xref: /openbmc/linux/arch/arm64/mm/fault.c (revision aeefc1a0)
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/signal.h>
14 #include <linux/mm.h>
15 #include <linux/hardirq.h>
16 #include <linux/init.h>
17 #include <linux/kprobes.h>
18 #include <linux/uaccess.h>
19 #include <linux/page-flags.h>
20 #include <linux/sched/signal.h>
21 #include <linux/sched/debug.h>
22 #include <linux/highmem.h>
23 #include <linux/perf_event.h>
24 #include <linux/preempt.h>
25 #include <linux/hugetlb.h>
26 
27 #include <asm/acpi.h>
28 #include <asm/bug.h>
29 #include <asm/cmpxchg.h>
30 #include <asm/cpufeature.h>
31 #include <asm/exception.h>
32 #include <asm/daifflags.h>
33 #include <asm/debug-monitors.h>
34 #include <asm/esr.h>
35 #include <asm/kprobes.h>
36 #include <asm/processor.h>
37 #include <asm/sysreg.h>
38 #include <asm/system_misc.h>
39 #include <asm/tlbflush.h>
40 #include <asm/traps.h>
41 
42 struct fault_info {
43 	int	(*fn)(unsigned long far, unsigned int esr,
44 		      struct pt_regs *regs);
45 	int	sig;
46 	int	code;
47 	const char *name;
48 };
49 
50 static const struct fault_info fault_info[];
51 static struct fault_info debug_fault_info[];
52 
53 static inline const struct fault_info *esr_to_fault_info(unsigned int esr)
54 {
55 	return fault_info + (esr & ESR_ELx_FSC);
56 }
57 
58 static inline const struct fault_info *esr_to_debug_fault_info(unsigned int esr)
59 {
60 	return debug_fault_info + DBG_ESR_EVT(esr);
61 }
62 
63 static void data_abort_decode(unsigned int esr)
64 {
65 	pr_alert("Data abort info:\n");
66 
67 	if (esr & ESR_ELx_ISV) {
68 		pr_alert("  Access size = %u byte(s)\n",
69 			 1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
70 		pr_alert("  SSE = %lu, SRT = %lu\n",
71 			 (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
72 			 (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
73 		pr_alert("  SF = %lu, AR = %lu\n",
74 			 (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
75 			 (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
76 	} else {
77 		pr_alert("  ISV = 0, ISS = 0x%08lx\n", esr & ESR_ELx_ISS_MASK);
78 	}
79 
80 	pr_alert("  CM = %lu, WnR = %lu\n",
81 		 (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
82 		 (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT);
83 }
84 
85 static void mem_abort_decode(unsigned int esr)
86 {
87 	pr_alert("Mem abort info:\n");
88 
89 	pr_alert("  ESR = 0x%08x\n", esr);
90 	pr_alert("  EC = 0x%02lx: %s, IL = %u bits\n",
91 		 ESR_ELx_EC(esr), esr_get_class_string(esr),
92 		 (esr & ESR_ELx_IL) ? 32 : 16);
93 	pr_alert("  SET = %lu, FnV = %lu\n",
94 		 (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
95 		 (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
96 	pr_alert("  EA = %lu, S1PTW = %lu\n",
97 		 (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
98 		 (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
99 
100 	if (esr_is_data_abort(esr))
101 		data_abort_decode(esr);
102 }
103 
104 static inline unsigned long mm_to_pgd_phys(struct mm_struct *mm)
105 {
106 	/* Either init_pg_dir or swapper_pg_dir */
107 	if (mm == &init_mm)
108 		return __pa_symbol(mm->pgd);
109 
110 	return (unsigned long)virt_to_phys(mm->pgd);
111 }
112 
113 /*
114  * Dump out the page tables associated with 'addr' in the currently active mm.
