xref: /linux/arch/arm64/mm/fault.c (revision 2993c9b04e616df0848b655d7202a707a70fc876)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Based on arch/arm/mm/fault.c
4  *
5  * Copyright (C) 1995  Linus Torvalds
6  * Copyright (C) 1995-2004 Russell King
7  * Copyright (C) 2012 ARM Ltd.
8  */
9 
10 #include <linux/acpi.h>
11 #include <linux/bitfield.h>
12 #include <linux/extable.h>
13 #include <linux/signal.h>
14 #include <linux/mm.h>
15 #include <linux/hardirq.h>
16 #include <linux/init.h>
17 #include <linux/kprobes.h>
18 #include <linux/uaccess.h>
19 #include <linux/page-flags.h>
20 #include <linux/sched/signal.h>
21 #include <linux/sched/debug.h>
22 #include <linux/highmem.h>
23 #include <linux/perf_event.h>
24 #include <linux/preempt.h>
25 #include <linux/hugetlb.h>
26 
27 #include <asm/acpi.h>
28 #include <asm/bug.h>
29 #include <asm/cmpxchg.h>
30 #include <asm/cpufeature.h>
31 #include <asm/exception.h>
32 #include <asm/daifflags.h>
33 #include <asm/debug-monitors.h>
34 #include <asm/esr.h>
35 #include <asm/kasan.h>
36 #include <asm/sysreg.h>
37 #include <asm/system_misc.h>
38 #include <asm/pgtable.h>
39 #include <asm/tlbflush.h>
40 #include <asm/traps.h>
41 
42 struct fault_info {
43 	int	(*fn)(unsigned long addr, unsigned int esr,
44 		      struct pt_regs *regs);
45 	int	sig;
46 	int	code;
47 	const char *name;
48 };
49 
50 static const struct fault_info fault_info[];
51 static struct fault_info debug_fault_info[];
52 
53 static inline const struct fault_info *esr_to_fault_info(unsigned int esr)
54 {
55 	return fault_info + (esr & ESR_ELx_FSC);
56 }
57 
58 static inline const struct fault_info *esr_to_debug_fault_info(unsigned int esr)
59 {
60 	return debug_fault_info + DBG_ESR_EVT(esr);
61 }
62 
63 static void data_abort_decode(unsigned int esr)
64 {
65 	pr_alert("Data abort info:\n");
66 
67 	if (esr & ESR_ELx_ISV) {
68 		pr_alert("  Access size = %u byte(s)\n",
69 			 1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
70 		pr_alert("  SSE = %lu, SRT = %lu\n",
71 			 (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
72 			 (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
73 		pr_alert("  SF = %lu, AR = %lu\n",
74 			 (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
75 			 (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
76 	} else {
77 		pr_alert("  ISV = 0, ISS = 0x%08lx\n", esr & ESR_ELx_ISS_MASK);
78 	}
79 
80 	pr_alert("  CM = %lu, WnR = %lu\n",
81 		 (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
82 		 (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT);
83 }
84 
85 static void mem_abort_decode(unsigned int esr)
86 {
87 	pr_alert("Mem abort info:\n");
88 
89 	pr_alert("  ESR = 0x%08x\n", esr);
90 	pr_alert("  EC = 0x%02lx: %s, IL = %u bits\n",
91 		 ESR_ELx_EC(esr), esr_get_class_string(esr),
92 		 (esr & ESR_ELx_IL) ? 32 : 16);
93 	pr_alert("  SET = %lu, FnV = %lu\n",
94 		 (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
95 		 (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
96 	pr_alert("  EA = %lu, S1PTW = %lu\n",
97 		 (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
98 		 (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
99 
100 	if (esr_is_data_abort(esr))
101 		data_abort_decode(esr);
102 }
103 
104 static inline bool is_ttbr0_addr(unsigned long addr)
105 {
106 	/* entry assembly clears tags for TTBR0 addrs */
107 	return addr < TASK_SIZE;
108 }
109 
110 static inline bool is_ttbr1_addr(unsigned long addr)
111 {
112 	/* TTBR1 addresses may have a tag if KASAN_SW_TAGS is in use */
113 	return arch_kasan_reset_tag(addr) >= PAGE_OFFSET;
114 }
115 
116 static inline unsigned long mm_to_pgd_phys(struct mm_struct *mm)
117 {
118 	/* Either init_pg_dir or swapper_pg_dir */
119 	if (mm == &init_mm)
120 		return __pa_symbol(mm->pgd);
121 
122 	return (unsigned long)virt_to_phys(mm->pgd);
123 }
124 
125 /*
126  * Dump out the page tables associated with 'addr' in the currently active mm.
