xref: /openbmc/linux/arch/arm64/mm/fault.c (revision dd1431e5)
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 int 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 int 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 int esr)
62 {
63 	return debug_fault_info + DBG_ESR_EVT(esr);
64 }
65 
66 static void data_abort_decode(unsigned int 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 int esr)
89 {
90 	pr_alert("Mem abort info:\n");
91 
92 	pr_alert("  ESR = 0x%08x\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 
103 	if (esr_is_data_abort(esr))
104 		data_abort_decode(esr);
105 }
106 
107 static inline unsigned long mm_to_pgd_phys(struct mm_struct *mm)
108 {
109 	/* Either init_pg_dir or swapper_pg_dir */
110 	if (mm == &init_mm)
111 		return __pa_symbol(mm->pgd);
112 
113 	return (unsigned long)virt_to_phys(mm->pgd);
114 }
115 
116 /*
117  * Dump out the page tables associated with 'addr' in the currently active mm.
118  */
119 static void show_pte(unsigned long addr)
120 {
121 	struct mm_struct *mm;
122 	pgd_t *pgdp;
123 	pgd_t pgd;
124 
125 	if (is_ttbr0_addr(addr)) {
126 		/* TTBR0 */
127 		mm = current->active_mm;
128 		if (mm == &init_mm) {
129 			pr_alert("[%016lx] user address but active_mm is swapper\n",
130 				 addr);
131 			return;
132 		}
133 	} else if (is_ttbr1_addr(addr)) {
134 		/* TTBR1 */
135 		mm = &init_mm;
136 	} else {
137 		pr_alert("[%016lx] address between user and kernel address ranges\n",
138 			 addr);
139 		return;
140 	}
141 
142 	pr_alert("%s pgtable: %luk pages, %llu-bit VAs, pgdp=%016lx\n",
143 		 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
144 		 vabits_actual, mm_to_pgd_phys(mm));
145 	pgdp = pgd_offset(mm, addr);
146 	pgd = READ_ONCE(*pgdp);
147 	pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
148 
149 	do {
150 		p4d_t *p4dp, p4d;
151 		pud_t *pudp, pud;
152 		pmd_t *pmdp, pmd;
153 		pte_t *ptep, pte;
154 
155 		if (pgd_none(pgd) || pgd_bad(pgd))
156 			break;
157 
158 		p4dp = p4d_offset(pgdp, addr);
159 		p4d = READ_ONCE(*p4dp);
160 		pr_cont(", p4d=%016llx", p4d_val(p4d));
161 		if (p4d_none(p4d) || p4d_bad(p4d))
162 			break;
163 
164 		pudp = pud_offset(p4dp, addr);
165 		pud = READ_ONCE(*pudp);
166 		pr_cont(", pud=%016llx", pud_val(pud));
167 		if (pud_none(pud) || pud_bad(pud))
168 			break;
169 
170 		pmdp = pmd_offset(pudp, addr);
171 		pmd = READ_ONCE(*pmdp);
172 		pr_cont(", pmd=%016llx", pmd_val(pmd));
173 		if (pmd_none(pmd) || pmd_bad(pmd))
174 			break;
175 
176 		ptep = pte_offset_map(pmdp, addr);
177 		pte = READ_ONCE(*ptep);
178 		pr_cont(", pte=%016llx", pte_val(pte));
179 		pte_unmap(ptep);
180 	} while(0);
181 
182 	pr_cont("\n");
183 }
184 
185 /*
186  * This function sets the access flags (dirty, accessed), as well as write
187  * permission, and only to a more permissive setting.
188  *
189  * It needs to cope with hardware update of the accessed/dirty state by other
190  * agents in the system and can safely skip the __sync_icache_dcache() call as,
191  * like set_pte_at(), the PTE is never changed from no-exec to exec here.
192  *
193  * Returns whether or not the PTE actually changed.
