arch/arm64/mm/fault.c

   1 /*
   2  * Based on arch/arm/mm/fault.c
   3  *
   4  * Copyright (C) 1995  Linus Torvalds
   5  * Copyright (C) 1995-2004 Russell King
   6  * Copyright (C) 2012 ARM Ltd.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 as
  10  * published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  19  */
  20
  21 #include <linux/extable.h>
  22 #include <linux/signal.h>
  23 #include <linux/mm.h>
  24 #include <linux/hardirq.h>
  25 #include <linux/init.h>
  26 #include <linux/kprobes.h>
  27 #include <linux/uaccess.h>
  28 #include <linux/page-flags.h>
  29 #include <linux/sched/signal.h>
  30 #include <linux/sched/debug.h>
  31 #include <linux/highmem.h>
  32 #include <linux/perf_event.h>
  33 #include <linux/preempt.h>
  34 #include <linux/hugetlb.h>
  35
  36 #include <asm/bug.h>
  37 #include <asm/cmpxchg.h>
  38 #include <asm/cpufeature.h>
  39 #include <asm/exception.h>
  40 #include <asm/debug-monitors.h>
  41 #include <asm/esr.h>
  42 #include <asm/sysreg.h>
  43 #include <asm/system_misc.h>
  44 #include <asm/pgtable.h>
  45 #include <asm/tlbflush.h>
  46
  47 #include <acpi/ghes.h>
  48
  49 struct fault_info {
  50         int     (*fn)(unsigned long addr, unsigned int esr,
  51                       struct pt_regs *regs);
  52         int     sig;
  53         int     code;
  54         const char *name;
  55 };
  56
  57 static const struct fault_info fault_info[];
  58
  59 static inline const struct fault_info *esr_to_fault_info(unsigned int esr)
  60 {
  61         return fault_info + (esr & 63);
  62 }
  63
  64 #ifdef CONFIG_KPROBES
  65 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
  66 {
  67         int ret = 0;
  68
  69         /* kprobe_running() needs smp_processor_id() */
  70         if (!user_mode(regs)) {
  71                 preempt_disable();
  72                 if (kprobe_running() && kprobe_fault_handler(regs, esr))
  73                         ret = 1;
  74                 preempt_enable();
  75         }
  76
  77         return ret;
  78 }
  79 #else
  80 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
  81 {
  82         return 0;
  83 }
  84 #endif
  85
  86 static void data_abort_decode(unsigned int esr)
  87 {
  88         pr_alert("Data abort info:\n");
  89
  90         if (esr & ESR_ELx_ISV) {
  91                 pr_alert("  Access size = %u byte(s)\n",
  92                          1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
  93                 pr_alert("  SSE = %lu, SRT = %lu\n",
  94                          (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
  95                          (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
  96                 pr_alert("  SF = %lu, AR = %lu\n",
  97                          (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
  98                          (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
  99         } else {
 100                 pr_alert("  ISV = 0, ISS = 0x%08lx\n", esr & ESR_ELx_ISS_MASK);
 101         }
 102
 103         pr_alert("  CM = %lu, WnR = %lu\n",
 104                  (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
 105                  (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT);
 106 }
 107
 108 /*
 109  * Decode mem abort information
 110  */
 111 static void mem_abort_decode(unsigned int esr)
 112 {
 113         pr_alert("Mem abort info:\n");
 114
 115         pr_alert("  Exception class = %s, IL = %u bits\n",
 116                  esr_get_class_string(esr),
 117                  (esr & ESR_ELx_IL) ? 32 : 16);
 118         pr_alert("  SET = %lu, FnV = %lu\n",
 119                  (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
 120                  (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
 121         pr_alert("  EA = %lu, S1PTW = %lu\n",
 122                  (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
 123                  (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
 124
 125         if (esr_is_data_abort(esr))
 126                 data_abort_decode(esr);
 127 }
 128
 129 /*
 130  * Dump out the page tables associated with 'addr' in the currently active mm.