127  */
128 static void show_pte(unsigned long addr)
129 {
130 	struct mm_struct *mm;
131 	pgd_t *pgdp;
132 	pgd_t pgd;
133 
134 	if (is_ttbr0_addr(addr)) {
135 		/* TTBR0 */
136 		mm = current->active_mm;
137 		if (mm == &init_mm) {
138 			pr_alert("[%016lx] user address but active_mm is swapper\n",
139 				 addr);
140 			return;
141 		}
142 	} else if (is_ttbr1_addr(addr)) {
143 		/* TTBR1 */
144 		mm = &init_mm;
145 	} else {
146 		pr_alert("[%016lx] address between user and kernel address ranges\n",
147 			 addr);
148 		return;
149 	}
150 
151 	pr_alert("%s pgtable: %luk pages, %llu-bit VAs, pgdp=%016lx\n",
152 		 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
153 		 vabits_actual, mm_to_pgd_phys(mm));
154 	pgdp = pgd_offset(mm, addr);
155 	pgd = READ_ONCE(*pgdp);
156 	pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
157 
158 	do {
159 		pud_t *pudp, pud;
160 		pmd_t *pmdp, pmd;
161 		pte_t *ptep, pte;
162 
163 		if (pgd_none(pgd) || pgd_bad(pgd))
164 			break;
165 
166 		pudp = pud_offset(pgdp, addr);
167 		pud = READ_ONCE(*pudp);
168 		pr_cont(", pud=%016llx", pud_val(pud));
169 		if (pud_none(pud) || pud_bad(pud))
170 			break;
171 
172 		pmdp = pmd_offset(pudp, addr);
173 		pmd = READ_ONCE(*pmdp);
174 		pr_cont(", pmd=%016llx", pmd_val(pmd));
175 		if (pmd_none(pmd) || pmd_bad(pmd))
176 			break;
177 
178 		ptep = pte_offset_map(pmdp, addr);
179 		pte = READ_ONCE(*ptep);
180 		pr_cont(", pte=%016llx", pte_val(pte));
181 		pte_unmap(ptep);
182 	} while(0);
183 
184 	pr_cont("\n");
185 }
186 
187 /*
188  * This function sets the access flags (dirty, accessed), as well as write
189  * permission, and only to a more permissive setting.
190  *
191  * It needs to cope with hardware update of the accessed/dirty state by other
192  * agents in the system and can safely skip the __sync_icache_dcache() call as,
193  * like set_pte_at(), the PTE is never changed from no-exec to exec here.
194  *
195  * Returns whether or not the PTE actually changed.
196  */
197 int ptep_set_access_flags(struct vm_area_struct *vma,
198 			  unsigned long address, pte_t *ptep,
199 			  pte_t entry, int dirty)
200 {
201 	pteval_t old_pteval, pteval;
202 	pte_t pte = READ_ONCE(*ptep);
203 
204 	if (pte_same(pte, entry))
205 		return 0;
206 
207 	/* only preserve the access flags and write permission */
208 	pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
209 
210 	/*
211 	 * Setting the flags must be done atomically to avoid racing with the
212 	 * hardware update of the access/dirty state. The PTE_RDONLY bit must
213 	 * be set to the most permissive (lowest value) of *ptep and entry
214 	 * (calculated as: a & b == ~(~a | ~b)).
