xref: /linux/arch/arm64/mm/fault.c (revision 78c3925c048c752334873f56c3a3d1c9d53e0416)
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/kfence.h>
14 #include <linux/signal.h>
15 #include <linux/mm.h>
16 #include <linux/hardirq.h>
17 #include <linux/init.h>
18 #include <linux/kasan.h>
19 #include <linux/kprobes.h>
20 #include <linux/uaccess.h>
21 #include <linux/page-flags.h>
22 #include <linux/sched/signal.h>
23 #include <linux/sched/debug.h>
24 #include <linux/highmem.h>
25 #include <linux/perf_event.h>
26 #include <linux/preempt.h>
27 #include <linux/hugetlb.h>
28 
29 #include <asm/acpi.h>
30 #include <asm/bug.h>
31 #include <asm/cmpxchg.h>
32 #include <asm/cpufeature.h>
33 #include <asm/efi.h>
34 #include <asm/exception.h>
35 #include <asm/daifflags.h>
36 #include <asm/debug-monitors.h>
37 #include <asm/esr.h>
38 #include <asm/kprobes.h>
39 #include <asm/mte.h>
40 #include <asm/processor.h>
41 #include <asm/sysreg.h>
42 #include <asm/system_misc.h>
43 #include <asm/tlbflush.h>
44 #include <asm/traps.h>
45 
46 struct fault_info {
47 	int	(*fn)(unsigned long far, unsigned long esr,
48 		      struct pt_regs *regs);
49 	int	sig;
50 	int	code;
51 	const char *name;
52 };
53 
54 static const struct fault_info fault_info[];
55 static struct fault_info debug_fault_info[];
56 
57 static inline const struct fault_info *esr_to_fault_info(unsigned long esr)
58 {
59 	return fault_info + (esr & ESR_ELx_FSC);
60 }
61 
62 static inline const struct fault_info *esr_to_debug_fault_info(unsigned long esr)
63 {
64 	return debug_fault_info + DBG_ESR_EVT(esr);
65 }
66 
67 static void data_abort_decode(unsigned long esr)
68 {
69 	unsigned long iss2 = ESR_ELx_ISS2(esr);
70 
71 	pr_alert("Data abort info:\n");
72 
73 	if (esr & ESR_ELx_ISV) {
74 		pr_alert("  Access size = %u byte(s)\n",
75 			 1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
76 		pr_alert("  SSE = %lu, SRT = %lu\n",
77 			 (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
78 			 (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
79 		pr_alert("  SF = %lu, AR = %lu\n",
80 			 (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
81 			 (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
82 	} else {
83 		pr_alert("  ISV = 0, ISS = 0x%08lx, ISS2 = 0x%08lx\n",
84 			 esr & ESR_ELx_ISS_MASK, iss2);
85 	}
86 
87 	pr_alert("  CM = %lu, WnR = %lu, TnD = %lu, TagAccess = %lu\n",
88 		 (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
89 		 (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT,
90 		 (iss2 & ESR_ELx_TnD) >> ESR_ELx_TnD_SHIFT,
91 		 (iss2 & ESR_ELx_TagAccess) >> ESR_ELx_TagAccess_SHIFT);
92 
93 	pr_alert("  GCS = %ld, Overlay = %lu, DirtyBit = %lu, Xs = %llu\n",
94 		 (iss2 & ESR_ELx_GCS) >> ESR_ELx_GCS_SHIFT,
95 		 (iss2 & ESR_ELx_Overlay) >> ESR_ELx_Overlay_SHIFT,
96 		 (iss2 & ESR_ELx_DirtyBit) >> ESR_ELx_DirtyBit_SHIFT,
97 		 (iss2 & ESR_ELx_Xs_MASK) >> ESR_ELx_Xs_SHIFT);
98 }
99 
100 static void mem_abort_decode(unsigned long esr)
101 {
102 	pr_alert("Mem abort info:\n");
103 
104 	pr_alert("  ESR = 0x%016lx\n", esr);
105 	pr_alert("  EC = 0x%02lx: %s, IL = %u bits\n",
106 		 ESR_ELx_EC(esr), esr_get_class_string(esr),
107 		 (esr & ESR_ELx_IL) ? 32 : 16);
108 	pr_alert("  SET = %lu, FnV = %lu\n",
109 		 (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
110 		 (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
111 	pr_alert("  EA = %lu, S1PTW = %lu\n",
112 		 (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
113 		 (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
114 	pr_alert("  FSC = 0x%02lx: %s\n", (esr & ESR_ELx_FSC),
115 		 esr_to_fault_info(esr)->name);
116 
117 	if (esr_is_data_abort(esr))
118 		data_abort_decode(esr);
119 }
120 
121 static inline unsigned long mm_to_pgd_phys(struct mm_struct *mm)
122 {
123 	/* Either init_pg_dir or swapper_pg_dir */
124 	if (mm == &init_mm)
125 		return __pa_symbol(mm->pgd);
126 
127 	return (unsigned long)virt_to_phys(mm->pgd);
128 }
129 
130 /*
131  * Dump out the page tables associated with 'addr' in the currently active mm.
