1=======================
2Kernel Probes (Kprobes)
3=======================
4
5:Author: Jim Keniston <jkenisto@us.ibm.com>
6:Author: Prasanna S Panchamukhi <prasanna.panchamukhi@gmail.com>
7:Author: Masami Hiramatsu <mhiramat@redhat.com>
8
9.. CONTENTS
10
11  1. Concepts: Kprobes, and Return Probes
12  2. Architectures Supported
13  3. Configuring Kprobes
14  4. API Reference
15  5. Kprobes Features and Limitations
16  6. Probe Overhead
17  7. TODO
18  8. Kprobes Example
19  9. Kretprobes Example
20  10. Deprecated Features
21  Appendix A: The kprobes debugfs interface
22  Appendix B: The kprobes sysctl interface
23  Appendix C: References
24
25Concepts: Kprobes and Return Probes
26=========================================
27
28Kprobes enables you to dynamically break into any kernel routine and
29collect debugging and performance information non-disruptively. You
30can trap at almost any kernel code address [1]_, specifying a handler
31routine to be invoked when the breakpoint is hit.
32
33.. [1] some parts of the kernel code can not be trapped, see
34       :ref:`kprobes_blacklist`)
35
36There are currently two types of probes: kprobes, and kretprobes
37(also called return probes).  A kprobe can be inserted on virtually
38any instruction in the kernel.  A return probe fires when a specified
39function returns.
40
41In the typical case, Kprobes-based instrumentation is packaged as
42a kernel module.  The module's init function installs ("registers")
43one or more probes, and the exit function unregisters them.  A
44registration function such as register_kprobe() specifies where
45the probe is to be inserted and what handler is to be called when
46the probe is hit.
47
48There are also ``register_/unregister_*probes()`` functions for batch
49registration/unregistration of a group of ``*probes``. These functions
50can speed up unregistration process when you have to unregister
51a lot of probes at once.
52
53The next four subsections explain how the different types of
54probes work and how jump optimization works.  They explain certain
55things that you'll need to know in order to make the best use of
56Kprobes -- e.g., the difference between a pre_handler and
57a post_handler, and how to use the maxactive and nmissed fields of
58a kretprobe.  But if you're in a hurry to start using Kprobes, you
59can skip ahead to :ref:`kprobes_archs_supported`.
60
61How Does a Kprobe Work?
62-----------------------
63
64When a kprobe is registered, Kprobes makes a copy of the probed
65instruction and replaces the first byte(s) of the probed instruction
66with a breakpoint instruction (e.g., int3 on i386 and x86_64).
67
68When a CPU hits the breakpoint instruction, a trap occurs, the CPU's
69registers are saved, and control passes to Kprobes via the
70notifier_call_chain mechanism.  Kprobes executes the "pre_handler"
71associated with the kprobe, passing the handler the addresses of the
72kprobe struct and the saved registers.
73
74Next, Kprobes single-steps its copy of the probed instruction.
75(It would be simpler to single-step the actual instruction in place,
76but then Kprobes would have to temporarily remove the breakpoint
77instruction.  This would open a small time window when another CPU
78could sail right past the probepoint.)
79
80After the instruction is single-stepped, Kprobes executes the
81"post_handler," if any, that is associated with the kprobe.
82Execution then continues with the instruction following the probepoint.
83
84Changing Execution Path
85-----------------------
86
87Since kprobes can probe into a running kernel code, it can change the
88register set, including instruction pointer. This operation requires
89maximum care, such as keeping the stack frame, recovering the execution
90path etc. Since it operates on a running kernel and needs deep knowledge
91of computer architecture and concurrent computing, you can easily shoot
92your foot.
93
94If you change the instruction pointer (and set up other related
95registers) in pre_handler, you must return !0 so that kprobes stops
96single stepping and just returns to the given address.
97This also means post_handler should not be called anymore.
98
99Note that this operation may be harder on some architectures which use
100TOC (Table of Contents) for function call, since you have to setup a new
101TOC for your function in your module, and recover the old one after
102returning from it.
103
104Return Probes
105-------------
106
107How Does a Return Probe Work?
108^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
109
110When you call register_kretprobe(), Kprobes establishes a kprobe at
111the entry to the function.  When the probed function is called and this
112probe is hit, Kprobes saves a copy of the return address, and replaces
113the return address with the address of a "trampoline."  The trampoline
114is an arbitrary piece of code -- typically just a nop instruction.