215 	 */
216 	pte_val(entry) ^= PTE_RDONLY;
217 	pteval = pte_val(pte);
218 	do {
219 		old_pteval = pteval;
220 		pteval ^= PTE_RDONLY;
221 		pteval |= pte_val(entry);
222 		pteval ^= PTE_RDONLY;
223 		pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
224 	} while (pteval != old_pteval);
225 
226 	flush_tlb_fix_spurious_fault(vma, address);
227 	return 1;
228 }
229 
230 static bool is_el1_instruction_abort(unsigned int esr)
231 {
232 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
233 }
234 
235 static inline bool is_el1_permission_fault(unsigned long addr, unsigned int esr,
236 					   struct pt_regs *regs)
237 {
238 	unsigned int ec       = ESR_ELx_EC(esr);
239 	unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
240 
241 	if (ec != ESR_ELx_EC_DABT_CUR && ec != ESR_ELx_EC_IABT_CUR)
242 		return false;
243 
244 	if (fsc_type == ESR_ELx_FSC_PERM)
245 		return true;
246 
247 	if (is_ttbr0_addr(addr) && system_uses_ttbr0_pan())
248 		return fsc_type == ESR_ELx_FSC_FAULT &&
249 			(regs->pstate & PSR_PAN_BIT);
250 
251 	return false;
252 }
253 
254 static bool __kprobes is_spurious_el1_translation_fault(unsigned long addr,
255 							unsigned int esr,
256 							struct pt_regs *regs)
257 {
258 	unsigned long flags;
259 	u64 par, dfsc;
260 
261 	if (ESR_ELx_EC(esr) != ESR_ELx_EC_DABT_CUR ||
262 	    (esr & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT)
263 		return false;
264 
265 	local_irq_save(flags);
266 	asm volatile("at s1e1r, %0" :: "r" (addr));
267 	isb();
268 	par = read_sysreg(par_el1);
269 	local_irq_restore(flags);
270 
271 	if (!(par & SYS_PAR_EL1_F))
272 		return false;
273 
274 	/*
275 	 * If we got a different type of fault from the AT instruction,
276 	 * treat the translation fault as spurious.
277 	 */
278 	dfsc = FIELD_GET(SYS_PAR_EL1_FST, par);
279 	return (dfsc & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT;
280 }
281 
282 static void die_kernel_fault(const char *msg, unsigned long addr,
283 			     unsigned int esr, struct pt_regs *regs)
284 {
285 	bust_spinlocks(1);
286 
287 	pr_alert("Unable to handle kernel %s at virtual address %016lx\n", msg,
288 		 addr);
289 
290 	mem_abort_decode(esr);
291 
292 	show_pte(addr);
293 	die("Oops", regs, esr);
294 	bust_spinlocks(0);
295 	do_exit(SIGKILL);
296 }
297 
298 static void __do_kernel_fault(unsigned long addr, unsigned int esr,
299 			      struct pt_regs *regs)
300 {
301 	const char *msg;
302 
303 	/*
304 	 * Are we prepared to handle this kernel fault?
305 	 * We are almost certainly not prepared to handle instruction faults.
306 	 */
307 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
308 		return;
309 
310 	if (WARN_RATELIMIT(is_spurious_el1_translation_fault(addr, esr, regs),
311 	    "Ignoring spurious kernel translation fault at virtual address %016lx\n", addr))
312 		return;
313 
314 	if (is_el1_permission_fault(addr, esr, regs)) {
315 		if (esr & ESR_ELx_WNR)
316 			msg = "write to read-only memory";
317 		else
318 			msg = "read from unreadable memory";
319 	} else if (addr < PAGE_SIZE) {
320 		msg = "NULL pointer dereference";
321 	} else {
322 		msg = "paging request";
323 	}
324 
325 	die_kernel_fault(msg, addr, esr, regs);
326 }
327 
328 static void set_thread_esr(unsigned long address, unsigned int esr)
329 {
330 	current->thread.fault_address = address;
331 
332 	/*
333 	 * If the faulting address is in the kernel, we must sanitize the ESR.
