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, 2006
20 * s390 port, used ppc64 as template. Mike Grundy <grundym@us.ibm.com>
23 #include <linux/kprobes.h>
24 #include <linux/ptrace.h>
25 #include <linux/preempt.h>
26 #include <linux/stop_machine.h>
27 #include <linux/kdebug.h>
28 #include <asm/cacheflush.h>
29 #include <asm/sections.h>
30 #include <asm/uaccess.h>
31 #include <linux/module.h>
33 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
34 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
36 int __kprobes arch_prepare_kprobe(struct kprobe *p)
38 /* Make sure the probe isn't going on a difficult instruction */
39 if (is_prohibited_opcode((kprobe_opcode_t *) p->addr))
42 if ((unsigned long)p->addr & 0x01) {
43 printk("Attempt to register kprobe at an unaligned address\n");
47 /* Use the get_insn_slot() facility for correctness */
48 if (!(p->ainsn.insn = get_insn_slot()))
51 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
53 get_instruction_type(&p->ainsn);
58 int __kprobes is_prohibited_opcode(kprobe_opcode_t *instruction)
60 switch (*(__u8 *) instruction) {
61 case 0x0c: /* bassm */
67 switch (*(__u16 *) instruction) {
69 case 0xb25a: /* bsa */
70 case 0xb240: /* bakr */
71 case 0xb258: /* bsg */
79 void __kprobes get_instruction_type(struct arch_specific_insn *ainsn)
81 /* default fixup method */
82 ainsn->fixup = FIXUP_PSW_NORMAL;
85 ainsn->reg = (*ainsn->insn & 0xf0) >> 4;
87 /* save the instruction length (pop 5-5) in bytes */
88 switch (*(__u8 *) (ainsn->insn) >> 4) {
101 switch (*(__u8 *) ainsn->insn) {
102 case 0x05: /* balr */
103 case 0x0d: /* basr */
104 ainsn->fixup = FIXUP_RETURN_REGISTER;
105 /* if r2 = 0, no branch will be taken */
106 if ((*ainsn->insn & 0x0f) == 0)
107 ainsn->fixup |= FIXUP_BRANCH_NOT_TAKEN;
109 case 0x06: /* bctr */
111 ainsn->fixup = FIXUP_BRANCH_NOT_TAKEN;
115 ainsn->fixup = FIXUP_RETURN_REGISTER;
120 case 0x87: /* bxle */
121 ainsn->fixup = FIXUP_BRANCH_NOT_TAKEN;
123 case 0x82: /* lpsw */
124 ainsn->fixup = FIXUP_NOT_REQUIRED;
126 case 0xb2: /* lpswe */
127 if (*(((__u8 *) ainsn->insn) + 1) == 0xb2) {
128 ainsn->fixup = FIXUP_NOT_REQUIRED;
131 case 0xa7: /* bras */
132 if ((*ainsn->insn & 0x0f) == 0x05) {
133 ainsn->fixup |= FIXUP_RETURN_REGISTER;
137 if ((*ainsn->insn & 0x0f) == 0x00 /* larl */
138 || (*ainsn->insn & 0x0f) == 0x05) /* brasl */
139 ainsn->fixup |= FIXUP_RETURN_REGISTER;
142 if (*(((__u8 *) ainsn->insn) + 5 ) == 0x44 || /* bxhg */
143 *(((__u8 *) ainsn->insn) + 5) == 0x45) {/* bxleg */
144 ainsn->fixup = FIXUP_BRANCH_NOT_TAKEN;
147 case 0xe3: /* bctg */
148 if (*(((__u8 *) ainsn->insn) + 5) == 0x46) {
149 ainsn->fixup = FIXUP_BRANCH_NOT_TAKEN;
155 static int __kprobes swap_instruction(void *aref)
157 struct ins_replace_args *args = aref;
163 * Text segment is read-only, hence we use stura to bypass dynamic
164 * address translation to exchange the instruction. Since stura
165 * always operates on four bytes, but we only want to exchange two
166 * bytes do some calculations to get things right. In addition we
167 * shall not cross any page boundaries (vmalloc area!) when writing
168 * the new instruction.