115At boot time, Kprobes registers a kprobe at the trampoline.
116
117When the probed function executes its return instruction, control
118passes to the trampoline and that probe is hit.  Kprobes' trampoline
119handler calls the user-specified return handler associated with the
120kretprobe, then sets the saved instruction pointer to the saved return
121address, and that's where execution resumes upon return from the trap.
122
123While the probed function is executing, its return address is
124stored in an object of type kretprobe_instance.  Before calling
125register_kretprobe(), the user sets the maxactive field of the
126kretprobe struct to specify how many instances of the specified
127function can be probed simultaneously.  register_kretprobe()
128pre-allocates the indicated number of kretprobe_instance objects.
129
130For example, if the function is non-recursive and is called with a
131spinlock held, maxactive = 1 should be enough.  If the function is
132non-recursive and can never relinquish the CPU (e.g., via a semaphore
133or preemption), NR_CPUS should be enough.  If maxactive <= 0, it is
134set to a default value.  If CONFIG_PREEMPT is enabled, the default
135is max(10, 2*NR_CPUS).  Otherwise, the default is NR_CPUS.
136
137It's not a disaster if you set maxactive too low; you'll just miss
138some probes.  In the kretprobe struct, the nmissed field is set to
139zero when the return probe is registered, and is incremented every
140time the probed function is entered but there is no kretprobe_instance
141object available for establishing the return probe.
142
143Kretprobe entry-handler
144^^^^^^^^^^^^^^^^^^^^^^^
145
146Kretprobes also provides an optional user-specified handler which runs
147on function entry. This handler is specified by setting the entry_handler
148field of the kretprobe struct. Whenever the kprobe placed by kretprobe at the
149function entry is hit, the user-defined entry_handler, if any, is invoked.
150If the entry_handler returns 0 (success) then a corresponding return handler
151is guaranteed to be called upon function return. If the entry_handler
152returns a non-zero error then Kprobes leaves the return address as is, and
153the kretprobe has no further effect for that particular function instance.
154
155Multiple entry and return handler invocations are matched using the unique
156kretprobe_instance object associated with them. Additionally, a user
157may also specify per return-instance private data to be part of each
158kretprobe_instance object. This is especially useful when sharing private
159data between corresponding user entry and return handlers. The size of each
160private data object can be specified at kretprobe registration time by
161setting the data_size field of the kretprobe struct. This data can be
162accessed through the data field of each kretprobe_instance object.
163
164In case probed function is entered but there is no kretprobe_instance
165object available, then in addition to incrementing the nmissed count,
166the user entry_handler invocation is also skipped.
167
168.. _kprobes_jump_optimization:
169
170How Does Jump Optimization Work?
171--------------------------------
172
173If your kernel is built with CONFIG_OPTPROBES=y (currently this flag
174is automatically set 'y' on x86/x86-64, non-preemptive kernel) and
175the "debug.kprobes_optimization" kernel parameter is set to 1 (see
176sysctl(8)), Kprobes tries to reduce probe-hit overhead by using a jump
177instruction instead of a breakpoint instruction at each probepoint.
178
179Init a Kprobe
180^^^^^^^^^^^^^
181
182When a probe is registered, before attempting this optimization,
183Kprobes inserts an ordinary, breakpoint-based kprobe at the specified
184address. So, even if it's not possible to optimize this particular
185probepoint, there'll be a probe there.
186
187Safety Check
188^^^^^^^^^^^^
189
190Before optimizing a probe, Kprobes performs the following safety checks:
191
192- Kprobes verifies that the region that will be replaced by the jump
193  instruction (the "optimized region") lies entirely within one function.
194  (A jump instruction is multiple bytes, and so may overlay multiple
195  instructions.)
196
197- Kprobes analyzes the entire function and verifies that there is no
198  jump into the optimized region.  Specifically:
199
200  - the function contains no indirect jump;
201  - the function contains no instruction that causes an exception (since
202    the fixup code triggered by the exception could jump back into the
203    optimized region -- Kprobes checks the exception tables to verify this);
204  - there is no near jump to the optimized region (other than to the first
205    byte).
206
207- For each instruction in the optimized region, Kprobes verifies that
208  the instruction can be executed out of line.
