xref: /openbmc/linux/arch/arm64/mm/fault.c (revision f125e2d4)
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/pgtable.h>
40 #include <asm/tlbflush.h>
41 #include <asm/traps.h>
42 
43 struct fault_info {
44 	int	(*fn)(unsigned long addr, unsigned int esr,
45 		      struct pt_regs *regs);
46 	int	sig;
47 	int	code;
48 	const char *name;
49 };
50 
51 static const struct fault_info fault_info[];
52 static struct fault_info debug_fault_info[];
53 
54 static inline const struct fault_info *esr_to_fault_info(unsigned int esr)
55 {
56 	return fault_info + (esr & ESR_ELx_FSC);
57 }
58 
59 static inline const struct fault_info *esr_to_debug_fault_info(unsigned int esr)
60 {
61 	return debug_fault_info + DBG_ESR_EVT(esr);
62 }
63 
64 static void data_abort_decode(unsigned int esr)
65 {
66 	pr_alert("Data abort info:\n");
67 
68 	if (esr & ESR_ELx_ISV) {
69 		pr_alert("  Access size = %u byte(s)\n",
70 			 1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
71 		pr_alert("  SSE = %lu, SRT = %lu\n",
72 			 (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
73 			 (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
74 		pr_alert("  SF = %lu, AR = %lu\n",
75 			 (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
76 			 (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
77 	} else {
78 		pr_alert("  ISV = 0, ISS = 0x%08lx\n", esr & ESR_ELx_ISS_MASK);
79 	}
80 
81 	pr_alert("  CM = %lu, WnR = %lu\n",
82 		 (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
83 		 (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT);
84 }
85 
86 static void mem_abort_decode(unsigned int esr)
87 {
88 	pr_alert("Mem abort info:\n");
89 
90 	pr_alert("  ESR = 0x%08x\n", esr);
91 	pr_alert("  EC = 0x%02lx: %s, IL = %u bits\n",
92 		 ESR_ELx_EC(esr), esr_get_class_string(esr),
93 		 (esr & ESR_ELx_IL) ? 32 : 16);
94 	pr_alert("  SET = %lu, FnV = %lu\n",
95 		 (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
96 		 (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
97 	pr_alert("  EA = %lu, S1PTW = %lu\n",
98 		 (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
99 		 (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
100 
101 	if (esr_is_data_abort(esr))
102 		data_abort_decode(esr);
103 }
104 
105 static inline unsigned long mm_to_pgd_phys(struct mm_struct *mm)
106 {
107 	/* Either init_pg_dir or swapper_pg_dir */
108 	if (mm == &init_mm)
109 		return __pa_symbol(mm->pgd);
110 
111 	return (unsigned long)virt_to_phys(mm->pgd);
112 }
113 
114 /*
115  * Dump out the page tables associated with 'addr' in the currently active mm.
116  */
117 static void show_pte(unsigned long addr)
118 {
119 	struct mm_struct *mm;
120 	pgd_t *pgdp;
121 	pgd_t pgd;
122 
123 	if (is_ttbr0_addr(addr)) {
124 		/* TTBR0 */
125 		mm = current->active_mm;
126 		if (mm == &init_mm) {
127 			pr_alert("[%016lx] user address but active_mm is swapper\n",
128 				 addr);
129 			return;
130 		}
131 	} else if (is_ttbr1_addr(addr)) {
132 		/* TTBR1 */
133 		mm = &init_mm;
134 	} else {
135 		pr_alert("[%016lx] address between user and kernel address ranges\n",
136 			 addr);
137 		return;
138 	}
139 
140 	pr_alert("%s pgtable: %luk pages, %llu-bit VAs, pgdp=%016lx\n",
141 		 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
142 		 vabits_actual, mm_to_pgd_phys(mm));
143 	pgdp = pgd_offset(mm, addr);
144 	pgd = READ_ONCE(*pgdp);
145 	pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
146 
147 	do {
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 		pudp = pud_offset(pgdp, addr);
156 		pud = READ_ONCE(*pudp);
157 		pr_cont(", pud=%016llx", pud_val(pud));
158 		if (pud_none(pud) || pud_bad(pud))
159 			break;
160 
161 		pmdp = pmd_offset(pudp, addr);
162 		pmd = READ_ONCE(*pmdp);
163 		pr_cont(", pmd=%016llx", pmd_val(pmd));
164 		if (pmd_none(pmd) || pmd_bad(pmd))
165 			break;
166 
167 		ptep = pte_offset_map(pmdp, addr);
168 		pte = READ_ONCE(*ptep);
169 		pr_cont(", pte=%016llx", pte_val(pte));
170 		pte_unmap(ptep);
171 	} while(0);
172 
173 	pr_cont("\n");
174 }
175 
176 /*
177  * This function sets the access flags (dirty, accessed), as well as write
178  * permission, and only to a more permissive setting.
