Loading...
1#
2# Architectures that offer an FUNCTION_TRACER implementation should
3# select HAVE_FUNCTION_TRACER:
4#
5
6config USER_STACKTRACE_SUPPORT
7 bool
8
9config NOP_TRACER
10 bool
11
12config HAVE_FTRACE_NMI_ENTER
13 bool
14 help
15 See Documentation/trace/ftrace-design.txt
16
17config HAVE_FUNCTION_TRACER
18 bool
19 help
20 See Documentation/trace/ftrace-design.txt
21
22config HAVE_FUNCTION_GRAPH_TRACER
23 bool
24 help
25 See Documentation/trace/ftrace-design.txt
26
27config HAVE_FUNCTION_GRAPH_FP_TEST
28 bool
29 help
30 See Documentation/trace/ftrace-design.txt
31
32config HAVE_FUNCTION_TRACE_MCOUNT_TEST
33 bool
34 help
35 See Documentation/trace/ftrace-design.txt
36
37config HAVE_DYNAMIC_FTRACE
38 bool
39 help
40 See Documentation/trace/ftrace-design.txt
41
42config HAVE_DYNAMIC_FTRACE_WITH_REGS
43 bool
44
45config HAVE_FTRACE_MCOUNT_RECORD
46 bool
47 help
48 See Documentation/trace/ftrace-design.txt
49
50config HAVE_SYSCALL_TRACEPOINTS
51 bool
52 help
53 See Documentation/trace/ftrace-design.txt
54
55config HAVE_FENTRY
56 bool
57 help
58 Arch supports the gcc options -pg with -mfentry
59
60config HAVE_C_RECORDMCOUNT
61 bool
62 help
63 C version of recordmcount available?
64
65config TRACER_MAX_TRACE
66 bool
67
68config TRACE_CLOCK
69 bool
70
71config RING_BUFFER
72 bool
73 select TRACE_CLOCK
74 select IRQ_WORK
75
76config FTRACE_NMI_ENTER
77 bool
78 depends on HAVE_FTRACE_NMI_ENTER
79 default y
80
81config EVENT_TRACING
82 select CONTEXT_SWITCH_TRACER
83 bool
84
85config CONTEXT_SWITCH_TRACER
86 bool
87
88config RING_BUFFER_ALLOW_SWAP
89 bool
90 help
91 Allow the use of ring_buffer_swap_cpu.
92 Adds a very slight overhead to tracing when enabled.
93
94# All tracer options should select GENERIC_TRACER. For those options that are
95# enabled by all tracers (context switch and event tracer) they select TRACING.
96# This allows those options to appear when no other tracer is selected. But the
97# options do not appear when something else selects it. We need the two options
98# GENERIC_TRACER and TRACING to avoid circular dependencies to accomplish the
99# hiding of the automatic options.
100
101config TRACING
102 bool
103 select DEBUG_FS
104 select RING_BUFFER
105 select STACKTRACE if STACKTRACE_SUPPORT
106 select TRACEPOINTS
107 select NOP_TRACER
108 select BINARY_PRINTF
109 select EVENT_TRACING
110 select TRACE_CLOCK
111
112config GENERIC_TRACER
113 bool
114 select TRACING
115
116#
117# Minimum requirements an architecture has to meet for us to
118# be able to offer generic tracing facilities:
119#
120config TRACING_SUPPORT
121 bool
122 # PPC32 has no irqflags tracing support, but it can use most of the
123 # tracers anyway, they were tested to build and work. Note that new
124 # exceptions to this list aren't welcomed, better implement the
125 # irqflags tracing for your architecture.
126 depends on TRACE_IRQFLAGS_SUPPORT || PPC32
127 depends on STACKTRACE_SUPPORT
128 default y
129
130if TRACING_SUPPORT
131
132menuconfig FTRACE
133 bool "Tracers"
134 default y if DEBUG_KERNEL
135 help
136 Enable the kernel tracing infrastructure.
137
138if FTRACE
139
140config FUNCTION_TRACER
141 bool "Kernel Function Tracer"
142 depends on HAVE_FUNCTION_TRACER
143 select KALLSYMS
144 select GENERIC_TRACER
145 select CONTEXT_SWITCH_TRACER
146 help
147 Enable the kernel to trace every kernel function. This is done
148 by using a compiler feature to insert a small, 5-byte No-Operation
149 instruction at the beginning of every kernel function, which NOP
150 sequence is then dynamically patched into a tracer call when
151 tracing is enabled by the administrator. If it's runtime disabled
152 (the bootup default), then the overhead of the instructions is very
153 small and not measurable even in micro-benchmarks.
154
155config FUNCTION_GRAPH_TRACER
156 bool "Kernel Function Graph Tracer"
157 depends on HAVE_FUNCTION_GRAPH_TRACER
158 depends on FUNCTION_TRACER
159 depends on !X86_32 || !CC_OPTIMIZE_FOR_SIZE
160 default y
161 help
162 Enable the kernel to trace a function at both its return
163 and its entry.
164 Its first purpose is to trace the duration of functions and
165 draw a call graph for each thread with some information like
166 the return value. This is done by setting the current return
167 address on the current task structure into a stack of calls.
168
169
170config IRQSOFF_TRACER
171 bool "Interrupts-off Latency Tracer"
172 default n
173 depends on TRACE_IRQFLAGS_SUPPORT
174 depends on !ARCH_USES_GETTIMEOFFSET
175 select TRACE_IRQFLAGS
176 select GENERIC_TRACER
177 select TRACER_MAX_TRACE
178 select RING_BUFFER_ALLOW_SWAP
179 select TRACER_SNAPSHOT
180 select TRACER_SNAPSHOT_PER_CPU_SWAP
181 help
182 This option measures the time spent in irqs-off critical
183 sections, with microsecond accuracy.
184
185 The default measurement method is a maximum search, which is
186 disabled by default and can be runtime (re-)started
187 via:
188
189 echo 0 > /sys/kernel/debug/tracing/tracing_max_latency
190
191 (Note that kernel size and overhead increase with this option
192 enabled. This option and the preempt-off timing option can be
193 used together or separately.)