115  */
116 static void show_pte(unsigned long addr)
117 {
118 	struct mm_struct *mm;
119 	pgd_t *pgdp;
120 	pgd_t pgd;
121 
122 	if (is_ttbr0_addr(addr)) {
123 		/* TTBR0 */
124 		mm = current->active_mm;
125 		if (mm == &init_mm) {
126 			pr_alert("[%016lx] user address but active_mm is swapper\n",
127 				 addr);
128 			return;
129 		}
130 	} else if (is_ttbr1_addr(addr)) {
131 		/* TTBR1 */
132 		mm = &init_mm;
133 	} else {
134 		pr_alert("[%016lx] address between user and kernel address ranges\n",
135 			 addr);
136 		return;
137 	}
138 
139 	pr_alert("%s pgtable: %luk pages, %llu-bit VAs, pgdp=%016lx\n",
140 		 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
141 		 vabits_actual, mm_to_pgd_phys(mm));
142 	pgdp = pgd_offset(mm, addr);
143 	pgd = READ_ONCE(*pgdp);
144 	pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
145 
146 	do {
147 		p4d_t *p4dp, p4d;
148 		pud_t *pudp, pud;
149 		pmd_t *pmdp, pmd;
150 		pte_t *ptep, pte;
151 
152 		if (pgd_none(pgd) || pgd_bad(pgd))
153 			break;
154 
155 		p4dp = p4d_offset(pgdp, addr);
156 		p4d = READ_ONCE(*p4dp);
157 		pr_cont(", p4d=%016llx", p4d_val(p4d));
158 		if (p4d_none(p4d) || p4d_bad(p4d))
159 			break;
160 
161 		pudp = pud_offset(p4dp, addr);
162 		pud = READ_ONCE(*pudp);
163 		pr_cont(", pud=%016llx", pud_val(pud));
164 		if (pud_none(pud) || pud_bad(pud))
165 			break;
166 
167 		pmdp = pmd_offset(pudp, addr);
168 		pmd = READ_ONCE(*pmdp);
169 		pr_cont(", pmd=%016llx", pmd_val(pmd));
170 		if (pmd_none(pmd) || pmd_bad(pmd))
171 			break;
172 
173 		ptep = pte_offset_map(pmdp, addr);
174 		pte = READ_ONCE(*ptep);
175 		pr_cont(", pte=%016llx", pte_val(pte));
176 		pte_unmap(ptep);
177 	} while(0);
178 
179 	pr_cont("\n");
180 }
181 
182 /*
183  * This function sets the access flags (dirty, accessed), as well as write
184  * permission, and only to a more permissive setting.
185  *
186  * It needs to cope with hardware update of the accessed/dirty state by other
187  * agents in the system and can safely skip the __sync_icache_dcache() call as,
188  * like set_pte_at(), the PTE is never changed from no-exec to exec here.
189  *
190  * Returns whether or not the PTE actually changed.
191  */
192 int ptep_set_access_flags(struct vm_area_struct *vma,
193 			  unsigned long address, pte_t *ptep,
194 			  pte_t entry, int dirty)
195 {
196 	pteval_t old_pteval, pteval;
197 	pte_t pte = READ_ONCE(*ptep);
198 
199 	if (pte_same(pte, entry))
200 		return 0;
201 
202 	/* only preserve the access flags and write permission */
203 	pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
204 
205 	/*
206 	 * Setting the flags must be done atomically to avoid racing with the
207 	 * hardware update of the access/dirty state. The PTE_RDONLY bit must
208 	 * be set to the most permissive (lowest value) of *ptep and entry
209 	 * (calculated as: a & b == ~(~a | ~b)).
210 	 */
211 	pte_val(entry) ^= PTE_RDONLY;
212 	pteval = pte_val(pte);
213 	do {
214 		old_pteval = pteval;
215 		pteval ^= PTE_RDONLY;
216 		pteval |= pte_val(entry);
217 		pteval ^= PTE_RDONLY;
218 		pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
219 	} while (pteval != old_pteval);
220 
221 	/* Invalidate a stale read-only entry */
222 	if (dirty)
223 		flush_tlb_page(vma, address);
224 	return 1;
225 }
226 
227 static bool is_el1_instruction_abort(unsigned int esr)
228 {
229 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
230 }
231 
232 static inline bool is_el1_permission_fault(unsigned long addr, unsigned int esr,
233 					   struct pt_regs *regs)
234 {
235 	unsigned int ec       = ESR_ELx_EC(esr);
236 	unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
237 
238 	if (ec != ESR_ELx_EC_DABT_CUR && ec != ESR_ELx_EC_IABT_CUR)
239 		return false;
240 
241 	if (fsc_type == ESR_ELx_FSC_PERM)
242 		return true;
243 
244 	if (is_ttbr0_addr(addr) && system_uses_ttbr0_pan())
245 		return fsc_type == ESR_ELx_FSC_FAULT &&
246 			(regs->pstate & PSR_PAN_BIT);
247 
248 	return false;
249 }
250 
251 static bool __kprobes is_spurious_el1_translation_fault(unsigned long addr,
252 							unsigned int esr,
253 							struct pt_regs *regs)
254 {
255 	unsigned long flags;
256 	u64 par, dfsc;
257 
258 	if (ESR_ELx_EC(esr) != ESR_ELx_EC_DABT_CUR ||
259 	    (esr & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT)
260 		return false;
261 
262 	local_irq_save(flags);
263 	asm volatile("at s1e1r, %0" :: "r" (addr));
264 	isb();
265 	par = read_sysreg_par();
266 	local_irq_restore(flags);
267 
268 	/*
269 	 * If we now have a valid translation, treat the translation fault as
270 	 * spurious.
