xref: /linux/arch/arm64/mm/fault.c (revision 320fefa9e2edc67011e235ea1d50f0d00ddfe004)
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 	pr_alert("Data abort info:\n");
70 
71 	if (esr & ESR_ELx_ISV) {
72 		pr_alert("  Access size = %u byte(s)\n",
73 			 1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
74 		pr_alert("  SSE = %lu, SRT = %lu\n",
75 			 (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
76 			 (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
77 		pr_alert("  SF = %lu, AR = %lu\n",
78 			 (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
79 			 (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
80 	} else {
81 		pr_alert("  ISV = 0, ISS = 0x%08lx\n", esr & ESR_ELx_ISS_MASK);
82 	}
83 
84 	pr_alert("  CM = %lu, WnR = %lu\n",
85 		 (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
86 		 (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT);
87 }
88 
89 static void mem_abort_decode(unsigned long esr)
90 {
91 	pr_alert("Mem abort info:\n");
92 
93 	pr_alert("  ESR = 0x%016lx\n", esr);
94 	pr_alert("  EC = 0x%02lx: %s, IL = %u bits\n",
95 		 ESR_ELx_EC(esr), esr_get_class_string(esr),
96 		 (esr & ESR_ELx_IL) ? 32 : 16);
97 	pr_alert("  SET = %lu, FnV = %lu\n",
98 		 (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
99 		 (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
100 	pr_alert("  EA = %lu, S1PTW = %lu\n",
101 		 (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
102 		 (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
103 	pr_alert("  FSC = 0x%02lx: %s\n", (esr & ESR_ELx_FSC),
104 		 esr_to_fault_info(esr)->name);
105 
106 	if (esr_is_data_abort(esr))
107 		data_abort_decode(esr);
108 }
109 
110 static inline unsigned long mm_to_pgd_phys(struct mm_struct *mm)
111 {
112 	/* Either init_pg_dir or swapper_pg_dir */
113 	if (mm == &init_mm)
114 		return __pa_symbol(mm->pgd);
115 
116 	return (unsigned long)virt_to_phys(mm->pgd);
117 }
118 
119 /*
120  * Dump out the page tables associated with 'addr' in the currently active mm.
121  */
122 static void show_pte(unsigned long addr)
123 {
124 	struct mm_struct *mm;
125 	pgd_t *pgdp;
126 	pgd_t pgd;
127 
128 	if (is_ttbr0_addr(addr)) {
129 		/* TTBR0 */
130 		mm = current->active_mm;
131 		if (mm == &init_mm) {
132 			pr_alert("[%016lx] user address but active_mm is swapper\n",
133 				 addr);
134 			return;
135 		}
136 	} else if (is_ttbr1_addr(addr)) {
137 		/* TTBR1 */
138 		mm = &init_mm;
139 	} else {
140 		pr_alert("[%016lx] address between user and kernel address ranges\n",
141 			 addr);
142 		return;
143 	}
144 
145 	pr_alert("%s pgtable: %luk pages, %llu-bit VAs, pgdp=%016lx\n",
146 		 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
147 		 vabits_actual, mm_to_pgd_phys(mm));
148 	pgdp = pgd_offset(mm, addr);
149 	pgd = READ_ONCE(*pgdp);
150 	pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
151 
152 	do {
153 		p4d_t *p4dp, p4d;
154 		pud_t *pudp, pud;
155 		pmd_t *pmdp, pmd;
156 		pte_t *ptep, pte;
157 
158 		if (pgd_none(pgd) || pgd_bad(pgd))
159 			break;
160 
161 		p4dp = p4d_offset(pgdp, addr);
162 		p4d = READ_ONCE(*p4dp);
163 		pr_cont(", p4d=%016llx", p4d_val(p4d));
164 		if (p4d_none(p4d) || p4d_bad(p4d))
165 			break;
166 
167 		pudp = pud_offset(p4dp, addr);
168 		pud = READ_ONCE(*pudp);
169 		pr_cont(", pud=%016llx", pud_val(pud));
170 		if (pud_none(pud) || pud_bad(pud))
171 			break;
172 
173 		pmdp = pmd_offset(pudp, addr);
174 		pmd = READ_ONCE(*pmdp);
175 		pr_cont(", pmd=%016llx", pmd_val(pmd));
176 		if (pmd_none(pmd) || pmd_bad(pmd))
177 			break;
178 
179 		ptep = pte_offset_map(pmdp, addr);
180 		pte = READ_ONCE(*ptep);
181 		pr_cont(", pte=%016llx", pte_val(pte));
182 		pte_unmap(ptep);
183 	} while(0);
184 
185 	pr_cont("\n");
186 }
187 
188 /*
189  * This function sets the access flags (dirty, accessed), as well as write
190  * permission, and only to a more permissive setting.