179  *
180  * It needs to cope with hardware update of the accessed/dirty state by other
181  * agents in the system and can safely skip the __sync_icache_dcache() call as,
182  * like set_pte_at(), the PTE is never changed from no-exec to exec here.
183  *
184  * Returns whether or not the PTE actually changed.
185  */
186 int ptep_set_access_flags(struct vm_area_struct *vma,
187 			  unsigned long address, pte_t *ptep,
188 			  pte_t entry, int dirty)
189 {
190 	pteval_t old_pteval, pteval;
191 	pte_t pte = READ_ONCE(*ptep);
192 
193 	if (pte_same(pte, entry))
194 		return 0;
195 
196 	/* only preserve the access flags and write permission */
197 	pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
198 
199 	/*
200 	 * Setting the flags must be done atomically to avoid racing with the
201 	 * hardware update of the access/dirty state. The PTE_RDONLY bit must
202 	 * be set to the most permissive (lowest value) of *ptep and entry
203 	 * (calculated as: a & b == ~(~a | ~b)).
204 	 */
205 	pte_val(entry) ^= PTE_RDONLY;
206 	pteval = pte_val(pte);
207 	do {
208 		old_pteval = pteval;
209 		pteval ^= PTE_RDONLY;
210 		pteval |= pte_val(entry);
211 		pteval ^= PTE_RDONLY;
212 		pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
213 	} while (pteval != old_pteval);
214 
215 	flush_tlb_fix_spurious_fault(vma, address);
216 	return 1;
217 }
218 
219 static bool is_el1_instruction_abort(unsigned int esr)
220 {
221 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
222 }
223 
224 static inline bool is_el1_permission_fault(unsigned long addr, unsigned int esr,
225 					   struct pt_regs *regs)
226 {
227 	unsigned int ec       = ESR_ELx_EC(esr);
228 	unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
229 
230 	if (ec != ESR_ELx_EC_DABT_CUR && ec != ESR_ELx_EC_IABT_CUR)
231 		return false;
232 
233 	if (fsc_type == ESR_ELx_FSC_PERM)
234 		return true;
235 
236 	if (is_ttbr0_addr(addr) && system_uses_ttbr0_pan())
237 		return fsc_type == ESR_ELx_FSC_FAULT &&
238 			(regs->pstate & PSR_PAN_BIT);
239 
240 	return false;
241 }
242 
243 static bool __kprobes is_spurious_el1_translation_fault(unsigned long addr,
244 							unsigned int esr,
245 							struct pt_regs *regs)
246 {
247 	unsigned long flags;
248 	u64 par, dfsc;
249 
250 	if (ESR_ELx_EC(esr) != ESR_ELx_EC_DABT_CUR ||
251 	    (esr & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT)
252 		return false;
253 
254 	local_irq_save(flags);
255 	asm volatile("at s1e1r, %0" :: "r" (addr));
256 	isb();
257 	par = read_sysreg(par_el1);
258 	local_irq_restore(flags);
259 
260 	/*
261 	 * If we now have a valid translation, treat the translation fault as
262 	 * spurious.
263 	 */
264 	if (!(par & SYS_PAR_EL1_F))
265 		return true;
266 
267 	/*
268 	 * If we got a different type of fault from the AT instruction,
269 	 * treat the translation fault as spurious.
