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