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 #include <linux/highmem.h>
  20
  21 #include <asm/system.h>
  22 #include <asm/pgtable.h>
  23 #include <asm/tlbflush.h>
  24
  25 #include "fault.h"
  26
  27
  28 #ifdef CONFIG_KPROBES
  29 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
  30 {
  31         int ret = 0;
  32
  33         if (!user_mode(regs)) {
  34                 /* kprobe_running() needs smp_processor_id() */
  35                 preempt_disable();
  36                 if (kprobe_running() && kprobe_fault_handler(regs, fsr))
  37                         ret = 1;
  38                 preempt_enable();
  39         }
  40
  41         return ret;
  42 }
  43 #else
  44 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
  45 {
  46         return 0;
  47 }
  48 #endif
  49
  50 /*
  51  * This is useful to dump out the page tables associated with
  52  * 'addr' in mm 'mm'.
  53  */
  54 void show_pte(struct mm_struct *mm, unsigned long addr)
  55 {
  56         pgd_t *pgd;
  57
  58         if (!mm)
  59                 mm = &init_mm;
  60
  61         printk(KERN_ALERT "pgd = %p\n", mm->pgd);
  62         pgd = pgd_offset(mm, addr);
  63         printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
  64
  65         do {
  66                 pmd_t *pmd;
  67                 pte_t *pte;
  68
  69                 if (pgd_none(*pgd))
  70                         break;
  71
  72                 if (pgd_bad(*pgd)) {
  73                         printk("(bad)");
  74                         break;
  75                 }
  76
  77                 pmd = pmd_offset(pgd, addr);
  78                 if (PTRS_PER_PMD != 1)
  79                         printk(", *pmd=%08lx", pmd_val(*pmd));
  80
  81                 if (pmd_none(*pmd))
  82                         break;
  83
  84                 if (pmd_bad(*pmd)) {
  85                         printk("(bad)");
  86                         break;
  87                 }
  88
  89                 /* We must not map this if we have highmem enabled */
  90                 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
  91                         break;
  92
  93                 pte = pte_offset_map(pmd, addr);
  94                 printk(", *pte=%08lx", pte_val(*pte));
  95                 printk(", *ppte=%08lx", pte_val(pte[-PTRS_PER_PTE]));
  96                 pte_unmap(pte);
  97         } while(0);
  98
  99         printk("\n");
 100 }
 101
 102 /*
 103  * Oops.  The kernel tried to access some page that wasn't present.
 104  */
 105 static void
 106 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
 107                   struct pt_regs *regs)
 108 {
 109         /*
 110          * Are we prepared to handle this kernel fault?
 111          */
 112         if (fixup_exception(regs))
 113                 return;
 114
 115         /*
 116          * No handler, we'll have to terminate things with extreme prejudice.
 117          */
 118         bust_spinlocks(1);
 119         printk(KERN_ALERT
 120                 "Unable to handle kernel %s at virtual address %08lx\n",
 121                 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
 122                 "paging request", addr);
 123
 124         show_pte(mm, addr);
 125         die("Oops", regs, fsr);
 126         bust_spinlocks(0);
 127         do_exit(SIGKILL);
 128 }
 129
 130 /*
 131  * Something tried to access memory that isn't in our memory map..
 132  * User mode accesses just cause a SIGSEGV
 133  */
 134 static void
 135 __do_user_fault(struct task_struct *tsk, unsigned long addr,
 136                 unsigned int fsr, unsigned int sig, int code,
 137                 struct pt_regs *regs)
 138 {
 139         struct siginfo si;
 140
 141 #ifdef CONFIG_DEBUG_USER
 142         if (user_debug & UDBG_SEGV) {
 143                 printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
 144                        tsk->comm, sig, addr, fsr);
 145                 show_pte(tsk->mm, addr);
 146                 show_regs(regs);
 147         }
 148 #endif
 149
 150         tsk->thread.address = addr;
 151         tsk->thread.error_code = fsr;
 152         tsk->thread.trap_no = 14;
 153         si.si_signo = sig;
 154         si.si_errno = 0;
 155         si.si_code = code;
 156         si.si_addr = (void __user *)addr;
 157         force_sig_info(sig, &si, tsk);
 158 }
 159
 160 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 161 {
 162         struct task_struct *tsk = current;
 163         struct mm_struct *mm = tsk->active_mm;
 164
 165         /*
 166          * If we are in kernel mode at this point, we
 167          * have no context to handle this fault with.
