xref: /openbmc/linux/arch/powerpc/kernel/kprobes.c (revision 82003e04)
1 /*
2  *  Kernel Probes (KProbes)
3  *
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.
8  *
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.
13  *
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.
17  *
18  * Copyright (C) IBM Corporation, 2002, 2004
19  *
20  * 2002-Oct	Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
21  *		Probes initial implementation ( includes contributions from
22  *		Rusty Russell).
23  * 2004-July	Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
24  *		interface to access function arguments.
25  * 2004-Nov	Ananth N Mavinakayanahalli <ananth@in.ibm.com> kprobes port
26  *		for PPC64
27  */
28 
29 #include <linux/kprobes.h>
30 #include <linux/ptrace.h>
31 #include <linux/preempt.h>
32 #include <linux/extable.h>
33 #include <linux/kdebug.h>
34 #include <linux/slab.h>
35 #include <asm/code-patching.h>
36 #include <asm/cacheflush.h>
37 #include <asm/sstep.h>
38 #include <asm/uaccess.h>
39 
40 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
41 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
42 
43 struct kretprobe_blackpoint kretprobe_blacklist[] = {{NULL, NULL}};
44 
45 int __kprobes arch_prepare_kprobe(struct kprobe *p)
46 {
47 	int ret = 0;
48 	kprobe_opcode_t insn = *p->addr;
49 
50 	if ((unsigned long)p->addr & 0x03) {
51 		printk("Attempt to register kprobe at an unaligned address\n");
52 		ret = -EINVAL;
53 	} else if (IS_MTMSRD(insn) || IS_RFID(insn) || IS_RFI(insn)) {
54 		printk("Cannot register a kprobe on rfi/rfid or mtmsr[d]\n");
55 		ret = -EINVAL;
56 	}
57 
58 	/* insn must be on a special executable page on ppc64.  This is
59 	 * not explicitly required on ppc32 (right now), but it doesn't hurt */
60 	if (!ret) {
61 		p->ainsn.insn = get_insn_slot();
62 		if (!p->ainsn.insn)
63 			ret = -ENOMEM;
64 	}
65 
66 	if (!ret) {
67 		memcpy(p->ainsn.insn, p->addr,
68 				MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
69 		p->opcode = *p->addr;
70 		flush_icache_range((unsigned long)p->ainsn.insn,
71 			(unsigned long)p->ainsn.insn + sizeof(kprobe_opcode_t));
72 	}
73 
74 	p->ainsn.boostable = 0;
75 	return ret;
76 }
77 
78 void __kprobes arch_arm_kprobe(struct kprobe *p)
79 {
80 	*p->addr = BREAKPOINT_INSTRUCTION;
81 	flush_icache_range((unsigned long) p->addr,
82 			   (unsigned long) p->addr + sizeof(kprobe_opcode_t));
83 }
84 
85 void __kprobes arch_disarm_kprobe(struct kprobe *p)
86 {
87 	*p->addr = p->opcode;
88 	flush_icache_range((unsigned long) p->addr,
89 			   (unsigned long) p->addr + sizeof(kprobe_opcode_t));
90 }
91 
92 void __kprobes arch_remove_kprobe(struct kprobe *p)
93 {
94 	if (p->ainsn.insn) {
95 		free_insn_slot(p->ainsn.insn, 0);
96 		p->ainsn.insn = NULL;
97 	}
98 }
99 
100 static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
101 {
102 	enable_single_step(regs);
103 
104 	/*
105 	 * On powerpc we should single step on the original
106 	 * instruction even if the probed insn is a trap
107 	 * variant as values in regs could play a part in
108 	 * if the trap is taken or not
109 	 */
110 	regs->nip = (unsigned long)p->ainsn.insn;
111 }
112 
113 static void __kprobes save_previous_kprobe(struct kprobe_ctlblk *kcb)
114 {
115 	kcb->prev_kprobe.kp = kprobe_running();
116 	kcb->prev_kprobe.status = kcb->kprobe_status;
117 	kcb->prev_kprobe.saved_msr = kcb->kprobe_saved_msr;