194  */
195 int ptep_set_access_flags(struct vm_area_struct *vma,
196 			  unsigned long address, pte_t *ptep,
197 			  pte_t entry, int dirty)
198 {
199 	pteval_t old_pteval, pteval;
200 	pte_t pte = READ_ONCE(*ptep);
201 
202 	if (pte_same(pte, entry))
203 		return 0;
204 
205 	/* only preserve the access flags and write permission */
206 	pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
207 
208 	/*
209 	 * Setting the flags must be done atomically to avoid racing with the
210 	 * hardware update of the access/dirty state. The PTE_RDONLY bit must
211 	 * be set to the most permissive (lowest value) of *ptep and entry
212 	 * (calculated as: a & b == ~(~a | ~b)).
213 	 */
214 	pte_val(entry) ^= PTE_RDONLY;
215 	pteval = pte_val(pte);
216 	do {
217 		old_pteval = pteval;
218 		pteval ^= PTE_RDONLY;
219 		pteval |= pte_val(entry);
220 		pteval ^= PTE_RDONLY;
221 		pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
222 	} while (pteval != old_pteval);
223 
224 	/* Invalidate a stale read-only entry */
225 	if (dirty)
226 		flush_tlb_page(vma, address);
227 	return 1;
228 }
229 
230 static bool is_el1_instruction_abort(unsigned int esr)
231 {
232 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
233 }
234 
235 static inline bool is_el1_permission_fault(unsigned long addr, unsigned int esr,
236 					   struct pt_regs *regs)
237 {
238 	unsigned int ec       = ESR_ELx_EC(esr);
239 	unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
240 
241 	if (ec != ESR_ELx_EC_DABT_CUR && ec != ESR_ELx_EC_IABT_CUR)
242 		return false;
243 
244 	if (fsc_type == ESR_ELx_FSC_PERM)
245 		return true;
246 
247 	if (is_ttbr0_addr(addr) && system_uses_ttbr0_pan())
248 		return fsc_type == ESR_ELx_FSC_FAULT &&
249 			(regs->pstate & PSR_PAN_BIT);
250 
251 	return false;
252 }
253 
254 static bool __kprobes is_spurious_el1_translation_fault(unsigned long addr,
255 							unsigned int esr,
256 							struct pt_regs *regs)
257 {
258 	unsigned long flags;
259 	u64 par, dfsc;
260 
261 	if (ESR_ELx_EC(esr) != ESR_ELx_EC_DABT_CUR ||
262 	    (esr & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT)
263 		return false;
264 
265 	local_irq_save(flags);
266 	asm volatile("at s1e1r, %0" :: "r" (addr));
267 	isb();
268 	par = read_sysreg_par();
269 	local_irq_restore(flags);
270 
271 	/*
272 	 * If we now have a valid translation, treat the translation fault as
273 	 * spurious.
274 	 */
275 	if (!(par & SYS_PAR_EL1_F))
276 		return true;
277 
278 	/*
279 	 * If we got a different type of fault from the AT instruction,
280 	 * treat the translation fault as spurious.
281 	 */
282 	dfsc = FIELD_GET(SYS_PAR_EL1_FST, par);
283 	return (dfsc & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT;
284 }
285 
286 static void die_kernel_fault(const char *msg, unsigned long addr,
287 			     unsigned int esr, struct pt_regs *regs)
288 {
289 	bust_spinlocks(1);
290 
291 	pr_alert("Unable to handle kernel %s at virtual address %016lx\n", msg,
292 		 addr);
293 
294 	mem_abort_decode(esr);
295 
296 	show_pte(addr);
297 	die("Oops", regs, esr);
298 	bust_spinlocks(0);
299 	do_exit(SIGKILL);
300 }
301 
302 #ifdef CONFIG_KASAN_HW_TAGS
303 static void report_tag_fault(unsigned long addr, unsigned int esr,
304 			     struct pt_regs *regs)
305 {
306 	static bool reported;
307 	bool is_write;
308 
309 	if (READ_ONCE(reported))
310 		return;
311 
312 	/*
313 	 * This is used for KASAN tests and assumes that no MTE faults
314 	 * happened before running the tests.
315 	 */
316 	if (mte_report_once())
317 		WRITE_ONCE(reported, true);
318 
319 	/*
320 	 * SAS bits aren't set for all faults reported in EL1, so we can't
321 	 * find out access size.
