xref: /openbmc/linux/arch/arm64/mm/fault.c (revision 60772e48)
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 
47 #include <acpi/ghes.h>
48 
49 struct fault_info {
50 	int	(*fn)(unsigned long addr, unsigned int esr,
51 		      struct pt_regs *regs);
52 	int	sig;
53 	int	code;
54 	const char *name;
55 };
56 
57 static const struct fault_info fault_info[];
58 
59 static inline const struct fault_info *esr_to_fault_info(unsigned int esr)
60 {
61 	return fault_info + (esr & 63);
62 }
63 
64 #ifdef CONFIG_KPROBES
65 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
66 {
67 	int ret = 0;
68 
69 	/* kprobe_running() needs smp_processor_id() */
70 	if (!user_mode(regs)) {
71 		preempt_disable();
72 		if (kprobe_running() && kprobe_fault_handler(regs, esr))
73 			ret = 1;
74 		preempt_enable();
75 	}
76 
77 	return ret;
78 }
79 #else
80 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
81 {
82 	return 0;
83 }
84 #endif
85 
86 static void data_abort_decode(unsigned int esr)
87 {
88 	pr_alert("Data abort info:\n");
89 
90 	if (esr & ESR_ELx_ISV) {
91 		pr_alert("  Access size = %u byte(s)\n",
92 			 1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
93 		pr_alert("  SSE = %lu, SRT = %lu\n",
94 			 (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
95 			 (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
96 		pr_alert("  SF = %lu, AR = %lu\n",
97 			 (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
98 			 (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
99 	} else {
100 		pr_alert("  ISV = 0, ISS = 0x%08lx\n", esr & ESR_ELx_ISS_MASK);
101 	}
102 
103 	pr_alert("  CM = %lu, WnR = %lu\n",
104 		 (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
105 		 (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT);
106 }
107 
108 static void mem_abort_decode(unsigned int esr)
109 {
110 	pr_alert("Mem abort info:\n");
111 
112 	pr_alert("  ESR = 0x%08x\n", esr);
113 	pr_alert("  Exception class = %s, IL = %u bits\n",
114 		 esr_get_class_string(esr),
115 		 (esr & ESR_ELx_IL) ? 32 : 16);
116 	pr_alert("  SET = %lu, FnV = %lu\n",
117 		 (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
118 		 (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
119 	pr_alert("  EA = %lu, S1PTW = %lu\n",
120 		 (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
121 		 (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
122 
123 	if (esr_is_data_abort(esr))
124 		data_abort_decode(esr);
125 }
126 
127 /*
128  * Dump out the page tables associated with 'addr' in the currently active mm.
129  */
130 void show_pte(unsigned long addr)
131 {
132 	struct mm_struct *mm;
133 	pgd_t *pgdp;
134 	pgd_t pgd;
135 
136 	if (addr < TASK_SIZE) {
137 		/* TTBR0 */
138 		mm = current->active_mm;
139 		if (mm == &init_mm) {
140 			pr_alert("[%016lx] user address but active_mm is swapper\n",
141 				 addr);
142 			return;
143 		}
144 	} else if (addr >= VA_START) {
145 		/* TTBR1 */
146 		mm = &init_mm;
147 	} else {
148 		pr_alert("[%016lx] address between user and kernel address ranges\n",
149 			 addr);
150 		return;
151 	}
152 
153 	pr_alert("%s pgtable: %luk pages, %u-bit VAs, pgdp = %p\n",
154 		 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
155 		 VA_BITS, mm->pgd);
156 	pgdp = pgd_offset(mm, addr);
157 	pgd = READ_ONCE(*pgdp);
158 	pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
159 
160 	do {
161 		pud_t *pudp, pud;
162 		pmd_t *pmdp, pmd;
163 		pte_t *ptep, pte;
164 
165 		if (pgd_none(pgd) || pgd_bad(pgd))
166 			break;
167 
168 		pudp = pud_offset(pgdp, addr);
169 		pud = READ_ONCE(*pudp);
170 		pr_cont(", pud=%016llx", pud_val(pud));
171 		if (pud_none(pud) || pud_bad(pud))
172 			break;
173 
174 		pmdp = pmd_offset(pudp, addr);
175 		pmd = READ_ONCE(*pmdp);
176 		pr_cont(", pmd=%016llx", pmd_val(pmd));
177 		if (pmd_none(pmd) || pmd_bad(pmd))
178 			break;
179 
180 		ptep = pte_offset_map(pmdp, addr);
181 		pte = READ_ONCE(*ptep);
182 		pr_cont(", pte=%016llx", pte_val(pte));
183 		pte_unmap(ptep);
184 	} while(0);
185 
186 	pr_cont("\n");
187 }
188 
189 /*
190  * This function sets the access flags (dirty, accessed), as well as write
191  * permission, and only to a more permissive setting.
