1 /*
   2  * Copyright 1999-2009 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 #include <sstream>
  58 #include <iostream>
  59 #include <fstream>
  60 
  61 #define MAX_PATH    (2 * K)
  62 
  63 // for timer info max values which include all bits
  64 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
  65 #define SEC_IN_NANOSECS  1000000000LL
  66 
  67 ////////////////////////////////////////////////////////////////////////////////
  68 // global variables
  69 julong os::Linux::_physical_memory = 0;
  70 
  71 address   os::Linux::_initial_thread_stack_bottom = NULL;
  72 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
  73 
  74 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
  75 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
  76 Mutex* os::Linux::_createThread_lock = NULL;
  77 pthread_t os::Linux::_main_thread;
  78 int os::Linux::_page_size = -1;
  79 bool os::Linux::_is_floating_stack = false;
  80 bool os::Linux::_is_NPTL = false;
  81 bool os::Linux::_supports_fast_thread_cpu_time = false;
  82 const char * os::Linux::_glibc_version = NULL;
  83 const char * os::Linux::_libpthread_version = NULL;
  84 
  85 static jlong initial_time_count=0;
  86 
  87 static int clock_tics_per_sec = 100;
  88 
  89 // For diagnostics to print a message once. see run_periodic_checks
  90 static sigset_t check_signal_done;
  91 static bool check_signals = true;;
  92 
  93 static pid_t _initial_pid = 0;
  94 
  95 /* Signal number used to suspend/resume a thread */
  96 
  97 /* do not use any signal number less than SIGSEGV, see 4355769 */
  98 static int SR_signum = SIGUSR2;
  99 sigset_t SR_sigset;
 100 
 101 /* Used to protect dlsym() calls */
 102 static pthread_mutex_t dl_mutex;
 103 
 104 ////////////////////////////////////////////////////////////////////////////////
 105 // utility functions
 106 
 107 static int SR_initialize();
 108 static int SR_finalize();
 109 
 110 julong os::available_memory() {
 111   return Linux::available_memory();
 112 }
 113 
 114 julong os::Linux::available_memory() {
 115   // values in struct sysinfo are "unsigned long"
 116   struct sysinfo si;
 117   sysinfo(&si);
 118 
 119   return (julong)si.freeram * si.mem_unit;
 120 }
 121 
 122 julong os::physical_memory() {
 123   return Linux::physical_memory();
 124 }
 125 
 126 julong os::allocatable_physical_memory(julong size) {
 127 #ifdef _LP64
 128   return size;
 129 #else
 130   julong result = MIN2(size, (julong)3800*M);
 131    if (!is_allocatable(result)) {
 132      // See comments under solaris for alignment considerations
 133      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 134      result =  MIN2(size, reasonable_size);
 135    }
 136    return result;
 137 #endif // _LP64
 138 }
 139 
 140 ////////////////////////////////////////////////////////////////////////////////
 141 // environment support
 142 
 143 bool os::getenv(const char* name, char* buf, int len) {
 144   const char* val = ::getenv(name);
 145   if (val != NULL && strlen(val) < (size_t)len) {
 146     strcpy(buf, val);
 147     return true;
 148   }
 149   if (len > 0) buf[0] = 0;  // return a null string
 150   return false;
 151 }
 152 
 153 
 154 // Return true if user is running as root.
 155 
 156 bool os::have_special_privileges() {
 157   static bool init = false;
 158   static bool privileges = false;
 159   if (!init) {
 160     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 161     init = true;
 162   }
 163   return privileges;
 164 }
 165 
 166 
 167 #ifndef SYS_gettid
 168 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 169 #ifdef __ia64__
 170 #define SYS_gettid 1105
 171 #elif __i386__
 172 #define SYS_gettid 224
 173 #elif __amd64__
 174 #define SYS_gettid 186
 175 #elif __sparc__
 176 #define SYS_gettid 143
 177 #else
 178 #error define gettid for the arch
 179 #endif
 180 #endif
 181 
 182 // Cpu architecture string
 183 #if   defined(ZERO)
 184 static char cpu_arch[] = ZERO_LIBARCH;
 185 #elif defined(IA64)
 186 static char cpu_arch[] = "ia64";
 187 #elif defined(IA32)
 188 static char cpu_arch[] = "i386";
 189 #elif defined(AMD64)
 190 static char cpu_arch[] = "amd64";
 191 #elif defined(SPARC)
 192 #  ifdef _LP64
 193 static char cpu_arch[] = "sparcv9";
 194 #  else
 195 static char cpu_arch[] = "sparc";
 196 #  endif
 197 #else
 198 #error Add appropriate cpu_arch setting
 199 #endif
 200 
 201 
 202 // pid_t gettid()
 203 //
 204 // Returns the kernel thread id of the currently running thread. Kernel
 205 // thread id is used to access /proc.
 206 //
 207 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 208 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 209 //
 210 pid_t os::Linux::gettid() {
 211   int rslt = syscall(SYS_gettid);
 212   if (rslt == -1) {
 213      // old kernel, no NPTL support
 214      return getpid();
 215   } else {
 216      return (pid_t)rslt;
 217   }
 218 }
 219 
 220 // Most versions of linux have a bug where the number of processors are
 221 // determined by looking at the /proc file system.  In a chroot environment,
 222 // the system call returns 1.  This causes the VM to act as if it is
 223 // a single processor and elide locking (see is_MP() call).
 224 static bool unsafe_chroot_detected = false;
 225 static const char *unstable_chroot_error = "/proc file system not found.\n"
 226                      "Java may be unstable running multithreaded in a chroot "
 227                      "environment on Linux when /proc filesystem is not mounted.";
 228 
 229 void os::Linux::initialize_system_info() {
 230   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 231   if (processor_count() == 1) {
 232     pid_t pid = os::Linux::gettid();
 233     char fname[32];
 234     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 235     FILE *fp = fopen(fname, "r");
 236     if (fp == NULL) {
 237       unsafe_chroot_detected = true;
 238     } else {
 239       fclose(fp);
 240     }
 241   }
 242   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 243   assert(processor_count() > 0, "linux error");
 244 }
 245 
 246 void os::init_system_properties_values() {
 247 //  char arch[12];
 248 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 249 
 250   // The next steps are taken in the product version:
 251   //
 252   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 253   // This library should be located at:
 254   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 255   //
 256   // If "/jre/lib/" appears at the right place in the path, then we
 257   // assume libjvm[_g].so is installed in a JDK and we use this path.
 258   //
 259   // Otherwise exit with message: "Could not create the Java virtual machine."
 260   //
 261   // The following extra steps are taken in the debugging version:
 262   //
 263   // If "/jre/lib/" does NOT appear at the right place in the path
 264   // instead of exit check for $JAVA_HOME environment variable.
 265   //
 266   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 267   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 268   // it looks like libjvm[_g].so is installed there
 269   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 270   //
 271   // Otherwise exit.
 272   //
 273   // Important note: if the location of libjvm.so changes this
 274   // code needs to be changed accordingly.
 275 
 276   // The next few definitions allow the code to be verbatim:
 277 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 278 #define getenv(n) ::getenv(n)
 279 
 280 /*
 281  * See ld(1):
 282  *      The linker uses the following search paths to locate required
 283  *      shared libraries:
 284  *        1: ...
 285  *        ...
 286  *        7: The default directories, normally /lib and /usr/lib.
 287  */
 288 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 289 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 290 #else
 291 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 292 #endif
 293 
 294 #define EXTENSIONS_DIR  "/lib/ext"
 295 #define ENDORSED_DIR    "/lib/endorsed"
 296 #define REG_DIR         "/usr/java/packages"
 297 
 298   {
 299     /* sysclasspath, java_home, dll_dir */
 300     {
 301         char *home_path;
 302         char *dll_path;
 303         char *pslash;
 304         char buf[MAXPATHLEN];
 305         os::jvm_path(buf, sizeof(buf));
 306 
 307         // Found the full path to libjvm.so.
 308         // Now cut the path to <java_home>/jre if we can.
 309         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 310         pslash = strrchr(buf, '/');
 311         if (pslash != NULL)
 312             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 313         dll_path = malloc(strlen(buf) + 1);
 314         if (dll_path == NULL)
 315             return;
 316         strcpy(dll_path, buf);
 317         Arguments::set_dll_dir(dll_path);
 318 
 319         if (pslash != NULL) {
 320             pslash = strrchr(buf, '/');
 321             if (pslash != NULL) {
 322                 *pslash = '\0';       /* get rid of /<arch> */
 323                 pslash = strrchr(buf, '/');
 324                 if (pslash != NULL)
 325                     *pslash = '\0';   /* get rid of /lib */
 326             }
 327         }
 328 
 329         home_path = malloc(strlen(buf) + 1);
 330         if (home_path == NULL)
 331             return;
 332         strcpy(home_path, buf);
 333         Arguments::set_java_home(home_path);
 334 
 335         if (!set_boot_path('/', ':'))
 336             return;
 337     }
 338 
 339     /*
 340      * Where to look for native libraries
 341      *
 342      * Note: Due to a legacy implementation, most of the library path
 343      * is set in the launcher.  This was to accomodate linking restrictions
 344      * on legacy Linux implementations (which are no longer supported).
 345      * Eventually, all the library path setting will be done here.
 346      *
 347      * However, to prevent the proliferation of improperly built native
 348      * libraries, the new path component /usr/java/packages is added here.
 349      * Eventually, all the library path setting will be done here.
 350      */
 351     {
 352         char *ld_library_path;
 353 
 354         /*
 355          * Construct the invariant part of ld_library_path. Note that the
 356          * space for the colon and the trailing null are provided by the
 357          * nulls included by the sizeof operator (so actually we allocate
 358          * a byte more than necessary).
 359          */
 360         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 361             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 362         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 363 
 364         /*
 365          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 366          * should always exist (until the legacy problem cited above is
 367          * addressed).
 368          */
 369         char *v = getenv("LD_LIBRARY_PATH");
 370         if (v != NULL) {
 371             char *t = ld_library_path;
 372             /* That's +1 for the colon and +1 for the trailing '\0' */
 373             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 374             sprintf(ld_library_path, "%s:%s", v, t);
 375         }
 376         Arguments::set_library_path(ld_library_path);
 377     }
 378 
 379     /*
 380      * Extensions directories.
 381      *
 382      * Note that the space for the colon and the trailing null are provided
 383      * by the nulls included by the sizeof operator (so actually one byte more
 384      * than necessary is allocated).
 385      */
 386     {
 387         char *buf = malloc(strlen(Arguments::get_java_home()) +
 388             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 389         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 390             Arguments::get_java_home());
 391         Arguments::set_ext_dirs(buf);
 392     }
 393 
 394     /* Endorsed standards default directory. */
 395     {
 396         char * buf;
 397         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 398         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 399         Arguments::set_endorsed_dirs(buf);
 400     }
 401   }
 402 
 403 #undef malloc
 404 #undef getenv
 405 #undef EXTENSIONS_DIR
 406 #undef ENDORSED_DIR
 407 
 408   // Done
 409   return;
 410 }
 411 
 412 ////////////////////////////////////////////////////////////////////////////////
 413 // breakpoint support
 414 
 415 void os::breakpoint() {
 416   BREAKPOINT;
 417 }
 418 
 419 extern "C" void breakpoint() {
 420   // use debugger to set breakpoint here
 421 }
 422 
 423 ////////////////////////////////////////////////////////////////////////////////
 424 // signal support
 425 
 426 debug_only(static bool signal_sets_initialized = false);
 427 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 428 
 429 bool os::Linux::is_sig_ignored(int sig) {
 430       struct sigaction oact;
 431       sigaction(sig, (struct sigaction*)NULL, &oact);
 432       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 433                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 434       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 435            return true;
 436       else
 437            return false;
 438 }
 439 
 440 void os::Linux::signal_sets_init() {
 441   // Should also have an assertion stating we are still single-threaded.
 442   assert(!signal_sets_initialized, "Already initialized");
 443   // Fill in signals that are necessarily unblocked for all threads in
 444   // the VM. Currently, we unblock the following signals:
 445   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 446   //                         by -Xrs (=ReduceSignalUsage));
 447   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 448   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 449   // the dispositions or masks wrt these signals.
 450   // Programs embedding the VM that want to use the above signals for their
 451   // own purposes must, at this time, use the "-Xrs" option to prevent
 452   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 453   // (See bug 4345157, and other related bugs).
 454   // In reality, though, unblocking these signals is really a nop, since
 455   // these signals are not blocked by default.
 456   sigemptyset(&unblocked_sigs);
 457   sigemptyset(&allowdebug_blocked_sigs);
 458   sigaddset(&unblocked_sigs, SIGILL);
 459   sigaddset(&unblocked_sigs, SIGSEGV);
 460   sigaddset(&unblocked_sigs, SIGBUS);
 461   sigaddset(&unblocked_sigs, SIGFPE);
 462   sigaddset(&unblocked_sigs, SR_signum);
 463 
 464   if (!ReduceSignalUsage) {
 465    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 466       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 467       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 468    }
 469    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 470       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 471       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 472    }
 473    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 474       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 475       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 476    }
 477   }
 478   // Fill in signals that are blocked by all but the VM thread.
 479   sigemptyset(&vm_sigs);
 480   if (!ReduceSignalUsage)
 481     sigaddset(&vm_sigs, BREAK_SIGNAL);
 482   debug_only(signal_sets_initialized = true);
 483 
 484 }
 485 
 486 // These are signals that are unblocked while a thread is running Java.
 487 // (For some reason, they get blocked by default.)
 488 sigset_t* os::Linux::unblocked_signals() {
 489   assert(signal_sets_initialized, "Not initialized");
 490   return &unblocked_sigs;
 491 }
 492 
 493 // These are the signals that are blocked while a (non-VM) thread is
 494 // running Java. Only the VM thread handles these signals.
 495 sigset_t* os::Linux::vm_signals() {
 496   assert(signal_sets_initialized, "Not initialized");
 497   return &vm_sigs;
 498 }
 499 
 500 // These are signals that are blocked during cond_wait to allow debugger in
 501 sigset_t* os::Linux::allowdebug_blocked_signals() {
 502   assert(signal_sets_initialized, "Not initialized");
 503   return &allowdebug_blocked_sigs;
 504 }
 505 
 506 void os::Linux::hotspot_sigmask(Thread* thread) {
 507 
 508   //Save caller's signal mask before setting VM signal mask
 509   sigset_t caller_sigmask;
 510   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 511 
 512   OSThread* osthread = thread->osthread();
 513   osthread->set_caller_sigmask(caller_sigmask);
 514 
 515   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 516 
 517   if (!ReduceSignalUsage) {
 518     if (thread->is_VM_thread()) {
 519       // Only the VM thread handles BREAK_SIGNAL ...
 520       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 521     } else {
 522       // ... all other threads block BREAK_SIGNAL
 523       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 524     }
 525   }
 526 }
 527 
 528 //////////////////////////////////////////////////////////////////////////////
 529 // detecting pthread library
 530 
 531 void os::Linux::libpthread_init() {
 532   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 533   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 534   // generic name for earlier versions.
 535   // Define macros here so we can build HotSpot on old systems.
 536 # ifndef _CS_GNU_LIBC_VERSION
 537 # define _CS_GNU_LIBC_VERSION 2
 538 # endif
 539 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 540 # define _CS_GNU_LIBPTHREAD_VERSION 3
 541 # endif
 542 
 543   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 544   if (n > 0) {
 545      char *str = (char *)malloc(n);
 546      confstr(_CS_GNU_LIBC_VERSION, str, n);
 547      os::Linux::set_glibc_version(str);
 548   } else {
 549      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 550      static char _gnu_libc_version[32];
 551      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 552               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 553      os::Linux::set_glibc_version(_gnu_libc_version);
 554   }
 555 
 556   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 557   if (n > 0) {
 558      char *str = (char *)malloc(n);
 559      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 560      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 561      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 562      // is the case. LinuxThreads has a hard limit on max number of threads.
 563      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 564      // On the other hand, NPTL does not have such a limit, sysconf()
 565      // will return -1 and errno is not changed. Check if it is really NPTL.
 566      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 567          strstr(str, "NPTL") &&
 568          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 569        free(str);
 570        os::Linux::set_libpthread_version("linuxthreads");
 571      } else {
 572        os::Linux::set_libpthread_version(str);
 573      }
 574   } else {
 575     // glibc before 2.3.2 only has LinuxThreads.
 576     os::Linux::set_libpthread_version("linuxthreads");
 577   }
 578 
 579   if (strstr(libpthread_version(), "NPTL")) {
 580      os::Linux::set_is_NPTL();
 581   } else {
 582      os::Linux::set_is_LinuxThreads();
 583   }
 584 
 585   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 586   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 587   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 588      os::Linux::set_is_floating_stack();
 589   }
 590 }
 591 
 592 /////////////////////////////////////////////////////////////////////////////
 593 // thread stack
 594 
 595 // Force Linux kernel to expand current thread stack. If "bottom" is close
 596 // to the stack guard, caller should block all signals.
 597 //
 598 // MAP_GROWSDOWN:
 599 //   A special mmap() flag that is used to implement thread stacks. It tells
 600 //   kernel that the memory region should extend downwards when needed. This
 601 //   allows early versions of LinuxThreads to only mmap the first few pages
 602 //   when creating a new thread. Linux kernel will automatically expand thread
 603 //   stack as needed (on page faults).
 604 //
 605 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 606 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 607 //   region, it's hard to tell if the fault is due to a legitimate stack
 608 //   access or because of reading/writing non-exist memory (e.g. buffer
 609 //   overrun). As a rule, if the fault happens below current stack pointer,
 610 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 611 //   application (see Linux kernel fault.c).
 612 //
 613 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 614 //   stack overflow detection.
 615 //
 616 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 617 //   not use this flag. However, the stack of initial thread is not created
 618 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 619 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 620 //   and then attach the thread to JVM.
 621 //
 622 // To get around the problem and allow stack banging on Linux, we need to
 623 // manually expand thread stack after receiving the SIGSEGV.
 624 //
 625 // There are two ways to expand thread stack to address "bottom", we used
 626 // both of them in JVM before 1.5:
 627 //   1. adjust stack pointer first so that it is below "bottom", and then
 628 //      touch "bottom"
 629 //   2. mmap() the page in question
 630 //
 631 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 632 // if current sp is already near the lower end of page 101, and we need to
 633 // call mmap() to map page 100, it is possible that part of the mmap() frame
 634 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 635 // That will destroy the mmap() frame and cause VM to crash.
 636 //
 637 // The following code works by adjusting sp first, then accessing the "bottom"
 638 // page to force a page fault. Linux kernel will then automatically expand the
 639 // stack mapping.
 640 //
 641 // _expand_stack_to() assumes its frame size is less than page size, which
 642 // should always be true if the function is not inlined.
 643 
 644 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 645 #define NOINLINE
 646 #else
 647 #define NOINLINE __attribute__ ((noinline))
 648 #endif
 649 
 650 static void _expand_stack_to(address bottom) NOINLINE;
 651 
 652 static void _expand_stack_to(address bottom) {
 653   address sp;
 654   size_t size;
 655   volatile char *p;
 656 
 657   // Adjust bottom to point to the largest address within the same page, it
 658   // gives us a one-page buffer if alloca() allocates slightly more memory.
