arch/arm/mm/fault.c

   1 /*
   2  *  linux/arch/arm/mm/fault.c
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *  Modifications for ARM processor (c) 1995-2004 Russell King
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/extable.h>
  12 #include <linux/signal.h>
  13 #include <linux/mm.h>
  14 #include <linux/hardirq.h>
  15 #include <linux/init.h>
  16 #include <linux/kprobes.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/page-flags.h>
  19 #include <linux/sched/signal.h>
  20 #include <linux/sched/debug.h>
  21 #include <linux/highmem.h>
  22 #include <linux/perf_event.h>
  23
  24 #include <asm/exception.h>
  25 #include <asm/pgtable.h>
  26 #include <asm/system_misc.h>
  27 #include <asm/system_info.h>
  28 #include <asm/tlbflush.h>
  29
  30 #include "fault.h"
  31
  32 #ifdef CONFIG_MMU
  33
  34 #ifdef CONFIG_KPROBES
  35 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
  36 {
  37         int ret = 0;
  38
  39         if (!user_mode(regs)) {
  40                 /* kprobe_running() needs smp_processor_id() */
  41                 preempt_disable();
  42                 if (kprobe_running() && kprobe_fault_handler(regs, fsr))
  43                         ret = 1;
  44                 preempt_enable();
  45         }
  46
  47         return ret;
  48 }
  49 #else
  50 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
  51 {
  52         return 0;
  53 }
  54 #endif
  55
  56 /*
  57  * This is useful to dump out the page tables associated with
  58  * 'addr' in mm 'mm'.
  59  */
  60 void show_pte(struct mm_struct *mm, unsigned long addr)
  61 {
  62         pgd_t *pgd;
  63
  64         if (!mm)
  65                 mm = &init_mm;
  66
  67         pr_alert("pgd = %p\n", mm->pgd);
  68         pgd = pgd_offset(mm, addr);
  69         pr_alert("[%08lx] *pgd=%08llx",
  70                         addr, (long long)pgd_val(*pgd));
  71
  72         do {
  73                 pud_t *pud;
  74                 pmd_t *pmd;
  75                 pte_t *pte;
  76
  77                 if (pgd_none(*pgd))
  78                         break;
  79
  80                 if (pgd_bad(*pgd)) {
  81                         pr_cont("(bad)");
  82                         break;
  83                 }
  84
  85                 pud = pud_offset(pgd, addr);
  86                 if (PTRS_PER_PUD != 1)
  87                         pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
  88
  89                 if (pud_none(*pud))
  90                         break;
  91
  92                 if (pud_bad(*pud)) {
  93                         pr_cont("(bad)");
  94                         break;
  95                 }
  96
  97                 pmd = pmd_offset(pud, addr);
  98                 if (PTRS_PER_PMD != 1)
  99                         pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
 100
 101                 if (pmd_none(*pmd))
 102                         break;
 103
 104                 if (pmd_bad(*pmd)) {
 105                         pr_cont("(bad)");
 106                         break;
 107                 }
 108
 109                 /* We must not map this if we have highmem enabled */
 110                 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
 111                         break;
 112
 113                 pte = pte_offset_map(pmd, addr);
 114                 pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
 115 #ifndef CONFIG_ARM_LPAE
 116                 pr_cont(", *ppte=%08llx",
 117                        (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
 118 #endif
 119                 pte_unmap(pte);
 120         } while(0);
 121
 122         pr_cont("\n");
 123 }
 124 #else                                   /* CONFIG_MMU */
 125 void show_pte(struct mm_struct *mm, unsigned long addr)
 126 { }
 127 #endif                                  /* CONFIG_MMU */
 128
 129 /*
 130  * Oops.  The kernel tried to access some page that wasn't present.
 131  */
 132 static void
 133 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
 134                   struct pt_regs *regs)
 135 {
 136         /*
 137          * Are we prepared to handle this kernel fault?
