src/runtime/bsd-os.c

   1 /*
   2  * OS-dependent routines for BSD-ish systems
   3  *
   4  * This file (along with os.h) exports an OS-independent interface to
   5  * the operating system VM facilities. This interface looks a lot like
   6  * the Mach interface (but simpler in some places). For some operating
   7  * systems, a subset of these functions will have to be emulated.
   8  */
   9
  10 /*
  11  * This software is part of the SBCL system. See the README file for
  12  * more information.
  13  *
  14  * This software is derived from the CMU CL system, which was
  15  * written at Carnegie Mellon University and released into the
  16  * public domain. The software is in the public domain and is
  17  * provided with absolutely no warranty. See the COPYING and CREDITS
  18  * files for more information.
  19  */
  20
  21 #include <stdio.h>
  22 #include <sys/param.h>
  23 #include <sys/file.h>
  24 #include <unistd.h>
  25 #include <utime.h>
  26 #include <assert.h>
  27 #include <errno.h>
  28 #include "sbcl.h"
  29 #include "./signal.h"
  30 #include "os.h"
  31 #include "arch.h"
  32 #include "globals.h"
  33 #include "interrupt.h"
  34 #include "interr.h"
  35 #include "lispregs.h"
  36 #include "thread.h"
  37 #include "runtime.h"
  38 #include "genesis/static-symbols.h"
  39 #include "genesis/fdefn.h"
  40
  41 #include <sys/types.h>
  42 #include <signal.h>
  43 /* #include <sys/sysinfo.h> */
  44 #include "validate.h"
  45 #if defined LISP_FEATURE_GENCGC
  46 #include "gencgc-internal.h"
  47 #endif
  48
  49 #if defined(LISP_FEATURE_SB_WTIMER) && !defined(LISP_FEATURE_DARWIN)
  50 # include <sys/event.h>
  51 #endif
  52
  53 \f
  54 os_vm_size_t os_vm_page_size;
  55
  56 #ifdef __NetBSD__
  57 #include <sys/resource.h>
  58 #include <sys/sysctl.h>
  59 #include <string.h>
  60 #include <sys/stat.h> /* For the stat-family wrappers. */
  61 #include <dirent.h>   /* For the opendir()/readdir() wrappers */
  62 #include <sys/socket.h> /* For the socket() wrapper */
  63 static void netbsd_init();
  64 static os_vm_size_t max_allocation_size;
  65 #endif /* __NetBSD__ */
  66
  67 #if defined LISP_FEATURE_FREEBSD
  68 #include <sys/sysctl.h>
  69 #if defined(LISP_FEATURE_SB_THREAD) && !defined(LISP_FEATURE_SB_PTHREAD_FUTEX)
  70 #include <sys/umtx.h>
  71 #endif
  72
  73 static void freebsd_init();
  74 #endif /* __FreeBSD__ */
  75
  76 #ifdef __DragonFly__
  77 #include <sys/sysctl.h>
  78
  79 static void dragonfly_init();
  80 #endif /* __DragonFly__ */
  81
  82 #ifdef __OpenBSD__
  83 #include <sys/types.h>
  84 #include <sys/resource.h>
  85 #include <sys/stat.h>
  86 #include <sys/sysctl.h>
  87 #include <dlfcn.h>
  88 #ifdef LISP_FEATURE_X86
  89 #include <machine/cpu.h>
  90 #endif
  91
  92 static void openbsd_init();
  93 #endif
  94
  95 void
  96 os_init(char *argv[], char *envp[])
  97 {
  98     os_vm_page_size = BACKEND_PAGE_BYTES;
  99
 100 #ifdef __NetBSD__
 101     netbsd_init();
 102 #elif defined(LISP_FEATURE_FREEBSD)
 103     freebsd_init();
 104 #elif defined(__OpenBSD__)
 105     openbsd_init();
 106 #elif defined(LISP_FEATURE_DARWIN)
 107     darwin_init();
 108 #elif defined(__DragonFly__)
 109     dragonfly_init();
 110 #endif
 111 }
 112
 113 sigset_t *
 114 os_context_sigmask_addr(os_context_t *context)
 115 {
 116     /* (Unlike most of the other context fields that we access, the
