1 /*
   2  * Copyright 1999-2008 Sun Microsystems, Inc.  All Rights Reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  20  * CA 95054 USA or visit www.sun.com if you need additional information or
  21  * have any questions.
  22  *
  23  */
  24 
  25 // do not include  precompiled  header file
  26 # include "incls/_os_linux.cpp.incl"
  27 
  28 // put OS-includes here
  29 # include <sys/types.h>
  30 # include <sys/mman.h>
  31 # include <pthread.h>
  32 # include <signal.h>
  33 # include <errno.h>
  34 # include <dlfcn.h>
  35 # include <stdio.h>
  36 # include <unistd.h>
  37 # include <sys/resource.h>
  38 # include <pthread.h>
  39 # include <sys/stat.h>
  40 # include <sys/time.h>
  41 # include <sys/times.h>
  42 # include <sys/utsname.h>
  43 # include <sys/socket.h>
  44 # include <sys/wait.h>
  45 # include <pwd.h>
  46 # include <poll.h>
  47 # include <semaphore.h>
  48 # include <fcntl.h>
  49 # include <string.h>
  50 # include <syscall.h>
  51 # include <sys/sysinfo.h>
  52 # include <gnu/libc-version.h>
  53 # include <sys/ipc.h>
  54 # include <sys/shm.h>
  55 # include <link.h>
  56 
  57 #define MAX_PATH    (2 * K)
  58 
  59 // for timer info max values which include all bits
  60 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
  61 #define SEC_IN_NANOSECS  1000000000LL
  62 
  63 ////////////////////////////////////////////////////////////////////////////////
  64 // global variables
  65 julong os::Linux::_physical_memory = 0;
  66 
  67 address   os::Linux::_initial_thread_stack_bottom = NULL;
  68 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
  69 
  70 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
  71 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
  72 Mutex* os::Linux::_createThread_lock = NULL;
  73 pthread_t os::Linux::_main_thread;
  74 int os::Linux::_page_size = -1;
  75 bool os::Linux::_is_floating_stack = false;
  76 bool os::Linux::_is_NPTL = false;
  77 bool os::Linux::_supports_fast_thread_cpu_time = false;
  78 const char * os::Linux::_glibc_version = NULL;
  79 const char * os::Linux::_libpthread_version = NULL;
  80 
  81 static jlong initial_time_count=0;
  82 
  83 static int clock_tics_per_sec = 100;
  84 
  85 // For diagnostics to print a message once. see run_periodic_checks
  86 static sigset_t check_signal_done;
  87 static bool check_signals = true;;
  88 
  89 static pid_t _initial_pid = 0;
  90 
  91 /* Signal number used to suspend/resume a thread */
  92 
  93 /* do not use any signal number less than SIGSEGV, see 4355769 */
  94 static int SR_signum = SIGUSR2;
  95 sigset_t SR_sigset;
  96 
  97 /* Used to protect dlsym() calls */
  98 static pthread_mutex_t dl_mutex;
  99 
 100 ////////////////////////////////////////////////////////////////////////////////
 101 // utility functions
 102 
 103 static int SR_initialize();
 104 static int SR_finalize();
 105 
 106 julong os::available_memory() {
 107   return Linux::available_memory();
 108 }
 109 
 110 julong os::Linux::available_memory() {
 111   // values in struct sysinfo are "unsigned long"
 112   struct sysinfo si;
 113   sysinfo(&si);
 114 
 115   return (julong)si.freeram * si.mem_unit;
 116 }
 117 
 118 julong os::physical_memory() {
 119   return Linux::physical_memory();
 120 }
 121 
 122 julong os::allocatable_physical_memory(julong size) {
 123 #ifdef _LP64
 124   return size;
 125 #else
 126   julong result = MIN2(size, (julong)3800*M);
 127    if (!is_allocatable(result)) {
 128      // See comments under solaris for alignment considerations
 129      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 130      result =  MIN2(size, reasonable_size);
 131    }
 132    return result;
 133 #endif // _LP64
 134 }
 135 
 136 ////////////////////////////////////////////////////////////////////////////////
 137 // environment support
 138 
 139 bool os::getenv(const char* name, char* buf, int len) {
 140   const char* val = ::getenv(name);
 141   if (val != NULL && strlen(val) < (size_t)len) {
 142     strcpy(buf, val);
 143     return true;
 144   }
 145   if (len > 0) buf[0] = 0;  // return a null string
 146   return false;
 147 }
 148 
 149 
 150 // Return true if user is running as root.
 151 
 152 bool os::have_special_privileges() {
 153   static bool init = false;
 154   static bool privileges = false;
 155   if (!init) {
 156     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 157     init = true;
 158   }
 159   return privileges;
 160 }
 161 
 162 
 163 #ifndef SYS_gettid
 164 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 165 #ifdef __ia64__
 166 #define SYS_gettid 1105
 167 #elif __i386__
 168 #define SYS_gettid 224
 169 #elif __amd64__
 170 #define SYS_gettid 186
 171 #elif __sparc__
 172 #define SYS_gettid 143
 173 #else
 174 #error define gettid for the arch
 175 #endif
 176 #endif
 177 
 178 // Cpu architecture string
 179 #if   defined(IA64)
 180 static char cpu_arch[] = "ia64";
 181 #elif defined(IA32)
 182 static char cpu_arch[] = "i386";
 183 #elif defined(AMD64)
 184 static char cpu_arch[] = "amd64";
 185 #elif defined(SPARC)
 186 #  ifdef _LP64
 187 static char cpu_arch[] = "sparcv9";
 188 #  else
 189 static char cpu_arch[] = "sparc";
 190 #  endif
 191 #else
 192 #error Add appropriate cpu_arch setting
 193 #endif
 194 
 195 
 196 // pid_t gettid()
 197 //
 198 // Returns the kernel thread id of the currently running thread. Kernel
 199 // thread id is used to access /proc.
 200 //
 201 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 202 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 203 //
 204 pid_t os::Linux::gettid() {
 205   int rslt = syscall(SYS_gettid);
 206   if (rslt == -1) {
 207      // old kernel, no NPTL support
 208      return getpid();
 209   } else {
 210      return (pid_t)rslt;
 211   }
 212 }
 213 
 214 // Most versions of linux have a bug where the number of processors are
 215 // determined by looking at the /proc file system.  In a chroot environment,
 216 // the system call returns 1.  This causes the VM to act as if it is
 217 // a single processor and elide locking (see is_MP() call).
 218 static bool unsafe_chroot_detected = false;
 219 static const char *unstable_chroot_error = "/proc file system not found.\n"
 220                      "Java may be unstable running multithreaded in a chroot "
 221                      "environment on Linux when /proc filesystem is not mounted.";
 222 
 223 void os::Linux::initialize_system_info() {
 224   _processor_count = sysconf(_SC_NPROCESSORS_CONF);
 225   if (_processor_count == 1) {
 226     pid_t pid = os::Linux::gettid();
 227     char fname[32];
 228     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 229     FILE *fp = fopen(fname, "r");
 230     if (fp == NULL) {
 231       unsafe_chroot_detected = true;
 232     } else {
 233       fclose(fp);
 234     }
 235   }
 236   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 237   assert(_processor_count > 0, "linux error");
 238 }
 239 
 240 void os::init_system_properties_values() {
 241 //  char arch[12];
 242 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 243 
 244   // The next steps are taken in the product version:
 245   //
 246   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 247   // This library should be located at:
 248   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 249   //
 250   // If "/jre/lib/" appears at the right place in the path, then we
 251   // assume libjvm[_g].so is installed in a JDK and we use this path.
 252   //
 253   // Otherwise exit with message: "Could not create the Java virtual machine."
 254   //
 255   // The following extra steps are taken in the debugging version:
 256   //
 257   // If "/jre/lib/" does NOT appear at the right place in the path
 258   // instead of exit check for $JAVA_HOME environment variable.
 259   //
 260   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 261   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 262   // it looks like libjvm[_g].so is installed there
 263   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 264   //
 265   // Otherwise exit.
 266   //
 267   // Important note: if the location of libjvm.so changes this
 268   // code needs to be changed accordingly.
 269 
 270   // The next few definitions allow the code to be verbatim:
 271 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 272 #define getenv(n) ::getenv(n)
 273 
 274 /*
 275  * See ld(1):
 276  *      The linker uses the following search paths to locate required
 277  *      shared libraries:
 278  *        1: ...
 279  *        ...
 280  *        7: The default directories, normally /lib and /usr/lib.
 281  */
 282 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 283 
 284 #define EXTENSIONS_DIR  "/lib/ext"
 285 #define ENDORSED_DIR    "/lib/endorsed"
 286 #define REG_DIR         "/usr/java/packages"
 287 
 288   {
 289     /* sysclasspath, java_home, dll_dir */
 290     {
 291         char *home_path;
 292         char *dll_path;
 293         char *pslash;
 294         char buf[MAXPATHLEN];
 295         os::jvm_path(buf, sizeof(buf));
 296 
 297         // Found the full path to libjvm.so.
 298         // Now cut the path to <java_home>/jre if we can.
 299         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 300         pslash = strrchr(buf, '/');
 301         if (pslash != NULL)
 302             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 303         dll_path = malloc(strlen(buf) + 1);
 304         if (dll_path == NULL)
 305             return;
 306         strcpy(dll_path, buf);
 307         Arguments::set_dll_dir(dll_path);
 308 
 309         if (pslash != NULL) {
 310             pslash = strrchr(buf, '/');
 311             if (pslash != NULL) {
 312                 *pslash = '\0';       /* get rid of /<arch> */
 313                 pslash = strrchr(buf, '/');
 314                 if (pslash != NULL)
 315                     *pslash = '\0';   /* get rid of /lib */
 316             }
 317         }
 318 
 319         home_path = malloc(strlen(buf) + 1);
 320         if (home_path == NULL)
 321             return;
 322         strcpy(home_path, buf);
 323         Arguments::set_java_home(home_path);
 324 
 325         if (!set_boot_path('/', ':'))
 326             return;
 327     }
 328 
 329     /*
 330      * Where to look for native libraries
 331      *
 332      * Note: Due to a legacy implementation, most of the library path
 333      * is set in the launcher.  This was to accomodate linking restrictions
 334      * on legacy Linux implementations (which are no longer supported).
 335      * Eventually, all the library path setting will be done here.
 336      *
 337      * However, to prevent the proliferation of improperly built native
 338      * libraries, the new path component /usr/java/packages is added here.
 339      * Eventually, all the library path setting will be done here.
 340      */
 341     {
 342         char *ld_library_path;
 343 
 344         /*
 345          * Construct the invariant part of ld_library_path. Note that the
 346          * space for the colon and the trailing null are provided by the
 347          * nulls included by the sizeof operator (so actually we allocate
 348          * a byte more than necessary).
 349          */
 350         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 351             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 352         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 353 
 354         /*
 355          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 356          * should always exist (until the legacy problem cited above is
 357          * addressed).
 358          */
 359         char *v = getenv("LD_LIBRARY_PATH");
 360         if (v != NULL) {
 361             char *t = ld_library_path;
 362             /* That's +1 for the colon and +1 for the trailing '\0' */
 363             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 364             sprintf(ld_library_path, "%s:%s", v, t);
 365         }
 366         Arguments::set_library_path(ld_library_path);
 367     }
 368 
 369     /*
 370      * Extensions directories.
 371      *
 372      * Note that the space for the colon and the trailing null are provided
 373      * by the nulls included by the sizeof operator (so actually one byte more
 374      * than necessary is allocated).
 375      */
 376     {
 377         char *buf = malloc(strlen(Arguments::get_java_home()) +
 378             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 379         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 380             Arguments::get_java_home());
 381         Arguments::set_ext_dirs(buf);
 382     }
 383 
 384     /* Endorsed standards default directory. */
 385     {
 386         char * buf;
 387         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 388         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 389         Arguments::set_endorsed_dirs(buf);
 390     }
 391   }
 392 
 393 #undef malloc
 394 #undef getenv
 395 #undef EXTENSIONS_DIR
 396 #undef ENDORSED_DIR
 397 
 398   // Done
 399   return;
 400 }
 401 
 402 ////////////////////////////////////////////////////////////////////////////////
 403 // breakpoint support
 404 
 405 void os::breakpoint() {
 406   BREAKPOINT;
 407 }
 408 
 409 extern "C" void breakpoint() {
 410   // use debugger to set breakpoint here
 411 }
 412 
 413 ////////////////////////////////////////////////////////////////////////////////
 414 // signal support
 415 
 416 debug_only(static bool signal_sets_initialized = false);
 417 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 418 
 419 bool os::Linux::is_sig_ignored(int sig) {
 420       struct sigaction oact;
 421       sigaction(sig, (struct sigaction*)NULL, &oact);
 422       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 423                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 424       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 425            return true;
 426       else
 427            return false;
 428 }
 429 
 430 void os::Linux::signal_sets_init() {
 431   // Should also have an assertion stating we are still single-threaded.
 432   assert(!signal_sets_initialized, "Already initialized");
 433   // Fill in signals that are necessarily unblocked for all threads in
 434   // the VM. Currently, we unblock the following signals:
 435   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 436   //                         by -Xrs (=ReduceSignalUsage));
 437   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 438   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 439   // the dispositions or masks wrt these signals.
 440   // Programs embedding the VM that want to use the above signals for their
 441   // own purposes must, at this time, use the "-Xrs" option to prevent
 442   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 443   // (See bug 4345157, and other related bugs).
 444   // In reality, though, unblocking these signals is really a nop, since
 445   // these signals are not blocked by default.
 446   sigemptyset(&unblocked_sigs);
 447   sigemptyset(&allowdebug_blocked_sigs);
 448   sigaddset(&unblocked_sigs, SIGILL);
 449   sigaddset(&unblocked_sigs, SIGSEGV);
 450   sigaddset(&unblocked_sigs, SIGBUS);
 451   sigaddset(&unblocked_sigs, SIGFPE);
 452   sigaddset(&unblocked_sigs, SR_signum);
 453 
 454   if (!ReduceSignalUsage) {
 455    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 456       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 457       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 458    }
 459    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 460       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 461       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 462    }
 463    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 464       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 465       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 466    }
 467   }
 468   // Fill in signals that are blocked by all but the VM thread.
 469   sigemptyset(&vm_sigs);
 470   if (!ReduceSignalUsage)
 471     sigaddset(&vm_sigs, BREAK_SIGNAL);
 472   debug_only(signal_sets_initialized = true);
 473 
 474 }
 475 
 476 // These are signals that are unblocked while a thread is running Java.
 477 // (For some reason, they get blocked by default.)
 478 sigset_t* os::Linux::unblocked_signals() {
 479   assert(signal_sets_initialized, "Not initialized");
 480   return &unblocked_sigs;
 481 }
 482 
 483 // These are the signals that are blocked while a (non-VM) thread is
 484 // running Java. Only the VM thread handles these signals.
 485 sigset_t* os::Linux::vm_signals() {
 486   assert(signal_sets_initialized, "Not initialized");
 487   return &vm_sigs;
 488 }
 489 
 490 // These are signals that are blocked during cond_wait to allow debugger in
 491 sigset_t* os::Linux::allowdebug_blocked_signals() {
 492   assert(signal_sets_initialized, "Not initialized");
 493   return &allowdebug_blocked_sigs;
 494 }
 495 
 496 void os::Linux::hotspot_sigmask(Thread* thread) {
 497 
 498   //Save caller's signal mask before setting VM signal mask
 499   sigset_t caller_sigmask;
 500   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 501 
 502   OSThread* osthread = thread->osthread();
 503   osthread->set_caller_sigmask(caller_sigmask);
 504 
 505   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 506 
 507   if (!ReduceSignalUsage) {
 508     if (thread->is_VM_thread()) {
 509       // Only the VM thread handles BREAK_SIGNAL ...
 510       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 511     } else {
 512       // ... all other threads block BREAK_SIGNAL
 513       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 514     }
 515   }
 516 }
 517 
 518 //////////////////////////////////////////////////////////////////////////////
 519 // detecting pthread library
 520 
 521 void os::Linux::libpthread_init() {
 522   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 523   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 524   // generic name for earlier versions.
 525   // Define macros here so we can build HotSpot on old systems.
 526 # ifndef _CS_GNU_LIBC_VERSION
 527 # define _CS_GNU_LIBC_VERSION 2
 528 # endif
 529 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 530 # define _CS_GNU_LIBPTHREAD_VERSION 3
 531 # endif
 532 
 533   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 534   if (n > 0) {
 535      char *str = (char *)malloc(n);
 536      confstr(_CS_GNU_LIBC_VERSION, str, n);
 537      os::Linux::set_glibc_version(str);
 538   } else {
 539      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 540      static char _gnu_libc_version[32];
 541      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 542               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 543      os::Linux::set_glibc_version(_gnu_libc_version);
 544   }
 545 
 546   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 547   if (n > 0) {
 548      char *str = (char *)malloc(n);
 549      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 550      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 551      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 552      // is the case. LinuxThreads has a hard limit on max number of threads.
 553      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 554      // On the other hand, NPTL does not have such a limit, sysconf()
 555      // will return -1 and errno is not changed. Check if it is really NPTL.
 556      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 557          strstr(str, "NPTL") &&
 558          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 559        free(str);
 560        os::Linux::set_libpthread_version("linuxthreads");
 561      } else {
 562        os::Linux::set_libpthread_version(str);
 563      }
 564   } else {
 565     // glibc before 2.3.2 only has LinuxThreads.
 566     os::Linux::set_libpthread_version("linuxthreads");
 567   }
 568 
 569   if (strstr(libpthread_version(), "NPTL")) {
 570      os::Linux::set_is_NPTL();
 571   } else {
 572      os::Linux::set_is_LinuxThreads();
 573   }
 574 
 575   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 576   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 577   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 578      os::Linux::set_is_floating_stack();
 579   }
 580 }
 581 
 582 /////////////////////////////////////////////////////////////////////////////
 583 // thread stack
 584 
 585 // Force Linux kernel to expand current thread stack. If "bottom" is close
 586 // to the stack guard, caller should block all signals.
 587 //
 588 // MAP_GROWSDOWN:
 589 //   A special mmap() flag that is used to implement thread stacks. It tells
 590 //   kernel that the memory region should extend downwards when needed. This
 591 //   allows early versions of LinuxThreads to only mmap the first few pages
 592 //   when creating a new thread. Linux kernel will automatically expand thread
 593 //   stack as needed (on page faults).
 594 //
 595 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 596 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 597 //   region, it's hard to tell if the fault is due to a legitimate stack
 598 //   access or because of reading/writing non-exist memory (e.g. buffer
 599 //   overrun). As a rule, if the fault happens below current stack pointer,
 600 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 601 //   application (see Linux kernel fault.c).
 602 //
 603 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 604 //   stack overflow detection.
