arch/ia64/mm/fault.c

   1 /*
   2  * MMU fault handling support.
   3  *
   4  * Copyright (C) 1998-2002 Hewlett-Packard Co
   5  *      David Mosberger-Tang <davidm@hpl.hp.com>
   6  */
   7 #include <linux/sched.h>
   8 #include <linux/kernel.h>
   9 #include <linux/mm.h>
  10 #include <linux/smp_lock.h>
  11 #include <linux/interrupt.h>
  12 #include <linux/kprobes.h>
  13
  14 #include <asm/pgtable.h>
  15 #include <asm/processor.h>
  16 #include <asm/system.h>
  17 #include <asm/uaccess.h>
  18 #include <asm/kdebug.h>
  19
  20 extern void die (char *, struct pt_regs *, long);
  21
  22 /*
  23  * This routine is analogous to expand_stack() but instead grows the
  24  * register backing store (which grows towards higher addresses).
  25  * Since the register backing store is access sequentially, we
  26  * disallow growing the RBS by more than a page at a time.  Note that
  27  * the VM_GROWSUP flag can be set on any VM area but that's fine
  28  * because the total process size is still limited by RLIMIT_STACK and
  29  * RLIMIT_AS.
  30  */
  31 static inline long
  32 expand_backing_store (struct vm_area_struct *vma, unsigned long address)
  33 {
  34         unsigned long grow;
  35
  36         grow = PAGE_SIZE >> PAGE_SHIFT;
  37         if (address - vma->vm_start > current->signal->rlim[RLIMIT_STACK].rlim_cur
  38             || (((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) > current->signal->rlim[RLIMIT_AS].rlim_cur))
  39                 return -ENOMEM;
  40         vma->vm_end += PAGE_SIZE;
  41         vma->vm_mm->total_vm += grow;
  42         if (vma->vm_flags & VM_LOCKED)
  43                 vma->vm_mm->locked_vm += grow;
  44         vm_stat_account(vma->vm_mm, vma->vm_flags, vma->vm_file, grow);
  45         return 0;
  46 }
  47
  48 /*
  49  * Return TRUE if ADDRESS points at a page in the kernel's mapped segment
  50  * (inside region 5, on ia64) and that page is present.
  51  */
  52 static int
  53 mapped_kernel_page_is_present (unsigned long address)
  54 {
  55         pgd_t *pgd;
  56         pud_t *pud;
  57         pmd_t *pmd;
  58         pte_t *ptep, pte;
  59
  60         pgd = pgd_offset_k(address);
  61         if (pgd_none(*pgd) || pgd_bad(*pgd))
  62                 return 0;
  63
  64         pud = pud_offset(pgd, address);
  65         if (pud_none(*pud) || pud_bad(*pud))
  66                 return 0;
  67
  68         pmd = pmd_offset(pud, address);
  69         if (pmd_none(*pmd) || pmd_bad(*pmd))
  70                 return 0;
  71
  72         ptep = pte_offset_kernel(pmd, address);
  73         if (!ptep)
  74                 return 0;
  75
  76         pte = *ptep;
  77         return pte_present(pte);
  78 }
  79
  80 void __kprobes
  81 ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs)
  82 {
  83         int signal = SIGSEGV, code = SEGV_MAPERR;
  84         struct vm_area_struct *vma, *prev_vma;
  85         struct mm_struct *mm = current->mm;
  86         struct siginfo si;
  87         unsigned long mask;
  88
  89         /*
  90          * If we're in an interrupt or have no user context, we must not take the fault..
  91          */
  92         if (in_atomic() || !mm)
  93                 goto no_context;
  94
  95 #ifdef CONFIG_VIRTUAL_MEM_MAP
  96         /*
  97          * If fault is in region 5 and we are in the kernel, we may already
  98          * have the mmap_sem (pfn_valid macro is called during mmap). There
  99          * is no vma for region 5 addr's anyway, so skip getting the semaphore
 100          * and go directly to the exception handling code.
 101          */
 102
 103         if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
 104                 goto bad_area_no_up;
 105 #endif
 106
 107         /*
 108          * This is to handle the kprobes on user space access instructions
 109          */
 110         if (notify_die(DIE_PAGE_FAULT, "page fault", regs, code, TRAP_BRKPT,
 111                                         SIGSEGV) == NOTIFY_STOP)
 112                 return;
 113
 114         down_read(&mm->mmap_sem);
 115
 116         vma = find_vma_prev(mm, address, &prev_vma);
 117         if (!vma)
 118                 goto bad_area;
 119
 120         /* find_vma_prev() returns vma such that address < vma->vm_end or NULL */
 121         if (address < vma->vm_start)
 122                 goto check_expansion;
 123
 124   good_area:
 125         code = SEGV_ACCERR;
 126
 127         /* OK, we've got a good vm_area for this memory area.  Check the access permissions: */
 128
 129 #       define VM_READ_BIT      0
 130 #       define VM_WRITE_BIT     1
 131 #       define VM_EXEC_BIT      2
 132
 133 #       if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
 134             || (1 << VM_EXEC_BIT) != VM_EXEC)
 135 #               error File is out of sync with <linux/mm.h>.  Please update.
