Based on kernel version 4.1. Page generated on 2015-06-28 12:13 EST.
1 Title : Kernel Probes (Kprobes) 2 Authors : Jim Keniston <firstname.lastname@example.org> 3 : Prasanna S Panchamukhi <email@example.com> 4 : Masami Hiramatsu <firstname.lastname@example.org> 5 6 CONTENTS 7 8 1. Concepts: Kprobes, Jprobes, Return Probes 9 2. Architectures Supported 10 3. Configuring Kprobes 11 4. API Reference 12 5. Kprobes Features and Limitations 13 6. Probe Overhead 14 7. TODO 15 8. Kprobes Example 16 9. Jprobes Example 17 10. Kretprobes Example 18 Appendix A: The kprobes debugfs interface 19 Appendix B: The kprobes sysctl interface 20 21 1. Concepts: Kprobes, Jprobes, Return Probes 22 23 Kprobes enables you to dynamically break into any kernel routine and 24 collect debugging and performance information non-disruptively. You 25 can trap at almost any kernel code address(*), specifying a handler 26 routine to be invoked when the breakpoint is hit. 27 (*: some parts of the kernel code can not be trapped, see 1.5 Blacklist) 28 29 There are currently three types of probes: kprobes, jprobes, and 30 kretprobes (also called return probes). A kprobe can be inserted 31 on virtually any instruction in the kernel. A jprobe is inserted at 32 the entry to a kernel function, and provides convenient access to the 33 function's arguments. A return probe fires when a specified function 34 returns. 35 36 In the typical case, Kprobes-based instrumentation is packaged as 37 a kernel module. The module's init function installs ("registers") 38 one or more probes, and the exit function unregisters them. A 39 registration function such as register_kprobe() specifies where 40 the probe is to be inserted and what handler is to be called when 41 the probe is hit. 42 43 There are also register_/unregister_*probes() functions for batch 44 registration/unregistration of a group of *probes. These functions 45 can speed up unregistration process when you have to unregister 46 a lot of probes at once. 47 48 The next four subsections explain how the different types of 49 probes work and how jump optimization works. They explain certain 50 things that you'll need to know in order to make the best use of 51 Kprobes -- e.g., the difference between a pre_handler and 52 a post_handler, and how to use the maxactive and nmissed fields of 53 a kretprobe. But if you're in a hurry to start using Kprobes, you 54 can skip ahead to section 2. 55 56 1.1 How Does a Kprobe Work? 57 58 When a kprobe is registered, Kprobes makes a copy of the probed 59 instruction and replaces the first byte(s) of the probed instruction 60 with a breakpoint instruction (e.g., int3 on i386 and x86_64). 61 62 When a CPU hits the breakpoint instruction, a trap occurs, the CPU's 63 registers are saved, and control passes to Kprobes via the 64 notifier_call_chain mechanism. Kprobes executes the "pre_handler" 65 associated with the kprobe, passing the handler the addresses of the 66 kprobe struct and the saved registers. 67 68 Next, Kprobes single-steps its copy of the probed instruction. 69 (It would be simpler to single-step the actual instruction in place, 70 but then Kprobes would have to temporarily remove the breakpoint 71 instruction. This would open a small time window when another CPU 72 could sail right past the probepoint.) 73 74 After the instruction is single-stepped, Kprobes executes the 75 "post_handler," if any, that is associated with the kprobe. 76 Execution then continues with the instruction following the probepoint. 77 78 1.2 How Does a Jprobe Work? 79 80 A jprobe is implemented using a kprobe that is placed on a function's 81 entry point. It employs a simple mirroring principle to allow 82 seamless access to the probed function's arguments. The jprobe 83 handler routine should have the same signature (arg list and return 84 type) as the function being probed, and must always end by calling 85 the Kprobes function jprobe_return(). 86 87 Here's how it works. When the probe is hit, Kprobes makes a copy of 88 the saved registers and a generous portion of the stack (see below). 