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 #endif /* __NetBSD__ */
  65
  66 #ifdef __FreeBSD__
  67 #include <sys/sysctl.h>
  68 #if defined(LISP_FEATURE_SB_THREAD) && !defined(LISP_FEATURE_SB_PTHREAD_FUTEX)
  69 #include <sys/umtx.h>
  70 #endif
  71
  72 static void freebsd_init();
  73 #endif /* __FreeBSD__ */
  74
  75 #ifdef __OpenBSD__
  76 #include <sys/types.h>
  77 #include <sys/resource.h>
  78 #include <sys/stat.h>
  79 #include <sys/sysctl.h>
  80 #include <dlfcn.h>
  81 #ifdef LISP_FEATURE_X86
  82 #include <machine/cpu.h>
  83 #endif
  84
  85 static void openbsd_init();
  86 #endif
  87
  88 void
  89 os_init(char *argv[], char *envp[])
  90 {
  91     os_vm_page_size = BACKEND_PAGE_BYTES;
  92
  93 #ifdef __NetBSD__
  94     netbsd_init();
  95 #elif defined(__FreeBSD__)
  96     freebsd_init();
  97 #elif defined(__OpenBSD__)
  98     openbsd_init();
  99 #elif defined(LISP_FEATURE_DARWIN)
 100     darwin_init();
 101 #endif
 102 }
 103
 104 sigset_t *
 105 os_context_sigmask_addr(os_context_t *context)
 106 {
 107     /* (Unlike most of the other context fields that we access, the
 108      * signal mask field is a field of the basic, outermost context
 109      * struct itself both in FreeBSD 4.0 and in OpenBSD 2.6.) */
 110 #if defined(__FreeBSD__)  || defined(__NetBSD__) || defined(LISP_FEATURE_DARWIN)
 111     return &context->uc_sigmask;
 112 #elif defined (__OpenBSD__)
 113     return &context->sc_mask;
 114 #else
 115 #error unsupported BSD variant
 116 #endif
 117 }
 118
 119 os_vm_address_t
 120 os_validate(os_vm_address_t addr, os_vm_size_t len)
 121 {
 122     int flags = MAP_PRIVATE | MAP_ANON;
 123
 124     if (addr)
 125         flags |= MAP_FIXED;
 126
 127     addr = mmap(addr, len, OS_VM_PROT_ALL, flags, -1, 0);
 128
 129     if (addr == MAP_FAILED) {
 130         perror("mmap");
 131         return NULL;
 132     }
 133
 134     return addr;
 135 }
 136
 137 void
 138 os_invalidate(os_vm_address_t addr, os_vm_size_t len)
 139 {
 140     if (munmap(addr, len) == -1)
 141         perror("munmap");
 142 }
 143
 144 os_vm_address_t
 145 os_map(int fd, int offset, os_vm_address_t addr, os_vm_size_t len)
 146 {
 147     addr = mmap(addr, len,
 148                 OS_VM_PROT_ALL,
 149                 MAP_PRIVATE | MAP_FILE | MAP_FIXED,
 150                 fd, (off_t) offset);
 151
 152     if (addr == MAP_FAILED) {
 153         perror("mmap");
 154         lose("unexpected mmap(..) failure\n");
 155     }
 156
 157     return addr;
 158 }
 159
 160 void
 161 os_protect(os_vm_address_t address, os_vm_size_t length, os_vm_prot_t prot)
 162 {
 163     if (mprotect(address, length, prot) == -1) {
 164         perror("mprotect");
 165     }
 166 }
 167 \f
 168 static boolean
 169 in_range_p(os_vm_address_t a, lispobj sbeg, size_t slen)
 170 {
 171     char* beg = (char*) sbeg;
 172     char* end = (char*) sbeg + slen;
 173     char* adr = (char*) a;
 174     return (adr >= beg && adr < end);
 175 }
 176
 177 boolean
 178 is_valid_lisp_addr(os_vm_address_t addr)
 179 {
 180     struct thread *th;
 181
 182     if (in_range_p(addr, READ_ONLY_SPACE_START, READ_ONLY_SPACE_SIZE) ||
 183         in_range_p(addr, STATIC_SPACE_START, STATIC_SPACE_SIZE) ||
 184         in_range_p(addr, DYNAMIC_SPACE_START, dynamic_space_size))
 185         return 1;
 186     for_each_thread(th) {
 187         if (((os_vm_address_t)th->control_stack_start <= addr) &&
 188             (addr < (os_vm_address_t)th->control_stack_end))
 189             return 1;
 190         if (in_range_p(addr, (lispobj) th->binding_stack_start,
 191                        BINDING_STACK_SIZE))
 192             return 1;
 193     }
 194     return 0;
 195 }
 196 \f
 197 /*
 198  * any OS-dependent special low-level handling for signals
 199  */
 200
 201 #if defined LISP_FEATURE_GENCGC
 202
 203 /*
 204  * The GENCGC needs to be hooked into whatever signal is raised for
 205  * page fault on this OS.
