[PATCH] kprobes: increment kprobe missed count for multiprobes
[linux-2.6/openmoko-kernel/knife-kernel.git] / arch / x86_64 / kernel / kprobes.c
blobafe11f4fbd1d49d068e7bc3fb64f6d759c342ef1
1 /*
2 * Kernel Probes (KProbes)
3 * arch/x86_64/kernel/kprobes.c
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19 * Copyright (C) IBM Corporation, 2002, 2004
21 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
22 * Probes initial implementation ( includes contributions from
23 * Rusty Russell).
24 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
25 * interface to access function arguments.
26 * 2004-Oct Jim Keniston <kenistoj@us.ibm.com> and Prasanna S Panchamukhi
27 * <prasanna@in.ibm.com> adapted for x86_64
28 * 2005-Mar Roland McGrath <roland@redhat.com>
29 * Fixed to handle %rip-relative addressing mode correctly.
30 * 2005-May Rusty Lynch <rusty.lynch@intel.com>
31 * Added function return probes functionality
34 #include <linux/config.h>
35 #include <linux/kprobes.h>
36 #include <linux/ptrace.h>
37 #include <linux/string.h>
38 #include <linux/slab.h>
39 #include <linux/preempt.h>
41 #include <asm/cacheflush.h>
42 #include <asm/pgtable.h>
43 #include <asm/kdebug.h>
45 static DECLARE_MUTEX(kprobe_mutex);
46 void jprobe_return_end(void);
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 inline int is_IF_modifier(kprobe_opcode_t *insn)
56 switch (*insn) {
57 case 0xfa: /* cli */
58 case 0xfb: /* sti */
59 case 0xcf: /* iret/iretd */
60 case 0x9d: /* popf/popfd */
61 return 1;
64 if (*insn >= 0x40 && *insn <= 0x4f && *++insn == 0xcf)
65 return 1;
66 return 0;
69 int __kprobes arch_prepare_kprobe(struct kprobe *p)
71 /* insn: must be on special executable page on x86_64. */
72 down(&kprobe_mutex);
73 p->ainsn.insn = get_insn_slot();
74 up(&kprobe_mutex);
75 if (!p->ainsn.insn) {
76 return -ENOMEM;
78 return 0;
82 * Determine if the instruction uses the %rip-relative addressing mode.
83 * If it does, return the address of the 32-bit displacement word.
84 * If not, return null.
86 static inline s32 *is_riprel(u8 *insn)
88 #define W(row,b0,b1,b2,b3,b4,b5,b6,b7,b8,b9,ba,bb,bc,bd,be,bf) \
89 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
90 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
91 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
92 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
93 << (row % 64))
94 static const u64 onebyte_has_modrm[256 / 64] = {
95 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
96 /* ------------------------------- */
97 W(0x00, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* 00 */
98 W(0x10, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* 10 */
99 W(0x20, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0)| /* 20 */
100 W(0x30, 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0), /* 30 */
101 W(0x40, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 40 */
102 W(0x50, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 50 */
103 W(0x60, 0,0,1,1,0,0,0,0,0,1,0,1,0,0,0,0)| /* 60 */
104 W(0x70, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* 70 */
105 W(0x80, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 80 */
106 W(0x90, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 90 */
107 W(0xa0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* a0 */
108 W(0xb0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* b0 */
109 W(0xc0, 1,1,0,0,1,1,1,1,0,0,0,0,0,0,0,0)| /* c0 */
110 W(0xd0, 1,1,1,1,0,0,0,0,1,1,1,1,1,1,1,1)| /* d0 */
111 W(0xe0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* e0 */
112 W(0xf0, 0,0,0,0,0,0,1,1,0,0,0,0,0,0,1,1) /* f0 */
113 /* ------------------------------- */
114 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
116 static const u64 twobyte_has_modrm[256 / 64] = {
117 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
118 /* ------------------------------- */
119 W(0x00, 1,1,1,1,0,0,0,0,0,0,0,0,0,1,0,1)| /* 0f */
120 W(0x10, 1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0)| /* 1f */