118 }
119 
120 static void __kprobes restore_previous_kprobe(struct kprobe_ctlblk *kcb)
121 {
122 	__this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
123 	kcb->kprobe_status = kcb->prev_kprobe.status;
124 	kcb->kprobe_saved_msr = kcb->prev_kprobe.saved_msr;
125 }
126 
127 static void __kprobes set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
128 				struct kprobe_ctlblk *kcb)
129 {
130 	__this_cpu_write(current_kprobe, p);
131 	kcb->kprobe_saved_msr = regs->msr;
132 }
133 
134 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
135 				      struct pt_regs *regs)
136 {
137 	ri->ret_addr = (kprobe_opcode_t *)regs->link;
138 
139 	/* Replace the return addr with trampoline addr */
140 	regs->link = (unsigned long)kretprobe_trampoline;
141 }
142 
143 static int __kprobes kprobe_handler(struct pt_regs *regs)
144 {
145 	struct kprobe *p;
146 	int ret = 0;
147 	unsigned int *addr = (unsigned int *)regs->nip;
148 	struct kprobe_ctlblk *kcb;
149 
150 	/*
151 	 * We don't want to be preempted for the entire
152 	 * duration of kprobe processing
153 	 */
154 	preempt_disable();
155 	kcb = get_kprobe_ctlblk();
156 
157 	/* Check we're not actually recursing */
158 	if (kprobe_running()) {
159 		p = get_kprobe(addr);
160 		if (p) {
161 			kprobe_opcode_t insn = *p->ainsn.insn;
162 			if (kcb->kprobe_status == KPROBE_HIT_SS &&
163 					is_trap(insn)) {
164 				/* Turn off 'trace' bits */
165 				regs->msr &= ~MSR_SINGLESTEP;
166 				regs->msr |= kcb->kprobe_saved_msr;
167 				goto no_kprobe;
168 			}
169 			/* We have reentered the kprobe_handler(), since
170 			 * another probe was hit while within the handler.
171 			 * We here save the original kprobes variables and
172 			 * just single step on the instruction of the new probe
173 			 * without calling any user handlers.
174 			 */
175 			save_previous_kprobe(kcb);
176 			set_current_kprobe(p, regs, kcb);
177 			kcb->kprobe_saved_msr = regs->msr;
178 			kprobes_inc_nmissed_count(p);
179 			prepare_singlestep(p, regs);
180 			kcb->kprobe_status = KPROBE_REENTER;
181 			return 1;
182 		} else {
183 			if (*addr != BREAKPOINT_INSTRUCTION) {
184 				/* If trap variant, then it belongs not to us */
185 				kprobe_opcode_t cur_insn = *addr;
186 				if (is_trap(cur_insn))
187 		       			goto no_kprobe;
188 				/* The breakpoint instruction was removed by
189 				 * another cpu right after we hit, no further
190 				 * handling of this interrupt is appropriate
191 				 */
192 				ret = 1;
193 				goto no_kprobe;
194 			}
195 			p = __this_cpu_read(current_kprobe);
196 			if (p->break_handler && p->break_handler(p, regs)) {
197 				goto ss_probe;
198 			}
199 		}
200 		goto no_kprobe;
201 	}
202 
203 	p = get_kprobe(addr);
204 	if (!p) {
205 		if (*addr != BREAKPOINT_INSTRUCTION) {
206 			/*
207 			 * PowerPC has multiple variants of the "trap"
208 			 * instruction. If the current instruction is a
209 			 * trap variant, it could belong to someone else
210 			 */
211 			kprobe_opcode_t cur_insn = *addr;
212 			if (is_trap(cur_insn))
213 		       		goto no_kprobe;
214 			/*
215 			 * The breakpoint instruction was removed right
216 			 * after we hit it.  Another cpu has removed
217 			 * either a probepoint or a debugger breakpoint
218 			 * at this address.  In either case, no further
219 			 * handling of this interrupt is appropriate.