 206  */
 207
 208 void
 209 memory_fault_handler(int signal, siginfo_t *siginfo, os_context_t *context)
 210 {
 211     void *fault_addr = arch_get_bad_addr(signal, siginfo, context);
 212
 213 #if defined(LISP_FEATURE_RESTORE_TLS_SEGMENT_REGISTER_FROM_CONTEXT)
 214     FSHOW_SIGNAL((stderr, "/ TLS: restoring fs: %p in memory_fault_handler\n",
 215                   *CONTEXT_ADDR_FROM_STEM(fs)));
 216     os_restore_tls_segment_register(context);
 217 #endif
 218
 219     FSHOW((stderr, "Memory fault at: %p, PC: %p\n", fault_addr, *os_context_pc_addr(context)));
 220
 221 #ifdef LISP_FEATURE_SB_SAFEPOINT
 222     if (!handle_safepoint_violation(context, fault_addr))
 223 #endif
 224
 225     if (!gencgc_handle_wp_violation(fault_addr))
 226         if(!handle_guard_page_triggered(context,fault_addr))
 227             lisp_memory_fault_error(context, fault_addr);
 228 }
 229
 230 #if defined(LISP_FEATURE_MACH_EXCEPTION_HANDLER)
 231 void
 232 mach_error_memory_fault_handler(int signal, siginfo_t *siginfo,
 233                                 os_context_t *context) {
 234     lose("Unhandled memory fault. Exiting.");
 235 }
 236 #endif
 237
 238 void
 239 os_install_interrupt_handlers(void)
 240 {
 241     SHOW("os_install_interrupt_handlers()/bsd-os/defined(GENCGC)");
 242 #if defined(LISP_FEATURE_MACH_EXCEPTION_HANDLER)
 243     undoably_install_low_level_interrupt_handler(SIG_MEMORY_FAULT,
 244                                                  mach_error_memory_fault_handler);
 245 #else
 246     undoably_install_low_level_interrupt_handler(SIG_MEMORY_FAULT,
 247 #ifdef LISP_FEATURE_FREEBSD
 248                                                  (__siginfohandler_t *)
 249 #endif
 250                                                  memory_fault_handler);
 251 #endif
 252
 253 #ifdef LISP_FEATURE_SB_THREAD
 254 # ifdef LISP_FEATURE_SB_SAFEPOINT
 255 #  ifdef LISP_FEATURE_SB_THRUPTION
 256     undoably_install_low_level_interrupt_handler(SIGPIPE, thruption_handler);
 257 #  endif
 258 # else
 259     undoably_install_low_level_interrupt_handler(SIG_STOP_FOR_GC,
 260                                                  sig_stop_for_gc_handler);
 261 # endif
 262 #endif
 263     SHOW("leaving os_install_interrupt_handlers()");
 264 }
 265
 266 #else /* Currently PPC/Darwin/Cheney only */
 267
 268 static void
 269 sigsegv_handler(int signal, siginfo_t *info, os_context_t *context)
 270 {
 271 #if 0
 272     unsigned int pc =  (unsigned int *)(*os_context_pc_addr(context));
 273 #endif
 274     os_vm_address_t addr;
 275
 276     addr = arch_get_bad_addr(signal, info, context);
 277     if (!cheneygc_handle_wp_violation(context, addr))
 278         if (!handle_guard_page_triggered(context, addr))
 279             interrupt_handle_now(signal, info, context);
 280 }
 281
 282 void
 283 os_install_interrupt_handlers(void)
 284 {
 285     SHOW("os_install_interrupt_handlers()/bsd-os/!defined(GENCGC)");
 286     undoably_install_low_level_interrupt_handler(SIG_MEMORY_FAULT,
 287                                                  sigsegv_handler);
 288 }
 289
 290 #endif /* defined GENCGC */
 291