 605 //
 606 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 607 //   not use this flag. However, the stack of initial thread is not created
 608 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 609 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 610 //   and then attach the thread to JVM.
 611 //
 612 // To get around the problem and allow stack banging on Linux, we need to
 613 // manually expand thread stack after receiving the SIGSEGV.
 614 //
 615 // There are two ways to expand thread stack to address "bottom", we used
 616 // both of them in JVM before 1.5:
 617 //   1. adjust stack pointer first so that it is below "bottom", and then
 618 //      touch "bottom"
 619 //   2. mmap() the page in question
 620 //
 621 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 622 // if current sp is already near the lower end of page 101, and we need to
 623 // call mmap() to map page 100, it is possible that part of the mmap() frame
 624 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 625 // That will destroy the mmap() frame and cause VM to crash.
 626 //
 627 // The following code works by adjusting sp first, then accessing the "bottom"
 628 // page to force a page fault. Linux kernel will then automatically expand the
 629 // stack mapping.
 630 //
 631 // _expand_stack_to() assumes its frame size is less than page size, which
 632 // should always be true if the function is not inlined.
 633 
 634 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 635 #define NOINLINE
 636 #else
 637 #define NOINLINE __attribute__ ((noinline))
 638 #endif
 639 
 640 static void _expand_stack_to(address bottom) NOINLINE;
 641 
 642 static void _expand_stack_to(address bottom) {
 643   address sp;
 644   size_t size;
 645   volatile char *p;
 646 
 647   // Adjust bottom to point to the largest address within the same page, it
 648   // gives us a one-page buffer if alloca() allocates slightly more memory.
 649   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 650   bottom += os::Linux::page_size() - 1;
 651 
 652   // sp might be slightly above current stack pointer; if that's the case, we
 653   // will alloca() a little more space than necessary, which is OK. Don't use
 654   // os::current_stack_pointer(), as its result can be slightly below current
 655   // stack pointer, causing us to not alloca enough to reach "bottom".
 656   sp = (address)&sp;
 657 
 658   if (sp > bottom) {
 659     size = sp - bottom;
 660     p = (volatile char *)alloca(size);
 661     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 662     p[0] = '\0';
 663   }
 664 }
 665 
 666 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 667   assert(t!=NULL, "just checking");
 668   assert(t->osthread()->expanding_stack(), "expand should be set");
 669   assert(t->stack_base() != NULL, "stack_base was not initialized");
 670 
 671   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 672     sigset_t mask_all, old_sigset;
 673     sigfillset(&mask_all);
 674     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 675     _expand_stack_to(addr);
 676     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 677     return true;
 678   }
 679   return false;
 680 }
 681 
 682 //////////////////////////////////////////////////////////////////////////////
 683 // create new thread
 684 
 685 static address highest_vm_reserved_address();
 686 
 687 // check if it's safe to start a new thread
 688 static bool _thread_safety_check(Thread* thread) {
 689   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 690     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 691     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 692     //   allocated (MAP_FIXED) from high address space. Every thread stack
 693     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 694     //   it to other values if they rebuild LinuxThreads).
 695     //
 696     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 697     // the memory region has already been mmap'ed. That means if we have too
 698     // many threads and/or very large heap, eventually thread stack will
 699     // collide with heap.
 700     //
 701     // Here we try to prevent heap/stack collision by comparing current
 702     // stack bottom with the highest address that has been mmap'ed by JVM
 703     // plus a safety margin for memory maps created by native code.
 704     //
 705     // This feature can be disabled by setting ThreadSafetyMargin to 0
 706     //
 707     if (ThreadSafetyMargin > 0) {
 708       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 709 
 710       // not safe if our stack extends below the safety margin
 711       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 712     } else {
 713       return true;
 714     }
 715   } else {
 716     // Floating stack LinuxThreads or NPTL:
 717     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 718     //   there's not enough space left, pthread_create() will fail. If we come
 719     //   here, that means enough space has been reserved for stack.
 720     return true;
 721   }
 722 }
 723 
 724 // Thread start routine for all newly created threads
 725 static void *java_start(Thread *thread) {
 726   // Try to randomize the cache line index of hot stack frames.
 727   // This helps when threads of the same stack traces evict each other's
 728   // cache lines. The threads can be either from the same JVM instance, or
 729   // from different JVM instances. The benefit is especially true for
 730   // processors with hyperthreading technology.
 731   static int counter = 0;
 732   int pid = os::current_process_id();
 733   alloca(((pid ^ counter++) & 7) * 128);
 734 
 735   ThreadLocalStorage::set_thread(thread);
 736 
 737   OSThread* osthread = thread->osthread();
 738   Monitor* sync = osthread->startThread_lock();
 739 
 740   // non floating stack LinuxThreads needs extra check, see above
 741   if (!_thread_safety_check(thread)) {
 742     // notify parent thread
 743     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 744     osthread->set_state(ZOMBIE);
 745     sync->notify_all();
 746     return NULL;
 747   }
 748 
 749   // thread_id is kernel thread id (similar to Solaris LWP id)
 750   osthread->set_thread_id(os::Linux::gettid());
 751 
 752   if (UseNUMA) {
 753     int lgrp_id = os::numa_get_group_id();
 754     if (lgrp_id != -1) {
 755       thread->set_lgrp_id(lgrp_id);
 756     }
 757   }
 758   // initialize signal mask for this thread
 759   os::Linux::hotspot_sigmask(thread);
 760 
 761   // initialize floating point control register
 762   os::Linux::init_thread_fpu_state();
 763 
 764   // handshaking with parent thread
 765   {
 766     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 767 
 768     // notify parent thread
 769     osthread->set_state(INITIALIZED);
 770     sync->notify_all();
 771 
 772     // wait until os::start_thread()
 773     while (osthread->get_state() == INITIALIZED) {
 774       sync->wait(Mutex::_no_safepoint_check_flag);
 775     }
 776   }
 777 
 778   // call one more level start routine
 779   thread->run();
 780 
 781   return 0;
 782 }
 783 
 784 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 785   assert(thread->osthread() == NULL, "caller responsible");
 786 
 787   // Allocate the OSThread object
 788   OSThread* osthread = new OSThread(NULL, NULL);
 789   if (osthread == NULL) {
 790     return false;
 791   }
 792 
 793   // set the correct thread state
 794   osthread->set_thread_type(thr_type);
 795 
 796   // Initial state is ALLOCATED but not INITIALIZED
 797   osthread->set_state(ALLOCATED);
 798 
 799   thread->set_osthread(osthread);
 800 
 801   // init thread attributes
 802   pthread_attr_t attr;
 803   pthread_attr_init(&attr);
 804   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 805 
 806   // stack size
 807   if (os::Linux::supports_variable_stack_size()) {
 808     // calculate stack size if it's not specified by caller
 809     if (stack_size == 0) {
 810       stack_size = os::Linux::default_stack_size(thr_type);
 811 
 812       switch (thr_type) {
 813       case os::java_thread:
 814         // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
 815         if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
 816         break;
 817       case os::compiler_thread:
 818         if (CompilerThreadStackSize > 0) {
 819           stack_size = (size_t)(CompilerThreadStackSize * K);
 820           break;
 821         } // else fall through:
 822           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 823       case os::vm_thread:
 824       case os::pgc_thread:
 825       case os::cgc_thread:
 826       case os::watcher_thread:
 827         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 828         break;
 829       }
 830     }
 831 
 832     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 833     pthread_attr_setstacksize(&attr, stack_size);
 834   } else {
 835     // let pthread_create() pick the default value.
 836   }
 837 
 838   // glibc guard page
 839   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 840 
 841   ThreadState state;
 842 
 843   {
 844     // Serialize thread creation if we are running with fixed stack LinuxThreads
 845     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 846     if (lock) {
 847       os::Linux::createThread_lock()->lock_without_safepoint_check();
 848     }
 849 
 850     pthread_t tid;
 851     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 852 
 853     pthread_attr_destroy(&attr);
 854 
 855     if (ret != 0) {
 856       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 857         perror("pthread_create()");
 858       }
 859       // Need to clean up stuff we've allocated so far
 860       thread->set_osthread(NULL);
 861       delete osthread;
 862       if (lock) os::Linux::createThread_lock()->unlock();
 863       return false;
 864     }
 865 
 866     // Store pthread info into the OSThread
 867     osthread->set_pthread_id(tid);
 868 
 869     // Wait until child thread is either initialized or aborted
 870     {
 871       Monitor* sync_with_child = osthread->startThread_lock();
 872       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 873       while ((state = osthread->get_state()) == ALLOCATED) {
 874         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 875       }
 876     }
 877 
 878     if (lock) {
 879       os::Linux::createThread_lock()->unlock();
 880     }
 881   }
 882 
 883   // Aborted due to thread limit being reached
 884   if (state == ZOMBIE) {
 885       thread->set_osthread(NULL);
 886       delete osthread;
 887       return false;
 888   }
 889 
 890   // The thread is returned suspended (in state INITIALIZED),
 891   // and is started higher up in the call chain
 892   assert(state == INITIALIZED, "race condition");
 893   return true;
 894 }
 895 
 896 /////////////////////////////////////////////////////////////////////////////
 897 // attach existing thread
 898 
 899 // bootstrap the main thread
 900 bool os::create_main_thread(JavaThread* thread) {
 901   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 902   return create_attached_thread(thread);
 903 }
 904 
 905 bool os::create_attached_thread(JavaThread* thread) {
 906 #ifdef ASSERT
 907     thread->verify_not_published();
 908 #endif
 909 
 910   // Allocate the OSThread object
 911   OSThread* osthread = new OSThread(NULL, NULL);
 912 
 913   if (osthread == NULL) {
 914     return false;
 915   }
 916 
 917   // Store pthread info into the OSThread
 918   osthread->set_thread_id(os::Linux::gettid());
 919   osthread->set_pthread_id(::pthread_self());
 920 
 921   // initialize floating point control register
 922   os::Linux::init_thread_fpu_state();
 923 
 924   // Initial thread state is RUNNABLE
 925   osthread->set_state(RUNNABLE);
 926 
 927   thread->set_osthread(osthread);
 928 
 929   if (UseNUMA) {
 930     int lgrp_id = os::numa_get_group_id();
 931     if (lgrp_id != -1) {
 932       thread->set_lgrp_id(lgrp_id);
 933     }
 934   }
 935 
 936   if (os::Linux::is_initial_thread()) {
 937     // If current thread is initial thread, its stack is mapped on demand,
 938     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 939     // the entire stack region to avoid SEGV in stack banging.
 940     // It is also useful to get around the heap-stack-gap problem on SuSE
 941     // kernel (see 4821821 for details). We first expand stack to the top
 942     // of yellow zone, then enable stack yellow zone (order is significant,
 943     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 944     // is no gap between the last two virtual memory regions.
 945 
 946     JavaThread *jt = (JavaThread *)thread;
 947     address addr = jt->stack_yellow_zone_base();
 948     assert(addr != NULL, "initialization problem?");
 949     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 950 
 951     osthread->set_expanding_stack();
 952     os::Linux::manually_expand_stack(jt, addr);
 953     osthread->clear_expanding_stack();
 954   }
 955 
 956   // initialize signal mask for this thread
 957   // and save the caller's signal mask
 958   os::Linux::hotspot_sigmask(thread);
 959 
 960   return true;
 961 }
 962 
 963 void os::pd_start_thread(Thread* thread) {
 964   OSThread * osthread = thread->osthread();
 965   assert(osthread->get_state() != INITIALIZED, "just checking");
 966   Monitor* sync_with_child = osthread->startThread_lock();
 967   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 968   sync_with_child->notify();
 969 }
 970 
 971 // Free Linux resources related to the OSThread
 972 void os::free_thread(OSThread* osthread) {
 973   assert(osthread != NULL, "osthread not set");
 974 
 975   if (Thread::current()->osthread() == osthread) {
 976     // Restore caller's signal mask
 977     sigset_t sigmask = osthread->caller_sigmask();
 978     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 979    }
 980 
 981   delete osthread;
 982 }
 983 
 984 //////////////////////////////////////////////////////////////////////////////
 985 // thread local storage
 986 
 987 int os::allocate_thread_local_storage() {
 988   pthread_key_t key;
 989   int rslt = pthread_key_create(&key, NULL);
 990   assert(rslt == 0, "cannot allocate thread local storage");
 991   return (int)key;
 992 }
 993 
 994 // Note: This is currently not used by VM, as we don't destroy TLS key
 995 // on VM exit.
 996 void os::free_thread_local_storage(int index) {
 997   int rslt = pthread_key_delete((pthread_key_t)index);
 998   assert(rslt == 0, "invalid index");
 999 }
1000 
1001 void os::thread_local_storage_at_put(int index, void* value) {
1002   int rslt = pthread_setspecific((pthread_key_t)index, value);
1003   assert(rslt == 0, "pthread_setspecific failed");
1004 }
1005 
1006 extern "C" Thread* get_thread() {
1007   return ThreadLocalStorage::thread();
1008 }
1009 
1010 //////////////////////////////////////////////////////////////////////////////
1011 // initial thread
1012 
1013 // Check if current thread is the initial thread, similar to Solaris thr_main.
1014 bool os::Linux::is_initial_thread(void) {
1015   char dummy;
1016   // If called before init complete, thread stack bottom will be null.
1017   // Can be called if fatal error occurs before initialization.
1018   if (initial_thread_stack_bottom() == NULL) return false;
1019   assert(initial_thread_stack_bottom() != NULL &&
1020          initial_thread_stack_size()   != 0,
1021          "os::init did not locate initial thread's stack region");
1022   if ((address)&dummy >= initial_thread_stack_bottom() &&
1023       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1024        return true;
1025   else return false;
1026 }
1027 
1028 // Find the virtual memory area that contains addr
1029 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1030   FILE *fp = fopen("/proc/self/maps", "r");
1031   if (fp) {
1032     address low, high;
1033     while (!feof(fp)) {
1034       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1035         if (low <= addr && addr < high) {
1036            if (vma_low)  *vma_low  = low;
1037            if (vma_high) *vma_high = high;
1038            fclose (fp);
1039            return true;
1040         }
1041       }
1042       for (;;) {
1043         int ch = fgetc(fp);
1044         if (ch == EOF || ch == (int)'\n') break;
1045       }
1046     }
1047     fclose(fp);
1048   }
1049   return false;
1050 }
1051 
1052 // Locate initial thread stack. This special handling of initial thread stack
1053 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1054 // bogus value for initial thread.
1055 void os::Linux::capture_initial_stack(size_t max_size) {
1056   // stack size is the easy part, get it from RLIMIT_STACK
1057   size_t stack_size;
1058   struct rlimit rlim;
1059   getrlimit(RLIMIT_STACK, &rlim);
1060   stack_size = rlim.rlim_cur;
1061 
1062   // 6308388: a bug in ld.so will relocate its own .data section to the
1063   //   lower end of primordial stack; reduce ulimit -s value a little bit
1064   //   so we won't install guard page on ld.so's data section.
1065   stack_size -= 2 * page_size();
1066 
1067   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1068   //   7.1, in both cases we will get 2G in return value.
1069   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1070   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1071   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1072   //   in case other parts in glibc still assumes 2M max stack size.
1073   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1074 #ifndef IA64
1075   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1076 #else
1077   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1078   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1079 #endif
1080 
1081   // Try to figure out where the stack base (top) is. This is harder.
1082   //
1083   // When an application is started, glibc saves the initial stack pointer in
1084   // a global variable "__libc_stack_end", which is then used by system
1085   // libraries. __libc_stack_end should be pretty close to stack top. The
1086   // variable is available since the very early days. However, because it is
1087   // a private interface, it could disappear in the future.
1088   //
1089   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1090   // to __libc_stack_end, it is very close to stack top, but isn't the real
1091   // stack top. Note that /proc may not exist if VM is running as a chroot
1092   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1093   // /proc/<pid>/stat could change in the future (though unlikely).
1094   //
1095   // We try __libc_stack_end first. If that doesn't work, look for
1096   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1097   // as a hint, which should work well in most cases.
1098 
1099   uintptr_t stack_start;
1100 
1101   // try __libc_stack_end first
1102   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1103   if (p && *p) {
1104     stack_start = *p;
1105   } else {
1106     // see if we can get the start_stack field from /proc/self/stat
1107     FILE *fp;
1108     int pid;
1109     char state;
1110     int ppid;
1111     int pgrp;
1112     int session;
1113     int nr;
1114     int tpgrp;
1115     unsigned long flags;
1116     unsigned long minflt;
1117     unsigned long cminflt;
1118     unsigned long majflt;
1119     unsigned long cmajflt;
1120     unsigned long utime;
1121     unsigned long stime;
1122     long cutime;
1123     long cstime;
1124     long prio;
1125     long nice;
1126     long junk;
1127     long it_real;
1128     uintptr_t start;
1129     uintptr_t vsize;
1130     uintptr_t rss;
1131     unsigned long rsslim;
1132     uintptr_t scodes;
1133     uintptr_t ecode;
1134     int i;
1135 
1136     // Figure what the primordial thread stack base is. Code is inspired
1137     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1138     // followed by command name surrounded by parentheses, state, etc.
1139     char stat[2048];
1140     int statlen;
1141 
1142     fp = fopen("/proc/self/stat", "r");
1143     if (fp) {
1144       statlen = fread(stat, 1, 2047, fp);
1145       stat[statlen] = '\0';
1146       fclose(fp);
1147 
1148       // Skip pid and the command string. Note that we could be dealing with
1149       // weird command names, e.g. user could decide to rename java launcher
1150       // to "java 1.4.2 :)", then the stat file would look like
1151       //                1234 (java 1.4.2 :)) R ... ...
1152       // We don't really need to know the command string, just find the last
1153       // occurrence of ")" and then start parsing from there. See bug 4726580.
1154       char * s = strrchr(stat, ')');
1155 
1156       i = 0;
1157       if (s) {
1158         // Skip blank chars
1159         do s++; while (isspace(*s));
1160 
1161         /*                                     1   1   1   1   1   1   1   1   1   1   2   2   2   2   2   2   2   2   2 */
1162         /*              3  4  5  6  7  8   9   0   1   2   3   4   5   6   7   8   9   0   1   2   3   4   5   6   7   8 */
1163         i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld %lu %lu %ld %lu %lu %lu %lu",
1164              &state,          /* 3  %c  */
1165              &ppid,           /* 4  %d  */
1166              &pgrp,           /* 5  %d  */
1167              &session,        /* 6  %d  */
1168              &nr,             /* 7  %d  */
1169              &tpgrp,          /* 8  %d  */
1170              &flags,          /* 9  %lu  */
1171              &minflt,         /* 10 %lu  */
1172              &cminflt,        /* 11 %lu  */
1173              &majflt,         /* 12 %lu  */
1174              &cmajflt,        /* 13 %lu  */
1175              &utime,          /* 14 %lu  */
1176              &stime,          /* 15 %lu  */
1177              &cutime,         /* 16 %ld  */
1178              &cstime,         /* 17 %ld  */
1179              &prio,           /* 18 %ld  */
1180              &nice,           /* 19 %ld  */
1181              &junk,           /* 20 %ld  */
1182              &it_real,        /* 21 %ld  */
1183              &start,          /* 22 %lu  */
1184              &vsize,          /* 23 %lu  */
1185              &rss,            /* 24 %ld  */
1186              &rsslim,         /* 25 %lu  */
1187              &scodes,         /* 26 %lu  */
1188              &ecode,          /* 27 %lu  */
1189              &stack_start);   /* 28 %lu  */
1190       }
1191 
1192       if (i != 28 - 2) {
1193          assert(false, "Bad conversion from /proc/self/stat");
1194          // product mode - assume we are the initial thread, good luck in the
1195          // embedded case.
