1 // SPDX-License-Identifier: GPL-2.0-only
3 * Based on arch/arm/mm/fault.c
5 * Copyright (C) 1995 Linus Torvalds
6 * Copyright (C) 1995-2004 Russell King
7 * Copyright (C) 2012 ARM Ltd.
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>
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>
31 #include <asm/cmpxchg.h>
32 #include <asm/cpufeature.h>
33 #include <asm/exception.h>
34 #include <asm/daifflags.h>
35 #include <asm/debug-monitors.h>
37 #include <asm/kprobes.h>
39 #include <asm/processor.h>
40 #include <asm/sysreg.h>
41 #include <asm/system_misc.h>
42 #include <asm/tlbflush.h>
43 #include <asm/traps.h>
46 int (*fn)(unsigned long far, unsigned int esr,
47 struct pt_regs *regs);
53 static const struct fault_info fault_info[];
54 static struct fault_info debug_fault_info[];
56 static inline const struct fault_info *esr_to_fault_info(unsigned int esr)
58 return fault_info + (esr & ESR_ELx_FSC);
61 static inline const struct fault_info *esr_to_debug_fault_info(unsigned int esr)
63 return debug_fault_info + DBG_ESR_EVT(esr);
66 static void data_abort_decode(unsigned int esr)
68 pr_alert("Data abort info:\n");
70 if (esr & ESR_ELx_ISV) {
71 pr_alert(" Access size = %u byte(s)\n",
72 1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
73 pr_alert(" SSE = %lu, SRT = %lu\n",
74 (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
75 (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
76 pr_alert(" SF = %lu, AR = %lu\n",
77 (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
78 (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
80 pr_alert(" ISV = 0, ISS = 0x%08lx\n", esr & ESR_ELx_ISS_MASK);
83 pr_alert(" CM = %lu, WnR = %lu\n",
84 (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
85 (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT);
88 static void mem_abort_decode(unsigned int esr)
90 pr_alert("Mem abort info:\n");
92 pr_alert(" ESR = 0x%08x\n", esr);
93 pr_alert(" EC = 0x%02lx: %s, IL = %u bits\n",
94 ESR_ELx_EC(esr), esr_get_class_string(esr),
95 (esr & ESR_ELx_IL) ? 32 : 16);
96 pr_alert(" SET = %lu, FnV = %lu\n",
97 (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
98 (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
99 pr_alert(" EA = %lu, S1PTW = %lu\n",
100 (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
101 (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
102 pr_alert(" FSC = 0x%02x: %s\n", (esr & ESR_ELx_FSC),
103 esr_to_fault_info(esr)->name);
105 if (esr_is_data_abort(esr))
106 data_abort_decode(esr);
109 static inline unsigned long mm_to_pgd_phys(struct mm_struct *mm)
111 /* Either init_pg_dir or swapper_pg_dir */
113 return __pa_symbol(mm->pgd);
115 return (unsigned long)virt_to_phys(mm->pgd);
119 * Dump out the page tables associated with 'addr' in the currently active mm.
121 static void show_pte(unsigned long addr)
123 struct mm_struct *mm;
127 if (is_ttbr0_addr(addr)) {
129 mm = current->active_mm;
130 if (mm == &init_mm) {
131 pr_alert("[%016lx] user address but active_mm is swapper\n",
135 } else if (is_ttbr1_addr(addr)) {
139 pr_alert("[%016lx] address between user and kernel address ranges\n",
144 pr_alert("%s pgtable: %luk pages, %llu-bit VAs, pgdp=%016lx\n",
145 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
146 vabits_actual, mm_to_pgd_phys(mm));
147 pgdp = pgd_offset(mm, addr);
148 pgd = READ_ONCE(*pgdp);
149 pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
157 if (pgd_none(pgd) || pgd_bad(pgd))
160 p4dp = p4d_offset(pgdp, addr);
161 p4d = READ_ONCE(*p4dp);
162 pr_cont(", p4d=%016llx", p4d_val(p4d));
163 if (p4d_none(p4d) || p4d_bad(p4d))
166 pudp = pud_offset(p4dp, addr);
167 pud = READ_ONCE(*pudp);
168 pr_cont(", pud=%016llx", pud_val(pud));
169 if (pud_none(pud) || pud_bad(pud))
172 pmdp = pmd_offset(pudp, addr);
173 pmd = READ_ONCE(*pmdp);
174 pr_cont(", pmd=%016llx", pmd_val(pmd));
175 if (pmd_none(pmd) || pmd_bad(pmd))
178 ptep = pte_offset_map(pmdp, addr);
179 pte = READ_ONCE(*ptep);
180 pr_cont(", pte=%016llx", pte_val(pte));
188 * This function sets the access flags (dirty, accessed), as well as write
189 * permission, and only to a more permissive setting.
