arch/cris/mm/fault.c

   1 /*
   2  *  arch/cris/mm/fault.c
   3  *
   4  *  Copyright (C) 2000-2010  Axis Communications AB
   5  */
   6
   7 #include <linux/mm.h>
   8 #include <linux/interrupt.h>
   9 #include <linux/module.h>
  10 #include <linux/wait.h>
  11 #include <asm/uaccess.h>
  12 #include <arch/system.h>
  13
  14 extern int find_fixup_code(struct pt_regs *);
  15 extern void die_if_kernel(const char *, struct pt_regs *, long);
  16 extern void show_registers(struct pt_regs *regs);
  17
  18 /* debug of low-level TLB reload */
  19 #undef DEBUG
  20
  21 #ifdef DEBUG
  22 #define D(x) x
  23 #else
  24 #define D(x)
  25 #endif
  26
  27 /* debug of higher-level faults */
  28 #define DPG(x)
  29
  30 /* current active page directory */
  31
  32 DEFINE_PER_CPU(pgd_t *, current_pgd);
  33 unsigned long cris_signal_return_page;
  34
  35 /*
  36  * This routine handles page faults.  It determines the address,
  37  * and the problem, and then passes it off to one of the appropriate
  38  * routines.
  39  *
  40  * Notice that the address we're given is aligned to the page the fault
  41  * occurred in, since we only get the PFN in R_MMU_CAUSE not the complete
  42  * address.
  43  *
  44  * error_code:
  45  *      bit 0 == 0 means no page found, 1 means protection fault
  46  *      bit 1 == 0 means read, 1 means write
  47  *
  48  * If this routine detects a bad access, it returns 1, otherwise it
  49  * returns 0.
  50  */
  51
  52 asmlinkage void
  53 do_page_fault(unsigned long address, struct pt_regs *regs,
  54               int protection, int writeaccess)
  55 {
  56         struct task_struct *tsk;
  57         struct mm_struct *mm;
  58         struct vm_area_struct * vma;
  59         siginfo_t info;
  60         int fault;
  61         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
  62                                 ((writeaccess & 1) ? FAULT_FLAG_WRITE : 0);
  63
  64         D(printk(KERN_DEBUG
  65                  "Page fault for %lX on %X at %lX, prot %d write %d\n",
  66                  address, smp_processor_id(), instruction_pointer(regs),
  67                  protection, writeaccess));
  68
  69         tsk = current;
  70
  71         /*
  72          * We fault-in kernel-space virtual memory on-demand. The
  73          * 'reference' page table is init_mm.pgd.
  74          *
  75          * NOTE! We MUST NOT take any locks for this case. We may
  76          * be in an interrupt or a critical region, and should
  77          * only copy the information from the master page table,
  78          * nothing more.
  79          *
  80          * NOTE2: This is done so that, when updating the vmalloc
  81          * mappings we don't have to walk all processes pgdirs and
  82          * add the high mappings all at once. Instead we do it as they
  83          * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
  84          * bit set so sometimes the TLB can use a lingering entry.
  85          *
  86          * This verifies that the fault happens in kernel space
  87          * and that the fault was not a protection error (error_code & 1).
  88          */
  89
  90         if (address >= VMALLOC_START &&
  91             !protection &&
  92             !user_mode(regs))
  93                 goto vmalloc_fault;
  94
  95         /* When stack execution is not allowed we store the signal
  96          * trampolines in the reserved cris_signal_return_page.
  97          * Handle this in the exact same way as vmalloc (we know
  98          * that the mapping is there and is valid so no need to
  99          * call handle_mm_fault).
 100          */
 101         if (cris_signal_return_page &&
 102             address == cris_signal_return_page &&
 103             !protection && user_mode(regs))
 104                 goto vmalloc_fault;
 105
 106         /* we can and should enable interrupts at this point */
 107         local_irq_enable();
 108
 109         mm = tsk->mm;
 110         info.si_code = SEGV_MAPERR;
 111
 112         /*
 113          * If we're in an interrupt or "atomic" operation or have no
 114          * user context, we must not take the fault.
 115          */
 116
 117         if (in_atomic() || !mm)
 118                 goto no_context;
 119
 120 retry:
 121         down_read(&mm->mmap_sem);
 122         vma = find_vma(mm, address);
 123         if (!vma)
 124                 goto bad_area;
 125         if (vma->vm_start <= address)
 126                 goto good_area;
 127         if (!(vma->vm_flags & VM_GROWSDOWN))
 128                 goto bad_area;
 129         if (user_mode(regs)) {
 130                 /*
 131                  * accessing the stack below usp is always a bug.
