xref: /linux/arch/arm64/mm/fault.c (revision e21f9e2e862e9eb3dd64eaddb6256b3e5098660f)
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_permission_fault(unsigned int esr, struct pt_regs *regs,
239 				       unsigned long addr)
240 {
241 	unsigned int ec       = ESR_ELx_EC(esr);
242 	unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
243 
244 	if (ec != ESR_ELx_EC_DABT_CUR && ec != ESR_ELx_EC_IABT_CUR)
245 		return false;
246 
247 	if (fsc_type == ESR_ELx_FSC_PERM)
248 		return true;
249 
250 	if (addr < TASK_SIZE && system_uses_ttbr0_pan())
251 		return fsc_type == ESR_ELx_FSC_FAULT &&
252 			(regs->pstate & PSR_PAN_BIT);
253 
254 	return false;
255 }
256 
257 static void __do_kernel_fault(unsigned long addr, unsigned int esr,
258 			      struct pt_regs *regs)
259 {
260 	const char *msg;
261 
262 	/*
263 	 * Are we prepared to handle this kernel fault?
264 	 * We are almost certainly not prepared to handle instruction faults.
265 	 */
266 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
267 		return;
268 
269 	bust_spinlocks(1);
270 
271 	if (is_permission_fault(esr, regs, addr)) {
272 		if (esr & ESR_ELx_WNR)
273 			msg = "write to read-only memory";
274 		else
275 			msg = "read from unreadable memory";
276 	} else if (addr < PAGE_SIZE) {
277 		msg = "NULL pointer dereference";
278 	} else {
279 		msg = "paging request";
280 	}
281 
282 	pr_alert("Unable to handle kernel %s at virtual address %08lx\n", msg,
283 		 addr);
284 
285 	mem_abort_decode(esr);
286 
287 	show_pte(addr);
288 	die("Oops", regs, esr);
289 	bust_spinlocks(0);
290 	do_exit(SIGKILL);
291 }
292 
293 static void __do_user_fault(struct siginfo *info, unsigned int esr)
294 {
295 	current->thread.fault_address = (unsigned long)info->si_addr;
296 	current->thread.fault_code = esr;
297 	arm64_force_sig_info(info, esr_to_fault_info(esr)->name, current);
298 }
299 
300 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
301 {
302 	/*
303 	 * If we are in kernel mode at this point, we have no context to
304 	 * handle this fault with.
305 	 */
306 	if (user_mode(regs)) {
307 		const struct fault_info *inf = esr_to_fault_info(esr);
308 		struct siginfo si = {
309 			.si_signo	= inf->sig,
310 			.si_code	= inf->code,
311 			.si_addr	= (void __user *)addr,
312 		};
313 
314 		__do_user_fault(&si, esr);
315 	} else {
316 		__do_kernel_fault(addr, esr, regs);
317 	}
318 }
319 
320 #define VM_FAULT_BADMAP		0x010000
321 #define VM_FAULT_BADACCESS	0x020000
322 
323 static int __do_page_fault(struct mm_struct *mm, unsigned long addr,
324 			   unsigned int mm_flags, unsigned long vm_flags,
325 			   struct task_struct *tsk)
326 {
327 	struct vm_area_struct *vma;
328 	int fault;
329 
330 	vma = find_vma(mm, addr);
331 	fault = VM_FAULT_BADMAP;
332 	if (unlikely(!vma))
333 		goto out;
334 	if (unlikely(vma->vm_start > addr))
335 		goto check_stack;
336 
337 	/*
338 	 * Ok, we have a good vm_area for this memory access, so we can handle
339 	 * it.
340 	 */
341 good_area:
342 	/*
343 	 * Check that the permissions on the VMA allow for the fault which
344 	 * occurred.
345 	 */
346 	if (!(vma->vm_flags & vm_flags)) {
347 		fault = VM_FAULT_BADACCESS;
348 		goto out;
349 	}
350 
351 	return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags);
352 
353 check_stack:
354 	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
355 		goto good_area;
356 out:
357 	return fault;
358 }
359 
360 static bool is_el0_instruction_abort(unsigned int esr)
361 {
362 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
363 }
364 
365 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
366 				   struct pt_regs *regs)
367 {
368 	struct task_struct *tsk;
369 	struct mm_struct *mm;
370 	struct siginfo si;
371 	int fault, major = 0;
372 	unsigned long vm_flags = VM_READ | VM_WRITE;
373 	unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
374 
375 	if (notify_page_fault(regs, esr))
376 		return 0;
377 
378 	tsk = current;
379 	mm  = tsk->mm;
380 
381 	/*
382 	 * If we're in an interrupt or have no user context, we must not take
383 	 * the fault.