209
210Preparing Detour Buffer
211^^^^^^^^^^^^^^^^^^^^^^^
212
213Next, Kprobes prepares a "detour" buffer, which contains the following
214instruction sequence:
215
216- code to push the CPU's registers (emulating a breakpoint trap)
217- a call to the trampoline code which calls user's probe handlers.
218- code to restore registers
219- the instructions from the optimized region
220- a jump back to the original execution path.
221
222Pre-optimization
223^^^^^^^^^^^^^^^^
224
225After preparing the detour buffer, Kprobes verifies that none of the
226following situations exist:
227
228- The probe has a post_handler.
229- Other instructions in the optimized region are probed.
230- The probe is disabled.
231
232In any of the above cases, Kprobes won't start optimizing the probe.
233Since these are temporary situations, Kprobes tries to start
234optimizing it again if the situation is changed.
235
236If the kprobe can be optimized, Kprobes enqueues the kprobe to an
237optimizing list, and kicks the kprobe-optimizer workqueue to optimize
238it.  If the to-be-optimized probepoint is hit before being optimized,
239Kprobes returns control to the original instruction path by setting
240the CPU's instruction pointer to the copied code in the detour buffer
241-- thus at least avoiding the single-step.
242
243Optimization
244^^^^^^^^^^^^
245
246The Kprobe-optimizer doesn't insert the jump instruction immediately;
247rather, it calls synchronize_rcu() for safety first, because it's
248possible for a CPU to be interrupted in the middle of executing the
249optimized region [3]_.  As you know, synchronize_rcu() can ensure
250that all interruptions that were active when synchronize_rcu()
251was called are done, but only if CONFIG_PREEMPT=n.  So, this version
252of kprobe optimization supports only kernels with CONFIG_PREEMPT=n [4]_.
253
254After that, the Kprobe-optimizer calls stop_machine() to replace
255the optimized region with a jump instruction to the detour buffer,
256using text_poke_smp().
257
258Unoptimization
259^^^^^^^^^^^^^^
260
261When an optimized kprobe is unregistered, disabled, or blocked by
262another kprobe, it will be unoptimized.  If this happens before
263the optimization is complete, the kprobe is just dequeued from the
264optimized list.  If the optimization has been done, the jump is
265replaced with the original code (except for an int3 breakpoint in
266the first byte) by using text_poke_smp().
267
268.. [3] Please imagine that the 2nd instruction is interrupted and then
269   the optimizer replaces the 2nd instruction with the jump *address*
270   while the interrupt handler is running. When the interrupt
271   returns to original address, there is no valid instruction,
272   and it causes an unexpected result.
273
274.. [4] This optimization-safety checking may be replaced with the
275   stop-machine method that ksplice uses for supporting a CONFIG_PREEMPT=y
276   kernel.
277
278NOTE for geeks:
279The jump optimization changes the kprobe's pre_handler behavior.
280Without optimization, the pre_handler can change the kernel's execution
281path by changing regs->ip and returning 1.  However, when the probe
282is optimized, that modification is ignored.  Thus, if you want to
283tweak the kernel's execution path, you need to suppress optimization,
284using one of the following techniques:
285
286- Specify an empty function for the kprobe's post_handler.
287
288or
289
290- Execute 'sysctl -w debug.kprobes_optimization=n'
291
292.. _kprobes_blacklist:
293
294Blacklist
295---------
296
297Kprobes can probe most of the kernel except itself. This means
298that there are some functions where kprobes cannot probe. Probing
299(trapping) such functions can cause a recursive trap (e.g. double
300fault) or the nested probe handler may never be called.
301Kprobes manages such functions as a blacklist.
302If you want to add a function into the blacklist, you just need
303to (1) include linux/kprobes.h and (2) use NOKPROBE_SYMBOL() macro
304to specify a blacklisted function.
305Kprobes checks the given probe address against the blacklist and
306rejects registering it, if the given address is in the blacklist.
307
308.. _kprobes_archs_supported:
309
310Architectures Supported
311=======================
312
313Kprobes and return probes are implemented on the following
314architectures:
315
316- i386 (Supports jump optimization)
317- x86_64 (AMD-64, EM64T) (Supports jump optimization)
318- ppc64
319- ia64 (Does not support probes on instruction slot1.)
320- sparc64 (Return probes not yet implemented.)
