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/extable.h>
30 #include <linux/module.h>
31 #include <linux/slab.h>
32 #include <linux/hardirq.h>
33 #include <linux/ftrace.h>
34 #include <asm/cacheflush.h>
35 #include <asm/sections.h>
36 #include <asm/uaccess.h>
39 DEFINE_PER_CPU(struct kprobe *, current_kprobe);
40 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
42 struct kretprobe_blackpoint kretprobe_blacklist[] = { };
44 DEFINE_INSN_CACHE_OPS(dmainsn);
46 static void *alloc_dmainsn_page(void)
48 return (void *)__get_free_page(GFP_KERNEL | GFP_DMA);
51 static void free_dmainsn_page(void *page)
53 free_page((unsigned long)page);
56 struct kprobe_insn_cache kprobe_dmainsn_slots = {
57 .mutex = __MUTEX_INITIALIZER(kprobe_dmainsn_slots.mutex),
58 .alloc = alloc_dmainsn_page,
59 .free = free_dmainsn_page,
60 .pages = LIST_HEAD_INIT(kprobe_dmainsn_slots.pages),
61 .insn_size = MAX_INSN_SIZE,
64 static void copy_instruction(struct kprobe *p)
66 unsigned long ip = (unsigned long) p->addr;
70 if (ftrace_location(ip) == ip) {
72 * If kprobes patches the instruction that is morphed by
73 * ftrace make sure that kprobes always sees the branch
74 * "jg .+24" that skips the mcount block or the "brcl 0,0"
75 * in case of hotpatch.
77 ftrace_generate_nop_insn((struct ftrace_insn *)p->ainsn.insn);
78 p->ainsn.is_ftrace_insn = 1;
80 memcpy(p->ainsn.insn, p->addr, insn_length(*p->addr >> 8));
81 p->opcode = p->ainsn.insn[0];
82 if (!probe_is_insn_relative_long(p->ainsn.insn))
85 * For pc-relative instructions in RIL-b or RIL-c format patch the
86 * RI2 displacement field. We have already made sure that the insn
87 * slot for the patched instruction is within the same 2GB area
88 * as the original instruction (either kernel image or module area).
89 * Therefore the new displacement will always fit.
91 disp = *(s32 *)&p->ainsn.insn[1];
92 addr = (u64)(unsigned long)p->addr;
93 new_addr = (u64)(unsigned long)p->ainsn.insn;
94 new_disp = ((addr + (disp * 2)) - new_addr) / 2;
95 *(s32 *)&p->ainsn.insn[1] = new_disp;
97 NOKPROBE_SYMBOL(copy_instruction);
99 static inline int is_kernel_addr(void *addr)
101 return addr < (void *)_end;
104 static int s390_get_insn_slot(struct kprobe *p)
107 * Get an insn slot that is within the same 2GB area like the original
108 * instruction. That way instructions with a 32bit signed displacement
109 * field can be patched and executed within the insn slot.
111 p->ainsn.insn = NULL;
112 if (is_kernel_addr(p->addr))
113 p->ainsn.insn = get_dmainsn_slot();
114 else if (is_module_addr(p->addr))
115 p->ainsn.insn = get_insn_slot();
116 return p->ainsn.insn ? 0 : -ENOMEM;
118 NOKPROBE_SYMBOL(s390_get_insn_slot);
120 static void s390_free_insn_slot(struct kprobe *p)
124 if (is_kernel_addr(p->addr))
125 free_dmainsn_slot(p->ainsn.insn, 0);
127 free_insn_slot(p->ainsn.insn, 0);
128 p->ainsn.insn = NULL;
130 NOKPROBE_SYMBOL(s390_free_insn_slot);
132 int arch_prepare_kprobe(struct kprobe *p)
134 if ((unsigned long) p->addr & 0x01)
136 /* Make sure the probe isn't going on a difficult instruction */
137 if (probe_is_prohibited_opcode(p->addr))
139 if (s390_get_insn_slot(p))
144 NOKPROBE_SYMBOL(arch_prepare_kprobe);
