arch/sh/mm/fault_32.c

   1 /*
   2  * Page fault handler for SH with an MMU.
   3  *
   4  *  Copyright (C) 1999  Niibe Yutaka
   5  *  Copyright (C) 2003 - 2008  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 <linux/marker.h>
  19 #include <asm/io_trapped.h>
  20 #include <asm/system.h>
  21 #include <asm/mmu_context.h>
  22 #include <asm/tlbflush.h>
  23
  24 /*
  25  * This routine handles page faults.  It determines the address,
  26  * and the problem, and then passes it off to one of the appropriate
  27  * routines.
  28  */
  29 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
  30                                         unsigned long writeaccess,
  31                                         unsigned long address)
  32 {
  33         struct task_struct *tsk;
  34         struct mm_struct *mm;
  35         struct vm_area_struct * vma;
  36         int si_code;
  37         int fault;
  38         siginfo_t info;
  39
  40         /*
  41          * We don't bother with any notifier callbacks here, as they are
  42          * all handled through the __do_page_fault() fast-path.
  43          */
  44
  45         tsk = current;
  46         si_code = SEGV_MAPERR;
  47
  48         if (unlikely(address >= TASK_SIZE)) {
  49                 /*
  50                  * Synchronize this task's top level page-table
  51                  * with the 'reference' page table.
  52                  *
  53                  * Do _not_ use "tsk" here. We might be inside
  54                  * an interrupt in the middle of a task switch..
  55                  */
  56                 int offset = pgd_index(address);
  57                 pgd_t *pgd, *pgd_k;
  58                 pud_t *pud, *pud_k;
  59                 pmd_t *pmd, *pmd_k;
  60
  61                 pgd = get_TTB() + offset;
  62                 pgd_k = swapper_pg_dir + offset;
  63
  64                 if (!pgd_present(*pgd)) {
  65                         if (!pgd_present(*pgd_k))
  66                                 goto bad_area_nosemaphore;
  67                         set_pgd(pgd, *pgd_k);
  68                         return;
  69                 }
  70
  71                 pud = pud_offset(pgd, address);
  72                 pud_k = pud_offset(pgd_k, address);
  73
  74                 if (!pud_present(*pud)) {
  75                         if (!pud_present(*pud_k))
  76                                 goto bad_area_nosemaphore;
  77                         set_pud(pud, *pud_k);
  78                         return;
  79                 }
  80
  81                 pmd = pmd_offset(pud, address);
  82                 pmd_k = pmd_offset(pud_k, address);
  83                 if (pmd_present(*pmd) || !pmd_present(*pmd_k))
  84                         goto bad_area_nosemaphore;
  85                 set_pmd(pmd, *pmd_k);
  86
  87                 return;
  88         }
  89
  90         /* Only enable interrupts if they were on before the fault */
  91         if ((regs->sr & SR_IMASK) != SR_IMASK) {
  92                 trace_hardirqs_on();
  93                 local_irq_enable();
  94         }
  95
  96         mm = tsk->mm;
  97
  98         /*
  99          * If we're in an interrupt or have no user
 100          * context, we must not take the fault..
 101          */
 102         if (in_atomic() || !mm)
 103                 goto no_context;
 104
 105         down_read(&mm->mmap_sem);
 106
 107         vma = find_vma(mm, address);
 108         if (!vma)
 109                 goto bad_area;
 110         if (vma->vm_start <= address)
 111                 goto good_area;
 112         if (!(vma->vm_flags & VM_GROWSDOWN))
 113                 goto bad_area;
 114         if (expand_stack(vma, address))
 115                 goto bad_area;
 116 /*
 117  * Ok, we have a good vm_area for this memory access, so
 118  * we can handle it..
 119  */
 120 good_area:
 121         si_code = SEGV_ACCERR;
 122         if (writeaccess) {
 123                 if (!(vma->vm_flags & VM_WRITE))
 124                         goto bad_area;
 125         } else {
 126                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 127                         goto bad_area;
 128         }
 129
 130         /*
 131          * If for any reason at all we couldn't handle the fault,
 132          * make sure we exit gracefully rather than endlessly redo
 133          * the fault.
 134          */
 135 survive:
 136         fault = handle_mm_fault(mm, vma, address, writeaccess);
 137         if (unlikely(fault & VM_FAULT_ERROR)) {
 138                 if (fault & VM_FAULT_OOM)
 139                         goto out_of_memory;
 140                 else if (fault & VM_FAULT_SIGBUS)
 141                         goto do_sigbus;
 142                 BUG();
 143         }
 144         if (fault & VM_FAULT_MAJOR)
 145                 tsk->maj_flt++;
 146         else
 147                 tsk->min_flt++;
 148
 149         up_read(&mm->mmap_sem);
 150         return;
 151
 152 /*
 153  * Something tried to access memory that isn't in our memory map..
 154  * Fix it, but check if it's kernel or user first..
 155  */
 156 bad_area:
 157         up_read(&mm->mmap_sem);
 158
 159 bad_area_nosemaphore:
 160         if (user_mode(regs)) {
 161                 info.si_signo = SIGSEGV;
 162                 info.si_errno = 0;
 163                 info.si_code = si_code;
 164                 info.si_addr = (void *) address;
 165                 force_sig_info(SIGSEGV, &info, tsk);
 166                 return;
 167         }
 168
 169 no_context:
 170         /* Are we prepared to handle this kernel fault?  */
 171         if (fixup_exception(regs))
 172                 return;
 173
 174         if (handle_trapped_io(regs, address))
 175                 return;
 176 /*
 177  * Oops. The kernel tried to access some bad page. We'll have to
 178  * terminate things with extreme prejudice.