89 Kprobes then points the saved instruction pointer at the jprobe's 90 handler routine, and returns from the trap. As a result, control 91 passes to the handler, which is presented with the same register and 92 stack contents as the probed function. When it is done, the handler 93 calls jprobe_return(), which traps again to restore the original stack 94 contents and processor state and switch to the probed function. 95 96 By convention, the callee owns its arguments, so gcc may produce code 97 that unexpectedly modifies that portion of the stack. This is why 98 Kprobes saves a copy of the stack and restores it after the jprobe 99 handler has run. Up to MAX_STACK_SIZE bytes are copied -- e.g., 100 64 bytes on i386. 101 102 Note that the probed function's args may be passed on the stack 103 or in registers. The jprobe will work in either case, so long as the 104 handler's prototype matches that of the probed function. 105 106 1.3 Return Probes 107 108 1.3.1 How Does a Return Probe Work? 109 110 When you call register_kretprobe(), Kprobes establishes a kprobe at 111 the entry to the function. When the probed function is called and this 112 probe is hit, Kprobes saves a copy of the return address, and replaces 113 the return address with the address of a "trampoline." The trampoline 114 is an arbitrary piece of code -- typically just a nop instruction. 115 At boot time, Kprobes registers a kprobe at the trampoline. 116 117 When the probed function executes its return instruction, control 118 passes to the trampoline and that probe is hit. Kprobes' trampoline 119 handler calls the user-specified return handler associated with the 120 kretprobe, then sets the saved instruction pointer to the saved return 121 address, and that's where execution resumes upon return from the trap. 122 123 While the probed function is executing, its return address is 124 stored in an object of type kretprobe_instance. Before calling 125 register_kretprobe(), the user sets the maxactive field of the 126 kretprobe struct to specify how many instances of the specified 127 function can be probed simultaneously. register_kretprobe() 128 pre-allocates the indicated number of kretprobe_instance objects. 129 130 For example, if the function is non-recursive and is called with a 131 spinlock held, maxactive = 1 should be enough. If the function is 132 non-recursive and can never relinquish the CPU (e.g., via a semaphore 133 or preemption), NR_CPUS should be enough. If maxactive <= 0, it is 134 set to a default value. If CONFIG_PREEMPT is enabled, the default 135 is max(10, 2*NR_CPUS). Otherwise, the default is NR_CPUS. 136 137 It's not a disaster if you set maxactive too low; you'll just miss 138 some probes. In the kretprobe struct, the nmissed field is set to 139 zero when the return probe is registered, and is incremented every 140 time the probed function is entered but there is no kretprobe_instance 141 object available for establishing the return probe. 142 143 1.3.2 Kretprobe entry-handler 144 145 Kretprobes also provides an optional user-specified handler which runs 146 on function entry. This handler is specified by setting the entry_handler 147 field of the kretprobe struct. Whenever the kprobe placed by kretprobe at the 148 function entry is hit, the user-defined entry_handler, if any, is invoked. 149 If the entry_handler returns 0 (success) then a corresponding return handler 150 is guaranteed to be called upon function return. If the entry_handler 151 returns a non-zero error then Kprobes leaves the return address as is, and 152 the kretprobe has no further effect for that particular function instance. 153 154 Multiple entry and return handler invocations are matched using the unique 155 kretprobe_instance object associated with them. Additionally, a user 156 may also specify per return-instance private data to be part of each 157 kretprobe_instance object. This is especially useful when sharing private 158 data between corresponding user entry and return handlers. The size of each 159 private data object can be specified at kretprobe registration time by 160 setting the data_size field of the kretprobe struct. This data can be 161 accessed through the data field of each kretprobe_instance object. 