aboutsummaryrefslogtreecommitdiffstats
path: root/Documentation/ftrace.txt
diff options
context:
space:
mode:
Diffstat (limited to 'Documentation/ftrace.txt')
-rw-r--r--Documentation/ftrace.txt1134
1 files changed, 769 insertions, 365 deletions
diff --git a/Documentation/ftrace.txt b/Documentation/ftrace.txt
index 803b1318b13d..fd9a3e693813 100644
--- a/Documentation/ftrace.txt
+++ b/Documentation/ftrace.txt
@@ -15,31 +15,31 @@ Introduction
15 15
16Ftrace is an internal tracer designed to help out developers and 16Ftrace is an internal tracer designed to help out developers and
17designers of systems to find what is going on inside the kernel. 17designers of systems to find what is going on inside the kernel.
18It can be used for debugging or analyzing latencies and performance 18It can be used for debugging or analyzing latencies and
19issues that take place outside of user-space. 19performance issues that take place outside of user-space.
20 20
21Although ftrace is the function tracer, it also includes an 21Although ftrace is the function tracer, it also includes an
22infrastructure that allows for other types of tracing. Some of the 22infrastructure that allows for other types of tracing. Some of
23tracers that are currently in ftrace include a tracer to trace 23the tracers that are currently in ftrace include a tracer to
24context switches, the time it takes for a high priority task to 24trace context switches, the time it takes for a high priority
25run after it was woken up, the time interrupts are disabled, and 25task to run after it was woken up, the time interrupts are
26more (ftrace allows for tracer plugins, which means that the list of 26disabled, and more (ftrace allows for tracer plugins, which
27tracers can always grow). 27means that the list of tracers can always grow).
28 28
29 29
30The File System 30The File System
31--------------- 31---------------
32 32
33Ftrace uses the debugfs file system to hold the control files as well 33Ftrace uses the debugfs file system to hold the control files as
34as the files to display output. 34well as the files to display output.
35 35
36To mount the debugfs system: 36To mount the debugfs system:
37 37
38 # mkdir /debug 38 # mkdir /debug
39 # mount -t debugfs nodev /debug 39 # mount -t debugfs nodev /debug
40 40
41(Note: it is more common to mount at /sys/kernel/debug, but for simplicity 41( Note: it is more common to mount at /sys/kernel/debug, but for
42 this document will use /debug) 42 simplicity this document will use /debug)
43 43
44That's it! (assuming that you have ftrace configured into your kernel) 44That's it! (assuming that you have ftrace configured into your kernel)
45 45
@@ -50,90 +50,124 @@ of ftrace. Here is a list of some of the key files:
50 50
51 Note: all time values are in microseconds. 51 Note: all time values are in microseconds.
52 52
53 current_tracer: This is used to set or display the current tracer 53 current_tracer:
54 that is configured. 54
55 55 This is used to set or display the current tracer
56 available_tracers: This holds the different types of tracers that 56 that is configured.
57 have been compiled into the kernel. The tracers 57
58 listed here can be configured by echoing their name 58 available_tracers:
59 into current_tracer. 59
60 60 This holds the different types of tracers that
61 tracing_enabled: This sets or displays whether the current_tracer 61 have been compiled into the kernel. The
62 is activated and tracing or not. Echo 0 into this 62 tracers listed here can be configured by
63 file to disable the tracer or 1 to enable it. 63 echoing their name into current_tracer.
64 64
65 trace: This file holds the output of the trace in a human readable 65 tracing_enabled:
66 format (described below). 66
67 67 This sets or displays whether the current_tracer
68 latency_trace: This file shows the same trace but the information 68 is activated and tracing or not. Echo 0 into this
69 is organized more to display possible latencies 69 file to disable the tracer or 1 to enable it.
70 in the system (described below). 70
71 71 trace:
72 trace_pipe: The output is the same as the "trace" file but this 72
73 file is meant to be streamed with live tracing. 73 This file holds the output of the trace in a human
74 Reads from this file will block until new data 74 readable format (described below).
75 is retrieved. Unlike the "trace" and "latency_trace" 75
76 files, this file is a consumer. This means reading 76 latency_trace:
77 from this file causes sequential reads to display 77
78 more current data. Once data is read from this 78 This file shows the same trace but the information
79 file, it is consumed, and will not be read 79 is organized more to display possible latencies
80 again with a sequential read. The "trace" and 80 in the system (described below).
81 "latency_trace" files are static, and if the 81
82 tracer is not adding more data, they will display 82 trace_pipe:
83 the same information every time they are read. 83
84 84 The output is the same as the "trace" file but this
85 trace_options: This file lets the user control the amount of data 85 file is meant to be streamed with live tracing.
86 that is displayed in one of the above output 86 Reads from this file will block until new data
87 files. 87 is retrieved. Unlike the "trace" and "latency_trace"
88 88 files, this file is a consumer. This means reading
89 trace_max_latency: Some of the tracers record the max latency. 89 from this file causes sequential reads to display
90 For example, the time interrupts are disabled. 90 more current data. Once data is read from this
91 This time is saved in this file. The max trace 91 file, it is consumed, and will not be read
92 will also be stored, and displayed by either 92 again with a sequential read. The "trace" and
93 "trace" or "latency_trace". A new max trace will 93 "latency_trace" files are static, and if the
94 only be recorded if the latency is greater than 94 tracer is not adding more data, they will display
95 the value in this file. (in microseconds) 95 the same information every time they are read.
96 96
97 buffer_size_kb: This sets or displays the number of kilobytes each CPU 97 trace_options:
98 buffer can hold. The tracer buffers are the same size 98
99 for each CPU. The displayed number is the size of the 99 This file lets the user control the amount of data
100 CPU buffer and not total size of all buffers. The 100 that is displayed in one of the above output
101 trace buffers are allocated in pages (blocks of memory 101 files.
102 that the kernel uses for allocation, usually 4 KB in size). 102
103 If the last page allocated has room for more bytes 103 tracing_max_latency:
104 than requested, the rest of the page will be used, 104
105 making the actual allocation bigger than requested. 105 Some of the tracers record the max latency.
106 (Note, the size may not be a multiple of the page size due 106 For example, the time interrupts are disabled.
107 to buffer managment overhead.) 107 This time is saved in this file. The max trace
108 108 will also be stored, and displayed by either
109 This can only be updated when the current_tracer 109 "trace" or "latency_trace". A new max trace will
110 is set to "nop". 110 only be recorded if the latency is greater than
111 111 the value in this file. (in microseconds)
112 tracing_cpumask: This is a mask that lets the user only trace 112
113 on specified CPUS. The format is a hex string 113 buffer_size_kb:
114 representing the CPUS. 114
115 115 This sets or displays the number of kilobytes each CPU
116 set_ftrace_filter: When dynamic ftrace is configured in (see the 116 buffer can hold. The tracer buffers are the same size
117 section below "dynamic ftrace"), the code is dynamically 117 for each CPU. The displayed number is the size of the
118 modified (code text rewrite) to disable calling of the 118 CPU buffer and not total size of all buffers. The
119 function profiler (mcount). This lets tracing be configured 119 trace buffers are allocated in pages (blocks of memory
120 in with practically no overhead in performance. This also 120 that the kernel uses for allocation, usually 4 KB in size).
121 has a side effect of enabling or disabling specific functions 121 If the last page allocated has room for more bytes
122 to be traced. Echoing names of functions into this file 122 than requested, the rest of the page will be used,
123 will limit the trace to only those functions. 123 making the actual allocation bigger than requested.
124 124 ( Note, the size may not be a multiple of the page size
125 set_ftrace_notrace: This has an effect opposite to that of 125 due to buffer managment overhead. )
126 set_ftrace_filter. Any function that is added here will not 126
127 be traced. If a function exists in both set_ftrace_filter 127 This can only be updated when the current_tracer
128 and set_ftrace_notrace, the function will _not_ be traced. 128 is set to "nop".
129 129
130 set_ftrace_pid: Have the function tracer only trace a single thread. 130 tracing_cpumask:
131 131
132 available_filter_functions: This lists the functions that ftrace 132 This is a mask that lets the user only trace
133 has processed and can trace. These are the function 133 on specified CPUS. The format is a hex string
134 names that you can pass to "set_ftrace_filter" or 134 representing the CPUS.
135 "set_ftrace_notrace". (See the section "dynamic ftrace" 135
136 below for more details.) 136 set_ftrace_filter:
137
138 When dynamic ftrace is configured in (see the
139 section below "dynamic ftrace"), the code is dynamically
140 modified (code text rewrite) to disable calling of the
141 function profiler (mcount). This lets tracing be configured
142 in with practically no overhead in performance. This also
143 has a side effect of enabling or disabling specific functions
144 to be traced. Echoing names of functions into this file
145 will limit the trace to only those functions.