 659   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 660   bottom += os::Linux::page_size() - 1;
 661 
 662   // sp might be slightly above current stack pointer; if that's the case, we
 663   // will alloca() a little more space than necessary, which is OK. Don't use
 664   // os::current_stack_pointer(), as its result can be slightly below current
 665   // stack pointer, causing us to not alloca enough to reach "bottom".
 666   sp = (address)&sp;
 667 
 668   if (sp > bottom) {
 669     size = sp - bottom;
 670     p = (volatile char *)alloca(size);
 671     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 672     p[0] = '\0';
 673   }
 674 }
 675 
 676 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 677   assert(t!=NULL, "just checking");
 678   assert(t->osthread()->expanding_stack(), "expand should be set");
 679   assert(t->stack_base() != NULL, "stack_base was not initialized");
 680 
 681   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 682     sigset_t mask_all, old_sigset;
 683     sigfillset(&mask_all);
 684     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 685     _expand_stack_to(addr);
 686     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 687     return true;
 688   }
 689   return false;
 690 }
 691 
 692 //////////////////////////////////////////////////////////////////////////////
 693 // create new thread
 694 
 695 static address highest_vm_reserved_address();
 696 
 697 // check if it's safe to start a new thread
 698 static bool _thread_safety_check(Thread* thread) {
 699   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 700     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 701     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 702     //   allocated (MAP_FIXED) from high address space. Every thread stack
 703     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 704     //   it to other values if they rebuild LinuxThreads).
 705     //
 706     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 707     // the memory region has already been mmap'ed. That means if we have too
 708     // many threads and/or very large heap, eventually thread stack will
 709     // collide with heap.
 710     //
 711     // Here we try to prevent heap/stack collision by comparing current
 712     // stack bottom with the highest address that has been mmap'ed by JVM
 713     // plus a safety margin for memory maps created by native code.
 714     //
 715     // This feature can be disabled by setting ThreadSafetyMargin to 0
 716     //
 717     if (ThreadSafetyMargin > 0) {
 718       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 719 
 720       // not safe if our stack extends below the safety margin
 721       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 722     } else {
 723       return true;
 724     }
 725   } else {
 726     // Floating stack LinuxThreads or NPTL:
 727     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 728     //   there's not enough space left, pthread_create() will fail. If we come
 729     //   here, that means enough space has been reserved for stack.
 730     return true;
 731   }
 732 }
 733 
 734 // Thread start routine for all newly created threads
 735 static void *java_start(Thread *thread) {
 736   // Try to randomize the cache line index of hot stack frames.
 737   // This helps when threads of the same stack traces evict each other's
 738   // cache lines. The threads can be either from the same JVM instance, or
 739   // from different JVM instances. The benefit is especially true for
 740   // processors with hyperthreading technology.
 741   static int counter = 0;
 742   int pid = os::current_process_id();
 743   alloca(((pid ^ counter++) & 7) * 128);
 744 
 745   ThreadLocalStorage::set_thread(thread);
 746 
 747   OSThread* osthread = thread->osthread();
 748   Monitor* sync = osthread->startThread_lock();
 749 
 750   // non floating stack LinuxThreads needs extra check, see above
 751   if (!_thread_safety_check(thread)) {
 752     // notify parent thread
 753     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 754     osthread->set_state(ZOMBIE);
 755     sync->notify_all();
 756     return NULL;
 757   }
 758 
 759   // thread_id is kernel thread id (similar to Solaris LWP id)
 760   osthread->set_thread_id(os::Linux::gettid());
 761 
 762   if (UseNUMA) {
 763     int lgrp_id = os::numa_get_group_id();
 764     if (lgrp_id != -1) {
 765       thread->set_lgrp_id(lgrp_id);
 766     }
 767   }
 768   // initialize signal mask for this thread
 769   os::Linux::hotspot_sigmask(thread);
 770 
 771   // initialize floating point control register
 772   os::Linux::init_thread_fpu_state();
 773 
 774   // handshaking with parent thread
 775   {
 776     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 777 
 778     // notify parent thread
 779     osthread->set_state(INITIALIZED);
 780     sync->notify_all();
 781 
 782     // wait until os::start_thread()
 783     while (osthread->get_state() == INITIALIZED) {
 784       sync->wait(Mutex::_no_safepoint_check_flag);
 785     }
 786   }
 787 
 788   // call one more level start routine
 789   thread->run();
 790 
 791   return 0;
 792 }
 793 
 794 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 795   assert(thread->osthread() == NULL, "caller responsible");
 796 
 797   // Allocate the OSThread object
 798   OSThread* osthread = new OSThread(NULL, NULL);
 799   if (osthread == NULL) {
 800     return false;
 801   }
 802 
 803   // set the correct thread state
 804   osthread->set_thread_type(thr_type);
 805 
 806   // Initial state is ALLOCATED but not INITIALIZED
 807   osthread->set_state(ALLOCATED);
 808 
 809   thread->set_osthread(osthread);
 810 
 811   // init thread attributes
 812   pthread_attr_t attr;
 813   pthread_attr_init(&attr);
 814   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 815 
 816   // stack size
 817   if (os::Linux::supports_variable_stack_size()) {
 818     // calculate stack size if it's not specified by caller
 819     if (stack_size == 0) {
 820       stack_size = os::Linux::default_stack_size(thr_type);
 821 
 822       switch (thr_type) {
 823       case os::java_thread:
 824         // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
 825         if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
 826         break;
 827       case os::compiler_thread:
 828         if (CompilerThreadStackSize > 0) {
 829           stack_size = (size_t)(CompilerThreadStackSize * K);
 830           break;
 831         } // else fall through:
 832           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 833       case os::vm_thread:
 834       case os::pgc_thread:
 835       case os::cgc_thread:
 836       case os::watcher_thread:
 837         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 838         break;
 839       }
 840     }
 841 
 842     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 843     pthread_attr_setstacksize(&attr, stack_size);
 844   } else {
 845     // let pthread_create() pick the default value.
 846   }
 847 
 848   // glibc guard page
 849   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 850 
 851   ThreadState state;
 852 
 853   {
 854     // Serialize thread creation if we are running with fixed stack LinuxThreads
 855     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 856     if (lock) {
 857       os::Linux::createThread_lock()->lock_without_safepoint_check();
 858     }
 859 
 860     pthread_t tid;
 861     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 862 
 863     pthread_attr_destroy(&attr);
 864 
 865     if (ret != 0) {
 866       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 867         perror("pthread_create()");
 868       }
 869       // Need to clean up stuff we've allocated so far
 870       thread->set_osthread(NULL);
 871       delete osthread;
 872       if (lock) os::Linux::createThread_lock()->unlock();
 873       return false;
 874     }
 875 
 876     // Store pthread info into the OSThread
 877     osthread->set_pthread_id(tid);
 878 
 879     // Wait until child thread is either initialized or aborted
 880     {
 881       Monitor* sync_with_child = osthread->startThread_lock();
 882       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 883       while ((state = osthread->get_state()) == ALLOCATED) {
 884         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 885       }
 886     }
 887 
 888     if (lock) {
 889       os::Linux::createThread_lock()->unlock();
 890     }
 891   }
 892 
 893   // Aborted due to thread limit being reached
 894   if (state == ZOMBIE) {
 895       thread->set_osthread(NULL);
 896       delete osthread;
 897       return false;
 898   }
 899 
 900   // The thread is returned suspended (in state INITIALIZED),
 901   // and is started higher up in the call chain
 902   assert(state == INITIALIZED, "race condition");
 903   return true;
 904 }
 905 
 906 /////////////////////////////////////////////////////////////////////////////
 907 // attach existing thread
 908 
 909 // bootstrap the main thread
 910 bool os::create_main_thread(JavaThread* thread) {
 911   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 912   return create_attached_thread(thread);
 913 }
 914 
 915 bool os::create_attached_thread(JavaThread* thread) {
 916 #ifdef ASSERT
 917     thread->verify_not_published();
 918 #endif
 919 
 920   // Allocate the OSThread object
 921   OSThread* osthread = new OSThread(NULL, NULL);
 922 
 923   if (osthread == NULL) {
 924     return false;
 925   }
 926 
 927   // Store pthread info into the OSThread
 928   osthread->set_thread_id(os::Linux::gettid());
 929   osthread->set_pthread_id(::pthread_self());
 930 
 931   // initialize floating point control register
 932   os::Linux::init_thread_fpu_state();
 933 
 934   // Initial thread state is RUNNABLE
 935   osthread->set_state(RUNNABLE);
 936 
 937   thread->set_osthread(osthread);
 938 
 939   if (UseNUMA) {
 940     int lgrp_id = os::numa_get_group_id();
 941     if (lgrp_id != -1) {
 942       thread->set_lgrp_id(lgrp_id);
 943     }
 944   }
 945 
 946   if (os::Linux::is_initial_thread()) {
 947     // If current thread is initial thread, its stack is mapped on demand,
 948     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 949     // the entire stack region to avoid SEGV in stack banging.
 950     // It is also useful to get around the heap-stack-gap problem on SuSE
 951     // kernel (see 4821821 for details). We first expand stack to the top
 952     // of yellow zone, then enable stack yellow zone (order is significant,
 953     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 954     // is no gap between the last two virtual memory regions.
 955 
 956     JavaThread *jt = (JavaThread *)thread;
 957     address addr = jt->stack_yellow_zone_base();
 958     assert(addr != NULL, "initialization problem?");
 959     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 960 
 961     osthread->set_expanding_stack();
 962     os::Linux::manually_expand_stack(jt, addr);
 963     osthread->clear_expanding_stack();
 964   }
 965 
 966   // initialize signal mask for this thread
 967   // and save the caller's signal mask
 968   os::Linux::hotspot_sigmask(thread);
 969 
 970   return true;
 971 }
 972 
 973 void os::pd_start_thread(Thread* thread) {
 974   OSThread * osthread = thread->osthread();
 975   assert(osthread->get_state() != INITIALIZED, "just checking");
 976   Monitor* sync_with_child = osthread->startThread_lock();
 977   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 978   sync_with_child->notify();
 979 }
 980 
 981 // Free Linux resources related to the OSThread
 982 void os::free_thread(OSThread* osthread) {
 983   assert(osthread != NULL, "osthread not set");
 984 
 985   if (Thread::current()->osthread() == osthread) {
 986     // Restore caller's signal mask
 987     sigset_t sigmask = osthread->caller_sigmask();
 988     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 989    }
 990 
 991   delete osthread;
 992 }
 993 
 994 //////////////////////////////////////////////////////////////////////////////
 995 // thread local storage
 996 
 997 int os::allocate_thread_local_storage() {
 998   pthread_key_t key;
 999   int rslt = pthread_key_create(&key, NULL);
1000   assert(rslt == 0, "cannot allocate thread local storage");
1001   return (int)key;
1002 }
1003 
1004 // Note: This is currently not used by VM, as we don't destroy TLS key
1005 // on VM exit.
1006 void os::free_thread_local_storage(int index) {
1007   int rslt = pthread_key_delete((pthread_key_t)index);
1008   assert(rslt == 0, "invalid index");
1009 }
1010 
1011 void os::thread_local_storage_at_put(int index, void* value) {
1012   int rslt = pthread_setspecific((pthread_key_t)index, value);
1013   assert(rslt == 0, "pthread_setspecific failed");
1014 }
1015 
1016 extern "C" Thread* get_thread() {
1017   return ThreadLocalStorage::thread();
1018 }
1019 
1020 //////////////////////////////////////////////////////////////////////////////
1021 // initial thread
1022 
1023 // Check if current thread is the initial thread, similar to Solaris thr_main.
1024 bool os::Linux::is_initial_thread(void) {
1025   char dummy;
1026   // If called before init complete, thread stack bottom will be null.
1027   // Can be called if fatal error occurs before initialization.
1028   if (initial_thread_stack_bottom() == NULL) return false;
1029   assert(initial_thread_stack_bottom() != NULL &&
1030          initial_thread_stack_size()   != 0,
1031          "os::init did not locate initial thread's stack region");
1032   if ((address)&dummy >= initial_thread_stack_bottom() &&
1033       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1034        return true;
1035   else return false;
1036 }
1037 
1038 // Find the virtual memory area that contains addr
1039 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1040   FILE *fp = fopen("/proc/self/maps", "r");
1041   if (fp) {
1042     address low, high;
1043     while (!feof(fp)) {
1044       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1045         if (low <= addr && addr < high) {
1046            if (vma_low)  *vma_low  = low;
1047            if (vma_high) *vma_high = high;
1048            fclose (fp);
1049            return true;
1050         }
1051       }
1052       for (;;) {
1053         int ch = fgetc(fp);
1054         if (ch == EOF || ch == (int)'\n') break;
1055       }
1056     }
1057     fclose(fp);
1058   }
1059   return false;
1060 }
1061 
1062 // Locate initial thread stack. This special handling of initial thread stack
1063 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1064 // bogus value for initial thread.
1065 void os::Linux::capture_initial_stack(size_t max_size) {
1066   // stack size is the easy part, get it from RLIMIT_STACK
1067   size_t stack_size;
1068   struct rlimit rlim;
1069   getrlimit(RLIMIT_STACK, &rlim);
1070   stack_size = rlim.rlim_cur;
1071 
1072   // 6308388: a bug in ld.so will relocate its own .data section to the
1073   //   lower end of primordial stack; reduce ulimit -s value a little bit
1074   //   so we won't install guard page on ld.so's data section.
1075   stack_size -= 2 * page_size();
1076 
1077   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1078   //   7.1, in both cases we will get 2G in return value.
1079   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1080   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1081   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1082   //   in case other parts in glibc still assumes 2M max stack size.
1083   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1084 #ifndef IA64
1085   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1086 #else
1087   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1088   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1089 #endif
1090 
1091   // Try to figure out where the stack base (top) is. This is harder.
1092   //
1093   // When an application is started, glibc saves the initial stack pointer in
1094   // a global variable "__libc_stack_end", which is then used by system
1095   // libraries. __libc_stack_end should be pretty close to stack top. The
1096   // variable is available since the very early days. However, because it is
1097   // a private interface, it could disappear in the future.
1098   //
1099   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1100   // to __libc_stack_end, it is very close to stack top, but isn't the real
1101   // stack top. Note that /proc may not exist if VM is running as a chroot
1102   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1103   // /proc/<pid>/stat could change in the future (though unlikely).
1104   //
1105   // We try __libc_stack_end first. If that doesn't work, look for
1106   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1107   // as a hint, which should work well in most cases.
1108 
1109   uintptr_t stack_start;
1110 
1111   // try __libc_stack_end first
1112   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1113   if (p && *p) {
1114     stack_start = *p;
1115   } else {
1116     // see if we can get the start_stack field from /proc/self/stat
1117     FILE *fp;
1118     int pid;
1119     char state;
1120     int ppid;
1121     int pgrp;
1122     int session;
1123     int nr;
1124     int tpgrp;
1125     unsigned long flags;
1126     unsigned long minflt;
1127     unsigned long cminflt;
1128     unsigned long majflt;
1129     unsigned long cmajflt;
1130     unsigned long utime;
1131     unsigned long stime;
1132     long cutime;
1133     long cstime;
1134     long prio;
1135     long nice;
1136     long junk;
1137     long it_real;
1138     uintptr_t start;
1139     uintptr_t vsize;
1140     uintptr_t rss;
1141     unsigned long rsslim;
1142     uintptr_t scodes;
1143     uintptr_t ecode;
1144     int i;
1145 
1146     // Figure what the primordial thread stack base is. Code is inspired
1147     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1148     // followed by command name surrounded by parentheses, state, etc.
1149     char stat[2048];
1150     int statlen;
1151 
1152     fp = fopen("/proc/self/stat", "r");
1153     if (fp) {
1154       statlen = fread(stat, 1, 2047, fp);
1155       stat[statlen] = '\0';
1156       fclose(fp);
1157 
1158       // Skip pid and the command string. Note that we could be dealing with
1159       // weird command names, e.g. user could decide to rename java launcher
1160       // to "java 1.4.2 :)", then the stat file would look like
1161       //                1234 (java 1.4.2 :)) R ... ...
1162       // We don't really need to know the command string, just find the last
1163       // occurrence of ")" and then start parsing from there. See bug 4726580.
1164       char * s = strrchr(stat, ')');
1165 
1166       i = 0;
1167       if (s) {
1168         // Skip blank chars
1169         do s++; while (isspace(*s));
1170 
1171         /*                                     1   1   1   1   1   1   1   1   1   1   2   2   2   2   2   2   2   2   2 */
1172         /*              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 */
1173         i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld "
1174                    UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT
1175                    " %lu "
1176                    UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT,
1177              &state,          /* 3  %c  */
1178              &ppid,           /* 4  %d  */
1179              &pgrp,           /* 5  %d  */
1180              &session,        /* 6  %d  */
1181              &nr,             /* 7  %d  */
1182              &tpgrp,          /* 8  %d  */
1183              &flags,          /* 9  %lu  */
1184              &minflt,         /* 10 %lu  */
1185              &cminflt,        /* 11 %lu  */
1186              &majflt,         /* 12 %lu  */
1187              &cmajflt,        /* 13 %lu  */
1188              &utime,          /* 14 %lu  */
1189              &stime,          /* 15 %lu  */
1190              &cutime,         /* 16 %ld  */
1191              &cstime,         /* 17 %ld  */
1192              &prio,           /* 18 %ld  */
1193              &nice,           /* 19 %ld  */
1194              &junk,           /* 20 %ld  */
1195              &it_real,        /* 21 %ld  */
1196              &start,          /* 22 UINTX_FORMAT  */
1197              &vsize,          /* 23 UINTX_FORMAT  */
1198              &rss,            /* 24 UINTX_FORMAT  */
1199              &rsslim,         /* 25 %lu  */
1200              &scodes,         /* 26 UINTX_FORMAT  */
1201              &ecode,          /* 27 UINTX_FORMAT  */
1202              &stack_start);   /* 28 UINTX_FORMAT  */
1203       }
1204 
1205       if (i != 28 - 2) {
1206          assert(false, "Bad conversion from /proc/self/stat");
1207          // product mode - assume we are the initial thread, good luck in the
1208          // embedded case.
1209          warning("Can't detect initial thread stack location - bad conversion");
1210          stack_start = (uintptr_t) &rlim;
1211       }
1212     } else {
1213       // For some reason we can't open /proc/self/stat (for example, running on
1214       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1215       // most cases, so don't abort:
1216       warning("Can't detect initial thread stack location - no /proc/self/stat");
1217       stack_start = (uintptr_t) &rlim;
1218     }
1219   }
1220 
1221   // Now we have a pointer (stack_start) very close to the stack top, the
1222   // next thing to do is to figure out the exact location of stack top. We
1223   // can find out the virtual memory area that contains stack_start by
1224   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1225   // and its upper limit is the real stack top. (again, this would fail if
1226   // running inside chroot, because /proc may not exist.)
