Root/
Source at commit b05a5adf03613de371c77c3235f7d970d7cd0c71 created 13 years 1 month ago. By Lars-Peter Clausen, NAND: Optimize reading the eec data for the JZ4740 (evil hack) | |
---|---|
1 | Title : Kernel Probes (Kprobes) |
2 | Authors : Jim Keniston <jkenisto@us.ibm.com> |
3 | : Prasanna S Panchamukhi <prasanna.panchamukhi@gmail.com> |
4 | : Masami Hiramatsu <mhiramat@redhat.com> |
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 | |
28 | There are currently three types of probes: kprobes, jprobes, and |
29 | kretprobes (also called return probes). A kprobe can be inserted |
30 | on virtually any instruction in the kernel. A jprobe is inserted at |
31 | the entry to a kernel function, and provides convenient access to the |
32 | function's arguments. A return probe fires when a specified function |
33 | returns. |
34 | |
35 | In the typical case, Kprobes-based instrumentation is packaged as |
36 | a kernel module. The module's init function installs ("registers") |
37 | one or more probes, and the exit function unregisters them. A |
38 | registration function such as register_kprobe() specifies where |
39 | the probe is to be inserted and what handler is to be called when |
40 | the probe is hit. |
41 | |
42 | There are also register_/unregister_*probes() functions for batch |
43 | registration/unregistration of a group of *probes. These functions |
44 | can speed up unregistration process when you have to unregister |
45 | a lot of probes at once. |
46 | |
47 | The next four subsections explain how the different types of |
48 | probes work and how jump optimization works. They explain certain |
49 | things that you'll need to know in order to make the best use of |
50 | Kprobes -- e.g., the difference between a pre_handler and |
51 | a post_handler, and how to use the maxactive and nmissed fields of |
52 | a kretprobe. But if you're in a hurry to start using Kprobes, you |
53 | can skip ahead to section 2. |
54 | |
55 | 1.1 How Does a Kprobe Work? |
56 | |
57 | When a kprobe is registered, Kprobes makes a copy of the probed |
58 | instruction and replaces the first byte(s) of the probed instruction |
59 | with a breakpoint instruction (e.g., int3 on i386 and x86_64). |
60 | |
61 | When a CPU hits the breakpoint instruction, a trap occurs, the CPU's |
62 | registers are saved, and control passes to Kprobes via the |
63 | notifier_call_chain mechanism. Kprobes executes the "pre_handler" |
64 | associated with the kprobe, passing the handler the addresses of the |
65 | kprobe struct and the saved registers. |
66 | |
67 | Next, Kprobes single-steps its copy of the probed instruction. |
68 | (It would be simpler to single-step the actual instruction in place, |
69 | but then Kprobes would have to temporarily remove the breakpoint |
70 | instruction. This would open a small time window when another CPU |
71 | could sail right past the probepoint.) |
72 | |
73 | After the instruction is single-stepped, Kprobes executes the |
74 | "post_handler," if any, that is associated with the kprobe. |
75 | Execution then continues with the instruction following the probepoint. |
76 | |
77 | 1.2 How Does a Jprobe Work? |
78 | |
79 | A jprobe is implemented using a kprobe that is placed on a function's |
80 | entry point. It employs a simple mirroring principle to allow |
81 | seamless access to the probed function's arguments. The jprobe |
82 | handler routine should have the same signature (arg list and return |
83 | type) as the function being probed, and must always end by calling |
84 | the Kprobes function jprobe_return(). |
85 | |
86 | Here's how it works. When the probe is hit, Kprobes makes a copy of |
87 | the saved registers and a generous portion of the stack (see below). |
88 | Kprobes then points the saved instruction pointer at the jprobe's |
89 | handler routine, and returns from the trap. As a result, control |
90 | passes to the handler, which is presented with the same register and |
91 | stack contents as the probed function. When it is done, the handler |
92 | calls jprobe_return(), which traps again to restore the original stack |
93 | contents and processor state and switch to the probed function. |
94 | |
95 | By convention, the callee owns its arguments, so gcc may produce code |
96 | that unexpectedly modifies that portion of the stack. This is why |
97 | Kprobes saves a copy of the stack and restores it after the jprobe |
