xref: /openbmc/linux/arch/arm64/mm/fault.c (revision cfdfc14e)
1 /*
2  * Based on arch/arm/mm/fault.c
3  *
4  * Copyright (C) 1995  Linus Torvalds
5  * Copyright (C) 1995-2004 Russell King
6  * Copyright (C) 2012 ARM Ltd.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 as
10  * published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License
18  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19  */
20 
21 #include <linux/extable.h>
22 #include <linux/signal.h>
23 #include <linux/mm.h>
24 #include <linux/hardirq.h>
25 #include <linux/init.h>
26 #include <linux/kprobes.h>
27 #include <linux/uaccess.h>
28 #include <linux/page-flags.h>
29 #include <linux/sched/signal.h>
30 #include <linux/sched/debug.h>
31 #include <linux/highmem.h>
32 #include <linux/perf_event.h>
33 #include <linux/preempt.h>
34 #include <linux/hugetlb.h>
35 
36 #include <asm/bug.h>
37 #include <asm/cmpxchg.h>
38 #include <asm/cpufeature.h>
39 #include <asm/exception.h>
40 #include <asm/debug-monitors.h>
41 #include <asm/esr.h>
42 #include <asm/sysreg.h>
43 #include <asm/system_misc.h>
44 #include <asm/pgtable.h>
45 #include <asm/tlbflush.h>
46 #include <asm/traps.h>
47 
48 #include <acpi/ghes.h>
49 
50 struct fault_info {
51 	int	(*fn)(unsigned long addr, unsigned int esr,
52 		      struct pt_regs *regs);
53 	int	sig;
54 	int	code;
55 	const char *name;
56 };
57 
58 static const struct fault_info fault_info[];
59 
60 static inline const struct fault_info *esr_to_fault_info(unsigned int esr)
61 {
62 	return fault_info + (esr & 63);
63 }
64 
65 #ifdef CONFIG_KPROBES
66 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
67 {
68 	int ret = 0;
69 
70 	/* kprobe_running() needs smp_processor_id() */
71 	if (!user_mode(regs)) {
72 		preempt_disable();
73 		if (kprobe_running() && kprobe_fault_handler(regs, esr))
74 			ret = 1;
75 		preempt_enable();
76 	}
77 
78 	return ret;
79 }
80 #else
81 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
82 {
83 	return 0;
84 }
85 #endif
86 
87 static void data_abort_decode(unsigned int esr)
88 {
89 	pr_alert("Data abort info:\n");
90 
91 	if (esr & ESR_ELx_ISV) {
92 		pr_alert("  Access size = %u byte(s)\n",
93 			 1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
94 		pr_alert("  SSE = %lu, SRT = %lu\n",
95 			 (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
96 			 (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
97 		pr_alert("  SF = %lu, AR = %lu\n",
98 			 (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
99 			 (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
100 	} else {
101 		pr_alert("  ISV = 0, ISS = 0x%08lx\n", esr & ESR_ELx_ISS_MASK);
102 	}
103 
104 	pr_alert("  CM = %lu, WnR = %lu\n",
105 		 (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
106 		 (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT);
107 }
108 
109 static void mem_abort_decode(unsigned int esr)
110 {
111 	pr_alert("Mem abort info:\n");
112 
113 	pr_alert("  ESR = 0x%08x\n", esr);
114 	pr_alert("  Exception class = %s, IL = %u bits\n",
115 		 esr_get_class_string(esr),
116 		 (esr & ESR_ELx_IL) ? 32 : 16);
117 	pr_alert("  SET = %lu, FnV = %lu\n",
118 		 (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
119 		 (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
120 	pr_alert("  EA = %lu, S1PTW = %lu\n",
121 		 (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
122 		 (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
123 
124 	if (esr_is_data_abort(esr))
125 		data_abort_decode(esr);
126 }
127 
128 /*
129  * Dump out the page tables associated with 'addr' in the currently active mm.
