2 * Kernel Probes (KProbes)
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright (C) IBM Corporation, 2002, 2004
20 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
21 * Probes initial implementation ( includes contributions from
23 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
24 * interface to access function arguments.
25 * 2004-Oct Jim Keniston <kenistoj@us.ibm.com> and Prasanna S Panchamukhi
26 * <prasanna@in.ibm.com> adapted for x86_64
27 * 2005-Mar Roland McGrath <roland@redhat.com>
28 * Fixed to handle %rip-relative addressing mode correctly.
29 * 2005-May Rusty Lynch <rusty.lynch@intel.com>
30 * Added function return probes functionality
33 #include <linux/kprobes.h>
34 #include <linux/ptrace.h>
35 #include <linux/string.h>
36 #include <linux/slab.h>
37 #include <linux/preempt.h>
38 #include <linux/module.h>
39 #include <linux/kdebug.h>
41 #include <asm/pgtable.h>
42 #include <asm/uaccess.h>
43 #include <asm/alternative.h>
45 void jprobe_return_end(void);
46 static void __kprobes
arch_copy_kprobe(struct kprobe
*p
);
48 DEFINE_PER_CPU(struct kprobe
*, current_kprobe
) = NULL
;
49 DEFINE_PER_CPU(struct kprobe_ctlblk
, kprobe_ctlblk
);
52 * returns non-zero if opcode modifies the interrupt flag.
54 static __always_inline
int is_IF_modifier(kprobe_opcode_t
*insn
)
59 case 0xcf: /* iret/iretd */
60 case 0x9d: /* popf/popfd */
64 if (*insn
>= 0x40 && *insn
<= 0x4f && *++insn
== 0xcf)
69 int __kprobes
arch_prepare_kprobe(struct kprobe
*p
)
71 /* insn: must be on special executable page on x86_64. */
72 p
->ainsn
.insn
= get_insn_slot();
81 * Determine if the instruction uses the %rip-relative addressing mode.
82 * If it does, return the address of the 32-bit displacement word.
83 * If not, return null.
85 static s32 __kprobes
*is_riprel(u8
*insn
)
87 #define W(row,b0,b1,b2,b3,b4,b5,b6,b7,b8,b9,ba,bb,bc,bd,be,bf) \
88 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
89 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
90 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
91 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
93 static const u64 onebyte_has_modrm
[256 / 64] = {
94 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
95 /* ------------------------------- */
96 W(0x00, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* 00 */
97 W(0x10, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* 10 */
98 W(0x20, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* 20 */
99 W(0x30, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0), /* 30 */
100 W(0x40, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 40 */
101 W(0x50, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 50 */
102 W(0x60, 0,0,1,1,0,0,0,0,0,1,0,1,0,0,0,0)| /* 60 */
103 W(0x70, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* 70 */
104 W(0x80, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 80 */
105 W(0x90, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 90 */
106 W(0xa0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* a0 */
107 W(0xb0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* b0 */
108 W(0xc0, 1,1,0,0,1,1,1,1,0,0,0,0,0,0,0,0)| /* c0 */
109 W(0xd0, 1,1,1,1,0,0,0,0,1,1,1,1,1,1,1,1)| /* d0 */
110 W(0xe0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* e0 */
111 W(0xf0, 0,0,0,0,0,0,1,1,0,0,0,0,0,0,1,1) /* f0 */
112 /* ------------------------------- */
113 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
115 static const u64 twobyte_has_modrm
