arch/powerpc/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/ptrace.h>
  28 #include <linux/mman.h>
  29 #include <linux/mm.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/highmem.h>
  32 #include <linux/module.h>
  33 #include <linux/kprobes.h>
  34
  35 #include <asm/page.h>
  36 #include <asm/pgtable.h>
  37 #include <asm/mmu.h>
  38 #include <asm/mmu_context.h>
  39 #include <asm/system.h>
  40 #include <asm/uaccess.h>
  41 #include <asm/tlbflush.h>
  42 #include <asm/kdebug.h>
  43 #include <asm/siginfo.h>
  44
  45 /*
  46  * Check whether the instruction at regs->nip is a store using
  47  * an update addressing form which will update r1.
  48  */
  49 static int store_updates_sp(struct pt_regs *regs)
  50 {
  51         unsigned int inst;
  52
  53         if (get_user(inst, (unsigned int __user *)regs->nip))
  54                 return 0;
  55         /* check for 1 in the rA field */
  56         if (((inst >> 16) & 0x1f) != 1)
  57                 return 0;
  58         /* check major opcode */
  59         switch (inst >> 26) {
  60         case 37:        /* stwu */
  61         case 39:        /* stbu */
  62         case 45:        /* sthu */
  63         case 53:        /* stfsu */
  64         case 55:        /* stfdu */
  65                 return 1;
  66         case 62:        /* std or stdu */
  67                 return (inst & 3) == 1;
  68         case 31:
  69                 /* check minor opcode */
  70                 switch ((inst >> 1) & 0x3ff) {
  71                 case 181:       /* stdux */
  72                 case 183:       /* stwux */
  73                 case 247:       /* stbux */
  74                 case 439:       /* sthux */
  75                 case 695:       /* stfsux */
  76                 case 759:       /* stfdux */
  77                         return 1;
  78                 }
  79         }
  80         return 0;
  81 }
  82
  83 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
  84 static void do_dabr(struct pt_regs *regs, unsigned long error_code)
  85 {
  86         siginfo_t info;
  87
  88         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
  89                         11, SIGSEGV) == NOTIFY_STOP)
  90                 return;
  91
  92         if (debugger_dabr_match(regs))
  93                 return;
  94
  95         /* Clear the DABR */
  96         set_dabr(0);
  97
  98         /* Deliver the signal to userspace */
  99         info.si_signo = SIGTRAP;
 100         info.si_errno = 0;
 101         info.si_code = TRAP_HWBKPT;
 102         info.si_addr = (void __user *)regs->nip;
 103         force_sig_info(SIGTRAP, &info, current);
 104 }
 105 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
 106
 107 /*
 108  * For 600- and 800-family processors, the error_code parameter is DSISR
 109  * for a data fault, SRR1 for an instruction fault. For 400-family processors
 110  * the error_code parameter is ESR for a data fault, 0 for an instruction
 111  * fault.
 112  * For 64-bit processors, the error_code parameter is
 113  *  - DSISR for a non-SLB data access fault,
 114  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
 115  *  - 0 any SLB fault.
 116  *
 117  * The return value is 0 if the fault was handled, or the signal
 118  * number if this is a kernel fault that can't be handled here.
 119  */
 120 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
 121                             unsigned long error_code)
 122 {
 123         struct vm_area_struct * vma;
 124         struct mm_struct *mm = current->mm;
 125         siginfo_t info;
 126         int code = SEGV_MAPERR;
 127         int is_write = 0;
 128         int trap = TRAP(regs);
 129         int is_exec = trap == 0x400;
 130
 131 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
 132         /*
 133          * Fortunately the bit assignments in SRR1 for an instruction
 134          * fault and DSISR for a data fault are mostly the same for the
 135          * bits we are interested in.  But there are some bits which
 136          * indicate errors in DSISR but can validly be set in SRR1.