170 addr = (u32 *)((unsigned long)args->ptr & -4UL);
171 if ((unsigned long)args->ptr & 2)
172 instr = ((*addr) & 0xffff0000) | args->new;
174 instr = ((*addr) & 0x0000ffff) | args->new << 16;
183 : "a" (addr), "d" (instr)
189 void __kprobes arch_arm_kprobe(struct kprobe *p)
191 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
192 unsigned long status = kcb->kprobe_status;
193 struct ins_replace_args args;
196 args.old = p->opcode;
197 args.new = BREAKPOINT_INSTRUCTION;
199 kcb->kprobe_status = KPROBE_SWAP_INST;
200 stop_machine_run(swap_instruction, &args, NR_CPUS);
201 kcb->kprobe_status = status;
204 void __kprobes arch_disarm_kprobe(struct kprobe *p)
206 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
207 unsigned long status = kcb->kprobe_status;
208 struct ins_replace_args args;
211 args.old = BREAKPOINT_INSTRUCTION;
212 args.new = p->opcode;
214 kcb->kprobe_status = KPROBE_SWAP_INST;
215 stop_machine_run(swap_instruction, &args, NR_CPUS);
216 kcb->kprobe_status = status;
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 prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
228 per_cr_bits kprobe_per_regs[1];
230 memset(kprobe_per_regs, 0, sizeof(per_cr_bits));
231 regs->psw.addr = (unsigned long)p->ainsn.insn | PSW_ADDR_AMODE;
233 /* Set up the per control reg info, will pass to lctl */
234 kprobe_per_regs[0].em_instruction_fetch = 1;
235 kprobe_per_regs[0].starting_addr = (unsigned long)p->ainsn.insn;
236 kprobe_per_regs[0].ending_addr = (unsigned long)p->ainsn.insn + 1;
238 /* Set the PER control regs, turns on single step for this address */
239 __ctl_load(kprobe_per_regs, 9, 11);
240 regs->psw.mask |= PSW_MASK_PER;
241 regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT | PSW_MASK_MCHECK);
244 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
246 kcb->prev_kprobe.kp = kprobe_running();
247 kcb->prev_kprobe.status = kcb->kprobe_status;
248 kcb->prev_kprobe.kprobe_saved_imask = kcb->kprobe_saved_imask;
249 memcpy(kcb->prev_kprobe.kprobe_saved_ctl, kcb->kprobe_saved_ctl,
250 sizeof(kcb->kprobe_saved_ctl));
253 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
255 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
256 kcb->kprobe_status = kcb->prev_kprobe.status;
257 kcb->kprobe_saved_imask = kcb->prev_kprobe.kprobe_saved_imask;
258 memcpy(kcb->kprobe_saved_ctl, kcb->prev_kprobe.kprobe_saved_ctl,
259 sizeof(kcb->kprobe_saved_ctl));
262 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
263 struct kprobe_ctlblk *kcb)
265 __get_cpu_var(current_kprobe) = p;
266 /* Save the interrupt and per flags */
267 kcb->kprobe_saved_imask = regs->psw.mask &
268 (PSW_MASK_PER | PSW_MASK_IO | PSW_MASK_EXT | PSW_MASK_MCHECK);
269 /* Save the control regs that govern PER */
270 __ctl_store(kcb->kprobe_saved_ctl, 9, 11);
273 /* Called with kretprobe_lock held */
274 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
275 struct pt_regs *regs)
277 ri->ret_addr = (kprobe_opcode_t *) regs->gprs[14];
279 /* Replace the return addr with trampoline addr */
280 regs->gprs[14] = (unsigned long)&kretprobe_trampoline;
283 static int __kprobes kprobe_handler(struct pt_regs *regs)
287 unsigned long *addr = (unsigned long *)
288 ((regs->psw.addr & PSW_ADDR_INSN) - 2);
289 struct kprobe_ctlblk *kcb;
292 * We don't want to be preempted for the entire
293 * duration of kprobe processing
296 kcb = get_kprobe_ctlblk();
298 /* Check we're not actually recursing */
299 if (kprobe_running()) {
300 p = get_kprobe(addr);
302 if (kcb->kprobe_status == KPROBE_HIT_SS &&
303 *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
304 regs->psw.mask &= ~PSW_MASK_PER;
305 regs->psw.mask |= kcb->kprobe_saved_imask;
308 /* We have reentered the kprobe_handler(), since
309 * another probe was hit while within the handler.
310 * We here save the original kprobes variables and
311 * just single step on the instruction of the new probe
312 * without calling any user handlers.
314 save_previous_kprobe(kcb);
315 set_current_kprobe(p, regs, kcb);
316 kprobes_inc_nmissed_count(p);
317 prepare_singlestep(p, regs);
318 kcb->kprobe_status = KPROBE_REENTER;
321 p = __get_cpu_var(current_kprobe);
322 if (p->break_handler && p->break_handler(p, regs)) {
329 p = get_kprobe(addr);
332 * No kprobe at this address. The fault has not been
333 * caused by a kprobe breakpoint. The race of breakpoint
334 * vs. kprobe remove does not exist because on s390 we
335 * use stop_machine_run to arm/disarm the breakpoints.
