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/module.h>
  12 #include <linux/signal.h>
  13 #include <linux/mm.h>
  14 #include <linux/init.h>
  15
  16 #include <asm/system.h>
  17 #include <asm/pgtable.h>
  18 #include <asm/tlbflush.h>
  19 #include <asm/uaccess.h>
  20
  21 #include "fault.h"
  22
  23 /*
  24  * This is useful to dump out the page tables associated with
  25  * 'addr' in mm 'mm'.
  26  */
  27 void show_pte(struct mm_struct *mm, unsigned long addr)
  28 {
  29         pgd_t *pgd;
  30
  31         if (!mm)
  32                 mm = &init_mm;
  33
  34         printk(KERN_ALERT "pgd = %p\n", mm->pgd);
  35         pgd = pgd_offset(mm, addr);
  36         printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
  37
  38         do {
  39                 pmd_t *pmd;
  40                 pte_t *pte;
  41
  42                 if (pgd_none(*pgd))
  43                         break;
  44
  45                 if (pgd_bad(*pgd)) {
  46                         printk("(bad)");
  47                         break;
  48                 }
  49
  50                 pmd = pmd_offset(pgd, addr);
  51 #if PTRS_PER_PMD != 1
  52                 printk(", *pmd=%08lx", pmd_val(*pmd));
  53 #endif
  54
  55                 if (pmd_none(*pmd))
  56                         break;
  57
  58                 if (pmd_bad(*pmd)) {
  59                         printk("(bad)");
  60                         break;
  61                 }
  62
  63 #ifndef CONFIG_HIGHMEM
  64                 /* We must not map this if we have highmem enabled */
  65                 pte = pte_offset_map(pmd, addr);
  66                 printk(", *pte=%08lx", pte_val(*pte));
  67                 printk(", *ppte=%08lx", pte_val(pte[-PTRS_PER_PTE]));
  68                 pte_unmap(pte);
  69 #endif
  70         } while(0);
  71
  72         printk("\n");
  73 }
  74
  75 /*
  76  * Oops.  The kernel tried to access some page that wasn't present.
  77  */
  78 static void
  79 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
  80                   struct pt_regs *regs)
  81 {
  82         /*
  83          * Are we prepared to handle this kernel fault?
  84          */
  85         if (fixup_exception(regs))
  86                 return;
  87
  88         /*
  89          * No handler, we'll have to terminate things with extreme prejudice.
  90          */
  91         bust_spinlocks(1);
  92         printk(KERN_ALERT
  93                 "Unable to handle kernel %s at virtual address %08lx\n",
  94                 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
  95                 "paging request", addr);
  96
  97         show_pte(mm, addr);
  98         die("Oops", regs, fsr);
  99         bust_spinlocks(0);
 100         do_exit(SIGKILL);
 101 }
 102
 103 /*
 104  * Something tried to access memory that isn't in our memory map..
 105  * User mode accesses just cause a SIGSEGV
 106  */
 107 static void
 108 __do_user_fault(struct task_struct *tsk, unsigned long addr,
 109                 unsigned int fsr, unsigned int sig, int code,
 110                 struct pt_regs *regs)
 111 {
 112         struct siginfo si;
 113
 114 #ifdef CONFIG_DEBUG_USER
 115         if (user_debug & UDBG_SEGV) {
 116                 printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
 117                        tsk->comm, sig, addr, fsr);
 118                 show_pte(tsk->mm, addr);
 119                 show_regs(regs);
 120         }
 121 #endif
 122
 123         tsk->thread.address = addr;
 124         tsk->thread.error_code = fsr;
 125         tsk->thread.trap_no = 14;
 126         si.si_signo = sig;
 127         si.si_errno = 0;
 128         si.si_code = code;
 129         si.si_addr = (void __user *)addr;
 130         force_sig_info(sig, &si, tsk);
 131 }
 132
 133 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 134 {
 135         struct task_struct *tsk = current;
 136         struct mm_struct *mm = tsk->active_mm;
 137
 138         /*
 139          * If we are in kernel mode at this point, we
 140          * have no context to handle this fault with.
