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 #include <linux/prefetch.h>
  14 #include <linux/uaccess.h>
  15
  16 #include <asm/pgtable.h>
  17 #include <asm/processor.h>
  18
  19 extern int die(char *, struct pt_regs *, long);
  20
  21 #ifdef CONFIG_KPROBES
  22 static inline int notify_page_fault(struct pt_regs *regs, int trap)
  23 {
  24         int ret = 0;
  25
  26         if (!user_mode(regs)) {
  27                 /* kprobe_running() needs smp_processor_id() */
  28                 preempt_disable();
  29                 if (kprobe_running() && kprobe_fault_handler(regs, trap))
  30                         ret = 1;
  31                 preempt_enable();
  32         }
  33
  34         return ret;
  35 }
  36 #else
  37 static inline int notify_page_fault(struct pt_regs *regs, int trap)
  38 {
  39         return 0;
  40 }
  41 #endif
  42
  43 /*
  44  * Return TRUE if ADDRESS points at a page in the kernel's mapped segment
  45  * (inside region 5, on ia64) and that page is present.
  46  */
  47 static int
  48 mapped_kernel_page_is_present (unsigned long address)
  49 {
  50         pgd_t *pgd;
  51         pud_t *pud;
  52         pmd_t *pmd;
  53         pte_t *ptep, pte;
  54
  55         pgd = pgd_offset_k(address);
  56         if (pgd_none(*pgd) || pgd_bad(*pgd))
  57                 return 0;
  58
  59         pud = pud_offset(pgd, address);
  60         if (pud_none(*pud) || pud_bad(*pud))
  61                 return 0;
  62
  63         pmd = pmd_offset(pud, address);
  64         if (pmd_none(*pmd) || pmd_bad(*pmd))
  65                 return 0;
  66
  67         ptep = pte_offset_kernel(pmd, address);
  68         if (!ptep)
  69                 return 0;
  70
  71         pte = *ptep;
  72         return pte_present(pte);
  73 }
  74
  75 #       define VM_READ_BIT      0
  76 #       define VM_WRITE_BIT     1
  77 #       define VM_EXEC_BIT      2
  78
  79 void __kprobes
  80 ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs)
  81 {
  82         int signal = SIGSEGV, code = SEGV_MAPERR;
  83         struct vm_area_struct *vma, *prev_vma;
  84         struct mm_struct *mm = current->mm;
  85         struct siginfo si;
  86         unsigned long mask;
  87         int fault;
  88         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
  89
  90         mask = ((((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
  91                 | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
  92
  93         /* mmap_sem is performance critical.... */
  94         prefetchw(&mm->mmap_sem);
  95
  96         /*
  97          * If we're in an interrupt or have no user context, we must not take the fault..
  98          */
  99         if (faulthandler_disabled() || !mm)
 100                 goto no_context;
 101
 102 #ifdef CONFIG_VIRTUAL_MEM_MAP
 103         /*
 104          * If fault is in region 5 and we are in the kernel, we may already
 105          * have the mmap_sem (pfn_valid macro is called during mmap). There
 106          * is no vma for region 5 addr's anyway, so skip getting the semaphore
 107          * and go directly to the exception handling code.
 108          */
 109
 110         if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
 111                 goto bad_area_no_up;
 112 #endif
 113
 114         /*
 115          * This is to handle the kprobes on user space access instructions
 116          */
 117         if (notify_page_fault(regs, TRAP_BRKPT))
 118                 return;
 119
 120         if (user_mode(regs))
 121                 flags |= FAULT_FLAG_USER;
 122         if (mask & VM_WRITE)
 123                 flags |= FAULT_FLAG_WRITE;
 124 retry:
 125         down_read(&mm->mmap_sem);
 126
 127         vma = find_vma_prev(mm, address, &prev_vma);
 128         if (!vma && !prev_vma )
 129                 goto bad_area;
 130
 131         /*
 132          * find_vma_prev() returns vma such that address < vma->vm_end or NULL
 133          *
 134          * May find no vma, but could be that the last vm area is the
 135          * register backing store that needs to expand upwards, in
 136          * this case vma will be null, but prev_vma will ne non-null
 137          */
 138         if (( !vma && prev_vma ) || (address < vma->vm_start) )
 139                 goto check_expansion;
 140
 141   good_area:
 142         code = SEGV_ACCERR;
 143
 144         /* OK, we've got a good vm_area for this memory area.  Check the access permissions: */
 145
 146 #       if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
 147             || (1 << VM_EXEC_BIT) != VM_EXEC)
 148 #               error File is out of sync with <linux/mm.h>.  Please update.
 149 #       endif
 150
 151         if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
 152                 goto bad_area;
 153
 154         if ((vma->vm_flags & mask) != mask)
 155                 goto bad_area;
 156
 157         /*
 158          * If for any reason at all we couldn't handle the fault, make
 159          * sure we exit gracefully rather than endlessly redo the
 160          * fault.