130  */
131 void show_pte(unsigned long addr)
132 {
133 	struct mm_struct *mm;
134 	pgd_t *pgdp;
135 	pgd_t pgd;
136 
137 	if (addr < TASK_SIZE) {
138 		/* TTBR0 */
139 		mm = current->active_mm;
140 		if (mm == &init_mm) {
141 			pr_alert("[%016lx] user address but active_mm is swapper\n",
142 				 addr);
143 			return;
144 		}
145 	} else if (addr >= VA_START) {
146 		/* TTBR1 */
147 		mm = &init_mm;
148 	} else {
149 		pr_alert("[%016lx] address between user and kernel address ranges\n",
150 			 addr);
151 		return;
152 	}
153 
154 	pr_alert("%s pgtable: %luk pages, %u-bit VAs, pgdp = %p\n",
155 		 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
156 		 VA_BITS, mm->pgd);
157 	pgdp = pgd_offset(mm, addr);
158 	pgd = READ_ONCE(*pgdp);
159 	pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
160 
161 	do {
162 		pud_t *pudp, pud;
163 		pmd_t *pmdp, pmd;
164 		pte_t *ptep, pte;
165 
166 		if (pgd_none(pgd) || pgd_bad(pgd))
167 			break;
168 
169 		pudp = pud_offset(pgdp, addr);
170 		pud = READ_ONCE(*pudp);
171 		pr_cont(", pud=%016llx", pud_val(pud));
172 		if (pud_none(pud) || pud_bad(pud))
173 			break;
174 
175 		pmdp = pmd_offset(pudp, addr);
176 		pmd = READ_ONCE(*pmdp);
177 		pr_cont(", pmd=%016llx", pmd_val(pmd));
178 		if (pmd_none(pmd) || pmd_bad(pmd))
179 			break;
180 
181 		ptep = pte_offset_map(pmdp, addr);
182 		pte = READ_ONCE(*ptep);
183 		pr_cont(", pte=%016llx", pte_val(pte));
184 		pte_unmap(ptep);
185 	} while(0);
186 
187 	pr_cont("\n");
188 }
189 
190 /*
191  * This function sets the access flags (dirty, accessed), as well as write
192  * permission, and only to a more permissive setting.
193  *
194  * It needs to cope with hardware update of the accessed/dirty state by other
195  * agents in the system and can safely skip the __sync_icache_dcache() call as,
196  * like set_pte_at(), the PTE is never changed from no-exec to exec here.
197  *
198  * Returns whether or not the PTE actually changed.
199  */
200 int ptep_set_access_flags(struct vm_area_struct *vma,
201 			  unsigned long address, pte_t *ptep,
202 			  pte_t entry, int dirty)
203 {
204 	pteval_t old_pteval, pteval;
205 	pte_t pte = READ_ONCE(*ptep);
206 
207 	if (pte_same(pte, entry))
208 		return 0;
209 
210 	/* only preserve the access flags and write permission */
211 	pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
212 
213 	/*
214 	 * Setting the flags must be done atomically to avoid racing with the
215 	 * hardware update of the access/dirty state. The PTE_RDONLY bit must
216 	 * be set to the most permissive (lowest value) of *ptep and entry
217 	 * (calculated as: a & b == ~(~a | ~b)).
218 	 */
219 	pte_val(entry) ^= PTE_RDONLY;
220 	pteval = pte_val(pte);
221 	do {
222 		old_pteval = pteval;
223 		pteval ^= PTE_RDONLY;
224 		pteval |= pte_val(entry);
225 		pteval ^= PTE_RDONLY;
226 		pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
227 	} while (pteval != old_pteval);
228 
229 	flush_tlb_fix_spurious_fault(vma, address);
230 	return 1;
231 }
232 
233 static bool is_el1_instruction_abort(unsigned int esr)
234 {
235 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
236 }
237 
238 static inline bool is_el1_permission_fault(unsigned int esr,
239 					   struct pt_regs *regs,
240 					   unsigned long addr)
241 {
242 	unsigned int ec       = ESR_ELx_EC(esr);
243 	unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
244 
245 	if (ec != ESR_ELx_EC_DABT_CUR && ec != ESR_ELx_EC_IABT_CUR)
246 		return false;
247 
248 	if (fsc_type == ESR_ELx_FSC_PERM)
249 		return true;
250 
251 	if (addr < TASK_SIZE && system_uses_ttbr0_pan())
252 		return fsc_type == ESR_ELx_FSC_FAULT &&
253 			(regs->pstate & PSR_PAN_BIT);
254 
255 	return false;
256 }
257 
258 static void die_kernel_fault(const char *msg, unsigned long addr,
259 			     unsigned int esr, struct pt_regs *regs)
260 {
261 	bust_spinlocks(1);
262 
263 	pr_alert("Unable to handle kernel %s at virtual address %016lx\n", msg,
264 		 addr);
265 
266 	mem_abort_decode(esr);
267 
268 	show_pte(addr);
269 	die("Oops", regs, esr);
270 	bust_spinlocks(0);
271 	do_exit(SIGKILL);
272 }
273 
274 static void __do_kernel_fault(unsigned long addr, unsigned int esr,
275 			      struct pt_regs *regs)
276 {
277 	const char *msg;
278 
279 	/*
280 	 * Are we prepared to handle this kernel fault?
