xref: /linux/arch/powerpc/mm/fault.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  PowerPC version
4  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
5  *
6  *  Derived from "arch/i386/mm/fault.c"
7  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
8  *
9  *  Modified by Cort Dougan and Paul Mackerras.
10  *
11  *  Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
12  */
13 
14 #include <linux/signal.h>
15 #include <linux/sched.h>
16 #include <linux/sched/task_stack.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/string.h>
20 #include <linux/types.h>
21 #include <linux/pagemap.h>
22 #include <linux/ptrace.h>
23 #include <linux/mman.h>
24 #include <linux/mm.h>
25 #include <linux/interrupt.h>
26 #include <linux/highmem.h>
27 #include <linux/extable.h>
28 #include <linux/kprobes.h>
29 #include <linux/kdebug.h>
30 #include <linux/perf_event.h>
31 #include <linux/ratelimit.h>
32 #include <linux/context_tracking.h>
33 #include <linux/hugetlb.h>
34 #include <linux/uaccess.h>
35 #include <linux/kfence.h>
36 #include <linux/pkeys.h>
37 
38 #include <asm/firmware.h>
39 #include <asm/interrupt.h>
40 #include <asm/page.h>
41 #include <asm/mmu.h>
42 #include <asm/mmu_context.h>
43 #include <asm/siginfo.h>
44 #include <asm/debug.h>
45 #include <asm/kup.h>
46 #include <asm/inst.h>
47 
48 
49 /*
50  * do_page_fault error handling helpers
51  */
52 
53 static int
54 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code)
55 {
56 	/*
57 	 * If we are in kernel mode, bail out with a SEGV, this will
58 	 * be caught by the assembly which will restore the non-volatile
59 	 * registers before calling bad_page_fault()
60 	 */
61 	if (!user_mode(regs))
62 		return SIGSEGV;
63 
64 	_exception(SIGSEGV, regs, si_code, address);
65 
66 	return 0;
67 }
68 
69 static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address)
70 {
71 	return __bad_area_nosemaphore(regs, address, SEGV_MAPERR);
72 }
73 
74 static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code,
75 		      struct mm_struct *mm, struct vm_area_struct *vma)
76 {
77 
78 	/*
79 	 * Something tried to access memory that isn't in our memory map..
80 	 * Fix it, but check if it's kernel or user first..
81 	 */
82 	if (mm)
83 		mmap_read_unlock(mm);
84 	else
85 		vma_end_read(vma);
86 
87 	return __bad_area_nosemaphore(regs, address, si_code);
88 }
89 
90 static noinline int bad_access_pkey(struct pt_regs *regs, unsigned long address,
91 				    struct mm_struct *mm,
92 				    struct vm_area_struct *vma)
93 {
94 	int pkey;
95 
96 	/*
97 	 * We don't try to fetch the pkey from page table because reading
98 	 * page table without locking doesn't guarantee stable pte value.
99 	 * Hence the pkey value that we return to userspace can be different
100 	 * from the pkey that actually caused access error.
101 	 *
102 	 * It does *not* guarantee that the VMA we find here
103 	 * was the one that we faulted on.
104 	 *
105 	 * 1. T1   : mprotect_key(foo, PAGE_SIZE, pkey=4);
106 	 * 2. T1   : set AMR to deny access to pkey=4, touches, page
107 	 * 3. T1   : faults...
108 	 * 4.    T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
109 	 * 5. T1   : enters fault handler, takes mmap_lock, etc...
110 	 * 6. T1   : reaches here, sees vma_pkey(vma)=5, when we really
111 	 *	     faulted on a pte with its pkey=4.