194
195config PREEMPT_TRACER
196 bool "Preemption-off Latency Tracer"
197 default n
198 depends on !ARCH_USES_GETTIMEOFFSET
199 depends on PREEMPT
200 select GENERIC_TRACER
201 select TRACER_MAX_TRACE
202 select RING_BUFFER_ALLOW_SWAP
203 select TRACER_SNAPSHOT
204 select TRACER_SNAPSHOT_PER_CPU_SWAP
205 help
206 This option measures the time spent in preemption-off critical
207 sections, with microsecond accuracy.
208
209 The default measurement method is a maximum search, which is
210 disabled by default and can be runtime (re-)started
211 via:
212
213 echo 0 > /sys/kernel/debug/tracing/tracing_max_latency
214
215 (Note that kernel size and overhead increase with this option
216 enabled. This option and the irqs-off timing option can be
217 used together or separately.)
218
219config SCHED_TRACER
220 bool "Scheduling Latency Tracer"
221 select GENERIC_TRACER
222 select CONTEXT_SWITCH_TRACER
223 select TRACER_MAX_TRACE
224 select TRACER_SNAPSHOT
225 help
226 This tracer tracks the latency of the highest priority task
227 to be scheduled in, starting from the point it has woken up.
228
229config ENABLE_DEFAULT_TRACERS
230 bool "Trace process context switches and events"
231 depends on !GENERIC_TRACER
232 select TRACING
233 help
234 This tracer hooks to various trace points in the kernel,
235 allowing the user to pick and choose which trace point they
236 want to trace. It also includes the sched_switch tracer plugin.
237
238config FTRACE_SYSCALLS
239 bool "Trace syscalls"
240 depends on HAVE_SYSCALL_TRACEPOINTS
241 select GENERIC_TRACER
242 select KALLSYMS
243 help
244 Basic tracer to catch the syscall entry and exit events.
245
246config TRACER_SNAPSHOT
247 bool "Create a snapshot trace buffer"
248 select TRACER_MAX_TRACE
249 help
250 Allow tracing users to take snapshot of the current buffer using the
251 ftrace interface, e.g.:
252
253 echo 1 > /sys/kernel/debug/tracing/snapshot
254 cat snapshot
255
256config TRACER_SNAPSHOT_PER_CPU_SWAP
257 bool "Allow snapshot to swap per CPU"
258 depends on TRACER_SNAPSHOT
259 select RING_BUFFER_ALLOW_SWAP
260 help
261 Allow doing a snapshot of a single CPU buffer instead of a
262 full swap (all buffers). If this is set, then the following is
263 allowed:
264
265 echo 1 > /sys/kernel/debug/tracing/per_cpu/cpu2/snapshot
266
267 After which, only the tracing buffer for CPU 2 was swapped with
268 the main tracing buffer, and the other CPU buffers remain the same.
269
270 When this is enabled, this adds a little more overhead to the
271 trace recording, as it needs to add some checks to synchronize
272 recording with swaps. But this does not affect the performance
273 of the overall system. This is enabled by default when the preempt
274 or irq latency tracers are enabled, as those need to swap as well
275 and already adds the overhead (plus a lot more).
276
277config TRACE_BRANCH_PROFILING
278 bool
279 select GENERIC_TRACER
280
281choice
282 prompt "Branch Profiling"
283 default BRANCH_PROFILE_NONE
284 help
285 The branch profiling is a software profiler. It will add hooks
286 into the C conditionals to test which path a branch takes.
287
288 The likely/unlikely profiler only looks at the conditions that
289 are annotated with a likely or unlikely macro.
290
291 The "all branch" profiler will profile every if-statement in the
292 kernel. This profiler will also enable the likely/unlikely
293 profiler.
294
295 Either of the above profilers adds a bit of overhead to the system.
296 If unsure, choose "No branch profiling".
297
298config BRANCH_PROFILE_NONE
299 bool "No branch profiling"
300 help
301 No branch profiling. Branch profiling adds a bit of overhead.
302 Only enable it if you want to analyse the branching behavior.
303 Otherwise keep it disabled.
304
305config PROFILE_ANNOTATED_BRANCHES
306 bool "Trace likely/unlikely profiler"
307 select TRACE_BRANCH_PROFILING
308 help
309 This tracer profiles all likely and unlikely macros
310 in the kernel. It will display the results in:
311
312 /sys/kernel/debug/tracing/trace_stat/branch_annotated
313
314 Note: this will add a significant overhead; only turn this
315 on if you need to profile the system's use of these macros.
316
317config PROFILE_ALL_BRANCHES
318 bool "Profile all if conditionals"
319 select TRACE_BRANCH_PROFILING
320 help
321 This tracer profiles all branch conditions. Every if ()
322 taken in the kernel is recorded whether it hit or miss.
323 The results will be displayed in:
324
325 /sys/kernel/debug/tracing/trace_stat/branch_all
326
327 This option also enables the likely/unlikely profiler.
328
329 This configuration, when enabled, will impose a great overhead
330 on the system. This should only be enabled when the system
331 is to be analyzed in much detail.
332endchoice
333
334config TRACING_BRANCHES
335 bool
336 help
337 Selected by tracers that will trace the likely and unlikely
338 conditions. This prevents the tracers themselves from being
339 profiled. Profiling the tracing infrastructure can only happen
340 when the likelys and unlikelys are not being traced.
341
342config BRANCH_TRACER
343 bool "Trace likely/unlikely instances"
344 depends on TRACE_BRANCH_PROFILING
345 select TRACING_BRANCHES
346 help
347 This traces the events of likely and unlikely condition
348 calls in the kernel. The difference between this and the
349 "Trace likely/unlikely profiler" is that this is not a
350 histogram of the callers, but actually places the calling
351 events into a running trace buffer to see when and where the
352 events happened, as well as their results.
353
354 Say N if unsure.
355
356config STACK_TRACER
357 bool "Trace max stack"
358 depends on HAVE_FUNCTION_TRACER
359 select FUNCTION_TRACER
360 select STACKTRACE
361 select KALLSYMS
362 help
363 This special tracer records the maximum stack footprint of the
364 kernel and displays it in /sys/kernel/debug/tracing/stack_trace.