220 			 */
221 			ret = 1;
222 		}
223 		/* Not one of ours: let kernel handle it */
224 		goto no_kprobe;
225 	}
226 
227 	kcb->kprobe_status = KPROBE_HIT_ACTIVE;
228 	set_current_kprobe(p, regs, kcb);
229 	if (p->pre_handler && p->pre_handler(p, regs))
230 		/* handler has already set things up, so skip ss setup */
231 		return 1;
232 
233 ss_probe:
234 	if (p->ainsn.boostable >= 0) {
235 		unsigned int insn = *p->ainsn.insn;
236 
237 		/* regs->nip is also adjusted if emulate_step returns 1 */
238 		ret = emulate_step(regs, insn);
239 		if (ret > 0) {
240 			/*
241 			 * Once this instruction has been boosted
242 			 * successfully, set the boostable flag
243 			 */
244 			if (unlikely(p->ainsn.boostable == 0))
245 				p->ainsn.boostable = 1;
246 
247 			if (p->post_handler)
248 				p->post_handler(p, regs, 0);
249 
250 			kcb->kprobe_status = KPROBE_HIT_SSDONE;
251 			reset_current_kprobe();
252 			preempt_enable_no_resched();
253 			return 1;
254 		} else if (ret < 0) {
255 			/*
256 			 * We don't allow kprobes on mtmsr(d)/rfi(d), etc.
257 			 * So, we should never get here... but, its still
258 			 * good to catch them, just in case...
259 			 */
260 			printk("Can't step on instruction %x\n", insn);
261 			BUG();
262 		} else if (ret == 0)
263 			/* This instruction can't be boosted */
264 			p->ainsn.boostable = -1;
265 	}
266 	prepare_singlestep(p, regs);
267 	kcb->kprobe_status = KPROBE_HIT_SS;
268 	return 1;
269 
270 no_kprobe:
271 	preempt_enable_no_resched();
272 	return ret;
273 }
274 
275 /*
276  * Function return probe trampoline:
277  * 	- init_kprobes() establishes a probepoint here
278  * 	- When the probed function returns, this probe
279  * 		causes the handlers to fire
280  */
281 asm(".global kretprobe_trampoline\n"
282 	".type kretprobe_trampoline, @function\n"
283 	"kretprobe_trampoline:\n"
284 	"nop\n"
285 	".size kretprobe_trampoline, .-kretprobe_trampoline\n");
286 
287 /*
288  * Called when the probe at kretprobe trampoline is hit
289  */
290 static int __kprobes trampoline_probe_handler(struct kprobe *p,
291 						struct pt_regs *regs)
292 {
293 	struct kretprobe_instance *ri = NULL;
294 	struct hlist_head *head, empty_rp;
295 	struct hlist_node *tmp;
296 	unsigned long flags, orig_ret_address = 0;
297 	unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
298 
299 	INIT_HLIST_HEAD(&empty_rp);
300 	kretprobe_hash_lock(current, &head, &flags);
301 
302 	/*
303 	 * It is possible to have multiple instances associated with a given
304 	 * task either because an multiple functions in the call path
305 	 * have a return probe installed on them, and/or more than one return
306 	 * return probe was registered for a target function.
