arch/sh/mm/fault.c

   1 /*
   2  * Page fault handler for SH with an MMU.
   3  *
   4  *  Copyright (C) 1999  Niibe Yutaka
   5  *  Copyright (C) 2003 - 2007  Paul Mundt
   6  *
   7  *  Based on linux/arch/i386/mm/fault.c:
   8  *   Copyright (C) 1995  Linus Torvalds
   9  *
  10  * This file is subject to the terms and conditions of the GNU General Public
  11  * License.  See the file "COPYING" in the main directory of this archive
  12  * for more details.
  13  */
  14 #include <linux/kernel.h>
  15 #include <linux/mm.h>
  16 #include <linux/hardirq.h>
  17 #include <linux/kprobes.h>
  18 #include <asm/kdebug.h>
  19 #include <asm/system.h>
  20 #include <asm/mmu_context.h>
  21 #include <asm/tlbflush.h>
  22 #include <asm/kgdb.h>
  23
  24 #ifdef CONFIG_KPROBES
  25 ATOMIC_NOTIFIER_HEAD(notify_page_fault_chain);
  26
  27 /* Hook to register for page fault notifications */
  28 int register_page_fault_notifier(struct notifier_block *nb)
  29 {
  30         return atomic_notifier_chain_register(&notify_page_fault_chain, nb);
  31 }
  32
  33 int unregister_page_fault_notifier(struct notifier_block *nb)
  34 {
  35         return atomic_notifier_chain_unregister(&notify_page_fault_chain, nb);
  36 }
  37
  38 static inline int notify_page_fault(enum die_val val, struct pt_regs *regs,
  39                                     int trap, int sig)
  40 {
  41         struct die_args args = {
  42                 .regs = regs,
  43                 .trapnr = trap,
  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, struct pt_regs *regs,
  49                                     int trap, int sig)
  50 {
  51         return NOTIFY_DONE;
  52 }
  53 #endif
  54
  55 /*
  56  * This routine handles page faults.  It determines the address,
  57  * and the problem, and then passes it off to one of the appropriate
  58  * routines.
  59  */
  60 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
  61                                         unsigned long writeaccess,
  62                                         unsigned long address)
  63 {
  64         struct task_struct *tsk;
  65         struct mm_struct *mm;
  66         struct vm_area_struct * vma;
  67         unsigned long page;
  68         int si_code;
  69         siginfo_t info;
  70
  71         trace_hardirqs_on();
  72
  73         if (notify_page_fault(DIE_PAGE_FAULT, regs,
  74                               writeaccess, SIGSEGV) == NOTIFY_STOP)
  75                 return;
  76
  77         local_irq_enable();
  78
  79 #ifdef CONFIG_SH_KGDB
  80         if (kgdb_nofault && kgdb_bus_err_hook)
  81                 kgdb_bus_err_hook();
  82 #endif
  83
  84         tsk = current;
  85         mm = tsk->mm;
  86         si_code = SEGV_MAPERR;
  87
  88         if (unlikely(address >= TASK_SIZE)) {
  89                 /*
  90                  * Synchronize this task's top level page-table
  91                  * with the 'reference' page table.
  92                  *
  93                  * Do _not_ use "tsk" here. We might be inside
  94                  * an interrupt in the middle of a task switch..
  95                  */
  96                 int offset = pgd_index(address);
  97                 pgd_t *pgd, *pgd_k;
  98                 pud_t *pud, *pud_k;
  99                 pmd_t *pmd, *pmd_k;
 100
 101                 pgd = get_TTB() + offset;
 102                 pgd_k = swapper_pg_dir + offset;
 103
 104                 /* This will never happen with the folded page table. */
 105                 if (!pgd_present(*pgd)) {
 106                         if (!pgd_present(*pgd_k))
 107                                 goto bad_area_nosemaphore;
 108                         set_pgd(pgd, *pgd_k);
 109                         return;
 110                 }
 111
 112                 pud = pud_offset(pgd, address);
 113                 pud_k = pud_offset(pgd_k, address);
 114                 if (pud_present(*pud) || !pud_present(*pud_k))
 115                         goto bad_area_nosemaphore;
 116                 set_pud(pud, *pud_k);
 117
 118                 pmd = pmd_offset(pud, address);
 119                 pmd_k = pmd_offset(pud_k, address);
 120                 if (pmd_present(*pmd) || !pmd_present(*pmd_k))
 121                         goto bad_area_nosemaphore;
 122                 set_pmd(pmd, *pmd_k);
 123
 124                 return;
 125         }
 126
 127         /*
 128          * If we're in an interrupt or have no user
 129          * context, we must not take the fault..
