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/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 #include <asm/tlbflush.h>
  40 #include <asm/siginfo.h>
  41
  42
  43 #ifdef CONFIG_KPROBES
  44 static inline int notify_page_fault(struct pt_regs *regs)
  45 {
  46         int ret = 0;
  47
  48         /* kprobe_running() needs smp_processor_id() */
  49         if (!user_mode(regs)) {
  50                 preempt_disable();
  51                 if (kprobe_running() && kprobe_fault_handler(regs, 11))
  52                         ret = 1;
  53                 preempt_enable();
  54         }
  55
  56         return ret;
  57 }
  58 #else
  59 static inline int notify_page_fault(struct pt_regs *regs)
  60 {
  61         return 0;
  62 }
  63 #endif
  64
  65 /*
  66  * Check whether the instruction at regs->nip is a store using
  67  * an update addressing form which will update r1.
  68  */
  69 static int store_updates_sp(struct pt_regs *regs)
  70 {
  71         unsigned int inst;
  72
  73         if (get_user(inst, (unsigned int __user *)regs->nip))
  74                 return 0;
  75         /* check for 1 in the rA field */
  76         if (((inst >> 16) & 0x1f) != 1)
  77                 return 0;
  78         /* check major opcode */
  79         switch (inst >> 26) {
  80         case 37:        /* stwu */
  81         case 39:        /* stbu */
  82         case 45:        /* sthu */
  83         case 53:        /* stfsu */
  84         case 55:        /* stfdu */
  85                 return 1;
  86         case 62:        /* std or stdu */
  87                 return (inst & 3) == 1;
  88         case 31:
  89                 /* check minor opcode */
  90                 switch ((inst >> 1) & 0x3ff) {
  91                 case 181:       /* stdux */
  92                 case 183:       /* stwux */
  93                 case 247:       /* stbux */
  94                 case 439:       /* sthux */
  95                 case 695:       /* stfsux */
  96                 case 759:       /* stfdux */
  97                         return 1;
  98                 }
  99         }
 100         return 0;
 101 }
 102
 103 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
 104 static void do_dabr(struct pt_regs *regs, unsigned long address,
 105                     unsigned long error_code)
 106 {
 107         siginfo_t info;
 108
 109         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 110                         11, SIGSEGV) == NOTIFY_STOP)
 111                 return;
 112
 113         if (debugger_dabr_match(regs))
 114                 return;
 115
 116         /* Clear the DABR */
 117         set_dabr(0);
 118
 119         /* Deliver the signal to userspace */
 120         info.si_signo = SIGTRAP;
 121         info.si_errno = 0;
 122         info.si_code = TRAP_HWBKPT;
 123         info.si_addr = (void __user *)address;
 124         force_sig_info(SIGTRAP, &info, current);
 125 }
 126 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
 127
 128 /*
 129  * For 600- and 800-family processors, the error_code parameter is DSISR
 130  * for a data fault, SRR1 for an instruction fault. For 400-family processors
 131  * the error_code parameter is ESR for a data fault, 0 for an instruction
 132  * fault.
 133  * For 64-bit processors, the error_code parameter is
 134  *  - DSISR for a non-SLB data access fault,
 135  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
 136  *  - 0 any SLB fault.
 137  *
 138  * The return value is 0 if the fault was handled, or the signal
 139  * number if this is a kernel fault that can't be handled here.
 140  */
 141 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
 142                             unsigned long error_code)
 143 {
 144         struct vm_area_struct * vma;
 145         struct mm_struct *mm = current->mm;
 146         siginfo_t info;
 147         int code = SEGV_MAPERR;
 148         int is_write = 0, ret;
 149         int trap = TRAP(regs);
 150         int is_exec = trap == 0x400;
 151
 152 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
 153         /*
 154          * Fortunately the bit assignments in SRR1 for an instruction
 155          * fault and DSISR for a data fault are mostly the same for the
 156          * bits we are interested in.  But there are some bits which
 157          * indicate errors in DSISR but can validly be set in SRR1.