 138          */
 139         if (fixup_exception(regs))
 140                 return;
 141
 142         /*
 143          * No handler, we'll have to terminate things with extreme prejudice.
 144          */
 145         bust_spinlocks(1);
 146         pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
 147                  (addr < PAGE_SIZE) ? "NULL pointer dereference" :
 148                  "paging request", addr);
 149
 150         show_pte(mm, addr);
 151         die("Oops", regs, fsr);
 152         bust_spinlocks(0);
 153         do_exit(SIGKILL);
 154 }
 155
 156 /*
 157  * Something tried to access memory that isn't in our memory map..
 158  * User mode accesses just cause a SIGSEGV
 159  */
 160 static void
 161 __do_user_fault(struct task_struct *tsk, unsigned long addr,
 162                 unsigned int fsr, unsigned int sig, int code,
 163                 struct pt_regs *regs)
 164 {
 165         struct siginfo si;
 166
 167 #ifdef CONFIG_DEBUG_USER
 168         if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
 169             ((user_debug & UDBG_BUS)  && (sig == SIGBUS))) {
 170                 printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
 171                        tsk->comm, sig, addr, fsr);
 172                 show_pte(tsk->mm, addr);
 173                 show_regs(regs);
 174         }
 175 #endif
 176
 177         tsk->thread.address = addr;
 178         tsk->thread.error_code = fsr;
 179         tsk->thread.trap_no = 14;
 180         si.si_signo = sig;
 181         si.si_errno = 0;
 182         si.si_code = code;
 183         si.si_addr = (void __user *)addr;
 184         force_sig_info(sig, &si, tsk);
 185 }
 186
 187 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 188 {
 189         struct task_struct *tsk = current;
 190         struct mm_struct *mm = tsk->active_mm;
 191
 192         /*
 193          * If we are in kernel mode at this point, we
 194          * have no context to handle this fault with.
 195          */
 196         if (user_mode(regs))
 197                 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
 198         else
 199                 __do_kernel_fault(mm, addr, fsr, regs);
 200 }
 201
 202 #ifdef CONFIG_MMU
 203 #define VM_FAULT_BADMAP         0x010000
 204 #define VM_FAULT_BADACCESS      0x020000
 205
 206 /*
 207  * Check that the permissions on the VMA allow for the fault which occurred.
 208  * If we encountered a write fault, we must have write permission, otherwise
 209  * we allow any permission.
 210  */
 211 static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
 212 {
 213         unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
 214
 215         if (fsr & FSR_WRITE)
 216                 mask = VM_WRITE;
 217         if (fsr & FSR_LNX_PF)
 218                 mask = VM_EXEC;
 219
 220         return vma->vm_flags & mask ? false : true;
 221 }
 222
 223 static int __kprobes
 224 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
 225                 unsigned int flags, struct task_struct *tsk)
 226 {
 227         struct vm_area_struct *vma;
 228         int fault;
 229
 230         vma = find_vma(mm, addr);
 231         fault = VM_FAULT_BADMAP;
 232         if (unlikely(!vma))
 233                 goto out;
 234         if (unlikely(vma->vm_start > addr))
 235                 goto check_stack;
 236
 237         /*
 238          * Ok, we have a good vm_area for this
 239          * memory access, so we can handle it.
 240          */
 241 good_area:
 242         if (access_error(fsr, vma)) {
 243                 fault = VM_FAULT_BADACCESS;
 244                 goto out;
 245         }
 246
 247         return handle_mm_fault(vma, addr & PAGE_MASK, flags);
 248
 249 check_stack:
 250         /* Don't allow expansion below FIRST_USER_ADDRESS */
 251         if (vma->vm_flags & VM_GROWSDOWN &&
 252             addr >= FIRST_USER_ADDRESS && !expand_stack(vma, addr))
 253                 goto good_area;
 254 out:
 255         return fault;
 256 }
 257
 258 static int __kprobes
 259 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 260 {
 261         struct task_struct *tsk;
 262         struct mm_struct *mm;
 263         int fault, sig, code;
 264         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 265
 266         if (notify_page_fault(regs, fsr))
 267                 return 0;
 268
 269         tsk = current;
 270         mm  = tsk->mm;
 271
 272         /* Enable interrupts if they were enabled in the parent context. */
 273         if (interrupts_enabled(regs))
 274                 local_irq_enable();
 275
 276         /*
 277          * If we're in an interrupt or have no user
 278          * context, we must not take the fault..