1227 
1228   uintptr_t stack_top;
1229   address low, high;
1230   if (find_vma((address)stack_start, &low, &high)) {
1231     // success, "high" is the true stack top. (ignore "low", because initial
1232     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1233     stack_top = (uintptr_t)high;
1234   } else {
1235     // failed, likely because /proc/self/maps does not exist
1236     warning("Can't detect initial thread stack location - find_vma failed");
1237     // best effort: stack_start is normally within a few pages below the real
1238     // stack top, use it as stack top, and reduce stack size so we won't put
1239     // guard page outside stack.
1240     stack_top = stack_start;
1241     stack_size -= 16 * page_size();
1242   }
1243 
1244   // stack_top could be partially down the page so align it
1245   stack_top = align_size_up(stack_top, page_size());
1246 
1247   if (max_size && stack_size > max_size) {
1248      _initial_thread_stack_size = max_size;
1249   } else {
1250      _initial_thread_stack_size = stack_size;
1251   }
1252 
1253   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1254   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1255 }
1256 
1257 ////////////////////////////////////////////////////////////////////////////////
1258 // time support
1259 
1260 // Time since start-up in seconds to a fine granularity.
1261 // Used by VMSelfDestructTimer and the MemProfiler.
1262 double os::elapsedTime() {
1263 
1264   return (double)(os::elapsed_counter()) * 0.000001;
1265 }
1266 
1267 jlong os::elapsed_counter() {
1268   timeval time;
1269   int status = gettimeofday(&time, NULL);
1270   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1271 }
1272 
1273 jlong os::elapsed_frequency() {
1274   return (1000 * 1000);
1275 }
1276 
1277 // For now, we say that linux does not support vtime.  I have no idea
1278 // whether it can actually be made to (DLD, 9/13/05).
1279 
1280 bool os::supports_vtime() { return false; }
1281 bool os::enable_vtime()   { return false; }
1282 bool os::vtime_enabled()  { return false; }
1283 double os::elapsedVTime() {
1284   // better than nothing, but not much
1285   return elapsedTime();
1286 }
1287 
1288 jlong os::javaTimeMillis() {
1289   timeval time;
1290   int status = gettimeofday(&time, NULL);
1291   assert(status != -1, "linux error");
1292   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1293 }
1294 
1295 #ifndef CLOCK_MONOTONIC
1296 #define CLOCK_MONOTONIC (1)
1297 #endif
1298 
1299 void os::Linux::clock_init() {
1300   // we do dlopen's in this particular order due to bug in linux
1301   // dynamical loader (see 6348968) leading to crash on exit
1302   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1303   if (handle == NULL) {
1304     handle = dlopen("librt.so", RTLD_LAZY);
1305   }
1306 
1307   if (handle) {
1308     int (*clock_getres_func)(clockid_t, struct timespec*) =
1309            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1310     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1311            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1312     if (clock_getres_func && clock_gettime_func) {
1313       // See if monotonic clock is supported by the kernel. Note that some
1314       // early implementations simply return kernel jiffies (updated every
1315       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1316       // for nano time (though the monotonic property is still nice to have).
1317       // It's fixed in newer kernels, however clock_getres() still returns
1318       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1319       // resolution for now. Hopefully as people move to new kernels, this
1320       // won't be a problem.
1321       struct timespec res;
1322       struct timespec tp;
1323       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1324           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1325         // yes, monotonic clock is supported
1326         _clock_gettime = clock_gettime_func;
1327       } else {
1328         // close librt if there is no monotonic clock
1329         dlclose(handle);
1330       }
1331     }
1332   }
1333 }
1334 
1335 #ifndef SYS_clock_getres
1336 
1337 #if defined(IA32) || defined(AMD64)
1338 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1339 #else
1340 #error Value of SYS_clock_getres not known on this platform
1341 #endif
1342 
1343 #endif
1344 
1345 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1346 
1347 void os::Linux::fast_thread_clock_init() {
1348   if (!UseLinuxPosixThreadCPUClocks) {
1349     return;
1350   }
1351   clockid_t clockid;
1352   struct timespec tp;
1353   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1354       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1355 
1356   // Switch to using fast clocks for thread cpu time if
1357   // the sys_clock_getres() returns 0 error code.
1358   // Note, that some kernels may support the current thread
1359   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1360   // returned by the pthread_getcpuclockid().
1361   // If the fast Posix clocks are supported then the sys_clock_getres()
1362   // must return at least tp.tv_sec == 0 which means a resolution
1363   // better than 1 sec. This is extra check for reliability.
1364 
1365   if(pthread_getcpuclockid_func &&
1366      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1367      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1368 
1369     _supports_fast_thread_cpu_time = true;
1370     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1371   }
1372 }
1373 
1374 jlong os::javaTimeNanos() {
1375   if (Linux::supports_monotonic_clock()) {
1376     struct timespec tp;
1377     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1378     assert(status == 0, "gettime error");
1379     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1380     return result;
1381   } else {
1382     timeval time;
1383     int status = gettimeofday(&time, NULL);
1384     assert(status != -1, "linux error");
1385     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1386     return 1000 * usecs;
1387   }
1388 }
1389 
1390 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1391   if (Linux::supports_monotonic_clock()) {
1392     info_ptr->max_value = ALL_64_BITS;
1393 
1394     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1395     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1396     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1397   } else {
1398     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1399     info_ptr->max_value = ALL_64_BITS;
1400 
1401     // gettimeofday is a real time clock so it skips
1402     info_ptr->may_skip_backward = true;
1403     info_ptr->may_skip_forward = true;
1404   }
1405 
1406   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1407 }
1408 
1409 // Return the real, user, and system times in seconds from an
1410 // arbitrary fixed point in the past.
1411 bool os::getTimesSecs(double* process_real_time,
1412                       double* process_user_time,
1413                       double* process_system_time) {
1414   struct tms ticks;
1415   clock_t real_ticks = times(&ticks);
1416 
1417   if (real_ticks == (clock_t) (-1)) {
1418     return false;
1419   } else {
1420     double ticks_per_second = (double) clock_tics_per_sec;
1421     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1422     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1423     *process_real_time = ((double) real_ticks) / ticks_per_second;
1424 
1425     return true;
1426   }
1427 }
1428 
1429 
1430 char * os::local_time_string(char *buf, size_t buflen) {
1431   struct tm t;
1432   time_t long_time;
1433   time(&long_time);
1434   localtime_r(&long_time, &t);
1435   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1436                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1437                t.tm_hour, t.tm_min, t.tm_sec);
1438   return buf;
1439 }
1440 
1441 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1442   return localtime_r(clock, res);
1443 }
1444 
1445 ////////////////////////////////////////////////////////////////////////////////
1446 // runtime exit support
1447 
1448 // Note: os::shutdown() might be called very early during initialization, or
1449 // called from signal handler. Before adding something to os::shutdown(), make
1450 // sure it is async-safe and can handle partially initialized VM.
1451 void os::shutdown() {
1452 
1453   // allow PerfMemory to attempt cleanup of any persistent resources
1454   perfMemory_exit();
1455 
1456   // needs to remove object in file system
1457   AttachListener::abort();
1458 
1459   // flush buffered output, finish log files
1460   ostream_abort();
1461 
1462   // Check for abort hook
1463   abort_hook_t abort_hook = Arguments::abort_hook();
1464   if (abort_hook != NULL) {
1465     abort_hook();
1466   }
1467 
1468 }
1469 
1470 // Note: os::abort() might be called very early during initialization, or
1471 // called from signal handler. Before adding something to os::abort(), make
1472 // sure it is async-safe and can handle partially initialized VM.
1473 void os::abort(bool dump_core) {
1474   os::shutdown();
1475   if (dump_core) {
1476 #ifndef PRODUCT
1477     fdStream out(defaultStream::output_fd());
1478     out.print_raw("Current thread is ");
1479     char buf[16];
1480     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1481     out.print_raw_cr(buf);
1482     out.print_raw_cr("Dumping core ...");
1483 #endif
1484     ::abort(); // dump core
1485   }
1486 
1487   ::exit(1);
1488 }
1489 
1490 // Die immediately, no exit hook, no abort hook, no cleanup.
1491 void os::die() {
1492   // _exit() on LinuxThreads only kills current thread
1493   ::abort();
1494 }
1495 
1496 // unused on linux for now.
1497 void os::set_error_file(const char *logfile) {}
1498 
1499 intx os::current_thread_id() { return (intx)pthread_self(); }
1500 int os::current_process_id() {
1501 
1502   // Under the old linux thread library, linux gives each thread
1503   // its own process id. Because of this each thread will return
1504   // a different pid if this method were to return the result
1505   // of getpid(2). Linux provides no api that returns the pid
1506   // of the launcher thread for the vm. This implementation
1507   // returns a unique pid, the pid of the launcher thread
1508   // that starts the vm 'process'.
1509 
1510   // Under the NPTL, getpid() returns the same pid as the
1511   // launcher thread rather than a unique pid per thread.
1512   // Use gettid() if you want the old pre NPTL behaviour.
1513 
1514   // if you are looking for the result of a call to getpid() that
1515   // returns a unique pid for the calling thread, then look at the
1516   // OSThread::thread_id() method in osThread_linux.hpp file
1517 
1518   return (int)(_initial_pid ? _initial_pid : getpid());
1519 }
1520 
1521 // DLL functions
1522 
1523 const char* os::dll_file_extension() { return ".so"; }
1524 
1525 const char* os::get_temp_directory() { return "/tmp/"; }
1526 
1527 static bool file_exists(const char* filename) {
1528   struct stat statbuf;
1529   if (filename == NULL || strlen(filename) == 0) {
1530     return false;
1531   }
1532   return os::stat(filename, &statbuf) == 0;
1533 }
1534 
1535 void os::dll_build_name(char* buffer, size_t buflen,
1536                         const char* pname, const char* fname) {
1537   // Copied from libhpi
1538   const size_t pnamelen = pname ? strlen(pname) : 0;
1539 
1540   // Quietly truncate on buffer overflow.  Should be an error.
1541   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1542       *buffer = '\0';
1543       return;
1544   }
1545 
1546   if (pnamelen == 0) {
1547     snprintf(buffer, buflen, "lib%s.so", fname);
1548   } else if (strchr(pname, *os::path_separator()) != NULL) {
1549     int n;
1550     char** pelements = split_path(pname, &n);
1551     for (int i = 0 ; i < n ; i++) {
1552       // Really shouldn't be NULL, but check can't hurt
1553       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1554         continue; // skip the empty path values
1555       }
1556       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1557       if (file_exists(buffer)) {
1558         break;
1559       }
1560     }
1561     // release the storage
1562     for (int i = 0 ; i < n ; i++) {
1563       if (pelements[i] != NULL) {
1564         FREE_C_HEAP_ARRAY(char, pelements[i]);
1565       }
1566     }
1567     if (pelements != NULL) {
1568       FREE_C_HEAP_ARRAY(char*, pelements);
1569     }
1570   } else {
1571     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1572   }
1573 }
1574 
1575 const char* os::get_current_directory(char *buf, int buflen) {
1576   return getcwd(buf, buflen);
1577 }
1578 
1579 // check if addr is inside libjvm[_g].so
1580 bool os::address_is_in_vm(address addr) {
1581   static address libjvm_base_addr;
1582   Dl_info dlinfo;
1583 
1584   if (libjvm_base_addr == NULL) {
1585     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1586     libjvm_base_addr = (address)dlinfo.dli_fbase;
1587     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1588   }
1589 
1590   if (dladdr((void *)addr, &dlinfo)) {
1591     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1592   }
1593 
1594   return false;
1595 }
1596 
1597 bool os::dll_address_to_function_name(address addr, char *buf,
1598                                       int buflen, int *offset) {
1599   Dl_info dlinfo;
1600 
1601   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1602     if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1603     if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1604     return true;
1605   } else {
1606     if (buf) buf[0] = '\0';
1607     if (offset) *offset = -1;
1608     return false;
1609   }
1610 }
1611 
1612 struct _address_to_library_name {
1613   address addr;          // input : memory address
1614   size_t  buflen;        //         size of fname
1615   char*   fname;         // output: library name
1616   address base;          //         library base addr
1617 };
1618 
1619 static int address_to_library_name_callback(struct dl_phdr_info *info,
1620                                             size_t size, void *data) {
1621   int i;
1622   bool found = false;
1623   address libbase = NULL;
1624   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1625 
1626   // iterate through all loadable segments
1627   for (i = 0; i < info->dlpi_phnum; i++) {
1628     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1629     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1630       // base address of a library is the lowest address of its loaded
1631       // segments.
1632       if (libbase == NULL || libbase > segbase) {
1633         libbase = segbase;
1634       }
1635       // see if 'addr' is within current segment
1636       if (segbase <= d->addr &&
1637           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1638         found = true;
1639       }
1640     }
1641   }
1642 
1643   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1644   // so dll_address_to_library_name() can fall through to use dladdr() which
1645   // can figure out executable name from argv[0].
1646   if (found && info->dlpi_name && info->dlpi_name[0]) {
1647     d->base = libbase;
1648     if (d->fname) {
1649       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1650     }
1651     return 1;
1652   }
1653   return 0;
1654 }
1655 
1656 bool os::dll_address_to_library_name(address addr, char* buf,
1657                                      int buflen, int* offset) {
1658   Dl_info dlinfo;
1659   struct _address_to_library_name data;
1660 
1661   // There is a bug in old glibc dladdr() implementation that it could resolve
1662   // to wrong library name if the .so file has a base address != NULL. Here
1663   // we iterate through the program headers of all loaded libraries to find
1664   // out which library 'addr' really belongs to. This workaround can be
1665   // removed once the minimum requirement for glibc is moved to 2.3.x.
1666   data.addr = addr;
1667   data.fname = buf;
1668   data.buflen = buflen;
1669   data.base = NULL;
1670   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1671 
1672   if (rslt) {
1673      // buf already contains library name
1674      if (offset) *offset = addr - data.base;
1675      return true;
1676   } else if (dladdr((void*)addr, &dlinfo)){
1677      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1678      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1679      return true;
1680   } else {
1681      if (buf) buf[0] = '\0';
1682      if (offset) *offset = -1;
1683      return false;
1684   }
1685 }
1686 
1687   // Loads .dll/.so and
1688   // in case of error it checks if .dll/.so was built for the
1689   // same architecture as Hotspot is running on
1690 
1691 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1692 {
1693   void * result= ::dlopen(filename, RTLD_LAZY);
1694   if (result != NULL) {
1695     // Successful loading
1696     return result;
1697   }
1698 
1699   Elf32_Ehdr elf_head;
1700 
1701   // Read system error message into ebuf
1702   // It may or may not be overwritten below
1703   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1704   ebuf[ebuflen-1]='\0';
1705   int diag_msg_max_length=ebuflen-strlen(ebuf);
1706   char* diag_msg_buf=ebuf+strlen(ebuf);
1707 
1708   if (diag_msg_max_length==0) {
1709     // No more space in ebuf for additional diagnostics message
1710     return NULL;
1711   }
1712 
1713 
1714   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1715 
1716   if (file_descriptor < 0) {
1717     // Can't open library, report dlerror() message
1718     return NULL;
1719   }
1720 
1721   bool failed_to_read_elf_head=
1722     (sizeof(elf_head)!=
1723         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1724 
1725   ::close(file_descriptor);
1726   if (failed_to_read_elf_head) {
1727     // file i/o error - report dlerror() msg
1728     return NULL;
1729   }
1730 
1731   typedef struct {
1732     Elf32_Half  code;         // Actual value as defined in elf.h
1733     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1734     char        elf_class;    // 32 or 64 bit
1735     char        endianess;    // MSB or LSB
1736     char*       name;         // String representation
1737   } arch_t;
1738 
1739   #ifndef EM_486
1740   #define EM_486          6               /* Intel 80486 */
1741   #endif
1742 
1743   static const arch_t arch_array[]={
1744     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1745     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1746     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1747     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1748     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1749     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1750     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1751     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1752     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1753     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1754     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1755     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1756     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1757     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1758     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1759     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1760   };
1761 
1762   #if  (defined IA32)
1763     static  Elf32_Half running_arch_code=EM_386;
1764   #elif   (defined AMD64)
1765     static  Elf32_Half running_arch_code=EM_X86_64;
1766   #elif  (defined IA64)
1767     static  Elf32_Half running_arch_code=EM_IA_64;
1768   #elif  (defined __sparc) && (defined _LP64)
1769     static  Elf32_Half running_arch_code=EM_SPARCV9;
1770   #elif  (defined __sparc) && (!defined _LP64)
1771     static  Elf32_Half running_arch_code=EM_SPARC;
1772   #elif  (defined __powerpc64__)
1773     static  Elf32_Half running_arch_code=EM_PPC64;
1774   #elif  (defined __powerpc__)
1775     static  Elf32_Half running_arch_code=EM_PPC;
1776   #elif  (defined ARM)
1777     static  Elf32_Half running_arch_code=EM_ARM;
1778   #elif  (defined S390)
1779     static  Elf32_Half running_arch_code=EM_S390;
1780   #elif  (defined ALPHA)
1781     static  Elf32_Half running_arch_code=EM_ALPHA;
1782   #elif  (defined MIPSEL)
1783     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1784   #elif  (defined PARISC)
1785     static  Elf32_Half running_arch_code=EM_PARISC;
1786   #elif  (defined MIPS)
1787     static  Elf32_Half running_arch_code=EM_MIPS;
1788   #elif  (defined M68K)
1789     static  Elf32_Half running_arch_code=EM_68K;
1790   #else
1791     #error Method os::dll_load requires that one of following is defined:\
1792          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1793   #endif
1794 
1795   // Identify compatability class for VM's architecture and library's architecture
1796   // Obtain string descriptions for architectures
1797 
1798   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1799   int running_arch_index=-1;
1800 
1801   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1802     if (running_arch_code == arch_array[i].code) {
1803       running_arch_index    = i;
1804     }
1805     if (lib_arch.code == arch_array[i].code) {
1806       lib_arch.compat_class = arch_array[i].compat_class;
1807       lib_arch.name         = arch_array[i].name;
1808     }
1809   }
1810 
1811   assert(running_arch_index != -1,
1812     "Didn't find running architecture code (running_arch_code) in arch_array");
1813   if (running_arch_index == -1) {
1814     // Even though running architecture detection failed
1815     // we may still continue with reporting dlerror() message
1816     return NULL;
1817   }
1818 
1819   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1820     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1821     return NULL;
1822   }
1823 
1824 #ifndef S390
1825   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1826     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1827     return NULL;
1828   }
1829 #endif // !S390
1830 
1831   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1832     if ( lib_arch.name!=NULL ) {
1833       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1834         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1835         lib_arch.name, arch_array[running_arch_index].name);
1836     } else {
1837       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1838       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1839         lib_arch.code,
1840         arch_array[running_arch_index].name);
1841     }
1842   }
1843 
1844   return NULL;
1845 }
1846 
1847 /*
1848  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
1849  * chances are you might want to run the generated bits against glibc-2.0
1850  * libdl.so, so always use locking for any version of glibc.
