1 /*
   2  * Copyright (c) 1999, 2013, Oracle and/or its affiliates. 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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "interpreter/interpreter.hpp"
  33 #include "jvm_bsd.h"
  34 #include "memory/allocation.inline.hpp"
  35 #include "memory/filemap.hpp"
  36 #include "mutex_bsd.inline.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "os_share_bsd.hpp"
  39 #include "prims/jniFastGetField.hpp"
  40 #include "prims/jvm.h"
  41 #include "prims/jvm_misc.hpp"
  42 #include "runtime/arguments.hpp"
  43 #include "runtime/extendedPC.hpp"
  44 #include "runtime/globals.hpp"
  45 #include "runtime/interfaceSupport.hpp"
  46 #include "runtime/java.hpp"
  47 #include "runtime/javaCalls.hpp"
  48 #include "runtime/mutexLocker.hpp"
  49 #include "runtime/objectMonitor.hpp"
  50 #include "runtime/osThread.hpp"
  51 #include "runtime/perfMemory.hpp"
  52 #include "runtime/sharedRuntime.hpp"
  53 #include "runtime/statSampler.hpp"
  54 #include "runtime/stubRoutines.hpp"
  55 #include "runtime/threadCritical.hpp"
  56 #include "runtime/timer.hpp"
  57 #include "services/attachListener.hpp"
  58 #include "services/memTracker.hpp"
  59 #include "services/runtimeService.hpp"
  60 #include "thread_bsd.inline.hpp"
  61 #include "utilities/decoder.hpp"
  62 #include "utilities/defaultStream.hpp"
  63 #include "utilities/events.hpp"
  64 #include "utilities/growableArray.hpp"
  65 #include "utilities/vmError.hpp"
  66 #ifdef TARGET_ARCH_x86
  67 # include "assembler_x86.inline.hpp"
  68 # include "nativeInst_x86.hpp"
  69 #endif
  70 #ifdef TARGET_ARCH_sparc
  71 # include "assembler_sparc.inline.hpp"
  72 # include "nativeInst_sparc.hpp"
  73 #endif
  74 #ifdef TARGET_ARCH_zero
  75 # include "assembler_zero.inline.hpp"
  76 # include "nativeInst_zero.hpp"
  77 #endif
  78 #ifdef TARGET_ARCH_arm
  79 # include "assembler_arm.inline.hpp"
  80 # include "nativeInst_arm.hpp"
  81 #endif
  82 #ifdef TARGET_ARCH_ppc
  83 # include "assembler_ppc.inline.hpp"
  84 # include "nativeInst_ppc.hpp"
  85 #endif
  86 
  87 // put OS-includes here
  88 # include <sys/types.h>
  89 # include <sys/mman.h>
  90 # include <sys/stat.h>
  91 # include <sys/select.h>
  92 # include <pthread.h>
  93 # include <signal.h>
  94 # include <errno.h>
  95 # include <dlfcn.h>
  96 # include <stdio.h>
  97 # include <unistd.h>
  98 # include <sys/resource.h>
  99 # include <pthread.h>
 100 # include <sys/stat.h>
 101 # include <sys/time.h>
 102 # include <sys/times.h>
 103 # include <sys/utsname.h>
 104 # include <sys/socket.h>
 105 # include <sys/wait.h>
 106 # include <time.h>
 107 # include <pwd.h>
 108 # include <poll.h>
 109 # include <semaphore.h>
 110 # include <fcntl.h>
 111 # include <string.h>
 112 #ifdef _ALLBSD_SOURCE
 113 # include <sys/param.h>
 114 # include <sys/sysctl.h>
 115 #else
 116 # include <syscall.h>
 117 # include <sys/sysinfo.h>
 118 # include <gnu/libc-version.h>
 119 #endif
 120 # include <sys/ipc.h>
 121 # include <sys/shm.h>
 122 #ifndef __APPLE__
 123 # include <link.h>
 124 #endif
 125 # include <stdint.h>
 126 # include <inttypes.h>
 127 # include <sys/ioctl.h>
 128 
 129 #if defined(__FreeBSD__) || defined(__NetBSD__)
 130 # include <elf.h>
 131 #endif
 132 
 133 #ifdef __APPLE__
 134 # include <mach/mach.h> // semaphore_* API
 135 # include <mach-o/dyld.h>
 136 # include <sys/proc_info.h>
 137 # include <objc/objc-auto.h>
 138 #endif
 139 
 140 #ifndef MAP_ANONYMOUS
 141 #define MAP_ANONYMOUS MAP_ANON
 142 #endif
 143 
 144 #define MAX_PATH    (2 * K)
 145 
 146 // for timer info max values which include all bits
 147 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 148 
 149 #define LARGEPAGES_BIT (1 << 6)
 150 ////////////////////////////////////////////////////////////////////////////////
 151 // global variables
 152 julong os::Bsd::_physical_memory = 0;
 153 
 154 #ifndef _ALLBSD_SOURCE
 155 address   os::Bsd::_initial_thread_stack_bottom = NULL;
 156 uintptr_t os::Bsd::_initial_thread_stack_size   = 0;
 157 #endif
 158 
 159 int (*os::Bsd::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 160 #ifndef _ALLBSD_SOURCE
 161 int (*os::Bsd::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 162 Mutex* os::Bsd::_createThread_lock = NULL;
 163 #endif
 164 pthread_t os::Bsd::_main_thread;
 165 int os::Bsd::_page_size = -1;
 166 #ifndef _ALLBSD_SOURCE
 167 bool os::Bsd::_is_floating_stack = false;
 168 bool os::Bsd::_is_NPTL = false;
 169 bool os::Bsd::_supports_fast_thread_cpu_time = false;
 170 const char * os::Bsd::_glibc_version = NULL;
 171 const char * os::Bsd::_libpthread_version = NULL;
 172 #endif
 173 
 174 static jlong initial_time_count=0;
 175 
 176 static int clock_tics_per_sec = 100;
 177 
 178 // For diagnostics to print a message once. see run_periodic_checks
 179 static sigset_t check_signal_done;
 180 static bool check_signals = true;;
 181 
 182 static pid_t _initial_pid = 0;
 183 
 184 /* Signal number used to suspend/resume a thread */
 185 
 186 /* do not use any signal number less than SIGSEGV, see 4355769 */
 187 static int SR_signum = SIGUSR2;
 188 sigset_t SR_sigset;
 189 
 190 
 191 ////////////////////////////////////////////////////////////////////////////////
 192 // utility functions
 193 
 194 static int SR_initialize();
 195 static int SR_finalize();
 196 
 197 julong os::available_memory() {
 198   return Bsd::available_memory();
 199 }
 200 
 201 julong os::Bsd::available_memory() {
 202 #ifdef _ALLBSD_SOURCE
 203   // XXXBSD: this is just a stopgap implementation
 204   return physical_memory() >> 2;
 205 #else
 206   // values in struct sysinfo are "unsigned long"
 207   struct sysinfo si;
 208   sysinfo(&si);
 209 
 210   return (julong)si.freeram * si.mem_unit;
 211 #endif
 212 }
 213 
 214 julong os::physical_memory() {
 215   return Bsd::physical_memory();
 216 }
 217 
 218 julong os::allocatable_physical_memory(julong size) {
 219 #ifdef _LP64
 220   return size;
 221 #else
 222   julong result = MIN2(size, (julong)3800*M);
 223    if (!is_allocatable(result)) {
 224      // See comments under solaris for alignment considerations
 225      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 226      result =  MIN2(size, reasonable_size);
 227    }
 228    return result;
 229 #endif // _LP64
 230 }
 231 
 232 ////////////////////////////////////////////////////////////////////////////////
 233 // environment support
 234 
 235 bool os::getenv(const char* name, char* buf, int len) {
 236   const char* val = ::getenv(name);
 237   if (val != NULL && strlen(val) < (size_t)len) {
 238     strcpy(buf, val);
 239     return true;
 240   }
 241   if (len > 0) buf[0] = 0;  // return a null string
 242   return false;
 243 }
 244 
 245 
 246 // Return true if user is running as root.
 247 
 248 bool os::have_special_privileges() {
 249   static bool init = false;
 250   static bool privileges = false;
 251   if (!init) {
 252     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 253     init = true;
 254   }
 255   return privileges;
 256 }
 257 
 258 
 259 #ifndef _ALLBSD_SOURCE
 260 #ifndef SYS_gettid
 261 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 262 #ifdef __ia64__
 263 #define SYS_gettid 1105
 264 #elif __i386__
 265 #define SYS_gettid 224
 266 #elif __amd64__
 267 #define SYS_gettid 186
 268 #elif __sparc__
 269 #define SYS_gettid 143
 270 #else
 271 #error define gettid for the arch
 272 #endif
 273 #endif
 274 #endif
 275 
 276 // Cpu architecture string
 277 #if   defined(ZERO)
 278 static char cpu_arch[] = ZERO_LIBARCH;
 279 #elif defined(IA64)
 280 static char cpu_arch[] = "ia64";
 281 #elif defined(IA32)
 282 static char cpu_arch[] = "i386";
 283 #elif defined(AMD64)
 284 static char cpu_arch[] = "amd64";
 285 #elif defined(ARM)
 286 static char cpu_arch[] = "arm";
 287 #elif defined(PPC)
 288 static char cpu_arch[] = "ppc";
 289 #elif defined(SPARC)
 290 #  ifdef _LP64
 291 static char cpu_arch[] = "sparcv9";
 292 #  else
 293 static char cpu_arch[] = "sparc";
 294 #  endif
 295 #else
 296 #error Add appropriate cpu_arch setting
 297 #endif
 298 
 299 // Compiler variant
 300 #ifdef COMPILER2
 301 #define COMPILER_VARIANT "server"
 302 #else
 303 #define COMPILER_VARIANT "client"
 304 #endif
 305 
 306 #ifndef _ALLBSD_SOURCE
 307 // pid_t gettid()
 308 //
 309 // Returns the kernel thread id of the currently running thread. Kernel
 310 // thread id is used to access /proc.
 311 //
 312 // (Note that getpid() on BsdThreads returns kernel thread id too; but
 313 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 314 //
 315 pid_t os::Bsd::gettid() {
 316   int rslt = syscall(SYS_gettid);
 317   if (rslt == -1) {
 318      // old kernel, no NPTL support
 319      return getpid();
 320   } else {
 321      return (pid_t)rslt;
 322   }
 323 }
 324 
 325 // Most versions of bsd have a bug where the number of processors are
 326 // determined by looking at the /proc file system.  In a chroot environment,
 327 // the system call returns 1.  This causes the VM to act as if it is
 328 // a single processor and elide locking (see is_MP() call).
 329 static bool unsafe_chroot_detected = false;
 330 static const char *unstable_chroot_error = "/proc file system not found.\n"
 331                      "Java may be unstable running multithreaded in a chroot "
 332                      "environment on Bsd when /proc filesystem is not mounted.";
 333 #endif
 334 
 335 #ifdef _ALLBSD_SOURCE
 336 void os::Bsd::initialize_system_info() {
 337   int mib[2];
 338   size_t len;
 339   int cpu_val;
 340   julong mem_val;
 341 
 342   /* get processors count via hw.ncpus sysctl */
 343   mib[0] = CTL_HW;
 344   mib[1] = HW_NCPU;
 345   len = sizeof(cpu_val);
 346   if (sysctl(mib, 2, &cpu_val, &len, NULL, 0) != -1 && cpu_val >= 1) {
 347        assert(len == sizeof(cpu_val), "unexpected data size");
 348        set_processor_count(cpu_val);
 349   }
 350   else {
 351        set_processor_count(1);   // fallback
 352   }
 353 
 354   /* get physical memory via hw.memsize sysctl (hw.memsize is used
 355    * since it returns a 64 bit value)
 356    */
 357   mib[0] = CTL_HW;
 358   mib[1] = HW_MEMSIZE;
 359   len = sizeof(mem_val);
 360   if (sysctl(mib, 2, &mem_val, &len, NULL, 0) != -1) {
 361        assert(len == sizeof(mem_val), "unexpected data size");
 362        _physical_memory = mem_val;
 363   } else {
 364        _physical_memory = 256*1024*1024;       // fallback (XXXBSD?)
 365   }
 366 
 367 #ifdef __OpenBSD__
 368   {
 369        // limit _physical_memory memory view on OpenBSD since
 370        // datasize rlimit restricts us anyway.
 371        struct rlimit limits;
 372        getrlimit(RLIMIT_DATA, &limits);
 373        _physical_memory = MIN2(_physical_memory, (julong)limits.rlim_cur);
 374   }
 375 #endif
 376 }
 377 #else
 378 void os::Bsd::initialize_system_info() {
 379   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 380   if (processor_count() == 1) {
 381     pid_t pid = os::Bsd::gettid();
 382     char fname[32];
 383     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 384     FILE *fp = fopen(fname, "r");
 385     if (fp == NULL) {
 386       unsafe_chroot_detected = true;
 387     } else {
 388       fclose(fp);
 389     }
 390   }
 391   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 392   assert(processor_count() > 0, "bsd error");
 393 }
 394 #endif
 395 
 396 #ifdef __APPLE__
 397 static const char *get_home() {
 398   const char *home_dir = ::getenv("HOME");
 399   if ((home_dir == NULL) || (*home_dir == '\0')) {
 400     struct passwd *passwd_info = getpwuid(geteuid());
 401     if (passwd_info != NULL) {
 402       home_dir = passwd_info->pw_dir;
 403     }
 404   }
 405 
 406   return home_dir;
 407 }
 408 #endif
 409 
 410 void os::init_system_properties_values() {
 411 //  char arch[12];
 412 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 413 
 414   // The next steps are taken in the product version:
 415   //
 416   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 417   // This library should be located at:
 418   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 419   //
 420   // If "/jre/lib/" appears at the right place in the path, then we
 421   // assume libjvm[_g].so is installed in a JDK and we use this path.
 422   //
 423   // Otherwise exit with message: "Could not create the Java virtual machine."
 424   //
 425   // The following extra steps are taken in the debugging version:
 426   //
 427   // If "/jre/lib/" does NOT appear at the right place in the path
 428   // instead of exit check for $JAVA_HOME environment variable.
 429   //
 430   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 431   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 432   // it looks like libjvm[_g].so is installed there
 433   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 434   //
 435   // Otherwise exit.
 436   //
 437   // Important note: if the location of libjvm.so changes this
 438   // code needs to be changed accordingly.
 439 
 440   // The next few definitions allow the code to be verbatim:
 441 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
 442 #define getenv(n) ::getenv(n)
 443 
 444 /*
 445  * See ld(1):
 446  *      The linker uses the following search paths to locate required
 447  *      shared libraries:
 448  *        1: ...
 449  *        ...
 450  *        7: The default directories, normally /lib and /usr/lib.
 451  */
 452 #ifndef DEFAULT_LIBPATH
 453 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 454 #endif
 455 
 456 #define EXTENSIONS_DIR  "/lib/ext"
 457 #define ENDORSED_DIR    "/lib/endorsed"
 458 #define REG_DIR         "/usr/java/packages"
 459 
 460 #ifdef __APPLE__
 461 #define SYS_EXTENSIONS_DIR   "/Library/Java/Extensions"
 462 #define SYS_EXTENSIONS_DIRS  SYS_EXTENSIONS_DIR ":/Network" SYS_EXTENSIONS_DIR ":/System" SYS_EXTENSIONS_DIR ":/usr/lib/java"
 463         const char *user_home_dir = get_home();
 464         // the null in SYS_EXTENSIONS_DIRS counts for the size of the colon after user_home_dir
 465         int system_ext_size = strlen(user_home_dir) + sizeof(SYS_EXTENSIONS_DIR) +
 466             sizeof(SYS_EXTENSIONS_DIRS);
 467 #endif
 468 
 469   {
 470     /* sysclasspath, java_home, dll_dir */
 471     {
 472         char *home_path;
 473         char *dll_path;
 474         char *pslash;
 475         char buf[MAXPATHLEN];
 476         os::jvm_path(buf, sizeof(buf));
 477 
 478         // Found the full path to libjvm.so.
 479         // Now cut the path to <java_home>/jre if we can.
 480         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 481         pslash = strrchr(buf, '/');
 482         if (pslash != NULL)
 483             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 484         dll_path = malloc(strlen(buf) + 1);
 485         if (dll_path == NULL)
 486             return;
 487         strcpy(dll_path, buf);
 488         Arguments::set_dll_dir(dll_path);
 489 
 490         if (pslash != NULL) {
 491             pslash = strrchr(buf, '/');
 492             if (pslash != NULL) {
 493                 *pslash = '\0';       /* get rid of /<arch> (/lib on macosx) */
 494 #ifndef __APPLE__
 495                 pslash = strrchr(buf, '/');
 496                 if (pslash != NULL)
 497                     *pslash = '\0';   /* get rid of /lib */
 498 #endif
 499             }
 500         }
 501 
 502         home_path = malloc(strlen(buf) + 1);
 503         if (home_path == NULL)
 504             return;
 505         strcpy(home_path, buf);
 506         Arguments::set_java_home(home_path);
 507 
 508         if (!set_boot_path('/', ':'))
 509             return;
 510     }
 511 
 512     /*
 513      * Where to look for native libraries
 514      *
 515      * Note: Due to a legacy implementation, most of the library path
 516      * is set in the launcher.  This was to accomodate linking restrictions
 517      * on legacy Bsd implementations (which are no longer supported).
 518      * Eventually, all the library path setting will be done here.
 519      *
 520      * However, to prevent the proliferation of improperly built native
 521      * libraries, the new path component /usr/java/packages is added here.
 522      * Eventually, all the library path setting will be done here.
