linux/Documentation/kprobes.txt
<<
>>
Prefs
   1Title   : Kernel Probes (Kprobes)
   2Authors : Jim Keniston <jkenisto@us.ibm.com>
   3        : Prasanna S Panchamukhi <prasanna.panchamukhi@gmail.com>
   4        : Masami Hiramatsu <mhiramat@redhat.com>
   5
   6CONTENTS
   7
   81. Concepts: Kprobes, Jprobes, Return Probes
   92. Architectures Supported
  103. Configuring Kprobes
  114. API Reference
  125. Kprobes Features and Limitations
  136. Probe Overhead
  147. TODO
  158. Kprobes Example
  169. Jprobes Example
  1710. Kretprobes Example
  18Appendix A: The kprobes debugfs interface
  19Appendix B: The kprobes sysctl interface
  20
  211. Concepts: Kprobes, Jprobes, Return Probes
  22
  23Kprobes enables you to dynamically break into any kernel routine and
  24collect debugging and performance information non-disruptively. You
  25can trap at almost any kernel code address(*), specifying a handler
  26routine to be invoked when the breakpoint is hit.
  27(*: some parts of the kernel code can not be trapped, see 1.5 Blacklist)
  28
  29There are currently three types of probes: kprobes, jprobes, and
  30kretprobes (also called return probes).  A kprobe can be inserted
  31on virtually any instruction in the kernel.  A jprobe is inserted at
  32the entry to a kernel function, and provides convenient access to the
  33function's arguments.  A return probe fires when a specified function
  34returns.
  35
  36In the typical case, Kprobes-based instrumentation is packaged as
  37a kernel module.  The module's init function installs ("registers")
  38one or more probes, and the exit function unregisters them.  A
  39registration function such as register_kprobe() specifies where
  40the probe is to be inserted and what handler is to be called when
  41the probe is hit.
  42
  43There are also register_/unregister_*probes() functions for batch
  44registration/unregistration of a group of *probes. These functions
  45can speed up unregistration process when you have to unregister
  46a lot of probes at once.
  47
  48The next four subsections explain how the different types of
  49probes work and how jump optimization works.  They explain certain
  50things that you'll need to know in order to make the best use of
  51Kprobes -- e.g., the difference between a pre_handler and
  52a post_handler, and how to use the maxactive and nmissed fields of
  53a kretprobe.  But if you're in a hurry to start using Kprobes, you
  54can skip ahead to section 2.
  55
  561.1 How Does a Kprobe Work?
  57
  58When a kprobe is registered, Kprobes makes a copy of the probed
  59instruction and replaces the first byte(s) of the probed instruction
  60with a breakpoint instruction (e.g., int3 on i386 and x86_64).
  61
  62When a CPU hits the breakpoint instruction, a trap occurs, the CPU's
  63registers are saved, and control passes to Kprobes via the
  64notifier_call_chain mechanism.  Kprobes executes the "pre_handler"
  65associated with the kprobe, passing the handler the addresses of the
  66kprobe struct and the saved registers.
  67
  68Next, Kprobes single-steps its copy of the probed instruction.
  69(It would be simpler to single-step the actual instruction in place,
  70but then Kprobes would have to temporarily remove the breakpoint
  71instruction.  This would open a small time window when another CPU
  72could sail right past the probepoint.)
  73
  74After the instruction is single-stepped, Kprobes executes the
  75"post_handler," if any, that is associated with the kprobe.
  76Execution then continues with the instruction following the probepoint.
  77
  781.2 How Does a Jprobe Work?
  79
  80A jprobe is implemented using a kprobe that is placed on a function's
  81entry point.  It employs a simple mirroring principle to allow
  82seamless access to the probed function's arguments.  The jprobe
  83handler routine should have the same signature (arg list and return
  84type) as the function being probed, and must always end by calling
  85the Kprobes function jprobe_return().
  86
  87Here's how it works.  When the probe is hit, Kprobes makes a copy of
  88the saved registers and a generous portion of the stack (see below).
