arch/sparc/kernel/sys_sparc_64.c

   1 /* linux/arch/sparc64/kernel/sys_sparc.c
   2  *
   3  * This file contains various random system calls that
   4  * have a non-standard calling sequence on the Linux/sparc
   5  * platform.
   6  */
   7
   8 #include <linux/errno.h>
   9 #include <linux/types.h>
  10 #include <linux/sched.h>
  11 #include <linux/fs.h>
  12 #include <linux/file.h>
  13 #include <linux/mm.h>
  14 #include <linux/sem.h>
  15 #include <linux/msg.h>
  16 #include <linux/shm.h>
  17 #include <linux/stat.h>
  18 #include <linux/mman.h>
  19 #include <linux/utsname.h>
  20 #include <linux/smp.h>
  21 #include <linux/slab.h>
  22 #include <linux/syscalls.h>
  23 #include <linux/ipc.h>
  24 #include <linux/personality.h>
  25 #include <linux/random.h>
  26 #include <linux/export.h>
  27 #include <linux/context_tracking.h>
  28
  29 #include <asm/uaccess.h>
  30 #include <asm/utrap.h>
  31 #include <asm/unistd.h>
  32
  33 #include "entry.h"
  34 #include "kernel.h"
  35 #include "systbls.h"
  36
  37 /* #define DEBUG_UNIMP_SYSCALL */
  38
  39 asmlinkage unsigned long sys_getpagesize(void)
  40 {
  41         return PAGE_SIZE;
  42 }
  43
  44 /* Does addr --> addr+len fall within 4GB of the VA-space hole or
  45  * overflow past the end of the 64-bit address space?
  46  */
  47 static inline int invalid_64bit_range(unsigned long addr, unsigned long len)
  48 {
  49         unsigned long va_exclude_start, va_exclude_end;
  50
  51         va_exclude_start = VA_EXCLUDE_START;
  52         va_exclude_end   = VA_EXCLUDE_END;
  53
  54         if (unlikely(len >= va_exclude_start))
  55                 return 1;
  56
  57         if (unlikely((addr + len) < addr))
  58                 return 1;
  59
  60         if (unlikely((addr >= va_exclude_start && addr < va_exclude_end) ||
  61                      ((addr + len) >= va_exclude_start &&
  62                       (addr + len) < va_exclude_end)))
  63                 return 1;
  64
  65         return 0;
  66 }
  67
  68 /* These functions differ from the default implementations in
  69  * mm/mmap.c in two ways:
  70  *
  71  * 1) For file backed MAP_SHARED mmap()'s we D-cache color align,
  72  *    for fixed such mappings we just validate what the user gave us.
  73  * 2) For 64-bit tasks we avoid mapping anything within 4GB of
  74  *    the spitfire/niagara VA-hole.
  75  */
  76
  77 static inline unsigned long COLOR_ALIGN(unsigned long addr,
  78                                          unsigned long pgoff)
  79 {
  80         unsigned long base = (addr+SHMLBA-1)&~(SHMLBA-1);
  81         unsigned long off = (pgoff<<PAGE_SHIFT) & (SHMLBA-1);
  82
  83         return base + off;
  84 }
  85
  86 unsigned long arch_get_unmapped_area(struct file *filp, unsigned long addr, unsigned long len, unsigned long pgoff, unsigned long flags)
  87 {
  88         struct mm_struct *mm = current->mm;
  89         struct vm_area_struct * vma;
  90         unsigned long task_size = TASK_SIZE;
  91         int do_color_align;
  92         struct vm_unmapped_area_info info;
  93
  94         if (flags & MAP_FIXED) {
  95                 /* We do not accept a shared mapping if it would violate
  96                  * cache aliasing constraints.