191 * It needs to cope with hardware update of the accessed/dirty state by other
192 * agents in the system and can safely skip the __sync_icache_dcache() call as,
193 * like set_pte_at(), the PTE is never changed from no-exec to exec here.
195 * Returns whether or not the PTE actually changed.
197 int ptep_set_access_flags(struct vm_area_struct *vma,
198 unsigned long address, pte_t *ptep,
199 pte_t entry, int dirty)
201 pteval_t old_pteval, pteval;
202 pte_t pte = READ_ONCE(*ptep);
204 if (pte_same(pte, entry))
207 /* only preserve the access flags and write permission */
208 pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
211 * Setting the flags must be done atomically to avoid racing with the
212 * hardware update of the access/dirty state. The PTE_RDONLY bit must
213 * be set to the most permissive (lowest value) of *ptep and entry
214 * (calculated as: a & b == ~(~a | ~b)).
216 pte_val(entry) ^= PTE_RDONLY;
217 pteval = pte_val(pte);
220 pteval ^= PTE_RDONLY;
221 pteval |= pte_val(entry);
222 pteval ^= PTE_RDONLY;
223 pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
224 } while (pteval != old_pteval);
226 /* Invalidate a stale read-only entry */
228 flush_tlb_page(vma, address);
232 static bool is_el1_instruction_abort(unsigned int esr)
234 return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
237 static bool is_el1_data_abort(unsigned int esr)
239 return ESR_ELx_EC(esr) == ESR_ELx_EC_DABT_CUR;
242 static inline bool is_el1_permission_fault(unsigned long addr, unsigned int esr,
243 struct pt_regs *regs)
245 unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
247 if (!is_el1_data_abort(esr) && !is_el1_instruction_abort(esr))
250 if (fsc_type == ESR_ELx_FSC_PERM)
253 if (is_ttbr0_addr(addr) && system_uses_ttbr0_pan())
254 return fsc_type == ESR_ELx_FSC_FAULT &&
255 (regs->pstate & PSR_PAN_BIT);
260 static bool __kprobes is_spurious_el1_translation_fault(unsigned long addr,
262 struct pt_regs *regs)
267 if (!is_el1_data_abort(esr) ||
268 (esr & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT)
271 local_irq_save(flags);
272 asm volatile("at s1e1r, %0" :: "r" (addr));
274 par = read_sysreg_par();
275 local_irq_restore(flags);
278 * If we now have a valid translation, treat the translation fault as
281 if (!(par & SYS_PAR_EL1_F))
285 * If we got a different type of fault from the AT instruction,
286 * treat the translation fault as spurious.
288 dfsc = FIELD_GET(SYS_PAR_EL1_FST, par);
289 return (dfsc & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT;
292 static void die_kernel_fault(const char *msg, unsigned long addr,
293 unsigned int esr, struct pt_regs *regs)
297 pr_alert("Unable to handle kernel %s at virtual address %016lx\n", msg,
300 kasan_non_canonical_hook(addr);
302 mem_abort_decode(esr);
305 die("Oops", regs, esr);
310 #ifdef CONFIG_KASAN_HW_TAGS
311 static void report_tag_fault(unsigned long addr, unsigned int esr,
312 struct pt_regs *regs)
315 * SAS bits aren't set for all faults reported in EL1, so we can't
316 * find out access size.
318 bool is_write = !!(esr & ESR_ELx_WNR);
319 kasan_report(addr, 0, is_write, regs->pc);
322 /* Tag faults aren't enabled without CONFIG_KASAN_HW_TAGS. */
323 static inline void report_tag_fault(unsigned long addr, unsigned int esr,
324 struct pt_regs *regs) { }
327 static void do_tag_recovery(unsigned long addr, unsigned int esr,
328 struct pt_regs *regs)
331 report_tag_fault(addr, esr, regs);
334 * Disable MTE Tag Checking on the local CPU for the current EL.
