arch/powerpc/mm/fault.c

   1 /*
   2  *  PowerPC version
   3  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
   4  *
   5  *  Derived from "arch/i386/mm/fault.c"
   6  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   7  *
   8  *  Modified by Cort Dougan and Paul Mackerras.
   9  *
  10  *  Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
  11  *
  12  *  This program is free software; you can redistribute it and/or
  13  *  modify it under the terms of the GNU General Public License
  14  *  as published by the Free Software Foundation; either version
  15  *  2 of the License, or (at your option) any later version.
  16  */
  17
  18 #include <linux/signal.h>
  19 #include <linux/sched.h>
  20 #include <linux/kernel.h>
  21 #include <linux/errno.h>
  22 #include <linux/string.h>
  23 #include <linux/types.h>
  24 #include <linux/ptrace.h>
  25 #include <linux/mman.h>
  26 #include <linux/mm.h>
  27 #include <linux/interrupt.h>
  28 #include <linux/highmem.h>
  29 #include <linux/module.h>
  30 #include <linux/kprobes.h>
  31 #include <linux/kdebug.h>
  32
  33 #include <asm/firmware.h>
  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/tlbflush.h>
  41 #include <asm/siginfo.h>
  42
  43
  44 #ifdef CONFIG_KPROBES
  45 static inline int notify_page_fault(struct pt_regs *regs)
  46 {
  47         int ret = 0;
  48
  49         /* kprobe_running() needs smp_processor_id() */
  50         if (!user_mode(regs)) {
  51                 preempt_disable();
  52                 if (kprobe_running() && kprobe_fault_handler(regs, 11))
  53                         ret = 1;
  54                 preempt_enable();
  55         }
  56
  57         return ret;
  58 }
  59 #else
  60 static inline int notify_page_fault(struct pt_regs *regs)
  61 {
  62         return 0;
  63 }
  64 #endif
  65
  66 /*
  67  * Check whether the instruction at regs->nip is a store using
  68  * an update addressing form which will update r1.
  69  */
  70 static int store_updates_sp(struct pt_regs *regs)
  71 {
  72         unsigned int inst;
  73
  74         if (get_user(inst, (unsigned int __user *)regs->nip))
  75                 return 0;
  76         /* check for 1 in the rA field */
  77         if (((inst >> 16) & 0x1f) != 1)
  78                 return 0;
  79         /* check major opcode */
  80         switch (inst >> 26) {
  81         case 37:        /* stwu */
  82         case 39:        /* stbu */
  83         case 45:        /* sthu */
  84         case 53:        /* stfsu */
  85         case 55:        /* stfdu */
  86                 return 1;
  87         case 62:        /* std or stdu */
  88                 return (inst & 3) == 1;
  89         case 31:
  90                 /* check minor opcode */
  91                 switch ((inst >> 1) & 0x3ff) {
  92                 case 181:       /* stdux */
  93                 case 183:       /* stwux */
  94                 case 247:       /* stbux */
  95                 case 439:       /* sthux */
  96                 case 695:       /* stfsux */
  97                 case 759:       /* stfdux */
  98                         return 1;
  99                 }
 100         }
 101         return 0;
 102 }
 103
 104 /*
 105  * For 600- and 800-family processors, the error_code parameter is DSISR
 106  * for a data fault, SRR1 for an instruction fault. For 400-family processors
 107  * the error_code parameter is ESR for a data fault, 0 for an instruction
 108  * fault.
 109  * For 64-bit processors, the error_code parameter is
 110  *  - DSISR for a non-SLB data access fault,
 111  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
 112  *  - 0 any SLB fault.
 113  *
 114  * The return value is 0 if the fault was handled, or the signal
 115  * number if this is a kernel fault that can't be handled here.
 116  */
 117 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
 118                             unsigned long error_code)
 119 {
 120         struct vm_area_struct * vma;
 121         struct mm_struct *mm = current->mm;
 122         siginfo_t info;
 123         int code = SEGV_MAPERR;
 124         int is_write = 0, ret;
 125         int trap = TRAP(regs);
 126         int is_exec = trap == 0x400;
 127
 128 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
 129         /*
 130          * Fortunately the bit assignments in SRR1 for an instruction
 131          * fault and DSISR for a data fault are mostly the same for the
 132          * bits we are interested in.  But there are some bits which
 133          * indicate errors in DSISR but can validly be set in SRR1.