334 	 * From userspace's point of view, kernel-only mappings don't exist
335 	 * at all, so we report them as level 0 translation faults.
336 	 * (This is not quite the way that "no mapping there at all" behaves:
337 	 * an alignment fault not caused by the memory type would take
338 	 * precedence over translation fault for a real access to empty
339 	 * space. Unfortunately we can't easily distinguish "alignment fault
340 	 * not caused by memory type" from "alignment fault caused by memory
341 	 * type", so we ignore this wrinkle and just return the translation
342 	 * fault.)
343 	 */
344 	if (!is_ttbr0_addr(current->thread.fault_address)) {
345 		switch (ESR_ELx_EC(esr)) {
346 		case ESR_ELx_EC_DABT_LOW:
347 			/*
348 			 * These bits provide only information about the
349 			 * faulting instruction, which userspace knows already.
350 			 * We explicitly clear bits which are architecturally
351 			 * RES0 in case they are given meanings in future.
352 			 * We always report the ESR as if the fault was taken
353 			 * to EL1 and so ISV and the bits in ISS[23:14] are
354 			 * clear. (In fact it always will be a fault to EL1.)
355 			 */
356 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL |
357 				ESR_ELx_CM | ESR_ELx_WNR;
358 			esr |= ESR_ELx_FSC_FAULT;
359 			break;
360 		case ESR_ELx_EC_IABT_LOW:
361 			/*
362 			 * Claim a level 0 translation fault.
363 			 * All other bits are architecturally RES0 for faults
364 			 * reported with that DFSC value, so we clear them.
365 			 */
366 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL;
367 			esr |= ESR_ELx_FSC_FAULT;
368 			break;
369 		default:
370 			/*
371 			 * This should never happen (entry.S only brings us
372 			 * into this code for insn and data aborts from a lower
373 			 * exception level). Fail safe by not providing an ESR
374 			 * context record at all.
375 			 */
376 			WARN(1, "ESR 0x%x is not DABT or IABT from EL0\n", esr);
377 			esr = 0;
378 			break;
379 		}
380 	}
381 
382 	current->thread.fault_code = esr;
383 }
384 
385 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
386 {
387 	/*
388 	 * If we are in kernel mode at this point, we have no context to
389 	 * handle this fault with.
390 	 */
391 	if (user_mode(regs)) {
392 		const struct fault_info *inf = esr_to_fault_info(esr);
393 
394 		set_thread_esr(addr, esr);
395 		arm64_force_sig_fault(inf->sig, inf->code, (void __user *)addr,
396 				      inf->name);
397 	} else {
398 		__do_kernel_fault(addr, esr, regs);
399 	}
400 }
401 
402 #define VM_FAULT_BADMAP		0x010000
403 #define VM_FAULT_BADACCESS	0x020000
404 
405 static vm_fault_t __do_page_fault(struct mm_struct *mm, unsigned long addr,
406 			   unsigned int mm_flags, unsigned long vm_flags)
407 {
408 	struct vm_area_struct *vma = find_vma(mm, addr);
409 
410 	if (unlikely(!vma))
411 		return VM_FAULT_BADMAP;
412 
413 	/*
414 	 * Ok, we have a good vm_area for this memory access, so we can handle
415 	 * it.
416 	 */
417 	if (unlikely(vma->vm_start > addr)) {
418 		if (!(vma->vm_flags & VM_GROWSDOWN))
419 			return VM_FAULT_BADMAP;
420 		if (expand_stack(vma, addr))
421 			return VM_FAULT_BADMAP;
422 	}
423 
424 	/*
425 	 * Check that the permissions on the VMA allow for the fault which
426 	 * occurred.
427 	 */
428 	if (!(vma->vm_flags & vm_flags))
429 		return VM_FAULT_BADACCESS;
430 	return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags);
431 }
432 
433 static bool is_el0_instruction_abort(unsigned int esr)
434 {
435 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
436 }
437 
438 /*
439  * Note: not valid for EL1 DC IVAC, but we never use that such that it
440  * should fault. EL0 cannot issue DC IVAC (undef).