162 163 In case probed function is entered but there is no kretprobe_instance 164 object available, then in addition to incrementing the nmissed count, 165 the user entry_handler invocation is also skipped. 166 167 1.4 How Does Jump Optimization Work? 168 169 If your kernel is built with CONFIG_OPTPROBES=y (currently this flag 170 is automatically set 'y' on x86/x86-64, non-preemptive kernel) and 171 the "debug.kprobes_optimization" kernel parameter is set to 1 (see 172 sysctl(8)), Kprobes tries to reduce probe-hit overhead by using a jump 173 instruction instead of a breakpoint instruction at each probepoint. 174 175 1.4.1 Init a Kprobe 176 177 When a probe is registered, before attempting this optimization, 178 Kprobes inserts an ordinary, breakpoint-based kprobe at the specified 179 address. So, even if it's not possible to optimize this particular 180 probepoint, there'll be a probe there. 181 182 1.4.2 Safety Check 183 184 Before optimizing a probe, Kprobes performs the following safety checks: 185 186 - Kprobes verifies that the region that will be replaced by the jump 187 instruction (the "optimized region") lies entirely within one function. 188 (A jump instruction is multiple bytes, and so may overlay multiple 189 instructions.) 190 191 - Kprobes analyzes the entire function and verifies that there is no 192 jump into the optimized region. Specifically: 193 - the function contains no indirect jump; 194 - the function contains no instruction that causes an exception (since 195 the fixup code triggered by the exception could jump back into the 196 optimized region -- Kprobes checks the exception tables to verify this); 197 and 198 - there is no near jump to the optimized region (other than to the first 199 byte). 200 201 - For each instruction in the optimized region, Kprobes verifies that 202 the instruction can be executed out of line. 203 204 1.4.3 Preparing Detour Buffer 205 206 Next, Kprobes prepares a "detour" buffer, which contains the following 207 instruction sequence: 208 - code to push the CPU's registers (emulating a breakpoint trap) 209 - a call to the trampoline code which calls user's probe handlers. 210 - code to restore registers 211 - the instructions from the optimized region 212 - a jump back to the original execution path. 213 214 1.4.4 Pre-optimization 215 216 After preparing the detour buffer, Kprobes verifies that none of the 217 following situations exist: 218 - The probe has either a break_handler (i.e., it's a jprobe) or a 219 post_handler. 220 - Other instructions in the optimized region are probed. 221 - The probe is disabled. 222 In any of the above cases, Kprobes won't start optimizing the probe. 223 Since these are temporary situations, Kprobes tries to start 224 optimizing it again if the situation is changed. 225 226 If the kprobe can be optimized, Kprobes enqueues the kprobe to an 227 optimizing list, and kicks the kprobe-optimizer workqueue to optimize 228 it. If the to-be-optimized probepoint is hit before being optimized, 229 Kprobes returns control to the original instruction path by setting 230 the CPU's instruction pointer to the copied code in the detour buffer 231 -- thus at least avoiding the single-step. 232 233 1.4.5 Optimization 234 235 The Kprobe-optimizer doesn't insert the jump instruction immediately; 236 rather, it calls synchronize_sched() for safety first, because it's 237 possible for a CPU to be interrupted in the middle of executing the 238 optimized region(*). As you know, synchronize_sched() can ensure 239 that all interruptions that were active when synchronize_sched() 240 was called are done, but only if CONFIG_PREEMPT=n. So, this version 241 of kprobe optimization supports only kernels with CONFIG_PREEMPT=n.(**) 242 243 After that, the Kprobe-optimizer calls stop_machine() to replace 244 the optimized region with a jump instruction to the detour buffer, 245 using text_poke_smp(). 246 247 1.4.6 Unoptimization 248 249 When an optimized kprobe is unregistered, disabled, or blocked by 250 another kprobe, it will be unoptimized. If this happens before 251 the optimization is complete, the kprobe is just dequeued from the 252 optimized list. If the optimization has been done, the jump is 253 replaced with the original code (except for an int3 breakpoint in 254 the first byte) by using text_poke_smp(). 