1 # SPDX-License-Identifier: GPL-2.0-only
3 # Architectures that offer an FUNCTION_TRACER implementation should
4 # select HAVE_FUNCTION_TRACER:
7 config USER_STACKTRACE_SUPPORT
18 depends on HAVE_RETHOOK
20 Enable generic return hooking feature. This is an internal
21 API, which will be used by other function-entry hooking
22 features like fprobe and kprobes.
24 config HAVE_FUNCTION_TRACER
27 See Documentation/trace/ftrace-design.rst
29 config HAVE_FUNCTION_GRAPH_TRACER
32 See Documentation/trace/ftrace-design.rst
34 config HAVE_DYNAMIC_FTRACE
37 See Documentation/trace/ftrace-design.rst
39 config HAVE_DYNAMIC_FTRACE_WITH_REGS
42 config HAVE_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
45 config HAVE_DYNAMIC_FTRACE_WITH_ARGS
48 If this is set, then arguments and stack can be found from
49 the pt_regs passed into the function callback regs parameter
50 by default, even without setting the REGS flag in the ftrace_ops.
51 This allows for use of regs_get_kernel_argument() and
52 kernel_stack_pointer().
54 config HAVE_FTRACE_MCOUNT_RECORD
57 See Documentation/trace/ftrace-design.rst
59 config HAVE_SYSCALL_TRACEPOINTS
62 See Documentation/trace/ftrace-design.rst
67 Arch supports the gcc options -pg with -mfentry
69 config HAVE_NOP_MCOUNT
72 Arch supports the gcc options -pg with -mrecord-mcount and -nop-mcount
74 config HAVE_OBJTOOL_MCOUNT
77 Arch supports objtool --mcount
79 config HAVE_C_RECORDMCOUNT
82 C version of recordmcount available?
84 config HAVE_BUILDTIME_MCOUNT_SORT
87 An architecture selects this if it sorts the mcount_loc section
90 config BUILDTIME_MCOUNT_SORT
93 depends on HAVE_BUILDTIME_MCOUNT_SORT && DYNAMIC_FTRACE
95 Sort the mcount_loc section at build time.
97 config TRACER_MAX_TRACE
109 select CONTEXT_SWITCH_TRACER
113 config CONTEXT_SWITCH_TRACER
116 config RING_BUFFER_ALLOW_SWAP
119 Allow the use of ring_buffer_swap_cpu.
120 Adds a very slight overhead to tracing when enabled.
122 config PREEMPTIRQ_TRACEPOINTS
124 depends on TRACE_PREEMPT_TOGGLE || TRACE_IRQFLAGS
128 Create preempt/irq toggle tracepoints if needed, so that other parts
129 of the kernel can use them to generate or add hooks to them.
131 # All tracer options should select GENERIC_TRACER. For those options that are
132 # enabled by all tracers (context switch and event tracer) they select TRACING.
133 # This allows those options to appear when no other tracer is selected. But the
134 # options do not appear when something else selects it. We need the two options
135 # GENERIC_TRACER and TRACING to avoid circular dependencies to accomplish the
136 # hiding of the automatic options.
141 select STACKTRACE if STACKTRACE_SUPPORT
147 select TASKS_RCU if PREEMPTION
149 config GENERIC_TRACER
154 # Minimum requirements an architecture has to meet for us to
155 # be able to offer generic tracing facilities:
157 config TRACING_SUPPORT
159 depends on TRACE_IRQFLAGS_SUPPORT
160 depends on STACKTRACE_SUPPORT
165 depends on TRACING_SUPPORT
166 default y if DEBUG_KERNEL
168 Enable the kernel tracing infrastructure.
172 config BOOTTIME_TRACING
173 bool "Boot-time Tracing support"
177 Enable developer to setup ftrace subsystem via supplemental
178 kernel cmdline at boot time for debugging (tracing) driver
179 initialization and boot process.
181 config FUNCTION_TRACER
182 bool "Kernel Function Tracer"
183 depends on HAVE_FUNCTION_TRACER
185 select GENERIC_TRACER
186 select CONTEXT_SWITCH_TRACER
188 select TASKS_RCU if PREEMPTION
189 select TASKS_RUDE_RCU
191 Enable the kernel to trace every kernel function. This is done
192 by using a compiler feature to insert a small, 5-byte No-Operation
193 instruction at the beginning of every kernel function, which NOP
194 sequence is then dynamically patched into a tracer call when
195 tracing is enabled by the administrator. If it's runtime disabled
196 (the bootup default), then the overhead of the instructions is very
197 small and not measurable even in micro-benchmarks (at least on
198 x86, but may have impact on other architectures).
