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/hardirq.h>
  15 #include <linux/init.h>
  16 #include <linux/kprobes.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/page-flags.h>
  19
  20 #include <asm/system.h>
  21 #include <asm/pgtable.h>
  22 #include <asm/tlbflush.h>
  23
  24 #include "fault.h"
  25
  26
  27 #ifdef CONFIG_KPROBES
  28 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
  29 {
  30         int ret = 0;
  31
  32         if (!user_mode(regs)) {
  33                 /* kprobe_running() needs smp_processor_id() */
  34                 preempt_disable();
  35                 if (kprobe_running() && kprobe_fault_handler(regs, fsr))
  36                         ret = 1;
  37                 preempt_enable();
  38         }
  39
  40         return ret;
  41 }
  42 #else
  43 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
  44 {
  45         return 0;
  46 }
  47 #endif
  48
  49 /*
  50  * This is useful to dump out the page tables associated with
  51  * 'addr' in mm 'mm'.
  52  */
  53 void show_pte(struct mm_struct *mm, unsigned long addr)
  54 {
  55         pgd_t *pgd;
  56
  57         if (!mm)
  58                 mm = &init_mm;
  59
  60         printk(KERN_ALERT "pgd = %p\n", mm->pgd);
  61         pgd = pgd_offset(mm, addr);
  62         printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
  63
  64         do {
  65                 pmd_t *pmd;
  66                 pte_t *pte;
  67
  68                 if (pgd_none(*pgd))
  69                         break;
  70
  71                 if (pgd_bad(*pgd)) {
  72                         printk("(bad)");
  73                         break;
  74                 }
  75
  76                 pmd = pmd_offset(pgd, addr);
  77                 if (PTRS_PER_PMD != 1)
  78                         printk(", *pmd=%08lx", pmd_val(*pmd));
  79
  80                 if (pmd_none(*pmd))
  81                         break;
  82
  83                 if (pmd_bad(*pmd)) {
  84                         printk("(bad)");
  85                         break;
  86                 }
  87
  88                 /* We must not map this if we have highmem enabled */
  89                 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
  90                         break;
  91
  92                 pte = pte_offset_map(pmd, addr);
  93                 printk(", *pte=%08lx", pte_val(*pte));
  94                 printk(", *ppte=%08lx", pte_val(pte[-PTRS_PER_PTE]));
  95                 pte_unmap(pte);
  96         } while(0);
  97
  98         printk("\n");
  99 }
 100
 101 /*
 102  * Oops.  The kernel tried to access some page that wasn't present.
 103  */
 104 static void
 105 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
 106                   struct pt_regs *regs)
 107 {
 108         /*
 109          * Are we prepared to handle this kernel fault?
 110          */
 111         if (fixup_exception(regs))
 112                 return;
 113
 114         /*
 115          * No handler, we'll have to terminate things with extreme prejudice.
 116          */
 117         bust_spinlocks(1);
 118         printk(KERN_ALERT
 119                 "Unable to handle kernel %s at virtual address %08lx\n",
 120                 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
 121                 "paging request", addr);
 122
 123         show_pte(mm, addr);
 124         die("Oops", regs, fsr);
 125         bust_spinlocks(0);
 126         do_exit(SIGKILL);
 127 }
 128
 129 /*
 130  * Something tried to access memory that isn't in our memory map..
 131  * User mode accesses just cause a SIGSEGV
 132  */
 133 static void
 134 __do_user_fault(struct task_struct *tsk, unsigned long addr,
 135                 unsigned int fsr, unsigned int sig, int code,
 136                 struct pt_regs *regs)
 137 {
 138         struct siginfo si;
 139
 140 #ifdef CONFIG_DEBUG_USER
 141         if (user_debug & UDBG_SEGV) {
 142                 printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
 143                        tsk->comm, sig, addr, fsr);
 144                 show_pte(tsk->mm, addr);
 145                 show_regs(regs);
 146         }
 147 #endif
 148
 149         tsk->thread.address = addr;
 150         tsk->thread.error_code = fsr;
 151         tsk->thread.trap_no = 14;
 152         si.si_signo = sig;
 153         si.si_errno = 0;
 154         si.si_code = code;
 155         si.si_addr = (void __user *)addr;
 156         force_sig_info(sig, &si, tsk);
 157 }
 158
 159 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 160 {
 161         struct task_struct *tsk = current;
 162         struct mm_struct *mm = tsk->active_mm;
 163
 164         /*
 165          * If we are in kernel mode at this point, we
 166          * have no context to handle this fault with.