 523      */
 524     {
 525         char *ld_library_path;
 526 
 527         /*
 528          * Construct the invariant part of ld_library_path. Note that the
 529          * space for the colon and the trailing null are provided by the
 530          * nulls included by the sizeof operator (so actually we allocate
 531          * a byte more than necessary).
 532          */
 533 #ifdef __APPLE__
 534         ld_library_path = (char *) malloc(system_ext_size);
 535         sprintf(ld_library_path, "%s" SYS_EXTENSIONS_DIR ":" SYS_EXTENSIONS_DIRS, user_home_dir);
 536 #else
 537         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 538             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 539         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 540 #endif
 541 
 542         /*
 543          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 544          * should always exist (until the legacy problem cited above is
 545          * addressed).
 546          */
 547 #ifdef __APPLE__
 548         // Prepend the default path with the JAVA_LIBRARY_PATH so that the app launcher code can specify a directory inside an app wrapper
 549         char *l = getenv("JAVA_LIBRARY_PATH");
 550         if (l != NULL) {
 551             char *t = ld_library_path;
 552             /* That's +1 for the colon and +1 for the trailing '\0' */
 553             ld_library_path = (char *) malloc(strlen(l) + 1 + strlen(t) + 1);
 554             sprintf(ld_library_path, "%s:%s", l, t);
 555             free(t);
 556         }
 557 
 558         char *v = getenv("DYLD_LIBRARY_PATH");
 559 #else
 560         char *v = getenv("LD_LIBRARY_PATH");
 561 #endif
 562         if (v != NULL) {
 563             char *t = ld_library_path;
 564             /* That's +1 for the colon and +1 for the trailing '\0' */
 565             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 566             sprintf(ld_library_path, "%s:%s", v, t);
 567             free(t);
 568         }
 569 
 570 #ifdef __APPLE__
 571         // Apple's Java6 has "." at the beginning of java.library.path.
 572         // OpenJDK on Windows has "." at the end of java.library.path.
 573         // OpenJDK on Linux and Solaris don't have "." in java.library.path
 574         // at all. To ease the transition from Apple's Java6 to OpenJDK7,
 575         // "." is appended to the end of java.library.path. Yes, this
 576         // could cause a change in behavior, but Apple's Java6 behavior
 577         // can be achieved by putting "." at the beginning of the
 578         // JAVA_LIBRARY_PATH environment variable.
 579         {
 580             char *t = ld_library_path;
 581             // that's +3 for appending ":." and the trailing '\0'
 582             ld_library_path = (char *) malloc(strlen(t) + 3);
 583             sprintf(ld_library_path, "%s:%s", t, ".");
 584             free(t);
 585         }
 586 #endif
 587 
 588         Arguments::set_library_path(ld_library_path);
 589     }
 590 
 591     /*
 592      * Extensions directories.
 593      *
 594      * Note that the space for the colon and the trailing null are provided
 595      * by the nulls included by the sizeof operator (so actually one byte more
 596      * than necessary is allocated).
 597      */
 598     {
 599 #ifdef __APPLE__
 600         char *buf = malloc(strlen(Arguments::get_java_home()) +
 601             sizeof(EXTENSIONS_DIR) + system_ext_size);
 602         sprintf(buf, "%s" SYS_EXTENSIONS_DIR ":%s" EXTENSIONS_DIR ":"
 603             SYS_EXTENSIONS_DIRS, user_home_dir, Arguments::get_java_home());
 604 #else
 605         char *buf = malloc(strlen(Arguments::get_java_home()) +
 606             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 607         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 608             Arguments::get_java_home());
 609 #endif
 610 
 611         Arguments::set_ext_dirs(buf);
 612     }
 613 
 614     /* Endorsed standards default directory. */
 615     {
 616         char * buf;
 617         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 618         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 619         Arguments::set_endorsed_dirs(buf);
 620     }
 621   }
 622 
 623 #ifdef __APPLE__
 624 #undef SYS_EXTENSIONS_DIR
 625 #endif
 626 #undef malloc
 627 #undef getenv
 628 #undef EXTENSIONS_DIR
 629 #undef ENDORSED_DIR
 630 
 631   // Done
 632   return;
 633 }
 634 
 635 ////////////////////////////////////////////////////////////////////////////////
 636 // breakpoint support
 637 
 638 void os::breakpoint() {
 639   BREAKPOINT;
 640 }
 641 
 642 extern "C" void breakpoint() {
 643   // use debugger to set breakpoint here
 644 }
 645 
 646 ////////////////////////////////////////////////////////////////////////////////
 647 // signal support
 648 
 649 debug_only(static bool signal_sets_initialized = false);
 650 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 651 
 652 bool os::Bsd::is_sig_ignored(int sig) {
 653       struct sigaction oact;
 654       sigaction(sig, (struct sigaction*)NULL, &oact);
 655       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 656                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 657       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 658            return true;
 659       else
 660            return false;
 661 }
 662 
 663 void os::Bsd::signal_sets_init() {
 664   // Should also have an assertion stating we are still single-threaded.
 665   assert(!signal_sets_initialized, "Already initialized");
 666   // Fill in signals that are necessarily unblocked for all threads in
 667   // the VM. Currently, we unblock the following signals:
 668   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 669   //                         by -Xrs (=ReduceSignalUsage));
 670   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 671   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 672   // the dispositions or masks wrt these signals.
 673   // Programs embedding the VM that want to use the above signals for their
 674   // own purposes must, at this time, use the "-Xrs" option to prevent
 675   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 676   // (See bug 4345157, and other related bugs).
 677   // In reality, though, unblocking these signals is really a nop, since
 678   // these signals are not blocked by default.
 679   sigemptyset(&unblocked_sigs);
 680   sigemptyset(&allowdebug_blocked_sigs);
 681   sigaddset(&unblocked_sigs, SIGILL);
 682   sigaddset(&unblocked_sigs, SIGSEGV);
 683   sigaddset(&unblocked_sigs, SIGBUS);
 684   sigaddset(&unblocked_sigs, SIGFPE);
 685   sigaddset(&unblocked_sigs, SR_signum);
 686 
 687   if (!ReduceSignalUsage) {
 688    if (!os::Bsd::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 689       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 690       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 691    }
 692    if (!os::Bsd::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 693       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 694       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 695    }
 696    if (!os::Bsd::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 697       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 698       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 699    }
 700   }
 701   // Fill in signals that are blocked by all but the VM thread.
 702   sigemptyset(&vm_sigs);
 703   if (!ReduceSignalUsage)
 704     sigaddset(&vm_sigs, BREAK_SIGNAL);
 705   debug_only(signal_sets_initialized = true);
 706 
 707 }
 708 
 709 // These are signals that are unblocked while a thread is running Java.
 710 // (For some reason, they get blocked by default.)
 711 sigset_t* os::Bsd::unblocked_signals() {
 712   assert(signal_sets_initialized, "Not initialized");
 713   return &unblocked_sigs;
 714 }
 715 
 716 // These are the signals that are blocked while a (non-VM) thread is
 717 // running Java. Only the VM thread handles these signals.
 718 sigset_t* os::Bsd::vm_signals() {
 719   assert(signal_sets_initialized, "Not initialized");
 720   return &vm_sigs;
 721 }
 722 
 723 // These are signals that are blocked during cond_wait to allow debugger in
 724 sigset_t* os::Bsd::allowdebug_blocked_signals() {
 725   assert(signal_sets_initialized, "Not initialized");
 726   return &allowdebug_blocked_sigs;
 727 }
 728 
 729 void os::Bsd::hotspot_sigmask(Thread* thread) {
 730 
 731   //Save caller's signal mask before setting VM signal mask
 732   sigset_t caller_sigmask;
 733   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 734 
 735   OSThread* osthread = thread->osthread();
 736   osthread->set_caller_sigmask(caller_sigmask);
 737 
 738   pthread_sigmask(SIG_UNBLOCK, os::Bsd::unblocked_signals(), NULL);
 739 
 740   if (!ReduceSignalUsage) {
 741     if (thread->is_VM_thread()) {
 742       // Only the VM thread handles BREAK_SIGNAL ...
 743       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 744     } else {
 745       // ... all other threads block BREAK_SIGNAL
 746       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 747     }
 748   }
 749 }
 750 
 751 #ifndef _ALLBSD_SOURCE
 752 //////////////////////////////////////////////////////////////////////////////
 753 // detecting pthread library
 754 
 755 void os::Bsd::libpthread_init() {
 756   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 757   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 758   // generic name for earlier versions.
 759   // Define macros here so we can build HotSpot on old systems.
 760 # ifndef _CS_GNU_LIBC_VERSION
 761 # define _CS_GNU_LIBC_VERSION 2
 762 # endif
 763 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 764 # define _CS_GNU_LIBPTHREAD_VERSION 3
 765 # endif
 766 
 767   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 768   if (n > 0) {
 769      char *str = (char *)malloc(n);
 770      confstr(_CS_GNU_LIBC_VERSION, str, n);
 771      os::Bsd::set_glibc_version(str);
 772   } else {
 773      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 774      static char _gnu_libc_version[32];
 775      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 776               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 777      os::Bsd::set_glibc_version(_gnu_libc_version);
 778   }
 779 
 780   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 781   if (n > 0) {
 782      char *str = (char *)malloc(n);
 783      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 784      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 785      // us "NPTL-0.29" even we are running with BsdThreads. Check if this
 786      // is the case. BsdThreads has a hard limit on max number of threads.
 787      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 788      // On the other hand, NPTL does not have such a limit, sysconf()
 789      // will return -1 and errno is not changed. Check if it is really NPTL.
 790      if (strcmp(os::Bsd::glibc_version(), "glibc 2.3.2") == 0 &&
 791          strstr(str, "NPTL") &&
 792          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 793        free(str);
 794        os::Bsd::set_libpthread_version("bsdthreads");
 795      } else {
 796        os::Bsd::set_libpthread_version(str);
 797      }
 798   } else {
 799     // glibc before 2.3.2 only has BsdThreads.
 800     os::Bsd::set_libpthread_version("bsdthreads");
 801   }
 802 
 803   if (strstr(libpthread_version(), "NPTL")) {
 804      os::Bsd::set_is_NPTL();
 805   } else {
 806      os::Bsd::set_is_BsdThreads();
 807   }
 808 
 809   // BsdThreads have two flavors: floating-stack mode, which allows variable
 810   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 811   if (os::Bsd::is_NPTL() || os::Bsd::supports_variable_stack_size()) {
 812      os::Bsd::set_is_floating_stack();
 813   }
 814 }
 815 
 816 /////////////////////////////////////////////////////////////////////////////
 817 // thread stack
 818 
 819 // Force Bsd kernel to expand current thread stack. If "bottom" is close
 820 // to the stack guard, caller should block all signals.
 821 //
 822 // MAP_GROWSDOWN:
 823 //   A special mmap() flag that is used to implement thread stacks. It tells
 824 //   kernel that the memory region should extend downwards when needed. This
 825 //   allows early versions of BsdThreads to only mmap the first few pages
 826 //   when creating a new thread. Bsd kernel will automatically expand thread
 827 //   stack as needed (on page faults).
 828 //
 829 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 830 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 831 //   region, it's hard to tell if the fault is due to a legitimate stack
 832 //   access or because of reading/writing non-exist memory (e.g. buffer
 833 //   overrun). As a rule, if the fault happens below current stack pointer,
 834 //   Bsd kernel does not expand stack, instead a SIGSEGV is sent to the
 835 //   application (see Bsd kernel fault.c).
 836 //
 837 //   This Bsd feature can cause SIGSEGV when VM bangs thread stack for
 838 //   stack overflow detection.
 839 //
 840 //   Newer version of BsdThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 841 //   not use this flag. However, the stack of initial thread is not created
 842 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 843 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 844 //   and then attach the thread to JVM.
 845 //
 846 // To get around the problem and allow stack banging on Bsd, we need to
 847 // manually expand thread stack after receiving the SIGSEGV.
 848 //
 849 // There are two ways to expand thread stack to address "bottom", we used
 850 // both of them in JVM before 1.5:
 851 //   1. adjust stack pointer first so that it is below "bottom", and then
 852 //      touch "bottom"
 853 //   2. mmap() the page in question
 854 //
 855 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 856 // if current sp is already near the lower end of page 101, and we need to
 857 // call mmap() to map page 100, it is possible that part of the mmap() frame
 858 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 859 // That will destroy the mmap() frame and cause VM to crash.
 860 //
 861 // The following code works by adjusting sp first, then accessing the "bottom"
 862 // page to force a page fault. Bsd kernel will then automatically expand the
 863 // stack mapping.
 864 //
 865 // _expand_stack_to() assumes its frame size is less than page size, which
 866 // should always be true if the function is not inlined.
 867 
 868 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 869 #define NOINLINE
 870 #else
 871 #define NOINLINE __attribute__ ((noinline))
 872 #endif
 873 
 874 static void _expand_stack_to(address bottom) NOINLINE;
 875 
 876 static void _expand_stack_to(address bottom) {
 877   address sp;
 878   size_t size;
 879   volatile char *p;
 880 
 881   // Adjust bottom to point to the largest address within the same page, it
 882   // gives us a one-page buffer if alloca() allocates slightly more memory.
 883   bottom = (address)align_size_down((uintptr_t)bottom, os::Bsd::page_size());
 884   bottom += os::Bsd::page_size() - 1;
 885 
 886   // sp might be slightly above current stack pointer; if that's the case, we
 887   // will alloca() a little more space than necessary, which is OK. Don't use
 888   // os::current_stack_pointer(), as its result can be slightly below current
 889   // stack pointer, causing us to not alloca enough to reach "bottom".
 890   sp = (address)&sp;
 891 
 892   if (sp > bottom) {
 893     size = sp - bottom;
 894     p = (volatile char *)alloca(size);
 895     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 896     p[0] = '\0';
 897   }
 898 }
 899 
 900 bool os::Bsd::manually_expand_stack(JavaThread * t, address addr) {
 901   assert(t!=NULL, "just checking");
 902   assert(t->osthread()->expanding_stack(), "expand should be set");
 903   assert(t->stack_base() != NULL, "stack_base was not initialized");
 904 
 905   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 906     sigset_t mask_all, old_sigset;
 907     sigfillset(&mask_all);
 908     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 909     _expand_stack_to(addr);
 910     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 911     return true;
 912   }
 913   return false;
 914 }
 915 #endif
 916 
 917 //////////////////////////////////////////////////////////////////////////////
 918 // create new thread
 919 
 920 static address highest_vm_reserved_address();
 921 
 922 // check if it's safe to start a new thread
 923 static bool _thread_safety_check(Thread* thread) {
 924 #ifdef _ALLBSD_SOURCE
 925     return true;
 926 #else
 927   if (os::Bsd::is_BsdThreads() && !os::Bsd::is_floating_stack()) {
 928     // Fixed stack BsdThreads (SuSE Bsd/x86, and some versions of Redhat)
 929     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 930     //   allocated (MAP_FIXED) from high address space. Every thread stack
 931     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 932     //   it to other values if they rebuild BsdThreads).
 933     //
 934     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 935     // the memory region has already been mmap'ed. That means if we have too
 936     // many threads and/or very large heap, eventually thread stack will
 937     // collide with heap.
 938     //
 939     // Here we try to prevent heap/stack collision by comparing current
 940     // stack bottom with the highest address that has been mmap'ed by JVM
 941     // plus a safety margin for memory maps created by native code.
 942     //
 943     // This feature can be disabled by setting ThreadSafetyMargin to 0
 944     //
 945     if (ThreadSafetyMargin > 0) {
 946       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 947 
 948       // not safe if our stack extends below the safety margin
 949       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 950     } else {
 951       return true;
 952     }
 953   } else {
 954     // Floating stack BsdThreads or NPTL:
 955     //   Unlike fixed stack BsdThreads, thread stacks are not MAP_FIXED. When
 956     //   there's not enough space left, pthread_create() will fail. If we come
 957     //   here, that means enough space has been reserved for stack.
 958     return true;
 959   }
 960 #endif
 961 }
 962 
 963 #ifdef __APPLE__
 964 // library handle for calling objc_registerThreadWithCollector()
 965 // without static linking to the libobjc library
 966 #define OBJC_LIB "/usr/lib/libobjc.dylib"
 967 #define OBJC_GCREGISTER "objc_registerThreadWithCollector"
 968 typedef void (*objc_registerThreadWithCollector_t)();
 969 extern "C" objc_registerThreadWithCollector_t objc_registerThreadWithCollectorFunction;
 970 objc_registerThreadWithCollector_t objc_registerThreadWithCollectorFunction = NULL;
 971 #endif
 972 
 973 #ifdef __APPLE__
 974 static uint64_t locate_unique_thread_id() {
 975   // Additional thread_id used to correlate threads in SA
 976   thread_identifier_info_data_t     m_ident_info;
 977   mach_msg_type_number_t            count = THREAD_IDENTIFIER_INFO_COUNT;
 978 
 979   thread_info(::mach_thread_self(), THREAD_IDENTIFIER_INFO,
 980               (thread_info_t) &m_ident_info, &count);
 981   return m_ident_info.thread_id;
 982 }
 983 #endif
 984 
 985 // Thread start routine for all newly created threads
 986 static void *java_start(Thread *thread) {
 987   // Try to randomize the cache line index of hot stack frames.
 988   // This helps when threads of the same stack traces evict each other's
 989   // cache lines. The threads can be either from the same JVM instance, or
 990   // from different JVM instances. The benefit is especially true for
 991   // processors with hyperthreading technology.