 292 #ifdef __NetBSD__
 293 static void netbsd_init()
 294 {
 295     struct rlimit rl;
 296     int mib[2], osrev;
 297     size_t len;
 298
 299     /* Are we running on a sufficiently functional kernel? */
 300     mib[0] = CTL_KERN;
 301     mib[1] = KERN_OSREV;
 302
 303     len = sizeof(osrev);
 304     sysctl(mib, 2, &osrev, &len, NULL, 0);
 305
 306     /* If we're older than 2.0... */
 307     if (osrev < 200000000) {
 308         fprintf(stderr, "osrev = %d (needed at least 200000000).\n", osrev);
 309         lose("NetBSD kernel too old to run sbcl.\n");
 310     }
 311
 312     /* NetBSD counts mmap()ed space against the process's data size limit,
 313      * so yank it up. This might be a nasty thing to do? */
 314     getrlimit (RLIMIT_DATA, &rl);
 315     /* Amazingly for such a new port, the provenance and meaning of
 316        this number are unknown.  It might just mean REALLY_BIG_LIMIT,
 317        or possibly it should be calculated from dynamic space size.
 318        -- CSR, 2004-04-08 */
 319     rl.rlim_cur = 1073741824;
 320     if (setrlimit (RLIMIT_DATA, &rl) < 0) {
 321         fprintf (stderr,
 322                  "RUNTIME WARNING: unable to raise process data size limit:\n\
 323   %s.\n\
 324 The system may fail to start.\n",
 325                  strerror(errno));
 326     }
 327 }
 328
 329 /* Various routines in NetBSD's C library are compatibility wrappers
 330    for old versions. Programs must be processed by the C toolchain in
 331    order to get up-to-date definitions of such routines. */
 332 /* The stat-family, opendir, and readdir are used only in sb-posix, as
 333    of 2007-01-16. -- RMK */
 334 int
 335 _stat(const char *path, struct stat *sb)
 336 {
 337     return stat(path, sb);
 338 }
 339 int
 340 _lstat(const char *path, struct stat *sb)
 341 {
 342     return lstat(path, sb);
 343 }
 344 int
 345 _fstat(int fd, struct stat *sb)
 346 {
 347     return fstat(fd, sb);
 348 }
 349
 350 DIR *
 351 _opendir(const char *filename)
 352 {
 353     return opendir(filename);
 354 }
 355 struct dirent *
 356 _readdir(DIR *dirp)
 357 {
 358     return readdir(dirp);
 359 }
 360
 361 int
 362 _utime(const char *file, const struct utimbuf *timep)
 363 {
 364     return utime(file, timep);
 365 }
 366
 367 /* Used in sb-bsd-sockets. */
 368 int
 369 _socket(int domain, int type, int protocol)
 370 {
 371     return socket(domain, type, protocol);
 372 }
 373 #endif /* __NetBSD__ */
 374
 375 #ifdef __FreeBSD__
 376 extern int getosreldate(void);
 377
 378 int sig_memory_fault;
 379
 380 static void freebsd_init()
 381 {
 382     /* Memory fault signal on FreeBSD was changed from SIGBUS to
 383      * SIGSEGV. */
 384     if (getosreldate() < 700004)
 385         sig_memory_fault = SIGBUS;
 386     else
 387         sig_memory_fault = SIGSEGV;
 388
 389     /* Quote from sbcl-devel (NIIMI Satoshi): "Some OSes, like FreeBSD
 390      * 4.x with GENERIC kernel, does not enable SSE support even on
 391      * SSE capable CPUs". Detect this situation and skip the
 392      * fast_bzero sse/base selection logic that's normally done in
 393      * x86-assem.S.