 117      * signal mask field is a field of the basic, outermost context
 118      * struct itself both in FreeBSD 4.0 and in OpenBSD 2.6.) */
 119 #if defined(LISP_FEATURE_FREEBSD) || defined(__NetBSD__) || defined(LISP_FEATURE_DARWIN) \
 120     || defined(__DragonFly__)
 121     return &context->uc_sigmask;
 122 #elif defined (__OpenBSD__)
 123     return &context->sc_mask;
 124 #else
 125 #error unsupported BSD variant
 126 #endif
 127 }
 128
 129 os_vm_address_t
 130 os_validate(os_vm_address_t addr, os_vm_size_t len)
 131 {
 132     int flags = MAP_PRIVATE | MAP_ANON;
 133
 134     if (addr)
 135         flags |= MAP_FIXED;
 136
 137 #ifdef __NetBSD__
 138     if (addr) {
 139         os_vm_address_t curaddr = addr;
 140
 141         while (len > 0) {
 142             os_vm_address_t resaddr;
 143             os_vm_size_t curlen = MIN(max_allocation_size, len);
 144
 145             resaddr = mmap(curaddr, curlen, OS_VM_PROT_ALL, flags, -1, 0);
 146
 147             if (resaddr == (os_vm_address_t) - 1) {
 148                 perror("mmap");
 149
 150                 while (curaddr > addr) {
 151                     curaddr -= max_allocation_size;
 152                     munmap(curaddr, max_allocation_size);
 153                 }
 154
 155                 return NULL;
 156             }
 157
 158             curaddr += curlen;
 159             len -= curlen;
 160         }
 161     } else
 162 #endif
 163     {
 164         addr = mmap(addr, len, OS_VM_PROT_ALL, flags, -1, 0);
 165     }
 166
 167     if (addr == MAP_FAILED) {
 168         perror("mmap");
 169         return NULL;
 170     }
 171
 172     return addr;
 173 }
 174
 175 void
 176 os_invalidate(os_vm_address_t addr, os_vm_size_t len)
 177 {
 178     if (munmap(addr, len) == -1)
 179         perror("munmap");
 180 }
 181
 182 void
 183 os_protect(os_vm_address_t address, os_vm_size_t length, os_vm_prot_t prot)
 184 {
 185     if (mprotect(address, length, prot) == -1) {
 186         perror("mprotect");
 187     }
 188 }
 189 \f
 190 static boolean
 191 in_range_p(os_vm_address_t a, lispobj sbeg, size_t slen)
 192 {
 193     char* beg = (char*) sbeg;
 194     char* end = (char*) sbeg + slen;
 195     char* adr = (char*) a;
 196     return (adr >= beg && adr < end);
 197 }
 198
 199 boolean
 200 is_valid_lisp_addr(os_vm_address_t addr)
 201 {
 202     struct thread *th;
 203
 204     if (in_range_p(addr, READ_ONLY_SPACE_START, READ_ONLY_SPACE_SIZE) ||
 205         in_range_p(addr, STATIC_SPACE_START, STATIC_SPACE_SIZE) ||
 206 #ifdef LISP_FEATURE_IMMOBILE_SPACE
 207         in_range_p(addr, IMMOBILE_SPACE_START, IMMOBILE_SPACE_SIZE) ||
 208 #endif
 209         in_range_p(addr, DYNAMIC_SPACE_START, dynamic_space_size))
 210         return 1;
 211     for_each_thread(th) {
 212         if (((os_vm_address_t)th->control_stack_start <= addr) &&
 213             (addr < (os_vm_address_t)th->control_stack_end))
 214             return 1;
 215         if (in_range_p(addr, (lispobj) th->binding_stack_start,
 216                        BINDING_STACK_SIZE))
 217             return 1;
 218     }
 219     return 0;
 220 }
 221 \f
 222 /*
 223  * any OS-dependent special low-level handling for signals
 224  */
 225
 226 #if defined LISP_FEATURE_GENCGC
 227
 228 /*
 229  * The GENCGC needs to be hooked into whatever signal is raised for
 230  * page fault on this OS.