441  */
442 static bool is_write_abort(unsigned int esr)
443 {
444 	return (esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM);
445 }
446 
447 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
448 				   struct pt_regs *regs)
449 {
450 	const struct fault_info *inf;
451 	struct mm_struct *mm = current->mm;
452 	vm_fault_t fault, major = 0;
453 	unsigned long vm_flags = VM_READ | VM_WRITE;
454 	unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
455 
456 	if (kprobe_page_fault(regs, esr))
457 		return 0;
458 
459 	/*
460 	 * If we're in an interrupt or have no user context, we must not take
461 	 * the fault.
462 	 */
463 	if (faulthandler_disabled() || !mm)
464 		goto no_context;
465 
466 	if (user_mode(regs))
467 		mm_flags |= FAULT_FLAG_USER;
468 
469 	if (is_el0_instruction_abort(esr)) {
470 		vm_flags = VM_EXEC;
471 		mm_flags |= FAULT_FLAG_INSTRUCTION;
472 	} else if (is_write_abort(esr)) {
473 		vm_flags = VM_WRITE;
474 		mm_flags |= FAULT_FLAG_WRITE;
475 	}
476 
477 	if (is_ttbr0_addr(addr) && is_el1_permission_fault(addr, esr, regs)) {
478 		/* regs->orig_addr_limit may be 0 if we entered from EL0 */
479 		if (regs->orig_addr_limit == KERNEL_DS)
480 			die_kernel_fault("access to user memory with fs=KERNEL_DS",
481 					 addr, esr, regs);
482 
483 		if (is_el1_instruction_abort(esr))
484 			die_kernel_fault("execution of user memory",
485 					 addr, esr, regs);
486 
487 		if (!search_exception_tables(regs->pc))
488 			die_kernel_fault("access to user memory outside uaccess routines",
489 					 addr, esr, regs);
490 	}
491 
492 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
493 
494 	/*
495 	 * As per x86, we may deadlock here. However, since the kernel only
496 	 * validly references user space from well defined areas of the code,
497 	 * we can bug out early if this is from code which shouldn't.
498 	 */
499 	if (!down_read_trylock(&mm->mmap_sem)) {
500 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
501 			goto no_context;
502 retry:
503 		down_read(&mm->mmap_sem);
504 	} else {
505 		/*
506 		 * The above down_read_trylock() might have succeeded in which
507 		 * case, we'll have missed the might_sleep() from down_read().
508 		 */
509 		might_sleep();
510 #ifdef CONFIG_DEBUG_VM
511 		if (!user_mode(regs) && !search_exception_tables(regs->pc)) {
512 			up_read(&mm->mmap_sem);
513 			goto no_context;
514 		}
515 #endif
516 	}
517 
518 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags);
519 	major |= fault & VM_FAULT_MAJOR;
520 
521 	if (fault & VM_FAULT_RETRY) {
522 		/*
523 		 * If we need to retry but a fatal signal is pending,
524 		 * handle the signal first. We do not need to release
525 		 * the mmap_sem because it would already be released
526 		 * in __lock_page_or_retry in mm/filemap.c.
527 		 */
528 		if (fatal_signal_pending(current)) {
529 			if (!user_mode(regs))
530 				goto no_context;
531 			return 0;
532 		}
533 
534 		/*
535 		 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
536 		 * starvation.
537 		 */
538 		if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
539 			mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
540 			mm_flags |= FAULT_FLAG_TRIED;
541 			goto retry;
542 		}
543 	}
544 	up_read(&mm->mmap_sem);
545 
546 	/*
547 	 * Handle the "normal" (no error) case first.
548 	 */
549 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
550 			      VM_FAULT_BADACCESS)))) {
551 		/*
552 		 * Major/minor page fault accounting is only done
553 		 * once. If we go through a retry, it is extremely
554 		 * likely that the page will be found in page cache at
555 		 * that point.