200 config FUNCTION_GRAPH_TRACER
201 bool "Kernel Function Graph Tracer"
202 depends on HAVE_FUNCTION_GRAPH_TRACER
203 depends on FUNCTION_TRACER
204 depends on !X86_32 || !CC_OPTIMIZE_FOR_SIZE
207 Enable the kernel to trace a function at both its return
209 Its first purpose is to trace the duration of functions and
210 draw a call graph for each thread with some information like
211 the return value. This is done by setting the current return
212 address on the current task structure into a stack of calls.
214 config DYNAMIC_FTRACE
215 bool "enable/disable function tracing dynamically"
216 depends on FUNCTION_TRACER
217 depends on HAVE_DYNAMIC_FTRACE
220 This option will modify all the calls to function tracing
221 dynamically (will patch them out of the binary image and
222 replace them with a No-Op instruction) on boot up. During
223 compile time, a table is made of all the locations that ftrace
224 can function trace, and this table is linked into the kernel
225 image. When this is enabled, functions can be individually
226 enabled, and the functions not enabled will not affect
227 performance of the system.
229 See the files in /sys/kernel/debug/tracing:
230 available_filter_functions
234 This way a CONFIG_FUNCTION_TRACER kernel is slightly larger, but
235 otherwise has native performance as long as no tracing is active.
237 config DYNAMIC_FTRACE_WITH_REGS
239 depends on DYNAMIC_FTRACE
240 depends on HAVE_DYNAMIC_FTRACE_WITH_REGS
242 config DYNAMIC_FTRACE_WITH_DIRECT_CALLS
244 depends on DYNAMIC_FTRACE_WITH_REGS
245 depends on HAVE_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
247 config DYNAMIC_FTRACE_WITH_ARGS
249 depends on DYNAMIC_FTRACE
250 depends on HAVE_DYNAMIC_FTRACE_WITH_ARGS
253 bool "Kernel Function Probe (fprobe)"
254 depends on FUNCTION_TRACER
255 depends on DYNAMIC_FTRACE_WITH_REGS
256 depends on HAVE_RETHOOK
260 This option enables kernel function probe (fprobe) based on ftrace.
261 The fprobe is similar to kprobes, but probes only for kernel function
262 entries and exits. This also can probe multiple functions by one
267 config FUNCTION_PROFILER
268 bool "Kernel function profiler"
269 depends on FUNCTION_TRACER
272 This option enables the kernel function profiler. A file is created
273 in debugfs called function_profile_enabled which defaults to zero.
274 When a 1 is echoed into this file profiling begins, and when a
275 zero is entered, profiling stops. A "functions" file is created in
276 the trace_stat directory; this file shows the list of functions that
277 have been hit and their counters.
282 bool "Trace max stack"
283 depends on HAVE_FUNCTION_TRACER
284 select FUNCTION_TRACER
288 This special tracer records the maximum stack footprint of the
289 kernel and displays it in /sys/kernel/debug/tracing/stack_trace.
291 This tracer works by hooking into every function call that the
292 kernel executes, and keeping a maximum stack depth value and
293 stack-trace saved. If this is configured with DYNAMIC_FTRACE
294 then it will not have any overhead while the stack tracer
297 To enable the stack tracer on bootup, pass in 'stacktrace'
298 on the kernel command line.
300 The stack tracer can also be enabled or disabled via the
301 sysctl kernel.stack_tracer_enabled
305 config TRACE_PREEMPT_TOGGLE
308 Enables hooks which will be called when preemption is first disabled,
311 config IRQSOFF_TRACER
312 bool "Interrupts-off Latency Tracer"
314 depends on TRACE_IRQFLAGS_SUPPORT
315 select TRACE_IRQFLAGS
316 select GENERIC_TRACER
317 select TRACER_MAX_TRACE
318 select RING_BUFFER_ALLOW_SWAP
319 select TRACER_SNAPSHOT
320 select TRACER_SNAPSHOT_PER_CPU_SWAP
322 This option measures the time spent in irqs-off critical
323 sections, with microsecond accuracy.