146
147 set_ftrace_notrace:
148
149 This has an effect opposite to that of
150 set_ftrace_filter. Any function that is added here will not
151 be traced. If a function exists in both set_ftrace_filter
152 and set_ftrace_notrace, the function will _not_ be traced.
153
154 set_ftrace_pid:
155
156 Have the function tracer only trace a single thread.
157
158 set_graph_function:
159
160 Set a "trigger" function where tracing should start
161 with the function graph tracer (See the section
162 "dynamic ftrace" for more details).
163
164 available_filter_functions:
165
166 This lists the functions that ftrace
167 has processed and can trace. These are the function
168 names that you can pass to "set_ftrace_filter" or
169 "set_ftrace_notrace". (See the section "dynamic ftrace"
170 below for more details.)
137 171
138 172
139The Tracers 173The Tracers
@@ -141,36 +175,66 @@ The Tracers
141 175
142Here is the list of current tracers that may be configured. 176Here is the list of current tracers that may be configured.
143 177
144 function - function tracer that uses mcount to trace all functions. 178 "function"
179
180 Function call tracer to trace all kernel functions.
181
182 "function_graph_tracer"
183
184 Similar to the function tracer except that the
185 function tracer probes the functions on their entry
186 whereas the function graph tracer traces on both entry
187 and exit of the functions. It then provides the ability
188 to draw a graph of function calls similar to C code
189 source.
145 190
146 sched_switch - traces the context switches between tasks. 191 "sched_switch"
147 192
148 irqsoff - traces the areas that disable interrupts and saves 193 Traces the context switches and wakeups between tasks.
149 the trace with the longest max latency.
150 See tracing_max_latency. When a new max is recorded,
151 it replaces the old trace. It is best to view this
152 trace via the latency_trace file.
153 194
154 preemptoff - Similar to irqsoff but traces and records the amount of 195 "irqsoff"
155 time for which preemption is disabled.
156 196
157 preemptirqsoff - Similar to irqsoff and preemptoff, but traces and 197 Traces the areas that disable interrupts and saves
158 records the largest time for which irqs and/or preemption 198 the trace with the longest max latency.
159 is disabled. 199 See tracing_max_latency. When a new max is recorded,
200 it replaces the old trace. It is best to view this
201 trace via the latency_trace file.
160 202
161 wakeup - Traces and records the max latency that it takes for 203 "preemptoff"
162 the highest priority task to get scheduled after
163 it has been woken up.
164 204
165 nop - This is not a tracer. To remove all tracers from tracing 205 Similar to irqsoff but traces and records the amount of
166 simply echo "nop" into current_tracer. 206 time for which preemption is disabled.
207
208 "preemptirqsoff"
209
210 Similar to irqsoff and preemptoff, but traces and
211 records the largest time for which irqs and/or preemption
212 is disabled.
213
214 "wakeup"
215
216 Traces and records the max latency that it takes for
217 the highest priority task to get scheduled after
218 it has been woken up.
219
220 "hw-branch-tracer"
221
222 Uses the BTS CPU feature on x86 CPUs to traces all
223 branches executed.
224
225 "nop"
226
227 This is the "trace nothing" tracer. To remove all
228 tracers from tracing simply echo "nop" into
229 current_tracer.
167 230
168 231
169Examples of using the tracer 232Examples of using the tracer
170---------------------------- 233----------------------------
171 234
172Here are typical examples of using the tracers when controlling them only 235Here are typical examples of using the tracers when controlling
173with the debugfs interface (without using any user-land utilities). 236them only with the debugfs interface (without using any
237user-land utilities).
174 238
175Output format: 239Output format:
176-------------- 240--------------
@@ -187,16 +251,16 @@ Here is an example of the output format of the file "trace"
187 bash-4251 [01] 10152.583855: _atomic_dec_and_lock <-dput 251 bash-4251 [01] 10152.583855: _atomic_dec_and_lock <-dput
188 -------- 252 --------
189 253
190A header is printed with the tracer name that is represented by the trace. 254A header is printed with the tracer name that is represented by
191In this case the tracer is "function". Then a header showing the format. Task 255the trace. In this case the tracer is "function". Then a header
192name "bash", the task PID "4251", the CPU that it was running on 256showing the format. Task name "bash", the task PID "4251", the
193"01", the timestamp in <secs>.<usecs> format, the function name that was 257CPU that it was running on "01", the timestamp in <secs>.<usecs>
194traced "path_put" and the parent function that called this function 258format, the function name that was traced "path_put" and the
195"path_walk". The timestamp is the time at which the function was 259parent function that called this function "path_walk". The
196entered. 260timestamp is the time at which the function was entered.
197 261
198The sched_switch tracer also includes tracing of task wakeups and 262The sched_switch tracer also includes tracing of task wakeups
199context switches. 263and context switches.
200 264
201 ksoftirqd/1-7 [01] 1453.070013: 7:115:R + 2916:115:S 265 ksoftirqd/1-7 [01] 1453.070013: 7:115:R + 2916:115:S
202 ksoftirqd/1-7 [01] 1453.070013: 7:115:R + 10:115:S 266 ksoftirqd/1-7 [01] 1453.070013: 7:115:R + 10:115:S
@@ -205,8 +269,8 @@ context switches.
205 kondemand/1-2916 [01] 1453.070013: 2916:115:S ==> 7:115:R 269 kondemand/1-2916 [01] 1453.070013: 2916:115:S ==> 7:115:R
206 ksoftirqd/1-7 [01] 1453.070013: 7:115:S ==> 0:140:R 270 ksoftirqd/1-7 [01] 1453.070013: 7:115:S ==> 0:140:R
207 271
208Wake ups are represented by a "+" and the context switches are shown as 272Wake ups are represented by a "+" and the context switches are
209"==>". The format is: 273shown as "==>". The format is:
210 274
211 Context switches: 275 Context switches:
212 276
@@ -220,19 +284,20 @@ Wake ups are represented by a "+" and the context switches are shown as
220 284
221 <pid>:<prio>:<state> + <pid>:<prio>:<state> 285 <pid>:<prio>:<state> + <pid>:<prio>:<state>
222 286
223The prio is the internal kernel priority, which is the inverse of the 287The prio is the internal kernel priority, which is the inverse
224priority that is usually displayed by user-space tools. Zero represents 288of the priority that is usually displayed by user-space tools.
225the highest priority (99). Prio 100 starts the "nice" priorities with 289Zero represents the highest priority (99). Prio 100 starts the
226100 being equal to nice -20 and 139 being nice 19. The prio "140" is 290"nice" priorities with 100 being equal to nice -20 and 139 being
227reserved for the idle task which is the lowest priority thread (pid 0). 291nice 19. The prio "140" is reserved for the idle task which is
292the lowest priority thread (pid 0).
228 293
229 294
230Latency trace format 295Latency trace format
231-------------------- 296--------------------
232 297
233For traces that display latency times, the latency_trace file gives 298For traces that display latency times, the latency_trace file
234somewhat more information to see why a latency happened. Here is a typical 299gives somewhat more information to see why a latency happened.
235trace. 300Here is a typical trace.
236 301
237# tracer: irqsoff 302# tracer: irqsoff
238# 303#
@@ -259,20 +324,20 @@ irqsoff latency trace v1.1.5 on 2.6.26-rc8
259 <idle>-0 0d.s1 98us : trace_hardirqs_on (do_softirq) 324 <idle>-0 0d.s1 98us : trace_hardirqs_on (do_softirq)
260 325
261 326
327This shows that the current tracer is "irqsoff" tracing the time
328for which interrupts were disabled. It gives the trace version
329and the version of the kernel upon which this was executed on
330(2.6.26-rc8). Then it displays the max latency in microsecs (97
331us). The number of trace entries displayed and the total number
332recorded (both are three: #3/3). The type of preemption that was
333used (PREEMPT). VP, KP, SP, and HP are always zero and are
334reserved for later use. #P is the number of online CPUS (#P:2).
262 335
263This shows that the current tracer is "irqsoff" tracing the time for which 336The task is the process that was running when the latency
264interrupts were disabled. It gives the trace version and the version 337occurred. (swapper pid: 0).
265of the kernel upon which this was executed on (2.6.26-rc8). Then it displays
266the max latency in microsecs (97 us). The number of trace entries displayed
267and the total number recorded (both are three: #3/3). The type of
268preemption that was used (PREEMPT). VP, KP, SP, and HP are always zero
269and are reserved for later use. #P is the number of online CPUS (#P:2).
