xref: /openbmc/linux/arch/arm64/mm/fault.c (revision 7ae5c03a)
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/exception.h>
34 #include <asm/daifflags.h>
35 #include <asm/debug-monitors.h>
36 #include <asm/esr.h>
37 #include <asm/kprobes.h>
38 #include <asm/mte.h>
39 #include <asm/processor.h>
40 #include <asm/sysreg.h>
41 #include <asm/system_misc.h>
42 #include <asm/tlbflush.h>
43 #include <asm/traps.h>
44 
45 struct fault_info {
46 	int	(*fn)(unsigned long far, unsigned long esr,
47 		      struct pt_regs *regs);
48 	int	sig;
49 	int	code;
50 	const char *name;
51 };
52 
53 static const struct fault_info fault_info[];
54 static struct fault_info debug_fault_info[];
55 
56 static inline const struct fault_info *esr_to_fault_info(unsigned long esr)
57 {
58 	return fault_info + (esr & ESR_ELx_FSC);
59 }
60 
61 static inline const struct fault_info *esr_to_debug_fault_info(unsigned long esr)
62 {
63 	return debug_fault_info + DBG_ESR_EVT(esr);
64 }
65 
66 static void data_abort_decode(unsigned long esr)
67 {
68 	pr_alert("Data abort info:\n");
69 
70 	if (esr & ESR_ELx_ISV) {
71 		pr_alert("  Access size = %u byte(s)\n",
72 			 1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
73 		pr_alert("  SSE = %lu, SRT = %lu\n",
74 			 (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
75 			 (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
76 		pr_alert("  SF = %lu, AR = %lu\n",
77 			 (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
78 			 (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
79 	} else {
80 		pr_alert("  ISV = 0, ISS = 0x%08lx\n", esr & ESR_ELx_ISS_MASK);
81 	}
82 
83 	pr_alert("  CM = %lu, WnR = %lu\n",
84 		 (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
85 		 (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT);
86 }
87 
88 static void mem_abort_decode(unsigned long esr)
89 {
90 	pr_alert("Mem abort info:\n");
91 
92 	pr_alert("  ESR = 0x%016lx\n", esr);
93 	pr_alert("  EC = 0x%02lx: %s, IL = %u bits\n",
94 		 ESR_ELx_EC(esr), esr_get_class_string(esr),
95 		 (esr & ESR_ELx_IL) ? 32 : 16);
96 	pr_alert("  SET = %lu, FnV = %lu\n",
97 		 (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
98 		 (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
99 	pr_alert("  EA = %lu, S1PTW = %lu\n",
100 		 (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
101 		 (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
102 	pr_alert("  FSC = 0x%02lx: %s\n", (esr & ESR_ELx_FSC),
103 		 esr_to_fault_info(esr)->name);
104 
105 	if (esr_is_data_abort(esr))
106 		data_abort_decode(esr);
107 }
108 
109 static inline unsigned long mm_to_pgd_phys(struct mm_struct *mm)
110 {
111 	/* Either init_pg_dir or swapper_pg_dir */
112 	if (mm == &init_mm)
113 		return __pa_symbol(mm->pgd);
114 
115 	return (unsigned long)virt_to_phys(mm->pgd);
116 }
117 
118 /*
119  * Dump out the page tables associated with 'addr' in the currently active mm.