 130          */
 131         if (in_atomic() || !mm)
 132                 goto no_context;
 133
 134         down_read(&mm->mmap_sem);
 135
 136         vma = find_vma(mm, address);
 137         if (!vma)
 138                 goto bad_area;
 139         if (vma->vm_start <= address)
 140                 goto good_area;
 141         if (!(vma->vm_flags & VM_GROWSDOWN))
 142                 goto bad_area;
 143         if (expand_stack(vma, address))
 144                 goto bad_area;
 145 /*
 146  * Ok, we have a good vm_area for this memory access, so
 147  * we can handle it..
 148  */
 149 good_area:
 150         si_code = SEGV_ACCERR;
 151         if (writeaccess) {
 152                 if (!(vma->vm_flags & VM_WRITE))
 153                         goto bad_area;
 154         } else {
 155                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 156                         goto bad_area;
 157         }
 158
 159         /*
 160          * If for any reason at all we couldn't handle the fault,
 161          * make sure we exit gracefully rather than endlessly redo
 162          * the fault.
 163          */
 164 survive:
 165         switch (handle_mm_fault(mm, vma, address, writeaccess)) {
 166                 case VM_FAULT_MINOR:
 167                         tsk->min_flt++;
 168                         break;
 169                 case VM_FAULT_MAJOR:
 170                         tsk->maj_flt++;
 171                         break;
 172                 case VM_FAULT_SIGBUS:
 173                         goto do_sigbus;
 174                 case VM_FAULT_OOM:
 175                         goto out_of_memory;
 176                 default:
 177                         BUG();
 178         }
 179
 180         up_read(&mm->mmap_sem);
 181         return;
 182
 183 /*
 184  * Something tried to access memory that isn't in our memory map..
 185  * Fix it, but check if it's kernel or user first..
 186  */
 187 bad_area:
 188         up_read(&mm->mmap_sem);
 189
 190 bad_area_nosemaphore:
 191         if (user_mode(regs)) {
 192                 info.si_signo = SIGSEGV;
 193                 info.si_errno = 0;
 194                 info.si_code = si_code;
 195                 info.si_addr = (void *) address;
 196                 force_sig_info(SIGSEGV, &info, tsk);
 197                 return;
 198         }
 199
 200 no_context:
 201         /* Are we prepared to handle this kernel fault?  */
 202         if (fixup_exception(regs))
 203                 return;
 204
 205 /*
 206  * Oops. The kernel tried to access some bad page. We'll have to
 207  * terminate things with extreme prejudice.
 208  *
 209  */
 210         if (address < PAGE_SIZE)
 211                 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference");
 212         else
 213                 printk(KERN_ALERT "Unable to handle kernel paging request");
 214         printk(" at virtual address %08lx\n", address);
 215         printk(KERN_ALERT "pc = %08lx\n", regs->pc);
 216         page = (unsigned long)get_TTB();
 217         if (page) {
 218                 page = ((unsigned long *) page)[address >> PGDIR_SHIFT];
 219                 printk(KERN_ALERT "*pde = %08lx\n", page);
 220                 if (page & _PAGE_PRESENT) {
 221                         page &= PAGE_MASK;
 222                         address &= 0x003ff000;
 223                         page = ((unsigned long *) __va(page))[address >> PAGE_SHIFT];
 224                         printk(KERN_ALERT "*pte = %08lx\n", page);
 225                 }
 226         }
 227         die("Oops", regs, writeaccess);
 228         do_exit(SIGKILL);
 229
 230 /*
 231  * We ran out of memory, or some other thing happened to us that made
 232  * us unable to handle the page fault gracefully.
