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 IBM Corp. 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 <linux/uaccess.h>
29 #include <linux/module.h>
30 #include <linux/slab.h>
31 #include <linux/hardirq.h>
32 #include <linux/ftrace.h>
33 #include <asm/cacheflush.h>
34 #include <asm/sections.h>
37 DEFINE_PER_CPU(struct kprobe *, current_kprobe);
38 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
40 struct kretprobe_blackpoint kretprobe_blacklist[] = { };
42 DEFINE_INSN_CACHE_OPS(dmainsn);
44 static void *alloc_dmainsn_page(void)
46 return (void *)__get_free_page(GFP_KERNEL | GFP_DMA);
49 static void free_dmainsn_page(void *page)
51 free_page((unsigned long)page);
54 struct kprobe_insn_cache kprobe_dmainsn_slots = {
55 .mutex = __MUTEX_INITIALIZER(kprobe_dmainsn_slots.mutex),
56 .alloc = alloc_dmainsn_page,
57 .free = free_dmainsn_page,
58 .pages = LIST_HEAD_INIT(kprobe_dmainsn_slots.pages),
59 .insn_size = MAX_INSN_SIZE,
62 static void copy_instruction(struct kprobe *p)
64 unsigned long ip = (unsigned long) p->addr;
68 if (ftrace_location(ip) == ip) {
70 * If kprobes patches the instruction that is morphed by
71 * ftrace make sure that kprobes always sees the branch
72 * "jg .+24" that skips the mcount block
74 ftrace_generate_nop_insn((struct ftrace_insn *)p->ainsn.insn);
75 p->ainsn.is_ftrace_insn = 1;
77 memcpy(p->ainsn.insn, p->addr, insn_length(*p->addr >> 8));
78 p->opcode = p->ainsn.insn[0];
79 if (!probe_is_insn_relative_long(p->ainsn.insn))
82 * For pc-relative instructions in RIL-b or RIL-c format patch the
83 * RI2 displacement field. We have already made sure that the insn
84 * slot for the patched instruction is within the same 2GB area
85 * as the original instruction (either kernel image or module area).
86 * Therefore the new displacement will always fit.
88 disp = *(s32 *)&p->ainsn.insn[1];
89 addr = (u64)(unsigned long)p->addr;
90 new_addr = (u64)(unsigned long)p->ainsn.insn;
91 new_disp = ((addr + (disp * 2)) - new_addr) / 2;
92 *(s32 *)&p->ainsn.insn[1] = new_disp;
94 NOKPROBE_SYMBOL(copy_instruction);
96 static inline int is_kernel_addr(void *addr)
98 return addr < (void *)_end;
101 static int s390_get_insn_slot(struct kprobe *p)
104 * Get an insn slot that is within the same 2GB area like the original
105 * instruction. That way instructions with a 32bit signed displacement
106 * field can be patched and executed within the insn slot.
108 p->ainsn.insn = NULL;
109 if (is_kernel_addr(p->addr))
110 p->ainsn.insn = get_dmainsn_slot();
111 else if (is_module_addr(p->addr))
112 p->ainsn.insn = get_insn_slot();
113 return p->ainsn.insn ? 0 : -ENOMEM;
115 NOKPROBE_SYMBOL(s390_get_insn_slot);
117 static void s390_free_insn_slot(struct kprobe *p)
121 if (is_kernel_addr(p->addr))
122 free_dmainsn_slot(p->ainsn.insn, 0);
124 free_insn_slot(p->ainsn.insn, 0);
125 p->ainsn.insn = NULL;
127 NOKPROBE_SYMBOL(s390_free_insn_slot);
129 int arch_prepare_kprobe(struct kprobe *p)
131 if ((unsigned long) p->addr & 0x01)
133 /* Make sure the probe isn't going on a difficult instruction */
134 if (probe_is_prohibited_opcode(p->addr))
136 if (s390_get_insn_slot(p))
141 NOKPROBE_SYMBOL(arch_prepare_kprobe);
143 int arch_check_ftrace_location(struct kprobe *p)
