2 * Kernel Probes (KProbes)
3 * arch/x86/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
23 * Dynamic Binary Instrumentation Module based on KProbes
24 * modules/kprobe/arch/asm-x86/dbi_kprobes.c
26 * This program is free software; you can redistribute it and/or modify
27 * it under the terms of the GNU General Public License as published by
28 * the Free Software Foundation; either version 2 of the License, or
29 * (at your option) any later version.
31 * This program is distributed in the hope that it will be useful,
32 * but WITHOUT ANY WARRANTY; without even the implied warranty of
33 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
34 * GNU General Public License for more details.
36 * You should have received a copy of the GNU General Public License
37 * along with this program; if not, write to the Free Software
38 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
40 * Copyright (C) Samsung Electronics, 2006-2010
42 * 2008-2009 Alexey Gerenkov <a.gerenkov@samsung.com> User-Space
43 * Probes initial implementation; Support x86/ARM/MIPS for both user and kernel spaces.
44 * 2010 Ekaterina Gorelkina <e.gorelkina@samsung.com>: redesign module for separating core and arch parts
45 * 2012 Stanislav Andreev <s.andreev@samsung.com>: added time debug profiling support; BUG() message fix
48 #include<linux/module.h>
49 #include <linux/kdebug.h>
51 #include "dbi_kprobes.h"
52 #include "../dbi_kprobes.h"
53 #include "../../dbi_kprobes.h"
55 #include "../../dbi_kdebug.h"
56 #include "../../dbi_insn_slots.h"
57 #include "../../dbi_kprobes_deps.h"
58 #define SUPRESS_BUG_MESSAGES
60 extern struct kprobe * per_cpu__current_kprobe;
61 extern struct kprobe * per_cpu__current_kprobe;
62 extern struct kprobe * current_kprobe;
64 DECLARE_MOD_FUNC_DEP(module_alloc, void *, unsigned long size);
65 DECLARE_MOD_FUNC_DEP(module_free, void, struct module *mod, void *module_region);
66 DECLARE_MOD_FUNC_DEP(fixup_exception, int, struct pt_regs * regs);
68 DECLARE_MOD_FUNC_DEP(freeze_processes, int, void);
69 DECLARE_MOD_FUNC_DEP(thaw_processes, void, void);
71 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 26))
72 DECLARE_MOD_FUNC_DEP(text_poke, void, void *addr, unsigned char *opcode, int len);
74 DECLARE_MOD_FUNC_DEP(text_poke, void *, void *addr, const void *opcode, size_t len);
76 DECLARE_MOD_FUNC_DEP(show_registers, void, struct pt_regs * regs);
78 DECLARE_MOD_DEP_WRAPPER (module_alloc, void *, unsigned long size)
79 IMP_MOD_DEP_WRAPPER (module_alloc, size)
81 DECLARE_MOD_DEP_WRAPPER (module_free, void, struct module *mod, void *module_region)
82 IMP_MOD_DEP_WRAPPER (module_free, mod, module_region)
84 DECLARE_MOD_DEP_WRAPPER (fixup_exception, int, struct pt_regs * regs)
85 IMP_MOD_DEP_WRAPPER (fixup_exception, regs)
87 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 26))
88 DECLARE_MOD_DEP_WRAPPER(text_poke, \
89 void, void *addr, unsigned char *opcode, int len)
91 DECLARE_MOD_DEP_WRAPPER(text_poke, \
92 void *, void *addr, const void *opcode, size_t len)
94 IMP_MOD_DEP_WRAPPER(text_poke, addr, opcode, len)
96 DECLARE_MOD_DEP_WRAPPER(show_registers, void, struct pt_regs * regs)
97 IMP_MOD_DEP_WRAPPER(show_registers, regs)
100 * Function return probe trampoline:
101 * - init_kprobes() establishes a probepoint here
102 * - When the probed function returns, this probe
103 * causes the handlers to fire
105 static __used void kretprobe_trampoline_holder(void)
107 asm volatile(".global kretprobe_trampoline\n"
108 "kretprobe_trampoline:\n"
110 /* skip cs, ip, orig_ax and gs. */
123 " call trampoline_probe_handler_x86\n"
124 /* move eflags to cs */
125 " movl 56(%esp), %edx\n"
126 " movl %edx, 52(%esp)\n"
127 /* replace saved flags with true return address. */
128 " movl %eax, 56(%esp)\n"
136 /* skip ds, es, fs, gs, orig_ax, and ip. Note: don't pop cs here*/
142 void kretprobe_trampoline(void);
144 /* insert a jmp code */
145 static __always_inline void set_jmp_op (void *from, void *to)
151 } __attribute__ ((packed)) * jop;
152 jop = (struct __arch_jmp_op *) from;
153 jop->raddr = (long) (to) - ((long) (from) + 5);
154 jop->op = RELATIVEJUMP_INSTRUCTION;
158 * returns non-zero if opcodes can be boosted.