384 	 */
385 	if (faulthandler_disabled() || !mm)
386 		goto no_context;
387 
388 	if (user_mode(regs))
389 		mm_flags |= FAULT_FLAG_USER;
390 
391 	if (is_el0_instruction_abort(esr)) {
392 		vm_flags = VM_EXEC;
393 	} else if ((esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM)) {
394 		vm_flags = VM_WRITE;
395 		mm_flags |= FAULT_FLAG_WRITE;
396 	}
397 
398 	if (addr < TASK_SIZE && is_permission_fault(esr, regs, addr)) {
399 		/* regs->orig_addr_limit may be 0 if we entered from EL0 */
400 		if (regs->orig_addr_limit == KERNEL_DS)
401 			die("Accessing user space memory with fs=KERNEL_DS", regs, esr);
402 
403 		if (is_el1_instruction_abort(esr))
404 			die("Attempting to execute userspace memory", regs, esr);
405 
406 		if (!search_exception_tables(regs->pc))
407 			die("Accessing user space memory outside uaccess.h routines", regs, esr);
408 	}
409 
410 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
411 
412 	/*
413 	 * As per x86, we may deadlock here. However, since the kernel only
414 	 * validly references user space from well defined areas of the code,
415 	 * we can bug out early if this is from code which shouldn't.
416 	 */
417 	if (!down_read_trylock(&mm->mmap_sem)) {
418 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
419 			goto no_context;
420 retry:
421 		down_read(&mm->mmap_sem);
422 	} else {
423 		/*
424 		 * The above down_read_trylock() might have succeeded in which
425 		 * case, we'll have missed the might_sleep() from down_read().
426 		 */
427 		might_sleep();
428 #ifdef CONFIG_DEBUG_VM
429 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
430 			goto no_context;
431 #endif
432 	}
433 
434 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, tsk);
435 	major |= fault & VM_FAULT_MAJOR;
436 
437 	if (fault & VM_FAULT_RETRY) {
438 		/*
439 		 * If we need to retry but a fatal signal is pending,
440 		 * handle the signal first. We do not need to release
441 		 * the mmap_sem because it would already be released
442 		 * in __lock_page_or_retry in mm/filemap.c.
443 		 */
444 		if (fatal_signal_pending(current)) {
445 			if (!user_mode(regs))
446 				goto no_context;
447 			return 0;
448 		}
449 
450 		/*
451 		 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
452 		 * starvation.
453 		 */
454 		if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
455 			mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
456 			mm_flags |= FAULT_FLAG_TRIED;
457 			goto retry;
458 		}
459 	}
460 	up_read(&mm->mmap_sem);
461 
462 	/*
463 	 * Handle the "normal" (no error) case first.
464 	 */
465 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
466 			      VM_FAULT_BADACCESS)))) {
467 		/*
468 		 * Major/minor page fault accounting is only done
469 		 * once. If we go through a retry, it is extremely
470 		 * likely that the page will be found in page cache at
471 		 * that point.
472 		 */
473 		if (major) {
474 			tsk->maj_flt++;
475 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
476 				      addr);
477 		} else {
478 			tsk->min_flt++;
479 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
480 				      addr);
481 		}
482 
483 		return 0;
484 	}
485 
486 	/*
487 	 * If we are in kernel mode at this point, we have no context to
488 	 * handle this fault with.
489 	 */
490 	if (!user_mode(regs))
491 		goto no_context;
492 
493 	if (fault & VM_FAULT_OOM) {
494 		/*
495 		 * We ran out of memory, call the OOM killer, and return to
496 		 * userspace (which will retry the fault, or kill us if we got
497 		 * oom-killed).
498 		 */
499 		pagefault_out_of_memory();
500 		return 0;
501 	}
502 
503 	clear_siginfo(&si);
504 	si.si_addr = (void __user *)addr;
505 
506 	if (fault & VM_FAULT_SIGBUS) {
507 		/*
508 		 * We had some memory, but were unable to successfully fix up
509 		 * this page fault.
510 		 */
511 		si.si_signo	= SIGBUS;
512 		si.si_code	= BUS_ADRERR;
513 	} else if (fault & VM_FAULT_HWPOISON_LARGE) {
514 		unsigned int hindex = VM_FAULT_GET_HINDEX(fault);
515 
516 		si.si_signo	= SIGBUS;
517 		si.si_code	= BUS_MCEERR_AR;
518 		si.si_addr_lsb	= hstate_index_to_shift(hindex);
519 	} else if (fault & VM_FAULT_HWPOISON) {
520 		si.si_signo	= SIGBUS;
521 		si.si_code	= BUS_MCEERR_AR;
522 		si.si_addr_lsb	= PAGE_SHIFT;
523 	} else {
524 		/*
525 		 * Something tried to access memory that isn't in our memory
526 		 * map.