335 * It will be done lazily on the other CPUs when they will hit a
338 sysreg_clear_set(sctlr_el1, SCTLR_ELx_TCF_MASK, SCTLR_ELx_TCF_NONE);
342 static bool is_el1_mte_sync_tag_check_fault(unsigned int esr)
344 unsigned int fsc = esr & ESR_ELx_FSC;
346 if (!is_el1_data_abort(esr))
349 if (fsc == ESR_ELx_FSC_MTE)
355 static void __do_kernel_fault(unsigned long addr, unsigned int esr,
356 struct pt_regs *regs)
361 * Are we prepared to handle this kernel fault?
362 * We are almost certainly not prepared to handle instruction faults.
364 if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
367 if (WARN_RATELIMIT(is_spurious_el1_translation_fault(addr, esr, regs),
368 "Ignoring spurious kernel translation fault at virtual address %016lx\n", addr))
371 if (is_el1_mte_sync_tag_check_fault(esr)) {
372 do_tag_recovery(addr, esr, regs);
377 if (is_el1_permission_fault(addr, esr, regs)) {
378 if (esr & ESR_ELx_WNR)
379 msg = "write to read-only memory";
380 else if (is_el1_instruction_abort(esr))
381 msg = "execute from non-executable memory";
383 msg = "read from unreadable memory";
384 } else if (addr < PAGE_SIZE) {
385 msg = "NULL pointer dereference";
387 if (kfence_handle_page_fault(addr, esr & ESR_ELx_WNR, regs))
390 msg = "paging request";
393 die_kernel_fault(msg, addr, esr, regs);
396 static void set_thread_esr(unsigned long address, unsigned int esr)
398 current->thread.fault_address = address;
401 * If the faulting address is in the kernel, we must sanitize the ESR.
402 * From userspace's point of view, kernel-only mappings don't exist
403 * at all, so we report them as level 0 translation faults.
404 * (This is not quite the way that "no mapping there at all" behaves:
405 * an alignment fault not caused by the memory type would take
406 * precedence over translation fault for a real access to empty
407 * space. Unfortunately we can't easily distinguish "alignment fault
408 * not caused by memory type" from "alignment fault caused by memory
409 * type", so we ignore this wrinkle and just return the translation
412 if (!is_ttbr0_addr(current->thread.fault_address)) {
413 switch (ESR_ELx_EC(esr)) {
414 case ESR_ELx_EC_DABT_LOW:
416 * These bits provide only information about the
417 * faulting instruction, which userspace knows already.
418 * We explicitly clear bits which are architecturally
419 * RES0 in case they are given meanings in future.
420 * We always report the ESR as if the fault was taken
421 * to EL1 and so ISV and the bits in ISS[23:14] are
422 * clear. (In fact it always will be a fault to EL1.)
424 esr &= ESR_ELx_EC_MASK | ESR_ELx_IL |
425 ESR_ELx_CM | ESR_ELx_WNR;
426 esr |= ESR_ELx_FSC_FAULT;
428 case ESR_ELx_EC_IABT_LOW:
430 * Claim a level 0 translation fault.
431 * All other bits are architecturally RES0 for faults
432 * reported with that DFSC value, so we clear them.
434 esr &= ESR_ELx_EC_MASK | ESR_ELx_IL;
435 esr |= ESR_ELx_FSC_FAULT;
439 * This should never happen (entry.S only brings us
440 * into this code for insn and data aborts from a lower
441 * exception level). Fail safe by not providing an ESR
442 * context record at all.
444 WARN(1, "ESR 0x%x is not DABT or IABT from EL0\n", esr);
450 current->thread.fault_code = esr;
453 static void do_bad_area(unsigned long far, unsigned int esr,
454 struct pt_regs *regs)
456 unsigned long addr = untagged_addr(far);
459 * If we are in kernel mode at this point, we have no context to
460 * handle this fault with.