339 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
340 set_current_kprobe(p, regs, kcb);
341 if (p->pre_handler && p->pre_handler(p, regs))
342 /* handler has already set things up, so skip ss setup */
346 prepare_singlestep(p, regs);
347 kcb->kprobe_status = KPROBE_HIT_SS;
351 preempt_enable_no_resched();
356 * Function return probe trampoline:
357 * - init_kprobes() establishes a probepoint here
358 * - When the probed function returns, this probe
359 * causes the handlers to fire
361 void kretprobe_trampoline_holder(void)
363 asm volatile(".global kretprobe_trampoline\n"
364 "kretprobe_trampoline: bcr 0,0\n");
368 * Called when the probe at kretprobe trampoline is hit
370 static int __kprobes trampoline_probe_handler(struct kprobe *p,
371 struct pt_regs *regs)
373 struct kretprobe_instance *ri = NULL;
374 struct hlist_head *head, empty_rp;
375 struct hlist_node *node, *tmp;
376 unsigned long flags, orig_ret_address = 0;
377 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
379 INIT_HLIST_HEAD(&empty_rp);
380 spin_lock_irqsave(&kretprobe_lock, flags);
381 head = kretprobe_inst_table_head(current);
384 * It is possible to have multiple instances associated with a given
385 * task either because an multiple functions in the call path
386 * have a return probe installed on them, and/or more then one return
387 * return probe was registered for a target function.
389 * We can handle this because:
390 * - instances are always inserted at the head of the list
391 * - when multiple return probes are registered for the same
392 * function, the first instance's ret_addr will point to the
393 * real return address, and all the rest will point to
394 * kretprobe_trampoline
396 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
397 if (ri->task != current)
398 /* another task is sharing our hash bucket */
401 if (ri->rp && ri->rp->handler)
402 ri->rp->handler(ri, regs);
404 orig_ret_address = (unsigned long)ri->ret_addr;
405 recycle_rp_inst(ri, &empty_rp);
407 if (orig_ret_address != trampoline_address) {
409 * This is the real return address. Any other
410 * instances associated with this task are for
411 * other calls deeper on the call stack
416 BUG_ON(!orig_ret_address || (orig_ret_address == trampoline_address));
417 regs->psw.addr = orig_ret_address | PSW_ADDR_AMODE;
419 reset_current_kprobe();
420 spin_unlock_irqrestore(&kretprobe_lock, flags);
421 preempt_enable_no_resched();
423 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
424 hlist_del(&ri->hlist);
428 * By returning a non-zero value, we are telling
429 * kprobe_handler() that we don't want the post_handler
430 * to run (and have re-enabled preemption)
436 * Called after single-stepping. p->addr is the address of the
437 * instruction whose first byte has been replaced by the "breakpoint"
438 * instruction. To avoid the SMP problems that can occur when we
439 * temporarily put back the original opcode to single-step, we
440 * single-stepped a copy of the instruction. The address of this
441 * copy is p->ainsn.insn.
443 static void __kprobes resume_execution(struct kprobe *p, struct pt_regs *regs)
445 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
447 regs->psw.addr &= PSW_ADDR_INSN;
449 if (p->ainsn.fixup & FIXUP_PSW_NORMAL)
450 regs->psw.addr = (unsigned long)p->addr +
451 ((unsigned long)regs->psw.addr -
452 (unsigned long)p->ainsn.insn);
454 if (p->ainsn.fixup & FIXUP_BRANCH_NOT_TAKEN)
455 if ((unsigned long)regs->psw.addr -
456 (unsigned long)p->ainsn.insn == p->ainsn.ilen)
457 regs->psw.addr = (unsigned long)p->addr + p->ainsn.ilen;
459 if (p->ainsn.fixup & FIXUP_RETURN_REGISTER)
460 regs->gprs[p->ainsn.reg] = ((unsigned long)p->addr +
461 (regs->gprs[p->ainsn.reg] -
462 (unsigned long)p->ainsn.insn))
465 regs->psw.addr |= PSW_ADDR_AMODE;
466 /* turn off PER mode */
467 regs->psw.mask &= ~PSW_MASK_PER;
468 /* Restore the original per control regs */
469 __ctl_load(kcb->kprobe_saved_ctl, 9, 11);
470 regs->psw.mask |= kcb->kprobe_saved_imask;
473 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
475 struct kprobe *cur = kprobe_running();
476 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
481 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
482 kcb->kprobe_status = KPROBE_HIT_SSDONE;
483 cur->post_handler(cur, regs, 0);
486 resume_execution(cur, regs);
488 /*Restore back the original saved kprobes variables and continue. */
489 if (kcb->kprobe_status == KPROBE_REENTER) {
490 restore_previous_kprobe(kcb);
493 reset_current_kprobe();
495 preempt_enable_no_resched();
498 * if somebody else is singlestepping across a probe point, psw mask
499 * will have PER set, in which case, continue the remaining processing
500 * of do_single_step, as if this is not a probe hit.