1196          warning("Can't detect initial thread stack location - bad conversion");
1197          stack_start = (uintptr_t) &rlim;
1198       }
1199     } else {
1200       // For some reason we can't open /proc/self/stat (for example, running on
1201       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1202       // most cases, so don't abort:
1203       warning("Can't detect initial thread stack location - no /proc/self/stat");
1204       stack_start = (uintptr_t) &rlim;
1205     }
1206   }
1207 
1208   // Now we have a pointer (stack_start) very close to the stack top, the
1209   // next thing to do is to figure out the exact location of stack top. We
1210   // can find out the virtual memory area that contains stack_start by
1211   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1212   // and its upper limit is the real stack top. (again, this would fail if
1213   // running inside chroot, because /proc may not exist.)
1214 
1215   uintptr_t stack_top;
1216   address low, high;
1217   if (find_vma((address)stack_start, &low, &high)) {
1218     // success, "high" is the true stack top. (ignore "low", because initial
1219     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1220     stack_top = (uintptr_t)high;
1221   } else {
1222     // failed, likely because /proc/self/maps does not exist
1223     warning("Can't detect initial thread stack location - find_vma failed");
1224     // best effort: stack_start is normally within a few pages below the real
1225     // stack top, use it as stack top, and reduce stack size so we won't put
1226     // guard page outside stack.
1227     stack_top = stack_start;
1228     stack_size -= 16 * page_size();
1229   }
1230 
1231   // stack_top could be partially down the page so align it
1232   stack_top = align_size_up(stack_top, page_size());
1233 
1234   if (max_size && stack_size > max_size) {
1235      _initial_thread_stack_size = max_size;
1236   } else {
1237      _initial_thread_stack_size = stack_size;
1238   }
1239 
1240   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1241   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1242 }
1243 
1244 ////////////////////////////////////////////////////////////////////////////////
1245 // time support
1246 
1247 // Time since start-up in seconds to a fine granularity.
1248 // Used by VMSelfDestructTimer and the MemProfiler.
1249 double os::elapsedTime() {
1250 
1251   return (double)(os::elapsed_counter()) * 0.000001;
1252 }
1253 
1254 jlong os::elapsed_counter() {
1255   timeval time;
1256   int status = gettimeofday(&time, NULL);
1257   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1258 }
1259 
1260 jlong os::elapsed_frequency() {
1261   return (1000 * 1000);
1262 }
1263 
1264 // For now, we say that linux does not support vtime.  I have no idea
1265 // whether it can actually be made to (DLD, 9/13/05).
1266 
1267 bool os::supports_vtime() { return false; }
1268 bool os::enable_vtime()   { return false; }
1269 bool os::vtime_enabled()  { return false; }
1270 double os::elapsedVTime() {
1271   // better than nothing, but not much
1272   return elapsedTime();
1273 }
1274 
1275 jlong os::javaTimeMillis() {
1276   timeval time;
1277   int status = gettimeofday(&time, NULL);
1278   assert(status != -1, "linux error");
1279   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1280 }
1281 
1282 #ifndef CLOCK_MONOTONIC
1283 #define CLOCK_MONOTONIC (1)
1284 #endif
1285 
1286 void os::Linux::clock_init() {
1287   // we do dlopen's in this particular order due to bug in linux
1288   // dynamical loader (see 6348968) leading to crash on exit
1289   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1290   if (handle == NULL) {
1291     handle = dlopen("librt.so", RTLD_LAZY);
1292   }
1293 
1294   if (handle) {
1295     int (*clock_getres_func)(clockid_t, struct timespec*) =
1296            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1297     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1298            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1299     if (clock_getres_func && clock_gettime_func) {
1300       // See if monotonic clock is supported by the kernel. Note that some
1301       // early implementations simply return kernel jiffies (updated every
1302       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1303       // for nano time (though the monotonic property is still nice to have).
1304       // It's fixed in newer kernels, however clock_getres() still returns
1305       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1306       // resolution for now. Hopefully as people move to new kernels, this
1307       // won't be a problem.
1308       struct timespec res;
1309       struct timespec tp;
1310       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1311           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1312         // yes, monotonic clock is supported
1313         _clock_gettime = clock_gettime_func;
1314       } else {
1315         // close librt if there is no monotonic clock
1316         dlclose(handle);
1317       }
1318     }
1319   }
1320 }
1321 
1322 #ifndef SYS_clock_getres
1323 
1324 #if defined(IA32) || defined(AMD64)
1325 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1326 #else
1327 #error Value of SYS_clock_getres not known on this platform
1328 #endif
1329 
1330 #endif
1331 
1332 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1333 
1334 void os::Linux::fast_thread_clock_init() {
1335   if (!UseLinuxPosixThreadCPUClocks) {
1336     return;
1337   }
1338   clockid_t clockid;
1339   struct timespec tp;
1340   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1341       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1342 
1343   // Switch to using fast clocks for thread cpu time if
1344   // the sys_clock_getres() returns 0 error code.
1345   // Note, that some kernels may support the current thread
1346   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1347   // returned by the pthread_getcpuclockid().
1348   // If the fast Posix clocks are supported then the sys_clock_getres()
1349   // must return at least tp.tv_sec == 0 which means a resolution
1350   // better than 1 sec. This is extra check for reliability.
1351 
1352   if(pthread_getcpuclockid_func &&
1353      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1354      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1355 
1356     _supports_fast_thread_cpu_time = true;
1357     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1358   }
1359 }
1360 
1361 jlong os::javaTimeNanos() {
1362   if (Linux::supports_monotonic_clock()) {
1363     struct timespec tp;
1364     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1365     assert(status == 0, "gettime error");
1366     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1367     return result;
1368   } else {
1369     timeval time;
1370     int status = gettimeofday(&time, NULL);
1371     assert(status != -1, "linux error");
1372     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1373     return 1000 * usecs;
1374   }
1375 }
1376 
1377 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1378   if (Linux::supports_monotonic_clock()) {
1379     info_ptr->max_value = ALL_64_BITS;
1380 
1381     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1382     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1383     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1384   } else {
1385     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1386     info_ptr->max_value = ALL_64_BITS;
1387 
1388     // gettimeofday is a real time clock so it skips
1389     info_ptr->may_skip_backward = true;
1390     info_ptr->may_skip_forward = true;
1391   }
1392 
1393   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1394 }
1395 
1396 // Return the real, user, and system times in seconds from an
1397 // arbitrary fixed point in the past.
1398 bool os::getTimesSecs(double* process_real_time,
1399                       double* process_user_time,
1400                       double* process_system_time) {
1401   struct tms ticks;
1402   clock_t real_ticks = times(&ticks);
1403 
1404   if (real_ticks == (clock_t) (-1)) {
1405     return false;
1406   } else {
1407     double ticks_per_second = (double) clock_tics_per_sec;
1408     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1409     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1410     *process_real_time = ((double) real_ticks) / ticks_per_second;
1411 
1412     return true;
1413   }
1414 }
1415 
1416 
1417 char * os::local_time_string(char *buf, size_t buflen) {
1418   struct tm t;
1419   time_t long_time;
1420   time(&long_time);
1421   localtime_r(&long_time, &t);
1422   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1423                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1424                t.tm_hour, t.tm_min, t.tm_sec);
1425   return buf;
1426 }
1427 
1428 ////////////////////////////////////////////////////////////////////////////////
1429 // runtime exit support
1430 
1431 // Note: os::shutdown() might be called very early during initialization, or
1432 // called from signal handler. Before adding something to os::shutdown(), make
1433 // sure it is async-safe and can handle partially initialized VM.
1434 void os::shutdown() {
1435 
1436   // allow PerfMemory to attempt cleanup of any persistent resources
1437   perfMemory_exit();
1438 
1439   // needs to remove object in file system
1440   AttachListener::abort();
1441 
1442   // flush buffered output, finish log files
1443   ostream_abort();
1444 
1445   // Check for abort hook
1446   abort_hook_t abort_hook = Arguments::abort_hook();
1447   if (abort_hook != NULL) {
1448     abort_hook();
1449   }
1450 
1451 }
1452 
1453 // Note: os::abort() might be called very early during initialization, or
1454 // called from signal handler. Before adding something to os::abort(), make
1455 // sure it is async-safe and can handle partially initialized VM.
1456 void os::abort(bool dump_core) {
1457   os::shutdown();
1458   if (dump_core) {
1459 #ifndef PRODUCT
1460     fdStream out(defaultStream::output_fd());
1461     out.print_raw("Current thread is ");
1462     char buf[16];
1463     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1464     out.print_raw_cr(buf);
1465     out.print_raw_cr("Dumping core ...");
1466 #endif
1467     ::abort(); // dump core
1468   }
1469 
1470   ::exit(1);
1471 }
1472 
1473 // Die immediately, no exit hook, no abort hook, no cleanup.
1474 void os::die() {
1475   // _exit() on LinuxThreads only kills current thread
1476   ::abort();
1477 }
1478 
1479 // unused on linux for now.
1480 void os::set_error_file(const char *logfile) {}
1481 
1482 intx os::current_thread_id() { return (intx)pthread_self(); }
1483 int os::current_process_id() {
1484 
1485   // Under the old linux thread library, linux gives each thread
1486   // its own process id. Because of this each thread will return
1487   // a different pid if this method were to return the result
1488   // of getpid(2). Linux provides no api that returns the pid
1489   // of the launcher thread for the vm. This implementation
1490   // returns a unique pid, the pid of the launcher thread
1491   // that starts the vm 'process'.
1492 
1493   // Under the NPTL, getpid() returns the same pid as the
1494   // launcher thread rather than a unique pid per thread.
1495   // Use gettid() if you want the old pre NPTL behaviour.
1496 
1497   // if you are looking for the result of a call to getpid() that
1498   // returns a unique pid for the calling thread, then look at the
1499   // OSThread::thread_id() method in osThread_linux.hpp file
1500 
1501   return (int)(_initial_pid ? _initial_pid : getpid());
1502 }
1503 
1504 // DLL functions
1505 
1506 const char* os::dll_file_extension() { return ".so"; }
1507 
1508 const char* os::get_temp_directory() { return "/tmp/"; }
1509 
1510 void os::dll_build_name(
1511     char* buffer, size_t buflen, const char* pname, const char* fname) {
1512   // copied from libhpi
1513   const size_t pnamelen = pname ? strlen(pname) : 0;
1514 
1515   /* Quietly truncate on buffer overflow.  Should be an error. */
1516   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1517       *buffer = '\0';
1518       return;
1519   }
1520 
1521   if (pnamelen == 0) {
1522       sprintf(buffer, "lib%s.so", fname);
1523   } else {
1524       sprintf(buffer, "%s/lib%s.so", pname, fname);
1525   }
1526 }
1527 
1528 const char* os::get_current_directory(char *buf, int buflen) {
1529   return getcwd(buf, buflen);
1530 }
1531 
1532 // check if addr is inside libjvm[_g].so
1533 bool os::address_is_in_vm(address addr) {
1534   static address libjvm_base_addr;
1535   Dl_info dlinfo;
1536 
1537   if (libjvm_base_addr == NULL) {
1538     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1539     libjvm_base_addr = (address)dlinfo.dli_fbase;
1540     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1541   }
1542 
1543   if (dladdr((void *)addr, &dlinfo)) {
1544     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1545   }
1546 
1547   return false;
1548 }
1549 
1550 bool os::dll_address_to_function_name(address addr, char *buf,
1551                                       int buflen, int *offset) {
1552   Dl_info dlinfo;
1553 
1554   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1555     if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1556     if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1557     return true;
1558   } else {
1559     if (buf) buf[0] = '\0';
1560     if (offset) *offset = -1;
1561     return false;
1562   }
1563 }
1564 
1565 struct _address_to_library_name {
1566   address addr;          // input : memory address
1567   size_t  buflen;        //         size of fname
1568   char*   fname;         // output: library name
1569   address base;          //         library base addr
1570 };
1571 
1572 static int address_to_library_name_callback(struct dl_phdr_info *info,
1573                                             size_t size, void *data) {
1574   int i;
1575   bool found = false;
1576   address libbase = NULL;
1577   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1578 
1579   // iterate through all loadable segments
1580   for (i = 0; i < info->dlpi_phnum; i++) {
1581     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1582     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1583       // base address of a library is the lowest address of its loaded
1584       // segments.
1585       if (libbase == NULL || libbase > segbase) {
1586         libbase = segbase;
1587       }
1588       // see if 'addr' is within current segment
1589       if (segbase <= d->addr &&
1590           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1591         found = true;
1592       }
1593     }
1594   }
1595 
1596   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1597   // so dll_address_to_library_name() can fall through to use dladdr() which
1598   // can figure out executable name from argv[0].
1599   if (found && info->dlpi_name && info->dlpi_name[0]) {
1600     d->base = libbase;
1601     if (d->fname) {
1602       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1603     }
1604     return 1;
1605   }
1606   return 0;
1607 }
1608 
1609 bool os::dll_address_to_library_name(address addr, char* buf,
1610                                      int buflen, int* offset) {
1611   Dl_info dlinfo;
1612   struct _address_to_library_name data;
1613 
1614   // There is a bug in old glibc dladdr() implementation that it could resolve
1615   // to wrong library name if the .so file has a base address != NULL. Here
1616   // we iterate through the program headers of all loaded libraries to find
1617   // out which library 'addr' really belongs to. This workaround can be
1618   // removed once the minimum requirement for glibc is moved to 2.3.x.
1619   data.addr = addr;
1620   data.fname = buf;
1621   data.buflen = buflen;
1622   data.base = NULL;
1623   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1624 
1625   if (rslt) {
1626      // buf already contains library name
1627      if (offset) *offset = addr - data.base;
1628      return true;
1629   } else if (dladdr((void*)addr, &dlinfo)){
1630      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1631      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1632      return true;
1633   } else {
1634      if (buf) buf[0] = '\0';
1635      if (offset) *offset = -1;
1636      return false;
1637   }
1638 }
1639 
1640   // Loads .dll/.so and
1641   // in case of error it checks if .dll/.so was built for the
1642   // same architecture as Hotspot is running on
1643 
1644 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1645 {
1646   void * result= ::dlopen(filename, RTLD_LAZY);
1647   if (result != NULL) {
1648     // Successful loading
1649     return result;
1650   }
1651 
1652   Elf32_Ehdr elf_head;
1653 
1654   // Read system error message into ebuf
1655   // It may or may not be overwritten below
1656   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1657   ebuf[ebuflen-1]='\0';
1658   int diag_msg_max_length=ebuflen-strlen(ebuf);
1659   char* diag_msg_buf=ebuf+strlen(ebuf);
1660 
1661   if (diag_msg_max_length==0) {
1662     // No more space in ebuf for additional diagnostics message
1663     return NULL;
1664   }
1665 
1666 
1667   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1668 
1669   if (file_descriptor < 0) {
1670     // Can't open library, report dlerror() message
1671     return NULL;
1672   }
1673 
1674   bool failed_to_read_elf_head=
1675     (sizeof(elf_head)!=
1676         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1677 
1678   ::close(file_descriptor);
1679   if (failed_to_read_elf_head) {
1680     // file i/o error - report dlerror() msg
1681     return NULL;
1682   }
1683 
1684   typedef struct {
1685     Elf32_Half  code;         // Actual value as defined in elf.h
1686     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1687     char        elf_class;    // 32 or 64 bit
1688     char        endianess;    // MSB or LSB
1689     char*       name;         // String representation
1690   } arch_t;
1691 
1692   #ifndef EM_486
1693   #define EM_486          6               /* Intel 80486 */
1694   #endif
1695 
1696   static const arch_t arch_array[]={
1697     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1698     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1699     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1700     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1701     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1702     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1703     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1704     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1705     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}
1706   };
1707 
1708   #if  (defined IA32)
1709     static  Elf32_Half running_arch_code=EM_386;
1710   #elif   (defined AMD64)
1711     static  Elf32_Half running_arch_code=EM_X86_64;
1712   #elif  (defined IA64)
1713     static  Elf32_Half running_arch_code=EM_IA_64;
1714   #elif  (defined __sparc) && (defined _LP64)
1715     static  Elf32_Half running_arch_code=EM_SPARCV9;
1716   #elif  (defined __sparc) && (!defined _LP64)
1717     static  Elf32_Half running_arch_code=EM_SPARC;
1718   #elif  (defined __powerpc64__)
1719     static  Elf32_Half running_arch_code=EM_PPC64;
1720   #elif  (defined __powerpc__)
1721     static  Elf32_Half running_arch_code=EM_PPC;
1722   #else
1723     #error Method os::dll_load requires that one of following is defined:\
1724          IA32, AMD64, IA64, __sparc, __powerpc__
1725   #endif
1726 
1727   // Identify compatability class for VM's architecture and library's architecture
1728   // Obtain string descriptions for architectures
1729 
1730   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1731   int running_arch_index=-1;
1732 
1733   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1734     if (running_arch_code == arch_array[i].code) {
1735       running_arch_index    = i;
1736     }
1737     if (lib_arch.code == arch_array[i].code) {
1738       lib_arch.compat_class = arch_array[i].compat_class;
1739       lib_arch.name         = arch_array[i].name;
1740     }
1741   }
1742 
1743   assert(running_arch_index != -1,
1744     "Didn't find running architecture code (running_arch_code) in arch_array");
1745   if (running_arch_index == -1) {
1746     // Even though running architecture detection failed
1747     // we may still continue with reporting dlerror() message
1748     return NULL;
1749   }
1750 
1751   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1752     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1753     return NULL;
1754   }
1755 
1756   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1757     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1758     return NULL;
1759   }
1760 
1761   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1762     if ( lib_arch.name!=NULL ) {
1763       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1764         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1765         lib_arch.name, arch_array[running_arch_index].name);
1766     } else {
1767       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1768       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1769         lib_arch.code,
1770         arch_array[running_arch_index].name);
1771     }
1772   }
1773 
1774   return NULL;
1775 }
1776 
1777 /*
1778  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
1779  * chances are you might want to run the generated bits against glibc-2.0
1780  * libdl.so, so always use locking for any version of glibc.