 131  */
 132 void show_pte(unsigned long addr)
 133 {
 134         struct mm_struct *mm;
 135         pgd_t *pgd;
 136
 137         if (addr < TASK_SIZE) {
 138                 /* TTBR0 */
 139                 mm = current->active_mm;
 140                 if (mm == &init_mm) {
 141                         pr_alert("[%016lx] user address but active_mm is swapper\n",
 142                                  addr);
 143                         return;
 144                 }
 145         } else if (addr >= VA_START) {
 146                 /* TTBR1 */
 147                 mm = &init_mm;
 148         } else {
 149                 pr_alert("[%016lx] address between user and kernel address ranges\n",
 150                          addr);
 151                 return;
 152         }
 153
 154         pr_alert("%s pgtable: %luk pages, %u-bit VAs, pgd = %p\n",
 155                  mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
 156                  VA_BITS, mm->pgd);
 157         pgd = pgd_offset(mm, addr);
 158         pr_alert("[%016lx] *pgd=%016llx", addr, pgd_val(*pgd));
 159
 160         do {
 161                 pud_t *pud;
 162                 pmd_t *pmd;
 163                 pte_t *pte;
 164
 165                 if (pgd_none(*pgd) || pgd_bad(*pgd))
 166                         break;
 167
 168                 pud = pud_offset(pgd, addr);
 169                 pr_cont(", *pud=%016llx", pud_val(*pud));
 170                 if (pud_none(*pud) || pud_bad(*pud))
 171                         break;
 172
 173                 pmd = pmd_offset(pud, addr);
 174                 pr_cont(", *pmd=%016llx", pmd_val(*pmd));
 175                 if (pmd_none(*pmd) || pmd_bad(*pmd))
 176                         break;
 177
 178                 pte = pte_offset_map(pmd, addr);
 179                 pr_cont(", *pte=%016llx", pte_val(*pte));
 180                 pte_unmap(pte);
 181         } while(0);
 182
 183         pr_cont("\n");
 184 }
 185
 186 /*
 187  * This function sets the access flags (dirty, accessed), as well as write
 188  * permission, and only to a more permissive setting.
 189  *
 190  * It needs to cope with hardware update of the accessed/dirty state by other
 191  * agents in the system and can safely skip the __sync_icache_dcache() call as,
 192  * like set_pte_at(), the PTE is never changed from no-exec to exec here.
 193  *
 194  * Returns whether or not the PTE actually changed.
 195  */
 196 int ptep_set_access_flags(struct vm_area_struct *vma,
 197                           unsigned long address, pte_t *ptep,
 198                           pte_t entry, int dirty)
 199 {
 200         pteval_t old_pteval, pteval;
 201
 202         if (pte_same(*ptep, entry))
 203                 return 0;
 204
 205         /* only preserve the access flags and write permission */
 206         pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
 207
 208         /*
 209          * Setting the flags must be done atomically to avoid racing with the
 210          * hardware update of the access/dirty state. The PTE_RDONLY bit must
 211          * be set to the most permissive (lowest value) of *ptep and entry
 212          * (calculated as: a & b == ~(~a | ~b)).
 213          */
 214         pte_val(entry) ^= PTE_RDONLY;
 215         pteval = READ_ONCE(pte_val(*ptep));
 216         do {
 217                 old_pteval = pteval;
 218                 pteval ^= PTE_RDONLY;
 219                 pteval |= pte_val(entry);
 220                 pteval ^= PTE_RDONLY;
 221                 pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
 222         } while (pteval != old_pteval);
 223
 224         flush_tlb_fix_spurious_fault(vma, address);
 225         return 1;
 226 }
 227
 228 static bool is_el1_instruction_abort(unsigned int esr)
 229 {
 230         return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
 231 }
 232
 233 static inline bool is_permission_fault(unsigned int esr, struct pt_regs *regs,
 234                                        unsigned long addr)
 235 {
 236         unsigned int ec       = ESR_ELx_EC(esr);
 237         unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
 238
 239         if (ec != ESR_ELx_EC_DABT_CUR && ec != ESR_ELx_EC_IABT_CUR)
 240                 return false;
 241
 242         if (fsc_type == ESR_ELx_FSC_PERM)
 243                 return true;
 244
 245         if (addr < USER_DS && system_uses_ttbr0_pan())
 246                 return fsc_type == ESR_ELx_FSC_FAULT &&
 247                         (regs->pstate & PSR_PAN_BIT);
 248
 249         return false;
 250 }
 251
 252 /*
 253  * The kernel tried to access some page that wasn't present.
