arch/ppc64/mm/fault.c

   1 /*
   2  *  arch/ppc/mm/fault.c
   3  *
   4  *  PowerPC version
   5  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
   6  *
   7  *  Derived from "arch/i386/mm/fault.c"
   8  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   9  *
  10  *  Modified by Cort Dougan and Paul Mackerras.
  11  *
  12  *  Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
  13  *
  14  *  This program is free software; you can redistribute it and/or
  15  *  modify it under the terms of the GNU General Public License
  16  *  as published by the Free Software Foundation; either version
  17  *  2 of the License, or (at your option) any later version.
  18  */
  19
  20 #include <linux/config.h>
  21 #include <linux/signal.h>
  22 #include <linux/sched.h>
  23 #include <linux/kernel.h>
  24 #include <linux/errno.h>
  25 #include <linux/string.h>
  26 #include <linux/types.h>
  27 #include <linux/mman.h>
  28 #include <linux/mm.h>
  29 #include <linux/interrupt.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/module.h>
  32
  33 #include <asm/page.h>
  34 #include <asm/pgtable.h>
  35 #include <asm/mmu.h>
  36 #include <asm/mmu_context.h>
  37 #include <asm/system.h>
  38 #include <asm/uaccess.h>
  39
  40 /*
  41  * Check whether the instruction at regs->nip is a store using
  42  * an update addressing form which will update r1.
  43  */
  44 static int store_updates_sp(struct pt_regs *regs)
  45 {
  46         unsigned int inst;
  47
  48         if (get_user(inst, (unsigned int __user *)regs->nip))
  49                 return 0;
  50         /* check for 1 in the rA field */
  51         if (((inst >> 16) & 0x1f) != 1)
  52                 return 0;
  53         /* check major opcode */
  54         switch (inst >> 26) {
  55         case 37:        /* stwu */
  56         case 39:        /* stbu */
  57         case 45:        /* sthu */
  58         case 53:        /* stfsu */
  59         case 55:        /* stfdu */
  60                 return 1;
  61         case 62:        /* std or stdu */
  62                 return (inst & 3) == 1;
  63         case 31:
  64                 /* check minor opcode */
  65                 switch ((inst >> 1) & 0x3ff) {
  66                 case 181:       /* stdux */
  67                 case 183:       /* stwux */
  68                 case 247:       /* stbux */
  69                 case 439:       /* sthux */
  70                 case 695:       /* stfsux */
  71                 case 759:       /* stfdux */
  72                         return 1;
  73                 }
  74         }
  75         return 0;
  76 }
  77
  78 /*
  79  * The error_code parameter is
  80  *  - DSISR for a non-SLB data access fault,
  81  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
  82  *  - 0 any SLB fault.
  83  * The return value is 0 if the fault was handled, or the signal
  84  * number if this is a kernel fault that can't be handled here.
  85  */
  86 int do_page_fault(struct pt_regs *regs, unsigned long address,
  87                   unsigned long error_code)
  88 {
  89         struct vm_area_struct * vma;
  90         struct mm_struct *mm = current->mm;
  91         siginfo_t info;
  92         unsigned long code = SEGV_MAPERR;
  93         unsigned long is_write = error_code & 0x02000000;
  94         unsigned long trap = TRAP(regs);
  95
  96         BUG_ON((trap == 0x380) || (trap == 0x480));
  97
  98         if (trap == 0x300) {
  99                 if (debugger_fault_handler(regs))
 100                         return 0;
 101         }
 102
 103         /* On a kernel SLB miss we can only check for a valid exception entry */
 104         if (!user_mode(regs) && (address >= TASK_SIZE))
 105                 return SIGSEGV;
 106
 107         if (error_code & 0x00400000) {
 108                 if (debugger_dabr_match(regs))
 109                         return 0;
 110         }
 111
 112         if (in_atomic() || mm == NULL) {
 113                 if (!user_mode(regs))
 114                         return SIGSEGV;
 115                 /* in_atomic() in user mode is really bad,
 116                    as is current->mm == NULL. */
 117                 printk(KERN_EMERG "Page fault in user mode with"
 118                        "in_atomic() = %d mm = %p\n", in_atomic(), mm);
 119                 printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
 120                        regs->nip, regs->msr);
 121                 die("Weird page fault", regs, SIGSEGV);
 122         }
 123
 124         /* When running in the kernel we expect faults to occur only to
 125          * addresses in user space.  All other faults represent errors in the
 126          * kernel and should generate an OOPS.  Unfortunatly, in the case of an
 127          * erroneous fault occuring in a code path which already holds mmap_sem
 128          * we will deadlock attempting to validate the fault against the
 129          * address space.  Luckily the kernel only validly references user
 130          * space from well defined areas of code, which are listed in the
 131          * exceptions table.
 132          *
 133          * As the vast majority of faults will be valid we will only perform
 134          * the source reference check when there is a possibilty of a deadlock.
 135          * Attempt to lock the address space, if we cannot we then validate the
 136          * source.  If this is invalid we can skip the address space check,
 137          * thus avoiding the deadlock.