 167          */
 168         if (user_mode(regs))
 169                 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
 170         else
 171                 __do_kernel_fault(mm, addr, fsr, regs);
 172 }
 173
 174 #define VM_FAULT_BADMAP         0x010000
 175 #define VM_FAULT_BADACCESS      0x020000
 176
 177 static int
 178 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
 179                 struct task_struct *tsk)
 180 {
 181         struct vm_area_struct *vma;
 182         int fault, mask;
 183
 184         vma = find_vma(mm, addr);
 185         fault = VM_FAULT_BADMAP;
 186         if (!vma)
 187                 goto out;
 188         if (vma->vm_start > addr)
 189                 goto check_stack;
 190
 191         /*
 192          * Ok, we have a good vm_area for this
 193          * memory access, so we can handle it.
 194          */
 195 good_area:
 196         if (fsr & (1 << 11)) /* write? */
 197                 mask = VM_WRITE;
 198         else
 199                 mask = VM_READ|VM_EXEC|VM_WRITE;
 200
 201         fault = VM_FAULT_BADACCESS;
 202         if (!(vma->vm_flags & mask))
 203                 goto out;
 204
 205         /*
 206          * If for any reason at all we couldn't handle
 207          * the fault, make sure we exit gracefully rather
 208          * than endlessly redo the fault.
 209          */
 210 survive:
 211         fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, (fsr & (1 << 11)) ? FAULT_FLAG_WRITE : 0);
 212         if (unlikely(fault & VM_FAULT_ERROR)) {
 213                 if (fault & VM_FAULT_OOM)
 214                         goto out_of_memory;
 215                 else if (fault & VM_FAULT_SIGBUS)
 216                         return fault;
 217                 BUG();
 218         }
 219         if (fault & VM_FAULT_MAJOR)
 220                 tsk->maj_flt++;
 221         else
 222                 tsk->min_flt++;
 223         return fault;
 224
 225 out_of_memory:
 226         if (!is_global_init(tsk))
 227                 goto out;
 228
 229         /*
 230          * If we are out of memory for pid1, sleep for a while and retry
 231          */
 232         up_read(&mm->mmap_sem);
 233         yield();
 234         down_read(&mm->mmap_sem);
 235         goto survive;
 236
 237 check_stack:
 238         if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
 239                 goto good_area;
 240 out:
 241         return fault;
 242 }
 243
 244 static int __kprobes
 245 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 246 {
 247         struct task_struct *tsk;
 248         struct mm_struct *mm;
 249         int fault, sig, code;
 250
 251         if (notify_page_fault(regs, fsr))
 252                 return 0;
 253
 254         tsk = current;
 255         mm  = tsk->mm;
 256
 257         /*
 258          * If we're in an interrupt or have no user
 259          * context, we must not take the fault..
 260          */
 261         if (in_atomic() || !mm)
 262                 goto no_context;
 263
 264         /*
 265          * As per x86, we may deadlock here.  However, since the kernel only
 266          * validly references user space from well defined areas of the code,
 267          * we can bug out early if this is from code which shouldn't.
 268          */
 269         if (!down_read_trylock(&mm->mmap_sem)) {
 270                 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
 271                         goto no_context;
 272                 down_read(&mm->mmap_sem);
 273         }
 274
 275         fault = __do_page_fault(mm, addr, fsr, tsk);
 276         up_read(&mm->mmap_sem);
 277
 278         /*
 279          * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
 280          */
 281         if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
 282                 return 0;
 283
 284         /*
 285          * If we are in kernel mode at this point, we
 286          * have no context to handle this fault with.