322 	 */
323 	is_write = !!(esr & ESR_ELx_WNR);
324 	kasan_report(addr, 0, is_write, regs->pc);
325 }
326 #else
327 /* Tag faults aren't enabled without CONFIG_KASAN_HW_TAGS. */
328 static inline void report_tag_fault(unsigned long addr, unsigned int esr,
329 				    struct pt_regs *regs) { }
330 #endif
331 
332 static void do_tag_recovery(unsigned long addr, unsigned int esr,
333 			   struct pt_regs *regs)
334 {
335 
336 	report_tag_fault(addr, esr, regs);
337 
338 	/*
339 	 * Disable MTE Tag Checking on the local CPU for the current EL.
340 	 * It will be done lazily on the other CPUs when they will hit a
341 	 * tag fault.
342 	 */
343 	sysreg_clear_set(sctlr_el1, SCTLR_ELx_TCF_MASK, SCTLR_ELx_TCF_NONE);
344 	isb();
345 }
346 
347 static bool is_el1_mte_sync_tag_check_fault(unsigned int esr)
348 {
349 	unsigned int ec = ESR_ELx_EC(esr);
350 	unsigned int fsc = esr & ESR_ELx_FSC;
351 
352 	if (ec != ESR_ELx_EC_DABT_CUR)
353 		return false;
354 
355 	if (fsc == ESR_ELx_FSC_MTE)
356 		return true;
357 
358 	return false;
359 }
360 
361 static void __do_kernel_fault(unsigned long addr, unsigned int esr,
362 			      struct pt_regs *regs)
363 {
364 	const char *msg;
365 
366 	/*
367 	 * Are we prepared to handle this kernel fault?
368 	 * We are almost certainly not prepared to handle instruction faults.
369 	 */
370 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
371 		return;
372 
373 	if (WARN_RATELIMIT(is_spurious_el1_translation_fault(addr, esr, regs),
374 	    "Ignoring spurious kernel translation fault at virtual address %016lx\n", addr))
375 		return;
376 
377 	if (is_el1_mte_sync_tag_check_fault(esr)) {
378 		do_tag_recovery(addr, esr, regs);
379 
380 		return;
381 	}
382 
383 	if (is_el1_permission_fault(addr, esr, regs)) {
384 		if (esr & ESR_ELx_WNR)
385 			msg = "write to read-only memory";
386 		else if (is_el1_instruction_abort(esr))
387 			msg = "execute from non-executable memory";
388 		else
389 			msg = "read from unreadable memory";
390 	} else if (addr < PAGE_SIZE) {
391 		msg = "NULL pointer dereference";
392 	} else {
393 		if (kfence_handle_page_fault(addr, esr & ESR_ELx_WNR, regs))
394 			return;
395 
396 		msg = "paging request";
397 	}
398 
399 	die_kernel_fault(msg, addr, esr, regs);
400 }
401 
402 static void set_thread_esr(unsigned long address, unsigned int esr)
403 {
404 	current->thread.fault_address = address;
405 
406 	/*
407 	 * If the faulting address is in the kernel, we must sanitize the ESR.
408 	 * From userspace's point of view, kernel-only mappings don't exist
409 	 * at all, so we report them as level 0 translation faults.
410 	 * (This is not quite the way that "no mapping there at all" behaves:
411 	 * an alignment fault not caused by the memory type would take
412 	 * precedence over translation fault for a real access to empty
413 	 * space. Unfortunately we can't easily distinguish "alignment fault
414 	 * not caused by memory type" from "alignment fault caused by memory
415 	 * type", so we ignore this wrinkle and just return the translation
416 	 * fault.)
417 	 */
418 	if (!is_ttbr0_addr(current->thread.fault_address)) {
419 		switch (ESR_ELx_EC(esr)) {
420 		case ESR_ELx_EC_DABT_LOW:
421 			/*
422 			 * These bits provide only information about the
423 			 * faulting instruction, which userspace knows already.
424 			 * We explicitly clear bits which are architecturally
425 			 * RES0 in case they are given meanings in future.
426 			 * We always report the ESR as if the fault was taken
427 			 * to EL1 and so ISV and the bits in ISS[23:14] are
428 			 * clear. (In fact it always will be a fault to EL1.)
429 			 */
430 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL |
431 				ESR_ELx_CM | ESR_ELx_WNR;
432 			esr |= ESR_ELx_FSC_FAULT;
433 			break;
434 		case ESR_ELx_EC_IABT_LOW:
435 			/*
436 			 * Claim a level 0 translation fault.