325 The default measurement method is a maximum search, which is
326 disabled by default and can be runtime (re-)started
329 echo 0 > /sys/kernel/debug/tracing/tracing_max_latency
331 (Note that kernel size and overhead increase with this option
332 enabled. This option and the preempt-off timing option can be
333 used together or separately.)
335 config PREEMPT_TRACER
336 bool "Preemption-off Latency Tracer"
338 depends on PREEMPTION
339 select GENERIC_TRACER
340 select TRACER_MAX_TRACE
341 select RING_BUFFER_ALLOW_SWAP
342 select TRACER_SNAPSHOT
343 select TRACER_SNAPSHOT_PER_CPU_SWAP
344 select TRACE_PREEMPT_TOGGLE
346 This option measures the time spent in preemption-off critical
347 sections, with microsecond accuracy.
349 The default measurement method is a maximum search, which is
350 disabled by default and can be runtime (re-)started
353 echo 0 > /sys/kernel/debug/tracing/tracing_max_latency
355 (Note that kernel size and overhead increase with this option
356 enabled. This option and the irqs-off timing option can be
357 used together or separately.)
360 bool "Scheduling Latency Tracer"
361 select GENERIC_TRACER
362 select CONTEXT_SWITCH_TRACER
363 select TRACER_MAX_TRACE
364 select TRACER_SNAPSHOT
366 This tracer tracks the latency of the highest priority task
367 to be scheduled in, starting from the point it has woken up.
370 bool "Tracer to detect hardware latencies (like SMIs)"
371 select GENERIC_TRACER
373 This tracer, when enabled will create one or more kernel threads,
374 depending on what the cpumask file is set to, which each thread
375 spinning in a loop looking for interruptions caused by
376 something other than the kernel. For example, if a
377 System Management Interrupt (SMI) takes a noticeable amount of
378 time, this tracer will detect it. This is useful for testing
379 if a system is reliable for Real Time tasks.
381 Some files are created in the tracing directory when this
384 hwlat_detector/width - time in usecs for how long to spin for
385 hwlat_detector/window - time in usecs between the start of each
388 A kernel thread is created that will spin with interrupts disabled
389 for "width" microseconds in every "window" cycle. It will not spin
390 for "window - width" microseconds, where the system can
393 The output will appear in the trace and trace_pipe files.
395 When the tracer is not running, it has no affect on the system,
396 but when it is running, it can cause the system to be
397 periodically non responsive. Do not run this tracer on a
400 To enable this tracer, echo in "hwlat" into the current_tracer
401 file. Every time a latency is greater than tracing_thresh, it will
402 be recorded into the ring buffer.
404 config OSNOISE_TRACER
405 bool "OS Noise tracer"
406 select GENERIC_TRACER
408 In the context of high-performance computing (HPC), the Operating
409 System Noise (osnoise) refers to the interference experienced by an
410 application due to activities inside the operating system. In the
411 context of Linux, NMIs, IRQs, SoftIRQs, and any other system thread
412 can cause noise to the system. Moreover, hardware-related jobs can
413 also cause noise, for example, via SMIs.
415 The osnoise tracer leverages the hwlat_detector by running a similar
416 loop with preemption, SoftIRQs and IRQs enabled, thus allowing all
417 the sources of osnoise during its execution. The osnoise tracer takes
418 note of the entry and exit point of any source of interferences,
419 increasing a per-cpu interference counter. It saves an interference
420 counter for each source of interference. The interference counter for
421 NMI, IRQs, SoftIRQs, and threads is increased anytime the tool
422 observes these interferences' entry events. When a noise happens
423 without any interference from the operating system level, the
424 hardware noise counter increases, pointing to a hardware-related
425 noise. In this way, osnoise can account for any source of
426 interference. At the end of the period, the osnoise tracer prints
427 the sum of all noise, the max single noise, the percentage of CPU
428 available for the thread, and the counters for the noise sources.
430 In addition to the tracer, a set of tracepoints were added to
431 facilitate the identification of the osnoise source.
433 The output will appear in the trace and trace_pipe files.
435 To enable this tracer, echo in "osnoise" into the current_tracer
438 config TIMERLAT_TRACER
439 bool "Timerlat tracer"
440 select OSNOISE_TRACER
441 select GENERIC_TRACER
443 The timerlat tracer aims to help the preemptive kernel developers
444 to find sources of wakeup latencies of real-time threads.