 134          */
 135         if (trap == 0x400)
 136                 error_code &= 0x48200000;
 137         else
 138                 is_write = error_code & DSISR_ISSTORE;
 139 #else
 140         is_write = error_code & ESR_DST;
 141 #endif /* CONFIG_4xx || CONFIG_BOOKE */
 142
 143         if (notify_page_fault(regs))
 144                 return 0;
 145
 146         if (unlikely(debugger_fault_handler(regs)))
 147                 return 0;
 148
 149         /* On a kernel SLB miss we can only check for a valid exception entry */
 150         if (!user_mode(regs) && (address >= TASK_SIZE))
 151                 return SIGSEGV;
 152
 153 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
 154         if (error_code & DSISR_DABRMATCH) {
 155                 /* DABR match */
 156                 do_dabr(regs, address, error_code);
 157                 return 0;
 158         }
 159 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
 160
 161         if (in_atomic() || mm == NULL) {
 162                 if (!user_mode(regs))
 163                         return SIGSEGV;
 164                 /* in_atomic() in user mode is really bad,
 165                    as is current->mm == NULL. */
 166                 printk(KERN_EMERG "Page fault in user mode with "
 167                        "in_atomic() = %d mm = %p\n", in_atomic(), mm);
 168                 printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
 169                        regs->nip, regs->msr);
 170                 die("Weird page fault", regs, SIGSEGV);
 171         }
 172
 173         /* When running in the kernel we expect faults to occur only to
 174          * addresses in user space.  All other faults represent errors in the
 175          * kernel and should generate an OOPS.  Unfortunately, in the case of an
 176          * erroneous fault occurring in a code path which already holds mmap_sem
 177          * we will deadlock attempting to validate the fault against the
 178          * address space.  Luckily the kernel only validly references user
 179          * space from well defined areas of code, which are listed in the
 180          * exceptions table.
 181          *
 182          * As the vast majority of faults will be valid we will only perform
 183          * the source reference check when there is a possibility of a deadlock.
 184          * Attempt to lock the address space, if we cannot we then validate the
 185          * source.  If this is invalid we can skip the address space check,
 186          * thus avoiding the deadlock.
 187          */
 188         if (!down_read_trylock(&mm->mmap_sem)) {
 189                 if (!user_mode(regs) && !search_exception_tables(regs->nip))
 190                         goto bad_area_nosemaphore;
 191
 192                 down_read(&mm->mmap_sem);
 193         }
 194
 195         vma = find_vma(mm, address);
 196         if (!vma)
 197                 goto bad_area;
 198         if (vma->vm_start <= address)
 199                 goto good_area;
 200         if (!(vma->vm_flags & VM_GROWSDOWN))
 201                 goto bad_area;
 202
 203         /*
 204          * N.B. The POWER/Open ABI allows programs to access up to
 205          * 288 bytes below the stack pointer.
 206          * The kernel signal delivery code writes up to about 1.5kB
 207          * below the stack pointer (r1) before decrementing it.
 208          * The exec code can write slightly over 640kB to the stack
 209          * before setting the user r1.  Thus we allow the stack to
 210          * expand to 1MB without further checks.
 211          */
 212         if (address + 0x100000 < vma->vm_end) {
 213                 /* get user regs even if this fault is in kernel mode */
 214                 struct pt_regs *uregs = current->thread.regs;
 215                 if (uregs == NULL)
 216                         goto bad_area;
 217
 218                 /*
 219                  * A user-mode access to an address a long way below
 220                  * the stack pointer is only valid if the instruction
 221                  * is one which would update the stack pointer to the
 222                  * address accessed if the instruction completed,
 223                  * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
 224                  * (or the byte, halfword, float or double forms).
 225                  *
 226                  * If we don't check this then any write to the area
 227                  * between the last mapped region and the stack will
 228                  * expand the stack rather than segfaulting.
 229                  */
 230                 if (address + 2048 < uregs->gpr[1]
 231                     && (!user_mode(regs) || !store_updates_sp(regs)))
 232                         goto bad_area;
 233         }
 234         if (expand_stack(vma, address))
 235                 goto bad_area;
 236
 237 good_area:
 238         code = SEGV_ACCERR;
 239 #if defined(CONFIG_6xx)
 240         if (error_code & 0x95700000)
 241                 /* an error such as lwarx to I/O controller space,
 242                    address matching DABR, eciwx, etc. */
 243                 goto bad_area;
 244 #endif /* CONFIG_6xx */
 245 #if defined(CONFIG_8xx)
 246         /* The MPC8xx seems to always set 0x80000000, which is
 247          * "undefined".  Of those that can be set, this is the only
 248          * one which seems bad.
 249          */
 250         if (error_code & 0x10000000)
 251                 /* Guarded storage error. */
 252                 goto bad_area;
 253 #endif /* CONFIG_8xx */
 254
 255         if (is_exec) {
 256 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
 257                 /* protection fault */
 258                 if (error_code & DSISR_PROTFAULT)
 259                         goto bad_area;
 260                 /*
 261                  * Allow execution from readable areas if the MMU does not
 262                  * provide separate controls over reading and executing.