1851  */
1852 void* os::dll_lookup(void* handle, const char* name) {
1853   pthread_mutex_lock(&dl_mutex);
1854   void* res = dlsym(handle, name);
1855   pthread_mutex_unlock(&dl_mutex);
1856   return res;
1857 }
1858 
1859 
1860 bool _print_ascii_file(const char* filename, outputStream* st) {
1861   int fd = open(filename, O_RDONLY);
1862   if (fd == -1) {
1863      return false;
1864   }
1865 
1866   char buf[32];
1867   int bytes;
1868   while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1869     st->print_raw(buf, bytes);
1870   }
1871 
1872   close(fd);
1873 
1874   return true;
1875 }
1876 
1877 void os::print_dll_info(outputStream *st) {
1878    st->print_cr("Dynamic libraries:");
1879 
1880    char fname[32];
1881    pid_t pid = os::Linux::gettid();
1882 
1883    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1884 
1885    if (!_print_ascii_file(fname, st)) {
1886      st->print("Can not get library information for pid = %d\n", pid);
1887    }
1888 }
1889 
1890 
1891 void os::print_os_info(outputStream* st) {
1892   st->print("OS:");
1893 
1894   // Try to identify popular distros.
1895   // Most Linux distributions have /etc/XXX-release file, which contains
1896   // the OS version string. Some have more than one /etc/XXX-release file
1897   // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1898   // so the order is important.
1899   if (!_print_ascii_file("/etc/mandrake-release", st) &&
1900       !_print_ascii_file("/etc/sun-release", st) &&
1901       !_print_ascii_file("/etc/redhat-release", st) &&
1902       !_print_ascii_file("/etc/SuSE-release", st) &&
1903       !_print_ascii_file("/etc/turbolinux-release", st) &&
1904       !_print_ascii_file("/etc/gentoo-release", st) &&
1905       !_print_ascii_file("/etc/debian_version", st)) {
1906       st->print("Linux");
1907   }
1908   st->cr();
1909 
1910   // kernel
1911   st->print("uname:");
1912   struct utsname name;
1913   uname(&name);
1914   st->print(name.sysname); st->print(" ");
1915   st->print(name.release); st->print(" ");
1916   st->print(name.version); st->print(" ");
1917   st->print(name.machine);
1918   st->cr();
1919 
1920   // Print warning if unsafe chroot environment detected
1921   if (unsafe_chroot_detected) {
1922     st->print("WARNING!! ");
1923     st->print_cr(unstable_chroot_error);
1924   }
1925 
1926   // libc, pthread
1927   st->print("libc:");
1928   st->print(os::Linux::glibc_version()); st->print(" ");
1929   st->print(os::Linux::libpthread_version()); st->print(" ");
1930   if (os::Linux::is_LinuxThreads()) {
1931      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
1932   }
1933   st->cr();
1934 
1935   // rlimit
1936   st->print("rlimit:");
1937   struct rlimit rlim;
1938 
1939   st->print(" STACK ");
1940   getrlimit(RLIMIT_STACK, &rlim);
1941   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1942   else st->print("%uk", rlim.rlim_cur >> 10);
1943 
1944   st->print(", CORE ");
1945   getrlimit(RLIMIT_CORE, &rlim);
1946   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1947   else st->print("%uk", rlim.rlim_cur >> 10);
1948 
1949   st->print(", NPROC ");
1950   getrlimit(RLIMIT_NPROC, &rlim);
1951   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1952   else st->print("%d", rlim.rlim_cur);
1953 
1954   st->print(", NOFILE ");
1955   getrlimit(RLIMIT_NOFILE, &rlim);
1956   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1957   else st->print("%d", rlim.rlim_cur);
1958 
1959   st->print(", AS ");
1960   getrlimit(RLIMIT_AS, &rlim);
1961   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1962   else st->print("%uk", rlim.rlim_cur >> 10);
1963   st->cr();
1964 
1965   // load average
1966   st->print("load average:");
1967   double loadavg[3];
1968   os::loadavg(loadavg, 3);
1969   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1970   st->cr();
1971 }
1972 
1973 void os::print_memory_info(outputStream* st) {
1974 
1975   st->print("Memory:");
1976   st->print(" %dk page", os::vm_page_size()>>10);
1977 
1978   // values in struct sysinfo are "unsigned long"
1979   struct sysinfo si;
1980   sysinfo(&si);
1981 
1982   st->print(", physical " UINT64_FORMAT "k",
1983             os::physical_memory() >> 10);
1984   st->print("(" UINT64_FORMAT "k free)",
1985             os::available_memory() >> 10);
1986   st->print(", swap " UINT64_FORMAT "k",
1987             ((jlong)si.totalswap * si.mem_unit) >> 10);
1988   st->print("(" UINT64_FORMAT "k free)",
1989             ((jlong)si.freeswap * si.mem_unit) >> 10);
1990   st->cr();
1991 }
1992 
1993 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
1994 // but they're the same for all the linux arch that we support
1995 // and they're the same for solaris but there's no common place to put this.
1996 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
1997                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
1998                           "ILL_COPROC", "ILL_BADSTK" };
1999 
2000 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2001                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2002                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2003 
2004 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2005 
2006 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2007 
2008 void os::print_siginfo(outputStream* st, void* siginfo) {
2009   st->print("siginfo:");
2010 
2011   const int buflen = 100;
2012   char buf[buflen];
2013   siginfo_t *si = (siginfo_t*)siginfo;
2014   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2015   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2016     st->print("si_errno=%s", buf);
2017   } else {
2018     st->print("si_errno=%d", si->si_errno);
2019   }
2020   const int c = si->si_code;
2021   assert(c > 0, "unexpected si_code");
2022   switch (si->si_signo) {
2023   case SIGILL:
2024     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2025     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2026     break;
2027   case SIGFPE:
2028     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2029     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2030     break;
2031   case SIGSEGV:
2032     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2033     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2034     break;
2035   case SIGBUS:
2036     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2037     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2038     break;
2039   default:
2040     st->print(", si_code=%d", si->si_code);
2041     // no si_addr
2042   }
2043 
2044   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2045       UseSharedSpaces) {
2046     FileMapInfo* mapinfo = FileMapInfo::current_info();
2047     if (mapinfo->is_in_shared_space(si->si_addr)) {
2048       st->print("\n\nError accessing class data sharing archive."   \
2049                 " Mapped file inaccessible during execution, "      \
2050                 " possible disk/network problem.");
2051     }
2052   }
2053   st->cr();
2054 }
2055 
2056 
2057 static void print_signal_handler(outputStream* st, int sig,
2058                                  char* buf, size_t buflen);
2059 
2060 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2061   st->print_cr("Signal Handlers:");
2062   print_signal_handler(st, SIGSEGV, buf, buflen);
2063   print_signal_handler(st, SIGBUS , buf, buflen);
2064   print_signal_handler(st, SIGFPE , buf, buflen);
2065   print_signal_handler(st, SIGPIPE, buf, buflen);
2066   print_signal_handler(st, SIGXFSZ, buf, buflen);
2067   print_signal_handler(st, SIGILL , buf, buflen);
2068   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2069   print_signal_handler(st, SR_signum, buf, buflen);
2070   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2071   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2072   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2073   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2074 }
2075 
2076 static char saved_jvm_path[MAXPATHLEN] = {0};
2077 
2078 // Find the full path to the current module, libjvm.so or libjvm_g.so
2079 void os::jvm_path(char *buf, jint len) {
2080   // Error checking.
2081   if (len < MAXPATHLEN) {
2082     assert(false, "must use a large-enough buffer");
2083     buf[0] = '\0';
2084     return;
2085   }
2086   // Lazy resolve the path to current module.
2087   if (saved_jvm_path[0] != 0) {
2088     strcpy(buf, saved_jvm_path);
2089     return;
2090   }
2091 
2092   char dli_fname[MAXPATHLEN];
2093   bool ret = dll_address_to_library_name(
2094                 CAST_FROM_FN_PTR(address, os::jvm_path),
2095                 dli_fname, sizeof(dli_fname), NULL);
2096   assert(ret != 0, "cannot locate libjvm");
2097   if (realpath(dli_fname, buf) == NULL)
2098     return;
2099 
2100   if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2101     // Support for the gamma launcher.  Typical value for buf is
2102     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2103     // the right place in the string, then assume we are installed in a JDK and
2104     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2105     // up the path so it looks like libjvm.so is installed there (append a
2106     // fake suffix hotspot/libjvm.so).
2107     const char *p = buf + strlen(buf) - 1;
2108     for (int count = 0; p > buf && count < 5; ++count) {
2109       for (--p; p > buf && *p != '/'; --p)
2110         /* empty */ ;
2111     }
2112 
2113     if (strncmp(p, "/jre/lib/", 9) != 0) {
2114       // Look for JAVA_HOME in the environment.
2115       char* java_home_var = ::getenv("JAVA_HOME");
2116       if (java_home_var != NULL && java_home_var[0] != 0) {
2117         // Check the current module name "libjvm.so" or "libjvm_g.so".
2118         p = strrchr(buf, '/');
2119         assert(strstr(p, "/libjvm") == p, "invalid library name");
2120         p = strstr(p, "_g") ? "_g" : "";
2121 
2122         if (realpath(java_home_var, buf) == NULL)
2123           return;
2124         sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch);
2125         if (0 == access(buf, F_OK)) {
2126           // Use current module name "libjvm[_g].so" instead of
2127           // "libjvm"debug_only("_g")".so" since for fastdebug version
2128           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2129           // It is used when we are choosing the HPI library's name
2130           // "libhpi[_g].so" in hpi::initialize_get_interface().
2131           sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p);
2132         } else {
2133           // Go back to path of .so
2134           if (realpath(dli_fname, buf) == NULL)
2135             return;
2136         }
2137       }
2138     }
2139   }
2140 
2141   strcpy(saved_jvm_path, buf);
2142 }
2143 
2144 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2145   // no prefix required, not even "_"
2146 }
2147 
2148 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2149   // no suffix required
2150 }
2151 
2152 ////////////////////////////////////////////////////////////////////////////////
2153 // sun.misc.Signal support
2154 
2155 static volatile jint sigint_count = 0;
2156 
2157 static void
2158 UserHandler(int sig, void *siginfo, void *context) {
2159   // 4511530 - sem_post is serialized and handled by the manager thread. When
2160   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2161   // don't want to flood the manager thread with sem_post requests.
2162   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2163       return;
2164 
2165   // Ctrl-C is pressed during error reporting, likely because the error
2166   // handler fails to abort. Let VM die immediately.
2167   if (sig == SIGINT && is_error_reported()) {
2168      os::die();
2169   }
2170 
2171   os::signal_notify(sig);
2172 }
2173 
2174 void* os::user_handler() {
2175   return CAST_FROM_FN_PTR(void*, UserHandler);
2176 }
2177 
2178 extern "C" {
2179   typedef void (*sa_handler_t)(int);
2180   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2181 }
2182 
2183 void* os::signal(int signal_number, void* handler) {
2184   struct sigaction sigAct, oldSigAct;
2185 
2186   sigfillset(&(sigAct.sa_mask));
2187   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2188   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2189 
2190   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2191     // -1 means registration failed
2192     return (void *)-1;
2193   }
2194 
2195   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2196 }
2197 
2198 void os::signal_raise(int signal_number) {
2199   ::raise(signal_number);
2200 }
2201 
2202 /*
2203  * The following code is moved from os.cpp for making this
2204  * code platform specific, which it is by its very nature.
2205  */
2206 
2207 // Will be modified when max signal is changed to be dynamic
2208 int os::sigexitnum_pd() {
2209   return NSIG;
2210 }
2211 
2212 // a counter for each possible signal value
2213 static volatile jint pending_signals[NSIG+1] = { 0 };
2214 
2215 // Linux(POSIX) specific hand shaking semaphore.
2216 static sem_t sig_sem;
2217 
2218 void os::signal_init_pd() {
2219   // Initialize signal structures
2220   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2221 
2222   // Initialize signal semaphore
2223   ::sem_init(&sig_sem, 0, 0);
2224 }
2225 
2226 void os::signal_notify(int sig) {
2227   Atomic::inc(&pending_signals[sig]);
2228   ::sem_post(&sig_sem);
2229 }
2230 
2231 static int check_pending_signals(bool wait) {
2232   Atomic::store(0, &sigint_count);
2233   for (;;) {
2234     for (int i = 0; i < NSIG + 1; i++) {
2235       jint n = pending_signals[i];
2236       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2237         return i;
2238       }
2239     }
2240     if (!wait) {
2241       return -1;
2242     }
2243     JavaThread *thread = JavaThread::current();
2244     ThreadBlockInVM tbivm(thread);
2245 
2246     bool threadIsSuspended;
2247     do {
2248       thread->set_suspend_equivalent();
2249       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2250       ::sem_wait(&sig_sem);
2251 
2252       // were we externally suspended while we were waiting?
2253       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2254       if (threadIsSuspended) {
2255         //
2256         // The semaphore has been incremented, but while we were waiting
2257         // another thread suspended us. We don't want to continue running
2258         // while suspended because that would surprise the thread that
2259         // suspended us.
2260         //
2261         ::sem_post(&sig_sem);
2262 
2263         thread->java_suspend_self();
2264       }
2265     } while (threadIsSuspended);
2266   }
2267 }
2268 
2269 int os::signal_lookup() {
2270   return check_pending_signals(false);
2271 }
2272 
2273 int os::signal_wait() {
2274   return check_pending_signals(true);
2275 }
2276 
2277 ////////////////////////////////////////////////////////////////////////////////
2278 // Virtual Memory
2279 
2280 int os::vm_page_size() {
2281   // Seems redundant as all get out
2282   assert(os::Linux::page_size() != -1, "must call os::init");
2283   return os::Linux::page_size();
2284 }
2285 
2286 // Solaris allocates memory by pages.
2287 int os::vm_allocation_granularity() {
2288   assert(os::Linux::page_size() != -1, "must call os::init");
2289   return os::Linux::page_size();
2290 }
2291 
2292 // Rationale behind this function:
2293 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2294 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2295 //  samples for JITted code. Here we create private executable mapping over the code cache
2296 //  and then we can use standard (well, almost, as mapping can change) way to provide
2297 //  info for the reporting script by storing timestamp and location of symbol
2298 void linux_wrap_code(char* base, size_t size) {
2299   static volatile jint cnt = 0;
2300 
2301   if (!UseOprofile) {
2302     return;
2303   }
2304 
2305   char buf[40];
2306   int num = Atomic::add(1, &cnt);
2307 
2308   sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num);
2309   unlink(buf);
2310 
2311   int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2312 
2313   if (fd != -1) {
2314     off_t rv = lseek(fd, size-2, SEEK_SET);
2315     if (rv != (off_t)-1) {
2316       if (write(fd, "", 1) == 1) {
2317         mmap(base, size,
2318              PROT_READ|PROT_WRITE|PROT_EXEC,
2319              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2320       }
2321     }
2322     close(fd);
2323     unlink(buf);
2324   }
2325 }
2326 
2327 // NOTE: Linux kernel does not really reserve the pages for us.
2328 //       All it does is to check if there are enough free pages
2329 //       left at the time of mmap(). This could be a potential
2330 //       problem.
2331 bool os::commit_memory(char* addr, size_t size, bool exec) {
2332   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2333   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2334                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2335   return res != (uintptr_t) MAP_FAILED;
2336 }
2337 
2338 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2339                        bool exec) {
2340   return commit_memory(addr, size, exec);
2341 }
2342 
2343 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2344 
2345 void os::free_memory(char *addr, size_t bytes) {
2346   ::mmap(addr, bytes, PROT_READ | PROT_WRITE,
2347          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2348 }
2349 
2350 void os::numa_make_global(char *addr, size_t bytes) {
2351   Linux::numa_interleave_memory(addr, bytes);
2352 }
2353 
2354 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2355   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2356 }
2357 
2358 bool os::numa_topology_changed()   { return false; }
2359 
2360 size_t os::numa_get_groups_num() {
2361   int max_node = Linux::numa_max_node();
2362   return max_node > 0 ? max_node + 1 : 1;
2363 }
2364 
2365 int os::numa_get_group_id() {
2366   int cpu_id = Linux::sched_getcpu();
2367   if (cpu_id != -1) {
2368     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2369     if (lgrp_id != -1) {
2370       return lgrp_id;
2371     }
2372   }
2373   return 0;
2374 }
2375 
2376 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2377   for (size_t i = 0; i < size; i++) {
2378     ids[i] = i;
2379   }
2380   return size;
2381 }
2382 
2383 bool os::get_page_info(char *start, page_info* info) {
2384   return false;
2385 }
2386 
2387 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2388   return end;
2389 }
2390 
2391 extern "C" void numa_warn(int number, char *where, ...) { }
2392 extern "C" void numa_error(char *where) { }
2393 
2394 
2395 // If we are running with libnuma version > 2, then we should
2396 // be trying to use symbols with versions 1.1
2397 // If we are running with earlier version, which did not have symbol versions,
2398 // we should use the base version.
2399 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2400   void *f = dlvsym(handle, name, "libnuma_1.1");
2401   if (f == NULL) {
2402     f = dlsym(handle, name);
2403   }
2404   return f;
2405 }
2406 
2407 bool os::Linux::libnuma_init() {
2408   // sched_getcpu() should be in libc.
2409   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2410                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2411 
2412   if (sched_getcpu() != -1) { // Does it work?
2413     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2414     if (handle != NULL) {
2415       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2416                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2417       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2418                                        libnuma_dlsym(handle, "numa_max_node")));
2419       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2420                                         libnuma_dlsym(handle, "numa_available")));
2421       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2422                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2423       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2424                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2425 
2426 
2427       if (numa_available() != -1) {
2428         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2429         // Create a cpu -> node mapping
2430         _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2431         rebuild_cpu_to_node_map();
2432         return true;
2433       }
2434     }
2435   }
2436   return false;
2437 }
2438 
2439 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2440 // The table is later used in get_node_by_cpu().
2441 void os::Linux::rebuild_cpu_to_node_map() {
2442   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2443                               // in libnuma (possible values are starting from 16,
2444                               // and continuing up with every other power of 2, but less
2445                               // than the maximum number of CPUs supported by kernel), and
2446                               // is a subject to change (in libnuma version 2 the requirements
2447                               // are more reasonable) we'll just hardcode the number they use
2448                               // in the library.