120  */
121 static void show_pte(unsigned long addr)
122 {
123 	struct mm_struct *mm;
124 	pgd_t *pgdp;
125 	pgd_t pgd;
126 
127 	if (is_ttbr0_addr(addr)) {
128 		/* TTBR0 */
129 		mm = current->active_mm;
130 		if (mm == &init_mm) {
131 			pr_alert("[%016lx] user address but active_mm is swapper\n",
132 				 addr);
133 			return;
134 		}
135 	} else if (is_ttbr1_addr(addr)) {
136 		/* TTBR1 */
137 		mm = &init_mm;
138 	} else {
139 		pr_alert("[%016lx] address between user and kernel address ranges\n",
140 			 addr);
141 		return;
142 	}
143 
144 	pr_alert("%s pgtable: %luk pages, %llu-bit VAs, pgdp=%016lx\n",
145 		 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
146 		 vabits_actual, mm_to_pgd_phys(mm));
147 	pgdp = pgd_offset(mm, addr);
148 	pgd = READ_ONCE(*pgdp);
149 	pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
150 
151 	do {
152 		p4d_t *p4dp, p4d;
153 		pud_t *pudp, pud;
154 		pmd_t *pmdp, pmd;
155 		pte_t *ptep, pte;
156 
157 		if (pgd_none(pgd) || pgd_bad(pgd))
158 			break;
159 
160 		p4dp = p4d_offset(pgdp, addr);
161 		p4d = READ_ONCE(*p4dp);
162 		pr_cont(", p4d=%016llx", p4d_val(p4d));
163 		if (p4d_none(p4d) || p4d_bad(p4d))
164 			break;
165 
166 		pudp = pud_offset(p4dp, 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 	/* Invalidate a stale read-only entry */
227 	if (dirty)
228 		flush_tlb_page(vma, address);
229 	return 1;
230 }
231 
232 static bool is_el1_instruction_abort(unsigned long esr)
233 {
234 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
235 }
236 
237 static bool is_el1_data_abort(unsigned long esr)
238 {
239 	return ESR_ELx_EC(esr) == ESR_ELx_EC_DABT_CUR;
240 }
241 
242 static inline bool is_el1_permission_fault(unsigned long addr, unsigned long esr,
243 					   struct pt_regs *regs)
244 {
245 	unsigned long fsc_type = esr & ESR_ELx_FSC_TYPE;
246 
247 	if (!is_el1_data_abort(esr) && !is_el1_instruction_abort(esr))
248 		return false;
249 
250 	if (fsc_type == ESR_ELx_FSC_PERM)
251 		return true;
252 
253 	if (is_ttbr0_addr(addr) && system_uses_ttbr0_pan())
254 		return fsc_type == ESR_ELx_FSC_FAULT &&
255 			(regs->pstate & PSR_PAN_BIT);
256 
257 	return false;
258 }
259 
260 static bool __kprobes is_spurious_el1_translation_fault(unsigned long addr,
261 							unsigned long esr,
262 							struct pt_regs *regs)
263 {
264 	unsigned long flags;
265 	u64 par, dfsc;
266 
267 	if (!is_el1_data_abort(esr) ||
268 	    (esr & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT)
269 		return false;
270 
271 	local_irq_save(flags);
272 	asm volatile("at s1e1r, %0" :: "r" (addr));
273 	isb();
274 	par = read_sysreg_par();
275 	local_irq_restore(flags);
276 
277 	/*
278 	 * If we now have a valid translation, treat the translation fault as
279 	 * spurious.
280 	 */
281 	if (!(par & SYS_PAR_EL1_F))
282 		return true;
283 
284 	/*
285 	 * If we got a different type of fault from the AT instruction,
286 	 * treat the translation fault as spurious.
287 	 */
288 	dfsc = FIELD_GET(SYS_PAR_EL1_FST, par);
289 	return (dfsc & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT;
290 }
291 
292 static void die_kernel_fault(const char *msg, unsigned long addr,
293 			     unsigned long esr, struct pt_regs *regs)
294 {
295 	bust_spinlocks(1);
296 
297 	pr_alert("Unable to handle kernel %s at virtual address %016lx\n", msg,
298 		 addr);
299 
300 	kasan_non_canonical_hook(addr);
301 
302 	mem_abort_decode(esr);
303 
304 	show_pte(addr);
305 	die("Oops", regs, esr);
306 	bust_spinlocks(0);
307 	make_task_dead(SIGKILL);
308 }
309 
310 #ifdef CONFIG_KASAN_HW_TAGS
311 static void report_tag_fault(unsigned long addr, unsigned long esr,
312 			     struct pt_regs *regs)
313 {
314 	/*
315 	 * SAS bits aren't set for all faults reported in EL1, so we can't
316 	 * find out access size.
317 	 */
318 	bool is_write = !!(esr & ESR_ELx_WNR);
319 	kasan_report(addr, 0, is_write, regs->pc);
320 }
321 #else
322 /* Tag faults aren't enabled without CONFIG_KASAN_HW_TAGS. */
323 static inline void report_tag_fault(unsigned long addr, unsigned long esr,
324 				    struct pt_regs *regs) { }
325 #endif
326 
327 static void do_tag_recovery(unsigned long addr, unsigned long esr,
328 			   struct pt_regs *regs)
329 {
330 
331 	report_tag_fault(addr, esr, regs);
332 
333 	/*
334 	 * Disable MTE Tag Checking on the local CPU for the current EL.