321- arm
322- ppc
323- mips
324- s390
325- parisc
326
327Configuring Kprobes
328===================
329
330When configuring the kernel using make menuconfig/xconfig/oldconfig,
331ensure that CONFIG_KPROBES is set to "y". Under "General setup", look
332for "Kprobes".
333
334So that you can load and unload Kprobes-based instrumentation modules,
335make sure "Loadable module support" (CONFIG_MODULES) and "Module
336unloading" (CONFIG_MODULE_UNLOAD) are set to "y".
337
338Also make sure that CONFIG_KALLSYMS and perhaps even CONFIG_KALLSYMS_ALL
339are set to "y", since kallsyms_lookup_name() is used by the in-kernel
340kprobe address resolution code.
341
342If you need to insert a probe in the middle of a function, you may find
343it useful to "Compile the kernel with debug info" (CONFIG_DEBUG_INFO),
344so you can use "objdump -d -l vmlinux" to see the source-to-object
345code mapping.
346
347API Reference
348=============
349
350The Kprobes API includes a "register" function and an "unregister"
351function for each type of probe. The API also includes "register_*probes"
352and "unregister_*probes" functions for (un)registering arrays of probes.
353Here are terse, mini-man-page specifications for these functions and
354the associated probe handlers that you'll write. See the files in the
355samples/kprobes/ sub-directory for examples.
356
357register_kprobe
358---------------
359
360::
361
362	#include <linux/kprobes.h>
363	int register_kprobe(struct kprobe *kp);
364
365Sets a breakpoint at the address kp->addr.  When the breakpoint is hit, Kprobes
366calls kp->pre_handler.  After the probed instruction is single-stepped, Kprobe
367calls kp->post_handler.  Any or all handlers can be NULL. If kp->flags is set
368KPROBE_FLAG_DISABLED, that kp will be registered but disabled, so, its handlers
369aren't hit until calling enable_kprobe(kp).
370
371.. note::
372
373   1. With the introduction of the "symbol_name" field to struct kprobe,
374      the probepoint address resolution will now be taken care of by the kernel.
375      The following will now work::
376
377	kp.symbol_name = "symbol_name";
378
379      (64-bit powerpc intricacies such as function descriptors are handled
380      transparently)
381
382   2. Use the "offset" field of struct kprobe if the offset into the symbol
383      to install a probepoint is known. This field is used to calculate the
384      probepoint.
385
386   3. Specify either the kprobe "symbol_name" OR the "addr". If both are
387      specified, kprobe registration will fail with -EINVAL.
388
389   4. With CISC architectures (such as i386 and x86_64), the kprobes code
390      does not validate if the kprobe.addr is at an instruction boundary.
391      Use "offset" with caution.
392
393register_kprobe() returns 0 on success, or a negative errno otherwise.
394
395User's pre-handler (kp->pre_handler)::
396
397	#include <linux/kprobes.h>
398	#include <linux/ptrace.h>
399	int pre_handler(struct kprobe *p, struct pt_regs *regs);
400
401Called with p pointing to the kprobe associated with the breakpoint,
402and regs pointing to the struct containing the registers saved when
403the breakpoint was hit.  Return 0 here unless you're a Kprobes geek.
404
405User's post-handler (kp->post_handler)::
406
407	#include <linux/kprobes.h>
408	#include <linux/ptrace.h>
409	void post_handler(struct kprobe *p, struct pt_regs *regs,
410			  unsigned long flags);
411
412p and regs are as described for the pre_handler.  flags always seems
413to be zero.
414
415register_kretprobe
416------------------
417
418::
419
420	#include <linux/kprobes.h>
421	int register_kretprobe(struct kretprobe *rp);
422
423Establishes a return probe for the function whose address is
424rp->kp.addr.  When that function returns, Kprobes calls rp->handler.
425You must set rp->maxactive appropriately before you call
426register_kretprobe(); see "How Does a Return Probe Work?" for details.
427
428register_kretprobe() returns 0 on success, or a negative errno
429otherwise.
430
431User's return-probe handler (rp->handler)::
432
433	#include <linux/kprobes.h>
434	#include <linux/ptrace.h>
435	int kretprobe_handler(struct kretprobe_instance *ri,
436			      struct pt_regs *regs);
437
438regs is as described for kprobe.pre_handler.  ri points to the
439kretprobe_instance object, of which the following fields may be
440of interest:
441
442- ret_addr: the return address
443- rp: points to the corresponding kretprobe object
444- task: points to the corresponding task struct
445- data: points to per return-instance private data; see "Kretprobe
446	entry-handler" for details.