 254  */
 255 static void __do_kernel_fault(unsigned long addr, unsigned int esr,
 256                               struct pt_regs *regs)
 257 {
 258         const char *msg;
 259
 260         /*
 261          * Are we prepared to handle this kernel fault?
 262          * We are almost certainly not prepared to handle instruction faults.
 263          */
 264         if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
 265                 return;
 266
 267         /*
 268          * No handler, we'll have to terminate things with extreme prejudice.
 269          */
 270         bust_spinlocks(1);
 271
 272         if (is_permission_fault(esr, regs, addr)) {
 273                 if (esr & ESR_ELx_WNR)
 274                         msg = "write to read-only memory";
 275                 else
 276                         msg = "read from unreadable memory";
 277         } else if (addr < PAGE_SIZE) {
 278                 msg = "NULL pointer dereference";
 279         } else {
 280                 msg = "paging request";
 281         }
 282
 283         pr_alert("Unable to handle kernel %s at virtual address %08lx\n", msg,
 284                  addr);
 285
 286         mem_abort_decode(esr);
 287
 288         show_pte(addr);
 289         die("Oops", regs, esr);
 290         bust_spinlocks(0);
 291         do_exit(SIGKILL);
 292 }
 293
 294 /*
 295  * Something tried to access memory that isn't in our memory map. User mode
 296  * accesses just cause a SIGSEGV
 297  */
 298 static void __do_user_fault(struct task_struct *tsk, unsigned long addr,
 299                             unsigned int esr, unsigned int sig, int code,
 300                             struct pt_regs *regs, int fault)
 301 {
 302         struct siginfo si;
 303         const struct fault_info *inf;
 304         unsigned int lsb = 0;
 305
 306         if (unhandled_signal(tsk, sig) && show_unhandled_signals_ratelimited()) {
 307                 inf = esr_to_fault_info(esr);
 308                 pr_info("%s[%d]: unhandled %s (%d) at 0x%08lx, esr 0x%03x",
 309                         tsk->comm, task_pid_nr(tsk), inf->name, sig,
 310                         addr, esr);
 311                 print_vma_addr(KERN_CONT ", in ", regs->pc);
 312                 pr_cont("\n");
 313                 __show_regs(regs);
 314         }
 315
 316         tsk->thread.fault_address = addr;
 317         tsk->thread.fault_code = esr;
 318         si.si_signo = sig;
 319         si.si_errno = 0;
 320         si.si_code = code;
 321         si.si_addr = (void __user *)addr;
 322         /*
 323          * Either small page or large page may be poisoned.
 324          * In other words, VM_FAULT_HWPOISON_LARGE and
 325          * VM_FAULT_HWPOISON are mutually exclusive.
 326          */
 327         if (fault & VM_FAULT_HWPOISON_LARGE)
 328                 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
 329         else if (fault & VM_FAULT_HWPOISON)
 330                 lsb = PAGE_SHIFT;
 331         si.si_addr_lsb = lsb;
 332
 333         force_sig_info(sig, &si, tsk);
 334 }
 335
 336 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
 337 {
 338         struct task_struct *tsk = current;
 339         const struct fault_info *inf;
 340
 341         /*
 342          * If we are in kernel mode at this point, we have no context to
 343          * handle this fault with.
 344          */
 345         if (user_mode(regs)) {
 346                 inf = esr_to_fault_info(esr);
 347                 __do_user_fault(tsk, addr, esr, inf->sig, inf->code, regs, 0);
 348         } else
 349                 __do_kernel_fault(addr, esr, regs);
 350 }
 351
 352 #define VM_FAULT_BADMAP         0x010000
 353 #define VM_FAULT_BADACCESS      0x020000
 354
 355 static int __do_page_fault(struct mm_struct *mm, unsigned long addr,
 356                            unsigned int mm_flags, unsigned long vm_flags,
 357                            struct task_struct *tsk)
 358 {
 359         struct vm_area_struct *vma;
 360         int fault;
 361
 362         vma = find_vma(mm, addr);
 363         fault = VM_FAULT_BADMAP;
 364         if (unlikely(!vma))
 365                 goto out;
 366         if (unlikely(vma->vm_start > addr))
 367                 goto check_stack;
 368
 369         /*
 370          * Ok, we have a good vm_area for this memory access, so we can handle
 371          * it.