146 int arch_check_ftrace_location(struct kprobe *p)
151 struct swap_insn_args {
153 unsigned int arm_kprobe : 1;
156 static int swap_instruction(void *data)
158 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
159 unsigned long status = kcb->kprobe_status;
160 struct swap_insn_args *args = data;
161 struct ftrace_insn new_insn, *insn;
162 struct kprobe *p = args->p;
165 new_insn.opc = args->arm_kprobe ? BREAKPOINT_INSTRUCTION : p->opcode;
166 len = sizeof(new_insn.opc);
167 if (!p->ainsn.is_ftrace_insn)
169 len = sizeof(new_insn);
170 insn = (struct ftrace_insn *) p->addr;
171 if (args->arm_kprobe) {
172 if (is_ftrace_nop(insn))
173 new_insn.disp = KPROBE_ON_FTRACE_NOP;
175 new_insn.disp = KPROBE_ON_FTRACE_CALL;
177 ftrace_generate_call_insn(&new_insn, (unsigned long)p->addr);
178 if (insn->disp == KPROBE_ON_FTRACE_NOP)
179 ftrace_generate_nop_insn(&new_insn);
182 kcb->kprobe_status = KPROBE_SWAP_INST;
183 s390_kernel_write(p->addr, &new_insn, len);
184 kcb->kprobe_status = status;
187 NOKPROBE_SYMBOL(swap_instruction);
189 void arch_arm_kprobe(struct kprobe *p)
191 struct swap_insn_args args = {.p = p, .arm_kprobe = 1};
193 stop_machine(swap_instruction, &args, NULL);
195 NOKPROBE_SYMBOL(arch_arm_kprobe);
197 void arch_disarm_kprobe(struct kprobe *p)
199 struct swap_insn_args args = {.p = p, .arm_kprobe = 0};
201 stop_machine(swap_instruction, &args, NULL);
203 NOKPROBE_SYMBOL(arch_disarm_kprobe);
205 void arch_remove_kprobe(struct kprobe *p)
207 s390_free_insn_slot(p);
209 NOKPROBE_SYMBOL(arch_remove_kprobe);
211 static void enable_singlestep(struct kprobe_ctlblk *kcb,
212 struct pt_regs *regs,
215 struct per_regs per_kprobe;
217 /* Set up the PER control registers %cr9-%cr11 */
218 per_kprobe.control = PER_EVENT_IFETCH;
219 per_kprobe.start = ip;
222 /* Save control regs and psw mask */
223 __ctl_store(kcb->kprobe_saved_ctl, 9, 11);
224 kcb->kprobe_saved_imask = regs->psw.mask &
225 (PSW_MASK_PER | PSW_MASK_IO | PSW_MASK_EXT);
227 /* Set PER control regs, turns on single step for the given address */
228 __ctl_load(per_kprobe, 9, 11);
229 regs->psw.mask |= PSW_MASK_PER;
230 regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
233 NOKPROBE_SYMBOL(enable_singlestep);
235 static void disable_singlestep(struct kprobe_ctlblk *kcb,
236 struct pt_regs *regs,
239 /* Restore control regs and psw mask, set new psw address */
240 __ctl_load(kcb->kprobe_saved_ctl, 9, 11);
241 regs->psw.mask &= ~PSW_MASK_PER;
242 regs->psw.mask |= kcb->kprobe_saved_imask;
245 NOKPROBE_SYMBOL(disable_singlestep);
248 * Activate a kprobe by storing its pointer to current_kprobe. The
249 * previous kprobe is stored in kcb->prev_kprobe. A stack of up to
250 * two kprobes can be active, see KPROBE_REENTER.
252 static void push_kprobe(struct kprobe_ctlblk *kcb, struct kprobe *p)
254 kcb->prev_kprobe.kp = __this_cpu_read(current_kprobe);
255 kcb->prev_kprobe.status = kcb->kprobe_status;
256 __this_cpu_write(current_kprobe, p);
258 NOKPROBE_SYMBOL(push_kprobe);
261 * Deactivate a kprobe by backing up to the previous state. If the
262 * current state is KPROBE_REENTER prev_kprobe.kp will be non-NULL,
263 * for any other state prev_kprobe.kp will be NULL.