2449   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2450 
2451   size_t cpu_num = os::active_processor_count();
2452   size_t cpu_map_size = NCPUS / BitsPerCLong;
2453   size_t cpu_map_valid_size =
2454     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2455 
2456   cpu_to_node()->clear();
2457   cpu_to_node()->at_grow(cpu_num - 1);
2458   size_t node_num = numa_get_groups_num();
2459 
2460   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2461   for (size_t i = 0; i < node_num; i++) {
2462     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2463       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2464         if (cpu_map[j] != 0) {
2465           for (size_t k = 0; k < BitsPerCLong; k++) {
2466             if (cpu_map[j] & (1UL << k)) {
2467               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2468             }
2469           }
2470         }
2471       }
2472     }
2473   }
2474   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2475 }
2476 
2477 int os::Linux::get_node_by_cpu(int cpu_id) {
2478   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2479     return cpu_to_node()->at(cpu_id);
2480   }
2481   return -1;
2482 }
2483 
2484 GrowableArray<int>* os::Linux::_cpu_to_node;
2485 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2486 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2487 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2488 os::Linux::numa_available_func_t os::Linux::_numa_available;
2489 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2490 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2491 unsigned long* os::Linux::_numa_all_nodes;
2492 
2493 bool os::uncommit_memory(char* addr, size_t size) {
2494   return ::mmap(addr, size, PROT_NONE,
2495                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0)
2496     != MAP_FAILED;
2497 }
2498 
2499 static bool
2500 get_stack_bounds(uintptr_t *bottom, uintptr_t *top)
2501 {
2502   using namespace std;
2503 
2504   ostringstream oss;
2505   oss << "/proc/" << syscall(SYS_gettid) << "/maps";
2506   ifstream cin(oss.str().c_str());
2507   while (!cin.eof())
2508   {
2509     string str;
2510     getline(cin,str);
2511     const string stack_str = "[stack]";
2512     if (str.length() > stack_str.length()
2513         && (str.compare(str.length() - stack_str.length(),
2514                        stack_str.length(), stack_str)
2515             == 0))
2516       {
2517         istringstream iss(str);
2518         iss.flags(ios::hex);
2519         iss >> *bottom;
2520         char c;
2521         iss >> c;
2522         iss >> *top;
2523         uintptr_t sp = (intptr_t)__builtin_frame_address(0);
2524         if (sp >= *bottom && sp <= *top)
2525           return true;
2526       }
2527   }
2528 
2529   return false;
2530 }
2531 
2532 bool os::create_stack_guard_pages(char* addr, size_t size) {
2533   uintptr_t stack_extent, stack_base;
2534   if (get_stack_bounds(&stack_extent, &stack_base)) {
2535     if (stack_extent < (uintptr_t)addr)
2536       ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2537   }
2538 
2539   return os::commit_memory(addr, size);
2540 }
2541 
2542 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2543   return ::munmap(addr, size) == 0;
2544 }
2545 
2546 static address _highest_vm_reserved_address = NULL;
2547 
2548 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2549 // at 'requested_addr'. If there are existing memory mappings at the same
2550 // location, however, they will be overwritten. If 'fixed' is false,
2551 // 'requested_addr' is only treated as a hint, the return value may or
2552 // may not start from the requested address. Unlike Linux mmap(), this
2553 // function returns NULL to indicate failure.
2554 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2555   char * addr;
2556   int flags;
2557 
2558   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2559   if (fixed) {
2560     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2561     flags |= MAP_FIXED;
2562   }
2563 
2564   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2565   // to PROT_EXEC if executable when we commit the page.
2566   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2567                        flags, -1, 0);
2568 
2569   if (addr != MAP_FAILED) {
2570     // anon_mmap() should only get called during VM initialization,
2571     // don't need lock (actually we can skip locking even it can be called
2572     // from multiple threads, because _highest_vm_reserved_address is just a
2573     // hint about the upper limit of non-stack memory regions.)
2574     if ((address)addr + bytes > _highest_vm_reserved_address) {
2575       _highest_vm_reserved_address = (address)addr + bytes;
2576     }
2577   }
2578 
2579   return addr == MAP_FAILED ? NULL : addr;
2580 }
2581 
2582 // Don't update _highest_vm_reserved_address, because there might be memory
2583 // regions above addr + size. If so, releasing a memory region only creates
2584 // a hole in the address space, it doesn't help prevent heap-stack collision.
2585 //
2586 static int anon_munmap(char * addr, size_t size) {
2587   return ::munmap(addr, size) == 0;
2588 }
2589 
2590 char* os::reserve_memory(size_t bytes, char* requested_addr,
2591                          size_t alignment_hint) {
2592   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2593 }
2594 
2595 bool os::release_memory(char* addr, size_t size) {
2596   return anon_munmap(addr, size);
2597 }
2598 
2599 static address highest_vm_reserved_address() {
2600   return _highest_vm_reserved_address;
2601 }
2602 
2603 static bool linux_mprotect(char* addr, size_t size, int prot) {
2604   // Linux wants the mprotect address argument to be page aligned.
2605   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2606 
2607   // According to SUSv3, mprotect() should only be used with mappings
2608   // established by mmap(), and mmap() always maps whole pages. Unaligned
2609   // 'addr' likely indicates problem in the VM (e.g. trying to change
2610   // protection of malloc'ed or statically allocated memory). Check the
2611   // caller if you hit this assert.
2612   assert(addr == bottom, "sanity check");
2613 
2614   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2615   return ::mprotect(bottom, size, prot) == 0;
2616 }
2617 
2618 // Set protections specified
2619 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2620                         bool is_committed) {
2621   unsigned int p = 0;
2622   switch (prot) {
2623   case MEM_PROT_NONE: p = PROT_NONE; break;
2624   case MEM_PROT_READ: p = PROT_READ; break;
2625   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2626   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2627   default:
2628     ShouldNotReachHere();
2629   }
2630   // is_committed is unused.
2631   return linux_mprotect(addr, bytes, p);
2632 }
2633 
2634 bool os::guard_memory(char* addr, size_t size) {
2635   return linux_mprotect(addr, size, PROT_NONE);
2636 }
2637 
2638 bool os::unguard_memory(char* addr, size_t size) {
2639   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2640 }
2641 
2642 // Large page support
2643 
2644 static size_t _large_page_size = 0;
2645 
2646 bool os::large_page_init() {
2647   if (!UseLargePages) return false;
2648 
2649   if (LargePageSizeInBytes) {
2650     _large_page_size = LargePageSizeInBytes;
2651   } else {
2652     // large_page_size on Linux is used to round up heap size. x86 uses either
2653     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2654     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2655     // page as large as 256M.
2656     //
2657     // Here we try to figure out page size by parsing /proc/meminfo and looking
2658     // for a line with the following format:
2659     //    Hugepagesize:     2048 kB
2660     //
2661     // If we can't determine the value (e.g. /proc is not mounted, or the text
2662     // format has been changed), we'll use the largest page size supported by
2663     // the processor.
2664 
2665 #ifndef ZERO
2666     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M);
2667 #endif // ZERO
2668 
2669     FILE *fp = fopen("/proc/meminfo", "r");
2670     if (fp) {
2671       while (!feof(fp)) {
2672         int x = 0;
2673         char buf[16];
2674         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2675           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2676             _large_page_size = x * K;
2677             break;
2678           }
2679         } else {
2680           // skip to next line
2681           for (;;) {
2682             int ch = fgetc(fp);
2683             if (ch == EOF || ch == (int)'\n') break;
2684           }
2685         }
2686       }
2687       fclose(fp);
2688     }
2689   }
2690 
2691   const size_t default_page_size = (size_t)Linux::page_size();
2692   if (_large_page_size > default_page_size) {
2693     _page_sizes[0] = _large_page_size;
2694     _page_sizes[1] = default_page_size;
2695     _page_sizes[2] = 0;
2696   }
2697 
2698   // Large page support is available on 2.6 or newer kernel, some vendors
2699   // (e.g. Redhat) have backported it to their 2.4 based distributions.
2700   // We optimistically assume the support is available. If later it turns out
2701   // not true, VM will automatically switch to use regular page size.
2702   return true;
2703 }
2704 
2705 #ifndef SHM_HUGETLB
2706 #define SHM_HUGETLB 04000
2707 #endif
2708 
2709 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
2710   // "exec" is passed in but not used.  Creating the shared image for
2711   // the code cache doesn't have an SHM_X executable permission to check.
2712   assert(UseLargePages, "only for large pages");
2713 
2714   key_t key = IPC_PRIVATE;
2715   char *addr;
2716 
2717   bool warn_on_failure = UseLargePages &&
2718                         (!FLAG_IS_DEFAULT(UseLargePages) ||
2719                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2720                         );
2721   char msg[128];
2722 
2723   // Create a large shared memory region to attach to based on size.
2724   // Currently, size is the total size of the heap
2725   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2726   if (shmid == -1) {
2727      // Possible reasons for shmget failure:
2728      // 1. shmmax is too small for Java heap.
2729      //    > check shmmax value: cat /proc/sys/kernel/shmmax
2730      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2731      // 2. not enough large page memory.
2732      //    > check available large pages: cat /proc/meminfo
2733      //    > increase amount of large pages:
2734      //          echo new_value > /proc/sys/vm/nr_hugepages
2735      //      Note 1: different Linux may use different name for this property,
2736      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
2737      //      Note 2: it's possible there's enough physical memory available but
2738      //            they are so fragmented after a long run that they can't
2739      //            coalesce into large pages. Try to reserve large pages when
2740      //            the system is still "fresh".
2741      if (warn_on_failure) {
2742        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2743        warning(msg);
2744      }
2745      return NULL;
2746   }
2747 
2748   // attach to the region
2749   addr = (char*)shmat(shmid, NULL, 0);
2750   int err = errno;
2751 
2752   // Remove shmid. If shmat() is successful, the actual shared memory segment
2753   // will be deleted when it's detached by shmdt() or when the process
2754   // terminates. If shmat() is not successful this will remove the shared
2755   // segment immediately.
2756   shmctl(shmid, IPC_RMID, NULL);
2757 
2758   if ((intptr_t)addr == -1) {
2759      if (warn_on_failure) {
2760        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2761        warning(msg);
2762      }
2763      return NULL;
2764   }
2765 
2766   return addr;
2767 }
2768 
2769 bool os::release_memory_special(char* base, size_t bytes) {
2770   // detaching the SHM segment will also delete it, see reserve_memory_special()
2771   int rslt = shmdt(base);
2772   return rslt == 0;
2773 }
2774 
2775 size_t os::large_page_size() {
2776   return _large_page_size;
2777 }
2778 
2779 // Linux does not support anonymous mmap with large page memory. The only way
2780 // to reserve large page memory without file backing is through SysV shared
2781 // memory API. The entire memory region is committed and pinned upfront.
2782 // Hopefully this will change in the future...
2783 bool os::can_commit_large_page_memory() {
2784   return false;
2785 }
2786 
2787 bool os::can_execute_large_page_memory() {
2788   return false;
2789 }
2790 
2791 // Reserve memory at an arbitrary address, only if that area is
2792 // available (and not reserved for something else).
2793 
2794 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2795   const int max_tries = 10;
2796   char* base[max_tries];
2797   size_t size[max_tries];
2798   const size_t gap = 0x000000;
2799 
2800   // Assert only that the size is a multiple of the page size, since
2801   // that's all that mmap requires, and since that's all we really know
2802   // about at this low abstraction level.  If we need higher alignment,
2803   // we can either pass an alignment to this method or verify alignment
2804   // in one of the methods further up the call chain.  See bug 5044738.
2805   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2806 
2807   // Repeatedly allocate blocks until the block is allocated at the
2808   // right spot. Give up after max_tries. Note that reserve_memory() will
2809   // automatically update _highest_vm_reserved_address if the call is
2810   // successful. The variable tracks the highest memory address every reserved
2811   // by JVM. It is used to detect heap-stack collision if running with
2812   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2813   // space than needed, it could confuse the collision detecting code. To
2814   // solve the problem, save current _highest_vm_reserved_address and
2815   // calculate the correct value before return.
2816   address old_highest = _highest_vm_reserved_address;
2817 
2818   // Linux mmap allows caller to pass an address as hint; give it a try first,
2819   // if kernel honors the hint then we can return immediately.
2820   char * addr = anon_mmap(requested_addr, bytes, false);
2821   if (addr == requested_addr) {
2822      return requested_addr;
2823   }
2824 
2825   if (addr != NULL) {
2826      // mmap() is successful but it fails to reserve at the requested address
2827      anon_munmap(addr, bytes);
2828   }
2829 
2830   int i;
2831   for (i = 0; i < max_tries; ++i) {
2832     base[i] = reserve_memory(bytes);
2833 
2834     if (base[i] != NULL) {
2835       // Is this the block we wanted?
2836       if (base[i] == requested_addr) {
2837         size[i] = bytes;
2838         break;
2839       }
2840 
2841       // Does this overlap the block we wanted? Give back the overlapped
2842       // parts and try again.
2843 
2844       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2845       if (top_overlap >= 0 && top_overlap < bytes) {
2846         unmap_memory(base[i], top_overlap);
2847         base[i] += top_overlap;
2848         size[i] = bytes - top_overlap;
2849       } else {
2850         size_t bottom_overlap = base[i] + bytes - requested_addr;
2851         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2852           unmap_memory(requested_addr, bottom_overlap);
2853           size[i] = bytes - bottom_overlap;
2854         } else {
2855           size[i] = bytes;
2856         }
2857       }
2858     }
2859   }
2860 
2861   // Give back the unused reserved pieces.
2862 
2863   for (int j = 0; j < i; ++j) {
2864     if (base[j] != NULL) {
2865       unmap_memory(base[j], size[j]);
2866     }
2867   }
2868 
2869   if (i < max_tries) {
2870     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2871     return requested_addr;
2872   } else {
2873     _highest_vm_reserved_address = old_highest;
2874     return NULL;
2875   }
2876 }
2877 
2878 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2879   return ::read(fd, buf, nBytes);
2880 }
2881 
2882 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
2883 // Solaris uses poll(), linux uses park().
2884 // Poll() is likely a better choice, assuming that Thread.interrupt()
2885 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2886 // SIGSEGV, see 4355769.
2887 
2888 const int NANOSECS_PER_MILLISECS = 1000000;
2889 
2890 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2891   assert(thread == Thread::current(),  "thread consistency check");
2892 
2893   ParkEvent * const slp = thread->_SleepEvent ;
2894   slp->reset() ;
2895   OrderAccess::fence() ;
2896 
2897   if (interruptible) {
2898     jlong prevtime = javaTimeNanos();
2899 
2900     for (;;) {
2901       if (os::is_interrupted(thread, true)) {
2902         return OS_INTRPT;
2903       }
2904 
2905       jlong newtime = javaTimeNanos();
2906 
2907       if (newtime - prevtime < 0) {
2908         // time moving backwards, should only happen if no monotonic clock
2909         // not a guarantee() because JVM should not abort on kernel/glibc bugs
2910         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2911       } else {
2912         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2913       }
2914 
2915       if(millis <= 0) {
2916         return OS_OK;
2917       }
2918 
2919       prevtime = newtime;
2920 
2921       {
2922         assert(thread->is_Java_thread(), "sanity check");
2923         JavaThread *jt = (JavaThread *) thread;
2924         ThreadBlockInVM tbivm(jt);
2925         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
2926 
2927         jt->set_suspend_equivalent();
2928         // cleared by handle_special_suspend_equivalent_condition() or
2929         // java_suspend_self() via check_and_wait_while_suspended()
2930 
2931         slp->park(millis);
2932 
2933         // were we externally suspended while we were waiting?
2934         jt->check_and_wait_while_suspended();
2935       }
2936     }
2937   } else {
2938     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2939     jlong prevtime = javaTimeNanos();
2940 
2941     for (;;) {
2942       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
2943       // the 1st iteration ...
2944       jlong newtime = javaTimeNanos();
2945 
2946       if (newtime - prevtime < 0) {
2947         // time moving backwards, should only happen if no monotonic clock
2948         // not a guarantee() because JVM should not abort on kernel/glibc bugs
2949         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2950       } else {
2951         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2952       }
2953 
2954       if(millis <= 0) break ;
2955 
2956       prevtime = newtime;
2957       slp->park(millis);
2958     }
2959     return OS_OK ;
2960   }
2961 }
2962 
2963 int os::naked_sleep() {
2964   // %% make the sleep time an integer flag. for now use 1 millisec.
2965   return os::sleep(Thread::current(), 1, false);
2966 }
2967 
2968 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2969 void os::infinite_sleep() {
2970   while (true) {    // sleep forever ...
2971     ::sleep(100);   // ... 100 seconds at a time
2972   }
2973 }
2974 
2975 // Used to convert frequent JVM_Yield() to nops
2976 bool os::dont_yield() {
2977   return DontYieldALot;
2978 }
2979 
2980 void os::yield() {
2981   sched_yield();
2982 }
2983 
2984 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
2985 
2986 void os::yield_all(int attempts) {
2987   // Yields to all threads, including threads with lower priorities
2988   // Threads on Linux are all with same priority. The Solaris style
2989   // os::yield_all() with nanosleep(1ms) is not necessary.
2990   sched_yield();
2991 }
2992 
2993 // Called from the tight loops to possibly influence time-sharing heuristics
2994 void os::loop_breaker(int attempts) {
2995   os::yield_all(attempts);
2996 }
2997 
2998 ////////////////////////////////////////////////////////////////////////////////
2999 // thread priority support
3000 
3001 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3002 // only supports dynamic priority, static priority must be zero. For real-time
3003 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3004 // However, for large multi-threaded applications, SCHED_RR is not only slower
3005 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3006 // of 5 runs - Sep 2005).
3007 //
3008 // The following code actually changes the niceness of kernel-thread/LWP. It
3009 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3010 // not the entire user process, and user level threads are 1:1 mapped to kernel
3011 // threads. It has always been the case, but could change in the future. For
3012 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3013 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3014 
3015 int os::java_to_os_priority[MaxPriority + 1] = {
3016   19,              // 0 Entry should never be used
3017 
3018    4,              // 1 MinPriority
3019    3,              // 2
3020    2,              // 3
3021 
3022    1,              // 4
3023    0,              // 5 NormPriority
3024   -1,              // 6
3025 
3026   -2,              // 7
3027   -3,              // 8
3028   -4,              // 9 NearMaxPriority
3029 
3030   -5               // 10 MaxPriority
3031 };
3032 
3033 static int prio_init() {
3034   if (ThreadPriorityPolicy == 1) {
3035     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3036     // if effective uid is not root. Perhaps, a more elegant way of doing
3037     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3038     if (geteuid() != 0) {
3039       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3040         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3041       }
3042       ThreadPriorityPolicy = 0;
3043     }
3044   }
3045   return 0;
3046 }
3047 
3048 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3049   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3050 
3051   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3052   return (ret == 0) ? OS_OK : OS_ERR;
3053 }
3054 
3055 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3056   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3057     *priority_ptr = java_to_os_priority[NormPriority];
3058     return OS_OK;
3059   }
3060 
3061   errno = 0;
3062   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3063   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3064 }
3065 
3066 // Hint to the underlying OS that a task switch would not be good.
3067 // Void return because it's a hint and can fail.
3068 void os::hint_no_preempt() {}
3069 
3070 ////////////////////////////////////////////////////////////////////////////////
3071 // suspend/resume support
3072 
3073 //  the low-level signal-based suspend/resume support is a remnant from the
3074 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3075 //  within hotspot. Now there is a single use-case for this:
3076 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3077 //      that runs in the watcher thread.
3078 //  The remaining code is greatly simplified from the more general suspension
3079 //  code that used to be used.
3080 //
3081 //  The protocol is quite simple:
3082 //  - suspend:
3083 //      - sends a signal to the target thread
3084 //      - polls the suspend state of the osthread using a yield loop
3085 //      - target thread signal handler (SR_handler) sets suspend state
3086 //        and blocks in sigsuspend until continued
3087 //  - resume:
3088 //      - sets target osthread state to continue
3089 //      - sends signal to end the sigsuspend loop in the SR_handler
3090 //
3091 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3092 //
3093 
3094 static void resume_clear_context(OSThread *osthread) {
3095   osthread->set_ucontext(NULL);
3096   osthread->set_siginfo(NULL);
3097 
3098   // notify the suspend action is completed, we have now resumed
3099   osthread->sr.clear_suspended();
3100 }
3101 
3102 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3103   osthread->set_ucontext(context);
3104   osthread->set_siginfo(siginfo);
3105 }
3106 
3107 //
3108 // Handler function invoked when a thread's execution is suspended or
3109 // resumed. We have to be careful that only async-safe functions are
3110 // called here (Note: most pthread functions are not async safe and
3111 // should be avoided.)