 141          */
 142         if (user_mode(regs))
 143                 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
 144         else
 145                 __do_kernel_fault(mm, addr, fsr, regs);
 146 }
 147
 148 #define VM_FAULT_BADMAP         (-20)
 149 #define VM_FAULT_BADACCESS      (-21)
 150
 151 static int
 152 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
 153                 struct task_struct *tsk)
 154 {
 155         struct vm_area_struct *vma;
 156         int fault, mask;
 157
 158         vma = find_vma(mm, addr);
 159         fault = VM_FAULT_BADMAP;
 160         if (!vma)
 161                 goto out;
 162         if (vma->vm_start > addr)
 163                 goto check_stack;
 164
 165         /*
 166          * Ok, we have a good vm_area for this
 167          * memory access, so we can handle it.
 168          */
 169 good_area:
 170         if (fsr & (1 << 11)) /* write? */
 171                 mask = VM_WRITE;
 172         else
 173                 mask = VM_READ|VM_EXEC|VM_WRITE;
 174
 175         fault = VM_FAULT_BADACCESS;
 176         if (!(vma->vm_flags & mask))
 177                 goto out;
 178
 179         /*
 180          * If for any reason at all we couldn't handle
 181          * the fault, make sure we exit gracefully rather
 182          * than endlessly redo the fault.
 183          */
 184 survive:
 185         fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, fsr & (1 << 11));
 186
 187         /*
 188          * Handle the "normal" cases first - successful and sigbus
 189          */
 190         switch (fault) {
 191         case VM_FAULT_MAJOR:
 192                 tsk->maj_flt++;
 193                 return fault;
 194         case VM_FAULT_MINOR:
 195                 tsk->min_flt++;
 196         case VM_FAULT_SIGBUS:
 197                 return fault;
 198         }
 199
 200         if (!is_init(tsk))
 201                 goto out;
 202
 203         /*
 204          * If we are out of memory for pid1, sleep for a while and retry
 205          */
 206         up_read(&mm->mmap_sem);
 207         yield();
 208         down_read(&mm->mmap_sem);
 209         goto survive;
 210
 211 check_stack:
 212         if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
 213                 goto good_area;
 214 out:
 215         return fault;
 216 }
 217
 218 static int
 219 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 220 {
 221         struct task_struct *tsk;
 222         struct mm_struct *mm;
 223         int fault, sig, code;
 224
 225         tsk = current;
 226         mm  = tsk->mm;
 227
 228         /*
 229          * If we're in an interrupt or have no user
 230          * context, we must not take the fault..
 231          */
 232         if (in_atomic() || !mm)
 233                 goto no_context;
 234
 235         /*
 236          * As per x86, we may deadlock here.  However, since the kernel only
 237          * validly references user space from well defined areas of the code,
 238          * we can bug out early if this is from code which shouldn't.
 239          */
 240         if (!down_read_trylock(&mm->mmap_sem)) {
 241                 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
 242                         goto no_context;
 243                 down_read(&mm->mmap_sem);
 244         }
 245
 246         fault = __do_page_fault(mm, addr, fsr, tsk);
 247         up_read(&mm->mmap_sem);
 248
 249         /*
 250          * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
 251          */
 252         if (fault >= VM_FAULT_MINOR)
 253                 return 0;
 254
 255         /*
 256          * If we are in kernel mode at this point, we
 257          * have no context to handle this fault with.
 258          */
 259         if (!user_mode(regs))
 260                 goto no_context;
 261
 262         switch (fault) {
 263         case VM_FAULT_OOM:
 264                 /*
 265                  * We ran out of memory, or some other thing
 266                  * happened to us that made us unable to handle
 267                  * the page fault gracefully.