270
271The task is the process that was running when the latency occurred.
272(swapper pid: 0).
273 338
274The start and stop (the functions in which the interrupts were disabled and 339The start and stop (the functions in which the interrupts were
275enabled respectively) that caused the latencies: 340disabled and enabled respectively) that caused the latencies:
276 341
277 apic_timer_interrupt is where the interrupts were disabled. 342 apic_timer_interrupt is where the interrupts were disabled.
278 do_softirq is where they were enabled again. 343 do_softirq is where they were enabled again.
@@ -308,12 +373,12 @@ The above is mostly meaningful for kernel developers.
308 latency_trace file is relative to the start of the trace. 373 latency_trace file is relative to the start of the trace.
309 374
310 delay: This is just to help catch your eye a bit better. And 375 delay: This is just to help catch your eye a bit better. And
311 needs to be fixed to be only relative to the same CPU. 376 needs to be fixed to be only relative to the same CPU.
312 The marks are determined by the difference between this 377 The marks are determined by the difference between this
313 current trace and the next trace. 378 current trace and the next trace.
314 '!' - greater than preempt_mark_thresh (default 100) 379 '!' - greater than preempt_mark_thresh (default 100)
315 '+' - greater than 1 microsecond 380 '+' - greater than 1 microsecond
316 ' ' - less than or equal to 1 microsecond. 381 ' ' - less than or equal to 1 microsecond.
317 382
318 The rest is the same as the 'trace' file. 383 The rest is the same as the 'trace' file.
319 384
@@ -321,14 +386,15 @@ The above is mostly meaningful for kernel developers.
321trace_options 386trace_options
322------------- 387-------------
323 388
324The trace_options file is used to control what gets printed in the trace 389The trace_options file is used to control what gets printed in
325output. To see what is available, simply cat the file: 390the trace output. To see what is available, simply cat the file:
326 391
327 cat /debug/tracing/trace_options 392 cat /debug/tracing/trace_options
328 print-parent nosym-offset nosym-addr noverbose noraw nohex nobin \ 393 print-parent nosym-offset nosym-addr noverbose noraw nohex nobin \
329 noblock nostacktrace nosched-tree nouserstacktrace nosym-userobj 394 noblock nostacktrace nosched-tree nouserstacktrace nosym-userobj
330 395
331To disable one of the options, echo in the option prepended with "no". 396To disable one of the options, echo in the option prepended with
397"no".
332 398
333 echo noprint-parent > /debug/tracing/trace_options 399 echo noprint-parent > /debug/tracing/trace_options
334 400
@@ -338,8 +404,8 @@ To enable an option, leave off the "no".
338 404
339Here are the available options: 405Here are the available options:
340 406
341 print-parent - On function traces, display the calling function 407 print-parent - On function traces, display the calling (parent)
342 as well as the function being traced. 408 function as well as the function being traced.
343 409
344 print-parent: 410 print-parent:
345 bash-4000 [01] 1477.606694: simple_strtoul <-strict_strtoul 411 bash-4000 [01] 1477.606694: simple_strtoul <-strict_strtoul
@@ -348,15 +414,16 @@ Here are the available options:
348 bash-4000 [01] 1477.606694: simple_strtoul 414 bash-4000 [01] 1477.606694: simple_strtoul
349 415
350 416
351 sym-offset - Display not only the function name, but also the offset 417 sym-offset - Display not only the function name, but also the
352 in the function. For example, instead of seeing just 418 offset in the function. For example, instead of
353 "ktime_get", you will see "ktime_get+0xb/0x20". 419 seeing just "ktime_get", you will see
420 "ktime_get+0xb/0x20".
354 421
355 sym-offset: 422 sym-offset:
356 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0 423 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
357 424
358 sym-addr - this will also display the function address as well as 425 sym-addr - this will also display the function address as well
359 the function name. 426 as the function name.
360 427
361 sym-addr: 428 sym-addr:
362 bash-4000 [01] 1477.606694: simple_strtoul <c0339346> 429 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
@@ -366,35 +433,41 @@ Here are the available options:
366 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \ 433 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
367 (+0.000ms): simple_strtoul (strict_strtoul) 434 (+0.000ms): simple_strtoul (strict_strtoul)
368 435
369 raw - This will display raw numbers. This option is best for use with 436 raw - This will display raw numbers. This option is best for
370 user applications that can translate the raw numbers better than 437 use with user applications that can translate the raw
371 having it done in the kernel. 438 numbers better than having it done in the kernel.
372 439
373 hex - Similar to raw, but the numbers will be in a hexadecimal format. 440 hex - Similar to raw, but the numbers will be in a hexadecimal
441 format.
374 442
375 bin - This will print out the formats in raw binary. 443 bin - This will print out the formats in raw binary.
376 444
377 block - TBD (needs update) 445 block - TBD (needs update)
378 446
379 stacktrace - This is one of the options that changes the trace itself. 447 stacktrace - This is one of the options that changes the trace
380 When a trace is recorded, so is the stack of functions. 448 itself. When a trace is recorded, so is the stack
381 This allows for back traces of trace sites. 449 of functions. This allows for back traces of
450 trace sites.
382 451
383 userstacktrace - This option changes the trace. 452 userstacktrace - This option changes the trace. It records a
384 It records a stacktrace of the current userspace thread. 453 stacktrace of the current userspace thread.
385 454
386 sym-userobj - when user stacktrace are enabled, look up which object the 455 sym-userobj - when user stacktrace are enabled, look up which
387 address belongs to, and print a relative address 456 object the address belongs to, and print a
388 This is especially useful when ASLR is on, otherwise you don't 457 relative address. This is especially useful when
389 get a chance to resolve the address to object/file/line after the app is no 458 ASLR is on, otherwise you don't get a chance to
390 longer running 459 resolve the address to object/file/line after
460 the app is no longer running
391 461
392 The lookup is performed when you read trace,trace_pipe,latency_trace. Example: 462 The lookup is performed when you read
463 trace,trace_pipe,latency_trace. Example:
393 464
394 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0 465 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
395x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6] 466x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
396 467
397 sched-tree - TBD (any users??) 468 sched-tree - trace all tasks that are on the runqueue, at
469 every scheduling event. Will add overhead if
470 there's a lot of tasks running at once.
398 471
399 472
400sched_switch 473sched_switch
@@ -431,18 +504,19 @@ of how to use it.
431 [...] 504 [...]
432 505
433 506
434As we have discussed previously about this format, the header shows 507As we have discussed previously about this format, the header
435the name of the trace and points to the options. The "FUNCTION" 508shows the name of the trace and points to the options. The
436is a misnomer since here it represents the wake ups and context 509"FUNCTION" is a misnomer since here it represents the wake ups
437switches. 510and context switches.
438 511
439The sched_switch file only lists the wake ups (represented with '+') 512The sched_switch file only lists the wake ups (represented with
440and context switches ('==>') with the previous task or current task 513'+') and context switches ('==>') with the previous task or
441first followed by the next task or task waking up. The format for both 514current task first followed by the next task or task waking up.
442of these is PID:KERNEL-PRIO:TASK-STATE. Remember that the KERNEL-PRIO 515The format for both of these is PID:KERNEL-PRIO:TASK-STATE.
443is the inverse of the actual priority with zero (0) being the highest 516Remember that the KERNEL-PRIO is the inverse of the actual
444priority and the nice values starting at 100 (nice -20). Below is 517priority with zero (0) being the highest priority and the nice
445a quick chart to map the kernel priority to user land priorities. 518values starting at 100 (nice -20). Below is a quick chart to map
519the kernel priority to user land priorities.
446 520
447 Kernel priority: 0 to 99 ==> user RT priority 99 to 0 521 Kernel priority: 0 to 99 ==> user RT priority 99 to 0
448 Kernel priority: 100 to 139 ==> user nice -20 to 19 522 Kernel priority: 100 to 139 ==> user nice -20 to 19
@@ -463,10 +537,10 @@ The task states are:
463ftrace_enabled 537ftrace_enabled
464-------------- 538--------------
465 539
466The following tracers (listed below) give different output depending 540The following tracers (listed below) give different output
467on whether or not the sysctl ftrace_enabled is set. To set ftrace_enabled, 541depending on whether or not the sysctl ftrace_enabled is set. To
468one can either use the sysctl function or set it via the proc 542set ftrace_enabled, one can either use the sysctl function or
469file system interface. 543set it via the proc file system interface.
470 544
471 sysctl kernel.ftrace_enabled=1 545 sysctl kernel.ftrace_enabled=1
472 546
@@ -474,12 +548,12 @@ file system interface.