446 The tracer creates a per-cpu kernel thread with real-time priority.
447 The tracer thread sets a periodic timer to wakeup itself, and goes
448 to sleep waiting for the timer to fire. At the wakeup, the thread
449 then computes a wakeup latency value as the difference between
450 the current time and the absolute time that the timer was set
453 The tracer prints two lines at every activation. The first is the
454 timer latency observed at the hardirq context before the
455 activation of the thread. The second is the timer latency observed
456 by the thread, which is the same level that cyclictest reports. The
457 ACTIVATION ID field serves to relate the irq execution to its
458 respective thread execution.
460 The tracer is build on top of osnoise tracer, and the osnoise:
461 events can be used to trace the source of interference from NMI,
462 IRQs and other threads. It also enables the capture of the
463 stacktrace at the IRQ context, which helps to identify the code
464 path that can cause thread delay.
467 bool "Memory mapped IO tracing"
468 depends on HAVE_MMIOTRACE_SUPPORT && PCI
469 select GENERIC_TRACER
471 Mmiotrace traces Memory Mapped I/O access and is meant for
472 debugging and reverse engineering. It is called from the ioremap
473 implementation and works via page faults. Tracing is disabled by
474 default and can be enabled at run-time.
476 See Documentation/trace/mmiotrace.rst.
477 If you are not helping to develop drivers, say N.
479 config ENABLE_DEFAULT_TRACERS
480 bool "Trace process context switches and events"
481 depends on !GENERIC_TRACER
484 This tracer hooks to various trace points in the kernel,
485 allowing the user to pick and choose which trace point they
486 want to trace. It also includes the sched_switch tracer plugin.
488 config FTRACE_SYSCALLS
489 bool "Trace syscalls"
490 depends on HAVE_SYSCALL_TRACEPOINTS
491 select GENERIC_TRACER
494 Basic tracer to catch the syscall entry and exit events.
496 config TRACER_SNAPSHOT
497 bool "Create a snapshot trace buffer"
498 select TRACER_MAX_TRACE
500 Allow tracing users to take snapshot of the current buffer using the
501 ftrace interface, e.g.:
503 echo 1 > /sys/kernel/debug/tracing/snapshot
506 config TRACER_SNAPSHOT_PER_CPU_SWAP
507 bool "Allow snapshot to swap per CPU"
508 depends on TRACER_SNAPSHOT
509 select RING_BUFFER_ALLOW_SWAP
511 Allow doing a snapshot of a single CPU buffer instead of a
512 full swap (all buffers). If this is set, then the following is
515 echo 1 > /sys/kernel/debug/tracing/per_cpu/cpu2/snapshot
517 After which, only the tracing buffer for CPU 2 was swapped with
518 the main tracing buffer, and the other CPU buffers remain the same.
520 When this is enabled, this adds a little more overhead to the
521 trace recording, as it needs to add some checks to synchronize
522 recording with swaps. But this does not affect the performance
523 of the overall system. This is enabled by default when the preempt
524 or irq latency tracers are enabled, as those need to swap as well
525 and already adds the overhead (plus a lot more).
527 config TRACE_BRANCH_PROFILING
529 select GENERIC_TRACER
532 prompt "Branch Profiling"
533 default BRANCH_PROFILE_NONE
535 The branch profiling is a software profiler. It will add hooks
536 into the C conditionals to test which path a branch takes.
538 The likely/unlikely profiler only looks at the conditions that
539 are annotated with a likely or unlikely macro.
541 The "all branch" profiler will profile every if-statement in the
542 kernel. This profiler will also enable the likely/unlikely
545 Either of the above profilers adds a bit of overhead to the system.
546 If unsure, choose "No branch profiling".
548 config BRANCH_PROFILE_NONE
549 bool "No branch profiling"
551 No branch profiling. Branch profiling adds a bit of overhead.
552 Only enable it if you want to analyse the branching behavior.
553 Otherwise keep it disabled.
555 config PROFILE_ANNOTATED_BRANCHES
556 bool "Trace likely/unlikely profiler"
557 select TRACE_BRANCH_PROFILING
559 This tracer profiles all likely and unlikely macros
560 in the kernel. It will display the results in:
562 /sys/kernel/debug/tracing/trace_stat/branch_annotated
564 Note: this will add a significant overhead; only turn this
565 on if you need to profile the system's use of these macros.
567 config PROFILE_ALL_BRANCHES
568 bool "Profile all if conditionals" if !FORTIFY_SOURCE
569 select TRACE_BRANCH_PROFILING
571 This tracer profiles all branch conditions. Every if ()
572 taken in the kernel is recorded whether it hit or miss.