 233  */
 234 out_of_memory:
 235         up_read(&mm->mmap_sem);
 236         if (is_init(current)) {
 237                 yield();
 238                 down_read(&mm->mmap_sem);
 239                 goto survive;
 240         }
 241         printk("VM: killing process %s\n", tsk->comm);
 242         if (user_mode(regs))
 243                 do_exit(SIGKILL);
 244         goto no_context;
 245
 246 do_sigbus:
 247         up_read(&mm->mmap_sem);
 248
 249         /*
 250          * Send a sigbus, regardless of whether we were in kernel
 251          * or user mode.
 252          */
 253         info.si_signo = SIGBUS;
 254         info.si_errno = 0;
 255         info.si_code = BUS_ADRERR;
 256         info.si_addr = (void *)address;
 257         force_sig_info(SIGBUS, &info, tsk);
 258
 259         /* Kernel mode? Handle exceptions or die */
 260         if (!user_mode(regs))
 261                 goto no_context;
 262 }
 263
 264 #ifdef CONFIG_SH_STORE_QUEUES
 265 /*
 266  * This is a special case for the SH-4 store queues, as pages for this
 267  * space still need to be faulted in before it's possible to flush the
 268  * store queue cache for writeout to the remapped region.
 269  */
 270 #define P3_ADDR_MAX             (P4SEG_STORE_QUE + 0x04000000)
 271 #else
 272 #define P3_ADDR_MAX             P4SEG
 273 #endif
 274
 275 /*
 276  * Called with interrupts disabled.
 277  */
 278 asmlinkage int __kprobes __do_page_fault(struct pt_regs *regs,
 279                                          unsigned long writeaccess,
 280                                          unsigned long address)
 281 {
 282         pgd_t *pgd;
 283         pud_t *pud;
 284         pmd_t *pmd;
 285         pte_t *pte;
 286         pte_t entry;
 287         struct mm_struct *mm = current->mm;
 288         spinlock_t *ptl;
 289         int ret = 1;
 290
 291 #ifdef CONFIG_SH_KGDB
 292         if (kgdb_nofault && kgdb_bus_err_hook)
 293                 kgdb_bus_err_hook();
 294 #endif
 295
 296         /*
 297          * We don't take page faults for P1, P2, and parts of P4, these
 298          * are always mapped, whether it be due to legacy behaviour in
 299          * 29-bit mode, or due to PMB configuration in 32-bit mode.
 300          */
 301         if (address >= P3SEG && address < P3_ADDR_MAX) {
 302                 pgd = pgd_offset_k(address);
 303                 mm = NULL;
 304         } else {
 305                 if (unlikely(address >= TASK_SIZE || !mm))
 306                         return 1;
 307
 308                 pgd = pgd_offset(mm, address);
 309         }
 310
 311         pud = pud_offset(pgd, address);
 312         if (pud_none_or_clear_bad(pud))
 313                 return 1;
 314         pmd = pmd_offset(pud, address);
 315         if (pmd_none_or_clear_bad(pmd))
 316                 return 1;
 317
 318         if (mm)
 319                 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
 320         else
 321                 pte = pte_offset_kernel(pmd, address);
 322
 323         entry = *pte;
 324         if (unlikely(pte_none(entry) || pte_not_present(entry)))
 325                 goto unlock;
 326         if (unlikely(writeaccess && !pte_write(entry)))
 327                 goto unlock;
 328
 329         if (writeaccess)
 330                 entry = pte_mkdirty(entry);
 331         entry = pte_mkyoung(entry);
 332
 333 #ifdef CONFIG_CPU_SH4
 334         /*
 335          * ITLB is not affected by "ldtlb" instruction.
 336          * So, we need to flush the entry by ourselves.
 337          */
 338         local_flush_tlb_one(get_asid(), address & PAGE_MASK);
 339 #endif
 340
 341         set_pte(pte, entry);
 342         update_mmu_cache(NULL, address, entry);
 343         ret = 0;
 344 unlock:
 345         if (mm)
 346                 pte_unmap_unlock(pte, ptl);
 347         return ret;
 348 }