 372          */
 373 good_area:
 374         /*
 375          * Check that the permissions on the VMA allow for the fault which
 376          * occurred.
 377          */
 378         if (!(vma->vm_flags & vm_flags)) {
 379                 fault = VM_FAULT_BADACCESS;
 380                 goto out;
 381         }
 382
 383         return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags);
 384
 385 check_stack:
 386         if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
 387                 goto good_area;
 388 out:
 389         return fault;
 390 }
 391
 392 static bool is_el0_instruction_abort(unsigned int esr)
 393 {
 394         return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
 395 }
 396
 397 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
 398                                    struct pt_regs *regs)
 399 {
 400         struct task_struct *tsk;
 401         struct mm_struct *mm;
 402         int fault, sig, code, major = 0;
 403         unsigned long vm_flags = VM_READ | VM_WRITE;
 404         unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 405
 406         if (notify_page_fault(regs, esr))
 407                 return 0;
 408
 409         tsk = current;
 410         mm  = tsk->mm;
 411
 412         /*
 413          * If we're in an interrupt or have no user context, we must not take
 414          * the fault.
 415          */
 416         if (faulthandler_disabled() || !mm)
 417                 goto no_context;
 418
 419         if (user_mode(regs))
 420                 mm_flags |= FAULT_FLAG_USER;
 421
 422         if (is_el0_instruction_abort(esr)) {
 423                 vm_flags = VM_EXEC;
 424         } else if ((esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM)) {
 425                 vm_flags = VM_WRITE;
 426                 mm_flags |= FAULT_FLAG_WRITE;
 427         }
 428
 429         if (addr < USER_DS && is_permission_fault(esr, regs, addr)) {
 430                 /* regs->orig_addr_limit may be 0 if we entered from EL0 */
 431                 if (regs->orig_addr_limit == KERNEL_DS)
 432                         die("Accessing user space memory with fs=KERNEL_DS", regs, esr);
 433
 434                 if (is_el1_instruction_abort(esr))
 435                         die("Attempting to execute userspace memory", regs, esr);
 436
 437                 if (!search_exception_tables(regs->pc))
 438                         die("Accessing user space memory outside uaccess.h routines", regs, esr);
 439         }
 440
 441         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
 442
 443         /*
 444          * As per x86, we may deadlock here. However, since the kernel only
 445          * validly references user space from well defined areas of the code,
 446          * we can bug out early if this is from code which shouldn't.
 447          */
 448         if (!down_read_trylock(&mm->mmap_sem)) {
 449                 if (!user_mode(regs) && !search_exception_tables(regs->pc))
 450                         goto no_context;
 451 retry:
 452                 down_read(&mm->mmap_sem);
 453         } else {
 454                 /*
 455                  * The above down_read_trylock() might have succeeded in which
 456                  * case, we'll have missed the might_sleep() from down_read().
 457                  */
 458                 might_sleep();
 459 #ifdef CONFIG_DEBUG_VM
 460                 if (!user_mode(regs) && !search_exception_tables(regs->pc))
 461                         goto no_context;
 462 #endif
 463         }
 464
 465         fault = __do_page_fault(mm, addr, mm_flags, vm_flags, tsk);
 466         major |= fault & VM_FAULT_MAJOR;
 467
 468         if (fault & VM_FAULT_RETRY) {
 469                 /*
 470                  * If we need to retry but a fatal signal is pending,
 471                  * handle the signal first. We do not need to release
 472                  * the mmap_sem because it would already be released
 473                  * in __lock_page_or_retry in mm/filemap.c.
 474                  */
 475                 if (fatal_signal_pending(current)) {
 476                         if (!user_mode(regs))
 477                                 goto no_context;
 478                         return 0;
 479                 }
 480
 481                 /*
 482                  * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
 483                  * starvation.