335 	 * It will be done lazily on the other CPUs when they will hit a
336 	 * tag fault.
337 	 */
338 	sysreg_clear_set(sctlr_el1, SCTLR_EL1_TCF_MASK,
339 			 SYS_FIELD_PREP_ENUM(SCTLR_EL1, TCF, NONE));
340 	isb();
341 }
342 
343 static bool is_el1_mte_sync_tag_check_fault(unsigned long esr)
344 {
345 	unsigned long fsc = esr & ESR_ELx_FSC;
346 
347 	if (!is_el1_data_abort(esr))
348 		return false;
349 
350 	if (fsc == ESR_ELx_FSC_MTE)
351 		return true;
352 
353 	return false;
354 }
355 
356 static void __do_kernel_fault(unsigned long addr, unsigned long esr,
357 			      struct pt_regs *regs)
358 {
359 	const char *msg;
360 
361 	/*
362 	 * Are we prepared to handle this kernel fault?
363 	 * We are almost certainly not prepared to handle instruction faults.
364 	 */
365 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
366 		return;
367 
368 	if (WARN_RATELIMIT(is_spurious_el1_translation_fault(addr, esr, regs),
369 	    "Ignoring spurious kernel translation fault at virtual address %016lx\n", addr))
370 		return;
371 
372 	if (is_el1_mte_sync_tag_check_fault(esr)) {
373 		do_tag_recovery(addr, esr, regs);
374 
375 		return;
376 	}
377 
378 	if (is_el1_permission_fault(addr, esr, regs)) {
379 		if (esr & ESR_ELx_WNR)
380 			msg = "write to read-only memory";
381 		else if (is_el1_instruction_abort(esr))
382 			msg = "execute from non-executable memory";
383 		else
384 			msg = "read from unreadable memory";
385 	} else if (addr < PAGE_SIZE) {
386 		msg = "NULL pointer dereference";
387 	} else {
388 		if (kfence_handle_page_fault(addr, esr & ESR_ELx_WNR, regs))
389 			return;
390 
391 		msg = "paging request";
392 	}
393 
394 	die_kernel_fault(msg, addr, esr, regs);
395 }
396 
397 static void set_thread_esr(unsigned long address, unsigned long esr)
398 {
399 	current->thread.fault_address = address;
400 
401 	/*
402 	 * If the faulting address is in the kernel, we must sanitize the ESR.
403 	 * From userspace's point of view, kernel-only mappings don't exist
404 	 * at all, so we report them as level 0 translation faults.
405 	 * (This is not quite the way that "no mapping there at all" behaves:
406 	 * an alignment fault not caused by the memory type would take
407 	 * precedence over translation fault for a real access to empty
408 	 * space. Unfortunately we can't easily distinguish "alignment fault
409 	 * not caused by memory type" from "alignment fault caused by memory
410 	 * type", so we ignore this wrinkle and just return the translation
411 	 * fault.)
412 	 */
413 	if (!is_ttbr0_addr(current->thread.fault_address)) {
414 		switch (ESR_ELx_EC(esr)) {
415 		case ESR_ELx_EC_DABT_LOW:
416 			/*
417 			 * These bits provide only information about the
418 			 * faulting instruction, which userspace knows already.
419 			 * We explicitly clear bits which are architecturally
420 			 * RES0 in case they are given meanings in future.
421 			 * We always report the ESR as if the fault was taken
422 			 * to EL1 and so ISV and the bits in ISS[23:14] are
423 			 * clear. (In fact it always will be a fault to EL1.)
424 			 */
425 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL |
426 				ESR_ELx_CM | ESR_ELx_WNR;
427 			esr |= ESR_ELx_FSC_FAULT;
428 			break;
429 		case ESR_ELx_EC_IABT_LOW:
430 			/*
431 			 * Claim a level 0 translation fault.
432 			 * All other bits are architecturally RES0 for faults
433 			 * reported with that DFSC value, so we clear them.
434 			 */
435 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL;
436 			esr |= ESR_ELx_FSC_FAULT;
437 			break;
438 		default:
439 			/*
440 			 * This should never happen (entry.S only brings us
441 			 * into this code for insn and data aborts from a lower
442 			 * exception level). Fail safe by not providing an ESR
443 			 * context record at all.