[256 / 64] = {
116 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
117 /* ------------------------------- */
118 W(0x00, 1,1,1,1,0,0,0,0,0,0,0,0,0,1,0,1)| /* 0f */
119 W(0x10, 1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0)| /* 1f */
120 W(0x20, 1,1,1,1,1,0,1,0,1,1,1,1,1,1,1,1)| /* 2f */
121 W(0x30, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* 3f */
122 W(0x40, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 4f */
123 W(0x50, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 5f */
124 W(0x60, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 6f */
125 W(0x70, 1,1,1,1,1,1,1,0,0,0,0,0,1,1,1,1), /* 7f */
126 W(0x80, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 8f */
127 W(0x90, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 9f */
128 W(0xa0, 0,0,0,1,1,1,1,1,0,0,0,1,1,1,1,1)| /* af */
129 W(0xb0, 1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1), /* bf */
130 W(0xc0, 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0)| /* cf */
131 W(0xd0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* df */
132 W(0xe0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* ef */
133 W(0xf0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0) /* ff */
134 /* ------------------------------- */
135 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
140 /* Skip legacy instruction prefixes. */
160 /* Skip REX instruction prefix. */
161 if ((*insn
& 0xf0) == 0x40)
164 if (*insn
== 0x0f) { /* Two-byte opcode. */
166 need_modrm
= test_bit(*insn
, twobyte_has_modrm
);
167 } else { /* One-byte opcode. */
168 need_modrm
= test_bit(*insn
, onebyte_has_modrm
);
173 if ((modrm
& 0xc7) == 0x05) { /* %rip+disp32 addressing mode */
174 /* Displacement follows ModRM byte. */
175 return (s32
*) ++insn
;
179 /* No %rip-relative addressing mode here. */
183 static void __kprobes
arch_copy_kprobe(struct kprobe
*p
)
186 memcpy(p
->ainsn
.insn
, p
->addr
, MAX_INSN_SIZE
);
187 ripdisp
= is_riprel(p
->ainsn
.insn
);
190 * The copied instruction uses the %rip-relative
191 * addressing mode. Adjust the displacement for the
192 * difference between the original location of this
193 * instruction and the location of the copy that will
194 * actually be run. The tricky bit here is making sure
195 * that the sign extension happens correctly in this
196 * calculation, since we need a signed 32-bit result to
197 * be sign-extended to 64 bits when it's added to the
198 * %rip value and yield the same 64-bit result that the
199 * sign-extension of the original signed 32-bit
200 * displacement would have given.
202 s64 disp
= (u8
*) p
->addr
+ *ripdisp
- (u8
*) p
->ainsn
.insn
;
203 BUG_ON((s64
) (s32
) disp
!= disp
); /* Sanity check. */
206 p
->opcode
= *p
->addr
;
209 void __kprobes
arch_arm_kprobe(struct kprobe
*p
)
211 text_poke(p
->addr
, ((unsigned char []){BREAKPOINT_INSTRUCTION
}), 1);
214 void __kprobes
arch_disarm_kprobe(struct kprobe
*p
)
216 text_poke(p
->addr
, &p
->opcode
, 1);
219 void __kprobes
arch_remove_kprobe(struct kprobe
*p
)
221 mutex_lock(&kprobe_mutex
);
222 free_insn_slot(p
->ainsn
.insn
, 0);
223 mutex_unlock(&kprobe_mutex
);
226 static void __kprobes
save_previous_kprobe(struct kprobe_ctlblk
*kcb
)
228 kcb
->prev_kprobe
.kp
= kprobe_running();
229 kcb
->prev_kprobe
.status
= kcb
->kprobe_status
;
230 kcb
->prev_kprobe
.old_rflags
= kcb
->kprobe_old_rflags
;
231 kcb
->prev_kprobe
.saved_rflags
= kcb
->kprobe_saved_rflags
