814 | # cmd pid ||||| time | caller | |
815 | # \ / ||||| \ | / | |
816 | ls-4473 0.N.. 0us : preempt_schedule (write_chan) | |
817 | ls-4473 0dN.1 1us : _spin_lock (schedule) | |
818 | ls-4473 0dN.1 2us : add_preempt_count (_spin_lock) | |
819 | ls-4473 0d..2 2us : put_prev_task_fair (schedule) | |
820 | [...] | |
821 | ls-4473 0d..2 13us : set_normalized_timespec (ktime_get_ts) | |
822 | ls-4473 0d..2 13us : __switch_to (schedule) | |
823 | sshd-4261 0d..2 14us : finish_task_switch (schedule) | |
824 | sshd-4261 0d..2 14us : _spin_unlock_irq (finish_task_switch) | |
825 | sshd-4261 0d..1 15us : add_preempt_count (_spin_lock_irqsave) | |
826 | sshd-4261 0d..2 16us : _spin_unlock_irqrestore (hrtick_set) | |
827 | sshd-4261 0d..2 16us : do_IRQ (common_interrupt) | |
828 | sshd-4261 0d..2 17us : irq_enter (do_IRQ) | |
829 | sshd-4261 0d..2 17us : idle_cpu (irq_enter) | |
830 | sshd-4261 0d..2 18us : add_preempt_count (irq_enter) | |
831 | sshd-4261 0d.h2 18us : idle_cpu (irq_enter) | |
832 | sshd-4261 0d.h. 18us : handle_fasteoi_irq (do_IRQ) | |
833 | sshd-4261 0d.h. 19us : _spin_lock (handle_fasteoi_irq) | |
834 | sshd-4261 0d.h. 19us : add_preempt_count (_spin_lock) | |
835 | sshd-4261 0d.h1 20us : _spin_unlock (handle_fasteoi_irq) | |
836 | sshd-4261 0d.h1 20us : sub_preempt_count (_spin_unlock) | |
837 | [...] | |
838 | sshd-4261 0d.h1 28us : _spin_unlock (handle_fasteoi_irq) | |
839 | sshd-4261 0d.h1 29us : sub_preempt_count (_spin_unlock) | |
840 | sshd-4261 0d.h2 29us : irq_exit (do_IRQ) | |
841 | sshd-4261 0d.h2 29us : sub_preempt_count (irq_exit) | |
842 | sshd-4261 0d..3 30us : do_softirq (irq_exit) | |
843 | sshd-4261 0d... 30us : __do_softirq (do_softirq) | |
844 | sshd-4261 0d... 31us : __local_bh_disable (__do_softirq) | |
845 | sshd-4261 0d... 31us+: add_preempt_count (__local_bh_disable) | |
846 | sshd-4261 0d.s4 34us : add_preempt_count (__local_bh_disable) | |
847 | [...] | |
848 | sshd-4261 0d.s3 43us : sub_preempt_count (local_bh_enable_ip) | |
849 | sshd-4261 0d.s4 44us : sub_preempt_count (local_bh_enable_ip) | |
850 | sshd-4261 0d.s3 44us : smp_apic_timer_interrupt (apic_timer_interrupt) | |
851 | sshd-4261 0d.s3 45us : irq_enter (smp_apic_timer_interrupt) | |
852 | sshd-4261 0d.s3 45us : idle_cpu (irq_enter) | |
853 | sshd-4261 0d.s3 46us : add_preempt_count (irq_enter) | |
854 | sshd-4261 0d.H3 46us : idle_cpu (irq_enter) | |
855 | sshd-4261 0d.H3 47us : hrtimer_interrupt (smp_apic_timer_interrupt) | |
856 | sshd-4261 0d.H3 47us : ktime_get (hrtimer_interrupt) | |
857 | [...] | |
858 | sshd-4261 0d.H3 81us : tick_program_event (hrtimer_interrupt) | |
859 | sshd-4261 0d.H3 82us : ktime_get (tick_program_event) | |
860 | sshd-4261 0d.H3 82us : ktime_get_ts (ktime_get) | |
861 | sshd-4261 0d.H3 83us : getnstimeofday (ktime_get_ts) | |
862 | sshd-4261 0d.H3 83us : set_normalized_timespec (ktime_get_ts) | |
863 | sshd-4261 0d.H3 84us : clockevents_program_event (tick_program_event) | |
864 | sshd-4261 0d.H3 84us : lapic_next_event (clockevents_program_event) | |
865 | sshd-4261 0d.H3 85us : irq_exit (smp_apic_timer_interrupt) | |
866 | sshd-4261 0d.H3 85us : sub_preempt_count (irq_exit) | |
867 | sshd-4261 0d.s4 86us : sub_preempt_count (irq_exit) | |
868 | sshd-4261 0d.s3 86us : add_preempt_count (__local_bh_disable) | |
869 | [...] | |
870 | sshd-4261 0d.s1 98us : sub_preempt_count (net_rx_action) | |
871 | sshd-4261 0d.s. 99us : add_preempt_count (_spin_lock_irq) | |
872 | sshd-4261 0d.s1 99us+: _spin_unlock_irq (run_timer_softirq) | |
873 | sshd-4261 0d.s. 104us : _local_bh_enable (__do_softirq) | |
874 | sshd-4261 0d.s. 104us : sub_preempt_count (_local_bh_enable) | |
875 | sshd-4261 0d.s. 105us : _local_bh_enable (__do_softirq) | |
876 | sshd-4261 0d.s1 105us : trace_preempt_on (__do_softirq) | |
877 | ||
878 | ||
5752674e IM |
879 | This is a very interesting trace. It started with the preemption |
880 | of the ls task. We see that the task had the "need_resched" bit | |
881 | set via the 'N' in the trace. Interrupts were disabled before | |
882 | the spin_lock at the beginning of the trace. We see that a | |
883 | schedule took place to run sshd. When the interrupts were | |
884 | enabled, we took an interrupt. On return from the interrupt | |
885 | handler, the softirq ran. We took another interrupt while | |
886 | running the softirq as we see from the capital 'H'. | |
eb6d42ea SR |
887 | |
888 | ||
889 | wakeup | |
890 | ------ | |
891 | ||
5752674e IM |
892 | In a Real-Time environment it is very important to know the |
893 | wakeup time it takes for the highest priority task that is woken | |
894 | up to the time that it executes. This is also known as "schedule | |
895 | latency". I stress the point that this is about RT tasks. It is | |
896 | also important to know the scheduling latency of non-RT tasks, | |
897 | but the average schedule latency is better for non-RT tasks. | |
898 | Tools like LatencyTop are more appropriate for such | |
899 | measurements. | |
eb6d42ea | 900 | |
a41eebab | 901 | Real-Time environments are interested in the worst case latency. |
5752674e IM |
902 | That is the longest latency it takes for something to happen, |
903 | and not the average. We can have a very fast scheduler that may | |
904 | only have a large latency once in a while, but that would not | |
905 | work well with Real-Time tasks. The wakeup tracer was designed | |
906 | to record the worst case wakeups of RT tasks. Non-RT tasks are | |
907 | not recorded because the tracer only records one worst case and | |
908 | tracing non-RT tasks that are unpredictable will overwrite the | |
909 | worst case latency of RT tasks. | |
910 | ||
911 | Since this tracer only deals with RT tasks, we will run this | |
912 | slightly differently than we did with the previous tracers. | |