191  *
192  * It needs to cope with hardware update of the accessed/dirty state by other
193  * agents in the system and can safely skip the __sync_icache_dcache() call as,
194  * like set_pte_at(), the PTE is never changed from no-exec to exec here.
195  *
196  * Returns whether or not the PTE actually changed.
197  */
198 int ptep_set_access_flags(struct vm_area_struct *vma,
199 			  unsigned long address, pte_t *ptep,
200 			  pte_t entry, int dirty)
201 {
202 	pteval_t old_pteval, pteval;
203 	pte_t pte = READ_ONCE(*ptep);
204 
205 	if (pte_same(pte, entry))
206 		return 0;
207 
208 	/* only preserve the access flags and write permission */
209 	pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
210 
211 	/*
212 	 * Setting the flags must be done atomically to avoid racing with the
213 	 * hardware update of the access/dirty state. The PTE_RDONLY bit must
214 	 * be set to the most permissive (lowest value) of *ptep and entry
215 	 * (calculated as: a & b == ~(~a | ~b)).
216 	 */
217 	pte_val(entry) ^= PTE_RDONLY;
218 	pteval = pte_val(pte);
219 	do {
220 		old_pteval = pteval;
221 		pteval ^= PTE_RDONLY;
222 		pteval |= pte_val(entry);
223 		pteval ^= PTE_RDONLY;
224 		pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
225 	} while (pteval != old_pteval);
226 
227 	/* Invalidate a stale read-only entry */
228 	if (dirty)
229 		flush_tlb_page(vma, address);
230 	return 1;
231 }
232 
233 static bool is_el1_instruction_abort(unsigned long esr)
234 {
235 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
236 }
237 
238 static bool is_el1_data_abort(unsigned long esr)
239 {
240 	return ESR_ELx_EC(esr) == ESR_ELx_EC_DABT_CUR;
241 }
242 
243 static inline bool is_el1_permission_fault(unsigned long addr, unsigned long esr,
244 					   struct pt_regs *regs)
245 {
246 	unsigned long fsc_type = esr & ESR_ELx_FSC_TYPE;
247 
248 	if (!is_el1_data_abort(esr) && !is_el1_instruction_abort(esr))
249 		return false;
250 
251 	if (fsc_type == ESR_ELx_FSC_PERM)
252 		return true;
253 
254 	if (is_ttbr0_addr(addr) && system_uses_ttbr0_pan())
255 		return fsc_type == ESR_ELx_FSC_FAULT &&
256 			(regs->pstate & PSR_PAN_BIT);
257 
258 	return false;
259 }
260 
261 static bool __kprobes is_spurious_el1_translation_fault(unsigned long addr,
262 							unsigned long esr,
263 							struct pt_regs *regs)
264 {
265 	unsigned long flags;
266 	u64 par, dfsc;
267 
268 	if (!is_el1_data_abort(esr) ||
269 	    (esr & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT)
270 		return false;
271 
272 	local_irq_save(flags);
273 	asm volatile("at s1e1r, %0" :: "r" (addr));
274 	isb();
275 	par = read_sysreg_par();
276 	local_irq_restore(flags);
277 
278 	/*
279 	 * If we now have a valid translation, treat the translation fault as
280 	 * spurious.
281 	 */
282 	if (!(par & SYS_PAR_EL1_F))
283 		return true;
284 
285 	/*
286 	 * If we got a different type of fault from the AT instruction,
287 	 * treat the translation fault as spurious.