447
448The regs_return_value(regs) macro provides a simple abstraction to
449extract the return value from the appropriate register as defined by
450the architecture's ABI.
451
452The handler's return value is currently ignored.
453
454unregister_*probe
455------------------
456
457::
458
459	#include <linux/kprobes.h>
460	void unregister_kprobe(struct kprobe *kp);
461	void unregister_kretprobe(struct kretprobe *rp);
462
463Removes the specified probe.  The unregister function can be called
464at any time after the probe has been registered.
465
466.. note::
467
468   If the functions find an incorrect probe (ex. an unregistered probe),
469   they clear the addr field of the probe.
470
471register_*probes
472----------------
473
474::
475
476	#include <linux/kprobes.h>
477	int register_kprobes(struct kprobe **kps, int num);
478	int register_kretprobes(struct kretprobe **rps, int num);
479
480Registers each of the num probes in the specified array.  If any
481error occurs during registration, all probes in the array, up to
482the bad probe, are safely unregistered before the register_*probes
483function returns.
484
485- kps/rps: an array of pointers to ``*probe`` data structures
486- num: the number of the array entries.
487
488.. note::
489
490   You have to allocate(or define) an array of pointers and set all
491   of the array entries before using these functions.
492
493unregister_*probes
494------------------
495
496::
497
498	#include <linux/kprobes.h>
499	void unregister_kprobes(struct kprobe **kps, int num);
500	void unregister_kretprobes(struct kretprobe **rps, int num);
501
502Removes each of the num probes in the specified array at once.
503
504.. note::
505
506   If the functions find some incorrect probes (ex. unregistered
507   probes) in the specified array, they clear the addr field of those
508   incorrect probes. However, other probes in the array are
509   unregistered correctly.
510
511disable_*probe
512--------------
513
514::
515
516	#include <linux/kprobes.h>
517	int disable_kprobe(struct kprobe *kp);
518	int disable_kretprobe(struct kretprobe *rp);
519
520Temporarily disables the specified ``*probe``. You can enable it again by using
521enable_*probe(). You must specify the probe which has been registered.
522
523enable_*probe
524-------------
525
526::
527
528	#include <linux/kprobes.h>
529	int enable_kprobe(struct kprobe *kp);
530	int enable_kretprobe(struct kretprobe *rp);
531
532Enables ``*probe`` which has been disabled by disable_*probe(). You must specify
533the probe which has been registered.
534
535Kprobes Features and Limitations
536================================
537
538Kprobes allows multiple probes at the same address. Also,
539a probepoint for which there is a post_handler cannot be optimized.
540So if you install a kprobe with a post_handler, at an optimized
541probepoint, the probepoint will be unoptimized automatically.
542
543In general, you can install a probe anywhere in the kernel.
544In particular, you can probe interrupt handlers.  Known exceptions
545are discussed in this section.
546
547The register_*probe functions will return -EINVAL if you attempt
548to install a probe in the code that implements Kprobes (mostly
549kernel/kprobes.c and ``arch/*/kernel/kprobes.c``, but also functions such
550as do_page_fault and notifier_call_chain).
551
552If you install a probe in an inline-able function, Kprobes makes
553no attempt to chase down all inline instances of the function and
554install probes there.  gcc may inline a function without being asked,
555so keep this in mind if you're not seeing the probe hits you expect.
556
557A probe handler can modify the environment of the probed function
558-- e.g., by modifying kernel data structures, or by modifying the
559contents of the pt_regs struct (which are restored to the registers
560upon return from the breakpoint).  So Kprobes can be used, for example,
561to install a bug fix or to inject faults for testing.  Kprobes, of
562course, has no way to distinguish the deliberately injected faults
563from the accidental ones.  Don't drink and probe.
564
565Kprobes makes no attempt to prevent probe handlers from stepping on
566each other -- e.g., probing printk() and then calling printk() from a
567probe handler.  If a probe handler hits a probe, that second probe's
568handlers won't be run in that instance, and the kprobe.nmissed member
569of the second probe will be incremented.
570
571As of Linux v2.6.15-rc1, multiple handlers (or multiple instances of
572the same handler) may run concurrently on different CPUs.
573
574Kprobes does not use mutexes or allocate memory except during
575registration and unregistration.