527 		 */
528 		si.si_signo	= SIGSEGV;
529 		si.si_code	= fault == VM_FAULT_BADACCESS ?
530 				  SEGV_ACCERR : SEGV_MAPERR;
531 	}
532 
533 	__do_user_fault(&si, esr);
534 	return 0;
535 
536 no_context:
537 	__do_kernel_fault(addr, esr, regs);
538 	return 0;
539 }
540 
541 static int __kprobes do_translation_fault(unsigned long addr,
542 					  unsigned int esr,
543 					  struct pt_regs *regs)
544 {
545 	if (addr < TASK_SIZE)
546 		return do_page_fault(addr, esr, regs);
547 
548 	do_bad_area(addr, esr, regs);
549 	return 0;
550 }
551 
552 static int do_alignment_fault(unsigned long addr, unsigned int esr,
553 			      struct pt_regs *regs)
554 {
555 	do_bad_area(addr, esr, regs);
556 	return 0;
557 }
558 
559 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
560 {
561 	return 1; /* "fault" */
562 }
563 
564 static int do_sea(unsigned long addr, unsigned int esr, struct pt_regs *regs)
565 {
566 	struct siginfo info;
567 	const struct fault_info *inf;
568 
569 	inf = esr_to_fault_info(esr);
570 
571 	/*
572 	 * Synchronous aborts may interrupt code which had interrupts masked.
573 	 * Before calling out into the wider kernel tell the interested
574 	 * subsystems.
575 	 */
576 	if (IS_ENABLED(CONFIG_ACPI_APEI_SEA)) {
577 		if (interrupts_enabled(regs))
578 			nmi_enter();
579 
580 		ghes_notify_sea();
581 
582 		if (interrupts_enabled(regs))
583 			nmi_exit();
584 	}
585 
586 	info.si_signo = inf->sig;
587 	info.si_errno = 0;
588 	info.si_code  = inf->code;
589 	if (esr & ESR_ELx_FnV)
590 		info.si_addr = NULL;
591 	else
592 		info.si_addr  = (void __user *)addr;
593 	arm64_notify_die(inf->name, regs, &info, esr);
594 
595 	return 0;
596 }
597 
598 static const struct fault_info fault_info[] = {
599 	{ do_bad,		SIGKILL, SI_KERNEL,	"ttbr address size fault"	},
600 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 1 address size fault"	},
601 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 2 address size fault"	},
602 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 3 address size fault"	},
603 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
604 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
605 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
606 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
607 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 8"			},
608 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
609 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
610 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
611 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 12"			},
612 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
613 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
614 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
615 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous external abort"	},
616 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 17"			},
617 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 18"			},
618 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 19"			},
619 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 (translation table walk)"	},
620 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 (translation table walk)"	},
621 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 (translation table walk)"	},
622 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 (translation table walk)"	},
623 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous parity or ECC error" },	// Reserved when RAS is implemented
624 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 25"			},
625 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 26"			},
626 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 27"			},
627 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
628 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
629 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
630 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
631 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 32"			},
632 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
633 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 34"			},
634 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 35"			},
635 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 36"			},
636 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 37"			},
637 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 38"			},
638 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 39"			},
639 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 40"			},
640 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 41"			},
641 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 42"			},
642 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 43"			},
643 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 44"			},
644 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 45"			},
645 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 46"			},
646 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 47"			},
647 	{ do_bad,		SIGKILL, SI_KERNEL,	"TLB conflict abort"		},
648 	{ do_bad,		SIGKILL, SI_KERNEL,	"Unsupported atomic hardware update fault"	},
649 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 50"			},
650 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 51"			},
651 	{ do_bad,		SIGKILL, SI_KERNEL,	"implementation fault (lockdown abort)" },
652 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"implementation fault (unsupported exclusive)" },
653 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 54"			},
654 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 55"			},
655 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 56"			},
656 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 57"			},
657 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 58" 			},
658 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 59"			},
659 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 60"			},
660 	{ do_bad,		SIGKILL, SI_KERNEL,	"section domain fault"		},
661 	{ do_bad,		SIGKILL, SI_KERNEL,	"page domain fault"		},
662 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 63"			},
663 };
664 
665 int handle_guest_sea(phys_addr_t addr, unsigned int esr)
666 {
667 	int ret = -ENOENT;
668 
669 	if (IS_ENABLED(CONFIG_ACPI_APEI_SEA))
670 		ret = ghes_notify_sea();
671 
672 	return ret;
673 }
674 
675 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
676 					 struct pt_regs *regs)
677 {
678 	const struct fault_info *inf = esr_to_fault_info(esr);
679 	struct siginfo info;
680 
681 	if (!inf->fn(addr, esr, regs))
682 		return;
683 
684 	if (!user_mode(regs)) {
685 		pr_alert("Unhandled fault at 0x%016lx\n", addr);
686 		mem_abort_decode(esr);
687 		show_pte(addr);
688 	}
689 
690 	info.si_signo = inf->sig;
691 	info.si_errno = 0;
692 	info.si_code  = inf->code;
693 	info.si_addr  = (void __user *)addr;
694 	arm64_notify_die(inf->name, regs, &info, esr);
695 }
696 
697 asmlinkage void __exception do_el0_irq_bp_hardening(void)
698 {
699 	/* PC has already been checked in entry.S */
700 	arm64_apply_bp_hardening();
701 }
702 
703 asmlinkage void __exception do_el0_ia_bp_hardening(unsigned long addr,
704 						   unsigned int esr,
705 						   struct pt_regs *regs)
706 {
707 	/*
708 	 * We've taken an instruction abort from userspace and not yet
709 	 * re-enabled IRQs. If the address is a kernel address, apply
710 	 * BP hardening prior to enabling IRQs and pre-emption.