3112 //
3113 // Note: sigwait() is a more natural fit than sigsuspend() from an
3114 // interface point of view, but sigwait() prevents the signal hander
3115 // from being run. libpthread would get very confused by not having
3116 // its signal handlers run and prevents sigwait()'s use with the
3117 // mutex granting granting signal.
3118 //
3119 // Currently only ever called on the VMThread
3120 //
3121 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3122   // Save and restore errno to avoid confusing native code with EINTR
3123   // after sigsuspend.
3124   int old_errno = errno;
3125 
3126   Thread* thread = Thread::current();
3127   OSThread* osthread = thread->osthread();
3128   assert(thread->is_VM_thread(), "Must be VMThread");
3129   // read current suspend action
3130   int action = osthread->sr.suspend_action();
3131   if (action == SR_SUSPEND) {
3132     suspend_save_context(osthread, siginfo, context);
3133 
3134     // Notify the suspend action is about to be completed. do_suspend()
3135     // waits until SR_SUSPENDED is set and then returns. We will wait
3136     // here for a resume signal and that completes the suspend-other
3137     // action. do_suspend/do_resume is always called as a pair from
3138     // the same thread - so there are no races
3139 
3140     // notify the caller
3141     osthread->sr.set_suspended();
3142 
3143     sigset_t suspend_set;  // signals for sigsuspend()
3144 
3145     // get current set of blocked signals and unblock resume signal
3146     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3147     sigdelset(&suspend_set, SR_signum);
3148 
3149     // wait here until we are resumed
3150     do {
3151       sigsuspend(&suspend_set);
3152       // ignore all returns until we get a resume signal
3153     } while (osthread->sr.suspend_action() != SR_CONTINUE);
3154 
3155     resume_clear_context(osthread);
3156 
3157   } else {
3158     assert(action == SR_CONTINUE, "unexpected sr action");
3159     // nothing special to do - just leave the handler
3160   }
3161 
3162   errno = old_errno;
3163 }
3164 
3165 
3166 static int SR_initialize() {
3167   struct sigaction act;
3168   char *s;
3169   /* Get signal number to use for suspend/resume */
3170   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3171     int sig = ::strtol(s, 0, 10);
3172     if (sig > 0 || sig < _NSIG) {
3173         SR_signum = sig;
3174     }
3175   }
3176 
3177   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3178         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3179 
3180   sigemptyset(&SR_sigset);
3181   sigaddset(&SR_sigset, SR_signum);
3182 
3183   /* Set up signal handler for suspend/resume */
3184   act.sa_flags = SA_RESTART|SA_SIGINFO;
3185   act.sa_handler = (void (*)(int)) SR_handler;
3186 
3187   // SR_signum is blocked by default.
3188   // 4528190 - We also need to block pthread restart signal (32 on all
3189   // supported Linux platforms). Note that LinuxThreads need to block
3190   // this signal for all threads to work properly. So we don't have
3191   // to use hard-coded signal number when setting up the mask.
3192   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3193 
3194   if (sigaction(SR_signum, &act, 0) == -1) {
3195     return -1;
3196   }
3197 
3198   // Save signal flag
3199   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3200   return 0;
3201 }
3202 
3203 static int SR_finalize() {
3204   return 0;
3205 }
3206 
3207 
3208 // returns true on success and false on error - really an error is fatal
3209 // but this seems the normal response to library errors
3210 static bool do_suspend(OSThread* osthread) {
3211   // mark as suspended and send signal
3212   osthread->sr.set_suspend_action(SR_SUSPEND);
3213   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3214   assert_status(status == 0, status, "pthread_kill");
3215 
3216   // check status and wait until notified of suspension
3217   if (status == 0) {
3218     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3219       os::yield_all(i);
3220     }
3221     osthread->sr.set_suspend_action(SR_NONE);
3222     return true;
3223   }
3224   else {
3225     osthread->sr.set_suspend_action(SR_NONE);
3226     return false;
3227   }
3228 }
3229 
3230 static void do_resume(OSThread* osthread) {
3231   assert(osthread->sr.is_suspended(), "thread should be suspended");
3232   osthread->sr.set_suspend_action(SR_CONTINUE);
3233 
3234   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3235   assert_status(status == 0, status, "pthread_kill");
3236   // check status and wait unit notified of resumption
3237   if (status == 0) {
3238     for (int i = 0; osthread->sr.is_suspended(); i++) {
3239       os::yield_all(i);
3240     }
3241   }
3242   osthread->sr.set_suspend_action(SR_NONE);
3243 }
3244 
3245 ////////////////////////////////////////////////////////////////////////////////
3246 // interrupt support
3247 
3248 void os::interrupt(Thread* thread) {
3249   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3250     "possibility of dangling Thread pointer");
3251 
3252   OSThread* osthread = thread->osthread();
3253 
3254   if (!osthread->interrupted()) {
3255     osthread->set_interrupted(true);
3256     // More than one thread can get here with the same value of osthread,
3257     // resulting in multiple notifications.  We do, however, want the store
3258     // to interrupted() to be visible to other threads before we execute unpark().
3259     OrderAccess::fence();
3260     ParkEvent * const slp = thread->_SleepEvent ;
3261     if (slp != NULL) slp->unpark() ;
3262   }
3263 
3264   // For JSR166. Unpark even if interrupt status already was set
3265   if (thread->is_Java_thread())
3266     ((JavaThread*)thread)->parker()->unpark();
3267 
3268   ParkEvent * ev = thread->_ParkEvent ;
3269   if (ev != NULL) ev->unpark() ;
3270 
3271 }
3272 
3273 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3274   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3275     "possibility of dangling Thread pointer");
3276 
3277   OSThread* osthread = thread->osthread();
3278 
3279   bool interrupted = osthread->interrupted();
3280 
3281   if (interrupted && clear_interrupted) {
3282     osthread->set_interrupted(false);
3283     // consider thread->_SleepEvent->reset() ... optional optimization
3284   }
3285 
3286   return interrupted;
3287 }
3288 
3289 ///////////////////////////////////////////////////////////////////////////////////
3290 // signal handling (except suspend/resume)
3291 
3292 // This routine may be used by user applications as a "hook" to catch signals.
3293 // The user-defined signal handler must pass unrecognized signals to this
3294 // routine, and if it returns true (non-zero), then the signal handler must
3295 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3296 // routine will never retun false (zero), but instead will execute a VM panic
3297 // routine kill the process.
3298 //
3299 // If this routine returns false, it is OK to call it again.  This allows
3300 // the user-defined signal handler to perform checks either before or after
3301 // the VM performs its own checks.  Naturally, the user code would be making
3302 // a serious error if it tried to handle an exception (such as a null check
3303 // or breakpoint) that the VM was generating for its own correct operation.
3304 //
3305 // This routine may recognize any of the following kinds of signals:
3306 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3307 // It should be consulted by handlers for any of those signals.
3308 //
3309 // The caller of this routine must pass in the three arguments supplied
3310 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3311 // field of the structure passed to sigaction().  This routine assumes that
3312 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3313 //
3314 // Note that the VM will print warnings if it detects conflicting signal
3315 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3316 //
3317 extern "C" int
3318 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3319                         void* ucontext, int abort_if_unrecognized);
3320 
3321 void signalHandler(int sig, siginfo_t* info, void* uc) {
3322   assert(info != NULL && uc != NULL, "it must be old kernel");
3323   JVM_handle_linux_signal(sig, info, uc, true);
3324 }
3325 
3326 
3327 // This boolean allows users to forward their own non-matching signals
3328 // to JVM_handle_linux_signal, harmlessly.
3329 bool os::Linux::signal_handlers_are_installed = false;
3330 
3331 // For signal-chaining
3332 struct sigaction os::Linux::sigact[MAXSIGNUM];
3333 unsigned int os::Linux::sigs = 0;
3334 bool os::Linux::libjsig_is_loaded = false;
3335 typedef struct sigaction *(*get_signal_t)(int);
3336 get_signal_t os::Linux::get_signal_action = NULL;
3337 
3338 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3339   struct sigaction *actp = NULL;
3340 
3341   if (libjsig_is_loaded) {
3342     // Retrieve the old signal handler from libjsig
3343     actp = (*get_signal_action)(sig);
3344   }
3345   if (actp == NULL) {
3346     // Retrieve the preinstalled signal handler from jvm
3347     actp = get_preinstalled_handler(sig);
3348   }
3349 
3350   return actp;
3351 }
3352 
3353 static bool call_chained_handler(struct sigaction *actp, int sig,
3354                                  siginfo_t *siginfo, void *context) {
3355   // Call the old signal handler
3356   if (actp->sa_handler == SIG_DFL) {
3357     // It's more reasonable to let jvm treat it as an unexpected exception
3358     // instead of taking the default action.
3359     return false;
3360   } else if (actp->sa_handler != SIG_IGN) {
3361     if ((actp->sa_flags & SA_NODEFER) == 0) {
3362       // automaticlly block the signal
3363       sigaddset(&(actp->sa_mask), sig);
3364     }
3365 
3366     sa_handler_t hand;
3367     sa_sigaction_t sa;
3368     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3369     // retrieve the chained handler
3370     if (siginfo_flag_set) {
3371       sa = actp->sa_sigaction;
3372     } else {
3373       hand = actp->sa_handler;
3374     }
3375 
3376     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3377       actp->sa_handler = SIG_DFL;
3378     }
3379 
3380     // try to honor the signal mask
3381     sigset_t oset;
3382     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3383 
3384     // call into the chained handler
3385     if (siginfo_flag_set) {
3386       (*sa)(sig, siginfo, context);
3387     } else {
3388       (*hand)(sig);
3389     }
3390 
3391     // restore the signal mask
3392     pthread_sigmask(SIG_SETMASK, &oset, 0);
3393   }
3394   // Tell jvm's signal handler the signal is taken care of.
3395   return true;
3396 }
3397 
3398 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3399   bool chained = false;
3400   // signal-chaining
3401   if (UseSignalChaining) {
3402     struct sigaction *actp = get_chained_signal_action(sig);
3403     if (actp != NULL) {
3404       chained = call_chained_handler(actp, sig, siginfo, context);
3405     }
3406   }
3407   return chained;
3408 }
3409 
3410 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3411   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3412     return &sigact[sig];
3413   }
3414   return NULL;
3415 }
3416 
3417 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3418   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3419   sigact[sig] = oldAct;
3420   sigs |= (unsigned int)1 << sig;
3421 }
3422 
3423 // for diagnostic
3424 int os::Linux::sigflags[MAXSIGNUM];
3425 
3426 int os::Linux::get_our_sigflags(int sig) {
3427   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3428   return sigflags[sig];
3429 }
3430 
3431 void os::Linux::set_our_sigflags(int sig, int flags) {
3432   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3433   sigflags[sig] = flags;
3434 }
3435 
3436 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3437   // Check for overwrite.
3438   struct sigaction oldAct;
3439   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3440 
3441   void* oldhand = oldAct.sa_sigaction
3442                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3443                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3444   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3445       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3446       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3447     if (AllowUserSignalHandlers || !set_installed) {
3448       // Do not overwrite; user takes responsibility to forward to us.
3449       return;
3450     } else if (UseSignalChaining) {
3451       // save the old handler in jvm
3452       save_preinstalled_handler(sig, oldAct);
3453       // libjsig also interposes the sigaction() call below and saves the
3454       // old sigaction on it own.
3455     } else {
3456       fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig);
3457     }
3458   }
3459 
3460   struct sigaction sigAct;
3461   sigfillset(&(sigAct.sa_mask));
3462   sigAct.sa_handler = SIG_DFL;
3463   if (!set_installed) {
3464     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3465   } else {
3466     sigAct.sa_sigaction = signalHandler;
3467     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3468   }
3469   // Save flags, which are set by ours
3470   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3471   sigflags[sig] = sigAct.sa_flags;
3472 
3473   int ret = sigaction(sig, &sigAct, &oldAct);
3474   assert(ret == 0, "check");
3475 
3476   void* oldhand2  = oldAct.sa_sigaction
3477                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3478                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3479   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3480 }
3481 
3482 // install signal handlers for signals that HotSpot needs to
3483 // handle in order to support Java-level exception handling.
3484 
3485 void os::Linux::install_signal_handlers() {
3486   if (!signal_handlers_are_installed) {
3487     signal_handlers_are_installed = true;
3488 
3489     // signal-chaining
3490     typedef void (*signal_setting_t)();
3491     signal_setting_t begin_signal_setting = NULL;
3492     signal_setting_t end_signal_setting = NULL;
3493     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3494                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3495     if (begin_signal_setting != NULL) {
3496       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3497                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3498       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3499                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3500       libjsig_is_loaded = true;
3501       assert(UseSignalChaining, "should enable signal-chaining");
3502     }
3503     if (libjsig_is_loaded) {
3504       // Tell libjsig jvm is setting signal handlers
3505       (*begin_signal_setting)();
3506     }
3507 
3508     set_signal_handler(SIGSEGV, true);
3509     set_signal_handler(SIGPIPE, true);
3510     set_signal_handler(SIGBUS, true);
3511     set_signal_handler(SIGILL, true);
3512     set_signal_handler(SIGFPE, true);
3513     set_signal_handler(SIGXFSZ, true);
3514 
3515     if (libjsig_is_loaded) {
3516       // Tell libjsig jvm finishes setting signal handlers
3517       (*end_signal_setting)();
3518     }
3519 
3520     // We don't activate signal checker if libjsig is in place, we trust ourselves
3521     // and if UserSignalHandler is installed all bets are off
3522     if (CheckJNICalls) {
3523       if (libjsig_is_loaded) {
3524         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3525         check_signals = false;
3526       }
3527       if (AllowUserSignalHandlers) {
3528         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3529         check_signals = false;
3530       }
3531     }
3532   }
3533 }
3534 
3535 // This is the fastest way to get thread cpu time on Linux.
3536 // Returns cpu time (user+sys) for any thread, not only for current.
3537 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3538 // It might work on 2.6.10+ with a special kernel/glibc patch.
3539 // For reference, please, see IEEE Std 1003.1-2004:
3540 //   http://www.unix.org/single_unix_specification
3541 
3542 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3543   struct timespec tp;
3544   int rc = os::Linux::clock_gettime(clockid, &tp);
3545   assert(rc == 0, "clock_gettime is expected to return 0 code");
3546 
3547   return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3548 }
3549 
3550 /////
3551 // glibc on Linux platform uses non-documented flag
3552 // to indicate, that some special sort of signal
3553 // trampoline is used.
3554 // We will never set this flag, and we should
3555 // ignore this flag in our diagnostic
3556 #ifdef SIGNIFICANT_SIGNAL_MASK
3557 #undef SIGNIFICANT_SIGNAL_MASK
3558 #endif
3559 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3560 
3561 static const char* get_signal_handler_name(address handler,
3562                                            char* buf, int buflen) {
3563   int offset;
3564   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3565   if (found) {
3566     // skip directory names
3567     const char *p1, *p2;
3568     p1 = buf;
3569     size_t len = strlen(os::file_separator());
3570     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3571     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3572   } else {
3573     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3574   }
3575   return buf;
3576 }
3577 
3578 static void print_signal_handler(outputStream* st, int sig,
3579                                  char* buf, size_t buflen) {
3580   struct sigaction sa;
3581 
3582   sigaction(sig, NULL, &sa);
3583 
3584   // See comment for SIGNIFICANT_SIGNAL_MASK define
3585   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3586 
3587   st->print("%s: ", os::exception_name(sig, buf, buflen));
3588 
3589   address handler = (sa.sa_flags & SA_SIGINFO)
3590     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3591     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3592 
3593   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3594     st->print("SIG_DFL");
3595   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3596     st->print("SIG_IGN");
3597   } else {
3598     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3599   }
3600 
3601   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3602 
3603   address rh = VMError::get_resetted_sighandler(sig);
3604   // May be, handler was resetted by VMError?
3605   if(rh != NULL) {
3606     handler = rh;
3607     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3608   }
3609 
3610   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
3611 
3612   // Check: is it our handler?
3613   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3614      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3615     // It is our signal handler
3616     // check for flags, reset system-used one!
3617     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3618       st->print(
3619                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3620                 os::Linux::get_our_sigflags(sig));
3621     }
3622   }
3623   st->cr();
3624 }
3625 
3626 
3627 #define DO_SIGNAL_CHECK(sig) \
3628   if (!sigismember(&check_signal_done, sig)) \
3629     os::Linux::check_signal_handler(sig)
3630 
3631 // This method is a periodic task to check for misbehaving JNI applications
3632 // under CheckJNI, we can add any periodic checks here
3633 
3634 void os::run_periodic_checks() {
3635 
3636   if (check_signals == false) return;
3637 
3638   // SEGV and BUS if overridden could potentially prevent
3639   // generation of hs*.log in the event of a crash, debugging
3640   // such a case can be very challenging, so we absolutely
3641   // check the following for a good measure:
3642   DO_SIGNAL_CHECK(SIGSEGV);
3643   DO_SIGNAL_CHECK(SIGILL);
3644   DO_SIGNAL_CHECK(SIGFPE);
3645   DO_SIGNAL_CHECK(SIGBUS);
3646   DO_SIGNAL_CHECK(SIGPIPE);
3647   DO_SIGNAL_CHECK(SIGXFSZ);
3648 
3649 
3650   // ReduceSignalUsage allows the user to override these handlers
3651   // see comments at the very top and jvm_solaris.h
3652   if (!ReduceSignalUsage) {
3653     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3654     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3655     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3656     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3657   }
3658 
3659   DO_SIGNAL_CHECK(SR_signum);
3660   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3661 }
3662 
3663 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3664 
3665 static os_sigaction_t os_sigaction = NULL;
3666 
3667 void os::Linux::check_signal_handler(int sig) {
3668   char buf[O_BUFLEN];
3669   address jvmHandler = NULL;
3670 
3671 
3672   struct sigaction act;
3673   if (os_sigaction == NULL) {
3674     // only trust the default sigaction, in case it has been interposed
3675     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3676     if (os_sigaction == NULL) return;
3677   }
3678 
3679   os_sigaction(sig, (struct sigaction*)NULL, &act);
3680 
3681 
3682   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3683 
3684   address thisHandler = (act.sa_flags & SA_SIGINFO)
3685     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3686     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3687 
3688 
3689   switch(sig) {
3690   case SIGSEGV:
3691   case SIGBUS:
3692   case SIGFPE:
3693   case SIGPIPE:
3694   case SIGILL:
3695   case SIGXFSZ:
3696     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3697     break;
3698 
3699   case SHUTDOWN1_SIGNAL:
3700   case SHUTDOWN2_SIGNAL:
3701   case SHUTDOWN3_SIGNAL:
3702   case BREAK_SIGNAL:
3703     jvmHandler = (address)user_handler();
3704     break;
3705 
3706   case INTERRUPT_SIGNAL:
3707     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3708     break;
3709 
3710   default:
3711     if (sig == SR_signum) {
3712       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3713     } else {
3714       return;
3715     }
3716     break;
3717   }
3718 
3719   if (thisHandler != jvmHandler) {
3720     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3721     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3722     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3723     // No need to check this sig any longer
3724     sigaddset(&check_signal_done, sig);
3725   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3726     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3727     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3728     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
3729     // No need to check this sig any longer
3730     sigaddset(&check_signal_done, sig);
3731   }
3732 
3733   // Dump all the signal
3734   if (sigismember(&check_signal_done, sig)) {
3735     print_signal_handlers(tty, buf, O_BUFLEN);
3736   }
3737 }
3738 
3739 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3740 
3741 extern bool signal_name(int signo, char* buf, size_t len);
3742 
3743 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3744   if (0 < exception_code && exception_code <= SIGRTMAX) {
3745     // signal
3746     if (!signal_name(exception_code, buf, size)) {
3747       jio_snprintf(buf, size, "SIG%d", exception_code);
3748     }
3749     return buf;
3750   } else {
3751     return NULL;
3752   }
3753 }
3754 
3755 // this is called _before_ the most of global arguments have been parsed
3756 void os::init(void) {
3757   char dummy;   /* used to get a guess on initial stack address */
3758 //  first_hrtime = gethrtime();
3759 
3760   // With LinuxThreads the JavaMain thread pid (primordial thread)
3761   // is different than the pid of the java launcher thread.