112 	 */
113 	pkey = vma_pkey(vma);
114 
115 	if (mm)
116 		mmap_read_unlock(mm);
117 	else
118 		vma_end_read(vma);
119 
120 	/*
121 	 * If we are in kernel mode, bail out with a SEGV, this will
122 	 * be caught by the assembly which will restore the non-volatile
123 	 * registers before calling bad_page_fault()
124 	 */
125 	if (!user_mode(regs))
126 		return SIGSEGV;
127 
128 	_exception_pkey(regs, address, pkey);
129 
130 	return 0;
131 }
132 
133 static noinline int bad_access(struct pt_regs *regs, unsigned long address,
134 			       struct mm_struct *mm, struct vm_area_struct *vma)
135 {
136 	return __bad_area(regs, address, SEGV_ACCERR, mm, vma);
137 }
138 
139 static int do_sigbus(struct pt_regs *regs, unsigned long address,
140 		     vm_fault_t fault)
141 {
142 	if (!user_mode(regs))
143 		return SIGBUS;
144 
145 	current->thread.trap_nr = BUS_ADRERR;
146 #ifdef CONFIG_MEMORY_FAILURE
147 	if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
148 		unsigned int lsb = 0; /* shutup gcc */
149 
150 		pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
151 			current->comm, current->pid, address);
152 
153 		if (fault & VM_FAULT_HWPOISON_LARGE)
154 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
155 		if (fault & VM_FAULT_HWPOISON)
156 			lsb = PAGE_SHIFT;
157 
158 		force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb);
159 		return 0;
160 	}
161 
162 #endif
163 	force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
164 	return 0;
165 }
166 
167 static int mm_fault_error(struct pt_regs *regs, unsigned long addr,
168 				vm_fault_t fault)
169 {
170 	/*
171 	 * Kernel page fault interrupted by SIGKILL. We have no reason to
172 	 * continue processing.
173 	 */
174 	if (fatal_signal_pending(current) && !user_mode(regs))
175 		return SIGKILL;
176 
177 	/* Out of memory */
178 	if (fault & VM_FAULT_OOM) {
179 		/*
180 		 * We ran out of memory, or some other thing happened to us that
181 		 * made us unable to handle the page fault gracefully.
182 		 */
183 		if (!user_mode(regs))
184 			return SIGSEGV;
185 		pagefault_out_of_memory();
186 	} else {
187 		if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
188 			     VM_FAULT_HWPOISON_LARGE))
189 			return do_sigbus(regs, addr, fault);
190 		else if (fault & VM_FAULT_SIGSEGV)
191 			return bad_area_nosemaphore(regs, addr);
192 		else
193 			BUG();
194 	}
195 	return 0;
196 }
197 
198 /* Is this a bad kernel fault ? */
199 static bool bad_kernel_fault(struct pt_regs *regs, unsigned long error_code,
200 			     unsigned long address, bool is_write)
201 {
202 	int is_exec = TRAP(regs) == INTERRUPT_INST_STORAGE;
203 
204 	if (is_exec) {
205 		pr_crit_ratelimited("kernel tried to execute %s page (%lx) - exploit attempt? (uid: %d)\n",
206 				    address >= TASK_SIZE ? "exec-protected" : "user",
207 				    address,
208 				    from_kuid(&init_user_ns, current_uid()));
209 
210 		// Kernel exec fault is always bad
211 		return true;
212 	}
213 
214 	// Kernel fault on kernel address is bad
215 	if (address >= TASK_SIZE)
216 		return true;
217 
218 	// Read/write fault blocked by KUAP is bad, it can never succeed.
219 	if (bad_kuap_fault(regs, address, is_write)) {
220 		pr_crit_ratelimited("Kernel attempted to %s user page (%lx) - exploit attempt? (uid: %d)\n",
221 				    is_write ? "write" : "read", address,
222 				    from_kuid(&init_user_ns, current_uid()));
223 
224 		// Fault on user outside of certain regions (eg. copy_tofrom_user()) is bad
225 		if (!search_exception_tables(regs->nip))
226 			return true;
227 
228 		// Read/write fault in a valid region (the exception table search passed
229 		// above), but blocked by KUAP is bad, it can never succeed.
230 		return WARN(true, "Bug: %s fault blocked by KUAP!", is_write ? "Write" : "Read");
231 	}
232 
233 	// What's left? Kernel fault on user and allowed by KUAP in the faulting context.
234 	return false;
235 }
236 
237 static bool access_pkey_error(bool is_write, bool is_exec, bool is_pkey,
238 			      struct vm_area_struct *vma)
239 {
240 	/*
241 	 * Make sure to check the VMA so that we do not perform
242 	 * faults just to hit a pkey fault as soon as we fill in a
243 	 * page. Only called for current mm, hence foreign == 0
244 	 */
245 	if (!arch_vma_access_permitted(vma, is_write, is_exec, 0))
246 		return true;
247 
248 	return false;
249 }
250 
251 static bool access_error(bool is_write, bool is_exec, struct vm_area_struct *vma)
252 {
253 	/*
254 	 * Allow execution from readable areas if the MMU does not
255 	 * provide separate controls over reading and executing.