288 	 */
289 	dfsc = FIELD_GET(SYS_PAR_EL1_FST, par);
290 	return (dfsc & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT;
291 }
292 
293 static void die_kernel_fault(const char *msg, unsigned long addr,
294 			     unsigned long esr, struct pt_regs *regs)
295 {
296 	bust_spinlocks(1);
297 
298 	pr_alert("Unable to handle kernel %s at virtual address %016lx\n", msg,
299 		 addr);
300 
301 	kasan_non_canonical_hook(addr);
302 
303 	mem_abort_decode(esr);
304 
305 	show_pte(addr);
306 	die("Oops", regs, esr);
307 	bust_spinlocks(0);
308 	make_task_dead(SIGKILL);
309 }
310 
311 #ifdef CONFIG_KASAN_HW_TAGS
312 static void report_tag_fault(unsigned long addr, unsigned long esr,
313 			     struct pt_regs *regs)
314 {
315 	/*
316 	 * SAS bits aren't set for all faults reported in EL1, so we can't
317 	 * find out access size.
318 	 */
319 	bool is_write = !!(esr & ESR_ELx_WNR);
320 	kasan_report(addr, 0, is_write, regs->pc);
321 }
322 #else
323 /* Tag faults aren't enabled without CONFIG_KASAN_HW_TAGS. */
324 static inline void report_tag_fault(unsigned long addr, unsigned long esr,
325 				    struct pt_regs *regs) { }
326 #endif
327 
328 static void do_tag_recovery(unsigned long addr, unsigned long esr,
329 			   struct pt_regs *regs)
330 {
331 
332 	report_tag_fault(addr, esr, regs);
333 
334 	/*
335 	 * Disable MTE Tag Checking on the local CPU for the current EL.
336 	 * It will be done lazily on the other CPUs when they will hit a
337 	 * tag fault.
338 	 */
339 	sysreg_clear_set(sctlr_el1, SCTLR_EL1_TCF_MASK,
340 			 SYS_FIELD_PREP_ENUM(SCTLR_EL1, TCF, NONE));
341 	isb();
342 }
343 
344 static bool is_el1_mte_sync_tag_check_fault(unsigned long esr)
345 {
346 	unsigned long fsc = esr & ESR_ELx_FSC;
347 
348 	if (!is_el1_data_abort(esr))
349 		return false;
350 
351 	if (fsc == ESR_ELx_FSC_MTE)
352 		return true;
353 
354 	return false;
355 }
356 
357 static void __do_kernel_fault(unsigned long addr, unsigned long esr,
358 			      struct pt_regs *regs)
359 {
360 	const char *msg;
361 
362 	/*
363 	 * Are we prepared to handle this kernel fault?
364 	 * We are almost certainly not prepared to handle instruction faults.
365 	 */
366 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
367 		return;
368 
369 	if (WARN_RATELIMIT(is_spurious_el1_translation_fault(addr, esr, regs),
370 	    "Ignoring spurious kernel translation fault at virtual address %016lx\n", addr))
371 		return;
372 
373 	if (is_el1_mte_sync_tag_check_fault(esr)) {
374 		do_tag_recovery(addr, esr, regs);
375 
376 		return;
377 	}
378 
379 	if (is_el1_permission_fault(addr, esr, regs)) {
380 		if (esr & ESR_ELx_WNR)
381 			msg = "write to read-only memory";
382 		else if (is_el1_instruction_abort(esr))
383 			msg = "execute from non-executable memory";
384 		else
385 			msg = "read from unreadable memory";
386 	} else if (addr < PAGE_SIZE) {
387 		msg = "NULL pointer dereference";
388 	} else {
389 		if (kfence_handle_page_fault(addr, esr & ESR_ELx_WNR, regs))
390 			return;
391 
392 		msg = "paging request";
393 	}
394 
395 	if (efi_runtime_fixup_exception(regs, msg))
396 		return;
397 
398 	die_kernel_fault(msg, addr, esr, regs);
399 }
400 
401 static void set_thread_esr(unsigned long address, unsigned long esr)
402 {
403 	current->thread.fault_address = address;
404 
405 	/*
406 	 * If the faulting address is in the kernel, we must sanitize the ESR.
407 	 * From userspace's point of view, kernel-only mappings don't exist
408 	 * at all, so we report them as level 0 translation faults.