913 | Instead of performing an 'ls', we will run 'sleep 1' under | |
914 | 'chrt' which changes the priority of the task. | |
eb6d42ea | 915 | |
156f5a78 | 916 | # echo wakeup > current_tracer |
4a88d44a | 917 | # echo latency-format > trace_options |
156f5a78 GL |
918 | # echo 0 > tracing_max_latency |
919 | # echo 1 > tracing_enabled | |
eb6d42ea | 920 | # chrt -f 5 sleep 1 |
156f5a78 | 921 | # echo 0 > tracing_enabled |
4a88d44a | 922 | # cat trace |
eb6d42ea SR |
923 | # tracer: wakeup |
924 | # | |
925 | wakeup latency trace v1.1.5 on 2.6.26-rc8 | |
926 | -------------------------------------------------------------------- | |
927 | latency: 4 us, #2/2, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
928 | ----------------- | |
929 | | task: sleep-4901 (uid:0 nice:0 policy:1 rt_prio:5) | |
930 | ----------------- | |
931 | ||
932 | # _------=> CPU# | |
933 | # / _-----=> irqs-off | |
934 | # | / _----=> need-resched | |
935 | # || / _---=> hardirq/softirq | |
936 | # ||| / _--=> preempt-depth | |
937 | # |||| / | |
938 | # ||||| delay | |
939 | # cmd pid ||||| time | caller | |
940 | # \ / ||||| \ | / | |
941 | <idle>-0 1d.h4 0us+: try_to_wake_up (wake_up_process) | |
942 | <idle>-0 1d..4 4us : schedule (cpu_idle) | |
943 | ||
944 | ||
5752674e IM |
945 | Running this on an idle system, we see that it only took 4 |
946 | microseconds to perform the task switch. Note, since the trace | |
947 | marker in the schedule is before the actual "switch", we stop | |
948 | the tracing when the recorded task is about to schedule in. This | |
949 | may change if we add a new marker at the end of the scheduler. | |
eb6d42ea | 950 | |
5752674e IM |
951 | Notice that the recorded task is 'sleep' with the PID of 4901 |
952 | and it has an rt_prio of 5. This priority is user-space priority | |
953 | and not the internal kernel priority. The policy is 1 for | |
954 | SCHED_FIFO and 2 for SCHED_RR. | |
eb6d42ea SR |
955 | |
956 | Doing the same with chrt -r 5 and ftrace_enabled set. | |
957 | ||
958 | # tracer: wakeup | |
959 | # | |
960 | wakeup latency trace v1.1.5 on 2.6.26-rc8 | |
961 | -------------------------------------------------------------------- | |
962 | latency: 50 us, #60/60, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:2) | |
963 | ----------------- | |
964 | | task: sleep-4068 (uid:0 nice:0 policy:2 rt_prio:5) | |
965 | ----------------- | |
966 | ||
967 | # _------=> CPU# | |
968 | # / _-----=> irqs-off | |
969 | # | / _----=> need-resched | |
970 | # || / _---=> hardirq/softirq | |
971 | # ||| / _--=> preempt-depth | |
972 | # |||| / | |
973 | # ||||| delay | |
974 | # cmd pid ||||| time | caller | |
975 | # \ / ||||| \ | / | |
976 | ksoftirq-7 1d.H3 0us : try_to_wake_up (wake_up_process) | |
977 | ksoftirq-7 1d.H4 1us : sub_preempt_count (marker_probe_cb) | |
978 | ksoftirq-7 1d.H3 2us : check_preempt_wakeup (try_to_wake_up) | |
979 | ksoftirq-7 1d.H3 3us : update_curr (check_preempt_wakeup) | |
980 | ksoftirq-7 1d.H3 4us : calc_delta_mine (update_curr) | |
981 | ksoftirq-7 1d.H3 5us : __resched_task (check_preempt_wakeup) | |
982 | ksoftirq-7 1d.H3 6us : task_wake_up_rt (try_to_wake_up) | |
983 | ksoftirq-7 1d.H3 7us : _spin_unlock_irqrestore (try_to_wake_up) | |
984 | [...] | |
985 | ksoftirq-7 1d.H2 17us : irq_exit (smp_apic_timer_interrupt) | |
986 | ksoftirq-7 1d.H2 18us : sub_preempt_count (irq_exit) | |
987 | ksoftirq-7 1d.s3 19us : sub_preempt_count (irq_exit) | |
988 | ksoftirq-7 1..s2 20us : rcu_process_callbacks (__do_softirq) | |
989 | [...] | |
990 | ksoftirq-7 1..s2 26us : __rcu_process_callbacks (rcu_process_callbacks) | |
991 | ksoftirq-7 1d.s2 27us : _local_bh_enable (__do_softirq) | |
992 | ksoftirq-7 1d.s2 28us : sub_preempt_count (_local_bh_enable) | |
993 | ksoftirq-7 1.N.3 29us : sub_preempt_count (ksoftirqd) | |
994 | ksoftirq-7 1.N.2 30us : _cond_resched (ksoftirqd) | |
995 | ksoftirq-7 1.N.2 31us : __cond_resched (_cond_resched) | |
996 | ksoftirq-7 1.N.2 32us : add_preempt_count (__cond_resched) | |
997 | ksoftirq-7 1.N.2 33us : schedule (__cond_resched) | |
998 | ksoftirq-7 1.N.2 33us : add_preempt_count (schedule) | |
999 | ksoftirq-7 1.N.3 34us : hrtick_clear (schedule) | |
1000 | ksoftirq-7 1dN.3 35us : _spin_lock (schedule) | |
1001 | ksoftirq-7 1dN.3 36us : add_preempt_count (_spin_lock) | |
1002 | ksoftirq-7 1d..4 37us : put_prev_task_fair (schedule) | |
1003 | ksoftirq-7 1d..4 38us : update_curr (put_prev_task_fair) | |
1004 | [...] | |
1005 | ksoftirq-7 1d..5 47us : _spin_trylock (tracing_record_cmdline) | |
1006 | ksoftirq-7 1d..5 48us : add_preempt_count (_spin_trylock) | |
1007 | ksoftirq-7 1d..6 49us : _spin_unlock (tracing_record_cmdline) | |
1008 | ksoftirq-7 1d..6 49us : sub_preempt_count (_spin_unlock) | |
1009 | ksoftirq-7 1d..4 50us : schedule (__cond_resched) | |
1010 | ||
5752674e IM |
1011 | The interrupt went off while running ksoftirqd. This task runs |
1012 | at SCHED_OTHER. Why did not we see the 'N' set early? This may | |
1013 | be a harmless bug with x86_32 and 4K stacks. On x86_32 with 4K | |
1014 | stacks configured, the interrupt and softirq run with their own | |
1015 | stack. Some information is held on the top of the task's stack | |
1016 | (need_resched and preempt_count are both stored there). The | |
1017 | setting of the NEED_RESCHED bit is done directly to the task's | |
1018 | stack, but the reading of the NEED_RESCHED is done by looking at | |
1019 | the current stack, which in this case is the stack for the hard | |
1020 | interrupt. This hides the fact that NEED_RESCHED has been set. | |
1021 | We do not see the 'N' until we switch back to the task's | |
a41eebab | 1022 | assigned stack. |
eb6d42ea | 1023 | |
9b803c0f SR |
1024 | function |