148 struct swap_insn_args {
150 unsigned int arm_kprobe : 1;
153 static int swap_instruction(void *data)
155 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
156 unsigned long status = kcb->kprobe_status;
157 struct swap_insn_args *args = data;
158 struct ftrace_insn new_insn, *insn;
159 struct kprobe *p = args->p;
162 new_insn.opc = args->arm_kprobe ? BREAKPOINT_INSTRUCTION : p->opcode;
163 len = sizeof(new_insn.opc);
164 if (!p->ainsn.is_ftrace_insn)
166 len = sizeof(new_insn);
167 insn = (struct ftrace_insn *) p->addr;
168 if (args->arm_kprobe) {
169 if (is_ftrace_nop(insn))
170 new_insn.disp = KPROBE_ON_FTRACE_NOP;
172 new_insn.disp = KPROBE_ON_FTRACE_CALL;
174 ftrace_generate_call_insn(&new_insn, (unsigned long)p->addr);
175 if (insn->disp == KPROBE_ON_FTRACE_NOP)
176 ftrace_generate_nop_insn(&new_insn);
179 kcb->kprobe_status = KPROBE_SWAP_INST;
180 probe_kernel_write(p->addr, &new_insn, len);
181 kcb->kprobe_status = status;
184 NOKPROBE_SYMBOL(swap_instruction);
186 void arch_arm_kprobe(struct kprobe *p)
188 struct swap_insn_args args = {.p = p, .arm_kprobe = 1};
190 stop_machine(swap_instruction, &args, NULL);
192 NOKPROBE_SYMBOL(arch_arm_kprobe);
194 void arch_disarm_kprobe(struct kprobe *p)
196 struct swap_insn_args args = {.p = p, .arm_kprobe = 0};
198 stop_machine(swap_instruction, &args, NULL);
200 NOKPROBE_SYMBOL(arch_disarm_kprobe);
202 void arch_remove_kprobe(struct kprobe *p)
204 s390_free_insn_slot(p);
206 NOKPROBE_SYMBOL(arch_remove_kprobe);
208 static void enable_singlestep(struct kprobe_ctlblk *kcb,
209 struct pt_regs *regs,
212 struct per_regs per_kprobe;
214 /* Set up the PER control registers %cr9-%cr11 */
215 per_kprobe.control = PER_EVENT_IFETCH;
216 per_kprobe.start = ip;
219 /* Save control regs and psw mask */
220 __ctl_store(kcb->kprobe_saved_ctl, 9, 11);
221 kcb->kprobe_saved_imask = regs->psw.mask &
222 (PSW_MASK_PER | PSW_MASK_IO | PSW_MASK_EXT);
224 /* Set PER control regs, turns on single step for the given address */
225 __ctl_load(per_kprobe, 9, 11);
226 regs->psw.mask |= PSW_MASK_PER;
227 regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
228 regs->psw.addr = ip | PSW_ADDR_AMODE;
230 NOKPROBE_SYMBOL(enable_singlestep);
232 static void disable_singlestep(struct kprobe_ctlblk *kcb,
233 struct pt_regs *regs,
236 /* Restore control regs and psw mask, set new psw address */
237 __ctl_load(kcb->kprobe_saved_ctl, 9, 11);
238 regs->psw.mask &= ~PSW_MASK_PER;
239 regs->psw.mask |= kcb->kprobe_saved_imask;
240 regs->psw.addr = ip | PSW_ADDR_AMODE;
242 NOKPROBE_SYMBOL(disable_singlestep);
245 * Activate a kprobe by storing its pointer to current_kprobe. The
246 * previous kprobe is stored in kcb->prev_kprobe. A stack of up to
247 * two kprobes can be active, see KPROBE_REENTER.
249 static void push_kprobe(struct kprobe_ctlblk *kcb, struct kprobe *p)
251 kcb->prev_kprobe.kp = __this_cpu_read(current_kprobe);
252 kcb->prev_kprobe.status = kcb->kprobe_status;
253 __this_cpu_write(current_kprobe, p);
255 NOKPROBE_SYMBOL(push_kprobe);
258 * Deactivate a kprobe by backing up to the previous state. If the
259 * current state is KPROBE_REENTER prev_kprobe.kp will be non-NULL,
260 * for any other state prev_kprobe.kp will be NULL.