  97                  */
  98                 if ((flags & MAP_SHARED) &&
  99                     ((addr - (pgoff << PAGE_SHIFT)) & (SHMLBA - 1)))
 100                         return -EINVAL;
 101                 return addr;
 102         }
 103
 104         if (test_thread_flag(TIF_32BIT))
 105                 task_size = STACK_TOP32;
 106         if (unlikely(len > task_size || len >= VA_EXCLUDE_START))
 107                 return -ENOMEM;
 108
 109         do_color_align = 0;
 110         if (filp || (flags & MAP_SHARED))
 111                 do_color_align = 1;
 112
 113         if (addr) {
 114                 if (do_color_align)
 115                         addr = COLOR_ALIGN(addr, pgoff);
 116                 else
 117                         addr = PAGE_ALIGN(addr);
 118
 119                 vma = find_vma(mm, addr);
 120                 if (task_size - len >= addr &&
 121                     (!vma || addr + len <= vma->vm_start))
 122                         return addr;
 123         }
 124
 125         info.flags = 0;
 126         info.length = len;
 127         info.low_limit = TASK_UNMAPPED_BASE;
 128         info.high_limit = min(task_size, VA_EXCLUDE_START);
 129         info.align_mask = do_color_align ? (PAGE_MASK & (SHMLBA - 1)) : 0;
 130         info.align_offset = pgoff << PAGE_SHIFT;
 131         addr = vm_unmapped_area(&info);
 132
 133         if ((addr & ~PAGE_MASK) && task_size > VA_EXCLUDE_END) {
 134                 VM_BUG_ON(addr != -ENOMEM);
 135                 info.low_limit = VA_EXCLUDE_END;
 136                 info.high_limit = task_size;
 137                 addr = vm_unmapped_area(&info);
 138         }
 139
 140         return addr;
 141 }
 142
 143 unsigned long
 144 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
 145                           const unsigned long len, const unsigned long pgoff,
 146                           const unsigned long flags)
 147 {
 148         struct vm_area_struct *vma;
 149         struct mm_struct *mm = current->mm;
 150         unsigned long task_size = STACK_TOP32;
 151         unsigned long addr = addr0;
 152         int do_color_align;
 153         struct vm_unmapped_area_info info;
 154
 155         /* This should only ever run for 32-bit processes.  */
 156         BUG_ON(!test_thread_flag(TIF_32BIT));
 157
 158         if (flags & MAP_FIXED) {
 159                 /* We do not accept a shared mapping if it would violate
 160                  * cache aliasing constraints.
 161                  */
 162                 if ((flags & MAP_SHARED) &&
 163                     ((addr - (pgoff << PAGE_SHIFT)) & (SHMLBA - 1)))
 164                         return -EINVAL;
 165                 return addr;
 166         }
 167
 168         if (unlikely(len > task_size))
 169                 return -ENOMEM;
 170
 171         do_color_align = 0;
 172         if (filp || (flags & MAP_SHARED))
 173                 do_color_align = 1;
 174
 175         /* requesting a specific address */
 176         if (addr) {
 177                 if (do_color_align)
 178                         addr = COLOR_ALIGN(addr, pgoff);
 179                 else
 180                         addr = PAGE_ALIGN(addr);
 181
 182                 vma = find_vma(mm, addr);
 183                 if (task_size - len >= addr &&
 184                     (!vma || addr + len <= vma->vm_start))
 185                         return addr;
 186         }
 187
 188         info.flags = VM_UNMAPPED_AREA_TOPDOWN;
 189         info.length = len;
 190         info.low_limit = PAGE_SIZE;
 191         info.high_limit = mm->mmap_base;
 192         info.align_mask = do_color_align ? (PAGE_MASK & (SHMLBA - 1)) : 0;
 193         info.align_offset = pgoff << PAGE_SHIFT;
 194         addr = vm_unmapped_area(&info);
 195
 196         /*
 197          * A failed mmap() very likely causes application failure,
 198          * so fall back to the bottom-up function here. This scenario
 199          * can happen with large stack limits and large mmap()
 200          * allocations.