 394      */
 395 #ifdef LISP_FEATURE_X86
 396     {
 397         size_t len;
 398         int instruction_sse;
 399
 400         len = sizeof(instruction_sse);
 401         if (sysctlbyname("hw.instruction_sse", &instruction_sse, &len,
 402                          NULL, 0) == 0 && instruction_sse != 0) {
 403             /* Use the SSE detector */
 404             fast_bzero_pointer = fast_bzero_detect;
 405         }
 406     }
 407 #endif /* LISP_FEATURE_X86 */
 408 }
 409
 410 #if defined(LISP_FEATURE_SB_THREAD) && defined(LISP_FEATURE_SB_FUTEX) \
 411     && !defined(LISP_FEATURE_SB_PTHREAD_FUTEX)
 412 int
 413 futex_wait(int *lock_word, long oldval, long sec, unsigned long usec)
 414 {
 415     struct timespec timeout;
 416     int ret;
 417
 418     if (sec < 0)
 419         ret = umtx_wait((void *)lock_word, oldval, NULL);
 420     else {
 421         timeout.tv_sec = sec;
 422         timeout.tv_nsec = usec * 1000;
 423         ret = umtx_wait((void *)lock_word, oldval, &timeout);
 424     }
 425
 426     switch (ret) {
 427     case 0:
 428         return 0;
 429     case ETIMEDOUT:
 430         return 1;
 431     case EINTR:
 432         return 2;
 433     default:
 434         /* EWOULDBLOCK and others, need to check the lock */
 435         return -1;
 436     }
 437 }
 438
 439 int
 440 futex_wake(int *lock_word, int n)
 441 {
 442     return umtx_wake((void *)lock_word, n);
 443 }
 444 #endif
 445 #endif /* __FreeBSD__ */
 446
 447 #ifdef LISP_FEATURE_DARWIN
 448 /* defined in ppc-darwin-os.c instead */
 449 #elif defined(LISP_FEATURE_FREEBSD)
 450 #ifndef KERN_PROC_PATHNAME
 451 #define KERN_PROC_PATHNAME 12
 452 #endif
 453
 454 char *
 455 os_get_runtime_executable_path(int external)
 456 {
 457     char path[PATH_MAX + 1];
 458
 459     if (getosreldate() >= 600024) {
 460         /* KERN_PROC_PATHNAME is available */
 461         size_t len = PATH_MAX + 1;
 462         int mib[4];
 463
 464         mib[0] = CTL_KERN;
 465         mib[1] = KERN_PROC;
 466         mib[2] = KERN_PROC_PATHNAME;
 467         mib[3] = -1;
 468         if (sysctl(mib, 4, &path, &len, NULL, 0) != 0)
 469             return NULL;
 470     } else {
 471         int size;
 472         size = readlink("/proc/curproc/file", path, sizeof(path) - 1);
 473         if (size < 0)
 474             return NULL;
 475         path[size] = '\0';
 476     }
 477     if (strcmp(path, "unknown") == 0)
 478         return NULL;
 479     return copied_string(path);
 480 }
 481 #elif defined(LISP_FEATURE_NETBSD) || defined(LISP_FEATURE_OPENBSD)
 482 char *
 483 os_get_runtime_executable_path(int external)
 484 {
 485     struct stat sb;
 486     if (!external && stat("/proc/curproc/file", &sb) == 0)
 487         return copied_string("/proc/curproc/file");
 488     return NULL;
 489 }
 490 #else /* Not DARWIN or FREEBSD or NETBSD or OPENBSD */
 491 char *
 492 os_get_runtime_executable_path(int external)
 493 {
 494     return NULL;
 495 }
 496 #endif
 497
 498 #ifdef __OpenBSD__
 499
 500 int openbsd_use_fxsave = 0;
 501
 502 void
 503 openbsd_init()
 504 {
 505 #ifdef LISP_FEATURE_X86
 506     int mib[2];
 507     size_t size;
 508 #endif
 509     /*
 510      * Show a warning if it looks like the memory available after
 511      * allocating the spaces won't be at least this much.
 512      */
 513 #ifdef LISP_FEATURE_X86_64
 514     const int wantfree = 64 * 1024 * 1024;
 515 #else
 516     const int wantfree = 32 * 1024 * 1024;
 517 #endif
 518     struct rlimit rl;
 519
 520 #ifdef LISP_FEATURE_X86
 521     /* Save the machdep.osfxsr sysctl for use by os_restore_fp_control() */
 522     mib[0] = CTL_MACHDEP;
 523     mib[1] = CPU_OSFXSR;
 524     size = sizeof (openbsd_use_fxsave);
 525     sysctl(mib, 2, &openbsd_use_fxsave, &size, NULL, 0);
 526 #endif
 527
 528     /* OpenBSD, like NetBSD, counts mmap()ed space against the
 529      * process's data size limit. If the soft limit is lower than the
 530      * hard limit then try to yank it up, this lets users in the
 531      * "staff" or "daemon" login classes run sbcl with larger dynamic
 532      * space sizes.