262 static void pop_kprobe(struct kprobe_ctlblk *kcb)
264 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
265 kcb->kprobe_status = kcb->prev_kprobe.status;
267 NOKPROBE_SYMBOL(pop_kprobe);
269 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
271 ri->ret_addr = (kprobe_opcode_t *) regs->gprs[14];
273 /* Replace the return addr with trampoline addr */
274 regs->gprs[14] = (unsigned long) &kretprobe_trampoline;
276 NOKPROBE_SYMBOL(arch_prepare_kretprobe);
278 static void kprobe_reenter_check(struct kprobe_ctlblk *kcb, struct kprobe *p)
280 switch (kcb->kprobe_status) {
281 case KPROBE_HIT_SSDONE:
282 case KPROBE_HIT_ACTIVE:
283 kprobes_inc_nmissed_count(p);
289 * A kprobe on the code path to single step an instruction
290 * is a BUG. The code path resides in the .kprobes.text
291 * section and is executed with interrupts disabled.
293 printk(KERN_EMERG "Invalid kprobe detected at %p.\n", p->addr);
298 NOKPROBE_SYMBOL(kprobe_reenter_check);
300 static int kprobe_handler(struct pt_regs *regs)
302 struct kprobe_ctlblk *kcb;
306 * We want to disable preemption for the entire duration of kprobe
307 * processing. That includes the calls to the pre/post handlers
308 * and single stepping the kprobe instruction.
311 kcb = get_kprobe_ctlblk();
312 p = get_kprobe((void *)((regs->psw.addr & PSW_ADDR_INSN) - 2));
315 if (kprobe_running()) {
317 * We have hit a kprobe while another is still
318 * active. This can happen in the pre and post
319 * handler. Single step the instruction of the
320 * new probe but do not call any handler function
321 * of this secondary kprobe.
322 * push_kprobe and pop_kprobe saves and restores
323 * the currently active kprobe.
325 kprobe_reenter_check(kcb, p);
327 kcb->kprobe_status = KPROBE_REENTER;
330 * If we have no pre-handler or it returned 0, we
331 * continue with single stepping. If we have a
332 * pre-handler and it returned non-zero, it prepped
333 * for calling the break_handler below on re-entry
334 * for jprobe processing, so get out doing nothing
338 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
339 if (p->pre_handler && p->pre_handler(p, regs))
341 kcb->kprobe_status = KPROBE_HIT_SS;
343 enable_singlestep(kcb, regs, (unsigned long) p->ainsn.insn);
345 } else if (kprobe_running()) {
346 p = __this_cpu_read(current_kprobe);
347 if (p->break_handler && p->break_handler(p, regs)) {
349 * Continuation after the jprobe completed and
350 * caused the jprobe_return trap. The jprobe
351 * break_handler "returns" to the original
352 * function that still has the kprobe breakpoint
353 * installed. We continue with single stepping.
355 kcb->kprobe_status = KPROBE_HIT_SS;
356 enable_singlestep(kcb, regs,
357 (unsigned long) p->ainsn.insn);
360 * No kprobe at this address and the current kprobe
361 * has no break handler (no jprobe!). The kernel just
362 * exploded, let the standard trap handler pick up the
366 * No kprobe at this address and no active kprobe. The trap has
367 * not been caused by a kprobe breakpoint. The race of breakpoint
368 * vs. kprobe remove does not exist because on s390 as we use
369 * stop_machine to arm/disarm the breakpoints.