160 int can_boost(kprobe_opcode_t *opcodes)
162 #define W(row,b0,b1,b2,b3,b4,b5,b6,b7,b8,b9,ba,bb,bc,bd,be,bf) \
163 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
164 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
165 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
166 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
169 * Undefined/reserved opcodes, conditional jump, Opcode Extension
170 * Groups, and some special opcodes can not be boost.
172 static const unsigned long twobyte_is_boostable[256 / 32] = {
173 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
174 /* ------------------------------- */
175 W (0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
176 W (0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0), /* 10 */
177 W (0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
178 W (0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0), /* 30 */
179 W (0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
180 W (0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0), /* 50 */
181 W (0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
182 W (0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1), /* 70 */
183 W (0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
184 W (0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1), /* 90 */
185 W (0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
186 W (0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1), /* b0 */
187 W (0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
188 W (0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1), /* d0 */
189 W (0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
190 W (0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0) /* f0 */
191 /* ------------------------------- */
192 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
195 kprobe_opcode_t opcode;
196 kprobe_opcode_t *orig_opcodes = opcodes;
198 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
200 opcode = *(opcodes++);
202 /* 2nd-byte opcode */
205 if (opcodes - orig_opcodes > MAX_INSN_SIZE - 1)
207 return test_bit (*opcodes, twobyte_is_boostable);
210 switch (opcode & 0xf0)
213 if (0x63 < opcode && opcode < 0x67)
214 goto retry; /* prefixes */
215 /* can't boost Address-size override and bound */
216 return (opcode != 0x62 && opcode != 0x67);
218 return 0; /* can't boost conditional jump */
220 /* can't boost software-interruptions */
221 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf;
223 /* can boost AA* and XLAT */
224 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7);
226 /* can boost in/out and absolute jmps */
227 return ((opcode & 0x04) || opcode == 0xea);
229 if ((opcode & 0x0c) == 0 && opcode != 0xf1)
230 goto retry; /* lock/rep(ne) prefix */
231 /* clear and set flags can be boost */
232 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe));
234 if (opcode == 0x26 || opcode == 0x36 || opcode == 0x3e)
235 goto retry; /* prefixes */
236 /* can't boost CS override and call */
237 return (opcode != 0x2e && opcode != 0x9a);
240 EXPORT_SYMBOL_GPL(can_boost);
243 * returns non-zero if opcode modifies the interrupt flag.