256 	 *
257 	 * Note: That code used to not be enabled for 4xx/BookE.
258 	 * It is now as I/D cache coherency for these is done at
259 	 * set_pte_at() time and I see no reason why the test
260 	 * below wouldn't be valid on those processors. This -may-
261 	 * break programs compiled with a really old ABI though.
262 	 */
263 	if (is_exec) {
264 		return !(vma->vm_flags & VM_EXEC) &&
265 			(cpu_has_feature(CPU_FTR_NOEXECUTE) ||
266 			 !(vma->vm_flags & (VM_READ | VM_WRITE)));
267 	}
268 
269 	if (is_write) {
270 		if (unlikely(!(vma->vm_flags & VM_WRITE)))
271 			return true;
272 		return false;
273 	}
274 
275 	/*
276 	 * VM_READ, VM_WRITE and VM_EXEC may imply read permissions, as
277 	 * defined in protection_map[].  In that case Read faults can only be
278 	 * caused by a PROT_NONE mapping. However a non exec access on a
279 	 * VM_EXEC only mapping is invalid anyway, so report it as such.
280 	 */
281 	if (unlikely(!vma_is_accessible(vma)))
282 		return true;
283 
284 	if ((vma->vm_flags & VM_ACCESS_FLAGS) == VM_EXEC)
285 		return true;
286 
287 	/*
288 	 * We should ideally do the vma pkey access check here. But in the
289 	 * fault path, handle_mm_fault() also does the same check. To avoid
290 	 * these multiple checks, we skip it here and handle access error due
291 	 * to pkeys later.
292 	 */
293 	return false;
294 }
295 
296 #ifdef CONFIG_PPC_SMLPAR
297 static inline void cmo_account_page_fault(void)
298 {
299 	if (firmware_has_feature(FW_FEATURE_CMO)) {
300 		u32 page_ins;
301 
302 		preempt_disable();
303 		page_ins = be32_to_cpu(get_lppaca()->page_ins);
304 		page_ins += 1 << PAGE_FACTOR;
305 		get_lppaca()->page_ins = cpu_to_be32(page_ins);
306 		preempt_enable();
307 	}
308 }
309 #else
310 static inline void cmo_account_page_fault(void) { }
311 #endif /* CONFIG_PPC_SMLPAR */
312 
313 static void sanity_check_fault(bool is_write, bool is_user,
314 			       unsigned long error_code, unsigned long address)
315 {
316 	/*
317 	 * Userspace trying to access kernel address, we get PROTFAULT for that.
318 	 */
319 	if (is_user && address >= TASK_SIZE) {
320 		if ((long)address == -1)
321 			return;
322 
323 		pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n",
324 				   current->comm, current->pid, address,
325 				   from_kuid(&init_user_ns, current_uid()));
326 		return;
327 	}
328 
329 	if (!IS_ENABLED(CONFIG_PPC_BOOK3S))
330 		return;
331 
332 	/*
333 	 * For hash translation mode, we should never get a
334 	 * PROTFAULT. Any update to pte to reduce access will result in us
335 	 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
336 	 * fault instead of DSISR_PROTFAULT.
337 	 *
338 	 * A pte update to relax the access will not result in a hash page table
339 	 * entry invalidate and hence can result in DSISR_PROTFAULT.
340 	 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
341 	 * the special !is_write in the below conditional.
342 	 *
343 	 * For platforms that doesn't supports coherent icache and do support
344 	 * per page noexec bit, we do setup things such that we do the
345 	 * sync between D/I cache via fault. But that is handled via low level
346 	 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
347 	 * here in such case.
348 	 *
349 	 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
350 	 * check should handle those and hence we should fall to the bad_area
351 	 * handling correctly.
352 	 *
353 	 * For embedded with per page exec support that doesn't support coherent
354 	 * icache we do get PROTFAULT and we handle that D/I cache sync in
355 	 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
356 	 * is conditional for server MMU.
357 	 *
358 	 * For radix, we can get prot fault for autonuma case, because radix
359 	 * page table will have them marked noaccess for user.