 484                  */
 485                 if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
 486                         mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
 487                         mm_flags |= FAULT_FLAG_TRIED;
 488                         goto retry;
 489                 }
 490         }
 491         up_read(&mm->mmap_sem);
 492
 493         /*
 494          * Handle the "normal" (no error) case first.
 495          */
 496         if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
 497                               VM_FAULT_BADACCESS)))) {
 498                 /*
 499                  * Major/minor page fault accounting is only done
 500                  * once. If we go through a retry, it is extremely
 501                  * likely that the page will be found in page cache at
 502                  * that point.
 503                  */
 504                 if (major) {
 505                         tsk->maj_flt++;
 506                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
 507                                       addr);
 508                 } else {
 509                         tsk->min_flt++;
 510                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
 511                                       addr);
 512                 }
 513
 514                 return 0;
 515         }
 516
 517         /*
 518          * If we are in kernel mode at this point, we have no context to
 519          * handle this fault with.
 520          */
 521         if (!user_mode(regs))
 522                 goto no_context;
 523
 524         if (fault & VM_FAULT_OOM) {
 525                 /*
 526                  * We ran out of memory, call the OOM killer, and return to
 527                  * userspace (which will retry the fault, or kill us if we got
 528                  * oom-killed).
 529                  */
 530                 pagefault_out_of_memory();
 531                 return 0;
 532         }
 533
 534         if (fault & VM_FAULT_SIGBUS) {
 535                 /*
 536                  * We had some memory, but were unable to successfully fix up
 537                  * this page fault.
 538                  */
 539                 sig = SIGBUS;
 540                 code = BUS_ADRERR;
 541         } else if (fault & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE)) {
 542                 sig = SIGBUS;
 543                 code = BUS_MCEERR_AR;
 544         } else {
 545                 /*
 546                  * Something tried to access memory that isn't in our memory
 547                  * map.
 548                  */
 549                 sig = SIGSEGV;
 550                 code = fault == VM_FAULT_BADACCESS ?
 551                         SEGV_ACCERR : SEGV_MAPERR;
 552         }
 553
 554         __do_user_fault(tsk, addr, esr, sig, code, regs, fault);
 555         return 0;
 556
 557 no_context:
 558         __do_kernel_fault(addr, esr, regs);
 559         return 0;
 560 }
 561
 562 /*
 563  * First Level Translation Fault Handler
 564  *
 565  * We enter here because the first level page table doesn't contain a valid
 566  * entry for the address.
 567  *
 568  * If the address is in kernel space (>= TASK_SIZE), then we are probably
 569  * faulting in the vmalloc() area.
 570  *
 571  * If the init_task's first level page tables contains the relevant entry, we
 572  * copy the it to this task.  If not, we send the process a signal, fixup the
 573  * exception, or oops the kernel.
 574  *
 575  * NOTE! We MUST NOT take any locks for this case. We may be in an interrupt
 576  * or a critical region, and should only copy the information from the master
 577  * page table, nothing more.
 578  */
 579 static int __kprobes do_translation_fault(unsigned long addr,
 580                                           unsigned int esr,
 581                                           struct pt_regs *regs)
 582 {
 583         if (addr < TASK_SIZE)
 584                 return do_page_fault(addr, esr, regs);
 585
 586         do_bad_area(addr, esr, regs);
 587         return 0;
 588 }
 589
 590 static int do_alignment_fault(unsigned long addr, unsigned int esr,
 591                               struct pt_regs *regs)
 592 {
 593         do_bad_area(addr, esr, regs);
 594         return 0;
 595 }
 596
 597 /*
 598  * This abort handler always returns "fault".
 599  */
 600 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
 601 {
 602         return 1;
 603 }
 604
 605 /*
 606  * This abort handler deals with Synchronous External Abort.
 607  * It calls notifiers, and then returns "fault".
 608  */
 609 static int do_sea(unsigned long addr, unsigned int esr, struct pt_regs *regs)
 610 {
 611         struct siginfo info;
 612         const struct fault_info *inf;
 613         int ret = 0;
 614
 615         inf = esr_to_fault_info(esr);
 616         pr_err("Synchronous External Abort: %s (0x%08x) at 0x%016lx\n",
 617                 inf->name, esr, addr);
 618
 619         /*
 620          * Synchronous aborts may interrupt code which had interrupts masked.