1025 | -------- | |
eb6d42ea | 1026 | |
9b803c0f | 1027 | This tracer is the function tracer. Enabling the function tracer |
5752674e IM |
1028 | can be done from the debug file system. Make sure the |
1029 | ftrace_enabled is set; otherwise this tracer is a nop. | |
eb6d42ea SR |
1030 | |
1031 | # sysctl kernel.ftrace_enabled=1 | |
156f5a78 GL |
1032 | # echo function > current_tracer |
1033 | # echo 1 > tracing_enabled | |
eb6d42ea | 1034 | # usleep 1 |
156f5a78 GL |
1035 | # echo 0 > tracing_enabled |
1036 | # cat trace | |
9b803c0f | 1037 | # tracer: function |
eb6d42ea SR |
1038 | # |
1039 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
1040 | # | | | | | | |
1041 | bash-4003 [00] 123.638713: finish_task_switch <-schedule | |
1042 | bash-4003 [00] 123.638714: _spin_unlock_irq <-finish_task_switch | |
1043 | bash-4003 [00] 123.638714: sub_preempt_count <-_spin_unlock_irq | |
1044 | bash-4003 [00] 123.638715: hrtick_set <-schedule | |
1045 | bash-4003 [00] 123.638715: _spin_lock_irqsave <-hrtick_set | |
1046 | bash-4003 [00] 123.638716: add_preempt_count <-_spin_lock_irqsave | |
1047 | bash-4003 [00] 123.638716: _spin_unlock_irqrestore <-hrtick_set | |
1048 | bash-4003 [00] 123.638717: sub_preempt_count <-_spin_unlock_irqrestore | |
1049 | bash-4003 [00] 123.638717: hrtick_clear <-hrtick_set | |
1050 | bash-4003 [00] 123.638718: sub_preempt_count <-schedule | |
1051 | bash-4003 [00] 123.638718: sub_preempt_count <-preempt_schedule | |
1052 | bash-4003 [00] 123.638719: wait_for_completion <-__stop_machine_run | |
1053 | bash-4003 [00] 123.638719: wait_for_common <-wait_for_completion | |
1054 | bash-4003 [00] 123.638720: _spin_lock_irq <-wait_for_common | |
1055 | bash-4003 [00] 123.638720: add_preempt_count <-_spin_lock_irq | |
1056 | [...] | |
1057 | ||
1058 | ||
5752674e IM |
1059 | Note: function tracer uses ring buffers to store the above |
1060 | entries. The newest data may overwrite the oldest data. | |
1061 | Sometimes using echo to stop the trace is not sufficient because | |
1062 | the tracing could have overwritten the data that you wanted to | |
1063 | record. For this reason, it is sometimes better to disable | |
1064 | tracing directly from a program. This allows you to stop the | |
1065 | tracing at the point that you hit the part that you are | |
1066 | interested in. To disable the tracing directly from a C program, | |
1067 | something like following code snippet can be used: | |
eb6d42ea SR |
1068 | |
1069 | int trace_fd; | |
1070 | [...] | |
1071 | int main(int argc, char *argv[]) { | |
1072 | [...] | |
156f5a78 | 1073 | trace_fd = open(tracing_file("tracing_enabled"), O_WRONLY); |
eb6d42ea SR |
1074 | [...] |
1075 | if (condition_hit()) { | |
f2d9c740 | 1076 | write(trace_fd, "0", 1); |
eb6d42ea SR |
1077 | } |
1078 | [...] | |
1079 | } | |
1080 | ||
df4fc315 SR |
1081 | |
1082 | Single thread tracing | |
1083 | --------------------- | |
1084 | ||
156f5a78 | 1085 | By writing into set_ftrace_pid you can trace a |
df4fc315 SR |
1086 | single thread. For example: |
1087 | ||
156f5a78 | 1088 | # cat set_ftrace_pid |
df4fc315 | 1089 | no pid |
156f5a78 GL |
1090 | # echo 3111 > set_ftrace_pid |
1091 | # cat set_ftrace_pid | |
df4fc315 | 1092 | 3111 |
156f5a78 GL |
1093 | # echo function > current_tracer |
1094 | # cat trace | head | |
df4fc315 SR |
1095 | # tracer: function |
1096 | # | |
1097 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
1098 | # | | | | | | |
1099 | yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return | |
1100 | yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range | |
1101 | yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel | |
1102 | yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel | |
1103 | yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll | |
1104 | yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll | |
156f5a78 GL |
1105 | # echo -1 > set_ftrace_pid |
1106 | # cat trace |head | |
df4fc315 SR |
1107 | # tracer: function |
1108 | # | |
1109 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
1110 | # | | | | | | |
1111 | ##### CPU 3 buffer started #### | |
1112 | yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait | |
1113 | yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry | |
1114 | yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry | |
1115 | yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit | |
1116 | yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit | |
1117 | ||
1118 | If you want to trace a function when executing, you could use | |
1119 | something like this simple program: | |
1120 | ||
1121 | #include <stdio.h> | |
1122 | #include <stdlib.h> | |
1123 | #include <sys/types.h> | |
1124 | #include <sys/stat.h> | |
1125 | #include <fcntl.h> | |
1126 | #include <unistd.h> | |
67b394f7 | 1127 | #include <string.h> |
df4fc315 | 1128 | |
156f5a78 GL |
1129 | #define _STR(x) #x |
1130 | #define STR(x) _STR(x) | |
1131 | #define MAX_PATH 256 | |
1132 | ||
1133 | const char *find_debugfs(void) | |
1134 | { | |
1135 | static char debugfs[MAX_PATH+1]; | |
1136 | static int debugfs_found; | |
1137 | char type[100]; | |
1138 | FILE *fp; | |
1139 | ||
1140 | if (debugfs_found) | |
1141 | return debugfs; | |
1142 | ||
1143 | if ((fp = fopen("/proc/mounts","r")) == NULL) { | |
1144 | perror("/proc/mounts"); | |
1145 | return NULL; | |
1146 | } | |
1147 | ||
1148 | while (fscanf(fp, "%*s %" | |
1149 | STR(MAX_PATH) | |
1150 | "s %99s %*s %*d %*d\n", | |
1151 | debugfs, type) == 2) { | |
1152 | if (strcmp(type, "debugfs") == 0) | |
1153 | break; | |
1154 | } | |
1155 | fclose(fp); | |
1156 | ||
1157 | if (strcmp(type, "debugfs") != 0) { | |
1158 | fprintf(stderr, "debugfs not mounted"); | |
1159 | return NULL; | |
1160 | } | |
1161 | ||