 179  *
 180  */
 181
 182         bust_spinlocks(1);
 183
 184         if (oops_may_print()) {
 185                 unsigned long page;
 186
 187                 if (address < PAGE_SIZE)
 188                         printk(KERN_ALERT "Unable to handle kernel NULL "
 189                                           "pointer dereference");
 190                 else
 191                         printk(KERN_ALERT "Unable to handle kernel paging "
 192                                           "request");
 193                 printk(" at virtual address %08lx\n", address);
 194                 printk(KERN_ALERT "pc = %08lx\n", regs->pc);
 195                 page = (unsigned long)get_TTB();
 196                 if (page) {
 197                         page = ((__typeof__(page) *)page)[address >> PGDIR_SHIFT];
 198                         printk(KERN_ALERT "*pde = %08lx\n", page);
 199                         if (page & _PAGE_PRESENT) {
 200                                 page &= PAGE_MASK;
 201                                 address &= 0x003ff000;
 202                                 page = ((__typeof__(page) *)
 203                                                 __va(page))[address >>
 204                                                             PAGE_SHIFT];
 205                                 printk(KERN_ALERT "*pte = %08lx\n", page);
 206                         }
 207                 }
 208         }
 209
 210         die("Oops", regs, writeaccess);
 211         bust_spinlocks(0);
 212         do_exit(SIGKILL);
 213
 214 /*
 215  * We ran out of memory, or some other thing happened to us that made
 216  * us unable to handle the page fault gracefully.
 217  */
 218 out_of_memory:
 219         up_read(&mm->mmap_sem);
 220         if (is_global_init(current)) {
 221                 yield();
 222                 down_read(&mm->mmap_sem);
 223                 goto survive;
 224         }
 225         printk("VM: killing process %s\n", tsk->comm);
 226         if (user_mode(regs))
 227                 do_group_exit(SIGKILL);
 228         goto no_context;
 229
 230 do_sigbus:
 231         up_read(&mm->mmap_sem);
 232
 233         /*
 234          * Send a sigbus, regardless of whether we were in kernel
 235          * or user mode.
 236          */
 237         info.si_signo = SIGBUS;
 238         info.si_errno = 0;
 239         info.si_code = BUS_ADRERR;
 240         info.si_addr = (void *)address;
 241         force_sig_info(SIGBUS, &info, tsk);
 242
 243         /* Kernel mode? Handle exceptions or die */
 244         if (!user_mode(regs))
 245                 goto no_context;
 246 }
 247
 248 static inline int notify_page_fault(struct pt_regs *regs, int trap)
 249 {
 250         int ret = 0;
 251
 252         trace_mark(kernel_arch_trap_entry, "trap_id %d ip #p%ld",
 253                    trap >> 5, instruction_pointer(regs));
 254
 255 #ifdef CONFIG_KPROBES
 256         if (!user_mode(regs)) {
 257                 preempt_disable();
 258                 if (kprobe_running() && kprobe_fault_handler(regs, trap))
 259                         ret = 1;
 260                 preempt_enable();
 261         }
 262 #endif
 263
 264         return ret;
 265 }
 266
 267 /*
 268  * Called with interrupts disabled.
 269  */
 270 asmlinkage int __kprobes __do_page_fault(struct pt_regs *regs,
 271                                          unsigned long writeaccess,
 272                                          unsigned long address)
 273 {
 274         pgd_t *pgd;
 275         pud_t *pud;
 276         pmd_t *pmd;
 277         pte_t *pte;
 278         pte_t entry;
 279         int ret = 0;
 280
 281         if (notify_page_fault(regs, lookup_exception_vector()))
 282                 goto out;
 283
 284         ret = 1;
 285
 286         /*
 287          * We don't take page faults for P1, P2, and parts of P4, these
 288          * are always mapped, whether it be due to legacy behaviour in
 289          * 29-bit mode, or due to PMB configuration in 32-bit mode.
 290          */
 291         if (address >= P3SEG && address < P3_ADDR_MAX) {
 292                 pgd = pgd_offset_k(address);
 293         } else {
 294                 if (unlikely(address >= TASK_SIZE || !current->mm))
 295                         goto out;
 296
 297                 pgd = pgd_offset(current->mm, address);
 298         }
 299
 300         pud = pud_offset(pgd, address);
 301         if (pud_none_or_clear_bad(pud))
 302                 goto out;
 303         pmd = pmd_offset(pud, address);
 304         if (pmd_none_or_clear_bad(pmd))
 305                 goto out;
 306         pte = pte_offset_kernel(pmd, address);
 307         entry = *pte;
 308         if (unlikely(pte_none(entry) || pte_not_present(entry)))
 309                 goto out;
 310         if (unlikely(writeaccess && !pte_write(entry)))
 311                 goto out;
 312
 313         if (writeaccess)
 314                 entry = pte_mkdirty(entry);
 315         entry = pte_mkyoung(entry);
 316
 317 #if defined(CONFIG_CPU_SH4) && !defined(CONFIG_SMP)
 318         /*
 319          * ITLB is not affected by "ldtlb" instruction.
 320          * So, we need to flush the entry by ourselves.
 321          */
 322         local_flush_tlb_one(get_asid(), address & PAGE_MASK);
 323 #endif
 324
 325         set_pte(pte, entry);
 326         update_mmu_cache(NULL, address, entry);
 327
 328         ret = 0;
 329 out:
 330         trace_mark(kernel_arch_trap_exit, MARK_NOARGS);
 331         return ret;
 332 }