192  *
193  * It needs to cope with hardware update of the accessed/dirty state by other
194  * agents in the system and can safely skip the __sync_icache_dcache() call as,
195  * like set_pte_at(), the PTE is never changed from no-exec to exec here.
196  *
197  * Returns whether or not the PTE actually changed.
198  */
199 int ptep_set_access_flags(struct vm_area_struct *vma,
200 			  unsigned long address, pte_t *ptep,
201 			  pte_t entry, int dirty)
202 {
203 	pteval_t old_pteval, pteval;
204 	pte_t pte = READ_ONCE(*ptep);
205 
206 	if (pte_same(pte, entry))
207 		return 0;
208 
209 	/* only preserve the access flags and write permission */
210 	pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
211 
212 	/*
213 	 * Setting the flags must be done atomically to avoid racing with the
214 	 * hardware update of the access/dirty state. The PTE_RDONLY bit must
215 	 * be set to the most permissive (lowest value) of *ptep and entry
216 	 * (calculated as: a & b == ~(~a | ~b)).
217 	 */
218 	pte_val(entry) ^= PTE_RDONLY;
219 	pteval = pte_val(pte);
220 	do {
221 		old_pteval = pteval;
222 		pteval ^= PTE_RDONLY;
223 		pteval |= pte_val(entry);
224 		pteval ^= PTE_RDONLY;
225 		pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
226 	} while (pteval != old_pteval);
227 
228 	flush_tlb_fix_spurious_fault(vma, address);
229 	return 1;
230 }
231 
232 static bool is_el1_instruction_abort(unsigned int esr)
233 {
234 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
235 }
236 
237 static inline bool is_permission_fault(unsigned int esr, struct pt_regs *regs,
238 				       unsigned long addr)
239 {
240 	unsigned int ec       = ESR_ELx_EC(esr);
241 	unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
242 
243 	if (ec != ESR_ELx_EC_DABT_CUR && ec != ESR_ELx_EC_IABT_CUR)
244 		return false;
245 
246 	if (fsc_type == ESR_ELx_FSC_PERM)
247 		return true;
248 
249 	if (addr < TASK_SIZE && system_uses_ttbr0_pan())
250 		return fsc_type == ESR_ELx_FSC_FAULT &&
251 			(regs->pstate & PSR_PAN_BIT);
252 
253 	return false;
254 }
255 
256 static void __do_kernel_fault(unsigned long addr, unsigned int esr,
257 			      struct pt_regs *regs)
258 {
259 	const char *msg;
260 
261 	/*
262 	 * Are we prepared to handle this kernel fault?
263 	 * We are almost certainly not prepared to handle instruction faults.