265 static void pop_kprobe(struct kprobe_ctlblk *kcb)
267 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
268 kcb->kprobe_status = kcb->prev_kprobe.status;
270 NOKPROBE_SYMBOL(pop_kprobe);
272 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
274 ri->ret_addr = (kprobe_opcode_t *) regs->gprs[14];
276 /* Replace the return addr with trampoline addr */
277 regs->gprs[14] = (unsigned long) &kretprobe_trampoline;
279 NOKPROBE_SYMBOL(arch_prepare_kretprobe);
281 static void kprobe_reenter_check(struct kprobe_ctlblk *kcb, struct kprobe *p)
283 switch (kcb->kprobe_status) {
284 case KPROBE_HIT_SSDONE:
285 case KPROBE_HIT_ACTIVE:
286 kprobes_inc_nmissed_count(p);
292 * A kprobe on the code path to single step an instruction
293 * is a BUG. The code path resides in the .kprobes.text
294 * section and is executed with interrupts disabled.
296 printk(KERN_EMERG "Invalid kprobe detected at %p.\n", p->addr);
301 NOKPROBE_SYMBOL(kprobe_reenter_check);
303 static int kprobe_handler(struct pt_regs *regs)
305 struct kprobe_ctlblk *kcb;
309 * We want to disable preemption for the entire duration of kprobe
310 * processing. That includes the calls to the pre/post handlers
311 * and single stepping the kprobe instruction.
314 kcb = get_kprobe_ctlblk();
315 p = get_kprobe((void *)(regs->psw.addr - 2));
318 if (kprobe_running()) {
320 * We have hit a kprobe while another is still
321 * active. This can happen in the pre and post
322 * handler. Single step the instruction of the
323 * new probe but do not call any handler function
324 * of this secondary kprobe.
325 * push_kprobe and pop_kprobe saves and restores
326 * the currently active kprobe.
328 kprobe_reenter_check(kcb, p);
330 kcb->kprobe_status = KPROBE_REENTER;
333 * If we have no pre-handler or it returned 0, we
334 * continue with single stepping. If we have a
335 * pre-handler and it returned non-zero, it prepped
336 * for calling the break_handler below on re-entry
337 * for jprobe processing, so get out doing nothing
341 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
342 if (p->pre_handler && p->pre_handler(p, regs))
344 kcb->kprobe_status = KPROBE_HIT_SS;
346 enable_singlestep(kcb, regs, (unsigned long) p->ainsn.insn);
348 } else if (kprobe_running()) {
349 p = __this_cpu_read(current_kprobe);
350 if (p->break_handler && p->break_handler(p, regs)) {
352 * Continuation after the jprobe completed and
353 * caused the jprobe_return trap. The jprobe
354 * break_handler "returns" to the original
355 * function that still has the kprobe breakpoint
356 * installed. We continue with single stepping.
358 kcb->kprobe_status = KPROBE_HIT_SS;
359 enable_singlestep(kcb, regs,
360 (unsigned long) p->ainsn.insn);
363 * No kprobe at this address and the current kprobe
364 * has no break handler (no jprobe!). The kernel just
365 * exploded, let the standard trap handler pick up the
369 * No kprobe at this address and no active kprobe. The trap has
370 * not been caused by a kprobe breakpoint. The race of breakpoint
371 * vs. kprobe remove does not exist because on s390 as we use
372 * stop_machine to arm/disarm the breakpoints.
374 preempt_enable_no_resched();
377 NOKPROBE_SYMBOL(kprobe_handler);
380 * Function return probe trampoline:
381 * - init_kprobes() establishes a probepoint here
382 * - When the probed function returns, this probe
383 * causes the handlers to fire
385 static void __used kretprobe_trampoline_holder(void)
387 asm volatile(".global kretprobe_trampoline\n"
388 "kretprobe_trampoline: bcr 0,0\n");
392 * Called when the probe at kretprobe trampoline is hit
394 static int trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
396 struct kretprobe_instance *ri;
397 struct hlist_head *head, empty_rp;
398 struct hlist_node *tmp;
399 unsigned long flags, orig_ret_address;
400 unsigned long trampoline_address;
401 kprobe_opcode_t *correct_ret_addr;
403 INIT_HLIST_HEAD(&empty_rp);
404 kretprobe_hash_lock(current, &head, &flags);
407 * It is possible to have multiple instances associated with a given
408 * task either because an multiple functions in the call path
409 * have a return probe installed on them, and/or more than one return
410 * return probe was registered for a target function.