365
366 This tracer works by hooking into every function call that the
367 kernel executes, and keeping a maximum stack depth value and
368 stack-trace saved. If this is configured with DYNAMIC_FTRACE
369 then it will not have any overhead while the stack tracer
370 is disabled.
371
372 To enable the stack tracer on bootup, pass in 'stacktrace'
373 on the kernel command line.
374
375 The stack tracer can also be enabled or disabled via the
376 sysctl kernel.stack_tracer_enabled
377
378 Say N if unsure.
379
380config BLK_DEV_IO_TRACE
381 bool "Support for tracing block IO actions"
382 depends on SYSFS
383 depends on BLOCK
384 select RELAY
385 select DEBUG_FS
386 select TRACEPOINTS
387 select GENERIC_TRACER
388 select STACKTRACE
389 help
390 Say Y here if you want to be able to trace the block layer actions
391 on a given queue. Tracing allows you to see any traffic happening
392 on a block device queue. For more information (and the userspace
393 support tools needed), fetch the blktrace tools from:
394
395 git://git.kernel.dk/blktrace.git
396
397 Tracing also is possible using the ftrace interface, e.g.:
398
399 echo 1 > /sys/block/sda/sda1/trace/enable
400 echo blk > /sys/kernel/debug/tracing/current_tracer
401 cat /sys/kernel/debug/tracing/trace_pipe
402
403 If unsure, say N.
404
405config KPROBE_EVENT
406 depends on KPROBES
407 depends on HAVE_REGS_AND_STACK_ACCESS_API
408 bool "Enable kprobes-based dynamic events"
409 select TRACING
410 select PROBE_EVENTS
411 default y
412 help
413 This allows the user to add tracing events (similar to tracepoints)
414 on the fly via the ftrace interface. See
415 Documentation/trace/kprobetrace.txt for more details.
416
417 Those events can be inserted wherever kprobes can probe, and record
418 various register and memory values.
419
420 This option is also required by perf-probe subcommand of perf tools.
421 If you want to use perf tools, this option is strongly recommended.
422
423config UPROBE_EVENT
424 bool "Enable uprobes-based dynamic events"
425 depends on ARCH_SUPPORTS_UPROBES
426 depends on MMU
427 depends on PERF_EVENTS
428 select UPROBES
429 select PROBE_EVENTS
430 select TRACING
431 default n
432 help
433 This allows the user to add tracing events on top of userspace
434 dynamic events (similar to tracepoints) on the fly via the trace
435 events interface. Those events can be inserted wherever uprobes
436 can probe, and record various registers.
437 This option is required if you plan to use perf-probe subcommand
438 of perf tools on user space applications.
439
440config PROBE_EVENTS
441 def_bool n
442
443config DYNAMIC_FTRACE
444 bool "enable/disable function tracing dynamically"
445 depends on FUNCTION_TRACER
446 depends on HAVE_DYNAMIC_FTRACE
447 default y
448 help
449 This option will modify all the calls to function tracing
450 dynamically (will patch them out of the binary image and
451 replace them with a No-Op instruction) on boot up. During
452 compile time, a table is made of all the locations that ftrace
453 can function trace, and this table is linked into the kernel
454 image. When this is enabled, functions can be individually
455 enabled, and the functions not enabled will not affect
456 performance of the system.
457
458 See the files in /sys/kernel/debug/tracing:
459 available_filter_functions
460 set_ftrace_filter
461 set_ftrace_notrace
462
463 This way a CONFIG_FUNCTION_TRACER kernel is slightly larger, but
464 otherwise has native performance as long as no tracing is active.
465
466config DYNAMIC_FTRACE_WITH_REGS
467 def_bool y
468 depends on DYNAMIC_FTRACE
469 depends on HAVE_DYNAMIC_FTRACE_WITH_REGS
470
471config FUNCTION_PROFILER
472 bool "Kernel function profiler"
473 depends on FUNCTION_TRACER
474 default n
475 help
476 This option enables the kernel function profiler. A file is created
477 in debugfs called function_profile_enabled which defaults to zero.
478 When a 1 is echoed into this file profiling begins, and when a
479 zero is entered, profiling stops. A "functions" file is created in
480 the trace_stats directory; this file shows the list of functions that
481 have been hit and their counters.
482
483 If in doubt, say N.
484
485config FTRACE_MCOUNT_RECORD
486 def_bool y
487 depends on DYNAMIC_FTRACE
488 depends on HAVE_FTRACE_MCOUNT_RECORD
489
490config FTRACE_SELFTEST
491 bool
492
493config FTRACE_STARTUP_TEST
494 bool "Perform a startup test on ftrace"
495 depends on GENERIC_TRACER
496 select FTRACE_SELFTEST
497 help
498 This option performs a series of startup tests on ftrace. On bootup
499 a series of tests are made to verify that the tracer is
500 functioning properly. It will do tests on all the configured
501 tracers of ftrace.
502
503config EVENT_TRACE_TEST_SYSCALLS
504 bool "Run selftest on syscall events"
505 depends on FTRACE_STARTUP_TEST
506 help
507 This option will also enable testing every syscall event.
508 It only enables the event and disables it and runs various loads
509 with the event enabled. This adds a bit more time for kernel boot
510 up since it runs this on every system call defined.
511
512 TBD - enable a way to actually call the syscalls as we test their
513 events
514
515config MMIOTRACE
516 bool "Memory mapped IO tracing"
517 depends on HAVE_MMIOTRACE_SUPPORT && PCI
518 select GENERIC_TRACER
519 help
520 Mmiotrace traces Memory Mapped I/O access and is meant for
521 debugging and reverse engineering. It is called from the ioremap
522 implementation and works via page faults. Tracing is disabled by
523 default and can be enabled at run-time.
524
525 See Documentation/trace/mmiotrace.txt.
526 If you are not helping to develop drivers, say N.
527
528config MMIOTRACE_TEST
529 tristate "Test module for mmiotrace"
530 depends on MMIOTRACE && m
531 help
532 This is a dumb module for testing mmiotrace. It is very dangerous
533 as it will write garbage to IO memory starting at a given address.