576
577Probe handlers are run with preemption disabled or interrupt disabled,
578which depends on the architecture and optimization state.  (e.g.,
579kretprobe handlers and optimized kprobe handlers run without interrupt
580disabled on x86/x86-64).  In any case, your handler should not yield
581the CPU (e.g., by attempting to acquire a semaphore, or waiting I/O).
582
583Since a return probe is implemented by replacing the return
584address with the trampoline's address, stack backtraces and calls
585to __builtin_return_address() will typically yield the trampoline's
586address instead of the real return address for kretprobed functions.
587(As far as we can tell, __builtin_return_address() is used only
588for instrumentation and error reporting.)
589
590If the number of times a function is called does not match the number
591of times it returns, registering a return probe on that function may
592produce undesirable results. In such a case, a line:
593kretprobe BUG!: Processing kretprobe d000000000041aa8 @ c00000000004f48c
594gets printed. With this information, one will be able to correlate the
595exact instance of the kretprobe that caused the problem. We have the
596do_exit() case covered. do_execve() and do_fork() are not an issue.
597We're unaware of other specific cases where this could be a problem.
598
599If, upon entry to or exit from a function, the CPU is running on
600a stack other than that of the current task, registering a return
601probe on that function may produce undesirable results.  For this
602reason, Kprobes doesn't support return probes (or kprobes)
603on the x86_64 version of __switch_to(); the registration functions
604return -EINVAL.
605
606On x86/x86-64, since the Jump Optimization of Kprobes modifies
607instructions widely, there are some limitations to optimization. To
608explain it, we introduce some terminology. Imagine a 3-instruction
609sequence consisting of a two 2-byte instructions and one 3-byte
610instruction.
611
612::
613
614		IA
615		|
616	[-2][-1][0][1][2][3][4][5][6][7]
617		[ins1][ins2][  ins3 ]
618		[<-     DCR       ->]
619		[<- JTPR ->]
620
621	ins1: 1st Instruction
622	ins2: 2nd Instruction
623	ins3: 3rd Instruction
624	IA:  Insertion Address
625	JTPR: Jump Target Prohibition Region
626	DCR: Detoured Code Region
627
628The instructions in DCR are copied to the out-of-line buffer
629of the kprobe, because the bytes in DCR are replaced by
630a 5-byte jump instruction. So there are several limitations.
631
632a) The instructions in DCR must be relocatable.
633b) The instructions in DCR must not include a call instruction.
634c) JTPR must not be targeted by any jump or call instruction.
635d) DCR must not straddle the border between functions.
636
637Anyway, these limitations are checked by the in-kernel instruction
638decoder, so you don't need to worry about that.
639
640Probe Overhead
641==============
642
643On a typical CPU in use in 2005, a kprobe hit takes 0.5 to 1.0
644microseconds to process.  Specifically, a benchmark that hits the same
645probepoint repeatedly, firing a simple handler each time, reports 1-2
646million hits per second, depending on the architecture.  A return-probe
647hit typically takes 50-75% longer than a kprobe hit.
648When you have a return probe set on a function, adding a kprobe at
649the entry to that function adds essentially no overhead.
650
651Here are sample overhead figures (in usec) for different architectures::
652
653  k = kprobe; r = return probe; kr = kprobe + return probe
654  on same function
655
656  i386: Intel Pentium M, 1495 MHz, 2957.31 bogomips
657  k = 0.57 usec; r = 0.92; kr = 0.99
658
659  x86_64: AMD Opteron 246, 1994 MHz, 3971.48 bogomips
660  k = 0.49 usec; r = 0.80; kr = 0.82
661
662  ppc64: POWER5 (gr), 1656 MHz (SMT disabled, 1 virtual CPU per physical CPU)
663  k = 0.77 usec; r = 1.26; kr = 1.45
664
665Optimized Probe Overhead
666------------------------
667
668Typically, an optimized kprobe hit takes 0.07 to 0.1 microseconds to
669process. Here are sample overhead figures (in usec) for x86 architectures::
670
671  k = unoptimized kprobe, b = boosted (single-step skipped), o = optimized kprobe,
672  r = unoptimized kretprobe, rb = boosted kretprobe, ro = optimized kretprobe.