 263                  */
 264                 if (!(vma->vm_flags & VM_EXEC) &&
 265                     (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
 266                      !(vma->vm_flags & (VM_READ | VM_WRITE))))
 267                         goto bad_area;
 268 #else
 269                 pte_t *ptep;
 270                 pmd_t *pmdp;
 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, &pmdp)) {
 276                         spinlock_t *ptl = pte_lockptr(mm, pmdp);
 277                         spin_lock(ptl);
 278                         if (pte_present(*ptep)) {
 279                                 struct page *page = pte_page(*ptep);
 280
 281                                 if (!test_bit(PG_arch_1, &page->flags)) {
 282                                         flush_dcache_icache_page(page);
 283                                         set_bit(PG_arch_1, &page->flags);
 284                                 }
 285                                 pte_update(ptep, 0, _PAGE_HWEXEC |
 286                                            _PAGE_ACCESSED);
 287                                 local_flush_tlb_page(vma, address);
 288                                 pte_unmap_unlock(ptep, ptl);
 289                                 up_read(&mm->mmap_sem);
 290                                 return 0;
 291                         }
 292                         pte_unmap_unlock(ptep, ptl);
 293                 }
 294 #endif
 295         /* a write */
 296         } else if (is_write) {
 297                 if (!(vma->vm_flags & VM_WRITE))
 298                         goto bad_area;
 299         /* a read */
 300         } else {
 301                 /* protection fault */
 302                 if (error_code & 0x08000000)
 303                         goto bad_area;
 304                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 305                         goto bad_area;
 306         }
 307
 308         /*
 309          * If for any reason at all we couldn't handle the fault,
 310          * make sure we exit gracefully rather than endlessly redo
 311          * the fault.
 312          */
 313  survive:
 314         ret = handle_mm_fault(mm, vma, address, is_write);
 315         if (unlikely(ret & VM_FAULT_ERROR)) {
 316                 if (ret & VM_FAULT_OOM)
 317                         goto out_of_memory;
 318                 else if (ret & VM_FAULT_SIGBUS)
 319                         goto do_sigbus;
 320                 BUG();
 321         }
 322         if (ret & VM_FAULT_MAJOR) {
 323                 current->maj_flt++;
 324 #ifdef CONFIG_PPC_SMLPAR
 325                 if (firmware_has_feature(FW_FEATURE_CMO)) {
 326                         preempt_disable();
 327                         get_lppaca()->page_ins += (1 << PAGE_FACTOR);
 328                         preempt_enable();
 329                 }
 330 #endif
 331         } else
 332                 current->min_flt++;
 333         up_read(&mm->mmap_sem);
 334         return 0;
 335
 336 bad_area:
 337         up_read(&mm->mmap_sem);
 338
 339 bad_area_nosemaphore:
 340         /* User mode accesses cause a SIGSEGV */
 341         if (user_mode(regs)) {
 342                 _exception(SIGSEGV, regs, code, address);
 343                 return 0;
 344         }
 345
 346         if (is_exec && (error_code & DSISR_PROTFAULT)
 347             && printk_ratelimit())
 348                 printk(KERN_CRIT "kernel tried to execute NX-protected"
 349                        " page (%lx) - exploit attempt? (uid: %d)\n",
 350                        address, current->uid);
 351
 352         return SIGSEGV;
 353
 354 /*
 355  * We ran out of memory, or some other thing happened to us that made
 356  * us unable to handle the page fault gracefully.
 357  */
 358 out_of_memory:
 359         up_read(&mm->mmap_sem);
 360         if (is_global_init(current)) {
 361                 yield();
 362                 down_read(&mm->mmap_sem);
 363                 goto survive;
 364         }
 365         printk("VM: killing process %s\n", current->comm);
 366         if (user_mode(regs))
 367                 do_group_exit(SIGKILL);
 368         return SIGKILL;
 369
 370 do_sigbus:
 371         up_read(&mm->mmap_sem);
 372         if (user_mode(regs)) {
 373                 info.si_signo = SIGBUS;
 374                 info.si_errno = 0;
 375                 info.si_code = BUS_ADRERR;
 376                 info.si_addr = (void __user *)address;
 377                 force_sig_info(SIGBUS, &info, current);
 378                 return 0;
 379         }
 380         return SIGBUS;
 381 }
 382
 383 /*
 384  * bad_page_fault is called when we have a bad access from the kernel.
 385  * It is called from the DSI and ISI handlers in head.S and from some
 386  * of the procedures in traps.c.
 387  */
 388 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
 389 {
 390         const struct exception_table_entry *entry;
 391
 392         /* Are we prepared to handle this fault?  */
 393         if ((entry = search_exception_tables(regs->nip)) != NULL) {
 394                 regs->nip = entry->fixup;
 395                 return;
 396         }
 397
 398         /* kernel has accessed a bad area */
 399
 400         switch (regs->trap) {
 401         case 0x300:
 402         case 0x380:
 403                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 404                         "data at address 0x%08lx\n", regs->dar);
 405                 break;
 406         case 0x400:
 407         case 0x480:
 408                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 409                         "instruction fetch\n");
 410                 break;
 411         default:
 412                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 413                         "unknown fault\n");
 414                 break;
 415         }
 416         printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
 417                 regs->nip);
 418
 419         die("Kernel access of bad area", regs, sig);
 420 }