;
234 static void __kprobes
restore_previous_kprobe(struct kprobe_ctlblk
*kcb
)
236 __get_cpu_var(current_kprobe
) = kcb
->prev_kprobe
.kp
;
237 kcb
->kprobe_status
= kcb
->prev_kprobe
.status
;
238 kcb
->kprobe_old_rflags
= kcb
->prev_kprobe
.old_rflags
;
239 kcb
->kprobe_saved_rflags
= kcb
->prev_kprobe
.saved_rflags
;
242 static void __kprobes
set_current_kprobe(struct kprobe
*p
, struct pt_regs
*regs
,
243 struct kprobe_ctlblk
*kcb
)
245 __get_cpu_var(current_kprobe
) = p
;
246 kcb
->kprobe_saved_rflags
= kcb
->kprobe_old_rflags
247 = (regs
->eflags
& (TF_MASK
| IF_MASK
));
248 if (is_IF_modifier(p
->ainsn
.insn
))
249 kcb
->kprobe_saved_rflags
&= ~IF_MASK
;
252 static void __kprobes
prepare_singlestep(struct kprobe
*p
, struct pt_regs
*regs
)
254 regs
->eflags
|= TF_MASK
;
255 regs
->eflags
&= ~IF_MASK
;
256 /*single step inline if the instruction is an int3*/
257 if (p
->opcode
== BREAKPOINT_INSTRUCTION
)
258 regs
->rip
= (unsigned long)p
->addr
;
260 regs
->rip
= (unsigned long)p
->ainsn
.insn
;
263 /* Called with kretprobe_lock held */
264 void __kprobes
arch_prepare_kretprobe(struct kretprobe_instance
*ri
,
265 struct pt_regs
*regs
)
267 unsigned long *sara
= (unsigned long *)regs
->rsp
;
269 ri
->ret_addr
= (kprobe_opcode_t
*) *sara
;
270 /* Replace the return addr with trampoline addr */
271 *sara
= (unsigned long) &kretprobe_trampoline
;
274 int __kprobes
kprobe_handler(struct pt_regs
*regs
)
278 kprobe_opcode_t
*addr
= (kprobe_opcode_t
*)(regs
->rip
- sizeof(kprobe_opcode_t
));
279 struct kprobe_ctlblk
*kcb
;
282 * We don't want to be preempted for the entire
283 * duration of kprobe processing
286 kcb
= get_kprobe_ctlblk();
288 /* Check we're not actually recursing */
289 if (kprobe_running()) {
290 p
= get_kprobe(addr
);
292 if (kcb
->kprobe_status
== KPROBE_HIT_SS
&&
293 *p
->ainsn
.insn
== BREAKPOINT_INSTRUCTION
) {
294 regs
->eflags
&= ~TF_MASK
;
295 regs
->eflags
|= kcb
->kprobe_saved_rflags
;
297 } else if (kcb
->kprobe_status
== KPROBE_HIT_SSDONE
) {
298 /* TODO: Provide re-entrancy from
299 * post_kprobes_handler() and avoid exception
300 * stack corruption while single-stepping on
301 * the instruction of the new probe.
303 arch_disarm_kprobe(p
);
304 regs
->rip
= (unsigned long)p
->addr
;
305 reset_current_kprobe();
308 /* We have reentered the kprobe_handler(), since
309 * another probe was hit while within the
310 * handler. We here save the original kprobe
311 * variables and just single step on instruction
312 * of the new probe without calling any user
315 save_previous_kprobe(kcb
);
316 set_current_kprobe(p
, regs
, kcb
);
317 kprobes_inc_nmissed_count(p
);
318 prepare_singlestep(p
, regs
);
319 kcb
->kprobe_status
= KPROBE_REENTER
;
323 if (*addr
!= BREAKPOINT_INSTRUCTION
) {
324 /* The breakpoint instruction was removed by
325 * another cpu right after we hit, no further
326 * handling of this interrupt is appropriate
328 regs
->rip
= (unsigned long)addr
;
332 p
= __get_cpu_var(current_kprobe
);
333 if (p
->break_handler
&& p
->break_handler(p
, regs
)) {
340 p
= get_kprobe(addr
);
342 if (*addr
!= BREAKPOINT_INSTRUCTION
) {
344 * The breakpoint instruction was removed right
345 * after we hit it. Another cpu has removed
346 * either a probepoint or a debugger breakpoint
347 * at this address. In either case, no further
348 * handling of this interrupt is appropriate.
349 * Back up over the (now missing) int3 and run
350 * the original instruction.