474 548
475 echo 1 > /proc/sys/kernel/ftrace_enabled 549 echo 1 > /proc/sys/kernel/ftrace_enabled
476 550
477To disable ftrace_enabled simply replace the '1' with '0' in 551To disable ftrace_enabled simply replace the '1' with '0' in the
478the above commands. 552above commands.
479 553
480When ftrace_enabled is set the tracers will also record the functions 554When ftrace_enabled is set the tracers will also record the
481that are within the trace. The descriptions of the tracers 555functions that are within the trace. The descriptions of the
482will also show an example with ftrace enabled. 556tracers will also show an example with ftrace enabled.
483 557
484 558
485irqsoff 559irqsoff
@@ -487,17 +561,18 @@ irqsoff
487 561
488When interrupts are disabled, the CPU can not react to any other 562When interrupts are disabled, the CPU can not react to any other
489external event (besides NMIs and SMIs). This prevents the timer 563external event (besides NMIs and SMIs). This prevents the timer
490interrupt from triggering or the mouse interrupt from letting the 564interrupt from triggering or the mouse interrupt from letting
491kernel know of a new mouse event. The result is a latency with the 565the kernel know of a new mouse event. The result is a latency
492reaction time. 566with the reaction time.
493 567
494The irqsoff tracer tracks the time for which interrupts are disabled. 568The irqsoff tracer tracks the time for which interrupts are
495When a new maximum latency is hit, the tracer saves the trace leading up 569disabled. When a new maximum latency is hit, the tracer saves
496to that latency point so that every time a new maximum is reached, the old 570the trace leading up to that latency point so that every time a
497saved trace is discarded and the new trace is saved. 571new maximum is reached, the old saved trace is discarded and the
572new trace is saved.
498 573
499To reset the maximum, echo 0 into tracing_max_latency. Here is an 574To reset the maximum, echo 0 into tracing_max_latency. Here is
500example: 575an example:
501 576
502 # echo irqsoff > /debug/tracing/current_tracer 577 # echo irqsoff > /debug/tracing/current_tracer
503 # echo 0 > /debug/tracing/tracing_max_latency 578 # echo 0 > /debug/tracing/tracing_max_latency
@@ -532,10 +607,11 @@ irqsoff latency trace v1.1.5 on 2.6.26
532 607
533 608
534Here we see that that we had a latency of 12 microsecs (which is 609Here we see that that we had a latency of 12 microsecs (which is
535very good). The _write_lock_irq in sys_setpgid disabled interrupts. 610very good). The _write_lock_irq in sys_setpgid disabled
536The difference between the 12 and the displayed timestamp 14us occurred 611interrupts. The difference between the 12 and the displayed
537because the clock was incremented between the time of recording the max 612timestamp 14us occurred because the clock was incremented
538latency and the time of recording the function that had that latency. 613between the time of recording the max latency and the time of
614recording the function that had that latency.
539 615
540Note the above example had ftrace_enabled not set. If we set the 616Note the above example had ftrace_enabled not set. If we set the
541ftrace_enabled, we get a much larger output: 617ftrace_enabled, we get a much larger output:
@@ -586,24 +662,24 @@ irqsoff latency trace v1.1.5 on 2.6.26-rc8
586 662
587 663
588Here we traced a 50 microsecond latency. But we also see all the 664Here we traced a 50 microsecond latency. But we also see all the
589functions that were called during that time. Note that by enabling 665functions that were called during that time. Note that by
590function tracing, we incur an added overhead. This overhead may 666enabling function tracing, we incur an added overhead. This
591extend the latency times. But nevertheless, this trace has provided 667overhead may extend the latency times. But nevertheless, this
592some very helpful debugging information. 668trace has provided some very helpful debugging information.
593 669
594 670
595preemptoff 671preemptoff
596---------- 672----------
597 673
598When preemption is disabled, we may be able to receive interrupts but 674When preemption is disabled, we may be able to receive
599the task cannot be preempted and a higher priority task must wait 675interrupts but the task cannot be preempted and a higher
600for preemption to be enabled again before it can preempt a lower 676priority task must wait for preemption to be enabled again
601priority task. 677before it can preempt a lower priority task.
602 678
603The preemptoff tracer traces the places that disable preemption. 679The preemptoff tracer traces the places that disable preemption.
604Like the irqsoff tracer, it records the maximum latency for which preemption 680Like the irqsoff tracer, it records the maximum latency for
605was disabled. The control of preemptoff tracer is much like the irqsoff 681which preemption was disabled. The control of preemptoff tracer
606tracer. 682is much like the irqsoff tracer.
607 683
608 # echo preemptoff > /debug/tracing/current_tracer 684 # echo preemptoff > /debug/tracing/current_tracer
609 # echo 0 > /debug/tracing/tracing_max_latency 685 # echo 0 > /debug/tracing/tracing_max_latency
@@ -637,11 +713,12 @@ preemptoff latency trace v1.1.5 on 2.6.26-rc8
637 sshd-4261 0d.s1 30us : trace_preempt_on (__do_softirq) 713 sshd-4261 0d.s1 30us : trace_preempt_on (__do_softirq)
638 714
639 715
640This has some more changes. Preemption was disabled when an interrupt 716This has some more changes. Preemption was disabled when an
641came in (notice the 'h'), and was enabled while doing a softirq. 717interrupt came in (notice the 'h'), and was enabled while doing
642(notice the 's'). But we also see that interrupts have been disabled 718a softirq. (notice the 's'). But we also see that interrupts
643when entering the preempt off section and leaving it (the 'd'). 719have been disabled when entering the preempt off section and
644We do not know if interrupts were enabled in the mean time. 720leaving it (the 'd'). We do not know if interrupts were enabled
721in the mean time.
645 722
646# tracer: preemptoff 723# tracer: preemptoff
647# 724#
@@ -700,28 +777,30 @@ preemptoff latency trace v1.1.5 on 2.6.26-rc8
700 sshd-4261 0d.s1 64us : trace_preempt_on (__do_softirq) 777 sshd-4261 0d.s1 64us : trace_preempt_on (__do_softirq)
701 778
702 779
703The above is an example of the preemptoff trace with ftrace_enabled 780The above is an example of the preemptoff trace with
704set. Here we see that interrupts were disabled the entire time. 781ftrace_enabled set. Here we see that interrupts were disabled
705The irq_enter code lets us know that we entered an interrupt 'h'. 782the entire time. The irq_enter code lets us know that we entered
706Before that, the functions being traced still show that it is not 783an interrupt 'h'. Before that, the functions being traced still
707in an interrupt, but we can see from the functions themselves that 784show that it is not in an interrupt, but we can see from the
708this is not the case. 785functions themselves that this is not the case.
709 786
710Notice that __do_softirq when called does not have a preempt_count. 787Notice that __do_softirq when called does not have a
711It may seem that we missed a preempt enabling. What really happened 788preempt_count. It may seem that we missed a preempt enabling.
712is that the preempt count is held on the thread's stack and we 789What really happened is that the preempt count is held on the
713switched to the softirq stack (4K stacks in effect). The code 790thread's stack and we switched to the softirq stack (4K stacks
714does not copy the preempt count, but because interrupts are disabled, 791in effect). The code does not copy the preempt count, but
715we do not need to worry about it. Having a tracer like this is good 792because interrupts are disabled, we do not need to worry about
716for letting people know what really happens inside the kernel. 793it. Having a tracer like this is good for letting people know
794what really happens inside the kernel.
717 795
718 796
719preemptirqsoff 797preemptirqsoff
720-------------- 798--------------
721 799
722Knowing the locations that have interrupts disabled or preemption 800Knowing the locations that have interrupts disabled or
723disabled for the longest times is helpful. But sometimes we would 801preemption disabled for the longest times is helpful. But
724like to know when either preemption and/or interrupts are disabled. 802sometimes we would like to know when either preemption and/or
803interrupts are disabled.
725 804
726Consider the following code: 805Consider the following code:
727 806
@@ -741,11 +820,13 @@ The preemptoff tracer will record the total length of
741call_function_with_irqs_and_preemption_off() and 820call_function_with_irqs_and_preemption_off() and
742call_function_with_preemption_off(). 821call_function_with_preemption_off().
743 822
744But neither will trace the time that interrupts and/or preemption 823But neither will trace the time that interrupts and/or
745is disabled. This total time is the time that we can not schedule. 824preemption is disabled. This total time is the time that we can
746To record this time, use the preemptirqsoff tracer. 825not schedule. To record this time, use the preemptirqsoff
826tracer.