98 | handler has run. Up to MAX_STACK_SIZE bytes are copied -- e.g., |
99 | 64 bytes on i386. |
100 | |
101 | Note that the probed function's args may be passed on the stack |
102 | or in registers. The jprobe will work in either case, so long as the |
103 | handler's prototype matches that of the probed function. |
104 | |
105 | 1.3 Return Probes |
106 | |
107 | 1.3.1 How Does a Return Probe Work? |
108 | |
109 | When you call register_kretprobe(), Kprobes establishes a kprobe at |
110 | the entry to the function. When the probed function is called and this |
111 | probe is hit, Kprobes saves a copy of the return address, and replaces |
112 | the return address with the address of a "trampoline." The trampoline |
113 | is an arbitrary piece of code -- typically just a nop instruction. |
114 | At boot time, Kprobes registers a kprobe at the trampoline. |
115 | |
116 | When the probed function executes its return instruction, control |
117 | passes to the trampoline and that probe is hit. Kprobes' trampoline |
118 | handler calls the user-specified return handler associated with the |
119 | kretprobe, then sets the saved instruction pointer to the saved return |
120 | address, and that's where execution resumes upon return from the trap. |
121 | |
122 | While the probed function is executing, its return address is |
123 | stored in an object of type kretprobe_instance. Before calling |
124 | register_kretprobe(), the user sets the maxactive field of the |
125 | kretprobe struct to specify how many instances of the specified |
126 | function can be probed simultaneously. register_kretprobe() |
127 | pre-allocates the indicated number of kretprobe_instance objects. |
128 | |
129 | For example, if the function is non-recursive and is called with a |
130 | spinlock held, maxactive = 1 should be enough. If the function is |
131 | non-recursive and can never relinquish the CPU (e.g., via a semaphore |
132 | or preemption), NR_CPUS should be enough. If maxactive <= 0, it is |
133 | set to a default value. If CONFIG_PREEMPT is enabled, the default |
134 | is max(10, 2*NR_CPUS). Otherwise, the default is NR_CPUS. |
135 | |
136 | It's not a disaster if you set maxactive too low; you'll just miss |
137 | some probes. In the kretprobe struct, the nmissed field is set to |
138 | zero when the return probe is registered, and is incremented every |
139 | time the probed function is entered but there is no kretprobe_instance |
140 | object available for establishing the return probe. |
141 | |
142 | 1.3.2 Kretprobe entry-handler |
143 | |
144 | Kretprobes also provides an optional user-specified handler which runs |
145 | on function entry. This handler is specified by setting the entry_handler |
146 | field of the kretprobe struct. Whenever the kprobe placed by kretprobe at the |
147 | function entry is hit, the user-defined entry_handler, if any, is invoked. |
148 | If the entry_handler returns 0 (success) then a corresponding return handler |
149 | is guaranteed to be called upon function return. If the entry_handler |
150 | returns a non-zero error then Kprobes leaves the return address as is, and |
151 | the kretprobe has no further effect for that particular function instance. |
152 | |
153 | Multiple entry and return handler invocations are matched using the unique |
154 | kretprobe_instance object associated with them. Additionally, a user |
155 | may also specify per return-instance private data to be part of each |
156 | kretprobe_instance object. This is especially useful when sharing private |
157 | data between corresponding user entry and return handlers. The size of each |
158 | private data object can be specified at kretprobe registration time by |
159 | setting the data_size field of the kretprobe struct. This data can be |
160 | accessed through the data field of each kretprobe_instance object. |
161 | |
162 | In case probed function is entered but there is no kretprobe_instance |
163 | object available, then in addition to incrementing the nmissed count, |
164 | the user entry_handler invocation is also skipped. |
165 | |
166 | 1.4 How Does Jump Optimization Work? |
167 | |
168 | If your kernel is built with CONFIG_OPTPROBES=y (currently this flag |
169 | is automatically set 'y' on x86/x86-64, non-preemptive kernel) and |
170 | the "debug.kprobes_optimization" kernel parameter is set to 1 (see |