352 regs
->rip
= (unsigned long)addr
;
355 /* Not one of ours: let kernel handle it */
359 set_current_kprobe(p
, regs
, kcb
);
360 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
362 if (p
->pre_handler
&& p
->pre_handler(p
, regs
))
363 /* handler has already set things up, so skip ss setup */
367 prepare_singlestep(p
, regs
);
368 kcb
->kprobe_status
= KPROBE_HIT_SS
;
372 preempt_enable_no_resched();
377 * For function-return probes, init_kprobes() establishes a probepoint
378 * here. When a retprobed function returns, this probe is hit and
379 * trampoline_probe_handler() runs, calling the kretprobe's handler.
381 void kretprobe_trampoline_holder(void)
383 asm volatile ( ".global kretprobe_trampoline\n"
384 "kretprobe_trampoline: \n"
389 * Called when we hit the probe point at kretprobe_trampoline
391 int __kprobes
trampoline_probe_handler(struct kprobe
*p
, struct pt_regs
*regs
)
393 struct kretprobe_instance
*ri
= NULL
;
394 struct hlist_head
*head
, empty_rp
;
395 struct hlist_node
*node
, *tmp
;
396 unsigned long flags
, orig_ret_address
= 0;
397 unsigned long trampoline_address
=(unsigned long)&kretprobe_trampoline
;
399 INIT_HLIST_HEAD(&empty_rp
);
400 spin_lock_irqsave(&kretprobe_lock
, flags
);
401 head
= kretprobe_inst_table_head(current
);
404 * It is possible to have multiple instances associated with a given
405 * task either because an multiple functions in the call path
406 * have a return probe installed on them, and/or more then one return
407 * return probe was registered for a target function.
409 * We can handle this because:
410 * - instances are always inserted at the head of the list
411 * - when multiple return probes are registered for the same
412 * function, the first instance's ret_addr will point to the
413 * real return address, and all the rest will point to
414 * kretprobe_trampoline
416 hlist_for_each_entry_safe(ri
, node
, tmp
, head
, hlist
) {
417 if (ri
->task
!= current
)
418 /* another task is sharing our hash bucket */
421 if (ri
->rp
&& ri
->rp
->handler
)
422 ri
->rp
->handler(ri
, regs
);
424 orig_ret_address
= (unsigned long)ri
->ret_addr
;
425 recycle_rp_inst(ri
, &empty_rp
);
427 if (orig_ret_address
!= trampoline_address
)
429 * This is the real return address. Any other
430 * instances associated with this task are for
431 * other calls deeper on the call stack
436 kretprobe_assert(ri
, orig_ret_address
, trampoline_address
);
437 regs
->rip
= orig_ret_address
;
439 reset_current_kprobe();
440 spin_unlock_irqrestore(&kretprobe_lock
, flags
);
441 preempt_enable_no_resched();
443 hlist_for_each_entry_safe(ri
, node
, tmp
, &empty_rp
, hlist
) {
444 hlist_del(&ri
->hlist
);
448 * By returning a non-zero value, we are telling
449 * kprobe_handler() that we don't want the post_handler
450 * to run (and have re-enabled preemption)
456 * Called after single-stepping. p->addr is the address of the
457 * instruction whose first byte has been replaced by the "int 3"
458 * instruction. To avoid the SMP problems that can occur when we
459 * temporarily put back the original opcode to single-step, we
460 * single-stepped a copy of the instruction. The address of this
461 * copy is p->ainsn.insn.
463 * This function prepares to return from the post-single-step
464 * interrupt. We have to fix up the stack as follows:
466 * 0) Except in the case of absolute or indirect jump or call instructions,
467 * the new rip is relative to the copied instruction. We need to make
468 * it relative to the original instruction.
470 * 1) If the single-stepped instruction was pushfl, then the TF and IF
471 * flags are set in the just-pushed eflags, and may need to be cleared.
473 * 2) If the single-stepped instruction was a call, the return address
474 * that is atop the stack is the address following the copied instruction.
475 * We need to make it the address following the original instruction.