556 		 */
557 		if (major) {
558 			current->maj_flt++;
559 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
560 				      addr);
561 		} else {
562 			current->min_flt++;
563 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
564 				      addr);
565 		}
566 
567 		return 0;
568 	}
569 
570 	/*
571 	 * If we are in kernel mode at this point, we have no context to
572 	 * handle this fault with.
573 	 */
574 	if (!user_mode(regs))
575 		goto no_context;
576 
577 	if (fault & VM_FAULT_OOM) {
578 		/*
579 		 * We ran out of memory, call the OOM killer, and return to
580 		 * userspace (which will retry the fault, or kill us if we got
581 		 * oom-killed).
582 		 */
583 		pagefault_out_of_memory();
584 		return 0;
585 	}
586 
587 	inf = esr_to_fault_info(esr);
588 	set_thread_esr(addr, esr);
589 	if (fault & VM_FAULT_SIGBUS) {
590 		/*
591 		 * We had some memory, but were unable to successfully fix up
592 		 * this page fault.
593 		 */
594 		arm64_force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)addr,
595 				      inf->name);
596 	} else if (fault & (VM_FAULT_HWPOISON_LARGE | VM_FAULT_HWPOISON)) {
597 		unsigned int lsb;
598 
599 		lsb = PAGE_SHIFT;
600 		if (fault & VM_FAULT_HWPOISON_LARGE)
601 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
602 
603 		arm64_force_sig_mceerr(BUS_MCEERR_AR, (void __user *)addr, lsb,
604 				       inf->name);
605 	} else {
606 		/*
607 		 * Something tried to access memory that isn't in our memory
608 		 * map.
609 		 */
610 		arm64_force_sig_fault(SIGSEGV,
611 				      fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR,
612 				      (void __user *)addr,
613 				      inf->name);
614 	}
615 
616 	return 0;
617 
618 no_context:
619 	__do_kernel_fault(addr, esr, regs);
620 	return 0;
621 }
622 
623 static int __kprobes do_translation_fault(unsigned long addr,
624 					  unsigned int esr,
625 					  struct pt_regs *regs)
626 {
627 	if (is_ttbr0_addr(addr))
628 		return do_page_fault(addr, esr, regs);
629 
630 	do_bad_area(addr, esr, regs);
631 	return 0;
632 }
633 
634 static int do_alignment_fault(unsigned long addr, unsigned int esr,
635 			      struct pt_regs *regs)
636 {
637 	do_bad_area(addr, esr, regs);
638 	return 0;
639 }
640 
641 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
642 {
643 	return 1; /* "fault" */
644 }
645 
646 static int do_sea(unsigned long addr, unsigned int esr, struct pt_regs *regs)
647 {
648 	const struct fault_info *inf;
649 	void __user *siaddr;
650 
651 	inf = esr_to_fault_info(esr);
652 
653 	/*
654 	 * Return value ignored as we rely on signal merging.
655 	 * Future patches will make this more robust.