171 | sysctl(8)), Kprobes tries to reduce probe-hit overhead by using a jump |
172 | instruction instead of a breakpoint instruction at each probepoint. |
173 | |
174 | 1.4.1 Init a Kprobe |
175 | |
176 | When a probe is registered, before attempting this optimization, |
177 | Kprobes inserts an ordinary, breakpoint-based kprobe at the specified |
178 | address. So, even if it's not possible to optimize this particular |
179 | probepoint, there'll be a probe there. |
180 | |
181 | 1.4.2 Safety Check |
182 | |
183 | Before optimizing a probe, Kprobes performs the following safety checks: |
184 | |
185 | - Kprobes verifies that the region that will be replaced by the jump |
186 | instruction (the "optimized region") lies entirely within one function. |
187 | (A jump instruction is multiple bytes, and so may overlay multiple |
188 | instructions.) |
189 | |
190 | - Kprobes analyzes the entire function and verifies that there is no |
191 | jump into the optimized region. Specifically: |
192 | - the function contains no indirect jump; |
193 | - the function contains no instruction that causes an exception (since |
194 | the fixup code triggered by the exception could jump back into the |
195 | optimized region -- Kprobes checks the exception tables to verify this); |
196 | and |
197 | - there is no near jump to the optimized region (other than to the first |
198 | byte). |
199 | |
200 | - For each instruction in the optimized region, Kprobes verifies that |
201 | the instruction can be executed out of line. |
202 | |
203 | 1.4.3 Preparing Detour Buffer |
204 | |
205 | Next, Kprobes prepares a "detour" buffer, which contains the following |
206 | instruction sequence: |
207 | - code to push the CPU's registers (emulating a breakpoint trap) |
208 | - a call to the trampoline code which calls user's probe handlers. |
209 | - code to restore registers |
210 | - the instructions from the optimized region |
211 | - a jump back to the original execution path. |
212 | |
213 | 1.4.4 Pre-optimization |
214 | |
215 | After preparing the detour buffer, Kprobes verifies that none of the |
216 | following situations exist: |
217 | - The probe has either a break_handler (i.e., it's a jprobe) or a |
218 | post_handler. |
219 | - Other instructions in the optimized region are probed. |
220 | - The probe is disabled. |
221 | In any of the above cases, Kprobes won't start optimizing the probe. |
222 | Since these are temporary situations, Kprobes tries to start |
223 | optimizing it again if the situation is changed. |
224 | |
225 | If the kprobe can be optimized, Kprobes enqueues the kprobe to an |
226 | optimizing list, and kicks the kprobe-optimizer workqueue to optimize |
227 | it. If the to-be-optimized probepoint is hit before being optimized, |
228 | Kprobes returns control to the original instruction path by setting |
229 | the CPU's instruction pointer to the copied code in the detour buffer |
230 | -- thus at least avoiding the single-step. |
231 | |
232 | 1.4.5 Optimization |
233 | |
234 | The Kprobe-optimizer doesn't insert the jump instruction immediately; |
235 | rather, it calls synchronize_sched() for safety first, because it's |
236 | possible for a CPU to be interrupted in the middle of executing the |
237 | optimized region(*). As you know, synchronize_sched() can ensure |
238 | that all interruptions that were active when synchronize_sched() |
239 | was called are done, but only if CONFIG_PREEMPT=n. So, this version |
240 | of kprobe optimization supports only kernels with CONFIG_PREEMPT=n.(**) |
241 | |
242 | After that, the Kprobe-optimizer calls stop_machine() to replace |
243 | the optimized region with a jump instruction to the detour buffer, |
244 | using text_poke_smp(). |
245 | |
246 | 1.4.6 Unoptimization |
247 | |
248 | When an optimized kprobe is unregistered, disabled, or blocked by |
249 | another kprobe, it will be unoptimized. If this happens before |
250 | the optimization is complete, the kprobe is just dequeued from the |
251 | optimized list. If the optimization has been done, the jump is |
252 | replaced with the original code (except for an int3 breakpoint in |
253 | the first byte) by using text_poke_smp(). |
254 | |
255 | (*)Please imagine that the 2nd instruction is interrupted and then |
256 | the optimizer replaces the 2nd instruction with the jump *address* |