360 	 */
361 	if (radix_enabled() || is_write)
362 		return;
363 
364 	WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
365 }
366 
367 /*
368  * Define the correct "is_write" bit in error_code based
369  * on the processor family
370  */
371 #ifdef CONFIG_BOOKE
372 #define page_fault_is_write(__err)	((__err) & ESR_DST)
373 #else
374 #define page_fault_is_write(__err)	((__err) & DSISR_ISSTORE)
375 #endif
376 
377 #ifdef CONFIG_BOOKE
378 #define page_fault_is_bad(__err)	(0)
379 #elif defined(CONFIG_PPC_8xx)
380 #define page_fault_is_bad(__err)	((__err) & DSISR_NOEXEC_OR_G)
381 #elif defined(CONFIG_PPC64)
382 static int page_fault_is_bad(unsigned long err)
383 {
384 	unsigned long flag = DSISR_BAD_FAULT_64S;
385 
386 	/*
387 	 * PAPR+ v2.11 § 14.15.3.4.1 (unreleased)
388 	 * If byte 0, bit 3 of pi-attribute-specifier-type in
389 	 * ibm,pi-features property is defined, ignore the DSI error
390 	 * which is caused by the paste instruction on the
391 	 * suspended NX window.
392 	 */
393 	if (mmu_has_feature(MMU_FTR_NX_DSI))
394 		flag &= ~DSISR_BAD_COPYPASTE;
395 
396 	return err & flag;
397 }
398 #else
399 #define page_fault_is_bad(__err)	((__err) & DSISR_BAD_FAULT_32S)
400 #endif
401 
402 /*
403  * For 600- and 800-family processors, the error_code parameter is DSISR
404  * for a data fault, SRR1 for an instruction fault.
405  * For 400-family processors the error_code parameter is ESR for a data fault,
406  * 0 for an instruction fault.
407  * For 64-bit processors, the error_code parameter is DSISR for a data access
408  * fault, SRR1 & 0x08000000 for an instruction access fault.
409  *
410  * The return value is 0 if the fault was handled, or the signal
411  * number if this is a kernel fault that can't be handled here.
412  */
413 static int ___do_page_fault(struct pt_regs *regs, unsigned long address,
414 			   unsigned long error_code)
415 {
416 	struct vm_area_struct * vma;
417 	struct mm_struct *mm = current->mm;
418 	unsigned int flags = FAULT_FLAG_DEFAULT;
419 	int is_exec = TRAP(regs) == INTERRUPT_INST_STORAGE;
420 	int is_user = user_mode(regs);
421 	int is_write = page_fault_is_write(error_code);
422 	vm_fault_t fault, major = 0;
423 	bool kprobe_fault = kprobe_page_fault(regs, 11);
424 
425 	if (unlikely(debugger_fault_handler(regs) || kprobe_fault))
426 		return 0;
427 
428 	if (unlikely(page_fault_is_bad(error_code))) {
429 		if (is_user) {
430 			_exception(SIGBUS, regs, BUS_OBJERR, address);
431 			return 0;
432 		}
433 		return SIGBUS;
434 	}
435 
436 	/* Additional sanity check(s) */
437 	sanity_check_fault(is_write, is_user, error_code, address);
438 
439 	/*
440 	 * The kernel should never take an execute fault nor should it
441 	 * take a page fault to a kernel address or a page fault to a user
442 	 * address outside of dedicated places.
443 	 *
444 	 * Rather than kfence directly reporting false negatives, search whether
445 	 * the NIP belongs to the fixup table for cases where fault could come
446 	 * from functions like copy_from_kernel_nofault().