67b394f7 | 1162 | strcat(debugfs, "/tracing/"); |
156f5a78 GL |
1163 | debugfs_found = 1; |
1164 | ||
1165 | return debugfs; | |
1166 | } | |
1167 | ||
1168 | const char *tracing_file(const char *file_name) | |
1169 | { | |
1170 | static char trace_file[MAX_PATH+1]; | |
1171 | snprintf(trace_file, MAX_PATH, "%s/%s", find_debugfs(), file_name); | |
1172 | return trace_file; | |
1173 | } | |
1174 | ||
df4fc315 SR |
1175 | int main (int argc, char **argv) |
1176 | { | |
1177 | if (argc < 1) | |
1178 | exit(-1); | |
1179 | ||
1180 | if (fork() > 0) { | |
1181 | int fd, ffd; | |
1182 | char line[64]; | |
1183 | int s; | |
1184 | ||
156f5a78 | 1185 | ffd = open(tracing_file("current_tracer"), O_WRONLY); |
df4fc315 SR |
1186 | if (ffd < 0) |
1187 | exit(-1); | |
1188 | write(ffd, "nop", 3); | |
1189 | ||
156f5a78 | 1190 | fd = open(tracing_file("set_ftrace_pid"), O_WRONLY); |
df4fc315 SR |
1191 | s = sprintf(line, "%d\n", getpid()); |
1192 | write(fd, line, s); | |
1193 | ||
1194 | write(ffd, "function", 8); | |
1195 | ||
1196 | close(fd); | |
1197 | close(ffd); | |
1198 | ||
1199 | execvp(argv[1], argv+1); | |
1200 | } | |
1201 | ||
1202 | return 0; | |
1203 | } | |
1204 | ||
e2ea5399 MM |
1205 | |
1206 | hw-branch-tracer (x86 only) | |
1207 | --------------------------- | |
1208 | ||
1209 | This tracer uses the x86 last branch tracing hardware feature to | |
1210 | collect a branch trace on all cpus with relatively low overhead. | |
1211 | ||
1212 | The tracer uses a fixed-size circular buffer per cpu and only | |
1213 | traces ring 0 branches. The trace file dumps that buffer in the | |
1214 | following format: | |
1215 | ||
1216 | # tracer: hw-branch-tracer | |
1217 | # | |
1218 | # CPU# TO <- FROM | |
1219 | 0 scheduler_tick+0xb5/0x1bf <- task_tick_idle+0x5/0x6 | |
1220 | 2 run_posix_cpu_timers+0x2b/0x72a <- run_posix_cpu_timers+0x25/0x72a | |
1221 | 0 scheduler_tick+0x139/0x1bf <- scheduler_tick+0xed/0x1bf | |
1222 | 0 scheduler_tick+0x17c/0x1bf <- scheduler_tick+0x148/0x1bf | |
1223 | 2 run_posix_cpu_timers+0x9e/0x72a <- run_posix_cpu_timers+0x5e/0x72a | |
1224 | 0 scheduler_tick+0x1b6/0x1bf <- scheduler_tick+0x1aa/0x1bf | |
1225 | ||
1226 | ||
5752674e IM |
1227 | The tracer may be used to dump the trace for the oops'ing cpu on |
1228 | a kernel oops into the system log. To enable this, | |
1229 | ftrace_dump_on_oops must be set. To set ftrace_dump_on_oops, one | |
1230 | can either use the sysctl function or set it via the proc system | |
1231 | interface. | |
e2ea5399 | 1232 | |
cecbca96 | 1233 | sysctl kernel.ftrace_dump_on_oops=n |
e2ea5399 MM |
1234 | |
1235 | or | |
1236 | ||
cecbca96 | 1237 | echo n > /proc/sys/kernel/ftrace_dump_on_oops |
e2ea5399 | 1238 | |
cecbca96 FW |
1239 | If n = 1, ftrace will dump buffers of all CPUs, if n = 2 ftrace will |
1240 | only dump the buffer of the CPU that triggered the oops. | |
e2ea5399 | 1241 | |
5752674e IM |
1242 | Here's an example of such a dump after a null pointer |
1243 | dereference in a kernel module: | |
e2ea5399 MM |
1244 | |
1245 | [57848.105921] BUG: unable to handle kernel NULL pointer dereference at 0000000000000000 | |
1246 | [57848.106019] IP: [<ffffffffa0000006>] open+0x6/0x14 [oops] | |
1247 | [57848.106019] PGD 2354e9067 PUD 2375e7067 PMD 0 | |
1248 | [57848.106019] Oops: 0002 [#1] SMP | |
1249 | [57848.106019] last sysfs file: /sys/devices/pci0000:00/0000:00:1e.0/0000:20:05.0/local_cpus | |
1250 | [57848.106019] Dumping ftrace buffer: | |
1251 | [57848.106019] --------------------------------- | |
1252 | [...] | |
1253 | [57848.106019] 0 chrdev_open+0xe6/0x165 <- cdev_put+0x23/0x24 | |
1254 | [57848.106019] 0 chrdev_open+0x117/0x165 <- chrdev_open+0xfa/0x165 | |
1255 | [57848.106019] 0 chrdev_open+0x120/0x165 <- chrdev_open+0x11c/0x165 | |
1256 | [57848.106019] 0 chrdev_open+0x134/0x165 <- chrdev_open+0x12b/0x165 | |
1257 | [57848.106019] 0 open+0x0/0x14 [oops] <- chrdev_open+0x144/0x165 | |
1258 | [57848.106019] 0 page_fault+0x0/0x30 <- open+0x6/0x14 [oops] | |
1259 | [57848.106019] 0 error_entry+0x0/0x5b <- page_fault+0x4/0x30 | |
1260 | [57848.106019] 0 error_kernelspace+0x0/0x31 <- error_entry+0x59/0x5b | |
1261 | [57848.106019] 0 error_sti+0x0/0x1 <- error_kernelspace+0x2d/0x31 | |
1262 | [57848.106019] 0 page_fault+0x9/0x30 <- error_sti+0x0/0x1 | |
1263 | [57848.106019] 0 do_page_fault+0x0/0x881 <- page_fault+0x1a/0x30 | |
1264 | [...] | |
1265 | [57848.106019] 0 do_page_fault+0x66b/0x881 <- is_prefetch+0x1ee/0x1f2 | |
1266 | [57848.106019] 0 do_page_fault+0x6e0/0x881 <- do_page_fault+0x67a/0x881 | |
1267 | [57848.106019] 0 oops_begin+0x0/0x96 <- do_page_fault+0x6e0/0x881 | |
1268 | [57848.106019] 0 trace_hw_branch_oops+0x0/0x2d <- oops_begin+0x9/0x96 | |
1269 | [...] | |
1270 | [57848.106019] 0 ds_suspend_bts+0x2a/0xe3 <- ds_suspend_bts+0x1a/0xe3 | |
1271 | [57848.106019] --------------------------------- | |
1272 | [57848.106019] CPU 0 | |
1273 | [57848.106019] Modules linked in: oops | |
1274 | [57848.106019] Pid: 5542, comm: cat Tainted: G W 2.6.28 #23 | |
1275 | [57848.106019] RIP: 0010:[<ffffffffa0000006>] [<ffffffffa0000006>] open+0x6/0x14 [oops] | |
1276 | [57848.106019] RSP: 0018:ffff880235457d48 EFLAGS: 00010246 | |
1277 | [...] | |
1278 | ||
1279 | ||
985ec20a FW |
1280 | function graph tracer |
1281 | --------------------------- | |
1282 | ||
5752674e IM |
1283 | This tracer is similar to the function tracer except that it |
1284 | probes a function on its entry and its exit. This is done by | |
1285 | using a dynamically allocated stack of return addresses in each | |
1286 | task_struct. On function entry the tracer overwrites the return | |
1287 | address of each function traced to set a custom probe. Thus the | |
1288 | original return address is stored on the stack of return address | |