281 	 * We are almost certainly not prepared to handle instruction faults.
282 	 */
283 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
284 		return;
285 
286 	if (is_el1_permission_fault(esr, regs, addr)) {
287 		if (esr & ESR_ELx_WNR)
288 			msg = "write to read-only memory";
289 		else
290 			msg = "read from unreadable memory";
291 	} else if (addr < PAGE_SIZE) {
292 		msg = "NULL pointer dereference";
293 	} else {
294 		msg = "paging request";
295 	}
296 
297 	die_kernel_fault(msg, addr, esr, regs);
298 }
299 
300 static void __do_user_fault(struct siginfo *info, unsigned int esr)
301 {
302 	current->thread.fault_address = (unsigned long)info->si_addr;
303 
304 	/*
305 	 * If the faulting address is in the kernel, we must sanitize the ESR.
306 	 * From userspace's point of view, kernel-only mappings don't exist
307 	 * at all, so we report them as level 0 translation faults.
308 	 * (This is not quite the way that "no mapping there at all" behaves:
309 	 * an alignment fault not caused by the memory type would take
310 	 * precedence over translation fault for a real access to empty
311 	 * space. Unfortunately we can't easily distinguish "alignment fault
312 	 * not caused by memory type" from "alignment fault caused by memory
313 	 * type", so we ignore this wrinkle and just return the translation
314 	 * fault.)
315 	 */
316 	if (current->thread.fault_address >= TASK_SIZE) {
317 		switch (ESR_ELx_EC(esr)) {
318 		case ESR_ELx_EC_DABT_LOW:
319 			/*
320 			 * These bits provide only information about the
321 			 * faulting instruction, which userspace knows already.
322 			 * We explicitly clear bits which are architecturally
323 			 * RES0 in case they are given meanings in future.
324 			 * We always report the ESR as if the fault was taken
325 			 * to EL1 and so ISV and the bits in ISS[23:14] are
326 			 * clear. (In fact it always will be a fault to EL1.)
327 			 */
328 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL |
329 				ESR_ELx_CM | ESR_ELx_WNR;
330 			esr |= ESR_ELx_FSC_FAULT;
331 			break;
332 		case ESR_ELx_EC_IABT_LOW:
333 			/*
334 			 * Claim a level 0 translation fault.
335 			 * All other bits are architecturally RES0 for faults
336 			 * reported with that DFSC value, so we clear them.
337 			 */
338 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL;
339 			esr |= ESR_ELx_FSC_FAULT;
340 			break;
341 		default:
342 			/*
343 			 * This should never happen (entry.S only brings us
344 			 * into this code for insn and data aborts from a lower
345 			 * exception level). Fail safe by not providing an ESR
346 			 * context record at all.
347 			 */
348 			WARN(1, "ESR 0x%x is not DABT or IABT from EL0\n", esr);
349 			esr = 0;
350 			break;
351 		}
352 	}
353 
354 	current->thread.fault_code = esr;
355 	arm64_force_sig_info(info, esr_to_fault_info(esr)->name, current);
356 }
357 
358 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
359 {
360 	/*
361 	 * If we are in kernel mode at this point, we have no context to
362 	 * handle this fault with.