121 W(0x20, 1,1,1,1,1,0,1,0,1,1,1,1,1,1,1,1)| /* 2f */
122 W(0x30, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0), /* 3f */
123 W(0x40, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 4f */
124 W(0x50, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 5f */
125 W(0x60, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 6f */
126 W(0x70, 1,1,1,1,1,1,1,0,0,0,0,0,1,1,1,1), /* 7f */
127 W(0x80, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)| /* 8f */
128 W(0x90, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* 9f */
129 W(0xa0, 0,0,0,1,1,1,1,1,0,0,0,1,1,1,1,1)| /* af */
130 W(0xb0, 1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1), /* bf */
131 W(0xc0, 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0)| /* cf */
132 W(0xd0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* df */
133 W(0xe0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1)| /* ef */
134 W(0xf0, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0) /* ff */
135 /* ------------------------------- */
136 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
138 #undef W
139 int need_modrm;
141 /* Skip legacy instruction prefixes. */
142 while (1) {
143 switch (*insn) {
144 case 0x66:
145 case 0x67:
146 case 0x2e:
147 case 0x3e:
148 case 0x26:
149 case 0x64:
150 case 0x65:
151 case 0x36:
152 case 0xf0:
153 case 0xf3:
154 case 0xf2:
155 ++insn;
156 continue;
158 break;
161 /* Skip REX instruction prefix. */
162 if ((*insn & 0xf0) == 0x40)
163 ++insn;
165 if (*insn == 0x0f) { /* Two-byte opcode. */
166 ++insn;
167 need_modrm = test_bit(*insn, twobyte_has_modrm);
168 } else { /* One-byte opcode. */
169 need_modrm = test_bit(*insn, onebyte_has_modrm);
172 if (need_modrm) {
173 u8 modrm = *++insn;
174 if ((modrm & 0xc7) == 0x05) { /* %rip+disp32 addressing mode */
175 /* Displacement follows ModRM byte. */
176 return (s32 *) ++insn;
180 /* No %rip-relative addressing mode here. */
181 return NULL;
184 void __kprobes arch_copy_kprobe(struct kprobe *p)
186 s32 *ripdisp;
187 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE);
188 ripdisp = is_riprel(p->ainsn.insn);
189 if (ripdisp) {
191 * The copied instruction uses the %rip-relative
192 * addressing mode. Adjust the displacement for the
193 * difference between the original location of this
194 * instruction and the location of the copy that will
195 * actually be run. The tricky bit here is making sure
196 * that the sign extension happens correctly in this
197 * calculation, since we need a signed 32-bit result to
198 * be sign-extended to 64 bits when it's added to the
199 * %rip value and yield the same 64-bit result that the
200 * sign-extension of the original signed 32-bit
201 * displacement would have given.
203 s64 disp = (u8 *) p->addr + *ripdisp - (u8 *) p->ainsn.insn;
204 BUG_ON((s64) (s32) disp != disp); /* Sanity check. */
205 *ripdisp = disp;
207 p->opcode = *p->addr;
210 void __kprobes arch_arm_kprobe(struct kprobe *p)
212 *p->addr = BREAKPOINT_INSTRUCTION;
213 flush_icache_range((unsigned long) p->addr,
214 (unsigned long) p->addr + sizeof(kprobe_opcode_t));
217 void __kprobes arch_disarm_kprobe(struct kprobe *p)
219 *p->addr = p->opcode;
220 flush_icache_range((unsigned long) p->addr,
221 (unsigned long) p->addr + sizeof(kprobe_opcode_t));
224 void __kprobes arch_remove_kprobe(struct kprobe *p)
226 down(&kprobe_mutex);
227 free_insn_slot(p->ainsn.insn);
228 up(&kprobe_mutex);
231 static inline void save_previous_kprobe(struct kprobe_ctlblk *kcb)
233 kcb->prev_kprobe.kp = kprobe_running();
234 kcb->prev_kprobe.status = kcb->kprobe_status;
235 kcb->prev_kprobe.old_rflags = kcb->kprobe_old_rflags;
236 kcb->prev_kprobe.saved_rflags = kcb->kprobe_saved_rflags;
239 static inline void restore_previous_kprobe(struct kprobe_ctlblk *kcb)
241 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
242 kcb->kprobe_status = kcb->prev_kprobe.status;
243 kcb->kprobe_old_rflags = kcb->prev_kprobe.old_rflags;