270 	 */
271 	dfsc = FIELD_GET(SYS_PAR_EL1_FST, par);
272 	return (dfsc & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT;
273 }
274 
275 static void die_kernel_fault(const char *msg, unsigned long addr,
276 			     unsigned int esr, struct pt_regs *regs)
277 {
278 	bust_spinlocks(1);
279 
280 	pr_alert("Unable to handle kernel %s at virtual address %016lx\n", msg,
281 		 addr);
282 
283 	mem_abort_decode(esr);
284 
285 	show_pte(addr);
286 	die("Oops", regs, esr);
287 	bust_spinlocks(0);
288 	do_exit(SIGKILL);
289 }
290 
291 static void __do_kernel_fault(unsigned long addr, unsigned int esr,
292 			      struct pt_regs *regs)
293 {
294 	const char *msg;
295 
296 	/*
297 	 * Are we prepared to handle this kernel fault?
298 	 * We are almost certainly not prepared to handle instruction faults.
299 	 */
300 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
301 		return;
302 
303 	if (WARN_RATELIMIT(is_spurious_el1_translation_fault(addr, esr, regs),
304 	    "Ignoring spurious kernel translation fault at virtual address %016lx\n", addr))
305 		return;
306 
307 	if (is_el1_permission_fault(addr, esr, regs)) {
308 		if (esr & ESR_ELx_WNR)
309 			msg = "write to read-only memory";
310 		else if (is_el1_instruction_abort(esr))
311 			msg = "execute from non-executable memory";
312 		else
313 			msg = "read from unreadable memory";
314 	} else if (addr < PAGE_SIZE) {
315 		msg = "NULL pointer dereference";
316 	} else {
317 		msg = "paging request";
318 	}
319 
320 	die_kernel_fault(msg, addr, esr, regs);
321 }
322 
323 static void set_thread_esr(unsigned long address, unsigned int esr)
324 {
325 	current->thread.fault_address = address;
326 
327 	/*
328 	 * If the faulting address is in the kernel, we must sanitize the ESR.
329 	 * From userspace's point of view, kernel-only mappings don't exist
330 	 * at all, so we report them as level 0 translation faults.
331 	 * (This is not quite the way that "no mapping there at all" behaves:
332 	 * an alignment fault not caused by the memory type would take
333 	 * precedence over translation fault for a real access to empty
334 	 * space. Unfortunately we can't easily distinguish "alignment fault
335 	 * not caused by memory type" from "alignment fault caused by memory
336 	 * type", so we ignore this wrinkle and just return the translation
337 	 * fault.)
338 	 */
339 	if (!is_ttbr0_addr(current->thread.fault_address)) {
340 		switch (ESR_ELx_EC(esr)) {
341 		case ESR_ELx_EC_DABT_LOW:
342 			/*
343 			 * These bits provide only information about the
344 			 * faulting instruction, which userspace knows already.
345 			 * We explicitly clear bits which are architecturally
346 			 * RES0 in case they are given meanings in future.
347 			 * We always report the ESR as if the fault was taken
348 			 * to EL1 and so ISV and the bits in ISS[23:14] are
349 			 * clear. (In fact it always will be a fault to EL1.)
350 			 */
351 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL |
352 				ESR_ELx_CM | ESR_ELx_WNR;
353 			esr |= ESR_ELx_FSC_FAULT;
354 			break;
355 		case ESR_ELx_EC_IABT_LOW:
356 			/*
357 			 * Claim a level 0 translation fault.
358 			 * All other bits are architecturally RES0 for faults
359 			 * reported with that DFSC value, so we clear them.
360 			 */
361 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL;
362 			esr |= ESR_ELx_FSC_FAULT;
363 			break;
364 		default:
365 			/*
366 			 * This should never happen (entry.S only brings us
367 			 * into this code for insn and data aborts from a lower
368 			 * exception level). Fail safe by not providing an ESR
369 			 * context record at all.
370 			 */
371 			WARN(1, "ESR 0x%x is not DABT or IABT from EL0\n", esr);
372 			esr = 0;
373 			break;
374 		}
375 	}
376 
377 	current->thread.fault_code = esr;
378 }
379 
380 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
381 {
382 	/*
383 	 * If we are in kernel mode at this point, we have no context to
384 	 * handle this fault with.