132  */
133 static void show_pte(unsigned long addr)
134 {
135 	struct mm_struct *mm;
136 	pgd_t *pgdp;
137 	pgd_t pgd;
138 
139 	if (is_ttbr0_addr(addr)) {
140 		/* TTBR0 */
141 		mm = current->active_mm;
142 		if (mm == &init_mm) {
143 			pr_alert("[%016lx] user address but active_mm is swapper\n",
144 				 addr);
145 			return;
146 		}
147 	} else if (is_ttbr1_addr(addr)) {
148 		/* TTBR1 */
149 		mm = &init_mm;
150 	} else {
151 		pr_alert("[%016lx] address between user and kernel address ranges\n",
152 			 addr);
153 		return;
154 	}
155 
156 	pr_alert("%s pgtable: %luk pages, %llu-bit VAs, pgdp=%016lx\n",
157 		 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
158 		 vabits_actual, mm_to_pgd_phys(mm));
159 	pgdp = pgd_offset(mm, addr);
160 	pgd = READ_ONCE(*pgdp);
161 	pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
162 
163 	do {
164 		p4d_t *p4dp, p4d;
165 		pud_t *pudp, pud;
166 		pmd_t *pmdp, pmd;
167 		pte_t *ptep, pte;
168 
169 		if (pgd_none(pgd) || pgd_bad(pgd))
170 			break;
171 
172 		p4dp = p4d_offset(pgdp, addr);
173 		p4d = READ_ONCE(*p4dp);
174 		pr_cont(", p4d=%016llx", p4d_val(p4d));
175 		if (p4d_none(p4d) || p4d_bad(p4d))
176 			break;
177 
178 		pudp = pud_offset(p4dp, addr);
179 		pud = READ_ONCE(*pudp);
180 		pr_cont(", pud=%016llx", pud_val(pud));
181 		if (pud_none(pud) || pud_bad(pud))
182 			break;
183 
184 		pmdp = pmd_offset(pudp, addr);
185 		pmd = READ_ONCE(*pmdp);
186 		pr_cont(", pmd=%016llx", pmd_val(pmd));
187 		if (pmd_none(pmd) || pmd_bad(pmd))
188 			break;
189 
190 		ptep = pte_offset_map(pmdp, addr);
191 		if (!ptep)
192 			break;
193 
194 		pte = __ptep_get(ptep);
195 		pr_cont(", pte=%016llx", pte_val(pte));
196 		pte_unmap(ptep);
197 	} while(0);
198 
199 	pr_cont("\n");
200 }
201 
202 /*
203  * This function sets the access flags (dirty, accessed), as well as write
204  * permission, and only to a more permissive setting.
205  *
206  * It needs to cope with hardware update of the accessed/dirty state by other
207  * agents in the system and can safely skip the __sync_icache_dcache() call as,
208  * like __set_ptes(), the PTE is never changed from no-exec to exec here.
209  *
210  * Returns whether or not the PTE actually changed.
211  */
212 int __ptep_set_access_flags(struct vm_area_struct *vma,
213 			    unsigned long address, pte_t *ptep,
214 			    pte_t entry, int dirty)
215 {
216 	pteval_t old_pteval, pteval;
217 	pte_t pte = __ptep_get(ptep);
218 
219 	if (pte_same(pte, entry))
220 		return 0;
221 
222 	/* only preserve the access flags and write permission */
223 	pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
224 
225 	/*
226 	 * Setting the flags must be done atomically to avoid racing with the
227 	 * hardware update of the access/dirty state. The PTE_RDONLY bit must
228 	 * be set to the most permissive (lowest value) of *ptep and entry
229 	 * (calculated as: a & b == ~(~a | ~b)).