573 The results will be displayed in:
575 /sys/kernel/debug/tracing/trace_stat/branch_all
577 This option also enables the likely/unlikely profiler.
579 This configuration, when enabled, will impose a great overhead
580 on the system. This should only be enabled when the system
581 is to be analyzed in much detail.
584 config TRACING_BRANCHES
587 Selected by tracers that will trace the likely and unlikely
588 conditions. This prevents the tracers themselves from being
589 profiled. Profiling the tracing infrastructure can only happen
590 when the likelys and unlikelys are not being traced.
593 bool "Trace likely/unlikely instances"
594 depends on TRACE_BRANCH_PROFILING
595 select TRACING_BRANCHES
597 This traces the events of likely and unlikely condition
598 calls in the kernel. The difference between this and the
599 "Trace likely/unlikely profiler" is that this is not a
600 histogram of the callers, but actually places the calling
601 events into a running trace buffer to see when and where the
602 events happened, as well as their results.
606 config BLK_DEV_IO_TRACE
607 bool "Support for tracing block IO actions"
613 select GENERIC_TRACER
616 Say Y here if you want to be able to trace the block layer actions
617 on a given queue. Tracing allows you to see any traffic happening
618 on a block device queue. For more information (and the userspace
619 support tools needed), fetch the blktrace tools from:
621 git://git.kernel.dk/blktrace.git
623 Tracing also is possible using the ftrace interface, e.g.:
625 echo 1 > /sys/block/sda/sda1/trace/enable
626 echo blk > /sys/kernel/debug/tracing/current_tracer
627 cat /sys/kernel/debug/tracing/trace_pipe
633 depends on HAVE_REGS_AND_STACK_ACCESS_API
634 bool "Enable kprobes-based dynamic events"
637 select DYNAMIC_EVENTS
640 This allows the user to add tracing events (similar to tracepoints)
641 on the fly via the ftrace interface. See
642 Documentation/trace/kprobetrace.rst for more details.
644 Those events can be inserted wherever kprobes can probe, and record
645 various register and memory values.
647 This option is also required by perf-probe subcommand of perf tools.
648 If you want to use perf tools, this option is strongly recommended.
650 config KPROBE_EVENTS_ON_NOTRACE
651 bool "Do NOT protect notrace function from kprobe events"
652 depends on KPROBE_EVENTS
653 depends on DYNAMIC_FTRACE
656 This is only for the developers who want to debug ftrace itself
659 If kprobes can use ftrace instead of breakpoint, ftrace related
660 functions are protected from kprobe-events to prevent an infinite
661 recursion or any unexpected execution path which leads to a kernel
664 This option disables such protection and allows you to put kprobe
665 events on ftrace functions for debugging ftrace by itself.
666 Note that this might let you shoot yourself in the foot.
671 bool "Enable uprobes-based dynamic events"
672 depends on ARCH_SUPPORTS_UPROBES
674 depends on PERF_EVENTS
677 select DYNAMIC_EVENTS
681 This allows the user to add tracing events on top of userspace
682 dynamic events (similar to tracepoints) on the fly via the trace
683 events interface. Those events can be inserted wherever uprobes
684 can probe, and record various registers.
685 This option is required if you plan to use perf-probe subcommand
686 of perf tools on user space applications.
689 depends on BPF_SYSCALL
690 depends on (KPROBE_EVENTS || UPROBE_EVENTS) && PERF_EVENTS
694 This allows the user to attach BPF programs to kprobe, uprobe, and
697 config DYNAMIC_EVENTS
703 config BPF_KPROBE_OVERRIDE
704 bool "Enable BPF programs to override a kprobed function"
705 depends on BPF_EVENTS
706 depends on FUNCTION_ERROR_INJECTION
709 Allows BPF to override the execution of a probed function and
710 set a different return value. This is used for error injection.
