xref: /openbmc/linux/arch/arm64/mm/fault.c (revision bbaf1ff0)
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/efi.h>
34 #include <asm/exception.h>
35 #include <asm/daifflags.h>
36 #include <asm/debug-monitors.h>
37 #include <asm/esr.h>
38 #include <asm/kprobes.h>
39 #include <asm/mte.h>
40 #include <asm/processor.h>
41 #include <asm/sysreg.h>
42 #include <asm/system_misc.h>
43 #include <asm/tlbflush.h>
44 #include <asm/traps.h>
45 
46 struct fault_info {
47 	int	(*fn)(unsigned long far, unsigned long esr,
48 		      struct pt_regs *regs);
49 	int	sig;
50 	int	code;
51 	const char *name;
52 };
53 
54 static const struct fault_info fault_info[];
55 static struct fault_info debug_fault_info[];
56 
57 static inline const struct fault_info *esr_to_fault_info(unsigned long esr)
58 {
59 	return fault_info + (esr & ESR_ELx_FSC);
60 }
61 
62 static inline const struct fault_info *esr_to_debug_fault_info(unsigned long esr)
63 {
64 	return debug_fault_info + DBG_ESR_EVT(esr);
65 }
66 
67 static void data_abort_decode(unsigned long esr)
68 {
69 	unsigned long iss2 = ESR_ELx_ISS2(esr);
70 
71 	pr_alert("Data abort info:\n");
72 
73 	if (esr & ESR_ELx_ISV) {
74 		pr_alert("  Access size = %u byte(s)\n",
75 			 1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
76 		pr_alert("  SSE = %lu, SRT = %lu\n",
77 			 (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
78 			 (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
79 		pr_alert("  SF = %lu, AR = %lu\n",
80 			 (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
81 			 (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
82 	} else {
83 		pr_alert("  ISV = 0, ISS = 0x%08lx, ISS2 = 0x%08lx\n",
84 			 esr & ESR_ELx_ISS_MASK, iss2);
85 	}
86 
87 	pr_alert("  CM = %lu, WnR = %lu, TnD = %lu, TagAccess = %lu\n",
88 		 (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
89 		 (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT,
90 		 (iss2 & ESR_ELx_TnD) >> ESR_ELx_TnD_SHIFT,
91 		 (iss2 & ESR_ELx_TagAccess) >> ESR_ELx_TagAccess_SHIFT);
92 
93 	pr_alert("  GCS = %ld, Overlay = %lu, DirtyBit = %lu, Xs = %llu\n",
94 		 (iss2 & ESR_ELx_GCS) >> ESR_ELx_GCS_SHIFT,
95 		 (iss2 & ESR_ELx_Overlay) >> ESR_ELx_Overlay_SHIFT,
96 		 (iss2 & ESR_ELx_DirtyBit) >> ESR_ELx_DirtyBit_SHIFT,
97 		 (iss2 & ESR_ELx_Xs_MASK) >> ESR_ELx_Xs_SHIFT);
98 }
99 
100 static void mem_abort_decode(unsigned long esr)
101 {
102 	pr_alert("Mem abort info:\n");
103 
104 	pr_alert("  ESR = 0x%016lx\n", esr);
105 	pr_alert("  EC = 0x%02lx: %s, IL = %u bits\n",
106 		 ESR_ELx_EC(esr), esr_get_class_string(esr),
107 		 (esr & ESR_ELx_IL) ? 32 : 16);
108 	pr_alert("  SET = %lu, FnV = %lu\n",
109 		 (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
110 		 (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
111 	pr_alert("  EA = %lu, S1PTW = %lu\n",
112 		 (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
113 		 (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
114 	pr_alert("  FSC = 0x%02lx: %s\n", (esr & ESR_ELx_FSC),
115 		 esr_to_fault_info(esr)->name);
116 
117 	if (esr_is_data_abort(esr))
118 		data_abort_decode(esr);
119 }
120 
121 static inline unsigned long mm_to_pgd_phys(struct mm_struct *mm)
122 {
123 	/* Either init_pg_dir or swapper_pg_dir */
124 	if (mm == &init_mm)
125 		return __pa_symbol(mm->pgd);
126 
127 	return (unsigned long)virt_to_phys(mm->pgd);
128 }
129 
130 /*
131  * Dump out the page tables associated with 'addr' in the currently active mm.