477 static void __kprobes
resume_execution(struct kprobe
*p
,
478 struct pt_regs
*regs
, struct kprobe_ctlblk
*kcb
)
480 unsigned long *tos
= (unsigned long *)regs
->rsp
;
481 unsigned long next_rip
= 0;
482 unsigned long copy_rip
= (unsigned long)p
->ainsn
.insn
;
483 unsigned long orig_rip
= (unsigned long)p
->addr
;
484 kprobe_opcode_t
*insn
= p
->ainsn
.insn
;
486 /*skip the REX prefix*/
487 if (*insn
>= 0x40 && *insn
<= 0x4f)
491 case 0x9c: /* pushfl */
492 *tos
&= ~(TF_MASK
| IF_MASK
);
493 *tos
|= kcb
->kprobe_old_rflags
;
495 case 0xc3: /* ret/lret */
499 regs
->eflags
&= ~TF_MASK
;
500 /* rip is already adjusted, no more changes required*/
502 case 0xe8: /* call relative - Fix return addr */
503 *tos
= orig_rip
+ (*tos
- copy_rip
);
506 if ((insn
[1] & 0x30) == 0x10) {
507 /* call absolute, indirect */
508 /* Fix return addr; rip is correct. */
509 next_rip
= regs
->rip
;
510 *tos
= orig_rip
+ (*tos
- copy_rip
);
511 } else if (((insn
[1] & 0x31) == 0x20) || /* jmp near, absolute indirect */
512 ((insn
[1] & 0x31) == 0x21)) { /* jmp far, absolute indirect */
513 /* rip is correct. */
514 next_rip
= regs
->rip
;
517 case 0xea: /* jmp absolute -- rip is correct */
518 next_rip
= regs
->rip
;
524 regs
->eflags
&= ~TF_MASK
;
526 regs
->rip
= next_rip
;
528 regs
->rip
= orig_rip
+ (regs
->rip
- copy_rip
);
532 int __kprobes
post_kprobe_handler(struct pt_regs
*regs
)
534 struct kprobe
*cur
= kprobe_running();
535 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
540 if ((kcb
->kprobe_status
!= KPROBE_REENTER
) && cur
->post_handler
) {
541 kcb
->kprobe_status
= KPROBE_HIT_SSDONE
;
542 cur
->post_handler(cur
, regs
, 0);
545 resume_execution(cur
, regs
, kcb
);
546 regs
->eflags
|= kcb
->kprobe_saved_rflags
;
547 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
548 if (raw_irqs_disabled_flags(regs
->eflags
))
549 trace_hardirqs_off();
554 /* Restore the original saved kprobes variables and continue. */
555 if (kcb
->kprobe_status
== KPROBE_REENTER
) {
556 restore_previous_kprobe(kcb
);
559 reset_current_kprobe();
561 preempt_enable_no_resched();
564 * if somebody else is singlestepping across a probe point, eflags
565 * will have TF set, in which case, continue the remaining processing
566 * of do_debug, as if this is not a probe hit.
568 if (regs
->eflags
& TF_MASK
)
574 int __kprobes
kprobe_fault_handler(struct pt_regs
*regs
, int trapnr
)
576 struct kprobe
*cur
= kprobe_running();
577 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
578 const struct exception_table_entry
*fixup
;
580 switch(kcb
->kprobe_status
) {
584 * We are here because the instruction being single
585 * stepped caused a page fault. We reset the current
586 * kprobe and the rip points back to the probe address
587 * and allow the page fault handler to continue as a
590 regs
->rip
= (unsigned long)cur
->addr
;
591 regs
->eflags
|= kcb
->kprobe_old_rflags
;
592 if (kcb
->kprobe_status
== KPROBE_REENTER
)
593 restore_previous_kprobe(kcb
);
595 reset_current_kprobe();
596 preempt_enable_no_resched();
598 case KPROBE_HIT_ACTIVE
:
599 case KPROBE_HIT_SSDONE
:
601 * We increment the nmissed count for accounting,
602 * we can also use npre/npostfault count for accouting
603 * these specific fault cases.