462 if (user_mode(regs)) {
463 const struct fault_info *inf = esr_to_fault_info(esr);
465 set_thread_esr(addr, esr);
466 arm64_force_sig_fault(inf->sig, inf->code, far, inf->name);
468 __do_kernel_fault(addr, esr, regs);
472 #define VM_FAULT_BADMAP 0x010000
473 #define VM_FAULT_BADACCESS 0x020000
475 static vm_fault_t __do_page_fault(struct mm_struct *mm, unsigned long addr,
476 unsigned int mm_flags, unsigned long vm_flags,
477 struct pt_regs *regs)
479 struct vm_area_struct *vma = find_vma(mm, addr);
482 return VM_FAULT_BADMAP;
485 * Ok, we have a good vm_area for this memory access, so we can handle
488 if (unlikely(vma->vm_start > addr)) {
489 if (!(vma->vm_flags & VM_GROWSDOWN))
490 return VM_FAULT_BADMAP;
491 if (expand_stack(vma, addr))
492 return VM_FAULT_BADMAP;
496 * Check that the permissions on the VMA allow for the fault which
499 if (!(vma->vm_flags & vm_flags))
500 return VM_FAULT_BADACCESS;
501 return handle_mm_fault(vma, addr, mm_flags, regs);
504 static bool is_el0_instruction_abort(unsigned int esr)
506 return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
510 * Note: not valid for EL1 DC IVAC, but we never use that such that it
511 * should fault. EL0 cannot issue DC IVAC (undef).
513 static bool is_write_abort(unsigned int esr)
515 return (esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM);
518 static int __kprobes do_page_fault(unsigned long far, unsigned int esr,
519 struct pt_regs *regs)
521 const struct fault_info *inf;
522 struct mm_struct *mm = current->mm;
524 unsigned long vm_flags;
525 unsigned int mm_flags = FAULT_FLAG_DEFAULT;
526 unsigned long addr = untagged_addr(far);
528 if (kprobe_page_fault(regs, esr))
532 * If we're in an interrupt or have no user context, we must not take
535 if (faulthandler_disabled() || !mm)
539 mm_flags |= FAULT_FLAG_USER;
542 * vm_flags tells us what bits we must have in vma->vm_flags
543 * for the fault to be benign, __do_page_fault() would check
544 * vma->vm_flags & vm_flags and returns an error if the
545 * intersection is empty
547 if (is_el0_instruction_abort(esr)) {
548 /* It was exec fault */
550 mm_flags |= FAULT_FLAG_INSTRUCTION;
551 } else if (is_write_abort(esr)) {
552 /* It was write fault */
554 mm_flags |= FAULT_FLAG_WRITE;
556 /* It was read fault */
558 /* Write implies read */
559 vm_flags |= VM_WRITE;
560 /* If EPAN is absent then exec implies read */
561 if (!cpus_have_const_cap(ARM64_HAS_EPAN))
565 if (is_ttbr0_addr(addr) && is_el1_permission_fault(addr, esr, regs)) {
566 if (is_el1_instruction_abort(esr))
567 die_kernel_fault("execution of user memory",
570 if (!search_exception_tables(regs->pc))
571 die_kernel_fault("access to user memory outside uaccess routines",
575 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
578 * As per x86, we may deadlock here. However, since the kernel only
579 * validly references user space from well defined areas of the code,
580 * we can bug out early if this is from code which shouldn't.
582 if (!mmap_read_trylock(mm)) {
583 if (!user_mode(regs) && !search_exception_tables(regs->pc))
589 * The above mmap_read_trylock() might have succeeded in which
590 * case, we'll have missed the might_sleep() from down_read().
593 #ifdef CONFIG_DEBUG_VM
594 if (!user_mode(regs) && !search_exception_tables(regs->pc)) {
595 mmap_read_unlock(mm);
601 fault = __do_page_fault(mm, addr, mm_flags, vm_flags, regs);
603 /* Quick path to respond to signals */
604 if (fault_signal_pending(fault, regs)) {
605 if (!user_mode(regs))
610 if (fault & VM_FAULT_RETRY) {
611 if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
612 mm_flags |= FAULT_FLAG_TRIED;
616 mmap_read_unlock(mm);
619 * Handle the "normal" (no error) case first.
621 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
622 VM_FAULT_BADACCESS))))
626 * If we are in kernel mode at this point, we have no context to
627 * handle this fault with.