271 	 */
272 	if (!(par & SYS_PAR_EL1_F))
273 		return true;
274 
275 	/*
276 	 * If we got a different type of fault from the AT instruction,
277 	 * treat the translation fault as spurious.
278 	 */
279 	dfsc = FIELD_GET(SYS_PAR_EL1_FST, par);
280 	return (dfsc & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT;
281 }
282 
283 static void die_kernel_fault(const char *msg, unsigned long addr,
284 			     unsigned int esr, struct pt_regs *regs)
285 {
286 	bust_spinlocks(1);
287 
288 	pr_alert("Unable to handle kernel %s at virtual address %016lx\n", msg,
289 		 addr);
290 
291 	mem_abort_decode(esr);
292 
293 	show_pte(addr);
294 	die("Oops", regs, esr);
295 	bust_spinlocks(0);
296 	do_exit(SIGKILL);
297 }
298 
299 static void __do_kernel_fault(unsigned long addr, unsigned int esr,
300 			      struct pt_regs *regs)
301 {
302 	const char *msg;
303 
304 	/*
305 	 * Are we prepared to handle this kernel fault?
306 	 * We are almost certainly not prepared to handle instruction faults.
307 	 */
308 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
309 		return;
310 
311 	if (WARN_RATELIMIT(is_spurious_el1_translation_fault(addr, esr, regs),
312 	    "Ignoring spurious kernel translation fault at virtual address %016lx\n", addr))
313 		return;
314 
315 	if (is_el1_permission_fault(addr, esr, regs)) {
316 		if (esr & ESR_ELx_WNR)
317 			msg = "write to read-only memory";
318 		else if (is_el1_instruction_abort(esr))
319 			msg = "execute from non-executable memory";
320 		else
321 			msg = "read from unreadable memory";
322 	} else if (addr < PAGE_SIZE) {
323 		msg = "NULL pointer dereference";
324 	} else {
325 		msg = "paging request";
326 	}
327 
328 	die_kernel_fault(msg, addr, esr, regs);
329 }
330 
331 static void set_thread_esr(unsigned long address, unsigned int esr)
332 {
333 	current->thread.fault_address = address;
334 
335 	/*
336 	 * If the faulting address is in the kernel, we must sanitize the ESR.
337 	 * From userspace's point of view, kernel-only mappings don't exist
338 	 * at all, so we report them as level 0 translation faults.
339 	 * (This is not quite the way that "no mapping there at all" behaves:
340 	 * an alignment fault not caused by the memory type would take
341 	 * precedence over translation fault for a real access to empty
342 	 * space. Unfortunately we can't easily distinguish "alignment fault
343 	 * not caused by memory type" from "alignment fault caused by memory
344 	 * type", so we ignore this wrinkle and just return the translation
345 	 * fault.)
346 	 */
347 	if (!is_ttbr0_addr(current->thread.fault_address)) {
348 		switch (ESR_ELx_EC(esr)) {
349 		case ESR_ELx_EC_DABT_LOW:
350 			/*
351 			 * These bits provide only information about the
352 			 * faulting instruction, which userspace knows already.
353 			 * We explicitly clear bits which are architecturally
354 			 * RES0 in case they are given meanings in future.
355 			 * We always report the ESR as if the fault was taken
356 			 * to EL1 and so ISV and the bits in ISS[23:14] are
357 			 * clear. (In fact it always will be a fault to EL1.)
358 			 */
359 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL |
360 				ESR_ELx_CM | ESR_ELx_WNR;
361 			esr |= ESR_ELx_FSC_FAULT;
362 			break;
363 		case ESR_ELx_EC_IABT_LOW:
364 			/*
365 			 * Claim a level 0 translation fault.