444 			 */
445 			WARN(1, "ESR 0x%lx is not DABT or IABT from EL0\n", esr);
446 			esr = 0;
447 			break;
448 		}
449 	}
450 
451 	current->thread.fault_code = esr;
452 }
453 
454 static void do_bad_area(unsigned long far, unsigned long esr,
455 			struct pt_regs *regs)
456 {
457 	unsigned long addr = untagged_addr(far);
458 
459 	/*
460 	 * If we are in kernel mode at this point, we have no context to
461 	 * handle this fault with.
462 	 */
463 	if (user_mode(regs)) {
464 		const struct fault_info *inf = esr_to_fault_info(esr);
465 
466 		set_thread_esr(addr, esr);
467 		arm64_force_sig_fault(inf->sig, inf->code, far, inf->name);
468 	} else {
469 		__do_kernel_fault(addr, esr, regs);
470 	}
471 }
472 
473 #define VM_FAULT_BADMAP		0x010000
474 #define VM_FAULT_BADACCESS	0x020000
475 
476 static vm_fault_t __do_page_fault(struct mm_struct *mm, unsigned long addr,
477 				  unsigned int mm_flags, unsigned long vm_flags,
478 				  struct pt_regs *regs)
479 {
480 	struct vm_area_struct *vma = find_vma(mm, addr);
481 
482 	if (unlikely(!vma))
483 		return VM_FAULT_BADMAP;
484 
485 	/*
486 	 * Ok, we have a good vm_area for this memory access, so we can handle
487 	 * it.
488 	 */
489 	if (unlikely(vma->vm_start > addr)) {
490 		if (!(vma->vm_flags & VM_GROWSDOWN))
491 			return VM_FAULT_BADMAP;
492 		if (expand_stack(vma, addr))
493 			return VM_FAULT_BADMAP;
494 	}
495 
496 	/*
497 	 * Check that the permissions on the VMA allow for the fault which
498 	 * occurred.
499 	 */
500 	if (!(vma->vm_flags & vm_flags))
501 		return VM_FAULT_BADACCESS;
502 	return handle_mm_fault(vma, addr, mm_flags, regs);
503 }
504 
505 static bool is_el0_instruction_abort(unsigned long esr)
506 {
507 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
508 }
509 
510 /*
511  * Note: not valid for EL1 DC IVAC, but we never use that such that it
512  * should fault. EL0 cannot issue DC IVAC (undef).
513  */
514 static bool is_write_abort(unsigned long esr)
515 {
516 	return (esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM);
517 }
518 
519 static int __kprobes do_page_fault(unsigned long far, unsigned long esr,
520 				   struct pt_regs *regs)
521 {
522 	const struct fault_info *inf;
523 	struct mm_struct *mm = current->mm;
524 	vm_fault_t fault;
525 	unsigned long vm_flags;
526 	unsigned int mm_flags = FAULT_FLAG_DEFAULT;
527 	unsigned long addr = untagged_addr(far);
528 
529 	if (kprobe_page_fault(regs, esr))
530 		return 0;
531 
532 	/*
533 	 * If we're in an interrupt or have no user context, we must not take
534 	 * the fault.
535 	 */
536 	if (faulthandler_disabled() || !mm)
537 		goto no_context;
538 
539 	if (user_mode(regs))
540 		mm_flags |= FAULT_FLAG_USER;
541 
542 	/*
543 	 * vm_flags tells us what bits we must have in vma->vm_flags
544 	 * for the fault to be benign, __do_page_fault() would check
545 	 * vma->vm_flags & vm_flags and returns an error if the
546 	 * intersection is empty
547 	 */
548 	if (is_el0_instruction_abort(esr)) {
549 		/* It was exec fault */
550 		vm_flags = VM_EXEC;
551 		mm_flags |= FAULT_FLAG_INSTRUCTION;
552 	} else if (is_write_abort(esr)) {
553 		/* It was write fault */
554 		vm_flags = VM_WRITE;
555 		mm_flags |= FAULT_FLAG_WRITE;
556 	} else {
557 		/* It was read fault */
558 		vm_flags = VM_READ;
559 		/* Write implies read */
560 		vm_flags |= VM_WRITE;
561 		/* If EPAN is absent then exec implies read */
562 		if (!cpus_have_const_cap(ARM64_HAS_EPAN))
563 			vm_flags |= VM_EXEC;
564 	}
565 
566 	if (is_ttbr0_addr(addr) && is_el1_permission_fault(addr, esr, regs)) {
567 		if (is_el1_instruction_abort(esr))
568 			die_kernel_fault("execution of user memory",
569 					 addr, esr, regs);
570 
571 		if (!search_exception_tables(regs->pc))
572 			die_kernel_fault("access to user memory outside uaccess routines",
573 					 addr, esr, regs);
574 	}
575 
576 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
577 
578 	/*
579 	 * As per x86, we may deadlock here. However, since the kernel only
580 	 * validly references user space from well defined areas of the code,
581 	 * we can bug out early if this is from code which shouldn't.