230 	 */
231 	pte_val(entry) ^= PTE_RDONLY;
232 	pteval = pte_val(pte);
233 	do {
234 		old_pteval = pteval;
235 		pteval ^= PTE_RDONLY;
236 		pteval |= pte_val(entry);
237 		pteval ^= PTE_RDONLY;
238 		pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
239 	} while (pteval != old_pteval);
240 
241 	/* Invalidate a stale read-only entry */
242 	if (dirty)
243 		flush_tlb_page(vma, address);
244 	return 1;
245 }
246 
247 static bool is_el1_instruction_abort(unsigned long esr)
248 {
249 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
250 }
251 
252 static bool is_el1_data_abort(unsigned long esr)
253 {
254 	return ESR_ELx_EC(esr) == ESR_ELx_EC_DABT_CUR;
255 }
256 
257 static inline bool is_el1_permission_fault(unsigned long addr, unsigned long esr,
258 					   struct pt_regs *regs)
259 {
260 	if (!is_el1_data_abort(esr) && !is_el1_instruction_abort(esr))
261 		return false;
262 
263 	if (esr_fsc_is_permission_fault(esr))
264 		return true;
265 
266 	if (is_ttbr0_addr(addr) && system_uses_ttbr0_pan())
267 		return esr_fsc_is_translation_fault(esr) &&
268 			(regs->pstate & PSR_PAN_BIT);
269 
270 	return false;
271 }
272 
273 static bool __kprobes is_spurious_el1_translation_fault(unsigned long addr,
274 							unsigned long esr,
275 							struct pt_regs *regs)
276 {
277 	unsigned long flags;
278 	u64 par, dfsc;
279 
280 	if (!is_el1_data_abort(esr) || !esr_fsc_is_translation_fault(esr))
281 		return false;
282 
283 	local_irq_save(flags);
284 	asm volatile("at s1e1r, %0" :: "r" (addr));
285 	isb();
286 	par = read_sysreg_par();
287 	local_irq_restore(flags);
288 
289 	/*
290 	 * If we now have a valid translation, treat the translation fault as
291 	 * spurious.
292 	 */
293 	if (!(par & SYS_PAR_EL1_F))
294 		return true;
295 
296 	/*
297 	 * If we got a different type of fault from the AT instruction,
298 	 * treat the translation fault as spurious.
299 	 */
300 	dfsc = FIELD_GET(SYS_PAR_EL1_FST, par);
301 	return !esr_fsc_is_translation_fault(dfsc);
302 }
303 
304 static void die_kernel_fault(const char *msg, unsigned long addr,
305 			     unsigned long esr, struct pt_regs *regs)
306 {
307 	bust_spinlocks(1);
308 
309 	pr_alert("Unable to handle kernel %s at virtual address %016lx\n", msg,
310 		 addr);
311 
312 	kasan_non_canonical_hook(addr);
313 
314 	mem_abort_decode(esr);
315 
316 	show_pte(addr);
317 	die("Oops", regs, esr);
318 	bust_spinlocks(0);
319 	make_task_dead(SIGKILL);
320 }
321 
322 #ifdef CONFIG_KASAN_HW_TAGS
323 static void report_tag_fault(unsigned long addr, unsigned long esr,
324 			     struct pt_regs *regs)
325 {
326 	/*
327 	 * SAS bits aren't set for all faults reported in EL1, so we can't
328 	 * find out access size.
329 	 */
330 	bool is_write = !!(esr & ESR_ELx_WNR);
331 	kasan_report((void *)addr, 0, is_write, regs->pc);
332 }
333 #else
334 /* Tag faults aren't enabled without CONFIG_KASAN_HW_TAGS. */
335 static inline void report_tag_fault(unsigned long addr, unsigned long esr,
336 				    struct pt_regs *regs) { }
337 #endif
338 
339 static void do_tag_recovery(unsigned long addr, unsigned long esr,
340 			   struct pt_regs *regs)
341 {
342 
343 	report_tag_fault(addr, esr, regs);
344 
345 	/*
346 	 * Disable MTE Tag Checking on the local CPU for the current EL.
347 	 * It will be done lazily on the other CPUs when they will hit a
348 	 * tag fault.