1289 | in the task_struct. | |
985ec20a | 1290 | |
5752674e IM |
1291 | Probing on both ends of a function leads to special features |
1292 | such as: | |
985ec20a | 1293 | |
5752674e IM |
1294 | - measure of a function's time execution |
1295 | - having a reliable call stack to draw function calls graph | |
985ec20a FW |
1296 | |
1297 | This tracer is useful in several situations: | |
1298 | ||
5752674e IM |
1299 | - you want to find the reason of a strange kernel behavior and |
1300 | need to see what happens in detail on any areas (or specific | |
1301 | ones). | |
1302 | ||
1303 | - you are experiencing weird latencies but it's difficult to | |
1304 | find its origin. | |
1305 | ||
1306 | - you want to find quickly which path is taken by a specific | |
1307 | function | |
1308 | ||
1309 | - you just want to peek inside a working kernel and want to see | |
1310 | what happens there. | |
985ec20a FW |
1311 | |
1312 | # tracer: function_graph | |
1313 | # | |
1314 | # CPU DURATION FUNCTION CALLS | |
1315 | # | | | | | | | | |
1316 | ||
1317 | 0) | sys_open() { | |
1318 | 0) | do_sys_open() { | |
1319 | 0) | getname() { | |
1320 | 0) | kmem_cache_alloc() { | |
1321 | 0) 1.382 us | __might_sleep(); | |
1322 | 0) 2.478 us | } | |
1323 | 0) | strncpy_from_user() { | |
1324 | 0) | might_fault() { | |
1325 | 0) 1.389 us | __might_sleep(); | |
1326 | 0) 2.553 us | } | |
1327 | 0) 3.807 us | } | |
1328 | 0) 7.876 us | } | |
1329 | 0) | alloc_fd() { | |
1330 | 0) 0.668 us | _spin_lock(); | |
1331 | 0) 0.570 us | expand_files(); | |
1332 | 0) 0.586 us | _spin_unlock(); | |
1333 | ||
1334 | ||
5752674e IM |
1335 | There are several columns that can be dynamically |
1336 | enabled/disabled. You can use every combination of options you | |
1337 | want, depending on your needs. | |
985ec20a | 1338 | |
5752674e IM |
1339 | - The cpu number on which the function executed is default |
1340 | enabled. It is sometimes better to only trace one cpu (see | |
1341 | tracing_cpu_mask file) or you might sometimes see unordered | |
1342 | function calls while cpu tracing switch. | |
985ec20a | 1343 | |
156f5a78 GL |
1344 | hide: echo nofuncgraph-cpu > trace_options |
1345 | show: echo funcgraph-cpu > trace_options | |
985ec20a | 1346 | |
5752674e IM |
1347 | - The duration (function's time of execution) is displayed on |
1348 | the closing bracket line of a function or on the same line | |
1349 | than the current function in case of a leaf one. It is default | |
1350 | enabled. | |
985ec20a | 1351 | |
156f5a78 GL |
1352 | hide: echo nofuncgraph-duration > trace_options |
1353 | show: echo funcgraph-duration > trace_options | |
985ec20a | 1354 | |
5752674e IM |
1355 | - The overhead field precedes the duration field in case of |
1356 | reached duration thresholds. | |
985ec20a | 1357 | |
156f5a78 GL |
1358 | hide: echo nofuncgraph-overhead > trace_options |
1359 | show: echo funcgraph-overhead > trace_options | |
985ec20a FW |
1360 | depends on: funcgraph-duration |
1361 | ||
1362 | ie: | |
1363 | ||
1364 | 0) | up_write() { | |
1365 | 0) 0.646 us | _spin_lock_irqsave(); | |
1366 | 0) 0.684 us | _spin_unlock_irqrestore(); | |
1367 | 0) 3.123 us | } | |
1368 | 0) 0.548 us | fput(); | |
1369 | 0) + 58.628 us | } | |
1370 | ||
1371 | [...] | |
1372 | ||
1373 | 0) | putname() { | |
1374 | 0) | kmem_cache_free() { | |
1375 | 0) 0.518 us | __phys_addr(); | |
1376 | 0) 1.757 us | } | |
1377 | 0) 2.861 us | } | |
1378 | 0) ! 115.305 us | } | |
1379 | 0) ! 116.402 us | } | |
1380 | ||
1381 | + means that the function exceeded 10 usecs. | |
1382 | ! means that the function exceeded 100 usecs. | |
1383 | ||
1384 | ||
5752674e IM |
1385 | - The task/pid field displays the thread cmdline and pid which |
1386 | executed the function. It is default disabled. | |
985ec20a | 1387 | |
156f5a78 GL |
1388 | hide: echo nofuncgraph-proc > trace_options |
1389 | show: echo funcgraph-proc > trace_options | |
985ec20a FW |
1390 | |
1391 | ie: | |
1392 | ||
1393 | # tracer: function_graph | |
1394 | # | |
1395 | # CPU TASK/PID DURATION FUNCTION CALLS | |
1396 | # | | | | | | | | | | |
1397 | 0) sh-4802 | | d_free() { | |
1398 | 0) sh-4802 | | call_rcu() { | |
1399 | 0) sh-4802 | | __call_rcu() { | |
1400 | 0) sh-4802 | 0.616 us | rcu_process_gp_end(); | |
1401 | 0) sh-4802 | 0.586 us | check_for_new_grace_period(); | |
1402 | 0) sh-4802 | 2.899 us | } | |
1403 | 0) sh-4802 | 4.040 us | } | |
1404 | 0) sh-4802 | 5.151 us | } | |
1405 | 0) sh-4802 | + 49.370 us | } | |
1406 | ||
1407 | ||
5752674e IM |
1408 | - The absolute time field is an absolute timestamp given by the |
1409 | system clock since it started. A snapshot of this time is | |
1410 | given on each entry/exit of functions | |
985ec20a | 1411 | |
156f5a78 GL |
1412 | hide: echo nofuncgraph-abstime > trace_options |
1413 | show: echo funcgraph-abstime > trace_options | |
985ec20a FW |
1414 | |
1415 | ie: | |
1416 | ||
1417 | # | |
1418 | # TIME CPU DURATION FUNCTION CALLS | |
1419 | # | | | | | | | | | |
1420 | 360.774522 | 1) 0.541 us | } | |
1421 | 360.774522 | 1) 4.663 us | } | |
1422 | 360.774523 | 1) 0.541 us | __wake_up_bit(); | |
1423 | 360.774524 | 1) 6.796 us | } | |
1424 | 360.774524 | 1) 7.952 us | } | |
1425 | 360.774525 | 1) 9.063 us | } | |
1426 | 360.774525 | 1) 0.615 us | journal_mark_dirty(); | |
1427 | 360.774527 | 1) 0.578 us | __brelse(); | |
1428 | 360.774528 | 1) | reiserfs_prepare_for_journal() { | |
1429 | 360.774528 | 1) | unlock_buffer() { | |
1430 | 360.774529 | 1) | wake_up_bit() { | |
1431 | 360.774529 | 1) | bit_waitqueue() { | |
1432 | 360.774530 | 1) 0.594 us | __phys_addr(); | |
1433 | ||
1434 | ||
5752674e | 1435 | You can put some comments on specific functions by using |