 231  */
 232
 233 void
 234 memory_fault_handler(int signal, siginfo_t *siginfo, os_context_t *context)
 235 {
 236     void *fault_addr = arch_get_bad_addr(signal, siginfo, context);
 237
 238 #if defined(LISP_FEATURE_RESTORE_TLS_SEGMENT_REGISTER_FROM_CONTEXT)
 239     FSHOW_SIGNAL((stderr, "/ TLS: restoring fs: %p in memory_fault_handler\n",
 240                   *CONTEXT_ADDR_FROM_STEM(fs)));
 241     os_restore_tls_segment_register(context);
 242 #endif
 243
 244     FSHOW((stderr, "Memory fault at: %p, PC: %p\n", fault_addr, *os_context_pc_addr(context)));
 245
 246 #ifdef LISP_FEATURE_SB_SAFEPOINT
 247     if (!handle_safepoint_violation(context, fault_addr))
 248 #endif
 249
 250     if (!gencgc_handle_wp_violation(fault_addr))
 251         if(!handle_guard_page_triggered(context,fault_addr))
 252             lisp_memory_fault_error(context, fault_addr);
 253 }
 254
 255 #if defined(LISP_FEATURE_MACH_EXCEPTION_HANDLER)
 256 void
 257 mach_error_memory_fault_handler(int signal, siginfo_t *siginfo,
 258                                 os_context_t *context) {
 259     lose("Unhandled memory fault. Exiting.");
 260 }
 261 #endif
 262
 263 void
 264 os_install_interrupt_handlers(void)
 265 {
 266     SHOW("os_install_interrupt_handlers()/bsd-os/defined(GENCGC)");
 267 #if defined(LISP_FEATURE_MACH_EXCEPTION_HANDLER)
 268     undoably_install_low_level_interrupt_handler(SIG_MEMORY_FAULT,
 269                                                  mach_error_memory_fault_handler);
 270 #else
 271     undoably_install_low_level_interrupt_handler(SIG_MEMORY_FAULT,
 272 #if defined(LISP_FEATURE_FREEBSD) && !defined(__GLIBC__)
 273                                                  (__siginfohandler_t *)
 274 #endif
 275                                                  memory_fault_handler);
 276 #endif
 277
 278 #ifdef LISP_FEATURE_SB_THREAD
 279 # ifdef LISP_FEATURE_SB_SAFEPOINT
 280 #  ifdef LISP_FEATURE_SB_THRUPTION
 281     undoably_install_low_level_interrupt_handler(SIGPIPE, thruption_handler);
 282 #  endif
 283 # else
 284     undoably_install_low_level_interrupt_handler(SIG_STOP_FOR_GC,
 285                                                  sig_stop_for_gc_handler);
 286 # endif
 287 #endif
 288     SHOW("leaving os_install_interrupt_handlers()");
 289 }
 290
 291 #else /* Currently PPC/Darwin/Cheney only */
 292
 293 static void
 294 sigsegv_handler(int signal, siginfo_t *info, os_context_t *context)
 295 {
 296     os_vm_address_t addr;
 297
 298     addr = arch_get_bad_addr(signal, info, context);
 299     if (!cheneygc_handle_wp_violation(context, addr))
 300         if (!handle_guard_page_triggered(context, addr))
 301             interrupt_handle_now(signal, info, context);
 302 }
 303
 304 void
 305 os_install_interrupt_handlers(void)
 306 {
 307     SHOW("os_install_interrupt_handlers()/bsd-os/!defined(GENCGC)");
 308     undoably_install_low_level_interrupt_handler(SIG_MEMORY_FAULT,
 309                                                  sigsegv_handler);
 310 }
 311
 312 #endif /* defined GENCGC */
 313
 314 #ifdef __NetBSD__
 315 static void netbsd_init()
 316 {
 317     struct rlimit rl;
 318     int mib[2], osrev;
 319     size_t len;
 320
 321     /* Are we running on a sufficiently functional kernel? */
 322     mib[0] = CTL_KERN;
 323     mib[1] = KERN_OSREV;
 324