 201          */
 202         if (addr & ~PAGE_MASK) {
 203                 VM_BUG_ON(addr != -ENOMEM);
 204                 info.flags = 0;
 205                 info.low_limit = TASK_UNMAPPED_BASE;
 206                 info.high_limit = STACK_TOP32;
 207                 addr = vm_unmapped_area(&info);
 208         }
 209
 210         return addr;
 211 }
 212
 213 /* Try to align mapping such that we align it as much as possible. */
 214 unsigned long get_fb_unmapped_area(struct file *filp, unsigned long orig_addr, unsigned long len, unsigned long pgoff, unsigned long flags)
 215 {
 216         unsigned long align_goal, addr = -ENOMEM;
 217         unsigned long (*get_area)(struct file *, unsigned long,
 218                                   unsigned long, unsigned long, unsigned long);
 219
 220         get_area = current->mm->get_unmapped_area;
 221
 222         if (flags & MAP_FIXED) {
 223                 /* Ok, don't mess with it. */
 224                 return get_area(NULL, orig_addr, len, pgoff, flags);
 225         }
 226         flags &= ~MAP_SHARED;
 227
 228         align_goal = PAGE_SIZE;
 229         if (len >= (4UL * 1024 * 1024))
 230                 align_goal = (4UL * 1024 * 1024);
 231         else if (len >= (512UL * 1024))
 232                 align_goal = (512UL * 1024);
 233         else if (len >= (64UL * 1024))
 234                 align_goal = (64UL * 1024);
 235
 236         do {
 237                 addr = get_area(NULL, orig_addr, len + (align_goal - PAGE_SIZE), pgoff, flags);
 238                 if (!(addr & ~PAGE_MASK)) {
 239                         addr = (addr + (align_goal - 1UL)) & ~(align_goal - 1UL);
 240                         break;
 241                 }
 242
 243                 if (align_goal == (4UL * 1024 * 1024))
 244                         align_goal = (512UL * 1024);
 245                 else if (align_goal == (512UL * 1024))
 246                         align_goal = (64UL * 1024);
 247                 else
 248                         align_goal = PAGE_SIZE;
 249         } while ((addr & ~PAGE_MASK) && align_goal > PAGE_SIZE);
 250
 251         /* Mapping is smaller than 64K or larger areas could not
 252          * be obtained.
 253          */
 254         if (addr & ~PAGE_MASK)
 255                 addr = get_area(NULL, orig_addr, len, pgoff, flags);
 256
 257         return addr;
 258 }
 259 EXPORT_SYMBOL(get_fb_unmapped_area);
 260
 261 /* Essentially the same as PowerPC.  */
 262 static unsigned long mmap_rnd(void)
 263 {
 264         unsigned long rnd = 0UL;
 265
 266         if (current->flags & PF_RANDOMIZE) {
 267                 unsigned long val = get_random_long();
 268                 if (test_thread_flag(TIF_32BIT))
 269                         rnd = (val % (1UL << (23UL-PAGE_SHIFT)));
 270                 else
 271                         rnd = (val % (1UL << (30UL-PAGE_SHIFT)));
 272         }
 273         return rnd << PAGE_SHIFT;
 274 }
 275
 276 void arch_pick_mmap_layout(struct mm_struct *mm)
 277 {
 278         unsigned long random_factor = mmap_rnd();
 279         unsigned long gap;
 280
 281         /*
 282          * Fall back to the standard layout if the personality
 283          * bit is set, or if the expected stack growth is unlimited:
 284          */
 285         gap = rlimit(RLIMIT_STACK);
 286         if (!test_thread_flag(TIF_32BIT) ||
 287             (current->personality & ADDR_COMPAT_LAYOUT) ||
 288             gap == RLIM_INFINITY ||
 289             sysctl_legacy_va_layout) {
 290                 mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
 291                 mm->get_unmapped_area = arch_get_unmapped_area;
 292         } else {
 293                 /* We know it's 32-bit */
 294                 unsigned long task_size = STACK_TOP32;
 295
 296                 if (gap < 128 * 1024 * 1024)
 297                         gap = 128 * 1024 * 1024;
 298                 if (gap > (task_size / 6 * 5))
 299                         gap = (task_size / 6 * 5);
 300
 301                 mm->mmap_base = PAGE_ALIGN(task_size - gap - random_factor);
 302                 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
 303         }
 304 }
 305
 306 /*
 307  * sys_pipe() is the normal C calling standard for creating
 308  * a pipe. It's not the way unix traditionally does this, though.
