xref: /openbmc/linux/arch/powerpc/mm/fault.c (revision 4089eef0)
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
__bad_area_nosemaphore(struct pt_regs * regs,unsigned long address,int si_code)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 
bad_area_nosemaphore(struct pt_regs * regs,unsigned long address)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 
__bad_area(struct pt_regs * regs,unsigned long address,int si_code)74 static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code)
75 {
76 	struct mm_struct *mm = current->mm;
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 	mmap_read_unlock(mm);
83 
84 	return __bad_area_nosemaphore(regs, address, si_code);
85 }
86 
bad_access_pkey(struct pt_regs * regs,unsigned long address,struct vm_area_struct * vma)87 static noinline int bad_access_pkey(struct pt_regs *regs, unsigned long address,
88 				    struct vm_area_struct *vma)
89 {
90 	struct mm_struct *mm = current->mm;
91 	int pkey;
92 
93 	/*
94 	 * We don't try to fetch the pkey from page table because reading
95 	 * page table without locking doesn't guarantee stable pte value.
96 	 * Hence the pkey value that we return to userspace can be different
97 	 * from the pkey that actually caused access error.
98 	 *
99 	 * It does *not* guarantee that the VMA we find here
100 	 * was the one that we faulted on.
101 	 *
102 	 * 1. T1   : mprotect_key(foo, PAGE_SIZE, pkey=4);
103 	 * 2. T1   : set AMR to deny access to pkey=4, touches, page
104 	 * 3. T1   : faults...
105 	 * 4.    T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
106 	 * 5. T1   : enters fault handler, takes mmap_lock, etc...
107 	 * 6. T1   : reaches here, sees vma_pkey(vma)=5, when we really
108 	 *	     faulted on a pte with its pkey=4.
109 	 */
110 	pkey = vma_pkey(vma);
111 
112 	mmap_read_unlock(mm);
113 
114 	/*
115 	 * If we are in kernel mode, bail out with a SEGV, this will
116 	 * be caught by the assembly which will restore the non-volatile
117 	 * registers before calling bad_page_fault()
118 	 */
119 	if (!user_mode(regs))
120 		return SIGSEGV;
121 
122 	_exception_pkey(regs, address, pkey);
123 
124 	return 0;
125 }
126 
bad_access(struct pt_regs * regs,unsigned long address)127 static noinline int bad_access(struct pt_regs *regs, unsigned long address)
128 {
129 	return __bad_area(regs, address, SEGV_ACCERR);
130 }
131 
do_sigbus(struct pt_regs * regs,unsigned long address,vm_fault_t fault)132 static int do_sigbus(struct pt_regs *regs, unsigned long address,
133 		     vm_fault_t fault)
134 {
135 	if (!user_mode(regs))
136 		return SIGBUS;
137 
138 	current->thread.trap_nr = BUS_ADRERR;
139 #ifdef CONFIG_MEMORY_FAILURE
140 	if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
141 		unsigned int lsb = 0; /* shutup gcc */
142 
143 		pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
144 			current->comm, current->pid, address);
145 
146 		if (fault & VM_FAULT_HWPOISON_LARGE)
147 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
148 		if (fault & VM_FAULT_HWPOISON)
149 			lsb = PAGE_SHIFT;
150 
151 		force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb);
152 		return 0;
153 	}
154 
155 #endif
156 	force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
157 	return 0;
158 }
159 
mm_fault_error(struct pt_regs * regs,unsigned long addr,vm_fault_t fault)160 static int mm_fault_error(struct pt_regs *regs, unsigned long addr,
161 				vm_fault_t fault)
162 {
163 	/*
164 	 * Kernel page fault interrupted by SIGKILL. We have no reason to
165 	 * continue processing.
166 	 */
167 	if (fatal_signal_pending(current) && !user_mode(regs))
168 		return SIGKILL;
169 
170 	/* Out of memory */
171 	if (fault & VM_FAULT_OOM) {
172 		/*
173 		 * We ran out of memory, or some other thing happened to us that
174 		 * made us unable to handle the page fault gracefully.