132  */
133 static void show_pte(unsigned long addr)
134 {
135 	struct mm_struct *mm;
136 	pgd_t *pgdp;
137 	pgd_t pgd;
138 
139 	if (is_ttbr0_addr(addr)) {
140 		/* TTBR0 */
141 		mm = current->active_mm;
142 		if (mm == &init_mm) {
143 			pr_alert("[%016lx] user address but active_mm is swapper\n",
144 				 addr);
145 			return;
146 		}
147 	} else if (is_ttbr1_addr(addr)) {
148 		/* TTBR1 */
149 		mm = &init_mm;
150 	} else {
151 		pr_alert("[%016lx] address between user and kernel address ranges\n",
152 			 addr);
153 		return;
154 	}
155 
156 	pr_alert("%s pgtable: %luk pages, %llu-bit VAs, pgdp=%016lx\n",
157 		 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
158 		 vabits_actual, mm_to_pgd_phys(mm));
159 	pgdp = pgd_offset(mm, addr);
160 	pgd = READ_ONCE(*pgdp);
161 	pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
162 
163 	do {
164 		p4d_t *p4dp, p4d;
165 		pud_t *pudp, pud;
166 		pmd_t *pmdp, pmd;
167 		pte_t *ptep, pte;
168 
169 		if (pgd_none(pgd) || pgd_bad(pgd))
170 			break;
171 
172 		p4dp = p4d_offset(pgdp, addr);
173 		p4d = READ_ONCE(*p4dp);
174 		pr_cont(", p4d=%016llx", p4d_val(p4d));
175 		if (p4d_none(p4d) || p4d_bad(p4d))
176 			break;
177 
178 		pudp = pud_offset(p4dp, addr);
179 		pud = READ_ONCE(*pudp);
180 		pr_cont(", pud=%016llx", pud_val(pud));
181 		if (pud_none(pud) || pud_bad(pud))
182 			break;
183 
184 		pmdp = pmd_offset(pudp, addr);
185 		pmd = READ_ONCE(*pmdp);
186 		pr_cont(", pmd=%016llx", pmd_val(pmd));
187 		if (pmd_none(pmd) || pmd_bad(pmd))
188 			break;
189 
190 		ptep = pte_offset_map(pmdp, addr);
191 		if (!ptep)
192 			break;
193 
194 		pte = READ_ONCE(*ptep);
195 		pr_cont(", pte=%016llx", pte_val(pte));
196 		pte_unmap(ptep);
197 	} while(0);
198 
199 	pr_cont("\n");
200 }
201 
202 /*
203  * This function sets the access flags (dirty, accessed), as well as write
204  * permission, and only to a more permissive setting.
205  *
206  * It needs to cope with hardware update of the accessed/dirty state by other
207  * agents in the system and can safely skip the __sync_icache_dcache() call as,
208  * like set_pte_at(), the PTE is never changed from no-exec to exec here.
209  *
210  * Returns whether or not the PTE actually changed.
211  */
212 int ptep_set_access_flags(struct vm_area_struct *vma,
213 			  unsigned long address, pte_t *ptep,
214 			  pte_t entry, int dirty)
215 {
216 	pteval_t old_pteval, pteval;
217 	pte_t pte = READ_ONCE(*ptep);
218 
219 	if (pte_same(pte, entry))
220 		return 0;
221 
222 	/* only preserve the access flags and write permission */
223 	pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
224 
225 	/*
226 	 * Setting the flags must be done atomically to avoid racing with the
227 	 * hardware update of the access/dirty state. The PTE_RDONLY bit must
228 	 * be set to the most permissive (lowest value) of *ptep and entry
229 	 * (calculated as: a & b == ~(~a | ~b)).
