arch/mips/mm/fault.c

   1 /*
   2  * This file is subject to the terms and conditions of the GNU General Public
   3  * License.  See the file "COPYING" in the main directory of this archive
   4  * for more details.
   5  *
   6  * Copyright (C) 1995 - 2000 by Ralf Baechle
   7  */
   8 #include <linux/signal.h>
   9 #include <linux/sched.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/kernel.h>
  12 #include <linux/errno.h>
  13 #include <linux/string.h>
  14 #include <linux/types.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/mman.h>
  17 #include <linux/mm.h>
  18 #include <linux/smp.h>
  19 #include <linux/module.h>
  20 #include <linux/kprobes.h>
  21 #include <linux/perf_event.h>
  22
  23 #include <asm/branch.h>
  24 #include <asm/mmu_context.h>
  25 #include <asm/uaccess.h>
  26 #include <asm/ptrace.h>
  27 #include <asm/highmem.h>                /* For VMALLOC_END */
  28 #include <linux/kdebug.h>
  29
  30 /*
  31  * This routine handles page faults.  It determines the address,
  32  * and the problem, and then passes it off to one of the appropriate
  33  * routines.
  34  */
  35 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs, unsigned long write,
  36                               unsigned long address)
  37 {
  38         struct vm_area_struct * vma = NULL;
  39         struct task_struct *tsk = current;
  40         struct mm_struct *mm = tsk->mm;
  41         const int field = sizeof(unsigned long) * 2;
  42         siginfo_t info;
  43         int fault;
  44         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
  45
  46 #if 0
  47         printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", raw_smp_processor_id(),
  48                current->comm, current->pid, field, address, write,
  49                field, regs->cp0_epc);
  50 #endif
  51
  52 #ifdef CONFIG_KPROBES
  53         /*
  54          * This is to notify the fault handler of the kprobes.  The
  55          * exception code is redundant as it is also carried in REGS,
  56          * but we pass it anyhow.
  57          */
  58         if (notify_die(DIE_PAGE_FAULT, "page fault", regs, -1,
  59                        (regs->cp0_cause >> 2) & 0x1f, SIGSEGV) == NOTIFY_STOP)
  60                 return;
  61 #endif
  62
  63         info.si_code = SEGV_MAPERR;
  64
  65         /*
  66          * We fault-in kernel-space virtual memory on-demand. The
  67          * 'reference' page table is init_mm.pgd.
  68          *
  69          * NOTE! We MUST NOT take any locks for this case. We may
  70          * be in an interrupt or a critical region, and should
  71          * only copy the information from the master page table,
  72          * nothing more.
  73          */
  74 #ifdef CONFIG_64BIT
  75 # define VMALLOC_FAULT_TARGET no_context
  76 #else
  77 # define VMALLOC_FAULT_TARGET vmalloc_fault
  78 #endif
  79
  80         if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END))
  81                 goto VMALLOC_FAULT_TARGET;
  82 #ifdef MODULE_START
  83         if (unlikely(address >= MODULE_START && address < MODULE_END))
  84                 goto VMALLOC_FAULT_TARGET;
  85 #endif
  86
  87         /*
  88          * If we're in an interrupt or have no user
  89          * context, we must not take the fault..
  90          */
  91         if (in_atomic() || !mm)
  92                 goto bad_area_nosemaphore;
  93
  94         if (user_mode(regs))
  95                 flags |= FAULT_FLAG_USER;
  96 retry:
  97         down_read(&mm->mmap_sem);
  98         vma = find_vma(mm, address);
  99         if (!vma)
 100                 goto bad_area;
 101         if (vma->vm_start <= address)
 102                 goto good_area;
 103         if (!(vma->vm_flags & VM_GROWSDOWN))
 104                 goto bad_area;
 105         if (expand_stack(vma, address))
 106                 goto bad_area;
 107 /*
 108  * Ok, we have a good vm_area for this memory access, so
 109  * we can handle it..
 110  */
 111 good_area:
 112         info.si_code = SEGV_ACCERR;
 113
 114         if (write) {
 115                 if (!(vma->vm_flags & VM_WRITE))
 116                         goto bad_area;
 117                 flags |= FAULT_FLAG_WRITE;
 118         } else {
 119                 if (cpu_has_rixi) {
 120                         if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
 121 #if 0
 122                                 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
 123                                           raw_smp_processor_id(),
 124                                           current->comm, current->pid,
 125                                           field, address, write,
 126                                           field, regs->cp0_epc);
 127 #endif
 128                                 goto bad_area;
 129                         }
 130                         if (!(vma->vm_flags & VM_READ)) {
 131 #if 0
 132                                 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] RI violation\n",
 133                                           raw_smp_processor_id(),
 134                                           current->comm, current->pid,
 135                                           field, address, write,
 136                                           field, regs->cp0_epc);
 137 #endif
 138                                 goto bad_area;
 139                         }
 140                 } else {
 141                         if (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)))
 142                                 goto bad_area;
 143                 }
 144         }
 145
 146         /*
 147          * If for any reason at all we couldn't handle the fault,
 148          * make sure we exit gracefully rather than endlessly redo
 149          * the fault.