 325     len = sizeof(osrev);
 326     sysctl(mib, 2, &osrev, &len, NULL, 0);
 327
 328     /* If we're older than 2.0... */
 329     if (osrev < 200000000) {
 330         fprintf(stderr, "osrev = %d (needed at least 200000000).\n", osrev);
 331         lose("NetBSD kernel too old to run sbcl.\n");
 332     }
 333
 334     /* NetBSD counts mmap()ed space against the process's data size limit,
 335      * so yank it up. This might be a nasty thing to do? */
 336     getrlimit (RLIMIT_DATA, &rl);
 337     if (rl.rlim_cur < rl.rlim_max) {
 338         rl.rlim_cur = rl.rlim_max;
 339         if (setrlimit (RLIMIT_DATA, &rl) < 0) {
 340             fprintf (stderr,
 341                      "RUNTIME WARNING: unable to raise process data size limit:\n\
 342 %s.\n\
 343 The system may fail to start.\n",
 344                      strerror(errno));
 345         }
 346     }
 347     max_allocation_size = (os_vm_size_t)((rl.rlim_cur / 2) &
 348       ~(32 * 1024 * 1024));
 349
 350 #ifdef LISP_FEATURE_X86
 351     {
 352         size_t len;
 353         int sse;
 354
 355         len = sizeof(sse);
 356         if (sysctlbyname("machdep.sse", &sse, &len,
 357                          NULL, 0) == 0 && sse != 0) {
 358             /* Use the SSE detector */
 359             fast_bzero_pointer = fast_bzero_detect;
 360         }
 361      }
 362 #endif /* LISP_FEATURE_X86 */
 363 }
 364
 365 /* Various routines in NetBSD's C library are compatibility wrappers
 366    for old versions. Programs must be processed by the C toolchain in
 367    order to get up-to-date definitions of such routines. */
 368 /* The stat-family, opendir, and readdir are used only in sb-posix, as
 369    of 2007-01-16. -- RMK */
 370 int
 371 _stat(const char *path, struct stat *sb)
 372 {
 373     return stat(path, sb);
 374 }
 375 int
 376 _lstat(const char *path, struct stat *sb)
 377 {
 378     return lstat(path, sb);
 379 }
 380 int
 381 _fstat(int fd, struct stat *sb)
 382 {
 383     return fstat(fd, sb);
 384 }
 385
 386 DIR *
 387 _opendir(const char *filename)
 388 {
 389     return opendir(filename);
 390 }
 391 struct dirent *
 392 _readdir(DIR *dirp)
 393 {
 394     return readdir(dirp);
 395 }
 396
 397 int
 398 _utime(const char *file, const struct utimbuf *timep)
 399 {
 400     return utime(file, timep);
 401 }
 402
 403 /* Used in sb-bsd-sockets. */
 404 int
 405 _socket(int domain, int type, int protocol)
 406 {
 407     return socket(domain, type, protocol);
 408 }
 409 #endif /* __NetBSD__ */
 410
 411 #if defined(LISP_FEATURE_FREEBSD)
 412 #ifndef __GLIBC__
 413 extern int getosreldate(void);
 414 #endif
 415
 416 int sig_memory_fault;
 417
 418 static void freebsd_init()
 419 {
 420     /* Memory fault signal on FreeBSD was changed from SIGBUS to
 421      * SIGSEGV. */
 422 #ifdef __GLIBC__
 423         sig_memory_fault = SIGSEGV;
 424 #else
 425     if (getosreldate() < 700004)
 426         sig_memory_fault = SIGBUS;
 427     else
 428         sig_memory_fault = SIGSEGV;
 429 #endif
 430
 431     /* Quote from sbcl-devel (NIIMI Satoshi): "Some OSes, like FreeBSD
 432      * 4.x with GENERIC kernel, does not enable SSE support even on
 433      * SSE capable CPUs". Detect this situation and skip the
 434      * fast_bzero sse/base selection logic that's normally done in
 435      * x86-assem.S.