 279          */
 280         if (faulthandler_disabled() || !mm)
 281                 goto no_context;
 282
 283         if (user_mode(regs))
 284                 flags |= FAULT_FLAG_USER;
 285         if (fsr & FSR_WRITE)
 286                 flags |= FAULT_FLAG_WRITE;
 287
 288         /*
 289          * As per x86, we may deadlock here.  However, since the kernel only
 290          * validly references user space from well defined areas of the code,
 291          * we can bug out early if this is from code which shouldn't.
 292          */
 293         if (!down_read_trylock(&mm->mmap_sem)) {
 294                 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
 295                         goto no_context;
 296 retry:
 297                 down_read(&mm->mmap_sem);
 298         } else {
 299                 /*
 300                  * The above down_read_trylock() might have succeeded in
 301                  * which case, we'll have missed the might_sleep() from
 302                  * down_read()
 303                  */
 304                 might_sleep();
 305 #ifdef CONFIG_DEBUG_VM
 306                 if (!user_mode(regs) &&
 307                     !search_exception_tables(regs->ARM_pc))
 308                         goto no_context;
 309 #endif
 310         }
 311
 312         fault = __do_page_fault(mm, addr, fsr, flags, tsk);
 313
 314         /* If we need to retry but a fatal signal is pending, handle the
 315          * signal first. We do not need to release the mmap_sem because
 316          * it would already be released in __lock_page_or_retry in
 317          * mm/filemap.c. */
 318         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 319                 return 0;
 320
 321         /*
 322          * Major/minor page fault accounting is only done on the
 323          * initial attempt. If we go through a retry, it is extremely
 324          * likely that the page will be found in page cache at that point.
 325          */
 326
 327         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
 328         if (!(fault & VM_FAULT_ERROR) && flags & FAULT_FLAG_ALLOW_RETRY) {
 329                 if (fault & VM_FAULT_MAJOR) {
 330                         tsk->maj_flt++;
 331                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
 332                                         regs, addr);
 333                 } else {
 334                         tsk->min_flt++;
 335                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
 336                                         regs, addr);
 337                 }
 338                 if (fault & VM_FAULT_RETRY) {
 339                         /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
 340                         * of starvation. */
 341                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 342                         flags |= FAULT_FLAG_TRIED;
 343                         goto retry;
 344                 }
 345         }
 346
 347         up_read(&mm->mmap_sem);
 348
 349         /*
 350          * Handle the "normal" case first - VM_FAULT_MAJOR
 351          */
 352         if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
 353                 return 0;
 354
 355         /*
 356          * If we are in kernel mode at this point, we
 357          * have no context to handle this fault with.
 358          */
 359         if (!user_mode(regs))
 360                 goto no_context;
 361
 362         if (fault & VM_FAULT_OOM) {
 363                 /*
 364                  * We ran out of memory, call the OOM killer, and return to
 365                  * userspace (which will retry the fault, or kill us if we
 366                  * got oom-killed)
 367                  */
 368                 pagefault_out_of_memory();
 369                 return 0;
 370         }
 371
 372         if (fault & VM_FAULT_SIGBUS) {
 373                 /*
 374                  * We had some memory, but were unable to
 375                  * successfully fix up this page fault.
 376                  */
 377                 sig = SIGBUS;
 378                 code = BUS_ADRERR;
 379         } else {
 380                 /*
 381                  * Something tried to access memory that
 382                  * isn't in our memory map..
 383                  */
 384                 sig = SIGSEGV;
 385                 code = fault == VM_FAULT_BADACCESS ?
