arch/mn10300/mm/fault.c

   1 /* MN10300 MMU Fault handler
   2  *
   3  * Copyright (C) 2007 Matsushita Electric Industrial Co., Ltd.
   4  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
   5  * Modified by David Howells (dhowells@redhat.com)
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public Licence
   9  * as published by the Free Software Foundation; either version
  10  * 2 of the Licence, or (at your option) any later version.
  11  */
  12
  13 #include <linux/signal.h>
  14 #include <linux/sched.h>
  15 #include <linux/kernel.h>
  16 #include <linux/errno.h>
  17 #include <linux/string.h>
  18 #include <linux/types.h>
  19 #include <linux/ptrace.h>
  20 #include <linux/mman.h>
  21 #include <linux/mm.h>
  22 #include <linux/smp.h>
  23 #include <linux/interrupt.h>
  24 #include <linux/init.h>
  25 #include <linux/vt_kern.h>              /* For unblank_screen() */
  26
  27 #include <asm/uaccess.h>
  28 #include <asm/pgalloc.h>
  29 #include <asm/hardirq.h>
  30 #include <asm/cpu-regs.h>
  31 #include <asm/debugger.h>
  32 #include <asm/gdb-stub.h>
  33
  34 /*
  35  * Unlock any spinlocks which will prevent us from getting the
  36  * message out
  37  */
  38 void bust_spinlocks(int yes)
  39 {
  40         if (yes) {
  41                 oops_in_progress = 1;
  42         } else {
  43                 int loglevel_save = console_loglevel;
  44 #ifdef CONFIG_VT
  45                 unblank_screen();
  46 #endif
  47                 oops_in_progress = 0;
  48                 /*
  49                  * OK, the message is on the console.  Now we call printk()
  50                  * without oops_in_progress set so that printk will give klogd
  51                  * a poke.  Hold onto your hats...
  52                  */
  53                 console_loglevel = 15;  /* NMI oopser may have shut the console
  54                                          * up */
  55                 printk(" ");
  56                 console_loglevel = loglevel_save;
  57         }
  58 }
  59
  60 void do_BUG(const char *file, int line)
  61 {
  62         bust_spinlocks(1);
  63         printk(KERN_EMERG "------------[ cut here ]------------\n");
  64         printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
  65 }
  66
  67 #if 0
  68 static void print_pagetable_entries(pgd_t *pgdir, unsigned long address)
  69 {
  70         pgd_t *pgd;
  71         pmd_t *pmd;
  72         pte_t *pte;
  73
  74         pgd = pgdir + __pgd_offset(address);
  75         printk(KERN_DEBUG "pgd entry %p: %016Lx\n",
  76                pgd, (long long) pgd_val(*pgd));
  77
  78         if (!pgd_present(*pgd)) {
  79                 printk(KERN_DEBUG "... pgd not present!\n");
  80                 return;
  81         }
  82         pmd = pmd_offset(pgd, address);
  83         printk(KERN_DEBUG "pmd entry %p: %016Lx\n",
  84                pmd, (long long)pmd_val(*pmd));
  85
  86         if (!pmd_present(*pmd)) {
  87                 printk(KERN_DEBUG "... pmd not present!\n");
  88                 return;
  89         }
  90         pte = pte_offset(pmd, address);
  91         printk(KERN_DEBUG "pte entry %p: %016Lx\n",
  92                pte, (long long) pte_val(*pte));
  93
  94         if (!pte_present(*pte))
  95                 printk(KERN_DEBUG "... pte not present!\n");
  96 }
  97 #endif
  98
  99 /*
 100  * This routine handles page faults.  It determines the address,
 101  * and the problem, and then passes it off to one of the appropriate
 102  * routines.
 103  *
 104  * fault_code:
 105  * - LSW: either MMUFCR_IFC or MMUFCR_DFC as appropriate
 106  * - MSW: 0 if data access, 1 if instruction access
 107  * - bit 0: TLB miss flag
 108  * - bit 1: initial write
 109  * - bit 2: page invalid
 110  * - bit 3: protection violation
 111  * - bit 4: accessor (0=user 1=kernel)
 112  * - bit 5: 0=read 1=write
 113  * - bit 6-8: page protection spec
 114  * - bit 9: illegal address
 115  * - bit 16: 0=data 1=ins
 116  *
 117  */
 118 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long fault_code,
 119                               unsigned long address)
 120 {
 121         struct vm_area_struct *vma;
 122         struct task_struct *tsk;
 123         struct mm_struct *mm;
 124         unsigned long page;
 125         siginfo_t info;
 126         int fault;
 127         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 128
 129 #ifdef CONFIG_GDBSTUB
 130         /* handle GDB stub causing a fault */
 131         if (gdbstub_busy) {
 132                 gdbstub_exception(regs, TBR & TBR_INT_CODE);
 133                 return;
 134         }
 135 #endif
 136
 137 #if 0
 138         printk(KERN_DEBUG "--- do_page_fault(%p,%s:%04lx,%08lx)\n",
 139                regs,
 140                fault_code & 0x10000 ? "ins" : "data",
 141                fault_code & 0xffff, address);
 142 #endif
 143
 144         tsk = current;
 145
 146         /*
 147          * We fault-in kernel-space virtual memory on-demand. The
 148          * 'reference' page table is init_mm.pgd.