366 			 * All other bits are architecturally RES0 for faults
367 			 * reported with that DFSC value, so we clear them.
368 			 */
369 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL;
370 			esr |= ESR_ELx_FSC_FAULT;
371 			break;
372 		default:
373 			/*
374 			 * This should never happen (entry.S only brings us
375 			 * into this code for insn and data aborts from a lower
376 			 * exception level). Fail safe by not providing an ESR
377 			 * context record at all.
378 			 */
379 			WARN(1, "ESR 0x%x is not DABT or IABT from EL0\n", esr);
380 			esr = 0;
381 			break;
382 		}
383 	}
384 
385 	current->thread.fault_code = esr;
386 }
387 
388 static void do_bad_area(unsigned long far, unsigned int esr,
389 			struct pt_regs *regs)
390 {
391 	unsigned long addr = untagged_addr(far);
392 
393 	/*
394 	 * If we are in kernel mode at this point, we have no context to
395 	 * handle this fault with.
396 	 */
397 	if (user_mode(regs)) {
398 		const struct fault_info *inf = esr_to_fault_info(esr);
399 
400 		set_thread_esr(addr, esr);
401 		arm64_force_sig_fault(inf->sig, inf->code, far, inf->name);
402 	} else {
403 		__do_kernel_fault(addr, esr, regs);
404 	}
405 }
406 
407 #define VM_FAULT_BADMAP		0x010000
408 #define VM_FAULT_BADACCESS	0x020000
409 
410 static vm_fault_t __do_page_fault(struct mm_struct *mm, unsigned long addr,
411 				  unsigned int mm_flags, unsigned long vm_flags,
412 				  struct pt_regs *regs)
413 {
414 	struct vm_area_struct *vma = find_vma(mm, addr);
415 
416 	if (unlikely(!vma))
417 		return VM_FAULT_BADMAP;
418 
419 	/*
420 	 * Ok, we have a good vm_area for this memory access, so we can handle
421 	 * it.
422 	 */
423 	if (unlikely(vma->vm_start > addr)) {
424 		if (!(vma->vm_flags & VM_GROWSDOWN))
425 			return VM_FAULT_BADMAP;
426 		if (expand_stack(vma, addr))
427 			return VM_FAULT_BADMAP;
428 	}
429 
430 	/*
431 	 * Check that the permissions on the VMA allow for the fault which
432 	 * occurred.
433 	 */
434 	if (!(vma->vm_flags & vm_flags))
435 		return VM_FAULT_BADACCESS;
436 	return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags, regs);
437 }
438 
439 static bool is_el0_instruction_abort(unsigned int esr)
440 {
441 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
442 }
443 
444 /*
445  * Note: not valid for EL1 DC IVAC, but we never use that such that it
446  * should fault. EL0 cannot issue DC IVAC (undef).
447  */
448 static bool is_write_abort(unsigned int esr)
449 {
450 	return (esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM);
451 }
452 
453 static int __kprobes do_page_fault(unsigned long far, unsigned int esr,
454 				   struct pt_regs *regs)
455 {
456 	const struct fault_info *inf;
457 	struct mm_struct *mm = current->mm;
458 	vm_fault_t fault;
459 	unsigned long vm_flags = VM_ACCESS_FLAGS;
460 	unsigned int mm_flags = FAULT_FLAG_DEFAULT;
461 	unsigned long addr = untagged_addr(far);
462 
463 	if (kprobe_page_fault(regs, esr))
464 		return 0;
465 
466 	/*
467 	 * If we're in an interrupt or have no user context, we must not take
468 	 * the fault.
469 	 */
470 	if (faulthandler_disabled() || !mm)
471 		goto no_context;
472 
473 	if (user_mode(regs))
474 		mm_flags |= FAULT_FLAG_USER;
475 
476 	if (is_el0_instruction_abort(esr)) {
477 		vm_flags = VM_EXEC;
478 		mm_flags |= FAULT_FLAG_INSTRUCTION;
479 	} else if (is_write_abort(esr)) {
480 		vm_flags = VM_WRITE;
481 		mm_flags |= FAULT_FLAG_WRITE;
482 	}
483 
484 	if (is_ttbr0_addr(addr) && is_el1_permission_fault(addr, esr, regs)) {
485 		if (is_el1_instruction_abort(esr))
486 			die_kernel_fault("execution of user memory",
487 					 addr, esr, regs);
488 
489 		if (!search_exception_tables(regs->pc))
490 			die_kernel_fault("access to user memory outside uaccess routines",
491 					 addr, esr, regs);
492 	}
493 
494 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
495 
496 	/*
497 	 * As per x86, we may deadlock here. However, since the kernel only
498 	 * validly references user space from well defined areas of the code,
499 	 * we can bug out early if this is from code which shouldn't.
