xref: /linux/arch/arm64/mm/fault.c (revision aa23aa55166c2865ac430168c4b9d405cf8c6980)
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/extable.h>
12 #include <linux/signal.h>
13 #include <linux/mm.h>
14 #include <linux/hardirq.h>
15 #include <linux/init.h>
16 #include <linux/kprobes.h>
17 #include <linux/uaccess.h>
18 #include <linux/page-flags.h>
19 #include <linux/sched/signal.h>
20 #include <linux/sched/debug.h>
21 #include <linux/highmem.h>
22 #include <linux/perf_event.h>
23 #include <linux/preempt.h>
24 #include <linux/hugetlb.h>
25 
26 #include <asm/acpi.h>
27 #include <asm/bug.h>
28 #include <asm/cmpxchg.h>
29 #include <asm/cpufeature.h>
30 #include <asm/exception.h>
31 #include <asm/daifflags.h>
32 #include <asm/debug-monitors.h>
33 #include <asm/esr.h>
34 #include <asm/kasan.h>
35 #include <asm/sysreg.h>
36 #include <asm/system_misc.h>
37 #include <asm/pgtable.h>
38 #include <asm/tlbflush.h>
39 #include <asm/traps.h>
40 
41 struct fault_info {
42 	int	(*fn)(unsigned long addr, unsigned int esr,
43 		      struct pt_regs *regs);
44 	int	sig;
45 	int	code;
46 	const char *name;
47 };
48 
49 static const struct fault_info fault_info[];
50 static struct fault_info debug_fault_info[];
51 
52 static inline const struct fault_info *esr_to_fault_info(unsigned int esr)
53 {
54 	return fault_info + (esr & ESR_ELx_FSC);
55 }
56 
57 static inline const struct fault_info *esr_to_debug_fault_info(unsigned int esr)
58 {
59 	return debug_fault_info + DBG_ESR_EVT(esr);
60 }
61 
62 #ifdef CONFIG_KPROBES
63 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
64 {
65 	int ret = 0;
66 
67 	/* kprobe_running() needs smp_processor_id() */
68 	if (!user_mode(regs)) {
69 		preempt_disable();
70 		if (kprobe_running() && kprobe_fault_handler(regs, esr))
71 			ret = 1;
72 		preempt_enable();
73 	}
74 
75 	return ret;
76 }
77 #else
78 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
79 {
80 	return 0;
81 }
82 #endif
83 
84 static void data_abort_decode(unsigned int esr)
85 {
86 	pr_alert("Data abort info:\n");
87 
88 	if (esr & ESR_ELx_ISV) {
89 		pr_alert("  Access size = %u byte(s)\n",
90 			 1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
91 		pr_alert("  SSE = %lu, SRT = %lu\n",
92 			 (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
93 			 (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
94 		pr_alert("  SF = %lu, AR = %lu\n",
95 			 (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
96 			 (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
97 	} else {
98 		pr_alert("  ISV = 0, ISS = 0x%08lx\n", esr & ESR_ELx_ISS_MASK);
99 	}
100 
101 	pr_alert("  CM = %lu, WnR = %lu\n",
102 		 (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
103 		 (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT);
104 }
105 
106 static void mem_abort_decode(unsigned int esr)
107 {
108 	pr_alert("Mem abort info:\n");
109 
110 	pr_alert("  ESR = 0x%08x\n", esr);
111 	pr_alert("  Exception class = %s, IL = %u bits\n",
112 		 esr_get_class_string(esr),
113 		 (esr & ESR_ELx_IL) ? 32 : 16);
114 	pr_alert("  SET = %lu, FnV = %lu\n",
115 		 (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
116 		 (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
117 	pr_alert("  EA = %lu, S1PTW = %lu\n",
118 		 (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
119 		 (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
120 
121 	if (esr_is_data_abort(esr))
122 		data_abort_decode(esr);
123 }
124 
125 static inline bool is_ttbr0_addr(unsigned long addr)
126 {
127 	/* entry assembly clears tags for TTBR0 addrs */
128 	return addr < TASK_SIZE;
129 }
130 
131 static inline bool is_ttbr1_addr(unsigned long addr)
132 {
133 	/* TTBR1 addresses may have a tag if KASAN_SW_TAGS is in use */
134 	return arch_kasan_reset_tag(addr) >= VA_START;
135 }
136 
137 /*
138  * Dump out the page tables associated with 'addr' in the currently active mm.