385 	 */
386 	if (user_mode(regs)) {
387 		const struct fault_info *inf = esr_to_fault_info(esr);
388 
389 		set_thread_esr(addr, esr);
390 		arm64_force_sig_fault(inf->sig, inf->code, (void __user *)addr,
391 				      inf->name);
392 	} else {
393 		__do_kernel_fault(addr, esr, regs);
394 	}
395 }
396 
397 #define VM_FAULT_BADMAP		0x010000
398 #define VM_FAULT_BADACCESS	0x020000
399 
400 static vm_fault_t __do_page_fault(struct mm_struct *mm, unsigned long addr,
401 			   unsigned int mm_flags, unsigned long vm_flags)
402 {
403 	struct vm_area_struct *vma = find_vma(mm, addr);
404 
405 	if (unlikely(!vma))
406 		return VM_FAULT_BADMAP;
407 
408 	/*
409 	 * Ok, we have a good vm_area for this memory access, so we can handle
410 	 * it.
411 	 */
412 	if (unlikely(vma->vm_start > addr)) {
413 		if (!(vma->vm_flags & VM_GROWSDOWN))
414 			return VM_FAULT_BADMAP;
415 		if (expand_stack(vma, addr))
416 			return VM_FAULT_BADMAP;
417 	}
418 
419 	/*
420 	 * Check that the permissions on the VMA allow for the fault which
421 	 * occurred.
422 	 */
423 	if (!(vma->vm_flags & vm_flags))
424 		return VM_FAULT_BADACCESS;
425 	return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags);
426 }
427 
428 static bool is_el0_instruction_abort(unsigned int esr)
429 {
430 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
431 }
432 
433 /*
434  * Note: not valid for EL1 DC IVAC, but we never use that such that it
435  * should fault. EL0 cannot issue DC IVAC (undef).
436  */
437 static bool is_write_abort(unsigned int esr)
438 {
439 	return (esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM);
440 }
441 
442 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
443 				   struct pt_regs *regs)
444 {
445 	const struct fault_info *inf;
446 	struct mm_struct *mm = current->mm;
447 	vm_fault_t fault, major = 0;
448 	unsigned long vm_flags = VM_READ | VM_WRITE | VM_EXEC;
449 	unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
450 
451 	if (kprobe_page_fault(regs, esr))
452 		return 0;
453 
454 	/*
455 	 * If we're in an interrupt or have no user context, we must not take
456 	 * the fault.
457 	 */
458 	if (faulthandler_disabled() || !mm)
459 		goto no_context;
460 
461 	if (user_mode(regs))
462 		mm_flags |= FAULT_FLAG_USER;
463 
464 	if (is_el0_instruction_abort(esr)) {
465 		vm_flags = VM_EXEC;
466 		mm_flags |= FAULT_FLAG_INSTRUCTION;
467 	} else if (is_write_abort(esr)) {
468 		vm_flags = VM_WRITE;
469 		mm_flags |= FAULT_FLAG_WRITE;
470 	}
471 
472 	if (is_ttbr0_addr(addr) && is_el1_permission_fault(addr, esr, regs)) {
473 		/* regs->orig_addr_limit may be 0 if we entered from EL0 */
474 		if (regs->orig_addr_limit == KERNEL_DS)
475 			die_kernel_fault("access to user memory with fs=KERNEL_DS",
476 					 addr, esr, regs);
477 
478 		if (is_el1_instruction_abort(esr))
479 			die_kernel_fault("execution of user memory",
480 					 addr, esr, regs);
481 
482 		if (!search_exception_tables(regs->pc))
483 			die_kernel_fault("access to user memory outside uaccess routines",
484 					 addr, esr, regs);
485 	}
486 
487 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
488 
489 	/*
490 	 * As per x86, we may deadlock here. However, since the kernel only
491 	 * validly references user space from well defined areas of the code,
492 	 * we can bug out early if this is from code which shouldn't.
493 	 */
494 	if (!down_read_trylock(&mm->mmap_sem)) {
495 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
496 			goto no_context;
497 retry:
498 		down_read(&mm->mmap_sem);
499 	} else {
500 		/*
501 		 * The above down_read_trylock() might have succeeded in which
502 		 * case, we'll have missed the might_sleep() from down_read().