437 			 * All other bits are architecturally RES0 for faults
438 			 * reported with that DFSC value, so we clear them.
439 			 */
440 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL;
441 			esr |= ESR_ELx_FSC_FAULT;
442 			break;
443 		default:
444 			/*
445 			 * This should never happen (entry.S only brings us
446 			 * into this code for insn and data aborts from a lower
447 			 * exception level). Fail safe by not providing an ESR
448 			 * context record at all.
449 			 */
450 			WARN(1, "ESR 0x%x is not DABT or IABT from EL0\n", esr);
451 			esr = 0;
452 			break;
453 		}
454 	}
455 
456 	current->thread.fault_code = esr;
457 }
458 
459 static void do_bad_area(unsigned long far, unsigned int esr,
460 			struct pt_regs *regs)
461 {
462 	unsigned long addr = untagged_addr(far);
463 
464 	/*
465 	 * If we are in kernel mode at this point, we have no context to
466 	 * handle this fault with.
467 	 */
468 	if (user_mode(regs)) {
469 		const struct fault_info *inf = esr_to_fault_info(esr);
470 
471 		set_thread_esr(addr, esr);
472 		arm64_force_sig_fault(inf->sig, inf->code, far, inf->name);
473 	} else {
474 		__do_kernel_fault(addr, esr, regs);
475 	}
476 }
477 
478 #define VM_FAULT_BADMAP		0x010000
479 #define VM_FAULT_BADACCESS	0x020000
480 
481 static vm_fault_t __do_page_fault(struct mm_struct *mm, unsigned long addr,
482 				  unsigned int mm_flags, unsigned long vm_flags,
483 				  struct pt_regs *regs)
484 {
485 	struct vm_area_struct *vma = find_vma(mm, addr);
486 
487 	if (unlikely(!vma))
488 		return VM_FAULT_BADMAP;
489 
490 	/*
491 	 * Ok, we have a good vm_area for this memory access, so we can handle
492 	 * it.
493 	 */
494 	if (unlikely(vma->vm_start > addr)) {
495 		if (!(vma->vm_flags & VM_GROWSDOWN))
496 			return VM_FAULT_BADMAP;
497 		if (expand_stack(vma, addr))
498 			return VM_FAULT_BADMAP;
499 	}
500 
501 	/*
502 	 * Check that the permissions on the VMA allow for the fault which
503 	 * occurred.
504 	 */
505 	if (!(vma->vm_flags & vm_flags))
506 		return VM_FAULT_BADACCESS;
507 	return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags, regs);
508 }
509 
510 static bool is_el0_instruction_abort(unsigned int esr)
511 {
512 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
513 }
514 
515 /*
516  * Note: not valid for EL1 DC IVAC, but we never use that such that it
517  * should fault. EL0 cannot issue DC IVAC (undef).
518  */
519 static bool is_write_abort(unsigned int esr)
520 {
521 	return (esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM);
522 }
523 
524 static int __kprobes do_page_fault(unsigned long far, unsigned int esr,
525 				   struct pt_regs *regs)
526 {
527 	const struct fault_info *inf;
528 	struct mm_struct *mm = current->mm;
529 	vm_fault_t fault;
530 	unsigned long vm_flags;
531 	unsigned int mm_flags = FAULT_FLAG_DEFAULT;
532 	unsigned long addr = untagged_addr(far);
533 
534 	if (kprobe_page_fault(regs, esr))
535 		return 0;
536 
537 	/*
538 	 * If we're in an interrupt or have no user context, we must not take
539 	 * the fault.