412 * We can handle this because:
413 * - instances are always inserted at the head of the list
414 * - when multiple return probes are registered for the same
415 * function, the first instance's ret_addr will point to the
416 * real return address, and all the rest will point to
417 * kretprobe_trampoline
420 orig_ret_address = 0;
421 correct_ret_addr = NULL;
422 trampoline_address = (unsigned long) &kretprobe_trampoline;
423 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
424 if (ri->task != current)
425 /* another task is sharing our hash bucket */
428 orig_ret_address = (unsigned long) ri->ret_addr;
430 if (orig_ret_address != trampoline_address)
432 * This is the real return address. Any other
433 * instances associated with this task are for
434 * other calls deeper on the call stack
439 kretprobe_assert(ri, orig_ret_address, trampoline_address);
441 correct_ret_addr = ri->ret_addr;
442 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
443 if (ri->task != current)
444 /* another task is sharing our hash bucket */
447 orig_ret_address = (unsigned long) ri->ret_addr;
449 if (ri->rp && ri->rp->handler) {
450 ri->ret_addr = correct_ret_addr;
451 ri->rp->handler(ri, regs);
454 recycle_rp_inst(ri, &empty_rp);
456 if (orig_ret_address != trampoline_address)
458 * This is the real return address. Any other
459 * instances associated with this task are for
460 * other calls deeper on the call stack
465 regs->psw.addr = orig_ret_address;
467 pop_kprobe(get_kprobe_ctlblk());
468 kretprobe_hash_unlock(current, &flags);
469 preempt_enable_no_resched();
471 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
472 hlist_del(&ri->hlist);
476 * By returning a non-zero value, we are telling
477 * kprobe_handler() that we don't want the post_handler
478 * to run (and have re-enabled preemption)
482 NOKPROBE_SYMBOL(trampoline_probe_handler);
485 * Called after single-stepping. p->addr is the address of the
486 * instruction whose first byte has been replaced by the "breakpoint"
487 * instruction. To avoid the SMP problems that can occur when we
488 * temporarily put back the original opcode to single-step, we
489 * single-stepped a copy of the instruction. The address of this
490 * copy is p->ainsn.insn.
492 static void resume_execution(struct kprobe *p, struct pt_regs *regs)
494 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
495 unsigned long ip = regs->psw.addr;
496 int fixup = probe_get_fixup_type(p->ainsn.insn);
498 /* Check if the kprobes location is an enabled ftrace caller */
499 if (p->ainsn.is_ftrace_insn) {
500 struct ftrace_insn *insn = (struct ftrace_insn *) p->addr;
501 struct ftrace_insn call_insn;
503 ftrace_generate_call_insn(&call_insn, (unsigned long) p->addr);
505 * A kprobe on an enabled ftrace call site actually single
506 * stepped an unconditional branch (ftrace nop equivalent).
507 * Now we need to fixup things and pretend that a brasl r0,...
508 * was executed instead.
510 if (insn->disp == KPROBE_ON_FTRACE_CALL) {
511 ip += call_insn.disp * 2 - MCOUNT_INSN_SIZE;
512 regs->gprs[0] = (unsigned long)p->addr + sizeof(*insn);
516 if (fixup & FIXUP_PSW_NORMAL)
517 ip += (unsigned long) p->addr - (unsigned long) p->ainsn.insn;
519 if (fixup & FIXUP_BRANCH_NOT_TAKEN) {
520 int ilen = insn_length(p->ainsn.insn[0] >> 8);
521 if (ip - (unsigned long) p->ainsn.insn == ilen)
522 ip = (unsigned long) p->addr + ilen;
525 if (fixup & FIXUP_RETURN_REGISTER) {
526 int reg = (p->ainsn.insn[0] & 0xf0) >> 4;
527 regs->gprs[reg] += (unsigned long) p->addr -
528 (unsigned long) p->ainsn.insn;
531 disable_singlestep(kcb, regs, ip);
533 NOKPROBE_SYMBOL(resume_execution);
535 static int post_kprobe_handler(struct pt_regs *regs)
537 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
538 struct kprobe *p = kprobe_running();
543 if (kcb->kprobe_status != KPROBE_REENTER && p->post_handler) {
544 kcb->kprobe_status = KPROBE_HIT_SSDONE;
545 p->post_handler(p, regs, 0);
548 resume_execution(p, regs);
550 preempt_enable_no_resched();
553 * if somebody else is singlestepping across a probe point, psw mask
554 * will have PER set, in which case, continue the remaining processing
555 * of do_single_step, as if this is not a probe hit.