255 256 (*)Please imagine that the 2nd instruction is interrupted and then 257 the optimizer replaces the 2nd instruction with the jump *address* 258 while the interrupt handler is running. When the interrupt 259 returns to original address, there is no valid instruction, 260 and it causes an unexpected result. 261 262 (**)This optimization-safety checking may be replaced with the 263 stop-machine method that ksplice uses for supporting a CONFIG_PREEMPT=y 264 kernel. 265 266 NOTE for geeks: 267 The jump optimization changes the kprobe's pre_handler behavior. 268 Without optimization, the pre_handler can change the kernel's execution 269 path by changing regs->ip and returning 1. However, when the probe 270 is optimized, that modification is ignored. Thus, if you want to 271 tweak the kernel's execution path, you need to suppress optimization, 272 using one of the following techniques: 273 - Specify an empty function for the kprobe's post_handler or break_handler. 274 or 275 - Execute 'sysctl -w debug.kprobes_optimization=n' 276 277 1.5 Blacklist 278 279 Kprobes can probe most of the kernel except itself. This means 280 that there are some functions where kprobes cannot probe. Probing 281 (trapping) such functions can cause a recursive trap (e.g. double 282 fault) or the nested probe handler may never be called. 283 Kprobes manages such functions as a blacklist. 284 If you want to add a function into the blacklist, you just need 285 to (1) include linux/kprobes.h and (2) use NOKPROBE_SYMBOL() macro 286 to specify a blacklisted function. 287 Kprobes checks the given probe address against the blacklist and 288 rejects registering it, if the given address is in the blacklist. 289 290 2. Architectures Supported 291 292 Kprobes, jprobes, and return probes are implemented on the following 293 architectures: 294 295 - i386 (Supports jump optimization) 296 - x86_64 (AMD-64, EM64T) (Supports jump optimization) 297 - ppc64 298 - ia64 (Does not support probes on instruction slot1.) 299 - sparc64 (Return probes not yet implemented.) 300 - arm 301 - ppc 302 - mips 303 - s390 304 305 3. Configuring Kprobes 306 307 When configuring the kernel using make menuconfig/xconfig/oldconfig, 308 ensure that CONFIG_KPROBES is set to "y". Under "General setup", look 309 for "Kprobes". 310 311 So that you can load and unload Kprobes-based instrumentation modules, 312 make sure "Loadable module support" (CONFIG_MODULES) and "Module 313 unloading" (CONFIG_MODULE_UNLOAD) are set to "y". 314 315 Also make sure that CONFIG_KALLSYMS and perhaps even CONFIG_KALLSYMS_ALL 316 are set to "y", since kallsyms_lookup_name() is used by the in-kernel 317 kprobe address resolution code. 318 319 If you need to insert a probe in the middle of a function, you may find 320 it useful to "Compile the kernel with debug info" (CONFIG_DEBUG_INFO), 321 so you can use "objdump -d -l vmlinux" to see the source-to-object 322 code mapping. 323 324 4. API Reference 325 326 The Kprobes API includes a "register" function and an "unregister" 327 function for each type of probe. The API also includes "register_*probes" 328 and "unregister_*probes" functions for (un)registering arrays of probes. 329 Here are terse, mini-man-page specifications for these functions and 330 the associated probe handlers that you'll write. See the files in the 331 samples/kprobes/ sub-directory for examples. 332 333 4.1 register_kprobe 334 335 #include <linux/kprobes.h> 336 int register_kprobe(struct kprobe *kp); 337 338 Sets a breakpoint at the address kp->addr. When the breakpoint is 339 hit, Kprobes calls kp->pre_handler. After the probed instruction 340 is single-stepped, Kprobe calls kp->post_handler. If a fault 341 occurs during execution of kp->pre_handler or kp->post_handler, 342 or during single-stepping of the probed instruction, Kprobes calls 343 kp->fault_handler. Any or all handlers can be NULL. If kp->flags 344 is set KPROBE_FLAG_DISABLED, that kp will be registered but disabled, 345 so, its handlers aren't hit until calling enable_kprobe(kp). 