 621          * Before calling out into the wider kernel tell the interested
 622          * subsystems.
 623          */
 624         if (IS_ENABLED(CONFIG_ACPI_APEI_SEA)) {
 625                 if (interrupts_enabled(regs))
 626                         nmi_enter();
 627
 628                 ret = ghes_notify_sea();
 629
 630                 if (interrupts_enabled(regs))
 631                         nmi_exit();
 632         }
 633
 634         info.si_signo = SIGBUS;
 635         info.si_errno = 0;
 636         info.si_code  = 0;
 637         if (esr & ESR_ELx_FnV)
 638                 info.si_addr = NULL;
 639         else
 640                 info.si_addr  = (void __user *)addr;
 641         arm64_notify_die("", regs, &info, esr);
 642
 643         return ret;
 644 }
 645
 646 static const struct fault_info fault_info[] = {
 647         { do_bad,               SIGBUS,  0,             "ttbr address size fault"       },
 648         { do_bad,               SIGBUS,  0,             "level 1 address size fault"    },
 649         { do_bad,               SIGBUS,  0,             "level 2 address size fault"    },
 650         { do_bad,               SIGBUS,  0,             "level 3 address size fault"    },
 651         { do_translation_fault, SIGSEGV, SEGV_MAPERR,   "level 0 translation fault"     },
 652         { do_translation_fault, SIGSEGV, SEGV_MAPERR,   "level 1 translation fault"     },
 653         { do_translation_fault, SIGSEGV, SEGV_MAPERR,   "level 2 translation fault"     },
 654         { do_translation_fault, SIGSEGV, SEGV_MAPERR,   "level 3 translation fault"     },
 655         { do_bad,               SIGBUS,  0,             "unknown 8"                     },
 656         { do_page_fault,        SIGSEGV, SEGV_ACCERR,   "level 1 access flag fault"     },
 657         { do_page_fault,        SIGSEGV, SEGV_ACCERR,   "level 2 access flag fault"     },
 658         { do_page_fault,        SIGSEGV, SEGV_ACCERR,   "level 3 access flag fault"     },
 659         { do_bad,               SIGBUS,  0,             "unknown 12"                    },
 660         { do_page_fault,        SIGSEGV, SEGV_ACCERR,   "level 1 permission fault"      },
 661         { do_page_fault,        SIGSEGV, SEGV_ACCERR,   "level 2 permission fault"      },
 662         { do_page_fault,        SIGSEGV, SEGV_ACCERR,   "level 3 permission fault"      },
 663         { do_sea,               SIGBUS,  0,             "synchronous external abort"    },
 664         { do_bad,               SIGBUS,  0,             "unknown 17"                    },
 665         { do_bad,               SIGBUS,  0,             "unknown 18"                    },
 666         { do_bad,               SIGBUS,  0,             "unknown 19"                    },
 667         { do_sea,               SIGBUS,  0,             "level 0 (translation table walk)"      },
 668         { do_sea,               SIGBUS,  0,             "level 1 (translation table walk)"      },
 669         { do_sea,               SIGBUS,  0,             "level 2 (translation table walk)"      },
 670         { do_sea,               SIGBUS,  0,             "level 3 (translation table walk)"      },
 671         { do_sea,               SIGBUS,  0,             "synchronous parity or ECC error" },
 672         { do_bad,               SIGBUS,  0,             "unknown 25"                    },
 673         { do_bad,               SIGBUS,  0,             "unknown 26"                    },
 674         { do_bad,               SIGBUS,  0,             "unknown 27"                    },
 675         { do_sea,               SIGBUS,  0,             "level 0 synchronous parity error (translation table walk)"     },
 676         { do_sea,               SIGBUS,  0,             "level 1 synchronous parity error (translation table walk)"     },
 677         { do_sea,               SIGBUS,  0,             "level 2 synchronous parity error (translation table walk)"     },
 678         { do_sea,               SIGBUS,  0,             "level 3 synchronous parity error (translation table walk)"     },
 679         { do_bad,               SIGBUS,  0,             "unknown 32"                    },
 680         { do_alignment_fault,   SIGBUS,  BUS_ADRALN,    "alignment fault"               },