257 | while the interrupt handler is running. When the interrupt |
258 | returns to original address, there is no valid instruction, |
259 | and it causes an unexpected result. |
260 | |
261 | (**)This optimization-safety checking may be replaced with the |
262 | stop-machine method that ksplice uses for supporting a CONFIG_PREEMPT=y |
263 | kernel. |
264 | |
265 | NOTE for geeks: |
266 | The jump optimization changes the kprobe's pre_handler behavior. |
267 | Without optimization, the pre_handler can change the kernel's execution |
268 | path by changing regs->ip and returning 1. However, when the probe |
269 | is optimized, that modification is ignored. Thus, if you want to |
270 | tweak the kernel's execution path, you need to suppress optimization, |
271 | using one of the following techniques: |
272 | - Specify an empty function for the kprobe's post_handler or break_handler. |
273 | or |
274 | - Execute 'sysctl -w debug.kprobes_optimization=n' |
275 | |
276 | 2. Architectures Supported |
277 | |
278 | Kprobes, jprobes, and return probes are implemented on the following |
279 | architectures: |
280 | |
281 | - i386 (Supports jump optimization) |
282 | - x86_64 (AMD-64, EM64T) (Supports jump optimization) |
283 | - ppc64 |
284 | - ia64 (Does not support probes on instruction slot1.) |
285 | - sparc64 (Return probes not yet implemented.) |
286 | - arm |
287 | - ppc |
288 | - mips |
289 | |
290 | 3. Configuring Kprobes |
291 | |
292 | When configuring the kernel using make menuconfig/xconfig/oldconfig, |
293 | ensure that CONFIG_KPROBES is set to "y". Under "Instrumentation |
294 | Support", look for "Kprobes". |
295 | |
296 | So that you can load and unload Kprobes-based instrumentation modules, |
297 | make sure "Loadable module support" (CONFIG_MODULES) and "Module |
298 | unloading" (CONFIG_MODULE_UNLOAD) are set to "y". |
299 | |
300 | Also make sure that CONFIG_KALLSYMS and perhaps even CONFIG_KALLSYMS_ALL |
301 | are set to "y", since kallsyms_lookup_name() is used by the in-kernel |
302 | kprobe address resolution code. |
303 | |
304 | If you need to insert a probe in the middle of a function, you may find |
305 | it useful to "Compile the kernel with debug info" (CONFIG_DEBUG_INFO), |
306 | so you can use "objdump -d -l vmlinux" to see the source-to-object |
307 | code mapping. |
308 | |
309 | 4. API Reference |
310 | |
311 | The Kprobes API includes a "register" function and an "unregister" |
312 | function for each type of probe. The API also includes "register_*probes" |
313 | and "unregister_*probes" functions for (un)registering arrays of probes. |
314 | Here are terse, mini-man-page specifications for these functions and |
315 | the associated probe handlers that you'll write. See the files in the |
316 | samples/kprobes/ sub-directory for examples. |
317 | |
318 | 4.1 register_kprobe |
319 | |
320 | #include <linux/kprobes.h> |
321 | int register_kprobe(struct kprobe *kp); |
322 | |
323 | Sets a breakpoint at the address kp->addr. When the breakpoint is |
324 | hit, Kprobes calls kp->pre_handler. After the probed instruction |
325 | is single-stepped, Kprobe calls kp->post_handler. If a fault |
326 | occurs during execution of kp->pre_handler or kp->post_handler, |
327 | or during single-stepping of the probed instruction, Kprobes calls |
328 | kp->fault_handler. Any or all handlers can be NULL. If kp->flags |
329 | is set KPROBE_FLAG_DISABLED, that kp will be registered but disabled, |
330 | so, its handlers aren't hit until calling enable_kprobe(kp). |
331 | |
332 | NOTE: |
333 | 1. With the introduction of the "symbol_name" field to struct kprobe, |
334 | the probepoint address resolution will now be taken care of by the kernel. |
335 | The following will now work: |
336 | |
337 | kp.symbol_name = "symbol_name"; |
338 | |
339 | (64-bit powerpc intricacies such as function descriptors are handled |
340 | transparently) |
341 | |
342 | 2. Use the "offset" field of struct kprobe if the offset into the symbol |
343 | to install a probepoint is known. This field is used to calculate the |
344 | probepoint. |
345 | |
346 | 3. Specify either the kprobe "symbol_name" OR the "addr". If both are |
347 | specified, kprobe registration will fail with -EINVAL. |
348 | |
349 | 4. With CISC architectures (such as i386 and x86_64), the kprobes code |
350 | does not validate if the kprobe.addr is at an instruction boundary. |