139  */
140 static void show_pte(unsigned long addr)
141 {
142 	struct mm_struct *mm;
143 	pgd_t *pgdp;
144 	pgd_t pgd;
145 
146 	if (is_ttbr0_addr(addr)) {
147 		/* TTBR0 */
148 		mm = current->active_mm;
149 		if (mm == &init_mm) {
150 			pr_alert("[%016lx] user address but active_mm is swapper\n",
151 				 addr);
152 			return;
153 		}
154 	} else if (is_ttbr1_addr(addr)) {
155 		/* TTBR1 */
156 		mm = &init_mm;
157 	} else {
158 		pr_alert("[%016lx] address between user and kernel address ranges\n",
159 			 addr);
160 		return;
161 	}
162 
163 	pr_alert("%s pgtable: %luk pages, %u-bit VAs, pgdp=%016lx\n",
164 		 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
165 		 mm == &init_mm ? VA_BITS : (int)vabits_user,
166 		 (unsigned long)virt_to_phys(mm->pgd));
167 	pgdp = pgd_offset(mm, addr);
168 	pgd = READ_ONCE(*pgdp);
169 	pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
170 
171 	do {
172 		pud_t *pudp, pud;
173 		pmd_t *pmdp, pmd;
174 		pte_t *ptep, pte;
175 
176 		if (pgd_none(pgd) || pgd_bad(pgd))
177 			break;
178 
179 		pudp = pud_offset(pgdp, addr);
180 		pud = READ_ONCE(*pudp);
181 		pr_cont(", pud=%016llx", pud_val(pud));
182 		if (pud_none(pud) || pud_bad(pud))
183 			break;
184 
185 		pmdp = pmd_offset(pudp, addr);
186 		pmd = READ_ONCE(*pmdp);
187 		pr_cont(", pmd=%016llx", pmd_val(pmd));
188 		if (pmd_none(pmd) || pmd_bad(pmd))
189 			break;
190 
191 		ptep = pte_offset_map(pmdp, addr);
192 		pte = READ_ONCE(*ptep);
193 		pr_cont(", pte=%016llx", pte_val(pte));
194 		pte_unmap(ptep);
195 	} while(0);
196 
197 	pr_cont("\n");
198 }
199 
200 /*
201  * This function sets the access flags (dirty, accessed), as well as write
202  * permission, and only to a more permissive setting.
203  *
204  * It needs to cope with hardware update of the accessed/dirty state by other
205  * agents in the system and can safely skip the __sync_icache_dcache() call as,
206  * like set_pte_at(), the PTE is never changed from no-exec to exec here.
207  *
208  * Returns whether or not the PTE actually changed.
209  */
210 int ptep_set_access_flags(struct vm_area_struct *vma,
211 			  unsigned long address, pte_t *ptep,
212 			  pte_t entry, int dirty)
213 {
214 	pteval_t old_pteval, pteval;
215 	pte_t pte = READ_ONCE(*ptep);
216 
217 	if (pte_same(pte, entry))
218 		return 0;
219 
220 	/* only preserve the access flags and write permission */
221 	pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
222 
223 	/*
224 	 * Setting the flags must be done atomically to avoid racing with the
225 	 * hardware update of the access/dirty state. The PTE_RDONLY bit must
226 	 * be set to the most permissive (lowest value) of *ptep and entry
227 	 * (calculated as: a & b == ~(~a | ~b)).