 137          */
 138         if (trap == 0x400)
 139                 error_code &= 0x48200000;
 140         else
 141                 is_write = error_code & DSISR_ISSTORE;
 142 #else
 143         is_write = error_code & ESR_DST;
 144 #endif /* CONFIG_4xx || CONFIG_BOOKE */
 145
 146         if (notify_die(DIE_PAGE_FAULT, "page_fault", regs, error_code,
 147                                 11, SIGSEGV) == NOTIFY_STOP)
 148                 return 0;
 149
 150         if (trap == 0x300) {
 151                 if (debugger_fault_handler(regs))
 152                         return 0;
 153         }
 154
 155         /* On a kernel SLB miss we can only check for a valid exception entry */
 156         if (!user_mode(regs) && (address >= TASK_SIZE))
 157                 return SIGSEGV;
 158
 159 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
 160         if (error_code & DSISR_DABRMATCH) {
 161                 /* DABR match */
 162                 do_dabr(regs, error_code);
 163                 return 0;
 164         }
 165 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
 166
 167         if (in_atomic() || mm == NULL) {
 168                 if (!user_mode(regs))
 169                         return SIGSEGV;
 170                 /* in_atomic() in user mode is really bad,
 171                    as is current->mm == NULL. */
 172                 printk(KERN_EMERG "Page fault in user mode with"
 173                        "in_atomic() = %d mm = %p\n", in_atomic(), mm);
 174                 printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
 175                        regs->nip, regs->msr);
 176                 die("Weird page fault", regs, SIGSEGV);
 177         }
 178
 179         /* When running in the kernel we expect faults to occur only to
 180          * addresses in user space.  All other faults represent errors in the
 181          * kernel and should generate an OOPS.  Unfortunatly, in the case of an
 182          * erroneous fault occuring in a code path which already holds mmap_sem
 183          * we will deadlock attempting to validate the fault against the
 184          * address space.  Luckily the kernel only validly references user
 185          * space from well defined areas of code, which are listed in the
 186          * exceptions table.
 187          *
 188          * As the vast majority of faults will be valid we will only perform
 189          * the source reference check when there is a possibilty of a deadlock.
 190          * Attempt to lock the address space, if we cannot we then validate the
 191          * source.  If this is invalid we can skip the address space check,
 192          * thus avoiding the deadlock.
 193          */
 194         if (!down_read_trylock(&mm->mmap_sem)) {
 195                 if (!user_mode(regs) && !search_exception_tables(regs->nip))
 196                         goto bad_area_nosemaphore;
 197
 198                 down_read(&mm->mmap_sem);
 199         }
 200
 201         vma = find_vma(mm, address);
 202         if (!vma)
 203                 goto bad_area;
 204         if (vma->vm_start <= address)
 205                 goto good_area;
 206         if (!(vma->vm_flags & VM_GROWSDOWN))
 207                 goto bad_area;
 208
 209         /*
 210          * N.B. The POWER/Open ABI allows programs to access up to
 211          * 288 bytes below the stack pointer.
 212          * The kernel signal delivery code writes up to about 1.5kB
 213          * below the stack pointer (r1) before decrementing it.
 214          * The exec code can write slightly over 640kB to the stack
 215          * before setting the user r1.  Thus we allow the stack to
 216          * expand to 1MB without further checks.
 217          */
 218         if (address + 0x100000 < vma->vm_end) {
 219                 /* get user regs even if this fault is in kernel mode */
 220                 struct pt_regs *uregs = current->thread.regs;
 221                 if (uregs == NULL)
 222                         goto bad_area;
 223
 224                 /*
 225                  * A user-mode access to an address a long way below
 226                  * the stack pointer is only valid if the instruction
 227                  * is one which would update the stack pointer to the
 228                  * address accessed if the instruction completed,
 229                  * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
 230                  * (or the byte, halfword, float or double forms).
 231                  *
 232                  * If we don't check this then any write to the area
 233                  * between the last mapped region and the stack will
 234                  * expand the stack rather than segfaulting.
 235                  */
 236                 if (address + 2048 < uregs->gpr[1]
 237                     && (!user_mode(regs) || !store_updates_sp(regs)))
 238                         goto bad_area;
 239         }
 240         if (expand_stack(vma, address))
 241                 goto bad_area;
 242
 243 good_area:
 244         code = SEGV_ACCERR;
 245 #if defined(CONFIG_6xx)
 246         if (error_code & 0x95700000)
 247                 /* an error such as lwarx to I/O controller space,
 248                    address matching DABR, eciwx, etc. */
 249                 goto bad_area;
 250 #endif /* CONFIG_6xx */
 251 #if defined(CONFIG_8xx)
 252         /* The MPC8xx seems to always set 0x80000000, which is
 253          * "undefined".  Of those that can be set, this is the only
 254          * one which seems bad.