540 	 */
541 	if (faulthandler_disabled() || !mm)
542 		goto no_context;
543 
544 	if (user_mode(regs))
545 		mm_flags |= FAULT_FLAG_USER;
546 
547 	/*
548 	 * vm_flags tells us what bits we must have in vma->vm_flags
549 	 * for the fault to be benign, __do_page_fault() would check
550 	 * vma->vm_flags & vm_flags and returns an error if the
551 	 * intersection is empty
552 	 */
553 	if (is_el0_instruction_abort(esr)) {
554 		/* It was exec fault */
555 		vm_flags = VM_EXEC;
556 		mm_flags |= FAULT_FLAG_INSTRUCTION;
557 	} else if (is_write_abort(esr)) {
558 		/* It was write fault */
559 		vm_flags = VM_WRITE;
560 		mm_flags |= FAULT_FLAG_WRITE;
561 	} else {
562 		/* It was read fault */
563 		vm_flags = VM_READ;
564 		/* Write implies read */
565 		vm_flags |= VM_WRITE;
566 		/* If EPAN is absent then exec implies read */
567 		if (!cpus_have_const_cap(ARM64_HAS_EPAN))
568 			vm_flags |= VM_EXEC;
569 	}
570 
571 	if (is_ttbr0_addr(addr) && is_el1_permission_fault(addr, esr, regs)) {
572 		if (is_el1_instruction_abort(esr))
573 			die_kernel_fault("execution of user memory",
574 					 addr, esr, regs);
575 
576 		if (!search_exception_tables(regs->pc))
577 			die_kernel_fault("access to user memory outside uaccess routines",
578 					 addr, esr, regs);
579 	}
580 
581 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
582 
583 	/*
584 	 * As per x86, we may deadlock here. However, since the kernel only
585 	 * validly references user space from well defined areas of the code,
586 	 * we can bug out early if this is from code which shouldn't.
587 	 */
588 	if (!mmap_read_trylock(mm)) {
589 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
590 			goto no_context;
591 retry:
592 		mmap_read_lock(mm);
593 	} else {
594 		/*
595 		 * The above mmap_read_trylock() might have succeeded in which
596 		 * case, we'll have missed the might_sleep() from down_read().
597 		 */
598 		might_sleep();
599 #ifdef CONFIG_DEBUG_VM
600 		if (!user_mode(regs) && !search_exception_tables(regs->pc)) {
601 			mmap_read_unlock(mm);
602 			goto no_context;
603 		}
604 #endif
605 	}
606 
607 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, regs);
608 
609 	/* Quick path to respond to signals */
610 	if (fault_signal_pending(fault, regs)) {
611 		if (!user_mode(regs))
612 			goto no_context;
613 		return 0;
614 	}
615 
616 	if (fault & VM_FAULT_RETRY) {
617 		if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
618 			mm_flags |= FAULT_FLAG_TRIED;
619 			goto retry;
620 		}
621 	}
622 	mmap_read_unlock(mm);
623 
624 	/*
625 	 * Handle the "normal" (no error) case first.
626 	 */
627 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
628 			      VM_FAULT_BADACCESS))))
629 		return 0;
630 
631 	/*
632 	 * If we are in kernel mode at this point, we have no context to
633 	 * handle this fault with.
634 	 */
635 	if (!user_mode(regs))
636 		goto no_context;
637 
638 	if (fault & VM_FAULT_OOM) {
639 		/*
640 		 * We ran out of memory, call the OOM killer, and return to
641 		 * userspace (which will retry the fault, or kill us if we got
642 		 * oom-killed).
643 		 */
644 		pagefault_out_of_memory();
645 		return 0;
646 	}
647 
648 	inf = esr_to_fault_info(esr);
649 	set_thread_esr(addr, esr);
650 	if (fault & VM_FAULT_SIGBUS) {
651 		/*
652 		 * We had some memory, but were unable to successfully fix up
653 		 * this page fault.
654 		 */
655 		arm64_force_sig_fault(SIGBUS, BUS_ADRERR, far, inf->name);
656 	} else if (fault & (VM_FAULT_HWPOISON_LARGE | VM_FAULT_HWPOISON)) {
657 		unsigned int lsb;
658 
659 		lsb = PAGE_SHIFT;
660 		if (fault & VM_FAULT_HWPOISON_LARGE)
661 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
662 
663 		arm64_force_sig_mceerr(BUS_MCEERR_AR, far, lsb, inf->name);
664 	} else {
665 		/*
666 		 * Something tried to access memory that isn't in our memory
667 		 * map.