656 	 */
657 	apei_claim_sea(regs);
658 
659 	if (esr & ESR_ELx_FnV)
660 		siaddr = NULL;
661 	else
662 		siaddr  = (void __user *)addr;
663 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
664 
665 	return 0;
666 }
667 
668 static const struct fault_info fault_info[] = {
669 	{ do_bad,		SIGKILL, SI_KERNEL,	"ttbr address size fault"	},
670 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 1 address size fault"	},
671 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 2 address size fault"	},
672 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 3 address size fault"	},
673 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
674 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
675 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
676 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
677 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 8"			},
678 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
679 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
680 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
681 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 12"			},
682 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
683 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
684 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
685 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous external abort"	},
686 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 17"			},
687 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 18"			},
688 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 19"			},
689 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 (translation table walk)"	},
690 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 (translation table walk)"	},
691 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 (translation table walk)"	},
692 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 (translation table walk)"	},
693 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous parity or ECC error" },	// Reserved when RAS is implemented
694 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 25"			},
695 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 26"			},
696 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 27"			},
697 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
698 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
699 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
700 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
701 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 32"			},
702 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
703 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 34"			},
704 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 35"			},
705 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 36"			},
706 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 37"			},
707 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 38"			},
708 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 39"			},
709 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 40"			},
710 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 41"			},
711 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 42"			},
712 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 43"			},
713 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 44"			},
714 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 45"			},
715 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 46"			},
716 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 47"			},
717 	{ do_bad,		SIGKILL, SI_KERNEL,	"TLB conflict abort"		},
718 	{ do_bad,		SIGKILL, SI_KERNEL,	"Unsupported atomic hardware update fault"	},
719 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 50"			},
720 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 51"			},
721 	{ do_bad,		SIGKILL, SI_KERNEL,	"implementation fault (lockdown abort)" },
722 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"implementation fault (unsupported exclusive)" },
723 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 54"			},
724 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 55"			},
725 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 56"			},
726 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 57"			},
727 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 58" 			},
728 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 59"			},
729 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 60"			},
730 	{ do_bad,		SIGKILL, SI_KERNEL,	"section domain fault"		},
731 	{ do_bad,		SIGKILL, SI_KERNEL,	"page domain fault"		},
732 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 63"			},
733 };
734 
735 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
736 					 struct pt_regs *regs)
737 {
738 	const struct fault_info *inf = esr_to_fault_info(esr);
739 
740 	if (!inf->fn(addr, esr, regs))
741 		return;
742 
743 	if (!user_mode(regs)) {
744 		pr_alert("Unhandled fault at 0x%016lx\n", addr);
745 		mem_abort_decode(esr);
746 		show_pte(addr);
747 	}
748 
749 	arm64_notify_die(inf->name, regs,
750 			 inf->sig, inf->code, (void __user *)addr, esr);
751 }
752 
753 asmlinkage void __exception do_el0_irq_bp_hardening(void)
754 {
755 	/* PC has already been checked in entry.S */
756 	arm64_apply_bp_hardening();
757 }
758 
759 asmlinkage void __exception do_el0_ia_bp_hardening(unsigned long addr,
760 						   unsigned int esr,
761 						   struct pt_regs *regs)
762 {
763 	/*
764 	 * We've taken an instruction abort from userspace and not yet
765 	 * re-enabled IRQs. If the address is a kernel address, apply
766 	 * BP hardening prior to enabling IRQs and pre-emption.
767 	 */
768 	if (!is_ttbr0_addr(addr))
769 		arm64_apply_bp_hardening();
770 
771 	local_daif_restore(DAIF_PROCCTX);
772 	do_mem_abort(addr, esr, regs);
773 }
774 
775 
776 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
777 					   unsigned int esr,
778 					   struct pt_regs *regs)
779 {
780 	if (user_mode(regs)) {
781 		if (!is_ttbr0_addr(instruction_pointer(regs)))
782 			arm64_apply_bp_hardening();
783 		local_daif_restore(DAIF_PROCCTX);
784 	}
785 
786 	arm64_notify_die("SP/PC alignment exception", regs,
787 			 SIGBUS, BUS_ADRALN, (void __user *)addr, esr);
788 }
789 
790 int __init early_brk64(unsigned long addr, unsigned int esr,
791 		       struct pt_regs *regs);
792 
793 /*
794  * __refdata because early_brk64 is __init, but the reference to it is
795  * clobbered at arch_initcall time.
796  * See traps.c and debug-monitors.c:debug_traps_init().