 161          */
 162         fault = handle_mm_fault(mm, vma, address, flags);
 163
 164         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 165                 return;
 166
 167         if (unlikely(fault & VM_FAULT_ERROR)) {
 168                 /*
 169                  * We ran out of memory, or some other thing happened
 170                  * to us that made us unable to handle the page fault
 171                  * gracefully.
 172                  */
 173                 if (fault & VM_FAULT_OOM) {
 174                         goto out_of_memory;
 175                 } else if (fault & VM_FAULT_SIGSEGV) {
 176                         goto bad_area;
 177                 } else if (fault & VM_FAULT_SIGBUS) {
 178                         signal = SIGBUS;
 179                         goto bad_area;
 180                 }
 181                 BUG();
 182         }
 183
 184         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 185                 if (fault & VM_FAULT_MAJOR)
 186                         current->maj_flt++;
 187                 else
 188                         current->min_flt++;
 189                 if (fault & VM_FAULT_RETRY) {
 190                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 191                         flags |= FAULT_FLAG_TRIED;
 192
 193                          /* No need to up_read(&mm->mmap_sem) as we would
 194                          * have already released it in __lock_page_or_retry
 195                          * in mm/filemap.c.
 196                          */
 197
 198                         goto retry;
 199                 }
 200         }
 201
 202         up_read(&mm->mmap_sem);
 203         return;
 204
 205   check_expansion:
 206         if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
 207                 if (!vma)
 208                         goto bad_area;
 209                 if (!(vma->vm_flags & VM_GROWSDOWN))
 210                         goto bad_area;
 211                 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
 212                     || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
 213                         goto bad_area;
 214                 if (expand_stack(vma, address))
 215                         goto bad_area;
 216         } else {
 217                 vma = prev_vma;
 218                 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
 219                     || REGION_OFFSET(address) >= RGN_MAP_LIMIT)
 220                         goto bad_area;
 221                 /*
 222                  * Since the register backing store is accessed sequentially,
 223                  * we disallow growing it by more than a page at a time.
 224                  */
 225                 if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
 226                         goto bad_area;
 227                 if (expand_upwards(vma, address))
 228                         goto bad_area;
 229         }
 230         goto good_area;
 231
 232   bad_area:
 233         up_read(&mm->mmap_sem);
 234 #ifdef CONFIG_VIRTUAL_MEM_MAP
 235   bad_area_no_up:
 236 #endif
 237         if ((isr & IA64_ISR_SP)
 238             || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
 239         {
 240                 /*
 241                  * This fault was due to a speculative load or lfetch.fault, set the "ed"
 242                  * bit in the psr to ensure forward progress.  (Target register will get a
 243                  * NaT for ld.s, lfetch will be canceled.)
 244                  */
 245                 ia64_psr(regs)->ed = 1;
 246                 return;
 247         }
 248         if (user_mode(regs)) {
 249                 si.si_signo = signal;
 250                 si.si_errno = 0;
 251                 si.si_code = code;
 252                 si.si_addr = (void __user *) address;
 253                 si.si_isr = isr;
 254                 si.si_flags = __ISR_VALID;
 255                 force_sig_info(signal, &si, current);
 256                 return;
 257         }
 258
 259   no_context:
 260         if ((isr & IA64_ISR_SP)
 261             || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
 262         {
 263                 /*
 264                  * This fault was due to a speculative load or lfetch.fault, set the "ed"
 265                  * bit in the psr to ensure forward progress.  (Target register will get a
 266                  * NaT for ld.s, lfetch will be canceled.)
 267                  */
 268                 ia64_psr(regs)->ed = 1;
 269                 return;
 270         }
 271
 272         /*
 273          * Since we have no vma's for region 5, we might get here even if the address is
 274          * valid, due to the VHPT walker inserting a non present translation that becomes
 275          * stale. If that happens, the non present fault handler already purged the stale
 276          * translation, which fixed the problem. So, we check to see if the translation is
 277          * valid, and return if it is.
 278          */
 279         if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
 280                 return;
 281
 282         if (ia64_done_with_exception(regs))
 283                 return;
 284
 285         /*
 286          * Oops. The kernel tried to access some bad page. We'll have to terminate things
 287          * with extreme prejudice.
 288          */
 289         bust_spinlocks(1);
 290
 291         if (address < PAGE_SIZE)
 292                 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
 293         else
 294                 printk(KERN_ALERT "Unable to handle kernel paging request at "
 295                        "virtual address %016lx\n", address);
 296         if (die("Oops", regs, isr))
 297                 regs = NULL;
 298         bust_spinlocks(0);
 299         if (regs)
 300                 do_exit(SIGKILL);
 301         return;
 302
 303   out_of_memory:
 304         up_read(&mm->mmap_sem);
 305         if (!user_mode(regs))
 306                 goto no_context;
 307         pagefault_out_of_memory();
 308 }