500 	 */
501 	if (!mmap_read_trylock(mm)) {
502 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
503 			goto no_context;
504 retry:
505 		mmap_read_lock(mm);
506 	} else {
507 		/*
508 		 * The above down_read_trylock() might have succeeded in which
509 		 * case, we'll have missed the might_sleep() from down_read().
510 		 */
511 		might_sleep();
512 #ifdef CONFIG_DEBUG_VM
513 		if (!user_mode(regs) && !search_exception_tables(regs->pc)) {
514 			mmap_read_unlock(mm);
515 			goto no_context;
516 		}
517 #endif
518 	}
519 
520 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, regs);
521 
522 	/* Quick path to respond to signals */
523 	if (fault_signal_pending(fault, regs)) {
524 		if (!user_mode(regs))
525 			goto no_context;
526 		return 0;
527 	}
528 
529 	if (fault & VM_FAULT_RETRY) {
530 		if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
531 			mm_flags |= FAULT_FLAG_TRIED;
532 			goto retry;
533 		}
534 	}
535 	mmap_read_unlock(mm);
536 
537 	/*
538 	 * Handle the "normal" (no error) case first.
539 	 */
540 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
541 			      VM_FAULT_BADACCESS))))
542 		return 0;
543 
544 	/*
545 	 * If we are in kernel mode at this point, we have no context to
546 	 * handle this fault with.
547 	 */
548 	if (!user_mode(regs))
549 		goto no_context;
550 
551 	if (fault & VM_FAULT_OOM) {
552 		/*
553 		 * We ran out of memory, call the OOM killer, and return to
554 		 * userspace (which will retry the fault, or kill us if we got
555 		 * oom-killed).
556 		 */
557 		pagefault_out_of_memory();
558 		return 0;
559 	}
560 
561 	inf = esr_to_fault_info(esr);
562 	set_thread_esr(addr, esr);
563 	if (fault & VM_FAULT_SIGBUS) {
564 		/*
565 		 * We had some memory, but were unable to successfully fix up
566 		 * this page fault.
567 		 */
568 		arm64_force_sig_fault(SIGBUS, BUS_ADRERR, far, inf->name);
569 	} else if (fault & (VM_FAULT_HWPOISON_LARGE | VM_FAULT_HWPOISON)) {
570 		unsigned int lsb;
571 
572 		lsb = PAGE_SHIFT;
573 		if (fault & VM_FAULT_HWPOISON_LARGE)
574 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
575 
576 		arm64_force_sig_mceerr(BUS_MCEERR_AR, far, lsb, inf->name);
577 	} else {
578 		/*
579 		 * Something tried to access memory that isn't in our memory
580 		 * map.
581 		 */
582 		arm64_force_sig_fault(SIGSEGV,
583 				      fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR,
584 				      far, inf->name);
585 	}
586 
587 	return 0;
588 
589 no_context:
590 	__do_kernel_fault(addr, esr, regs);
591 	return 0;
592 }
593 
594 static int __kprobes do_translation_fault(unsigned long far,
595 					  unsigned int esr,
596 					  struct pt_regs *regs)
597 {
598 	unsigned long addr = untagged_addr(far);
599 
600 	if (is_ttbr0_addr(addr))
601 		return do_page_fault(far, esr, regs);
602 
603 	do_bad_area(far, esr, regs);
604 	return 0;
605 }
606 
607 static int do_alignment_fault(unsigned long far, unsigned int esr,
608 			      struct pt_regs *regs)
609 {
610 	do_bad_area(far, esr, regs);
611 	return 0;
612 }
613 
614 static int do_bad(unsigned long far, unsigned int esr, struct pt_regs *regs)
615 {
616 	return 1; /* "fault" */
617 }
618 
619 static int do_sea(unsigned long far, unsigned int esr, struct pt_regs *regs)
620 {
621 	const struct fault_info *inf;
622 	unsigned long siaddr;
623 
624 	inf = esr_to_fault_info(esr);
625 
626 	if (user_mode(regs) && apei_claim_sea(regs) == 0) {
627 		/*
628 		 * APEI claimed this as a firmware-first notification.