5e1607a0 | 1436 | trace_printk() For example, if you want to put a comment inside |
5752674e | 1437 | the __might_sleep() function, you just have to include |
5e1607a0 | 1438 | <linux/ftrace.h> and call trace_printk() inside __might_sleep() |
985ec20a | 1439 | |
5e1607a0 | 1440 | trace_printk("I'm a comment!\n") |
985ec20a FW |
1441 | |
1442 | will produce: | |
1443 | ||
1444 | 1) | __might_sleep() { | |
1445 | 1) | /* I'm a comment! */ | |
1446 | 1) 1.449 us | } | |
1447 | ||
1448 | ||
5752674e IM |
1449 | You might find other useful features for this tracer in the |
1450 | following "dynamic ftrace" section such as tracing only specific | |
1451 | functions or tasks. | |
985ec20a | 1452 | |
eb6d42ea SR |
1453 | dynamic ftrace |
1454 | -------------- | |
1455 | ||
f2d9c740 | 1456 | If CONFIG_DYNAMIC_FTRACE is set, the system will run with |
eb6d42ea SR |
1457 | virtually no overhead when function tracing is disabled. The way |
1458 | this works is the mcount function call (placed at the start of | |
5752674e IM |
1459 | every kernel function, produced by the -pg switch in gcc), |
1460 | starts of pointing to a simple return. (Enabling FTRACE will | |
1461 | include the -pg switch in the compiling of the kernel.) | |
eb6d42ea | 1462 | |
9b803c0f SR |
1463 | At compile time every C file object is run through the |
1464 | recordmcount.pl script (located in the scripts directory). This | |
1465 | script will process the C object using objdump to find all the | |
5752674e IM |
1466 | locations in the .text section that call mcount. (Note, only the |
1467 | .text section is processed, since processing other sections like | |
1468 | .init.text may cause races due to those sections being freed). | |
9b803c0f | 1469 | |
5752674e IM |
1470 | A new section called "__mcount_loc" is created that holds |
1471 | references to all the mcount call sites in the .text section. | |
1472 | This section is compiled back into the original object. The | |
1473 | final linker will add all these references into a single table. | |
9b803c0f SR |
1474 | |
1475 | On boot up, before SMP is initialized, the dynamic ftrace code | |
5752674e IM |
1476 | scans this table and updates all the locations into nops. It |
1477 | also records the locations, which are added to the | |
1478 | available_filter_functions list. Modules are processed as they | |
1479 | are loaded and before they are executed. When a module is | |
1480 | unloaded, it also removes its functions from the ftrace function | |
1481 | list. This is automatic in the module unload code, and the | |
1482 | module author does not need to worry about it. | |
1483 | ||
1484 | When tracing is enabled, kstop_machine is called to prevent | |
1485 | races with the CPUS executing code being modified (which can | |
3ad2f3fb | 1486 | cause the CPU to do undesirable things), and the nops are |
5752674e IM |
1487 | patched back to calls. But this time, they do not call mcount |
1488 | (which is just a function stub). They now call into the ftrace | |
1489 | infrastructure. | |
eb6d42ea SR |
1490 | |
1491 | One special side-effect to the recording of the functions being | |
f2d9c740 | 1492 | traced is that we can now selectively choose which functions we |
5752674e IM |
1493 | wish to trace and which ones we want the mcount calls to remain |
1494 | as nops. | |
eb6d42ea | 1495 | |
5752674e IM |
1496 | Two files are used, one for enabling and one for disabling the |
1497 | tracing of specified functions. They are: | |
eb6d42ea SR |
1498 | |
1499 | set_ftrace_filter | |
1500 | ||
1501 | and | |
1502 | ||
1503 | set_ftrace_notrace | |
1504 | ||
5752674e IM |
1505 | A list of available functions that you can add to these files is |
1506 | listed in: | |
eb6d42ea SR |
1507 | |
1508 | available_filter_functions | |
1509 | ||
156f5a78 | 1510 | # cat available_filter_functions |
eb6d42ea SR |
1511 | put_prev_task_idle |
1512 | kmem_cache_create | |
1513 | pick_next_task_rt | |
1514 | get_online_cpus | |
1515 | pick_next_task_fair | |
1516 | mutex_lock | |
1517 | [...] | |
1518 | ||
f2d9c740 | 1519 | If I am only interested in sys_nanosleep and hrtimer_interrupt: |
eb6d42ea SR |
1520 | |
1521 | # echo sys_nanosleep hrtimer_interrupt \ | |
156f5a78 | 1522 | > set_ftrace_filter |
6993b1bb | 1523 | # echo function > current_tracer |
156f5a78 | 1524 | # echo 1 > tracing_enabled |
eb6d42ea | 1525 | # usleep 1 |
156f5a78 GL |
1526 | # echo 0 > tracing_enabled |
1527 | # cat trace | |
eb6d42ea SR |
1528 | # tracer: ftrace |
1529 | # | |
1530 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
1531 | # | | | | | | |
1532 | usleep-4134 [00] 1317.070017: hrtimer_interrupt <-smp_apic_timer_interrupt | |
1533 | usleep-4134 [00] 1317.070111: sys_nanosleep <-syscall_call | |
1534 | <idle>-0 [00] 1317.070115: hrtimer_interrupt <-smp_apic_timer_interrupt | |
1535 | ||
f2d9c740 | 1536 | To see which functions are being traced, you can cat the file: |
eb6d42ea | 1537 | |
156f5a78 | 1538 | # cat set_ftrace_filter |
eb6d42ea SR |
1539 | hrtimer_interrupt |
1540 | sys_nanosleep | |
1541 | ||
1542 | ||
5752674e IM |
1543 | Perhaps this is not enough. The filters also allow simple wild |
1544 | cards. Only the following are currently available | |
eb6d42ea | 1545 | |
a41eebab | 1546 | <match>* - will match functions that begin with <match> |
eb6d42ea SR |
1547 | *<match> - will match functions that end with <match> |
1548 | *<match>* - will match functions that have <match> in it | |
1549 | ||
f2d9c740 | 1550 | These are the only wild cards which are supported. |
eb6d42ea SR |
1551 | |
1552 | <match>*<match> will not work. | |
1553 | ||
5752674e IM |
1554 | Note: It is better to use quotes to enclose the wild cards, |
1555 | otherwise the shell may expand the parameters into names | |
1556 | of files in the local directory. | |