409 	 * (This is not quite the way that "no mapping there at all" behaves:
410 	 * an alignment fault not caused by the memory type would take
411 	 * precedence over translation fault for a real access to empty
412 	 * space. Unfortunately we can't easily distinguish "alignment fault
413 	 * not caused by memory type" from "alignment fault caused by memory
414 	 * type", so we ignore this wrinkle and just return the translation
415 	 * fault.)
416 	 */
417 	if (!is_ttbr0_addr(current->thread.fault_address)) {
418 		switch (ESR_ELx_EC(esr)) {
419 		case ESR_ELx_EC_DABT_LOW:
420 			/*
421 			 * These bits provide only information about the
422 			 * faulting instruction, which userspace knows already.
423 			 * We explicitly clear bits which are architecturally
424 			 * RES0 in case they are given meanings in future.
425 			 * We always report the ESR as if the fault was taken
426 			 * to EL1 and so ISV and the bits in ISS[23:14] are
427 			 * clear. (In fact it always will be a fault to EL1.)
428 			 */
429 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL |
430 				ESR_ELx_CM | ESR_ELx_WNR;
431 			esr |= ESR_ELx_FSC_FAULT;
432 			break;
433 		case ESR_ELx_EC_IABT_LOW:
434 			/*
435 			 * Claim a level 0 translation fault.
436 			 * All other bits are architecturally RES0 for faults
437 			 * reported with that DFSC value, so we clear them.
438 			 */
439 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL;
440 			esr |= ESR_ELx_FSC_FAULT;
441 			break;
442 		default:
443 			/*
444 			 * This should never happen (entry.S only brings us
445 			 * into this code for insn and data aborts from a lower
446 			 * exception level). Fail safe by not providing an ESR
447 			 * context record at all.
448 			 */
449 			WARN(1, "ESR 0x%lx is not DABT or IABT from EL0\n", esr);
450 			esr = 0;
451 			break;
452 		}
453 	}
454 
455 	current->thread.fault_code = esr;
456 }
457 
458 static void do_bad_area(unsigned long far, unsigned long esr,
459 			struct pt_regs *regs)
460 {
461 	unsigned long addr = untagged_addr(far);
462 
463 	/*
464 	 * If we are in kernel mode at this point, we have no context to
465 	 * handle this fault with.
466 	 */
467 	if (user_mode(regs)) {
468 		const struct fault_info *inf = esr_to_fault_info(esr);
469 
470 		set_thread_esr(addr, esr);
471 		arm64_force_sig_fault(inf->sig, inf->code, far, inf->name);
472 	} else {
473 		__do_kernel_fault(addr, esr, regs);
474 	}
475 }
476 
477 #define VM_FAULT_BADMAP		0x010000
478 #define VM_FAULT_BADACCESS	0x020000
479 
480 static vm_fault_t __do_page_fault(struct mm_struct *mm, unsigned long addr,
481 				  unsigned int mm_flags, unsigned long vm_flags,
482 				  struct pt_regs *regs)
483 {
484 	struct vm_area_struct *vma = find_vma(mm, addr);
485 
486 	if (unlikely(!vma))
487 		return VM_FAULT_BADMAP;
488 
489 	/*
490 	 * Ok, we have a good vm_area for this memory access, so we can handle
491 	 * it.
492 	 */
493 	if (unlikely(vma->vm_start > addr)) {
494 		if (!(vma->vm_flags & VM_GROWSDOWN))
495 			return VM_FAULT_BADMAP;
496 		if (expand_stack(vma, addr))
497 			return VM_FAULT_BADMAP;
498 	}
499 
500 	/*
501 	 * Check that the permissions on the VMA allow for the fault which
502 	 * occurred.
503 	 */
504 	if (!(vma->vm_flags & vm_flags))
505 		return VM_FAULT_BADACCESS;
506 	return handle_mm_fault(vma, addr, mm_flags, regs);
507 }
508 
509 static bool is_el0_instruction_abort(unsigned long esr)
510 {
511 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
512 }
513 
514 /*
515  * Note: not valid for EL1 DC IVAC, but we never use that such that it
516  * should fault. EL0 cannot issue DC IVAC (undef).