673
674  i386: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips
675  k = 0.80 usec; b = 0.33; o = 0.05; r = 1.10; rb = 0.61; ro = 0.33
676
677  x86-64: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips
678  k = 0.99 usec; b = 0.43; o = 0.06; r = 1.24; rb = 0.68; ro = 0.30
679
680TODO
681====
682
683a. SystemTap (http://sourceware.org/systemtap): Provides a simplified
684   programming interface for probe-based instrumentation.  Try it out.
685b. Kernel return probes for sparc64.
686c. Support for other architectures.
687d. User-space probes.
688e. Watchpoint probes (which fire on data references).
689
690Kprobes Example
691===============
692
693See samples/kprobes/kprobe_example.c
694
695Kretprobes Example
696==================
697
698See samples/kprobes/kretprobe_example.c
699
700Deprecated Features
701===================
702
703Jprobes is now a deprecated feature. People who are depending on it should
704migrate to other tracing features or use older kernels. Please consider to
705migrate your tool to one of the following options:
706
707- Use trace-event to trace target function with arguments.
708
709  trace-event is a low-overhead (and almost no visible overhead if it
710  is off) statically defined event interface. You can define new events
711  and trace it via ftrace or any other tracing tools.
712
713  See the following urls:
714
715    - https://lwn.net/Articles/379903/
716    - https://lwn.net/Articles/381064/
717    - https://lwn.net/Articles/383362/
718
719- Use ftrace dynamic events (kprobe event) with perf-probe.
720
721  If you build your kernel with debug info (CONFIG_DEBUG_INFO=y), you can
722  find which register/stack is assigned to which local variable or arguments
723  by using perf-probe and set up new event to trace it.
724
725  See following documents:
726
727  - Documentation/trace/kprobetrace.rst
728  - Documentation/trace/events.rst
729  - tools/perf/Documentation/perf-probe.txt
730
731
732The kprobes debugfs interface
733=============================
734
735
736With recent kernels (> 2.6.20) the list of registered kprobes is visible
737under the /sys/kernel/debug/kprobes/ directory (assuming debugfs is mounted at //sys/kernel/debug).
738
739/sys/kernel/debug/kprobes/list: Lists all registered probes on the system::
740
741	c015d71a  k  vfs_read+0x0
742	c03dedc5  r  tcp_v4_rcv+0x0
743
744The first column provides the kernel address where the probe is inserted.
745The second column identifies the type of probe (k - kprobe and r - kretprobe)
746while the third column specifies the symbol+offset of the probe.
747If the probed function belongs to a module, the module name is also
748specified. Following columns show probe status. If the probe is on
749a virtual address that is no longer valid (module init sections, module
750virtual addresses that correspond to modules that've been unloaded),
751such probes are marked with [GONE]. If the probe is temporarily disabled,
752such probes are marked with [DISABLED]. If the probe is optimized, it is
753marked with [OPTIMIZED]. If the probe is ftrace-based, it is marked with
754[FTRACE].
755
756/sys/kernel/debug/kprobes/enabled: Turn kprobes ON/OFF forcibly.
757
758Provides a knob to globally and forcibly turn registered kprobes ON or OFF.
759By default, all kprobes are enabled. By echoing "0" to this file, all
760registered probes will be disarmed, till such time a "1" is echoed to this
761file. Note that this knob just disarms and arms all kprobes and doesn't
762change each probe's disabling state. This means that disabled kprobes (marked
763[DISABLED]) will be not enabled if you turn ON all kprobes by this knob.
764
765
766The kprobes sysctl interface
767============================
768
769/proc/sys/debug/kprobes-optimization: Turn kprobes optimization ON/OFF.
770
771When CONFIG_OPTPROBES=y, this sysctl interface appears and it provides
772a knob to globally and forcibly turn jump optimization (see section
773:ref:`kprobes_jump_optimization`) ON or OFF. By default, jump optimization
774is allowed (ON). If you echo "0" to this file or set
775"debug.kprobes_optimization" to 0 via sysctl, all optimized probes will be
776unoptimized, and any new probes registered after that will not be optimized.
777
778Note that this knob *changes* the optimized state. This means that optimized
779probes (marked [OPTIMIZED]) will be unoptimized ([OPTIMIZED] tag will be
780removed). If the knob is turned on, they will be optimized again.
781
782References
783==========
784
785For additional information on Kprobes, refer to the following URLs:
786
787- https://lwn.net/Articles/132196/
788- https://www.kernel.org/doc/ols/2006/ols2006v2-pages-109-124.pdf
789
790