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/interrupt.h>
  11 #include <linux/kprobes.h>
  12 #include <linux/kdebug.h>
  13
  14 #include <asm/pgtable.h>
  15 #include <asm/processor.h>
  16 #include <asm/system.h>
  17 #include <asm/uaccess.h>
  18
  19 extern void die (char *, struct pt_regs *, long);
  20
  21 #ifdef CONFIG_KPROBES
  22 ATOMIC_NOTIFIER_HEAD(notify_page_fault_chain);
  23
  24 /* Hook to register for page fault notifications */
  25 int register_page_fault_notifier(struct notifier_block *nb)
  26 {
  27         return atomic_notifier_chain_register(&notify_page_fault_chain, nb);
  28 }
  29
  30 int unregister_page_fault_notifier(struct notifier_block *nb)
  31 {
  32         return atomic_notifier_chain_unregister(&notify_page_fault_chain, nb);
  33 }
  34
  35 static inline int notify_page_fault(enum die_val val, const char *str,
  36                         struct pt_regs *regs, long err, int trap, int sig)
  37 {
  38         struct die_args args = {
  39                 .regs = regs,
  40                 .str = str,
  41                 .err = err,
  42                 .trapnr = trap,
  43                 .signr = sig
  44         };
  45         return atomic_notifier_call_chain(&notify_page_fault_chain, val, &args);
  46 }
  47 #else
  48 static inline int notify_page_fault(enum die_val val, const char *str,
  49                         struct pt_regs *regs, long err, int trap, int sig)
  50 {
  51         return NOTIFY_DONE;
  52 }
  53 #endif
  54
  55 /*
  56  * Return TRUE if ADDRESS points at a page in the kernel's mapped segment
  57  * (inside region 5, on ia64) and that page is present.
  58  */
  59 static int
  60 mapped_kernel_page_is_present (unsigned long address)
  61 {
  62         pgd_t *pgd;
  63         pud_t *pud;
  64         pmd_t *pmd;
  65         pte_t *ptep, pte;
  66
  67         pgd = pgd_offset_k(address);
  68         if (pgd_none(*pgd) || pgd_bad(*pgd))
  69                 return 0;
  70
  71         pud = pud_offset(pgd, address);
  72         if (pud_none(*pud) || pud_bad(*pud))
  73                 return 0;
  74
  75         pmd = pmd_offset(pud, address);
  76         if (pmd_none(*pmd) || pmd_bad(*pmd))
  77                 return 0;
  78
  79         ptep = pte_offset_kernel(pmd, address);
  80         if (!ptep)
  81                 return 0;
  82
  83         pte = *ptep;
  84         return pte_present(pte);
  85 }
  86
  87 void __kprobes
  88 ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs)
  89 {
  90         int signal = SIGSEGV, code = SEGV_MAPERR;
  91         struct vm_area_struct *vma, *prev_vma;
  92         struct mm_struct *mm = current->mm;
  93         struct siginfo si;
  94         unsigned long mask;
  95
  96         /* mmap_sem is performance critical.... */
  97         prefetchw(&mm->mmap_sem);
  98
  99         /*
 100          * If we're in an interrupt or have no user context, we must not take the fault..
 101          */
 102         if (in_atomic() || !mm)
 103                 goto no_context;
 104
 105 #ifdef CONFIG_VIRTUAL_MEM_MAP
 106         /*
 107          * If fault is in region 5 and we are in the kernel, we may already
 108          * have the mmap_sem (pfn_valid macro is called during mmap). There
 109          * is no vma for region 5 addr's anyway, so skip getting the semaphore
 110          * and go directly to the exception handling code.
 111          */
 112
 113         if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
 114                 goto bad_area_no_up;
 115 #endif
 116
 117         /*
 118          * This is to handle the kprobes on user space access instructions
 119          */
 120         if (notify_page_fault(DIE_PAGE_FAULT, "page fault", regs, code, TRAP_BRKPT,
 121                                         SIGSEGV) == NOTIFY_STOP)
 122                 return;
 123
 124         down_read(&mm->mmap_sem);
 125
 126         vma = find_vma_prev(mm, address, &prev_vma);
 127         if (!vma)
 128                 goto bad_area;
 129
 130         /* find_vma_prev() returns vma such that address < vma->vm_end or NULL */
 131         if (address < vma->vm_start)
 132                 goto check_expansion;
 133
 134   good_area:
 135         code = SEGV_ACCERR;
 136
 137         /* OK, we've got a good vm_area for this memory area.  Check the access permissions: */
 138
 139 #       define VM_READ_BIT      0
 140 #       define VM_WRITE_BIT     1
 141 #       define VM_EXEC_BIT      2
 142
 143 #       if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
 144             || (1 << VM_EXEC_BIT) != VM_EXEC)
 145 #               error File is out of sync with <linux/mm.h>.  Please update.