 158          */
 159         if (trap == 0x400)
 160                 error_code &= 0x48200000;
 161         else
 162                 is_write = error_code & DSISR_ISSTORE;
 163 #else
 164         is_write = error_code & ESR_DST;
 165 #endif /* CONFIG_4xx || CONFIG_BOOKE */
 166
 167         if (notify_page_fault(regs))
 168                 return 0;
 169
 170         if (unlikely(debugger_fault_handler(regs)))
 171                 return 0;
 172
 173         /* On a kernel SLB miss we can only check for a valid exception entry */
 174         if (!user_mode(regs) && (address >= TASK_SIZE))
 175                 return SIGSEGV;
 176
 177 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
 178         if (error_code & DSISR_DABRMATCH) {
 179                 /* DABR match */
 180                 do_dabr(regs, address, error_code);
 181                 return 0;
 182         }
 183 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
 184
 185         if (in_atomic() || mm == NULL) {
 186                 if (!user_mode(regs))
 187                         return SIGSEGV;
 188                 /* in_atomic() in user mode is really bad,
 189                    as is current->mm == NULL. */
 190                 printk(KERN_EMERG "Page fault in user mode with "
 191                        "in_atomic() = %d mm = %p\n", in_atomic(), mm);
 192                 printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
 193                        regs->nip, regs->msr);
 194                 die("Weird page fault", regs, SIGSEGV);
 195         }
 196
 197         /* When running in the kernel we expect faults to occur only to
 198          * addresses in user space.  All other faults represent errors in the
 199          * kernel and should generate an OOPS.  Unfortunately, in the case of an
 200          * erroneous fault occurring in a code path which already holds mmap_sem
 201          * we will deadlock attempting to validate the fault against the
 202          * address space.  Luckily the kernel only validly references user
 203          * space from well defined areas of code, which are listed in the
 204          * exceptions table.
 205          *
 206          * As the vast majority of faults will be valid we will only perform
 207          * the source reference check when there is a possibility of a deadlock.
 208          * Attempt to lock the address space, if we cannot we then validate the
 209          * source.  If this is invalid we can skip the address space check,
 210          * thus avoiding the deadlock.
 211          */
 212         if (!down_read_trylock(&mm->mmap_sem)) {
 213                 if (!user_mode(regs) && !search_exception_tables(regs->nip))
 214                         goto bad_area_nosemaphore;
 215
 216                 down_read(&mm->mmap_sem);
 217         }
 218
 219         vma = find_vma(mm, address);
 220         if (!vma)
 221                 goto bad_area;
 222         if (vma->vm_start <= address)
 223                 goto good_area;
 224         if (!(vma->vm_flags & VM_GROWSDOWN))
 225                 goto bad_area;
 226
 227         /*
 228          * N.B. The POWER/Open ABI allows programs to access up to
 229          * 288 bytes below the stack pointer.
 230          * The kernel signal delivery code writes up to about 1.5kB
 231          * below the stack pointer (r1) before decrementing it.
 232          * The exec code can write slightly over 640kB to the stack
 233          * before setting the user r1.  Thus we allow the stack to
 234          * expand to 1MB without further checks.
 235          */
 236         if (address + 0x100000 < vma->vm_end) {
 237                 /* get user regs even if this fault is in kernel mode */
 238                 struct pt_regs *uregs = current->thread.regs;
 239                 if (uregs == NULL)
 240                         goto bad_area;
 241
 242                 /*
 243                  * A user-mode access to an address a long way below
 244                  * the stack pointer is only valid if the instruction
 245                  * is one which would update the stack pointer to the
 246                  * address accessed if the instruction completed,
 247                  * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
 248                  * (or the byte, halfword, float or double forms).
 249                  *
 250                  * If we don't check this then any write to the area
 251                  * between the last mapped region and the stack will
 252                  * expand the stack rather than segfaulting.
 253                  */
 254                 if (address + 2048 < uregs->gpr[1]
 255                     && (!user_mode(regs) || !store_updates_sp(regs)))
 256                         goto bad_area;
 257         }
 258         if (expand_stack(vma, address))
 259                 goto bad_area;
 260
 261 good_area:
 262         code = SEGV_ACCERR;
 263 #if defined(CONFIG_6xx)
 264         if (error_code & 0x95700000)
 265                 /* an error such as lwarx to I/O controller space,
 266                    address matching DABR, eciwx, etc. */
 267                 goto bad_area;
 268 #endif /* CONFIG_6xx */
 269 #if defined(CONFIG_8xx)
 270         /* The MPC8xx seems to always set 0x80000000, which is
 271          * "undefined".  Of those that can be set, this is the only
 272          * one which seems bad.
 273          */
 274         if (error_code & 0x10000000)
 275                 /* Guarded storage error. */
 276                 goto bad_area;
 277 #endif /* CONFIG_8xx */
 278
 279         if (is_exec) {
 280 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
 281                 /* protection fault */
 282                 if (error_code & DSISR_PROTFAULT)
 283                         goto bad_area;
 284                 /*
 285                  * Allow execution from readable areas if the MMU does not
 286                  * provide separate controls over reading and executing.