264 	 */
265 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
266 		return;
267 
268 	bust_spinlocks(1);
269 
270 	if (is_permission_fault(esr, regs, addr)) {
271 		if (esr & ESR_ELx_WNR)
272 			msg = "write to read-only memory";
273 		else
274 			msg = "read from unreadable memory";
275 	} else if (addr < PAGE_SIZE) {
276 		msg = "NULL pointer dereference";
277 	} else {
278 		msg = "paging request";
279 	}
280 
281 	pr_alert("Unable to handle kernel %s at virtual address %08lx\n", msg,
282 		 addr);
283 
284 	mem_abort_decode(esr);
285 
286 	show_pte(addr);
287 	die("Oops", regs, esr);
288 	bust_spinlocks(0);
289 	do_exit(SIGKILL);
290 }
291 
292 static void __do_user_fault(struct task_struct *tsk, unsigned long addr,
293 			    unsigned int esr, unsigned int sig, int code,
294 			    struct pt_regs *regs, int fault)
295 {
296 	struct siginfo si;
297 	const struct fault_info *inf;
298 	unsigned int lsb = 0;
299 
300 	if (unhandled_signal(tsk, sig) && show_unhandled_signals_ratelimited()) {
301 		inf = esr_to_fault_info(esr);
302 		pr_info("%s[%d]: unhandled %s (%d) at 0x%08lx, esr 0x%03x",
303 			tsk->comm, task_pid_nr(tsk), inf->name, sig,
304 			addr, esr);
305 		print_vma_addr(KERN_CONT ", in ", regs->pc);
306 		pr_cont("\n");
307 		__show_regs(regs);
308 	}
309 
310 	tsk->thread.fault_address = addr;
311 	tsk->thread.fault_code = esr;
312 	si.si_signo = sig;
313 	si.si_errno = 0;
314 	si.si_code = code;
315 	si.si_addr = (void __user *)addr;
316 	/*
317 	 * Either small page or large page may be poisoned.
318 	 * In other words, VM_FAULT_HWPOISON_LARGE and
319 	 * VM_FAULT_HWPOISON are mutually exclusive.
320 	 */
321 	if (fault & VM_FAULT_HWPOISON_LARGE)
322 		lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
323 	else if (fault & VM_FAULT_HWPOISON)
324 		lsb = PAGE_SHIFT;
325 	si.si_addr_lsb = lsb;
326 
327 	force_sig_info(sig, &si, tsk);
328 }
329 
330 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
331 {
332 	struct task_struct *tsk = current;
333 	const struct fault_info *inf;
334 
335 	/*
336 	 * If we are in kernel mode at this point, we have no context to
337 	 * handle this fault with.
338 	 */
339 	if (user_mode(regs)) {
340 		inf = esr_to_fault_info(esr);
341 		__do_user_fault(tsk, addr, esr, inf->sig, inf->code, regs, 0);
342 	} else
343 		__do_kernel_fault(addr, esr, regs);
344 }
345 
346 #define VM_FAULT_BADMAP		0x010000
347 #define VM_FAULT_BADACCESS	0x020000
348 
349 static int __do_page_fault(struct mm_struct *mm, unsigned long addr,
350 			   unsigned int mm_flags, unsigned long vm_flags,
351 			   struct task_struct *tsk)
352 {
353 	struct vm_area_struct *vma;
354 	int fault;
355 
356 	vma = find_vma(mm, addr);
357 	fault = VM_FAULT_BADMAP;
358 	if (unlikely(!vma))
359 		goto out;
360 	if (unlikely(vma->vm_start > addr))
361 		goto check_stack;
362 
363 	/*
364 	 * Ok, we have a good vm_area for this memory access, so we can handle
365 	 * it.
366 	 */
367 good_area:
368 	/*
369 	 * Check that the permissions on the VMA allow for the fault which
370 	 * occurred.
371 	 */
372 	if (!(vma->vm_flags & vm_flags)) {
373 		fault = VM_FAULT_BADACCESS;
374 		goto out;
375 	}
376 
377 	return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags);
378 
379 check_stack:
380 	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
381 		goto good_area;
382 out:
383 	return fault;
384 }
385 
386 static bool is_el0_instruction_abort(unsigned int esr)
387 {
388 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
389 }
390 
391 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
392 				   struct pt_regs *regs)
393 {
394 	struct task_struct *tsk;
395 	struct mm_struct *mm;
396 	int fault, sig, code, major = 0;
397 	unsigned long vm_flags = VM_READ | VM_WRITE;
398 	unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
399 
400 	if (notify_page_fault(regs, esr))
401 		return 0;
402 
403 	tsk = current;
404 	mm  = tsk->mm;
405 
406 	/*
407 	 * If we're in an interrupt or have no user context, we must not take
408 	 * the fault.