517  */
518 static bool is_write_abort(unsigned long esr)
519 {
520 	return (esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM);
521 }
522 
523 static int __kprobes do_page_fault(unsigned long far, unsigned long esr,
524 				   struct pt_regs *regs)
525 {
526 	const struct fault_info *inf;
527 	struct mm_struct *mm = current->mm;
528 	vm_fault_t fault;
529 	unsigned long vm_flags;
530 	unsigned int mm_flags = FAULT_FLAG_DEFAULT;
531 	unsigned long addr = untagged_addr(far);
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 (!cpus_have_const_cap(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 	/*
583 	 * As per x86, we may deadlock here. However, since the kernel only
584 	 * validly references user space from well defined areas of the code,
585 	 * we can bug out early if this is from code which shouldn't.
586 	 */
587 	if (!mmap_read_trylock(mm)) {
588 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
589 			goto no_context;
590 retry:
591 		mmap_read_lock(mm);
592 	} else {
593 		/*
594 		 * The above mmap_read_trylock() might have succeeded in which
595 		 * case, we'll have missed the might_sleep() from down_read().
596 		 */
597 		might_sleep();
598 #ifdef CONFIG_DEBUG_VM
599 		if (!user_mode(regs) && !search_exception_tables(regs->pc)) {
600 			mmap_read_unlock(mm);
601 			goto no_context;
602 		}
603 #endif
604 	}
605 
606 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, regs);
607 
608 	/* Quick path to respond to signals */
609 	if (fault_signal_pending(fault, regs)) {
610 		if (!user_mode(regs))
611 			goto no_context;
612 		return 0;
613 	}
614 
615 	/* The fault is fully completed (including releasing mmap lock) */
616 	if (fault & VM_FAULT_COMPLETED)
617 		return 0;
618 
619 	if (fault & VM_FAULT_RETRY) {
620 		mm_flags |= FAULT_FLAG_TRIED;
621 		goto retry;
622 	}
623 	mmap_read_unlock(mm);
624 
625 	/*
626 	 * Handle the "normal" (no error) case first.
627 	 */
628 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
629 			      VM_FAULT_BADACCESS))))
630 		return 0;
631 
632 	/*
633 	 * If we are in kernel mode at this point, we have no context to
634 	 * handle this fault with.
635 	 */
636 	if (!user_mode(regs))
637 		goto no_context;
638 
639 	if (fault & VM_FAULT_OOM) {
640 		/*
641 		 * We ran out of memory, call the OOM killer, and return to
642 		 * userspace (which will retry the fault, or kill us if we got
643 		 * oom-killed).
644 		 */
645 		pagefault_out_of_memory();
646 		return 0;
647 	}
648 
649 	inf = esr_to_fault_info(esr);
650 	set_thread_esr(addr, esr);
651 	if (fault & VM_FAULT_SIGBUS) {
652 		/*
653 		 * We had some memory, but were unable to successfully fix up
654 		 * this page fault.
655 		 */
656 		arm64_force_sig_fault(SIGBUS, BUS_ADRERR, far, inf->name);
657 	} else if (fault & (VM_FAULT_HWPOISON_LARGE | VM_FAULT_HWPOISON)) {
658 		unsigned int lsb;
659 
660 		lsb = PAGE_SHIFT;
661 		if (fault & VM_FAULT_HWPOISON_LARGE)
662 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
663 
664 		arm64_force_sig_mceerr(BUS_MCEERR_AR, far, lsb, inf->name);
665 	} else {
666 		/*
667 		 * Something tried to access memory that isn't in our memory
668 		 * map.
669 		 */
670 		arm64_force_sig_fault(SIGSEGV,
671 				      fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR,
672 				      far, inf->name);
673 	}
674 
675 	return 0;
676 
677 no_context:
678 	__do_kernel_fault(addr, esr, regs);
679 	return 0;
680 }
681 
682 static int __kprobes do_translation_fault(unsigned long far,
683 					  unsigned long esr,
684 					  struct pt_regs *regs)
685 {
686 	unsigned long addr = untagged_addr(far);
687 
688 	if (is_ttbr0_addr(addr))
689 		return do_page_fault(far, esr, regs);
690 
691 	do_bad_area(far, esr, regs);
692 	return 0;
693 }
694 
695 static int do_alignment_fault(unsigned long far, unsigned long esr,
696 			      struct pt_regs *regs)
697 {
698 	if (IS_ENABLED(CONFIG_COMPAT_ALIGNMENT_FIXUPS) &&
699 	    compat_user_mode(regs))
700 		return do_compat_alignment_fixup(far, regs);
701 	do_bad_area(far, esr, regs);
702 	return 0;
703 }
704 
705 static int do_bad(unsigned long far, unsigned long esr, struct pt_regs *regs)
706 {
707 	return 1; /* "fault" */
708 }
709 
710 static int do_sea(unsigned long far, unsigned long esr, struct pt_regs *regs)
711 {
712 	const struct fault_info *inf;
713 	unsigned long siaddr;
714 
715 	inf = esr_to_fault_info(esr);
716 
717 	if (user_mode(regs) && apei_claim_sea(regs) == 0) {
718 		/*
719 		 * APEI claimed this as a firmware-first notification.