 255          */
 256         if (error_code & 0x10000000)
 257                 /* Guarded storage error. */
 258                 goto bad_area;
 259 #endif /* CONFIG_8xx */
 260
 261         if (is_exec) {
 262 #ifdef CONFIG_PPC64
 263                 /* protection fault */
 264                 if (error_code & DSISR_PROTFAULT)
 265                         goto bad_area;
 266                 if (!(vma->vm_flags & VM_EXEC))
 267                         goto bad_area;
 268 #endif
 269 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
 270                 pte_t *ptep;
 271
 272                 /* Since 4xx/Book-E supports per-page execute permission,
 273                  * we lazily flush dcache to icache. */
 274                 ptep = NULL;
 275                 if (get_pteptr(mm, address, &ptep) && pte_present(*ptep)) {
 276                         struct page *page = pte_page(*ptep);
 277
 278                         if (! test_bit(PG_arch_1, &page->flags)) {
 279                                 flush_dcache_icache_page(page);
 280                                 set_bit(PG_arch_1, &page->flags);
 281                         }
 282                         pte_update(ptep, 0, _PAGE_HWEXEC);
 283                         _tlbie(address);
 284                         pte_unmap(ptep);
 285                         up_read(&mm->mmap_sem);
 286                         return 0;
 287                 }
 288                 if (ptep != NULL)
 289                         pte_unmap(ptep);
 290 #endif
 291         /* a write */
 292         } else if (is_write) {
 293                 if (!(vma->vm_flags & VM_WRITE))
 294                         goto bad_area;
 295         /* a read */
 296         } else {
 297                 /* protection fault */
 298                 if (error_code & 0x08000000)
 299                         goto bad_area;
 300                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 301                         goto bad_area;
 302         }
 303
 304         /*
 305          * If for any reason at all we couldn't handle the fault,
 306          * make sure we exit gracefully rather than endlessly redo
 307          * the fault.
 308          */
 309  survive:
 310         switch (handle_mm_fault(mm, vma, address, is_write)) {
 311
 312         case VM_FAULT_MINOR:
 313                 current->min_flt++;
 314                 break;
 315         case VM_FAULT_MAJOR:
 316                 current->maj_flt++;
 317                 break;
 318         case VM_FAULT_SIGBUS:
 319                 goto do_sigbus;
 320         case VM_FAULT_OOM:
 321                 goto out_of_memory;
 322         default:
 323                 BUG();
 324         }
 325
 326         up_read(&mm->mmap_sem);
 327         return 0;
 328
 329 bad_area:
 330         up_read(&mm->mmap_sem);
 331
 332 bad_area_nosemaphore:
 333         /* User mode accesses cause a SIGSEGV */
 334         if (user_mode(regs)) {
 335                 _exception(SIGSEGV, regs, code, address);
 336                 return 0;
 337         }
 338
 339         if (is_exec && (error_code & DSISR_PROTFAULT)
 340             && printk_ratelimit())
 341                 printk(KERN_CRIT "kernel tried to execute NX-protected"
 342                        " page (%lx) - exploit attempt? (uid: %d)\n",
 343                        address, current->uid);
 344
 345         return SIGSEGV;
 346
 347 /*
 348  * We ran out of memory, or some other thing happened to us that made
 349  * us unable to handle the page fault gracefully.
 350  */
 351 out_of_memory:
 352         up_read(&mm->mmap_sem);
 353         if (current->pid == 1) {
 354                 yield();
 355                 down_read(&mm->mmap_sem);
 356                 goto survive;
 357         }
 358         printk("VM: killing process %s\n", current->comm);
 359         if (user_mode(regs))
 360                 do_exit(SIGKILL);
 361         return SIGKILL;
 362
 363 do_sigbus:
 364         up_read(&mm->mmap_sem);
 365         if (user_mode(regs)) {
 366                 info.si_signo = SIGBUS;
 367                 info.si_errno = 0;
 368                 info.si_code = BUS_ADRERR;
 369                 info.si_addr = (void __user *)address;
 370                 force_sig_info(SIGBUS, &info, current);
 371                 return 0;
 372         }
 373         return SIGBUS;
 374 }
 375
 376 /*
 377  * bad_page_fault is called when we have a bad access from the kernel.
 378  * It is called from the DSI and ISI handlers in head.S and from some
 379  * of the procedures in traps.c.
 380  */
 381 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
 382 {
 383         const struct exception_table_entry *entry;
 384
 385         /* Are we prepared to handle this fault?  */
 386         if ((entry = search_exception_tables(regs->nip)) != NULL) {
 387                 regs->nip = entry->fixup;
 388                 return;
 389         }
 390
 391         /* kernel has accessed a bad area */
 392
 393         printk(KERN_ALERT "Unable to handle kernel paging request for ");
 394         switch (regs->trap) {
 395                 case 0x300:
 396                 case 0x380:
 397                         printk("data at address 0x%08lx\n", regs->dar);
 398                         break;
 399                 case 0x400:
 400                 case 0x480:
 401                         printk("instruction fetch\n");
 402                         break;
 403                 default:
 404                         printk("unknown fault\n");
 405         }
 406         printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
 407                 regs->nip);
 408
 409         die("Kernel access of bad area", regs, sig);
 410 }