 309  */
 310 SYSCALL_DEFINE1(sparc_pipe_real, struct pt_regs *, regs)
 311 {
 312         int fd[2];
 313         int error;
 314
 315         error = do_pipe_flags(fd, 0);
 316         if (error)
 317                 goto out;
 318         regs->u_regs[UREG_I1] = fd[1];
 319         error = fd[0];
 320 out:
 321         return error;
 322 }
 323
 324 /*
 325  * sys_ipc() is the de-multiplexer for the SysV IPC calls..
 326  *
 327  * This is really horribly ugly.
 328  */
 329
 330 SYSCALL_DEFINE6(sparc_ipc, unsigned int, call, int, first, unsigned long, second,
 331                 unsigned long, third, void __user *, ptr, long, fifth)
 332 {
 333         long err;
 334
 335         /* No need for backward compatibility. We can start fresh... */
 336         if (call <= SEMTIMEDOP) {
 337                 switch (call) {
 338                 case SEMOP:
 339                         err = sys_semtimedop(first, ptr,
 340                                              (unsigned int)second, NULL);
 341                         goto out;
 342                 case SEMTIMEDOP:
 343                         err = sys_semtimedop(first, ptr, (unsigned int)second,
 344                                 (const struct timespec __user *)
 345                                              (unsigned long) fifth);
 346                         goto out;
 347                 case SEMGET:
 348                         err = sys_semget(first, (int)second, (int)third);
 349                         goto out;
 350                 case SEMCTL: {
 351                         err = sys_semctl(first, second,
 352                                          (int)third | IPC_64,
 353                                          (unsigned long) ptr);
 354                         goto out;
 355                 }
 356                 default:
 357                         err = -ENOSYS;
 358                         goto out;
 359                 }
 360         }
 361         if (call <= MSGCTL) {
 362                 switch (call) {
 363                 case MSGSND:
 364                         err = sys_msgsnd(first, ptr, (size_t)second,
 365                                          (int)third);
 366                         goto out;
 367                 case MSGRCV:
 368                         err = sys_msgrcv(first, ptr, (size_t)second, fifth,
 369                                          (int)third);
 370                         goto out;
 371                 case MSGGET:
 372                         err = sys_msgget((key_t)first, (int)second);
 373                         goto out;
 374                 case MSGCTL:
 375                         err = sys_msgctl(first, (int)second | IPC_64, ptr);
 376                         goto out;
 377                 default:
 378                         err = -ENOSYS;
 379                         goto out;
 380                 }
 381         }
 382         if (call <= SHMCTL) {
 383                 switch (call) {
 384                 case SHMAT: {
 385                         ulong raddr;
 386                         err = do_shmat(first, ptr, (int)second, &raddr, SHMLBA);
 387                         if (!err) {
 388                                 if (put_user(raddr,
 389                                              (ulong __user *) third))
 390                                         err = -EFAULT;
 391                         }
 392                         goto out;
 393                 }
 394                 case SHMDT:
 395                         err = sys_shmdt(ptr);
 396                         goto out;
 397                 case SHMGET:
 398                         err = sys_shmget(first, (size_t)second, (int)third);
 399                         goto out;
 400                 case SHMCTL:
 401                         err = sys_shmctl(first, (int)second | IPC_64, ptr);
 402                         goto out;
 403                 default:
 404                         err = -ENOSYS;
 405                         goto out;
 406                 }
 407         } else {
 408                 err = -ENOSYS;
 409         }
 410 out:
 411         return err;
 412 }
 413
 414 SYSCALL_DEFINE1(sparc64_personality, unsigned long, personality)
 415 {
 416         long ret;
