arch/unicore32/mm/fault.c

   1 /*
   2  * linux/arch/unicore32/mm/fault.c
   3  *
   4  * Code specific to PKUnity SoC and UniCore ISA
   5  *
   6  * Copyright (C) 2001-2010 GUAN Xue-tao
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 as
  10  * published by the Free Software Foundation.
  11  */
  12 #include <linux/module.h>
  13 #include <linux/signal.h>
  14 #include <linux/mm.h>
  15 #include <linux/hardirq.h>
  16 #include <linux/init.h>
  17 #include <linux/kprobes.h>
  18 #include <linux/uaccess.h>
  19 #include <linux/page-flags.h>
  20 #include <linux/sched.h>
  21 #include <linux/io.h>
  22
  23 #include <asm/pgtable.h>
  24 #include <asm/tlbflush.h>
  25
  26 /*
  27  * Fault status register encodings.  We steal bit 31 for our own purposes.
  28  */
  29 #define FSR_LNX_PF              (1 << 31)
  30
  31 static inline int fsr_fs(unsigned int fsr)
  32 {
  33         /* xyabcde will be abcde+xy */
  34         return (fsr & 31) + ((fsr & (3 << 5)) >> 5);
  35 }
  36
  37 /*
  38  * This is useful to dump out the page tables associated with
  39  * 'addr' in mm 'mm'.
  40  */
  41 void show_pte(struct mm_struct *mm, unsigned long addr)
  42 {
  43         pgd_t *pgd;
  44
  45         if (!mm)
  46                 mm = &init_mm;
  47
  48         printk(KERN_ALERT "pgd = %p\n", mm->pgd);
  49         pgd = pgd_offset(mm, addr);
  50         printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
  51
  52         do {
  53                 pmd_t *pmd;
  54                 pte_t *pte;
  55
  56                 if (pgd_none(*pgd))
  57                         break;
  58
  59                 if (pgd_bad(*pgd)) {
  60                         printk("(bad)");
  61                         break;
  62                 }
  63
  64                 pmd = pmd_offset((pud_t *) pgd, addr);
  65                 if (PTRS_PER_PMD != 1)
  66                         printk(", *pmd=%08lx", pmd_val(*pmd));
  67
  68                 if (pmd_none(*pmd))
  69                         break;
  70
  71                 if (pmd_bad(*pmd)) {
  72                         printk("(bad)");
  73                         break;
  74                 }
  75
  76                 /* We must not map this if we have highmem enabled */
  77                 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
  78                         break;
  79
  80                 pte = pte_offset_map(pmd, addr);
  81                 printk(", *pte=%08lx", pte_val(*pte));
  82                 pte_unmap(pte);
  83         } while (0);
  84
  85         printk("\n");
  86 }
  87
  88 /*
  89  * Oops.  The kernel tried to access some page that wasn't present.
  90  */
  91 static void __do_kernel_fault(struct mm_struct *mm, unsigned long addr,
  92                 unsigned int fsr, struct pt_regs *regs)
  93 {
  94         /*
  95          * Are we prepared to handle this kernel fault?
  96          */
  97         if (fixup_exception(regs))
  98                 return;
  99
 100         /*
 101          * No handler, we'll have to terminate things with extreme prejudice.
 102          */
 103         bust_spinlocks(1);
 104         printk(KERN_ALERT
 105                "Unable to handle kernel %s at virtual address %08lx\n",
 106                (addr < PAGE_SIZE) ? "NULL pointer dereference" :
 107                "paging request", addr);
 108
 109         show_pte(mm, addr);
 110         die("Oops", regs, fsr);
 111         bust_spinlocks(0);
 112         do_exit(SIGKILL);
 113 }
 114
 115 /*
 116  * Something tried to access memory that isn't in our memory map..
 117  * User mode accesses just cause a SIGSEGV
 118  */
 119 static void __do_user_fault(struct task_struct *tsk, unsigned long addr,
 120                 unsigned int fsr, unsigned int sig, int code,
 121                 struct pt_regs *regs)
 122 {
 123         struct siginfo si;
 124
 125         tsk->thread.address = addr;
 126         tsk->thread.error_code = fsr;
 127         tsk->thread.trap_no = 14;
 128         si.si_signo = sig;
 129         si.si_errno = 0;
 130         si.si_code = code;
 131         si.si_addr = (void __user *)addr;
 132         force_sig_info(sig, &si, tsk);
 133 }
 134
 135 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 136 {
 137         struct task_struct *tsk = current;
 138         struct mm_struct *mm = tsk->active_mm;
 139
 140         /*
 141          * If we are in kernel mode at this point, we
 142          * have no context to handle this fault with.