 992   static int counter = 0;
 993   int pid = os::current_process_id();
 994   alloca(((pid ^ counter++) & 7) * 128);
 995 
 996   ThreadLocalStorage::set_thread(thread);
 997 
 998   OSThread* osthread = thread->osthread();
 999   Monitor* sync = osthread->startThread_lock();
1000 
1001   // non floating stack BsdThreads needs extra check, see above
1002   if (!_thread_safety_check(thread)) {
1003     // notify parent thread
1004     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
1005     osthread->set_state(ZOMBIE);
1006     sync->notify_all();
1007     return NULL;
1008   }
1009 
1010 #ifdef _ALLBSD_SOURCE
1011 #ifdef __APPLE__
1012   // thread_id is mach thread on macos
1013   osthread->set_thread_id(::mach_thread_self());
1014   osthread->set_unique_thread_id(locate_unique_thread_id());
1015 #else
1016   // thread_id is pthread_id on BSD
1017   osthread->set_thread_id(::pthread_self());
1018 #endif
1019 #else
1020   // thread_id is kernel thread id (similar to Solaris LWP id)
1021   osthread->set_thread_id(os::Bsd::gettid());
1022 
1023   if (UseNUMA) {
1024     int lgrp_id = os::numa_get_group_id();
1025     if (lgrp_id != -1) {
1026       thread->set_lgrp_id(lgrp_id);
1027     }
1028   }
1029 #endif
1030   // initialize signal mask for this thread
1031   os::Bsd::hotspot_sigmask(thread);
1032 
1033   // initialize floating point control register
1034   os::Bsd::init_thread_fpu_state();
1035 
1036 #ifdef __APPLE__
1037   // register thread with objc gc
1038   if (objc_registerThreadWithCollectorFunction != NULL) {
1039     objc_registerThreadWithCollectorFunction();
1040   }
1041 #endif
1042 
1043   // handshaking with parent thread
1044   {
1045     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
1046 
1047     // notify parent thread
1048     osthread->set_state(INITIALIZED);
1049     sync->notify_all();
1050 
1051     // wait until os::start_thread()
1052     while (osthread->get_state() == INITIALIZED) {
1053       sync->wait(Mutex::_no_safepoint_check_flag);
1054     }
1055   }
1056 
1057   // call one more level start routine
1058   thread->run();
1059 
1060   return 0;
1061 }
1062 
1063 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
1064   assert(thread->osthread() == NULL, "caller responsible");
1065 
1066   // Allocate the OSThread object
1067   OSThread* osthread = new OSThread(NULL, NULL);
1068   if (osthread == NULL) {
1069     return false;
1070   }
1071 
1072   // set the correct thread state
1073   osthread->set_thread_type(thr_type);
1074 
1075   // Initial state is ALLOCATED but not INITIALIZED
1076   osthread->set_state(ALLOCATED);
1077 
1078   thread->set_osthread(osthread);
1079 
1080   // init thread attributes
1081   pthread_attr_t attr;
1082   pthread_attr_init(&attr);
1083   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1084 
1085   // stack size
1086   if (os::Bsd::supports_variable_stack_size()) {
1087     // calculate stack size if it's not specified by caller
1088     if (stack_size == 0) {
1089       stack_size = os::Bsd::default_stack_size(thr_type);
1090 
1091       switch (thr_type) {
1092       case os::java_thread:
1093         // Java threads use ThreadStackSize which default value can be
1094         // changed with the flag -Xss
1095         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
1096         stack_size = JavaThread::stack_size_at_create();
1097         break;
1098       case os::compiler_thread:
1099         if (CompilerThreadStackSize > 0) {
1100           stack_size = (size_t)(CompilerThreadStackSize * K);
1101           break;
1102         } // else fall through:
1103           // use VMThreadStackSize if CompilerThreadStackSize is not defined
1104       case os::vm_thread:
1105       case os::pgc_thread:
1106       case os::cgc_thread:
1107       case os::watcher_thread:
1108         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1109         break;
1110       }
1111     }
1112 
1113     stack_size = MAX2(stack_size, os::Bsd::min_stack_allowed);
1114     pthread_attr_setstacksize(&attr, stack_size);
1115   } else {
1116     // let pthread_create() pick the default value.
1117   }
1118 
1119 #ifndef _ALLBSD_SOURCE
1120   // glibc guard page
1121   pthread_attr_setguardsize(&attr, os::Bsd::default_guard_size(thr_type));
1122 #endif
1123 
1124   ThreadState state;
1125 
1126   {
1127 
1128 #ifndef _ALLBSD_SOURCE
1129     // Serialize thread creation if we are running with fixed stack BsdThreads
1130     bool lock = os::Bsd::is_BsdThreads() && !os::Bsd::is_floating_stack();
1131     if (lock) {
1132       os::Bsd::createThread_lock()->lock_without_safepoint_check();
1133     }
1134 #endif
1135 
1136     pthread_t tid;
1137     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
1138 
1139     pthread_attr_destroy(&attr);
1140 
1141     if (ret != 0) {
1142       if (PrintMiscellaneous && (Verbose || WizardMode)) {
1143         perror("pthread_create()");
1144       }
1145       // Need to clean up stuff we've allocated so far
1146       thread->set_osthread(NULL);
1147       delete osthread;
1148 #ifndef _ALLBSD_SOURCE
1149       if (lock) os::Bsd::createThread_lock()->unlock();
1150 #endif
1151       return false;
1152     }
1153 
1154     // Store pthread info into the OSThread
1155     osthread->set_pthread_id(tid);
1156 
1157     // Wait until child thread is either initialized or aborted
1158     {
1159       Monitor* sync_with_child = osthread->startThread_lock();
1160       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1161       while ((state = osthread->get_state()) == ALLOCATED) {
1162         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
1163       }
1164     }
1165 
1166 #ifndef _ALLBSD_SOURCE
1167     if (lock) {
1168       os::Bsd::createThread_lock()->unlock();
1169     }
1170 #endif
1171   }
1172 
1173   // Aborted due to thread limit being reached
1174   if (state == ZOMBIE) {
1175       thread->set_osthread(NULL);
1176       delete osthread;
1177       return false;
1178   }
1179 
1180   // The thread is returned suspended (in state INITIALIZED),
1181   // and is started higher up in the call chain
1182   assert(state == INITIALIZED, "race condition");
1183   return true;
1184 }
1185 
1186 /////////////////////////////////////////////////////////////////////////////
1187 // attach existing thread
1188 
1189 // bootstrap the main thread
1190 bool os::create_main_thread(JavaThread* thread) {
1191   assert(os::Bsd::_main_thread == pthread_self(), "should be called inside main thread");
1192   return create_attached_thread(thread);
1193 }
1194 
1195 bool os::create_attached_thread(JavaThread* thread) {
1196 #ifdef ASSERT
1197     thread->verify_not_published();
1198 #endif
1199 
1200   // Allocate the OSThread object
1201   OSThread* osthread = new OSThread(NULL, NULL);
1202 
1203   if (osthread == NULL) {
1204     return false;
1205   }
1206 
1207   // Store pthread info into the OSThread
1208 #ifdef _ALLBSD_SOURCE
1209 #ifdef __APPLE__
1210   osthread->set_thread_id(::mach_thread_self());
1211   osthread->set_unique_thread_id(locate_unique_thread_id());
1212 #else
1213   osthread->set_thread_id(::pthread_self());
1214 #endif
1215 #else
1216   osthread->set_thread_id(os::Bsd::gettid());
1217 #endif
1218   osthread->set_pthread_id(::pthread_self());
1219 
1220   // initialize floating point control register
1221   os::Bsd::init_thread_fpu_state();
1222 
1223   // Initial thread state is RUNNABLE
1224   osthread->set_state(RUNNABLE);
1225 
1226   thread->set_osthread(osthread);
1227 
1228 #ifndef _ALLBSD_SOURCE
1229   if (UseNUMA) {
1230     int lgrp_id = os::numa_get_group_id();
1231     if (lgrp_id != -1) {
1232       thread->set_lgrp_id(lgrp_id);
1233     }
1234   }
1235 
1236   if (os::Bsd::is_initial_thread()) {
1237     // If current thread is initial thread, its stack is mapped on demand,
1238     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1239     // the entire stack region to avoid SEGV in stack banging.
1240     // It is also useful to get around the heap-stack-gap problem on SuSE
1241     // kernel (see 4821821 for details). We first expand stack to the top
1242     // of yellow zone, then enable stack yellow zone (order is significant,
1243     // enabling yellow zone first will crash JVM on SuSE Bsd), so there
1244     // is no gap between the last two virtual memory regions.
1245 
1246     JavaThread *jt = (JavaThread *)thread;
1247     address addr = jt->stack_yellow_zone_base();
1248     assert(addr != NULL, "initialization problem?");
1249     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1250 
1251     osthread->set_expanding_stack();
1252     os::Bsd::manually_expand_stack(jt, addr);
1253     osthread->clear_expanding_stack();
1254   }
1255 #endif
1256 
1257   // initialize signal mask for this thread
1258   // and save the caller's signal mask
1259   os::Bsd::hotspot_sigmask(thread);
1260 
1261   return true;
1262 }
1263 
1264 void os::pd_start_thread(Thread* thread) {
1265   OSThread * osthread = thread->osthread();
1266   assert(osthread->get_state() != INITIALIZED, "just checking");
1267   Monitor* sync_with_child = osthread->startThread_lock();
1268   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1269   sync_with_child->notify();
1270 }
1271 
1272 // Free Bsd resources related to the OSThread
1273 void os::free_thread(OSThread* osthread) {
1274   assert(osthread != NULL, "osthread not set");
1275 
1276   if (Thread::current()->osthread() == osthread) {
1277     // Restore caller's signal mask
1278     sigset_t sigmask = osthread->caller_sigmask();
1279     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1280    }
1281 
1282   delete osthread;
1283 }
1284 
1285 //////////////////////////////////////////////////////////////////////////////
1286 // thread local storage
1287 
1288 int os::allocate_thread_local_storage() {
1289   pthread_key_t key;
1290   int rslt = pthread_key_create(&key, NULL);
1291   assert(rslt == 0, "cannot allocate thread local storage");
1292   return (int)key;
1293 }
1294 
1295 // Note: This is currently not used by VM, as we don't destroy TLS key
1296 // on VM exit.
1297 void os::free_thread_local_storage(int index) {
1298   int rslt = pthread_key_delete((pthread_key_t)index);
1299   assert(rslt == 0, "invalid index");
1300 }
1301 
1302 void os::thread_local_storage_at_put(int index, void* value) {
1303   int rslt = pthread_setspecific((pthread_key_t)index, value);
1304   assert(rslt == 0, "pthread_setspecific failed");
1305 }
1306 
1307 extern "C" Thread* get_thread() {
1308   return ThreadLocalStorage::thread();
1309 }
1310 
1311 //////////////////////////////////////////////////////////////////////////////
1312 // initial thread
1313 
1314 #ifndef _ALLBSD_SOURCE
1315 // Check if current thread is the initial thread, similar to Solaris thr_main.
1316 bool os::Bsd::is_initial_thread(void) {
1317   char dummy;
1318   // If called before init complete, thread stack bottom will be null.
1319   // Can be called if fatal error occurs before initialization.
1320   if (initial_thread_stack_bottom() == NULL) return false;
1321   assert(initial_thread_stack_bottom() != NULL &&
1322          initial_thread_stack_size()   != 0,
1323          "os::init did not locate initial thread's stack region");
1324   if ((address)&dummy >= initial_thread_stack_bottom() &&
1325       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1326        return true;
1327   else return false;
1328 }
1329 
1330 // Find the virtual memory area that contains addr
1331 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1332   FILE *fp = fopen("/proc/self/maps", "r");
1333   if (fp) {
1334     address low, high;
1335     while (!feof(fp)) {
1336       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1337         if (low <= addr && addr < high) {
1338            if (vma_low)  *vma_low  = low;
1339            if (vma_high) *vma_high = high;
1340            fclose (fp);
1341            return true;
1342         }
1343       }
1344       for (;;) {
1345         int ch = fgetc(fp);
1346         if (ch == EOF || ch == (int)'\n') break;
1347       }
1348     }
1349     fclose(fp);
1350   }
1351   return false;
1352 }
1353 
1354 // Locate initial thread stack. This special handling of initial thread stack
1355 // is needed because pthread_getattr_np() on most (all?) Bsd distros returns
1356 // bogus value for initial thread.
1357 void os::Bsd::capture_initial_stack(size_t max_size) {
1358   // stack size is the easy part, get it from RLIMIT_STACK
1359   size_t stack_size;
1360   struct rlimit rlim;
1361   getrlimit(RLIMIT_STACK, &rlim);
1362   stack_size = rlim.rlim_cur;
1363 
1364   // 6308388: a bug in ld.so will relocate its own .data section to the
1365   //   lower end of primordial stack; reduce ulimit -s value a little bit
1366   //   so we won't install guard page on ld.so's data section.
1367   stack_size -= 2 * page_size();
1368 
1369   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1370   //   7.1, in both cases we will get 2G in return value.
1371   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1372   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1373   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1374   //   in case other parts in glibc still assumes 2M max stack size.
1375   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1376 #ifndef IA64
1377   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1378 #else
1379   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1380   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1381 #endif
1382 
1383   // Try to figure out where the stack base (top) is. This is harder.
1384   //
1385   // When an application is started, glibc saves the initial stack pointer in
1386   // a global variable "__libc_stack_end", which is then used by system
1387   // libraries. __libc_stack_end should be pretty close to stack top. The
1388   // variable is available since the very early days. However, because it is
1389   // a private interface, it could disappear in the future.
1390   //
1391   // Bsd kernel saves start_stack information in /proc/<pid>/stat. Similar
1392   // to __libc_stack_end, it is very close to stack top, but isn't the real
1393   // stack top. Note that /proc may not exist if VM is running as a chroot
1394   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1395   // /proc/<pid>/stat could change in the future (though unlikely).
1396   //
1397   // We try __libc_stack_end first. If that doesn't work, look for
1398   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1399   // as a hint, which should work well in most cases.
1400 
1401   uintptr_t stack_start;
1402 
1403   // try __libc_stack_end first
1404   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1405   if (p && *p) {
1406     stack_start = *p;
1407   } else {
1408     // see if we can get the start_stack field from /proc/self/stat
1409     FILE *fp;
1410     int pid;
1411     char state;
1412     int ppid;
1413     int pgrp;
1414     int session;
1415     int nr;
1416     int tpgrp;
1417     unsigned long flags;
1418     unsigned long minflt;
1419     unsigned long cminflt;
1420     unsigned long majflt;
1421     unsigned long cmajflt;
1422     unsigned long utime;
1423     unsigned long stime;
1424     long cutime;
1425     long cstime;
1426     long prio;
1427     long nice;
1428     long junk;
1429     long it_real;
1430     uintptr_t start;
1431     uintptr_t vsize;
1432     intptr_t rss;
1433     uintptr_t rsslim;
1434     uintptr_t scodes;
1435     uintptr_t ecode;
1436     int i;
1437 
1438     // Figure what the primordial thread stack base is. Code is inspired
1439     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1440     // followed by command name surrounded by parentheses, state, etc.
1441     char stat[2048];
1442     int statlen;
1443 
1444     fp = fopen("/proc/self/stat", "r");
1445     if (fp) {
1446       statlen = fread(stat, 1, 2047, fp);
1447       stat[statlen] = '\0';
1448       fclose(fp);
1449 
1450       // Skip pid and the command string. Note that we could be dealing with
1451       // weird command names, e.g. user could decide to rename java launcher
1452       // to "java 1.4.2 :)", then the stat file would look like
1453       //                1234 (java 1.4.2 :)) R ... ...
1454       // We don't really need to know the command string, just find the last
1455       // occurrence of ")" and then start parsing from there. See bug 4726580.
1456       char * s = strrchr(stat, ')');
1457 
1458       i = 0;
1459       if (s) {
1460         // Skip blank chars
1461         do s++; while (isspace(*s));
1462 
1463 #define _UFM UINTX_FORMAT
1464 #define _DFM INTX_FORMAT
1465 
1466         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1467         /*              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 */
1468         i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1469              &state,          /* 3  %c  */
1470              &ppid,           /* 4  %d  */
1471              &pgrp,           /* 5  %d  */
1472              &session,        /* 6  %d  */
1473              &nr,             /* 7  %d  */
1474              &tpgrp,          /* 8  %d  */
1475              &flags,          /* 9  %lu  */
1476              &minflt,         /* 10 %lu  */
1477              &cminflt,        /* 11 %lu  */
1478              &majflt,         /* 12 %lu  */
1479              &cmajflt,        /* 13 %lu  */
1480              &utime,          /* 14 %lu  */
1481              &stime,          /* 15 %lu  */
1482              &cutime,         /* 16 %ld  */
1483              &cstime,         /* 17 %ld  */
1484              &prio,           /* 18 %ld  */
1485              &nice,           /* 19 %ld  */
1486              &junk,           /* 20 %ld  */
1487              &it_real,        /* 21 %ld  */
1488              &start,          /* 22 UINTX_FORMAT */
1489              &vsize,          /* 23 UINTX_FORMAT */
1490              &rss,            /* 24 INTX_FORMAT  */
1491              &rsslim,         /* 25 UINTX_FORMAT */
1492              &scodes,         /* 26 UINTX_FORMAT */
1493              &ecode,          /* 27 UINTX_FORMAT */
1494              &stack_start);   /* 28 UINTX_FORMAT */
1495       }
1496 
1497 #undef _UFM
1498 #undef _DFM
1499 
1500       if (i != 28 - 2) {
1501          assert(false, "Bad conversion from /proc/self/stat");
1502          // product mode - assume we are the initial thread, good luck in the
1503          // embedded case.