346 347 NOTE: 348 1. With the introduction of the "symbol_name" field to struct kprobe, 349 the probepoint address resolution will now be taken care of by the kernel. 350 The following will now work: 351 352 kp.symbol_name = "symbol_name"; 353 354 (64-bit powerpc intricacies such as function descriptors are handled 355 transparently) 356 357 2. Use the "offset" field of struct kprobe if the offset into the symbol 358 to install a probepoint is known. This field is used to calculate the 359 probepoint. 360 361 3. Specify either the kprobe "symbol_name" OR the "addr". If both are 362 specified, kprobe registration will fail with -EINVAL. 363 364 4. With CISC architectures (such as i386 and x86_64), the kprobes code 365 does not validate if the kprobe.addr is at an instruction boundary. 366 Use "offset" with caution. 367 368 register_kprobe() returns 0 on success, or a negative errno otherwise. 369 370 User's pre-handler (kp->pre_handler): 371 #include <linux/kprobes.h> 372 #include <linux/ptrace.h> 373 int pre_handler(struct kprobe *p, struct pt_regs *regs); 374 375 Called with p pointing to the kprobe associated with the breakpoint, 376 and regs pointing to the struct containing the registers saved when 377 the breakpoint was hit. Return 0 here unless you're a Kprobes geek. 378 379 User's post-handler (kp->post_handler): 380 #include <linux/kprobes.h> 381 #include <linux/ptrace.h> 382 void post_handler(struct kprobe *p, struct pt_regs *regs, 383 unsigned long flags); 384 385 p and regs are as described for the pre_handler. flags always seems 386 to be zero. 387 388 User's fault-handler (kp->fault_handler): 389 #include <linux/kprobes.h> 390 #include <linux/ptrace.h> 391 int fault_handler(struct kprobe *p, struct pt_regs *regs, int trapnr); 392 393 p and regs are as described for the pre_handler. trapnr is the 394 architecture-specific trap number associated with the fault (e.g., 395 on i386, 13 for a general protection fault or 14 for a page fault). 396 Returns 1 if it successfully handled the exception. 397 398 4.2 register_jprobe 399 400 #include <linux/kprobes.h> 401 int register_jprobe(struct jprobe *jp) 402 403 Sets a breakpoint at the address jp->kp.addr, which must be the address 404 of the first instruction of a function. When the breakpoint is hit, 405 Kprobes runs the handler whose address is jp->entry. 406 407 The handler should have the same arg list and return type as the probed 408 function; and just before it returns, it must call jprobe_return(). 409 (The handler never actually returns, since jprobe_return() returns 410 control to Kprobes.) If the probed function is declared asmlinkage 411 or anything else that affects how args are passed, the handler's 412 declaration must match. 413 414 register_jprobe() returns 0 on success, or a negative errno otherwise. 415 416 4.3 register_kretprobe 417 418 #include <linux/kprobes.h> 419 int register_kretprobe(struct kretprobe *rp); 420 421 Establishes a return probe for the function whose address is 422 rp->kp.addr. When that function returns, Kprobes calls rp->handler. 423 You must set rp->maxactive appropriately before you call 424 register_kretprobe(); see "How Does a Return Probe Work?" for details. 425 426 register_kretprobe() returns 0 on success, or a negative errno 427 otherwise. 428 429 User's return-probe handler (rp->handler): 430 #include <linux/kprobes.h> 431 #include <linux/ptrace.h> 432 int kretprobe_handler(struct kretprobe_instance *ri, struct pt_regs *regs); 433 434 regs is as described for kprobe.pre_handler. ri points to the 435 kretprobe_instance object, of which the following fields may be 436 of interest: 437 - ret_addr: the return address 438 - rp: points to the corresponding kretprobe object 439 - task: points to the corresponding task struct 440 - data: points to per return-instance private data; see "Kretprobe 441 entry-handler" for details. 442 443 The regs_return_value(regs) macro provides a simple abstraction to 444 extract the return value from the appropriate register as defined by 445 the architecture's ABI. 446 447 The handler's return value is currently ignored. 