534 However, it should be safe to use on e.g. unused portion of VRAM.
535
536 Say N, unless you absolutely know what you are doing.
537
538config RING_BUFFER_BENCHMARK
539 tristate "Ring buffer benchmark stress tester"
540 depends on RING_BUFFER
541 help
542 This option creates a test to stress the ring buffer and benchmark it.
543 It creates its own ring buffer such that it will not interfere with
544 any other users of the ring buffer (such as ftrace). It then creates
545 a producer and consumer that will run for 10 seconds and sleep for
546 10 seconds. Each interval it will print out the number of events
547 it recorded and give a rough estimate of how long each iteration took.
548
549 It does not disable interrupts or raise its priority, so it may be
550 affected by processes that are running.
551
552 If unsure, say N.
553
554config RING_BUFFER_STARTUP_TEST
555 bool "Ring buffer startup self test"
556 depends on RING_BUFFER
557 help
558 Run a simple self test on the ring buffer on boot up. Late in the
559 kernel boot sequence, the test will start that kicks off
560 a thread per cpu. Each thread will write various size events
561 into the ring buffer. Another thread is created to send IPIs
562 to each of the threads, where the IPI handler will also write
563 to the ring buffer, to test/stress the nesting ability.
564 If any anomalies are discovered, a warning will be displayed
565 and all ring buffers will be disabled.
566
567 The test runs for 10 seconds. This will slow your boot time
568 by at least 10 more seconds.
569
570 At the end of the test, statics and more checks are done.
571 It will output the stats of each per cpu buffer. What
572 was written, the sizes, what was read, what was lost, and
573 other similar details.
574
575 If unsure, say N
576
577endif # FTRACE
578
579endif # TRACING_SUPPORT
580
1# SPDX-License-Identifier: GPL-2.0-only
2#
3# Architectures that offer an FUNCTION_TRACER implementation should
4# select HAVE_FUNCTION_TRACER:
5#
6
7config USER_STACKTRACE_SUPPORT
8 bool
9
10config NOP_TRACER
11 bool
12
13config HAVE_FTRACE_NMI_ENTER
14 bool
15 help
16 See Documentation/trace/ftrace-design.rst
17
18config HAVE_FUNCTION_TRACER
19 bool
20 help
21 See Documentation/trace/ftrace-design.rst
22
23config HAVE_FUNCTION_GRAPH_TRACER
24 bool
25 help
26 See Documentation/trace/ftrace-design.rst
27
28config HAVE_DYNAMIC_FTRACE
29 bool
30 help
31 See Documentation/trace/ftrace-design.rst
32
33config HAVE_DYNAMIC_FTRACE_WITH_REGS
34 bool
35
36config HAVE_FTRACE_MCOUNT_RECORD
37 bool
38 help
39 See Documentation/trace/ftrace-design.rst
40
41config HAVE_SYSCALL_TRACEPOINTS
42 bool
43 help
44 See Documentation/trace/ftrace-design.rst
45
46config HAVE_FENTRY
47 bool
48 help
49 Arch supports the gcc options -pg with -mfentry
50
51config HAVE_NOP_MCOUNT
52 bool
53 help
54 Arch supports the gcc options -pg with -mrecord-mcount and -nop-mcount
55
56config HAVE_C_RECORDMCOUNT
57 bool
58 help
59 C version of recordmcount available?
60
61config TRACER_MAX_TRACE
62 bool
63
64config TRACE_CLOCK
65 bool
66
67config RING_BUFFER
68 bool
69 select TRACE_CLOCK
70 select IRQ_WORK
71
72config FTRACE_NMI_ENTER
73 bool
74 depends on HAVE_FTRACE_NMI_ENTER
75 default y
76
77config EVENT_TRACING
78 select CONTEXT_SWITCH_TRACER
79 select GLOB
80 bool
81
82config CONTEXT_SWITCH_TRACER
83 bool
84
85config RING_BUFFER_ALLOW_SWAP
86 bool
87 help
88 Allow the use of ring_buffer_swap_cpu.
89 Adds a very slight overhead to tracing when enabled.
90
91config PREEMPTIRQ_TRACEPOINTS
92 bool
93 depends on TRACE_PREEMPT_TOGGLE || TRACE_IRQFLAGS
94 select TRACING
95 default y
96 help
97 Create preempt/irq toggle tracepoints if needed, so that other parts
98 of the kernel can use them to generate or add hooks to them.
99
100# All tracer options should select GENERIC_TRACER. For those options that are
101# enabled by all tracers (context switch and event tracer) they select TRACING.
102# This allows those options to appear when no other tracer is selected. But the
103# options do not appear when something else selects it. We need the two options
104# GENERIC_TRACER and TRACING to avoid circular dependencies to accomplish the
105# hiding of the automatic options.
106
107config TRACING
108 bool
109 select DEBUG_FS
110 select RING_BUFFER
111 select STACKTRACE if STACKTRACE_SUPPORT
112 select TRACEPOINTS
113 select NOP_TRACER
114 select BINARY_PRINTF
115 select EVENT_TRACING
116 select TRACE_CLOCK
117
118config GENERIC_TRACER
119 bool
120 select TRACING
121
122#
123# Minimum requirements an architecture has to meet for us to
124# be able to offer generic tracing facilities:
125#
126config TRACING_SUPPORT
127 bool
128 depends on TRACE_IRQFLAGS_SUPPORT
129 depends on STACKTRACE_SUPPORT
130 default y
131
132if TRACING_SUPPORT
133
134menuconfig FTRACE
135 bool "Tracers"
136 default y if DEBUG_KERNEL
137 help
138 Enable the kernel tracing infrastructure.
139
140if FTRACE
141
142config FUNCTION_TRACER
143 bool "Kernel Function Tracer"
144 depends on HAVE_FUNCTION_TRACER
145 select KALLSYMS
146 select GENERIC_TRACER
147 select CONTEXT_SWITCH_TRACER
148 select GLOB
149 select TASKS_RCU if PREEMPTION
150 help
151 Enable the kernel to trace every kernel function. This is done
152 by using a compiler feature to insert a small, 5-byte No-Operation
153 instruction at the beginning of every kernel function, which NOP
154 sequence is then dynamically patched into a tracer call when
155 tracing is enabled by the administrator. If it's runtime disabled
156 (the bootup default), then the overhead of the instructions is very
157 small and not measurable even in micro-benchmarks.