  89Kprobes then points the saved instruction pointer at the jprobe's
  90handler routine, and returns from the trap.  As a result, control
  91passes to the handler, which is presented with the same register and
  92stack contents as the probed function.  When it is done, the handler
  93calls jprobe_return(), which traps again to restore the original stack
  94contents and processor state and switch to the probed function.
  95
  96By convention, the callee owns its arguments, so gcc may produce code
  97that unexpectedly modifies that portion of the stack.  This is why
  98Kprobes saves a copy of the stack and restores it after the jprobe
  99handler has run.  Up to MAX_STACK_SIZE bytes are copied -- e.g.,
 10064 bytes on i386.
 101
 102Note that the probed function's args may be passed on the stack
 103or in registers.  The jprobe will work in either case, so long as the
 104handler's prototype matches that of the probed function.
 105
 106Note that in some architectures (e.g.: arm64 and sparc64) the stack
 107copy is not done, as the actual location of stacked parameters may be
 108outside of a reasonable MAX_STACK_SIZE value and because that location
 109cannot be determined by the jprobes code. In this case the jprobes
 110user must be careful to make certain the calling signature of the
 111function does not cause parameters to be passed on the stack (e.g.:
 112more than eight function arguments, an argument of more than sixteen
 113bytes, or more than 64 bytes of argument data, depending on
 114architecture).
 115
 1161.3 Return Probes
 117
 1181.3.1 How Does a Return Probe Work?
 119
 120When you call register_kretprobe(), Kprobes establishes a kprobe at
 121the entry to the function.  When the probed function is called and this
 122probe is hit, Kprobes saves a copy of the return address, and replaces
 123the return address with the address of a "trampoline."  The trampoline
 124is an arbitrary piece of code -- typically just a nop instruction.
 125At boot time, Kprobes registers a kprobe at the trampoline.
 126
 127When the probed function executes its return instruction, control
 128passes to the trampoline and that probe is hit.  Kprobes' trampoline
 129handler calls the user-specified return handler associated with the
 130kretprobe, then sets the saved instruction pointer to the saved return
 131address, and that's where execution resumes upon return from the trap.
 132
 133While the probed function is executing, its return address is
 134stored in an object of type kretprobe_instance.  Before calling
 135register_kretprobe(), the user sets the maxactive field of the
 136kretprobe struct to specify how many instances of the specified
 137function can be probed simultaneously.  register_kretprobe()
 138pre-allocates the indicated number of kretprobe_instance objects.
 139
 140For example, if the function is non-recursive and is called with a
 141spinlock held, maxactive = 1 should be enough.  If the function is
 142non-recursive and can never relinquish the CPU (e.g., via a semaphore
 143or preemption), NR_CPUS should be enough.  If maxactive <= 0, it is
 144set to a default value.  If CONFIG_PREEMPT is enabled, the default
 145is max(10, 2*NR_CPUS).  Otherwise, the default is NR_CPUS.
 146
 147It's not a disaster if you set maxactive too low; you'll just miss
 148some probes.  In the kretprobe struct, the nmissed field is set to
 149zero when the return probe is registered, and is incremented every
 150time the probed function is entered but there is no kretprobe_instance
 151object available for establishing the return probe.
 152
 1531.3.2 Kretprobe entry-handler
 154
 155Kretprobes also provides an optional user-specified handler which runs
 156on function entry. This handler is specified by setting the entry_handler
 157field of the kretprobe struct. Whenever the kprobe placed by kretprobe at the
 158function entry is hit, the user-defined entry_handler, if any, is invoked.
 159If the entry_handler returns 0 (success) then a corresponding return handler
 160is guaranteed to be called upon function return. If the entry_handler
 161returns a non-zero error then Kprobes leaves the return address as is, and
 162the kretprobe has no further effect for that particular function instance.
 163
 164Multiple entry and return handler invocations are matched using the unique
 165kretprobe_instance object associated with them. Additionally, a user
 166may also specify per return-instance private data to be part of each
 167kretprobe_instance object. This is especially useful when sharing private
 168data between corresponding user entry and return handlers. The size of each
 169private data object can be specified at kretprobe registration time by
 170setting the data_size field of the kretprobe struct. This data can be
 171accessed through the data field of each kretprobe_instance object.
