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 #include <linux/kprobes.h>
  33
  34 #include <asm/page.h>
  35 #include <asm/pgtable.h>
  36 #include <asm/mmu.h>
  37 #include <asm/mmu_context.h>
  38 #include <asm/system.h>
  39 #include <asm/uaccess.h>
  40 #include <asm/kdebug.h>
  41 #include <asm/siginfo.h>
  42
  43 /*
  44  * Check whether the instruction at regs->nip is a store using
  45  * an update addressing form which will update r1.
  46  */
  47 static int store_updates_sp(struct pt_regs *regs)
  48 {
  49         unsigned int inst;
  50
  51         if (get_user(inst, (unsigned int __user *)regs->nip))
  52                 return 0;
  53         /* check for 1 in the rA field */
  54         if (((inst >> 16) & 0x1f) != 1)
  55                 return 0;
  56         /* check major opcode */
  57         switch (inst >> 26) {
  58         case 37:        /* stwu */
  59         case 39:        /* stbu */
  60         case 45:        /* sthu */
  61         case 53:        /* stfsu */
  62         case 55:        /* stfdu */
  63                 return 1;
  64         case 62:        /* std or stdu */
  65                 return (inst & 3) == 1;
  66         case 31:
  67                 /* check minor opcode */
  68                 switch ((inst >> 1) & 0x3ff) {
  69                 case 181:       /* stdux */
  70                 case 183:       /* stwux */
  71                 case 247:       /* stbux */
  72                 case 439:       /* sthux */
  73                 case 695:       /* stfsux */
  74                 case 759:       /* stfdux */
  75                         return 1;
  76                 }
  77         }
  78         return 0;
  79 }
  80
  81 static void do_dabr(struct pt_regs *regs, unsigned long error_code)
  82 {
  83         siginfo_t info;
  84
  85         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
  86                         11, SIGSEGV) == NOTIFY_STOP)
  87                 return;
  88
  89         if (debugger_dabr_match(regs))
  90                 return;
  91
  92         /* Clear the DABR */
  93         set_dabr(0);
  94
  95         /* Deliver the signal to userspace */
  96         info.si_signo = SIGTRAP;
  97         info.si_errno = 0;
  98         info.si_code = TRAP_HWBKPT;
  99         info.si_addr = (void __user *)regs->nip;
 100         force_sig_info(SIGTRAP, &info, current);
 101 }
 102
 103 /*
 104  * The error_code parameter is
 105  *  - DSISR for a non-SLB data access fault,
 106  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
 107  *  - 0 any SLB fault.
 108  * The return value is 0 if the fault was handled, or the signal
 109  * number if this is a kernel fault that can't be handled here.
 110  */
 111 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
 112                             unsigned long error_code)
 113 {
 114         struct vm_area_struct * vma;
 115         struct mm_struct *mm = current->mm;
 116         siginfo_t info;
 117         unsigned long code = SEGV_MAPERR;
 118         unsigned long is_write = error_code & DSISR_ISSTORE;
 119         unsigned long trap = TRAP(regs);
 120         unsigned long is_exec = trap == 0x400;
 121
 122         BUG_ON((trap == 0x380) || (trap == 0x480));
 123
 124         if (notify_die(DIE_PAGE_FAULT, "page_fault", regs, error_code,
 125                                 11, SIGSEGV) == NOTIFY_STOP)
 126                 return 0;
 127
 128         if (trap == 0x300) {
 129                 if (debugger_fault_handler(regs))
 130                         return 0;
 131         }
 132
 133         /* On a kernel SLB miss we can only check for a valid exception entry */
 134         if (!user_mode(regs) && (address >= TASK_SIZE))
 135                 return SIGSEGV;
 136
 137         if (error_code & DSISR_DABRMATCH) {
 138                 do_dabr(regs, error_code);
 139                 return 0;
 140         }
 141
 142         if (in_atomic() || mm == NULL) {
 143                 if (!user_mode(regs))
 144                         return SIGSEGV;
 145                 /* in_atomic() in user mode is really bad,
 146                    as is current->mm == NULL. */
 147                 printk(KERN_EMERG "Page fault in user mode with"
 148                        "in_atomic() = %d mm = %p\n", in_atomic(), mm);
 149                 printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
 150                        regs->nip, regs->msr);
 151                 die("Weird page fault", regs, SIGSEGV);
 152         }
 153
 154         /* When running in the kernel we expect faults to occur only to
 155          * addresses in user space.  All other faults represent errors in the
 156          * kernel and should generate an OOPS.  Unfortunatly, in the case of an
 157          * erroneous fault occuring in a code path which already holds mmap_sem
 158          * we will deadlock attempting to validate the fault against the
 159          * address space.  Luckily the kernel only validly references user
 160          * space from well defined areas of code, which are listed in the
 161          * exceptions table.
 162          *
 163          * As the vast majority of faults will be valid we will only perform
 164          * the source reference check when there is a possibilty of a deadlock.
 165          * Attempt to lock the address space, if we cannot we then validate the
 166          * source.  If this is invalid we can skip the address space check,
 167          * thus avoiding the deadlock.