 436      */
 437 #ifdef LISP_FEATURE_X86
 438     {
 439         size_t len;
 440         int instruction_sse;
 441
 442         len = sizeof(instruction_sse);
 443         if (sysctlbyname("hw.instruction_sse", &instruction_sse, &len,
 444                          NULL, 0) == 0 && instruction_sse != 0) {
 445             /* Use the SSE detector */
 446             fast_bzero_pointer = fast_bzero_detect;
 447         }
 448     }
 449 #endif /* LISP_FEATURE_X86 */
 450 }
 451
 452 #if defined(LISP_FEATURE_SB_THREAD) && defined(LISP_FEATURE_SB_FUTEX) \
 453     && !defined(LISP_FEATURE_SB_PTHREAD_FUTEX)
 454 int
 455 futex_wait(int *lock_word, long oldval, long sec, unsigned long usec)
 456 {
 457     struct timespec timeout;
 458     int ret;
 459
 460     if (sec < 0)
 461         ret = umtx_wait((void *)lock_word, oldval, NULL);
 462     else {
 463         timeout.tv_sec = sec;
 464         timeout.tv_nsec = usec * 1000;
 465         ret = umtx_wait((void *)lock_word, oldval, &timeout);
 466     }
 467
 468     switch (ret) {
 469     case 0:
 470         return 0;
 471     case ETIMEDOUT:
 472         return 1;
 473     case EINTR:
 474         return 2;
 475     default:
 476         /* EWOULDBLOCK and others, need to check the lock */
 477         return -1;
 478     }
 479 }
 480
 481 int
 482 futex_wake(int *lock_word, int n)
 483 {
 484     return umtx_wake((void *)lock_word, n);
 485 }
 486 #endif
 487 #endif /* __FreeBSD__ */
 488
 489 #ifdef __DragonFly__
 490 static void dragonfly_init()
 491 {
 492 #ifdef LISP_FEATURE_X86
 493     size_t len;
 494     int instruction_sse;
 495
 496     len = sizeof(instruction_sse);
 497     if (sysctlbyname("hw.instruction_sse", &instruction_sse, &len,
 498                      NULL, 0) == 0 && instruction_sse != 0) {
 499         /* Use the SSE detector */
 500         fast_bzero_pointer = fast_bzero_detect;
 501     }
 502 #endif /* LISP_FEATURE_X86 */
 503 }
 504
 505
 506 #if defined(LISP_FEATURE_SB_THREAD) && defined(LISP_FEATURE_SB_FUTEX) \
 507     && !defined(LISP_FEATURE_SB_PTHREAD_FUTEX)
 508 int
 509 futex_wait(int *lock_word, long oldval, long sec, unsigned long usec)
 510 {
 511     int ret;
 512
 513     if (sec < 0)
 514         ret = umtx_sleep(lock_word, oldval, 0);
 515     else {
 516         int count = usec + 1000000 * sec;
 517         ret = umtx_sleep(lock_word, oldval, count);
 518     }
 519
 520     if (ret == 0) return 0;
 521     else {
 522         switch (errno) {
 523         case EWOULDBLOCK: // Operation timed out
 524             return 1;
 525         case EINTR:
 526             return 2;
 527         default: // Such as EINVAL or EBUSY
 528             return -1;
 529         }
 530     }
 531 }
 532
 533 int
 534 futex_wake(int *lock_word, int n)
 535 {
 536     return umtx_wakeup(lock_word, n);
 537 }
 538 #endif
 539 #endif /* __DragonFly__ */
 540
 541 #ifdef LISP_FEATURE_DARWIN
 542 /* defined in ppc-darwin-os.c instead */
 543 #elif defined(LISP_FEATURE_FREEBSD)