244 kcb->kprobe_saved_rflags = kcb->prev_kprobe.saved_rflags;
247 static inline void set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
248 struct kprobe_ctlblk *kcb)
250 __get_cpu_var(current_kprobe) = p;
251 kcb->kprobe_saved_rflags = kcb->kprobe_old_rflags
252 = (regs->eflags & (TF_MASK | IF_MASK));
253 if (is_IF_modifier(p->ainsn.insn))
254 kcb->kprobe_saved_rflags &= ~IF_MASK;
257 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
259 regs->eflags |= TF_MASK;
260 regs->eflags &= ~IF_MASK;
261 /*single step inline if the instruction is an int3*/
262 if (p->opcode == BREAKPOINT_INSTRUCTION)
263 regs->rip = (unsigned long)p->addr;
264 else
265 regs->rip = (unsigned long)p->ainsn.insn;
268 /* Called with kretprobe_lock held */
269 void __kprobes arch_prepare_kretprobe(struct kretprobe *rp,
270 struct pt_regs *regs)
272 unsigned long *sara = (unsigned long *)regs->rsp;
273 struct kretprobe_instance *ri;
275 if ((ri = get_free_rp_inst(rp)) != NULL) {
276 ri->rp = rp;
277 ri->task = current;
278 ri->ret_addr = (kprobe_opcode_t *) *sara;
280 /* Replace the return addr with trampoline addr */
281 *sara = (unsigned long) &kretprobe_trampoline;
283 add_rp_inst(ri);
284 } else {
285 rp->nmissed++;
289 int __kprobes kprobe_handler(struct pt_regs *regs)
291 struct kprobe *p;
292 int ret = 0;
293 kprobe_opcode_t *addr = (kprobe_opcode_t *)(regs->rip - sizeof(kprobe_opcode_t));
294 struct kprobe_ctlblk *kcb;
297 * We don't want to be preempted for the entire
298 * duration of kprobe processing
300 preempt_disable();
301 kcb = get_kprobe_ctlblk();
303 /* Check we're not actually recursing */
304 if (kprobe_running()) {
305 p = get_kprobe(addr);
306 if (p) {
307 if (kcb->kprobe_status == KPROBE_HIT_SS &&
308 *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
309 regs->eflags &= ~TF_MASK;
310 regs->eflags |= kcb->kprobe_saved_rflags;
311 goto no_kprobe;
312 } else if (kcb->kprobe_status == KPROBE_HIT_SSDONE) {
313 /* TODO: Provide re-entrancy from
314 * post_kprobes_handler() and avoid exception
315 * stack corruption while single-stepping on
316 * the instruction of the new probe.
318 arch_disarm_kprobe(p);
319 regs->rip = (unsigned long)p->addr;
320 reset_current_kprobe();
321 ret = 1;
322 } else {
323 /* We have reentered the kprobe_handler(), since
324 * another probe was hit while within the
325 * handler. We here save the original kprobe
326 * variables and just single step on instruction
327 * of the new probe without calling any user
328 * handlers.
330 save_previous_kprobe(kcb);
331 set_current_kprobe(p, regs, kcb);
332 kprobes_inc_nmissed_count(p);
333 prepare_singlestep(p, regs);
334 kcb->kprobe_status = KPROBE_REENTER;
335 return 1;
337 } else {
338 p = __get_cpu_var(current_kprobe);
339 if (p->break_handler && p->break_handler(p, regs)) {
340 goto ss_probe;
343 goto no_kprobe;
346 p = get_kprobe(addr);
347 if (!p) {
348 if (*addr != BREAKPOINT_INSTRUCTION) {
350 * The breakpoint instruction was removed right
351 * after we hit it. Another cpu has removed
352 * either a probepoint or a debugger breakpoint
353 * at this address. In either case, no further
354 * handling of this interrupt is appropriate.
355 * Back up over the (now missing) int3 and run
356 * the original instruction.
358 regs->rip = (unsigned long)addr;
359 ret = 1;
361 /* Not one of ours: let kernel handle it */
362 goto no_kprobe;
365 set_current_kprobe(p, regs, kcb);
366 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
368 if (p->pre_handler && p->pre_handler(p, regs))
369 /* handler has already set things up, so skip ss setup */
370 return 1;
372 ss_probe:
373 prepare_singlestep(p, regs);
374 kcb->kprobe_status = KPROBE_HIT_SS;
375 return 1;
377 no_kprobe:
378 preempt_enable_no_resched();
379 return ret;
383 * For function-return probes, init_kprobes() establishes a probepoint
384 * here. When a retprobed function returns, this probe is hit and
385 * trampoline_probe_handler() runs, calling the kretprobe's handler.