 168          */
 169         if (user_mode(regs))
 170                 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
 171         else
 172                 __do_kernel_fault(mm, addr, fsr, regs);
 173 }
 174
 175 #define VM_FAULT_BADMAP         0x010000
 176 #define VM_FAULT_BADACCESS      0x020000
 177
 178 static int
 179 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
 180                 struct task_struct *tsk)
 181 {
 182         struct vm_area_struct *vma;
 183         int fault, mask;
 184
 185         vma = find_vma(mm, addr);
 186         fault = VM_FAULT_BADMAP;
 187         if (!vma)
 188                 goto out;
 189         if (vma->vm_start > addr)
 190                 goto check_stack;
 191
 192         /*
 193          * Ok, we have a good vm_area for this
 194          * memory access, so we can handle it.
 195          */
 196 good_area:
 197         if (fsr & (1 << 11)) /* write? */
 198                 mask = VM_WRITE;
 199         else
 200                 mask = VM_READ|VM_EXEC|VM_WRITE;
 201
 202         fault = VM_FAULT_BADACCESS;
 203         if (!(vma->vm_flags & mask))
 204                 goto out;
 205
 206         /*
 207          * If for any reason at all we couldn't handle
 208          * the fault, make sure we exit gracefully rather
 209          * than endlessly redo the fault.
 210          */
 211 survive:
 212         fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, (fsr & (1 << 11)) ? FAULT_FLAG_WRITE : 0);
 213         if (unlikely(fault & VM_FAULT_ERROR)) {
 214                 if (fault & VM_FAULT_OOM)
 215                         goto out_of_memory;
 216                 else if (fault & VM_FAULT_SIGBUS)
 217                         return fault;
 218                 BUG();
 219         }
 220         if (fault & VM_FAULT_MAJOR)
 221                 tsk->maj_flt++;
 222         else
 223                 tsk->min_flt++;
 224         return fault;
 225
 226 out_of_memory:
 227         if (!is_global_init(tsk))
 228                 goto out;
 229
 230         /*
 231          * If we are out of memory for pid1, sleep for a while and retry
 232          */
 233         up_read(&mm->mmap_sem);
 234         yield();
 235         down_read(&mm->mmap_sem);
 236         goto survive;
 237
 238 check_stack:
 239         if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
 240                 goto good_area;
 241 out:
 242         return fault;
 243 }
 244
 245 static int __kprobes
 246 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 247 {
 248         struct task_struct *tsk;
 249         struct mm_struct *mm;
 250         int fault, sig, code;
 251
 252         if (notify_page_fault(regs, fsr))
 253                 return 0;
 254
 255         tsk = current;
 256         mm  = tsk->mm;
 257
 258         /*
 259          * If we're in an interrupt or have no user
 260          * context, we must not take the fault..
 261          */
 262         if (in_atomic() || !mm)
 263                 goto no_context;
 264
 265         /*
 266          * As per x86, we may deadlock here.  However, since the kernel only
 267          * validly references user space from well defined areas of the code,
 268          * we can bug out early if this is from code which shouldn't.
 269          */
 270         if (!down_read_trylock(&mm->mmap_sem)) {
 271                 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
 272                         goto no_context;
 273                 down_read(&mm->mmap_sem);
 274         }
 275
 276         fault = __do_page_fault(mm, addr, fsr, tsk);
 277         up_read(&mm->mmap_sem);
 278
 279         /*
 280          * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
 281          */
 282         if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
 283                 return 0;
 284
 285         /*
 286          * If we are in kernel mode at this point, we
 287          * have no context to handle this fault with.
 288          */
 289         if (!user_mode(regs))
 290                 goto no_context;
 291
 292         if (fault & VM_FAULT_OOM) {
 293                 /*
 294                  * We ran out of memory, or some other thing
 295                  * happened to us that made us unable to handle
 296                  * the page fault gracefully.