175 		 */
176 		if (!user_mode(regs))
177 			return SIGSEGV;
178 		pagefault_out_of_memory();
179 	} else {
180 		if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
181 			     VM_FAULT_HWPOISON_LARGE))
182 			return do_sigbus(regs, addr, fault);
183 		else if (fault & VM_FAULT_SIGSEGV)
184 			return bad_area_nosemaphore(regs, addr);
185 		else
186 			BUG();
187 	}
188 	return 0;
189 }
190 
191 /* Is this a bad kernel fault ? */
bad_kernel_fault(struct pt_regs * regs,unsigned long error_code,unsigned long address,bool is_write)192 static bool bad_kernel_fault(struct pt_regs *regs, unsigned long error_code,
193 			     unsigned long address, bool is_write)
194 {
195 	int is_exec = TRAP(regs) == INTERRUPT_INST_STORAGE;
196 
197 	if (is_exec) {
198 		pr_crit_ratelimited("kernel tried to execute %s page (%lx) - exploit attempt? (uid: %d)\n",
199 				    address >= TASK_SIZE ? "exec-protected" : "user",
200 				    address,
201 				    from_kuid(&init_user_ns, current_uid()));
202 
203 		// Kernel exec fault is always bad
204 		return true;
205 	}
206 
207 	// Kernel fault on kernel address is bad
208 	if (address >= TASK_SIZE)
209 		return true;
210 
211 	// Read/write fault blocked by KUAP is bad, it can never succeed.
212 	if (bad_kuap_fault(regs, address, is_write)) {
213 		pr_crit_ratelimited("Kernel attempted to %s user page (%lx) - exploit attempt? (uid: %d)\n",
214 				    is_write ? "write" : "read", address,
215 				    from_kuid(&init_user_ns, current_uid()));
216 
217 		// Fault on user outside of certain regions (eg. copy_tofrom_user()) is bad
218 		if (!search_exception_tables(regs->nip))
219 			return true;
220 
221 		// Read/write fault in a valid region (the exception table search passed
222 		// above), but blocked by KUAP is bad, it can never succeed.
223 		return WARN(true, "Bug: %s fault blocked by KUAP!", is_write ? "Write" : "Read");
224 	}
225 
226 	// What's left? Kernel fault on user and allowed by KUAP in the faulting context.
227 	return false;
228 }
229 
access_pkey_error(bool is_write,bool is_exec,bool is_pkey,struct vm_area_struct * vma)230 static bool access_pkey_error(bool is_write, bool is_exec, bool is_pkey,
231 			      struct vm_area_struct *vma)
232 {
233 	/*
234 	 * Make sure to check the VMA so that we do not perform
235 	 * faults just to hit a pkey fault as soon as we fill in a
236 	 * page. Only called for current mm, hence foreign == 0
237 	 */
238 	if (!arch_vma_access_permitted(vma, is_write, is_exec, 0))
239 		return true;
240 
241 	return false;
242 }
243 
access_error(bool is_write,bool is_exec,struct vm_area_struct * vma)244 static bool access_error(bool is_write, bool is_exec, struct vm_area_struct *vma)
245 {
246 	/*
247 	 * Allow execution from readable areas if the MMU does not
248 	 * provide separate controls over reading and executing.
249 	 *
250 	 * Note: That code used to not be enabled for 4xx/BookE.
251 	 * It is now as I/D cache coherency for these is done at
252 	 * set_pte_at() time and I see no reason why the test
253 	 * below wouldn't be valid on those processors. This -may-
254 	 * break programs compiled with a really old ABI though.
255 	 */
256 	if (is_exec) {
257 		return !(vma->vm_flags & VM_EXEC) &&
258 			(cpu_has_feature(CPU_FTR_NOEXECUTE) ||
259 			 !(vma->vm_flags & (VM_READ | VM_WRITE)));
260 	}
261 
262 	if (is_write) {
263 		if (unlikely(!(vma->vm_flags & VM_WRITE)))
264 			return true;
265 		return false;
266 	}
267 
268 	/*
269 	 * VM_READ, VM_WRITE and VM_EXEC all imply read permissions, as
270 	 * defined in protection_map[].  Read faults can only be caused by
271 	 * a PROT_NONE mapping, or with a PROT_EXEC-only mapping on Radix.