712 config FTRACE_MCOUNT_RECORD
714 depends on DYNAMIC_FTRACE
715 depends on HAVE_FTRACE_MCOUNT_RECORD
717 config FTRACE_MCOUNT_USE_PATCHABLE_FUNCTION_ENTRY
719 depends on FTRACE_MCOUNT_RECORD
721 config FTRACE_MCOUNT_USE_CC
723 depends on $(cc-option,-mrecord-mcount)
724 depends on !FTRACE_MCOUNT_USE_PATCHABLE_FUNCTION_ENTRY
725 depends on FTRACE_MCOUNT_RECORD
727 config FTRACE_MCOUNT_USE_OBJTOOL
729 depends on HAVE_OBJTOOL_MCOUNT
730 depends on !FTRACE_MCOUNT_USE_PATCHABLE_FUNCTION_ENTRY
731 depends on !FTRACE_MCOUNT_USE_CC
732 depends on FTRACE_MCOUNT_RECORD
735 config FTRACE_MCOUNT_USE_RECORDMCOUNT
737 depends on !FTRACE_MCOUNT_USE_PATCHABLE_FUNCTION_ENTRY
738 depends on !FTRACE_MCOUNT_USE_CC
739 depends on !FTRACE_MCOUNT_USE_OBJTOOL
740 depends on FTRACE_MCOUNT_RECORD
744 depends on ARCH_HAVE_NMI_SAFE_CMPXCHG
746 tracing_map is a special-purpose lock-free map for tracing,
747 separated out as a stand-alone facility in order to allow it
748 to be shared between multiple tracers. It isn't meant to be
749 generally used outside of that context, and is normally
750 selected by tracers that use it.
753 bool "Synthetic trace events"
755 select DYNAMIC_EVENTS
758 Synthetic events are user-defined trace events that can be
759 used to combine data from other trace events or in fact any
760 data source. Synthetic events can be generated indirectly
761 via the trace() action of histogram triggers or directly
762 by way of an in-kernel API.
764 See Documentation/trace/events.rst or
765 Documentation/trace/histogram.rst for details and examples.
770 bool "User trace events"
772 select DYNAMIC_EVENTS
773 depends on BROKEN || COMPILE_TEST # API needs to be straighten out
775 User trace events are user-defined trace events that
776 can be used like an existing kernel trace event. User trace
777 events are generated by writing to a tracefs file. User
778 processes can determine if their tracing events should be
779 generated by memory mapping a tracefs file and checking for
780 an associated byte being non-zero.
785 bool "Histogram triggers"
786 depends on ARCH_HAVE_NMI_SAFE_CMPXCHG
789 select DYNAMIC_EVENTS
793 Hist triggers allow one or more arbitrary trace event fields
794 to be aggregated into hash tables and dumped to stdout by
795 reading a debugfs/tracefs file. They're useful for
796 gathering quick and dirty (though precise) summaries of
797 event activity as an initial guide for further investigation
798 using more advanced tools.
800 Inter-event tracing of quantities such as latencies is also
801 supported using hist triggers under this option.
803 See Documentation/trace/histogram.rst.
806 config TRACE_EVENT_INJECT
807 bool "Trace event injection"
810 Allow user-space to inject a specific trace event into the ring
811 buffer. This is mainly used for testing purpose.
815 config TRACEPOINT_BENCHMARK
816 bool "Add tracepoint that benchmarks tracepoints"
818 This option creates the tracepoint "benchmark:benchmark_event".
819 When the tracepoint is enabled, it kicks off a kernel thread that
820 goes into an infinite loop (calling cond_resched() to let other tasks
821 run), and calls the tracepoint. Each iteration will record the time
822 it took to write to the tracepoint and the next iteration that
823 data will be passed to the tracepoint itself. That is, the tracepoint
824 will report the time it took to do the previous tracepoint.
825 The string written to the tracepoint is a static string of 128 bytes
826 to keep the time the same. The initial string is simply a write of
827 "START". The second string records the cold cache time of the first
828 write which is not added to the rest of the calculations.
830 As it is a tight loop, it benchmarks as hot cache. That's fine because
831 we care most about hot paths that are probably in cache already.
833 An example of the output:
836 first=3672 [COLD CACHED]
837 last=632 first=3672 max=632 min=632 avg=316 std=446 std^2=199712
838 last=278 first=3672 max=632 min=278 avg=303 std=316 std^2=100337
839 last=277 first=3672 max=632 min=277 avg=296 std=258 std^2=67064
840 last=273 first=3672 max=632 min=273 avg=292 std=224 std^2=50411
841 last=273 first=3672 max=632 min=273 avg=288 std=200 std^2=40389
842 last=281 first=3672 max=632 min=273 avg=287 std=183 std^2=33666
845 config RING_BUFFER_BENCHMARK
846 tristate "Ring buffer benchmark stress tester"
847 depends on RING_BUFFER
849 This option creates a test to stress the ring buffer and benchmark it.
850 It creates its own ring buffer such that it will not interfere with
851 any other users of the ring buffer (such as ftrace). It then creates
852 a producer and consumer that will run for 10 seconds and sleep for
853 10 seconds. Each interval it will print out the number of events
854 it recorded and give a rough estimate of how long each iteration took.