 172
 173In case probed function is entered but there is no kretprobe_instance
 174object available, then in addition to incrementing the nmissed count,
 175the user entry_handler invocation is also skipped.
 176
 1771.4 How Does Jump Optimization Work?
 178
 179If your kernel is built with CONFIG_OPTPROBES=y (currently this flag
 180is automatically set 'y' on x86/x86-64, non-preemptive kernel) and
 181the "debug.kprobes_optimization" kernel parameter is set to 1 (see
 182sysctl(8)), Kprobes tries to reduce probe-hit overhead by using a jump
 183instruction instead of a breakpoint instruction at each probepoint.
 184
 1851.4.1 Init a Kprobe
 186
 187When a probe is registered, before attempting this optimization,
 188Kprobes inserts an ordinary, breakpoint-based kprobe at the specified
 189address. So, even if it's not possible to optimize this particular
 190probepoint, there'll be a probe there.
 191
 1921.4.2 Safety Check
 193
 194Before optimizing a probe, Kprobes performs the following safety checks:
 195
 196- Kprobes verifies that the region that will be replaced by the jump
 197instruction (the "optimized region") lies entirely within one function.
 198(A jump instruction is multiple bytes, and so may overlay multiple
 199instructions.)
 200
 201- Kprobes analyzes the entire function and verifies that there is no
 202jump into the optimized region.  Specifically:
 203  - the function contains no indirect jump;
 204  - the function contains no instruction that causes an exception (since
 205  the fixup code triggered by the exception could jump back into the
 206  optimized region -- Kprobes checks the exception tables to verify this);
 207  and
 208  - there is no near jump to the optimized region (other than to the first
 209  byte).
 210
 211- For each instruction in the optimized region, Kprobes verifies that
 212the instruction can be executed out of line.
 213
 2141.4.3 Preparing Detour Buffer
 215
 216Next, Kprobes prepares a "detour" buffer, which contains the following
 217instruction sequence:
 218- code to push the CPU's registers (emulating a breakpoint trap)
 219- a call to the trampoline code which calls user's probe handlers.
 220- code to restore registers
 221- the instructions from the optimized region
 222- a jump back to the original execution path.
 223
 2241.4.4 Pre-optimization
 225
 226After preparing the detour buffer, Kprobes verifies that none of the
 227following situations exist:
 228- The probe has either a break_handler (i.e., it's a jprobe) or a
 229post_handler.
 230- Other instructions in the optimized region are probed.
 231- The probe is disabled.
 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
 2431.4.5 Optimization
 244
 245The Kprobe-optimizer doesn't insert the jump instruction immediately;
 246rather, it calls synchronize_sched() for safety first, because it's
 247possible for a CPU to be interrupted in the middle of executing the
 248optimized region(*).  As you know, synchronize_sched() can ensure
 249that all interruptions that were active when synchronize_sched()
 250was called are done, but only if CONFIG_PREEMPT=n.  So, this version
 251of kprobe optimization supports only kernels with CONFIG_PREEMPT=n.(**)
 252
 253After that, the Kprobe-optimizer calls stop_machine() to replace
 254the optimized region with a jump instruction to the detour buffer,
 255using text_poke_smp().
 256
 2571.4.6 Unoptimization
 258
 259When an optimized kprobe is unregistered, disabled, or blocked by
 260another kprobe, it will be unoptimized.  If this happens before
 261the optimization is complete, the kprobe is just dequeued from the
 262optimized list.  If the optimization has been done, the jump is
 263replaced with the original code (except for an int3 breakpoint in
 264the first byte) by using text_poke_smp().
 265
 266(*)Please imagine that the 2nd instruction is interrupted and then
 267the optimizer replaces the 2nd instruction with the jump *address*
 268while the interrupt handler is running. When the interrupt
 269returns to original address, there is no valid instruction,
 270and it causes an unexpected result.
 271
 272(**)This optimization-safety checking may be replaced with the
 273stop-machine method that ksplice uses for supporting a CONFIG_PREEMPT=y
 274kernel.
 275
 276NOTE for geeks:
 277The jump optimization changes the kprobe's pre_handler behavior.