3762   // So, on Linux, the launcher thread pid is passed to the VM
3763   // via the sun.java.launcher.pid property.
3764   // Use this property instead of getpid() if it was correctly passed.
3765   // See bug 6351349.
3766   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3767 
3768   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3769 
3770   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3771 
3772   init_random(1234567);
3773 
3774   ThreadCritical::initialize();
3775 
3776   Linux::set_page_size(sysconf(_SC_PAGESIZE));
3777   if (Linux::page_size() == -1) {
3778     fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno));
3779   }
3780   init_page_sizes((size_t) Linux::page_size());
3781 
3782   Linux::initialize_system_info();
3783 
3784   // main_thread points to the aboriginal thread
3785   Linux::_main_thread = pthread_self();
3786 
3787   Linux::clock_init();
3788   initial_time_count = os::elapsed_counter();
3789   pthread_mutex_init(&dl_mutex, NULL);
3790 }
3791 
3792 // To install functions for atexit system call
3793 extern "C" {
3794   static void perfMemory_exit_helper() {
3795     perfMemory_exit();
3796   }
3797 }
3798 
3799 // this is called _after_ the global arguments have been parsed
3800 jint os::init_2(void)
3801 {
3802   Linux::fast_thread_clock_init();
3803 
3804   // Allocate a single page and mark it as readable for safepoint polling
3805   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3806   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3807 
3808   os::set_polling_page( polling_page );
3809 
3810 #ifndef PRODUCT
3811   if(Verbose && PrintMiscellaneous)
3812     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3813 #endif
3814 
3815   if (!UseMembar) {
3816     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3817     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3818     os::set_memory_serialize_page( mem_serialize_page );
3819 
3820 #ifndef PRODUCT
3821     if(Verbose && PrintMiscellaneous)
3822       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3823 #endif
3824   }
3825 
3826   FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3827 
3828   // initialize suspend/resume support - must do this before signal_sets_init()
3829   if (SR_initialize() != 0) {
3830     perror("SR_initialize failed");
3831     return JNI_ERR;
3832   }
3833 
3834   Linux::signal_sets_init();
3835   Linux::install_signal_handlers();
3836 
3837   size_t threadStackSizeInBytes = ThreadStackSize * K;
3838   if (threadStackSizeInBytes != 0 &&
3839       threadStackSizeInBytes < Linux::min_stack_allowed) {
3840         tty->print_cr("\nThe stack size specified is too small, "
3841                       "Specify at least %dk",
3842                       Linux::min_stack_allowed / K);
3843         return JNI_ERR;
3844   }
3845 
3846   // Make the stack size a multiple of the page size so that
3847   // the yellow/red zones can be guarded.
3848   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3849         vm_page_size()));
3850 
3851   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
3852 
3853   Linux::libpthread_init();
3854   if (PrintMiscellaneous && (Verbose || WizardMode)) {
3855      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
3856           Linux::glibc_version(), Linux::libpthread_version(),
3857           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
3858   }
3859 
3860   if (UseNUMA) {
3861     if (!Linux::libnuma_init()) {
3862       UseNUMA = false;
3863     } else {
3864       if ((Linux::numa_max_node() < 1)) {
3865         // There's only one node(they start from 0), disable NUMA.
3866         UseNUMA = false;
3867       }
3868     }
3869     if (!UseNUMA && ForceNUMA) {
3870       UseNUMA = true;
3871     }
3872   }
3873 
3874   if (MaxFDLimit) {
3875     // set the number of file descriptors to max. print out error
3876     // if getrlimit/setrlimit fails but continue regardless.
3877     struct rlimit nbr_files;
3878     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3879     if (status != 0) {
3880       if (PrintMiscellaneous && (Verbose || WizardMode))
3881         perror("os::init_2 getrlimit failed");
3882     } else {
3883       nbr_files.rlim_cur = nbr_files.rlim_max;
3884       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3885       if (status != 0) {
3886         if (PrintMiscellaneous && (Verbose || WizardMode))
3887           perror("os::init_2 setrlimit failed");
3888       }
3889     }
3890   }
3891 
3892   // Initialize lock used to serialize thread creation (see os::create_thread)
3893   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
3894 
3895   // Initialize HPI.
3896   jint hpi_result = hpi::initialize();
3897   if (hpi_result != JNI_OK) {
3898     tty->print_cr("There was an error trying to initialize the HPI library.");
3899     return hpi_result;
3900   }
3901 
3902   // at-exit methods are called in the reverse order of their registration.
3903   // atexit functions are called on return from main or as a result of a
3904   // call to exit(3C). There can be only 32 of these functions registered
3905   // and atexit() does not set errno.
3906 
3907   if (PerfAllowAtExitRegistration) {
3908     // only register atexit functions if PerfAllowAtExitRegistration is set.
3909     // atexit functions can be delayed until process exit time, which
3910     // can be problematic for embedded VM situations. Embedded VMs should
3911     // call DestroyJavaVM() to assure that VM resources are released.
3912 
3913     // note: perfMemory_exit_helper atexit function may be removed in
3914     // the future if the appropriate cleanup code can be added to the
3915     // VM_Exit VMOperation's doit method.
3916     if (atexit(perfMemory_exit_helper) != 0) {
3917       warning("os::init2 atexit(perfMemory_exit_helper) failed");
3918     }
3919   }
3920 
3921   // initialize thread priority policy
3922   prio_init();
3923 
3924   return JNI_OK;
3925 }
3926 
3927 // Mark the polling page as unreadable
3928 void os::make_polling_page_unreadable(void) {
3929   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
3930     fatal("Could not disable polling page");
3931 };
3932 
3933 // Mark the polling page as readable
3934 void os::make_polling_page_readable(void) {
3935   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
3936     fatal("Could not enable polling page");
3937   }
3938 };
3939 
3940 int os::active_processor_count() {
3941   // Linux doesn't yet have a (official) notion of processor sets,
3942   // so just return the number of online processors.
3943   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
3944   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
3945   return online_cpus;
3946 }
3947 
3948 bool os::distribute_processes(uint length, uint* distribution) {
3949   // Not yet implemented.
3950   return false;
3951 }
3952 
3953 bool os::bind_to_processor(uint processor_id) {
3954   // Not yet implemented.
3955   return false;
3956 }
3957 
3958 ///
3959 
3960 // Suspends the target using the signal mechanism and then grabs the PC before
3961 // resuming the target. Used by the flat-profiler only
3962 ExtendedPC os::get_thread_pc(Thread* thread) {
3963   // Make sure that it is called by the watcher for the VMThread
3964   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
3965   assert(thread->is_VM_thread(), "Can only be called for VMThread");
3966 
3967   ExtendedPC epc;
3968 
3969   OSThread* osthread = thread->osthread();
3970   if (do_suspend(osthread)) {
3971     if (osthread->ucontext() != NULL) {
3972       epc = os::Linux::ucontext_get_pc(osthread->ucontext());
3973     } else {
3974       // NULL context is unexpected, double-check this is the VMThread
3975       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3976     }
3977     do_resume(osthread);
3978   }
3979   // failure means pthread_kill failed for some reason - arguably this is
3980   // a fatal problem, but such problems are ignored elsewhere
3981 
3982   return epc;
3983 }
3984 
3985 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
3986 {
3987    if (is_NPTL()) {
3988       return pthread_cond_timedwait(_cond, _mutex, _abstime);
3989    } else {
3990 #ifndef IA64
3991       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
3992       // word back to default 64bit precision if condvar is signaled. Java
3993       // wants 53bit precision.  Save and restore current value.
3994       int fpu = get_fpu_control_word();
3995 #endif // IA64
3996       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
3997 #ifndef IA64
3998       set_fpu_control_word(fpu);
3999 #endif // IA64
4000       return status;
4001    }
4002 }
4003 
4004 ////////////////////////////////////////////////////////////////////////////////
4005 // debug support
4006 
4007 #ifndef PRODUCT
4008 static address same_page(address x, address y) {
4009   int page_bits = -os::vm_page_size();
4010   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4011     return x;
4012   else if (x > y)
4013     return (address)(intptr_t(y) | ~page_bits) + 1;
4014   else
4015     return (address)(intptr_t(y) & page_bits);
4016 }
4017 
4018 bool os::find(address addr) {
4019   Dl_info dlinfo;
4020   memset(&dlinfo, 0, sizeof(dlinfo));
4021   if (dladdr(addr, &dlinfo)) {
4022     tty->print(PTR_FORMAT ": ", addr);
4023     if (dlinfo.dli_sname != NULL) {
4024       tty->print("%s+%#x", dlinfo.dli_sname,
4025                  addr - (intptr_t)dlinfo.dli_saddr);
4026     } else if (dlinfo.dli_fname) {
4027       tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4028     } else {
4029       tty->print("<absolute address>");
4030     }
4031     if (dlinfo.dli_fname) {
4032       tty->print(" in %s", dlinfo.dli_fname);
4033     }
4034     if (dlinfo.dli_fbase) {
4035       tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4036     }
4037     tty->cr();
4038 
4039     if (Verbose) {
4040       // decode some bytes around the PC
4041       address begin = same_page(addr-40, addr);
4042       address end   = same_page(addr+40, addr);
4043       address       lowest = (address) dlinfo.dli_sname;
4044       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4045       if (begin < lowest)  begin = lowest;
4046       Dl_info dlinfo2;
4047       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4048           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4049         end = (address) dlinfo2.dli_saddr;
4050       Disassembler::decode(begin, end);
4051     }
4052     return true;
4053   }
4054   return false;
4055 }
4056 
4057 #endif
4058 
4059 ////////////////////////////////////////////////////////////////////////////////
4060 // misc
4061 
4062 // This does not do anything on Linux. This is basically a hook for being
4063 // able to use structured exception handling (thread-local exception filters)
4064 // on, e.g., Win32.
4065 void
4066 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4067                          JavaCallArguments* args, Thread* thread) {
4068   f(value, method, args, thread);
4069 }
4070 
4071 void os::print_statistics() {
4072 }
4073 
4074 int os::message_box(const char* title, const char* message) {
4075   int i;
4076   fdStream err(defaultStream::error_fd());
4077   for (i = 0; i < 78; i++) err.print_raw("=");
4078   err.cr();
4079   err.print_raw_cr(title);
4080   for (i = 0; i < 78; i++) err.print_raw("-");
4081   err.cr();
4082   err.print_raw_cr(message);
4083   for (i = 0; i < 78; i++) err.print_raw("=");
4084   err.cr();
4085 
4086   char buf[16];
4087   // Prevent process from exiting upon "read error" without consuming all CPU
4088   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4089 
4090   return buf[0] == 'y' || buf[0] == 'Y';
4091 }
4092 
4093 int os::stat(const char *path, struct stat *sbuf) {
4094   char pathbuf[MAX_PATH];
4095   if (strlen(path) > MAX_PATH - 1) {
4096     errno = ENAMETOOLONG;
4097     return -1;
4098   }
4099   hpi::native_path(strcpy(pathbuf, path));
4100   return ::stat(pathbuf, sbuf);
4101 }
4102 
4103 bool os::check_heap(bool force) {
4104   return true;
4105 }
4106 
4107 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4108   return ::vsnprintf(buf, count, format, args);
4109 }
4110 
4111 // Is a (classpath) directory empty?
4112 bool os::dir_is_empty(const char* path) {
4113   DIR *dir = NULL;
4114   struct dirent *ptr;
4115 
4116   dir = opendir(path);
4117   if (dir == NULL) return true;
4118 
4119   /* Scan the directory */
4120   bool result = true;
4121   char buf[sizeof(struct dirent) + MAX_PATH];
4122   while (result && (ptr = ::readdir(dir)) != NULL) {
4123     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4124       result = false;
4125     }
4126   }
4127   closedir(dir);
4128   return result;
4129 }
4130 
4131 // create binary file, rewriting existing file if required
4132 int os::create_binary_file(const char* path, bool rewrite_existing) {
4133   int oflags = O_WRONLY | O_CREAT;
4134   if (!rewrite_existing) {
4135     oflags |= O_EXCL;
4136   }
4137   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4138 }
4139 
4140 // return current position of file pointer
4141 jlong os::current_file_offset(int fd) {
4142   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4143 }
4144 
4145 // move file pointer to the specified offset
4146 jlong os::seek_to_file_offset(int fd, jlong offset) {
4147   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4148 }
4149 
4150 // Map a block of memory.
4151 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4152                      char *addr, size_t bytes, bool read_only,
4153                      bool allow_exec) {
4154   int prot;
4155   int flags;
4156 
4157   if (read_only) {
4158     prot = PROT_READ;
4159     flags = MAP_SHARED;
4160   } else {
4161     prot = PROT_READ | PROT_WRITE;
4162     flags = MAP_PRIVATE;
4163   }
4164 
4165   if (allow_exec) {
4166     prot |= PROT_EXEC;
4167   }
4168 
4169   if (addr != NULL) {
4170     flags |= MAP_FIXED;
4171   }
4172 
4173   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4174                                      fd, file_offset);
4175   if (mapped_address == MAP_FAILED) {
4176     return NULL;
4177   }
4178   return mapped_address;
4179 }
4180 
4181 
4182 // Remap a block of memory.
4183 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4184                        char *addr, size_t bytes, bool read_only,
4185                        bool allow_exec) {
4186   // same as map_memory() on this OS
4187   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4188                         allow_exec);
4189 }
4190 
4191 
4192 // Unmap a block of memory.
4193 bool os::unmap_memory(char* addr, size_t bytes) {
4194   return munmap(addr, bytes) == 0;
4195 }
4196 
4197 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4198 
4199 static clockid_t thread_cpu_clockid(Thread* thread) {
4200   pthread_t tid = thread->osthread()->pthread_id();
4201   clockid_t clockid;
4202 
4203   // Get thread clockid
4204   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4205   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4206   return clockid;
4207 }
4208 
4209 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4210 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4211 // of a thread.
4212 //
4213 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4214 // the fast estimate available on the platform.
4215 
4216 jlong os::current_thread_cpu_time() {
4217   if (os::Linux::supports_fast_thread_cpu_time()) {
4218     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4219   } else {
4220     // return user + sys since the cost is the same
4221     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4222   }
4223 }
4224 
4225 jlong os::thread_cpu_time(Thread* thread) {
4226   // consistent with what current_thread_cpu_time() returns
4227   if (os::Linux::supports_fast_thread_cpu_time()) {
4228     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4229   } else {
4230     return slow_thread_cpu_time(thread, true /* user + sys */);
4231   }
4232 }
4233 
4234 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4235   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4236     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4237   } else {
4238     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4239   }
4240 }
4241 
4242 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4243   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4244     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4245   } else {
4246     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4247   }
4248 }
4249 
4250 //
4251 //  -1 on error.
4252 //
4253 
4254 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4255   static bool proc_pid_cpu_avail = true;
4256   static bool proc_task_unchecked = true;
4257   static const char *proc_stat_path = "/proc/%d/stat";
4258   pid_t  tid = thread->osthread()->thread_id();
4259   int i;
4260   char *s;
4261   char stat[2048];
4262   int statlen;
4263   char proc_name[64];
4264   int count;
4265   long sys_time, user_time;
4266   char string[64];
4267   int idummy;
4268   long ldummy;
4269   FILE *fp;
4270 
4271   // We first try accessing /proc/<pid>/cpu since this is faster to
4272   // process.  If this file is not present (linux kernels 2.5 and above)
4273   // then we open /proc/<pid>/stat.
4274   if ( proc_pid_cpu_avail ) {
4275     sprintf(proc_name, "/proc/%d/cpu", tid);
4276     fp =  fopen(proc_name, "r");
4277     if ( fp != NULL ) {
4278       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4279       fclose(fp);
4280       if ( count != 3 ) return -1;
4281 
4282       if (user_sys_cpu_time) {
4283         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4284       } else {
4285         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4286       }
4287     }
4288     else proc_pid_cpu_avail = false;
4289   }
4290 
4291   // The /proc/<tid>/stat aggregates per-process usage on
4292   // new Linux kernels 2.6+ where NPTL is supported.
4293   // The /proc/self/task/<tid>/stat still has the per-thread usage.
4294   // See bug 6328462.
4295   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4296   // and possibly in some other cases, so we check its availability.
4297   if (proc_task_unchecked && os::Linux::is_NPTL()) {
4298     // This is executed only once
4299     proc_task_unchecked = false;
4300     fp = fopen("/proc/self/task", "r");
4301     if (fp != NULL) {
4302       proc_stat_path = "/proc/self/task/%d/stat";
4303       fclose(fp);
4304     }
4305   }
4306 
4307   sprintf(proc_name, proc_stat_path, tid);
4308   fp = fopen(proc_name, "r");
4309   if ( fp == NULL ) return -1;
4310   statlen = fread(stat, 1, 2047, fp);
4311   stat[statlen] = '\0';
4312   fclose(fp);
4313 
4314   // Skip pid and the command string. Note that we could be dealing with
4315   // weird command names, e.g. user could decide to rename java launcher
4316   // to "java 1.4.2 :)", then the stat file would look like
4317   //                1234 (java 1.4.2 :)) R ... ...
4318   // We don't really need to know the command string, just find the last
4319   // occurrence of ")" and then start parsing from there. See bug 4726580.