1781  */
1782 void* os::dll_lookup(void* handle, const char* name) {
1783   pthread_mutex_lock(&dl_mutex);
1784   void* res = dlsym(handle, name);
1785   pthread_mutex_unlock(&dl_mutex);
1786   return res;
1787 }
1788 
1789 
1790 bool _print_ascii_file(const char* filename, outputStream* st) {
1791   int fd = open(filename, O_RDONLY);
1792   if (fd == -1) {
1793      return false;
1794   }
1795 
1796   char buf[32];
1797   int bytes;
1798   while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1799     st->print_raw(buf, bytes);
1800   }
1801 
1802   close(fd);
1803 
1804   return true;
1805 }
1806 
1807 void os::print_dll_info(outputStream *st) {
1808    st->print_cr("Dynamic libraries:");
1809 
1810    char fname[32];
1811    pid_t pid = os::Linux::gettid();
1812 
1813    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1814 
1815    if (!_print_ascii_file(fname, st)) {
1816      st->print("Can not get library information for pid = %d\n", pid);
1817    }
1818 }
1819 
1820 
1821 void os::print_os_info(outputStream* st) {
1822   st->print("OS:");
1823 
1824   // Try to identify popular distros.
1825   // Most Linux distributions have /etc/XXX-release file, which contains
1826   // the OS version string. Some have more than one /etc/XXX-release file
1827   // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1828   // so the order is important.
1829   if (!_print_ascii_file("/etc/mandrake-release", st) &&
1830       !_print_ascii_file("/etc/sun-release", st) &&
1831       !_print_ascii_file("/etc/redhat-release", st) &&
1832       !_print_ascii_file("/etc/SuSE-release", st) &&
1833       !_print_ascii_file("/etc/turbolinux-release", st) &&
1834       !_print_ascii_file("/etc/gentoo-release", st) &&
1835       !_print_ascii_file("/etc/debian_version", st)) {
1836       st->print("Linux");
1837   }
1838   st->cr();
1839 
1840   // kernel
1841   st->print("uname:");
1842   struct utsname name;
1843   uname(&name);
1844   st->print(name.sysname); st->print(" ");
1845   st->print(name.release); st->print(" ");
1846   st->print(name.version); st->print(" ");
1847   st->print(name.machine);
1848   st->cr();
1849 
1850   // Print warning if unsafe chroot environment detected
1851   if (unsafe_chroot_detected) {
1852     st->print("WARNING!! ");
1853     st->print_cr(unstable_chroot_error);
1854   }
1855 
1856   // libc, pthread
1857   st->print("libc:");
1858   st->print(os::Linux::glibc_version()); st->print(" ");
1859   st->print(os::Linux::libpthread_version()); st->print(" ");
1860   if (os::Linux::is_LinuxThreads()) {
1861      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
1862   }
1863   st->cr();
1864 
1865   // rlimit
1866   st->print("rlimit:");
1867   struct rlimit rlim;
1868 
1869   st->print(" STACK ");
1870   getrlimit(RLIMIT_STACK, &rlim);
1871   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1872   else st->print("%uk", rlim.rlim_cur >> 10);
1873 
1874   st->print(", CORE ");
1875   getrlimit(RLIMIT_CORE, &rlim);
1876   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1877   else st->print("%uk", rlim.rlim_cur >> 10);
1878 
1879   st->print(", NPROC ");
1880   getrlimit(RLIMIT_NPROC, &rlim);
1881   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1882   else st->print("%d", rlim.rlim_cur);
1883 
1884   st->print(", NOFILE ");
1885   getrlimit(RLIMIT_NOFILE, &rlim);
1886   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1887   else st->print("%d", rlim.rlim_cur);
1888 
1889   st->print(", AS ");
1890   getrlimit(RLIMIT_AS, &rlim);
1891   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1892   else st->print("%uk", rlim.rlim_cur >> 10);
1893   st->cr();
1894 
1895   // load average
1896   st->print("load average:");
1897   double loadavg[3];
1898   os::loadavg(loadavg, 3);
1899   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1900   st->cr();
1901 }
1902 
1903 void os::print_memory_info(outputStream* st) {
1904 
1905   st->print("Memory:");
1906   st->print(" %dk page", os::vm_page_size()>>10);
1907 
1908   // values in struct sysinfo are "unsigned long"
1909   struct sysinfo si;
1910   sysinfo(&si);
1911 
1912   st->print(", physical " UINT64_FORMAT "k",
1913             os::physical_memory() >> 10);
1914   st->print("(" UINT64_FORMAT "k free)",
1915             os::available_memory() >> 10);
1916   st->print(", swap " UINT64_FORMAT "k",
1917             ((jlong)si.totalswap * si.mem_unit) >> 10);
1918   st->print("(" UINT64_FORMAT "k free)",
1919             ((jlong)si.freeswap * si.mem_unit) >> 10);
1920   st->cr();
1921 }
1922 
1923 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
1924 // but they're the same for all the linux arch that we support
1925 // and they're the same for solaris but there's no common place to put this.
1926 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
1927                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
1928                           "ILL_COPROC", "ILL_BADSTK" };
1929 
1930 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
1931                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
1932                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
1933 
1934 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
1935 
1936 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
1937 
1938 void os::print_siginfo(outputStream* st, void* siginfo) {
1939   st->print("siginfo:");
1940 
1941   const int buflen = 100;
1942   char buf[buflen];
1943   siginfo_t *si = (siginfo_t*)siginfo;
1944   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
1945   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
1946     st->print("si_errno=%s", buf);
1947   } else {
1948     st->print("si_errno=%d", si->si_errno);
1949   }
1950   const int c = si->si_code;
1951   assert(c > 0, "unexpected si_code");
1952   switch (si->si_signo) {
1953   case SIGILL:
1954     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
1955     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1956     break;
1957   case SIGFPE:
1958     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
1959     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1960     break;
1961   case SIGSEGV:
1962     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
1963     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1964     break;
1965   case SIGBUS:
1966     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
1967     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1968     break;
1969   default:
1970     st->print(", si_code=%d", si->si_code);
1971     // no si_addr
1972   }
1973 
1974   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
1975       UseSharedSpaces) {
1976     FileMapInfo* mapinfo = FileMapInfo::current_info();
1977     if (mapinfo->is_in_shared_space(si->si_addr)) {
1978       st->print("\n\nError accessing class data sharing archive."   \
1979                 " Mapped file inaccessible during execution, "      \
1980                 " possible disk/network problem.");
1981     }
1982   }
1983   st->cr();
1984 }
1985 
1986 
1987 static void print_signal_handler(outputStream* st, int sig,
1988                                  char* buf, size_t buflen);
1989 
1990 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1991   st->print_cr("Signal Handlers:");
1992   print_signal_handler(st, SIGSEGV, buf, buflen);
1993   print_signal_handler(st, SIGBUS , buf, buflen);
1994   print_signal_handler(st, SIGFPE , buf, buflen);
1995   print_signal_handler(st, SIGPIPE, buf, buflen);
1996   print_signal_handler(st, SIGXFSZ, buf, buflen);
1997   print_signal_handler(st, SIGILL , buf, buflen);
1998   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
1999   print_signal_handler(st, SR_signum, buf, buflen);
2000   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2001   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2002   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2003   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2004 }
2005 
2006 static char saved_jvm_path[MAXPATHLEN] = {0};
2007 
2008 // Find the full path to the current module, libjvm.so or libjvm_g.so
2009 void os::jvm_path(char *buf, jint len) {
2010   // Error checking.
2011   if (len < MAXPATHLEN) {
2012     assert(false, "must use a large-enough buffer");
2013     buf[0] = '\0';
2014     return;
2015   }
2016   // Lazy resolve the path to current module.
2017   if (saved_jvm_path[0] != 0) {
2018     strcpy(buf, saved_jvm_path);
2019     return;
2020   }
2021 
2022   char dli_fname[MAXPATHLEN];
2023   bool ret = dll_address_to_library_name(
2024                 CAST_FROM_FN_PTR(address, os::jvm_path),
2025                 dli_fname, sizeof(dli_fname), NULL);
2026   assert(ret != 0, "cannot locate libjvm");
2027   realpath(dli_fname, buf);
2028 
2029   if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2030     // Support for the gamma launcher.  Typical value for buf is
2031     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2032     // the right place in the string, then assume we are installed in a JDK and
2033     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2034     // up the path so it looks like libjvm.so is installed there (append a
2035     // fake suffix hotspot/libjvm.so).
2036     const char *p = buf + strlen(buf) - 1;
2037     for (int count = 0; p > buf && count < 5; ++count) {
2038       for (--p; p > buf && *p != '/'; --p)
2039         /* empty */ ;
2040     }
2041 
2042     if (strncmp(p, "/jre/lib/", 9) != 0) {
2043       // Look for JAVA_HOME in the environment.
2044       char* java_home_var = ::getenv("JAVA_HOME");
2045       if (java_home_var != NULL && java_home_var[0] != 0) {
2046         // Check the current module name "libjvm.so" or "libjvm_g.so".
2047         p = strrchr(buf, '/');
2048         assert(strstr(p, "/libjvm") == p, "invalid library name");
2049         p = strstr(p, "_g") ? "_g" : "";
2050 
2051         realpath(java_home_var, buf);
2052         sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch);
2053         if (0 == access(buf, F_OK)) {
2054           // Use current module name "libjvm[_g].so" instead of
2055           // "libjvm"debug_only("_g")".so" since for fastdebug version
2056           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2057           // It is used when we are choosing the HPI library's name
2058           // "libhpi[_g].so" in hpi::initialize_get_interface().
2059           sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p);
2060         } else {
2061           // Go back to path of .so
2062           realpath(dli_fname, buf);
2063         }
2064       }
2065     }
2066   }
2067 
2068   strcpy(saved_jvm_path, buf);
2069 }
2070 
2071 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2072   // no prefix required, not even "_"
2073 }
2074 
2075 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2076   // no suffix required
2077 }
2078 
2079 ////////////////////////////////////////////////////////////////////////////////
2080 // sun.misc.Signal support
2081 
2082 static volatile jint sigint_count = 0;
2083 
2084 static void
2085 UserHandler(int sig, void *siginfo, void *context) {
2086   // 4511530 - sem_post is serialized and handled by the manager thread. When
2087   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2088   // don't want to flood the manager thread with sem_post requests.
2089   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2090       return;
2091 
2092   // Ctrl-C is pressed during error reporting, likely because the error
2093   // handler fails to abort. Let VM die immediately.
2094   if (sig == SIGINT && is_error_reported()) {
2095      os::die();
2096   }
2097 
2098   os::signal_notify(sig);
2099 }
2100 
2101 void* os::user_handler() {
2102   return CAST_FROM_FN_PTR(void*, UserHandler);
2103 }
2104 
2105 extern "C" {
2106   typedef void (*sa_handler_t)(int);
2107   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2108 }
2109 
2110 void* os::signal(int signal_number, void* handler) {
2111   struct sigaction sigAct, oldSigAct;
2112 
2113   sigfillset(&(sigAct.sa_mask));
2114   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2115   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2116 
2117   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2118     // -1 means registration failed
2119     return (void *)-1;
2120   }
2121 
2122   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2123 }
2124 
2125 void os::signal_raise(int signal_number) {
2126   ::raise(signal_number);
2127 }
2128 
2129 /*
2130  * The following code is moved from os.cpp for making this
2131  * code platform specific, which it is by its very nature.
2132  */
2133 
2134 // Will be modified when max signal is changed to be dynamic
2135 int os::sigexitnum_pd() {
2136   return NSIG;
2137 }
2138 
2139 // a counter for each possible signal value
2140 static volatile jint pending_signals[NSIG+1] = { 0 };
2141 
2142 // Linux(POSIX) specific hand shaking semaphore.
2143 static sem_t sig_sem;
2144 
2145 void os::signal_init_pd() {
2146   // Initialize signal structures
2147   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2148 
2149   // Initialize signal semaphore
2150   ::sem_init(&sig_sem, 0, 0);
2151 }
2152 
2153 void os::signal_notify(int sig) {
2154   Atomic::inc(&pending_signals[sig]);
2155   ::sem_post(&sig_sem);
2156 }
2157 
2158 static int check_pending_signals(bool wait) {
2159   Atomic::store(0, &sigint_count);
2160   for (;;) {
2161     for (int i = 0; i < NSIG + 1; i++) {
2162       jint n = pending_signals[i];
2163       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2164         return i;
2165       }
2166     }
2167     if (!wait) {
2168       return -1;
2169     }
2170     JavaThread *thread = JavaThread::current();
2171     ThreadBlockInVM tbivm(thread);
2172 
2173     bool threadIsSuspended;
2174     do {
2175       thread->set_suspend_equivalent();
2176       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2177       ::sem_wait(&sig_sem);
2178 
2179       // were we externally suspended while we were waiting?
2180       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2181       if (threadIsSuspended) {
2182         //
2183         // The semaphore has been incremented, but while we were waiting
2184         // another thread suspended us. We don't want to continue running
2185         // while suspended because that would surprise the thread that
2186         // suspended us.
2187         //
2188         ::sem_post(&sig_sem);
2189 
2190         thread->java_suspend_self();
2191       }
2192     } while (threadIsSuspended);
2193   }
2194 }
2195 
2196 int os::signal_lookup() {
2197   return check_pending_signals(false);
2198 }
2199 
2200 int os::signal_wait() {
2201   return check_pending_signals(true);
2202 }
2203 
2204 ////////////////////////////////////////////////////////////////////////////////
2205 // Virtual Memory
2206 
2207 int os::vm_page_size() {
2208   // Seems redundant as all get out
2209   assert(os::Linux::page_size() != -1, "must call os::init");
2210   return os::Linux::page_size();
2211 }
2212 
2213 // Solaris allocates memory by pages.
2214 int os::vm_allocation_granularity() {
2215   assert(os::Linux::page_size() != -1, "must call os::init");
2216   return os::Linux::page_size();
2217 }
2218 
2219 // Rationale behind this function:
2220 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2221 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2222 //  samples for JITted code. Here we create private executable mapping over the code cache
2223 //  and then we can use standard (well, almost, as mapping can change) way to provide
2224 //  info for the reporting script by storing timestamp and location of symbol
2225 void linux_wrap_code(char* base, size_t size) {
2226   static volatile jint cnt = 0;
2227 
2228   if (!UseOprofile) {
2229     return;
2230   }
2231 
2232   char buf[40];
2233   int num = Atomic::add(1, &cnt);
2234 
2235   sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num);
2236   unlink(buf);
2237 
2238   int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2239 
2240   if (fd != -1) {
2241     off_t rv = lseek(fd, size-2, SEEK_SET);
2242     if (rv != (off_t)-1) {
2243       if (write(fd, "", 1) == 1) {
2244         mmap(base, size,
2245              PROT_READ|PROT_WRITE|PROT_EXEC,
2246              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2247       }
2248     }
2249     close(fd);
2250     unlink(buf);
2251   }
2252 }
2253 
2254 // NOTE: Linux kernel does not really reserve the pages for us.
2255 //       All it does is to check if there are enough free pages
2256 //       left at the time of mmap(). This could be a potential
2257 //       problem.
2258 bool os::commit_memory(char* addr, size_t size) {
2259   uintptr_t res = (uintptr_t) ::mmap(addr, size,
2260                                    PROT_READ|PROT_WRITE|PROT_EXEC,
2261                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2262   return res != (uintptr_t) MAP_FAILED;
2263 }
2264 
2265 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint) {
2266   return commit_memory(addr, size);
2267 }
2268 
2269 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2270 
2271 void os::free_memory(char *addr, size_t bytes) {
2272   uncommit_memory(addr, bytes);
2273 }
2274 
2275 void os::numa_make_global(char *addr, size_t bytes)    { }
2276 
2277 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2278   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2279 }
2280 
2281 bool os::numa_topology_changed()   { return false; }
2282 
2283 size_t os::numa_get_groups_num() {
2284   int max_node = Linux::numa_max_node();
2285   return max_node > 0 ? max_node + 1 : 1;
2286 }
2287 
2288 int os::numa_get_group_id() {
2289   int cpu_id = Linux::sched_getcpu();
2290   if (cpu_id != -1) {
2291     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2292     if (lgrp_id != -1) {
2293       return lgrp_id;
2294     }
2295   }
2296   return 0;
2297 }
2298 
2299 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2300   for (size_t i = 0; i < size; i++) {
2301     ids[i] = i;
2302   }
2303   return size;
2304 }
2305 
2306 bool os::get_page_info(char *start, page_info* info) {
2307   return false;
2308 }
2309 
2310 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2311   return end;
2312 }
2313 
2314 extern "C" void numa_warn(int number, char *where, ...) { }
2315 extern "C" void numa_error(char *where) { }
2316 
2317 void os::Linux::libnuma_init() {
2318   // sched_getcpu() should be in libc.
2319   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2320                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2321 
2322   if (sched_getcpu() != -1) { // Does it work?
2323     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2324     if (handle != NULL) {
2325       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2326                                            dlsym(handle, "numa_node_to_cpus")));
2327       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2328                                        dlsym(handle, "numa_max_node")));
2329       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2330                                         dlsym(handle, "numa_available")));
2331       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2332                                             dlsym(handle, "numa_tonode_memory")));
2333       if (numa_available() != -1) {
2334         // Create a cpu -> node mapping
2335         _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2336         rebuild_cpu_to_node_map();
2337       }
2338     }
2339   }
2340 }
2341 
2342 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2343 // The table is later used in get_node_by_cpu().
2344 void os::Linux::rebuild_cpu_to_node_map() {
2345   int cpu_num = os::active_processor_count();
2346   cpu_to_node()->clear();
2347   cpu_to_node()->at_grow(cpu_num - 1);
2348   int node_num = numa_get_groups_num();
2349   int cpu_map_size = (cpu_num + BitsPerLong - 1) / BitsPerLong;
2350   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2351   for (int i = 0; i < node_num; i++) {
2352     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2353       for (int j = 0; j < cpu_map_size; j++) {
2354         if (cpu_map[j] != 0) {
2355           for (int k = 0; k < BitsPerLong; k++) {
2356             if (cpu_map[j] & (1UL << k)) {
2357               cpu_to_node()->at_put(j * BitsPerLong + k, i);
2358             }
2359           }
2360         }
2361       }
2362     }
2363   }
2364   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2365 }
2366 
2367 int os::Linux::get_node_by_cpu(int cpu_id) {
2368   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2369     return cpu_to_node()->at(cpu_id);
2370   }
2371   return -1;
2372 }
2373 
2374 GrowableArray<int>* os::Linux::_cpu_to_node;
2375 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2376 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2377 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2378 os::Linux::numa_available_func_t os::Linux::_numa_available;
2379 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2380 
2381 
2382 bool os::uncommit_memory(char* addr, size_t size) {
2383   return ::mmap(addr, size,
2384                 PROT_READ|PROT_WRITE|PROT_EXEC,
2385                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0)
2386     != MAP_FAILED;
2387 }
2388 
2389 static address _highest_vm_reserved_address = NULL;
2390 
2391 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2392 // at 'requested_addr'. If there are existing memory mappings at the same
2393 // location, however, they will be overwritten. If 'fixed' is false,
2394 // 'requested_addr' is only treated as a hint, the return value may or
2395 // may not start from the requested address. Unlike Linux mmap(), this
2396 // function returns NULL to indicate failure.