272 	 */
273 	if (unlikely(!vma_is_accessible(vma)))
274 		return true;
275 
276 	if (unlikely(radix_enabled() && ((vma->vm_flags & VM_ACCESS_FLAGS) == VM_EXEC)))
277 		return true;
278 
279 	/*
280 	 * We should ideally do the vma pkey access check here. But in the
281 	 * fault path, handle_mm_fault() also does the same check. To avoid
282 	 * these multiple checks, we skip it here and handle access error due
283 	 * to pkeys later.
284 	 */
285 	return false;
286 }
287 
288 #ifdef CONFIG_PPC_SMLPAR
cmo_account_page_fault(void)289 static inline void cmo_account_page_fault(void)
290 {
291 	if (firmware_has_feature(FW_FEATURE_CMO)) {
292 		u32 page_ins;
293 
294 		preempt_disable();
295 		page_ins = be32_to_cpu(get_lppaca()->page_ins);
296 		page_ins += 1 << PAGE_FACTOR;
297 		get_lppaca()->page_ins = cpu_to_be32(page_ins);
298 		preempt_enable();
299 	}
300 }
301 #else
cmo_account_page_fault(void)302 static inline void cmo_account_page_fault(void) { }
303 #endif /* CONFIG_PPC_SMLPAR */
304 
sanity_check_fault(bool is_write,bool is_user,unsigned long error_code,unsigned long address)305 static void sanity_check_fault(bool is_write, bool is_user,
306 			       unsigned long error_code, unsigned long address)
307 {
308 	/*
309 	 * Userspace trying to access kernel address, we get PROTFAULT for that.
310 	 */
311 	if (is_user && address >= TASK_SIZE) {
312 		if ((long)address == -1)
313 			return;
314 
315 		pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n",
316 				   current->comm, current->pid, address,
317 				   from_kuid(&init_user_ns, current_uid()));
318 		return;
319 	}
320 
321 	if (!IS_ENABLED(CONFIG_PPC_BOOK3S))
322 		return;
323 
324 	/*
325 	 * For hash translation mode, we should never get a
326 	 * PROTFAULT. Any update to pte to reduce access will result in us
327 	 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
328 	 * fault instead of DSISR_PROTFAULT.
329 	 *
330 	 * A pte update to relax the access will not result in a hash page table
331 	 * entry invalidate and hence can result in DSISR_PROTFAULT.
332 	 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
333 	 * the special !is_write in the below conditional.
334 	 *
335 	 * For platforms that doesn't supports coherent icache and do support
336 	 * per page noexec bit, we do setup things such that we do the
337 	 * sync between D/I cache via fault. But that is handled via low level
338 	 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
339 	 * here in such case.
340 	 *
341 	 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
342 	 * check should handle those and hence we should fall to the bad_area
343 	 * handling correctly.
344 	 *
345 	 * For embedded with per page exec support that doesn't support coherent
346 	 * icache we do get PROTFAULT and we handle that D/I cache sync in
347 	 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
348 	 * is conditional for server MMU.
349 	 *
350 	 * For radix, we can get prot fault for autonuma case, because radix
351 	 * page table will have them marked noaccess for user.
352 	 */
353 	if (radix_enabled() || is_write)
354 		return;
355 
356 	WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
357 }
358 
359 /*
360  * Define the correct "is_write" bit in error_code based
361  * on the processor family
362  */
363 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
364 #define page_fault_is_write(__err)	((__err) & ESR_DST)
365 #else
366 #define page_fault_is_write(__err)	((__err) & DSISR_ISSTORE)
367 #endif
368 
369 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
370 #define page_fault_is_bad(__err)	(0)
371 #elif defined(CONFIG_PPC_8xx)
372 #define page_fault_is_bad(__err)	((__err) & DSISR_NOEXEC_OR_G)
373 #elif defined(CONFIG_PPC64)
page_fault_is_bad(unsigned long err)374 static int page_fault_is_bad(unsigned long err)
375 {
376 	unsigned long flag = DSISR_BAD_FAULT_64S;
377 
378 	/*
379 	 * PAPR+ v2.11 § 14.15.3.4.1 (unreleased)
380 	 * If byte 0, bit 3 of pi-attribute-specifier-type in
381 	 * ibm,pi-features property is defined, ignore the DSI error
382 	 * which is caused by the paste instruction on the
383 	 * suspended NX window.