409 	 */
410 	if (faulthandler_disabled() || !mm)
411 		goto no_context;
412 
413 	if (user_mode(regs))
414 		mm_flags |= FAULT_FLAG_USER;
415 
416 	if (is_el0_instruction_abort(esr)) {
417 		vm_flags = VM_EXEC;
418 	} else if ((esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM)) {
419 		vm_flags = VM_WRITE;
420 		mm_flags |= FAULT_FLAG_WRITE;
421 	}
422 
423 	if (addr < TASK_SIZE && is_permission_fault(esr, regs, addr)) {
424 		/* regs->orig_addr_limit may be 0 if we entered from EL0 */
425 		if (regs->orig_addr_limit == KERNEL_DS)
426 			die("Accessing user space memory with fs=KERNEL_DS", regs, esr);
427 
428 		if (is_el1_instruction_abort(esr))
429 			die("Attempting to execute userspace memory", regs, esr);
430 
431 		if (!search_exception_tables(regs->pc))
432 			die("Accessing user space memory outside uaccess.h routines", regs, esr);
433 	}
434 
435 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
436 
437 	/*
438 	 * As per x86, we may deadlock here. However, since the kernel only
439 	 * validly references user space from well defined areas of the code,
440 	 * we can bug out early if this is from code which shouldn't.
441 	 */
442 	if (!down_read_trylock(&mm->mmap_sem)) {
443 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
444 			goto no_context;
445 retry:
446 		down_read(&mm->mmap_sem);
447 	} else {
448 		/*
449 		 * The above down_read_trylock() might have succeeded in which
450 		 * case, we'll have missed the might_sleep() from down_read().
451 		 */
452 		might_sleep();
453 #ifdef CONFIG_DEBUG_VM
454 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
455 			goto no_context;
456 #endif
457 	}
458 
459 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, tsk);
460 	major |= fault & VM_FAULT_MAJOR;
461 
462 	if (fault & VM_FAULT_RETRY) {
463 		/*
464 		 * If we need to retry but a fatal signal is pending,
465 		 * handle the signal first. We do not need to release
466 		 * the mmap_sem because it would already be released
467 		 * in __lock_page_or_retry in mm/filemap.c.
468 		 */
469 		if (fatal_signal_pending(current)) {
470 			if (!user_mode(regs))
471 				goto no_context;
472 			return 0;
473 		}
474 
475 		/*
476 		 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
477 		 * starvation.
478 		 */
479 		if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
480 			mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
481 			mm_flags |= FAULT_FLAG_TRIED;
482 			goto retry;
483 		}
484 	}
485 	up_read(&mm->mmap_sem);
486 
487 	/*
488 	 * Handle the "normal" (no error) case first.
489 	 */
490 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
491 			      VM_FAULT_BADACCESS)))) {
492 		/*
493 		 * Major/minor page fault accounting is only done
494 		 * once. If we go through a retry, it is extremely
495 		 * likely that the page will be found in page cache at
496 		 * that point.
497 		 */
498 		if (major) {
499 			tsk->maj_flt++;
500 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
501 				      addr);
502 		} else {
503 			tsk->min_flt++;
504 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
505 				      addr);
506 		}
507 
508 		return 0;
509 	}
510 
511 	/*
512 	 * If we are in kernel mode at this point, we have no context to
513 	 * handle this fault with.
514 	 */
515 	if (!user_mode(regs))
516 		goto no_context;
517 
518 	if (fault & VM_FAULT_OOM) {
519 		/*
520 		 * We ran out of memory, call the OOM killer, and return to
521 		 * userspace (which will retry the fault, or kill us if we got
522 		 * oom-killed).