230 	 */
231 	pte_val(entry) ^= PTE_RDONLY;
232 	pteval = pte_val(pte);
233 	do {
234 		old_pteval = pteval;
235 		pteval ^= PTE_RDONLY;
236 		pteval |= pte_val(entry);
237 		pteval ^= PTE_RDONLY;
238 		pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
239 	} while (pteval != old_pteval);
240 
241 	/* Invalidate a stale read-only entry */
242 	if (dirty)
243 		flush_tlb_page(vma, address);
244 	return 1;
245 }
246 
247 static bool is_el1_instruction_abort(unsigned long esr)
248 {
249 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
250 }
251 
252 static bool is_el1_data_abort(unsigned long esr)
253 {
254 	return ESR_ELx_EC(esr) == ESR_ELx_EC_DABT_CUR;
255 }
256 
257 static inline bool is_el1_permission_fault(unsigned long addr, unsigned long esr,
258 					   struct pt_regs *regs)
259 {
260 	unsigned long fsc_type = esr & ESR_ELx_FSC_TYPE;
261 
262 	if (!is_el1_data_abort(esr) && !is_el1_instruction_abort(esr))
263 		return false;
264 
265 	if (fsc_type == ESR_ELx_FSC_PERM)
266 		return true;
267 
268 	if (is_ttbr0_addr(addr) && system_uses_ttbr0_pan())
269 		return fsc_type == ESR_ELx_FSC_FAULT &&
270 			(regs->pstate & PSR_PAN_BIT);
271 
272 	return false;
273 }
274 
275 static bool __kprobes is_spurious_el1_translation_fault(unsigned long addr,
276 							unsigned long esr,
277 							struct pt_regs *regs)
278 {
279 	unsigned long flags;
280 	u64 par, dfsc;
281 
282 	if (!is_el1_data_abort(esr) ||
283 	    (esr & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT)
284 		return false;
285 
286 	local_irq_save(flags);
287 	asm volatile("at s1e1r, %0" :: "r" (addr));
288 	isb();
289 	par = read_sysreg_par();
290 	local_irq_restore(flags);
291 
292 	/*
293 	 * If we now have a valid translation, treat the translation fault as
294 	 * spurious.
295 	 */
296 	if (!(par & SYS_PAR_EL1_F))
297 		return true;
298 
299 	/*
300 	 * If we got a different type of fault from the AT instruction,
301 	 * treat the translation fault as spurious.
302 	 */
303 	dfsc = FIELD_GET(SYS_PAR_EL1_FST, par);
304 	return (dfsc & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT;
305 }
306 
307 static void die_kernel_fault(const char *msg, unsigned long addr,
308 			     unsigned long esr, struct pt_regs *regs)
309 {
310 	bust_spinlocks(1);
311 
312 	pr_alert("Unable to handle kernel %s at virtual address %016lx\n", msg,
313 		 addr);
314 
315 	kasan_non_canonical_hook(addr);
316 
317 	mem_abort_decode(esr);
318 
319 	show_pte(addr);
320 	die("Oops", regs, esr);
321 	bust_spinlocks(0);
322 	make_task_dead(SIGKILL);
323 }
324 
325 #ifdef CONFIG_KASAN_HW_TAGS
326 static void report_tag_fault(unsigned long addr, unsigned long esr,
327 			     struct pt_regs *regs)
328 {
329 	/*
330 	 * SAS bits aren't set for all faults reported in EL1, so we can't
331 	 * find out access size.
332 	 */
333 	bool is_write = !!(esr & ESR_ELx_WNR);
334 	kasan_report((void *)addr, 0, is_write, regs->pc);
335 }
336 #else
337 /* Tag faults aren't enabled without CONFIG_KASAN_HW_TAGS. */
338 static inline void report_tag_fault(unsigned long addr, unsigned long esr,
339 				    struct pt_regs *regs) { }
340 #endif
341 
342 static void do_tag_recovery(unsigned long addr, unsigned long esr,
343 			   struct pt_regs *regs)
344 {
345 
346 	report_tag_fault(addr, esr, regs);
347 
348 	/*
349 	 * Disable MTE Tag Checking on the local CPU for the current EL.
350 	 * It will be done lazily on the other CPUs when they will hit a
351 	 * tag fault.
352 	 */
353 	sysreg_clear_set(sctlr_el1, SCTLR_EL1_TCF_MASK,
354 			 SYS_FIELD_PREP_ENUM(SCTLR_EL1, TCF, NONE));
355 	isb();
356 }
357 
358 static bool is_el1_mte_sync_tag_check_fault(unsigned long esr)
359 {
360 	unsigned long fsc = esr & ESR_ELx_FSC;
361 
362 	if (!is_el1_data_abort(esr))
363 		return false;
364 
365 	if (fsc == ESR_ELx_FSC_MTE)
366 		return true;
367 
368 	return false;
369 }
370 
371 static bool is_translation_fault(unsigned long esr)
372 {
373 	return (esr & ESR_ELx_FSC_TYPE) == ESR_ELx_FSC_FAULT;
374 }
375 
376 static void __do_kernel_fault(unsigned long addr, unsigned long esr,
377 			      struct pt_regs *regs)
378 {
379 	const char *msg;
380 
381 	/*
382 	 * Are we prepared to handle this kernel fault?
383 	 * We are almost certainly not prepared to handle instruction faults.
