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 #include <linux/perf_event.h>
  33 #include <linux/magic.h>
  34 #include <linux/ratelimit.h>
  35
  36 #include <asm/firmware.h>
  37 #include <asm/page.h>
  38 #include <asm/pgtable.h>
  39 #include <asm/mmu.h>
  40 #include <asm/mmu_context.h>
  41 #include <asm/system.h>
  42 #include <asm/uaccess.h>
  43 #include <asm/tlbflush.h>
  44 #include <asm/siginfo.h>
  45 #include <mm/mmu_decl.h>
  46
  47 #ifdef CONFIG_KPROBES
  48 static inline int notify_page_fault(struct pt_regs *regs)
  49 {
  50         int ret = 0;
  51
  52         /* kprobe_running() needs smp_processor_id() */
  53         if (!user_mode(regs)) {
  54                 preempt_disable();
  55                 if (kprobe_running() && kprobe_fault_handler(regs, 11))
  56                         ret = 1;
  57                 preempt_enable();
  58         }
  59
  60         return ret;
  61 }
  62 #else
  63 static inline int notify_page_fault(struct pt_regs *regs)
  64 {
  65         return 0;
  66 }
  67 #endif
  68
  69 /*
  70  * Check whether the instruction at regs->nip is a store using
  71  * an update addressing form which will update r1.
  72  */
  73 static int store_updates_sp(struct pt_regs *regs)
  74 {
  75         unsigned int inst;
  76
  77         if (get_user(inst, (unsigned int __user *)regs->nip))
  78                 return 0;
  79         /* check for 1 in the rA field */
  80         if (((inst >> 16) & 0x1f) != 1)
  81                 return 0;
  82         /* check major opcode */
  83         switch (inst >> 26) {
  84         case 37:        /* stwu */
  85         case 39:        /* stbu */
  86         case 45:        /* sthu */
  87         case 53:        /* stfsu */
  88         case 55:        /* stfdu */
  89                 return 1;
  90         case 62:        /* std or stdu */
  91                 return (inst & 3) == 1;
  92         case 31:
  93                 /* check minor opcode */
  94                 switch ((inst >> 1) & 0x3ff) {
  95                 case 181:       /* stdux */
  96                 case 183:       /* stwux */
  97                 case 247:       /* stbux */
  98                 case 439:       /* sthux */
  99                 case 695:       /* stfsux */
 100                 case 759:       /* stfdux */
 101                         return 1;
 102                 }
 103         }
 104         return 0;
 105 }
 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, ret;
 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_page_fault(regs))
 147                 return 0;
 148
 149         if (unlikely(debugger_fault_handler(regs)))
 150                 return 0;
 151
 152         /* On a kernel SLB miss we can only check for a valid exception entry */
 153         if (!user_mode(regs) && (address >= TASK_SIZE))
 154                 return SIGSEGV;
 155
 156 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \
 157                              defined(CONFIG_PPC_BOOK3S_64))
 158         if (error_code & DSISR_DABRMATCH) {
 159                 /* DABR match */
 160                 do_dabr(regs, address, error_code);
 161                 return 0;
 162         }
 163 #endif
 164
 165         if (in_atomic() || mm == NULL) {
 166                 if (!user_mode(regs))
 167                         return SIGSEGV;
 168                 /* in_atomic() in user mode is really bad,
 169                    as is current->mm == NULL. */
 170                 printk(KERN_EMERG "Page fault in user mode with "
 171                        "in_atomic() = %d mm = %p\n", in_atomic(), mm);
 172                 printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
 173                        regs->nip, regs->msr);
 174                 die("Weird page fault", regs, SIGSEGV);
 175         }
 176
 177         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
 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.  Unfortunately, in the case of an
 182          * erroneous fault occurring 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 possibility 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         /* 8xx sometimes need to load a invalid/non-present TLBs.
 253          * These must be invalidated separately as linux mm don't.
 254          */
 255         if (error_code & 0x40000000) /* no translation? */
 256                 _tlbil_va(address, 0, 0, 0);
 257
 258         /* The MPC8xx seems to always set 0x80000000, which is
 259          * "undefined".  Of those that can be set, this is the only
 260          * one which seems bad.
 261          */
 262         if (error_code & 0x10000000)
 263                 /* Guarded storage error. */
 264                 goto bad_area;
 265 #endif /* CONFIG_8xx */
 266
 267         if (is_exec) {
 268 #ifdef CONFIG_PPC_STD_MMU
 269                 /* Protection fault on exec go straight to failure on
 270                  * Hash based MMUs as they either don't support per-page
 271                  * execute permission, or if they do, it's handled already
 272                  * at the hash level. This test would probably have to
 273                  * be removed if we change the way this works to make hash
 274                  * processors use the same I/D cache coherency mechanism
 275                  * as embedded.