747 827
748Again, using this trace is much like the irqsoff and preemptoff tracers. 828Again, using this trace is much like the irqsoff and preemptoff
829tracers.
749 830
750 # echo preemptirqsoff > /debug/tracing/current_tracer 831 # echo preemptirqsoff > /debug/tracing/current_tracer
751 # echo 0 > /debug/tracing/tracing_max_latency 832 # echo 0 > /debug/tracing/tracing_max_latency
@@ -781,9 +862,10 @@ preemptirqsoff latency trace v1.1.5 on 2.6.26-rc8
781 862
782 863
783The trace_hardirqs_off_thunk is called from assembly on x86 when 864The trace_hardirqs_off_thunk is called from assembly on x86 when
784interrupts are disabled in the assembly code. Without the function 865interrupts are disabled in the assembly code. Without the
785tracing, we do not know if interrupts were enabled within the preemption 866function tracing, we do not know if interrupts were enabled
786points. We do see that it started with preemption enabled. 867within the preemption points. We do see that it started with
868preemption enabled.
787 869
788Here is a trace with ftrace_enabled set: 870Here is a trace with ftrace_enabled set:
789 871
@@ -871,40 +953,42 @@ preemptirqsoff latency trace v1.1.5 on 2.6.26-rc8
871 sshd-4261 0d.s1 105us : trace_preempt_on (__do_softirq) 953 sshd-4261 0d.s1 105us : trace_preempt_on (__do_softirq)
872 954
873 955
874This is a very interesting trace. It started with the preemption of 956This is a very interesting trace. It started with the preemption
875the ls task. We see that the task had the "need_resched" bit set 957of the ls task. We see that the task had the "need_resched" bit
876via the 'N' in the trace. Interrupts were disabled before the spin_lock 958set via the 'N' in the trace. Interrupts were disabled before
877at the beginning of the trace. We see that a schedule took place to run 959the spin_lock at the beginning of the trace. We see that a
878sshd. When the interrupts were enabled, we took an interrupt. 960schedule took place to run sshd. When the interrupts were
879On return from the interrupt handler, the softirq ran. We took another 961enabled, we took an interrupt. On return from the interrupt
880interrupt while running the softirq as we see from the capital 'H'. 962handler, the softirq ran. We took another interrupt while
963running the softirq as we see from the capital 'H'.
881 964
882 965
883wakeup 966wakeup
884------ 967------
885 968
886In a Real-Time environment it is very important to know the wakeup 969In a Real-Time environment it is very important to know the
887time it takes for the highest priority task that is woken up to the 970wakeup time it takes for the highest priority task that is woken
888time that it executes. This is also known as "schedule latency". 971up to the time that it executes. This is also known as "schedule
889I stress the point that this is about RT tasks. It is also important 972latency". I stress the point that this is about RT tasks. It is
890to know the scheduling latency of non-RT tasks, but the average 973also important to know the scheduling latency of non-RT tasks,
891schedule latency is better for non-RT tasks. Tools like 974but the average schedule latency is better for non-RT tasks.
892LatencyTop are more appropriate for such measurements. 975Tools like LatencyTop are more appropriate for such
976measurements.
893 977
894Real-Time environments are interested in the worst case latency. 978Real-Time environments are interested in the worst case latency.
895That is the longest latency it takes for something to happen, and 979That is the longest latency it takes for something to happen,
896not the average. We can have a very fast scheduler that may only 980and not the average. We can have a very fast scheduler that may
897have a large latency once in a while, but that would not work well 981only have a large latency once in a while, but that would not
898with Real-Time tasks. The wakeup tracer was designed to record 982work well with Real-Time tasks. The wakeup tracer was designed
899the worst case wakeups of RT tasks. Non-RT tasks are not recorded 983to record the worst case wakeups of RT tasks. Non-RT tasks are
900because the tracer only records one worst case and tracing non-RT 984not recorded because the tracer only records one worst case and
901tasks that are unpredictable will overwrite the worst case latency 985tracing non-RT tasks that are unpredictable will overwrite the
902of RT tasks. 986worst case latency of RT tasks.
903 987
904Since this tracer only deals with RT tasks, we will run this slightly 988Since this tracer only deals with RT tasks, we will run this
905differently than we did with the previous tracers. Instead of performing 989slightly differently than we did with the previous tracers.
906an 'ls', we will run 'sleep 1' under 'chrt' which changes the 990Instead of performing an 'ls', we will run 'sleep 1' under
907priority of the task. 991'chrt' which changes the priority of the task.
908 992
909 # echo wakeup > /debug/tracing/current_tracer 993 # echo wakeup > /debug/tracing/current_tracer
910 # echo 0 > /debug/tracing/tracing_max_latency 994 # echo 0 > /debug/tracing/tracing_max_latency
@@ -934,17 +1018,16 @@ wakeup latency trace v1.1.5 on 2.6.26-rc8
934 <idle>-0 1d..4 4us : schedule (cpu_idle) 1018 <idle>-0 1d..4 4us : schedule (cpu_idle)
935 1019
936 1020
1021Running this on an idle system, we see that it only took 4
1022microseconds to perform the task switch. Note, since the trace
1023marker in the schedule is before the actual "switch", we stop
1024the tracing when the recorded task is about to schedule in. This
1025may change if we add a new marker at the end of the scheduler.
937 1026
938Running this on an idle system, we see that it only took 4 microseconds 1027Notice that the recorded task is 'sleep' with the PID of 4901
939to perform the task switch. Note, since the trace marker in the 1028and it has an rt_prio of 5. This priority is user-space priority
940schedule is before the actual "switch", we stop the tracing when 1029and not the internal kernel priority. The policy is 1 for
941the recorded task is about to schedule in. This may change if 1030SCHED_FIFO and 2 for SCHED_RR.
942we add a new marker at the end of the scheduler.
943
944Notice that the recorded task is 'sleep' with the PID of 4901 and it
945has an rt_prio of 5. This priority is user-space priority and not
946the internal kernel priority. The policy is 1 for SCHED_FIFO and 2
947for SCHED_RR.
948 1031
949Doing the same with chrt -r 5 and ftrace_enabled set. 1032Doing the same with chrt -r 5 and ftrace_enabled set.
950 1033
@@ -1001,24 +1084,25 @@ ksoftirq-7 1d..6 49us : _spin_unlock (tracing_record_cmdline)
1001ksoftirq-7 1d..6 49us : sub_preempt_count (_spin_unlock) 1084ksoftirq-7 1d..6 49us : sub_preempt_count (_spin_unlock)
1002ksoftirq-7 1d..4 50us : schedule (__cond_resched) 1085ksoftirq-7 1d..4 50us : schedule (__cond_resched)
1003 1086
1004The interrupt went off while running ksoftirqd. This task runs at 1087The interrupt went off while running ksoftirqd. This task runs
1005SCHED_OTHER. Why did not we see the 'N' set early? This may be 1088at SCHED_OTHER. Why did not we see the 'N' set early? This may
1006a harmless bug with x86_32 and 4K stacks. On x86_32 with 4K stacks 1089be a harmless bug with x86_32 and 4K stacks. On x86_32 with 4K
1007configured, the interrupt and softirq run with their own stack. 1090stacks configured, the interrupt and softirq run with their own
1008Some information is held on the top of the task's stack (need_resched 1091stack. Some information is held on the top of the task's stack
1009and preempt_count are both stored there). The setting of the NEED_RESCHED 1092(need_resched and preempt_count are both stored there). The
1010bit is done directly to the task's stack, but the reading of the 1093setting of the NEED_RESCHED bit is done directly to the task's
1011NEED_RESCHED is done by looking at the current stack, which in this case 1094stack, but the reading of the NEED_RESCHED is done by looking at
1012is the stack for the hard interrupt. This hides the fact that NEED_RESCHED 1095the current stack, which in this case is the stack for the hard
1013has been set. We do not see the 'N' until we switch back to the task's 1096interrupt. This hides the fact that NEED_RESCHED has been set.
1097We do not see the 'N' until we switch back to the task's
1014assigned stack. 1098assigned stack.
1015 1099
1016function 1100function
1017-------- 1101--------
1018 1102
1019This tracer is the function tracer. Enabling the function tracer 1103This tracer is the function tracer. Enabling the function tracer
1020can be done from the debug file system. Make sure the ftrace_enabled is 1104can be done from the debug file system. Make sure the
1021set; otherwise this tracer is a nop. 1105ftrace_enabled is set; otherwise this tracer is a nop.
1022 1106
1023 # sysctl kernel.ftrace_enabled=1 1107 # sysctl kernel.ftrace_enabled=1
1024 # echo function > /debug/tracing/current_tracer 1108 # echo function > /debug/tracing/current_tracer
@@ -1048,14 +1132,15 @@ set; otherwise this tracer is a nop.