720 		 * Some processing deferred to task_work before ret_to_user().
721 		 */
722 		return 0;
723 	}
724 
725 	if (esr & ESR_ELx_FnV) {
726 		siaddr = 0;
727 	} else {
728 		/*
729 		 * The architecture specifies that the tag bits of FAR_EL1 are
730 		 * UNKNOWN for synchronous external aborts. Mask them out now
731 		 * so that userspace doesn't see them.
732 		 */
733 		siaddr  = untagged_addr(far);
734 	}
735 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
736 
737 	return 0;
738 }
739 
740 static int do_tag_check_fault(unsigned long far, unsigned long esr,
741 			      struct pt_regs *regs)
742 {
743 	/*
744 	 * The architecture specifies that bits 63:60 of FAR_EL1 are UNKNOWN
745 	 * for tag check faults. Set them to corresponding bits in the untagged
746 	 * address.
747 	 */
748 	far = (__untagged_addr(far) & ~MTE_TAG_MASK) | (far & MTE_TAG_MASK);
749 	do_bad_area(far, esr, regs);
750 	return 0;
751 }
752 
753 static const struct fault_info fault_info[] = {
754 	{ do_bad,		SIGKILL, SI_KERNEL,	"ttbr address size fault"	},
755 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 1 address size fault"	},
756 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 2 address size fault"	},
757 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 3 address size fault"	},
758 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
759 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
760 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
761 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
762 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 8"			},
763 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
764 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
765 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
766 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 12"			},
767 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
768 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
769 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
770 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous external abort"	},
771 	{ do_tag_check_fault,	SIGSEGV, SEGV_MTESERR,	"synchronous tag check fault"	},
772 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 18"			},
773 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 19"			},
774 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 (translation table walk)"	},
775 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 (translation table walk)"	},
776 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 (translation table walk)"	},
777 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 (translation table walk)"	},
778 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous parity or ECC error" },	// Reserved when RAS is implemented
779 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 25"			},
780 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 26"			},
781 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 27"			},
782 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
783 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
784 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
785 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
786 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 32"			},
787 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
788 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 34"			},
789 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 35"			},
790 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 36"			},
791 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 37"			},
792 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 38"			},
793 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 39"			},
794 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 40"			},
795 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 41"			},
796 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 42"			},
797 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 43"			},
798 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 44"			},
799 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 45"			},
800 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 46"			},
801 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 47"			},
802 	{ do_bad,		SIGKILL, SI_KERNEL,	"TLB conflict abort"		},
803 	{ do_bad,		SIGKILL, SI_KERNEL,	"Unsupported atomic hardware update fault"	},
804 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 50"			},
805 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 51"			},
806 	{ do_bad,		SIGKILL, SI_KERNEL,	"implementation fault (lockdown abort)" },
807 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"implementation fault (unsupported exclusive)" },
808 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 54"			},
809 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 55"			},
810 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 56"			},
811 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 57"			},
812 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 58" 			},
813 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 59"			},
814 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 60"			},
815 	{ do_bad,		SIGKILL, SI_KERNEL,	"section domain fault"		},
816 	{ do_bad,		SIGKILL, SI_KERNEL,	"page domain fault"		},
817 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 63"			},
818 };
819 
820 void do_mem_abort(unsigned long far, unsigned long esr, struct pt_regs *regs)
821 {
822 	const struct fault_info *inf = esr_to_fault_info(esr);
823 	unsigned long addr = untagged_addr(far);
824 
825 	if (!inf->fn(far, esr, regs))
826 		return;
827 
828 	if (!user_mode(regs))
829 		die_kernel_fault(inf->name, addr, esr, regs);
830 
831 	/*
832 	 * At this point we have an unrecognized fault type whose tag bits may
833 	 * have been defined as UNKNOWN. Therefore we only expose the untagged
834 	 * address to the signal handler.