 278Without optimization, the pre_handler can change the kernel's execution
 279path by changing regs->ip and returning 1.  However, when the probe
 280is optimized, that modification is ignored.  Thus, if you want to
 281tweak the kernel's execution path, you need to suppress optimization,
 282using one of the following techniques:
 283- Specify an empty function for the kprobe's post_handler or break_handler.
 284 or
 285- Execute 'sysctl -w debug.kprobes_optimization=n'
 286
 2871.5 Blacklist
 288
 289Kprobes can probe most of the kernel except itself. This means
 290that there are some functions where kprobes cannot probe. Probing
 291(trapping) such functions can cause a recursive trap (e.g. double
 292fault) or the nested probe handler may never be called.
 293Kprobes manages such functions as a blacklist.
 294If you want to add a function into the blacklist, you just need
 295to (1) include linux/kprobes.h and (2) use NOKPROBE_SYMBOL() macro
 296to specify a blacklisted function.
 297Kprobes checks the given probe address against the blacklist and
 298rejects registering it, if the given address is in the blacklist.
 299
 3002. Architectures Supported
 301
 302Kprobes, jprobes, and return probes are implemented on the following
 303architectures:
 304
 305- i386 (Supports jump optimization)
 306- x86_64 (AMD-64, EM64T) (Supports jump optimization)
 307- ppc64
 308- ia64 (Does not support probes on instruction slot1.)
 309- sparc64 (Return probes not yet implemented.)
 310- arm
 311- ppc
 312- mips
 313- s390
 314
 3153. Configuring Kprobes
 316
 317When configuring the kernel using make menuconfig/xconfig/oldconfig,
 318ensure that CONFIG_KPROBES is set to "y". Under "General setup", look
 319for "Kprobes".
 320
 321So that you can load and unload Kprobes-based instrumentation modules,
 322make sure "Loadable module support" (CONFIG_MODULES) and "Module
 323unloading" (CONFIG_MODULE_UNLOAD) are set to "y".
 324
 325Also make sure that CONFIG_KALLSYMS and perhaps even CONFIG_KALLSYMS_ALL
 326are set to "y", since kallsyms_lookup_name() is used by the in-kernel
 327kprobe address resolution code.
 328
 329If you need to insert a probe in the middle of a function, you may find
 330it useful to "Compile the kernel with debug info" (CONFIG_DEBUG_INFO),
 331so you can use "objdump -d -l vmlinux" to see the source-to-object
 332code mapping.
 333
 3344. API Reference
 335
 336The Kprobes API includes a "register" function and an "unregister"
 337function for each type of probe. The API also includes "register_*probes"
 338and "unregister_*probes" functions for (un)registering arrays of probes.
 339Here are terse, mini-man-page specifications for these functions and
 340the associated probe handlers that you'll write. See the files in the
 341samples/kprobes/ sub-directory for examples.
 342
 3434.1 register_kprobe
 344
 345#include <linux/kprobes.h>
 346int register_kprobe(struct kprobe *kp);
 347
 348Sets a breakpoint at the address kp->addr.  When the breakpoint is
 349hit, Kprobes calls kp->pre_handler.  After the probed instruction
 350is single-stepped, Kprobe calls kp->post_handler.  If a fault
 351occurs during execution of kp->pre_handler or kp->post_handler,
 352or during single-stepping of the probed instruction, Kprobes calls
 353kp->fault_handler.  Any or all handlers can be NULL. If kp->flags
 354is set KPROBE_FLAG_DISABLED, that kp will be registered but disabled,
 355so, its handlers aren't hit until calling enable_kprobe(kp).
 356
 357NOTE:
 3581. With the introduction of the "symbol_name" field to struct kprobe,
 359the probepoint address resolution will now be taken care of by the kernel.
 360The following will now work:
 361
 362        kp.symbol_name = "symbol_name";
 363
 364(64-bit powerpc intricacies such as function descriptors are handled
 365transparently)
 366
 3672. Use the "offset" field of struct kprobe if the offset into the symbol
 368to install a probepoint is known. This field is used to calculate the
 369probepoint.