228 	 */
229 	pte_val(entry) ^= PTE_RDONLY;
230 	pteval = pte_val(pte);
231 	do {
232 		old_pteval = pteval;
233 		pteval ^= PTE_RDONLY;
234 		pteval |= pte_val(entry);
235 		pteval ^= PTE_RDONLY;
236 		pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
237 	} while (pteval != old_pteval);
238 
239 	flush_tlb_fix_spurious_fault(vma, address);
240 	return 1;
241 }
242 
243 static bool is_el1_instruction_abort(unsigned int esr)
244 {
245 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
246 }
247 
248 static inline bool is_el1_permission_fault(unsigned long addr, unsigned int esr,
249 					   struct pt_regs *regs)
250 {
251 	unsigned int ec       = ESR_ELx_EC(esr);
252 	unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
253 
254 	if (ec != ESR_ELx_EC_DABT_CUR && ec != ESR_ELx_EC_IABT_CUR)
255 		return false;
256 
257 	if (fsc_type == ESR_ELx_FSC_PERM)
258 		return true;
259 
260 	if (is_ttbr0_addr(addr) && system_uses_ttbr0_pan())
261 		return fsc_type == ESR_ELx_FSC_FAULT &&
262 			(regs->pstate & PSR_PAN_BIT);
263 
264 	return false;
265 }
266 
267 static void die_kernel_fault(const char *msg, unsigned long addr,
268 			     unsigned int esr, struct pt_regs *regs)
269 {
270 	bust_spinlocks(1);
271 
272 	pr_alert("Unable to handle kernel %s at virtual address %016lx\n", msg,
273 		 addr);
274 
275 	mem_abort_decode(esr);
276 
277 	show_pte(addr);
278 	die("Oops", regs, esr);
279 	bust_spinlocks(0);
280 	do_exit(SIGKILL);
281 }
282 
283 static void __do_kernel_fault(unsigned long addr, unsigned int esr,
284 			      struct pt_regs *regs)
285 {
286 	const char *msg;
287 
288 	/*
289 	 * Are we prepared to handle this kernel fault?
290 	 * We are almost certainly not prepared to handle instruction faults.
291 	 */
292 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
293 		return;
294 
295 	if (is_el1_permission_fault(addr, esr, regs)) {
296 		if (esr & ESR_ELx_WNR)
297 			msg = "write to read-only memory";
298 		else
299 			msg = "read from unreadable memory";
300 	} else if (addr < PAGE_SIZE) {
301 		msg = "NULL pointer dereference";
302 	} else {
303 		msg = "paging request";
304 	}
305 
306 	die_kernel_fault(msg, addr, esr, regs);
307 }
308 
309 static void set_thread_esr(unsigned long address, unsigned int esr)
310 {
311 	current->thread.fault_address = address;
312 
313 	/*
314 	 * If the faulting address is in the kernel, we must sanitize the ESR.
315 	 * From userspace's point of view, kernel-only mappings don't exist
316 	 * at all, so we report them as level 0 translation faults.
317 	 * (This is not quite the way that "no mapping there at all" behaves:
318 	 * an alignment fault not caused by the memory type would take
319 	 * precedence over translation fault for a real access to empty
320 	 * space. Unfortunately we can't easily distinguish "alignment fault
321 	 * not caused by memory type" from "alignment fault caused by memory
322 	 * type", so we ignore this wrinkle and just return the translation
323 	 * fault.)
324 	 */
325 	if (!is_ttbr0_addr(current->thread.fault_address)) {
326 		switch (ESR_ELx_EC(esr)) {
327 		case ESR_ELx_EC_DABT_LOW:
328 			/*
329 			 * These bits provide only information about the
330 			 * faulting instruction, which userspace knows already.
331 			 * We explicitly clear bits which are architecturally
332 			 * RES0 in case they are given meanings in future.
333 			 * We always report the ESR as if the fault was taken
334 			 * to EL1 and so ISV and the bits in ISS[23:14] are
335 			 * clear. (In fact it always will be a fault to EL1.)
336 			 */
337 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL |
338 				ESR_ELx_CM | ESR_ELx_WNR;
339 			esr |= ESR_ELx_FSC_FAULT;
340 			break;
341 		case ESR_ELx_EC_IABT_LOW:
342 			/*
343 			 * Claim a level 0 translation fault.
344 			 * All other bits are architecturally RES0 for faults
345 			 * reported with that DFSC value, so we clear them.