 297                  */
 298                 printk("VM: killing process %s\n", tsk->comm);
 299                 do_group_exit(SIGKILL);
 300                 return 0;
 301         }
 302         if (fault & VM_FAULT_SIGBUS) {
 303                 /*
 304                  * We had some memory, but were unable to
 305                  * successfully fix up this page fault.
 306                  */
 307                 sig = SIGBUS;
 308                 code = BUS_ADRERR;
 309         } else {
 310                 /*
 311                  * Something tried to access memory that
 312                  * isn't in our memory map..
 313                  */
 314                 sig = SIGSEGV;
 315                 code = fault == VM_FAULT_BADACCESS ?
 316                         SEGV_ACCERR : SEGV_MAPERR;
 317         }
 318
 319         __do_user_fault(tsk, addr, fsr, sig, code, regs);
 320         return 0;
 321
 322 no_context:
 323         __do_kernel_fault(mm, addr, fsr, regs);
 324         return 0;
 325 }
 326
 327 /*
 328  * First Level Translation Fault Handler
 329  *
 330  * We enter here because the first level page table doesn't contain
 331  * a valid entry for the address.
 332  *
 333  * If the address is in kernel space (>= TASK_SIZE), then we are
 334  * probably faulting in the vmalloc() area.
 335  *
 336  * If the init_task's first level page tables contains the relevant
 337  * entry, we copy the it to this task.  If not, we send the process
 338  * a signal, fixup the exception, or oops the kernel.
 339  *
 340  * NOTE! We MUST NOT take any locks for this case. We may be in an
 341  * interrupt or a critical region, and should only copy the information
 342  * from the master page table, nothing more.
 343  */
 344 static int __kprobes
 345 do_translation_fault(unsigned long addr, unsigned int fsr,
 346                      struct pt_regs *regs)
 347 {
 348         unsigned int index;
 349         pgd_t *pgd, *pgd_k;
 350         pmd_t *pmd, *pmd_k;
 351
 352         if (addr < TASK_SIZE)
 353                 return do_page_fault(addr, fsr, regs);
 354
 355         index = pgd_index(addr);
 356
 357         /*
 358          * FIXME: CP15 C1 is write only on ARMv3 architectures.
 359          */
 360         pgd = cpu_get_pgd() + index;
 361         pgd_k = init_mm.pgd + index;
 362
 363         if (pgd_none(*pgd_k))
 364                 goto bad_area;
 365
 366         if (!pgd_present(*pgd))
 367                 set_pgd(pgd, *pgd_k);
 368
 369         pmd_k = pmd_offset(pgd_k, addr);
 370         pmd   = pmd_offset(pgd, addr);
 371
 372         if (pmd_none(*pmd_k))
 373                 goto bad_area;
 374
 375         copy_pmd(pmd, pmd_k);
 376         return 0;
 377
 378 bad_area:
 379         do_bad_area(addr, fsr, regs);
 380         return 0;
 381 }
 382
 383 /*
 384  * Some section permission faults need to be handled gracefully.
 385  * They can happen due to a __{get,put}_user during an oops.
 386  */
 387 static int
 388 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 389 {
 390         do_bad_area(addr, fsr, regs);
 391         return 0;
 392 }
 393
 394 /*
 395  * This abort handler always returns "fault".
 396  */
 397 static int
 398 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 399 {
 400         return 1;
 401 }
 402
 403 static struct fsr_info {
 404         int     (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
 405         int     sig;
 406         int     code;
 407         const char *name;
 408 } fsr_info[] = {
 409         /*
 410          * The following are the standard ARMv3 and ARMv4 aborts.  ARMv5
 411          * defines these to be "precise" aborts.