 132                  * we get page-aligned addresses so we can only check
 133                  * if we're within a page from usp, but that might be
 134                  * enough to catch brutal errors at least.
 135                  */
 136                 if (address + PAGE_SIZE < rdusp())
 137                         goto bad_area;
 138         }
 139         if (expand_stack(vma, address))
 140                 goto bad_area;
 141
 142         /*
 143          * Ok, we have a good vm_area for this memory access, so
 144          * we can handle it..
 145          */
 146
 147  good_area:
 148         info.si_code = SEGV_ACCERR;
 149
 150         /* first do some preliminary protection checks */
 151
 152         if (writeaccess == 2){
 153                 if (!(vma->vm_flags & VM_EXEC))
 154                         goto bad_area;
 155         } else if (writeaccess == 1) {
 156                 if (!(vma->vm_flags & VM_WRITE))
 157                         goto bad_area;
 158         } else {
 159                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 160                         goto bad_area;
 161         }
 162
 163         /*
 164          * If for any reason at all we couldn't handle the fault,
 165          * make sure we exit gracefully rather than endlessly redo
 166          * the fault.
 167          */
 168
 169         fault = handle_mm_fault(mm, vma, address, flags);
 170
 171         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 172                 return;
 173
 174         if (unlikely(fault & VM_FAULT_ERROR)) {
 175                 if (fault & VM_FAULT_OOM)
 176                         goto out_of_memory;
 177                 else if (fault & VM_FAULT_SIGBUS)
 178                         goto do_sigbus;
 179                 BUG();
 180         }
 181
 182         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 183                 if (fault & VM_FAULT_MAJOR)
 184                         tsk->maj_flt++;
 185                 else
 186                         tsk->min_flt++;
 187                 if (fault & VM_FAULT_RETRY) {
 188                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 189                         flags |= FAULT_FLAG_TRIED;
 190
 191                         /*
 192                          * No need to up_read(&mm->mmap_sem) as we would
 193                          * have already released it in __lock_page_or_retry
 194                          * in mm/filemap.c.
 195                          */
 196
 197                         goto retry;
 198                 }
 199         }
 200
 201         up_read(&mm->mmap_sem);
 202         return;
 203
 204         /*
 205          * Something tried to access memory that isn't in our memory map..
 206          * Fix it, but check if it's kernel or user first..
 207          */
 208
 209  bad_area:
 210         up_read(&mm->mmap_sem);
 211
 212  bad_area_nosemaphore:
 213         DPG(show_registers(regs));
 214
 215         /* User mode accesses just cause a SIGSEGV */
 216
 217         if (user_mode(regs)) {
 218                 printk(KERN_NOTICE "%s (pid %d) segfaults for page "
 219                         "address %08lx at pc %08lx\n",
 220                         tsk->comm, tsk->pid,
 221                         address, instruction_pointer(regs));
 222
 223                 /* With DPG on, we've already dumped registers above.  */
 224                 DPG(if (0))
 225                         show_registers(regs);
 226
 227 #ifdef CONFIG_NO_SEGFAULT_TERMINATION
 228                 DECLARE_WAIT_QUEUE_HEAD(wq);
 229                 wait_event_interruptible(wq, 0 == 1);
 230 #else
 231                 info.si_signo = SIGSEGV;
 232                 info.si_errno = 0;
 233                 /* info.si_code has been set above */
 234                 info.si_addr = (void *)address;
 235                 force_sig_info(SIGSEGV, &info, tsk);
 236 #endif
 237                 return;
 238         }
 239
 240  no_context:
 241
 242         /* Are we prepared to handle this kernel fault?
 243          *
 244          * (The kernel has valid exception-points in the source
 245          *  when it accesses user-memory. When it fails in one
 246          *  of those points, we find it in a table and do a jump
 247          *  to some fixup code that loads an appropriate error
 248          *  code)
 249          */
 250
 251         if (find_fixup_code(regs))
 252                 return;
 253
 254         /*
 255          * Oops. The kernel tried to access some bad page. We'll have to
 256          * terminate things with extreme prejudice.