 143          */
 144         if (user_mode(regs))
 145                 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
 146         else
 147                 __do_kernel_fault(mm, addr, fsr, regs);
 148 }
 149
 150 #define VM_FAULT_BADMAP         0x010000
 151 #define VM_FAULT_BADACCESS      0x020000
 152
 153 /*
 154  * Check that the permissions on the VMA allow for the fault which occurred.
 155  * If we encountered a write fault, we must have write permission, otherwise
 156  * we allow any permission.
 157  */
 158 static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
 159 {
 160         unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
 161
 162         if (!(fsr ^ 0x12))      /* write? */
 163                 mask = VM_WRITE;
 164         if (fsr & FSR_LNX_PF)
 165                 mask = VM_EXEC;
 166
 167         return vma->vm_flags & mask ? false : true;
 168 }
 169
 170 static int __do_pf(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
 171                 unsigned int flags, struct task_struct *tsk)
 172 {
 173         struct vm_area_struct *vma;
 174         int fault;
 175
 176         vma = find_vma(mm, addr);
 177         fault = VM_FAULT_BADMAP;
 178         if (unlikely(!vma))
 179                 goto out;
 180         if (unlikely(vma->vm_start > addr))
 181                 goto check_stack;
 182
 183         /*
 184          * Ok, we have a good vm_area for this
 185          * memory access, so we can handle it.
 186          */
 187 good_area:
 188         if (access_error(fsr, vma)) {
 189                 fault = VM_FAULT_BADACCESS;
 190                 goto out;
 191         }
 192
 193         /*
 194          * If for any reason at all we couldn't handle the fault, make
 195          * sure we exit gracefully rather than endlessly redo the fault.
 196          */
 197         fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, flags);
 198         return fault;
 199
 200 check_stack:
 201         if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
 202                 goto good_area;
 203 out:
 204         return fault;
 205 }
 206
 207 static int do_pf(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 208 {
 209         struct task_struct *tsk;
 210         struct mm_struct *mm;
 211         int fault, sig, code;
 212         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
 213                                  ((!(fsr ^ 0x12)) ? FAULT_FLAG_WRITE : 0);
 214
 215         tsk = current;
 216         mm = tsk->mm;
 217
 218         /*
 219          * If we're in an interrupt or have no user
 220          * context, we must not take the fault..
 221          */
 222         if (in_atomic() || !mm)
 223                 goto no_context;
 224
 225         /*
 226          * As per x86, we may deadlock here.  However, since the kernel only
 227          * validly references user space from well defined areas of the code,
 228          * we can bug out early if this is from code which shouldn't.
 229          */
 230         if (!down_read_trylock(&mm->mmap_sem)) {
 231                 if (!user_mode(regs)
 232                     && !search_exception_tables(regs->UCreg_pc))
 233                         goto no_context;
 234 retry:
 235                 down_read(&mm->mmap_sem);
 236         } else {
 237                 /*
 238                  * The above down_read_trylock() might have succeeded in
 239                  * which case, we'll have missed the might_sleep() from
 240                  * down_read()
 241                  */
 242                 might_sleep();
 243 #ifdef CONFIG_DEBUG_VM
 244                 if (!user_mode(regs) &&
 245                     !search_exception_tables(regs->UCreg_pc))
 246                         goto no_context;
 247 #endif
 248         }
 249
 250         fault = __do_pf(mm, addr, fsr, flags, tsk);
 251
 252         /* If we need to retry but a fatal signal is pending, handle the
 253          * signal first. We do not need to release the mmap_sem because
 254          * it would already be released in __lock_page_or_retry in
 255          * mm/filemap.c. */
 256         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 257                 return 0;
 258
 259         if (!(fault & VM_FAULT_ERROR) && (flags & FAULT_FLAG_ALLOW_RETRY)) {
 260                 if (fault & VM_FAULT_MAJOR)
 261                         tsk->maj_flt++;
 262                 else
 263                         tsk->min_flt++;
 264                 if (fault & VM_FAULT_RETRY) {
 265                         /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
 266                         * of starvation. */
 267                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 268                         goto retry;
 269                 }
 270         }
 271
 272         up_read(&mm->mmap_sem);
 273
 274         /*
 275          * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
 276          */
 277         if (likely(!(fault &
 278                (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
 279                 return 0;
 280
 281         if (fault & VM_FAULT_OOM) {
 282                 /*
 283                  * We ran out of memory, call the OOM killer, and return to
 284                  * userspace (which will retry the fault, or kill us if we
 285                  * got oom-killed)
 286                  */
 287                 pagefault_out_of_memory();
 288                 return 0;
 289         }
 290
 291         /*
 292          * If we are in kernel mode at this point, we
 293          * have no context to handle this fault with.