245 static int is_IF_modifier (kprobe_opcode_t opcode)
251 case 0xcf: /* iret/iretd */
252 case 0x9d: /* popf/popfd */
258 int arch_prepare_kprobe(struct kprobe *p, struct slot_manager *sm)
260 kprobe_opcode_t insns[KPROBES_TRAMP_LEN];
264 if ((unsigned long) p->addr & 0x01)
266 DBPRINTF ("Attempt to register kprobe at an unaligned address\n");
273 kprobe_opcode_t insn[MAX_INSN_SIZE];
274 struct arch_specific_insn ainsn;
275 /* insn: must be on special executable page on i386. */
276 p->ainsn.insn = alloc_insn_slot(sm);
279 memcpy (insn, p->addr, MAX_INSN_SIZE * sizeof (kprobe_opcode_t));
281 ret = arch_check_insn (&ainsn);
284 p->opcode = *p->addr;
287 if (can_boost (p->addr))
288 p->ainsn.boostable = 0;
290 p->ainsn.boostable = -1;
291 memcpy (p->ainsn.insn, insn, MAX_INSN_SIZE * sizeof (kprobe_opcode_t));
295 free_insn_slot(sm, p->ainsn.insn);
301 void prepare_singlestep (struct kprobe *p, struct pt_regs *regs)
305 regs->EREG (ip) = (unsigned long)p->ss_addr;
310 regs->EREG (flags) |= TF_MASK;
311 regs->EREG (flags) &= ~IF_MASK;
312 /*single step inline if the instruction is an int3 */
313 if (p->opcode == BREAKPOINT_INSTRUCTION){
314 regs->EREG (ip) = (unsigned long) p->addr;
315 //printk("break_insn!!!\n");
318 regs->EREG (ip) = (unsigned long) p->ainsn.insn;
321 EXPORT_SYMBOL_GPL(prepare_singlestep);
323 void save_previous_kprobe (struct kprobe_ctlblk *kcb, struct kprobe *cur_p)
325 if (kcb->prev_kprobe.kp != NULL)
327 panic("no space to save new probe[]: task = %d/%s, prev %p, current %p, new %p,",
328 current->pid, current->comm, kcb->prev_kprobe.kp->addr,
329 kprobe_running()->addr, cur_p->addr);
333 kcb->prev_kprobe.kp = kprobe_running();
334 kcb->prev_kprobe.status = kcb->kprobe_status;
338 void restore_previous_kprobe (struct kprobe_ctlblk *kcb)
340 __get_cpu_var (current_kprobe) = kcb->prev_kprobe.kp;
341 kcb->kprobe_status = kcb->prev_kprobe.status;
342 kcb->prev_kprobe.kp = NULL;
343 kcb->prev_kprobe.status = 0;
346 void set_current_kprobe (struct kprobe *p, struct pt_regs *regs, struct kprobe_ctlblk *kcb)
348 __get_cpu_var (current_kprobe) = p;
349 DBPRINTF ("set_current_kprobe[]: p=%p addr=%p\n", p, p->addr);
350 kcb->kprobe_saved_eflags = kcb->kprobe_old_eflags = (regs->EREG (flags) & (TF_MASK | IF_MASK));
351 if (is_IF_modifier (p->opcode))
352 kcb->kprobe_saved_eflags &= ~IF_MASK;
355 int kprobe_handler (struct pt_regs *regs)
357 struct kprobe *p = 0;
358 int ret = 0, reenter = 0;
359 kprobe_opcode_t *addr = NULL;
360 struct kprobe_ctlblk *kcb;
361 #ifdef SUPRESS_BUG_MESSAGES
362 int swap_oops_in_progress;
365 /* We're in an interrupt, but this is clear and BUG()-safe. */
366 addr = (kprobe_opcode_t *) (regs->EREG (ip) - sizeof (kprobe_opcode_t));
367 DBPRINTF ("KPROBE: regs->eip = 0x%lx addr = 0x%p\n", regs->EREG (ip), addr);
368 #ifdef SUPRESS_BUG_MESSAGES
369 // oops_in_progress used to avoid BUG() messages that slow down kprobe_handler() execution
370 swap_oops_in_progress = oops_in_progress;
371 oops_in_progress = 1;
375 kcb = get_kprobe_ctlblk ();
377 /* Check we're not actually recursing */
378 if (kprobe_running()) {
379 p = get_kprobe(addr);
381 if (kcb->kprobe_status == KPROBE_HIT_SS && *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
382 regs->EREG(flags) &= ~TF_MASK;
383 regs->EREG(flags) |= kcb->kprobe_saved_eflags;
388 /* We have reentered the kprobe_handler(), since
389 * another probe was hit while within the handler.