582 	 */
583 	if (!mmap_read_trylock(mm)) {
584 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
585 			goto no_context;
586 retry:
587 		mmap_read_lock(mm);
588 	} else {
589 		/*
590 		 * The above mmap_read_trylock() might have succeeded in which
591 		 * case, we'll have missed the might_sleep() from down_read().
592 		 */
593 		might_sleep();
594 #ifdef CONFIG_DEBUG_VM
595 		if (!user_mode(regs) && !search_exception_tables(regs->pc)) {
596 			mmap_read_unlock(mm);
597 			goto no_context;
598 		}
599 #endif
600 	}
601 
602 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, regs);
603 
604 	/* Quick path to respond to signals */
605 	if (fault_signal_pending(fault, regs)) {
606 		if (!user_mode(regs))
607 			goto no_context;
608 		return 0;
609 	}
610 
611 	/* The fault is fully completed (including releasing mmap lock) */
612 	if (fault & VM_FAULT_COMPLETED)
613 		return 0;
614 
615 	if (fault & VM_FAULT_RETRY) {
616 		mm_flags |= FAULT_FLAG_TRIED;
617 		goto retry;
618 	}
619 	mmap_read_unlock(mm);
620 
621 	/*
622 	 * Handle the "normal" (no error) case first.
623 	 */
624 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
625 			      VM_FAULT_BADACCESS))))
626 		return 0;
627 
628 	/*
629 	 * If we are in kernel mode at this point, we have no context to
630 	 * handle this fault with.
631 	 */
632 	if (!user_mode(regs))
633 		goto no_context;
634 
635 	if (fault & VM_FAULT_OOM) {
636 		/*
637 		 * We ran out of memory, call the OOM killer, and return to
638 		 * userspace (which will retry the fault, or kill us if we got
639 		 * oom-killed).
640 		 */
641 		pagefault_out_of_memory();
642 		return 0;
643 	}
644 
645 	inf = esr_to_fault_info(esr);
646 	set_thread_esr(addr, esr);
647 	if (fault & VM_FAULT_SIGBUS) {
648 		/*
649 		 * We had some memory, but were unable to successfully fix up
650 		 * this page fault.
651 		 */
652 		arm64_force_sig_fault(SIGBUS, BUS_ADRERR, far, inf->name);
653 	} else if (fault & (VM_FAULT_HWPOISON_LARGE | VM_FAULT_HWPOISON)) {
654 		unsigned int lsb;
655 
656 		lsb = PAGE_SHIFT;
657 		if (fault & VM_FAULT_HWPOISON_LARGE)
658 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
659 
660 		arm64_force_sig_mceerr(BUS_MCEERR_AR, far, lsb, inf->name);
661 	} else {
662 		/*
663 		 * Something tried to access memory that isn't in our memory
664 		 * map.