384 	 */
385 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
386 		return;
387 
388 	if (WARN_RATELIMIT(is_spurious_el1_translation_fault(addr, esr, regs),
389 	    "Ignoring spurious kernel translation fault at virtual address %016lx\n", addr))
390 		return;
391 
392 	if (is_el1_mte_sync_tag_check_fault(esr)) {
393 		do_tag_recovery(addr, esr, regs);
394 
395 		return;
396 	}
397 
398 	if (is_el1_permission_fault(addr, esr, regs)) {
399 		if (esr & ESR_ELx_WNR)
400 			msg = "write to read-only memory";
401 		else if (is_el1_instruction_abort(esr))
402 			msg = "execute from non-executable memory";
403 		else
404 			msg = "read from unreadable memory";
405 	} else if (addr < PAGE_SIZE) {
406 		msg = "NULL pointer dereference";
407 	} else {
408 		if (is_translation_fault(esr) &&
409 		    kfence_handle_page_fault(addr, esr & ESR_ELx_WNR, regs))
410 			return;
411 
412 		msg = "paging request";
413 	}
414 
415 	if (efi_runtime_fixup_exception(regs, msg))
416 		return;
417 
418 	die_kernel_fault(msg, addr, esr, regs);
419 }
420 
421 static void set_thread_esr(unsigned long address, unsigned long esr)
422 {
423 	current->thread.fault_address = address;
424 
425 	/*
426 	 * If the faulting address is in the kernel, we must sanitize the ESR.
427 	 * From userspace's point of view, kernel-only mappings don't exist
428 	 * at all, so we report them as level 0 translation faults.
429 	 * (This is not quite the way that "no mapping there at all" behaves:
430 	 * an alignment fault not caused by the memory type would take
431 	 * precedence over translation fault for a real access to empty
432 	 * space. Unfortunately we can't easily distinguish "alignment fault
433 	 * not caused by memory type" from "alignment fault caused by memory
434 	 * type", so we ignore this wrinkle and just return the translation
435 	 * fault.)
436 	 */
437 	if (!is_ttbr0_addr(current->thread.fault_address)) {
438 		switch (ESR_ELx_EC(esr)) {
439 		case ESR_ELx_EC_DABT_LOW:
440 			/*
441 			 * These bits provide only information about the
442 			 * faulting instruction, which userspace knows already.
443 			 * We explicitly clear bits which are architecturally
444 			 * RES0 in case they are given meanings in future.
445 			 * We always report the ESR as if the fault was taken
446 			 * to EL1 and so ISV and the bits in ISS[23:14] are
447 			 * clear. (In fact it always will be a fault to EL1.)
448 			 */
449 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL |
450 				ESR_ELx_CM | ESR_ELx_WNR;
451 			esr |= ESR_ELx_FSC_FAULT;
452 			break;
453 		case ESR_ELx_EC_IABT_LOW:
454 			/*
455 			 * Claim a level 0 translation fault.
456 			 * All other bits are architecturally RES0 for faults
457 			 * reported with that DFSC value, so we clear them.
458 			 */
459 			esr &= ESR_ELx_EC_MASK | ESR_ELx_IL;
460 			esr |= ESR_ELx_FSC_FAULT;
461 			break;
462 		default:
463 			/*
464 			 * This should never happen (entry.S only brings us
465 			 * into this code for insn and data aborts from a lower
466 			 * exception level). Fail safe by not providing an ESR
467 			 * context record at all.
468 			 */
469 			WARN(1, "ESR 0x%lx is not DABT or IABT from EL0\n", esr);
470 			esr = 0;
471 			break;
472 		}
473 	}
474 
475 	current->thread.fault_code = esr;
476 }
477 
478 static void do_bad_area(unsigned long far, unsigned long esr,
479 			struct pt_regs *regs)
480 {
481 	unsigned long addr = untagged_addr(far);
482 
483 	/*
484 	 * If we are in kernel mode at this point, we have no context to
485 	 * handle this fault with.
486 	 */
487 	if (user_mode(regs)) {
488 		const struct fault_info *inf = esr_to_fault_info(esr);
489 
490 		set_thread_esr(addr, esr);
491 		arm64_force_sig_fault(inf->sig, inf->code, far, inf->name);
492 	} else {
493 		__do_kernel_fault(addr, esr, regs);
494 	}
495 }
496 
497 #define VM_FAULT_BADMAP		((__force vm_fault_t)0x010000)
498 #define VM_FAULT_BADACCESS	((__force vm_fault_t)0x020000)
499 
500 static vm_fault_t __do_page_fault(struct mm_struct *mm, unsigned long addr,
501 				  unsigned int mm_flags, unsigned long vm_flags,
502 				  struct pt_regs *regs)
503 {
504 	struct vm_area_struct *vma = find_vma(mm, addr);
505 
506 	if (unlikely(!vma))
507 		return VM_FAULT_BADMAP;
508 
509 	/*
510 	 * Ok, we have a good vm_area for this memory access, so we can handle
511 	 * it.