447 	 */
448 	if (unlikely(!is_user && bad_kernel_fault(regs, error_code, address, is_write))) {
449 		if (is_kfence_address((void *)address) &&
450 		    !search_exception_tables(instruction_pointer(regs)) &&
451 		    kfence_handle_page_fault(address, is_write, regs))
452 			return 0;
453 
454 		return SIGSEGV;
455 	}
456 
457 	/*
458 	 * If we're in an interrupt, have no user context or are running
459 	 * in a region with pagefaults disabled then we must not take the fault
460 	 */
461 	if (unlikely(faulthandler_disabled() || !mm)) {
462 		if (is_user)
463 			printk_ratelimited(KERN_ERR "Page fault in user mode"
464 					   " with faulthandler_disabled()=%d"
465 					   " mm=%p\n",
466 					   faulthandler_disabled(), mm);
467 		return bad_area_nosemaphore(regs, address);
468 	}
469 
470 	interrupt_cond_local_irq_enable(regs);
471 
472 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
473 
474 	/*
475 	 * We want to do this outside mmap_lock, because reading code around nip
476 	 * can result in fault, which will cause a deadlock when called with
477 	 * mmap_lock held
478 	 */
479 	if (is_user)
480 		flags |= FAULT_FLAG_USER;
481 	if (is_write)
482 		flags |= FAULT_FLAG_WRITE;
483 	if (is_exec)
484 		flags |= FAULT_FLAG_INSTRUCTION;
485 
486 	if (!(flags & FAULT_FLAG_USER))
487 		goto lock_mmap;
488 
489 	vma = lock_vma_under_rcu(mm, address);
490 	if (!vma)
491 		goto lock_mmap;
492 
493 	if (unlikely(access_pkey_error(is_write, is_exec,
494 				       (error_code & DSISR_KEYFAULT), vma))) {
495 		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
496 		return bad_access_pkey(regs, address, NULL, vma);
497 	}
498 
499 	if (unlikely(access_error(is_write, is_exec, vma))) {
500 		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
501 		return bad_access(regs, address, NULL, vma);
502 	}
503 
504 	fault = handle_mm_fault(vma, address, flags | FAULT_FLAG_VMA_LOCK, regs);
505 	if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
506 		vma_end_read(vma);
507 
508 	if (!(fault & VM_FAULT_RETRY)) {
509 		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
510 		goto done;
511 	}
512 	count_vm_vma_lock_event(VMA_LOCK_RETRY);
513 	if (fault & VM_FAULT_MAJOR)
514 		flags |= FAULT_FLAG_TRIED;
515 
516 	if (fault_signal_pending(fault, regs))
517 		return user_mode(regs) ? 0 : SIGBUS;
518 
519 lock_mmap:
520 
521 	/* When running in the kernel we expect faults to occur only to
522 	 * addresses in user space.  All other faults represent errors in the
523 	 * kernel and should generate an OOPS.  Unfortunately, in the case of an
524 	 * erroneous fault occurring in a code path which already holds mmap_lock
525 	 * we will deadlock attempting to validate the fault against the
526 	 * address space.  Luckily the kernel only validly references user
527 	 * space from well defined areas of code, which are listed in the
528 	 * exceptions table. lock_mm_and_find_vma() handles that logic.
529 	 */
530 retry:
531 	vma = lock_mm_and_find_vma(mm, address, regs);
532 	if (unlikely(!vma))
533 		return bad_area_nosemaphore(regs, address);
534 
535 	if (unlikely(access_pkey_error(is_write, is_exec,
536 				       (error_code & DSISR_KEYFAULT), vma)))
537 		return bad_access_pkey(regs, address, mm, vma);
538 
539 	if (unlikely(access_error(is_write, is_exec, vma)))
540 		return bad_access(regs, address, mm, vma);
541 
542 	/*
543 	 * If for any reason at all we couldn't handle the fault,
544 	 * make sure we exit gracefully rather than endlessly redo
545 	 * the fault.
546 	 */
547 	fault = handle_mm_fault(vma, address, flags, regs);
548 
549 	major |= fault & VM_FAULT_MAJOR;
550 
551 	if (fault_signal_pending(fault, regs))
552 		return user_mode(regs) ? 0 : SIGBUS;
553 
554 	/* The fault is fully completed (including releasing mmap lock) */
555 	if (fault & VM_FAULT_COMPLETED)
556 		goto out;
557 
558 	/*
559 	 * Handle the retry right now, the mmap_lock has been released in that
560 	 * case.
561 	 */
562 	if (unlikely(fault & VM_FAULT_RETRY)) {
563 		flags |= FAULT_FLAG_TRIED;
564 		goto retry;
565 	}
566 
567 	mmap_read_unlock(current->mm);
568 
569 done:
570 	if (unlikely(fault & VM_FAULT_ERROR))
571 		return mm_fault_error(regs, address, fault);
572 
573 out:
574 	/*
575 	 * Major/minor page fault accounting.