387 void kretprobe_trampoline_holder(void)
389 asm volatile ( ".global kretprobe_trampoline\n"
390 "kretprobe_trampoline: \n"
391 "nop\n");
395 * Called when we hit the probe point at kretprobe_trampoline
397 int __kprobes trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
399 struct kretprobe_instance *ri = NULL;
400 struct hlist_head *head;
401 struct hlist_node *node, *tmp;
402 unsigned long flags, orig_ret_address = 0;
403 unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
405 spin_lock_irqsave(&kretprobe_lock, flags);
406 head = kretprobe_inst_table_head(current);
409 * It is possible to have multiple instances associated with a given
410 * task either because an multiple functions in the call path
411 * have a return probe installed on them, and/or more then one return
412 * return probe was registered for a target function.
414 * We can handle this because:
415 * - instances are always inserted at the head of the list
416 * - when multiple return probes are registered for the same
417 * function, the first instance's ret_addr will point to the
418 * real return address, and all the rest will point to
419 * kretprobe_trampoline
421 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
422 if (ri->task != current)
423 /* another task is sharing our hash bucket */
424 continue;
426 if (ri->rp && ri->rp->handler)
427 ri->rp->handler(ri, regs);
429 orig_ret_address = (unsigned long)ri->ret_addr;
430 recycle_rp_inst(ri);
432 if (orig_ret_address != trampoline_address)
434 * This is the real return address. Any other
435 * instances associated with this task are for
436 * other calls deeper on the call stack
438 break;
441 BUG_ON(!orig_ret_address || (orig_ret_address == trampoline_address));
442 regs->rip = orig_ret_address;
444 reset_current_kprobe();
445 spin_unlock_irqrestore(&kretprobe_lock, flags);
446 preempt_enable_no_resched();
449 * By returning a non-zero value, we are telling
450 * kprobe_handler() that we don't want the post_handler
451 * to run (and have re-enabled preemption)
453 return 1;
457 * Called after single-stepping. p->addr is the address of the
458 * instruction whose first byte has been replaced by the "int 3"
459 * instruction. To avoid the SMP problems that can occur when we
460 * temporarily put back the original opcode to single-step, we
461 * single-stepped a copy of the instruction. The address of this
462 * copy is p->ainsn.insn.
464 * This function prepares to return from the post-single-step
465 * interrupt. We have to fix up the stack as follows:
467 * 0) Except in the case of absolute or indirect jump or call instructions,
468 * the new rip is relative to the copied instruction. We need to make
469 * it relative to the original instruction.
471 * 1) If the single-stepped instruction was pushfl, then the TF and IF
472 * flags are set in the just-pushed eflags, and may need to be cleared.
474 * 2) If the single-stepped instruction was a call, the return address
475 * that is atop the stack is the address following the copied instruction.
476 * We need to make it the address following the original instruction.
478 static void __kprobes resume_execution(struct kprobe *p,
479 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
481 unsigned long *tos = (unsigned long *)regs->rsp;
482 unsigned long next_rip = 0;
483 unsigned long copy_rip = (unsigned long)p->ainsn.insn;
484 unsigned long orig_rip = (unsigned long)p->addr;
485 kprobe_opcode_t *insn = p->ainsn.insn;
487 /*skip the REX prefix*/
488 if (*insn >= 0x40 && *insn <= 0x4f)
489 insn++;
491 switch (*insn) {
492 case 0x9c: /* pushfl */
493 *tos &= ~(TF_MASK | IF_MASK);
494 *tos |= kcb->kprobe_old_rflags;
495 break;
496 case 0xc3: /* ret/lret */
497 case 0xcb:
498 case 0xc2:
499 case 0xca:
500 regs->eflags &= ~TF_MASK;
501 /* rip is already adjusted, no more changes required*/
502 return;
503 case 0xe8: /* call relative - Fix return addr */
504 *tos = orig_rip + (*tos - copy_rip);
505 break;
506 case 0xff:
507 if ((*insn & 0x30) == 0x10) {
508 /* call absolute, indirect */
509 /* Fix return addr; rip is correct. */
510 next_rip = regs->rip;
511 *tos = orig_rip + (*tos - copy_rip);
512 } else if (((*insn & 0x31) == 0x20) || /* jmp near, absolute indirect */
513 ((*insn & 0x31) == 0x21)) { /* jmp far, absolute indirect */
514 /* rip is correct. */
515 next_rip = regs->rip;
517 break;
518 case 0xea: /* jmp absolute -- rip is correct */
519 next_rip = regs->rip;
520 break;
521 default:
522 break;
525 regs->eflags &= ~TF_MASK;
526 if (next_rip) {
527 regs->rip = next_rip;
528 } else {
529 regs->rip = orig_rip + (regs->rip - copy_rip);
533 int __kprobes post_kprobe_handler(struct pt_regs *regs)
535 struct kprobe *cur = kprobe_running();
536 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
538 if (!cur)
539 return 0;
541 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
542 kcb->kprobe_status = KPROBE_HIT_SSDONE;
543 cur->post_handler(cur, regs, 0);
546 resume_execution(cur, regs, kcb);
547 regs->eflags |= kcb->kprobe_saved_rflags;
549 /* Restore the original saved kprobes variables and continue. */
550 if (kcb->kprobe_status == KPROBE_REENTER) {
551 restore_previous_kprobe(kcb);
552 goto out;
554 reset_current_kprobe();
555 out:
556 preempt_enable_no_resched();
559 * if somebody else is singlestepping across a probe point, eflags
560 * will have TF set, in which case, continue the remaining processing
561 * of do_debug, as if this is not a probe hit.