349 	 */
350 	sysreg_clear_set(sctlr_el1, SCTLR_EL1_TCF_MASK,
351 			 SYS_FIELD_PREP_ENUM(SCTLR_EL1, TCF, NONE));
352 	isb();
353 }
354 
355 static bool is_el1_mte_sync_tag_check_fault(unsigned long esr)
356 {
357 	unsigned long fsc = esr & ESR_ELx_FSC;
358 
359 	if (!is_el1_data_abort(esr))
360 		return false;
361 
362 	if (fsc == ESR_ELx_FSC_MTE)
363 		return true;
364 
365 	return false;
366 }
367 
368 static void __do_kernel_fault(unsigned long addr, unsigned long esr,
369 			      struct pt_regs *regs)
370 {
371 	const char *msg;
372 
373 	/*
374 	 * Are we prepared to handle this kernel fault?
375 	 * We are almost certainly not prepared to handle instruction faults.
376 	 */
377 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
378 		return;
379 
380 	if (WARN_RATELIMIT(is_spurious_el1_translation_fault(addr, esr, regs),
381 	    "Ignoring spurious kernel translation fault at virtual address %016lx\n", addr))
382 		return;
383 
384 	if (is_el1_mte_sync_tag_check_fault(esr)) {
385 		do_tag_recovery(addr, esr, regs);
386 
387 		return;
388 	}
389 
390 	if (is_el1_permission_fault(addr, esr, regs)) {
391 		if (esr & ESR_ELx_WNR)
392 			msg = "write to read-only memory";
393 		else if (is_el1_instruction_abort(esr))
394 			msg = "execute from non-executable memory";
395 		else
396 			msg = "read from unreadable memory";
397 	} else if (addr < PAGE_SIZE) {
398 		msg = "NULL pointer dereference";
399 	} else {
400 		if (esr_fsc_is_translation_fault(esr) &&
401 		    kfence_handle_page_fault(addr, esr & ESR_ELx_WNR, regs))
402 			return;
403 
404 		msg = "paging request";
405 	}
406 
407 	if (efi_runtime_fixup_exception(regs, msg))
408 		return;
409 
410 	die_kernel_fault(msg, addr, esr, regs);
411 }
412 
413 static void set_thread_esr(unsigned long address, unsigned long esr)
414 {
415 	current->thread.fault_address = address;
416 
417 	/*
418 	 * If the faulting address is in the kernel, we must sanitize the ESR.
419 	 * From userspace's point of view, kernel-only mappings don't exist
420 	 * at all, so we report them as level 0 translation faults.
421 	 * (This is not quite the way that "no mapping there at all" behaves:
422 	 * an alignment fault not caused by the memory type would take
423 	 * precedence over translation fault for a real access to empty
424 	 * space. Unfortunately we can't easily distinguish "alignment fault
425 	 * not caused by memory type" from "alignment fault caused by memory
426 	 * type", so we ignore this wrinkle and just return the translation
427 	 * fault.)
428 	 */
429 	if (!is_ttbr0_addr(current->thread.fault_address)) {
430 		switch (ESR_ELx_EC(esr)) {
431 		case ESR_ELx_EC_DABT_LOW:
432 			/*
433 			 * These bits provide only information about the
434 			 * faulting instruction, which userspace knows already.
435 			 * We explicitly clear bits which are architecturally
436 			 * RES0 in case they are given meanings in future.
437 			 * We always report the ESR as if the fault was taken
438 			 * to EL1 and so ISV and the bits in ISS[23:14] are
439 			 * clear. (In fact it always will be a fault to EL1.)
440 			 */
441 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL |
442 				ESR_ELx_CM | ESR_ELx_WNR;
443 			esr |= ESR_ELx_FSC_FAULT;
444 			break;
445 		case ESR_ELx_EC_IABT_LOW:
446 			/*
447 			 * Claim a level 0 translation fault.
448 			 * All other bits are architecturally RES0 for faults
449 			 * reported with that DFSC value, so we clear them.
450 			 */
451 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL;
452 			esr |= ESR_ELx_FSC_FAULT;
453 			break;
454 		default:
455 			/*
456 			 * This should never happen (entry.S only brings us
457 			 * into this code for insn and data aborts from a lower
458 			 * exception level). Fail safe by not providing an ESR
459 			 * context record at all.
460 			 */
461 			WARN(1, "ESR 0x%lx is not DABT or IABT from EL0\n", esr);
462 			esr = 0;
463 			break;
464 		}
465 	}
466 
467 	current->thread.fault_code = esr;
468 }
469 
470 static void do_bad_area(unsigned long far, unsigned long esr,
471 			struct pt_regs *regs)
472 {
473 	unsigned long addr = untagged_addr(far);
474 
475 	/*
476 	 * If we are in kernel mode at this point, we have no context to
477 	 * handle this fault with.