665 		 */
666 		arm64_force_sig_fault(SIGSEGV,
667 				      fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR,
668 				      far, inf->name);
669 	}
670 
671 	return 0;
672 
673 no_context:
674 	__do_kernel_fault(addr, esr, regs);
675 	return 0;
676 }
677 
678 static int __kprobes do_translation_fault(unsigned long far,
679 					  unsigned long esr,
680 					  struct pt_regs *regs)
681 {
682 	unsigned long addr = untagged_addr(far);
683 
684 	if (is_ttbr0_addr(addr))
685 		return do_page_fault(far, esr, regs);
686 
687 	do_bad_area(far, esr, regs);
688 	return 0;
689 }
690 
691 static int do_alignment_fault(unsigned long far, unsigned long esr,
692 			      struct pt_regs *regs)
693 {
694 	do_bad_area(far, esr, regs);
695 	return 0;
696 }
697 
698 static int do_bad(unsigned long far, unsigned long esr, struct pt_regs *regs)
699 {
700 	return 1; /* "fault" */
701 }
702 
703 static int do_sea(unsigned long far, unsigned long esr, struct pt_regs *regs)
704 {
705 	const struct fault_info *inf;
706 	unsigned long siaddr;
707 
708 	inf = esr_to_fault_info(esr);
709 
710 	if (user_mode(regs) && apei_claim_sea(regs) == 0) {
711 		/*
712 		 * APEI claimed this as a firmware-first notification.
713 		 * Some processing deferred to task_work before ret_to_user().
714 		 */
715 		return 0;
716 	}
717 
718 	if (esr & ESR_ELx_FnV) {
719 		siaddr = 0;
720 	} else {
721 		/*
722 		 * The architecture specifies that the tag bits of FAR_EL1 are
723 		 * UNKNOWN for synchronous external aborts. Mask them out now
724 		 * so that userspace doesn't see them.
725 		 */
726 		siaddr  = untagged_addr(far);
727 	}
728 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
729 
730 	return 0;
731 }
732 
733 static int do_tag_check_fault(unsigned long far, unsigned long esr,
734 			      struct pt_regs *regs)
735 {
736 	/*
737 	 * The architecture specifies that bits 63:60 of FAR_EL1 are UNKNOWN
738 	 * for tag check faults. Set them to corresponding bits in the untagged
739 	 * address.
740 	 */
741 	far = (__untagged_addr(far) & ~MTE_TAG_MASK) | (far & MTE_TAG_MASK);
742 	do_bad_area(far, esr, regs);
743 	return 0;
744 }
745 
746 static const struct fault_info fault_info[] = {
747 	{ do_bad,		SIGKILL, SI_KERNEL,	"ttbr address size fault"	},
748 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 1 address size fault"	},
749 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 2 address size fault"	},
750 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 3 address size fault"	},
751 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
752 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
753 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
754 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
755 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 8"			},
756 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
757 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
758 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
759 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 12"			},
760 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
761 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
762 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
763 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous external abort"	},
764 	{ do_tag_check_fault,	SIGSEGV, SEGV_MTESERR,	"synchronous tag check fault"	},
765 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 18"			},
766 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 19"			},
767 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 (translation table walk)"	},
768 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 (translation table walk)"	},
769 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 (translation table walk)"	},
770 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 (translation table walk)"	},
771 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous parity or ECC error" },	// Reserved when RAS is implemented
772 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 25"			},
773 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 26"			},
774 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 27"			},
775 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
776 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
777 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
778 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
779 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 32"			},
780 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
781 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 34"			},
782 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 35"			},
783 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 36"			},
784 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 37"			},
785 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 38"			},
786 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 39"			},
787 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 40"			},
788 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 41"			},
789 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 42"			},
790 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 43"			},
791 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 44"			},
792 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 45"			},
793 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 46"			},
794 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 47"			},
795 	{ do_bad,		SIGKILL, SI_KERNEL,	"TLB conflict abort"		},
796 	{ do_bad,		SIGKILL, SI_KERNEL,	"Unsupported atomic hardware update fault"	},
797 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 50"			},
798 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 51"			},
799 	{ do_bad,		SIGKILL, SI_KERNEL,	"implementation fault (lockdown abort)" },
800 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"implementation fault (unsupported exclusive)" },
801 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 54"			},
802 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 55"			},
803 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 56"			},
804 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 57"			},
805 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 58" 			},
806 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 59"			},
807 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 60"			},
808 	{ do_bad,		SIGKILL, SI_KERNEL,	"section domain fault"		},
809 	{ do_bad,		SIGKILL, SI_KERNEL,	"page domain fault"		},
810 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 63"			},
811 };
812 
813 void do_mem_abort(unsigned long far, unsigned long esr, struct pt_regs *regs)
814 {
815 	const struct fault_info *inf = esr_to_fault_info(esr);
816 	unsigned long addr = untagged_addr(far);
817 
818 	if (!inf->fn(far, esr, regs))
819 		return;
820 
821 	if (!user_mode(regs))
822 		die_kernel_fault(inf->name, addr, esr, regs);
823 
824 	/*
825 	 * At this point we have an unrecognized fault type whose tag bits may
826 	 * have been defined as UNKNOWN. Therefore we only expose the untagged
827 	 * address to the signal handler.