390 * We here save the original kprobes variables and
391 * just single step on the instruction of the new probe
392 * without calling any user handlers.
394 save_previous_kprobe (kcb, p);
395 set_current_kprobe (p, regs, kcb);
396 kprobes_inc_nmissed_count (p);
397 prepare_singlestep (p, regs);
398 kcb->kprobe_status = KPROBE_REENTER;
399 // FIXME should we enable preemption here??...
400 //preempt_enable_no_resched ();
401 #ifdef SUPRESS_BUG_MESSAGES
402 oops_in_progress = swap_oops_in_progress;
406 if (*addr != BREAKPOINT_INSTRUCTION) {
407 /* The breakpoint instruction was removed by
408 * another cpu right after we hit, no further
409 * handling of this interrupt is appropriate
411 regs->EREG(ip) -= sizeof(kprobe_opcode_t);
416 p = __get_cpu_var(current_kprobe);
417 if (p->break_handler && p->break_handler(p, regs))
424 DBPRINTF ("get_kprobe %p", addr);
426 p = get_kprobe(addr);
429 if (*addr != BREAKPOINT_INSTRUCTION) {
431 * The breakpoint instruction was removed right
432 * after we hit it. Another cpu has removed
433 * either a probepoint or a debugger breakpoint
434 * at this address. In either case, no further
435 * handling of this interrupt is appropriate.
436 * Back up over the (now missing) int3 and run
437 * the original instruction.
439 regs->EREG(ip) -= sizeof(kprobe_opcode_t);
444 /* Not one of ours: let kernel handle it */
445 DBPRINTF ("no_kprobe");
450 set_current_kprobe (p, regs, kcb);
453 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
456 ret = p->pre_handler(p, regs);
460 if (ret == 2) { // we have alreadyc called the handler, so just single step the instruction
461 DBPRINTF ("p->pre_handler[] 2");
464 DBPRINTF ("p->pre_handler[] 1");
465 // FIXME should we enable preemption here??...
466 //preempt_enable_no_resched ();
467 #ifdef SUPRESS_BUG_MESSAGES
468 oops_in_progress = swap_oops_in_progress;
470 /* handler has already set things up, so skip ss setup */
471 prepare_singlestep(p, regs);
474 DBPRINTF ("p->pre_handler[] 0");
477 DBPRINTF ("p = %p\n", p);
478 DBPRINTF ("p->opcode = 0x%lx *p->addr = 0x%lx p->addr = 0x%p\n", (unsigned long) p->opcode, p->tgid ? 0 : (unsigned long) (*p->addr), p->addr);
480 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PM)
481 if (p->ainsn.boostable == 1 && !p->post_handler)
483 /* Boost up -- we can execute copied instructions directly */
484 reset_current_kprobe ();
485 regs->EREG (ip) = (unsigned long) p->ainsn.insn;
486 preempt_enable_no_resched ();
487 #ifdef SUPRESS_BUG_MESSAGES
488 oops_in_progress = swap_oops_in_progress;
492 #endif // !CONFIG_PREEMPT
493 prepare_singlestep (p, regs);
494 kcb->kprobe_status = KPROBE_HIT_SS;
495 // FIXME should we enable preemption here??...