668 		 */
669 		arm64_force_sig_fault(SIGSEGV,
670 				      fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR,
671 				      far, inf->name);
672 	}
673 
674 	return 0;
675 
676 no_context:
677 	__do_kernel_fault(addr, esr, regs);
678 	return 0;
679 }
680 
681 static int __kprobes do_translation_fault(unsigned long far,
682 					  unsigned int esr,
683 					  struct pt_regs *regs)
684 {
685 	unsigned long addr = untagged_addr(far);
686 
687 	if (is_ttbr0_addr(addr))
688 		return do_page_fault(far, esr, regs);
689 
690 	do_bad_area(far, esr, regs);
691 	return 0;
692 }
693 
694 static int do_alignment_fault(unsigned long far, unsigned int esr,
695 			      struct pt_regs *regs)
696 {
697 	do_bad_area(far, esr, regs);
698 	return 0;
699 }
700 
701 static int do_bad(unsigned long far, unsigned int esr, struct pt_regs *regs)
702 {
703 	return 1; /* "fault" */
704 }
705 
706 static int do_sea(unsigned long far, unsigned int esr, struct pt_regs *regs)
707 {
708 	const struct fault_info *inf;
709 	unsigned long siaddr;
710 
711 	inf = esr_to_fault_info(esr);
712 
713 	if (user_mode(regs) && apei_claim_sea(regs) == 0) {
714 		/*
715 		 * APEI claimed this as a firmware-first notification.
716 		 * Some processing deferred to task_work before ret_to_user().
717 		 */
718 		return 0;
719 	}
720 
721 	if (esr & ESR_ELx_FnV) {
722 		siaddr = 0;
723 	} else {
724 		/*
725 		 * The architecture specifies that the tag bits of FAR_EL1 are
726 		 * UNKNOWN for synchronous external aborts. Mask them out now
727 		 * so that userspace doesn't see them.
728 		 */
729 		siaddr  = untagged_addr(far);
730 	}
731 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
732 
733 	return 0;
734 }
735 
736 static int do_tag_check_fault(unsigned long far, unsigned int esr,
737 			      struct pt_regs *regs)
738 {
739 	/*
740 	 * The architecture specifies that bits 63:60 of FAR_EL1 are UNKNOWN
741 	 * for tag check faults. Set them to corresponding bits in the untagged
742 	 * address.
743 	 */
744 	far = (__untagged_addr(far) & ~MTE_TAG_MASK) | (far & MTE_TAG_MASK);
745 	do_bad_area(far, esr, regs);
746 	return 0;
747 }
748 
749 static const struct fault_info fault_info[] = {
750 	{ do_bad,		SIGKILL, SI_KERNEL,	"ttbr address size fault"	},
751 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 1 address size fault"	},
752 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 2 address size fault"	},
753 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 3 address size fault"	},
754 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
755 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
756 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
757 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
758 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 8"			},
759 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
760 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
761 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
762 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 12"			},
763 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
764 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
765 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
766 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous external abort"	},
767 	{ do_tag_check_fault,	SIGSEGV, SEGV_MTESERR,	"synchronous tag check fault"	},
768 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 18"			},
769 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 19"			},
770 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 (translation table walk)"	},
771 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 (translation table walk)"	},
772 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 (translation table walk)"	},
773 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 (translation table walk)"	},
774 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous parity or ECC error" },	// Reserved when RAS is implemented
775 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 25"			},
776 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 26"			},
777 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 27"			},
778 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
779 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
780 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
781 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
782 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 32"			},
783 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
784 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 34"			},
785 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 35"			},
786 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 36"			},
787 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 37"			},
788 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 38"			},
789 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 39"			},
790 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 40"			},
791 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 41"			},
792 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 42"			},
793 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 43"			},
794 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 44"			},
795 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 45"			},
796 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 46"			},
797 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 47"			},
798 	{ do_bad,		SIGKILL, SI_KERNEL,	"TLB conflict abort"		},
799 	{ do_bad,		SIGKILL, SI_KERNEL,	"Unsupported atomic hardware update fault"	},
800 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 50"			},
801 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 51"			},
802 	{ do_bad,		SIGKILL, SI_KERNEL,	"implementation fault (lockdown abort)" },
803 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"implementation fault (unsupported exclusive)" },
804 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 54"			},
805 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 55"			},
806 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 56"			},
807 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 57"			},
808 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 58" 			},
809 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 59"			},
810 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 60"			},
811 	{ do_bad,		SIGKILL, SI_KERNEL,	"section domain fault"		},
812 	{ do_bad,		SIGKILL, SI_KERNEL,	"page domain fault"		},
813 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 63"			},
814 };
815 
816 void do_mem_abort(unsigned long far, unsigned int esr, struct pt_regs *regs)
817 {
818 	const struct fault_info *inf = esr_to_fault_info(esr);
819 	unsigned long addr = untagged_addr(far);
820 
821 	if (!inf->fn(far, esr, regs))
822 		return;
823 
824 	if (!user_mode(regs)) {
825 		pr_alert("Unhandled fault at 0x%016lx\n", addr);
826 		mem_abort_decode(esr);
827 		show_pte(addr);
828 	}
829 
830 	/*
831 	 * At this point we have an unrecognized fault type whose tag bits may
832 	 * have been defined as UNKNOWN. Therefore we only expose the untagged
833 	 * address to the signal handler.