 370
 3713. Specify either the kprobe "symbol_name" OR the "addr". If both are
 372specified, kprobe registration will fail with -EINVAL.
 373
 3744. With CISC architectures (such as i386 and x86_64), the kprobes code
 375does not validate if the kprobe.addr is at an instruction boundary.
 376Use "offset" with caution.
 377
 378register_kprobe() returns 0 on success, or a negative errno otherwise.
 379
 380User's pre-handler (kp->pre_handler):
 381#include <linux/kprobes.h>
 382#include <linux/ptrace.h>
 383int pre_handler(struct kprobe *p, struct pt_regs *regs);
 384
 385Called with p pointing to the kprobe associated with the breakpoint,
 386and regs pointing to the struct containing the registers saved when
 387the breakpoint was hit.  Return 0 here unless you're a Kprobes geek.
 388
 389User's post-handler (kp->post_handler):
 390#include <linux/kprobes.h>
 391#include <linux/ptrace.h>
 392void post_handler(struct kprobe *p, struct pt_regs *regs,
 393        unsigned long flags);
 394
 395p and regs are as described for the pre_handler.  flags always seems
 396to be zero.
 397
 398User's fault-handler (kp->fault_handler):
 399#include <linux/kprobes.h>
 400#include <linux/ptrace.h>
 401int fault_handler(struct kprobe *p, struct pt_regs *regs, int trapnr);
 402
 403p and regs are as described for the pre_handler.  trapnr is the
 404architecture-specific trap number associated with the fault (e.g.,
 405on i386, 13 for a general protection fault or 14 for a page fault).
 406Returns 1 if it successfully handled the exception.
 407
 4084.2 register_jprobe
 409
 410#include <linux/kprobes.h>
 411int register_jprobe(struct jprobe *jp)
 412
 413Sets a breakpoint at the address jp->kp.addr, which must be the address
 414of the first instruction of a function.  When the breakpoint is hit,
 415Kprobes runs the handler whose address is jp->entry.
 416
 417The handler should have the same arg list and return type as the probed
 418function; and just before it returns, it must call jprobe_return().
 419(The handler never actually returns, since jprobe_return() returns
 420control to Kprobes.)  If the probed function is declared asmlinkage
 421or anything else that affects how args are passed, the handler's
 422declaration must match.
 423
 424register_jprobe() returns 0 on success, or a negative errno otherwise.
 425
 4264.3 register_kretprobe
 427
 428#include <linux/kprobes.h>
 429int register_kretprobe(struct kretprobe *rp);
 430
 431Establishes a return probe for the function whose address is
 432rp->kp.addr.  When that function returns, Kprobes calls rp->handler.
 433You must set rp->maxactive appropriately before you call
 434register_kretprobe(); see "How Does a Return Probe Work?" for details.
 435
 436register_kretprobe() returns 0 on success, or a negative errno
 437otherwise.
 438
 439User's return-probe handler (rp->handler):
 440#include <linux/kprobes.h>
 441#include <linux/ptrace.h>
 442int kretprobe_handler(struct kretprobe_instance *ri, struct pt_regs *regs);
 443
 444regs is as described for kprobe.pre_handler.  ri points to the
 445kretprobe_instance object, of which the following fields may be
 446of interest:
 447- ret_addr: the return address
 448- rp: points to the corresponding kretprobe object
 449- task: points to the corresponding task struct
 450- data: points to per return-instance private data; see "Kretprobe
 451        entry-handler" for details.
 452
 453The regs_return_value(regs) macro provides a simple abstraction to
 454extract the return value from the appropriate register as defined by
 455the architecture's ABI.
 456
 457The handler's return value is currently ignored.
 458
 4594.4 unregister_*probe
 460
 461#include <linux/kprobes.h>
 462void unregister_kprobe(struct kprobe *kp);
 463void unregister_jprobe(struct jprobe *jp);
 464void unregister_kretprobe(struct kretprobe *rp);
 465
 466Removes the specified probe.  The unregister function can be called
 467at any time after the probe has been registered.
 468
 469NOTE:
 470If the functions find an incorrect probe (ex. an unregistered probe),
 471they clear the addr field of the probe.