496 //preempt_enable_no_resched ();
497 #ifdef SUPRESS_BUG_MESSAGES
498 oops_in_progress = swap_oops_in_progress;
504 preempt_enable_no_resched ();
505 #ifdef SUPRESS_BUG_MESSAGES
506 oops_in_progress = swap_oops_in_progress;
511 int setjmp_pre_handler (struct kprobe *p, struct pt_regs *regs)
513 struct jprobe *jp = container_of (p, struct jprobe, kp);
514 kprobe_pre_entry_handler_t pre_entry;
517 unsigned long addr, args[6];
518 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk ();
520 DBPRINTF ("setjmp_pre_handler %p:%d", p->addr, p->tgid);
521 pre_entry = (kprobe_pre_entry_handler_t) jp->pre_entry;
522 entry = (entry_point_t) jp->entry;
524 /* handle __switch_to probe */
525 if ((p->addr == sched_addr) && sched_rp) {
526 /* FIXME: Actually 2nd parameter is not used for x86 */
527 patch_suspended_task(sched_rp, (struct task_struct *)regs->dx, regs);
530 kcb->jprobe_saved_regs = *regs;
531 kcb->jprobe_saved_esp = ®s->EREG(sp);
532 addr = (unsigned long)(kcb->jprobe_saved_esp);
534 /* TBD: As Linus pointed out, gcc assumes that the callee
535 * owns the argument space and could overwrite it, e.g.
536 * tailcall optimization. So, to be absolutely safe
537 * we also save and restore enough stack bytes to cover
538 * the argument area. */
539 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr, MIN_STACK_SIZE (addr));
540 regs->EREG(flags) &= ~IF_MASK;
541 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
542 trace_hardirqs_off();
545 p->ss_addr = pre_entry(jp->priv_arg, regs);
547 regs->EREG(ip) = (unsigned long)(jp->entry);
552 void dbi_jprobe_return (void)
554 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk ();
556 asm volatile(" xchgl %%ebx,%%esp \n"
558 " .globl dbi_jprobe_return_end \n"
559 " dbi_jprobe_return_end: \n"
560 " nop \n"::"b" (kcb->jprobe_saved_esp):"memory");
563 void arch_ujprobe_return(void)
568 * Called after single-stepping. p->addr is the address of the
569 * instruction whose first byte has been replaced by the "int 3"
570 * instruction. To avoid the SMP problems that can occur when we
571 * temporarily put back the original opcode to single-step, we
572 * single-stepped a copy of the instruction. The address of this
573 * copy is p->ainsn.insn.
575 * This function prepares to return from the post-single-step
576 * interrupt. We have to fix up the stack as follows:
578 * 0) Except in the case of absolute or indirect jump or call instructions,
579 * the new eip is relative to the copied instruction. We need to make
580 * it relative to the original instruction.
582 * 1) If the single-stepped instruction was pushfl, then the TF and IF
583 * flags are set in the just-pushed eflags, and may need to be cleared.
585 * 2) If the single-stepped instruction was a call, the return address
586 * that is atop the stack is the address following the copied instruction.
587 * We need to make it the address following the original instruction.
589 * This function also checks instruction size for preparing direct execution.
591 static void resume_execution (struct kprobe *p, struct pt_regs *regs, struct kprobe_ctlblk *kcb)
593 unsigned long *tos, tos_dword = 0;
594 unsigned long copy_eip = (unsigned long) p->ainsn.insn;
595 unsigned long orig_eip = (unsigned long) p->addr;
596 kprobe_opcode_t insns[2];
598 regs->EREG (flags) &= ~TF_MASK;
600 tos = (unsigned long *)®s->EREG(sp);
601 insns[0] = p->ainsn.insn[0];
602 insns[1] = p->ainsn.insn[1];
606 case 0x9c: /* pushfl */
607 *tos &= ~(TF_MASK | IF_MASK);
608 *tos |= kcb->kprobe_old_eflags;
610 case 0xc2: /* iret/ret/lret */
615 case 0xea: /* jmp absolute -- eip is correct */
616 /* eip is already adjusted, no more changes required */
617 p->ainsn.boostable = 1;
619 case 0xe8: /* call relative - Fix return addr */
620 *tos = orig_eip + (*tos - copy_eip);
622 case 0x9a: /* call absolute -- same as call absolute, indirect */
623 *tos = orig_eip + (*tos - copy_eip);
626 if ((insns[1] & 0x30) == 0x10)
629 * call absolute, indirect
630 * Fix return addr; eip is correct.