512 	 */
513 	if (unlikely(vma->vm_start > addr)) {
514 		if (!(vma->vm_flags & VM_GROWSDOWN))
515 			return VM_FAULT_BADMAP;
516 		if (expand_stack(vma, addr))
517 			return VM_FAULT_BADMAP;
518 	}
519 
520 	/*
521 	 * Check that the permissions on the VMA allow for the fault which
522 	 * occurred.
523 	 */
524 	if (!(vma->vm_flags & vm_flags))
525 		return VM_FAULT_BADACCESS;
526 	return handle_mm_fault(vma, addr, mm_flags, regs);
527 }
528 
529 static bool is_el0_instruction_abort(unsigned long esr)
530 {
531 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
532 }
533 
534 /*
535  * Note: not valid for EL1 DC IVAC, but we never use that such that it
536  * should fault. EL0 cannot issue DC IVAC (undef).
537  */
538 static bool is_write_abort(unsigned long esr)
539 {
540 	return (esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM);
541 }
542 
543 static int __kprobes do_page_fault(unsigned long far, unsigned long esr,
544 				   struct pt_regs *regs)
545 {
546 	const struct fault_info *inf;
547 	struct mm_struct *mm = current->mm;
548 	vm_fault_t fault;
549 	unsigned long vm_flags;
550 	unsigned int mm_flags = FAULT_FLAG_DEFAULT;
551 	unsigned long addr = untagged_addr(far);
552 #ifdef CONFIG_PER_VMA_LOCK
553 	struct vm_area_struct *vma;
554 #endif
555 
556 	if (kprobe_page_fault(regs, esr))
557 		return 0;
558 
559 	/*
560 	 * If we're in an interrupt or have no user context, we must not take
561 	 * the fault.
562 	 */
563 	if (faulthandler_disabled() || !mm)
564 		goto no_context;
565 
566 	if (user_mode(regs))
567 		mm_flags |= FAULT_FLAG_USER;
568 
569 	/*
570 	 * vm_flags tells us what bits we must have in vma->vm_flags
571 	 * for the fault to be benign, __do_page_fault() would check
572 	 * vma->vm_flags & vm_flags and returns an error if the
573 	 * intersection is empty
574 	 */
575 	if (is_el0_instruction_abort(esr)) {
576 		/* It was exec fault */
577 		vm_flags = VM_EXEC;
578 		mm_flags |= FAULT_FLAG_INSTRUCTION;
579 	} else if (is_write_abort(esr)) {
580 		/* It was write fault */
581 		vm_flags = VM_WRITE;
582 		mm_flags |= FAULT_FLAG_WRITE;
583 	} else {
584 		/* It was read fault */
585 		vm_flags = VM_READ;
586 		/* Write implies read */
587 		vm_flags |= VM_WRITE;
588 		/* If EPAN is absent then exec implies read */
589 		if (!cpus_have_const_cap(ARM64_HAS_EPAN))
590 			vm_flags |= VM_EXEC;
591 	}
592 
593 	if (is_ttbr0_addr(addr) && is_el1_permission_fault(addr, esr, regs)) {
594 		if (is_el1_instruction_abort(esr))
595 			die_kernel_fault("execution of user memory",
596 					 addr, esr, regs);
597 
598 		if (!search_exception_tables(regs->pc))
599 			die_kernel_fault("access to user memory outside uaccess routines",
600 					 addr, esr, regs);
601 	}
602 
603 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
604 
605 #ifdef CONFIG_PER_VMA_LOCK
606 	if (!(mm_flags & FAULT_FLAG_USER))
607 		goto lock_mmap;
608 
609 	vma = lock_vma_under_rcu(mm, addr);
610 	if (!vma)
611 		goto lock_mmap;
612 
613 	if (!(vma->vm_flags & vm_flags)) {
614 		vma_end_read(vma);
615 		goto lock_mmap;
616 	}
617 	fault = handle_mm_fault(vma, addr, mm_flags | FAULT_FLAG_VMA_LOCK, regs);
618 	vma_end_read(vma);
619 
620 	if (!(fault & VM_FAULT_RETRY)) {
621 		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
622 		goto done;
623 	}
624 	count_vm_vma_lock_event(VMA_LOCK_RETRY);
625 
626 	/* Quick path to respond to signals */
627 	if (fault_signal_pending(fault, regs)) {
628 		if (!user_mode(regs))
629 			goto no_context;
630 		return 0;
631 	}
632 lock_mmap:
633 #endif /* CONFIG_PER_VMA_LOCK */
634 	/*
635 	 * As per x86, we may deadlock here. However, since the kernel only
636 	 * validly references user space from well defined areas of the code,
637 	 * we can bug out early if this is from code which shouldn't.
