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/extable.h>
  10 #include <linux/wait.h>
  11 #include <linux/sched/signal.h>
  12 #include <linux/uaccess.h>
  13 #include <arch/system.h>
  14
  15 extern int find_fixup_code(struct pt_regs *);
  16 extern void die_if_kernel(const char *, struct pt_regs *, long);
  17 extern void show_registers(struct pt_regs *regs);
  18
  19 /* debug of low-level TLB reload */
  20 #undef DEBUG
  21
  22 #ifdef DEBUG
  23 #define D(x) x
  24 #else
  25 #define D(x)
  26 #endif
  27
  28 /* debug of higher-level faults */
  29 #define DPG(x)
  30
  31 /* current active page directory */
  32
  33 DEFINE_PER_CPU(pgd_t *, current_pgd);
  34 unsigned long cris_signal_return_page;
  35
  36 /*
  37  * This routine handles page faults.  It determines the address,
  38  * and the problem, and then passes it off to one of the appropriate
  39  * routines.
  40  *
  41  * Notice that the address we're given is aligned to the page the fault
  42  * occurred in, since we only get the PFN in R_MMU_CAUSE not the complete
  43  * address.
  44  *
  45  * error_code:
  46  *      bit 0 == 0 means no page found, 1 means protection fault
  47  *      bit 1 == 0 means read, 1 means write
  48  *
  49  * If this routine detects a bad access, it returns 1, otherwise it
  50  * returns 0.
  51  */
  52
  53 asmlinkage void
  54 do_page_fault(unsigned long address, struct pt_regs *regs,
  55               int protection, int writeaccess)
  56 {
  57         struct task_struct *tsk;
  58         struct mm_struct *mm;
  59         struct vm_area_struct * vma;
  60         siginfo_t info;
  61         int fault;
  62         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
  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, have pagefaults disabled or have no
 114          * user context, we must not take the fault.
 115          */
 116
 117         if (faulthandler_disabled() || !mm)
 118                 goto no_context;
 119
 120         if (user_mode(regs))
 121                 flags |= FAULT_FLAG_USER;
 122 retry:
 123         down_read(&mm->mmap_sem);
 124         vma = find_vma(mm, address);
 125         if (!vma)
 126                 goto bad_area;
 127         if (vma->vm_start <= address)
 128                 goto good_area;
 129         if (!(vma->vm_flags & VM_GROWSDOWN))
 130                 goto bad_area;
 131         if (user_mode(regs)) {
 132                 /*
 133                  * accessing the stack below usp is always a bug.
 134                  * we get page-aligned addresses so we can only check
 135                  * if we're within a page from usp, but that might be
 136                  * enough to catch brutal errors at least.
 137                  */
 138                 if (address + PAGE_SIZE < rdusp())
 139                         goto bad_area;
 140         }
 141         if (expand_stack(vma, address))
 142                 goto bad_area;
 143
 144         /*
 145          * Ok, we have a good vm_area for this memory access, so
 146          * we can handle it..
 147          */
 148
 149  good_area:
 150         info.si_code = SEGV_ACCERR;
 151
 152         /* first do some preliminary protection checks */
 153
 154         if (writeaccess == 2){
 155                 if (!(vma->vm_flags & VM_EXEC))
 156                         goto bad_area;
 157         } else if (writeaccess == 1) {
 158                 if (!(vma->vm_flags & VM_WRITE))
 159                         goto bad_area;
 160                 flags |= FAULT_FLAG_WRITE;
 161         } else {
 162                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 163                         goto bad_area;
 164         }
 165
 166         /*
 167          * If for any reason at all we couldn't handle the fault,
 168          * make sure we exit gracefully rather than endlessly redo
 169          * the fault.
 170          */
 171
 172         fault = handle_mm_fault(vma, address, flags);
 173
 174         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 175                 return;
 176
 177         if (unlikely(fault & VM_FAULT_ERROR)) {
 178                 if (fault & VM_FAULT_OOM)
 179                         goto out_of_memory;
 180                 else if (fault & VM_FAULT_SIGSEGV)
 181                         goto bad_area;
 182                 else if (fault & VM_FAULT_SIGBUS)
 183                         goto do_sigbus;
 184                 BUG();
 185         }
 186
 187         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 188                 if (fault & VM_FAULT_MAJOR)
 189                         tsk->maj_flt++;
 190                 else
 191                         tsk->min_flt++;
 192                 if (fault & VM_FAULT_RETRY) {
 193                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 194                         flags |= FAULT_FLAG_TRIED;
 195
 196                         /*
 197                          * No need to up_read(&mm->mmap_sem) as we would
 198                          * have already released it in __lock_page_or_retry
 199                          * in mm/filemap.c.
 200                          */
 201
 202                         goto retry;
 203                 }
 204         }
 205
 206         up_read(&mm->mmap_sem);
 207         return;
 208
 209         /*
 210          * Something tried to access memory that isn't in our memory map..
 211          * Fix it, but check if it's kernel or user first..