631 * But this is not boostable
633 *tos = orig_eip + (*tos - copy_eip);
636 else if (((insns[1] & 0x31) == 0x20) || /* jmp near, absolute indirect */
637 ((insns[1] & 0x31) == 0x21))
638 { /* jmp far, absolute indirect */
639 /* eip is correct. And this is boostable */
640 p->ainsn.boostable = 1;
647 if (p->ainsn.boostable == 0)
649 if ((regs->EREG (ip) > copy_eip) && (regs->EREG (ip) - copy_eip) + 5 < MAX_INSN_SIZE)
652 * These instructions can be executed directly if it
653 * jumps back to correct address.
655 set_jmp_op((void *)regs->EREG(ip), (void *)orig_eip + (regs->EREG(ip) - copy_eip));
656 p->ainsn.boostable = 1;
660 p->ainsn.boostable = -1;
664 regs->EREG (ip) = orig_eip + (regs->EREG (ip) - copy_eip);
671 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
672 * remain disabled thoroughout this function.
674 static int post_kprobe_handler (struct pt_regs *regs)
676 struct kprobe *cur = kprobe_running ();
677 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk ();
681 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler)
683 kcb->kprobe_status = KPROBE_HIT_SSDONE;
684 cur->post_handler (cur, regs, 0);
687 resume_execution (cur, regs, kcb);
688 regs->EREG (flags) |= kcb->kprobe_saved_eflags;
690 trace_hardirqs_fixup_flags (regs->EREG (flags));
692 /*Restore back the original saved kprobes variables and continue. */
693 if (kcb->kprobe_status == KPROBE_REENTER)
695 restore_previous_kprobe (kcb);
698 reset_current_kprobe ();
700 preempt_enable_no_resched ();
703 * if somebody else is singlestepping across a probe point, eflags
704 * will have TF set, in which case, continue the remaining processing
705 * of do_debug, as if this is not a probe hit.
707 if (regs->EREG (flags) & TF_MASK)
713 int kprobe_fault_handler (struct pt_regs *regs, int trapnr)
715 struct kprobe *cur = kprobe_running ();
716 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk ();
718 switch (kcb->kprobe_status)
723 * We are here because the instruction being single
724 * stepped caused a page fault. We reset the current
725 * kprobe and the eip points back to the probe address
726 * and allow the page fault handler to continue as a
729 regs->EREG (ip) = (unsigned long) cur->addr;
730 regs->EREG (flags) |= kcb->kprobe_old_eflags;
731 if (kcb->kprobe_status == KPROBE_REENTER)
732 restore_previous_kprobe (kcb);
734 reset_current_kprobe ();
735 preempt_enable_no_resched ();
737 case KPROBE_HIT_ACTIVE:
738 case KPROBE_HIT_SSDONE:
740 * We increment the nmissed count for accounting,
741 * we can also use npre/npostfault count for accouting
742 * these specific fault cases.
744 kprobes_inc_nmissed_count (cur);
747 * We come here because instructions in the pre/post
748 * handler caused the page_fault, this could happen
749 * if handler tries to access user space by
750 * copy_from_user(), get_user() etc. Let the
751 * user-specified handler try to fix it first.
753 if (cur->fault_handler && cur->fault_handler (cur, regs, trapnr))
757 * In case the user-specified fault handler returned
758 * zero, try to fix up.
760 if (fixup_exception (regs))
764 * fixup_exception() could not handle it,
765 * Let do_page_fault() fix it.