797  */
798 static struct fault_info __refdata debug_fault_info[] = {
799 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
800 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
801 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
802 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 3"		},
803 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
804 	{ do_bad,	SIGKILL,	SI_KERNEL,	"aarch32 vector catch"	},
805 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
806 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 7"		},
807 };
808 
809 void __init hook_debug_fault_code(int nr,
810 				  int (*fn)(unsigned long, unsigned int, struct pt_regs *),
811 				  int sig, int code, const char *name)
812 {
813 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
814 
815 	debug_fault_info[nr].fn		= fn;
816 	debug_fault_info[nr].sig	= sig;
817 	debug_fault_info[nr].code	= code;
818 	debug_fault_info[nr].name	= name;
819 }
820 
821 /*
822  * In debug exception context, we explicitly disable preemption despite
823  * having interrupts disabled.
824  * This serves two purposes: it makes it much less likely that we would
825  * accidentally schedule in exception context and it will force a warning
826  * if we somehow manage to schedule by accident.
827  */
828 static void debug_exception_enter(struct pt_regs *regs)
829 {
830 	/*
831 	 * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
832 	 * already disabled to preserve the last enabled/disabled addresses.
833 	 */
834 	if (interrupts_enabled(regs))
835 		trace_hardirqs_off();
836 
837 	if (user_mode(regs)) {
838 		RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
839 	} else {
840 		/*
841 		 * We might have interrupted pretty much anything.  In
842 		 * fact, if we're a debug exception, we can even interrupt
843 		 * NMI processing. We don't want this code makes in_nmi()
844 		 * to return true, but we need to notify RCU.
845 		 */
846 		rcu_nmi_enter();
847 	}
848 
849 	preempt_disable();
850 
851 	/* This code is a bit fragile.  Test it. */
852 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "exception_enter didn't work");
853 }
854 NOKPROBE_SYMBOL(debug_exception_enter);
855 
856 static void debug_exception_exit(struct pt_regs *regs)
857 {
858 	preempt_enable_no_resched();
859 
860 	if (!user_mode(regs))
861 		rcu_nmi_exit();
862 
863 	if (interrupts_enabled(regs))
864 		trace_hardirqs_on();
865 }
866 NOKPROBE_SYMBOL(debug_exception_exit);
867 
868 #ifdef CONFIG_ARM64_ERRATUM_1463225
869 DECLARE_PER_CPU(int, __in_cortex_a76_erratum_1463225_wa);
870 
871 static int __exception
872 cortex_a76_erratum_1463225_debug_handler(struct pt_regs *regs)
873 {
874 	if (user_mode(regs))
875 		return 0;
876 
877 	if (!__this_cpu_read(__in_cortex_a76_erratum_1463225_wa))
878 		return 0;
879 
880 	/*
881 	 * We've taken a dummy step exception from the kernel to ensure
882 	 * that interrupts are re-enabled on the syscall path. Return back
883 	 * to cortex_a76_erratum_1463225_svc_handler() with debug exceptions
884 	 * masked so that we can safely restore the mdscr and get on with
885 	 * handling the syscall.
886 	 */
887 	regs->pstate |= PSR_D_BIT;
888 	return 1;
889 }
890 #else
891 static int __exception
892 cortex_a76_erratum_1463225_debug_handler(struct pt_regs *regs)
893 {
894 	return 0;
895 }
896 #endif /* CONFIG_ARM64_ERRATUM_1463225 */
897 
898 asmlinkage void __exception do_debug_exception(unsigned long addr_if_watchpoint,
899 					       unsigned int esr,
900 					       struct pt_regs *regs)
901 {
902 	const struct fault_info *inf = esr_to_debug_fault_info(esr);
903 	unsigned long pc = instruction_pointer(regs);
904 
905 	if (cortex_a76_erratum_1463225_debug_handler(regs))
906 		return;
907 
908 	debug_exception_enter(regs);
909 
910 	if (user_mode(regs) && !is_ttbr0_addr(pc))
911 		arm64_apply_bp_hardening();
912 
913 	if (inf->fn(addr_if_watchpoint, esr, regs)) {
914 		arm64_notify_die(inf->name, regs,
915 				 inf->sig, inf->code, (void __user *)pc, esr);
916 	}
917 
918 	debug_exception_exit(regs);
919 }
920 NOKPROBE_SYMBOL(do_debug_exception);
921