 544 #ifndef KERN_PROC_PATHNAME
 545 #define KERN_PROC_PATHNAME 12
 546 #endif
 547
 548 char *
 549 os_get_runtime_executable_path(int external)
 550 {
 551     char path[PATH_MAX + 1];
 552
 553 #ifndef __GLIBC__
 554     if (getosreldate() >= 600024) {
 555 #endif
 556         /* KERN_PROC_PATHNAME is available */
 557         size_t len = PATH_MAX + 1;
 558         int mib[4];
 559
 560         mib[0] = CTL_KERN;
 561         mib[1] = KERN_PROC;
 562         mib[2] = KERN_PROC_PATHNAME;
 563         mib[3] = -1;
 564         if (sysctl(mib, 4, &path, &len, NULL, 0) != 0)
 565             return NULL;
 566 #ifndef __GLIBC__
 567     } else {
 568         int size;
 569         size = readlink("/proc/curproc/file", path, sizeof(path) - 1);
 570         if (size < 0)
 571             return NULL;
 572         path[size] = '\0';
 573     }
 574 #endif
 575     if (strcmp(path, "unknown") == 0)
 576         return NULL;
 577     return copied_string(path);
 578 }
 579 #elif defined(LISP_FEATURE_DRAGONFLY)
 580 char *
 581 os_get_runtime_executable_path(int external)
 582 {
 583     char path[PATH_MAX + 1];
 584     int size = readlink("/proc/curproc/file", path, sizeof(path) - 1);
 585     if (size < 0)
 586         return NULL;
 587     path[size] = '\0';
 588
 589     if (strcmp(path, "unknown") == 0)
 590         return NULL;
 591     return copied_string(path);
 592 }
 593 #elif defined(LISP_FEATURE_NETBSD) || defined(LISP_FEATURE_OPENBSD)
 594 char *
 595 os_get_runtime_executable_path(int external)
 596 {
 597     struct stat sb;
 598     if (!external && stat("/proc/curproc/file", &sb) == 0)
 599         return copied_string("/proc/curproc/file");
 600     return NULL;
 601 }
 602 #else /* Not DARWIN or FREEBSD or NETBSD or OPENBSD or DragonFly */
 603 char *
 604 os_get_runtime_executable_path(int external)
 605 {
 606     return NULL;
 607 }
 608 #endif
 609
 610 #ifdef __OpenBSD__
 611
 612 int openbsd_use_fxsave = 0;
 613
 614 void
 615 openbsd_init()
 616 {
 617 #ifdef LISP_FEATURE_X86
 618     int mib[2];
 619     size_t size;
 620 #endif
 621     /*
 622      * Show a warning if it looks like the memory available after
 623      * allocating the spaces won't be at least this much.
 624      */
 625 #ifdef LISP_FEATURE_64_BIT
 626     const int wantfree = 64 * 1024 * 1024;
 627 #else
 628     const int wantfree = 32 * 1024 * 1024;
 629 #endif
 630     struct rlimit rl;
 631
 632 #ifdef LISP_FEATURE_X86
 633     /* Save the machdep.osfxsr sysctl for use by os_restore_fp_control() */
 634     mib[0] = CTL_MACHDEP;
 635     mib[1] = CPU_OSFXSR;
 636     size = sizeof (openbsd_use_fxsave);
 637     sysctl(mib, 2, &openbsd_use_fxsave, &size, NULL, 0);
 638 #endif
 639
 640     /* OpenBSD, like NetBSD, counts mmap()ed space against the
 641      * process's data size limit. If the soft limit is lower than the
 642      * hard limit then try to yank it up, this lets users in the
 643      * "staff" or "daemon" login classes run sbcl with larger dynamic
 644      * space sizes.