158
159config FUNCTION_GRAPH_TRACER
160 bool "Kernel Function Graph Tracer"
161 depends on HAVE_FUNCTION_GRAPH_TRACER
162 depends on FUNCTION_TRACER
163 depends on !X86_32 || !CC_OPTIMIZE_FOR_SIZE
164 default y
165 help
166 Enable the kernel to trace a function at both its return
167 and its entry.
168 Its first purpose is to trace the duration of functions and
169 draw a call graph for each thread with some information like
170 the return value. This is done by setting the current return
171 address on the current task structure into a stack of calls.
172
173config TRACE_PREEMPT_TOGGLE
174 bool
175 help
176 Enables hooks which will be called when preemption is first disabled,
177 and last enabled.
178
179config PREEMPTIRQ_EVENTS
180 bool "Enable trace events for preempt and irq disable/enable"
181 select TRACE_IRQFLAGS
182 select TRACE_PREEMPT_TOGGLE if PREEMPTION
183 select GENERIC_TRACER
184 default n
185 help
186 Enable tracing of disable and enable events for preemption and irqs.
187
188config IRQSOFF_TRACER
189 bool "Interrupts-off Latency Tracer"
190 default n
191 depends on TRACE_IRQFLAGS_SUPPORT
192 depends on !ARCH_USES_GETTIMEOFFSET
193 select TRACE_IRQFLAGS
194 select GENERIC_TRACER
195 select TRACER_MAX_TRACE
196 select RING_BUFFER_ALLOW_SWAP
197 select TRACER_SNAPSHOT
198 select TRACER_SNAPSHOT_PER_CPU_SWAP
199 help
200 This option measures the time spent in irqs-off critical
201 sections, with microsecond accuracy.
202
203 The default measurement method is a maximum search, which is
204 disabled by default and can be runtime (re-)started
205 via:
206
207 echo 0 > /sys/kernel/debug/tracing/tracing_max_latency
208
209 (Note that kernel size and overhead increase with this option
210 enabled. This option and the preempt-off timing option can be
211 used together or separately.)
212
213config PREEMPT_TRACER
214 bool "Preemption-off Latency Tracer"
215 default n
216 depends on !ARCH_USES_GETTIMEOFFSET
217 depends on PREEMPTION
218 select GENERIC_TRACER
219 select TRACER_MAX_TRACE
220 select RING_BUFFER_ALLOW_SWAP
221 select TRACER_SNAPSHOT
222 select TRACER_SNAPSHOT_PER_CPU_SWAP
223 select TRACE_PREEMPT_TOGGLE
224 help
225 This option measures the time spent in preemption-off critical
226 sections, with microsecond accuracy.
227
228 The default measurement method is a maximum search, which is
229 disabled by default and can be runtime (re-)started
230 via:
231
232 echo 0 > /sys/kernel/debug/tracing/tracing_max_latency
233
234 (Note that kernel size and overhead increase with this option
235 enabled. This option and the irqs-off timing option can be
236 used together or separately.)
237
238config SCHED_TRACER
239 bool "Scheduling Latency Tracer"
240 select GENERIC_TRACER
241 select CONTEXT_SWITCH_TRACER
242 select TRACER_MAX_TRACE
243 select TRACER_SNAPSHOT
244 help
245 This tracer tracks the latency of the highest priority task
246 to be scheduled in, starting from the point it has woken up.
247
248config HWLAT_TRACER
249 bool "Tracer to detect hardware latencies (like SMIs)"
250 select GENERIC_TRACER
251 help
252 This tracer, when enabled will create one or more kernel threads,
253 depending on what the cpumask file is set to, which each thread
254 spinning in a loop looking for interruptions caused by
255 something other than the kernel. For example, if a
256 System Management Interrupt (SMI) takes a noticeable amount of
257 time, this tracer will detect it. This is useful for testing
258 if a system is reliable for Real Time tasks.
259
260 Some files are created in the tracing directory when this
261 is enabled:
262
263 hwlat_detector/width - time in usecs for how long to spin for
264 hwlat_detector/window - time in usecs between the start of each
265 iteration
266
267 A kernel thread is created that will spin with interrupts disabled
268 for "width" microseconds in every "window" cycle. It will not spin
269 for "window - width" microseconds, where the system can
270 continue to operate.
271
272 The output will appear in the trace and trace_pipe files.
273
274 When the tracer is not running, it has no affect on the system,
275 but when it is running, it can cause the system to be
276 periodically non responsive. Do not run this tracer on a
277 production system.
278
279 To enable this tracer, echo in "hwlat" into the current_tracer
280 file. Every time a latency is greater than tracing_thresh, it will
281 be recorded into the ring buffer.
282
283config ENABLE_DEFAULT_TRACERS
284 bool "Trace process context switches and events"
285 depends on !GENERIC_TRACER
286 select TRACING
287 help
288 This tracer hooks to various trace points in the kernel,
289 allowing the user to pick and choose which trace point they
290 want to trace. It also includes the sched_switch tracer plugin.
291
292config FTRACE_SYSCALLS
293 bool "Trace syscalls"
294 depends on HAVE_SYSCALL_TRACEPOINTS
295 select GENERIC_TRACER
296 select KALLSYMS
297 help
298 Basic tracer to catch the syscall entry and exit events.
299
300config TRACER_SNAPSHOT
301 bool "Create a snapshot trace buffer"
302 select TRACER_MAX_TRACE
303 help
304 Allow tracing users to take snapshot of the current buffer using the
305 ftrace interface, e.g.:
306
307 echo 1 > /sys/kernel/debug/tracing/snapshot
308 cat snapshot
309
310config TRACER_SNAPSHOT_PER_CPU_SWAP
311 bool "Allow snapshot to swap per CPU"
312 depends on TRACER_SNAPSHOT
313 select RING_BUFFER_ALLOW_SWAP
314 help
315 Allow doing a snapshot of a single CPU buffer instead of a
316 full swap (all buffers). If this is set, then the following is
317 allowed:
318
319 echo 1 > /sys/kernel/debug/tracing/per_cpu/cpu2/snapshot
320
321 After which, only the tracing buffer for CPU 2 was swapped with
322 the main tracing buffer, and the other CPU buffers remain the same.