503 		 */
504 		might_sleep();
505 #ifdef CONFIG_DEBUG_VM
506 		if (!user_mode(regs) && !search_exception_tables(regs->pc)) {
507 			up_read(&mm->mmap_sem);
508 			goto no_context;
509 		}
510 #endif
511 	}
512 
513 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags);
514 	major |= fault & VM_FAULT_MAJOR;
515 
516 	if (fault & VM_FAULT_RETRY) {
517 		/*
518 		 * If we need to retry but a fatal signal is pending,
519 		 * handle the signal first. We do not need to release
520 		 * the mmap_sem because it would already be released
521 		 * in __lock_page_or_retry in mm/filemap.c.
522 		 */
523 		if (fatal_signal_pending(current)) {
524 			if (!user_mode(regs))
525 				goto no_context;
526 			return 0;
527 		}
528 
529 		/*
530 		 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
531 		 * starvation.
532 		 */
533 		if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
534 			mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
535 			mm_flags |= FAULT_FLAG_TRIED;
536 			goto retry;
537 		}
538 	}
539 	up_read(&mm->mmap_sem);
540 
541 	/*
542 	 * Handle the "normal" (no error) case first.
543 	 */
544 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
545 			      VM_FAULT_BADACCESS)))) {
546 		/*
547 		 * Major/minor page fault accounting is only done
548 		 * once. If we go through a retry, it is extremely
549 		 * likely that the page will be found in page cache at
550 		 * that point.
551 		 */
552 		if (major) {
553 			current->maj_flt++;
554 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
555 				      addr);
556 		} else {
557 			current->min_flt++;
558 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
559 				      addr);
560 		}
561 
562 		return 0;
563 	}
564 
565 	/*
566 	 * If we are in kernel mode at this point, we have no context to
567 	 * handle this fault with.
568 	 */
569 	if (!user_mode(regs))
570 		goto no_context;
571 
572 	if (fault & VM_FAULT_OOM) {
573 		/*
574 		 * We ran out of memory, call the OOM killer, and return to
575 		 * userspace (which will retry the fault, or kill us if we got
576 		 * oom-killed).
577 		 */
578 		pagefault_out_of_memory();
579 		return 0;
580 	}
581 
582 	inf = esr_to_fault_info(esr);
583 	set_thread_esr(addr, esr);
584 	if (fault & VM_FAULT_SIGBUS) {
585 		/*
586 		 * We had some memory, but were unable to successfully fix up
587 		 * this page fault.
588 		 */
589 		arm64_force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)addr,
590 				      inf->name);
591 	} else if (fault & (VM_FAULT_HWPOISON_LARGE | VM_FAULT_HWPOISON)) {
592 		unsigned int lsb;
593 
594 		lsb = PAGE_SHIFT;
595 		if (fault & VM_FAULT_HWPOISON_LARGE)
596 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
597 
598 		arm64_force_sig_mceerr(BUS_MCEERR_AR, (void __user *)addr, lsb,
599 				       inf->name);
600 	} else {
601 		/*
602 		 * Something tried to access memory that isn't in our memory
603 		 * map.
604 		 */
605 		arm64_force_sig_fault(SIGSEGV,
606 				      fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR,
607 				      (void __user *)addr,
608 				      inf->name);
609 	}
610 
611 	return 0;
612 
613 no_context:
614 	__do_kernel_fault(addr, esr, regs);
615 	return 0;
616 }
617 
618 static int __kprobes do_translation_fault(unsigned long addr,
619 					  unsigned int esr,
620 					  struct pt_regs *regs)
621 {
622 	if (is_ttbr0_addr(addr))
623 		return do_page_fault(addr, esr, regs);
624 
625 	do_bad_area(addr, esr, regs);
626 	return 0;
627 }
628 
629 static int do_alignment_fault(unsigned long addr, unsigned int esr,
630 			      struct pt_regs *regs)
631 {
632 	do_bad_area(addr, esr, regs);
633 	return 0;
634 }
635 
636 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
637 {
638 	return 1; /* "fault" */
639 }
640 
641 static int do_sea(unsigned long addr, unsigned int esr, struct pt_regs *regs)
642 {
643 	const struct fault_info *inf;
644 	void __user *siaddr;
645 
646 	inf = esr_to_fault_info(esr);
647 
648 	/*
649 	 * Return value ignored as we rely on signal merging.