629 if (!user_mode(regs))
632 if (fault & VM_FAULT_OOM) {
634 * We ran out of memory, call the OOM killer, and return to
635 * userspace (which will retry the fault, or kill us if we got
638 pagefault_out_of_memory();
642 inf = esr_to_fault_info(esr);
643 set_thread_esr(addr, esr);
644 if (fault & VM_FAULT_SIGBUS) {
646 * We had some memory, but were unable to successfully fix up
649 arm64_force_sig_fault(SIGBUS, BUS_ADRERR, far, inf->name);
650 } else if (fault & (VM_FAULT_HWPOISON_LARGE | VM_FAULT_HWPOISON)) {
654 if (fault & VM_FAULT_HWPOISON_LARGE)
655 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
657 arm64_force_sig_mceerr(BUS_MCEERR_AR, far, lsb, inf->name);
660 * Something tried to access memory that isn't in our memory
663 arm64_force_sig_fault(SIGSEGV,
664 fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR,
671 __do_kernel_fault(addr, esr, regs);
675 static int __kprobes do_translation_fault(unsigned long far,
677 struct pt_regs *regs)
679 unsigned long addr = untagged_addr(far);
681 if (is_ttbr0_addr(addr))
682 return do_page_fault(far, esr, regs);
684 do_bad_area(far, esr, regs);
688 static int do_alignment_fault(unsigned long far, unsigned int esr,
689 struct pt_regs *regs)
691 do_bad_area(far, esr, regs);
695 static int do_bad(unsigned long far, unsigned int esr, struct pt_regs *regs)
697 return 1; /* "fault" */
700 static int do_sea(unsigned long far, unsigned int esr, struct pt_regs *regs)
702 const struct fault_info *inf;
703 unsigned long siaddr;
705 inf = esr_to_fault_info(esr);
707 if (user_mode(regs) && apei_claim_sea(regs) == 0) {
709 * APEI claimed this as a firmware-first notification.
710 * Some processing deferred to task_work before ret_to_user().
715 if (esr & ESR_ELx_FnV) {
719 * The architecture specifies that the tag bits of FAR_EL1 are
720 * UNKNOWN for synchronous external aborts. Mask them out now
721 * so that userspace doesn't see them.
723 siaddr = untagged_addr(far);
725 arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
730 static int do_tag_check_fault(unsigned long far, unsigned int esr,
731 struct pt_regs *regs)
734 * The architecture specifies that bits 63:60 of FAR_EL1 are UNKNOWN
735 * for tag check faults. Set them to corresponding bits in the untagged
738 far = (__untagged_addr(far) & ~MTE_TAG_MASK) | (far & MTE_TAG_MASK);
739 do_bad_area(far, esr, regs);
743 static const struct fault_info fault_info[] = {
744 { do_bad, SIGKILL, SI_KERNEL, "ttbr address size fault" },
745 { do_bad, SIGKILL, SI_KERNEL, "level 1 address size fault" },
746 { do_bad, SIGKILL, SI_KERNEL, "level 2 address size fault" },
747 { do_bad, SIGKILL, SI_KERNEL, "level 3 address size fault" },
748 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 0 translation fault" },
749 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 1 translation fault" },
750 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 2 translation fault" },
751 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 3 translation fault" },
752 { do_bad, SIGKILL, SI_KERNEL, "unknown 8" },
753 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 1 access flag fault" },
754 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 2 access flag fault" },
755 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 3 access flag fault" },
756 { do_bad, SIGKILL, SI_KERNEL, "unknown 12" },
757 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 1 permission fault" },
758 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 2 permission fault" },
759 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 3 permission fault" },