363 	 */
364 	if (user_mode(regs)) {
365 		const struct fault_info *inf = esr_to_fault_info(esr);
366 		struct siginfo si;
367 
368 		clear_siginfo(&si);
369 		si.si_signo	= inf->sig;
370 		si.si_code	= inf->code;
371 		si.si_addr	= (void __user *)addr;
372 
373 		__do_user_fault(&si, esr);
374 	} else {
375 		__do_kernel_fault(addr, esr, regs);
376 	}
377 }
378 
379 #define VM_FAULT_BADMAP		0x010000
380 #define VM_FAULT_BADACCESS	0x020000
381 
382 static int __do_page_fault(struct mm_struct *mm, unsigned long addr,
383 			   unsigned int mm_flags, unsigned long vm_flags,
384 			   struct task_struct *tsk)
385 {
386 	struct vm_area_struct *vma;
387 	int fault;
388 
389 	vma = find_vma(mm, addr);
390 	fault = VM_FAULT_BADMAP;
391 	if (unlikely(!vma))
392 		goto out;
393 	if (unlikely(vma->vm_start > addr))
394 		goto check_stack;
395 
396 	/*
397 	 * Ok, we have a good vm_area for this memory access, so we can handle
398 	 * it.
399 	 */
400 good_area:
401 	/*
402 	 * Check that the permissions on the VMA allow for the fault which
403 	 * occurred.
404 	 */
405 	if (!(vma->vm_flags & vm_flags)) {
406 		fault = VM_FAULT_BADACCESS;
407 		goto out;
408 	}
409 
410 	return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags);
411 
412 check_stack:
413 	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
414 		goto good_area;
415 out:
416 	return fault;
417 }
418 
419 static bool is_el0_instruction_abort(unsigned int esr)
420 {
421 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
422 }
423 
424 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
425 				   struct pt_regs *regs)
426 {
427 	struct task_struct *tsk;
428 	struct mm_struct *mm;
429 	struct siginfo si;
430 	int fault, major = 0;
431 	unsigned long vm_flags = VM_READ | VM_WRITE;
432 	unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
433 
434 	if (notify_page_fault(regs, esr))
435 		return 0;
436 
437 	tsk = current;
438 	mm  = tsk->mm;
439 
440 	/*
441 	 * If we're in an interrupt or have no user context, we must not take
442 	 * the fault.
443 	 */
444 	if (faulthandler_disabled() || !mm)
445 		goto no_context;
446 
447 	if (user_mode(regs))
448 		mm_flags |= FAULT_FLAG_USER;
449 
450 	if (is_el0_instruction_abort(esr)) {
451 		vm_flags = VM_EXEC;
452 	} else if ((esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM)) {
453 		vm_flags = VM_WRITE;
454 		mm_flags |= FAULT_FLAG_WRITE;
455 	}
456 
457 	if (addr < TASK_SIZE && is_el1_permission_fault(esr, regs, addr)) {
458 		/* regs->orig_addr_limit may be 0 if we entered from EL0 */
459 		if (regs->orig_addr_limit == KERNEL_DS)
460 			die_kernel_fault("access to user memory with fs=KERNEL_DS",
461 					 addr, esr, regs);
462 
463 		if (is_el1_instruction_abort(esr))
464 			die_kernel_fault("execution of user memory",
465 					 addr, esr, regs);
466 
467 		if (!search_exception_tables(regs->pc))
468 			die_kernel_fault("access to user memory outside uaccess routines",
469 					 addr, esr, regs);
470 	}
471 
472 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
473 
474 	/*
475 	 * As per x86, we may deadlock here. However, since the kernel only
476 	 * validly references user space from well defined areas of the code,
477 	 * we can bug out early if this is from code which shouldn't.
478 	 */
479 	if (!down_read_trylock(&mm->mmap_sem)) {
480 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
481 			goto no_context;
482 retry:
483 		down_read(&mm->mmap_sem);
484 	} else {
485 		/*
486 		 * The above down_read_trylock() might have succeeded in which
487 		 * case, we'll have missed the might_sleep() from down_read().