 417
 418         if (personality(current->personality) == PER_LINUX32 &&
 419             personality(personality) == PER_LINUX)
 420                 personality |= PER_LINUX32;
 421         ret = sys_personality(personality);
 422         if (personality(ret) == PER_LINUX32)
 423                 ret &= ~PER_LINUX32;
 424
 425         return ret;
 426 }
 427
 428 int sparc_mmap_check(unsigned long addr, unsigned long len)
 429 {
 430         if (test_thread_flag(TIF_32BIT)) {
 431                 if (len >= STACK_TOP32)
 432                         return -EINVAL;
 433
 434                 if (addr > STACK_TOP32 - len)
 435                         return -EINVAL;
 436         } else {
 437                 if (len >= VA_EXCLUDE_START)
 438                         return -EINVAL;
 439
 440                 if (invalid_64bit_range(addr, len))
 441                         return -EINVAL;
 442         }
 443
 444         return 0;
 445 }
 446
 447 /* Linux version of mmap */
 448 SYSCALL_DEFINE6(mmap, unsigned long, addr, unsigned long, len,
 449                 unsigned long, prot, unsigned long, flags, unsigned long, fd,
 450                 unsigned long, off)
 451 {
 452         unsigned long retval = -EINVAL;
 453
 454         if ((off + PAGE_ALIGN(len)) < off)
 455                 goto out;
 456         if (off & ~PAGE_MASK)
 457                 goto out;
 458         retval = sys_mmap_pgoff(addr, len, prot, flags, fd, off >> PAGE_SHIFT);
 459 out:
 460         return retval;
 461 }
 462
 463 SYSCALL_DEFINE2(64_munmap, unsigned long, addr, size_t, len)
 464 {
 465         if (invalid_64bit_range(addr, len))
 466                 return -EINVAL;
 467
 468         return vm_munmap(addr, len);
 469 }
 470
 471 SYSCALL_DEFINE5(64_mremap, unsigned long, addr, unsigned long, old_len,
 472                 unsigned long, new_len, unsigned long, flags,
 473                 unsigned long, new_addr)
 474 {
 475         if (test_thread_flag(TIF_32BIT))
 476                 return -EINVAL;
 477         return sys_mremap(addr, old_len, new_len, flags, new_addr);
 478 }
 479
 480 /* we come to here via sys_nis_syscall so it can setup the regs argument */
 481 asmlinkage unsigned long c_sys_nis_syscall(struct pt_regs *regs)
 482 {
 483         static int count;
 484
 485         /* Don't make the system unusable, if someone goes stuck */
 486         if (count++ > 5)
 487                 return -ENOSYS;
 488
 489         printk ("Unimplemented SPARC system call %ld\n",regs->u_regs[1]);
 490 #ifdef DEBUG_UNIMP_SYSCALL
 491         show_regs (regs);
 492 #endif
 493
 494         return -ENOSYS;
 495 }
 496
 497 /* #define DEBUG_SPARC_BREAKPOINT */
 498
 499 asmlinkage void sparc_breakpoint(struct pt_regs *regs)
 500 {
 501         enum ctx_state prev_state = exception_enter();
 502         siginfo_t info;
 503
 504         if (test_thread_flag(TIF_32BIT)) {
 505                 regs->tpc &= 0xffffffff;
 506                 regs->tnpc &= 0xffffffff;
 507         }
 508 #ifdef DEBUG_SPARC_BREAKPOINT
 509         printk ("TRAP: Entering kernel PC=%lx, nPC=%lx\n", regs->tpc, regs->tnpc);
 510 #endif
 511         info.si_signo = SIGTRAP;
 512         info.si_errno = 0;
 513         info.si_code = TRAP_BRKPT;
 514         info.si_addr = (void __user *)regs->tpc;
 515         info.si_trapno = 0;
 516         force_sig_info(SIGTRAP, &info, current);
 517 #ifdef DEBUG_SPARC_BREAKPOINT
 518         printk ("TRAP: Returning to space: PC=%lx nPC=%lx\n", regs->tpc, regs->tnpc);
 519 #endif
 520         exception_exit(prev_state);
 521 }
 522
 523 extern void check_pending(int signum);
 524
 525 SYSCALL_DEFINE2(getdomainname, char __user *, name, int, len)
 526 {
 527         int nlen, err;
 528
 529         if (len < 0)
 530                 return -EINVAL;
 531
 532         down_read(&uts_sem);
 533
 534         nlen = strlen(utsname()->domainname) + 1;
 535         err = -EINVAL;
 536         if (nlen > len)
 537                 goto out;
 538
 539         err = -EFAULT;
 540         if (!copy_to_user(name, utsname()->domainname, nlen))
 541                 err = 0;
 542
 543 out:
 544         up_read(&uts_sem);
 545         return err;
 546 }
 547