 287          */
 288         if (!user_mode(regs))
 289                 goto no_context;
 290
 291         if (fault & VM_FAULT_OOM) {
 292                 /*
 293                  * We ran out of memory, or some other thing
 294                  * happened to us that made us unable to handle
 295                  * the page fault gracefully.
 296                  */
 297                 printk("VM: killing process %s\n", tsk->comm);
 298                 do_group_exit(SIGKILL);
 299                 return 0;
 300         }
 301         if (fault & VM_FAULT_SIGBUS) {
 302                 /*
 303                  * We had some memory, but were unable to
 304                  * successfully fix up this page fault.
 305                  */
 306                 sig = SIGBUS;
 307                 code = BUS_ADRERR;
 308         } else {
 309                 /*
 310                  * Something tried to access memory that
 311                  * isn't in our memory map..
 312                  */
 313                 sig = SIGSEGV;
 314                 code = fault == VM_FAULT_BADACCESS ?
 315                         SEGV_ACCERR : SEGV_MAPERR;
 316         }
 317
 318         __do_user_fault(tsk, addr, fsr, sig, code, regs);
 319         return 0;
 320
 321 no_context:
 322         __do_kernel_fault(mm, addr, fsr, regs);
 323         return 0;
 324 }
 325
 326 /*
 327  * First Level Translation Fault Handler
 328  *
 329  * We enter here because the first level page table doesn't contain
 330  * a valid entry for the address.
 331  *
 332  * If the address is in kernel space (>= TASK_SIZE), then we are
 333  * probably faulting in the vmalloc() area.
 334  *
 335  * If the init_task's first level page tables contains the relevant
 336  * entry, we copy the it to this task.  If not, we send the process
 337  * a signal, fixup the exception, or oops the kernel.
 338  *
 339  * NOTE! We MUST NOT take any locks for this case. We may be in an
 340  * interrupt or a critical region, and should only copy the information
 341  * from the master page table, nothing more.
 342  */
 343 static int __kprobes
 344 do_translation_fault(unsigned long addr, unsigned int fsr,
 345                      struct pt_regs *regs)
 346 {
 347         unsigned int index;
 348         pgd_t *pgd, *pgd_k;
 349         pmd_t *pmd, *pmd_k;
 350
 351         if (addr < TASK_SIZE)
 352                 return do_page_fault(addr, fsr, regs);
 353
 354         index = pgd_index(addr);
 355
 356         /*
 357          * FIXME: CP15 C1 is write only on ARMv3 architectures.
 358          */
 359         pgd = cpu_get_pgd() + index;
 360         pgd_k = init_mm.pgd + index;
 361
 362         if (pgd_none(*pgd_k))
 363                 goto bad_area;
 364
 365         if (!pgd_present(*pgd))
 366                 set_pgd(pgd, *pgd_k);
 367
 368         pmd_k = pmd_offset(pgd_k, addr);
 369         pmd   = pmd_offset(pgd, addr);
 370
 371         if (pmd_none(*pmd_k))
 372                 goto bad_area;
 373
 374         copy_pmd(pmd, pmd_k);
 375         return 0;
 376
 377 bad_area:
 378         do_bad_area(addr, fsr, regs);
 379         return 0;
 380 }
 381
 382 /*
 383  * Some section permission faults need to be handled gracefully.
 384  * They can happen due to a __{get,put}_user during an oops.
 385  */
 386 static int
 387 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 388 {
 389         do_bad_area(addr, fsr, regs);
 390         return 0;
 391 }
 392
 393 /*
 394  * This abort handler always returns "fault".
 395  */
 396 static int
 397 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 398 {
 399         return 1;
 400 }
 401
 402 static struct fsr_info {
 403         int     (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
 404         int     sig;
 405         int     code;
 406         const char *name;
 407 } fsr_info[] = {
 408         /*
 409          * The following are the standard ARMv3 and ARMv4 aborts.  ARMv5
 410          * defines these to be "precise" aborts.