856 It does not disable interrupts or raise its priority, so it may be
857 affected by processes that are running.
861 config TRACE_EVAL_MAP_FILE
862 bool "Show eval mappings for trace events"
865 The "print fmt" of the trace events will show the enum/sizeof names
866 instead of their values. This can cause problems for user space tools
867 that use this string to parse the raw data as user space does not know
868 how to convert the string to its value.
870 To fix this, there's a special macro in the kernel that can be used
871 to convert an enum/sizeof into its value. If this macro is used, then
872 the print fmt strings will be converted to their values.
874 If something does not get converted properly, this option can be
875 used to show what enums/sizeof the kernel tried to convert.
877 This option is for debugging the conversions. A file is created
878 in the tracing directory called "eval_map" that will show the
879 names matched with their values and what trace event system they
882 Normally, the mapping of the strings to values will be freed after
883 boot up or module load. With this option, they will not be freed, as
884 they are needed for the "eval_map" file. Enabling this option will
885 increase the memory footprint of the running kernel.
889 config FTRACE_RECORD_RECURSION
890 bool "Record functions that recurse in function tracing"
891 depends on FUNCTION_TRACER
893 All callbacks that attach to the function tracing have some sort
894 of protection against recursion. Even though the protection exists,
895 it adds overhead. This option will create a file in the tracefs
896 file system called "recursed_functions" that will list the functions
897 that triggered a recursion.
899 This will add more overhead to cases that have recursion.
903 config FTRACE_RECORD_RECURSION_SIZE
904 int "Max number of recursed functions to record"
906 depends on FTRACE_RECORD_RECURSION
908 This defines the limit of number of functions that can be
909 listed in the "recursed_functions" file, that lists all
910 the functions that caused a recursion to happen.
911 This file can be reset, but the limit can not change in
914 config RING_BUFFER_RECORD_RECURSION
915 bool "Record functions that recurse in the ring buffer"
916 depends on FTRACE_RECORD_RECURSION
917 # default y, because it is coupled with FTRACE_RECORD_RECURSION
920 The ring buffer has its own internal recursion. Although when
921 recursion happens it wont cause harm because of the protection,
922 but it does cause an unwanted overhead. Enabling this option will
923 place where recursion was detected into the ftrace "recursed_functions"
926 This will add more overhead to cases that have recursion.
928 config GCOV_PROFILE_FTRACE
929 bool "Enable GCOV profiling on ftrace subsystem"
930 depends on GCOV_KERNEL
932 Enable GCOV profiling on ftrace subsystem for checking
933 which functions/lines are tested.
937 Note that on a kernel compiled with this config, ftrace will
938 run significantly slower.
940 config FTRACE_SELFTEST
943 config FTRACE_STARTUP_TEST
944 bool "Perform a startup test on ftrace"
945 depends on GENERIC_TRACER
946 select FTRACE_SELFTEST
948 This option performs a series of startup tests on ftrace. On bootup
949 a series of tests are made to verify that the tracer is
950 functioning properly. It will do tests on all the configured
953 config EVENT_TRACE_STARTUP_TEST
954 bool "Run selftest on trace events"
955 depends on FTRACE_STARTUP_TEST
958 This option performs a test on all trace events in the system.
959 It basically just enables each event and runs some code that
960 will trigger events (not necessarily the event it enables)
961 This may take some time run as there are a lot of events.
963 config EVENT_TRACE_TEST_SYSCALLS
964 bool "Run selftest on syscall events"
965 depends on EVENT_TRACE_STARTUP_TEST
967 This option will also enable testing every syscall event.
968 It only enables the event and disables it and runs various loads
969 with the event enabled. This adds a bit more time for kernel boot
970 up since it runs this on every system call defined.
972 TBD - enable a way to actually call the syscalls as we test their
975 config FTRACE_SORT_STARTUP_TEST
976 bool "Verify compile time sorting of ftrace functions"
977 depends on DYNAMIC_FTRACE
978 depends on BUILDTIME_MCOUNT_SORT
980 Sorting of the mcount_loc sections that is used to find the
981 where the ftrace knows where to patch functions for tracing
982 and other callbacks is done at compile time. But if the sort
983 is not done correctly, it will cause non-deterministic failures.