523 		 */
524 		pagefault_out_of_memory();
525 		return 0;
526 	}
527 
528 	if (fault & VM_FAULT_SIGBUS) {
529 		/*
530 		 * We had some memory, but were unable to successfully fix up
531 		 * this page fault.
532 		 */
533 		sig = SIGBUS;
534 		code = BUS_ADRERR;
535 	} else if (fault & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE)) {
536 		sig = SIGBUS;
537 		code = BUS_MCEERR_AR;
538 	} else {
539 		/*
540 		 * Something tried to access memory that isn't in our memory
541 		 * map.
542 		 */
543 		sig = SIGSEGV;
544 		code = fault == VM_FAULT_BADACCESS ?
545 			SEGV_ACCERR : SEGV_MAPERR;
546 	}
547 
548 	__do_user_fault(tsk, addr, esr, sig, code, regs, fault);
549 	return 0;
550 
551 no_context:
552 	__do_kernel_fault(addr, esr, regs);
553 	return 0;
554 }
555 
556 static int __kprobes do_translation_fault(unsigned long addr,
557 					  unsigned int esr,
558 					  struct pt_regs *regs)
559 {
560 	if (addr < TASK_SIZE)
561 		return do_page_fault(addr, esr, regs);
562 
563 	do_bad_area(addr, esr, regs);
564 	return 0;
565 }
566 
567 static int do_alignment_fault(unsigned long addr, unsigned int esr,
568 			      struct pt_regs *regs)
569 {
570 	do_bad_area(addr, esr, regs);
571 	return 0;
572 }
573 
574 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
575 {
576 	return 1; /* "fault" */
577 }
578 
579 static int do_sea(unsigned long addr, unsigned int esr, struct pt_regs *regs)
580 {
581 	struct siginfo info;
582 	const struct fault_info *inf;
583 
584 	inf = esr_to_fault_info(esr);
585 	pr_err("Synchronous External Abort: %s (0x%08x) at 0x%016lx\n",
586 		inf->name, esr, addr);
587 
588 	/*
589 	 * Synchronous aborts may interrupt code which had interrupts masked.
590 	 * Before calling out into the wider kernel tell the interested
591 	 * subsystems.
592 	 */
593 	if (IS_ENABLED(CONFIG_ACPI_APEI_SEA)) {
594 		if (interrupts_enabled(regs))
595 			nmi_enter();
596 
597 		ghes_notify_sea();
598 
599 		if (interrupts_enabled(regs))
600 			nmi_exit();
601 	}
602 
603 	info.si_signo = SIGBUS;
604 	info.si_errno = 0;
605 	info.si_code  = BUS_FIXME;
606 	if (esr & ESR_ELx_FnV)
607 		info.si_addr = NULL;
608 	else
609 		info.si_addr  = (void __user *)addr;
610 	arm64_notify_die("", regs, &info, esr);
611 
612 	return 0;
613 }
614 
615 static const struct fault_info fault_info[] = {
616 	{ do_bad,		SIGBUS,  BUS_FIXME,	"ttbr address size fault"	},
617 	{ do_bad,		SIGBUS,  BUS_FIXME,	"level 1 address size fault"	},
618 	{ do_bad,		SIGBUS,  BUS_FIXME,	"level 2 address size fault"	},
619 	{ do_bad,		SIGBUS,  BUS_FIXME,	"level 3 address size fault"	},
620 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
621 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
622 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
623 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
624 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 8"			},
625 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
626 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
627 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
628 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 12"			},
629 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
630 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
631 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
632 	{ do_sea,		SIGBUS,  BUS_FIXME,	"synchronous external abort"	},
633 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 17"			},
634 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 18"			},
635 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 19"			},
636 	{ do_sea,		SIGBUS,  BUS_FIXME,	"level 0 (translation table walk)"	},
637 	{ do_sea,		SIGBUS,  BUS_FIXME,	"level 1 (translation table walk)"	},
638 	{ do_sea,		SIGBUS,  BUS_FIXME,	"level 2 (translation table walk)"	},
639 	{ do_sea,		SIGBUS,  BUS_FIXME,	"level 3 (translation table walk)"	},
640 	{ do_sea,		SIGBUS,  BUS_FIXME,	"synchronous parity or ECC error" },	// Reserved when RAS is implemented
641 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 25"			},