 472
 4734.5 register_*probes
 474
 475#include <linux/kprobes.h>
 476int register_kprobes(struct kprobe **kps, int num);
 477int register_kretprobes(struct kretprobe **rps, int num);
 478int register_jprobes(struct jprobe **jps, 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- kps/rps/jps: an array of pointers to *probe data structures
 485- num: the number of the array entries.
 486
 487NOTE:
 488You have to allocate(or define) an array of pointers and set all
 489of the array entries before using these functions.
 490
 4914.6 unregister_*probes
 492
 493#include <linux/kprobes.h>
 494void unregister_kprobes(struct kprobe **kps, int num);
 495void unregister_kretprobes(struct kretprobe **rps, int num);
 496void unregister_jprobes(struct jprobe **jps, int num);
 497
 498Removes each of the num probes in the specified array at once.
 499
 500NOTE:
 501If the functions find some incorrect probes (ex. unregistered
 502probes) in the specified array, they clear the addr field of those
 503incorrect probes. However, other probes in the array are
 504unregistered correctly.
 505
 5064.7 disable_*probe
 507
 508#include <linux/kprobes.h>
 509int disable_kprobe(struct kprobe *kp);
 510int disable_kretprobe(struct kretprobe *rp);
 511int disable_jprobe(struct jprobe *jp);
 512
 513Temporarily disables the specified *probe. You can enable it again by using
 514enable_*probe(). You must specify the probe which has been registered.
 515
 5164.8 enable_*probe
 517
 518#include <linux/kprobes.h>
 519int enable_kprobe(struct kprobe *kp);
 520int enable_kretprobe(struct kretprobe *rp);
 521int enable_jprobe(struct jprobe *jp);
 522
 523Enables *probe which has been disabled by disable_*probe(). You must specify
 524the probe which has been registered.
 525
 5265. Kprobes Features and Limitations
 527
 528Kprobes allows multiple probes at the same address.  Currently,
 529however, there cannot be multiple jprobes on the same function at
 530the same time.  Also, a probepoint for which there is a jprobe or
 531a post_handler cannot be optimized.  So if you install a jprobe,
 532or a kprobe with a post_handler, at an optimized probepoint, the
 533probepoint will be unoptimized automatically.
 534
 535In general, you can install a probe anywhere in the kernel.
 536In particular, you can probe interrupt handlers.  Known exceptions
 537are discussed in this section.
 538
 539The register_*probe functions will return -EINVAL if you attempt
 540to install a probe in the code that implements Kprobes (mostly
 541kernel/kprobes.c and arch/*/kernel/kprobes.c, but also functions such
 542as do_page_fault and notifier_call_chain).
 543
 544If you install a probe in an inline-able function, Kprobes makes
 545no attempt to chase down all inline instances of the function and
 546install probes there.  gcc may inline a function without being asked,
 547so keep this in mind if you're not seeing the probe hits you expect.
 548
 549A probe handler can modify the environment of the probed function
 550-- e.g., by modifying kernel data structures, or by modifying the
 551contents of the pt_regs struct (which are restored to the registers
 552upon return from the breakpoint).  So Kprobes can be used, for example,
 553to install a bug fix or to inject faults for testing.  Kprobes, of
 554course, has no way to distinguish the deliberately injected faults
 555from the accidental ones.  Don't drink and probe.
 556
 557Kprobes makes no attempt to prevent probe handlers from stepping on
 558each other -- e.g., probing printk() and then calling printk() from a
 559probe handler.  If a probe handler hits a probe, that second probe's
 560handlers won't be run in that instance, and the kprobe.nmissed member
 561of the second probe will be incremented.
 562
 563As of Linux v2.6.15-rc1, multiple handlers (or multiple instances of
 564the same handler) may run concurrently on different CPUs.
 565
 566Kprobes does not use mutexes or allocate memory except during
 567registration and unregistration.
 568
 569Probe handlers are run with preemption disabled.  Depending on the
 570architecture and optimization state, handlers may also run with
 571interrupts disabled (e.g., kretprobe handlers and optimized kprobe
 572handlers run without interrupt disabled on x86/x86-64).  In any case,
 573your handler should not yield the CPU (e.g., by attempting to acquire
 574a semaphore).