 681         { do_bad,               SIGBUS,  0,             "unknown 34"                    },
 682         { do_bad,               SIGBUS,  0,             "unknown 35"                    },
 683         { do_bad,               SIGBUS,  0,             "unknown 36"                    },
 684         { do_bad,               SIGBUS,  0,             "unknown 37"                    },
 685         { do_bad,               SIGBUS,  0,             "unknown 38"                    },
 686         { do_bad,               SIGBUS,  0,             "unknown 39"                    },
 687         { do_bad,               SIGBUS,  0,             "unknown 40"                    },
 688         { do_bad,               SIGBUS,  0,             "unknown 41"                    },
 689         { do_bad,               SIGBUS,  0,             "unknown 42"                    },
 690         { do_bad,               SIGBUS,  0,             "unknown 43"                    },
 691         { do_bad,               SIGBUS,  0,             "unknown 44"                    },
 692         { do_bad,               SIGBUS,  0,             "unknown 45"                    },
 693         { do_bad,               SIGBUS,  0,             "unknown 46"                    },
 694         { do_bad,               SIGBUS,  0,             "unknown 47"                    },
 695         { do_bad,               SIGBUS,  0,             "TLB conflict abort"            },
 696         { do_bad,               SIGBUS,  0,             "unknown 49"                    },
 697         { do_bad,               SIGBUS,  0,             "unknown 50"                    },
 698         { do_bad,               SIGBUS,  0,             "unknown 51"                    },
 699         { do_bad,               SIGBUS,  0,             "implementation fault (lockdown abort)" },
 700         { do_bad,               SIGBUS,  0,             "implementation fault (unsupported exclusive)" },
 701         { do_bad,               SIGBUS,  0,             "unknown 54"                    },
 702         { do_bad,               SIGBUS,  0,             "unknown 55"                    },
 703         { do_bad,               SIGBUS,  0,             "unknown 56"                    },
 704         { do_bad,               SIGBUS,  0,             "unknown 57"                    },
 705         { do_bad,               SIGBUS,  0,             "unknown 58"                    },
 706         { do_bad,               SIGBUS,  0,             "unknown 59"                    },
 707         { do_bad,               SIGBUS,  0,             "unknown 60"                    },
 708         { do_bad,               SIGBUS,  0,             "section domain fault"          },
 709         { do_bad,               SIGBUS,  0,             "page domain fault"             },
 710         { do_bad,               SIGBUS,  0,             "unknown 63"                    },
 711 };
 712
 713 /*
 714  * Handle Synchronous External Aborts that occur in a guest kernel.
 715  *
 716  * The return value will be zero if the SEA was successfully handled
 717  * and non-zero if there was an error processing the error or there was
 718  * no error to process.
 719  */
 720 int handle_guest_sea(phys_addr_t addr, unsigned int esr)
 721 {
 722         int ret = -ENOENT;
 723
 724         if (IS_ENABLED(CONFIG_ACPI_APEI_SEA))
 725                 ret = ghes_notify_sea();
 726
 727         return ret;
 728 }
 729
 730 /*
 731  * Dispatch a data abort to the relevant handler.
 732  */
 733 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
 734                                          struct pt_regs *regs)
 735 {
 736         const struct fault_info *inf = esr_to_fault_info(esr);
 737         struct siginfo info;
 738
 739         if (!inf->fn(addr, esr, regs))
 740                 return;
 741
 742         pr_alert("Unhandled fault: %s (0x%08x) at 0x%016lx\n",
 743                  inf->name, esr, addr);
 744
 745         mem_abort_decode(esr);
 746
 747         info.si_signo = inf->sig;
 748         info.si_errno = 0;
 749         info.si_code  = inf->code;
 750         info.si_addr  = (void __user *)addr;
 751         arm64_notify_die("", regs, &info, esr);
 752 }
 753
 754 /*
 755  * Handle stack alignment exceptions.