346 			 */
347 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL;
348 			esr |= ESR_ELx_FSC_FAULT;
349 			break;
350 		default:
351 			/*
352 			 * This should never happen (entry.S only brings us
353 			 * into this code for insn and data aborts from a lower
354 			 * exception level). Fail safe by not providing an ESR
355 			 * context record at all.
356 			 */
357 			WARN(1, "ESR 0x%x is not DABT or IABT from EL0\n", esr);
358 			esr = 0;
359 			break;
360 		}
361 	}
362 
363 	current->thread.fault_code = esr;
364 }
365 
366 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
367 {
368 	/*
369 	 * If we are in kernel mode at this point, we have no context to
370 	 * handle this fault with.
371 	 */
372 	if (user_mode(regs)) {
373 		const struct fault_info *inf = esr_to_fault_info(esr);
374 
375 		set_thread_esr(addr, esr);
376 		arm64_force_sig_fault(inf->sig, inf->code, (void __user *)addr,
377 				      inf->name);
378 	} else {
379 		__do_kernel_fault(addr, esr, regs);
380 	}
381 }
382 
383 #define VM_FAULT_BADMAP		0x010000
384 #define VM_FAULT_BADACCESS	0x020000
385 
386 static vm_fault_t __do_page_fault(struct mm_struct *mm, unsigned long addr,
387 			   unsigned int mm_flags, unsigned long vm_flags,
388 			   struct task_struct *tsk)
389 {
390 	struct vm_area_struct *vma;
391 	vm_fault_t fault;
392 
393 	vma = find_vma(mm, addr);
394 	fault = VM_FAULT_BADMAP;
395 	if (unlikely(!vma))
396 		goto out;
397 	if (unlikely(vma->vm_start > addr))
398 		goto check_stack;
399 
400 	/*
401 	 * Ok, we have a good vm_area for this memory access, so we can handle
402 	 * it.
403 	 */
404 good_area:
405 	/*
406 	 * Check that the permissions on the VMA allow for the fault which
407 	 * occurred.
408 	 */
409 	if (!(vma->vm_flags & vm_flags)) {
410 		fault = VM_FAULT_BADACCESS;
411 		goto out;
412 	}
413 
414 	return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags);
415 
416 check_stack:
417 	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
418 		goto good_area;
419 out:
420 	return fault;
421 }
422 
423 static bool is_el0_instruction_abort(unsigned int esr)
424 {
425 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
426 }
427 
428 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
429 				   struct pt_regs *regs)
430 {
431 	const struct fault_info *inf;
432 	struct task_struct *tsk;
433 	struct mm_struct *mm;
434 	vm_fault_t fault, major = 0;
435 	unsigned long vm_flags = VM_READ | VM_WRITE;
436 	unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
437 
438 	if (notify_page_fault(regs, esr))
439 		return 0;
440 
441 	tsk = current;
442 	mm  = tsk->mm;
443 
444 	/*
445 	 * If we're in an interrupt or have no user context, we must not take
446 	 * the fault.
447 	 */
448 	if (faulthandler_disabled() || !mm)
449 		goto no_context;
450 
451 	if (user_mode(regs))
452 		mm_flags |= FAULT_FLAG_USER;
453 
454 	if (is_el0_instruction_abort(esr)) {
455 		vm_flags = VM_EXEC;
456 	} else if ((esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM)) {
457 		vm_flags = VM_WRITE;
458 		mm_flags |= FAULT_FLAG_WRITE;
459 	}
460 
461 	if (is_ttbr0_addr(addr) && is_el1_permission_fault(addr, esr, regs)) {
462 		/* regs->orig_addr_limit may be 0 if we entered from EL0 */
463 		if (regs->orig_addr_limit == KERNEL_DS)
464 			die_kernel_fault("access to user memory with fs=KERNEL_DS",
465 					 addr, esr, regs);
466 
467 		if (is_el1_instruction_abort(esr))
468 			die_kernel_fault("execution of user memory",
469 					 addr, esr, regs);
470 
471 		if (!search_exception_tables(regs->pc))
472 			die_kernel_fault("access to user memory outside uaccess routines",
473 					 addr, esr, regs);
474 	}
475 
476 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
477 
478 	/*
479 	 * As per x86, we may deadlock here. However, since the kernel only
480 	 * validly references user space from well defined areas of the code,
481 	 * we can bug out early if this is from code which shouldn't.