1504          warning("Can't detect initial thread stack location - bad conversion");
1505          stack_start = (uintptr_t) &rlim;
1506       }
1507     } else {
1508       // For some reason we can't open /proc/self/stat (for example, running on
1509       // FreeBSD with a Bsd emulator, or inside chroot), this should work for
1510       // most cases, so don't abort:
1511       warning("Can't detect initial thread stack location - no /proc/self/stat");
1512       stack_start = (uintptr_t) &rlim;
1513     }
1514   }
1515 
1516   // Now we have a pointer (stack_start) very close to the stack top, the
1517   // next thing to do is to figure out the exact location of stack top. We
1518   // can find out the virtual memory area that contains stack_start by
1519   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1520   // and its upper limit is the real stack top. (again, this would fail if
1521   // running inside chroot, because /proc may not exist.)
1522 
1523   uintptr_t stack_top;
1524   address low, high;
1525   if (find_vma((address)stack_start, &low, &high)) {
1526     // success, "high" is the true stack top. (ignore "low", because initial
1527     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1528     stack_top = (uintptr_t)high;
1529   } else {
1530     // failed, likely because /proc/self/maps does not exist
1531     warning("Can't detect initial thread stack location - find_vma failed");
1532     // best effort: stack_start is normally within a few pages below the real
1533     // stack top, use it as stack top, and reduce stack size so we won't put
1534     // guard page outside stack.
1535     stack_top = stack_start;
1536     stack_size -= 16 * page_size();
1537   }
1538 
1539   // stack_top could be partially down the page so align it
1540   stack_top = align_size_up(stack_top, page_size());
1541 
1542   if (max_size && stack_size > max_size) {
1543      _initial_thread_stack_size = max_size;
1544   } else {
1545      _initial_thread_stack_size = stack_size;
1546   }
1547 
1548   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1549   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1550 }
1551 #endif
1552 
1553 ////////////////////////////////////////////////////////////////////////////////
1554 // time support
1555 
1556 // Time since start-up in seconds to a fine granularity.
1557 // Used by VMSelfDestructTimer and the MemProfiler.
1558 double os::elapsedTime() {
1559 
1560   return (double)(os::elapsed_counter()) * 0.000001;
1561 }
1562 
1563 jlong os::elapsed_counter() {
1564   timeval time;
1565   int status = gettimeofday(&time, NULL);
1566   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1567 }
1568 
1569 jlong os::elapsed_frequency() {
1570   return (1000 * 1000);
1571 }
1572 
1573 // XXX: For now, code this as if BSD does not support vtime.
1574 bool os::supports_vtime() { return false; }
1575 bool os::enable_vtime()   { return false; }
1576 bool os::vtime_enabled()  { return false; }
1577 double os::elapsedVTime() {
1578   // better than nothing, but not much
1579   return elapsedTime();
1580 }
1581 
1582 jlong os::javaTimeMillis() {
1583   timeval time;
1584   int status = gettimeofday(&time, NULL);
1585   assert(status != -1, "bsd error");
1586   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1587 }
1588 
1589 #ifndef CLOCK_MONOTONIC
1590 #define CLOCK_MONOTONIC (1)
1591 #endif
1592 
1593 #ifdef __APPLE__
1594 void os::Bsd::clock_init() {
1595         // XXXDARWIN: Investigate replacement monotonic clock
1596 }
1597 #elif defined(_ALLBSD_SOURCE)
1598 void os::Bsd::clock_init() {
1599   struct timespec res;
1600   struct timespec tp;
1601   if (::clock_getres(CLOCK_MONOTONIC, &res) == 0 &&
1602       ::clock_gettime(CLOCK_MONOTONIC, &tp)  == 0) {
1603     // yes, monotonic clock is supported
1604     _clock_gettime = ::clock_gettime;
1605   }
1606 }
1607 #else
1608 void os::Bsd::clock_init() {
1609   // we do dlopen's in this particular order due to bug in bsd
1610   // dynamical loader (see 6348968) leading to crash on exit
1611   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1612   if (handle == NULL) {
1613     handle = dlopen("librt.so", RTLD_LAZY);
1614   }
1615 
1616   if (handle) {
1617     int (*clock_getres_func)(clockid_t, struct timespec*) =
1618            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1619     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1620            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1621     if (clock_getres_func && clock_gettime_func) {
1622       // See if monotonic clock is supported by the kernel. Note that some
1623       // early implementations simply return kernel jiffies (updated every
1624       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1625       // for nano time (though the monotonic property is still nice to have).
1626       // It's fixed in newer kernels, however clock_getres() still returns
1627       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1628       // resolution for now. Hopefully as people move to new kernels, this
1629       // won't be a problem.
1630       struct timespec res;
1631       struct timespec tp;
1632       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1633           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1634         // yes, monotonic clock is supported
1635         _clock_gettime = clock_gettime_func;
1636       } else {
1637         // close librt if there is no monotonic clock
1638         dlclose(handle);
1639       }
1640     }
1641   }
1642 }
1643 #endif
1644 
1645 #ifndef _ALLBSD_SOURCE
1646 #ifndef SYS_clock_getres
1647 
1648 #if defined(IA32) || defined(AMD64)
1649 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1650 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1651 #else
1652 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1653 #define sys_clock_getres(x,y)  -1
1654 #endif
1655 
1656 #else
1657 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1658 #endif
1659 
1660 void os::Bsd::fast_thread_clock_init() {
1661   if (!UseBsdPosixThreadCPUClocks) {
1662     return;
1663   }
1664   clockid_t clockid;
1665   struct timespec tp;
1666   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1667       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1668 
1669   // Switch to using fast clocks for thread cpu time if
1670   // the sys_clock_getres() returns 0 error code.
1671   // Note, that some kernels may support the current thread
1672   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1673   // returned by the pthread_getcpuclockid().
1674   // If the fast Posix clocks are supported then the sys_clock_getres()
1675   // must return at least tp.tv_sec == 0 which means a resolution
1676   // better than 1 sec. This is extra check for reliability.
1677 
1678   if(pthread_getcpuclockid_func &&
1679      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1680      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1681 
1682     _supports_fast_thread_cpu_time = true;
1683     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1684   }
1685 }
1686 #endif
1687 
1688 jlong os::javaTimeNanos() {
1689   if (Bsd::supports_monotonic_clock()) {
1690     struct timespec tp;
1691     int status = Bsd::clock_gettime(CLOCK_MONOTONIC, &tp);
1692     assert(status == 0, "gettime error");
1693     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1694     return result;
1695   } else {
1696     timeval time;
1697     int status = gettimeofday(&time, NULL);
1698     assert(status != -1, "bsd error");
1699     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1700     return 1000 * usecs;
1701   }
1702 }
1703 
1704 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1705   if (Bsd::supports_monotonic_clock()) {
1706     info_ptr->max_value = ALL_64_BITS;
1707 
1708     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1709     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1710     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1711   } else {
1712     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1713     info_ptr->max_value = ALL_64_BITS;
1714 
1715     // gettimeofday is a real time clock so it skips
1716     info_ptr->may_skip_backward = true;
1717     info_ptr->may_skip_forward = true;
1718   }
1719 
1720   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1721 }
1722 
1723 // Return the real, user, and system times in seconds from an
1724 // arbitrary fixed point in the past.
1725 bool os::getTimesSecs(double* process_real_time,
1726                       double* process_user_time,
1727                       double* process_system_time) {
1728   struct tms ticks;
1729   clock_t real_ticks = times(&ticks);
1730 
1731   if (real_ticks == (clock_t) (-1)) {
1732     return false;
1733   } else {
1734     double ticks_per_second = (double) clock_tics_per_sec;
1735     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1736     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1737     *process_real_time = ((double) real_ticks) / ticks_per_second;
1738 
1739     return true;
1740   }
1741 }
1742 
1743 
1744 char * os::local_time_string(char *buf, size_t buflen) {
1745   struct tm t;
1746   time_t long_time;
1747   time(&long_time);
1748   localtime_r(&long_time, &t);
1749   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1750                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1751                t.tm_hour, t.tm_min, t.tm_sec);
1752   return buf;
1753 }
1754 
1755 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1756   return localtime_r(clock, res);
1757 }
1758 
1759 ////////////////////////////////////////////////////////////////////////////////
1760 // runtime exit support
1761 
1762 // Note: os::shutdown() might be called very early during initialization, or
1763 // called from signal handler. Before adding something to os::shutdown(), make
1764 // sure it is async-safe and can handle partially initialized VM.
1765 void os::shutdown() {
1766 
1767   // allow PerfMemory to attempt cleanup of any persistent resources
1768   perfMemory_exit();
1769 
1770   // needs to remove object in file system
1771   AttachListener::abort();
1772 
1773   // flush buffered output, finish log files
1774   ostream_abort();
1775 
1776   // Check for abort hook
1777   abort_hook_t abort_hook = Arguments::abort_hook();
1778   if (abort_hook != NULL) {
1779     abort_hook();
1780   }
1781 
1782 }
1783 
1784 // Note: os::abort() might be called very early during initialization, or
1785 // called from signal handler. Before adding something to os::abort(), make
1786 // sure it is async-safe and can handle partially initialized VM.
1787 void os::abort(bool dump_core) {
1788   os::shutdown();
1789   if (dump_core) {
1790 #ifndef PRODUCT
1791     fdStream out(defaultStream::output_fd());
1792     out.print_raw("Current thread is ");
1793     char buf[16];
1794     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1795     out.print_raw_cr(buf);
1796     out.print_raw_cr("Dumping core ...");
1797 #endif
1798     ::abort(); // dump core
1799   }
1800 
1801   ::exit(1);
1802 }
1803 
1804 // Die immediately, no exit hook, no abort hook, no cleanup.
1805 void os::die() {
1806   // _exit() on BsdThreads only kills current thread
1807   ::abort();
1808 }
1809 
1810 // unused on bsd for now.
1811 void os::set_error_file(const char *logfile) {}
1812 
1813 
1814 // This method is a copy of JDK's sysGetLastErrorString
1815 // from src/solaris/hpi/src/system_md.c
1816 
1817 size_t os::lasterror(char *buf, size_t len) {
1818 
1819   if (errno == 0)  return 0;
1820 
1821   const char *s = ::strerror(errno);
1822   size_t n = ::strlen(s);
1823   if (n >= len) {
1824     n = len - 1;
1825   }
1826   ::strncpy(buf, s, n);
1827   buf[n] = '\0';
1828   return n;
1829 }
1830 
1831 intx os::current_thread_id() {
1832 #ifdef __APPLE__
1833   return (intx)::mach_thread_self();
1834 #else
1835   return (intx)::pthread_self();
1836 #endif
1837 }
1838 int os::current_process_id() {
1839 
1840   // Under the old bsd thread library, bsd gives each thread
1841   // its own process id. Because of this each thread will return
1842   // a different pid if this method were to return the result
1843   // of getpid(2). Bsd provides no api that returns the pid
1844   // of the launcher thread for the vm. This implementation
1845   // returns a unique pid, the pid of the launcher thread
1846   // that starts the vm 'process'.
1847 
1848   // Under the NPTL, getpid() returns the same pid as the
1849   // launcher thread rather than a unique pid per thread.
1850   // Use gettid() if you want the old pre NPTL behaviour.
1851 
1852   // if you are looking for the result of a call to getpid() that
1853   // returns a unique pid for the calling thread, then look at the
1854   // OSThread::thread_id() method in osThread_bsd.hpp file
1855 
1856   return (int)(_initial_pid ? _initial_pid : getpid());
1857 }
1858 
1859 // DLL functions
1860 
1861 #define JNI_LIB_PREFIX "lib"
1862 #ifdef __APPLE__
1863 #define JNI_LIB_SUFFIX ".dylib"
1864 #else
1865 #define JNI_LIB_SUFFIX ".so"
1866 #endif
1867 
1868 const char* os::dll_file_extension() { return JNI_LIB_SUFFIX; }
1869 
1870 // This must be hard coded because it's the system's temporary
1871 // directory not the java application's temp directory, ala java.io.tmpdir.
1872 #ifdef __APPLE__
1873 // macosx has a secure per-user temporary directory
1874 char temp_path_storage[PATH_MAX];
1875 const char* os::get_temp_directory() {
1876   static char *temp_path = NULL;
1877   if (temp_path == NULL) {
1878     int pathSize = confstr(_CS_DARWIN_USER_TEMP_DIR, temp_path_storage, PATH_MAX);
1879     if (pathSize == 0 || pathSize > PATH_MAX) {
1880       strlcpy(temp_path_storage, "/tmp/", sizeof(temp_path_storage));
1881     }
1882     temp_path = temp_path_storage;
1883   }
1884   return temp_path;
1885 }
1886 #else /* __APPLE__ */
1887 const char* os::get_temp_directory() { return "/tmp"; }
1888 #endif /* __APPLE__ */
1889 
1890 static bool file_exists(const char* filename) {
1891   struct stat statbuf;
1892   if (filename == NULL || strlen(filename) == 0) {
1893     return false;
1894   }
1895   return os::stat(filename, &statbuf) == 0;
1896 }
1897 
1898 void os::dll_build_name(char* buffer, size_t buflen,
1899                         const char* pname, const char* fname) {
1900   // Copied from libhpi
1901   const size_t pnamelen = pname ? strlen(pname) : 0;
1902 
1903   // Quietly truncate on buffer overflow.  Should be an error.
1904   if (pnamelen + strlen(fname) + strlen(JNI_LIB_PREFIX) + strlen(JNI_LIB_SUFFIX) + 2 > buflen) {
1905       *buffer = '\0';
1906       return;
1907   }
1908 
1909   if (pnamelen == 0) {
1910     snprintf(buffer, buflen, JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX, fname);
1911   } else if (strchr(pname, *os::path_separator()) != NULL) {
1912     int n;
1913     char** pelements = split_path(pname, &n);
1914     for (int i = 0 ; i < n ; i++) {
1915       // Really shouldn't be NULL, but check can't hurt
1916       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1917         continue; // skip the empty path values
1918       }
1919       snprintf(buffer, buflen, "%s/" JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX,
1920           pelements[i], fname);
1921       if (file_exists(buffer)) {
1922         break;
1923       }
1924     }
1925     // release the storage
1926     for (int i = 0 ; i < n ; i++) {
1927       if (pelements[i] != NULL) {
1928         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1929       }
1930     }
1931     if (pelements != NULL) {
1932       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1933     }
1934   } else {
1935     snprintf(buffer, buflen, "%s/" JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX, pname, fname);
1936   }
1937 }
1938 
1939 const char* os::get_current_directory(char *buf, int buflen) {
1940   return getcwd(buf, buflen);
1941 }
1942 
1943 // check if addr is inside libjvm[_g].so
1944 bool os::address_is_in_vm(address addr) {
1945   static address libjvm_base_addr;
1946   Dl_info dlinfo;
1947 
1948   if (libjvm_base_addr == NULL) {
1949     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1950     libjvm_base_addr = (address)dlinfo.dli_fbase;
1951     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1952   }
1953 
1954   if (dladdr((void *)addr, &dlinfo)) {
1955     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1956   }
1957 
1958   return false;
1959 }
1960 
1961 
1962 #define MACH_MAXSYMLEN 256
1963 
1964 bool os::dll_address_to_function_name(address addr, char *buf,
1965                                       int buflen, int *offset) {
1966   Dl_info dlinfo;
1967   char localbuf[MACH_MAXSYMLEN];
1968 
1969   // dladdr will find names of dynamic functions only, but does
1970   // it set dli_fbase with mach_header address when it "fails" ?
1971   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1972     if (buf != NULL) {
1973       if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1974         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1975       }
1976     }
1977     if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1978     return true;
1979   } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1980     if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1981        buf, buflen, offset, dlinfo.dli_fname)) {
1982        return true;
1983     }
1984   }
1985 
1986   // Handle non-dymanic manually:
1987   if (dlinfo.dli_fbase != NULL &&
1988       Decoder::decode(addr, localbuf, MACH_MAXSYMLEN, offset, dlinfo.dli_fbase)) {
1989     if(!Decoder::demangle(localbuf, buf, buflen)) {
1990       jio_snprintf(buf, buflen, "%s", localbuf);
1991     }
1992     return true;
1993   }
1994   if (buf != NULL) buf[0] = '\0';
1995   if (offset != NULL) *offset = -1;
1996   return false;
1997 }
1998 
1999 #ifdef _ALLBSD_SOURCE
2000 // ported from solaris version
2001 bool os::dll_address_to_library_name(address addr, char* buf,
2002                                      int buflen, int* offset) {
2003   Dl_info dlinfo;
2004 
2005   if (dladdr((void*)addr, &dlinfo)){
2006      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
2007      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
2008      return true;
2009   } else {
2010      if (buf) buf[0] = '\0';
2011      if (offset) *offset = -1;
2012      return false;
2013   }
2014 }
2015 #else
2016 struct _address_to_library_name {
2017   address addr;          // input : memory address
2018   size_t  buflen;        //         size of fname
2019   char*   fname;         // output: library name
2020   address base;          //         library base addr
2021 };
2022 
2023 static int address_to_library_name_callback(struct dl_phdr_info *info,
2024                                             size_t size, void *data) {
2025   int i;
2026   bool found = false;
2027   address libbase = NULL;
2028   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
2029 
2030   // iterate through all loadable segments
2031   for (i = 0; i < info->dlpi_phnum; i++) {
2032     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
2033     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
2034       // base address of a library is the lowest address of its loaded
2035       // segments.
2036       if (libbase == NULL || libbase > segbase) {
2037         libbase = segbase;
2038       }
2039       // see if 'addr' is within current segment
2040       if (segbase <= d->addr &&
2041           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
2042         found = true;
2043       }
2044     }
2045   }
2046 
2047   // dlpi_name is NULL or empty if the ELF file is executable, return 0
2048   // so dll_address_to_library_name() can fall through to use dladdr() which
2049   // can figure out executable name from argv[0].