307 	 *
308 	 * We can handle this because:
309 	 *     - instances are always inserted at the head of the list
310 	 *     - when multiple return probes are registered for the same
311 	 *       function, the first instance's ret_addr will point to the
312 	 *       real return address, and all the rest will point to
313 	 *       kretprobe_trampoline
314 	 */
315 	hlist_for_each_entry_safe(ri, tmp, head, hlist) {
316 		if (ri->task != current)
317 			/* another task is sharing our hash bucket */
318 			continue;
319 
320 		if (ri->rp && ri->rp->handler)
321 			ri->rp->handler(ri, regs);
322 
323 		orig_ret_address = (unsigned long)ri->ret_addr;
324 		recycle_rp_inst(ri, &empty_rp);
325 
326 		if (orig_ret_address != trampoline_address)
327 			/*
328 			 * This is the real return address. Any other
329 			 * instances associated with this task are for
330 			 * other calls deeper on the call stack
331 			 */
332 			break;
333 	}
334 
335 	kretprobe_assert(ri, orig_ret_address, trampoline_address);
336 	regs->nip = orig_ret_address;
337 
338 	reset_current_kprobe();
339 	kretprobe_hash_unlock(current, &flags);
340 	preempt_enable_no_resched();
341 
342 	hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
343 		hlist_del(&ri->hlist);
344 		kfree(ri);
345 	}
346 	/*
347 	 * By returning a non-zero value, we are telling
348 	 * kprobe_handler() that we don't want the post_handler
349 	 * to run (and have re-enabled preemption)
350 	 */
351 	return 1;
352 }
353 
354 /*
355  * Called after single-stepping.  p->addr is the address of the
356  * instruction whose first byte has been replaced by the "breakpoint"
357  * instruction.  To avoid the SMP problems that can occur when we
358  * temporarily put back the original opcode to single-step, we
359  * single-stepped a copy of the instruction.  The address of this
360  * copy is p->ainsn.insn.
361  */
362 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
363 {
364 	struct kprobe *cur = kprobe_running();
365 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
366 
367 	if (!cur)
368 		return 0;
369 
370 	/* make sure we got here for instruction we have a kprobe on */
371 	if (((unsigned long)cur->ainsn.insn + 4) != regs->nip)
372 		return 0;
373 
374 	if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
375 		kcb->kprobe_status = KPROBE_HIT_SSDONE;
376 		cur->post_handler(cur, regs, 0);
377 	}
378 
379 	/* Adjust nip to after the single-stepped instruction */
380 	regs->nip = (unsigned long)cur->addr + 4;
381 	regs->msr |= kcb->kprobe_saved_msr;
382 
383 	/*Restore back the original saved kprobes variables and continue. */
384 	if (kcb->kprobe_status == KPROBE_REENTER) {
385 		restore_previous_kprobe(kcb);
386 		goto out;
387 	}
388 	reset_current_kprobe();
389 out:
390 	preempt_enable_no_resched();
391 
392 	/*
393 	 * if somebody else is singlestepping across a probe point, msr
394 	 * will have DE/SE set, in which case, continue the remaining processing
395 	 * of do_debug, as if this is not a probe hit.
396 	 */
397 	if (regs->msr & MSR_SINGLESTEP)
398 		return 0;
399 
400 	return 1;
401 }
402 
403 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
404 {
405 	struct kprobe *cur = kprobe_running();
406 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
407 	const struct exception_table_entry *entry;
408 
409 	switch(kcb->kprobe_status) {
410 	case KPROBE_HIT_SS:
411 	case KPROBE_REENTER:
412 		/*
413 		 * We are here because the instruction being single
414 		 * stepped caused a page fault. We reset the current
415 		 * kprobe and the nip points back to the probe address
416 		 * and allow the page fault handler to continue as a
417 		 * normal page fault.
418 		 */
419 		regs->nip = (unsigned long)cur->addr;
420 		regs->msr &= ~MSR_SINGLESTEP; /* Turn off 'trace' bits */
421 		regs->msr |= kcb->kprobe_saved_msr;
422 		if (kcb->kprobe_status == KPROBE_REENTER)
423 			restore_previous_kprobe(kcb);
424 		else
425 			reset_current_kprobe();
426 		preempt_enable_no_resched();
427 		break;
428 	case KPROBE_HIT_ACTIVE:
429 	case KPROBE_HIT_SSDONE:
430 		/*
431 		 * We increment the nmissed count for accounting,
432 		 * we can also use npre/npostfault count for accounting
433 		 * these specific fault cases.