502 if (regs->psw.mask & PSW_MASK_PER) {
509 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
511 struct kprobe *cur = kprobe_running();
512 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
513 const struct exception_table_entry *entry;
515 switch(kcb->kprobe_status) {
516 case KPROBE_SWAP_INST:
517 /* We are here because the instruction replacement failed */
522 * We are here because the instruction being single
523 * stepped caused a page fault. We reset the current
524 * kprobe and the nip points back to the probe address
525 * and allow the page fault handler to continue as a
528 regs->psw.addr = (unsigned long)cur->addr | PSW_ADDR_AMODE;
529 regs->psw.mask &= ~PSW_MASK_PER;
530 regs->psw.mask |= kcb->kprobe_saved_imask;
531 if (kcb->kprobe_status == KPROBE_REENTER)
532 restore_previous_kprobe(kcb);
534 reset_current_kprobe();
535 preempt_enable_no_resched();
537 case KPROBE_HIT_ACTIVE:
538 case KPROBE_HIT_SSDONE:
540 * We increment the nmissed count for accounting,
541 * we can also use npre/npostfault count for accouting
542 * these specific fault cases.
544 kprobes_inc_nmissed_count(cur);
547 * We come here because instructions in the pre/post
548 * handler caused the page_fault, this could happen
549 * if handler tries to access user space by
550 * copy_from_user(), get_user() etc. Let the
551 * user-specified handler try to fix it first.
553 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
557 * In case the user-specified fault handler returned
558 * zero, try to fix up.
560 entry = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
562 regs->psw.addr = entry->fixup | PSW_ADDR_AMODE;
567 * fixup_exception() could not handle it,
568 * Let do_page_fault() fix it.
578 * Wrapper routine to for handling exceptions.
580 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
581 unsigned long val, void *data)
583 struct die_args *args = (struct die_args *)data;
584 int ret = NOTIFY_DONE;
588 if (kprobe_handler(args->regs))
592 if (post_kprobe_handler(args->regs))
596 /* kprobe_running() needs smp_processor_id() */
598 if (kprobe_running() &&
599 kprobe_fault_handler(args->regs, args->trapnr))
609 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
611 struct jprobe *jp = container_of(p, struct jprobe, kp);
613 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
615 memcpy(&kcb->jprobe_saved_regs, regs, sizeof(struct pt_regs));
617 /* setup return addr to the jprobe handler routine */
618 regs->psw.addr = (unsigned long)(jp->entry) | PSW_ADDR_AMODE;
620 /* r14 is the function return address */
621 kcb->jprobe_saved_r14 = (unsigned long)regs->gprs[14];
622 /* r15 is the stack pointer */
623 kcb->jprobe_saved_r15 = (unsigned long)regs->gprs[15];
624 addr = (unsigned long)kcb->jprobe_saved_r15;
626 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *) addr,
627 MIN_STACK_SIZE(addr));
631 void __kprobes jprobe_return(void)
633 asm volatile(".word 0x0002");
636 void __kprobes jprobe_return_end(void)
638 asm volatile("bcr 0,0");
641 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
643 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
644 unsigned long stack_addr = (unsigned long)(kcb->jprobe_saved_r15);
646 /* Put the regs back */
647 memcpy(regs, &kcb->jprobe_saved_regs, sizeof(struct pt_regs));
648 /* put the stack back */
649 memcpy((kprobe_opcode_t *) stack_addr, kcb->jprobes_stack,
650 MIN_STACK_SIZE(stack_addr));
651 preempt_enable_no_resched();
655 static struct kprobe trampoline_p = {
656 .addr = (kprobe_opcode_t *) & kretprobe_trampoline,
657 .pre_handler = trampoline_probe_handler
660 int __init arch_init_kprobes(void)
662 return register_kprobe(&trampoline_p);
665 int __kprobes arch_trampoline_kprobe(struct kprobe *p)
667 if (p->addr == (kprobe_opcode_t *) & kretprobe_trampoline)