563 if (regs->eflags & TF_MASK)
564 return 0;
566 return 1;
569 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
571 struct kprobe *cur = kprobe_running();
572 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
574 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
575 return 1;
577 if (kcb->kprobe_status & KPROBE_HIT_SS) {
578 resume_execution(cur, regs, kcb);
579 regs->eflags |= kcb->kprobe_old_rflags;
581 reset_current_kprobe();
582 preempt_enable_no_resched();
584 return 0;
588 * Wrapper routine for handling exceptions.
590 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
591 unsigned long val, void *data)
593 struct die_args *args = (struct die_args *)data;
594 int ret = NOTIFY_DONE;
596 switch (val) {
597 case DIE_INT3:
598 if (kprobe_handler(args->regs))
599 ret = NOTIFY_STOP;
600 break;
601 case DIE_DEBUG:
602 if (post_kprobe_handler(args->regs))
603 ret = NOTIFY_STOP;
604 break;
605 case DIE_GPF:
606 case DIE_PAGE_FAULT:
607 /* kprobe_running() needs smp_processor_id() */
608 preempt_disable();
609 if (kprobe_running() &&
610 kprobe_fault_handler(args->regs, args->trapnr))
611 ret = NOTIFY_STOP;
612 preempt_enable();
613 break;
614 default:
615 break;
617 return ret;
620 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
622 struct jprobe *jp = container_of(p, struct jprobe, kp);
623 unsigned long addr;
624 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
626 kcb->jprobe_saved_regs = *regs;
627 kcb->jprobe_saved_rsp = (long *) regs->rsp;
628 addr = (unsigned long)(kcb->jprobe_saved_rsp);
630 * As Linus pointed out, gcc assumes that the callee
631 * owns the argument space and could overwrite it, e.g.
632 * tailcall optimization. So, to be absolutely safe
633 * we also save and restore enough stack bytes to cover
634 * the argument area.
636 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
637 MIN_STACK_SIZE(addr));
638 regs->eflags &= ~IF_MASK;
639 regs->rip = (unsigned long)(jp->entry);
640 return 1;
643 void __kprobes jprobe_return(void)
645 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
647 asm volatile (" xchg %%rbx,%%rsp \n"
648 " int3 \n"
649 " .globl jprobe_return_end \n"
650 " jprobe_return_end: \n"
651 " nop \n"::"b"
652 (kcb->jprobe_saved_rsp):"memory");
655 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
657 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
658 u8 *addr = (u8 *) (regs->rip - 1);
659 unsigned long stack_addr = (unsigned long)(kcb->jprobe_saved_rsp);
660 struct jprobe *jp = container_of(p, struct jprobe, kp);
662 if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
663 if ((long *)regs->rsp != kcb->jprobe_saved_rsp) {
664 struct pt_regs *saved_regs =
665 container_of(kcb->jprobe_saved_rsp,
666 struct pt_regs, rsp);
667 printk("current rsp %p does not match saved rsp %p\n",
668 (long *)regs->rsp, kcb->jprobe_saved_rsp);
669 printk("Saved registers for jprobe %p\n", jp);
670 show_registers(saved_regs);
671 printk("Current registers\n");
672 show_registers(regs);
673 BUG();
675 *regs = kcb->jprobe_saved_regs;
676 memcpy((kprobe_opcode_t *) stack_addr, kcb->jprobes_stack,
677 MIN_STACK_SIZE(stack_addr));
678 preempt_enable_no_resched();
679 return 1;
681 return 0;
684 static struct kprobe trampoline_p = {
685 .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
686 .pre_handler = trampoline_probe_handler
689 int __init arch_init_kprobes(void)
691 return register_kprobe(&trampoline_p);