 268                  */
 269                 printk("VM: killing process %s\n", tsk->comm);
 270                 do_exit(SIGKILL);
 271                 return 0;
 272
 273         case VM_FAULT_SIGBUS:
 274                 /*
 275                  * We had some memory, but were unable to
 276                  * successfully fix up this page fault.
 277                  */
 278                 sig = SIGBUS;
 279                 code = BUS_ADRERR;
 280                 break;
 281
 282         default:
 283                 /*
 284                  * Something tried to access memory that
 285                  * isn't in our memory map..
 286                  */
 287                 sig = SIGSEGV;
 288                 code = fault == VM_FAULT_BADACCESS ?
 289                         SEGV_ACCERR : SEGV_MAPERR;
 290                 break;
 291         }
 292
 293         __do_user_fault(tsk, addr, fsr, sig, code, regs);
 294         return 0;
 295
 296 no_context:
 297         __do_kernel_fault(mm, addr, fsr, regs);
 298         return 0;
 299 }
 300
 301 /*
 302  * First Level Translation Fault Handler
 303  *
 304  * We enter here because the first level page table doesn't contain
 305  * a valid entry for the address.
 306  *
 307  * If the address is in kernel space (>= TASK_SIZE), then we are
 308  * probably faulting in the vmalloc() area.
 309  *
 310  * If the init_task's first level page tables contains the relevant
 311  * entry, we copy the it to this task.  If not, we send the process
 312  * a signal, fixup the exception, or oops the kernel.
 313  *
 314  * NOTE! We MUST NOT take any locks for this case. We may be in an
 315  * interrupt or a critical region, and should only copy the information
 316  * from the master page table, nothing more.
 317  */
 318 static int
 319 do_translation_fault(unsigned long addr, unsigned int fsr,
 320                      struct pt_regs *regs)
 321 {
 322         unsigned int index;
 323         pgd_t *pgd, *pgd_k;
 324         pmd_t *pmd, *pmd_k;
 325
 326         if (addr < TASK_SIZE)
 327                 return do_page_fault(addr, fsr, regs);
 328
 329         index = pgd_index(addr);
 330
 331         /*
 332          * FIXME: CP15 C1 is write only on ARMv3 architectures.
 333          */
 334         pgd = cpu_get_pgd() + index;
 335         pgd_k = init_mm.pgd + index;
 336
 337         if (pgd_none(*pgd_k))
 338                 goto bad_area;
 339
 340         if (!pgd_present(*pgd))
 341                 set_pgd(pgd, *pgd_k);
 342
 343         pmd_k = pmd_offset(pgd_k, addr);
 344         pmd   = pmd_offset(pgd, addr);
 345
 346         if (pmd_none(*pmd_k))
 347                 goto bad_area;
 348
 349         copy_pmd(pmd, pmd_k);
 350         return 0;
 351
 352 bad_area:
 353         do_bad_area(addr, fsr, regs);
 354         return 0;
 355 }
 356
 357 /*
 358  * Some section permission faults need to be handled gracefully.
 359  * They can happen due to a __{get,put}_user during an oops.
 360  */
 361 static int
 362 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 363 {
 364         do_bad_area(addr, fsr, regs);
 365         return 0;
 366 }
 367
 368 /*
 369  * This abort handler always returns "fault".
 370  */
 371 static int
 372 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 373 {
 374         return 1;
 375 }
 376
 377 static struct fsr_info {
 378         int     (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
 379         int     sig;
 380         int     code;
 381         const char *name;
 382 } fsr_info[] = {
 383         /*
 384          * The following are the standard ARMv3 and ARMv4 aborts.  ARMv5
 385          * defines these to be "precise" aborts.