371 preempt_enable_no_resched();
374 NOKPROBE_SYMBOL(kprobe_handler);
377 * Function return probe trampoline:
378 * - init_kprobes() establishes a probepoint here
379 * - When the probed function returns, this probe
380 * causes the handlers to fire
382 static void __used kretprobe_trampoline_holder(void)
384 asm volatile(".global kretprobe_trampoline\n"
385 "kretprobe_trampoline: bcr 0,0\n");
389 * Called when the probe at kretprobe trampoline is hit
391 static int trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
393 struct kretprobe_instance *ri;
394 struct hlist_head *head, empty_rp;
395 struct hlist_node *tmp;
396 unsigned long flags, orig_ret_address;
397 unsigned long trampoline_address;
398 kprobe_opcode_t *correct_ret_addr;
400 INIT_HLIST_HEAD(&empty_rp);
401 kretprobe_hash_lock(current, &head, &flags);
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 than 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
417 orig_ret_address = 0;
418 correct_ret_addr = NULL;
419 trampoline_address = (unsigned long) &kretprobe_trampoline;
420 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
421 if (ri->task != current)
422 /* another task is sharing our hash bucket */
425 orig_ret_address = (unsigned long) ri->ret_addr;
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);
438 correct_ret_addr = ri->ret_addr;
439 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
440 if (ri->task != current)
441 /* another task is sharing our hash bucket */
444 orig_ret_address = (unsigned long) ri->ret_addr;
446 if (ri->rp && ri->rp->handler) {
447 ri->ret_addr = correct_ret_addr;
448 ri->rp->handler(ri, regs);
451 recycle_rp_inst(ri, &empty_rp);
453 if (orig_ret_address != trampoline_address)
455 * This is the real return address. Any other
456 * instances associated with this task are for
457 * other calls deeper on the call stack
462 regs->psw.addr = orig_ret_address | PSW_ADDR_AMODE;
464 pop_kprobe(get_kprobe_ctlblk());
465 kretprobe_hash_unlock(current, &flags);
466 preempt_enable_no_resched();
468 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
469 hlist_del(&ri->hlist);
473 * By returning a non-zero value, we are telling
474 * kprobe_handler() that we don't want the post_handler
475 * to run (and have re-enabled preemption)
479 NOKPROBE_SYMBOL(trampoline_probe_handler);
482 * Called after single-stepping. p->addr is the address of the
483 * instruction whose first byte has been replaced by the "breakpoint"
484 * instruction. To avoid the SMP problems that can occur when we
485 * temporarily put back the original opcode to single-step, we
486 * single-stepped a copy of the instruction. The address of this
487 * copy is p->ainsn.insn.
489 static void resume_execution(struct kprobe *p, struct pt_regs *regs)
491 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
492 unsigned long ip = regs->psw.addr & PSW_ADDR_INSN;
493 int fixup = probe_get_fixup_type(p->ainsn.insn);
495 /* Check if the kprobes location is an enabled ftrace caller */
496 if (p->ainsn.is_ftrace_insn) {
497 struct ftrace_insn *insn = (struct ftrace_insn *) p->addr;
498 struct ftrace_insn call_insn;
500 ftrace_generate_call_insn(&call_insn, (unsigned long) p->addr);
502 * A kprobe on an enabled ftrace call site actually single
503 * stepped an unconditional branch (ftrace nop equivalent).
504 * Now we need to fixup things and pretend that a brasl r0,...
505 * was executed instead.
507 if (insn->disp == KPROBE_ON_FTRACE_CALL) {
508 ip += call_insn.disp * 2 - MCOUNT_INSN_SIZE;
509 regs->gprs[0] = (unsigned long)p->addr + sizeof(*insn);
513 if (fixup & FIXUP_PSW_NORMAL)
514 ip += (unsigned long) p->addr - (unsigned long) p->ainsn.insn;
516 if (fixup & FIXUP_BRANCH_NOT_TAKEN) {
517 int ilen = insn_length(p->ainsn.insn[0] >> 8);
518 if (ip - (unsigned long) p->ainsn.insn == ilen)
519 ip = (unsigned long) p->addr + ilen;
522 if (fixup & FIXUP_RETURN_REGISTER) {
523 int reg = (p->ainsn.insn[0] & 0xf0) >> 4;
524 regs->gprs[reg] += (unsigned long) p->addr -
525 (unsigned long) p->ainsn.insn;
528 disable_singlestep(kcb, regs, ip);
530 NOKPROBE_SYMBOL(resume_execution);
532 static int post_kprobe_handler(struct pt_regs *regs)
534 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
535 struct kprobe *p = kprobe_running();
540 if (kcb->kprobe_status != KPROBE_REENTER && p->post_handler) {
541 kcb->kprobe_status = KPROBE_HIT_SSDONE;
542 p->post_handler(p, regs, 0);
545 resume_execution(p, regs);
547 preempt_enable_no_resched();
550 * if somebody else is singlestepping across a probe point, psw mask
551 * will have PER set, in which case, continue the remaining processing
552 * of do_single_step, as if this is not a probe hit.