1048[...] 1132[...]
1049 1133
1050 1134
1051Note: function tracer uses ring buffers to store the above entries. 1135Note: function tracer uses ring buffers to store the above
1052The newest data may overwrite the oldest data. Sometimes using echo to 1136entries. The newest data may overwrite the oldest data.
1053stop the trace is not sufficient because the tracing could have overwritten 1137Sometimes using echo to stop the trace is not sufficient because
1054the data that you wanted to record. For this reason, it is sometimes better to 1138the tracing could have overwritten the data that you wanted to
1055disable tracing directly from a program. This allows you to stop the 1139record. For this reason, it is sometimes better to disable
1056tracing at the point that you hit the part that you are interested in. 1140tracing directly from a program. This allows you to stop the
1057To disable the tracing directly from a C program, something like following 1141tracing at the point that you hit the part that you are
1058code snippet can be used: 1142interested in. To disable the tracing directly from a C program,
1143something like following code snippet can be used:
1059 1144
1060int trace_fd; 1145int trace_fd;
1061[...] 1146[...]
@@ -1070,10 +1155,10 @@ int main(int argc, char *argv[]) {
1070} 1155}
1071 1156
1072Note: Here we hard coded the path name. The debugfs mount is not 1157Note: Here we hard coded the path name. The debugfs mount is not
1073guaranteed to be at /debug (and is more commonly at /sys/kernel/debug). 1158guaranteed to be at /debug (and is more commonly at
1074For simple one time traces, the above is sufficent. For anything else, 1159/sys/kernel/debug). For simple one time traces, the above is
1075a search through /proc/mounts may be needed to find where the debugfs 1160sufficent. For anything else, a search through /proc/mounts may
1076file-system is mounted. 1161be needed to find where the debugfs file-system is mounted.
1077 1162
1078 1163
1079Single thread tracing 1164Single thread tracing
@@ -1152,49 +1237,297 @@ int main (int argc, char **argv)
1152 return 0; 1237 return 0;
1153} 1238}
1154 1239
1240
1241hw-branch-tracer (x86 only)
1242---------------------------
1243
1244This tracer uses the x86 last branch tracing hardware feature to
1245collect a branch trace on all cpus with relatively low overhead.
1246
1247The tracer uses a fixed-size circular buffer per cpu and only
1248traces ring 0 branches. The trace file dumps that buffer in the
1249following format:
1250
1251# tracer: hw-branch-tracer
1252#
1253# CPU# TO <- FROM
1254 0 scheduler_tick+0xb5/0x1bf <- task_tick_idle+0x5/0x6
1255 2 run_posix_cpu_timers+0x2b/0x72a <- run_posix_cpu_timers+0x25/0x72a
1256 0 scheduler_tick+0x139/0x1bf <- scheduler_tick+0xed/0x1bf
1257 0 scheduler_tick+0x17c/0x1bf <- scheduler_tick+0x148/0x1bf
1258 2 run_posix_cpu_timers+0x9e/0x72a <- run_posix_cpu_timers+0x5e/0x72a
1259 0 scheduler_tick+0x1b6/0x1bf <- scheduler_tick+0x1aa/0x1bf
1260
1261
1262The tracer may be used to dump the trace for the oops'ing cpu on
1263a kernel oops into the system log. To enable this,
1264ftrace_dump_on_oops must be set. To set ftrace_dump_on_oops, one
1265can either use the sysctl function or set it via the proc system
1266interface.
1267
1268 sysctl kernel.ftrace_dump_on_oops=1
1269
1270or
1271
1272 echo 1 > /proc/sys/kernel/ftrace_dump_on_oops
1273
1274
1275Here's an example of such a dump after a null pointer
1276dereference in a kernel module:
1277
1278[57848.105921] BUG: unable to handle kernel NULL pointer dereference at 0000000000000000
1279[57848.106019] IP: [<ffffffffa0000006>] open+0x6/0x14 [oops]
1280[57848.106019] PGD 2354e9067 PUD 2375e7067 PMD 0
1281[57848.106019] Oops: 0002 [#1] SMP
1282[57848.106019] last sysfs file: /sys/devices/pci0000:00/0000:00:1e.0/0000:20:05.0/local_cpus
1283[57848.106019] Dumping ftrace buffer:
1284[57848.106019] ---------------------------------
1285[...]
1286[57848.106019] 0 chrdev_open+0xe6/0x165 <- cdev_put+0x23/0x24
1287[57848.106019] 0 chrdev_open+0x117/0x165 <- chrdev_open+0xfa/0x165
1288[57848.106019] 0 chrdev_open+0x120/0x165 <- chrdev_open+0x11c/0x165
1289[57848.106019] 0 chrdev_open+0x134/0x165 <- chrdev_open+0x12b/0x165
1290[57848.106019] 0 open+0x0/0x14 [oops] <- chrdev_open+0x144/0x165
1291[57848.106019] 0 page_fault+0x0/0x30 <- open+0x6/0x14 [oops]
1292[57848.106019] 0 error_entry+0x0/0x5b <- page_fault+0x4/0x30
1293[57848.106019] 0 error_kernelspace+0x0/0x31 <- error_entry+0x59/0x5b
1294[57848.106019] 0 error_sti+0x0/0x1 <- error_kernelspace+0x2d/0x31
1295[57848.106019] 0 page_fault+0x9/0x30 <- error_sti+0x0/0x1
1296[57848.106019] 0 do_page_fault+0x0/0x881 <- page_fault+0x1a/0x30
1297[...]
1298[57848.106019] 0 do_page_fault+0x66b/0x881 <- is_prefetch+0x1ee/0x1f2
1299[57848.106019] 0 do_page_fault+0x6e0/0x881 <- do_page_fault+0x67a/0x881
1300[57848.106019] 0 oops_begin+0x0/0x96 <- do_page_fault+0x6e0/0x881
1301[57848.106019] 0 trace_hw_branch_oops+0x0/0x2d <- oops_begin+0x9/0x96
1302[...]
1303[57848.106019] 0 ds_suspend_bts+0x2a/0xe3 <- ds_suspend_bts+0x1a/0xe3
1304[57848.106019] ---------------------------------
1305[57848.106019] CPU 0
1306[57848.106019] Modules linked in: oops
1307[57848.106019] Pid: 5542, comm: cat Tainted: G W 2.6.28 #23
1308[57848.106019] RIP: 0010:[<ffffffffa0000006>] [<ffffffffa0000006>] open+0x6/0x14 [oops]
1309[57848.106019] RSP: 0018:ffff880235457d48 EFLAGS: 00010246
1310[...]
1311
1312
1313function graph tracer
1314---------------------------
1315
1316This tracer is similar to the function tracer except that it
1317probes a function on its entry and its exit. This is done by
1318using a dynamically allocated stack of return addresses in each
1319task_struct. On function entry the tracer overwrites the return
1320address of each function traced to set a custom probe. Thus the
1321original return address is stored on the stack of return address
1322in the task_struct.
1323
1324Probing on both ends of a function leads to special features
1325such as:
1326
1327- measure of a function's time execution
1328- having a reliable call stack to draw function calls graph
1329
1330This tracer is useful in several situations:
1331
1332- you want to find the reason of a strange kernel behavior and
1333 need to see what happens in detail on any areas (or specific
1334 ones).
1335
1336- you are experiencing weird latencies but it's difficult to
1337 find its origin.
1338
1339- you want to find quickly which path is taken by a specific
1340 function
1341
1342- you just want to peek inside a working kernel and want to see
1343 what happens there.
1344
1345# tracer: function_graph
1346#
1347# CPU DURATION FUNCTION CALLS
1348# | | | | | | |
1349
1350 0) | sys_open() {
1351 0) | do_sys_open() {
1352 0) | getname() {
1353 0) | kmem_cache_alloc() {
1354 0) 1.382 us | __might_sleep();
1355 0) 2.478 us | }
1356 0) | strncpy_from_user() {
1357 0) | might_fault() {
1358 0) 1.389 us | __might_sleep();
1359 0) 2.553 us | }
1360 0) 3.807 us | }
1361 0) 7.876 us | }
1362 0) | alloc_fd() {
1363 0) 0.668 us | _spin_lock();
1364 0) 0.570 us | expand_files();
1365 0) 0.586 us | _spin_unlock();
1366
1367
1368There are several columns that can be dynamically
1369enabled/disabled. You can use every combination of options you
1370want, depending on your needs.