 150          */
 151         fault = handle_mm_fault(mm, vma, address, flags);
 152
 153         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 154                 return;
 155
 156         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 157         if (unlikely(fault & VM_FAULT_ERROR)) {
 158                 if (fault & VM_FAULT_OOM)
 159                         goto out_of_memory;
 160                 else if (fault & VM_FAULT_SIGSEGV)
 161                         goto bad_area;
 162                 else if (fault & VM_FAULT_SIGBUS)
 163                         goto do_sigbus;
 164                 BUG();
 165         }
 166         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 167                 if (fault & VM_FAULT_MAJOR) {
 168                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
 169                                                   regs, address);
 170                         tsk->maj_flt++;
 171                 } else {
 172                         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
 173                                                   regs, address);
 174                         tsk->min_flt++;
 175                 }
 176                 if (fault & VM_FAULT_RETRY) {
 177                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 178                         flags |= FAULT_FLAG_TRIED;
 179
 180                         /*
 181                          * No need to up_read(&mm->mmap_sem) as we would
 182                          * have already released it in __lock_page_or_retry
 183                          * in mm/filemap.c.
 184                          */
 185
 186                         goto retry;
 187                 }
 188         }
 189
 190         up_read(&mm->mmap_sem);
 191         return;
 192
 193 /*
 194  * Something tried to access memory that isn't in our memory map..
 195  * Fix it, but check if it's kernel or user first..
 196  */
 197 bad_area:
 198         up_read(&mm->mmap_sem);
 199
 200 bad_area_nosemaphore:
 201         /* User mode accesses just cause a SIGSEGV */
 202         if (user_mode(regs)) {
 203                 tsk->thread.cp0_badvaddr = address;
 204                 tsk->thread.error_code = write;
 205 #if 0
 206                 printk("do_page_fault() #2: sending SIGSEGV to %s for "
 207                        "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
 208                        tsk->comm,
 209                        write ? "write access to" : "read access from",
 210                        field, address,
 211                        field, (unsigned long) regs->cp0_epc,
 212                        field, (unsigned long) regs->regs[31]);
 213 #endif
 214                 info.si_signo = SIGSEGV;
 215                 info.si_errno = 0;
 216                 /* info.si_code has been set above */
 217                 info.si_addr = (void __user *) address;
 218                 force_sig_info(SIGSEGV, &info, tsk);
 219                 return;
 220         }
 221
 222 no_context:
 223         /* Are we prepared to handle this kernel fault?  */
 224         if (fixup_exception(regs)) {
 225                 current->thread.cp0_baduaddr = address;
 226                 return;
 227         }
 228
 229         /*
 230          * Oops. The kernel tried to access some bad page. We'll have to
 231          * terminate things with extreme prejudice.
 232          */
 233         bust_spinlocks(1);
 234
 235         printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
 236                "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
 237                raw_smp_processor_id(), field, address, field, regs->cp0_epc,
 238                field,  regs->regs[31]);
 239         die("Oops", regs);
 240
 241 out_of_memory:
 242         /*
 243          * We ran out of memory, call the OOM killer, and return the userspace
 244          * (which will retry the fault, or kill us if we got oom-killed).
 245          */
 246         up_read(&mm->mmap_sem);
 247         if (!user_mode(regs))
 248                 goto no_context;
 249         pagefault_out_of_memory();
 250         return;
 251
 252 do_sigbus:
 253         up_read(&mm->mmap_sem);
 254
 255         /* Kernel mode? Handle exceptions or die */
 256         if (!user_mode(regs))
 257                 goto no_context;
 258         else
 259         /*
 260          * Send a sigbus, regardless of whether we were in kernel
 261          * or user mode.
 262          */
 263 #if 0
 264                 printk("do_page_fault() #3: sending SIGBUS to %s for "
 265                        "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
 266                        tsk->comm,
 267                        write ? "write access to" : "read access from",
 268                        field, address,
 269                        field, (unsigned long) regs->cp0_epc,
 270                        field, (unsigned long) regs->regs[31]);
 271 #endif
 272         tsk->thread.cp0_badvaddr = address;
 273         info.si_signo = SIGBUS;
 274         info.si_errno = 0;
 275         info.si_code = BUS_ADRERR;
 276         info.si_addr = (void __user *) address;
 277         force_sig_info(SIGBUS, &info, tsk);
 278
 279         return;
 280 #ifndef CONFIG_64BIT
 281 vmalloc_fault:
 282         {
 283                 /*
 284                  * Synchronize this task's top level page-table
 285                  * with the 'reference' page table.
 286                  *
 287                  * Do _not_ use "tsk" here. We might be inside
 288                  * an interrupt in the middle of a task switch..
 289                  */
 290                 int offset = __pgd_offset(address);
 291                 pgd_t *pgd, *pgd_k;
 292                 pud_t *pud, *pud_k;
 293                 pmd_t *pmd, *pmd_k;
 294                 pte_t *pte_k;
 295
 296                 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
 297                 pgd_k = init_mm.pgd + offset;
 298
 299                 if (!pgd_present(*pgd_k))
 300                         goto no_context;
 301                 set_pgd(pgd, *pgd_k);
 302
 303                 pud = pud_offset(pgd, address);
 304                 pud_k = pud_offset(pgd_k, address);
 305                 if (!pud_present(*pud_k))
 306                         goto no_context;
 307
 308                 pmd = pmd_offset(pud, address);
 309                 pmd_k = pmd_offset(pud_k, address);
 310                 if (!pmd_present(*pmd_k))
 311                         goto no_context;
 312                 set_pmd(pmd, *pmd_k);
 313
 314                 pte_k = pte_offset_kernel(pmd_k, address);
 315                 if (!pte_present(*pte_k))
 316                         goto no_context;
 317                 return;
 318         }
 319 #endif
 320 }