828 	 */
829 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, addr, esr);
830 }
831 NOKPROBE_SYMBOL(do_mem_abort);
832 
833 void do_sp_pc_abort(unsigned long addr, unsigned long esr, struct pt_regs *regs)
834 {
835 	arm64_notify_die("SP/PC alignment exception", regs, SIGBUS, BUS_ADRALN,
836 			 addr, esr);
837 }
838 NOKPROBE_SYMBOL(do_sp_pc_abort);
839 
840 int __init early_brk64(unsigned long addr, unsigned long esr,
841 		       struct pt_regs *regs);
842 
843 /*
844  * __refdata because early_brk64 is __init, but the reference to it is
845  * clobbered at arch_initcall time.
846  * See traps.c and debug-monitors.c:debug_traps_init().
847  */
848 static struct fault_info __refdata debug_fault_info[] = {
849 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
850 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
851 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
852 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 3"		},
853 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
854 	{ do_bad,	SIGKILL,	SI_KERNEL,	"aarch32 vector catch"	},
855 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
856 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 7"		},
857 };
858 
859 void __init hook_debug_fault_code(int nr,
860 				  int (*fn)(unsigned long, unsigned long, struct pt_regs *),
861 				  int sig, int code, const char *name)
862 {
863 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
864 
865 	debug_fault_info[nr].fn		= fn;
866 	debug_fault_info[nr].sig	= sig;
867 	debug_fault_info[nr].code	= code;
868 	debug_fault_info[nr].name	= name;
869 }
870 
871 /*
872  * In debug exception context, we explicitly disable preemption despite
873  * having interrupts disabled.
874  * This serves two purposes: it makes it much less likely that we would
875  * accidentally schedule in exception context and it will force a warning
876  * if we somehow manage to schedule by accident.
877  */
878 static void debug_exception_enter(struct pt_regs *regs)
879 {
880 	preempt_disable();
881 
882 	/* This code is a bit fragile.  Test it. */
883 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "exception_enter didn't work");
884 }
885 NOKPROBE_SYMBOL(debug_exception_enter);
886 
887 static void debug_exception_exit(struct pt_regs *regs)
888 {
889 	preempt_enable_no_resched();
890 }
891 NOKPROBE_SYMBOL(debug_exception_exit);
892 
893 void do_debug_exception(unsigned long addr_if_watchpoint, unsigned long esr,
894 			struct pt_regs *regs)
895 {
896 	const struct fault_info *inf = esr_to_debug_fault_info(esr);
897 	unsigned long pc = instruction_pointer(regs);
898 
899 	debug_exception_enter(regs);
900 
901 	if (user_mode(regs) && !is_ttbr0_addr(pc))
902 		arm64_apply_bp_hardening();
903 
904 	if (inf->fn(addr_if_watchpoint, esr, regs)) {
905 		arm64_notify_die(inf->name, regs, inf->sig, inf->code, pc, esr);
906 	}
907 
908 	debug_exception_exit(regs);
909 }
910 NOKPROBE_SYMBOL(do_debug_exception);
911 
912 /*
913  * Used during anonymous page fault handling.
914  */
915 struct page *alloc_zeroed_user_highpage_movable(struct vm_area_struct *vma,
916 						unsigned long vaddr)
917 {
918 	gfp_t flags = GFP_HIGHUSER_MOVABLE | __GFP_ZERO;
919 
920 	/*
921 	 * If the page is mapped with PROT_MTE, initialise the tags at the
922 	 * point of allocation and page zeroing as this is usually faster than
923 	 * separate DC ZVA and STGM.
924 	 */
925 	if (vma->vm_flags & VM_MTE)
926 		flags |= __GFP_ZEROTAGS;
927 
928 	return alloc_page_vma(flags, vma, vaddr);
929 }
930 
931 void tag_clear_highpage(struct page *page)
932 {
933 	mte_zero_clear_page_tags(page_address(page));
934 	set_bit(PG_mte_tagged, &page->flags);
935 }
936