2050   if (found && info->dlpi_name && info->dlpi_name[0]) {
2051     d->base = libbase;
2052     if (d->fname) {
2053       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
2054     }
2055     return 1;
2056   }
2057   return 0;
2058 }
2059 
2060 bool os::dll_address_to_library_name(address addr, char* buf,
2061                                      int buflen, int* offset) {
2062   Dl_info dlinfo;
2063   struct _address_to_library_name data;
2064 
2065   // There is a bug in old glibc dladdr() implementation that it could resolve
2066   // to wrong library name if the .so file has a base address != NULL. Here
2067   // we iterate through the program headers of all loaded libraries to find
2068   // out which library 'addr' really belongs to. This workaround can be
2069   // removed once the minimum requirement for glibc is moved to 2.3.x.
2070   data.addr = addr;
2071   data.fname = buf;
2072   data.buflen = buflen;
2073   data.base = NULL;
2074   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
2075 
2076   if (rslt) {
2077      // buf already contains library name
2078      if (offset) *offset = addr - data.base;
2079      return true;
2080   } else if (dladdr((void*)addr, &dlinfo)){
2081      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
2082      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
2083      return true;
2084   } else {
2085      if (buf) buf[0] = '\0';
2086      if (offset) *offset = -1;
2087      return false;
2088   }
2089 }
2090 #endif
2091 
2092   // Loads .dll/.so and
2093   // in case of error it checks if .dll/.so was built for the
2094   // same architecture as Hotspot is running on
2095 
2096 #ifdef __APPLE__
2097 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
2098   void * result= ::dlopen(filename, RTLD_LAZY);
2099   if (result != NULL) {
2100     // Successful loading
2101     return result;
2102   }
2103 
2104   // Read system error message into ebuf
2105   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2106   ebuf[ebuflen-1]='\0';
2107 
2108   return NULL;
2109 }
2110 #else
2111 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
2112 {
2113   void * result= ::dlopen(filename, RTLD_LAZY);
2114   if (result != NULL) {
2115     // Successful loading
2116     return result;
2117   }
2118 
2119   Elf32_Ehdr elf_head;
2120 
2121   // Read system error message into ebuf
2122   // It may or may not be overwritten below
2123   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2124   ebuf[ebuflen-1]='\0';
2125   int diag_msg_max_length=ebuflen-strlen(ebuf);
2126   char* diag_msg_buf=ebuf+strlen(ebuf);
2127 
2128   if (diag_msg_max_length==0) {
2129     // No more space in ebuf for additional diagnostics message
2130     return NULL;
2131   }
2132 
2133 
2134   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
2135 
2136   if (file_descriptor < 0) {
2137     // Can't open library, report dlerror() message
2138     return NULL;
2139   }
2140 
2141   bool failed_to_read_elf_head=
2142     (sizeof(elf_head)!=
2143         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
2144 
2145   ::close(file_descriptor);
2146   if (failed_to_read_elf_head) {
2147     // file i/o error - report dlerror() msg
2148     return NULL;
2149   }
2150 
2151   typedef struct {
2152     Elf32_Half  code;         // Actual value as defined in elf.h
2153     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
2154     char        elf_class;    // 32 or 64 bit
2155     char        endianess;    // MSB or LSB
2156     char*       name;         // String representation
2157   } arch_t;
2158 
2159   #ifndef EM_486
2160   #define EM_486          6               /* Intel 80486 */
2161   #endif
2162 
2163   #ifndef EM_MIPS_RS3_LE
2164   #define EM_MIPS_RS3_LE  10              /* MIPS */
2165   #endif
2166 
2167   #ifndef EM_PPC64
2168   #define EM_PPC64        21              /* PowerPC64 */
2169   #endif
2170 
2171   #ifndef EM_S390
2172   #define EM_S390         22              /* IBM System/390 */
2173   #endif
2174 
2175   #ifndef EM_IA_64
2176   #define EM_IA_64        50              /* HP/Intel IA-64 */
2177   #endif
2178 
2179   #ifndef EM_X86_64
2180   #define EM_X86_64       62              /* AMD x86-64 */
2181   #endif
2182 
2183   static const arch_t arch_array[]={
2184     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2185     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2186     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2187     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2188     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2189     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2190     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2191     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2192     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
2193     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
2194     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
2195     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
2196     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
2197     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
2198     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
2199     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
2200   };
2201 
2202   #if  (defined IA32)
2203     static  Elf32_Half running_arch_code=EM_386;
2204   #elif   (defined AMD64)
2205     static  Elf32_Half running_arch_code=EM_X86_64;
2206   #elif  (defined IA64)
2207     static  Elf32_Half running_arch_code=EM_IA_64;
2208   #elif  (defined __sparc) && (defined _LP64)
2209     static  Elf32_Half running_arch_code=EM_SPARCV9;
2210   #elif  (defined __sparc) && (!defined _LP64)
2211     static  Elf32_Half running_arch_code=EM_SPARC;
2212   #elif  (defined __powerpc64__)
2213     static  Elf32_Half running_arch_code=EM_PPC64;
2214   #elif  (defined __powerpc__)
2215     static  Elf32_Half running_arch_code=EM_PPC;
2216   #elif  (defined ARM)
2217     static  Elf32_Half running_arch_code=EM_ARM;
2218   #elif  (defined S390)
2219     static  Elf32_Half running_arch_code=EM_S390;
2220   #elif  (defined ALPHA)
2221     static  Elf32_Half running_arch_code=EM_ALPHA;
2222   #elif  (defined MIPSEL)
2223     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
2224   #elif  (defined PARISC)
2225     static  Elf32_Half running_arch_code=EM_PARISC;
2226   #elif  (defined MIPS)
2227     static  Elf32_Half running_arch_code=EM_MIPS;
2228   #elif  (defined M68K)
2229     static  Elf32_Half running_arch_code=EM_68K;
2230   #else
2231     #error Method os::dll_load requires that one of following is defined:\
2232          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2233   #endif
2234 
2235   // Identify compatability class for VM's architecture and library's architecture
2236   // Obtain string descriptions for architectures
2237 
2238   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2239   int running_arch_index=-1;
2240 
2241   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2242     if (running_arch_code == arch_array[i].code) {
2243       running_arch_index    = i;
2244     }
2245     if (lib_arch.code == arch_array[i].code) {
2246       lib_arch.compat_class = arch_array[i].compat_class;
2247       lib_arch.name         = arch_array[i].name;
2248     }
2249   }
2250 
2251   assert(running_arch_index != -1,
2252     "Didn't find running architecture code (running_arch_code) in arch_array");
2253   if (running_arch_index == -1) {
2254     // Even though running architecture detection failed
2255     // we may still continue with reporting dlerror() message
2256     return NULL;
2257   }
2258 
2259   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2260     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2261     return NULL;
2262   }
2263 
2264 #ifndef S390
2265   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2266     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2267     return NULL;
2268   }
2269 #endif // !S390
2270 
2271   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2272     if ( lib_arch.name!=NULL ) {
2273       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2274         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2275         lib_arch.name, arch_array[running_arch_index].name);
2276     } else {
2277       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2278       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2279         lib_arch.code,
2280         arch_array[running_arch_index].name);
2281     }
2282   }
2283 
2284   return NULL;
2285 }
2286 #endif /* !__APPLE__ */
2287 
2288 // XXX: Do we need a lock around this as per Linux?
2289 void* os::dll_lookup(void* handle, const char* name) {
2290   return dlsym(handle, name);
2291 }
2292 
2293 
2294 static bool _print_ascii_file(const char* filename, outputStream* st) {
2295   int fd = ::open(filename, O_RDONLY);
2296   if (fd == -1) {
2297      return false;
2298   }
2299 
2300   char buf[32];
2301   int bytes;
2302   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2303     st->print_raw(buf, bytes);
2304   }
2305 
2306   ::close(fd);
2307 
2308   return true;
2309 }
2310 
2311 void os::print_dll_info(outputStream *st) {
2312    st->print_cr("Dynamic libraries:");
2313 #ifdef _ALLBSD_SOURCE
2314 #ifdef RTLD_DI_LINKMAP
2315     Dl_info dli;
2316     void *handle;
2317     Link_map *map;
2318     Link_map *p;
2319 
2320     if (!dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli)) {
2321         st->print_cr("Error: Cannot print dynamic libraries.");
2322         return;
2323     }
2324     handle = dlopen(dli.dli_fname, RTLD_LAZY);
2325     if (handle == NULL) {
2326         st->print_cr("Error: Cannot print dynamic libraries.");
2327         return;
2328     }
2329     dlinfo(handle, RTLD_DI_LINKMAP, &map);
2330     if (map == NULL) {
2331         st->print_cr("Error: Cannot print dynamic libraries.");
2332         return;
2333     }
2334 
2335     while (map->l_prev != NULL)
2336         map = map->l_prev;
2337 
2338     while (map != NULL) {
2339         st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
2340         map = map->l_next;
2341     }
2342 
2343     dlclose(handle);
2344 #elif defined(__APPLE__)
2345     uint32_t count;
2346     uint32_t i;
2347 
2348     count = _dyld_image_count();
2349     for (i = 1; i < count; i++) {
2350         const char *name = _dyld_get_image_name(i);
2351         intptr_t slide = _dyld_get_image_vmaddr_slide(i);
2352         st->print_cr(PTR_FORMAT " \t%s", slide, name);
2353     }
2354 #else
2355    st->print_cr("Error: Cannot print dynamic libraries.");
2356 #endif
2357 #else
2358    char fname[32];
2359    pid_t pid = os::Bsd::gettid();
2360 
2361    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2362 
2363    if (!_print_ascii_file(fname, st)) {
2364      st->print("Can not get library information for pid = %d\n", pid);
2365    }
2366 #endif
2367 }
2368 
2369 void os::print_os_info_brief(outputStream* st) {
2370   st->print("Bsd");
2371 
2372   os::Posix::print_uname_info(st);
2373 }
2374 
2375 void os::print_os_info(outputStream* st) {
2376   st->print("OS:");
2377   st->print("Bsd");
2378 
2379   os::Posix::print_uname_info(st);
2380 
2381   os::Posix::print_rlimit_info(st);
2382 
2383   os::Posix::print_load_average(st);
2384 }
2385 
2386 void os::pd_print_cpu_info(outputStream* st) {
2387   // Nothing to do for now.
2388 }
2389 
2390 void os::print_memory_info(outputStream* st) {
2391 
2392   st->print("Memory:");
2393   st->print(" %dk page", os::vm_page_size()>>10);
2394 
2395 #ifndef _ALLBSD_SOURCE
2396   // values in struct sysinfo are "unsigned long"
2397   struct sysinfo si;
2398   sysinfo(&si);
2399 #endif
2400 
2401   st->print(", physical " UINT64_FORMAT "k",
2402             os::physical_memory() >> 10);
2403   st->print("(" UINT64_FORMAT "k free)",
2404             os::available_memory() >> 10);
2405 #ifndef _ALLBSD_SOURCE
2406   st->print(", swap " UINT64_FORMAT "k",
2407             ((jlong)si.totalswap * si.mem_unit) >> 10);
2408   st->print("(" UINT64_FORMAT "k free)",
2409             ((jlong)si.freeswap * si.mem_unit) >> 10);
2410 #endif
2411   st->cr();
2412 
2413   // meminfo
2414   st->print("\n/proc/meminfo:\n");
2415   _print_ascii_file("/proc/meminfo", st);
2416   st->cr();
2417 }
2418 
2419 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2420 // but they're the same for all the bsd arch that we support
2421 // and they're the same for solaris but there's no common place to put this.
2422 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2423                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2424                           "ILL_COPROC", "ILL_BADSTK" };
2425 
2426 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2427                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2428                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2429 
2430 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2431 
2432 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2433 
2434 void os::print_siginfo(outputStream* st, void* siginfo) {
2435   st->print("siginfo:");
2436 
2437   const int buflen = 100;
2438   char buf[buflen];
2439   siginfo_t *si = (siginfo_t*)siginfo;
2440   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2441   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2442     st->print("si_errno=%s", buf);
2443   } else {
2444     st->print("si_errno=%d", si->si_errno);
2445   }
2446   const int c = si->si_code;
2447   assert(c > 0, "unexpected si_code");
2448   switch (si->si_signo) {
2449   case SIGILL:
2450     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2451     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2452     break;
2453   case SIGFPE:
2454     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2455     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2456     break;
2457   case SIGSEGV:
2458     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2459     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2460     break;
2461   case SIGBUS:
2462     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2463     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2464     break;
2465   default:
2466     st->print(", si_code=%d", si->si_code);
2467     // no si_addr
2468   }
2469 
2470   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2471       UseSharedSpaces) {
2472     FileMapInfo* mapinfo = FileMapInfo::current_info();
2473     if (mapinfo->is_in_shared_space(si->si_addr)) {
2474       st->print("\n\nError accessing class data sharing archive."   \
2475                 " Mapped file inaccessible during execution, "      \
2476                 " possible disk/network problem.");
2477     }
2478   }
2479   st->cr();
2480 }
2481 
2482 
2483 static void print_signal_handler(outputStream* st, int sig,
2484                                  char* buf, size_t buflen);
2485 
2486 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2487   st->print_cr("Signal Handlers:");
2488   print_signal_handler(st, SIGSEGV, buf, buflen);
2489   print_signal_handler(st, SIGBUS , buf, buflen);
2490   print_signal_handler(st, SIGFPE , buf, buflen);
2491   print_signal_handler(st, SIGPIPE, buf, buflen);
2492   print_signal_handler(st, SIGXFSZ, buf, buflen);
2493   print_signal_handler(st, SIGILL , buf, buflen);
2494   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2495   print_signal_handler(st, SR_signum, buf, buflen);
2496   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2497   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2498   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2499   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2500 }
2501 
2502 static char saved_jvm_path[MAXPATHLEN] = {0};
2503 
2504 // Find the full path to the current module, libjvm or libjvm_g
2505 void os::jvm_path(char *buf, jint buflen) {
2506   // Error checking.
2507   if (buflen < MAXPATHLEN) {
2508     assert(false, "must use a large-enough buffer");
2509     buf[0] = '\0';
2510     return;
2511   }
2512   // Lazy resolve the path to current module.
2513   if (saved_jvm_path[0] != 0) {
2514     strcpy(buf, saved_jvm_path);
2515     return;
2516   }
2517 
2518   char dli_fname[MAXPATHLEN];
2519   bool ret = dll_address_to_library_name(
2520                 CAST_FROM_FN_PTR(address, os::jvm_path),
2521                 dli_fname, sizeof(dli_fname), NULL);
2522   assert(ret != 0, "cannot locate libjvm");
2523   char *rp = realpath(dli_fname, buf);
2524   if (rp == NULL)
2525     return;
2526 
2527   if (Arguments::created_by_gamma_launcher()) {
2528     // Support for the gamma launcher.  Typical value for buf is
2529     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm".  If "/jre/lib/" appears at
2530     // the right place in the string, then assume we are installed in a JDK and
2531     // we're done.  Otherwise, check for a JAVA_HOME environment variable and
2532     // construct a path to the JVM being overridden.
2533 
2534     const char *p = buf + strlen(buf) - 1;
2535     for (int count = 0; p > buf && count < 5; ++count) {
2536       for (--p; p > buf && *p != '/'; --p)
2537         /* empty */ ;
2538     }
2539 
2540     if (strncmp(p, "/jre/lib/", 9) != 0) {
2541       // Look for JAVA_HOME in the environment.
2542       char* java_home_var = ::getenv("JAVA_HOME");
2543       if (java_home_var != NULL && java_home_var[0] != 0) {
2544         char* jrelib_p;
2545         int len;
2546 
2547         // Check the current module name "libjvm" or "libjvm_g".
2548         p = strrchr(buf, '/');
2549         assert(strstr(p, "/libjvm") == p, "invalid library name");
2550         p = strstr(p, "_g") ? "_g" : "";
2551 
2552         rp = realpath(java_home_var, buf);
2553         if (rp == NULL)
2554           return;
2555 
2556         // determine if this is a legacy image or modules image
2557         // modules image doesn't have "jre" subdirectory
2558         len = strlen(buf);
2559         jrelib_p = buf + len;
2560 
2561         // Add the appropriate library subdir
2562         snprintf(jrelib_p, buflen-len, "/jre/lib");
2563         if (0 != access(buf, F_OK)) {
2564           snprintf(jrelib_p, buflen-len, "/lib");
2565         }
2566 
2567         // Add the appropriate client or server subdir
2568         len = strlen(buf);
2569         jrelib_p = buf + len;
2570         snprintf(jrelib_p, buflen-len, "/%s", COMPILER_VARIANT);
2571         if (0 != access(buf, F_OK)) {
2572           snprintf(jrelib_p, buflen-len, "");
2573         }
2574 
2575         // If the path exists within JAVA_HOME, add the JVM library name
2576         // to complete the path to JVM being overridden.  Otherwise fallback
2577         // to the path to the current library.
2578         if (0 == access(buf, F_OK)) {
2579           // Use current module name "libjvm[_g]" instead of
2580           // "libjvm"debug_only("_g")"" since for fastdebug version
2581           // we should have "libjvm" but debug_only("_g") adds "_g"!
2582           len = strlen(buf);
2583           snprintf(buf + len, buflen-len, "/libjvm%s%s", p, JNI_LIB_SUFFIX);
2584         } else {
2585           // Fall back to path of current library
2586           rp = realpath(dli_fname, buf);
2587           if (rp == NULL)
2588             return;
2589         }
2590       }
2591     }
2592   }
2593 
2594   strcpy(saved_jvm_path, buf);
2595 }
2596 
2597 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2598   // no prefix required, not even "_"
2599 }
2600 
2601 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2602   // no suffix required
2603 }
2604 
2605 ////////////////////////////////////////////////////////////////////////////////
2606 // sun.misc.Signal support
2607 
2608 static volatile jint sigint_count = 0;
2609 
2610 static void
2611 UserHandler(int sig, void *siginfo, void *context) {
2612   // 4511530 - sem_post is serialized and handled by the manager thread. When
2613   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2614   // don't want to flood the manager thread with sem_post requests.