1371
1372- The cpu number on which the function executed is default
1373 enabled. It is sometimes better to only trace one cpu (see
1374 tracing_cpu_mask file) or you might sometimes see unordered
1375 function calls while cpu tracing switch.
1376
1377 hide: echo nofuncgraph-cpu > /debug/tracing/trace_options
1378 show: echo funcgraph-cpu > /debug/tracing/trace_options
1379
1380- The duration (function's time of execution) is displayed on
1381 the closing bracket line of a function or on the same line
1382 than the current function in case of a leaf one. It is default
1383 enabled.
1384
1385 hide: echo nofuncgraph-duration > /debug/tracing/trace_options
1386 show: echo funcgraph-duration > /debug/tracing/trace_options
1387
1388- The overhead field precedes the duration field in case of
1389 reached duration thresholds.
1390
1391 hide: echo nofuncgraph-overhead > /debug/tracing/trace_options
1392 show: echo funcgraph-overhead > /debug/tracing/trace_options
1393 depends on: funcgraph-duration
1394
1395 ie:
1396
1397 0) | up_write() {
1398 0) 0.646 us | _spin_lock_irqsave();
1399 0) 0.684 us | _spin_unlock_irqrestore();
1400 0) 3.123 us | }
1401 0) 0.548 us | fput();
1402 0) + 58.628 us | }
1403
1404 [...]
1405
1406 0) | putname() {
1407 0) | kmem_cache_free() {
1408 0) 0.518 us | __phys_addr();
1409 0) 1.757 us | }
1410 0) 2.861 us | }
1411 0) ! 115.305 us | }
1412 0) ! 116.402 us | }
1413
1414 + means that the function exceeded 10 usecs.
1415 ! means that the function exceeded 100 usecs.
1416
1417
1418- The task/pid field displays the thread cmdline and pid which
1419 executed the function. It is default disabled.
1420
1421 hide: echo nofuncgraph-proc > /debug/tracing/trace_options
1422 show: echo funcgraph-proc > /debug/tracing/trace_options
1423
1424 ie:
1425
1426 # tracer: function_graph
1427 #
1428 # CPU TASK/PID DURATION FUNCTION CALLS
1429 # | | | | | | | | |
1430 0) sh-4802 | | d_free() {
1431 0) sh-4802 | | call_rcu() {
1432 0) sh-4802 | | __call_rcu() {
1433 0) sh-4802 | 0.616 us | rcu_process_gp_end();
1434 0) sh-4802 | 0.586 us | check_for_new_grace_period();
1435 0) sh-4802 | 2.899 us | }
1436 0) sh-4802 | 4.040 us | }
1437 0) sh-4802 | 5.151 us | }
1438 0) sh-4802 | + 49.370 us | }
1439
1440
1441- The absolute time field is an absolute timestamp given by the
1442 system clock since it started. A snapshot of this time is
1443 given on each entry/exit of functions
1444
1445 hide: echo nofuncgraph-abstime > /debug/tracing/trace_options
1446 show: echo funcgraph-abstime > /debug/tracing/trace_options
1447
1448 ie:
1449
1450 #
1451 # TIME CPU DURATION FUNCTION CALLS
1452 # | | | | | | | |
1453 360.774522 | 1) 0.541 us | }
1454 360.774522 | 1) 4.663 us | }
1455 360.774523 | 1) 0.541 us | __wake_up_bit();
1456 360.774524 | 1) 6.796 us | }
1457 360.774524 | 1) 7.952 us | }
1458 360.774525 | 1) 9.063 us | }
1459 360.774525 | 1) 0.615 us | journal_mark_dirty();
1460 360.774527 | 1) 0.578 us | __brelse();
1461 360.774528 | 1) | reiserfs_prepare_for_journal() {
1462 360.774528 | 1) | unlock_buffer() {
1463 360.774529 | 1) | wake_up_bit() {
1464 360.774529 | 1) | bit_waitqueue() {
1465 360.774530 | 1) 0.594 us | __phys_addr();
1466
1467
1468You can put some comments on specific functions by using
1469trace_printk() For example, if you want to put a comment inside
1470the __might_sleep() function, you just have to include
1471<linux/ftrace.h> and call trace_printk() inside __might_sleep()
1472
1473trace_printk("I'm a comment!\n")
1474
1475will produce:
1476
1477 1) | __might_sleep() {
1478 1) | /* I'm a comment! */
1479 1) 1.449 us | }
1480
1481
1482You might find other useful features for this tracer in the
1483following "dynamic ftrace" section such as tracing only specific
1484functions or tasks.
1485
1155dynamic ftrace 1486dynamic ftrace
1156-------------- 1487--------------
1157 1488
1158If CONFIG_DYNAMIC_FTRACE is set, the system will run with 1489If CONFIG_DYNAMIC_FTRACE is set, the system will run with
1159virtually no overhead when function tracing is disabled. The way 1490virtually no overhead when function tracing is disabled. The way
1160this works is the mcount function call (placed at the start of 1491this works is the mcount function call (placed at the start of
1161every kernel function, produced by the -pg switch in gcc), starts 1492every kernel function, produced by the -pg switch in gcc),
1162of pointing to a simple return. (Enabling FTRACE will include the 1493starts of pointing to a simple return. (Enabling FTRACE will
1163-pg switch in the compiling of the kernel.) 1494include the -pg switch in the compiling of the kernel.)
1164 1495
1165At compile time every C file object is run through the 1496At compile time every C file object is run through the
1166recordmcount.pl script (located in the scripts directory). This 1497recordmcount.pl script (located in the scripts directory). This
1167script will process the C object using objdump to find all the 1498script will process the C object using objdump to find all the
1168locations in the .text section that call mcount. (Note, only 1499locations in the .text section that call mcount. (Note, only the
1169the .text section is processed, since processing other sections 1500.text section is processed, since processing other sections like
1170like .init.text may cause races due to those sections being freed). 1501.init.text may cause races due to those sections being freed).
1171 1502
1172A new section called "__mcount_loc" is created that holds references 1503A new section called "__mcount_loc" is created that holds
1173to all the mcount call sites in the .text section. This section is 1504references to all the mcount call sites in the .text section.
1174compiled back into the original object. The final linker will add 1505This section is compiled back into the original object. The
1175all these references into a single table. 1506final linker will add all these references into a single table.
1176 1507
1177On boot up, before SMP is initialized, the dynamic ftrace code 1508On boot up, before SMP is initialized, the dynamic ftrace code
1178scans this table and updates all the locations into nops. It also 1509scans this table and updates all the locations into nops. It
1179records the locations, which are added to the available_filter_functions 1510also records the locations, which are added to the
1180list. Modules are processed as they are loaded and before they are 1511available_filter_functions list. Modules are processed as they
1181executed. When a module is unloaded, it also removes its functions from 1512are loaded and before they are executed. When a module is
1182the ftrace function list. This is automatic in the module unload 1513unloaded, it also removes its functions from the ftrace function
1183code, and the module author does not need to worry about it. 1514list. This is automatic in the module unload code, and the
1184 1515module author does not need to worry about it.
1185When tracing is enabled, kstop_machine is called to prevent races 1516
1186with the CPUS executing code being modified (which can cause the 1517When tracing is enabled, kstop_machine is called to prevent
1187CPU to do undesireable things), and the nops are patched back 1518races with the CPUS executing code being modified (which can
1188to calls. But this time, they do not call mcount (which is just 1519cause the CPU to do undesireable things), and the nops are
1189a function stub). They now call into the ftrace infrastructure. 1520patched back to calls. But this time, they do not call mcount
1521(which is just a function stub). They now call into the ftrace
1522infrastructure.
1190 1523
1191One special side-effect to the recording of the functions being 1524One special side-effect to the recording of the functions being
1192traced is that we can now selectively choose which functions we 1525traced is that we can now selectively choose which functions we
1193wish to trace and which ones we want the mcount calls to remain as 1526wish to trace and which ones we want the mcount calls to remain
1194nops. 1527as nops.
1195 1528
1196Two files are used, one for enabling and one for disabling the tracing 1529Two files are used, one for enabling and one for disabling the
1197of specified functions. They are: 1530tracing of specified functions. They are:
1198 1531
1199 set_ftrace_filter 1532 set_ftrace_filter
1200 1533
@@ -1202,8 +1535,8 @@ and
1202 1535
1203 set_ftrace_notrace 1536 set_ftrace_notrace
1204 1537
1205A list of available functions that you can add to these files is listed 1538A list of available functions that you can add to these files is
1206in: 1539listed in:
1207 1540
1208 available_filter_functions 1541 available_filter_functions
1209 1542
@@ -1240,8 +1573,8 @@ hrtimer_interrupt
1240sys_nanosleep 1573sys_nanosleep
1241 1574
1242 1575
1243Perhaps this is not enough. The filters also allow simple wild cards. 1576Perhaps this is not enough. The filters also allow simple wild
1244Only the following are currently available 1577cards. Only the following are currently available
1245 1578
1246 <match>* - will match functions that begin with <match> 1579 <match>* - will match functions that begin with <match>
1247 *<match> - will match functions that end with <match> 1580 *<match> - will match functions that end with <match>
@@ -1251,9 +1584,9 @@ These are the only wild cards which are supported.