 386                         SEGV_ACCERR : SEGV_MAPERR;
 387         }
 388
 389         __do_user_fault(tsk, addr, fsr, sig, code, regs);
 390         return 0;
 391
 392 no_context:
 393         __do_kernel_fault(mm, addr, fsr, regs);
 394         return 0;
 395 }
 396 #else                                   /* CONFIG_MMU */
 397 static int
 398 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 399 {
 400         return 0;
 401 }
 402 #endif                                  /* CONFIG_MMU */
 403
 404 /*
 405  * First Level Translation Fault Handler
 406  *
 407  * We enter here because the first level page table doesn't contain
 408  * a valid entry for the address.
 409  *
 410  * If the address is in kernel space (>= TASK_SIZE), then we are
 411  * probably faulting in the vmalloc() area.
 412  *
 413  * If the init_task's first level page tables contains the relevant
 414  * entry, we copy the it to this task.  If not, we send the process
 415  * a signal, fixup the exception, or oops the kernel.
 416  *
 417  * NOTE! We MUST NOT take any locks for this case. We may be in an
 418  * interrupt or a critical region, and should only copy the information
 419  * from the master page table, nothing more.
 420  */
 421 #ifdef CONFIG_MMU
 422 static int __kprobes
 423 do_translation_fault(unsigned long addr, unsigned int fsr,
 424                      struct pt_regs *regs)
 425 {
 426         unsigned int index;
 427         pgd_t *pgd, *pgd_k;
 428         pud_t *pud, *pud_k;
 429         pmd_t *pmd, *pmd_k;
 430
 431         if (addr < TASK_SIZE)
 432                 return do_page_fault(addr, fsr, regs);
 433
 434         if (user_mode(regs))
 435                 goto bad_area;
 436
 437         index = pgd_index(addr);
 438
 439         pgd = cpu_get_pgd() + index;
 440         pgd_k = init_mm.pgd + index;
 441
 442         if (pgd_none(*pgd_k))
 443                 goto bad_area;
 444         if (!pgd_present(*pgd))
 445                 set_pgd(pgd, *pgd_k);
 446
 447         pud = pud_offset(pgd, addr);
 448         pud_k = pud_offset(pgd_k, addr);
 449
 450         if (pud_none(*pud_k))
 451                 goto bad_area;
 452         if (!pud_present(*pud))
 453                 set_pud(pud, *pud_k);
 454
 455         pmd = pmd_offset(pud, addr);
 456         pmd_k = pmd_offset(pud_k, addr);
 457
 458 #ifdef CONFIG_ARM_LPAE
 459         /*
 460          * Only one hardware entry per PMD with LPAE.
 461          */
 462         index = 0;
 463 #else
 464         /*
 465          * On ARM one Linux PGD entry contains two hardware entries (see page
 466          * tables layout in pgtable.h). We normally guarantee that we always
 467          * fill both L1 entries. But create_mapping() doesn't follow the rule.
 468          * It can create inidividual L1 entries, so here we have to call
 469          * pmd_none() check for the entry really corresponded to address, not
 470          * for the first of pair.
 471          */
 472         index = (addr >> SECTION_SHIFT) & 1;
 473 #endif
 474         if (pmd_none(pmd_k[index]))
 475                 goto bad_area;
 476
 477         copy_pmd(pmd, pmd_k);
 478         return 0;
 479
 480 bad_area:
 481         do_bad_area(addr, fsr, regs);
 482         return 0;
 483 }
 484 #else                                   /* CONFIG_MMU */
 485 static int
 486 do_translation_fault(unsigned long addr, unsigned int fsr,
 487                      struct pt_regs *regs)
 488 {
 489         return 0;
 490 }
 491 #endif                                  /* CONFIG_MMU */
 492
 493 /*
 494  * Some section permission faults need to be handled gracefully.
 495  * They can happen due to a __{get,put}_user during an oops.