 294          */
 295         if (!user_mode(regs))
 296                 goto no_context;
 297
 298         if (fault & VM_FAULT_SIGBUS) {
 299                 /*
 300                  * We had some memory, but were unable to
 301                  * successfully fix up this page fault.
 302                  */
 303                 sig = SIGBUS;
 304                 code = BUS_ADRERR;
 305         } else {
 306                 /*
 307                  * Something tried to access memory that
 308                  * isn't in our memory map..
 309                  */
 310                 sig = SIGSEGV;
 311                 code = fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR;
 312         }
 313
 314         __do_user_fault(tsk, addr, fsr, sig, code, regs);
 315         return 0;
 316
 317 no_context:
 318         __do_kernel_fault(mm, addr, fsr, regs);
 319         return 0;
 320 }
 321
 322 /*
 323  * First Level Translation Fault Handler
 324  *
 325  * We enter here because the first level page table doesn't contain
 326  * a valid entry for the address.
 327  *
 328  * If the address is in kernel space (>= TASK_SIZE), then we are
 329  * probably faulting in the vmalloc() area.
 330  *
 331  * If the init_task's first level page tables contains the relevant
 332  * entry, we copy the it to this task.  If not, we send the process
 333  * a signal, fixup the exception, or oops the kernel.
 334  *
 335  * NOTE! We MUST NOT take any locks for this case. We may be in an
 336  * interrupt or a critical region, and should only copy the information
 337  * from the master page table, nothing more.
 338  */
 339 static int do_ifault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 340 {
 341         unsigned int index;
 342         pgd_t *pgd, *pgd_k;
 343         pmd_t *pmd, *pmd_k;
 344
 345         if (addr < TASK_SIZE)
 346                 return do_pf(addr, fsr, regs);
 347
 348         if (user_mode(regs))
 349                 goto bad_area;
 350
 351         index = pgd_index(addr);
 352
 353         pgd = cpu_get_pgd() + index;
 354         pgd_k = init_mm.pgd + index;
 355
 356         if (pgd_none(*pgd_k))
 357                 goto bad_area;
 358
 359         pmd_k = pmd_offset((pud_t *) pgd_k, addr);
 360         pmd = pmd_offset((pud_t *) pgd, addr);
 361
 362         if (pmd_none(*pmd_k))
 363                 goto bad_area;
 364
 365         set_pmd(pmd, *pmd_k);
 366         flush_pmd_entry(pmd);
 367         return 0;
 368
 369 bad_area:
 370         do_bad_area(addr, fsr, regs);
 371         return 0;
 372 }
 373
 374 /*
 375  * This abort handler always returns "fault".
 376  */
 377 static int do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 378 {
 379         return 1;
 380 }
 381
 382 static int do_good(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 383 {
 384         unsigned int res1, res2;
 385
 386         printk("dabt exception but no error!\n");
 387
 388         __asm__ __volatile__(
 389                         "mff %0,f0\n"
 390                         "mff %1,f1\n"
 391                         : "=r"(res1), "=r"(res2)
 392                         :
 393                         : "memory");
 394
 395         printk(KERN_EMERG "r0 :%08x  r1 :%08x\n", res1, res2);
 396         panic("shut up\n");
 397         return 0;
 398 }
 399
 400 static struct fsr_info {
 401         int (*fn) (unsigned long addr, unsigned int fsr, struct pt_regs *regs);
 402         int sig;
 403         int code;
 404         const char *name;
 405 } fsr_info[] = {
 406         /*
 407          * The following are the standard Unicore-I and UniCore-II aborts.