 533      */
 534     getrlimit (RLIMIT_DATA, &rl);
 535     if (rl.rlim_cur < rl.rlim_max) {
 536         rl.rlim_cur = rl.rlim_max;
 537         if (setrlimit (RLIMIT_DATA, &rl) < 0) {
 538             fprintf (stderr,
 539                      "RUNTIME WARNING: unable to raise process data size limit:\n\
 540   %s.\n\
 541 The system may fail to start.\n",
 542                      strerror(errno));
 543         }
 544     }
 545
 546     /*
 547      * Display a (hopefully) helpful warning if it looks like we won't
 548      * be able to allocate enough memory.
 549      */
 550     getrlimit (RLIMIT_DATA, &rl);
 551     if (dynamic_space_size + READ_ONLY_SPACE_SIZE + STATIC_SPACE_SIZE +
 552         LINKAGE_TABLE_SPACE_SIZE + wantfree > rl.rlim_cur)
 553         fprintf (stderr,
 554                  "RUNTIME WARNING: data size resource limit may be too low,\n"
 555                  "  try decreasing the dynamic space size with --dynamic-space-size\n"
 556                  "  or raising the datasize or datasize-max limits in /etc/login.conf\n");
 557 }
 558
 559 /* OpenBSD's dlsym() relies on the gcc bulitin
 560  * __builtin_return_address(0) returning an address in the
 561  * executable's text segment, but when called from lisp it will return
 562  * an address in the dynamic space.  Work around this by calling this
 563  * wrapper function instead. Note that tail-call optimization will
 564  * defeat this, disable it by saving the dlsym() return value in a
 565  * volatile variable.
 566 */
 567 void *
 568 os_dlsym(void *handle, const char *symbol)
 569 {
 570     void * volatile ret = dlsym(handle, symbol);
 571     return ret;
 572 }
 573
 574 #endif
 575
 576 #if defined(LISP_FEATURE_SB_WTIMER) && !defined(LISP_FEATURE_DARWIN)
 577 /*
 578  * Waitable timer implementation for the safepoint-based (SIGALRM-free)
 579  * timer facility using kqueue.
 580  */
 581 int
 582 os_create_wtimer()
 583 {
 584     int kq = kqueue();
 585     if (kq == -1)
 586         lose("os_create_wtimer: kqueue");
 587     return kq;
 588 }
 589
 590 int
 591 os_wait_for_wtimer(int kq)
 592 {
 593     struct kevent ev;
 594     int n;
 595     if ( (n = kevent(kq, 0, 0, &ev, 1, 0)) == -1) {
 596         if (errno != EINTR)
 597             lose("os_wtimer_listen failed");
 598         n = 0;
 599     }
 600     return n != 1;
 601 }
 602
 603 void
 604 os_close_wtimer(int kq)
 605 {
 606     if (close(kq) == -1)
 607         lose("os_close_wtimer failed");
 608 }
 609
 610 void
 611 os_set_wtimer(int kq, int sec, int nsec)
 612 {
 613     long long msec
 614         = ((long long) sec) * 1000 + (long long) (nsec+999999) / 1000000;
 615     if (msec > INT_MAX) msec = INT_MAX;
 616
 617     struct kevent ev;
 618     EV_SET(&ev, 1, EVFILT_TIMER, EV_ADD|EV_ENABLE|EV_ONESHOT, 0, (int)msec, 0);
 619     if (kevent(kq, &ev, 1, 0, 0, 0) == -1)
 620         perror("os_set_wtimer: kevent");
 621 }
 622
 623 void
 624 os_cancel_wtimer(int kq)
 625 {
 626     struct kevent ev;
 627     EV_SET(&ev, 1, EVFILT_TIMER, EV_DISABLE, 0, 0, 0);
 628     if (kevent(kq, &ev, 1, 0, 0, 0) == -1 && errno != ENOENT)
 629         perror("os_cancel_wtimer: kevent");
 630 }
 631 #endif