 138          */
 139         if (!down_read_trylock(&mm->mmap_sem)) {
 140                 if (!user_mode(regs) && !search_exception_tables(regs->nip))
 141                         goto bad_area_nosemaphore;
 142
 143                 down_read(&mm->mmap_sem);
 144         }
 145
 146         vma = find_vma(mm, address);
 147         if (!vma)
 148                 goto bad_area;
 149
 150         if (vma->vm_start <= address) {
 151                 goto good_area;
 152         }
 153         if (!(vma->vm_flags & VM_GROWSDOWN))
 154                 goto bad_area;
 155
 156         /*
 157          * N.B. The POWER/Open ABI allows programs to access up to
 158          * 288 bytes below the stack pointer.
 159          * The kernel signal delivery code writes up to about 1.5kB
 160          * below the stack pointer (r1) before decrementing it.
 161          * The exec code can write slightly over 640kB to the stack
 162          * before setting the user r1.  Thus we allow the stack to
 163          * expand to 1MB without further checks.
 164          */
 165         if (address + 0x100000 < vma->vm_end) {
 166                 /* get user regs even if this fault is in kernel mode */
 167                 struct pt_regs *uregs = current->thread.regs;
 168                 if (uregs == NULL)
 169                         goto bad_area;
 170
 171                 /*
 172                  * A user-mode access to an address a long way below
 173                  * the stack pointer is only valid if the instruction
 174                  * is one which would update the stack pointer to the
 175                  * address accessed if the instruction completed,
 176                  * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
 177                  * (or the byte, halfword, float or double forms).
 178                  *
 179                  * If we don't check this then any write to the area
 180                  * between the last mapped region and the stack will
 181                  * expand the stack rather than segfaulting.
 182                  */
 183                 if (address + 2048 < uregs->gpr[1]
 184                     && (!user_mode(regs) || !store_updates_sp(regs)))
 185                         goto bad_area;
 186         }
 187
 188         if (expand_stack(vma, address))
 189                 goto bad_area;
 190
 191 good_area:
 192         code = SEGV_ACCERR;
 193
 194         /* a write */
 195         if (is_write) {
 196                 if (!(vma->vm_flags & VM_WRITE))
 197                         goto bad_area;
 198         /* a read */
 199         } else {
 200                 /* protection fault */
 201                 if (error_code & 0x08000000)
 202                         goto bad_area;
 203                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 204                         goto bad_area;
 205         }
 206
 207  survive:
 208         /*
 209          * If for any reason at all we couldn't handle the fault,
 210          * make sure we exit gracefully rather than endlessly redo
 211          * the fault.
 212          */
 213         switch (handle_mm_fault(mm, vma, address, is_write)) {
 214
 215         case VM_FAULT_MINOR:
 216                 current->min_flt++;
 217                 break;
 218         case VM_FAULT_MAJOR:
 219                 current->maj_flt++;
 220                 break;
 221         case VM_FAULT_SIGBUS:
 222                 goto do_sigbus;
 223         case VM_FAULT_OOM:
 224                 goto out_of_memory;
 225         default:
 226                 BUG();
 227         }
 228
 229         up_read(&mm->mmap_sem);
 230         return 0;
 231
 232 bad_area:
 233         up_read(&mm->mmap_sem);
 234
 235 bad_area_nosemaphore:
 236         /* User mode accesses cause a SIGSEGV */
 237         if (user_mode(regs)) {
 238                 info.si_signo = SIGSEGV;
 239                 info.si_errno = 0;
 240                 info.si_code = code;
 241                 info.si_addr = (void __user *) address;
 242                 force_sig_info(SIGSEGV, &info, current);
 243                 return 0;
 244         }
 245
 246         return SIGSEGV;
 247
 248 /*
 249  * We ran out of memory, or some other thing happened to us that made
 250  * us unable to handle the page fault gracefully.
 251  */
 252 out_of_memory:
 253         up_read(&mm->mmap_sem);
 254         if (current->pid == 1) {
 255                 yield();
 256                 down_read(&mm->mmap_sem);
 257                 goto survive;
 258         }
 259         printk("VM: killing process %s\n", current->comm);
 260         if (user_mode(regs))
 261                 do_exit(SIGKILL);
 262         return SIGKILL;
 263
 264 do_sigbus:
 265         up_read(&mm->mmap_sem);
 266         if (user_mode(regs)) {
 267                 info.si_signo = SIGBUS;
 268                 info.si_errno = 0;
 269                 info.si_code = BUS_ADRERR;
 270                 info.si_addr = (void __user *)address;
 271                 force_sig_info(SIGBUS, &info, current);
 272                 return 0;
 273         }
 274         return SIGBUS;
 275 }
 276
 277 /*
 278  * bad_page_fault is called when we have a bad access from the kernel.
 279  * It is called from do_page_fault above and from some of the procedures
 280  * in traps.c.
 281  */
 282 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
 283 {
 284         const struct exception_table_entry *entry;
 285
 286         /* Are we prepared to handle this fault?  */
 287         if ((entry = search_exception_tables(regs->nip)) != NULL) {
 288                 regs->nip = entry->fixup;
 289                 return;
 290         }
 291
 292         /* kernel has accessed a bad area */
 293         die("Kernel access of bad area", regs, sig);
 294 }