c072c249 | 1557 | |
156f5a78 | 1558 | # echo 'hrtimer_*' > set_ftrace_filter |
eb6d42ea SR |
1559 | |
1560 | Produces: | |
1561 | ||
1562 | # tracer: ftrace | |
1563 | # | |
1564 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
1565 | # | | | | | | |
1566 | bash-4003 [00] 1480.611794: hrtimer_init <-copy_process | |
1567 | bash-4003 [00] 1480.611941: hrtimer_start <-hrtick_set | |
1568 | bash-4003 [00] 1480.611956: hrtimer_cancel <-hrtick_clear | |
1569 | bash-4003 [00] 1480.611956: hrtimer_try_to_cancel <-hrtimer_cancel | |
1570 | <idle>-0 [00] 1480.612019: hrtimer_get_next_event <-get_next_timer_interrupt | |
1571 | <idle>-0 [00] 1480.612025: hrtimer_get_next_event <-get_next_timer_interrupt | |
1572 | <idle>-0 [00] 1480.612032: hrtimer_get_next_event <-get_next_timer_interrupt | |
1573 | <idle>-0 [00] 1480.612037: hrtimer_get_next_event <-get_next_timer_interrupt | |
1574 | <idle>-0 [00] 1480.612382: hrtimer_get_next_event <-get_next_timer_interrupt | |
1575 | ||
1576 | ||
1577 | Notice that we lost the sys_nanosleep. | |
1578 | ||
156f5a78 | 1579 | # cat set_ftrace_filter |
eb6d42ea SR |
1580 | hrtimer_run_queues |
1581 | hrtimer_run_pending | |
1582 | hrtimer_init | |
1583 | hrtimer_cancel | |
1584 | hrtimer_try_to_cancel | |
1585 | hrtimer_forward | |
1586 | hrtimer_start | |
1587 | hrtimer_reprogram | |
1588 | hrtimer_force_reprogram | |
1589 | hrtimer_get_next_event | |
1590 | hrtimer_interrupt | |
1591 | hrtimer_nanosleep | |
1592 | hrtimer_wakeup | |
1593 | hrtimer_get_remaining | |
1594 | hrtimer_get_res | |
1595 | hrtimer_init_sleeper | |
1596 | ||
1597 | ||
1598 | This is because the '>' and '>>' act just like they do in bash. | |
1599 | To rewrite the filters, use '>' | |
1600 | To append to the filters, use '>>' | |
1601 | ||
5752674e IM |
1602 | To clear out a filter so that all functions will be recorded |
1603 | again: | |
eb6d42ea | 1604 | |
156f5a78 GL |
1605 | # echo > set_ftrace_filter |
1606 | # cat set_ftrace_filter | |
eb6d42ea SR |
1607 | # |
1608 | ||
1609 | Again, now we want to append. | |
1610 | ||
156f5a78 GL |
1611 | # echo sys_nanosleep > set_ftrace_filter |
1612 | # cat set_ftrace_filter | |
eb6d42ea | 1613 | sys_nanosleep |
156f5a78 GL |
1614 | # echo 'hrtimer_*' >> set_ftrace_filter |
1615 | # cat set_ftrace_filter | |
eb6d42ea SR |
1616 | hrtimer_run_queues |
1617 | hrtimer_run_pending | |
1618 | hrtimer_init | |
1619 | hrtimer_cancel | |
1620 | hrtimer_try_to_cancel | |
1621 | hrtimer_forward | |
1622 | hrtimer_start | |
1623 | hrtimer_reprogram | |
1624 | hrtimer_force_reprogram | |
1625 | hrtimer_get_next_event | |
1626 | hrtimer_interrupt | |
1627 | sys_nanosleep | |
1628 | hrtimer_nanosleep | |
1629 | hrtimer_wakeup | |
1630 | hrtimer_get_remaining | |
1631 | hrtimer_get_res | |
1632 | hrtimer_init_sleeper | |
1633 | ||
1634 | ||
5752674e IM |
1635 | The set_ftrace_notrace prevents those functions from being |
1636 | traced. | |
eb6d42ea | 1637 | |
156f5a78 | 1638 | # echo '*preempt*' '*lock*' > set_ftrace_notrace |
eb6d42ea SR |
1639 | |
1640 | Produces: | |
1641 | ||
1642 | # tracer: ftrace | |
1643 | # | |
1644 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
1645 | # | | | | | | |
1646 | bash-4043 [01] 115.281644: finish_task_switch <-schedule | |
1647 | bash-4043 [01] 115.281645: hrtick_set <-schedule | |
1648 | bash-4043 [01] 115.281645: hrtick_clear <-hrtick_set | |
1649 | bash-4043 [01] 115.281646: wait_for_completion <-__stop_machine_run | |
1650 | bash-4043 [01] 115.281647: wait_for_common <-wait_for_completion | |
1651 | bash-4043 [01] 115.281647: kthread_stop <-stop_machine_run | |
1652 | bash-4043 [01] 115.281648: init_waitqueue_head <-kthread_stop | |
1653 | bash-4043 [01] 115.281648: wake_up_process <-kthread_stop | |
1654 | bash-4043 [01] 115.281649: try_to_wake_up <-wake_up_process | |
1655 | ||
1656 | We can see that there's no more lock or preempt tracing. | |
1657 | ||
985ec20a | 1658 | |
5752674e IM |
1659 | Dynamic ftrace with the function graph tracer |
1660 | --------------------------------------------- | |
985ec20a | 1661 | |
5752674e IM |
1662 | Although what has been explained above concerns both the |
1663 | function tracer and the function-graph-tracer, there are some | |
1664 | special features only available in the function-graph tracer. | |
985ec20a | 1665 | |
5752674e IM |
1666 | If you want to trace only one function and all of its children, |
1667 | you just have to echo its name into set_graph_function: | |
985ec20a | 1668 | |
5752674e | 1669 | echo __do_fault > set_graph_function |
985ec20a | 1670 | |
5752674e IM |
1671 | will produce the following "expanded" trace of the __do_fault() |
1672 | function: | |
985ec20a FW |
1673 | |
1674 | 0) | __do_fault() { | |
1675 | 0) | filemap_fault() { | |
1676 | 0) | find_lock_page() { | |
1677 | 0) 0.804 us | find_get_page(); | |
1678 | 0) | __might_sleep() { | |
1679 | 0) 1.329 us | } | |
1680 | 0) 3.904 us | } | |
1681 | 0) 4.979 us | } | |
1682 | 0) 0.653 us | _spin_lock(); | |
1683 | 0) 0.578 us | page_add_file_rmap(); | |
1684 | 0) 0.525 us | native_set_pte_at(); | |
1685 | 0) 0.585 us | _spin_unlock(); | |
1686 | 0) | unlock_page() { | |
1687 | 0) 0.541 us | page_waitqueue(); | |
1688 | 0) 0.639 us | __wake_up_bit(); | |
1689 | 0) 2.786 us | } | |
1690 | 0) + 14.237 us | } | |
1691 | 0) | __do_fault() { | |
1692 | 0) | filemap_fault() { | |
1693 | 0) | find_lock_page() { | |
1694 | 0) 0.698 us | find_get_page(); | |
1695 | 0) | __might_sleep() { | |
1696 | 0) 1.412 us | } | |
1697 | 0) 3.950 us | } | |
1698 | 0) 5.098 us | } | |
1699 | 0) 0.631 us | _spin_lock(); | |
1700 | 0) 0.571 us | page_add_file_rmap(); | |
1701 | 0) 0.526 us | native_set_pte_at(); | |
1702 | 0) 0.586 us | _spin_unlock(); | |
1703 | 0) | unlock_page() { | |