 149          *
 150          * NOTE! We MUST NOT take any locks for this case. We may
 151          * be in an interrupt or a critical region, and should
 152          * only copy the information from the master page table,
 153          * nothing more.
 154          *
 155          * This verifies that the fault happens in kernel space
 156          * and that the fault was a page not present (invalid) error
 157          */
 158         if (address >= VMALLOC_START && address < VMALLOC_END &&
 159             (fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_SR &&
 160             (fault_code & MMUFCR_xFC_PGINVAL) == MMUFCR_xFC_PGINVAL
 161             )
 162                 goto vmalloc_fault;
 163
 164         mm = tsk->mm;
 165         info.si_code = SEGV_MAPERR;
 166
 167         /*
 168          * If we're in an interrupt or have no user
 169          * context, we must not take the fault..
 170          */
 171         if (in_atomic() || !mm)
 172                 goto no_context;
 173
 174 retry:
 175         down_read(&mm->mmap_sem);
 176
 177         vma = find_vma(mm, address);
 178         if (!vma)
 179                 goto bad_area;
 180         if (vma->vm_start <= address)
 181                 goto good_area;
 182         if (!(vma->vm_flags & VM_GROWSDOWN))
 183                 goto bad_area;
 184
 185         if ((fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_USR) {
 186                 /* accessing the stack below the stack pointer is always a
 187                  * bug */
 188                 if ((address & PAGE_MASK) + 2 * PAGE_SIZE < regs->sp) {
 189 #if 0
 190                         printk(KERN_WARNING
 191                                "[%d] ### Access below stack @%lx (sp=%lx)\n",
 192                                current->pid, address, regs->sp);
 193                         printk(KERN_WARNING
 194                                "vma [%08x - %08x]\n",
 195                                vma->vm_start, vma->vm_end);
 196                         show_registers(regs);
 197                         printk(KERN_WARNING
 198                                "[%d] ### Code: [%08lx]"
 199                                " %02x %02x %02x %02x %02x %02x %02x %02x\n",
 200                                current->pid,
 201                                regs->pc,
 202                                ((u8 *) regs->pc)[0],
 203                                ((u8 *) regs->pc)[1],
 204                                ((u8 *) regs->pc)[2],
 205                                ((u8 *) regs->pc)[3],
 206                                ((u8 *) regs->pc)[4],
 207                                ((u8 *) regs->pc)[5],
 208                                ((u8 *) regs->pc)[6],
 209                                ((u8 *) regs->pc)[7]
 210                                );
 211 #endif
 212                         goto bad_area;
 213                 }
 214         }
 215
 216         if (expand_stack(vma, address))
 217                 goto bad_area;
 218
 219 /*
 220  * Ok, we have a good vm_area for this memory access, so
 221  * we can handle it..
 222  */
 223 good_area:
 224         info.si_code = SEGV_ACCERR;
 225         switch (fault_code & (MMUFCR_xFC_PGINVAL|MMUFCR_xFC_TYPE)) {
 226         default:        /* 3: write, present */
 227         case MMUFCR_xFC_TYPE_WRITE:
 228 #ifdef TEST_VERIFY_AREA
 229                 if ((fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_SR)
 230                         printk(KERN_DEBUG "WP fault at %08lx\n", regs->pc);
 231 #endif
 232                 /* write to absent page */
 233         case MMUFCR_xFC_PGINVAL | MMUFCR_xFC_TYPE_WRITE:
 234                 if (!(vma->vm_flags & VM_WRITE))
 235                         goto bad_area;
 236                 flags |= FAULT_FLAG_WRITE;
 237                 break;
 238
 239                 /* read from protected page */
 240         case MMUFCR_xFC_TYPE_READ:
 241                 goto bad_area;
 242
 243                 /* read from absent page present */
 244         case MMUFCR_xFC_PGINVAL | MMUFCR_xFC_TYPE_READ:
 245                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 246                         goto bad_area;
 247                 break;
 248         }
 249
 250         /*
 251          * If for any reason at all we couldn't handle the fault,
 252          * make sure we exit gracefully rather than endlessly redo
 253          * the fault.
 254          */
 255         fault = handle_mm_fault(mm, vma, address, flags);
 256
 257         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 258                 return;
 259
 260         if (unlikely(fault & VM_FAULT_ERROR)) {
 261                 if (fault & VM_FAULT_OOM)
 262                         goto out_of_memory;
 263                 else if (fault & VM_FAULT_SIGBUS)
 264                         goto do_sigbus;
 265                 BUG();
 266         }
 267         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 268                 if (fault & VM_FAULT_MAJOR)
 269                         current->maj_flt++;
 270                 else
 271                         current->min_flt++;
 272                 if (fault & VM_FAULT_RETRY) {
 273                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 274
 275                          /* No need to up_read(&mm->mmap_sem) as we would
 276                          * have already released it in __lock_page_or_retry
 277                          * in mm/filemap.c.