488 		 */
489 		might_sleep();
490 #ifdef CONFIG_DEBUG_VM
491 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
492 			goto no_context;
493 #endif
494 	}
495 
496 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, tsk);
497 	major |= fault & VM_FAULT_MAJOR;
498 
499 	if (fault & VM_FAULT_RETRY) {
500 		/*
501 		 * If we need to retry but a fatal signal is pending,
502 		 * handle the signal first. We do not need to release
503 		 * the mmap_sem because it would already be released
504 		 * in __lock_page_or_retry in mm/filemap.c.
505 		 */
506 		if (fatal_signal_pending(current)) {
507 			if (!user_mode(regs))
508 				goto no_context;
509 			return 0;
510 		}
511 
512 		/*
513 		 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
514 		 * starvation.
515 		 */
516 		if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
517 			mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
518 			mm_flags |= FAULT_FLAG_TRIED;
519 			goto retry;
520 		}
521 	}
522 	up_read(&mm->mmap_sem);
523 
524 	/*
525 	 * Handle the "normal" (no error) case first.
526 	 */
527 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
528 			      VM_FAULT_BADACCESS)))) {
529 		/*
530 		 * Major/minor page fault accounting is only done
531 		 * once. If we go through a retry, it is extremely
532 		 * likely that the page will be found in page cache at
533 		 * that point.
534 		 */
535 		if (major) {
536 			tsk->maj_flt++;
537 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
538 				      addr);
539 		} else {
540 			tsk->min_flt++;
541 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
542 				      addr);
543 		}
544 
545 		return 0;
546 	}
547 
548 	/*
549 	 * If we are in kernel mode at this point, we have no context to
550 	 * handle this fault with.
551 	 */
552 	if (!user_mode(regs))
553 		goto no_context;
554 
555 	if (fault & VM_FAULT_OOM) {
556 		/*
557 		 * We ran out of memory, call the OOM killer, and return to
558 		 * userspace (which will retry the fault, or kill us if we got
559 		 * oom-killed).
560 		 */
561 		pagefault_out_of_memory();
562 		return 0;
563 	}
564 
565 	clear_siginfo(&si);
566 	si.si_addr = (void __user *)addr;
567 
568 	if (fault & VM_FAULT_SIGBUS) {
569 		/*
570 		 * We had some memory, but were unable to successfully fix up
571 		 * this page fault.
572 		 */
573 		si.si_signo	= SIGBUS;
574 		si.si_code	= BUS_ADRERR;
575 	} else if (fault & VM_FAULT_HWPOISON_LARGE) {
576 		unsigned int hindex = VM_FAULT_GET_HINDEX(fault);
577 
578 		si.si_signo	= SIGBUS;
579 		si.si_code	= BUS_MCEERR_AR;
580 		si.si_addr_lsb	= hstate_index_to_shift(hindex);
581 	} else if (fault & VM_FAULT_HWPOISON) {
582 		si.si_signo	= SIGBUS;
583 		si.si_code	= BUS_MCEERR_AR;
584 		si.si_addr_lsb	= PAGE_SHIFT;
585 	} else {
586 		/*
587 		 * Something tried to access memory that isn't in our memory
588 		 * map.
589 		 */
590 		si.si_signo	= SIGSEGV;
591 		si.si_code	= fault == VM_FAULT_BADACCESS ?
592 				  SEGV_ACCERR : SEGV_MAPERR;
593 	}
594 
595 	__do_user_fault(&si, esr);
596 	return 0;
597 
598 no_context:
599 	__do_kernel_fault(addr, esr, regs);
600 	return 0;
601 }
602 
603 static int __kprobes do_translation_fault(unsigned long addr,
604 					  unsigned int esr,
605 					  struct pt_regs *regs)
606 {
607 	if (addr < TASK_SIZE)
608 		return do_page_fault(addr, esr, regs);
609 
610 	do_bad_area(addr, esr, regs);
611 	return 0;
612 }
613 
614 static int do_alignment_fault(unsigned long addr, unsigned int esr,
615 			      struct pt_regs *regs)
616 {
617 	do_bad_area(addr, esr, regs);
618 	return 0;
619 }
620 
621 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
622 {
623 	return 1; /* "fault" */
624 }
625 
626 static int do_sea(unsigned long addr, unsigned int esr, struct pt_regs *regs)
627 {
628 	struct siginfo info;
629 	const struct fault_info *inf;
630 
631 	inf = esr_to_fault_info(esr);
632 
633 	/*
634 	 * Synchronous aborts may interrupt code which had interrupts masked.