 408          */
 409         { do_good,      SIGBUS,  0,             "no error"              },
 410         { do_bad,       SIGBUS,  BUS_ADRALN,    "alignment exception"   },
 411         { do_bad,       SIGBUS,  BUS_OBJERR,    "external exception"    },
 412         { do_bad,       SIGBUS,  0,             "burst operation"       },
 413         { do_bad,       SIGBUS,  0,             "unknown 00100"         },
 414         { do_ifault,    SIGSEGV, SEGV_MAPERR,   "2nd level pt non-exist"},
 415         { do_bad,       SIGBUS,  0,             "2nd lvl large pt non-exist" },
 416         { do_bad,       SIGBUS,  0,             "invalid pte"           },
 417         { do_pf,        SIGSEGV, SEGV_MAPERR,   "page miss"             },
 418         { do_bad,       SIGBUS,  0,             "middle page miss"      },
 419         { do_bad,       SIGBUS,  0,             "large page miss"       },
 420         { do_pf,        SIGSEGV, SEGV_MAPERR,   "super page (section) miss" },
 421         { do_bad,       SIGBUS,  0,             "unknown 01100"         },
 422         { do_bad,       SIGBUS,  0,             "unknown 01101"         },
 423         { do_bad,       SIGBUS,  0,             "unknown 01110"         },
 424         { do_bad,       SIGBUS,  0,             "unknown 01111"         },
 425         { do_bad,       SIGBUS,  0,             "addr: up 3G or IO"     },
 426         { do_pf,        SIGSEGV, SEGV_ACCERR,   "read unreadable addr"  },
 427         { do_pf,        SIGSEGV, SEGV_ACCERR,   "write unwriteable addr"},
 428         { do_pf,        SIGSEGV, SEGV_ACCERR,   "exec unexecutable addr"},
 429         { do_bad,       SIGBUS,  0,             "unknown 10100"         },
 430         { do_bad,       SIGBUS,  0,             "unknown 10101"         },
 431         { do_bad,       SIGBUS,  0,             "unknown 10110"         },
 432         { do_bad,       SIGBUS,  0,             "unknown 10111"         },
 433         { do_bad,       SIGBUS,  0,             "unknown 11000"         },
 434         { do_bad,       SIGBUS,  0,             "unknown 11001"         },
 435         { do_bad,       SIGBUS,  0,             "unknown 11010"         },
 436         { do_bad,       SIGBUS,  0,             "unknown 11011"         },
 437         { do_bad,       SIGBUS,  0,             "unknown 11100"         },
 438         { do_bad,       SIGBUS,  0,             "unknown 11101"         },
 439         { do_bad,       SIGBUS,  0,             "unknown 11110"         },
 440         { do_bad,       SIGBUS,  0,             "unknown 11111"         }
 441 };
 442
 443 void __init hook_fault_code(int nr,
 444                 int (*fn) (unsigned long, unsigned int, struct pt_regs *),
 445                 int sig, int code, const char *name)
 446 {
 447         if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
 448                 BUG();
 449
 450         fsr_info[nr].fn   = fn;
 451         fsr_info[nr].sig  = sig;
 452         fsr_info[nr].code = code;
 453         fsr_info[nr].name = name;
 454 }
 455
 456 /*
 457  * Dispatch a data abort to the relevant handler.
 458  */
 459 asmlinkage void do_DataAbort(unsigned long addr, unsigned int fsr,
 460                         struct pt_regs *regs)
 461 {
 462         const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
 463         struct siginfo info;
 464
 465         if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
 466                 return;
 467
 468         printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
 469                inf->name, fsr, addr);
 470
 471         info.si_signo = inf->sig;
 472         info.si_errno = 0;
 473         info.si_code = inf->code;
 474         info.si_addr = (void __user *)addr;
 475         uc32_notify_die("", regs, &info, fsr, 0);
 476 }
 477
 478 asmlinkage void do_PrefetchAbort(unsigned long addr,
 479                         unsigned int ifsr, struct pt_regs *regs)
 480 {
 481         const struct fsr_info *inf = fsr_info + fsr_fs(ifsr);
 482         struct siginfo info;
 483
 484         if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
 485                 return;
 486
 487         printk(KERN_ALERT "Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
 488                inf->name, ifsr, addr);
 489
 490         info.si_signo = inf->sig;
 491         info.si_errno = 0;
 492         info.si_code = inf->code;
 493         info.si_addr = (void __user *)addr;
 494         uc32_notify_die("", regs, &info, ifsr, 0);
 495 }