834 	 */
835 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, addr, esr);
836 }
837 NOKPROBE_SYMBOL(do_mem_abort);
838 
839 void do_el0_irq_bp_hardening(void)
840 {
841 	/* PC has already been checked in entry.S */
842 	arm64_apply_bp_hardening();
843 }
844 NOKPROBE_SYMBOL(do_el0_irq_bp_hardening);
845 
846 void do_sp_pc_abort(unsigned long addr, unsigned int esr, struct pt_regs *regs)
847 {
848 	arm64_notify_die("SP/PC alignment exception", regs, SIGBUS, BUS_ADRALN,
849 			 addr, esr);
850 }
851 NOKPROBE_SYMBOL(do_sp_pc_abort);
852 
853 int __init early_brk64(unsigned long addr, unsigned int esr,
854 		       struct pt_regs *regs);
855 
856 /*
857  * __refdata because early_brk64 is __init, but the reference to it is
858  * clobbered at arch_initcall time.
859  * See traps.c and debug-monitors.c:debug_traps_init().
860  */
861 static struct fault_info __refdata debug_fault_info[] = {
862 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
863 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
864 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
865 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 3"		},
866 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
867 	{ do_bad,	SIGKILL,	SI_KERNEL,	"aarch32 vector catch"	},
868 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
869 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 7"		},
870 };
871 
872 void __init hook_debug_fault_code(int nr,
873 				  int (*fn)(unsigned long, unsigned int, struct pt_regs *),
874 				  int sig, int code, const char *name)
875 {
876 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
877 
878 	debug_fault_info[nr].fn		= fn;
879 	debug_fault_info[nr].sig	= sig;
880 	debug_fault_info[nr].code	= code;
881 	debug_fault_info[nr].name	= name;
882 }
883 
884 /*
885  * In debug exception context, we explicitly disable preemption despite
886  * having interrupts disabled.
887  * This serves two purposes: it makes it much less likely that we would
888  * accidentally schedule in exception context and it will force a warning
889  * if we somehow manage to schedule by accident.
890  */
891 static void debug_exception_enter(struct pt_regs *regs)
892 {
893 	preempt_disable();
894 
895 	/* This code is a bit fragile.  Test it. */
896 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "exception_enter didn't work");
897 }
898 NOKPROBE_SYMBOL(debug_exception_enter);
899 
900 static void debug_exception_exit(struct pt_regs *regs)
901 {
902 	preempt_enable_no_resched();
903 }
904 NOKPROBE_SYMBOL(debug_exception_exit);
905 
906 void do_debug_exception(unsigned long addr_if_watchpoint, unsigned int esr,
907 			struct pt_regs *regs)
908 {
909 	const struct fault_info *inf = esr_to_debug_fault_info(esr);
910 	unsigned long pc = instruction_pointer(regs);
911 
912 	debug_exception_enter(regs);
913 
914 	if (user_mode(regs) && !is_ttbr0_addr(pc))
915 		arm64_apply_bp_hardening();
916 
917 	if (inf->fn(addr_if_watchpoint, esr, regs)) {
918 		arm64_notify_die(inf->name, regs, inf->sig, inf->code, pc, esr);
919 	}
920 
921 	debug_exception_exit(regs);
922 }
923 NOKPROBE_SYMBOL(do_debug_exception);
924