835 	 */
836 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, addr, esr);
837 }
838 NOKPROBE_SYMBOL(do_mem_abort);
839 
840 void do_sp_pc_abort(unsigned long addr, unsigned long esr, struct pt_regs *regs)
841 {
842 	arm64_notify_die("SP/PC alignment exception", regs, SIGBUS, BUS_ADRALN,
843 			 addr, esr);
844 }
845 NOKPROBE_SYMBOL(do_sp_pc_abort);
846 
847 int __init early_brk64(unsigned long addr, unsigned long esr,
848 		       struct pt_regs *regs);
849 
850 /*
851  * __refdata because early_brk64 is __init, but the reference to it is
852  * clobbered at arch_initcall time.
853  * See traps.c and debug-monitors.c:debug_traps_init().
854  */
855 static struct fault_info __refdata debug_fault_info[] = {
856 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
857 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
858 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
859 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 3"		},
860 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
861 	{ do_bad,	SIGKILL,	SI_KERNEL,	"aarch32 vector catch"	},
862 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
863 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 7"		},
864 };
865 
866 void __init hook_debug_fault_code(int nr,
867 				  int (*fn)(unsigned long, unsigned long, struct pt_regs *),
868 				  int sig, int code, const char *name)
869 {
870 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
871 
872 	debug_fault_info[nr].fn		= fn;
873 	debug_fault_info[nr].sig	= sig;
874 	debug_fault_info[nr].code	= code;
875 	debug_fault_info[nr].name	= name;
876 }
877 
878 /*
879  * In debug exception context, we explicitly disable preemption despite
880  * having interrupts disabled.
881  * This serves two purposes: it makes it much less likely that we would
882  * accidentally schedule in exception context and it will force a warning
883  * if we somehow manage to schedule by accident.
884  */
885 static void debug_exception_enter(struct pt_regs *regs)
886 {
887 	preempt_disable();
888 
889 	/* This code is a bit fragile.  Test it. */
890 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "exception_enter didn't work");
891 }
892 NOKPROBE_SYMBOL(debug_exception_enter);
893 
894 static void debug_exception_exit(struct pt_regs *regs)
895 {
896 	preempt_enable_no_resched();
897 }
898 NOKPROBE_SYMBOL(debug_exception_exit);
899 
900 void do_debug_exception(unsigned long addr_if_watchpoint, unsigned long esr,
901 			struct pt_regs *regs)
902 {
903 	const struct fault_info *inf = esr_to_debug_fault_info(esr);
904 	unsigned long pc = instruction_pointer(regs);
905 
906 	debug_exception_enter(regs);
907 
908 	if (user_mode(regs) && !is_ttbr0_addr(pc))
909 		arm64_apply_bp_hardening();
910 
911 	if (inf->fn(addr_if_watchpoint, esr, regs)) {
912 		arm64_notify_die(inf->name, regs, inf->sig, inf->code, pc, esr);
913 	}
914 
915 	debug_exception_exit(regs);
916 }
917 NOKPROBE_SYMBOL(do_debug_exception);
918 
919 /*
920  * Used during anonymous page fault handling.
921  */
922 struct page *alloc_zeroed_user_highpage_movable(struct vm_area_struct *vma,
923 						unsigned long vaddr)
924 {
925 	gfp_t flags = GFP_HIGHUSER_MOVABLE | __GFP_ZERO;
926 
927 	/*
928 	 * If the page is mapped with PROT_MTE, initialise the tags at the
929 	 * point of allocation and page zeroing as this is usually faster than
930 	 * separate DC ZVA and STGM.
931 	 */
932 	if (vma->vm_flags & VM_MTE)
933 		flags |= __GFP_ZEROTAGS;
934 
935 	return alloc_page_vma(flags, vma, vaddr);
936 }
937 
938 void tag_clear_highpage(struct page *page)
939 {
940 	mte_zero_clear_page_tags(page_address(page));
941 	set_bit(PG_mte_tagged, &page->flags);
942 }
943