482 	 */
483 	if (!down_read_trylock(&mm->mmap_sem)) {
484 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
485 			goto no_context;
486 retry:
487 		down_read(&mm->mmap_sem);
488 	} else {
489 		/*
490 		 * The above down_read_trylock() might have succeeded in which
491 		 * case, we'll have missed the might_sleep() from down_read().
492 		 */
493 		might_sleep();
494 #ifdef CONFIG_DEBUG_VM
495 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
496 			goto no_context;
497 #endif
498 	}
499 
500 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, tsk);
501 	major |= fault & VM_FAULT_MAJOR;
502 
503 	if (fault & VM_FAULT_RETRY) {
504 		/*
505 		 * If we need to retry but a fatal signal is pending,
506 		 * handle the signal first. We do not need to release
507 		 * the mmap_sem because it would already be released
508 		 * in __lock_page_or_retry in mm/filemap.c.
509 		 */
510 		if (fatal_signal_pending(current)) {
511 			if (!user_mode(regs))
512 				goto no_context;
513 			return 0;
514 		}
515 
516 		/*
517 		 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
518 		 * starvation.
519 		 */
520 		if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
521 			mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
522 			mm_flags |= FAULT_FLAG_TRIED;
523 			goto retry;
524 		}
525 	}
526 	up_read(&mm->mmap_sem);
527 
528 	/*
529 	 * Handle the "normal" (no error) case first.
530 	 */
531 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
532 			      VM_FAULT_BADACCESS)))) {
533 		/*
534 		 * Major/minor page fault accounting is only done
535 		 * once. If we go through a retry, it is extremely
536 		 * likely that the page will be found in page cache at
537 		 * that point.
538 		 */
539 		if (major) {
540 			tsk->maj_flt++;
541 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
542 				      addr);
543 		} else {
544 			tsk->min_flt++;
545 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
546 				      addr);
547 		}
548 
549 		return 0;
550 	}
551 
552 	/*
553 	 * If we are in kernel mode at this point, we have no context to
554 	 * handle this fault with.
555 	 */
556 	if (!user_mode(regs))
557 		goto no_context;
558 
559 	if (fault & VM_FAULT_OOM) {
560 		/*
561 		 * We ran out of memory, call the OOM killer, and return to
562 		 * userspace (which will retry the fault, or kill us if we got
563 		 * oom-killed).
564 		 */
565 		pagefault_out_of_memory();
566 		return 0;
567 	}
568 
569 	inf = esr_to_fault_info(esr);
570 	set_thread_esr(addr, esr);
571 	if (fault & VM_FAULT_SIGBUS) {
572 		/*
573 		 * We had some memory, but were unable to successfully fix up
574 		 * this page fault.
575 		 */
576 		arm64_force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)addr,
577 				      inf->name);
578 	} else if (fault & (VM_FAULT_HWPOISON_LARGE | VM_FAULT_HWPOISON)) {
579 		unsigned int lsb;
580 
581 		lsb = PAGE_SHIFT;
582 		if (fault & VM_FAULT_HWPOISON_LARGE)
583 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
584 
585 		arm64_force_sig_mceerr(BUS_MCEERR_AR, (void __user *)addr, lsb,
586 				       inf->name);
587 	} else {
588 		/*
589 		 * Something tried to access memory that isn't in our memory
590 		 * map.