629 		 * Some processing deferred to task_work before ret_to_user().
630 		 */
631 		return 0;
632 	}
633 
634 	if (esr & ESR_ELx_FnV) {
635 		siaddr = 0;
636 	} else {
637 		/*
638 		 * The architecture specifies that the tag bits of FAR_EL1 are
639 		 * UNKNOWN for synchronous external aborts. Mask them out now
640 		 * so that userspace doesn't see them.
641 		 */
642 		siaddr  = untagged_addr(far);
643 	}
644 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
645 
646 	return 0;
647 }
648 
649 static int do_tag_check_fault(unsigned long far, unsigned int esr,
650 			      struct pt_regs *regs)
651 {
652 	/*
653 	 * The architecture specifies that bits 63:60 of FAR_EL1 are UNKNOWN for tag
654 	 * check faults. Mask them out now so that userspace doesn't see them.
655 	 */
656 	far &= (1UL << 60) - 1;
657 	do_bad_area(far, esr, regs);
658 	return 0;
659 }
660 
661 static const struct fault_info fault_info[] = {
662 	{ do_bad,		SIGKILL, SI_KERNEL,	"ttbr address size fault"	},
663 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 1 address size fault"	},
664 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 2 address size fault"	},
665 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 3 address size fault"	},
666 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
667 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
668 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
669 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
670 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 8"			},
671 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
672 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
673 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
674 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 12"			},
675 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
676 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
677 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
678 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous external abort"	},
679 	{ do_tag_check_fault,	SIGSEGV, SEGV_MTESERR,	"synchronous tag check fault"	},
680 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 18"			},
681 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 19"			},
682 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 (translation table walk)"	},
683 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 (translation table walk)"	},
684 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 (translation table walk)"	},
685 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 (translation table walk)"	},
686 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous parity or ECC error" },	// Reserved when RAS is implemented
687 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 25"			},
688 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 26"			},
689 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 27"			},
690 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
691 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
692 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
693 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
694 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 32"			},
695 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
696 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 34"			},
697 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 35"			},
698 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 36"			},
699 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 37"			},
700 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 38"			},
701 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 39"			},
702 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 40"			},
703 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 41"			},
704 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 42"			},
705 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 43"			},
706 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 44"			},
707 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 45"			},
708 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 46"			},
709 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 47"			},
710 	{ do_bad,		SIGKILL, SI_KERNEL,	"TLB conflict abort"		},
711 	{ do_bad,		SIGKILL, SI_KERNEL,	"Unsupported atomic hardware update fault"	},
712 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 50"			},
713 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 51"			},
714 	{ do_bad,		SIGKILL, SI_KERNEL,	"implementation fault (lockdown abort)" },
715 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"implementation fault (unsupported exclusive)" },
716 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 54"			},
717 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 55"			},
718 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 56"			},
719 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 57"			},
720 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 58" 			},
721 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 59"			},
722 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 60"			},
723 	{ do_bad,		SIGKILL, SI_KERNEL,	"section domain fault"		},
724 	{ do_bad,		SIGKILL, SI_KERNEL,	"page domain fault"		},
725 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 63"			},
726 };
727 
728 void do_mem_abort(unsigned long far, unsigned int esr, struct pt_regs *regs)
729 {
730 	const struct fault_info *inf = esr_to_fault_info(esr);
731 	unsigned long addr = untagged_addr(far);
732 
733 	if (!inf->fn(far, esr, regs))
734 		return;
735 
736 	if (!user_mode(regs)) {
737 		pr_alert("Unhandled fault at 0x%016lx\n", addr);
738 		mem_abort_decode(esr);
739 		show_pte(addr);
740 	}
741 
742 	/*
743 	 * At this point we have an unrecognized fault type whose tag bits may
744 	 * have been defined as UNKNOWN. Therefore we only expose the untagged
745 	 * address to the signal handler.