2615   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2616       return;
2617 
2618   // Ctrl-C is pressed during error reporting, likely because the error
2619   // handler fails to abort. Let VM die immediately.
2620   if (sig == SIGINT && is_error_reported()) {
2621      os::die();
2622   }
2623 
2624   os::signal_notify(sig);
2625 }
2626 
2627 void* os::user_handler() {
2628   return CAST_FROM_FN_PTR(void*, UserHandler);
2629 }
2630 
2631 extern "C" {
2632   typedef void (*sa_handler_t)(int);
2633   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2634 }
2635 
2636 void* os::signal(int signal_number, void* handler) {
2637   struct sigaction sigAct, oldSigAct;
2638 
2639   sigfillset(&(sigAct.sa_mask));
2640   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2641   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2642 
2643   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2644     // -1 means registration failed
2645     return (void *)-1;
2646   }
2647 
2648   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2649 }
2650 
2651 void os::signal_raise(int signal_number) {
2652   ::raise(signal_number);
2653 }
2654 
2655 /*
2656  * The following code is moved from os.cpp for making this
2657  * code platform specific, which it is by its very nature.
2658  */
2659 
2660 // Will be modified when max signal is changed to be dynamic
2661 int os::sigexitnum_pd() {
2662   return NSIG;
2663 }
2664 
2665 // a counter for each possible signal value
2666 static volatile jint pending_signals[NSIG+1] = { 0 };
2667 
2668 // Bsd(POSIX) specific hand shaking semaphore.
2669 #ifdef __APPLE__
2670 typedef semaphore_t os_semaphore_t;
2671 #define SEM_INIT(sem, value)    semaphore_create(mach_task_self(), &sem, SYNC_POLICY_FIFO, value)
2672 #define SEM_WAIT(sem)           semaphore_wait(sem)
2673 #define SEM_POST(sem)           semaphore_signal(sem)
2674 #define SEM_DESTROY(sem)        semaphore_destroy(mach_task_self(), sem)
2675 #else
2676 typedef sem_t os_semaphore_t;
2677 #define SEM_INIT(sem, value)    sem_init(&sem, 0, value)
2678 #define SEM_WAIT(sem)           sem_wait(&sem)
2679 #define SEM_POST(sem)           sem_post(&sem)
2680 #define SEM_DESTROY(sem)        sem_destroy(&sem)
2681 #endif
2682 
2683 class Semaphore : public StackObj {
2684   public:
2685     Semaphore();
2686     ~Semaphore();
2687     void signal();
2688     void wait();
2689     bool trywait();
2690     bool timedwait(unsigned int sec, int nsec);
2691   private:
2692     jlong currenttime() const;
2693     semaphore_t _semaphore;
2694 };
2695 
2696 Semaphore::Semaphore() : _semaphore(0) {
2697   SEM_INIT(_semaphore, 0);
2698 }
2699 
2700 Semaphore::~Semaphore() {
2701   SEM_DESTROY(_semaphore);
2702 }
2703 
2704 void Semaphore::signal() {
2705   SEM_POST(_semaphore);
2706 }
2707 
2708 void Semaphore::wait() {
2709   SEM_WAIT(_semaphore);
2710 }
2711 
2712 jlong Semaphore::currenttime() const {
2713     struct timeval tv;
2714     gettimeofday(&tv, NULL);
2715     return (tv.tv_sec * NANOSECS_PER_SEC) + (tv.tv_usec * 1000);
2716 }
2717 
2718 #ifdef __APPLE__
2719 bool Semaphore::trywait() {
2720   return timedwait(0, 0);
2721 }
2722 
2723 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2724   kern_return_t kr = KERN_ABORTED;
2725   mach_timespec_t waitspec;
2726   waitspec.tv_sec = sec;
2727   waitspec.tv_nsec = nsec;
2728 
2729   jlong starttime = currenttime();
2730 
2731   kr = semaphore_timedwait(_semaphore, waitspec);
2732   while (kr == KERN_ABORTED) {
2733     jlong totalwait = (sec * NANOSECS_PER_SEC) + nsec;
2734 
2735     jlong current = currenttime();
2736     jlong passedtime = current - starttime;
2737 
2738     if (passedtime >= totalwait) {
2739       waitspec.tv_sec = 0;
2740       waitspec.tv_nsec = 0;
2741     } else {
2742       jlong waittime = totalwait - (current - starttime);
2743       waitspec.tv_sec = waittime / NANOSECS_PER_SEC;
2744       waitspec.tv_nsec = waittime % NANOSECS_PER_SEC;
2745     }
2746 
2747     kr = semaphore_timedwait(_semaphore, waitspec);
2748   }
2749 
2750   return kr == KERN_SUCCESS;
2751 }
2752 
2753 #else
2754 
2755 bool Semaphore::trywait() {
2756   return sem_trywait(&_semaphore) == 0;
2757 }
2758 
2759 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2760   struct timespec ts;
2761   jlong endtime = unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2762 
2763   while (1) {
2764     int result = sem_timedwait(&_semaphore, &ts);
2765     if (result == 0) {
2766       return true;
2767     } else if (errno == EINTR) {
2768       continue;
2769     } else if (errno == ETIMEDOUT) {
2770       return false;
2771     } else {
2772       return false;
2773     }
2774   }
2775 }
2776 
2777 #endif // __APPLE__
2778 
2779 static os_semaphore_t sig_sem;
2780 static Semaphore sr_semaphore;
2781 
2782 void os::signal_init_pd() {
2783   // Initialize signal structures
2784   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2785 
2786   // Initialize signal semaphore
2787   ::SEM_INIT(sig_sem, 0);
2788 }
2789 
2790 void os::signal_notify(int sig) {
2791   Atomic::inc(&pending_signals[sig]);
2792   ::SEM_POST(sig_sem);
2793 }
2794 
2795 static int check_pending_signals(bool wait) {
2796   Atomic::store(0, &sigint_count);
2797   for (;;) {
2798     for (int i = 0; i < NSIG + 1; i++) {
2799       jint n = pending_signals[i];
2800       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2801         return i;
2802       }
2803     }
2804     if (!wait) {
2805       return -1;
2806     }
2807     JavaThread *thread = JavaThread::current();
2808     ThreadBlockInVM tbivm(thread);
2809 
2810     bool threadIsSuspended;
2811     do {
2812       thread->set_suspend_equivalent();
2813       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2814       ::SEM_WAIT(sig_sem);
2815 
2816       // were we externally suspended while we were waiting?
2817       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2818       if (threadIsSuspended) {
2819         //
2820         // The semaphore has been incremented, but while we were waiting
2821         // another thread suspended us. We don't want to continue running
2822         // while suspended because that would surprise the thread that
2823         // suspended us.
2824         //
2825         ::SEM_POST(sig_sem);
2826 
2827         thread->java_suspend_self();
2828       }
2829     } while (threadIsSuspended);
2830   }
2831 }
2832 
2833 int os::signal_lookup() {
2834   return check_pending_signals(false);
2835 }
2836 
2837 int os::signal_wait() {
2838   return check_pending_signals(true);
2839 }
2840 
2841 ////////////////////////////////////////////////////////////////////////////////
2842 // Virtual Memory
2843 
2844 int os::vm_page_size() {
2845   // Seems redundant as all get out
2846   assert(os::Bsd::page_size() != -1, "must call os::init");
2847   return os::Bsd::page_size();
2848 }
2849 
2850 // Solaris allocates memory by pages.
2851 int os::vm_allocation_granularity() {
2852   assert(os::Bsd::page_size() != -1, "must call os::init");
2853   return os::Bsd::page_size();
2854 }
2855 
2856 // Rationale behind this function:
2857 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2858 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2859 //  samples for JITted code. Here we create private executable mapping over the code cache
2860 //  and then we can use standard (well, almost, as mapping can change) way to provide
2861 //  info for the reporting script by storing timestamp and location of symbol
2862 void bsd_wrap_code(char* base, size_t size) {
2863   static volatile jint cnt = 0;
2864 
2865   if (!UseOprofile) {
2866     return;
2867   }
2868 
2869   char buf[PATH_MAX + 1];
2870   int num = Atomic::add(1, &cnt);
2871 
2872   snprintf(buf, PATH_MAX + 1, "%s/hs-vm-%d-%d",
2873            os::get_temp_directory(), os::current_process_id(), num);
2874   unlink(buf);
2875 
2876   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2877 
2878   if (fd != -1) {
2879     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2880     if (rv != (off_t)-1) {
2881       if (::write(fd, "", 1) == 1) {
2882         mmap(base, size,
2883              PROT_READ|PROT_WRITE|PROT_EXEC,
2884              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2885       }
2886     }
2887     ::close(fd);
2888     unlink(buf);
2889   }
2890 }
2891 
2892 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2893                                     int err) {
2894   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2895           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2896           strerror(err), err);
2897 }
2898 
2899 // NOTE: Bsd kernel does not really reserve the pages for us.
2900 //       All it does is to check if there are enough free pages
2901 //       left at the time of mmap(). This could be a potential
2902 //       problem.
2903 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2904   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2905 #ifdef __OpenBSD__
2906   // XXX: Work-around mmap/MAP_FIXED bug temporarily on OpenBSD
2907   if (::mprotect(addr, size, prot) == 0) {
2908     return true;
2909   }
2910 #else
2911   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2912                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2913   if (res != (uintptr_t) MAP_FAILED) {
2914     return true;
2915   }
2916 #endif
2917 
2918   // Warn about any commit errors we see in non-product builds just
2919   // in case mmap() doesn't work as described on the man page.
2920   NOT_PRODUCT(warn_fail_commit_memory(addr, size, exec, errno);)
2921 
2922   return false;
2923 }
2924 
2925 #ifndef _ALLBSD_SOURCE
2926 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2927 #ifndef MAP_HUGETLB
2928 #define MAP_HUGETLB 0x40000
2929 #endif
2930 
2931 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2932 #ifndef MADV_HUGEPAGE
2933 #define MADV_HUGEPAGE 14
2934 #endif
2935 #endif
2936 
2937 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2938                        bool exec) {
2939 #ifndef _ALLBSD_SOURCE
2940   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2941     int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2942     uintptr_t res =
2943       (uintptr_t) ::mmap(addr, size, prot,
2944                          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2945                          -1, 0);
2946     return res != (uintptr_t) MAP_FAILED;
2947   }
2948 #endif
2949 
2950   // alignment_hint is ignored on this OS
2951   return pd_commit_memory(addr, size, exec);
2952 }
2953 
2954 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2955                                   const char* mesg) {
2956   assert(mesg != NULL, "mesg must be specified");
2957   if (!pd_commit_memory(addr, size, exec)) {
2958     // add extra info in product mode for vm_exit_out_of_memory():
2959     PRODUCT_ONLY(warn_fail_commit_memory(addr, size, exec, errno);)
2960     vm_exit_out_of_memory(size, mesg);
2961   }
2962 }
2963 
2964 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2965                                   size_t alignment_hint, bool exec,
2966                                   const char* mesg) {
2967   // alignment_hint is ignored on this OS
2968   pd_commit_memory_or_exit(addr, size, exec, mesg);
2969 }
2970 
2971 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2972 #ifndef _ALLBSD_SOURCE
2973   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2974     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2975     // be supported or the memory may already be backed by huge pages.
2976     ::madvise(addr, bytes, MADV_HUGEPAGE);
2977   }
2978 #endif
2979 }
2980 
2981 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2982   ::madvise(addr, bytes, MADV_DONTNEED);
2983 }
2984 
2985 void os::numa_make_global(char *addr, size_t bytes) {
2986 }
2987 
2988 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2989 }
2990 
2991 bool os::numa_topology_changed()   { return false; }
2992 
2993 size_t os::numa_get_groups_num() {
2994   return 1;
2995 }
2996 
2997 int os::numa_get_group_id() {
2998   return 0;
2999 }
3000 
3001 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
3002   if (size > 0) {
3003     ids[0] = 0;
3004     return 1;
3005   }
3006   return 0;
3007 }
3008 
3009 bool os::get_page_info(char *start, page_info* info) {
3010   return false;
3011 }
3012 
3013 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
3014   return end;
3015 }
3016 
3017 #ifndef _ALLBSD_SOURCE
3018 // Something to do with the numa-aware allocator needs these symbols
3019 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3020 extern "C" JNIEXPORT void numa_error(char *where) { }
3021 extern "C" JNIEXPORT int fork1() { return fork(); }
3022 
3023 
3024 // If we are running with libnuma version > 2, then we should
3025 // be trying to use symbols with versions 1.1
3026 // If we are running with earlier version, which did not have symbol versions,
3027 // we should use the base version.
3028 void* os::Bsd::libnuma_dlsym(void* handle, const char *name) {
3029   void *f = dlvsym(handle, name, "libnuma_1.1");
3030   if (f == NULL) {
3031     f = dlsym(handle, name);
3032   }
3033   return f;
3034 }
3035 
3036 bool os::Bsd::libnuma_init() {
3037   // sched_getcpu() should be in libc.
3038   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3039                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
3040 
3041   if (sched_getcpu() != -1) { // Does it work?
3042     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3043     if (handle != NULL) {
3044       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3045                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
3046       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3047                                        libnuma_dlsym(handle, "numa_max_node")));
3048       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3049                                         libnuma_dlsym(handle, "numa_available")));
3050       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3051                                             libnuma_dlsym(handle, "numa_tonode_memory")));
3052       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3053                                             libnuma_dlsym(handle, "numa_interleave_memory")));
3054 
3055 
3056       if (numa_available() != -1) {
3057         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3058         // Create a cpu -> node mapping
3059         _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
3060         rebuild_cpu_to_node_map();
3061         return true;
3062       }
3063     }
3064   }
3065   return false;
3066 }
3067 
3068 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3069 // The table is later used in get_node_by_cpu().
3070 void os::Bsd::rebuild_cpu_to_node_map() {
3071   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3072                               // in libnuma (possible values are starting from 16,
3073                               // and continuing up with every other power of 2, but less
3074                               // than the maximum number of CPUs supported by kernel), and
3075                               // is a subject to change (in libnuma version 2 the requirements
3076                               // are more reasonable) we'll just hardcode the number they use
3077                               // in the library.
3078   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3079 
3080   size_t cpu_num = os::active_processor_count();
3081   size_t cpu_map_size = NCPUS / BitsPerCLong;
3082   size_t cpu_map_valid_size =
3083     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3084 
3085   cpu_to_node()->clear();
3086   cpu_to_node()->at_grow(cpu_num - 1);
3087   size_t node_num = numa_get_groups_num();
3088 
3089   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
3090   for (size_t i = 0; i < node_num; i++) {
3091     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3092       for (size_t j = 0; j < cpu_map_valid_size; j++) {
3093         if (cpu_map[j] != 0) {
3094           for (size_t k = 0; k < BitsPerCLong; k++) {
3095             if (cpu_map[j] & (1UL << k)) {
3096               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
3097             }
3098           }
3099         }
3100       }
3101     }
3102   }
3103   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3104 }
3105 
3106 int os::Bsd::get_node_by_cpu(int cpu_id) {
3107   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3108     return cpu_to_node()->at(cpu_id);
3109   }
3110   return -1;
3111 }
3112 
3113 GrowableArray<int>* os::Bsd::_cpu_to_node;
3114 os::Bsd::sched_getcpu_func_t os::Bsd::_sched_getcpu;
3115 os::Bsd::numa_node_to_cpus_func_t os::Bsd::_numa_node_to_cpus;
3116 os::Bsd::numa_max_node_func_t os::Bsd::_numa_max_node;
3117 os::Bsd::numa_available_func_t os::Bsd::_numa_available;
3118 os::Bsd::numa_tonode_memory_func_t os::Bsd::_numa_tonode_memory;
3119 os::Bsd::numa_interleave_memory_func_t os::Bsd::_numa_interleave_memory;
3120 unsigned long* os::Bsd::_numa_all_nodes;
3121 #endif
3122 
3123 bool os::pd_uncommit_memory(char* addr, size_t size) {
3124 #ifdef __OpenBSD__
3125   // XXX: Work-around mmap/MAP_FIXED bug temporarily on OpenBSD
3126   return ::mprotect(addr, size, PROT_NONE) == 0;
3127 #else
3128   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3129                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3130   return res  != (uintptr_t) MAP_FAILED;
3131 #endif
3132 }
3133 
3134 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3135   return os::commit_memory(addr, size, !ExecMem);
3136 }
3137 
3138 // If this is a growable mapping, remove the guard pages entirely by
3139 // munmap()ping them.  If not, just call uncommit_memory().
3140 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3141   return os::uncommit_memory(addr, size);
3142 }
3143 
3144 static address _highest_vm_reserved_address = NULL;
3145 
3146 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3147 // at 'requested_addr'. If there are existing memory mappings at the same
3148 // location, however, they will be overwritten. If 'fixed' is false,
3149 // 'requested_addr' is only treated as a hint, the return value may or
3150 // may not start from the requested address. Unlike Bsd mmap(), this
3151 // function returns NULL to indicate failure.
3152 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3153   char * addr;
3154   int flags;
3155 
3156   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3157   if (fixed) {
3158     assert((uintptr_t)requested_addr % os::Bsd::page_size() == 0, "unaligned address");
3159     flags |= MAP_FIXED;
3160   }
3161 
3162   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
3163   // to PROT_EXEC if executable when we commit the page.
3164   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
3165                        flags, -1, 0);
3166 
3167   if (addr != MAP_FAILED) {
3168     // anon_mmap() should only get called during VM initialization,
3169     // don't need lock (actually we can skip locking even it can be called
3170     // from multiple threads, because _highest_vm_reserved_address is just a
3171     // hint about the upper limit of non-stack memory regions.)