478 	 */
479 	if (user_mode(regs)) {
480 		const struct fault_info *inf = esr_to_fault_info(esr);
481 
482 		set_thread_esr(addr, esr);
483 		arm64_force_sig_fault(inf->sig, inf->code, far, inf->name);
484 	} else {
485 		__do_kernel_fault(addr, esr, regs);
486 	}
487 }
488 
489 #define VM_FAULT_BADMAP		((__force vm_fault_t)0x010000)
490 #define VM_FAULT_BADACCESS	((__force vm_fault_t)0x020000)
491 
492 static vm_fault_t __do_page_fault(struct mm_struct *mm,
493 				  struct vm_area_struct *vma, unsigned long addr,
494 				  unsigned int mm_flags, unsigned long vm_flags,
495 				  struct pt_regs *regs)
496 {
497 	/*
498 	 * Ok, we have a good vm_area for this memory access, so we can handle
499 	 * it.
500 	 * Check that the permissions on the VMA allow for the fault which
501 	 * occurred.
502 	 */
503 	if (!(vma->vm_flags & vm_flags))
504 		return VM_FAULT_BADACCESS;
505 	return handle_mm_fault(vma, addr, mm_flags, regs);
506 }
507 
508 static bool is_el0_instruction_abort(unsigned long esr)
509 {
510 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
511 }
512 
513 /*
514  * Note: not valid for EL1 DC IVAC, but we never use that such that it
515  * should fault. EL0 cannot issue DC IVAC (undef).
516  */
517 static bool is_write_abort(unsigned long esr)
518 {
519 	return (esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM);
520 }
521 
522 static int __kprobes do_page_fault(unsigned long far, unsigned long esr,
523 				   struct pt_regs *regs)
524 {
525 	const struct fault_info *inf;
526 	struct mm_struct *mm = current->mm;
527 	vm_fault_t fault;
528 	unsigned long vm_flags;
529 	unsigned int mm_flags = FAULT_FLAG_DEFAULT;
530 	unsigned long addr = untagged_addr(far);
531 	struct vm_area_struct *vma;
532 
533 	if (kprobe_page_fault(regs, esr))
534 		return 0;
535 
536 	/*
537 	 * If we're in an interrupt or have no user context, we must not take
538 	 * the fault.
539 	 */
540 	if (faulthandler_disabled() || !mm)
541 		goto no_context;
542 
543 	if (user_mode(regs))
544 		mm_flags |= FAULT_FLAG_USER;
545 
546 	/*
547 	 * vm_flags tells us what bits we must have in vma->vm_flags
548 	 * for the fault to be benign, __do_page_fault() would check
549 	 * vma->vm_flags & vm_flags and returns an error if the
550 	 * intersection is empty
551 	 */
552 	if (is_el0_instruction_abort(esr)) {
553 		/* It was exec fault */
554 		vm_flags = VM_EXEC;
555 		mm_flags |= FAULT_FLAG_INSTRUCTION;
556 	} else if (is_write_abort(esr)) {
557 		/* It was write fault */
558 		vm_flags = VM_WRITE;
559 		mm_flags |= FAULT_FLAG_WRITE;
560 	} else {
561 		/* It was read fault */
562 		vm_flags = VM_READ;
563 		/* Write implies read */
564 		vm_flags |= VM_WRITE;
565 		/* If EPAN is absent then exec implies read */
566 		if (!alternative_has_cap_unlikely(ARM64_HAS_EPAN))
567 			vm_flags |= VM_EXEC;
568 	}
569 
570 	if (is_ttbr0_addr(addr) && is_el1_permission_fault(addr, esr, regs)) {
571 		if (is_el1_instruction_abort(esr))
572 			die_kernel_fault("execution of user memory",
573 					 addr, esr, regs);
574 
575 		if (!search_exception_tables(regs->pc))
576 			die_kernel_fault("access to user memory outside uaccess routines",
577 					 addr, esr, regs);
578 	}
579 
580 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
581 
582 	if (!(mm_flags & FAULT_FLAG_USER))
583 		goto lock_mmap;
584 
585 	vma = lock_vma_under_rcu(mm, addr);
586 	if (!vma)
587 		goto lock_mmap;
588 
589 	if (!(vma->vm_flags & vm_flags)) {
590 		vma_end_read(vma);
591 		goto lock_mmap;
592 	}
593 	fault = handle_mm_fault(vma, addr, mm_flags | FAULT_FLAG_VMA_LOCK, regs);
594 	if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
595 		vma_end_read(vma);
596 
597 	if (!(fault & VM_FAULT_RETRY)) {