 412          */
 413         { do_bad,               SIGSEGV, 0,             "vector exception"                 },
 414         { do_bad,               SIGILL,  BUS_ADRALN,    "alignment exception"              },
 415         { do_bad,               SIGKILL, 0,             "terminal exception"               },
 416         { do_bad,               SIGILL,  BUS_ADRALN,    "alignment exception"              },
 417         { do_bad,               SIGBUS,  0,             "external abort on linefetch"      },
 418         { do_translation_fault, SIGSEGV, SEGV_MAPERR,   "section translation fault"        },
 419         { do_bad,               SIGBUS,  0,             "external abort on linefetch"      },
 420         { do_page_fault,        SIGSEGV, SEGV_MAPERR,   "page translation fault"           },
 421         { do_bad,               SIGBUS,  0,             "external abort on non-linefetch"  },
 422         { do_bad,               SIGSEGV, SEGV_ACCERR,   "section domain fault"             },
 423         { do_bad,               SIGBUS,  0,             "external abort on non-linefetch"  },
 424         { do_bad,               SIGSEGV, SEGV_ACCERR,   "page domain fault"                },
 425         { do_bad,               SIGBUS,  0,             "external abort on translation"    },
 426         { do_sect_fault,        SIGSEGV, SEGV_ACCERR,   "section permission fault"         },
 427         { do_bad,               SIGBUS,  0,             "external abort on translation"    },
 428         { do_page_fault,        SIGSEGV, SEGV_ACCERR,   "page permission fault"            },
 429         /*
 430          * The following are "imprecise" aborts, which are signalled by bit
 431          * 10 of the FSR, and may not be recoverable.  These are only
 432          * supported if the CPU abort handler supports bit 10.
 433          */
 434         { do_bad,               SIGBUS,  0,             "unknown 16"                       },
 435         { do_bad,               SIGBUS,  0,             "unknown 17"                       },
 436         { do_bad,               SIGBUS,  0,             "unknown 18"                       },
 437         { do_bad,               SIGBUS,  0,             "unknown 19"                       },
 438         { do_bad,               SIGBUS,  0,             "lock abort"                       }, /* xscale */
 439         { do_bad,               SIGBUS,  0,             "unknown 21"                       },
 440         { do_bad,               SIGBUS,  BUS_OBJERR,    "imprecise external abort"         }, /* xscale */
 441         { do_bad,               SIGBUS,  0,             "unknown 23"                       },
 442         { do_bad,               SIGBUS,  0,             "dcache parity error"              }, /* xscale */
 443         { do_bad,               SIGBUS,  0,             "unknown 25"                       },
 444         { do_bad,               SIGBUS,  0,             "unknown 26"                       },
 445         { do_bad,               SIGBUS,  0,             "unknown 27"                       },
 446         { do_bad,               SIGBUS,  0,             "unknown 28"                       },
 447         { do_bad,               SIGBUS,  0,             "unknown 29"                       },
 448         { do_bad,               SIGBUS,  0,             "unknown 30"                       },
 449         { do_bad,               SIGBUS,  0,             "unknown 31"                       }
 450 };
 451
 452 void __init
 453 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
 454                 int sig, const char *name)
 455 {
 456         if (nr >= 0 && nr < ARRAY_SIZE(fsr_info)) {
 457                 fsr_info[nr].fn   = fn;
 458                 fsr_info[nr].sig  = sig;
 459                 fsr_info[nr].name = name;
 460         }
 461 }
 462
 463 /*
 464  * Dispatch a data abort to the relevant handler.
 465  */
 466 asmlinkage void __exception
 467 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 468 {
 469         const struct fsr_info *inf = fsr_info + (fsr & 15) + ((fsr & (1 << 10)) >> 6);
 470         struct siginfo info;
 471
 472         if (!inf->fn(addr, fsr, regs))
 473                 return;
 474
 475         printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
 476                 inf->name, fsr, addr);
 477
 478         info.si_signo = inf->sig;
 479         info.si_errno = 0;
 480         info.si_code  = inf->code;
 481         info.si_addr  = (void __user *)addr;
 482         arm_notify_die("", regs, &info, fsr, 0);
 483 }
 484
 485 asmlinkage void __exception
 486 do_PrefetchAbort(unsigned long addr, struct pt_regs *regs)
 487 {
 488         do_translation_fault(addr, 0, regs);
 489 }
 490