591 		 */
592 		arm64_force_sig_fault(SIGSEGV,
593 				      fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR,
594 				      (void __user *)addr,
595 				      inf->name);
596 	}
597 
598 	return 0;
599 
600 no_context:
601 	__do_kernel_fault(addr, esr, regs);
602 	return 0;
603 }
604 
605 static int __kprobes do_translation_fault(unsigned long addr,
606 					  unsigned int esr,
607 					  struct pt_regs *regs)
608 {
609 	if (is_ttbr0_addr(addr))
610 		return do_page_fault(addr, esr, regs);
611 
612 	do_bad_area(addr, esr, regs);
613 	return 0;
614 }
615 
616 static int do_alignment_fault(unsigned long addr, unsigned int esr,
617 			      struct pt_regs *regs)
618 {
619 	do_bad_area(addr, esr, regs);
620 	return 0;
621 }
622 
623 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
624 {
625 	return 1; /* "fault" */
626 }
627 
628 static int do_sea(unsigned long addr, unsigned int esr, struct pt_regs *regs)
629 {
630 	const struct fault_info *inf;
631 	void __user *siaddr;
632 
633 	inf = esr_to_fault_info(esr);
634 
635 	/*
636 	 * Return value ignored as we rely on signal merging.
637 	 * Future patches will make this more robust.
638 	 */
639 	apei_claim_sea(regs);
640 
641 	if (esr & ESR_ELx_FnV)
642 		siaddr = NULL;
643 	else
644 		siaddr  = (void __user *)addr;
645 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
646 
647 	return 0;
648 }
649 
650 static const struct fault_info fault_info[] = {
651 	{ do_bad,		SIGKILL, SI_KERNEL,	"ttbr address size fault"	},
652 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 1 address size fault"	},
653 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 2 address size fault"	},
654 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 3 address size fault"	},
655 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
656 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
657 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
658 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
659 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 8"			},
660 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
661 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
662 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
663 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 12"			},
664 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
665 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
666 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
667 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous external abort"	},
668 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 17"			},
669 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 18"			},
670 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 19"			},
671 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 (translation table walk)"	},
672 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 (translation table walk)"	},
673 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 (translation table walk)"	},
674 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 (translation table walk)"	},
675 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous parity or ECC error" },	// Reserved when RAS is implemented
676 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 25"			},
677 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 26"			},
678 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 27"			},
679 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
680 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
681 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
682 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
683 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 32"			},
684 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
685 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 34"			},
686 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 35"			},
687 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 36"			},
688 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 37"			},
689 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 38"			},
690 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 39"			},
691 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 40"			},
692 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 41"			},
693 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 42"			},
694 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 43"			},
695 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 44"			},
696 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 45"			},
697 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 46"			},
698 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 47"			},
699 	{ do_bad,		SIGKILL, SI_KERNEL,	"TLB conflict abort"		},
700 	{ do_bad,		SIGKILL, SI_KERNEL,	"Unsupported atomic hardware update fault"	},
701 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 50"			},
702 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 51"			},
703 	{ do_bad,		SIGKILL, SI_KERNEL,	"implementation fault (lockdown abort)" },
704 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"implementation fault (unsupported exclusive)" },
705 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 54"			},
706 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 55"			},
707 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 56"			},
708 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 57"			},
709 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 58" 			},
710 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 59"			},
711 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 60"			},
712 	{ do_bad,		SIGKILL, SI_KERNEL,	"section domain fault"		},
713 	{ do_bad,		SIGKILL, SI_KERNEL,	"page domain fault"		},
714 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 63"			},
715 };
716 
717 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
718 					 struct pt_regs *regs)
719 {
720 	const struct fault_info *inf = esr_to_fault_info(esr);
721 
722 	if (!inf->fn(addr, esr, regs))
723 		return;
724 
725 	if (!user_mode(regs)) {
726 		pr_alert("Unhandled fault at 0x%016lx\n", addr);
727 		mem_abort_decode(esr);
728 		show_pte(addr);
729 	}
730 
731 	arm64_notify_die(inf->name, regs,
732 			 inf->sig, inf->code, (void __user *)addr, esr);
733 }
734 
735 asmlinkage void __exception do_el0_irq_bp_hardening(void)
736 {
737 	/* PC has already been checked in entry.S */
738 	arm64_apply_bp_hardening();
739 }
740 
741 asmlinkage void __exception do_el0_ia_bp_hardening(unsigned long addr,
742 						   unsigned int esr,
743 						   struct pt_regs *regs)
744 {
745 	/*
746 	 * We've taken an instruction abort from userspace and not yet
747 	 * re-enabled IRQs. If the address is a kernel address, apply
748 	 * BP hardening prior to enabling IRQs and pre-emption.