635 	 * Before calling out into the wider kernel tell the interested
636 	 * subsystems.
637 	 */
638 	if (IS_ENABLED(CONFIG_ACPI_APEI_SEA)) {
639 		if (interrupts_enabled(regs))
640 			nmi_enter();
641 
642 		ghes_notify_sea();
643 
644 		if (interrupts_enabled(regs))
645 			nmi_exit();
646 	}
647 
648 	clear_siginfo(&info);
649 	info.si_signo = inf->sig;
650 	info.si_errno = 0;
651 	info.si_code  = inf->code;
652 	if (esr & ESR_ELx_FnV)
653 		info.si_addr = NULL;
654 	else
655 		info.si_addr  = (void __user *)addr;
656 	arm64_notify_die(inf->name, regs, &info, esr);
657 
658 	return 0;
659 }
660 
661 static const struct fault_info fault_info[] = {
662 	{ do_bad,		SIGKILL, SI_KERNEL,	"ttbr address size fault"	},
663 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 1 address size fault"	},
664 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 2 address size fault"	},
665 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 3 address size fault"	},
666 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
667 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
668 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
669 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
670 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 8"			},
671 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
672 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
673 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
674 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 12"			},
675 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
676 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
677 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
678 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous external abort"	},
679 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 17"			},
680 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 18"			},
681 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 19"			},
682 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 (translation table walk)"	},
683 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 (translation table walk)"	},
684 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 (translation table walk)"	},
685 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 (translation table walk)"	},
686 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous parity or ECC error" },	// Reserved when RAS is implemented
687 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 25"			},
688 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 26"			},
689 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 27"			},
690 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
691 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
692 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
693 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
694 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 32"			},
695 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
696 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 34"			},
697 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 35"			},
698 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 36"			},
699 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 37"			},
700 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 38"			},
701 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 39"			},
702 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 40"			},
703 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 41"			},
704 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 42"			},
705 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 43"			},
706 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 44"			},
707 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 45"			},
708 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 46"			},
709 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 47"			},
710 	{ do_bad,		SIGKILL, SI_KERNEL,	"TLB conflict abort"		},
711 	{ do_bad,		SIGKILL, SI_KERNEL,	"Unsupported atomic hardware update fault"	},
712 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 50"			},
713 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 51"			},
714 	{ do_bad,		SIGKILL, SI_KERNEL,	"implementation fault (lockdown abort)" },
715 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"implementation fault (unsupported exclusive)" },
716 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 54"			},
717 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 55"			},
718 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 56"			},
719 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 57"			},
720 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 58" 			},
721 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 59"			},
722 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 60"			},
723 	{ do_bad,		SIGKILL, SI_KERNEL,	"section domain fault"		},
724 	{ do_bad,		SIGKILL, SI_KERNEL,	"page domain fault"		},
725 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 63"			},
726 };
727 
728 int handle_guest_sea(phys_addr_t addr, unsigned int esr)
729 {
730 	int ret = -ENOENT;
731 
732 	if (IS_ENABLED(CONFIG_ACPI_APEI_SEA))
733 		ret = ghes_notify_sea();
734 
735 	return ret;
736 }
737 
738 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
739 					 struct pt_regs *regs)
740 {
741 	const struct fault_info *inf = esr_to_fault_info(esr);
742 	struct siginfo info;
743 
744 	if (!inf->fn(addr, esr, regs))
745 		return;
746 
747 	if (!user_mode(regs)) {
748 		pr_alert("Unhandled fault at 0x%016lx\n", addr);
749 		mem_abort_decode(esr);
750 		show_pte(addr);
751 	}
752 
753 	clear_siginfo(&info);
754 	info.si_signo = inf->sig;
755 	info.si_errno = 0;
756 	info.si_code  = inf->code;
757 	info.si_addr  = (void __user *)addr;
758 	arm64_notify_die(inf->name, regs, &info, esr);
759 }
760 
761 asmlinkage void __exception do_el0_irq_bp_hardening(void)
762 {
763 	/* PC has already been checked in entry.S */
764 	arm64_apply_bp_hardening();
765 }
766 
767 asmlinkage void __exception do_el0_ia_bp_hardening(unsigned long addr,
768 						   unsigned int esr,
769 						   struct pt_regs *regs)
770 {
771 	/*
772 	 * We've taken an instruction abort from userspace and not yet
773 	 * re-enabled IRQs. If the address is a kernel address, apply
774 	 * BP hardening prior to enabling IRQs and pre-emption.