774 int kprobe_exceptions_notify (struct notifier_block *self, unsigned long val, void *data)
776 struct die_args *args = (struct die_args *) data;
777 int ret = NOTIFY_DONE;
779 DBPRINTF ("val = %ld, data = 0x%X", val, (unsigned int) data);
781 if (args->regs && user_mode_vm(args->regs))
784 DBPRINTF ("switch (val) %lu %d %d", val, DIE_INT3, DIE_TRAP);
787 #ifdef CONFIG_KPROBES
792 DBPRINTF ("before kprobe_handler ret=%d %p", ret, args->regs);
793 if (kprobe_handler (args->regs))
795 DBPRINTF ("after kprobe_handler ret=%d %p", ret, args->regs);
798 if (post_kprobe_handler (args->regs))
802 // kprobe_running() needs smp_processor_id()
804 if (kprobe_running () && kprobe_fault_handler (args->regs, args->trapnr))
811 DBPRINTF ("ret=%d", ret);
812 /* if(ret == NOTIFY_STOP) */
813 /* handled_exceptions++; */
818 static struct notifier_block kprobe_exceptions_nb = {
819 .notifier_call = kprobe_exceptions_notify,
823 int longjmp_break_handler (struct kprobe *p, struct pt_regs *regs)
825 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk ();
826 u8 *addr = (u8 *) (regs->EREG (ip) - 1);
827 unsigned long stack_addr = (unsigned long) (kcb->jprobe_saved_esp);
828 struct jprobe *jp = container_of (p, struct jprobe, kp);
830 DBPRINTF ("p = %p\n", p);
832 if ((addr > (u8 *) dbi_jprobe_return) && (addr < (u8 *) dbi_jprobe_return_end))
834 if ((unsigned long *)(®s->EREG(sp)) != kcb->jprobe_saved_esp)
836 struct pt_regs *saved_regs = &kcb->jprobe_saved_regs;
837 printk ("current esp %p does not match saved esp %p\n", ®s->EREG (sp), kcb->jprobe_saved_esp);
838 printk ("Saved registers for jprobe %p\n", jp);
839 show_registers (saved_regs);
840 printk ("Current registers\n");
841 show_registers (regs);
845 *regs = kcb->jprobe_saved_regs;
846 memcpy ((kprobe_opcode_t *) stack_addr, kcb->jprobes_stack, MIN_STACK_SIZE (stack_addr));
847 preempt_enable_no_resched ();
852 void arch_arm_kprobe (struct kprobe *p)
854 text_poke (p->addr, ((unsigned char[])
855 {BREAKPOINT_INSTRUCTION}), 1);
858 void arch_disarm_kprobe (struct kprobe *p)
860 text_poke (p->addr, &p->opcode, 1);
863 static __used void *trampoline_probe_handler_x86(struct pt_regs *regs)
865 return (void *)trampoline_probe_handler(NULL, regs);
868 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
870 unsigned long *sara = (unsigned long *)®s->EREG(sp);
871 ri->ret_addr = (kprobe_opcode_t *)*sara;
872 ri->sp = ®s->EREG(sp);
874 /* Replace the return addr with trampoline addr */
875 *sara = (unsigned long)&kretprobe_trampoline;
878 int arch_init_module_deps()
880 INIT_MOD_DEP_VAR(module_alloc, module_alloc);
881 INIT_MOD_DEP_VAR(module_free, module_free);
882 INIT_MOD_DEP_VAR(fixup_exception, fixup_exception);
883 #if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 23)
884 # error this kernel version has no text_poke function which is necessaryf for x86 ach!!!
886 INIT_MOD_DEP_VAR(text_poke, text_poke);
888 INIT_MOD_DEP_VAR(show_registers, show_registers);
889 #if defined(CONFIG_PREEMPT) && defined(CONFIG_PM)
890 INIT_MOD_DEP_VAR(freeze_processes, freeze_processes);
891 INIT_MOD_DEP_VAR(thaw_processes, thaw_processes);
897 int arch_init_kprobes(void)
899 return register_die_notifier (&kprobe_exceptions_nb);
902 void arch_exit_kprobes(void)
904 unregister_die_notifier (&kprobe_exceptions_nb);