598 		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
599 		goto done;
600 	}
601 	count_vm_vma_lock_event(VMA_LOCK_RETRY);
602 	if (fault & VM_FAULT_MAJOR)
603 		mm_flags |= FAULT_FLAG_TRIED;
604 
605 	/* Quick path to respond to signals */
606 	if (fault_signal_pending(fault, regs)) {
607 		if (!user_mode(regs))
608 			goto no_context;
609 		return 0;
610 	}
611 lock_mmap:
612 
613 retry:
614 	vma = lock_mm_and_find_vma(mm, addr, regs);
615 	if (unlikely(!vma)) {
616 		fault = VM_FAULT_BADMAP;
617 		goto done;
618 	}
619 
620 	fault = __do_page_fault(mm, vma, addr, mm_flags, vm_flags, regs);
621 
622 	/* Quick path to respond to signals */
623 	if (fault_signal_pending(fault, regs)) {
624 		if (!user_mode(regs))
625 			goto no_context;
626 		return 0;
627 	}
628 
629 	/* The fault is fully completed (including releasing mmap lock) */
630 	if (fault & VM_FAULT_COMPLETED)
631 		return 0;
632 
633 	if (fault & VM_FAULT_RETRY) {
634 		mm_flags |= FAULT_FLAG_TRIED;
635 		goto retry;
636 	}
637 	mmap_read_unlock(mm);
638 
639 done:
640 	/*
641 	 * Handle the "normal" (no error) case first.
642 	 */
643 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
644 			      VM_FAULT_BADACCESS))))
645 		return 0;
646 
647 	/*
648 	 * If we are in kernel mode at this point, we have no context to
649 	 * handle this fault with.
650 	 */
651 	if (!user_mode(regs))
652 		goto no_context;
653 
654 	if (fault & VM_FAULT_OOM) {
655 		/*
656 		 * We ran out of memory, call the OOM killer, and return to
657 		 * userspace (which will retry the fault, or kill us if we got
658 		 * oom-killed).
659 		 */
660 		pagefault_out_of_memory();
661 		return 0;
662 	}
663 
664 	inf = esr_to_fault_info(esr);
665 	set_thread_esr(addr, esr);
666 	if (fault & VM_FAULT_SIGBUS) {
667 		/*
668 		 * We had some memory, but were unable to successfully fix up
669 		 * this page fault.
670 		 */
671 		arm64_force_sig_fault(SIGBUS, BUS_ADRERR, far, inf->name);
672 	} else if (fault & (VM_FAULT_HWPOISON_LARGE | VM_FAULT_HWPOISON)) {
673 		unsigned int lsb;
674 
675 		lsb = PAGE_SHIFT;
676 		if (fault & VM_FAULT_HWPOISON_LARGE)
677 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
678 
679 		arm64_force_sig_mceerr(BUS_MCEERR_AR, far, lsb, inf->name);
680 	} else {
681 		/*
682 		 * Something tried to access memory that isn't in our memory
683 		 * map.
684 		 */
685 		arm64_force_sig_fault(SIGSEGV,
686 				      fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR,
687 				      far, inf->name);
688 	}
689 
690 	return 0;
691 
692 no_context:
693 	__do_kernel_fault(addr, esr, regs);
694 	return 0;
695 }
696 
697 static int __kprobes do_translation_fault(unsigned long far,
698 					  unsigned long esr,
699 					  struct pt_regs *regs)
700 {
701 	unsigned long addr = untagged_addr(far);
702 
703 	if (is_ttbr0_addr(addr))
704 		return do_page_fault(far, esr, regs);
705 
706 	do_bad_area(far, esr, regs);
707 	return 0;
708 }
709 
710 static int do_alignment_fault(unsigned long far, unsigned long esr,
711 			      struct pt_regs *regs)
712 {
713 	if (IS_ENABLED(CONFIG_COMPAT_ALIGNMENT_FIXUPS) &&
714 	    compat_user_mode(regs))
715 		return do_compat_alignment_fixup(far, regs);
716 	do_bad_area(far, esr, regs);
717 	return 0;
718 }
719 
720 static int do_bad(unsigned long far, unsigned long esr, struct pt_regs *regs)
721 {
722 	return 1; /* "fault" */
723 }
724 
725 static int do_sea(unsigned long far, unsigned long esr, struct pt_regs *regs)
726 {
727 	const struct fault_info *inf;
728 	unsigned long siaddr;
729 
730 	inf = esr_to_fault_info(esr);
731 
732 	if (user_mode(regs) && apei_claim_sea(regs) == 0) {
733 		/*
734 		 * APEI claimed this as a firmware-first notification.