 136 #       endif
 137
 138         mask = (  (((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
 139                 | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT)
 140                 | (((isr >> IA64_ISR_R_BIT) & 1UL) << VM_READ_BIT));
 141
 142         if ((vma->vm_flags & mask) != mask)
 143                 goto bad_area;
 144
 145   survive:
 146         /*
 147          * If for any reason at all we couldn't handle the fault, make
 148          * sure we exit gracefully rather than endlessly redo the
 149          * fault.
 150          */
 151         switch (handle_mm_fault(mm, vma, address, (mask & VM_WRITE) != 0)) {
 152               case VM_FAULT_MINOR:
 153                 ++current->min_flt;
 154                 break;
 155               case VM_FAULT_MAJOR:
 156                 ++current->maj_flt;
 157                 break;
 158               case VM_FAULT_SIGBUS:
 159                 /*
 160                  * We ran out of memory, or some other thing happened
 161                  * to us that made us unable to handle the page fault
 162                  * gracefully.
 163                  */
 164                 signal = SIGBUS;
 165                 goto bad_area;
 166               case VM_FAULT_OOM:
 167                 goto out_of_memory;
 168               default:
 169                 BUG();
 170         }
 171         up_read(&mm->mmap_sem);
 172         return;
 173
 174   check_expansion:
 175         if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
 176                 if (!(vma->vm_flags & VM_GROWSDOWN))
 177                         goto bad_area;
 178                 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
 179                     || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
 180                         goto bad_area;
 181                 if (expand_stack(vma, address))
 182                         goto bad_area;
 183         } else {
 184                 vma = prev_vma;
 185                 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
 186                     || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
 187                         goto bad_area;
 188                 if (expand_backing_store(vma, address))
 189                         goto bad_area;
 190         }
 191         goto good_area;
 192
 193   bad_area:
 194         up_read(&mm->mmap_sem);
 195 #ifdef CONFIG_VIRTUAL_MEM_MAP
 196   bad_area_no_up:
 197 #endif
 198         if ((isr & IA64_ISR_SP)
 199             || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
 200         {
 201                 /*
 202                  * This fault was due to a speculative load or lfetch.fault, set the "ed"
 203                  * bit in the psr to ensure forward progress.  (Target register will get a
 204                  * NaT for ld.s, lfetch will be canceled.)
 205                  */
 206                 ia64_psr(regs)->ed = 1;
 207                 return;
 208         }
 209         if (user_mode(regs)) {
 210                 si.si_signo = signal;
 211                 si.si_errno = 0;
 212                 si.si_code = code;
 213                 si.si_addr = (void __user *) address;
 214                 si.si_isr = isr;
 215                 si.si_flags = __ISR_VALID;
 216                 force_sig_info(signal, &si, current);
 217                 return;
 218         }
 219
 220   no_context:
 221         if ((isr & IA64_ISR_SP)
 222             || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
 223         {
 224                 /*
 225                  * This fault was due to a speculative load or lfetch.fault, set the "ed"
 226                  * bit in the psr to ensure forward progress.  (Target register will get a
 227                  * NaT for ld.s, lfetch will be canceled.)
 228                  */
 229                 ia64_psr(regs)->ed = 1;
 230                 return;
 231         }
 232
 233         /*
 234          * Since we have no vma's for region 5, we might get here even if the address is
 235          * valid, due to the VHPT walker inserting a non present translation that becomes
 236          * stale. If that happens, the non present fault handler already purged the stale
 237          * translation, which fixed the problem. So, we check to see if the translation is
 238          * valid, and return if it is.
 239          */
 240         if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
 241                 return;
 242
 243         if (ia64_done_with_exception(regs))
 244                 return;
 245
 246         /*
 247          * Oops. The kernel tried to access some bad page. We'll have to terminate things
 248          * with extreme prejudice.
 249          */
 250         bust_spinlocks(1);
 251
 252         if (address < PAGE_SIZE)
 253                 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
 254         else
 255                 printk(KERN_ALERT "Unable to handle kernel paging request at "
 256                        "virtual address %016lx\n", address);
 257         die("Oops", regs, isr);
 258         bust_spinlocks(0);
 259         do_exit(SIGKILL);
 260         return;
 261
 262   out_of_memory:
 263         up_read(&mm->mmap_sem);
 264         if (current->pid == 1) {
 265                 yield();
 266                 down_read(&mm->mmap_sem);
 267                 goto survive;
 268         }
 269         printk(KERN_CRIT "VM: killing process %s\n", current->comm);
 270         if (user_mode(regs))
 271                 do_exit(SIGKILL);
 272         goto no_context;
 273 }