650 	 * Future patches will make this more robust.
651 	 */
652 	apei_claim_sea(regs);
653 
654 	if (esr & ESR_ELx_FnV)
655 		siaddr = NULL;
656 	else
657 		siaddr  = (void __user *)addr;
658 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
659 
660 	return 0;
661 }
662 
663 static const struct fault_info fault_info[] = {
664 	{ do_bad,		SIGKILL, SI_KERNEL,	"ttbr address size fault"	},
665 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 1 address size fault"	},
666 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 2 address size fault"	},
667 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 3 address size fault"	},
668 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
669 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
670 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
671 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
672 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 8"			},
673 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
674 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
675 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
676 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 12"			},
677 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
678 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
679 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
680 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous external abort"	},
681 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 17"			},
682 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 18"			},
683 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 19"			},
684 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 (translation table walk)"	},
685 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 (translation table walk)"	},
686 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 (translation table walk)"	},
687 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 (translation table walk)"	},
688 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous parity or ECC error" },	// Reserved when RAS is implemented
689 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 25"			},
690 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 26"			},
691 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 27"			},
692 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
693 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
694 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
695 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
696 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 32"			},
697 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
698 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 34"			},
699 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 35"			},
700 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 36"			},
701 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 37"			},
702 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 38"			},
703 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 39"			},
704 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 40"			},
705 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 41"			},
706 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 42"			},
707 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 43"			},
708 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 44"			},
709 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 45"			},
710 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 46"			},
711 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 47"			},
712 	{ do_bad,		SIGKILL, SI_KERNEL,	"TLB conflict abort"		},
713 	{ do_bad,		SIGKILL, SI_KERNEL,	"Unsupported atomic hardware update fault"	},
714 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 50"			},
715 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 51"			},
716 	{ do_bad,		SIGKILL, SI_KERNEL,	"implementation fault (lockdown abort)" },
717 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"implementation fault (unsupported exclusive)" },
718 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 54"			},
719 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 55"			},
720 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 56"			},
721 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 57"			},
722 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 58" 			},
723 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 59"			},
724 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 60"			},
725 	{ do_bad,		SIGKILL, SI_KERNEL,	"section domain fault"		},
726 	{ do_bad,		SIGKILL, SI_KERNEL,	"page domain fault"		},
727 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 63"			},
728 };
729 
730 void do_mem_abort(unsigned long addr, unsigned int esr, struct pt_regs *regs)
731 {
732 	const struct fault_info *inf = esr_to_fault_info(esr);
733 
734 	if (!inf->fn(addr, esr, regs))
735 		return;
736 
737 	if (!user_mode(regs)) {
738 		pr_alert("Unhandled fault at 0x%016lx\n", addr);
739 		mem_abort_decode(esr);
740 		show_pte(addr);
741 	}
742 
743 	arm64_notify_die(inf->name, regs,
744 			 inf->sig, inf->code, (void __user *)addr, esr);
745 }
746 NOKPROBE_SYMBOL(do_mem_abort);
747 
748 void do_el0_irq_bp_hardening(void)
749 {
750 	/* PC has already been checked in entry.S */
751 	arm64_apply_bp_hardening();
752 }
753 NOKPROBE_SYMBOL(do_el0_irq_bp_hardening);
754 
755 void do_sp_pc_abort(unsigned long addr, unsigned int esr, struct pt_regs *regs)
756 {
757 	arm64_notify_die("SP/PC alignment exception", regs,
758 			 SIGBUS, BUS_ADRALN, (void __user *)addr, esr);
759 }
760 NOKPROBE_SYMBOL(do_sp_pc_abort);
761 
762 int __init early_brk64(unsigned long addr, unsigned int esr,
763 		       struct pt_regs *regs);
764 
765 /*
766  * __refdata because early_brk64 is __init, but the reference to it is
767  * clobbered at arch_initcall time.
768  * See traps.c and debug-monitors.c:debug_traps_init().