576 	 */
577 	if (major)
578 		cmo_account_page_fault();
579 
580 	return 0;
581 }
582 NOKPROBE_SYMBOL(___do_page_fault);
583 
584 static __always_inline void __do_page_fault(struct pt_regs *regs)
585 {
586 	long err;
587 
588 	err = ___do_page_fault(regs, regs->dar, regs->dsisr);
589 	if (unlikely(err))
590 		bad_page_fault(regs, err);
591 }
592 
593 DEFINE_INTERRUPT_HANDLER(do_page_fault)
594 {
595 	__do_page_fault(regs);
596 }
597 
598 #ifdef CONFIG_PPC_BOOK3S_64
599 /* Same as do_page_fault but interrupt entry has already run in do_hash_fault */
600 void hash__do_page_fault(struct pt_regs *regs)
601 {
602 	__do_page_fault(regs);
603 }
604 NOKPROBE_SYMBOL(hash__do_page_fault);
605 #endif
606 
607 /*
608  * bad_page_fault is called when we have a bad access from the kernel.
609  * It is called from the DSI and ISI handlers in head.S and from some
610  * of the procedures in traps.c.
611  */
612 static void __bad_page_fault(struct pt_regs *regs, int sig)
613 {
614 	int is_write = page_fault_is_write(regs->dsisr);
615 	const char *msg;
616 
617 	/* kernel has accessed a bad area */
618 
619 	if (regs->dar < PAGE_SIZE)
620 		msg = "Kernel NULL pointer dereference";
621 	else
622 		msg = "Unable to handle kernel data access";
623 
624 	switch (TRAP(regs)) {
625 	case INTERRUPT_DATA_STORAGE:
626 	case INTERRUPT_H_DATA_STORAGE:
627 		pr_alert("BUG: %s on %s at 0x%08lx\n", msg,
628 			 is_write ? "write" : "read", regs->dar);
629 		break;
630 	case INTERRUPT_DATA_SEGMENT:
631 		pr_alert("BUG: %s at 0x%08lx\n", msg, regs->dar);
632 		break;
633 	case INTERRUPT_INST_STORAGE:
634 	case INTERRUPT_INST_SEGMENT:
635 		pr_alert("BUG: Unable to handle kernel instruction fetch%s",
636 			 regs->nip < PAGE_SIZE ? " (NULL pointer?)\n" : "\n");
637 		break;
638 	case INTERRUPT_ALIGNMENT:
639 		pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n",
640 			 regs->dar);
641 		break;
642 	default:
643 		pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n",
644 			 regs->dar);
645 		break;
646 	}
647 	printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
648 		regs->nip);
649 
650 	if (task_stack_end_corrupted(current))
651 		printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
652 
653 	die("Kernel access of bad area", regs, sig);
654 }
655 
656 void bad_page_fault(struct pt_regs *regs, int sig)
657 {
658 	const struct exception_table_entry *entry;
659 
660 	/* Are we prepared to handle this fault?  */
661 	entry = search_exception_tables(instruction_pointer(regs));
662 	if (entry)
663 		instruction_pointer_set(regs, extable_fixup(entry));
664 	else
665 		__bad_page_fault(regs, sig);
666 }
667 
668 #ifdef CONFIG_PPC_BOOK3S_64
669 DEFINE_INTERRUPT_HANDLER(do_bad_page_fault_segv)
670 {
671 	bad_page_fault(regs, SIGSEGV);
672 }
673 
674 /*
675  * In radix, segment interrupts indicate the EA is not addressable by the
676  * page table geometry, so they are always sent here.
677  *
678  * In hash, this is called if do_slb_fault returns error. Typically it is
679  * because the EA was outside the region allowed by software.
680  */
681 DEFINE_INTERRUPT_HANDLER(do_bad_segment_interrupt)
682 {
683 	int err = regs->result;
684 
685 	if (err == -EFAULT) {
686 		if (user_mode(regs))
687 			_exception(SIGSEGV, regs, SEGV_BNDERR, regs->dar);
688 		else
689 			bad_page_fault(regs, SIGSEGV);
690 	} else if (err == -EINVAL) {
691 		unrecoverable_exception(regs);
692 	} else {
693 		BUG();
694 	}
695 }
696 #endif
697