638 	 */
639 	if (!mmap_read_trylock(mm)) {
640 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
641 			goto no_context;
642 retry:
643 		mmap_read_lock(mm);
644 	} else {
645 		/*
646 		 * The above mmap_read_trylock() might have succeeded in which
647 		 * case, we'll have missed the might_sleep() from down_read().
648 		 */
649 		might_sleep();
650 #ifdef CONFIG_DEBUG_VM
651 		if (!user_mode(regs) && !search_exception_tables(regs->pc)) {
652 			mmap_read_unlock(mm);
653 			goto no_context;
654 		}
655 #endif
656 	}
657 
658 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, regs);
659 
660 	/* Quick path to respond to signals */
661 	if (fault_signal_pending(fault, regs)) {
662 		if (!user_mode(regs))
663 			goto no_context;
664 		return 0;
665 	}
666 
667 	/* The fault is fully completed (including releasing mmap lock) */
668 	if (fault & VM_FAULT_COMPLETED)
669 		return 0;
670 
671 	if (fault & VM_FAULT_RETRY) {
672 		mm_flags |= FAULT_FLAG_TRIED;
673 		goto retry;
674 	}
675 	mmap_read_unlock(mm);
676 
677 #ifdef CONFIG_PER_VMA_LOCK
678 done:
679 #endif
680 	/*
681 	 * Handle the "normal" (no error) case first.
682 	 */
683 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
684 			      VM_FAULT_BADACCESS))))
685 		return 0;
686 
687 	/*
688 	 * If we are in kernel mode at this point, we have no context to
689 	 * handle this fault with.
690 	 */
691 	if (!user_mode(regs))
692 		goto no_context;
693 
694 	if (fault & VM_FAULT_OOM) {
695 		/*
696 		 * We ran out of memory, call the OOM killer, and return to
697 		 * userspace (which will retry the fault, or kill us if we got
698 		 * oom-killed).
699 		 */
700 		pagefault_out_of_memory();
701 		return 0;
702 	}
703 
704 	inf = esr_to_fault_info(esr);
705 	set_thread_esr(addr, esr);
706 	if (fault & VM_FAULT_SIGBUS) {
707 		/*
708 		 * We had some memory, but were unable to successfully fix up
709 		 * this page fault.
710 		 */
711 		arm64_force_sig_fault(SIGBUS, BUS_ADRERR, far, inf->name);
712 	} else if (fault & (VM_FAULT_HWPOISON_LARGE | VM_FAULT_HWPOISON)) {
713 		unsigned int lsb;
714 
715 		lsb = PAGE_SHIFT;
716 		if (fault & VM_FAULT_HWPOISON_LARGE)
717 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
718 
719 		arm64_force_sig_mceerr(BUS_MCEERR_AR, far, lsb, inf->name);
720 	} else {
721 		/*
722 		 * Something tried to access memory that isn't in our memory
723 		 * map.
724 		 */
725 		arm64_force_sig_fault(SIGSEGV,
726 				      fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR,
727 				      far, inf->name);
728 	}
729 
730 	return 0;
731 
732 no_context:
733 	__do_kernel_fault(addr, esr, regs);
734 	return 0;
735 }
736 
737 static int __kprobes do_translation_fault(unsigned long far,
738 					  unsigned long esr,
739 					  struct pt_regs *regs)
740 {
741 	unsigned long addr = untagged_addr(far);
742 
743 	if (is_ttbr0_addr(addr))
744 		return do_page_fault(far, esr, regs);
745 
746 	do_bad_area(far, esr, regs);
747 	return 0;
748 }
749 
750 static int do_alignment_fault(unsigned long far, unsigned long esr,
751 			      struct pt_regs *regs)
752 {
753 	if (IS_ENABLED(CONFIG_COMPAT_ALIGNMENT_FIXUPS) &&
754 	    compat_user_mode(regs))
755 		return do_compat_alignment_fixup(far, regs);
756 	do_bad_area(far, esr, regs);
757 	return 0;
758 }
759 
760 static int do_bad(unsigned long far, unsigned long esr, struct pt_regs *regs)
761 {
762 	return 1; /* "fault" */
763 }
764 
765 static int do_sea(unsigned long far, unsigned long esr, struct pt_regs *regs)
766 {
767 	const struct fault_info *inf;
768 	unsigned long siaddr;
769 
770 	inf = esr_to_fault_info(esr);
771 
772 	if (user_mode(regs) && apei_claim_sea(regs) == 0) {
773 		/*
774 		 * APEI claimed this as a firmware-first notification.