760 { do_sea, SIGBUS, BUS_OBJERR, "synchronous external abort" },
761 { do_tag_check_fault, SIGSEGV, SEGV_MTESERR, "synchronous tag check fault" },
762 { do_bad, SIGKILL, SI_KERNEL, "unknown 18" },
763 { do_bad, SIGKILL, SI_KERNEL, "unknown 19" },
764 { do_sea, SIGKILL, SI_KERNEL, "level 0 (translation table walk)" },
765 { do_sea, SIGKILL, SI_KERNEL, "level 1 (translation table walk)" },
766 { do_sea, SIGKILL, SI_KERNEL, "level 2 (translation table walk)" },
767 { do_sea, SIGKILL, SI_KERNEL, "level 3 (translation table walk)" },
768 { do_sea, SIGBUS, BUS_OBJERR, "synchronous parity or ECC error" }, // Reserved when RAS is implemented
769 { do_bad, SIGKILL, SI_KERNEL, "unknown 25" },
770 { do_bad, SIGKILL, SI_KERNEL, "unknown 26" },
771 { do_bad, SIGKILL, SI_KERNEL, "unknown 27" },
772 { do_sea, SIGKILL, SI_KERNEL, "level 0 synchronous parity error (translation table walk)" }, // Reserved when RAS is implemented
773 { do_sea, SIGKILL, SI_KERNEL, "level 1 synchronous parity error (translation table walk)" }, // Reserved when RAS is implemented
774 { do_sea, SIGKILL, SI_KERNEL, "level 2 synchronous parity error (translation table walk)" }, // Reserved when RAS is implemented
775 { do_sea, SIGKILL, SI_KERNEL, "level 3 synchronous parity error (translation table walk)" }, // Reserved when RAS is implemented
776 { do_bad, SIGKILL, SI_KERNEL, "unknown 32" },
777 { do_alignment_fault, SIGBUS, BUS_ADRALN, "alignment fault" },
778 { do_bad, SIGKILL, SI_KERNEL, "unknown 34" },
779 { do_bad, SIGKILL, SI_KERNEL, "unknown 35" },
780 { do_bad, SIGKILL, SI_KERNEL, "unknown 36" },
781 { do_bad, SIGKILL, SI_KERNEL, "unknown 37" },
782 { do_bad, SIGKILL, SI_KERNEL, "unknown 38" },
783 { do_bad, SIGKILL, SI_KERNEL, "unknown 39" },
784 { do_bad, SIGKILL, SI_KERNEL, "unknown 40" },
785 { do_bad, SIGKILL, SI_KERNEL, "unknown 41" },
786 { do_bad, SIGKILL, SI_KERNEL, "unknown 42" },
787 { do_bad, SIGKILL, SI_KERNEL, "unknown 43" },
788 { do_bad, SIGKILL, SI_KERNEL, "unknown 44" },
789 { do_bad, SIGKILL, SI_KERNEL, "unknown 45" },
790 { do_bad, SIGKILL, SI_KERNEL, "unknown 46" },
791 { do_bad, SIGKILL, SI_KERNEL, "unknown 47" },
792 { do_bad, SIGKILL, SI_KERNEL, "TLB conflict abort" },
793 { do_bad, SIGKILL, SI_KERNEL, "Unsupported atomic hardware update fault" },
794 { do_bad, SIGKILL, SI_KERNEL, "unknown 50" },
795 { do_bad, SIGKILL, SI_KERNEL, "unknown 51" },
796 { do_bad, SIGKILL, SI_KERNEL, "implementation fault (lockdown abort)" },
797 { do_bad, SIGBUS, BUS_OBJERR, "implementation fault (unsupported exclusive)" },
798 { do_bad, SIGKILL, SI_KERNEL, "unknown 54" },
799 { do_bad, SIGKILL, SI_KERNEL, "unknown 55" },
800 { do_bad, SIGKILL, SI_KERNEL, "unknown 56" },
801 { do_bad, SIGKILL, SI_KERNEL, "unknown 57" },
802 { do_bad, SIGKILL, SI_KERNEL, "unknown 58" },
803 { do_bad, SIGKILL, SI_KERNEL, "unknown 59" },
804 { do_bad, SIGKILL, SI_KERNEL, "unknown 60" },
805 { do_bad, SIGKILL, SI_KERNEL, "section domain fault" },
806 { do_bad, SIGKILL, SI_KERNEL, "page domain fault" },
807 { do_bad, SIGKILL, SI_KERNEL, "unknown 63" },
810 void do_mem_abort(unsigned long far, unsigned int esr, struct pt_regs *regs)
812 const struct fault_info *inf = esr_to_fault_info(esr);
813 unsigned long addr = untagged_addr(far);
815 if (!inf->fn(far, esr, regs))
818 if (!user_mode(regs))
819 die_kernel_fault(inf->name, addr, esr, regs);
822 * At this point we have an unrecognized fault type whose tag bits may
823 * have been defined as UNKNOWN. Therefore we only expose the untagged
824 * address to the signal handler.