 278                          */
 279
 280                         goto retry;
 281                 }
 282         }
 283
 284         up_read(&mm->mmap_sem);
 285         return;
 286
 287 /*
 288  * Something tried to access memory that isn't in our memory map..
 289  * Fix it, but check if it's kernel or user first..
 290  */
 291 bad_area:
 292         up_read(&mm->mmap_sem);
 293
 294         /* User mode accesses just cause a SIGSEGV */
 295         if ((fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_USR) {
 296                 info.si_signo = SIGSEGV;
 297                 info.si_errno = 0;
 298                 /* info.si_code has been set above */
 299                 info.si_addr = (void *)address;
 300                 force_sig_info(SIGSEGV, &info, tsk);
 301                 return;
 302         }
 303
 304 no_context:
 305         /* Are we prepared to handle this kernel fault?  */
 306         if (fixup_exception(regs))
 307                 return;
 308
 309 /*
 310  * Oops. The kernel tried to access some bad page. We'll have to
 311  * terminate things with extreme prejudice.
 312  */
 313
 314         bust_spinlocks(1);
 315
 316         if (address < PAGE_SIZE)
 317                 printk(KERN_ALERT
 318                        "Unable to handle kernel NULL pointer dereference");
 319         else
 320                 printk(KERN_ALERT
 321                        "Unable to handle kernel paging request");
 322         printk(" at virtual address %08lx\n", address);
 323         printk(" printing pc:\n");
 324         printk(KERN_ALERT "%08lx\n", regs->pc);
 325
 326         debugger_intercept(fault_code & 0x00010000 ? EXCEP_IAERROR : EXCEP_DAERROR,
 327                            SIGSEGV, SEGV_ACCERR, regs);
 328
 329         page = PTBR;
 330         page = ((unsigned long *) __va(page))[address >> 22];
 331         printk(KERN_ALERT "*pde = %08lx\n", page);
 332         if (page & 1) {
 333                 page &= PAGE_MASK;
 334                 address &= 0x003ff000;
 335                 page = ((unsigned long *) __va(page))[address >> PAGE_SHIFT];
 336                 printk(KERN_ALERT "*pte = %08lx\n", page);
 337         }
 338
 339         die("Oops", regs, fault_code);
 340         do_exit(SIGKILL);
 341
 342 /*
 343  * We ran out of memory, or some other thing happened to us that made
 344  * us unable to handle the page fault gracefully.
 345  */
 346 out_of_memory:
 347         up_read(&mm->mmap_sem);
 348         if ((fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_USR) {
 349                 pagefault_out_of_memory();
 350                 return;
 351         }
 352         goto no_context;
 353
 354 do_sigbus:
 355         up_read(&mm->mmap_sem);
 356
 357         /*
 358          * Send a sigbus, regardless of whether we were in kernel
 359          * or user mode.
 360          */
 361         info.si_signo = SIGBUS;
 362         info.si_errno = 0;
 363         info.si_code = BUS_ADRERR;
 364         info.si_addr = (void *)address;
 365         force_sig_info(SIGBUS, &info, tsk);
 366
 367         /* Kernel mode? Handle exceptions or die */
 368         if ((fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_SR)
 369                 goto no_context;
 370         return;
 371
 372 vmalloc_fault:
 373         {
 374                 /*
 375                  * Synchronize this task's top level page-table
 376                  * with the 'reference' page table.
 377                  *
 378                  * Do _not_ use "tsk" here. We might be inside
 379                  * an interrupt in the middle of a task switch..
 380                  */
 381                 int index = pgd_index(address);
 382                 pgd_t *pgd, *pgd_k;
 383                 pud_t *pud, *pud_k;
 384                 pmd_t *pmd, *pmd_k;
 385                 pte_t *pte_k;
 386
 387                 pgd_k = init_mm.pgd + index;
 388
 389                 if (!pgd_present(*pgd_k))
 390                         goto no_context;
 391
 392                 pud_k = pud_offset(pgd_k, address);
 393                 if (!pud_present(*pud_k))
 394                         goto no_context;
 395
 396                 pmd_k = pmd_offset(pud_k, address);
 397                 if (!pmd_present(*pmd_k))
 398                         goto no_context;
 399
 400                 pgd = (pgd_t *) PTBR + index;
 401                 pud = pud_offset(pgd, address);
 402                 pmd = pmd_offset(pud, address);
 403                 set_pmd(pmd, *pmd_k);
 404
 405                 pte_k = pte_offset_kernel(pmd_k, address);
 406                 if (!pte_present(*pte_k))
 407                         goto no_context;
 408                 return;
 409         }
 410 }