984 When this is set, the sorted sections will be verified that they
985 are in deed sorted and will warn if they are not.
989 config RING_BUFFER_STARTUP_TEST
990 bool "Ring buffer startup self test"
991 depends on RING_BUFFER
993 Run a simple self test on the ring buffer on boot up. Late in the
994 kernel boot sequence, the test will start that kicks off
995 a thread per cpu. Each thread will write various size events
996 into the ring buffer. Another thread is created to send IPIs
997 to each of the threads, where the IPI handler will also write
998 to the ring buffer, to test/stress the nesting ability.
999 If any anomalies are discovered, a warning will be displayed
1000 and all ring buffers will be disabled.
1002 The test runs for 10 seconds. This will slow your boot time
1003 by at least 10 more seconds.
1005 At the end of the test, statics and more checks are done.
1006 It will output the stats of each per cpu buffer. What
1007 was written, the sizes, what was read, what was lost, and
1008 other similar details.
1012 config RING_BUFFER_VALIDATE_TIME_DELTAS
1013 bool "Verify ring buffer time stamp deltas"
1014 depends on RING_BUFFER
1016 This will audit the time stamps on the ring buffer sub
1017 buffer to make sure that all the time deltas for the
1018 events on a sub buffer matches the current time stamp.
1019 This audit is performed for every event that is not
1020 interrupted, or interrupting another event. A check
1021 is also made when traversing sub buffers to make sure
1022 that all the deltas on the previous sub buffer do not
1023 add up to be greater than the current time stamp.
1025 NOTE: This adds significant overhead to recording of events,
1026 and should only be used to test the logic of the ring buffer.
1027 Do not use it on production systems.
1029 Only say Y if you understand what this does, and you
1030 still want it enabled. Otherwise say N
1032 config MMIOTRACE_TEST
1033 tristate "Test module for mmiotrace"
1034 depends on MMIOTRACE && m
1036 This is a dumb module for testing mmiotrace. It is very dangerous
1037 as it will write garbage to IO memory starting at a given address.
1038 However, it should be safe to use on e.g. unused portion of VRAM.
1040 Say N, unless you absolutely know what you are doing.
1042 config PREEMPTIRQ_DELAY_TEST
1043 tristate "Test module to create a preempt / IRQ disable delay thread to test latency tracers"
1046 Select this option to build a test module that can help test latency
1047 tracers by executing a preempt or irq disable section with a user
1048 configurable delay. The module busy waits for the duration of the
1051 For example, the following invocation generates a burst of three
1052 irq-disabled critical sections for 500us:
1053 modprobe preemptirq_delay_test test_mode=irq delay=500 burst_size=3
1055 What's more, if you want to attach the test on the cpu which the latency
1056 tracer is running on, specify cpu_affinity=cpu_num at the end of the
1061 config SYNTH_EVENT_GEN_TEST
1062 tristate "Test module for in-kernel synthetic event generation"
1063 depends on SYNTH_EVENTS
1065 This option creates a test module to check the base
1066 functionality of in-kernel synthetic event definition and
1069 To test, insert the module, and then check the trace buffer
1070 for the generated sample events.
1074 config KPROBE_EVENT_GEN_TEST
1075 tristate "Test module for in-kernel kprobe event generation"
1076 depends on KPROBE_EVENTS
1078 This option creates a test module to check the base
1079 functionality of in-kernel kprobe event definition.
1081 To test, insert the module, and then check the trace buffer
1082 for the generated kprobe events.
1086 config HIST_TRIGGERS_DEBUG
1087 bool "Hist trigger debug support"
1088 depends on HIST_TRIGGERS
1090 Add "hist_debug" file for each event, which when read will
1091 dump out a bunch of internal details about the hist triggers
1092 defined on that event.
1094 The hist_debug file serves a couple of purposes:
1096 - Helps developers verify that nothing is broken.
1098 - Provides educational information to support the details
1099 of the hist trigger internals as described by
1100 Documentation/trace/histogram-design.rst.
1102 The hist_debug output only covers the data structures
1103 related to the histogram definitions themselves and doesn't
1104 display the internals of map buckets or variable values of
1109 source "kernel/trace/rv/Kconfig"