 756  */
 757 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
 758                                            unsigned int esr,
 759                                            struct pt_regs *regs)
 760 {
 761         struct siginfo info;
 762         struct task_struct *tsk = current;
 763
 764         if (show_unhandled_signals && unhandled_signal(tsk, SIGBUS))
 765                 pr_info_ratelimited("%s[%d]: %s exception: pc=%p sp=%p\n",
 766                                     tsk->comm, task_pid_nr(tsk),
 767                                     esr_get_class_string(esr), (void *)regs->pc,
 768                                     (void *)regs->sp);
 769
 770         info.si_signo = SIGBUS;
 771         info.si_errno = 0;
 772         info.si_code  = BUS_ADRALN;
 773         info.si_addr  = (void __user *)addr;
 774         arm64_notify_die("Oops - SP/PC alignment exception", regs, &info, esr);
 775 }
 776
 777 int __init early_brk64(unsigned long addr, unsigned int esr,
 778                        struct pt_regs *regs);
 779
 780 /*
 781  * __refdata because early_brk64 is __init, but the reference to it is
 782  * clobbered at arch_initcall time.
 783  * See traps.c and debug-monitors.c:debug_traps_init().
 784  */
 785 static struct fault_info __refdata debug_fault_info[] = {
 786         { do_bad,       SIGTRAP,        TRAP_HWBKPT,    "hardware breakpoint"   },
 787         { do_bad,       SIGTRAP,        TRAP_HWBKPT,    "hardware single-step"  },
 788         { do_bad,       SIGTRAP,        TRAP_HWBKPT,    "hardware watchpoint"   },
 789         { do_bad,       SIGBUS,         0,              "unknown 3"             },
 790         { do_bad,       SIGTRAP,        TRAP_BRKPT,     "aarch32 BKPT"          },
 791         { do_bad,       SIGTRAP,        0,              "aarch32 vector catch"  },
 792         { early_brk64,  SIGTRAP,        TRAP_BRKPT,     "aarch64 BRK"           },
 793         { do_bad,       SIGBUS,         0,              "unknown 7"             },
 794 };
 795
 796 void __init hook_debug_fault_code(int nr,
 797                                   int (*fn)(unsigned long, unsigned int, struct pt_regs *),
 798                                   int sig, int code, const char *name)
 799 {
 800         BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
 801
 802         debug_fault_info[nr].fn         = fn;
 803         debug_fault_info[nr].sig        = sig;
 804         debug_fault_info[nr].code       = code;
 805         debug_fault_info[nr].name       = name;
 806 }
 807
 808 asmlinkage int __exception do_debug_exception(unsigned long addr,
 809                                               unsigned int esr,
 810                                               struct pt_regs *regs)
 811 {
 812         const struct fault_info *inf = debug_fault_info + DBG_ESR_EVT(esr);
 813         struct siginfo info;
 814         int rv;
 815
 816         /*
 817          * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
 818          * already disabled to preserve the last enabled/disabled addresses.
 819          */
 820         if (interrupts_enabled(regs))
 821                 trace_hardirqs_off();
 822
 823         if (!inf->fn(addr, esr, regs)) {
 824                 rv = 1;
 825         } else {
 826                 pr_alert("Unhandled debug exception: %s (0x%08x) at 0x%016lx\n",
 827                          inf->name, esr, addr);
 828
 829                 info.si_signo = inf->sig;
 830                 info.si_errno = 0;
 831                 info.si_code  = inf->code;
 832                 info.si_addr  = (void __user *)addr;
 833                 arm64_notify_die("", regs, &info, 0);
 834                 rv = 0;
 835         }
 836
 837         if (interrupts_enabled(regs))
 838                 trace_hardirqs_on();
 839
 840         return rv;
 841 }
 842 NOKPROBE_SYMBOL(do_debug_exception);
 843
 844 #ifdef CONFIG_ARM64_PAN
 845 int cpu_enable_pan(void *__unused)
 846 {
 847         /*
 848          * We modify PSTATE. This won't work from irq context as the PSTATE
 849          * is discarded once we return from the exception.
 850          */
 851         WARN_ON_ONCE(in_interrupt());
 852
 853         config_sctlr_el1(SCTLR_EL1_SPAN, 0);
 854         asm(SET_PSTATE_PAN(1));
 855         return 0;
 856 }
 857 #endif /* CONFIG_ARM64_PAN */