 168          */
 169         if (!down_read_trylock(&mm->mmap_sem)) {
 170                 if (!user_mode(regs) && !search_exception_tables(regs->nip))
 171                         goto bad_area_nosemaphore;
 172
 173                 down_read(&mm->mmap_sem);
 174         }
 175
 176         vma = find_vma(mm, address);
 177         if (!vma)
 178                 goto bad_area;
 179
 180         if (vma->vm_start <= address) {
 181                 goto good_area;
 182         }
 183         if (!(vma->vm_flags & VM_GROWSDOWN))
 184                 goto bad_area;
 185
 186         /*
 187          * N.B. The POWER/Open ABI allows programs to access up to
 188          * 288 bytes below the stack pointer.
 189          * The kernel signal delivery code writes up to about 1.5kB
 190          * below the stack pointer (r1) before decrementing it.
 191          * The exec code can write slightly over 640kB to the stack
 192          * before setting the user r1.  Thus we allow the stack to
 193          * expand to 1MB without further checks.
 194          */
 195         if (address + 0x100000 < vma->vm_end) {
 196                 /* get user regs even if this fault is in kernel mode */
 197                 struct pt_regs *uregs = current->thread.regs;
 198                 if (uregs == NULL)
 199                         goto bad_area;
 200
 201                 /*
 202                  * A user-mode access to an address a long way below
 203                  * the stack pointer is only valid if the instruction
 204                  * is one which would update the stack pointer to the
 205                  * address accessed if the instruction completed,
 206                  * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
 207                  * (or the byte, halfword, float or double forms).
 208                  *
 209                  * If we don't check this then any write to the area
 210                  * between the last mapped region and the stack will
 211                  * expand the stack rather than segfaulting.
 212                  */
 213                 if (address + 2048 < uregs->gpr[1]
 214                     && (!user_mode(regs) || !store_updates_sp(regs)))
 215                         goto bad_area;
 216         }
 217
 218         if (expand_stack(vma, address))
 219                 goto bad_area;
 220
 221 good_area:
 222         code = SEGV_ACCERR;
 223
 224         if (is_exec) {
 225                 /* protection fault */
 226                 if (error_code & DSISR_PROTFAULT)
 227                         goto bad_area;
 228                 if (!(vma->vm_flags & VM_EXEC))
 229                         goto bad_area;
 230         /* a write */
 231         } else if (is_write) {
 232                 if (!(vma->vm_flags & VM_WRITE))
 233                         goto bad_area;
 234         /* a read */
 235         } else {
 236                 if (!(vma->vm_flags & VM_READ))
 237                         goto bad_area;
 238         }
 239
 240  survive:
 241         /*
 242          * If for any reason at all we couldn't handle the fault,
 243          * make sure we exit gracefully rather than endlessly redo
 244          * the fault.
 245          */
 246         switch (handle_mm_fault(mm, vma, address, is_write)) {
 247
 248         case VM_FAULT_MINOR:
 249                 current->min_flt++;
 250                 break;
 251         case VM_FAULT_MAJOR:
 252                 current->maj_flt++;
 253                 break;
 254         case VM_FAULT_SIGBUS:
 255                 goto do_sigbus;
 256         case VM_FAULT_OOM:
 257                 goto out_of_memory;
 258         default:
 259                 BUG();
 260         }
 261
 262         up_read(&mm->mmap_sem);
 263         return 0;
 264
 265 bad_area:
 266         up_read(&mm->mmap_sem);
 267
 268 bad_area_nosemaphore:
 269         /* User mode accesses cause a SIGSEGV */
 270         if (user_mode(regs)) {
 271                 info.si_signo = SIGSEGV;
 272                 info.si_errno = 0;
 273                 info.si_code = code;
 274                 info.si_addr = (void __user *) address;
 275                 force_sig_info(SIGSEGV, &info, current);
 276                 return 0;
 277         }
 278
 279         if (trap == 0x400 && (error_code & DSISR_PROTFAULT)
 280             && printk_ratelimit())
 281                 printk(KERN_CRIT "kernel tried to execute NX-protected"
 282                        " page (%lx) - exploit attempt? (uid: %d)\n",
 283                        address, current->uid);
 284
 285         return SIGSEGV;
 286
 287 /*
 288  * We ran out of memory, or some other thing happened to us that made
 289  * us unable to handle the page fault gracefully.
 290  */
 291 out_of_memory:
 292         up_read(&mm->mmap_sem);
 293         if (current->pid == 1) {
 294                 yield();
 295                 down_read(&mm->mmap_sem);
 296                 goto survive;
 297         }
 298         printk("VM: killing process %s\n", current->comm);
 299         if (user_mode(regs))
 300                 do_exit(SIGKILL);
 301         return SIGKILL;
 302
 303 do_sigbus:
 304         up_read(&mm->mmap_sem);
 305         if (user_mode(regs)) {
 306                 info.si_signo = SIGBUS;
 307                 info.si_errno = 0;
 308                 info.si_code = BUS_ADRERR;
 309                 info.si_addr = (void __user *)address;
 310                 force_sig_info(SIGBUS, &info, current);
 311                 return 0;
 312         }
 313         return SIGBUS;
 314 }
 315
 316 /*
 317  * bad_page_fault is called when we have a bad access from the kernel.
 318  * It is called from do_page_fault above and from some of the procedures
 319  * in traps.c.
 320  */
 321 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
 322 {
 323         const struct exception_table_entry *entry;
 324
 325         /* Are we prepared to handle this fault?  */
 326         if ((entry = search_exception_tables(regs->nip)) != NULL) {
 327                 regs->nip = entry->fixup;
 328                 return;
 329         }
 330
 331         /* kernel has accessed a bad area */
 332         die("Kernel access of bad area", regs, sig);
 333 }