384 	 */
385 	if (mmu_has_feature(MMU_FTR_NX_DSI))
386 		flag &= ~DSISR_BAD_COPYPASTE;
387 
388 	return err & flag;
389 }
390 #else
391 #define page_fault_is_bad(__err)	((__err) & DSISR_BAD_FAULT_32S)
392 #endif
393 
394 /*
395  * For 600- and 800-family processors, the error_code parameter is DSISR
396  * for a data fault, SRR1 for an instruction fault.
397  * For 400-family processors the error_code parameter is ESR for a data fault,
398  * 0 for an instruction fault.
399  * For 64-bit processors, the error_code parameter is DSISR for a data access
400  * fault, SRR1 & 0x08000000 for an instruction access fault.
401  *
402  * The return value is 0 if the fault was handled, or the signal
403  * number if this is a kernel fault that can't be handled here.
404  */
___do_page_fault(struct pt_regs * regs,unsigned long address,unsigned long error_code)405 static int ___do_page_fault(struct pt_regs *regs, unsigned long address,
406 			   unsigned long error_code)
407 {
408 	struct vm_area_struct * vma;
409 	struct mm_struct *mm = current->mm;
410 	unsigned int flags = FAULT_FLAG_DEFAULT;
411 	int is_exec = TRAP(regs) == INTERRUPT_INST_STORAGE;
412 	int is_user = user_mode(regs);
413 	int is_write = page_fault_is_write(error_code);
414 	vm_fault_t fault, major = 0;
415 	bool kprobe_fault = kprobe_page_fault(regs, 11);
416 
417 	if (unlikely(debugger_fault_handler(regs) || kprobe_fault))
418 		return 0;
419 
420 	if (unlikely(page_fault_is_bad(error_code))) {
421 		if (is_user) {
422 			_exception(SIGBUS, regs, BUS_OBJERR, address);
423 			return 0;
424 		}
425 		return SIGBUS;
426 	}
427 
428 	/* Additional sanity check(s) */
429 	sanity_check_fault(is_write, is_user, error_code, address);
430 
431 	/*
432 	 * The kernel should never take an execute fault nor should it
433 	 * take a page fault to a kernel address or a page fault to a user
434 	 * address outside of dedicated places
435 	 */
436 	if (unlikely(!is_user && bad_kernel_fault(regs, error_code, address, is_write))) {
437 		if (kfence_handle_page_fault(address, is_write, regs))
438 			return 0;
439 
440 		return SIGSEGV;
441 	}
442 
443 	/*
444 	 * If we're in an interrupt, have no user context or are running
445 	 * in a region with pagefaults disabled then we must not take the fault
446 	 */
447 	if (unlikely(faulthandler_disabled() || !mm)) {
448 		if (is_user)
449 			printk_ratelimited(KERN_ERR "Page fault in user mode"
450 					   " with faulthandler_disabled()=%d"
451 					   " mm=%p\n",
452 					   faulthandler_disabled(), mm);
453 		return bad_area_nosemaphore(regs, address);
454 	}
455 
456 	interrupt_cond_local_irq_enable(regs);
457 
458 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
459 
460 	/*
461 	 * We want to do this outside mmap_lock, because reading code around nip
462 	 * can result in fault, which will cause a deadlock when called with
463 	 * mmap_lock held
464 	 */
465 	if (is_user)
466 		flags |= FAULT_FLAG_USER;
467 	if (is_write)
468 		flags |= FAULT_FLAG_WRITE;
469 	if (is_exec)
470 		flags |= FAULT_FLAG_INSTRUCTION;
471 
472 	if (!(flags & FAULT_FLAG_USER))
473 		goto lock_mmap;
474 
475 	vma = lock_vma_under_rcu(mm, address);
476 	if (!vma)
477 		goto lock_mmap;
478 
479 	if (unlikely(access_pkey_error(is_write, is_exec,
480 				       (error_code & DSISR_KEYFAULT), vma))) {
481 		vma_end_read(vma);
482 		goto lock_mmap;
483 	}
484 