642 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 26"			},
643 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 27"			},
644 	{ do_sea,		SIGBUS,  BUS_FIXME,	"level 0 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
645 	{ do_sea,		SIGBUS,  BUS_FIXME,	"level 1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
646 	{ do_sea,		SIGBUS,  BUS_FIXME,	"level 2 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
647 	{ do_sea,		SIGBUS,  BUS_FIXME,	"level 3 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
648 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 32"			},
649 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
650 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 34"			},
651 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 35"			},
652 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 36"			},
653 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 37"			},
654 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 38"			},
655 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 39"			},
656 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 40"			},
657 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 41"			},
658 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 42"			},
659 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 43"			},
660 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 44"			},
661 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 45"			},
662 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 46"			},
663 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 47"			},
664 	{ do_bad,		SIGBUS,  BUS_FIXME,	"TLB conflict abort"		},
665 	{ do_bad,		SIGBUS,  BUS_FIXME,	"Unsupported atomic hardware update fault"	},
666 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 50"			},
667 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 51"			},
668 	{ do_bad,		SIGBUS,  BUS_FIXME,	"implementation fault (lockdown abort)" },
669 	{ do_bad,		SIGBUS,  BUS_FIXME,	"implementation fault (unsupported exclusive)" },
670 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 54"			},
671 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 55"			},
672 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 56"			},
673 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 57"			},
674 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 58" 			},
675 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 59"			},
676 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 60"			},
677 	{ do_bad,		SIGBUS,  BUS_FIXME,	"section domain fault"		},
678 	{ do_bad,		SIGBUS,  BUS_FIXME,	"page domain fault"		},
679 	{ do_bad,		SIGBUS,  BUS_FIXME,	"unknown 63"			},
680 };
681 
682 int handle_guest_sea(phys_addr_t addr, unsigned int esr)
683 {
684 	int ret = -ENOENT;
685 
686 	if (IS_ENABLED(CONFIG_ACPI_APEI_SEA))
687 		ret = ghes_notify_sea();
688 
689 	return ret;
690 }
691 
692 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
693 					 struct pt_regs *regs)
694 {
695 	const struct fault_info *inf = esr_to_fault_info(esr);
696 	struct siginfo info;
697 
698 	if (!inf->fn(addr, esr, regs))
699 		return;
700 
701 	pr_alert("Unhandled fault: %s at 0x%016lx\n",
702 		 inf->name, addr);
703 
704 	mem_abort_decode(esr);
705 
706 	if (!user_mode(regs))
707 		show_pte(addr);
708 
709 	info.si_signo = inf->sig;
710 	info.si_errno = 0;
711 	info.si_code  = inf->code;
712 	info.si_addr  = (void __user *)addr;
713 	arm64_notify_die("", regs, &info, esr);
714 }
715 
716 asmlinkage void __exception do_el0_irq_bp_hardening(void)
717 {
718 	/* PC has already been checked in entry.S */
719 	arm64_apply_bp_hardening();
720 }
721 
722 asmlinkage void __exception do_el0_ia_bp_hardening(unsigned long addr,
723 						   unsigned int esr,
724 						   struct pt_regs *regs)
725 {
726 	/*
727 	 * We've taken an instruction abort from userspace and not yet
728 	 * re-enabled IRQs. If the address is a kernel address, apply
729 	 * BP hardening prior to enabling IRQs and pre-emption.