749 	 */
750 	if (!is_ttbr0_addr(addr))
751 		arm64_apply_bp_hardening();
752 
753 	local_daif_restore(DAIF_PROCCTX);
754 	do_mem_abort(addr, esr, regs);
755 }
756 
757 
758 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
759 					   unsigned int esr,
760 					   struct pt_regs *regs)
761 {
762 	if (user_mode(regs)) {
763 		if (!is_ttbr0_addr(instruction_pointer(regs)))
764 			arm64_apply_bp_hardening();
765 		local_daif_restore(DAIF_PROCCTX);
766 	}
767 
768 	arm64_notify_die("SP/PC alignment exception", regs,
769 			 SIGBUS, BUS_ADRALN, (void __user *)addr, esr);
770 }
771 
772 int __init early_brk64(unsigned long addr, unsigned int esr,
773 		       struct pt_regs *regs);
774 
775 /*
776  * __refdata because early_brk64 is __init, but the reference to it is
777  * clobbered at arch_initcall time.
778  * See traps.c and debug-monitors.c:debug_traps_init().
779  */
780 static struct fault_info __refdata debug_fault_info[] = {
781 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
782 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
783 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
784 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 3"		},
785 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
786 	{ do_bad,	SIGKILL,	SI_KERNEL,	"aarch32 vector catch"	},
787 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
788 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 7"		},
789 };
790 
791 void __init hook_debug_fault_code(int nr,
792 				  int (*fn)(unsigned long, unsigned int, struct pt_regs *),
793 				  int sig, int code, const char *name)
794 {
795 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
796 
797 	debug_fault_info[nr].fn		= fn;
798 	debug_fault_info[nr].sig	= sig;
799 	debug_fault_info[nr].code	= code;
800 	debug_fault_info[nr].name	= name;
801 }
802 
803 #ifdef CONFIG_ARM64_ERRATUM_1463225
804 DECLARE_PER_CPU(int, __in_cortex_a76_erratum_1463225_wa);
805 
806 static int __exception
807 cortex_a76_erratum_1463225_debug_handler(struct pt_regs *regs)
808 {
809 	if (user_mode(regs))
810 		return 0;
811 
812 	if (!__this_cpu_read(__in_cortex_a76_erratum_1463225_wa))
813 		return 0;
814 
815 	/*
816 	 * We've taken a dummy step exception from the kernel to ensure
817 	 * that interrupts are re-enabled on the syscall path. Return back
818 	 * to cortex_a76_erratum_1463225_svc_handler() with debug exceptions
819 	 * masked so that we can safely restore the mdscr and get on with
820 	 * handling the syscall.
821 	 */
822 	regs->pstate |= PSR_D_BIT;
823 	return 1;
824 }
825 #else
826 static int __exception
827 cortex_a76_erratum_1463225_debug_handler(struct pt_regs *regs)
828 {
829 	return 0;
830 }
831 #endif /* CONFIG_ARM64_ERRATUM_1463225 */
832 
833 asmlinkage void __exception do_debug_exception(unsigned long addr_if_watchpoint,
834 					       unsigned int esr,
835 					       struct pt_regs *regs)
836 {
837 	const struct fault_info *inf = esr_to_debug_fault_info(esr);
838 	unsigned long pc = instruction_pointer(regs);
839 
840 	if (cortex_a76_erratum_1463225_debug_handler(regs))
841 		return;
842 
843 	/*
844 	 * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
845 	 * already disabled to preserve the last enabled/disabled addresses.
846 	 */
847 	if (interrupts_enabled(regs))
848 		trace_hardirqs_off();
849 
850 	if (user_mode(regs) && !is_ttbr0_addr(pc))
851 		arm64_apply_bp_hardening();
852 
853 	if (inf->fn(addr_if_watchpoint, esr, regs)) {
854 		arm64_notify_die(inf->name, regs,
855 				 inf->sig, inf->code, (void __user *)pc, esr);
856 	}
857 
858 	if (interrupts_enabled(regs))
859 		trace_hardirqs_on();
860 }
861 NOKPROBE_SYMBOL(do_debug_exception);
862