485 	if (unlikely(access_error(is_write, is_exec, vma))) {
486 		vma_end_read(vma);
487 		goto lock_mmap;
488 	}
489 
490 	fault = handle_mm_fault(vma, address, flags | FAULT_FLAG_VMA_LOCK, regs);
491 	if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
492 		vma_end_read(vma);
493 
494 	if (!(fault & VM_FAULT_RETRY)) {
495 		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
496 		goto done;
497 	}
498 	count_vm_vma_lock_event(VMA_LOCK_RETRY);
499 
500 	if (fault_signal_pending(fault, regs))
501 		return user_mode(regs) ? 0 : SIGBUS;
502 
503 lock_mmap:
504 
505 	/* When running in the kernel we expect faults to occur only to
506 	 * addresses in user space.  All other faults represent errors in the
507 	 * kernel and should generate an OOPS.  Unfortunately, in the case of an
508 	 * erroneous fault occurring in a code path which already holds mmap_lock
509 	 * we will deadlock attempting to validate the fault against the
510 	 * address space.  Luckily the kernel only validly references user
511 	 * space from well defined areas of code, which are listed in the
512 	 * exceptions table. lock_mm_and_find_vma() handles that logic.
513 	 */
514 retry:
515 	vma = lock_mm_and_find_vma(mm, address, regs);
516 	if (unlikely(!vma))
517 		return bad_area_nosemaphore(regs, address);
518 
519 	if (unlikely(access_pkey_error(is_write, is_exec,
520 				       (error_code & DSISR_KEYFAULT), vma)))
521 		return bad_access_pkey(regs, address, vma);
522 
523 	if (unlikely(access_error(is_write, is_exec, vma)))
524 		return bad_access(regs, address);
525 
526 	/*
527 	 * If for any reason at all we couldn't handle the fault,
528 	 * make sure we exit gracefully rather than endlessly redo
529 	 * the fault.
530 	 */
531 	fault = handle_mm_fault(vma, address, flags, regs);
532 
533 	major |= fault & VM_FAULT_MAJOR;
534 
535 	if (fault_signal_pending(fault, regs))
536 		return user_mode(regs) ? 0 : SIGBUS;
537 
538 	/* The fault is fully completed (including releasing mmap lock) */
539 	if (fault & VM_FAULT_COMPLETED)
540 		goto out;
541 
542 	/*
543 	 * Handle the retry right now, the mmap_lock has been released in that
544 	 * case.
545 	 */
546 	if (unlikely(fault & VM_FAULT_RETRY)) {
547 		flags |= FAULT_FLAG_TRIED;
548 		goto retry;
549 	}
550 
551 	mmap_read_unlock(current->mm);
552 
553 done:
554 	if (unlikely(fault & VM_FAULT_ERROR))
555 		return mm_fault_error(regs, address, fault);
556 
557 out:
558 	/*
559 	 * Major/minor page fault accounting.
560 	 */
561 	if (major)
562 		cmo_account_page_fault();
563 
564 	return 0;
565 }
566 NOKPROBE_SYMBOL(___do_page_fault);
567 
__do_page_fault(struct pt_regs * regs)568 static __always_inline void __do_page_fault(struct pt_regs *regs)
569 {
570 	long err;
571 
572 	err = ___do_page_fault(regs, regs->dar, regs->dsisr);
573 	if (unlikely(err))
574 		bad_page_fault(regs, err);
575 }
576 
DEFINE_INTERRUPT_HANDLER(do_page_fault)577 DEFINE_INTERRUPT_HANDLER(do_page_fault)
578 {
579 	__do_page_fault(regs);
580 }
581 
582 #ifdef CONFIG_PPC_BOOK3S_64
583 /* Same as do_page_fault but interrupt entry has already run in do_hash_fault */
hash__do_page_fault(struct pt_regs * regs)584 void hash__do_page_fault(struct pt_regs *regs)
585 {
586 	__do_page_fault(regs);
587 }
588 NOKPROBE_SYMBOL(hash__do_page_fault);
589 #endif
590 
591 /*
592  * bad_page_fault is called when we have a bad access from the kernel.