735 		 * Some processing deferred to task_work before ret_to_user().
736 		 */
737 		return 0;
738 	}
739 
740 	if (esr & ESR_ELx_FnV) {
741 		siaddr = 0;
742 	} else {
743 		/*
744 		 * The architecture specifies that the tag bits of FAR_EL1 are
745 		 * UNKNOWN for synchronous external aborts. Mask them out now
746 		 * so that userspace doesn't see them.
747 		 */
748 		siaddr  = untagged_addr(far);
749 	}
750 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
751 
752 	return 0;
753 }
754 
755 static int do_tag_check_fault(unsigned long far, unsigned long esr,
756 			      struct pt_regs *regs)
757 {
758 	/*
759 	 * The architecture specifies that bits 63:60 of FAR_EL1 are UNKNOWN
760 	 * for tag check faults. Set them to corresponding bits in the untagged
761 	 * address.
762 	 */
763 	far = (__untagged_addr(far) & ~MTE_TAG_MASK) | (far & MTE_TAG_MASK);
764 	do_bad_area(far, esr, regs);
765 	return 0;
766 }
767 
768 static const struct fault_info fault_info[] = {
769 	{ do_bad,		SIGKILL, SI_KERNEL,	"ttbr address size fault"	},
770 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 1 address size fault"	},
771 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 2 address size fault"	},
772 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 3 address size fault"	},
773 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
774 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
775 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
776 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
777 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 0 access flag fault"	},
778 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
779 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
780 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
781 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 0 permission fault"	},
782 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
783 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
784 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
785 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous external abort"	},
786 	{ do_tag_check_fault,	SIGSEGV, SEGV_MTESERR,	"synchronous tag check fault"	},
787 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 18"			},
788 	{ do_sea,		SIGKILL, SI_KERNEL,	"level -1 (translation table walk)"	},
789 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 (translation table walk)"	},
790 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 (translation table walk)"	},
791 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 (translation table walk)"	},
792 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 (translation table walk)"	},
793 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous parity or ECC error" },	// Reserved when RAS is implemented
794 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 25"			},
795 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 26"			},
796 	{ do_sea,		SIGKILL, SI_KERNEL,	"level -1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
797 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
798 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
799 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
800 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
801 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 32"			},
802 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
803 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 34"			},
804 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 35"			},
805 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 36"			},
806 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 37"			},
807 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 38"			},
808 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 39"			},
809 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 40"			},
810 	{ do_bad,		SIGKILL, SI_KERNEL,	"level -1 address size fault"	},
811 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 42"			},
812 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level -1 translation fault"	},
813 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 44"			},
814 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 45"			},
815 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 46"			},
816 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 47"			},
817 	{ do_bad,		SIGKILL, SI_KERNEL,	"TLB conflict abort"		},
818 	{ do_bad,		SIGKILL, SI_KERNEL,	"Unsupported atomic hardware update fault"	},
819 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 50"			},
820 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 51"			},
821 	{ do_bad,		SIGKILL, SI_KERNEL,	"implementation fault (lockdown abort)" },
822 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"implementation fault (unsupported exclusive)" },
823 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 54"			},
824 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 55"			},
825 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 56"			},
826 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 57"			},
827 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 58" 			},
828 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 59"			},
829 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 60"			},
830 	{ do_bad,		SIGKILL, SI_KERNEL,	"section domain fault"		},
831 	{ do_bad,		SIGKILL, SI_KERNEL,	"page domain fault"		},
832 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 63"			},
833 };
834 
835 void do_mem_abort(unsigned long far, unsigned long esr, struct pt_regs *regs)
836 {
837 	const struct fault_info *inf = esr_to_fault_info(esr);
838 	unsigned long addr = untagged_addr(far);
839 
840 	if (!inf->fn(far, esr, regs))
841 		return;
842 
843 	if (!user_mode(regs))
844 		die_kernel_fault(inf->name, addr, esr, regs);
845 
846 	/*
847 	 * At this point we have an unrecognized fault type whose tag bits may
848 	 * have been defined as UNKNOWN. Therefore we only expose the untagged
849 	 * address to the signal handler.
