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