2397 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2398   char * addr;
2399   int flags;
2400 
2401   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2402   if (fixed) {
2403     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2404     flags |= MAP_FIXED;
2405   }
2406 
2407   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE|PROT_EXEC,
2408                        flags, -1, 0);
2409 
2410   if (addr != MAP_FAILED) {
2411     // anon_mmap() should only get called during VM initialization,
2412     // don't need lock (actually we can skip locking even it can be called
2413     // from multiple threads, because _highest_vm_reserved_address is just a
2414     // hint about the upper limit of non-stack memory regions.)
2415     if ((address)addr + bytes > _highest_vm_reserved_address) {
2416       _highest_vm_reserved_address = (address)addr + bytes;
2417     }
2418   }
2419 
2420   return addr == MAP_FAILED ? NULL : addr;
2421 }
2422 
2423 // Don't update _highest_vm_reserved_address, because there might be memory
2424 // regions above addr + size. If so, releasing a memory region only creates
2425 // a hole in the address space, it doesn't help prevent heap-stack collision.
2426 //
2427 static int anon_munmap(char * addr, size_t size) {
2428   return ::munmap(addr, size) == 0;
2429 }
2430 
2431 char* os::reserve_memory(size_t bytes, char* requested_addr,
2432                          size_t alignment_hint) {
2433   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2434 }
2435 
2436 bool os::release_memory(char* addr, size_t size) {
2437   return anon_munmap(addr, size);
2438 }
2439 
2440 static address highest_vm_reserved_address() {
2441   return _highest_vm_reserved_address;
2442 }
2443 
2444 static bool linux_mprotect(char* addr, size_t size, int prot) {
2445   // Linux wants the mprotect address argument to be page aligned.
2446   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2447 
2448   // According to SUSv3, mprotect() should only be used with mappings
2449   // established by mmap(), and mmap() always maps whole pages. Unaligned
2450   // 'addr' likely indicates problem in the VM (e.g. trying to change
2451   // protection of malloc'ed or statically allocated memory). Check the
2452   // caller if you hit this assert.
2453   assert(addr == bottom, "sanity check");
2454 
2455   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2456   return ::mprotect(bottom, size, prot) == 0;
2457 }
2458 
2459 // Set protections specified
2460 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2461                         bool is_committed) {
2462   unsigned int p = 0;
2463   switch (prot) {
2464   case MEM_PROT_NONE: p = PROT_NONE; break;
2465   case MEM_PROT_READ: p = PROT_READ; break;
2466   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2467   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2468   default:
2469     ShouldNotReachHere();
2470   }
2471   // is_committed is unused.
2472   return linux_mprotect(addr, bytes, p);
2473 }
2474 
2475 bool os::guard_memory(char* addr, size_t size) {
2476   return linux_mprotect(addr, size, PROT_NONE);
2477 }
2478 
2479 bool os::unguard_memory(char* addr, size_t size) {
2480   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE|PROT_EXEC);
2481 }
2482 
2483 // Large page support
2484 
2485 static size_t _large_page_size = 0;
2486 
2487 bool os::large_page_init() {
2488   if (!UseLargePages) return false;
2489 
2490   if (LargePageSizeInBytes) {
2491     _large_page_size = LargePageSizeInBytes;
2492   } else {
2493     // large_page_size on Linux is used to round up heap size. x86 uses either
2494     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2495     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2496     // page as large as 256M.
2497     //
2498     // Here we try to figure out page size by parsing /proc/meminfo and looking
2499     // for a line with the following format:
2500     //    Hugepagesize:     2048 kB
2501     //
2502     // If we can't determine the value (e.g. /proc is not mounted, or the text
2503     // format has been changed), we'll use the largest page size supported by
2504     // the processor.
2505 
2506     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M);
2507 
2508     FILE *fp = fopen("/proc/meminfo", "r");
2509     if (fp) {
2510       while (!feof(fp)) {
2511         int x = 0;
2512         char buf[16];
2513         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2514           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2515             _large_page_size = x * K;
2516             break;
2517           }
2518         } else {
2519           // skip to next line
2520           for (;;) {
2521             int ch = fgetc(fp);
2522             if (ch == EOF || ch == (int)'\n') break;
2523           }
2524         }
2525       }
2526       fclose(fp);
2527     }
2528   }
2529 
2530   const size_t default_page_size = (size_t)Linux::page_size();
2531   if (_large_page_size > default_page_size) {
2532     _page_sizes[0] = _large_page_size;
2533     _page_sizes[1] = default_page_size;
2534     _page_sizes[2] = 0;
2535   }
2536 
2537   // Large page support is available on 2.6 or newer kernel, some vendors
2538   // (e.g. Redhat) have backported it to their 2.4 based distributions.
2539   // We optimistically assume the support is available. If later it turns out
2540   // not true, VM will automatically switch to use regular page size.
2541   return true;
2542 }
2543 
2544 #ifndef SHM_HUGETLB
2545 #define SHM_HUGETLB 04000
2546 #endif
2547 
2548 char* os::reserve_memory_special(size_t bytes) {
2549   assert(UseLargePages, "only for large pages");
2550 
2551   key_t key = IPC_PRIVATE;
2552   char *addr;
2553 
2554   bool warn_on_failure = UseLargePages &&
2555                         (!FLAG_IS_DEFAULT(UseLargePages) ||
2556                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2557                         );
2558   char msg[128];
2559 
2560   // Create a large shared memory region to attach to based on size.
2561   // Currently, size is the total size of the heap
2562   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2563   if (shmid == -1) {
2564      // Possible reasons for shmget failure:
2565      // 1. shmmax is too small for Java heap.
2566      //    > check shmmax value: cat /proc/sys/kernel/shmmax
2567      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2568      // 2. not enough large page memory.
2569      //    > check available large pages: cat /proc/meminfo
2570      //    > increase amount of large pages:
2571      //          echo new_value > /proc/sys/vm/nr_hugepages
2572      //      Note 1: different Linux may use different name for this property,
2573      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
2574      //      Note 2: it's possible there's enough physical memory available but
2575      //            they are so fragmented after a long run that they can't
2576      //            coalesce into large pages. Try to reserve large pages when
2577      //            the system is still "fresh".
2578      if (warn_on_failure) {
2579        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2580        warning(msg);
2581      }
2582      return NULL;
2583   }
2584 
2585   // attach to the region
2586   addr = (char*)shmat(shmid, NULL, 0);
2587   int err = errno;
2588 
2589   // Remove shmid. If shmat() is successful, the actual shared memory segment
2590   // will be deleted when it's detached by shmdt() or when the process
2591   // terminates. If shmat() is not successful this will remove the shared
2592   // segment immediately.
2593   shmctl(shmid, IPC_RMID, NULL);
2594 
2595   if ((intptr_t)addr == -1) {
2596      if (warn_on_failure) {
2597        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2598        warning(msg);
2599      }
2600      return NULL;
2601   }
2602 
2603   return addr;
2604 }
2605 
2606 bool os::release_memory_special(char* base, size_t bytes) {
2607   // detaching the SHM segment will also delete it, see reserve_memory_special()
2608   int rslt = shmdt(base);
2609   return rslt == 0;
2610 }
2611 
2612 size_t os::large_page_size() {
2613   return _large_page_size;
2614 }
2615 
2616 // Linux does not support anonymous mmap with large page memory. The only way
2617 // to reserve large page memory without file backing is through SysV shared
2618 // memory API. The entire memory region is committed and pinned upfront.
2619 // Hopefully this will change in the future...
2620 bool os::can_commit_large_page_memory() {
2621   return false;
2622 }
2623 
2624 bool os::can_execute_large_page_memory() {
2625   return false;
2626 }
2627 
2628 // Reserve memory at an arbitrary address, only if that area is
2629 // available (and not reserved for something else).
2630 
2631 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2632   const int max_tries = 10;
2633   char* base[max_tries];
2634   size_t size[max_tries];
2635   const size_t gap = 0x000000;
2636 
2637   // Assert only that the size is a multiple of the page size, since
2638   // that's all that mmap requires, and since that's all we really know
2639   // about at this low abstraction level.  If we need higher alignment,
2640   // we can either pass an alignment to this method or verify alignment
2641   // in one of the methods further up the call chain.  See bug 5044738.
2642   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2643 
2644   // Repeatedly allocate blocks until the block is allocated at the
2645   // right spot. Give up after max_tries. Note that reserve_memory() will
2646   // automatically update _highest_vm_reserved_address if the call is
2647   // successful. The variable tracks the highest memory address every reserved
2648   // by JVM. It is used to detect heap-stack collision if running with
2649   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2650   // space than needed, it could confuse the collision detecting code. To
2651   // solve the problem, save current _highest_vm_reserved_address and
2652   // calculate the correct value before return.
2653   address old_highest = _highest_vm_reserved_address;
2654 
2655   // Linux mmap allows caller to pass an address as hint; give it a try first,
2656   // if kernel honors the hint then we can return immediately.
2657   char * addr = anon_mmap(requested_addr, bytes, false);
2658   if (addr == requested_addr) {
2659      return requested_addr;
2660   }
2661 
2662   if (addr != NULL) {
2663      // mmap() is successful but it fails to reserve at the requested address
2664      anon_munmap(addr, bytes);
2665   }
2666 
2667   int i;
2668   for (i = 0; i < max_tries; ++i) {
2669     base[i] = reserve_memory(bytes);
2670 
2671     if (base[i] != NULL) {
2672       // Is this the block we wanted?
2673       if (base[i] == requested_addr) {
2674         size[i] = bytes;
2675         break;
2676       }
2677 
2678       // Does this overlap the block we wanted? Give back the overlapped
2679       // parts and try again.
2680 
2681       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2682       if (top_overlap >= 0 && top_overlap < bytes) {
2683         unmap_memory(base[i], top_overlap);
2684         base[i] += top_overlap;
2685         size[i] = bytes - top_overlap;
2686       } else {
2687         size_t bottom_overlap = base[i] + bytes - requested_addr;
2688         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2689           unmap_memory(requested_addr, bottom_overlap);
2690           size[i] = bytes - bottom_overlap;
2691         } else {
2692           size[i] = bytes;
2693         }
2694       }
2695     }
2696   }
2697 
2698   // Give back the unused reserved pieces.
2699 
2700   for (int j = 0; j < i; ++j) {
2701     if (base[j] != NULL) {
2702       unmap_memory(base[j], size[j]);
2703     }
2704   }
2705 
2706   if (i < max_tries) {
2707     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2708     return requested_addr;
2709   } else {
2710     _highest_vm_reserved_address = old_highest;
2711     return NULL;
2712   }
2713 }
2714 
2715 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2716   return ::read(fd, buf, nBytes);
2717 }
2718 
2719 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
2720 // Solaris uses poll(), linux uses park().
2721 // Poll() is likely a better choice, assuming that Thread.interrupt()
2722 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2723 // SIGSEGV, see 4355769.
2724 
2725 const int NANOSECS_PER_MILLISECS = 1000000;
2726 
2727 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2728   assert(thread == Thread::current(),  "thread consistency check");
2729 
2730   ParkEvent * const slp = thread->_SleepEvent ;
2731   slp->reset() ;
2732   OrderAccess::fence() ;
2733 
2734   if (interruptible) {
2735     jlong prevtime = javaTimeNanos();
2736 
2737     for (;;) {
2738       if (os::is_interrupted(thread, true)) {
2739         return OS_INTRPT;
2740       }
2741 
2742       jlong newtime = javaTimeNanos();
2743 
2744       if (newtime - prevtime < 0) {
2745         // time moving backwards, should only happen if no monotonic clock
2746         // not a guarantee() because JVM should not abort on kernel/glibc bugs
2747         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2748       } else {
2749         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2750       }
2751 
2752       if(millis <= 0) {
2753         return OS_OK;
2754       }
2755 
2756       prevtime = newtime;
2757 
2758       {
2759         assert(thread->is_Java_thread(), "sanity check");
2760         JavaThread *jt = (JavaThread *) thread;
2761         ThreadBlockInVM tbivm(jt);
2762         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
2763 
2764         jt->set_suspend_equivalent();
2765         // cleared by handle_special_suspend_equivalent_condition() or
2766         // java_suspend_self() via check_and_wait_while_suspended()
2767 
2768         slp->park(millis);
2769 
2770         // were we externally suspended while we were waiting?
2771         jt->check_and_wait_while_suspended();
2772       }
2773     }
2774   } else {
2775     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2776     jlong prevtime = javaTimeNanos();
2777 
2778     for (;;) {
2779       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
2780       // the 1st iteration ...
2781       jlong newtime = javaTimeNanos();
2782 
2783       if (newtime - prevtime < 0) {
2784         // time moving backwards, should only happen if no monotonic clock
2785         // not a guarantee() because JVM should not abort on kernel/glibc bugs
2786         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2787       } else {
2788         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2789       }
2790 
2791       if(millis <= 0) break ;
2792 
2793       prevtime = newtime;
2794       slp->park(millis);
2795     }
2796     return OS_OK ;
2797   }
2798 }
2799 
2800 int os::naked_sleep() {
2801   // %% make the sleep time an integer flag. for now use 1 millisec.
2802   return os::sleep(Thread::current(), 1, false);
2803 }
2804 
2805 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2806 void os::infinite_sleep() {
2807   while (true) {    // sleep forever ...
2808     ::sleep(100);   // ... 100 seconds at a time
2809   }
2810 }
2811 
2812 // Used to convert frequent JVM_Yield() to nops
2813 bool os::dont_yield() {
2814   return DontYieldALot;
2815 }
2816 
2817 void os::yield() {
2818   sched_yield();
2819 }
2820 
2821 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
2822 
2823 void os::yield_all(int attempts) {
2824   // Yields to all threads, including threads with lower priorities
2825   // Threads on Linux are all with same priority. The Solaris style
2826   // os::yield_all() with nanosleep(1ms) is not necessary.
2827   sched_yield();
2828 }
2829 
2830 // Called from the tight loops to possibly influence time-sharing heuristics
2831 void os::loop_breaker(int attempts) {
2832   os::yield_all(attempts);
2833 }
2834 
2835 ////////////////////////////////////////////////////////////////////////////////
2836 // thread priority support
2837 
2838 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
2839 // only supports dynamic priority, static priority must be zero. For real-time
2840 // applications, Linux supports SCHED_RR which allows static priority (1-99).
2841 // However, for large multi-threaded applications, SCHED_RR is not only slower
2842 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
2843 // of 5 runs - Sep 2005).
2844 //
2845 // The following code actually changes the niceness of kernel-thread/LWP. It
2846 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
2847 // not the entire user process, and user level threads are 1:1 mapped to kernel
2848 // threads. It has always been the case, but could change in the future. For
2849 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
2850 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
2851 
2852 int os::java_to_os_priority[MaxPriority + 1] = {
2853   19,              // 0 Entry should never be used
2854 
2855    4,              // 1 MinPriority
2856    3,              // 2
2857    2,              // 3
2858 
2859    1,              // 4
2860    0,              // 5 NormPriority
2861   -1,              // 6
2862 
2863   -2,              // 7
2864   -3,              // 8
2865   -4,              // 9 NearMaxPriority
2866 
2867   -5               // 10 MaxPriority
2868 };
2869 
2870 static int prio_init() {
2871   if (ThreadPriorityPolicy == 1) {
2872     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
2873     // if effective uid is not root. Perhaps, a more elegant way of doing
2874     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
2875     if (geteuid() != 0) {
2876       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
2877         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
2878       }
2879       ThreadPriorityPolicy = 0;
2880     }
2881   }
2882   return 0;
2883 }
2884 
2885 OSReturn os::set_native_priority(Thread* thread, int newpri) {
2886   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
2887 
2888   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
2889   return (ret == 0) ? OS_OK : OS_ERR;
2890 }
2891 
2892 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
2893   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
2894     *priority_ptr = java_to_os_priority[NormPriority];
2895     return OS_OK;
2896   }
2897 
2898   errno = 0;
2899   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
2900   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
2901 }
2902 
2903 // Hint to the underlying OS that a task switch would not be good.
2904 // Void return because it's a hint and can fail.
2905 void os::hint_no_preempt() {}
2906 
2907 ////////////////////////////////////////////////////////////////////////////////
2908 // suspend/resume support
2909 
2910 //  the low-level signal-based suspend/resume support is a remnant from the
2911 //  old VM-suspension that used to be for java-suspension, safepoints etc,
2912 //  within hotspot. Now there is a single use-case for this:
2913 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
2914 //      that runs in the watcher thread.
2915 //  The remaining code is greatly simplified from the more general suspension
2916 //  code that used to be used.
2917 //
2918 //  The protocol is quite simple:
2919 //  - suspend:
2920 //      - sends a signal to the target thread
2921 //      - polls the suspend state of the osthread using a yield loop
2922 //      - target thread signal handler (SR_handler) sets suspend state
2923 //        and blocks in sigsuspend until continued
2924 //  - resume:
2925 //      - sets target osthread state to continue
2926 //      - sends signal to end the sigsuspend loop in the SR_handler
2927 //
2928 //  Note that the SR_lock plays no role in this suspend/resume protocol.
2929 //
2930 
2931 static void resume_clear_context(OSThread *osthread) {
2932   osthread->set_ucontext(NULL);
2933   osthread->set_siginfo(NULL);
2934 
2935   // notify the suspend action is completed, we have now resumed
2936   osthread->sr.clear_suspended();
2937 }
2938 
2939 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
2940   osthread->set_ucontext(context);
2941   osthread->set_siginfo(siginfo);
2942 }
2943 
2944 //
2945 // Handler function invoked when a thread's execution is suspended or
2946 // resumed. We have to be careful that only async-safe functions are
2947 // called here (Note: most pthread functions are not async safe and
2948 // should be avoided.)
2949 //
2950 // Note: sigwait() is a more natural fit than sigsuspend() from an
2951 // interface point of view, but sigwait() prevents the signal hander
2952 // from being run. libpthread would get very confused by not having
2953 // its signal handlers run and prevents sigwait()'s use with the
2954 // mutex granting granting signal.