351 | Use "offset" with caution. |
352 | |
353 | register_kprobe() returns 0 on success, or a negative errno otherwise. |
354 | |
355 | User's pre-handler (kp->pre_handler): |
356 | #include <linux/kprobes.h> |
357 | #include <linux/ptrace.h> |
358 | int pre_handler(struct kprobe *p, struct pt_regs *regs); |
359 | |
360 | Called with p pointing to the kprobe associated with the breakpoint, |
361 | and regs pointing to the struct containing the registers saved when |
362 | the breakpoint was hit. Return 0 here unless you're a Kprobes geek. |
363 | |
364 | User's post-handler (kp->post_handler): |
365 | #include <linux/kprobes.h> |
366 | #include <linux/ptrace.h> |
367 | void post_handler(struct kprobe *p, struct pt_regs *regs, |
368 | unsigned long flags); |
369 | |
370 | p and regs are as described for the pre_handler. flags always seems |
371 | to be zero. |
372 | |
373 | User's fault-handler (kp->fault_handler): |
374 | #include <linux/kprobes.h> |
375 | #include <linux/ptrace.h> |
376 | int fault_handler(struct kprobe *p, struct pt_regs *regs, int trapnr); |
377 | |
378 | p and regs are as described for the pre_handler. trapnr is the |
379 | architecture-specific trap number associated with the fault (e.g., |
380 | on i386, 13 for a general protection fault or 14 for a page fault). |
381 | Returns 1 if it successfully handled the exception. |
382 | |
383 | 4.2 register_jprobe |
384 | |
385 | #include <linux/kprobes.h> |
386 | int register_jprobe(struct jprobe *jp) |
387 | |
388 | Sets a breakpoint at the address jp->kp.addr, which must be the address |
389 | of the first instruction of a function. When the breakpoint is hit, |
390 | Kprobes runs the handler whose address is jp->entry. |
391 | |
392 | The handler should have the same arg list and return type as the probed |
393 | function; and just before it returns, it must call jprobe_return(). |
394 | (The handler never actually returns, since jprobe_return() returns |
395 | control to Kprobes.) If the probed function is declared asmlinkage |
396 | or anything else that affects how args are passed, the handler's |
397 | declaration must match. |
398 | |
399 | register_jprobe() returns 0 on success, or a negative errno otherwise. |
400 | |
401 | 4.3 register_kretprobe |
402 | |
403 | #include <linux/kprobes.h> |
404 | int register_kretprobe(struct kretprobe *rp); |
405 | |
406 | Establishes a return probe for the function whose address is |
407 | rp->kp.addr. When that function returns, Kprobes calls rp->handler. |
408 | You must set rp->maxactive appropriately before you call |
409 | register_kretprobe(); see "How Does a Return Probe Work?" for details. |
410 | |
411 | register_kretprobe() returns 0 on success, or a negative errno |
412 | otherwise. |
413 | |
414 | User's return-probe handler (rp->handler): |
415 | #include <linux/kprobes.h> |
416 | #include <linux/ptrace.h> |
417 | int kretprobe_handler(struct kretprobe_instance *ri, struct pt_regs *regs); |
418 | |
419 | regs is as described for kprobe.pre_handler. ri points to the |
420 | kretprobe_instance object, of which the following fields may be |
421 | of interest: |
422 | - ret_addr: the return address |
423 | - rp: points to the corresponding kretprobe object |
424 | - task: points to the corresponding task struct |
425 | - data: points to per return-instance private data; see "Kretprobe |
426 | entry-handler" for details. |
427 | |
428 | The regs_return_value(regs) macro provides a simple abstraction to |
429 | extract the return value from the appropriate register as defined by |
430 | the architecture's ABI. |
431 | |
432 | The handler's return value is currently ignored. |
433 | |
434 | 4.4 unregister_*probe |
435 | |
436 | #include <linux/kprobes.h> |
437 | void unregister_kprobe(struct kprobe *kp); |
438 | void unregister_jprobe(struct jprobe *jp); |
439 | void unregister_kretprobe(struct kretprobe *rp); |
440 | |
441 | Removes the specified probe. The unregister function can be called |
442 | at any time after the probe has been registered. |
443 | |
444 | NOTE: |
445 | If the functions find an incorrect probe (ex. an unregistered probe), |
446 | they clear the addr field of the probe. |
447 | |
448 | 4.5 register_*probes |
449 | |
450 | #include <linux/kprobes.h> |
451 | int register_kprobes(struct kprobe **kps, int num); |
452 | int register_kretprobes(struct kretprobe **rps, int num); |