850 	 */
851 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, addr, esr);
852 }
853 NOKPROBE_SYMBOL(do_mem_abort);
854 
855 void do_sp_pc_abort(unsigned long addr, unsigned long esr, struct pt_regs *regs)
856 {
857 	arm64_notify_die("SP/PC alignment exception", regs, SIGBUS, BUS_ADRALN,
858 			 addr, esr);
859 }
860 NOKPROBE_SYMBOL(do_sp_pc_abort);
861 
862 /*
863  * __refdata because early_brk64 is __init, but the reference to it is
864  * clobbered at arch_initcall time.
865  * See traps.c and debug-monitors.c:debug_traps_init().
866  */
867 static struct fault_info __refdata debug_fault_info[] = {
868 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
869 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
870 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
871 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 3"		},
872 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
873 	{ do_bad,	SIGKILL,	SI_KERNEL,	"aarch32 vector catch"	},
874 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
875 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 7"		},
876 };
877 
878 void __init hook_debug_fault_code(int nr,
879 				  int (*fn)(unsigned long, unsigned long, struct pt_regs *),
880 				  int sig, int code, const char *name)
881 {
882 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
883 
884 	debug_fault_info[nr].fn		= fn;
885 	debug_fault_info[nr].sig	= sig;
886 	debug_fault_info[nr].code	= code;
887 	debug_fault_info[nr].name	= name;
888 }
889 
890 /*
891  * In debug exception context, we explicitly disable preemption despite
892  * having interrupts disabled.
893  * This serves two purposes: it makes it much less likely that we would
894  * accidentally schedule in exception context and it will force a warning
895  * if we somehow manage to schedule by accident.
896  */
897 static void debug_exception_enter(struct pt_regs *regs)
898 {
899 	preempt_disable();
900 
901 	/* This code is a bit fragile.  Test it. */
902 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "exception_enter didn't work");
903 }
904 NOKPROBE_SYMBOL(debug_exception_enter);
905 
906 static void debug_exception_exit(struct pt_regs *regs)
907 {
908 	preempt_enable_no_resched();
909 }
910 NOKPROBE_SYMBOL(debug_exception_exit);
911 
912 void do_debug_exception(unsigned long addr_if_watchpoint, unsigned long esr,
913 			struct pt_regs *regs)
914 {
915 	const struct fault_info *inf = esr_to_debug_fault_info(esr);
916 	unsigned long pc = instruction_pointer(regs);
917 
918 	debug_exception_enter(regs);
919 
920 	if (user_mode(regs) && !is_ttbr0_addr(pc))
921 		arm64_apply_bp_hardening();
922 
923 	if (inf->fn(addr_if_watchpoint, esr, regs)) {
924 		arm64_notify_die(inf->name, regs, inf->sig, inf->code, pc, esr);
925 	}
926 
927 	debug_exception_exit(regs);
928 }
929 NOKPROBE_SYMBOL(do_debug_exception);
930 
931 /*
932  * Used during anonymous page fault handling.
933  */
934 struct folio *vma_alloc_zeroed_movable_folio(struct vm_area_struct *vma,
935 						unsigned long vaddr)
936 {
937 	gfp_t flags = GFP_HIGHUSER_MOVABLE | __GFP_ZERO;
938 
939 	/*
940 	 * If the page is mapped with PROT_MTE, initialise the tags at the
941 	 * point of allocation and page zeroing as this is usually faster than
942 	 * separate DC ZVA and STGM.
943 	 */
944 	if (vma->vm_flags & VM_MTE)
945 		flags |= __GFP_ZEROTAGS;
946 
947 	return vma_alloc_folio(flags, 0, vma, vaddr, false);
948 }
949 
950 void tag_clear_highpage(struct page *page)
951 {
952 	/* Newly allocated page, shouldn't have been tagged yet */
953 	WARN_ON_ONCE(!try_page_mte_tagging(page));
954 	mte_zero_clear_page_tags(page_address(page));
955 	set_page_mte_tagged(page);
956 }
957