448 449 4.4 unregister_*probe 450 451 #include <linux/kprobes.h> 452 void unregister_kprobe(struct kprobe *kp); 453 void unregister_jprobe(struct jprobe *jp); 454 void unregister_kretprobe(struct kretprobe *rp); 455 456 Removes the specified probe. The unregister function can be called 457 at any time after the probe has been registered. 458 459 NOTE: 460 If the functions find an incorrect probe (ex. an unregistered probe), 461 they clear the addr field of the probe. 462 463 4.5 register_*probes 464 465 #include <linux/kprobes.h> 466 int register_kprobes(struct kprobe **kps, int num); 467 int register_kretprobes(struct kretprobe **rps, int num); 468 int register_jprobes(struct jprobe **jps, int num); 469 470 Registers each of the num probes in the specified array. If any 471 error occurs during registration, all probes in the array, up to 472 the bad probe, are safely unregistered before the register_*probes 473 function returns. 474 - kps/rps/jps: an array of pointers to *probe data structures 475 - num: the number of the array entries. 476 477 NOTE: 478 You have to allocate(or define) an array of pointers and set all 479 of the array entries before using these functions. 480 481 4.6 unregister_*probes 482 483 #include <linux/kprobes.h> 484 void unregister_kprobes(struct kprobe **kps, int num); 485 void unregister_kretprobes(struct kretprobe **rps, int num); 486 void unregister_jprobes(struct jprobe **jps, int num); 487 488 Removes each of the num probes in the specified array at once. 489 490 NOTE: 491 If the functions find some incorrect probes (ex. unregistered 492 probes) in the specified array, they clear the addr field of those 493 incorrect probes. However, other probes in the array are 494 unregistered correctly. 495 496 4.7 disable_*probe 497 498 #include <linux/kprobes.h> 499 int disable_kprobe(struct kprobe *kp); 500 int disable_kretprobe(struct kretprobe *rp); 501 int disable_jprobe(struct jprobe *jp); 502 503 Temporarily disables the specified *probe. You can enable it again by using 504 enable_*probe(). You must specify the probe which has been registered. 505 506 4.8 enable_*probe 507 508 #include <linux/kprobes.h> 509 int enable_kprobe(struct kprobe *kp); 510 int enable_kretprobe(struct kretprobe *rp); 511 int enable_jprobe(struct jprobe *jp); 512 513 Enables *probe which has been disabled by disable_*probe(). You must specify 514 the probe which has been registered. 515 516 5. Kprobes Features and Limitations 517 518 Kprobes allows multiple probes at the same address. Currently, 519 however, there cannot be multiple jprobes on the same function at 520 the same time. Also, a probepoint for which there is a jprobe or 521 a post_handler cannot be optimized. So if you install a jprobe, 522 or a kprobe with a post_handler, at an optimized probepoint, the 523 probepoint will be unoptimized automatically. 524 525 In general, you can install a probe anywhere in the kernel. 526 In particular, you can probe interrupt handlers. Known exceptions 527 are discussed in this section. 528 529 The register_*probe functions will return -EINVAL if you attempt 530 to install a probe in the code that implements Kprobes (mostly 531 kernel/kprobes.c and arch/*/kernel/kprobes.c, but also functions such 532 as do_page_fault and notifier_call_chain). 533 534 If you install a probe in an inline-able function, Kprobes makes 535 no attempt to chase down all inline instances of the function and 536 install probes there. gcc may inline a function without being asked, 537 so keep this in mind if you're not seeing the probe hits you expect. 538 539 A probe handler can modify the environment of the probed function 540 -- e.g., by modifying kernel data structures, or by modifying the 541 contents of the pt_regs struct (which are restored to the registers 542 upon return from the breakpoint). So Kprobes can be used, for example, 543 to install a bug fix or to inject faults for testing. Kprobes, of 544 course, has no way to distinguish the deliberately injected faults 545 from the accidental ones. Don't drink and probe. 546 547 Kprobes makes no attempt to prevent probe handlers from stepping on 548 each other -- e.g., probing printk() and then calling printk() from a 549 probe handler. If a probe handler hits a probe, that second probe's 550 handlers won't be run in that instance, and the kprobe.nmissed member 551 of the second probe will be incremented. 