453 | int register_jprobes(struct jprobe **jps, int num); |
454 | |
455 | Registers each of the num probes in the specified array. If any |
456 | error occurs during registration, all probes in the array, up to |
457 | the bad probe, are safely unregistered before the register_*probes |
458 | function returns. |
459 | - kps/rps/jps: an array of pointers to *probe data structures |
460 | - num: the number of the array entries. |
461 | |
462 | NOTE: |
463 | You have to allocate(or define) an array of pointers and set all |
464 | of the array entries before using these functions. |
465 | |
466 | 4.6 unregister_*probes |
467 | |
468 | #include <linux/kprobes.h> |
469 | void unregister_kprobes(struct kprobe **kps, int num); |
470 | void unregister_kretprobes(struct kretprobe **rps, int num); |
471 | void unregister_jprobes(struct jprobe **jps, int num); |
472 | |
473 | Removes each of the num probes in the specified array at once. |
474 | |
475 | NOTE: |
476 | If the functions find some incorrect probes (ex. unregistered |
477 | probes) in the specified array, they clear the addr field of those |
478 | incorrect probes. However, other probes in the array are |
479 | unregistered correctly. |
480 | |
481 | 4.7 disable_*probe |
482 | |
483 | #include <linux/kprobes.h> |
484 | int disable_kprobe(struct kprobe *kp); |
485 | int disable_kretprobe(struct kretprobe *rp); |
486 | int disable_jprobe(struct jprobe *jp); |
487 | |
488 | Temporarily disables the specified *probe. You can enable it again by using |
489 | enable_*probe(). You must specify the probe which has been registered. |
490 | |
491 | 4.8 enable_*probe |
492 | |
493 | #include <linux/kprobes.h> |
494 | int enable_kprobe(struct kprobe *kp); |
495 | int enable_kretprobe(struct kretprobe *rp); |
496 | int enable_jprobe(struct jprobe *jp); |
497 | |
498 | Enables *probe which has been disabled by disable_*probe(). You must specify |
499 | the probe which has been registered. |
500 | |
501 | 5. Kprobes Features and Limitations |
502 | |
503 | Kprobes allows multiple probes at the same address. Currently, |
504 | however, there cannot be multiple jprobes on the same function at |
505 | the same time. Also, a probepoint for which there is a jprobe or |
506 | a post_handler cannot be optimized. So if you install a jprobe, |
507 | or a kprobe with a post_handler, at an optimized probepoint, the |
508 | probepoint will be unoptimized automatically. |
509 | |
510 | In general, you can install a probe anywhere in the kernel. |
511 | In particular, you can probe interrupt handlers. Known exceptions |
512 | are discussed in this section. |
513 | |
514 | The register_*probe functions will return -EINVAL if you attempt |
515 | to install a probe in the code that implements Kprobes (mostly |
516 | kernel/kprobes.c and arch/*/kernel/kprobes.c, but also functions such |
517 | as do_page_fault and notifier_call_chain). |
518 | |
519 | If you install a probe in an inline-able function, Kprobes makes |
520 | no attempt to chase down all inline instances of the function and |
521 | install probes there. gcc may inline a function without being asked, |
522 | so keep this in mind if you're not seeing the probe hits you expect. |
523 | |
524 | A probe handler can modify the environment of the probed function |
525 | -- e.g., by modifying kernel data structures, or by modifying the |
526 | contents of the pt_regs struct (which are restored to the registers |
527 | upon return from the breakpoint). So Kprobes can be used, for example, |
528 | to install a bug fix or to inject faults for testing. Kprobes, of |
529 | course, has no way to distinguish the deliberately injected faults |
530 | from the accidental ones. Don't drink and probe. |
531 | |
532 | Kprobes makes no attempt to prevent probe handlers from stepping on |
533 | each other -- e.g., probing printk() and then calling printk() from a |
534 | probe handler. If a probe handler hits a probe, that second probe's |
535 | handlers won't be run in that instance, and the kprobe.nmissed member |
536 | of the second probe will be incremented. |
537 | |
538 | As of Linux v2.6.15-rc1, multiple handlers (or multiple instances of |
539 | the same handler) may run concurrently on different CPUs. |
540 | |
541 | Kprobes does not use mutexes or allocate memory except during |
542 | registration and unregistration. |