557 if (regs->psw.mask & PSW_MASK_PER)
562 NOKPROBE_SYMBOL(post_kprobe_handler);
564 static int kprobe_trap_handler(struct pt_regs *regs, int trapnr)
566 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
567 struct kprobe *p = kprobe_running();
568 const struct exception_table_entry *entry;
570 switch(kcb->kprobe_status) {
571 case KPROBE_SWAP_INST:
572 /* We are here because the instruction replacement failed */
577 * We are here because the instruction being single
578 * stepped caused a page fault. We reset the current
579 * kprobe and the nip points back to the probe address
580 * and allow the page fault handler to continue as a
583 disable_singlestep(kcb, regs, (unsigned long) p->addr);
585 preempt_enable_no_resched();
587 case KPROBE_HIT_ACTIVE:
588 case KPROBE_HIT_SSDONE:
590 * We increment the nmissed count for accounting,
591 * we can also use npre/npostfault count for accounting
592 * these specific fault cases.
594 kprobes_inc_nmissed_count(p);
597 * We come here because instructions in the pre/post
598 * handler caused the page_fault, this could happen
599 * if handler tries to access user space by
600 * copy_from_user(), get_user() etc. Let the
601 * user-specified handler try to fix it first.
603 if (p->fault_handler && p->fault_handler(p, regs, trapnr))
607 * In case the user-specified fault handler returned
608 * zero, try to fix up.
610 entry = search_exception_tables(regs->psw.addr);
612 regs->psw.addr = extable_fixup(entry);
617 * fixup_exception() could not handle it,
618 * Let do_page_fault() fix it.
626 NOKPROBE_SYMBOL(kprobe_trap_handler);
628 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
632 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
634 ret = kprobe_trap_handler(regs, trapnr);
635 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
636 local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
639 NOKPROBE_SYMBOL(kprobe_fault_handler);
642 * Wrapper routine to for handling exceptions.
644 int kprobe_exceptions_notify(struct notifier_block *self,
645 unsigned long val, void *data)
647 struct die_args *args = (struct die_args *) data;
648 struct pt_regs *regs = args->regs;
649 int ret = NOTIFY_DONE;
651 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
656 if (kprobe_handler(regs))
660 if (post_kprobe_handler(regs))
664 if (!preemptible() && kprobe_running() &&
665 kprobe_trap_handler(regs, args->trapnr))
672 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
673 local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
677 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
679 int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
681 struct jprobe *jp = container_of(p, struct jprobe, kp);
682 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
685 memcpy(&kcb->jprobe_saved_regs, regs, sizeof(struct pt_regs));
687 /* setup return addr to the jprobe handler routine */
688 regs->psw.addr = (unsigned long) jp->entry;
689 regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
691 /* r15 is the stack pointer */
692 stack = (unsigned long) regs->gprs[15];
694 memcpy(kcb->jprobes_stack, (void *) stack, MIN_STACK_SIZE(stack));
697 * jprobes use jprobe_return() which skips the normal return
698 * path of the function, and this messes up the accounting of the
699 * function graph tracer to get messed up.
701 * Pause function graph tracing while performing the jprobe function.
703 pause_graph_tracing();
706 NOKPROBE_SYMBOL(setjmp_pre_handler);
708 void jprobe_return(void)
710 asm volatile(".word 0x0002");
712 NOKPROBE_SYMBOL(jprobe_return);
714 int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
716 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
719 /* It's OK to start function graph tracing again */
720 unpause_graph_tracing();
722 stack = (unsigned long) kcb->jprobe_saved_regs.gprs[15];
724 /* Put the regs back */
725 memcpy(regs, &kcb->jprobe_saved_regs, sizeof(struct pt_regs));
726 /* put the stack back */
727 memcpy((void *) stack, kcb->jprobes_stack, MIN_STACK_SIZE(stack));
728 preempt_enable_no_resched();
731 NOKPROBE_SYMBOL(longjmp_break_handler);
733 static struct kprobe trampoline = {
734 .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
735 .pre_handler = trampoline_probe_handler
738 int __init arch_init_kprobes(void)
740 return register_kprobe(&trampoline);
743 int arch_trampoline_kprobe(struct kprobe *p)
745 return p->addr == (kprobe_opcode_t *) &kretprobe_trampoline;
747 NOKPROBE_SYMBOL(arch_trampoline_kprobe);