2955 //
2956 // Currently only ever called on the VMThread
2957 //
2958 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
2959   // Save and restore errno to avoid confusing native code with EINTR
2960   // after sigsuspend.
2961   int old_errno = errno;
2962 
2963   Thread* thread = Thread::current();
2964   OSThread* osthread = thread->osthread();
2965   assert(thread->is_VM_thread(), "Must be VMThread");
2966   // read current suspend action
2967   int action = osthread->sr.suspend_action();
2968   if (action == SR_SUSPEND) {
2969     suspend_save_context(osthread, siginfo, context);
2970 
2971     // Notify the suspend action is about to be completed. do_suspend()
2972     // waits until SR_SUSPENDED is set and then returns. We will wait
2973     // here for a resume signal and that completes the suspend-other
2974     // action. do_suspend/do_resume is always called as a pair from
2975     // the same thread - so there are no races
2976 
2977     // notify the caller
2978     osthread->sr.set_suspended();
2979 
2980     sigset_t suspend_set;  // signals for sigsuspend()
2981 
2982     // get current set of blocked signals and unblock resume signal
2983     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
2984     sigdelset(&suspend_set, SR_signum);
2985 
2986     // wait here until we are resumed
2987     do {
2988       sigsuspend(&suspend_set);
2989       // ignore all returns until we get a resume signal
2990     } while (osthread->sr.suspend_action() != SR_CONTINUE);
2991 
2992     resume_clear_context(osthread);
2993 
2994   } else {
2995     assert(action == SR_CONTINUE, "unexpected sr action");
2996     // nothing special to do - just leave the handler
2997   }
2998 
2999   errno = old_errno;
3000 }
3001 
3002 
3003 static int SR_initialize() {
3004   struct sigaction act;
3005   char *s;
3006   /* Get signal number to use for suspend/resume */
3007   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3008     int sig = ::strtol(s, 0, 10);
3009     if (sig > 0 || sig < _NSIG) {
3010         SR_signum = sig;
3011     }
3012   }
3013 
3014   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3015         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3016 
3017   sigemptyset(&SR_sigset);
3018   sigaddset(&SR_sigset, SR_signum);
3019 
3020   /* Set up signal handler for suspend/resume */
3021   act.sa_flags = SA_RESTART|SA_SIGINFO;
3022   act.sa_handler = (void (*)(int)) SR_handler;
3023 
3024   // SR_signum is blocked by default.
3025   // 4528190 - We also need to block pthread restart signal (32 on all
3026   // supported Linux platforms). Note that LinuxThreads need to block
3027   // this signal for all threads to work properly. So we don't have
3028   // to use hard-coded signal number when setting up the mask.
3029   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3030 
3031   if (sigaction(SR_signum, &act, 0) == -1) {
3032     return -1;
3033   }
3034 
3035   // Save signal flag
3036   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3037   return 0;
3038 }
3039 
3040 static int SR_finalize() {
3041   return 0;
3042 }
3043 
3044 
3045 // returns true on success and false on error - really an error is fatal
3046 // but this seems the normal response to library errors
3047 static bool do_suspend(OSThread* osthread) {
3048   // mark as suspended and send signal
3049   osthread->sr.set_suspend_action(SR_SUSPEND);
3050   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3051   assert_status(status == 0, status, "pthread_kill");
3052 
3053   // check status and wait until notified of suspension
3054   if (status == 0) {
3055     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3056       os::yield_all(i);
3057     }
3058     osthread->sr.set_suspend_action(SR_NONE);
3059     return true;
3060   }
3061   else {
3062     osthread->sr.set_suspend_action(SR_NONE);
3063     return false;
3064   }
3065 }
3066 
3067 static void do_resume(OSThread* osthread) {
3068   assert(osthread->sr.is_suspended(), "thread should be suspended");
3069   osthread->sr.set_suspend_action(SR_CONTINUE);
3070 
3071   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3072   assert_status(status == 0, status, "pthread_kill");
3073   // check status and wait unit notified of resumption
3074   if (status == 0) {
3075     for (int i = 0; osthread->sr.is_suspended(); i++) {
3076       os::yield_all(i);
3077     }
3078   }
3079   osthread->sr.set_suspend_action(SR_NONE);
3080 }
3081 
3082 ////////////////////////////////////////////////////////////////////////////////
3083 // interrupt support
3084 
3085 void os::interrupt(Thread* thread) {
3086   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3087     "possibility of dangling Thread pointer");
3088 
3089   OSThread* osthread = thread->osthread();
3090 
3091   if (!osthread->interrupted()) {
3092     osthread->set_interrupted(true);
3093     // More than one thread can get here with the same value of osthread,
3094     // resulting in multiple notifications.  We do, however, want the store
3095     // to interrupted() to be visible to other threads before we execute unpark().
3096     OrderAccess::fence();
3097     ParkEvent * const slp = thread->_SleepEvent ;
3098     if (slp != NULL) slp->unpark() ;
3099   }
3100 
3101   // For JSR166. Unpark even if interrupt status already was set
3102   if (thread->is_Java_thread())
3103     ((JavaThread*)thread)->parker()->unpark();
3104 
3105   ParkEvent * ev = thread->_ParkEvent ;
3106   if (ev != NULL) ev->unpark() ;
3107 
3108 }
3109 
3110 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3111   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3112     "possibility of dangling Thread pointer");
3113 
3114   OSThread* osthread = thread->osthread();
3115 
3116   bool interrupted = osthread->interrupted();
3117 
3118   if (interrupted && clear_interrupted) {
3119     osthread->set_interrupted(false);
3120     // consider thread->_SleepEvent->reset() ... optional optimization
3121   }
3122 
3123   return interrupted;
3124 }
3125 
3126 ///////////////////////////////////////////////////////////////////////////////////
3127 // signal handling (except suspend/resume)
3128 
3129 // This routine may be used by user applications as a "hook" to catch signals.
3130 // The user-defined signal handler must pass unrecognized signals to this
3131 // routine, and if it returns true (non-zero), then the signal handler must
3132 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3133 // routine will never retun false (zero), but instead will execute a VM panic
3134 // routine kill the process.
3135 //
3136 // If this routine returns false, it is OK to call it again.  This allows
3137 // the user-defined signal handler to perform checks either before or after
3138 // the VM performs its own checks.  Naturally, the user code would be making
3139 // a serious error if it tried to handle an exception (such as a null check
3140 // or breakpoint) that the VM was generating for its own correct operation.
3141 //
3142 // This routine may recognize any of the following kinds of signals:
3143 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3144 // It should be consulted by handlers for any of those signals.
3145 //
3146 // The caller of this routine must pass in the three arguments supplied
3147 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3148 // field of the structure passed to sigaction().  This routine assumes that
3149 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3150 //
3151 // Note that the VM will print warnings if it detects conflicting signal
3152 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3153 //
3154 extern "C" int
3155 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3156                         void* ucontext, int abort_if_unrecognized);
3157 
3158 void signalHandler(int sig, siginfo_t* info, void* uc) {
3159   assert(info != NULL && uc != NULL, "it must be old kernel");
3160   JVM_handle_linux_signal(sig, info, uc, true);
3161 }
3162 
3163 
3164 // This boolean allows users to forward their own non-matching signals
3165 // to JVM_handle_linux_signal, harmlessly.
3166 bool os::Linux::signal_handlers_are_installed = false;
3167 
3168 // For signal-chaining
3169 struct sigaction os::Linux::sigact[MAXSIGNUM];
3170 unsigned int os::Linux::sigs = 0;
3171 bool os::Linux::libjsig_is_loaded = false;
3172 typedef struct sigaction *(*get_signal_t)(int);
3173 get_signal_t os::Linux::get_signal_action = NULL;
3174 
3175 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3176   struct sigaction *actp = NULL;
3177 
3178   if (libjsig_is_loaded) {
3179     // Retrieve the old signal handler from libjsig
3180     actp = (*get_signal_action)(sig);
3181   }
3182   if (actp == NULL) {
3183     // Retrieve the preinstalled signal handler from jvm
3184     actp = get_preinstalled_handler(sig);
3185   }
3186 
3187   return actp;
3188 }
3189 
3190 static bool call_chained_handler(struct sigaction *actp, int sig,
3191                                  siginfo_t *siginfo, void *context) {
3192   // Call the old signal handler
3193   if (actp->sa_handler == SIG_DFL) {
3194     // It's more reasonable to let jvm treat it as an unexpected exception
3195     // instead of taking the default action.
3196     return false;
3197   } else if (actp->sa_handler != SIG_IGN) {
3198     if ((actp->sa_flags & SA_NODEFER) == 0) {
3199       // automaticlly block the signal
3200       sigaddset(&(actp->sa_mask), sig);
3201     }
3202 
3203     sa_handler_t hand;
3204     sa_sigaction_t sa;
3205     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3206     // retrieve the chained handler
3207     if (siginfo_flag_set) {
3208       sa = actp->sa_sigaction;
3209     } else {
3210       hand = actp->sa_handler;
3211     }
3212 
3213     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3214       actp->sa_handler = SIG_DFL;
3215     }
3216 
3217     // try to honor the signal mask
3218     sigset_t oset;
3219     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3220 
3221     // call into the chained handler
3222     if (siginfo_flag_set) {
3223       (*sa)(sig, siginfo, context);
3224     } else {
3225       (*hand)(sig);
3226     }
3227 
3228     // restore the signal mask
3229     pthread_sigmask(SIG_SETMASK, &oset, 0);
3230   }
3231   // Tell jvm's signal handler the signal is taken care of.
3232   return true;
3233 }
3234 
3235 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3236   bool chained = false;
3237   // signal-chaining
3238   if (UseSignalChaining) {
3239     struct sigaction *actp = get_chained_signal_action(sig);
3240     if (actp != NULL) {
3241       chained = call_chained_handler(actp, sig, siginfo, context);
3242     }
3243   }
3244   return chained;
3245 }
3246 
3247 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3248   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3249     return &sigact[sig];
3250   }
3251   return NULL;
3252 }
3253 
3254 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3255   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3256   sigact[sig] = oldAct;
3257   sigs |= (unsigned int)1 << sig;
3258 }
3259 
3260 // for diagnostic
3261 int os::Linux::sigflags[MAXSIGNUM];
3262 
3263 int os::Linux::get_our_sigflags(int sig) {
3264   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3265   return sigflags[sig];
3266 }
3267 
3268 void os::Linux::set_our_sigflags(int sig, int flags) {
3269   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3270   sigflags[sig] = flags;
3271 }
3272 
3273 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3274   // Check for overwrite.
3275   struct sigaction oldAct;
3276   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3277 
3278   void* oldhand = oldAct.sa_sigaction
3279                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3280                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3281   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3282       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3283       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3284     if (AllowUserSignalHandlers || !set_installed) {
3285       // Do not overwrite; user takes responsibility to forward to us.
3286       return;
3287     } else if (UseSignalChaining) {
3288       // save the old handler in jvm
3289       save_preinstalled_handler(sig, oldAct);
3290       // libjsig also interposes the sigaction() call below and saves the
3291       // old sigaction on it own.
3292     } else {
3293       fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig);
3294     }
3295   }
3296 
3297   struct sigaction sigAct;
3298   sigfillset(&(sigAct.sa_mask));
3299   sigAct.sa_handler = SIG_DFL;
3300   if (!set_installed) {
3301     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3302   } else {
3303     sigAct.sa_sigaction = signalHandler;
3304     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3305   }
3306   // Save flags, which are set by ours
3307   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3308   sigflags[sig] = sigAct.sa_flags;
3309 
3310   int ret = sigaction(sig, &sigAct, &oldAct);
3311   assert(ret == 0, "check");
3312 
3313   void* oldhand2  = oldAct.sa_sigaction
3314                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3315                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3316   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3317 }
3318 
3319 // install signal handlers for signals that HotSpot needs to
3320 // handle in order to support Java-level exception handling.
3321 
3322 void os::Linux::install_signal_handlers() {
3323   if (!signal_handlers_are_installed) {
3324     signal_handlers_are_installed = true;
3325 
3326     // signal-chaining
3327     typedef void (*signal_setting_t)();
3328     signal_setting_t begin_signal_setting = NULL;
3329     signal_setting_t end_signal_setting = NULL;
3330     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3331                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3332     if (begin_signal_setting != NULL) {
3333       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3334                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3335       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3336                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3337       libjsig_is_loaded = true;
3338       assert(UseSignalChaining, "should enable signal-chaining");
3339     }
3340     if (libjsig_is_loaded) {
3341       // Tell libjsig jvm is setting signal handlers
3342       (*begin_signal_setting)();
3343     }
3344 
3345     set_signal_handler(SIGSEGV, true);
3346     set_signal_handler(SIGPIPE, true);
3347     set_signal_handler(SIGBUS, true);
3348     set_signal_handler(SIGILL, true);
3349     set_signal_handler(SIGFPE, true);
3350     set_signal_handler(SIGXFSZ, true);
3351 
3352     if (libjsig_is_loaded) {
3353       // Tell libjsig jvm finishes setting signal handlers
3354       (*end_signal_setting)();
3355     }
3356 
3357     // We don't activate signal checker if libjsig is in place, we trust ourselves
3358     // and if UserSignalHandler is installed all bets are off
3359     if (CheckJNICalls) {
3360       if (libjsig_is_loaded) {
3361         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3362         check_signals = false;
3363       }
3364       if (AllowUserSignalHandlers) {
3365         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3366         check_signals = false;
3367       }
3368     }
3369   }
3370 }
3371 
3372 // This is the fastest way to get thread cpu time on Linux.
3373 // Returns cpu time (user+sys) for any thread, not only for current.
3374 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3375 // It might work on 2.6.10+ with a special kernel/glibc patch.
3376 // For reference, please, see IEEE Std 1003.1-2004:
3377 //   http://www.unix.org/single_unix_specification
3378 
3379 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3380   struct timespec tp;
3381   int rc = os::Linux::clock_gettime(clockid, &tp);
3382   assert(rc == 0, "clock_gettime is expected to return 0 code");
3383 
3384   return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3385 }
3386 
3387 /////
3388 // glibc on Linux platform uses non-documented flag
3389 // to indicate, that some special sort of signal
3390 // trampoline is used.
3391 // We will never set this flag, and we should
3392 // ignore this flag in our diagnostic
3393 #ifdef SIGNIFICANT_SIGNAL_MASK
3394 #undef SIGNIFICANT_SIGNAL_MASK
3395 #endif
3396 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3397 
3398 static const char* get_signal_handler_name(address handler,
3399                                            char* buf, int buflen) {
3400   int offset;
3401   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3402   if (found) {
3403     // skip directory names
3404     const char *p1, *p2;
3405     p1 = buf;
3406     size_t len = strlen(os::file_separator());
3407     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3408     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3409   } else {
3410     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3411   }
3412   return buf;
3413 }
3414 
3415 static void print_signal_handler(outputStream* st, int sig,
3416                                  char* buf, size_t buflen) {
3417   struct sigaction sa;
3418 
3419   sigaction(sig, NULL, &sa);
3420 
3421   // See comment for SIGNIFICANT_SIGNAL_MASK define
3422   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3423 
3424   st->print("%s: ", os::exception_name(sig, buf, buflen));
3425 
3426   address handler = (sa.sa_flags & SA_SIGINFO)
3427     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3428     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3429 
3430   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3431     st->print("SIG_DFL");
3432   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3433     st->print("SIG_IGN");
3434   } else {
3435     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3436   }
3437 
3438   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3439 
3440   address rh = VMError::get_resetted_sighandler(sig);
3441   // May be, handler was resetted by VMError?
3442   if(rh != NULL) {
3443     handler = rh;
3444     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3445   }
3446 
3447   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
3448 
3449   // Check: is it our handler?
3450   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3451      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3452     // It is our signal handler
3453     // check for flags, reset system-used one!
3454     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3455       st->print(
3456                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3457                 os::Linux::get_our_sigflags(sig));
3458     }
3459   }
3460   st->cr();
3461 }
3462 
3463 
3464 #define DO_SIGNAL_CHECK(sig) \
3465   if (!sigismember(&check_signal_done, sig)) \
3466     os::Linux::check_signal_handler(sig)
3467 
3468 // This method is a periodic task to check for misbehaving JNI applications
3469 // under CheckJNI, we can add any periodic checks here
3470 
3471 void os::run_periodic_checks() {
3472 
3473   if (check_signals == false) return;
3474 
3475   // SEGV and BUS if overridden could potentially prevent
3476   // generation of hs*.log in the event of a crash, debugging
3477   // such a case can be very challenging, so we absolutely
3478   // check the following for a good measure:
3479   DO_SIGNAL_CHECK(SIGSEGV);
3480   DO_SIGNAL_CHECK(SIGILL);
3481   DO_SIGNAL_CHECK(SIGFPE);
3482   DO_SIGNAL_CHECK(SIGBUS);
3483   DO_SIGNAL_CHECK(SIGPIPE);
3484   DO_SIGNAL_CHECK(SIGXFSZ);
3485 
3486 
3487   // ReduceSignalUsage allows the user to override these handlers
3488   // see comments at the very top and jvm_solaris.h
3489   if (!ReduceSignalUsage) {
3490     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3491     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3492     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3493     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3494   }
3495 
3496   DO_SIGNAL_CHECK(SR_signum);
3497   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3498 }
3499 
3500 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3501 
3502 static os_sigaction_t os_sigaction = NULL;
3503 
3504 void os::Linux::check_signal_handler(int sig) {
3505   char buf[O_BUFLEN];
3506   address jvmHandler = NULL;
3507 
3508 
3509   struct sigaction act;
3510   if (os_sigaction == NULL) {
3511     // only trust the default sigaction, in case it has been interposed
3512     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3513     if (os_sigaction == NULL) return;
3514   }
3515 
3516   os_sigaction(sig, (struct sigaction*)NULL, &act);
3517 
3518 
3519   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3520 
3521   address thisHandler = (act.sa_flags & SA_SIGINFO)
3522     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3523     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3524 
3525 
3526   switch(sig) {
3527   case SIGSEGV:
3528   case SIGBUS:
3529   case SIGFPE:
3530   case SIGPIPE:
3531   case SIGILL:
3532   case SIGXFSZ:
3533     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3534     break;
3535 
3536   case SHUTDOWN1_SIGNAL:
3537   case SHUTDOWN2_SIGNAL:
3538   case SHUTDOWN3_SIGNAL:
3539   case BREAK_SIGNAL:
3540     jvmHandler = (address)user_handler();
3541     break;
3542 
3543   case INTERRUPT_SIGNAL:
3544     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3545     break;
3546 
3547   default:
3548     if (sig == SR_signum) {
3549       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3550     } else {
3551       return;
3552     }
3553     break;
3554   }
3555 
3556   if (thisHandler != jvmHandler) {
3557     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3558     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3559     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3560     // No need to check this sig any longer
3561     sigaddset(&check_signal_done, sig);
3562   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3563     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3564     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3565     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
3566     // No need to check this sig any longer
3567     sigaddset(&check_signal_done, sig);
3568   }
3569 
3570   // Dump all the signal
3571   if (sigismember(&check_signal_done, sig)) {
3572     print_signal_handlers(tty, buf, O_BUFLEN);
3573   }
3574 }
3575 
3576 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3577 
3578 extern bool signal_name(int signo, char* buf, size_t len);
3579 
3580 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3581   if (0 < exception_code && exception_code <= SIGRTMAX) {
3582     // signal
3583     if (!signal_name(exception_code, buf, size)) {
3584       jio_snprintf(buf, size, "SIG%d", exception_code);
3585     }
3586     return buf;
3587   } else {
3588     return NULL;
3589   }
3590 }
3591 
3592 // this is called _before_ the most of global arguments have been parsed
3593 void os::init(void) {
3594   char dummy;   /* used to get a guess on initial stack address */
3595 //  first_hrtime = gethrtime();
3596 
3597   // With LinuxThreads the JavaMain thread pid (primordial thread)
3598   // is different than the pid of the java launcher thread.