 645      */
 646     getrlimit (RLIMIT_DATA, &rl);
 647     if (rl.rlim_cur < rl.rlim_max) {
 648         rl.rlim_cur = rl.rlim_max;
 649         if (setrlimit (RLIMIT_DATA, &rl) < 0) {
 650             fprintf (stderr,
 651                      "RUNTIME WARNING: unable to raise process data size limit:\n\
 652   %s.\n\
 653 The system may fail to start.\n",
 654                      strerror(errno));
 655         }
 656     }
 657
 658     /*
 659      * Display a (hopefully) helpful warning if it looks like we won't
 660      * be able to allocate enough memory.
 661      */
 662     getrlimit (RLIMIT_DATA, &rl);
 663     if (dynamic_space_size + READ_ONLY_SPACE_SIZE + STATIC_SPACE_SIZE +
 664         LINKAGE_TABLE_SPACE_SIZE + wantfree > rl.rlim_cur)
 665         fprintf (stderr,
 666                  "RUNTIME WARNING: data size resource limit may be too low,\n"
 667                  "  try decreasing the dynamic space size with --dynamic-space-size\n"
 668                  "  or raising the datasize or datasize-max limits in /etc/login.conf\n");
 669 }
 670
 671 /* OpenBSD's dlsym() relies on the gcc bulitin
 672  * __builtin_return_address(0) returning an address in the
 673  * executable's text segment, but when called from lisp it will return
 674  * an address in the dynamic space.  Work around this by calling this
 675  * wrapper function instead. Note that tail-call optimization will
 676  * defeat this, disable it by saving the dlsym() return value in a
 677  * volatile variable.
 678 */
 679 void *
 680 os_dlsym(void *handle, const char *symbol)
 681 {
 682     void * volatile ret = dlsym(handle, symbol);
 683     return ret;
 684 }
 685
 686 #endif
 687
 688 #if defined(LISP_FEATURE_SB_WTIMER) && !defined(LISP_FEATURE_DARWIN)
 689 /*
 690  * Waitable timer implementation for the safepoint-based (SIGALRM-free)
 691  * timer facility using kqueue.
 692  */
 693 int
 694 os_create_wtimer()
 695 {
 696     int kq = kqueue();
 697     if (kq == -1)
 698         lose("os_create_wtimer: kqueue");
 699     return kq;
 700 }
 701
 702 int
 703 os_wait_for_wtimer(int kq)
 704 {
 705     struct kevent ev;
 706     int n;
 707     if ( (n = kevent(kq, 0, 0, &ev, 1, 0)) == -1) {
 708         if (errno != EINTR)
 709             lose("os_wtimer_listen failed");
 710         n = 0;
 711     }
 712     return n != 1;
 713 }
 714
 715 void
 716 os_close_wtimer(int kq)
 717 {
 718     if (close(kq) == -1)
 719         lose("os_close_wtimer failed");
 720 }
 721
 722 void
 723 os_set_wtimer(int kq, int sec, int nsec)
 724 {
 725     long long msec
 726         = ((long long) sec) * 1000 + (long long) (nsec+999999) / 1000000;
 727     if (msec > INT_MAX) msec = INT_MAX;
 728
 729     struct kevent ev;
 730     EV_SET(&ev, 1, EVFILT_TIMER, EV_ADD|EV_ENABLE|EV_ONESHOT, 0, (int)msec, 0);
 731     if (kevent(kq, &ev, 1, 0, 0, 0) == -1)
 732         perror("os_set_wtimer: kevent");
 733 }
 734
 735 void
 736 os_cancel_wtimer(int kq)
 737 {
 738     struct kevent ev;
 739     EV_SET(&ev, 1, EVFILT_TIMER, EV_DISABLE, 0, 0, 0);
 740     if (kevent(kq, &ev, 1, 0, 0, 0) == -1 && errno != ENOENT)
 741         perror("os_cancel_wtimer: kevent");
 742 }
 743 #endif