775 	 */
776 	if (addr > TASK_SIZE)
777 		arm64_apply_bp_hardening();
778 
779 	local_irq_enable();
780 	do_mem_abort(addr, esr, regs);
781 }
782 
783 
784 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
785 					   unsigned int esr,
786 					   struct pt_regs *regs)
787 {
788 	struct siginfo info;
789 
790 	if (user_mode(regs)) {
791 		if (instruction_pointer(regs) > TASK_SIZE)
792 			arm64_apply_bp_hardening();
793 		local_irq_enable();
794 	}
795 
796 	clear_siginfo(&info);
797 	info.si_signo = SIGBUS;
798 	info.si_errno = 0;
799 	info.si_code  = BUS_ADRALN;
800 	info.si_addr  = (void __user *)addr;
801 	arm64_notify_die("SP/PC alignment exception", regs, &info, esr);
802 }
803 
804 int __init early_brk64(unsigned long addr, unsigned int esr,
805 		       struct pt_regs *regs);
806 
807 /*
808  * __refdata because early_brk64 is __init, but the reference to it is
809  * clobbered at arch_initcall time.
810  * See traps.c and debug-monitors.c:debug_traps_init().
811  */
812 static struct fault_info __refdata debug_fault_info[] = {
813 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
814 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
815 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
816 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 3"		},
817 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
818 	{ do_bad,	SIGKILL,	SI_KERNEL,	"aarch32 vector catch"	},
819 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
820 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 7"		},
821 };
822 
823 void __init hook_debug_fault_code(int nr,
824 				  int (*fn)(unsigned long, unsigned int, struct pt_regs *),
825 				  int sig, int code, const char *name)
826 {
827 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
828 
829 	debug_fault_info[nr].fn		= fn;
830 	debug_fault_info[nr].sig	= sig;
831 	debug_fault_info[nr].code	= code;
832 	debug_fault_info[nr].name	= name;
833 }
834 
835 asmlinkage int __exception do_debug_exception(unsigned long addr,
836 					      unsigned int esr,
837 					      struct pt_regs *regs)
838 {
839 	const struct fault_info *inf = debug_fault_info + DBG_ESR_EVT(esr);
840 	int rv;
841 
842 	/*
843 	 * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
844 	 * already disabled to preserve the last enabled/disabled addresses.
845 	 */
846 	if (interrupts_enabled(regs))
847 		trace_hardirqs_off();
848 
849 	if (user_mode(regs) && instruction_pointer(regs) > TASK_SIZE)
850 		arm64_apply_bp_hardening();
851 
852 	if (!inf->fn(addr, esr, regs)) {
853 		rv = 1;
854 	} else {
855 		struct siginfo info;
856 
857 		clear_siginfo(&info);
858 		info.si_signo = inf->sig;
859 		info.si_errno = 0;
860 		info.si_code  = inf->code;
861 		info.si_addr  = (void __user *)addr;
862 		arm64_notify_die(inf->name, regs, &info, esr);
863 		rv = 0;
864 	}
865 
866 	if (interrupts_enabled(regs))
867 		trace_hardirqs_on();
868 
869 	return rv;
870 }
871 NOKPROBE_SYMBOL(do_debug_exception);
872 
873 #ifdef CONFIG_ARM64_PAN
874 void cpu_enable_pan(const struct arm64_cpu_capabilities *__unused)
875 {
876 	/*
877 	 * We modify PSTATE. This won't work from irq context as the PSTATE
878 	 * is discarded once we return from the exception.
879 	 */
880 	WARN_ON_ONCE(in_interrupt());
881 
882 	config_sctlr_el1(SCTLR_EL1_SPAN, 0);
883 	asm(SET_PSTATE_PAN(1));
884 }
885 #endif /* CONFIG_ARM64_PAN */
886