434 		 */
435 		kprobes_inc_nmissed_count(cur);
436 
437 		/*
438 		 * We come here because instructions in the pre/post
439 		 * handler caused the page_fault, this could happen
440 		 * if handler tries to access user space by
441 		 * copy_from_user(), get_user() etc. Let the
442 		 * user-specified handler try to fix it first.
443 		 */
444 		if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
445 			return 1;
446 
447 		/*
448 		 * In case the user-specified fault handler returned
449 		 * zero, try to fix up.
450 		 */
451 		if ((entry = search_exception_tables(regs->nip)) != NULL) {
452 			regs->nip = entry->fixup;
453 			return 1;
454 		}
455 
456 		/*
457 		 * fixup_exception() could not handle it,
458 		 * Let do_page_fault() fix it.
459 		 */
460 		break;
461 	default:
462 		break;
463 	}
464 	return 0;
465 }
466 
467 /*
468  * Wrapper routine to for handling exceptions.
469  */
470 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
471 				       unsigned long val, void *data)
472 {
473 	struct die_args *args = (struct die_args *)data;
474 	int ret = NOTIFY_DONE;
475 
476 	if (args->regs && user_mode(args->regs))
477 		return ret;
478 
479 	switch (val) {
480 	case DIE_BPT:
481 		if (kprobe_handler(args->regs))
482 			ret = NOTIFY_STOP;
483 		break;
484 	case DIE_SSTEP:
485 		if (post_kprobe_handler(args->regs))
486 			ret = NOTIFY_STOP;
487 		break;
488 	default:
489 		break;
490 	}
491 	return ret;
492 }
493 
494 unsigned long arch_deref_entry_point(void *entry)
495 {
496 	return ppc_global_function_entry(entry);
497 }
498 
499 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
500 {
501 	struct jprobe *jp = container_of(p, struct jprobe, kp);
502 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
503 
504 	memcpy(&kcb->jprobe_saved_regs, regs, sizeof(struct pt_regs));
505 
506 	/* setup return addr to the jprobe handler routine */
507 	regs->nip = arch_deref_entry_point(jp->entry);
508 #ifdef PPC64_ELF_ABI_v2
509 	regs->gpr[12] = (unsigned long)jp->entry;
510 #elif defined(PPC64_ELF_ABI_v1)
511 	regs->gpr[2] = (unsigned long)(((func_descr_t *)jp->entry)->toc);
512 #endif
513 
514 	return 1;
515 }
516 
517 void __used __kprobes jprobe_return(void)
518 {
519 	asm volatile("trap" ::: "memory");
520 }
521 
522 static void __used __kprobes jprobe_return_end(void)
523 {
524 };
525 
526 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
527 {
528 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
529 
530 	/*
531 	 * FIXME - we should ideally be validating that we got here 'cos
532 	 * of the "trap" in jprobe_return() above, before restoring the
533 	 * saved regs...
534 	 */
535 	memcpy(regs, &kcb->jprobe_saved_regs, sizeof(struct pt_regs));
536 	preempt_enable_no_resched();
537 	return 1;
538 }
539 
540 static struct kprobe trampoline_p = {
541 	.addr = (kprobe_opcode_t *) &kretprobe_trampoline,
542 	.pre_handler = trampoline_probe_handler
543 };
544 
545 int __init arch_init_kprobes(void)
546 {
547 	return register_kprobe(&trampoline_p);
548 }
549 
550 int __kprobes arch_trampoline_kprobe(struct kprobe *p)
551 {
552 	if (p->addr == (kprobe_opcode_t *)&kretprobe_trampoline)
553 		return 1;
554 
555 	return 0;
556 }
557