 287                  */
 288                 if (!(vma->vm_flags & VM_EXEC) &&
 289                     (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
 290                      !(vma->vm_flags & (VM_READ | VM_WRITE))))
 291                         goto bad_area;
 292 #else
 293                 pte_t *ptep;
 294                 pmd_t *pmdp;
 295
 296                 /* Since 4xx/Book-E supports per-page execute permission,
 297                  * we lazily flush dcache to icache. */
 298                 ptep = NULL;
 299                 if (get_pteptr(mm, address, &ptep, &pmdp)) {
 300                         spinlock_t *ptl = pte_lockptr(mm, pmdp);
 301                         spin_lock(ptl);
 302                         if (pte_present(*ptep)) {
 303                                 struct page *page = pte_page(*ptep);
 304
 305                                 if (!test_bit(PG_arch_1, &page->flags)) {
 306                                         flush_dcache_icache_page(page);
 307                                         set_bit(PG_arch_1, &page->flags);
 308                                 }
 309                                 pte_update(ptep, 0, _PAGE_HWEXEC);
 310                                 _tlbie(address, mm->context.id);
 311                                 pte_unmap_unlock(ptep, ptl);
 312                                 up_read(&mm->mmap_sem);
 313                                 return 0;
 314                         }
 315                         pte_unmap_unlock(ptep, ptl);
 316                 }
 317 #endif
 318         /* a write */
 319         } else if (is_write) {
 320                 if (!(vma->vm_flags & VM_WRITE))
 321                         goto bad_area;
 322         /* a read */
 323         } else {
 324                 /* protection fault */
 325                 if (error_code & 0x08000000)
 326                         goto bad_area;
 327                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 328                         goto bad_area;
 329         }
 330
 331         /*
 332          * If for any reason at all we couldn't handle the fault,
 333          * make sure we exit gracefully rather than endlessly redo
 334          * the fault.
 335          */
 336  survive:
 337         ret = handle_mm_fault(mm, vma, address, is_write);
 338         if (unlikely(ret & VM_FAULT_ERROR)) {
 339                 if (ret & VM_FAULT_OOM)
 340                         goto out_of_memory;
 341                 else if (ret & VM_FAULT_SIGBUS)
 342                         goto do_sigbus;
 343                 BUG();
 344         }
 345         if (ret & VM_FAULT_MAJOR)
 346                 current->maj_flt++;
 347         else
 348                 current->min_flt++;
 349         up_read(&mm->mmap_sem);
 350         return 0;
 351
 352 bad_area:
 353         up_read(&mm->mmap_sem);
 354
 355 bad_area_nosemaphore:
 356         /* User mode accesses cause a SIGSEGV */
 357         if (user_mode(regs)) {
 358                 _exception(SIGSEGV, regs, code, address);
 359                 return 0;
 360         }
 361
 362         if (is_exec && (error_code & DSISR_PROTFAULT)
 363             && printk_ratelimit())
 364                 printk(KERN_CRIT "kernel tried to execute NX-protected"
 365                        " page (%lx) - exploit attempt? (uid: %d)\n",
 366                        address, current->uid);
 367
 368         return SIGSEGV;
 369
 370 /*
 371  * We ran out of memory, or some other thing happened to us that made
 372  * us unable to handle the page fault gracefully.
 373  */
 374 out_of_memory:
 375         up_read(&mm->mmap_sem);
 376         if (is_global_init(current)) {
 377                 yield();
 378                 down_read(&mm->mmap_sem);
 379                 goto survive;
 380         }
 381         printk("VM: killing process %s\n", current->comm);
 382         if (user_mode(regs))
 383                 do_group_exit(SIGKILL);
 384         return SIGKILL;
 385
 386 do_sigbus:
 387         up_read(&mm->mmap_sem);
 388         if (user_mode(regs)) {
 389                 info.si_signo = SIGBUS;
 390                 info.si_errno = 0;
 391                 info.si_code = BUS_ADRERR;
 392                 info.si_addr = (void __user *)address;
 393                 force_sig_info(SIGBUS, &info, current);
 394                 return 0;
 395         }
 396         return SIGBUS;
 397 }
 398
 399 /*
 400  * bad_page_fault is called when we have a bad access from the kernel.
 401  * It is called from the DSI and ISI handlers in head.S and from some
 402  * of the procedures in traps.c.
 403  */
 404 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
 405 {
 406         const struct exception_table_entry *entry;
 407
 408         /* Are we prepared to handle this fault?  */
 409         if ((entry = search_exception_tables(regs->nip)) != NULL) {
 410                 regs->nip = entry->fixup;
 411                 return;
 412         }
 413
 414         /* kernel has accessed a bad area */
 415
 416         switch (regs->trap) {
 417         case 0x300:
 418         case 0x380:
 419                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 420                         "data at address 0x%08lx\n", regs->dar);
 421                 break;
 422         case 0x400:
 423         case 0x480:
 424                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 425                         "instruction fetch\n");
 426                 break;
 427         default:
 428                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 429                         "unknown fault\n");
 430                 break;
 431         }
 432         printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
 433                 regs->nip);
 434
 435         die("Kernel access of bad area", regs, sig);
 436 }