323
324 When this is enabled, this adds a little more overhead to the
325 trace recording, as it needs to add some checks to synchronize
326 recording with swaps. But this does not affect the performance
327 of the overall system. This is enabled by default when the preempt
328 or irq latency tracers are enabled, as those need to swap as well
329 and already adds the overhead (plus a lot more).
330
331config TRACE_BRANCH_PROFILING
332 bool
333 select GENERIC_TRACER
334
335choice
336 prompt "Branch Profiling"
337 default BRANCH_PROFILE_NONE
338 help
339 The branch profiling is a software profiler. It will add hooks
340 into the C conditionals to test which path a branch takes.
341
342 The likely/unlikely profiler only looks at the conditions that
343 are annotated with a likely or unlikely macro.
344
345 The "all branch" profiler will profile every if-statement in the
346 kernel. This profiler will also enable the likely/unlikely
347 profiler.
348
349 Either of the above profilers adds a bit of overhead to the system.
350 If unsure, choose "No branch profiling".
351
352config BRANCH_PROFILE_NONE
353 bool "No branch profiling"
354 help
355 No branch profiling. Branch profiling adds a bit of overhead.
356 Only enable it if you want to analyse the branching behavior.
357 Otherwise keep it disabled.
358
359config PROFILE_ANNOTATED_BRANCHES
360 bool "Trace likely/unlikely profiler"
361 select TRACE_BRANCH_PROFILING
362 help
363 This tracer profiles all likely and unlikely macros
364 in the kernel. It will display the results in:
365
366 /sys/kernel/debug/tracing/trace_stat/branch_annotated
367
368 Note: this will add a significant overhead; only turn this
369 on if you need to profile the system's use of these macros.
370
371config PROFILE_ALL_BRANCHES
372 bool "Profile all if conditionals" if !FORTIFY_SOURCE
373 select TRACE_BRANCH_PROFILING
374 imply CC_DISABLE_WARN_MAYBE_UNINITIALIZED # avoid false positives
375 help
376 This tracer profiles all branch conditions. Every if ()
377 taken in the kernel is recorded whether it hit or miss.
378 The results will be displayed in:
379
380 /sys/kernel/debug/tracing/trace_stat/branch_all
381
382 This option also enables the likely/unlikely profiler.
383
384 This configuration, when enabled, will impose a great overhead
385 on the system. This should only be enabled when the system
386 is to be analyzed in much detail.
387endchoice
388
389config TRACING_BRANCHES
390 bool
391 help
392 Selected by tracers that will trace the likely and unlikely
393 conditions. This prevents the tracers themselves from being
394 profiled. Profiling the tracing infrastructure can only happen
395 when the likelys and unlikelys are not being traced.
396
397config BRANCH_TRACER
398 bool "Trace likely/unlikely instances"
399 depends on TRACE_BRANCH_PROFILING
400 select TRACING_BRANCHES
401 help
402 This traces the events of likely and unlikely condition
403 calls in the kernel. The difference between this and the
404 "Trace likely/unlikely profiler" is that this is not a
405 histogram of the callers, but actually places the calling
406 events into a running trace buffer to see when and where the
407 events happened, as well as their results.
408
409 Say N if unsure.
410
411config STACK_TRACER
412 bool "Trace max stack"
413 depends on HAVE_FUNCTION_TRACER
414 select FUNCTION_TRACER
415 select STACKTRACE
416 select KALLSYMS
417 help
418 This special tracer records the maximum stack footprint of the
419 kernel and displays it in /sys/kernel/debug/tracing/stack_trace.
420
421 This tracer works by hooking into every function call that the
422 kernel executes, and keeping a maximum stack depth value and
423 stack-trace saved. If this is configured with DYNAMIC_FTRACE
424 then it will not have any overhead while the stack tracer
425 is disabled.
426
427 To enable the stack tracer on bootup, pass in 'stacktrace'
428 on the kernel command line.
429
430 The stack tracer can also be enabled or disabled via the
431 sysctl kernel.stack_tracer_enabled
432
433 Say N if unsure.
434
435config BLK_DEV_IO_TRACE
436 bool "Support for tracing block IO actions"
437 depends on SYSFS
438 depends on BLOCK
439 select RELAY
440 select DEBUG_FS
441 select TRACEPOINTS
442 select GENERIC_TRACER
443 select STACKTRACE
444 help
445 Say Y here if you want to be able to trace the block layer actions
446 on a given queue. Tracing allows you to see any traffic happening
447 on a block device queue. For more information (and the userspace
448 support tools needed), fetch the blktrace tools from:
449
450 git://git.kernel.dk/blktrace.git
451
452 Tracing also is possible using the ftrace interface, e.g.:
453
454 echo 1 > /sys/block/sda/sda1/trace/enable
455 echo blk > /sys/kernel/debug/tracing/current_tracer
456 cat /sys/kernel/debug/tracing/trace_pipe
457
458 If unsure, say N.
459
460config KPROBE_EVENTS
461 depends on KPROBES
462 depends on HAVE_REGS_AND_STACK_ACCESS_API
463 bool "Enable kprobes-based dynamic events"
464 select TRACING
465 select PROBE_EVENTS
466 select DYNAMIC_EVENTS
467 default y
468 help
469 This allows the user to add tracing events (similar to tracepoints)
470 on the fly via the ftrace interface. See
471 Documentation/trace/kprobetrace.rst for more details.
472
473 Those events can be inserted wherever kprobes can probe, and record
474 various register and memory values.
