fault.c

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