543 | |
544 | Probe handlers are run with preemption disabled. Depending on the |
545 | architecture and optimization state, handlers may also run with |
546 | interrupts disabled (e.g., kretprobe handlers and optimized kprobe |
547 | handlers run without interrupt disabled on x86/x86-64). In any case, |
548 | your handler should not yield the CPU (e.g., by attempting to acquire |
549 | a semaphore). |
550 | |
551 | Since a return probe is implemented by replacing the return |
552 | address with the trampoline's address, stack backtraces and calls |
553 | to __builtin_return_address() will typically yield the trampoline's |
554 | address instead of the real return address for kretprobed functions. |
555 | (As far as we can tell, __builtin_return_address() is used only |
556 | for instrumentation and error reporting.) |
557 | |
558 | If the number of times a function is called does not match the number |
559 | of times it returns, registering a return probe on that function may |
560 | produce undesirable results. In such a case, a line: |
561 | kretprobe BUG!: Processing kretprobe d000000000041aa8 @ c00000000004f48c |
562 | gets printed. With this information, one will be able to correlate the |
563 | exact instance of the kretprobe that caused the problem. We have the |
564 | do_exit() case covered. do_execve() and do_fork() are not an issue. |
565 | We're unaware of other specific cases where this could be a problem. |
566 | |
567 | If, upon entry to or exit from a function, the CPU is running on |
568 | a stack other than that of the current task, registering a return |
569 | probe on that function may produce undesirable results. For this |
570 | reason, Kprobes doesn't support return probes (or kprobes or jprobes) |
571 | on the x86_64 version of __switch_to(); the registration functions |
572 | return -EINVAL. |
573 | |
574 | On x86/x86-64, since the Jump Optimization of Kprobes modifies |
575 | instructions widely, there are some limitations to optimization. To |
576 | explain it, we introduce some terminology. Imagine a 3-instruction |
577 | sequence consisting of a two 2-byte instructions and one 3-byte |
578 | instruction. |
579 | |
580 | IA |
581 | | |
582 | [-2][-1][0][1][2][3][4][5][6][7] |
583 | [ins1][ins2][ ins3 ] |
584 | [<- DCR ->] |
585 | [<- JTPR ->] |
586 | |
587 | ins1: 1st Instruction |
588 | ins2: 2nd Instruction |
589 | ins3: 3rd Instruction |
590 | IA: Insertion Address |
591 | JTPR: Jump Target Prohibition Region |
592 | DCR: Detoured Code Region |
593 | |
594 | The instructions in DCR are copied to the out-of-line buffer |
595 | of the kprobe, because the bytes in DCR are replaced by |
596 | a 5-byte jump instruction. So there are several limitations. |
597 | |
598 | a) The instructions in DCR must be relocatable. |
599 | b) The instructions in DCR must not include a call instruction. |
600 | c) JTPR must not be targeted by any jump or call instruction. |
601 | d) DCR must not straddle the border between functions. |
602 | |
603 | Anyway, these limitations are checked by the in-kernel instruction |
604 | decoder, so you don't need to worry about that. |
605 | |
606 | 6. Probe Overhead |
607 | |
608 | On a typical CPU in use in 2005, a kprobe hit takes 0.5 to 1.0 |
609 | microseconds to process. Specifically, a benchmark that hits the same |
610 | probepoint repeatedly, firing a simple handler each time, reports 1-2 |
611 | million hits per second, depending on the architecture. A jprobe or |
612 | return-probe hit typically takes 50-75% longer than a kprobe hit. |
613 | When you have a return probe set on a function, adding a kprobe at |
614 | the entry to that function adds essentially no overhead. |
615 | |
616 | Here are sample overhead figures (in usec) for different architectures. |
617 | k = kprobe; j = jprobe; r = return probe; kr = kprobe + return probe |
618 | on same function; jr = jprobe + return probe on same function |
619 | |
620 | i386: Intel Pentium M, 1495 MHz, 2957.31 bogomips |
621 | k = 0.57 usec; j = 1.00; r = 0.92; kr = 0.99; jr = 1.40 |
622 | |
623 | x86_64: AMD Opteron 246, 1994 MHz, 3971.48 bogomips |
624 | k = 0.49 usec; j = 0.76; r = 0.80; kr = 0.82; jr = 1.07 |
625 | |
626 | ppc64: POWER5 (gr), 1656 MHz (SMT disabled, 1 virtual CPU per physical CPU) |
627 | k = 0.77 usec; j = 1.31; r = 1.26; kr = 1.45; jr = 1.99 |
628 | |
629 | 6.1 Optimized Probe Overhead |