 212          */
 213
 214  bad_area:
 215         up_read(&mm->mmap_sem);
 216
 217  bad_area_nosemaphore:
 218         DPG(show_registers(regs));
 219
 220         /* User mode accesses just cause a SIGSEGV */
 221
 222         if (user_mode(regs)) {
 223 #ifdef CONFIG_NO_SEGFAULT_TERMINATION
 224                 DECLARE_WAIT_QUEUE_HEAD(wq);
 225 #endif
 226                 printk(KERN_NOTICE "%s (pid %d) segfaults for page "
 227                         "address %08lx at pc %08lx\n",
 228                         tsk->comm, tsk->pid,
 229                         address, instruction_pointer(regs));
 230
 231                 /* With DPG on, we've already dumped registers above.  */
 232                 DPG(if (0))
 233                         show_registers(regs);
 234
 235 #ifdef CONFIG_NO_SEGFAULT_TERMINATION
 236                 wait_event_interruptible(wq, 0 == 1);
 237 #else
 238                 info.si_signo = SIGSEGV;
 239                 info.si_errno = 0;
 240                 /* info.si_code has been set above */
 241                 info.si_addr = (void *)address;
 242                 force_sig_info(SIGSEGV, &info, tsk);
 243 #endif
 244                 return;
 245         }
 246
 247  no_context:
 248
 249         /* Are we prepared to handle this kernel fault?
 250          *
 251          * (The kernel has valid exception-points in the source
 252          *  when it accesses user-memory. When it fails in one
 253          *  of those points, we find it in a table and do a jump
 254          *  to some fixup code that loads an appropriate error
 255          *  code)
 256          */
 257
 258         if (find_fixup_code(regs))
 259                 return;
 260
 261         /*
 262          * Oops. The kernel tried to access some bad page. We'll have to
 263          * terminate things with extreme prejudice.
 264          */
 265
 266         if (!oops_in_progress) {
 267                 oops_in_progress = 1;
 268                 if ((unsigned long) (address) < PAGE_SIZE)
 269                         printk(KERN_ALERT "Unable to handle kernel NULL "
 270                                 "pointer dereference");
 271                 else
 272                         printk(KERN_ALERT "Unable to handle kernel access"
 273                                 " at virtual address %08lx\n", address);
 274
 275                 die_if_kernel("Oops", regs, (writeaccess << 1) | protection);
 276                 oops_in_progress = 0;
 277         }
 278
 279         do_exit(SIGKILL);
 280
 281         /*
 282          * We ran out of memory, or some other thing happened to us that made
 283          * us unable to handle the page fault gracefully.
 284          */
 285
 286  out_of_memory:
 287         up_read(&mm->mmap_sem);
 288         if (!user_mode(regs))
 289                 goto no_context;
 290         pagefault_out_of_memory();
 291         return;
 292
 293  do_sigbus:
 294         up_read(&mm->mmap_sem);
 295
 296         /*
 297          * Send a sigbus, regardless of whether we were in kernel
 298          * or user mode.
 299          */
 300         info.si_signo = SIGBUS;
 301         info.si_errno = 0;
 302         info.si_code = BUS_ADRERR;
 303         info.si_addr = (void *)address;
 304         force_sig_info(SIGBUS, &info, tsk);
 305
 306         /* Kernel mode? Handle exceptions or die */
 307         if (!user_mode(regs))
 308                 goto no_context;
 309         return;
 310
 311 vmalloc_fault:
 312         {
 313                 /*
 314                  * Synchronize this task's top level page-table
 315                  * with the 'reference' page table.
 316                  *
 317                  * Use current_pgd instead of tsk->active_mm->pgd
 318                  * since the latter might be unavailable if this
 319                  * code is executed in a misfortunately run irq
 320                  * (like inside schedule() between switch_mm and
 321                  *  switch_to...).
 322                  */
 323
 324                 int offset = pgd_index(address);
 325                 pgd_t *pgd, *pgd_k;
 326                 pud_t *pud, *pud_k;
 327                 pmd_t *pmd, *pmd_k;
 328                 pte_t *pte_k;
 329
 330                 pgd = (pgd_t *)per_cpu(current_pgd, smp_processor_id()) + offset;
 331                 pgd_k = init_mm.pgd + offset;
 332
 333                 /* Since we're two-level, we don't need to do both
 334                  * set_pgd and set_pmd (they do the same thing). If
 335                  * we go three-level at some point, do the right thing
 336                  * with pgd_present and set_pgd here.
 337                  *
 338                  * Also, since the vmalloc area is global, we don't
 339                  * need to copy individual PTE's, it is enough to
 340                  * copy the pgd pointer into the pte page of the
 341                  * root task. If that is there, we'll find our pte if
 342                  * it exists.
 343                  */
 344
 345                 pud = pud_offset(pgd, address);
 346                 pud_k = pud_offset(pgd_k, address);
 347                 if (!pud_present(*pud_k))
 348                         goto no_context;
 349
 350                 pmd = pmd_offset(pud, address);
 351                 pmd_k = pmd_offset(pud_k, address);
 352
 353                 if (!pmd_present(*pmd_k))
 354                         goto bad_area_nosemaphore;
 355
 356                 set_pmd(pmd, *pmd_k);
 357
 358                 /* Make sure the actual PTE exists as well to
 359                  * catch kernel vmalloc-area accesses to non-mapped
 360                  * addresses. If we don't do this, this will just
 361                  * silently loop forever.
 362                  */
 363
 364                 pte_k = pte_offset_kernel(pmd_k, address);
 365                 if (!pte_present(*pte_k))
 366                         goto no_context;
 367
 368                 return;
 369         }
 370 }
 371
 372 /* Find fixup code. */
 373 int
 374 find_fixup_code(struct pt_regs *regs)
 375 {
 376         const struct exception_table_entry *fixup;
 377         /* in case of delay slot fault (v32) */
 378         unsigned long ip = (instruction_pointer(regs) & ~0x1);
 379
 380         fixup = search_exception_tables(ip);
 381         if (fixup != 0) {
 382                 /* Adjust the instruction pointer in the stackframe. */
 383                 instruction_pointer(regs) = fixup->fixup;
 384                 arch_fixup(regs);
 385                 return 1;
 386         }
 387
 388         return 0;
 389 }