475
476 This option is also required by perf-probe subcommand of perf tools.
477 If you want to use perf tools, this option is strongly recommended.
478
479config KPROBE_EVENTS_ON_NOTRACE
480 bool "Do NOT protect notrace function from kprobe events"
481 depends on KPROBE_EVENTS
482 depends on KPROBES_ON_FTRACE
483 default n
484 help
485 This is only for the developers who want to debug ftrace itself
486 using kprobe events.
487
488 If kprobes can use ftrace instead of breakpoint, ftrace related
489 functions are protected from kprobe-events to prevent an infinit
490 recursion or any unexpected execution path which leads to a kernel
491 crash.
492
493 This option disables such protection and allows you to put kprobe
494 events on ftrace functions for debugging ftrace by itself.
495 Note that this might let you shoot yourself in the foot.
496
497 If unsure, say N.
498
499config UPROBE_EVENTS
500 bool "Enable uprobes-based dynamic events"
501 depends on ARCH_SUPPORTS_UPROBES
502 depends on MMU
503 depends on PERF_EVENTS
504 select UPROBES
505 select PROBE_EVENTS
506 select DYNAMIC_EVENTS
507 select TRACING
508 default y
509 help
510 This allows the user to add tracing events on top of userspace
511 dynamic events (similar to tracepoints) on the fly via the trace
512 events interface. Those events can be inserted wherever uprobes
513 can probe, and record various registers.
514 This option is required if you plan to use perf-probe subcommand
515 of perf tools on user space applications.
516
517config BPF_EVENTS
518 depends on BPF_SYSCALL
519 depends on (KPROBE_EVENTS || UPROBE_EVENTS) && PERF_EVENTS
520 bool
521 default y
522 help
523 This allows the user to attach BPF programs to kprobe, uprobe, and
524 tracepoint events.
525
526config DYNAMIC_EVENTS
527 def_bool n
528
529config PROBE_EVENTS
530 def_bool n
531
532config DYNAMIC_FTRACE
533 bool "enable/disable function tracing dynamically"
534 depends on FUNCTION_TRACER
535 depends on HAVE_DYNAMIC_FTRACE
536 default y
537 help
538 This option will modify all the calls to function tracing
539 dynamically (will patch them out of the binary image and
540 replace them with a No-Op instruction) on boot up. During
541 compile time, a table is made of all the locations that ftrace
542 can function trace, and this table is linked into the kernel
543 image. When this is enabled, functions can be individually
544 enabled, and the functions not enabled will not affect
545 performance of the system.
546
547 See the files in /sys/kernel/debug/tracing:
548 available_filter_functions
549 set_ftrace_filter
550 set_ftrace_notrace
551
552 This way a CONFIG_FUNCTION_TRACER kernel is slightly larger, but
553 otherwise has native performance as long as no tracing is active.
554
555config DYNAMIC_FTRACE_WITH_REGS
556 def_bool y
557 depends on DYNAMIC_FTRACE
558 depends on HAVE_DYNAMIC_FTRACE_WITH_REGS
559
560config FUNCTION_PROFILER
561 bool "Kernel function profiler"
562 depends on FUNCTION_TRACER
563 default n
564 help
565 This option enables the kernel function profiler. A file is created
566 in debugfs called function_profile_enabled which defaults to zero.
567 When a 1 is echoed into this file profiling begins, and when a
568 zero is entered, profiling stops. A "functions" file is created in
569 the trace_stat directory; this file shows the list of functions that
570 have been hit and their counters.
571
572 If in doubt, say N.
573
574config BPF_KPROBE_OVERRIDE
575 bool "Enable BPF programs to override a kprobed function"
576 depends on BPF_EVENTS
577 depends on FUNCTION_ERROR_INJECTION
578 default n
579 help
580 Allows BPF to override the execution of a probed function and
581 set a different return value. This is used for error injection.
582
583config FTRACE_MCOUNT_RECORD
584 def_bool y
585 depends on DYNAMIC_FTRACE
586 depends on HAVE_FTRACE_MCOUNT_RECORD
587
588config FTRACE_SELFTEST
589 bool
590
591config FTRACE_STARTUP_TEST
592 bool "Perform a startup test on ftrace"
593 depends on GENERIC_TRACER
594 select FTRACE_SELFTEST
595 help
596 This option performs a series of startup tests on ftrace. On bootup
597 a series of tests are made to verify that the tracer is
598 functioning properly. It will do tests on all the configured
599 tracers of ftrace.
600
601config EVENT_TRACE_STARTUP_TEST
602 bool "Run selftest on trace events"
603 depends on FTRACE_STARTUP_TEST
604 default y
605 help
606 This option performs a test on all trace events in the system.
607 It basically just enables each event and runs some code that
608 will trigger events (not necessarily the event it enables)
609 This may take some time run as there are a lot of events.
610
611config EVENT_TRACE_TEST_SYSCALLS
612 bool "Run selftest on syscall events"
613 depends on EVENT_TRACE_STARTUP_TEST
614 help
615 This option will also enable testing every syscall event.
616 It only enables the event and disables it and runs various loads
617 with the event enabled. This adds a bit more time for kernel boot
618 up since it runs this on every system call defined.
619
620 TBD - enable a way to actually call the syscalls as we test their
621 events
622
623config MMIOTRACE
624 bool "Memory mapped IO tracing"
625 depends on HAVE_MMIOTRACE_SUPPORT && PCI
626 select GENERIC_TRACER
627 help
628 Mmiotrace traces Memory Mapped I/O access and is meant for
629 debugging and reverse engineering. It is called from the ioremap
630 implementation and works via page faults. Tracing is disabled by
631 default and can be enabled at run-time.
632
633 See Documentation/trace/mmiotrace.rst.
634 If you are not helping to develop drivers, say N.
635
636config TRACING_MAP
637 bool
638 depends on ARCH_HAVE_NMI_SAFE_CMPXCHG
639 help
640 tracing_map is a special-purpose lock-free map for tracing,
641 separated out as a stand-alone facility in order to allow it
642 to be shared between multiple tracers. It isn't meant to be
643 generally used outside of that context, and is normally
644 selected by tracers that use it.