3172     if ((address)addr + bytes > _highest_vm_reserved_address) {
3173       _highest_vm_reserved_address = (address)addr + bytes;
3174     }
3175   }
3176 
3177   return addr == MAP_FAILED ? NULL : addr;
3178 }
3179 
3180 // Don't update _highest_vm_reserved_address, because there might be memory
3181 // regions above addr + size. If so, releasing a memory region only creates
3182 // a hole in the address space, it doesn't help prevent heap-stack collision.
3183 //
3184 static int anon_munmap(char * addr, size_t size) {
3185   return ::munmap(addr, size) == 0;
3186 }
3187 
3188 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3189                          size_t alignment_hint) {
3190   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3191 }
3192 
3193 bool os::pd_release_memory(char* addr, size_t size) {
3194   return anon_munmap(addr, size);
3195 }
3196 
3197 static address highest_vm_reserved_address() {
3198   return _highest_vm_reserved_address;
3199 }
3200 
3201 static bool bsd_mprotect(char* addr, size_t size, int prot) {
3202   // Bsd wants the mprotect address argument to be page aligned.
3203   char* bottom = (char*)align_size_down((intptr_t)addr, os::Bsd::page_size());
3204 
3205   // According to SUSv3, mprotect() should only be used with mappings
3206   // established by mmap(), and mmap() always maps whole pages. Unaligned
3207   // 'addr' likely indicates problem in the VM (e.g. trying to change
3208   // protection of malloc'ed or statically allocated memory). Check the
3209   // caller if you hit this assert.
3210   assert(addr == bottom, "sanity check");
3211 
3212   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Bsd::page_size());
3213   return ::mprotect(bottom, size, prot) == 0;
3214 }
3215 
3216 // Set protections specified
3217 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3218                         bool is_committed) {
3219   unsigned int p = 0;
3220   switch (prot) {
3221   case MEM_PROT_NONE: p = PROT_NONE; break;
3222   case MEM_PROT_READ: p = PROT_READ; break;
3223   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3224   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3225   default:
3226     ShouldNotReachHere();
3227   }
3228   // is_committed is unused.
3229   return bsd_mprotect(addr, bytes, p);
3230 }
3231 
3232 bool os::guard_memory(char* addr, size_t size) {
3233   return bsd_mprotect(addr, size, PROT_NONE);
3234 }
3235 
3236 bool os::unguard_memory(char* addr, size_t size) {
3237   return bsd_mprotect(addr, size, PROT_READ|PROT_WRITE);
3238 }
3239 
3240 bool os::Bsd::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3241   bool result = false;
3242 #ifndef _ALLBSD_SOURCE
3243   void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
3244                   MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3245                   -1, 0);
3246 
3247   if (p != (void *) -1) {
3248     // We don't know if this really is a huge page or not.
3249     FILE *fp = fopen("/proc/self/maps", "r");
3250     if (fp) {
3251       while (!feof(fp)) {
3252         char chars[257];
3253         long x = 0;
3254         if (fgets(chars, sizeof(chars), fp)) {
3255           if (sscanf(chars, "%lx-%*x", &x) == 1
3256               && x == (long)p) {
3257             if (strstr (chars, "hugepage")) {
3258               result = true;
3259               break;
3260             }
3261           }
3262         }
3263       }
3264       fclose(fp);
3265     }
3266     munmap (p, page_size);
3267     if (result)
3268       return true;
3269   }
3270 
3271   if (warn) {
3272     warning("HugeTLBFS is not supported by the operating system.");
3273   }
3274 #endif
3275 
3276   return result;
3277 }
3278 
3279 /*
3280 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3281 *
3282 * From the coredump_filter documentation:
3283 *
3284 * - (bit 0) anonymous private memory
3285 * - (bit 1) anonymous shared memory
3286 * - (bit 2) file-backed private memory
3287 * - (bit 3) file-backed shared memory
3288 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3289 *           effective only if the bit 2 is cleared)
3290 * - (bit 5) hugetlb private memory
3291 * - (bit 6) hugetlb shared memory
3292 */
3293 static void set_coredump_filter(void) {
3294   FILE *f;
3295   long cdm;
3296 
3297   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3298     return;
3299   }
3300 
3301   if (fscanf(f, "%lx", &cdm) != 1) {
3302     fclose(f);
3303     return;
3304   }
3305 
3306   rewind(f);
3307 
3308   if ((cdm & LARGEPAGES_BIT) == 0) {
3309     cdm |= LARGEPAGES_BIT;
3310     fprintf(f, "%#lx", cdm);
3311   }
3312 
3313   fclose(f);
3314 }
3315 
3316 // Large page support
3317 
3318 static size_t _large_page_size = 0;
3319 
3320 void os::large_page_init() {
3321 #ifndef _ALLBSD_SOURCE
3322   if (!UseLargePages) {
3323     UseHugeTLBFS = false;
3324     UseSHM = false;
3325     return;
3326   }
3327 
3328   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3329     // If UseLargePages is specified on the command line try both methods,
3330     // if it's default, then try only HugeTLBFS.
3331     if (FLAG_IS_DEFAULT(UseLargePages)) {
3332       UseHugeTLBFS = true;
3333     } else {
3334       UseHugeTLBFS = UseSHM = true;
3335     }
3336   }
3337 
3338   if (LargePageSizeInBytes) {
3339     _large_page_size = LargePageSizeInBytes;
3340   } else {
3341     // large_page_size on Bsd is used to round up heap size. x86 uses either
3342     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3343     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3344     // page as large as 256M.
3345     //
3346     // Here we try to figure out page size by parsing /proc/meminfo and looking
3347     // for a line with the following format:
3348     //    Hugepagesize:     2048 kB
3349     //
3350     // If we can't determine the value (e.g. /proc is not mounted, or the text
3351     // format has been changed), we'll use the largest page size supported by
3352     // the processor.
3353 
3354 #ifndef ZERO
3355     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3356                        ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3357 #endif // ZERO
3358 
3359     FILE *fp = fopen("/proc/meminfo", "r");
3360     if (fp) {
3361       while (!feof(fp)) {
3362         int x = 0;
3363         char buf[16];
3364         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3365           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3366             _large_page_size = x * K;
3367             break;
3368           }
3369         } else {
3370           // skip to next line
3371           for (;;) {
3372             int ch = fgetc(fp);
3373             if (ch == EOF || ch == (int)'\n') break;
3374           }
3375         }
3376       }
3377       fclose(fp);
3378     }
3379   }
3380 
3381   // print a warning if any large page related flag is specified on command line
3382   bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3383 
3384   const size_t default_page_size = (size_t)Bsd::page_size();
3385   if (_large_page_size > default_page_size) {
3386     _page_sizes[0] = _large_page_size;
3387     _page_sizes[1] = default_page_size;
3388     _page_sizes[2] = 0;
3389   }
3390   UseHugeTLBFS = UseHugeTLBFS &&
3391                  Bsd::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3392 
3393   if (UseHugeTLBFS)
3394     UseSHM = false;
3395 
3396   UseLargePages = UseHugeTLBFS || UseSHM;
3397 
3398   set_coredump_filter();
3399 #endif
3400 }
3401 
3402 #ifndef _ALLBSD_SOURCE
3403 #ifndef SHM_HUGETLB
3404 #define SHM_HUGETLB 04000
3405 #endif
3406 #endif
3407 
3408 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3409   fatal("This code is not used or maintained.");
3410 
3411   // "exec" is passed in but not used.  Creating the shared image for
3412   // the code cache doesn't have an SHM_X executable permission to check.
3413   assert(UseLargePages && UseSHM, "only for SHM large pages");
3414 
3415   key_t key = IPC_PRIVATE;
3416   char *addr;
3417 
3418   bool warn_on_failure = UseLargePages &&
3419                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3420                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3421                         );
3422   char msg[128];
3423 
3424   // Create a large shared memory region to attach to based on size.
3425   // Currently, size is the total size of the heap
3426 #ifndef _ALLBSD_SOURCE
3427   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3428 #else
3429   int shmid = shmget(key, bytes, IPC_CREAT|SHM_R|SHM_W);
3430 #endif
3431   if (shmid == -1) {
3432      // Possible reasons for shmget failure:
3433      // 1. shmmax is too small for Java heap.
3434      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3435      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3436      // 2. not enough large page memory.
3437      //    > check available large pages: cat /proc/meminfo
3438      //    > increase amount of large pages:
3439      //          echo new_value > /proc/sys/vm/nr_hugepages
3440      //      Note 1: different Bsd may use different name for this property,
3441      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3442      //      Note 2: it's possible there's enough physical memory available but
3443      //            they are so fragmented after a long run that they can't
3444      //            coalesce into large pages. Try to reserve large pages when
3445      //            the system is still "fresh".
3446      if (warn_on_failure) {
3447        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3448        warning(msg);
3449      }
3450      return NULL;
3451   }
3452 
3453   // attach to the region
3454   addr = (char*)shmat(shmid, req_addr, 0);
3455   int err = errno;
3456 
3457   // Remove shmid. If shmat() is successful, the actual shared memory segment
3458   // will be deleted when it's detached by shmdt() or when the process
3459   // terminates. If shmat() is not successful this will remove the shared
3460   // segment immediately.
3461   shmctl(shmid, IPC_RMID, NULL);
3462 
3463   if ((intptr_t)addr == -1) {
3464      if (warn_on_failure) {
3465        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3466        warning(msg);
3467      }
3468      return NULL;
3469   }
3470 
3471   // The memory is committed
3472   MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3473 
3474   return addr;
3475 }
3476 
3477 bool os::release_memory_special(char* base, size_t bytes) {
3478   MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3479   // detaching the SHM segment will also delete it, see reserve_memory_special()
3480   int rslt = shmdt(base);
3481   if (rslt == 0) {
3482     tkr.record((address)base, bytes);
3483     return true;
3484   } else {
3485     tkr.discard();
3486     return false;
3487   }
3488 
3489 }
3490 
3491 size_t os::large_page_size() {
3492   return _large_page_size;
3493 }
3494 
3495 // HugeTLBFS allows application to commit large page memory on demand;
3496 // with SysV SHM the entire memory region must be allocated as shared
3497 // memory.
3498 bool os::can_commit_large_page_memory() {
3499   return UseHugeTLBFS;
3500 }
3501 
3502 bool os::can_execute_large_page_memory() {
3503   return UseHugeTLBFS;
3504 }
3505 
3506 // Reserve memory at an arbitrary address, only if that area is
3507 // available (and not reserved for something else).
3508 
3509 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3510   const int max_tries = 10;
3511   char* base[max_tries];
3512   size_t size[max_tries];
3513   const size_t gap = 0x000000;
3514 
3515   // Assert only that the size is a multiple of the page size, since
3516   // that's all that mmap requires, and since that's all we really know
3517   // about at this low abstraction level.  If we need higher alignment,
3518   // we can either pass an alignment to this method or verify alignment
3519   // in one of the methods further up the call chain.  See bug 5044738.
3520   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3521 
3522   // Repeatedly allocate blocks until the block is allocated at the
3523   // right spot. Give up after max_tries. Note that reserve_memory() will
3524   // automatically update _highest_vm_reserved_address if the call is
3525   // successful. The variable tracks the highest memory address every reserved
3526   // by JVM. It is used to detect heap-stack collision if running with
3527   // fixed-stack BsdThreads. Because here we may attempt to reserve more
3528   // space than needed, it could confuse the collision detecting code. To
3529   // solve the problem, save current _highest_vm_reserved_address and
3530   // calculate the correct value before return.
3531   address old_highest = _highest_vm_reserved_address;
3532 
3533   // Bsd mmap allows caller to pass an address as hint; give it a try first,
3534   // if kernel honors the hint then we can return immediately.
3535   char * addr = anon_mmap(requested_addr, bytes, false);
3536   if (addr == requested_addr) {
3537      return requested_addr;
3538   }
3539 
3540   if (addr != NULL) {
3541      // mmap() is successful but it fails to reserve at the requested address
3542      anon_munmap(addr, bytes);
3543   }
3544 
3545   int i;
3546   for (i = 0; i < max_tries; ++i) {
3547     base[i] = reserve_memory(bytes);
3548 
3549     if (base[i] != NULL) {
3550       // Is this the block we wanted?
3551       if (base[i] == requested_addr) {
3552         size[i] = bytes;
3553         break;
3554       }
3555 
3556       // Does this overlap the block we wanted? Give back the overlapped
3557       // parts and try again.
3558 
3559       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3560       if (top_overlap >= 0 && top_overlap < bytes) {
3561         unmap_memory(base[i], top_overlap);
3562         base[i] += top_overlap;
3563         size[i] = bytes - top_overlap;
3564       } else {
3565         size_t bottom_overlap = base[i] + bytes - requested_addr;
3566         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3567           unmap_memory(requested_addr, bottom_overlap);
3568           size[i] = bytes - bottom_overlap;
3569         } else {
3570           size[i] = bytes;
3571         }
3572       }
3573     }
3574   }
3575 
3576   // Give back the unused reserved pieces.
3577 
3578   for (int j = 0; j < i; ++j) {
3579     if (base[j] != NULL) {
3580       unmap_memory(base[j], size[j]);
3581     }
3582   }
3583 
3584   if (i < max_tries) {
3585     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3586     return requested_addr;
3587   } else {
3588     _highest_vm_reserved_address = old_highest;
3589     return NULL;
3590   }
3591 }
3592 
3593 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3594   RESTARTABLE_RETURN_INT(::read(fd, buf, nBytes));
3595 }
3596 
3597 // TODO-FIXME: reconcile Solaris' os::sleep with the bsd variation.
3598 // Solaris uses poll(), bsd uses park().
3599 // Poll() is likely a better choice, assuming that Thread.interrupt()
3600 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3601 // SIGSEGV, see 4355769.
3602 
3603 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3604   assert(thread == Thread::current(),  "thread consistency check");
3605 
3606   ParkEvent * const slp = thread->_SleepEvent ;
3607   slp->reset() ;
3608   OrderAccess::fence() ;
3609 
3610   if (interruptible) {
3611     jlong prevtime = javaTimeNanos();
3612 
3613     for (;;) {
3614       if (os::is_interrupted(thread, true)) {
3615         return OS_INTRPT;
3616       }
3617 
3618       jlong newtime = javaTimeNanos();
3619 
3620       if (newtime - prevtime < 0) {
3621         // time moving backwards, should only happen if no monotonic clock
3622         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3623         assert(!Bsd::supports_monotonic_clock(), "time moving backwards");
3624       } else {
3625         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3626       }
3627 
3628       if(millis <= 0) {
3629         return OS_OK;
3630       }
3631 
3632       prevtime = newtime;
3633 
3634       {
3635         assert(thread->is_Java_thread(), "sanity check");
3636         JavaThread *jt = (JavaThread *) thread;
3637         ThreadBlockInVM tbivm(jt);
3638         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3639 
3640         jt->set_suspend_equivalent();
3641         // cleared by handle_special_suspend_equivalent_condition() or
3642         // java_suspend_self() via check_and_wait_while_suspended()
3643 
3644         slp->park(millis);
3645 
3646         // were we externally suspended while we were waiting?
3647         jt->check_and_wait_while_suspended();
3648       }
3649     }
3650   } else {
3651     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3652     jlong prevtime = javaTimeNanos();
3653 
3654     for (;;) {
3655       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3656       // the 1st iteration ...
3657       jlong newtime = javaTimeNanos();
3658 
3659       if (newtime - prevtime < 0) {
3660         // time moving backwards, should only happen if no monotonic clock
3661         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3662         assert(!Bsd::supports_monotonic_clock(), "time moving backwards");
3663       } else {
3664         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3665       }
3666 
3667       if(millis <= 0) break ;
3668 
3669       prevtime = newtime;
3670       slp->park(millis);
3671     }
3672     return OS_OK ;
3673   }
3674 }
3675 
3676 int os::naked_sleep() {
3677   // %% make the sleep time an integer flag. for now use 1 millisec.
3678   return os::sleep(Thread::current(), 1, false);
3679 }
3680 
3681 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3682 void os::infinite_sleep() {
3683   while (true) {    // sleep forever ...
3684     ::sleep(100);   // ... 100 seconds at a time
3685   }
3686 }
3687 
3688 // Used to convert frequent JVM_Yield() to nops
3689 bool os::dont_yield() {
3690   return DontYieldALot;
3691 }
3692 
3693 void os::yield() {
3694   sched_yield();
3695 }
3696 
3697 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3698 
3699 void os::yield_all(int attempts) {
3700   // Yields to all threads, including threads with lower priorities
3701   // Threads on Bsd are all with same priority. The Solaris style
3702   // os::yield_all() with nanosleep(1ms) is not necessary.
3703   sched_yield();
3704 }
3705 
3706 // Called from the tight loops to possibly influence time-sharing heuristics
3707 void os::loop_breaker(int attempts) {
3708   os::yield_all(attempts);
3709 }
3710 
3711 ////////////////////////////////////////////////////////////////////////////////
3712 // thread priority support
3713 
3714 // Note: Normal Bsd applications are run with SCHED_OTHER policy. SCHED_OTHER
3715 // only supports dynamic priority, static priority must be zero. For real-time
3716 // applications, Bsd supports SCHED_RR which allows static priority (1-99).
3717 // However, for large multi-threaded applications, SCHED_RR is not only slower
3718 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3719 // of 5 runs - Sep 2005).