769  */
770 static struct fault_info __refdata debug_fault_info[] = {
771 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
772 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
773 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
774 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 3"		},
775 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
776 	{ do_bad,	SIGKILL,	SI_KERNEL,	"aarch32 vector catch"	},
777 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
778 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 7"		},
779 };
780 
781 void __init hook_debug_fault_code(int nr,
782 				  int (*fn)(unsigned long, unsigned int, struct pt_regs *),
783 				  int sig, int code, const char *name)
784 {
785 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
786 
787 	debug_fault_info[nr].fn		= fn;
788 	debug_fault_info[nr].sig	= sig;
789 	debug_fault_info[nr].code	= code;
790 	debug_fault_info[nr].name	= name;
791 }
792 
793 /*
794  * In debug exception context, we explicitly disable preemption despite
795  * having interrupts disabled.
796  * This serves two purposes: it makes it much less likely that we would
797  * accidentally schedule in exception context and it will force a warning
798  * if we somehow manage to schedule by accident.
799  */
800 static void debug_exception_enter(struct pt_regs *regs)
801 {
802 	/*
803 	 * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
804 	 * already disabled to preserve the last enabled/disabled addresses.
805 	 */
806 	if (interrupts_enabled(regs))
807 		trace_hardirqs_off();
808 
809 	if (user_mode(regs)) {
810 		RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
811 	} else {
812 		/*
813 		 * We might have interrupted pretty much anything.  In
814 		 * fact, if we're a debug exception, we can even interrupt
815 		 * NMI processing. We don't want this code makes in_nmi()
816 		 * to return true, but we need to notify RCU.
817 		 */
818 		rcu_nmi_enter();
819 	}
820 
821 	preempt_disable();
822 
823 	/* This code is a bit fragile.  Test it. */
824 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "exception_enter didn't work");
825 }
826 NOKPROBE_SYMBOL(debug_exception_enter);
827 
828 static void debug_exception_exit(struct pt_regs *regs)
829 {
830 	preempt_enable_no_resched();
831 
832 	if (!user_mode(regs))
833 		rcu_nmi_exit();
834 
835 	if (interrupts_enabled(regs))
836 		trace_hardirqs_on();
837 }
838 NOKPROBE_SYMBOL(debug_exception_exit);
839 
840 #ifdef CONFIG_ARM64_ERRATUM_1463225
841 DECLARE_PER_CPU(int, __in_cortex_a76_erratum_1463225_wa);
842 
843 static int cortex_a76_erratum_1463225_debug_handler(struct pt_regs *regs)
844 {
845 	if (user_mode(regs))
846 		return 0;
847 
848 	if (!__this_cpu_read(__in_cortex_a76_erratum_1463225_wa))
849 		return 0;
850 
851 	/*
852 	 * We've taken a dummy step exception from the kernel to ensure
853 	 * that interrupts are re-enabled on the syscall path. Return back
854 	 * to cortex_a76_erratum_1463225_svc_handler() with debug exceptions
855 	 * masked so that we can safely restore the mdscr and get on with
856 	 * handling the syscall.
857 	 */
858 	regs->pstate |= PSR_D_BIT;
859 	return 1;
860 }
861 #else
862 static int cortex_a76_erratum_1463225_debug_handler(struct pt_regs *regs)
863 {
864 	return 0;
865 }
866 #endif /* CONFIG_ARM64_ERRATUM_1463225 */
867 NOKPROBE_SYMBOL(cortex_a76_erratum_1463225_debug_handler);
868 
869 void do_debug_exception(unsigned long addr_if_watchpoint, unsigned int esr,
870 			struct pt_regs *regs)
871 {
872 	const struct fault_info *inf = esr_to_debug_fault_info(esr);
873 	unsigned long pc = instruction_pointer(regs);
874 
875 	if (cortex_a76_erratum_1463225_debug_handler(regs))
876 		return;
877 
878 	debug_exception_enter(regs);
879 
880 	if (user_mode(regs) && !is_ttbr0_addr(pc))
881 		arm64_apply_bp_hardening();
882 
883 	if (inf->fn(addr_if_watchpoint, esr, regs)) {
884 		arm64_notify_die(inf->name, regs,
885 				 inf->sig, inf->code, (void __user *)pc, esr);
886 	}
887 
888 	debug_exception_exit(regs);
889 }
890 NOKPROBE_SYMBOL(do_debug_exception);
891