 411          */
 412         { do_bad,               SIGSEGV, 0,             "vector exception"                 },
 413         { do_bad,               SIGILL,  BUS_ADRALN,    "alignment exception"              },
 414         { do_bad,               SIGKILL, 0,             "terminal exception"               },
 415         { do_bad,               SIGILL,  BUS_ADRALN,    "alignment exception"              },
 416         { do_bad,               SIGBUS,  0,             "external abort on linefetch"      },
 417         { do_translation_fault, SIGSEGV, SEGV_MAPERR,   "section translation fault"        },
 418         { do_bad,               SIGBUS,  0,             "external abort on linefetch"      },
 419         { do_page_fault,        SIGSEGV, SEGV_MAPERR,   "page translation fault"           },
 420         { do_bad,               SIGBUS,  0,             "external abort on non-linefetch"  },
 421         { do_bad,               SIGSEGV, SEGV_ACCERR,   "section domain fault"             },
 422         { do_bad,               SIGBUS,  0,             "external abort on non-linefetch"  },
 423         { do_bad,               SIGSEGV, SEGV_ACCERR,   "page domain fault"                },
 424         { do_bad,               SIGBUS,  0,             "external abort on translation"    },
 425         { do_sect_fault,        SIGSEGV, SEGV_ACCERR,   "section permission fault"         },
 426         { do_bad,               SIGBUS,  0,             "external abort on translation"    },
 427         { do_page_fault,        SIGSEGV, SEGV_ACCERR,   "page permission fault"            },
 428         /*
 429          * The following are "imprecise" aborts, which are signalled by bit
 430          * 10 of the FSR, and may not be recoverable.  These are only
 431          * supported if the CPU abort handler supports bit 10.
 432          */
 433         { do_bad,               SIGBUS,  0,             "unknown 16"                       },
 434         { do_bad,               SIGBUS,  0,             "unknown 17"                       },
 435         { do_bad,               SIGBUS,  0,             "unknown 18"                       },
 436         { do_bad,               SIGBUS,  0,             "unknown 19"                       },
 437         { do_bad,               SIGBUS,  0,             "lock abort"                       }, /* xscale */
 438         { do_bad,               SIGBUS,  0,             "unknown 21"                       },
 439         { do_bad,               SIGBUS,  BUS_OBJERR,    "imprecise external abort"         }, /* xscale */
 440         { do_bad,               SIGBUS,  0,             "unknown 23"                       },
 441         { do_bad,               SIGBUS,  0,             "dcache parity error"              }, /* xscale */
 442         { do_bad,               SIGBUS,  0,             "unknown 25"                       },
 443         { do_bad,               SIGBUS,  0,             "unknown 26"                       },
 444         { do_bad,               SIGBUS,  0,             "unknown 27"                       },
 445         { do_bad,               SIGBUS,  0,             "unknown 28"                       },
 446         { do_bad,               SIGBUS,  0,             "unknown 29"                       },
 447         { do_bad,               SIGBUS,  0,             "unknown 30"                       },
 448         { do_bad,               SIGBUS,  0,             "unknown 31"                       }
 449 };
 450
 451 void __init
 452 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
 453                 int sig, const char *name)
 454 {
 455         if (nr >= 0 && nr < ARRAY_SIZE(fsr_info)) {
 456                 fsr_info[nr].fn   = fn;
 457                 fsr_info[nr].sig  = sig;
 458                 fsr_info[nr].name = name;
 459         }
 460 }
 461
 462 /*
 463  * Dispatch a data abort to the relevant handler.
 464  */
 465 asmlinkage void __exception
 466 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 467 {
 468         const struct fsr_info *inf = fsr_info + (fsr & 15) + ((fsr & (1 << 10)) >> 6);
 469         struct siginfo info;
 470
 471         if (!inf->fn(addr, fsr, regs))
 472                 return;
 473
 474         printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
 475                 inf->name, fsr, addr);
 476
 477         info.si_signo = inf->sig;
 478         info.si_errno = 0;
 479         info.si_code  = inf->code;
 480         info.si_addr  = (void __user *)addr;
 481         arm_notify_die("", regs, &info, fsr, 0);
 482 }
 483
 484 asmlinkage void __exception
 485 do_PrefetchAbort(unsigned long addr, struct pt_regs *regs)
 486 {
 487         do_translation_fault(addr, 0, regs);
 488 }
 489