1704 | 0) 0.533 us | page_waitqueue(); | |
1705 | 0) 0.638 us | __wake_up_bit(); | |
1706 | 0) 2.793 us | } | |
1707 | 0) + 14.012 us | } | |
1708 | ||
5752674e | 1709 | You can also expand several functions at once: |
985ec20a | 1710 | |
5752674e IM |
1711 | echo sys_open > set_graph_function |
1712 | echo sys_close >> set_graph_function | |
985ec20a | 1713 | |
5752674e IM |
1714 | Now if you want to go back to trace all functions you can clear |
1715 | this special filter via: | |
985ec20a | 1716 | |
5752674e | 1717 | echo > set_graph_function |
985ec20a FW |
1718 | |
1719 | ||
07271aa4 CD |
1720 | Filter commands |
1721 | --------------- | |
1722 | ||
1723 | A few commands are supported by the set_ftrace_filter interface. | |
1724 | Trace commands have the following format: | |
1725 | ||
1726 | <function>:<command>:<parameter> | |
1727 | ||
1728 | The following commands are supported: | |
1729 | ||
1730 | - mod | |
1731 | This command enables function filtering per module. The | |
1732 | parameter defines the module. For example, if only the write* | |
1733 | functions in the ext3 module are desired, run: | |
1734 | ||
1735 | echo 'write*:mod:ext3' > set_ftrace_filter | |
1736 | ||
1737 | This command interacts with the filter in the same way as | |
1738 | filtering based on function names. Thus, adding more functions | |
1739 | in a different module is accomplished by appending (>>) to the | |
1740 | filter file. Remove specific module functions by prepending | |
1741 | '!': | |
1742 | ||
1743 | echo '!writeback*:mod:ext3' >> set_ftrace_filter | |
1744 | ||
1745 | - traceon/traceoff | |
1746 | These commands turn tracing on and off when the specified | |
1747 | functions are hit. The parameter determines how many times the | |
1748 | tracing system is turned on and off. If unspecified, there is | |
1749 | no limit. For example, to disable tracing when a schedule bug | |
1750 | is hit the first 5 times, run: | |
1751 | ||
1752 | echo '__schedule_bug:traceoff:5' > set_ftrace_filter | |
1753 | ||
1754 | These commands are cumulative whether or not they are appended | |
1755 | to set_ftrace_filter. To remove a command, prepend it by '!' | |
1756 | and drop the parameter: | |
1757 | ||
1758 | echo '!__schedule_bug:traceoff' > set_ftrace_filter | |
1759 | ||
1760 | ||
eb6d42ea SR |
1761 | trace_pipe |
1762 | ---------- | |
1763 | ||
5752674e IM |
1764 | The trace_pipe outputs the same content as the trace file, but |
1765 | the effect on the tracing is different. Every read from | |
1766 | trace_pipe is consumed. This means that subsequent reads will be | |
1767 | different. The trace is live. | |
eb6d42ea | 1768 | |
156f5a78 GL |
1769 | # echo function > current_tracer |
1770 | # cat trace_pipe > /tmp/trace.out & | |
eb6d42ea | 1771 | [1] 4153 |
156f5a78 | 1772 | # echo 1 > tracing_enabled |
eb6d42ea | 1773 | # usleep 1 |
156f5a78 GL |
1774 | # echo 0 > tracing_enabled |
1775 | # cat trace | |
9b803c0f | 1776 | # tracer: function |
eb6d42ea SR |
1777 | # |
1778 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
1779 | # | | | | | | |
1780 | ||
1781 | # | |
1782 | # cat /tmp/trace.out | |
1783 | bash-4043 [00] 41.267106: finish_task_switch <-schedule | |
1784 | bash-4043 [00] 41.267106: hrtick_set <-schedule | |
1785 | bash-4043 [00] 41.267107: hrtick_clear <-hrtick_set | |
1786 | bash-4043 [00] 41.267108: wait_for_completion <-__stop_machine_run | |
1787 | bash-4043 [00] 41.267108: wait_for_common <-wait_for_completion | |
1788 | bash-4043 [00] 41.267109: kthread_stop <-stop_machine_run | |
1789 | bash-4043 [00] 41.267109: init_waitqueue_head <-kthread_stop | |
1790 | bash-4043 [00] 41.267110: wake_up_process <-kthread_stop | |
1791 | bash-4043 [00] 41.267110: try_to_wake_up <-wake_up_process | |
1792 | bash-4043 [00] 41.267111: select_task_rq_rt <-try_to_wake_up | |
1793 | ||
1794 | ||
5752674e IM |
1795 | Note, reading the trace_pipe file will block until more input is |
1796 | added. By changing the tracer, trace_pipe will issue an EOF. We | |
1797 | needed to set the function tracer _before_ we "cat" the | |
1798 | trace_pipe file. | |
eb6d42ea SR |
1799 | |
1800 | ||
1801 | trace entries | |
1802 | ------------- | |
1803 | ||
5752674e IM |
1804 | Having too much or not enough data can be troublesome in |
1805 | diagnosing an issue in the kernel. The file buffer_size_kb is | |
1806 | used to modify the size of the internal trace buffers. The | |
1807 | number listed is the number of entries that can be recorded per | |
1808 | CPU. To know the full size, multiply the number of possible CPUS | |
1809 | with the number of entries. | |
eb6d42ea | 1810 | |
156f5a78 | 1811 | # cat buffer_size_kb |
1696b2b0 | 1812 | 1408 (units kilobytes) |
eb6d42ea | 1813 | |
5752674e IM |
1814 | Note, to modify this, you must have tracing completely disabled. |
1815 | To do that, echo "nop" into the current_tracer. If the | |
1816 | current_tracer is not set to "nop", an EINVAL error will be | |
1817 | returned. | |
eb6d42ea | 1818 | |
156f5a78 GL |
1819 | # echo nop > current_tracer |
1820 | # echo 10000 > buffer_size_kb | |
1821 | # cat buffer_size_kb | |
1696b2b0 | 1822 | 10000 (units kilobytes) |
eb6d42ea | 1823 | |
5752674e IM |
1824 | The number of pages which will be allocated is limited to a |
1825 | percentage of available memory. Allocating too much will produce | |
1826 | an error. | |
eb6d42ea | 1827 | |
156f5a78 | 1828 | # echo 1000000000000 > buffer_size_kb |
eb6d42ea | 1829 | -bash: echo: write error: Cannot allocate memory |
156f5a78 | 1830 | # cat buffer_size_kb |
eb6d42ea SR |
1831 | 85 |
1832 | ||
5752674e IM |
1833 | ----------- |
1834 | ||
1835 | More details can be found in the source code, in the | |
baf20b3e | 1836 | kernel/trace/*.c files. |