775 		 * Some processing deferred to task_work before ret_to_user().
776 		 */
777 		return 0;
778 	}
779 
780 	if (esr & ESR_ELx_FnV) {
781 		siaddr = 0;
782 	} else {
783 		/*
784 		 * The architecture specifies that the tag bits of FAR_EL1 are
785 		 * UNKNOWN for synchronous external aborts. Mask them out now
786 		 * so that userspace doesn't see them.
787 		 */
788 		siaddr  = untagged_addr(far);
789 	}
790 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
791 
792 	return 0;
793 }
794 
795 static int do_tag_check_fault(unsigned long far, unsigned long esr,
796 			      struct pt_regs *regs)
797 {
798 	/*
799 	 * The architecture specifies that bits 63:60 of FAR_EL1 are UNKNOWN
800 	 * for tag check faults. Set them to corresponding bits in the untagged
801 	 * address.
802 	 */
803 	far = (__untagged_addr(far) & ~MTE_TAG_MASK) | (far & MTE_TAG_MASK);
804 	do_bad_area(far, esr, regs);
805 	return 0;
806 }
807 
808 static const struct fault_info fault_info[] = {
809 	{ do_bad,		SIGKILL, SI_KERNEL,	"ttbr address size fault"	},
810 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 1 address size fault"	},
811 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 2 address size fault"	},
812 	{ do_bad,		SIGKILL, SI_KERNEL,	"level 3 address size fault"	},
813 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
814 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
815 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
816 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
817 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 8"			},
818 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
819 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
820 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
821 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 12"			},
822 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
823 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
824 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
825 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous external abort"	},
826 	{ do_tag_check_fault,	SIGSEGV, SEGV_MTESERR,	"synchronous tag check fault"	},
827 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 18"			},
828 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 19"			},
829 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 (translation table walk)"	},
830 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 (translation table walk)"	},
831 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 (translation table walk)"	},
832 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 (translation table walk)"	},
833 	{ do_sea,		SIGBUS,  BUS_OBJERR,	"synchronous parity or ECC error" },	// Reserved when RAS is implemented
834 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 25"			},
835 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 26"			},
836 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 27"			},
837 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 0 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
838 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 1 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
839 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 2 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
840 	{ do_sea,		SIGKILL, SI_KERNEL,	"level 3 synchronous parity error (translation table walk)"	},	// Reserved when RAS is implemented
841 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 32"			},
842 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
843 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 34"			},
844 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 35"			},
845 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 36"			},
846 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 37"			},
847 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 38"			},
848 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 39"			},
849 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 40"			},
850 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 41"			},
851 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 42"			},
852 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 43"			},
853 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 44"			},
854 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 45"			},
855 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 46"			},
856 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 47"			},
857 	{ do_bad,		SIGKILL, SI_KERNEL,	"TLB conflict abort"		},
858 	{ do_bad,		SIGKILL, SI_KERNEL,	"Unsupported atomic hardware update fault"	},
859 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 50"			},
860 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 51"			},
861 	{ do_bad,		SIGKILL, SI_KERNEL,	"implementation fault (lockdown abort)" },
862 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"implementation fault (unsupported exclusive)" },
863 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 54"			},
864 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 55"			},
865 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 56"			},
866 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 57"			},
867 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 58" 			},
868 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 59"			},
869 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 60"			},
870 	{ do_bad,		SIGKILL, SI_KERNEL,	"section domain fault"		},
871 	{ do_bad,		SIGKILL, SI_KERNEL,	"page domain fault"		},
872 	{ do_bad,		SIGKILL, SI_KERNEL,	"unknown 63"			},
873 };
874 
875 void do_mem_abort(unsigned long far, unsigned long esr, struct pt_regs *regs)
876 {
877 	const struct fault_info *inf = esr_to_fault_info(esr);
878 	unsigned long addr = untagged_addr(far);
879 
880 	if (!inf->fn(far, esr, regs))
881 		return;
882 
883 	if (!user_mode(regs))
884 		die_kernel_fault(inf->name, addr, esr, regs);
885 
886 	/*
887 	 * At this point we have an unrecognized fault type whose tag bits may
888 	 * have been defined as UNKNOWN. Therefore we only expose the untagged
889 	 * address to the signal handler.