 548 SYSCALL_DEFINE5(utrap_install, utrap_entry_t, type,
 549                 utrap_handler_t, new_p, utrap_handler_t, new_d,
 550                 utrap_handler_t __user *, old_p,
 551                 utrap_handler_t __user *, old_d)
 552 {
 553         if (type < UT_INSTRUCTION_EXCEPTION || type > UT_TRAP_INSTRUCTION_31)
 554                 return -EINVAL;
 555         if (new_p == (utrap_handler_t)(long)UTH_NOCHANGE) {
 556                 if (old_p) {
 557                         if (!current_thread_info()->utraps) {
 558                                 if (put_user(NULL, old_p))
 559                                         return -EFAULT;
 560                         } else {
 561                                 if (put_user((utrap_handler_t)(current_thread_info()->utraps[type]), old_p))
 562                                         return -EFAULT;
 563                         }
 564                 }
 565                 if (old_d) {
 566                         if (put_user(NULL, old_d))
 567                                 return -EFAULT;
 568                 }
 569                 return 0;
 570         }
 571         if (!current_thread_info()->utraps) {
 572                 current_thread_info()->utraps =
 573                         kzalloc((UT_TRAP_INSTRUCTION_31+1)*sizeof(long), GFP_KERNEL);
 574                 if (!current_thread_info()->utraps)
 575                         return -ENOMEM;
 576                 current_thread_info()->utraps[0] = 1;
 577         } else {
 578                 if ((utrap_handler_t)current_thread_info()->utraps[type] != new_p &&
 579                     current_thread_info()->utraps[0] > 1) {
 580                         unsigned long *p = current_thread_info()->utraps;
 581
 582                         current_thread_info()->utraps =
 583                                 kmalloc((UT_TRAP_INSTRUCTION_31+1)*sizeof(long),
 584                                         GFP_KERNEL);
 585                         if (!current_thread_info()->utraps) {
 586                                 current_thread_info()->utraps = p;
 587                                 return -ENOMEM;
 588                         }
 589                         p[0]--;
 590                         current_thread_info()->utraps[0] = 1;
 591                         memcpy(current_thread_info()->utraps+1, p+1,
 592                                UT_TRAP_INSTRUCTION_31*sizeof(long));
 593                 }
 594         }
 595         if (old_p) {
 596                 if (put_user((utrap_handler_t)(current_thread_info()->utraps[type]), old_p))
 597                         return -EFAULT;
 598         }
 599         if (old_d) {
 600                 if (put_user(NULL, old_d))
 601                         return -EFAULT;
 602         }
 603         current_thread_info()->utraps[type] = (long)new_p;
 604
 605         return 0;
 606 }
 607
 608 asmlinkage long sparc_memory_ordering(unsigned long model,
 609                                       struct pt_regs *regs)
 610 {
 611         if (model >= 3)
 612                 return -EINVAL;
 613         regs->tstate = (regs->tstate & ~TSTATE_MM) | (model << 14);
 614         return 0;
 615 }
 616
 617 SYSCALL_DEFINE5(rt_sigaction, int, sig, const struct sigaction __user *, act,
 618                 struct sigaction __user *, oact, void __user *, restorer,
 619                 size_t, sigsetsize)
 620 {
 621         struct k_sigaction new_ka, old_ka;
 622         int ret;
 623
 624         /* XXX: Don't preclude handling different sized sigset_t's.  */
 625         if (sigsetsize != sizeof(sigset_t))
 626                 return -EINVAL;
 627
 628         if (act) {
 629                 new_ka.ka_restorer = restorer;
 630                 if (copy_from_user(&new_ka.sa, act, sizeof(*act)))
 631                         return -EFAULT;
 632         }
 633
 634         ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
 635
 636         if (!ret && oact) {
 637                 if (copy_to_user(oact, &old_ka.sa, sizeof(*oact)))
 638                         return -EFAULT;
 639         }
 640
 641         return ret;
 642 }
 643
 644 asmlinkage long sys_kern_features(void)
 645 {
 646         return KERN_FEATURE_MIXED_MODE_STACK;
 647 }