3599   // So, on Linux, the launcher thread pid is passed to the VM
3600   // via the sun.java.launcher.pid property.
3601   // Use this property instead of getpid() if it was correctly passed.
3602   // See bug 6351349.
3603   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3604 
3605   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3606 
3607   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3608 
3609   init_random(1234567);
3610 
3611   ThreadCritical::initialize();
3612 
3613   Linux::set_page_size(sysconf(_SC_PAGESIZE));
3614   if (Linux::page_size() == -1) {
3615     fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno));
3616   }
3617   init_page_sizes((size_t) Linux::page_size());
3618 
3619   Linux::initialize_system_info();
3620 
3621   // main_thread points to the aboriginal thread
3622   Linux::_main_thread = pthread_self();
3623 
3624   Linux::clock_init();
3625   initial_time_count = os::elapsed_counter();
3626   pthread_mutex_init(&dl_mutex, NULL);
3627 }
3628 
3629 // To install functions for atexit system call
3630 extern "C" {
3631   static void perfMemory_exit_helper() {
3632     perfMemory_exit();
3633   }
3634 }
3635 
3636 // this is called _after_ the global arguments have been parsed
3637 jint os::init_2(void)
3638 {
3639   Linux::fast_thread_clock_init();
3640 
3641   // Allocate a single page and mark it as readable for safepoint polling
3642   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3643   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3644 
3645   os::set_polling_page( polling_page );
3646 
3647 #ifndef PRODUCT
3648   if(Verbose && PrintMiscellaneous)
3649     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3650 #endif
3651 
3652   if (!UseMembar) {
3653     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3654     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3655     os::set_memory_serialize_page( mem_serialize_page );
3656 
3657 #ifndef PRODUCT
3658     if(Verbose && PrintMiscellaneous)
3659       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3660 #endif
3661   }
3662 
3663   FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3664 
3665   // initialize suspend/resume support - must do this before signal_sets_init()
3666   if (SR_initialize() != 0) {
3667     perror("SR_initialize failed");
3668     return JNI_ERR;
3669   }
3670 
3671   Linux::signal_sets_init();
3672   Linux::install_signal_handlers();
3673 
3674   size_t threadStackSizeInBytes = ThreadStackSize * K;
3675   if (threadStackSizeInBytes != 0 &&
3676       threadStackSizeInBytes < Linux::min_stack_allowed) {
3677         tty->print_cr("\nThe stack size specified is too small, "
3678                       "Specify at least %dk",
3679                       Linux::min_stack_allowed / K);
3680         return JNI_ERR;
3681   }
3682 
3683   // Make the stack size a multiple of the page size so that
3684   // the yellow/red zones can be guarded.
3685   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3686         vm_page_size()));
3687 
3688   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
3689 
3690   Linux::libpthread_init();
3691   if (PrintMiscellaneous && (Verbose || WizardMode)) {
3692      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
3693           Linux::glibc_version(), Linux::libpthread_version(),
3694           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
3695   }
3696 
3697   if (UseNUMA) {
3698     Linux::libnuma_init();
3699   }
3700 
3701   if (MaxFDLimit) {
3702     // set the number of file descriptors to max. print out error
3703     // if getrlimit/setrlimit fails but continue regardless.
3704     struct rlimit nbr_files;
3705     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3706     if (status != 0) {
3707       if (PrintMiscellaneous && (Verbose || WizardMode))
3708         perror("os::init_2 getrlimit failed");
3709     } else {
3710       nbr_files.rlim_cur = nbr_files.rlim_max;
3711       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3712       if (status != 0) {
3713         if (PrintMiscellaneous && (Verbose || WizardMode))
3714           perror("os::init_2 setrlimit failed");
3715       }
3716     }
3717   }
3718 
3719   // Initialize lock used to serialize thread creation (see os::create_thread)
3720   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
3721 
3722   // Initialize HPI.
3723   jint hpi_result = hpi::initialize();
3724   if (hpi_result != JNI_OK) {
3725     tty->print_cr("There was an error trying to initialize the HPI library.");
3726     return hpi_result;
3727   }
3728 
3729   // at-exit methods are called in the reverse order of their registration.
3730   // atexit functions are called on return from main or as a result of a
3731   // call to exit(3C). There can be only 32 of these functions registered
3732   // and atexit() does not set errno.
3733 
3734   if (PerfAllowAtExitRegistration) {
3735     // only register atexit functions if PerfAllowAtExitRegistration is set.
3736     // atexit functions can be delayed until process exit time, which
3737     // can be problematic for embedded VM situations. Embedded VMs should
3738     // call DestroyJavaVM() to assure that VM resources are released.
3739 
3740     // note: perfMemory_exit_helper atexit function may be removed in
3741     // the future if the appropriate cleanup code can be added to the
3742     // VM_Exit VMOperation's doit method.
3743     if (atexit(perfMemory_exit_helper) != 0) {
3744       warning("os::init2 atexit(perfMemory_exit_helper) failed");
3745     }
3746   }
3747 
3748   // initialize thread priority policy
3749   prio_init();
3750 
3751   return JNI_OK;
3752 }
3753 
3754 // Mark the polling page as unreadable
3755 void os::make_polling_page_unreadable(void) {
3756   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
3757     fatal("Could not disable polling page");
3758 };
3759 
3760 // Mark the polling page as readable
3761 void os::make_polling_page_readable(void) {
3762   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
3763     fatal("Could not enable polling page");
3764   }
3765 };
3766 
3767 int os::active_processor_count() {
3768   // Linux doesn't yet have a (official) notion of processor sets,
3769   // so just return the number of online processors.
3770   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
3771   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
3772   return online_cpus;
3773 }
3774 
3775 bool os::distribute_processes(uint length, uint* distribution) {
3776   // Not yet implemented.
3777   return false;
3778 }
3779 
3780 bool os::bind_to_processor(uint processor_id) {
3781   // Not yet implemented.
3782   return false;
3783 }
3784 
3785 ///
3786 
3787 // Suspends the target using the signal mechanism and then grabs the PC before
3788 // resuming the target. Used by the flat-profiler only
3789 ExtendedPC os::get_thread_pc(Thread* thread) {
3790   // Make sure that it is called by the watcher for the VMThread
3791   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
3792   assert(thread->is_VM_thread(), "Can only be called for VMThread");
3793 
3794   ExtendedPC epc;
3795 
3796   OSThread* osthread = thread->osthread();
3797   if (do_suspend(osthread)) {
3798     if (osthread->ucontext() != NULL) {
3799       epc = os::Linux::ucontext_get_pc(osthread->ucontext());
3800     } else {
3801       // NULL context is unexpected, double-check this is the VMThread
3802       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3803     }
3804     do_resume(osthread);
3805   }
3806   // failure means pthread_kill failed for some reason - arguably this is
3807   // a fatal problem, but such problems are ignored elsewhere
3808 
3809   return epc;
3810 }
3811 
3812 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
3813 {
3814    if (is_NPTL()) {
3815       return pthread_cond_timedwait(_cond, _mutex, _abstime);
3816    } else {
3817 #ifndef IA64
3818       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
3819       // word back to default 64bit precision if condvar is signaled. Java
3820       // wants 53bit precision.  Save and restore current value.
3821       int fpu = get_fpu_control_word();
3822 #endif // IA64
3823       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
3824 #ifndef IA64
3825       set_fpu_control_word(fpu);
3826 #endif // IA64
3827       return status;
3828    }
3829 }
3830 
3831 ////////////////////////////////////////////////////////////////////////////////
3832 // debug support
3833 
3834 #ifndef PRODUCT
3835 static address same_page(address x, address y) {
3836   int page_bits = -os::vm_page_size();
3837   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
3838     return x;
3839   else if (x > y)
3840     return (address)(intptr_t(y) | ~page_bits) + 1;
3841   else
3842     return (address)(intptr_t(y) & page_bits);
3843 }
3844 
3845 bool os::find(address addr) {
3846   Dl_info dlinfo;
3847   memset(&dlinfo, 0, sizeof(dlinfo));
3848   if (dladdr(addr, &dlinfo)) {
3849     tty->print(PTR_FORMAT ": ", addr);
3850     if (dlinfo.dli_sname != NULL) {
3851       tty->print("%s+%#x", dlinfo.dli_sname,
3852                  addr - (intptr_t)dlinfo.dli_saddr);
3853     } else if (dlinfo.dli_fname) {
3854       tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
3855     } else {
3856       tty->print("<absolute address>");
3857     }
3858     if (dlinfo.dli_fname) {
3859       tty->print(" in %s", dlinfo.dli_fname);
3860     }
3861     if (dlinfo.dli_fbase) {
3862       tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
3863     }
3864     tty->cr();
3865 
3866     if (Verbose) {
3867       // decode some bytes around the PC
3868       address begin = same_page(addr-40, addr);
3869       address end   = same_page(addr+40, addr);
3870       address       lowest = (address) dlinfo.dli_sname;
3871       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
3872       if (begin < lowest)  begin = lowest;
3873       Dl_info dlinfo2;
3874       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
3875           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
3876         end = (address) dlinfo2.dli_saddr;
3877       Disassembler::decode(begin, end);
3878     }
3879     return true;
3880   }
3881   return false;
3882 }
3883 
3884 #endif
3885 
3886 ////////////////////////////////////////////////////////////////////////////////
3887 // misc
3888 
3889 // This does not do anything on Linux. This is basically a hook for being
3890 // able to use structured exception handling (thread-local exception filters)
3891 // on, e.g., Win32.
3892 void
3893 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
3894                          JavaCallArguments* args, Thread* thread) {
3895   f(value, method, args, thread);
3896 }
3897 
3898 void os::print_statistics() {
3899 }
3900 
3901 int os::message_box(const char* title, const char* message) {
3902   int i;
3903   fdStream err(defaultStream::error_fd());
3904   for (i = 0; i < 78; i++) err.print_raw("=");
3905   err.cr();
3906   err.print_raw_cr(title);
3907   for (i = 0; i < 78; i++) err.print_raw("-");
3908   err.cr();
3909   err.print_raw_cr(message);
3910   for (i = 0; i < 78; i++) err.print_raw("=");
3911   err.cr();
3912 
3913   char buf[16];
3914   // Prevent process from exiting upon "read error" without consuming all CPU
3915   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3916 
3917   return buf[0] == 'y' || buf[0] == 'Y';
3918 }
3919 
3920 int os::stat(const char *path, struct stat *sbuf) {
3921   char pathbuf[MAX_PATH];
3922   if (strlen(path) > MAX_PATH - 1) {
3923     errno = ENAMETOOLONG;
3924     return -1;
3925   }
3926   hpi::native_path(strcpy(pathbuf, path));
3927   return ::stat(pathbuf, sbuf);
3928 }
3929 
3930 bool os::check_heap(bool force) {
3931   return true;
3932 }
3933 
3934 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
3935   return ::vsnprintf(buf, count, format, args);
3936 }
3937 
3938 // Is a (classpath) directory empty?
3939 bool os::dir_is_empty(const char* path) {
3940   DIR *dir = NULL;
3941   struct dirent *ptr;
3942 
3943   dir = opendir(path);
3944   if (dir == NULL) return true;
3945 
3946   /* Scan the directory */
3947   bool result = true;
3948   char buf[sizeof(struct dirent) + MAX_PATH];
3949   while (result && (ptr = ::readdir(dir)) != NULL) {
3950     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
3951       result = false;
3952     }
3953   }
3954   closedir(dir);
3955   return result;
3956 }
3957 
3958 // create binary file, rewriting existing file if required
3959 int os::create_binary_file(const char* path, bool rewrite_existing) {
3960   int oflags = O_WRONLY | O_CREAT;
3961   if (!rewrite_existing) {
3962     oflags |= O_EXCL;
3963   }
3964   return ::open64(path, oflags, S_IREAD | S_IWRITE);
3965 }
3966 
3967 // return current position of file pointer
3968 jlong os::current_file_offset(int fd) {
3969   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
3970 }
3971 
3972 // move file pointer to the specified offset
3973 jlong os::seek_to_file_offset(int fd, jlong offset) {
3974   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
3975 }
3976 
3977 // Map a block of memory.
3978 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
3979                      char *addr, size_t bytes, bool read_only,
3980                      bool allow_exec) {
3981   int prot;
3982   int flags;
3983 
3984   if (read_only) {
3985     prot = PROT_READ;
3986     flags = MAP_SHARED;
3987   } else {
3988     prot = PROT_READ | PROT_WRITE;
3989     flags = MAP_PRIVATE;
3990   }
3991 
3992   if (allow_exec) {
3993     prot |= PROT_EXEC;
3994   }
3995 
3996   if (addr != NULL) {
3997     flags |= MAP_FIXED;
3998   }
3999 
4000   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4001                                      fd, file_offset);
4002   if (mapped_address == MAP_FAILED) {
4003     return NULL;
4004   }
4005   return mapped_address;
4006 }
4007 
4008 
4009 // Remap a block of memory.
4010 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4011                        char *addr, size_t bytes, bool read_only,
4012                        bool allow_exec) {
4013   // same as map_memory() on this OS
4014   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4015                         allow_exec);
4016 }
4017 
4018 
4019 // Unmap a block of memory.
4020 bool os::unmap_memory(char* addr, size_t bytes) {
4021   return munmap(addr, bytes) == 0;
4022 }
4023 
4024 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4025 
4026 static clockid_t thread_cpu_clockid(Thread* thread) {
4027   pthread_t tid = thread->osthread()->pthread_id();
4028   clockid_t clockid;
4029 
4030   // Get thread clockid
4031   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4032   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4033   return clockid;
4034 }
4035 
4036 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4037 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4038 // of a thread.
4039 //
4040 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4041 // the fast estimate available on the platform.
4042 
4043 jlong os::current_thread_cpu_time() {
4044   if (os::Linux::supports_fast_thread_cpu_time()) {
4045     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4046   } else {
4047     // return user + sys since the cost is the same
4048     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4049   }
4050 }
4051 
4052 jlong os::thread_cpu_time(Thread* thread) {
4053   // consistent with what current_thread_cpu_time() returns
4054   if (os::Linux::supports_fast_thread_cpu_time()) {
4055     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4056   } else {
4057     return slow_thread_cpu_time(thread, true /* user + sys */);
4058   }
4059 }
4060 
4061 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4062   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4063     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4064   } else {
4065     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4066   }
4067 }
4068 
4069 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4070   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4071     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4072   } else {
4073     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4074   }
4075 }
4076 
4077 //
4078 //  -1 on error.
4079 //
4080 
4081 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4082   static bool proc_pid_cpu_avail = true;
4083   static bool proc_task_unchecked = true;
4084   static const char *proc_stat_path = "/proc/%d/stat";
4085   pid_t  tid = thread->osthread()->thread_id();
4086   int i;
4087   char *s;
4088   char stat[2048];
4089   int statlen;
4090   char proc_name[64];
4091   int count;
4092   long sys_time, user_time;
4093   char string[64];
4094   int idummy;
4095   long ldummy;
4096   FILE *fp;
4097 
4098   // We first try accessing /proc/<pid>/cpu since this is faster to
4099   // process.  If this file is not present (linux kernels 2.5 and above)
4100   // then we open /proc/<pid>/stat.
4101   if ( proc_pid_cpu_avail ) {
4102     sprintf(proc_name, "/proc/%d/cpu", tid);
4103     fp =  fopen(proc_name, "r");
4104     if ( fp != NULL ) {
4105       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4106       fclose(fp);
4107       if ( count != 3 ) return -1;
4108 
4109       if (user_sys_cpu_time) {
4110         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4111       } else {
4112         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4113       }
4114     }
4115     else proc_pid_cpu_avail = false;
4116   }
4117 
4118   // The /proc/<tid>/stat aggregates per-process usage on
4119   // new Linux kernels 2.6+ where NPTL is supported.
4120   // The /proc/self/task/<tid>/stat still has the per-thread usage.
4121   // See bug 6328462.
4122   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4123   // and possibly in some other cases, so we check its availability.
4124   if (proc_task_unchecked && os::Linux::is_NPTL()) {
4125     // This is executed only once
4126     proc_task_unchecked = false;
4127     fp = fopen("/proc/self/task", "r");
4128     if (fp != NULL) {
4129       proc_stat_path = "/proc/self/task/%d/stat";
4130       fclose(fp);
4131     }
4132   }
4133 
4134   sprintf(proc_name, proc_stat_path, tid);
4135   fp = fopen(proc_name, "r");
4136   if ( fp == NULL ) return -1;
4137   statlen = fread(stat, 1, 2047, fp);
4138   stat[statlen] = '\0';
4139   fclose(fp);
4140 
4141   // Skip pid and the command string. Note that we could be dealing with
4142   // weird command names, e.g. user could decide to rename java launcher
4143   // to "java 1.4.2 :)", then the stat file would look like
4144   //                1234 (java 1.4.2 :)) R ... ...
4145   // We don't really need to know the command string, just find the last
4146   // occurrence of ")" and then start parsing from there. See bug 4726580.
4147   s = strrchr(stat, ')');
4148   i = 0;
4149   if (s == NULL ) return -1;
4150 
4151   // Skip blank chars
4152   do s++; while (isspace(*s));
4153 
4154   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4155                  &idummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4156                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4157                  &user_time, &sys_time);
4158   if ( count != 13 ) return -1;
4159   if (user_sys_cpu_time) {
4160     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4161   } else {
4162     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4163   }
4164 }
4165 
4166 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4167   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4168   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4169   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4170   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4171 }
4172 
4173 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4174   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4175   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4176   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4177   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4178 }
4179 
4180 bool os::is_thread_cpu_time_supported() {
4181   return true;
4182 }
4183 
4184 // System loadavg support.  Returns -1 if load average cannot be obtained.