593  * It is called from the DSI and ISI handlers in head.S and from some
594  * of the procedures in traps.c.
595  */
__bad_page_fault(struct pt_regs * regs,int sig)596 static void __bad_page_fault(struct pt_regs *regs, int sig)
597 {
598 	int is_write = page_fault_is_write(regs->dsisr);
599 	const char *msg;
600 
601 	/* kernel has accessed a bad area */
602 
603 	if (regs->dar < PAGE_SIZE)
604 		msg = "Kernel NULL pointer dereference";
605 	else
606 		msg = "Unable to handle kernel data access";
607 
608 	switch (TRAP(regs)) {
609 	case INTERRUPT_DATA_STORAGE:
610 	case INTERRUPT_H_DATA_STORAGE:
611 		pr_alert("BUG: %s on %s at 0x%08lx\n", msg,
612 			 is_write ? "write" : "read", regs->dar);
613 		break;
614 	case INTERRUPT_DATA_SEGMENT:
615 		pr_alert("BUG: %s at 0x%08lx\n", msg, regs->dar);
616 		break;
617 	case INTERRUPT_INST_STORAGE:
618 	case INTERRUPT_INST_SEGMENT:
619 		pr_alert("BUG: Unable to handle kernel instruction fetch%s",
620 			 regs->nip < PAGE_SIZE ? " (NULL pointer?)\n" : "\n");
621 		break;
622 	case INTERRUPT_ALIGNMENT:
623 		pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n",
624 			 regs->dar);
625 		break;
626 	default:
627 		pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n",
628 			 regs->dar);
629 		break;
630 	}
631 	printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
632 		regs->nip);
633 
634 	if (task_stack_end_corrupted(current))
635 		printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
636 
637 	die("Kernel access of bad area", regs, sig);
638 }
639 
bad_page_fault(struct pt_regs * regs,int sig)640 void bad_page_fault(struct pt_regs *regs, int sig)
641 {
642 	const struct exception_table_entry *entry;
643 
644 	/* Are we prepared to handle this fault?  */
645 	entry = search_exception_tables(instruction_pointer(regs));
646 	if (entry)
647 		instruction_pointer_set(regs, extable_fixup(entry));
648 	else
649 		__bad_page_fault(regs, sig);
650 }
651 
652 #ifdef CONFIG_PPC_BOOK3S_64
DEFINE_INTERRUPT_HANDLER(do_bad_page_fault_segv)653 DEFINE_INTERRUPT_HANDLER(do_bad_page_fault_segv)
654 {
655 	bad_page_fault(regs, SIGSEGV);
656 }
657 
658 /*
659  * In radix, segment interrupts indicate the EA is not addressable by the
660  * page table geometry, so they are always sent here.
661  *
662  * In hash, this is called if do_slb_fault returns error. Typically it is
663  * because the EA was outside the region allowed by software.
664  */
DEFINE_INTERRUPT_HANDLER(do_bad_segment_interrupt)665 DEFINE_INTERRUPT_HANDLER(do_bad_segment_interrupt)
666 {
667 	int err = regs->result;
668 
669 	if (err == -EFAULT) {
670 		if (user_mode(regs))
671 			_exception(SIGSEGV, regs, SEGV_BNDERR, regs->dar);
672 		else
673 			bad_page_fault(regs, SIGSEGV);
674 	} else if (err == -EINVAL) {
675 		unrecoverable_exception(regs);
676 	} else {
677 		BUG();
678 	}
679 }
680 #endif
681