552 553 As of Linux v2.6.15-rc1, multiple handlers (or multiple instances of 554 the same handler) may run concurrently on different CPUs. 555 556 Kprobes does not use mutexes or allocate memory except during 557 registration and unregistration. 558 559 Probe handlers are run with preemption disabled. Depending on the 560 architecture and optimization state, handlers may also run with 561 interrupts disabled (e.g., kretprobe handlers and optimized kprobe 562 handlers run without interrupt disabled on x86/x86-64). In any case, 563 your handler should not yield the CPU (e.g., by attempting to acquire 564 a semaphore). 565 566 Since a return probe is implemented by replacing the return 567 address with the trampoline's address, stack backtraces and calls 568 to __builtin_return_address() will typically yield the trampoline's 569 address instead of the real return address for kretprobed functions. 570 (As far as we can tell, __builtin_return_address() is used only 571 for instrumentation and error reporting.) 572 573 If the number of times a function is called does not match the number 574 of times it returns, registering a return probe on that function may 575 produce undesirable results. In such a case, a line: 576 kretprobe BUG!: Processing kretprobe d000000000041aa8 @ c00000000004f48c 577 gets printed. With this information, one will be able to correlate the 578 exact instance of the kretprobe that caused the problem. We have the 579 do_exit() case covered. do_execve() and do_fork() are not an issue. 580 We're unaware of other specific cases where this could be a problem. 581 582 If, upon entry to or exit from a function, the CPU is running on 583 a stack other than that of the current task, registering a return 584 probe on that function may produce undesirable results. For this 585 reason, Kprobes doesn't support return probes (or kprobes or jprobes) 586 on the x86_64 version of __switch_to(); the registration functions 587 return -EINVAL. 588 589 On x86/x86-64, since the Jump Optimization of Kprobes modifies 590 instructions widely, there are some limitations to optimization. To 591 explain it, we introduce some terminology. Imagine a 3-instruction 592 sequence consisting of a two 2-byte instructions and one 3-byte 593 instruction. 594 595 IA 596 | 597 [-2][-1] 598 [ins1][ins2][ ins3 ] 599 [<- DCR ->] 600 [<- JTPR ->] 601 602 ins1: 1st Instruction 603 ins2: 2nd Instruction 604 ins3: 3rd Instruction 605 IA: Insertion Address 606 JTPR: Jump Target Prohibition Region 607 DCR: Detoured Code Region 608 609 The instructions in DCR are copied to the out-of-line buffer 610 of the kprobe, because the bytes in DCR are replaced by 611 a 5-byte jump instruction. So there are several limitations. 612 613 a) The instructions in DCR must be relocatable. 614 b) The instructions in DCR must not include a call instruction. 615 c) JTPR must not be targeted by any jump or call instruction. 616 d) DCR must not straddle the border between functions. 617 618 Anyway, these limitations are checked by the in-kernel instruction 619 decoder, so you don't need to worry about that. 620 621 6. Probe Overhead 622 623 On a typical CPU in use in 2005, a kprobe hit takes 0.5 to 1.0 624 microseconds to process. Specifically, a benchmark that hits the same 625 probepoint repeatedly, firing a simple handler each time, reports 1-2 626 million hits per second, depending on the architecture. A jprobe or 627 return-probe hit typically takes 50-75% longer than a kprobe hit. 628 When you have a return probe set on a function, adding a kprobe at 629 the entry to that function adds essentially no overhead. 630 631 Here are sample overhead figures (in usec) for different architectures. 632 k = kprobe; j = jprobe; r = return probe; kr = kprobe + return probe 633 on same function; jr = jprobe + return probe on same function 634 635 i386: Intel Pentium M, 1495 MHz, 2957.31 bogomips 636 k = 0.57 usec; j = 1.00; r = 0.92; kr = 0.99; jr = 1.40 637 638 x86_64: AMD Opteron 246, 1994 MHz, 3971.48 bogomips 639 k = 0.49 usec; j = 0.76; r = 0.80; kr = 0.82; jr = 1.07 640 641 ppc64: POWER5 (gr), 1656 MHz (SMT disabled, 1 virtual CPU per physical CPU) 642 k = 0.77 usec; j = 1.31; r = 1.26; kr = 1.45; jr = 1.99 643 644 6.1 Optimized Probe Overhead 645 646 Typically, an optimized kprobe hit takes 0.07 to 0.1 microseconds to 647 process. Here are sample overhead figures (in usec) for x86 architectures. 