 276                  */
 277                 if (error_code & DSISR_PROTFAULT)
 278                         goto bad_area;
 279 #endif /* CONFIG_PPC_STD_MMU */
 280
 281                 /*
 282                  * Allow execution from readable areas if the MMU does not
 283                  * provide separate controls over reading and executing.
 284                  *
 285                  * Note: That code used to not be enabled for 4xx/BookE.
 286                  * It is now as I/D cache coherency for these is done at
 287                  * set_pte_at() time and I see no reason why the test
 288                  * below wouldn't be valid on those processors. This -may-
 289                  * break programs compiled with a really old ABI though.
 290                  */
 291                 if (!(vma->vm_flags & VM_EXEC) &&
 292                     (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
 293                      !(vma->vm_flags & (VM_READ | VM_WRITE))))
 294                         goto bad_area;
 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         ret = handle_mm_fault(mm, vma, address, is_write ? FAULT_FLAG_WRITE : 0);
 314         if (unlikely(ret & VM_FAULT_ERROR)) {
 315                 if (ret & VM_FAULT_OOM)
 316                         goto out_of_memory;
 317                 else if (ret & VM_FAULT_SIGBUS)
 318                         goto do_sigbus;
 319                 BUG();
 320         }
 321         if (ret & VM_FAULT_MAJOR) {
 322                 current->maj_flt++;
 323                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
 324                                      regs, address);
 325 #ifdef CONFIG_PPC_SMLPAR
 326                 if (firmware_has_feature(FW_FEATURE_CMO)) {
 327                         preempt_disable();
 328                         get_lppaca()->page_ins += (1 << PAGE_FACTOR);
 329                         preempt_enable();
 330                 }
 331 #endif
 332         } else {
 333                 current->min_flt++;
 334                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
 335                                      regs, address);
 336         }
 337         up_read(&mm->mmap_sem);
 338         return 0;
 339
 340 bad_area:
 341         up_read(&mm->mmap_sem);
 342
 343 bad_area_nosemaphore:
 344         /* User mode accesses cause a SIGSEGV */
 345         if (user_mode(regs)) {
 346                 _exception(SIGSEGV, regs, code, address);
 347                 return 0;
 348         }
 349
 350         if (is_exec && (error_code & DSISR_PROTFAULT))
 351                 printk_ratelimited(KERN_CRIT "kernel tried to execute NX-protected"
 352                                    " page (%lx) - exploit attempt? (uid: %d)\n",
 353                                    address, current_uid());
 354
 355         return SIGSEGV;
 356
 357 /*
 358  * We ran out of memory, or some other thing happened to us that made
 359  * us unable to handle the page fault gracefully.
 360  */
 361 out_of_memory:
 362         up_read(&mm->mmap_sem);
 363         if (!user_mode(regs))
 364                 return SIGKILL;
 365         pagefault_out_of_memory();
 366         return 0;
 367
 368 do_sigbus:
 369         up_read(&mm->mmap_sem);
 370         if (user_mode(regs)) {
 371                 info.si_signo = SIGBUS;
 372                 info.si_errno = 0;
 373                 info.si_code = BUS_ADRERR;
 374                 info.si_addr = (void __user *)address;
 375                 force_sig_info(SIGBUS, &info, current);
 376                 return 0;
 377         }
 378         return SIGBUS;
 379 }
 380
 381 /*
 382  * bad_page_fault is called when we have a bad access from the kernel.
 383  * It is called from the DSI and ISI handlers in head.S and from some
 384  * of the procedures in traps.c.
 385  */
 386 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
 387 {
 388         const struct exception_table_entry *entry;
 389         unsigned long *stackend;
 390
 391         /* Are we prepared to handle this fault?  */
 392         if ((entry = search_exception_tables(regs->nip)) != NULL) {
 393                 regs->nip = entry->fixup;
 394                 return;
 395         }
 396
 397         /* kernel has accessed a bad area */
 398
 399         switch (regs->trap) {
 400         case 0x300:
 401         case 0x380:
 402                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 403                         "data at address 0x%08lx\n", regs->dar);
 404                 break;
 405         case 0x400:
 406         case 0x480:
 407                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 408                         "instruction fetch\n");
 409                 break;
 410         default:
 411                 printk(KERN_ALERT "Unable to handle kernel paging request for "
 412                         "unknown fault\n");
 413                 break;
 414         }
 415         printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
 416                 regs->nip);
 417
 418         stackend = end_of_stack(current);
 419         if (current != &init_task && *stackend != STACK_END_MAGIC)
 420                 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
 421
 422         die("Kernel access of bad area", regs, sig);
 423 }