 257          */
 258
 259         if (!oops_in_progress) {
 260                 oops_in_progress = 1;
 261                 if ((unsigned long) (address) < PAGE_SIZE)
 262                         printk(KERN_ALERT "Unable to handle kernel NULL "
 263                                 "pointer dereference");
 264                 else
 265                         printk(KERN_ALERT "Unable to handle kernel access"
 266                                 " at virtual address %08lx\n", address);
 267
 268                 die_if_kernel("Oops", regs, (writeaccess << 1) | protection);
 269                 oops_in_progress = 0;
 270         }
 271
 272         do_exit(SIGKILL);
 273
 274         /*
 275          * We ran out of memory, or some other thing happened to us that made
 276          * us unable to handle the page fault gracefully.
 277          */
 278
 279  out_of_memory:
 280         up_read(&mm->mmap_sem);
 281         if (!user_mode(regs))
 282                 goto no_context;
 283         pagefault_out_of_memory();
 284         return;
 285
 286  do_sigbus:
 287         up_read(&mm->mmap_sem);
 288
 289         /*
 290          * Send a sigbus, regardless of whether we were in kernel
 291          * or user mode.
 292          */
 293         info.si_signo = SIGBUS;
 294         info.si_errno = 0;
 295         info.si_code = BUS_ADRERR;
 296         info.si_addr = (void *)address;
 297         force_sig_info(SIGBUS, &info, tsk);
 298
 299         /* Kernel mode? Handle exceptions or die */
 300         if (!user_mode(regs))
 301                 goto no_context;
 302         return;
 303
 304 vmalloc_fault:
 305         {
 306                 /*
 307                  * Synchronize this task's top level page-table
 308                  * with the 'reference' page table.
 309                  *
 310                  * Use current_pgd instead of tsk->active_mm->pgd
 311                  * since the latter might be unavailable if this
 312                  * code is executed in a misfortunately run irq
 313                  * (like inside schedule() between switch_mm and
 314                  *  switch_to...).
 315                  */
 316
 317                 int offset = pgd_index(address);
 318                 pgd_t *pgd, *pgd_k;
 319                 pud_t *pud, *pud_k;
 320                 pmd_t *pmd, *pmd_k;
 321                 pte_t *pte_k;
 322
 323                 pgd = (pgd_t *)per_cpu(current_pgd, smp_processor_id()) + offset;
 324                 pgd_k = init_mm.pgd + offset;
 325
 326                 /* Since we're two-level, we don't need to do both
 327                  * set_pgd and set_pmd (they do the same thing). If
 328                  * we go three-level at some point, do the right thing
 329                  * with pgd_present and set_pgd here.
 330                  *
 331                  * Also, since the vmalloc area is global, we don't
 332                  * need to copy individual PTE's, it is enough to
 333                  * copy the pgd pointer into the pte page of the
 334                  * root task. If that is there, we'll find our pte if
 335                  * it exists.
 336                  */
 337
 338                 pud = pud_offset(pgd, address);
 339                 pud_k = pud_offset(pgd_k, address);
 340                 if (!pud_present(*pud_k))
 341                         goto no_context;
 342
 343                 pmd = pmd_offset(pud, address);
 344                 pmd_k = pmd_offset(pud_k, address);
 345
 346                 if (!pmd_present(*pmd_k))
 347                         goto bad_area_nosemaphore;
 348
 349                 set_pmd(pmd, *pmd_k);
 350
 351                 /* Make sure the actual PTE exists as well to
 352                  * catch kernel vmalloc-area accesses to non-mapped
 353                  * addresses. If we don't do this, this will just
 354                  * silently loop forever.
 355                  */
 356
 357                 pte_k = pte_offset_kernel(pmd_k, address);
 358                 if (!pte_present(*pte_k))
 359                         goto no_context;
 360
 361                 return;
 362         }
 363 }
 364
 365 /* Find fixup code. */
 366 int
 367 find_fixup_code(struct pt_regs *regs)
 368 {
 369         const struct exception_table_entry *fixup;
 370         /* in case of delay slot fault (v32) */
 371         unsigned long ip = (instruction_pointer(regs) & ~0x1);
 372
 373         fixup = search_exception_tables(ip);
 374         if (fixup != 0) {
 375                 /* Adjust the instruction pointer in the stackframe. */
 376                 instruction_pointer(regs) = fixup->fixup;
 377                 arch_fixup(regs);
 378                 return 1;
 379         }
 380
 381         return 0;
 382 }