 496  */
 497 #ifndef CONFIG_ARM_LPAE
 498 static int
 499 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 500 {
 501         do_bad_area(addr, fsr, regs);
 502         return 0;
 503 }
 504 #endif /* CONFIG_ARM_LPAE */
 505
 506 /*
 507  * This abort handler always returns "fault".
 508  */
 509 static int
 510 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 511 {
 512         return 1;
 513 }
 514
 515 struct fsr_info {
 516         int     (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
 517         int     sig;
 518         int     code;
 519         const char *name;
 520 };
 521
 522 /* FSR definition */
 523 #ifdef CONFIG_ARM_LPAE
 524 #include "fsr-3level.c"
 525 #else
 526 #include "fsr-2level.c"
 527 #endif
 528
 529 void __init
 530 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
 531                 int sig, int code, const char *name)
 532 {
 533         if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
 534                 BUG();
 535
 536         fsr_info[nr].fn   = fn;
 537         fsr_info[nr].sig  = sig;
 538         fsr_info[nr].code = code;
 539         fsr_info[nr].name = name;
 540 }
 541
 542 /*
 543  * Dispatch a data abort to the relevant handler.
 544  */
 545 asmlinkage void __exception
 546 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 547 {
 548         const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
 549         struct siginfo info;
 550
 551         if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
 552                 return;
 553
 554         pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
 555                 inf->name, fsr, addr);
 556         show_pte(current->mm, addr);
 557
 558         info.si_signo = inf->sig;
 559         info.si_errno = 0;
 560         info.si_code  = inf->code;
 561         info.si_addr  = (void __user *)addr;
 562         arm_notify_die("", regs, &info, fsr, 0);
 563 }
 564
 565 void __init
 566 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
 567                  int sig, int code, const char *name)
 568 {
 569         if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
 570                 BUG();
 571
 572         ifsr_info[nr].fn   = fn;
 573         ifsr_info[nr].sig  = sig;
 574         ifsr_info[nr].code = code;
 575         ifsr_info[nr].name = name;
 576 }
 577
 578 asmlinkage void __exception
 579 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
 580 {
 581         const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
 582         struct siginfo info;
 583
 584         if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
 585                 return;
 586
 587         pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
 588                 inf->name, ifsr, addr);
 589
 590         info.si_signo = inf->sig;
 591         info.si_errno = 0;
 592         info.si_code  = inf->code;
 593         info.si_addr  = (void __user *)addr;
 594         arm_notify_die("", regs, &info, ifsr, 0);
 595 }
 596
 597 /*
 598  * Abort handler to be used only during first unmasking of asynchronous aborts
 599  * on the boot CPU. This makes sure that the machine will not die if the
 600  * firmware/bootloader left an imprecise abort pending for us to trip over.
 601  */
 602 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
 603                                       struct pt_regs *regs)
 604 {
 605         pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
 606                 "first unmask, this is most likely caused by a "
 607                 "firmware/bootloader bug.\n", fsr);
 608
 609         return 0;
 610 }
 611
 612 void __init early_abt_enable(void)
 613 {
 614         fsr_info[FSR_FS_AEA].fn = early_abort_handler;
 615         local_abt_enable();
 616         fsr_info[FSR_FS_AEA].fn = do_bad;
 617 }
 618
 619 #ifndef CONFIG_ARM_LPAE
 620 static int __init exceptions_init(void)
 621 {
 622         if (cpu_architecture() >= CPU_ARCH_ARMv6) {
 623                 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
 624                                 "I-cache maintenance fault");
 625         }
 626
 627         if (cpu_architecture() >= CPU_ARCH_ARMv7) {
 628                 /*
 629                  * TODO: Access flag faults introduced in ARMv6K.
 630                  * Runtime check for 'K' extension is needed
 631                  */
 632                 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
 633                                 "section access flag fault");
 634                 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
 635                                 "section access flag fault");
 636         }
 637
 638         return 0;
 639 }
 640
 641 arch_initcall(exceptions_init);
 642 #endif