4185 // Linux doesn't yet have a (official) notion of processor sets,
4186 // so just return the system wide load average.
4187 int os::loadavg(double loadavg[], int nelem) {
4188   return ::getloadavg(loadavg, nelem);
4189 }
4190 
4191 void os::pause() {
4192   char filename[MAX_PATH];
4193   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4194     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4195   } else {
4196     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4197   }
4198 
4199   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4200   if (fd != -1) {
4201     struct stat buf;
4202     close(fd);
4203     while (::stat(filename, &buf) == 0) {
4204       (void)::poll(NULL, 0, 100);
4205     }
4206   } else {
4207     jio_fprintf(stderr,
4208       "Could not open pause file '%s', continuing immediately.\n", filename);
4209   }
4210 }
4211 
4212 extern "C" {
4213 
4214 /**
4215  * NOTE: the following code is to keep the green threads code
4216  * in the libjava.so happy. Once the green threads is removed,
4217  * these code will no longer be needed.
4218  */
4219 int
4220 jdk_waitpid(pid_t pid, int* status, int options) {
4221     return waitpid(pid, status, options);
4222 }
4223 
4224 int
4225 fork1() {
4226     return fork();
4227 }
4228 
4229 int
4230 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4231     return sem_init(sem, pshared, value);
4232 }
4233 
4234 int
4235 jdk_sem_post(sem_t *sem) {
4236     return sem_post(sem);
4237 }
4238 
4239 int
4240 jdk_sem_wait(sem_t *sem) {
4241     return sem_wait(sem);
4242 }
4243 
4244 int
4245 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4246     return pthread_sigmask(how , newmask, oldmask);
4247 }
4248 
4249 }
4250 
4251 // Refer to the comments in os_solaris.cpp park-unpark.
4252 //
4253 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4254 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4255 // For specifics regarding the bug see GLIBC BUGID 261237 :
4256 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4257 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4258 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4259 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4260 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4261 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4262 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4263 // of libpthread avoids the problem, but isn't practical.
4264 //
4265 // Possible remedies:
4266 //
4267 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4268 //      This is palliative and probabilistic, however.  If the thread is preempted
4269 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4270 //      than the minimum period may have passed, and the abstime may be stale (in the
4271 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
4272 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4273 //
4274 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4275 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
4276 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4277 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
4278 //      thread.
4279 //
4280 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4281 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
4282 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4283 //      This also works well.  In fact it avoids kernel-level scalability impediments
4284 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4285 //      timers in a graceful fashion.
4286 //
4287 // 4.   When the abstime value is in the past it appears that control returns
4288 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4289 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
4290 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
4291 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
4292 //      It may be possible to avoid reinitialization by checking the return
4293 //      value from pthread_cond_timedwait().  In addition to reinitializing the
4294 //      condvar we must establish the invariant that cond_signal() is only called
4295 //      within critical sections protected by the adjunct mutex.  This prevents
4296 //      cond_signal() from "seeing" a condvar that's in the midst of being
4297 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
4298 //      desirable signal-after-unlock optimization that avoids futile context switching.
4299 //
4300 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
4301 //      structure when a condvar is used or initialized.  cond_destroy()  would
4302 //      release the helper structure.  Our reinitialize-after-timedwait fix
4303 //      put excessive stress on malloc/free and locks protecting the c-heap.
4304 //
4305 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
4306 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4307 // and only enabling the work-around for vulnerable environments.
4308 
4309 // utility to compute the abstime argument to timedwait:
4310 // millis is the relative timeout time
4311 // abstime will be the absolute timeout time
4312 // TODO: replace compute_abstime() with unpackTime()
4313 
4314 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4315   if (millis < 0)  millis = 0;
4316   struct timeval now;
4317   int status = gettimeofday(&now, NULL);
4318   assert(status == 0, "gettimeofday");
4319   jlong seconds = millis / 1000;
4320   millis %= 1000;
4321   if (seconds > 50000000) { // see man cond_timedwait(3T)
4322     seconds = 50000000;
4323   }
4324   abstime->tv_sec = now.tv_sec  + seconds;
4325   long       usec = now.tv_usec + millis * 1000;
4326   if (usec >= 1000000) {
4327     abstime->tv_sec += 1;
4328     usec -= 1000000;
4329   }
4330   abstime->tv_nsec = usec * 1000;
4331   return abstime;
4332 }
4333 
4334 
4335 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4336 // Conceptually TryPark() should be equivalent to park(0).
4337 
4338 int os::PlatformEvent::TryPark() {
4339   for (;;) {
4340     const int v = _Event ;
4341     guarantee ((v == 0) || (v == 1), "invariant") ;
4342     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
4343   }
4344 }
4345 
4346 void os::PlatformEvent::park() {       // AKA "down()"
4347   // Invariant: Only the thread associated with the Event/PlatformEvent
4348   // may call park().
4349   // TODO: assert that _Assoc != NULL or _Assoc == Self
4350   int v ;
4351   for (;;) {
4352       v = _Event ;
4353       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4354   }
4355   guarantee (v >= 0, "invariant") ;
4356   if (v == 0) {
4357      // Do this the hard way by blocking ...
4358      int status = pthread_mutex_lock(_mutex);
4359      assert_status(status == 0, status, "mutex_lock");
4360      guarantee (_nParked == 0, "invariant") ;
4361      ++ _nParked ;
4362      while (_Event < 0) {
4363         status = pthread_cond_wait(_cond, _mutex);
4364         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4365         // Treat this the same as if the wait was interrupted
4366         if (status == ETIME) { status = EINTR; }
4367         assert_status(status == 0 || status == EINTR, status, "cond_wait");
4368      }
4369      -- _nParked ;
4370 
4371     // In theory we could move the ST of 0 into _Event past the unlock(),
4372     // but then we'd need a MEMBAR after the ST.
4373     _Event = 0 ;
4374      status = pthread_mutex_unlock(_mutex);
4375      assert_status(status == 0, status, "mutex_unlock");
4376   }
4377   guarantee (_Event >= 0, "invariant") ;
4378 }
4379 
4380 int os::PlatformEvent::park(jlong millis) {
4381   guarantee (_nParked == 0, "invariant") ;
4382 
4383   int v ;
4384   for (;;) {
4385       v = _Event ;
4386       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4387   }
4388   guarantee (v >= 0, "invariant") ;
4389   if (v != 0) return OS_OK ;
4390 
4391   // We do this the hard way, by blocking the thread.
4392   // Consider enforcing a minimum timeout value.
4393   struct timespec abst;
4394   compute_abstime(&abst, millis);
4395 
4396   int ret = OS_TIMEOUT;
4397   int status = pthread_mutex_lock(_mutex);
4398   assert_status(status == 0, status, "mutex_lock");
4399   guarantee (_nParked == 0, "invariant") ;
4400   ++_nParked ;
4401 
4402   // Object.wait(timo) will return because of
4403   // (a) notification
4404   // (b) timeout
4405   // (c) thread.interrupt
4406   //
4407   // Thread.interrupt and object.notify{All} both call Event::set.
4408   // That is, we treat thread.interrupt as a special case of notification.
4409   // The underlying Solaris implementation, cond_timedwait, admits
4410   // spurious/premature wakeups, but the JLS/JVM spec prevents the
4411   // JVM from making those visible to Java code.  As such, we must
4412   // filter out spurious wakeups.  We assume all ETIME returns are valid.
4413   //
4414   // TODO: properly differentiate simultaneous notify+interrupt.
4415   // In that case, we should propagate the notify to another waiter.
4416 
4417   while (_Event < 0) {
4418     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4419     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4420       pthread_cond_destroy (_cond);
4421       pthread_cond_init (_cond, NULL) ;
4422     }
4423     assert_status(status == 0 || status == EINTR ||
4424                   status == ETIME || status == ETIMEDOUT,
4425                   status, "cond_timedwait");
4426     if (!FilterSpuriousWakeups) break ;                 // previous semantics
4427     if (status == ETIME || status == ETIMEDOUT) break ;
4428     // We consume and ignore EINTR and spurious wakeups.
4429   }
4430   --_nParked ;
4431   if (_Event >= 0) {
4432      ret = OS_OK;
4433   }
4434   _Event = 0 ;
4435   status = pthread_mutex_unlock(_mutex);
4436   assert_status(status == 0, status, "mutex_unlock");
4437   assert (_nParked == 0, "invariant") ;
4438   return ret;
4439 }
4440 
4441 void os::PlatformEvent::unpark() {
4442   int v, AnyWaiters ;
4443   for (;;) {
4444       v = _Event ;
4445       if (v > 0) {
4446          // The LD of _Event could have reordered or be satisfied
4447          // by a read-aside from this processor's write buffer.
4448          // To avoid problems execute a barrier and then
4449          // ratify the value.
4450          OrderAccess::fence() ;
4451          if (_Event == v) return ;
4452          continue ;
4453       }
4454       if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4455   }
4456   if (v < 0) {
4457      // Wait for the thread associated with the event to vacate
4458      int status = pthread_mutex_lock(_mutex);
4459      assert_status(status == 0, status, "mutex_lock");
4460      AnyWaiters = _nParked ;
4461      assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4462      if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4463         AnyWaiters = 0 ;
4464         pthread_cond_signal (_cond);
4465      }
4466      status = pthread_mutex_unlock(_mutex);
4467      assert_status(status == 0, status, "mutex_unlock");
4468      if (AnyWaiters != 0) {
4469         status = pthread_cond_signal(_cond);
4470         assert_status(status == 0, status, "cond_signal");
4471      }
4472   }
4473 
4474   // Note that we signal() _after dropping the lock for "immortal" Events.
4475   // This is safe and avoids a common class of  futile wakeups.  In rare
4476   // circumstances this can cause a thread to return prematurely from
4477   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4478   // simply re-test the condition and re-park itself.
4479 }
4480 
4481 
4482 // JSR166
4483 // -------------------------------------------------------
4484 
4485 /*
4486  * The solaris and linux implementations of park/unpark are fairly
4487  * conservative for now, but can be improved. They currently use a
4488  * mutex/condvar pair, plus a a count.
4489  * Park decrements count if > 0, else does a condvar wait.  Unpark
4490  * sets count to 1 and signals condvar.  Only one thread ever waits
4491  * on the condvar. Contention seen when trying to park implies that someone
4492  * is unparking you, so don't wait. And spurious returns are fine, so there
4493  * is no need to track notifications.
4494  */
4495 
4496 
4497 #define NANOSECS_PER_SEC 1000000000
4498 #define NANOSECS_PER_MILLISEC 1000000
4499 #define MAX_SECS 100000000
4500 /*
4501  * This code is common to linux and solaris and will be moved to a
4502  * common place in dolphin.
4503  *
4504  * The passed in time value is either a relative time in nanoseconds
4505  * or an absolute time in milliseconds. Either way it has to be unpacked
4506  * into suitable seconds and nanoseconds components and stored in the
4507  * given timespec structure.
4508  * Given time is a 64-bit value and the time_t used in the timespec is only
4509  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4510  * overflow if times way in the future are given. Further on Solaris versions
4511  * prior to 10 there is a restriction (see cond_timedwait) that the specified
4512  * number of seconds, in abstime, is less than current_time  + 100,000,000.
4513  * As it will be 28 years before "now + 100000000" will overflow we can
4514  * ignore overflow and just impose a hard-limit on seconds using the value
4515  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4516  * years from "now".
4517  */
4518 
4519 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4520   assert (time > 0, "convertTime");
4521 
4522   struct timeval now;
4523   int status = gettimeofday(&now, NULL);
4524   assert(status == 0, "gettimeofday");
4525 
4526   time_t max_secs = now.tv_sec + MAX_SECS;
4527 
4528   if (isAbsolute) {
4529     jlong secs = time / 1000;
4530     if (secs > max_secs) {
4531       absTime->tv_sec = max_secs;
4532     }
4533     else {
4534       absTime->tv_sec = secs;
4535     }
4536     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4537   }
4538   else {
4539     jlong secs = time / NANOSECS_PER_SEC;
4540     if (secs >= MAX_SECS) {
4541       absTime->tv_sec = max_secs;
4542       absTime->tv_nsec = 0;
4543     }
4544     else {
4545       absTime->tv_sec = now.tv_sec + secs;
4546       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4547       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4548         absTime->tv_nsec -= NANOSECS_PER_SEC;
4549         ++absTime->tv_sec; // note: this must be <= max_secs
4550       }
4551     }
4552   }
4553   assert(absTime->tv_sec >= 0, "tv_sec < 0");
4554   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4555   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4556   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4557 }
4558 
4559 void Parker::park(bool isAbsolute, jlong time) {
4560   // Optional fast-path check:
4561   // Return immediately if a permit is available.
4562   if (_counter > 0) {
4563       _counter = 0 ;
4564       return ;
4565   }
4566 
4567   Thread* thread = Thread::current();
4568   assert(thread->is_Java_thread(), "Must be JavaThread");
4569   JavaThread *jt = (JavaThread *)thread;
4570 
4571   // Optional optimization -- avoid state transitions if there's an interrupt pending.
4572   // Check interrupt before trying to wait
4573   if (Thread::is_interrupted(thread, false)) {
4574     return;
4575   }
4576 
4577   // Next, demultiplex/decode time arguments
4578   timespec absTime;
4579   if (time < 0) { // don't wait at all
4580     return;
4581   }
4582   if (time > 0) {
4583     unpackTime(&absTime, isAbsolute, time);
4584   }
4585 
4586 
4587   // Enter safepoint region
4588   // Beware of deadlocks such as 6317397.
4589   // The per-thread Parker:: mutex is a classic leaf-lock.
4590   // In particular a thread must never block on the Threads_lock while
4591   // holding the Parker:: mutex.  If safepoints are pending both the
4592   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4593   ThreadBlockInVM tbivm(jt);
4594 
4595   // Don't wait if cannot get lock since interference arises from
4596   // unblocking.  Also. check interrupt before trying wait
4597   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4598     return;
4599   }
4600 
4601   int status ;
4602   if (_counter > 0)  { // no wait needed
4603     _counter = 0;
4604     status = pthread_mutex_unlock(_mutex);
4605     assert (status == 0, "invariant") ;
4606     return;
4607   }
4608 
4609 #ifdef ASSERT
4610   // Don't catch signals while blocked; let the running threads have the signals.
4611   // (This allows a debugger to break into the running thread.)
4612   sigset_t oldsigs;
4613   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4614   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4615 #endif
4616 
4617   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4618   jt->set_suspend_equivalent();
4619   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4620 
4621   if (time == 0) {
4622     status = pthread_cond_wait (_cond, _mutex) ;
4623   } else {
4624     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4625     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4626       pthread_cond_destroy (_cond) ;
4627       pthread_cond_init    (_cond, NULL);
4628     }
4629   }
4630   assert_status(status == 0 || status == EINTR ||
4631                 status == ETIME || status == ETIMEDOUT,
4632                 status, "cond_timedwait");
4633 
4634 #ifdef ASSERT
4635   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4636 #endif
4637 
4638   _counter = 0 ;
4639   status = pthread_mutex_unlock(_mutex) ;
4640   assert_status(status == 0, status, "invariant") ;
4641   // If externally suspended while waiting, re-suspend
4642   if (jt->handle_special_suspend_equivalent_condition()) {
4643     jt->java_suspend_self();
4644   }
4645 
4646 }
4647 
4648 void Parker::unpark() {
4649   int s, status ;
4650   status = pthread_mutex_lock(_mutex);
4651   assert (status == 0, "invariant") ;
4652   s = _counter;
4653   _counter = 1;
4654   if (s < 1) {
4655      if (WorkAroundNPTLTimedWaitHang) {
4656         status = pthread_cond_signal (_cond) ;
4657         assert (status == 0, "invariant") ;
4658         status = pthread_mutex_unlock(_mutex);
4659         assert (status == 0, "invariant") ;
4660      } else {
4661         status = pthread_mutex_unlock(_mutex);
4662         assert (status == 0, "invariant") ;
4663         status = pthread_cond_signal (_cond) ;
4664         assert (status == 0, "invariant") ;
4665      }
4666   } else {
4667     pthread_mutex_unlock(_mutex);
4668     assert (status == 0, "invariant") ;
4669   }
4670 }
4671 
4672 
4673 extern char** environ;
4674 
4675 #ifndef __NR_fork
4676 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
4677 #endif
4678 
4679 #ifndef __NR_execve
4680 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
4681 #endif
4682 
4683 // Run the specified command in a separate process. Return its exit value,
4684 // or -1 on failure (e.g. can't fork a new process).
4685 // Unlike system(), this function can be called from signal handler. It
4686 // doesn't block SIGINT et al.
4687 int os::fork_and_exec(char* cmd) {
4688   const char * argv[4] = {"sh", "-c", cmd, NULL};
4689 
4690   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
4691   // pthread_atfork handlers and reset pthread library. All we need is a
4692   // separate process to execve. Make a direct syscall to fork process.
4693   // On IA64 there's no fork syscall, we have to use fork() and hope for
4694   // the best...
4695   pid_t pid = NOT_IA64(syscall(__NR_fork);)
4696               IA64_ONLY(fork();)
4697 
4698   if (pid < 0) {
4699     // fork failed
4700     return -1;
4701 
4702   } else if (pid == 0) {
4703     // child process
4704 
4705     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
4706     // first to kill every thread on the thread list. Because this list is
4707     // not reset by fork() (see notes above), execve() will instead kill
4708     // every thread in the parent process. We know this is the only thread
4709     // in the new process, so make a system call directly.
4710     // IA64 should use normal execve() from glibc to match the glibc fork()
4711     // above.
4712     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
4713     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
4714 
4715     // execve failed
4716     _exit(-1);
4717 
4718   } else  {
4719     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4720     // care about the actual exit code, for now.
4721 
4722     int status;
4723 
4724     // Wait for the child process to exit.  This returns immediately if
4725     // the child has already exited. */
4726     while (waitpid(pid, &status, 0) < 0) {
4727         switch (errno) {
4728         case ECHILD: return 0;
4729         case EINTR: break;
4730         default: return -1;
4731         }
4732     }
4733 
4734     if (WIFEXITED(status)) {
4735        // The child exited normally; get its exit code.
4736        return WEXITSTATUS(status);
4737     } else if (WIFSIGNALED(status)) {
4738        // The child exited because of a signal
4739        // The best value to return is 0x80 + signal number,
4740        // because that is what all Unix shells do, and because
4741        // it allows callers to distinguish between process exit and
4742        // process death by signal.
4743        return 0x80 + WTERMSIG(status);
4744     } else {
4745        // Unknown exit code; pass it through
4746        return status;
4747     }
4748   }
4749 }