826 arm64_notify_die(inf->name, regs, inf->sig, inf->code, addr, esr);
828 NOKPROBE_SYMBOL(do_mem_abort);
830 void do_sp_pc_abort(unsigned long addr, unsigned int esr, struct pt_regs *regs)
832 arm64_notify_die("SP/PC alignment exception", regs, SIGBUS, BUS_ADRALN,
835 NOKPROBE_SYMBOL(do_sp_pc_abort);
837 int __init early_brk64(unsigned long addr, unsigned int esr,
838 struct pt_regs *regs);
841 * __refdata because early_brk64 is __init, but the reference to it is
842 * clobbered at arch_initcall time.
843 * See traps.c and debug-monitors.c:debug_traps_init().
845 static struct fault_info __refdata debug_fault_info[] = {
846 { do_bad, SIGTRAP, TRAP_HWBKPT, "hardware breakpoint" },
847 { do_bad, SIGTRAP, TRAP_HWBKPT, "hardware single-step" },
848 { do_bad, SIGTRAP, TRAP_HWBKPT, "hardware watchpoint" },
849 { do_bad, SIGKILL, SI_KERNEL, "unknown 3" },
850 { do_bad, SIGTRAP, TRAP_BRKPT, "aarch32 BKPT" },
851 { do_bad, SIGKILL, SI_KERNEL, "aarch32 vector catch" },
852 { early_brk64, SIGTRAP, TRAP_BRKPT, "aarch64 BRK" },
853 { do_bad, SIGKILL, SI_KERNEL, "unknown 7" },
856 void __init hook_debug_fault_code(int nr,
857 int (*fn)(unsigned long, unsigned int, struct pt_regs *),
858 int sig, int code, const char *name)
860 BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
862 debug_fault_info[nr].fn = fn;
863 debug_fault_info[nr].sig = sig;
864 debug_fault_info[nr].code = code;
865 debug_fault_info[nr].name = name;
869 * In debug exception context, we explicitly disable preemption despite
870 * having interrupts disabled.
871 * This serves two purposes: it makes it much less likely that we would
872 * accidentally schedule in exception context and it will force a warning
873 * if we somehow manage to schedule by accident.
875 static void debug_exception_enter(struct pt_regs *regs)
879 /* This code is a bit fragile. Test it. */
880 RCU_LOCKDEP_WARN(!rcu_is_watching(), "exception_enter didn't work");
882 NOKPROBE_SYMBOL(debug_exception_enter);
884 static void debug_exception_exit(struct pt_regs *regs)
886 preempt_enable_no_resched();
888 NOKPROBE_SYMBOL(debug_exception_exit);
890 void do_debug_exception(unsigned long addr_if_watchpoint, unsigned int esr,
891 struct pt_regs *regs)
893 const struct fault_info *inf = esr_to_debug_fault_info(esr);
894 unsigned long pc = instruction_pointer(regs);
896 debug_exception_enter(regs);
898 if (user_mode(regs) && !is_ttbr0_addr(pc))
899 arm64_apply_bp_hardening();
901 if (inf->fn(addr_if_watchpoint, esr, regs)) {
902 arm64_notify_die(inf->name, regs, inf->sig, inf->code, pc, esr);
905 debug_exception_exit(regs);
907 NOKPROBE_SYMBOL(do_debug_exception);
910 * Used during anonymous page fault handling.
912 struct page *alloc_zeroed_user_highpage_movable(struct vm_area_struct *vma,
915 gfp_t flags = GFP_HIGHUSER_MOVABLE | __GFP_ZERO;
918 * If the page is mapped with PROT_MTE, initialise the tags at the
919 * point of allocation and page zeroing as this is usually faster than
920 * separate DC ZVA and STGM.
922 if (vma->vm_flags & VM_MTE)
923 flags |= __GFP_ZEROTAGS;
925 return alloc_page_vma(flags, vma, vaddr);
928 void tag_clear_highpage(struct page *page)
930 mte_zero_clear_page_tags(page_address(page));
931 page_kasan_tag_reset(page);
932 set_bit(PG_mte_tagged, &page->flags);