 146 #       endif
 147
 148         if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
 149                 goto bad_area;
 150
 151         mask = (  (((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
 152                 | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
 153
 154         if ((vma->vm_flags & mask) != mask)
 155                 goto bad_area;
 156
 157   survive:
 158         /*
 159          * If for any reason at all we couldn't handle the fault, make
 160          * sure we exit gracefully rather than endlessly redo the
 161          * fault.
 162          */
 163         switch (handle_mm_fault(mm, vma, address, (mask & VM_WRITE) != 0)) {
 164               case VM_FAULT_MINOR:
 165                 ++current->min_flt;
 166                 break;
 167               case VM_FAULT_MAJOR:
 168                 ++current->maj_flt;
 169                 break;
 170               case VM_FAULT_SIGBUS:
 171                 /*
 172                  * We ran out of memory, or some other thing happened
 173                  * to us that made us unable to handle the page fault
 174                  * gracefully.
 175                  */
 176                 signal = SIGBUS;
 177                 goto bad_area;
 178               case VM_FAULT_OOM:
 179                 goto out_of_memory;
 180               default:
 181                 BUG();
 182         }
 183         up_read(&mm->mmap_sem);
 184         return;
 185
 186   check_expansion:
 187         if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
 188                 if (!(vma->vm_flags & VM_GROWSDOWN))
 189                         goto bad_area;
 190                 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
 191                     || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
 192                         goto bad_area;
 193                 if (expand_stack(vma, address))
 194                         goto bad_area;
 195         } else {
 196                 vma = prev_vma;
 197                 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
 198                     || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
 199                         goto bad_area;
 200                 /*
 201                  * Since the register backing store is accessed sequentially,
 202                  * we disallow growing it by more than a page at a time.
 203                  */
 204                 if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
 205                         goto bad_area;
 206                 if (expand_upwards(vma, address))
 207                         goto bad_area;
 208         }
 209         goto good_area;
 210
 211   bad_area:
 212         up_read(&mm->mmap_sem);
 213 #ifdef CONFIG_VIRTUAL_MEM_MAP
 214   bad_area_no_up:
 215 #endif
 216         if ((isr & IA64_ISR_SP)
 217             || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
 218         {
 219                 /*
 220                  * This fault was due to a speculative load or lfetch.fault, set the "ed"
 221                  * bit in the psr to ensure forward progress.  (Target register will get a
 222                  * NaT for ld.s, lfetch will be canceled.)
 223                  */
 224                 ia64_psr(regs)->ed = 1;
 225                 return;
 226         }
 227         if (user_mode(regs)) {
 228                 si.si_signo = signal;
 229                 si.si_errno = 0;
 230                 si.si_code = code;
 231                 si.si_addr = (void __user *) address;
 232                 si.si_isr = isr;
 233                 si.si_flags = __ISR_VALID;
 234                 force_sig_info(signal, &si, current);
 235                 return;
 236         }
 237
 238   no_context:
 239         if ((isr & IA64_ISR_SP)
 240             || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
 241         {
 242                 /*
 243                  * This fault was due to a speculative load or lfetch.fault, set the "ed"
 244                  * bit in the psr to ensure forward progress.  (Target register will get a
 245                  * NaT for ld.s, lfetch will be canceled.)
 246                  */
 247                 ia64_psr(regs)->ed = 1;
 248                 return;
 249         }
 250
 251         /*
 252          * Since we have no vma's for region 5, we might get here even if the address is
 253          * valid, due to the VHPT walker inserting a non present translation that becomes
 254          * stale. If that happens, the non present fault handler already purged the stale
 255          * translation, which fixed the problem. So, we check to see if the translation is
 256          * valid, and return if it is.
 257          */
 258         if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
 259                 return;
 260
 261         if (ia64_done_with_exception(regs))
 262                 return;
 263
 264         /*
 265          * Oops. The kernel tried to access some bad page. We'll have to terminate things
 266          * with extreme prejudice.
 267          */
 268         bust_spinlocks(1);
 269
 270         if (address < PAGE_SIZE)
 271                 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
 272         else
 273                 printk(KERN_ALERT "Unable to handle kernel paging request at "
 274                        "virtual address %016lx\n", address);
 275         die("Oops", regs, isr);
 276         bust_spinlocks(0);
 277         do_exit(SIGKILL);
 278         return;
 279
 280   out_of_memory:
 281         up_read(&mm->mmap_sem);
 282         if (is_init(current)) {
 283                 yield();
 284                 down_read(&mm->mmap_sem);
 285                 goto survive;
 286         }
 287         printk(KERN_CRIT "VM: killing process %s\n", current->comm);
 288         if (user_mode(regs))
 289                 do_exit(SIGKILL);
 290         goto no_context;
 291 }