890 	 */
891 	arm64_notify_die(inf->name, regs, inf->sig, inf->code, addr, esr);
892 }
893 NOKPROBE_SYMBOL(do_mem_abort);
894 
895 void do_sp_pc_abort(unsigned long addr, unsigned long esr, struct pt_regs *regs)
896 {
897 	arm64_notify_die("SP/PC alignment exception", regs, SIGBUS, BUS_ADRALN,
898 			 addr, esr);
899 }
900 NOKPROBE_SYMBOL(do_sp_pc_abort);
901 
902 /*
903  * __refdata because early_brk64 is __init, but the reference to it is
904  * clobbered at arch_initcall time.
905  * See traps.c and debug-monitors.c:debug_traps_init().
906  */
907 static struct fault_info __refdata debug_fault_info[] = {
908 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
909 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
910 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
911 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 3"		},
912 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
913 	{ do_bad,	SIGKILL,	SI_KERNEL,	"aarch32 vector catch"	},
914 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
915 	{ do_bad,	SIGKILL,	SI_KERNEL,	"unknown 7"		},
916 };
917 
918 void __init hook_debug_fault_code(int nr,
919 				  int (*fn)(unsigned long, unsigned long, struct pt_regs *),
920 				  int sig, int code, const char *name)
921 {
922 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
923 
924 	debug_fault_info[nr].fn		= fn;
925 	debug_fault_info[nr].sig	= sig;
926 	debug_fault_info[nr].code	= code;
927 	debug_fault_info[nr].name	= name;
928 }
929 
930 /*
931  * In debug exception context, we explicitly disable preemption despite
932  * having interrupts disabled.
933  * This serves two purposes: it makes it much less likely that we would
934  * accidentally schedule in exception context and it will force a warning
935  * if we somehow manage to schedule by accident.
936  */
937 static void debug_exception_enter(struct pt_regs *regs)
938 {
939 	preempt_disable();
940 
941 	/* This code is a bit fragile.  Test it. */
942 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "exception_enter didn't work");
943 }
944 NOKPROBE_SYMBOL(debug_exception_enter);
945 
946 static void debug_exception_exit(struct pt_regs *regs)
947 {
948 	preempt_enable_no_resched();
949 }
950 NOKPROBE_SYMBOL(debug_exception_exit);
951 
952 void do_debug_exception(unsigned long addr_if_watchpoint, unsigned long esr,
953 			struct pt_regs *regs)
954 {
955 	const struct fault_info *inf = esr_to_debug_fault_info(esr);
956 	unsigned long pc = instruction_pointer(regs);
957 
958 	debug_exception_enter(regs);
959 
960 	if (user_mode(regs) && !is_ttbr0_addr(pc))
961 		arm64_apply_bp_hardening();
962 
963 	if (inf->fn(addr_if_watchpoint, esr, regs)) {
964 		arm64_notify_die(inf->name, regs, inf->sig, inf->code, pc, esr);
965 	}
966 
967 	debug_exception_exit(regs);
968 }
969 NOKPROBE_SYMBOL(do_debug_exception);
970 
971 /*
972  * Used during anonymous page fault handling.
973  */
974 struct folio *vma_alloc_zeroed_movable_folio(struct vm_area_struct *vma,
975 						unsigned long vaddr)
976 {
977 	gfp_t flags = GFP_HIGHUSER_MOVABLE | __GFP_ZERO;
978 
979 	/*
980 	 * If the page is mapped with PROT_MTE, initialise the tags at the
981 	 * point of allocation and page zeroing as this is usually faster than
982 	 * separate DC ZVA and STGM.
983 	 */
984 	if (vma->vm_flags & VM_MTE)
985 		flags |= __GFP_ZEROTAGS;
986 
987 	return vma_alloc_folio(flags, 0, vma, vaddr, false);
988 }
989 
990 void tag_clear_highpage(struct page *page)
991 {
992 	/* Newly allocated page, shouldn't have been tagged yet */
993 	WARN_ON_ONCE(!try_page_mte_tagging(page));
994 	mte_zero_clear_page_tags(page_address(page));
995 	set_page_mte_tagged(page);
996 }
997