 386          */
 387         { do_bad,               SIGSEGV, 0,             "vector exception"                 },
 388         { do_bad,               SIGILL,  BUS_ADRALN,    "alignment exception"              },
 389         { do_bad,               SIGKILL, 0,             "terminal exception"               },
 390         { do_bad,               SIGILL,  BUS_ADRALN,    "alignment exception"              },
 391         { do_bad,               SIGBUS,  0,             "external abort on linefetch"      },
 392         { do_translation_fault, SIGSEGV, SEGV_MAPERR,   "section translation fault"        },
 393         { do_bad,               SIGBUS,  0,             "external abort on linefetch"      },
 394         { do_page_fault,        SIGSEGV, SEGV_MAPERR,   "page translation fault"           },
 395         { do_bad,               SIGBUS,  0,             "external abort on non-linefetch"  },
 396         { do_bad,               SIGSEGV, SEGV_ACCERR,   "section domain fault"             },
 397         { do_bad,               SIGBUS,  0,             "external abort on non-linefetch"  },
 398         { do_bad,               SIGSEGV, SEGV_ACCERR,   "page domain fault"                },
 399         { do_bad,               SIGBUS,  0,             "external abort on translation"    },
 400         { do_sect_fault,        SIGSEGV, SEGV_ACCERR,   "section permission fault"         },
 401         { do_bad,               SIGBUS,  0,             "external abort on translation"    },
 402         { do_page_fault,        SIGSEGV, SEGV_ACCERR,   "page permission fault"            },
 403         /*
 404          * The following are "imprecise" aborts, which are signalled by bit
 405          * 10 of the FSR, and may not be recoverable.  These are only
 406          * supported if the CPU abort handler supports bit 10.
 407          */
 408         { do_bad,               SIGBUS,  0,             "unknown 16"                       },
 409         { do_bad,               SIGBUS,  0,             "unknown 17"                       },
 410         { do_bad,               SIGBUS,  0,             "unknown 18"                       },
 411         { do_bad,               SIGBUS,  0,             "unknown 19"                       },
 412         { do_bad,               SIGBUS,  0,             "lock abort"                       }, /* xscale */
 413         { do_bad,               SIGBUS,  0,             "unknown 21"                       },
 414         { do_bad,               SIGBUS,  BUS_OBJERR,    "imprecise external abort"         }, /* xscale */
 415         { do_bad,               SIGBUS,  0,             "unknown 23"                       },
 416         { do_bad,               SIGBUS,  0,             "dcache parity error"              }, /* xscale */
 417         { do_bad,               SIGBUS,  0,             "unknown 25"                       },
 418         { do_bad,               SIGBUS,  0,             "unknown 26"                       },
 419         { do_bad,               SIGBUS,  0,             "unknown 27"                       },
 420         { do_bad,               SIGBUS,  0,             "unknown 28"                       },
 421         { do_bad,               SIGBUS,  0,             "unknown 29"                       },
 422         { do_bad,               SIGBUS,  0,             "unknown 30"                       },
 423         { do_bad,               SIGBUS,  0,             "unknown 31"                       }
 424 };
 425
 426 void __init
 427 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
 428                 int sig, const char *name)
 429 {
 430         if (nr >= 0 && nr < ARRAY_SIZE(fsr_info)) {
 431                 fsr_info[nr].fn   = fn;
 432                 fsr_info[nr].sig  = sig;
 433                 fsr_info[nr].name = name;
 434         }
 435 }
 436
 437 /*
 438  * Dispatch a data abort to the relevant handler.
 439  */
 440 asmlinkage void __exception
 441 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 442 {
 443         const struct fsr_info *inf = fsr_info + (fsr & 15) + ((fsr & (1 << 10)) >> 6);
 444         struct siginfo info;
 445
 446         if (!inf->fn(addr, fsr, regs))
 447                 return;
 448
 449         printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
 450                 inf->name, fsr, addr);
 451
 452         info.si_signo = inf->sig;
 453         info.si_errno = 0;
 454         info.si_code  = inf->code;
 455         info.si_addr  = (void __user *)addr;
 456         arm_notify_die("", regs, &info, fsr, 0);
 457 }
 458
 459 asmlinkage void __exception
 460 do_PrefetchAbort(unsigned long addr, struct pt_regs *regs)
 461 {
 462         do_translation_fault(addr, 0, regs);
 463 }
 464