4320   s = strrchr(stat, ')');
4321   i = 0;
4322   if (s == NULL ) return -1;
4323 
4324   // Skip blank chars
4325   do s++; while (isspace(*s));
4326 
4327   count = sscanf(s,"%*c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4328                  &idummy, &idummy, &idummy, &idummy, &idummy,
4329                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4330                  &user_time, &sys_time);
4331   if ( count != 12 ) return -1;
4332   if (user_sys_cpu_time) {
4333     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4334   } else {
4335     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4336   }
4337 }
4338 
4339 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4340   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4341   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4342   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4343   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4344 }
4345 
4346 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4347   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4348   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4349   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4350   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4351 }
4352 
4353 bool os::is_thread_cpu_time_supported() {
4354   return true;
4355 }
4356 
4357 // System loadavg support.  Returns -1 if load average cannot be obtained.
4358 // Linux doesn't yet have a (official) notion of processor sets,
4359 // so just return the system wide load average.
4360 int os::loadavg(double loadavg[], int nelem) {
4361   return ::getloadavg(loadavg, nelem);
4362 }
4363 
4364 void os::pause() {
4365   char filename[MAX_PATH];
4366   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4367     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4368   } else {
4369     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4370   }
4371 
4372   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4373   if (fd != -1) {
4374     struct stat buf;
4375     close(fd);
4376     while (::stat(filename, &buf) == 0) {
4377       (void)::poll(NULL, 0, 100);
4378     }
4379   } else {
4380     jio_fprintf(stderr,
4381       "Could not open pause file '%s', continuing immediately.\n", filename);
4382   }
4383 }
4384 
4385 extern "C" {
4386 
4387 /**
4388  * NOTE: the following code is to keep the green threads code
4389  * in the libjava.so happy. Once the green threads is removed,
4390  * these code will no longer be needed.
4391  */
4392 int
4393 jdk_waitpid(pid_t pid, int* status, int options) {
4394     return waitpid(pid, status, options);
4395 }
4396 
4397 int
4398 fork1() {
4399     return fork();
4400 }
4401 
4402 int
4403 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4404     return sem_init(sem, pshared, value);
4405 }
4406 
4407 int
4408 jdk_sem_post(sem_t *sem) {
4409     return sem_post(sem);
4410 }
4411 
4412 int
4413 jdk_sem_wait(sem_t *sem) {
4414     return sem_wait(sem);
4415 }
4416 
4417 int
4418 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4419     return pthread_sigmask(how , newmask, oldmask);
4420 }
4421 
4422 }
4423 
4424 // Refer to the comments in os_solaris.cpp park-unpark.
4425 //
4426 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4427 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4428 // For specifics regarding the bug see GLIBC BUGID 261237 :
4429 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4430 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4431 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4432 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4433 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4434 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4435 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4436 // of libpthread avoids the problem, but isn't practical.
4437 //
4438 // Possible remedies:
4439 //
4440 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4441 //      This is palliative and probabilistic, however.  If the thread is preempted
4442 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4443 //      than the minimum period may have passed, and the abstime may be stale (in the
4444 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
4445 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4446 //
4447 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4448 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
4449 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4450 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
4451 //      thread.
4452 //
4453 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4454 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
4455 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4456 //      This also works well.  In fact it avoids kernel-level scalability impediments
4457 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4458 //      timers in a graceful fashion.
4459 //
4460 // 4.   When the abstime value is in the past it appears that control returns
4461 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4462 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
4463 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
4464 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
4465 //      It may be possible to avoid reinitialization by checking the return
4466 //      value from pthread_cond_timedwait().  In addition to reinitializing the
4467 //      condvar we must establish the invariant that cond_signal() is only called
4468 //      within critical sections protected by the adjunct mutex.  This prevents
4469 //      cond_signal() from "seeing" a condvar that's in the midst of being
4470 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
4471 //      desirable signal-after-unlock optimization that avoids futile context switching.
4472 //
4473 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
4474 //      structure when a condvar is used or initialized.  cond_destroy()  would
4475 //      release the helper structure.  Our reinitialize-after-timedwait fix
4476 //      put excessive stress on malloc/free and locks protecting the c-heap.
4477 //
4478 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
4479 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4480 // and only enabling the work-around for vulnerable environments.
4481 
4482 // utility to compute the abstime argument to timedwait:
4483 // millis is the relative timeout time
4484 // abstime will be the absolute timeout time
4485 // TODO: replace compute_abstime() with unpackTime()
4486 
4487 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4488   if (millis < 0)  millis = 0;
4489   struct timeval now;
4490   int status = gettimeofday(&now, NULL);
4491   assert(status == 0, "gettimeofday");
4492   jlong seconds = millis / 1000;
4493   millis %= 1000;
4494   if (seconds > 50000000) { // see man cond_timedwait(3T)
4495     seconds = 50000000;
4496   }
4497   abstime->tv_sec = now.tv_sec  + seconds;
4498   long       usec = now.tv_usec + millis * 1000;
4499   if (usec >= 1000000) {
4500     abstime->tv_sec += 1;
4501     usec -= 1000000;
4502   }
4503   abstime->tv_nsec = usec * 1000;
4504   return abstime;
4505 }
4506 
4507 
4508 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4509 // Conceptually TryPark() should be equivalent to park(0).
4510 
4511 int os::PlatformEvent::TryPark() {
4512   for (;;) {
4513     const int v = _Event ;
4514     guarantee ((v == 0) || (v == 1), "invariant") ;
4515     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
4516   }
4517 }
4518 
4519 void os::PlatformEvent::park() {       // AKA "down()"
4520   // Invariant: Only the thread associated with the Event/PlatformEvent
4521   // may call park().
4522   // TODO: assert that _Assoc != NULL or _Assoc == Self
4523   int v ;
4524   for (;;) {
4525       v = _Event ;
4526       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4527   }
4528   guarantee (v >= 0, "invariant") ;
4529   if (v == 0) {
4530      // Do this the hard way by blocking ...
4531      int status = pthread_mutex_lock(_mutex);
4532      assert_status(status == 0, status, "mutex_lock");
4533      guarantee (_nParked == 0, "invariant") ;
4534      ++ _nParked ;
4535      while (_Event < 0) {
4536         status = pthread_cond_wait(_cond, _mutex);
4537         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4538         // Treat this the same as if the wait was interrupted
4539         if (status == ETIME) { status = EINTR; }
4540         assert_status(status == 0 || status == EINTR, status, "cond_wait");
4541      }
4542      -- _nParked ;
4543 
4544     // In theory we could move the ST of 0 into _Event past the unlock(),
4545     // but then we'd need a MEMBAR after the ST.
4546     _Event = 0 ;
4547      status = pthread_mutex_unlock(_mutex);
4548      assert_status(status == 0, status, "mutex_unlock");
4549   }
4550   guarantee (_Event >= 0, "invariant") ;
4551 }
4552 
4553 int os::PlatformEvent::park(jlong millis) {
4554   guarantee (_nParked == 0, "invariant") ;
4555 
4556   int v ;
4557   for (;;) {
4558       v = _Event ;
4559       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4560   }
4561   guarantee (v >= 0, "invariant") ;
4562   if (v != 0) return OS_OK ;
4563 
4564   // We do this the hard way, by blocking the thread.
4565   // Consider enforcing a minimum timeout value.
4566   struct timespec abst;
4567   compute_abstime(&abst, millis);
4568 
4569   int ret = OS_TIMEOUT;
4570   int status = pthread_mutex_lock(_mutex);
4571   assert_status(status == 0, status, "mutex_lock");
4572   guarantee (_nParked == 0, "invariant") ;
4573   ++_nParked ;
4574 
4575   // Object.wait(timo) will return because of
4576   // (a) notification
4577   // (b) timeout
4578   // (c) thread.interrupt
4579   //
4580   // Thread.interrupt and object.notify{All} both call Event::set.
4581   // That is, we treat thread.interrupt as a special case of notification.
4582   // The underlying Solaris implementation, cond_timedwait, admits
4583   // spurious/premature wakeups, but the JLS/JVM spec prevents the
4584   // JVM from making those visible to Java code.  As such, we must
4585   // filter out spurious wakeups.  We assume all ETIME returns are valid.
4586   //
4587   // TODO: properly differentiate simultaneous notify+interrupt.
4588   // In that case, we should propagate the notify to another waiter.
4589 
4590   while (_Event < 0) {
4591     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4592     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4593       pthread_cond_destroy (_cond);
4594       pthread_cond_init (_cond, NULL) ;
4595     }
4596     assert_status(status == 0 || status == EINTR ||
4597                   status == ETIME || status == ETIMEDOUT,
4598                   status, "cond_timedwait");
4599     if (!FilterSpuriousWakeups) break ;                 // previous semantics
4600     if (status == ETIME || status == ETIMEDOUT) break ;
4601     // We consume and ignore EINTR and spurious wakeups.
4602   }
4603   --_nParked ;
4604   if (_Event >= 0) {
4605      ret = OS_OK;
4606   }
4607   _Event = 0 ;
4608   status = pthread_mutex_unlock(_mutex);
4609   assert_status(status == 0, status, "mutex_unlock");
4610   assert (_nParked == 0, "invariant") ;
4611   return ret;
4612 }
4613 
4614 void os::PlatformEvent::unpark() {
4615   int v, AnyWaiters ;
4616   for (;;) {
4617       v = _Event ;
4618       if (v > 0) {
4619          // The LD of _Event could have reordered or be satisfied
4620          // by a read-aside from this processor's write buffer.
4621          // To avoid problems execute a barrier and then
4622          // ratify the value.
4623          OrderAccess::fence() ;
4624          if (_Event == v) return ;
4625          continue ;
4626       }
4627       if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4628   }
4629   if (v < 0) {
4630      // Wait for the thread associated with the event to vacate
4631      int status = pthread_mutex_lock(_mutex);
4632      assert_status(status == 0, status, "mutex_lock");
4633      AnyWaiters = _nParked ;
4634      assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4635      if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4636         AnyWaiters = 0 ;
4637         pthread_cond_signal (_cond);
4638      }
4639      status = pthread_mutex_unlock(_mutex);
4640      assert_status(status == 0, status, "mutex_unlock");
4641      if (AnyWaiters != 0) {
4642         status = pthread_cond_signal(_cond);
4643         assert_status(status == 0, status, "cond_signal");
4644      }
4645   }
4646 
4647   // Note that we signal() _after dropping the lock for "immortal" Events.
4648   // This is safe and avoids a common class of  futile wakeups.  In rare
4649   // circumstances this can cause a thread to return prematurely from
4650   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4651   // simply re-test the condition and re-park itself.
4652 }
4653 
4654 
4655 // JSR166
4656 // -------------------------------------------------------
4657 
4658 /*
4659  * The solaris and linux implementations of park/unpark are fairly
4660  * conservative for now, but can be improved. They currently use a
4661  * mutex/condvar pair, plus a a count.
4662  * Park decrements count if > 0, else does a condvar wait.  Unpark
4663  * sets count to 1 and signals condvar.  Only one thread ever waits
4664  * on the condvar. Contention seen when trying to park implies that someone
4665  * is unparking you, so don't wait. And spurious returns are fine, so there
4666  * is no need to track notifications.
4667  */
4668 
4669 
4670 #define NANOSECS_PER_SEC 1000000000
4671 #define NANOSECS_PER_MILLISEC 1000000
4672 #define MAX_SECS 100000000
4673 /*
4674  * This code is common to linux and solaris and will be moved to a
4675  * common place in dolphin.
4676  *
4677  * The passed in time value is either a relative time in nanoseconds
4678  * or an absolute time in milliseconds. Either way it has to be unpacked
4679  * into suitable seconds and nanoseconds components and stored in the
4680  * given timespec structure.
4681  * Given time is a 64-bit value and the time_t used in the timespec is only
4682  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4683  * overflow if times way in the future are given. Further on Solaris versions
4684  * prior to 10 there is a restriction (see cond_timedwait) that the specified
4685  * number of seconds, in abstime, is less than current_time  + 100,000,000.
4686  * As it will be 28 years before "now + 100000000" will overflow we can
4687  * ignore overflow and just impose a hard-limit on seconds using the value
4688  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4689  * years from "now".
4690  */
4691 
4692 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4693   assert (time > 0, "convertTime");
4694 
4695   struct timeval now;
4696   int status = gettimeofday(&now, NULL);
4697   assert(status == 0, "gettimeofday");
4698 
4699   time_t max_secs = now.tv_sec + MAX_SECS;
4700 
4701   if (isAbsolute) {
4702     jlong secs = time / 1000;
4703     if (secs > max_secs) {
4704       absTime->tv_sec = max_secs;
4705     }
4706     else {
4707       absTime->tv_sec = secs;
4708     }
4709     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4710   }
4711   else {
4712     jlong secs = time / NANOSECS_PER_SEC;
4713     if (secs >= MAX_SECS) {
4714       absTime->tv_sec = max_secs;
4715       absTime->tv_nsec = 0;
4716     }
4717     else {
4718       absTime->tv_sec = now.tv_sec + secs;
4719       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4720       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4721         absTime->tv_nsec -= NANOSECS_PER_SEC;
4722         ++absTime->tv_sec; // note: this must be <= max_secs
4723       }
4724     }
4725   }
4726   assert(absTime->tv_sec >= 0, "tv_sec < 0");
4727   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4728   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4729   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4730 }
4731 
4732 void Parker::park(bool isAbsolute, jlong time) {
4733   // Optional fast-path check:
4734   // Return immediately if a permit is available.
4735   if (_counter > 0) {
4736       _counter = 0 ;
4737       OrderAccess::fence();
4738       return ;
4739   }
4740 
4741   Thread* thread = Thread::current();
4742   assert(thread->is_Java_thread(), "Must be JavaThread");
4743   JavaThread *jt = (JavaThread *)thread;
4744 
4745   // Optional optimization -- avoid state transitions if there's an interrupt pending.
4746   // Check interrupt before trying to wait
4747   if (Thread::is_interrupted(thread, false)) {
4748     return;
4749   }
4750 
4751   // Next, demultiplex/decode time arguments
4752   timespec absTime;
4753   if (time < 0) { // don't wait at all
4754     return;
4755   }
4756   if (time > 0) {
4757     unpackTime(&absTime, isAbsolute, time);
4758   }
4759 
4760 
4761   // Enter safepoint region
4762   // Beware of deadlocks such as 6317397.
4763   // The per-thread Parker:: mutex is a classic leaf-lock.
4764   // In particular a thread must never block on the Threads_lock while
4765   // holding the Parker:: mutex.  If safepoints are pending both the
4766   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4767   ThreadBlockInVM tbivm(jt);
4768 
4769   // Don't wait if cannot get lock since interference arises from
4770   // unblocking.  Also. check interrupt before trying wait
4771   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4772     return;
4773   }
4774 
4775   int status ;
4776   if (_counter > 0)  { // no wait needed
4777     _counter = 0;
4778     status = pthread_mutex_unlock(_mutex);
4779     assert (status == 0, "invariant") ;
4780     OrderAccess::fence();
4781     return;
4782   }
4783 
4784 #ifdef ASSERT
4785   // Don't catch signals while blocked; let the running threads have the signals.
4786   // (This allows a debugger to break into the running thread.)
4787   sigset_t oldsigs;
4788   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4789   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4790 #endif
4791 
4792   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4793   jt->set_suspend_equivalent();
4794   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4795 
4796   if (time == 0) {
4797     status = pthread_cond_wait (_cond, _mutex) ;
4798   } else {
4799     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4800     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4801       pthread_cond_destroy (_cond) ;
4802       pthread_cond_init    (_cond, NULL);
4803     }
4804   }
4805   assert_status(status == 0 || status == EINTR ||
4806                 status == ETIME || status == ETIMEDOUT,
4807                 status, "cond_timedwait");
4808 
4809 #ifdef ASSERT
4810   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4811 #endif
4812 
4813   _counter = 0 ;
4814   status = pthread_mutex_unlock(_mutex) ;
4815   assert_status(status == 0, status, "invariant") ;
4816   // If externally suspended while waiting, re-suspend
4817   if (jt->handle_special_suspend_equivalent_condition()) {
4818     jt->java_suspend_self();
4819   }
4820 
4821   OrderAccess::fence();
4822 }
4823 
4824 void Parker::unpark() {
4825   int s, status ;
4826   status = pthread_mutex_lock(_mutex);
4827   assert (status == 0, "invariant") ;
4828   s = _counter;
4829   _counter = 1;
4830   if (s < 1) {
4831      if (WorkAroundNPTLTimedWaitHang) {
4832         status = pthread_cond_signal (_cond) ;
4833         assert (status == 0, "invariant") ;
4834         status = pthread_mutex_unlock(_mutex);
4835         assert (status == 0, "invariant") ;
4836      } else {
4837         status = pthread_mutex_unlock(_mutex);
4838         assert (status == 0, "invariant") ;
4839         status = pthread_cond_signal (_cond) ;
4840         assert (status == 0, "invariant") ;
4841      }
4842   } else {
4843     pthread_mutex_unlock(_mutex);
4844     assert (status == 0, "invariant") ;
4845   }
4846 }
4847 
4848 
4849 extern char** environ;
4850 
4851 #ifndef __NR_fork
4852 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
4853 #endif
4854 
4855 #ifndef __NR_execve
4856 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
4857 #endif
4858 
4859 // Run the specified command in a separate process. Return its exit value,
4860 // or -1 on failure (e.g. can't fork a new process).
4861 // Unlike system(), this function can be called from signal handler. It
4862 // doesn't block SIGINT et al.
4863 int os::fork_and_exec(char* cmd) {
4864   const char * argv[4] = {"sh", "-c", cmd, NULL};
4865 
4866   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
4867   // pthread_atfork handlers and reset pthread library. All we need is a
4868   // separate process to execve. Make a direct syscall to fork process.
4869   // On IA64 there's no fork syscall, we have to use fork() and hope for
4870   // the best...
4871   pid_t pid = NOT_IA64(syscall(__NR_fork);)
4872               IA64_ONLY(fork();)
4873 
4874   if (pid < 0) {
4875     // fork failed
4876     return -1;
4877 
4878   } else if (pid == 0) {
4879     // child process
4880 
4881     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
4882     // first to kill every thread on the thread list. Because this list is
4883     // not reset by fork() (see notes above), execve() will instead kill
4884     // every thread in the parent process. We know this is the only thread
4885     // in the new process, so make a system call directly.
4886     // IA64 should use normal execve() from glibc to match the glibc fork()
4887     // above.
4888     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
4889     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
4890 
4891     // execve failed
4892     _exit(-1);
4893 
4894   } else  {
4895     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4896     // care about the actual exit code, for now.
4897 
4898     int status;
4899 
4900     // Wait for the child process to exit.  This returns immediately if
4901     // the child has already exited. */
4902     while (waitpid(pid, &status, 0) < 0) {
4903         switch (errno) {
4904         case ECHILD: return 0;
4905         case EINTR: break;
4906         default: return -1;
4907         }
4908     }
4909 
4910     if (WIFEXITED(status)) {
4911        // The child exited normally; get its exit code.
4912        return WEXITSTATUS(status);
4913     } else if (WIFSIGNALED(status)) {
4914        // The child exited because of a signal
4915        // The best value to return is 0x80 + signal number,
4916        // because that is what all Unix shells do, and because
4917        // it allows callers to distinguish between process exit and
4918        // process death by signal.
4919        return 0x80 + WTERMSIG(status);
4920     } else {
4921        // Unknown exit code; pass it through
4922        return status;
4923     }
4924   }
4925 }