554 if (regs->psw.mask & PSW_MASK_PER)
559 NOKPROBE_SYMBOL(post_kprobe_handler);
561 static int kprobe_trap_handler(struct pt_regs *regs, int trapnr)
563 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
564 struct kprobe *p = kprobe_running();
565 const struct exception_table_entry *entry;
567 switch(kcb->kprobe_status) {
568 case KPROBE_SWAP_INST:
569 /* We are here because the instruction replacement failed */
574 * We are here because the instruction being single
575 * stepped caused a page fault. We reset the current
576 * kprobe and the nip points back to the probe address
577 * and allow the page fault handler to continue as a
580 disable_singlestep(kcb, regs, (unsigned long) p->addr);
582 preempt_enable_no_resched();
584 case KPROBE_HIT_ACTIVE:
585 case KPROBE_HIT_SSDONE:
587 * We increment the nmissed count for accounting,
588 * we can also use npre/npostfault count for accounting
589 * these specific fault cases.
591 kprobes_inc_nmissed_count(p);
594 * We come here because instructions in the pre/post
595 * handler caused the page_fault, this could happen
596 * if handler tries to access user space by
597 * copy_from_user(), get_user() etc. Let the
598 * user-specified handler try to fix it first.
600 if (p->fault_handler && p->fault_handler(p, regs, trapnr))
604 * In case the user-specified fault handler returned
605 * zero, try to fix up.
607 entry = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
609 regs->psw.addr = extable_fixup(entry) | PSW_ADDR_AMODE;
614 * fixup_exception() could not handle it,
615 * Let do_page_fault() fix it.
623 NOKPROBE_SYMBOL(kprobe_trap_handler);
625 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
629 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
631 ret = kprobe_trap_handler(regs, trapnr);
632 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
633 local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
636 NOKPROBE_SYMBOL(kprobe_fault_handler);
639 * Wrapper routine to for handling exceptions.
641 int kprobe_exceptions_notify(struct notifier_block *self,
642 unsigned long val, void *data)
644 struct die_args *args = (struct die_args *) data;
645 struct pt_regs *regs = args->regs;
646 int ret = NOTIFY_DONE;
648 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
653 if (kprobe_handler(regs))
657 if (post_kprobe_handler(regs))
661 if (!preemptible() && kprobe_running() &&
662 kprobe_trap_handler(regs, args->trapnr))
669 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
670 local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
674 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
676 int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
678 struct jprobe *jp = container_of(p, struct jprobe, kp);
679 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
682 memcpy(&kcb->jprobe_saved_regs, regs, sizeof(struct pt_regs));
684 /* setup return addr to the jprobe handler routine */
685 regs->psw.addr = (unsigned long) jp->entry | PSW_ADDR_AMODE;
686 regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
688 /* r15 is the stack pointer */
689 stack = (unsigned long) regs->gprs[15];
691 memcpy(kcb->jprobes_stack, (void *) stack, MIN_STACK_SIZE(stack));
694 NOKPROBE_SYMBOL(setjmp_pre_handler);
696 void jprobe_return(void)
698 asm volatile(".word 0x0002");
700 NOKPROBE_SYMBOL(jprobe_return);
702 int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
704 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
707 stack = (unsigned long) kcb->jprobe_saved_regs.gprs[15];
709 /* Put the regs back */
710 memcpy(regs, &kcb->jprobe_saved_regs, sizeof(struct pt_regs));
711 /* put the stack back */
712 memcpy((void *) stack, kcb->jprobes_stack, MIN_STACK_SIZE(stack));
713 preempt_enable_no_resched();
716 NOKPROBE_SYMBOL(longjmp_break_handler);
718 static struct kprobe trampoline = {
719 .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
720 .pre_handler = trampoline_probe_handler
723 int __init arch_init_kprobes(void)
725 return register_kprobe(&trampoline);
728 int arch_trampoline_kprobe(struct kprobe *p)
730 return p->addr == (kprobe_opcode_t *) &kretprobe_trampoline;
732 NOKPROBE_SYMBOL(arch_trampoline_kprobe);