711 	 */
712 	if (addr > TASK_SIZE)
713 		arm64_apply_bp_hardening();
714 
715 	local_irq_enable();
716 	do_mem_abort(addr, esr, regs);
717 }
718 
719 
720 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
721 					   unsigned int esr,
722 					   struct pt_regs *regs)
723 {
724 	struct siginfo info;
725 
726 	if (user_mode(regs)) {
727 		if (instruction_pointer(regs) > TASK_SIZE)
728 			arm64_apply_bp_hardening();
729 		local_irq_enable();
730 	}
731 
732 	info.si_signo = SIGBUS;
733 	info.si_errno = 0;
734 	info.si_code  = BUS_ADRALN;
735 	info.si_addr  = (void __user *)addr;
736 	arm64_notify_die("SP/PC alignment exception", regs, &info, esr);
737 }
738 
739 int __init early_brk64(unsigned long addr, unsigned int esr,
740 		       struct pt_regs *regs);
741 
742 /*
743  * __refdata because early_brk64 is __init, but the reference to it is
744  * clobbered at arch_initcall time.
745  * See traps.c and debug-monitors.c:debug_traps_init().
746  */
747 static struct fault_info __refdata debug_fault_info[] = {
748 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
749 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
750 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
751 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 3"		},
752 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
753 	{ do_bad,	SIGKILL,	SI_KERNEL,	"aarch32 vector catch"	},
754 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
755 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 7"		},
756 };
757 
758 void __init hook_debug_fault_code(int nr,
759 				  int (*fn)(unsigned long, unsigned int, struct pt_regs *),
760 				  int sig, int code, const char *name)
761 {
762 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
763 
764 	debug_fault_info[nr].fn		= fn;
765 	debug_fault_info[nr].sig	= sig;
766 	debug_fault_info[nr].code	= code;
767 	debug_fault_info[nr].name	= name;
768 }
769 
770 asmlinkage int __exception do_debug_exception(unsigned long addr,
771 					      unsigned int esr,
772 					      struct pt_regs *regs)
773 {
774 	const struct fault_info *inf = debug_fault_info + DBG_ESR_EVT(esr);
775 	struct siginfo info;
776 	int rv;
777 
778 	/*
779 	 * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
780 	 * already disabled to preserve the last enabled/disabled addresses.
781 	 */
782 	if (interrupts_enabled(regs))
783 		trace_hardirqs_off();
784 
785 	if (user_mode(regs) && instruction_pointer(regs) > TASK_SIZE)
786 		arm64_apply_bp_hardening();
787 
788 	if (!inf->fn(addr, esr, regs)) {
789 		rv = 1;
790 	} else {
791 		info.si_signo = inf->sig;
792 		info.si_errno = 0;
793 		info.si_code  = inf->code;
794 		info.si_addr  = (void __user *)addr;
795 		arm64_notify_die(inf->name, regs, &info, esr);
796 		rv = 0;
797 	}
798 
799 	if (interrupts_enabled(regs))
800 		trace_hardirqs_on();
801 
802 	return rv;
803 }
804 NOKPROBE_SYMBOL(do_debug_exception);
805 
806 #ifdef CONFIG_ARM64_PAN
807 void cpu_enable_pan(const struct arm64_cpu_capabilities *__unused)
808 {
809 	/*
810 	 * We modify PSTATE. This won't work from irq context as the PSTATE
811 	 * is discarded once we return from the exception.
812 	 */
813 	WARN_ON_ONCE(in_interrupt());
814 
815 	config_sctlr_el1(SCTLR_EL1_SPAN, 0);
816 	asm(SET_PSTATE_PAN(1));
817 }
818 #endif /* CONFIG_ARM64_PAN */
819