730 	 */
731 	if (addr > TASK_SIZE)
732 		arm64_apply_bp_hardening();
733 
734 	local_irq_enable();
735 	do_mem_abort(addr, esr, regs);
736 }
737 
738 
739 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
740 					   unsigned int esr,
741 					   struct pt_regs *regs)
742 {
743 	struct siginfo info;
744 	struct task_struct *tsk = current;
745 
746 	if (user_mode(regs)) {
747 		if (instruction_pointer(regs) > TASK_SIZE)
748 			arm64_apply_bp_hardening();
749 		local_irq_enable();
750 	}
751 
752 	if (show_unhandled_signals && unhandled_signal(tsk, SIGBUS))
753 		pr_info_ratelimited("%s[%d]: %s exception: pc=%p sp=%p\n",
754 				    tsk->comm, task_pid_nr(tsk),
755 				    esr_get_class_string(esr), (void *)regs->pc,
756 				    (void *)regs->sp);
757 
758 	info.si_signo = SIGBUS;
759 	info.si_errno = 0;
760 	info.si_code  = BUS_ADRALN;
761 	info.si_addr  = (void __user *)addr;
762 	arm64_notify_die("Oops - SP/PC alignment exception", regs, &info, esr);
763 }
764 
765 int __init early_brk64(unsigned long addr, unsigned int esr,
766 		       struct pt_regs *regs);
767 
768 /*
769  * __refdata because early_brk64 is __init, but the reference to it is
770  * clobbered at arch_initcall time.
771  * See traps.c and debug-monitors.c:debug_traps_init().
772  */
773 static struct fault_info __refdata debug_fault_info[] = {
774 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
775 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
776 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
777 	{ do_bad,	SIGBUS,		BUS_FIXME,	"unknown 3"		},
778 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
779 	{ do_bad,	SIGTRAP,	TRAP_FIXME,	"aarch32 vector catch"	},
780 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
781 	{ do_bad,	SIGBUS,		BUS_FIXME,	"unknown 7"		},
782 };
783 
784 void __init hook_debug_fault_code(int nr,
785 				  int (*fn)(unsigned long, unsigned int, struct pt_regs *),
786 				  int sig, int code, const char *name)
787 {
788 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
789 
790 	debug_fault_info[nr].fn		= fn;
791 	debug_fault_info[nr].sig	= sig;
792 	debug_fault_info[nr].code	= code;
793 	debug_fault_info[nr].name	= name;
794 }
795 
796 asmlinkage int __exception do_debug_exception(unsigned long addr,
797 					      unsigned int esr,
798 					      struct pt_regs *regs)
799 {
800 	const struct fault_info *inf = debug_fault_info + DBG_ESR_EVT(esr);
801 	struct siginfo info;
802 	int rv;
803 
804 	/*
805 	 * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
806 	 * already disabled to preserve the last enabled/disabled addresses.
807 	 */
808 	if (interrupts_enabled(regs))
809 		trace_hardirqs_off();
810 
811 	if (user_mode(regs) && instruction_pointer(regs) > TASK_SIZE)
812 		arm64_apply_bp_hardening();
813 
814 	if (!inf->fn(addr, esr, regs)) {
815 		rv = 1;
816 	} else {
817 		pr_alert("Unhandled debug exception: %s (0x%08x) at 0x%016lx\n",
818 			 inf->name, esr, addr);
819 
820 		info.si_signo = inf->sig;
821 		info.si_errno = 0;
822 		info.si_code  = inf->code;
823 		info.si_addr  = (void __user *)addr;
824 		arm64_notify_die("", regs, &info, 0);
825 		rv = 0;
826 	}
827 
828 	if (interrupts_enabled(regs))
829 		trace_hardirqs_on();
830 
831 	return rv;
832 }
833 NOKPROBE_SYMBOL(do_debug_exception);
834 
835 #ifdef CONFIG_ARM64_PAN
836 int cpu_enable_pan(void *__unused)
837 {
838 	/*
839 	 * We modify PSTATE. This won't work from irq context as the PSTATE
840 	 * is discarded once we return from the exception.
841 	 */
842 	WARN_ON_ONCE(in_interrupt());
843 
844 	config_sctlr_el1(SCTLR_EL1_SPAN, 0);
845 	asm(SET_PSTATE_PAN(1));
846 	return 0;
847 }
848 #endif /* CONFIG_ARM64_PAN */
849