645
646config HIST_TRIGGERS
647 bool "Histogram triggers"
648 depends on ARCH_HAVE_NMI_SAFE_CMPXCHG
649 select TRACING_MAP
650 select TRACING
651 select DYNAMIC_EVENTS
652 default n
653 help
654 Hist triggers allow one or more arbitrary trace event fields
655 to be aggregated into hash tables and dumped to stdout by
656 reading a debugfs/tracefs file. They're useful for
657 gathering quick and dirty (though precise) summaries of
658 event activity as an initial guide for further investigation
659 using more advanced tools.
660
661 Inter-event tracing of quantities such as latencies is also
662 supported using hist triggers under this option.
663
664 See Documentation/trace/histogram.rst.
665 If in doubt, say N.
666
667config MMIOTRACE_TEST
668 tristate "Test module for mmiotrace"
669 depends on MMIOTRACE && m
670 help
671 This is a dumb module for testing mmiotrace. It is very dangerous
672 as it will write garbage to IO memory starting at a given address.
673 However, it should be safe to use on e.g. unused portion of VRAM.
674
675 Say N, unless you absolutely know what you are doing.
676
677config TRACEPOINT_BENCHMARK
678 bool "Add tracepoint that benchmarks tracepoints"
679 help
680 This option creates the tracepoint "benchmark:benchmark_event".
681 When the tracepoint is enabled, it kicks off a kernel thread that
682 goes into an infinite loop (calling cond_sched() to let other tasks
683 run), and calls the tracepoint. Each iteration will record the time
684 it took to write to the tracepoint and the next iteration that
685 data will be passed to the tracepoint itself. That is, the tracepoint
686 will report the time it took to do the previous tracepoint.
687 The string written to the tracepoint is a static string of 128 bytes
688 to keep the time the same. The initial string is simply a write of
689 "START". The second string records the cold cache time of the first
690 write which is not added to the rest of the calculations.
691
692 As it is a tight loop, it benchmarks as hot cache. That's fine because
693 we care most about hot paths that are probably in cache already.
694
695 An example of the output:
696
697 START
698 first=3672 [COLD CACHED]
699 last=632 first=3672 max=632 min=632 avg=316 std=446 std^2=199712
700 last=278 first=3672 max=632 min=278 avg=303 std=316 std^2=100337
701 last=277 first=3672 max=632 min=277 avg=296 std=258 std^2=67064
702 last=273 first=3672 max=632 min=273 avg=292 std=224 std^2=50411
703 last=273 first=3672 max=632 min=273 avg=288 std=200 std^2=40389
704 last=281 first=3672 max=632 min=273 avg=287 std=183 std^2=33666
705
706
707config RING_BUFFER_BENCHMARK
708 tristate "Ring buffer benchmark stress tester"
709 depends on RING_BUFFER
710 help
711 This option creates a test to stress the ring buffer and benchmark it.
712 It creates its own ring buffer such that it will not interfere with
713 any other users of the ring buffer (such as ftrace). It then creates
714 a producer and consumer that will run for 10 seconds and sleep for
715 10 seconds. Each interval it will print out the number of events
716 it recorded and give a rough estimate of how long each iteration took.
717
718 It does not disable interrupts or raise its priority, so it may be
719 affected by processes that are running.
720
721 If unsure, say N.
722
723config RING_BUFFER_STARTUP_TEST
724 bool "Ring buffer startup self test"
725 depends on RING_BUFFER
726 help
727 Run a simple self test on the ring buffer on boot up. Late in the
728 kernel boot sequence, the test will start that kicks off
729 a thread per cpu. Each thread will write various size events
730 into the ring buffer. Another thread is created to send IPIs
731 to each of the threads, where the IPI handler will also write
732 to the ring buffer, to test/stress the nesting ability.
733 If any anomalies are discovered, a warning will be displayed
734 and all ring buffers will be disabled.
735
736 The test runs for 10 seconds. This will slow your boot time
737 by at least 10 more seconds.
738
739 At the end of the test, statics and more checks are done.
740 It will output the stats of each per cpu buffer. What
741 was written, the sizes, what was read, what was lost, and
742 other similar details.
743
744 If unsure, say N
745
746config PREEMPTIRQ_DELAY_TEST
747 tristate "Preempt / IRQ disable delay thread to test latency tracers"
748 depends on m
749 help
750 Select this option to build a test module that can help test latency
751 tracers by executing a preempt or irq disable section with a user
752 configurable delay. The module busy waits for the duration of the
753 critical section.
754
755 For example, the following invocation forces a one-time irq-disabled
756 critical section for 500us:
757 modprobe preemptirq_delay_test test_mode=irq delay=500000
758
759 If unsure, say N
760
761config TRACE_EVAL_MAP_FILE
762 bool "Show eval mappings for trace events"
763 depends on TRACING
764 help
765 The "print fmt" of the trace events will show the enum/sizeof names
766 instead of their values. This can cause problems for user space tools
767 that use this string to parse the raw data as user space does not know
768 how to convert the string to its value.
769
770 To fix this, there's a special macro in the kernel that can be used
771 to convert an enum/sizeof into its value. If this macro is used, then
772 the print fmt strings will be converted to their values.
773
774 If something does not get converted properly, this option can be
775 used to show what enums/sizeof the kernel tried to convert.
776
777 This option is for debugging the conversions. A file is created
778 in the tracing directory called "eval_map" that will show the
779 names matched with their values and what trace event system they
780 belong too.
781
782 Normally, the mapping of the strings to values will be freed after
783 boot up or module load. With this option, they will not be freed, as
784 they are needed for the "eval_map" file. Enabling this option will
785 increase the memory footprint of the running kernel.
786
787 If unsure, say N
788
789config GCOV_PROFILE_FTRACE
790 bool "Enable GCOV profiling on ftrace subsystem"
791 depends on GCOV_KERNEL
792 help
793 Enable GCOV profiling on ftrace subsystem for checking
794 which functions/lines are tested.
795
796 If unsure, say N.
797
798 Note that on a kernel compiled with this config, ftrace will
799 run significantly slower.
800
801endif # FTRACE
802
803endif # TRACING_SUPPORT
804