746 	 */
747 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, addr, esr);
748 }
749 NOKPROBE_SYMBOL(do_mem_abort);
750 
751 void do_el0_irq_bp_hardening(void)
752 {
753 	/* PC has already been checked in entry.S */
754 	arm64_apply_bp_hardening();
755 }
756 NOKPROBE_SYMBOL(do_el0_irq_bp_hardening);
757 
758 void do_sp_pc_abort(unsigned long addr, unsigned int esr, struct pt_regs *regs)
759 {
760 	arm64_notify_die("SP/PC alignment exception", regs, SIGBUS, BUS_ADRALN,
761 			 addr, esr);
762 }
763 NOKPROBE_SYMBOL(do_sp_pc_abort);
764 
765 int __init early_brk64(unsigned long addr, unsigned int esr,
766 		       struct pt_regs *regs);
767 
768 /*
769  * __refdata because early_brk64 is __init, but the reference to it is
770  * clobbered at arch_initcall time.
771  * See traps.c and debug-monitors.c:debug_traps_init().
772  */
773 static struct fault_info __refdata debug_fault_info[] = {
774 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
775 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
776 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
777 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 3"		},
778 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
779 	{ do_bad,	SIGKILL,	SI_KERNEL,	"aarch32 vector catch"	},
780 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
781 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 7"		},
782 };
783 
784 void __init hook_debug_fault_code(int nr,
785 				  int (*fn)(unsigned long, unsigned int, struct pt_regs *),
786 				  int sig, int code, const char *name)
787 {
788 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
789 
790 	debug_fault_info[nr].fn		= fn;
791 	debug_fault_info[nr].sig	= sig;
792 	debug_fault_info[nr].code	= code;
793 	debug_fault_info[nr].name	= name;
794 }
795 
796 /*
797  * In debug exception context, we explicitly disable preemption despite
798  * having interrupts disabled.
799  * This serves two purposes: it makes it much less likely that we would
800  * accidentally schedule in exception context and it will force a warning
801  * if we somehow manage to schedule by accident.
802  */
803 static void debug_exception_enter(struct pt_regs *regs)
804 {
805 	preempt_disable();
806 
807 	/* This code is a bit fragile.  Test it. */
808 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "exception_enter didn't work");
809 }
810 NOKPROBE_SYMBOL(debug_exception_enter);
811 
812 static void debug_exception_exit(struct pt_regs *regs)
813 {
814 	preempt_enable_no_resched();
815 }
816 NOKPROBE_SYMBOL(debug_exception_exit);
817 
818 #ifdef CONFIG_ARM64_ERRATUM_1463225
819 DECLARE_PER_CPU(int, __in_cortex_a76_erratum_1463225_wa);
820 
821 static int cortex_a76_erratum_1463225_debug_handler(struct pt_regs *regs)
822 {
823 	if (user_mode(regs))
824 		return 0;
825 
826 	if (!__this_cpu_read(__in_cortex_a76_erratum_1463225_wa))
827 		return 0;
828 
829 	/*
830 	 * We've taken a dummy step exception from the kernel to ensure
831 	 * that interrupts are re-enabled on the syscall path. Return back
832 	 * to cortex_a76_erratum_1463225_svc_handler() with debug exceptions
833 	 * masked so that we can safely restore the mdscr and get on with
834 	 * handling the syscall.
835 	 */
836 	regs->pstate |= PSR_D_BIT;
837 	return 1;
838 }
839 #else
840 static int cortex_a76_erratum_1463225_debug_handler(struct pt_regs *regs)
841 {
842 	return 0;
843 }
844 #endif /* CONFIG_ARM64_ERRATUM_1463225 */
845 NOKPROBE_SYMBOL(cortex_a76_erratum_1463225_debug_handler);
846 
847 void do_debug_exception(unsigned long addr_if_watchpoint, unsigned int esr,
848 			struct pt_regs *regs)
849 {
850 	const struct fault_info *inf = esr_to_debug_fault_info(esr);
851 	unsigned long pc = instruction_pointer(regs);
852 
853 	if (cortex_a76_erratum_1463225_debug_handler(regs))
854 		return;
855 
856 	debug_exception_enter(regs);
857 
858 	if (user_mode(regs) && !is_ttbr0_addr(pc))
859 		arm64_apply_bp_hardening();
860 
861 	if (inf->fn(addr_if_watchpoint, esr, regs)) {
862 		arm64_notify_die(inf->name, regs, inf->sig, inf->code, pc, esr);
863 	}
864 
865 	debug_exception_exit(regs);
866 }
867 NOKPROBE_SYMBOL(do_debug_exception);
868