 575
 576Since a return probe is implemented by replacing the return
 577address with the trampoline's address, stack backtraces and calls
 578to __builtin_return_address() will typically yield the trampoline's
 579address instead of the real return address for kretprobed functions.
 580(As far as we can tell, __builtin_return_address() is used only
 581for instrumentation and error reporting.)
 582
 583If the number of times a function is called does not match the number
 584of times it returns, registering a return probe on that function may
 585produce undesirable results. In such a case, a line:
 586kretprobe BUG!: Processing kretprobe d000000000041aa8 @ c00000000004f48c
 587gets printed. With this information, one will be able to correlate the
 588exact instance of the kretprobe that caused the problem. We have the
 589do_exit() case covered. do_execve() and do_fork() are not an issue.
 590We're unaware of other specific cases where this could be a problem.
 591
 592If, upon entry to or exit from a function, the CPU is running on
 593a stack other than that of the current task, registering a return
 594probe on that function may produce undesirable results.  For this
 595reason, Kprobes doesn't support return probes (or kprobes or jprobes)
 596on the x86_64 version of __switch_to(); the registration functions
 597return -EINVAL.
 598
 599On x86/x86-64, since the Jump Optimization of Kprobes modifies
 600instructions widely, there are some limitations to optimization. To
 601explain it, we introduce some terminology. Imagine a 3-instruction
 602sequence consisting of a two 2-byte instructions and one 3-byte
 603instruction.
 604
 605        IA
 606         |
 607[-2][-1][0][1][2][3][4][5][6][7]
 608        [ins1][ins2][  ins3 ]
 609        [<-     DCR       ->]
 610           [<- JTPR ->]
 611
 612ins1: 1st Instruction
 613ins2: 2nd Instruction
 614ins3: 3rd Instruction
 615IA:  Insertion Address
 616JTPR: Jump Target Prohibition Region
 617DCR: Detoured Code Region
 618
 619The instructions in DCR are copied to the out-of-line buffer
 620of the kprobe, because the bytes in DCR are replaced by
 621a 5-byte jump instruction. So there are several limitations.
 622
 623a) The instructions in DCR must be relocatable.
 624b) The instructions in DCR must not include a call instruction.
 625c) JTPR must not be targeted by any jump or call instruction.
 626d) DCR must not straddle the border between functions.
 627
 628Anyway, these limitations are checked by the in-kernel instruction
 629decoder, so you don't need to worry about that.
 630
 6316. Probe Overhead
 632
 633On a typical CPU in use in 2005, a kprobe hit takes 0.5 to 1.0
 634microseconds to process.  Specifically, a benchmark that hits the same
 635probepoint repeatedly, firing a simple handler each time, reports 1-2
 636million hits per second, depending on the architecture.  A jprobe or
 637return-probe hit typically takes 50-75% longer than a kprobe hit.
 638When you have a return probe set on a function, adding a kprobe at
 639the entry to that function adds essentially no overhead.
 640
 641Here are sample overhead figures (in usec) for different architectures.
 642k = kprobe; j = jprobe; r = return probe; kr = kprobe + return probe
 643on same function; jr = jprobe + return probe on same function
 644
 645i386: Intel Pentium M, 1495 MHz, 2957.31 bogomips
 646k = 0.57 usec; j = 1.00; r = 0.92; kr = 0.99; jr = 1.40
 647
 648x86_64: AMD Opteron 246, 1994 MHz, 3971.48 bogomips
 649k = 0.49 usec; j = 0.76; r = 0.80; kr = 0.82; jr = 1.07
 650
 651ppc64: POWER5 (gr), 1656 MHz (SMT disabled, 1 virtual CPU per physical CPU)
 652k = 0.77 usec; j = 1.31; r = 1.26; kr = 1.45; jr = 1.99
 653
 6546.1 Optimized Probe Overhead
 655
 656Typically, an optimized kprobe hit takes 0.07 to 0.1 microseconds to
 657process. Here are sample overhead figures (in usec) for x86 architectures.