605 kprobes_inc_nmissed_count(cur
);
608 * We come here because instructions in the pre/post
609 * handler caused the page_fault, this could happen
610 * if handler tries to access user space by
611 * copy_from_user(), get_user() etc. Let the
612 * user-specified handler try to fix it first.
614 if (cur
->fault_handler
&& cur
->fault_handler(cur
, regs
, trapnr
))
618 * In case the user-specified fault handler returned
619 * zero, try to fix up.
621 fixup
= search_exception_tables(regs
->rip
);
623 regs
->rip
= fixup
->fixup
;
628 * fixup() could not handle it,
629 * Let do_page_fault() fix it.
639 * Wrapper routine for handling exceptions.
641 int __kprobes
kprobe_exceptions_notify(struct notifier_block
*self
,
642 unsigned long val
, void *data
)
644 struct die_args
*args
= (struct die_args
*)data
;
645 int ret
= NOTIFY_DONE
;
647 if (args
->regs
&& user_mode(args
->regs
))
652 if (kprobe_handler(args
->regs
))
656 if (post_kprobe_handler(args
->regs
))
660 /* kprobe_running() needs smp_processor_id() */
662 if (kprobe_running() &&
663 kprobe_fault_handler(args
->regs
, args
->trapnr
))
673 int __kprobes
setjmp_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
675 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
677 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
679 kcb
->jprobe_saved_regs
= *regs
;
680 kcb
->jprobe_saved_rsp
= (long *) regs
->rsp
;
681 addr
= (unsigned long)(kcb
->jprobe_saved_rsp
);
683 * As Linus pointed out, gcc assumes that the callee
684 * owns the argument space and could overwrite it, e.g.
685 * tailcall optimization. So, to be absolutely safe
686 * we also save and restore enough stack bytes to cover
689 memcpy(kcb
->jprobes_stack
, (kprobe_opcode_t
*)addr
,
690 MIN_STACK_SIZE(addr
));
691 regs
->eflags
&= ~IF_MASK
;
692 trace_hardirqs_off();
693 regs
->rip
= (unsigned long)(jp
->entry
);
697 void __kprobes
jprobe_return(void)
699 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
701 asm volatile (" xchg %%rbx,%%rsp \n"
703 " .globl jprobe_return_end \n"
704 " jprobe_return_end: \n"
706 (kcb
->jprobe_saved_rsp
):"memory");
709 int __kprobes
longjmp_break_handler(struct kprobe
*p
, struct pt_regs
*regs
)
711 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
712 u8
*addr
= (u8
*) (regs
->rip
- 1);
713 unsigned long stack_addr
= (unsigned long)(kcb
->jprobe_saved_rsp
);
714 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
716 if ((addr
> (u8
*) jprobe_return
) && (addr
< (u8
*) jprobe_return_end
)) {
717 if ((long *)regs
->rsp
!= kcb
->jprobe_saved_rsp
) {
718 struct pt_regs
*saved_regs
=
719 container_of(kcb
->jprobe_saved_rsp
,
720 struct pt_regs
, rsp
);
721 printk("current rsp %p does not match saved rsp %p\n",
722 (long *)regs
->rsp
, kcb
->jprobe_saved_rsp
);
723 printk("Saved registers for jprobe %p\n", jp
);
724 show_registers(saved_regs
);
725 printk("Current registers\n");
726 show_registers(regs
);
729 *regs
= kcb
->jprobe_saved_regs
;
730 memcpy((kprobe_opcode_t
*) stack_addr
, kcb
->jprobes_stack
,
731 MIN_STACK_SIZE(stack_addr
));
732 preempt_enable_no_resched();
738 static struct kprobe trampoline_p
= {
739 .addr
= (kprobe_opcode_t
*) &kretprobe_trampoline
,
740 .pre_handler
= trampoline_probe_handler
743 int __init
arch_init_kprobes(void)
745 return register_kprobe(&trampoline_p
);
748 int __kprobes
arch_trampoline_kprobe(struct kprobe
*p
)
750 if (p
->addr
== (kprobe_opcode_t
*)&kretprobe_trampoline
)