648 k = unoptimized kprobe, b = boosted (single-step skipped), o = optimized kprobe, 649 r = unoptimized kretprobe, rb = boosted kretprobe, ro = optimized kretprobe. 650 651 i386: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips 652 k = 0.80 usec; b = 0.33; o = 0.05; r = 1.10; rb = 0.61; ro = 0.33 653 654 x86-64: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips 655 k = 0.99 usec; b = 0.43; o = 0.06; r = 1.24; rb = 0.68; ro = 0.30 656 657 7. TODO 658 659 a. SystemTap (http://sourceware.org/systemtap): Provides a simplified 660 programming interface for probe-based instrumentation. Try it out. 661 b. Kernel return probes for sparc64. 662 c. Support for other architectures. 663 d. User-space probes. 664 e. Watchpoint probes (which fire on data references). 665 666 8. Kprobes Example 667 668 See samples/kprobes/kprobe_example.c 669 670 9. Jprobes Example 671 672 See samples/kprobes/jprobe_example.c 673 674 10. Kretprobes Example 675 676 See samples/kprobes/kretprobe_example.c 677 678 For additional information on Kprobes, refer to the following URLs: 679 http://www-106.ibm.com/developerworks/library/l-kprobes.html?ca=dgr-lnxw42Kprobe 680 http://www.redhat.com/magazine/005mar05/features/kprobes/ 681 http://www-users.cs.umn.edu/~boutcher/kprobes/ 682 http://www.linuxsymposium.org/2006/linuxsymposium_procv2.pdf (pages 101-115) 683 684 685 Appendix A: The kprobes debugfs interface 686 687 With recent kernels (> 2.6.20) the list of registered kprobes is visible 688 under the /sys/kernel/debug/kprobes/ directory (assuming debugfs is mounted at //sys/kernel/debug). 689 690 /sys/kernel/debug/kprobes/list: Lists all registered probes on the system 691 692 c015d71a k vfs_read+0x0 693 c011a316 j do_fork+0x0 694 c03dedc5 r tcp_v4_rcv+0x0 695 696 The first column provides the kernel address where the probe is inserted. 697 The second column identifies the type of probe (k - kprobe, r - kretprobe 698 and j - jprobe), while the third column specifies the symbol+offset of 699 the probe. If the probed function belongs to a module, the module name 700 is also specified. Following columns show probe status. If the probe is on 701 a virtual address that is no longer valid (module init sections, module 702 virtual addresses that correspond to modules that've been unloaded), 703 such probes are marked with [GONE]. If the probe is temporarily disabled, 704 such probes are marked with [DISABLED]. If the probe is optimized, it is 705 marked with [OPTIMIZED]. If the probe is ftrace-based, it is marked with 706 [FTRACE]. 707 708 /sys/kernel/debug/kprobes/enabled: Turn kprobes ON/OFF forcibly. 709 710 Provides a knob to globally and forcibly turn registered kprobes ON or OFF. 711 By default, all kprobes are enabled. By echoing "0" to this file, all 712 registered probes will be disarmed, till such time a "1" is echoed to this 713 file. Note that this knob just disarms and arms all kprobes and doesn't 714 change each probe's disabling state. This means that disabled kprobes (marked 715 [DISABLED]) will be not enabled if you turn ON all kprobes by this knob. 716 717 718 Appendix B: The kprobes sysctl interface 719 720 /proc/sys/debug/kprobes-optimization: Turn kprobes optimization ON/OFF. 721 722 When CONFIG_OPTPROBES=y, this sysctl interface appears and it provides 723 a knob to globally and forcibly turn jump optimization (see section 724 1.4) ON or OFF. By default, jump optimization is allowed (ON). 725 If you echo "0" to this file or set "debug.kprobes_optimization" to 726 0 via sysctl, all optimized probes will be unoptimized, and any new 727 probes registered after that will not be optimized. Note that this 728 knob *changes* the optimized state. This means that optimized probes 729 (marked [OPTIMIZED]) will be unoptimized ([OPTIMIZED] tag will be 730 removed). If the knob is turned on, they will be optimized again.