 658k = unoptimized kprobe, b = boosted (single-step skipped), o = optimized kprobe,
 659r = unoptimized kretprobe, rb = boosted kretprobe, ro = optimized kretprobe.
 660
 661i386: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips
 662k = 0.80 usec; b = 0.33; o = 0.05; r = 1.10; rb = 0.61; ro = 0.33
 663
 664x86-64: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips
 665k = 0.99 usec; b = 0.43; o = 0.06; r = 1.24; rb = 0.68; ro = 0.30
 666
 6677. TODO
 668
 669a. SystemTap (http://sourceware.org/systemtap): Provides a simplified
 670programming interface for probe-based instrumentation.  Try it out.
 671b. Kernel return probes for sparc64.
 672c. Support for other architectures.
 673d. User-space probes.
 674e. Watchpoint probes (which fire on data references).
 675
 6768. Kprobes Example
 677
 678See samples/kprobes/kprobe_example.c
 679
 6809. Jprobes Example
 681
 682See samples/kprobes/jprobe_example.c
 683
 68410. Kretprobes Example
 685
 686See samples/kprobes/kretprobe_example.c
 687
 688For additional information on Kprobes, refer to the following URLs:
 689http://www-106.ibm.com/developerworks/library/l-kprobes.html?ca=dgr-lnxw42Kprobe
 690http://www.redhat.com/magazine/005mar05/features/kprobes/
 691http://www-users.cs.umn.edu/~boutcher/kprobes/
 692http://www.linuxsymposium.org/2006/linuxsymposium_procv2.pdf (pages 101-115)
 693
 694
 695Appendix A: The kprobes debugfs interface
 696
 697With recent kernels (> 2.6.20) the list of registered kprobes is visible
 698under the /sys/kernel/debug/kprobes/ directory (assuming debugfs is mounted at //sys/kernel/debug).
 699
 700/sys/kernel/debug/kprobes/list: Lists all registered probes on the system
 701
 702c015d71a  k  vfs_read+0x0
 703c011a316  j  do_fork+0x0
 704c03dedc5  r  tcp_v4_rcv+0x0
 705
 706The first column provides the kernel address where the probe is inserted.
 707The second column identifies the type of probe (k - kprobe, r - kretprobe
 708and j - jprobe), while the third column specifies the symbol+offset of
 709the probe. If the probed function belongs to a module, the module name
 710is also specified. Following columns show probe status. If the probe is on
 711a virtual address that is no longer valid (module init sections, module
 712virtual addresses that correspond to modules that've been unloaded),
 713such probes are marked with [GONE]. If the probe is temporarily disabled,
 714such probes are marked with [DISABLED]. If the probe is optimized, it is
 715marked with [OPTIMIZED]. If the probe is ftrace-based, it is marked with
 716[FTRACE].
 717
 718/sys/kernel/debug/kprobes/enabled: Turn kprobes ON/OFF forcibly.
 719
 720Provides a knob to globally and forcibly turn registered kprobes ON or OFF.
 721By default, all kprobes are enabled. By echoing "0" to this file, all
 722registered probes will be disarmed, till such time a "1" is echoed to this
 723file. Note that this knob just disarms and arms all kprobes and doesn't
 724change each probe's disabling state. This means that disabled kprobes (marked
 725[DISABLED]) will be not enabled if you turn ON all kprobes by this knob.
 726
 727
 728Appendix B: The kprobes sysctl interface
 729
 730/proc/sys/debug/kprobes-optimization: Turn kprobes optimization ON/OFF.
 731
 732When CONFIG_OPTPROBES=y, this sysctl interface appears and it provides
 733a knob to globally and forcibly turn jump optimization (see section
 7341.4) ON or OFF. By default, jump optimization is allowed (ON).
 735If you echo "0" to this file or set "debug.kprobes_optimization" to
 7360 via sysctl, all optimized probes will be unoptimized, and any new
 737probes registered after that will not be optimized.  Note that this
 738knob *changes* the optimized state. This means that optimized probes
 739(marked [OPTIMIZED]) will be unoptimized ([OPTIMIZED] tag will be
 740removed). If the knob is turned on, they will be optimized again.
 741
 742