630 | |
631 | Typically, an optimized kprobe hit takes 0.07 to 0.1 microseconds to |
632 | process. Here are sample overhead figures (in usec) for x86 architectures. |
633 | k = unoptimized kprobe, b = boosted (single-step skipped), o = optimized kprobe, |
634 | r = unoptimized kretprobe, rb = boosted kretprobe, ro = optimized kretprobe. |
635 | |
636 | i386: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips |
637 | k = 0.80 usec; b = 0.33; o = 0.05; r = 1.10; rb = 0.61; ro = 0.33 |
638 | |
639 | x86-64: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips |
640 | k = 0.99 usec; b = 0.43; o = 0.06; r = 1.24; rb = 0.68; ro = 0.30 |
641 | |
642 | 7. TODO |
643 | |
644 | a. SystemTap (http://sourceware.org/systemtap): Provides a simplified |
645 | programming interface for probe-based instrumentation. Try it out. |
646 | b. Kernel return probes for sparc64. |
647 | c. Support for other architectures. |
648 | d. User-space probes. |
649 | e. Watchpoint probes (which fire on data references). |
650 | |
651 | 8. Kprobes Example |
652 | |
653 | See samples/kprobes/kprobe_example.c |
654 | |
655 | 9. Jprobes Example |
656 | |
657 | See samples/kprobes/jprobe_example.c |
658 | |
659 | 10. Kretprobes Example |
660 | |
661 | See samples/kprobes/kretprobe_example.c |
662 | |
663 | For additional information on Kprobes, refer to the following URLs: |
664 | http://www-106.ibm.com/developerworks/library/l-kprobes.html?ca=dgr-lnxw42Kprobe |
665 | http://www.redhat.com/magazine/005mar05/features/kprobes/ |
666 | http://www-users.cs.umn.edu/~boutcher/kprobes/ |
667 | http://www.linuxsymposium.org/2006/linuxsymposium_procv2.pdf (pages 101-115) |
668 | |
669 | |
670 | Appendix A: The kprobes debugfs interface |
671 | |
672 | With recent kernels (> 2.6.20) the list of registered kprobes is visible |
673 | under the /sys/kernel/debug/kprobes/ directory (assuming debugfs is mounted at //sys/kernel/debug). |
674 | |
675 | /sys/kernel/debug/kprobes/list: Lists all registered probes on the system |
676 | |
677 | c015d71a k vfs_read+0x0 |
678 | c011a316 j do_fork+0x0 |
679 | c03dedc5 r tcp_v4_rcv+0x0 |
680 | |
681 | The first column provides the kernel address where the probe is inserted. |
682 | The second column identifies the type of probe (k - kprobe, r - kretprobe |
683 | and j - jprobe), while the third column specifies the symbol+offset of |
684 | the probe. If the probed function belongs to a module, the module name |
685 | is also specified. Following columns show probe status. If the probe is on |
686 | a virtual address that is no longer valid (module init sections, module |
687 | virtual addresses that correspond to modules that've been unloaded), |
688 | such probes are marked with [GONE]. If the probe is temporarily disabled, |
689 | such probes are marked with [DISABLED]. If the probe is optimized, it is |
690 | marked with [OPTIMIZED]. |
691 | |
692 | /sys/kernel/debug/kprobes/enabled: Turn kprobes ON/OFF forcibly. |
693 | |
694 | Provides a knob to globally and forcibly turn registered kprobes ON or OFF. |
695 | By default, all kprobes are enabled. By echoing "0" to this file, all |
696 | registered probes will be disarmed, till such time a "1" is echoed to this |
697 | file. Note that this knob just disarms and arms all kprobes and doesn't |
698 | change each probe's disabling state. This means that disabled kprobes (marked |
699 | [DISABLED]) will be not enabled if you turn ON all kprobes by this knob. |
700 | |
701 | |
702 | Appendix B: The kprobes sysctl interface |
703 | |
704 | /proc/sys/debug/kprobes-optimization: Turn kprobes optimization ON/OFF. |
705 | |
706 | When CONFIG_OPTPROBES=y, this sysctl interface appears and it provides |
707 | a knob to globally and forcibly turn jump optimization (see section |
708 | 1.4) ON or OFF. By default, jump optimization is allowed (ON). |
709 | If you echo "0" to this file or set "debug.kprobes_optimization" to |
710 | 0 via sysctl, all optimized probes will be unoptimized, and any new |
711 | probes registered after that will not be optimized. Note that this |
712 | knob *changes* the optimized state. This means that optimized probes |
713 | (marked [OPTIMIZED]) will be unoptimized ([OPTIMIZED] tag will be |
714 | removed). If the knob is turned on, they will be optimized again. |
715 | |
716 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9