1251 1584
1252 <match>*<match> will not work. 1585 <match>*<match> will not work.
1253 1586
1254Note: It is better to use quotes to enclose the wild cards, otherwise 1587Note: It is better to use quotes to enclose the wild cards,
1255 the shell may expand the parameters into names of files in the local 1588 otherwise the shell may expand the parameters into names
1256 directory. 1589 of files in the local directory.
1257 1590
1258 # echo 'hrtimer_*' > /debug/tracing/set_ftrace_filter 1591 # echo 'hrtimer_*' > /debug/tracing/set_ftrace_filter
1259 1592
@@ -1299,7 +1632,8 @@ This is because the '>' and '>>' act just like they do in bash.
1299To rewrite the filters, use '>' 1632To rewrite the filters, use '>'
1300To append to the filters, use '>>' 1633To append to the filters, use '>>'
1301 1634
1302To clear out a filter so that all functions will be recorded again: 1635To clear out a filter so that all functions will be recorded
1636again:
1303 1637
1304 # echo > /debug/tracing/set_ftrace_filter 1638 # echo > /debug/tracing/set_ftrace_filter
1305 # cat /debug/tracing/set_ftrace_filter 1639 # cat /debug/tracing/set_ftrace_filter
@@ -1331,7 +1665,8 @@ hrtimer_get_res
1331hrtimer_init_sleeper 1665hrtimer_init_sleeper
1332 1666
1333 1667
1334The set_ftrace_notrace prevents those functions from being traced. 1668The set_ftrace_notrace prevents those functions from being
1669traced.
1335 1670
1336 # echo '*preempt*' '*lock*' > /debug/tracing/set_ftrace_notrace 1671 # echo '*preempt*' '*lock*' > /debug/tracing/set_ftrace_notrace
1337 1672
@@ -1353,13 +1688,75 @@ Produces:
1353 1688
1354We can see that there's no more lock or preempt tracing. 1689We can see that there's no more lock or preempt tracing.
1355 1690
1691
1692Dynamic ftrace with the function graph tracer
1693---------------------------------------------
1694
1695Although what has been explained above concerns both the
1696function tracer and the function-graph-tracer, there are some
1697special features only available in the function-graph tracer.
1698
1699If you want to trace only one function and all of its children,
1700you just have to echo its name into set_graph_function:
1701
1702 echo __do_fault > set_graph_function
1703
1704will produce the following "expanded" trace of the __do_fault()
1705function:
1706
1707 0) | __do_fault() {
1708 0) | filemap_fault() {
1709 0) | find_lock_page() {
1710 0) 0.804 us | find_get_page();
1711 0) | __might_sleep() {
1712 0) 1.329 us | }
1713 0) 3.904 us | }
1714 0) 4.979 us | }
1715 0) 0.653 us | _spin_lock();
1716 0) 0.578 us | page_add_file_rmap();
1717 0) 0.525 us | native_set_pte_at();
1718 0) 0.585 us | _spin_unlock();
1719 0) | unlock_page() {
1720 0) 0.541 us | page_waitqueue();
1721 0) 0.639 us | __wake_up_bit();
1722 0) 2.786 us | }
1723 0) + 14.237 us | }
1724 0) | __do_fault() {
1725 0) | filemap_fault() {
1726 0) | find_lock_page() {
1727 0) 0.698 us | find_get_page();
1728 0) | __might_sleep() {
1729 0) 1.412 us | }
1730 0) 3.950 us | }
1731 0) 5.098 us | }
1732 0) 0.631 us | _spin_lock();
1733 0) 0.571 us | page_add_file_rmap();
1734 0) 0.526 us | native_set_pte_at();
1735 0) 0.586 us | _spin_unlock();
1736 0) | unlock_page() {
1737 0) 0.533 us | page_waitqueue();
1738 0) 0.638 us | __wake_up_bit();
1739 0) 2.793 us | }
1740 0) + 14.012 us | }
1741
1742You can also expand several functions at once:
1743
1744 echo sys_open > set_graph_function
1745 echo sys_close >> set_graph_function
1746
1747Now if you want to go back to trace all functions you can clear
1748this special filter via:
1749
1750 echo > set_graph_function
1751
1752
1356trace_pipe 1753trace_pipe
1357---------- 1754----------
1358 1755
1359The trace_pipe outputs the same content as the trace file, but the effect 1756The trace_pipe outputs the same content as the trace file, but
1360on the tracing is different. Every read from trace_pipe is consumed. 1757the effect on the tracing is different. Every read from
1361This means that subsequent reads will be different. The trace 1758trace_pipe is consumed. This means that subsequent reads will be
1362is live. 1759different. The trace is live.
1363 1760
1364 # echo function > /debug/tracing/current_tracer 1761 # echo function > /debug/tracing/current_tracer
1365 # cat /debug/tracing/trace_pipe > /tmp/trace.out & 1762 # cat /debug/tracing/trace_pipe > /tmp/trace.out &
@@ -1387,38 +1784,45 @@ is live.
1387 bash-4043 [00] 41.267111: select_task_rq_rt <-try_to_wake_up 1784 bash-4043 [00] 41.267111: select_task_rq_rt <-try_to_wake_up
1388 1785
1389 1786
1390Note, reading the trace_pipe file will block until more input is added. 1787Note, reading the trace_pipe file will block until more input is
1391By changing the tracer, trace_pipe will issue an EOF. We needed 1788added. By changing the tracer, trace_pipe will issue an EOF. We
1392to set the function tracer _before_ we "cat" the trace_pipe file. 1789needed to set the function tracer _before_ we "cat" the
1790trace_pipe file.
1393 1791
1394 1792
1395trace entries 1793trace entries
1396------------- 1794-------------
1397 1795
1398Having too much or not enough data can be troublesome in diagnosing 1796Having too much or not enough data can be troublesome in
1399an issue in the kernel. The file buffer_size_kb is used to modify 1797diagnosing an issue in the kernel. The file buffer_size_kb is
1400the size of the internal trace buffers. The number listed 1798used to modify the size of the internal trace buffers. The
1401is the number of entries that can be recorded per CPU. To know 1799number listed is the number of entries that can be recorded per
1402the full size, multiply the number of possible CPUS with the 1800CPU. To know the full size, multiply the number of possible CPUS
1403number of entries. 1801with the number of entries.
1404 1802
1405 # cat /debug/tracing/buffer_size_kb 1803 # cat /debug/tracing/buffer_size_kb
14061408 (units kilobytes) 18041408 (units kilobytes)
1407 1805
1408Note, to modify this, you must have tracing completely disabled. To do that, 1806Note, to modify this, you must have tracing completely disabled.
1409echo "nop" into the current_tracer. If the current_tracer is not set 1807To do that, echo "nop" into the current_tracer. If the
1410to "nop", an EINVAL error will be returned. 1808current_tracer is not set to "nop", an EINVAL error will be
1809returned.
1411 1810
1412 # echo nop > /debug/tracing/current_tracer 1811 # echo nop > /debug/tracing/current_tracer
1413 # echo 10000 > /debug/tracing/buffer_size_kb 1812 # echo 10000 > /debug/tracing/buffer_size_kb
1414 # cat /debug/tracing/buffer_size_kb 1813 # cat /debug/tracing/buffer_size_kb
141510000 (units kilobytes) 181410000 (units kilobytes)
1416 1815
1417The number of pages which will be allocated is limited to a percentage 1816The number of pages which will be allocated is limited to a
1418of available memory. Allocating too much will produce an error. 1817percentage of available memory. Allocating too much will produce
1818an error.
1419 1819
1420 # echo 1000000000000 > /debug/tracing/buffer_size_kb 1820 # echo 1000000000000 > /debug/tracing/buffer_size_kb
1421-bash: echo: write error: Cannot allocate memory 1821-bash: echo: write error: Cannot allocate memory
1422 # cat /debug/tracing/buffer_size_kb 1822 # cat /debug/tracing/buffer_size_kb
142385 182385
1424 1824
1825-----------
1826
1827More details can be found in the source code, in the
1828kernel/tracing/*.c files.