3720 //
3721 // The following code actually changes the niceness of kernel-thread/LWP. It
3722 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3723 // not the entire user process, and user level threads are 1:1 mapped to kernel
3724 // threads. It has always been the case, but could change in the future. For
3725 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3726 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3727 
3728 #if defined(_ALLBSD_SOURCE) && !defined(__APPLE__)
3729 int os::java_to_os_priority[CriticalPriority + 1] = {
3730   19,              // 0 Entry should never be used
3731 
3732    0,              // 1 MinPriority
3733    3,              // 2
3734    6,              // 3
3735 
3736   10,              // 4
3737   15,              // 5 NormPriority
3738   18,              // 6
3739 
3740   21,              // 7
3741   25,              // 8
3742   28,              // 9 NearMaxPriority
3743 
3744   31,              // 10 MaxPriority
3745 
3746   31               // 11 CriticalPriority
3747 };
3748 #elif defined(__APPLE__)
3749 /* Using Mach high-level priority assignments */
3750 int os::java_to_os_priority[CriticalPriority + 1] = {
3751    0,              // 0 Entry should never be used (MINPRI_USER)
3752 
3753   27,              // 1 MinPriority
3754   28,              // 2
3755   29,              // 3
3756 
3757   30,              // 4
3758   31,              // 5 NormPriority (BASEPRI_DEFAULT)
3759   32,              // 6
3760 
3761   33,              // 7
3762   34,              // 8
3763   35,              // 9 NearMaxPriority
3764 
3765   36,              // 10 MaxPriority
3766 
3767   36               // 11 CriticalPriority
3768 };
3769 #else
3770 int os::java_to_os_priority[CriticalPriority + 1] = {
3771   19,              // 0 Entry should never be used
3772 
3773    4,              // 1 MinPriority
3774    3,              // 2
3775    2,              // 3
3776 
3777    1,              // 4
3778    0,              // 5 NormPriority
3779   -1,              // 6
3780 
3781   -2,              // 7
3782   -3,              // 8
3783   -4,              // 9 NearMaxPriority
3784 
3785   -5,              // 10 MaxPriority
3786 
3787   -5               // 11 CriticalPriority
3788 };
3789 #endif
3790 
3791 static int prio_init() {
3792   if (ThreadPriorityPolicy == 1) {
3793     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3794     // if effective uid is not root. Perhaps, a more elegant way of doing
3795     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3796     if (geteuid() != 0) {
3797       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3798         warning("-XX:ThreadPriorityPolicy requires root privilege on Bsd");
3799       }
3800       ThreadPriorityPolicy = 0;
3801     }
3802   }
3803   if (UseCriticalJavaThreadPriority) {
3804     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3805   }
3806   return 0;
3807 }
3808 
3809 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3810   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3811 
3812 #ifdef __OpenBSD__
3813   // OpenBSD pthread_setprio starves low priority threads
3814   return OS_OK;
3815 #elif defined(__FreeBSD__)
3816   int ret = pthread_setprio(thread->osthread()->pthread_id(), newpri);
3817 #elif defined(__APPLE__) || defined(__NetBSD__)
3818   struct sched_param sp;
3819   int policy;
3820   pthread_t self = pthread_self();
3821 
3822   if (pthread_getschedparam(self, &policy, &sp) != 0)
3823     return OS_ERR;
3824 
3825   sp.sched_priority = newpri;
3826   if (pthread_setschedparam(self, policy, &sp) != 0)
3827     return OS_ERR;
3828 
3829   return OS_OK;
3830 #else
3831   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3832   return (ret == 0) ? OS_OK : OS_ERR;
3833 #endif
3834 }
3835 
3836 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3837   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3838     *priority_ptr = java_to_os_priority[NormPriority];
3839     return OS_OK;
3840   }
3841 
3842   errno = 0;
3843 #if defined(__OpenBSD__) || defined(__FreeBSD__)
3844   *priority_ptr = pthread_getprio(thread->osthread()->pthread_id());
3845 #elif defined(__APPLE__) || defined(__NetBSD__)
3846   int policy;
3847   struct sched_param sp;
3848 
3849   pthread_getschedparam(pthread_self(), &policy, &sp);
3850   *priority_ptr = sp.sched_priority;
3851 #else
3852   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3853 #endif
3854   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3855 }
3856 
3857 // Hint to the underlying OS that a task switch would not be good.
3858 // Void return because it's a hint and can fail.
3859 void os::hint_no_preempt() {}
3860 
3861 ////////////////////////////////////////////////////////////////////////////////
3862 // suspend/resume support
3863 
3864 //  the low-level signal-based suspend/resume support is a remnant from the
3865 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3866 //  within hotspot. Now there is a single use-case for this:
3867 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3868 //      that runs in the watcher thread.
3869 //  The remaining code is greatly simplified from the more general suspension
3870 //  code that used to be used.
3871 //
3872 //  The protocol is quite simple:
3873 //  - suspend:
3874 //      - sends a signal to the target thread
3875 //      - polls the suspend state of the osthread using a yield loop
3876 //      - target thread signal handler (SR_handler) sets suspend state
3877 //        and blocks in sigsuspend until continued
3878 //  - resume:
3879 //      - sets target osthread state to continue
3880 //      - sends signal to end the sigsuspend loop in the SR_handler
3881 //
3882 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3883 //
3884 
3885 static void resume_clear_context(OSThread *osthread) {
3886   osthread->set_ucontext(NULL);
3887   osthread->set_siginfo(NULL);
3888 }
3889 
3890 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3891   osthread->set_ucontext(context);
3892   osthread->set_siginfo(siginfo);
3893 }
3894 
3895 //
3896 // Handler function invoked when a thread's execution is suspended or
3897 // resumed. We have to be careful that only async-safe functions are
3898 // called here (Note: most pthread functions are not async safe and
3899 // should be avoided.)
3900 //
3901 // Note: sigwait() is a more natural fit than sigsuspend() from an
3902 // interface point of view, but sigwait() prevents the signal hander
3903 // from being run. libpthread would get very confused by not having
3904 // its signal handlers run and prevents sigwait()'s use with the
3905 // mutex granting granting signal.
3906 //
3907 // Currently only ever called on the VMThread or JavaThread
3908 //
3909 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3910   // Save and restore errno to avoid confusing native code with EINTR
3911   // after sigsuspend.
3912   int old_errno = errno;
3913 
3914   Thread* thread = Thread::current();
3915   OSThread* osthread = thread->osthread();
3916   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3917 
3918   os::SuspendResume::State current = osthread->sr.state();
3919   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3920     suspend_save_context(osthread, siginfo, context);
3921 
3922     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3923     os::SuspendResume::State state = osthread->sr.suspended();
3924     if (state == os::SuspendResume::SR_SUSPENDED) {
3925       sigset_t suspend_set;  // signals for sigsuspend()
3926 
3927       // get current set of blocked signals and unblock resume signal
3928       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3929       sigdelset(&suspend_set, SR_signum);
3930 
3931       sr_semaphore.signal();
3932       // wait here until we are resumed
3933       while (1) {
3934         sigsuspend(&suspend_set);
3935 
3936         os::SuspendResume::State result = osthread->sr.running();
3937         if (result == os::SuspendResume::SR_RUNNING) {
3938           sr_semaphore.signal();
3939           break;
3940         } else if (result != os::SuspendResume::SR_SUSPENDED) {
3941           ShouldNotReachHere();
3942         }
3943       }
3944 
3945     } else if (state == os::SuspendResume::SR_RUNNING) {
3946       // request was cancelled, continue
3947     } else {
3948       ShouldNotReachHere();
3949     }
3950 
3951     resume_clear_context(osthread);
3952   } else if (current == os::SuspendResume::SR_RUNNING) {
3953     // request was cancelled, continue
3954   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3955     // ignore
3956   } else {
3957     // ignore
3958   }
3959 
3960   errno = old_errno;
3961 }
3962 
3963 
3964 static int SR_initialize() {
3965   struct sigaction act;
3966   char *s;
3967   /* Get signal number to use for suspend/resume */
3968   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3969     int sig = ::strtol(s, 0, 10);
3970     if (sig > 0 || sig < NSIG) {
3971         SR_signum = sig;
3972     }
3973   }
3974 
3975   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3976         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3977 
3978   sigemptyset(&SR_sigset);
3979   sigaddset(&SR_sigset, SR_signum);
3980 
3981   /* Set up signal handler for suspend/resume */
3982   act.sa_flags = SA_RESTART|SA_SIGINFO;
3983   act.sa_handler = (void (*)(int)) SR_handler;
3984 
3985   // SR_signum is blocked by default.
3986   // 4528190 - We also need to block pthread restart signal (32 on all
3987   // supported Bsd platforms). Note that BsdThreads need to block
3988   // this signal for all threads to work properly. So we don't have
3989   // to use hard-coded signal number when setting up the mask.
3990   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3991 
3992   if (sigaction(SR_signum, &act, 0) == -1) {
3993     return -1;
3994   }
3995 
3996   // Save signal flag
3997   os::Bsd::set_our_sigflags(SR_signum, act.sa_flags);
3998   return 0;
3999 }
4000 
4001 static int SR_finalize() {
4002   return 0;
4003 }
4004 
4005 static int sr_notify(OSThread* osthread) {
4006   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4007   assert_status(status == 0, status, "pthread_kill");
4008   return status;
4009 }
4010 
4011 // "Randomly" selected value for how long we want to spin
4012 // before bailing out on suspending a thread, also how often
4013 // we send a signal to a thread we want to resume
4014 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4015 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4016 
4017 // returns true on success and false on error - really an error is fatal
4018 // but this seems the normal response to library errors
4019 static bool do_suspend(OSThread* osthread) {
4020   assert(osthread->sr.is_running(), "thread should be running");
4021   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4022 
4023   // mark as suspended and send signal
4024   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4025     // failed to switch, state wasn't running?
4026     ShouldNotReachHere();
4027     return false;
4028   }
4029 
4030   if (sr_notify(osthread) != 0) {
4031     ShouldNotReachHere();
4032   }
4033 
4034   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4035   while (true) {
4036     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4037       break;
4038     } else {
4039       // timeout
4040       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4041       if (cancelled == os::SuspendResume::SR_RUNNING) {
4042         return false;
4043       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4044         // make sure that we consume the signal on the semaphore as well
4045         sr_semaphore.wait();
4046         break;
4047       } else {
4048         ShouldNotReachHere();
4049         return false;
4050       }
4051     }
4052   }
4053 
4054   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4055   return true;
4056 }
4057 
4058 static void do_resume(OSThread* osthread) {
4059   assert(osthread->sr.is_suspended(), "thread should be suspended");
4060   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4061 
4062   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4063     // failed to switch to WAKEUP_REQUEST
4064     ShouldNotReachHere();
4065     return;
4066   }
4067 
4068   while (true) {
4069     if (sr_notify(osthread) == 0) {
4070       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4071         if (osthread->sr.is_running()) {
4072           return;
4073         }
4074       }
4075     } else {
4076       ShouldNotReachHere();
4077     }
4078   }
4079 
4080   guarantee(osthread->sr.is_running(), "Must be running!");
4081 }
4082 
4083 ////////////////////////////////////////////////////////////////////////////////
4084 // interrupt support
4085 
4086 void os::interrupt(Thread* thread) {
4087   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4088     "possibility of dangling Thread pointer");
4089 
4090   OSThread* osthread = thread->osthread();
4091 
4092   if (!osthread->interrupted()) {
4093     osthread->set_interrupted(true);
4094     // More than one thread can get here with the same value of osthread,
4095     // resulting in multiple notifications.  We do, however, want the store
4096     // to interrupted() to be visible to other threads before we execute unpark().
4097     OrderAccess::fence();
4098     ParkEvent * const slp = thread->_SleepEvent ;
4099     if (slp != NULL) slp->unpark() ;
4100   }
4101 
4102   // For JSR166. Unpark even if interrupt status already was set
4103   if (thread->is_Java_thread())
4104     ((JavaThread*)thread)->parker()->unpark();
4105 
4106   ParkEvent * ev = thread->_ParkEvent ;
4107   if (ev != NULL) ev->unpark() ;
4108 
4109 }
4110 
4111 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4112   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4113     "possibility of dangling Thread pointer");
4114 
4115   OSThread* osthread = thread->osthread();
4116 
4117   bool interrupted = osthread->interrupted();
4118 
4119   if (interrupted && clear_interrupted) {
4120     osthread->set_interrupted(false);
4121     // consider thread->_SleepEvent->reset() ... optional optimization
4122   }
4123 
4124   return interrupted;
4125 }
4126 
4127 ///////////////////////////////////////////////////////////////////////////////////
4128 // signal handling (except suspend/resume)
4129 
4130 // This routine may be used by user applications as a "hook" to catch signals.
4131 // The user-defined signal handler must pass unrecognized signals to this
4132 // routine, and if it returns true (non-zero), then the signal handler must
4133 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4134 // routine will never retun false (zero), but instead will execute a VM panic
4135 // routine kill the process.
4136 //
4137 // If this routine returns false, it is OK to call it again.  This allows
4138 // the user-defined signal handler to perform checks either before or after
4139 // the VM performs its own checks.  Naturally, the user code would be making
4140 // a serious error if it tried to handle an exception (such as a null check
4141 // or breakpoint) that the VM was generating for its own correct operation.
4142 //
4143 // This routine may recognize any of the following kinds of signals:
4144 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4145 // It should be consulted by handlers for any of those signals.
4146 //
4147 // The caller of this routine must pass in the three arguments supplied
4148 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4149 // field of the structure passed to sigaction().  This routine assumes that
4150 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4151 //
4152 // Note that the VM will print warnings if it detects conflicting signal
4153 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4154 //
4155 extern "C" JNIEXPORT int
4156 JVM_handle_bsd_signal(int signo, siginfo_t* siginfo,
4157                         void* ucontext, int abort_if_unrecognized);
4158 
4159 void signalHandler(int sig, siginfo_t* info, void* uc) {
4160   assert(info != NULL && uc != NULL, "it must be old kernel");
4161   JVM_handle_bsd_signal(sig, info, uc, true);
4162 }
4163 
4164 
4165 // This boolean allows users to forward their own non-matching signals
4166 // to JVM_handle_bsd_signal, harmlessly.
4167 bool os::Bsd::signal_handlers_are_installed = false;
4168 
4169 // For signal-chaining
4170 struct sigaction os::Bsd::sigact[MAXSIGNUM];
4171 unsigned int os::Bsd::sigs = 0;
4172 bool os::Bsd::libjsig_is_loaded = false;
4173 typedef struct sigaction *(*get_signal_t)(int);
4174 get_signal_t os::Bsd::get_signal_action = NULL;
4175 
4176 struct sigaction* os::Bsd::get_chained_signal_action(int sig) {
4177   struct sigaction *actp = NULL;
4178 
4179   if (libjsig_is_loaded) {
4180     // Retrieve the old signal handler from libjsig
4181     actp = (*get_signal_action)(sig);
4182   }
4183   if (actp == NULL) {
4184     // Retrieve the preinstalled signal handler from jvm
4185     actp = get_preinstalled_handler(sig);
4186   }
4187 
4188   return actp;
4189 }
4190 
4191 static bool call_chained_handler(struct sigaction *actp, int sig,
4192                                  siginfo_t *siginfo, void *context) {
4193   // Call the old signal handler
4194   if (actp->sa_handler == SIG_DFL) {
4195     // It's more reasonable to let jvm treat it as an unexpected exception
4196     // instead of taking the default action.
4197     return false;
4198   } else if (actp->sa_handler != SIG_IGN) {
4199     if ((actp->sa_flags & SA_NODEFER) == 0) {
4200       // automaticlly block the signal
4201       sigaddset(&(actp->sa_mask), sig);
4202     }
4203 
4204     sa_handler_t hand;
4205     sa_sigaction_t sa;
4206     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4207     // retrieve the chained handler
4208     if (siginfo_flag_set) {
4209       sa = actp->sa_sigaction;
4210     } else {
4211       hand = actp->sa_handler;
4212     }
4213 
4214     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4215       actp->sa_handler = SIG_DFL;
4216     }
4217 
4218     // try to honor the signal mask
4219     sigset_t oset;
4220     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4221 
4222     // call into the chained handler
4223     if (siginfo_flag_set) {
4224       (*sa)(sig, siginfo, context);
4225     } else {
4226       (*hand)(sig);
4227     }
4228 
4229     // restore the signal mask
4230     pthread_sigmask(SIG_SETMASK, &oset, 0);
4231   }
4232   // Tell jvm's signal handler the signal is taken care of.
4233   return true;
4234 }
4235 
4236 bool os::Bsd::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4237   bool chained = false;
4238   // signal-chaining
4239   if (UseSignalChaining) {
4240     struct sigaction *actp = get_chained_signal_action(sig);
4241     if (actp != NULL) {
4242       chained = call_chained_handler(actp, sig, siginfo, context);
4243     }
4244   }
4245   return chained;
4246 }
4247 
4248 struct sigaction* os::Bsd::get_preinstalled_handler(int sig) {
4249   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4250     return &sigact[sig];
4251   }
4252   return NULL;
4253 }
4254 
4255 void os::Bsd::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4256   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4257   sigact[sig] = oldAct;
4258   sigs |= (unsigned int)1 << sig;
4259 }
4260 
4261 // for diagnostic
4262 int os::Bsd::sigflags[MAXSIGNUM];
4263 
4264 int os::Bsd::get_our_sigflags(int sig) {
4265   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4266   return sigflags[sig];
4267 }
4268 
4269 void os::Bsd::set_our_sigflags(int sig, int flags) {
4270   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4271   sigflags[sig] = flags;
4272 }
4273 
4274 void os::Bsd::set_signal_handler(int sig, bool set_installed) {
4275   // Check for overwrite.
4276   struct sigaction oldAct;
4277   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4278 
4279   void* oldhand = oldAct.sa_sigaction
4280                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4281                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4282   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4283       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4284       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4285     if (AllowUserSignalHandlers || !set_installed) {
4286       // Do not overwrite; user takes responsibility to forward to us.
4287       return;
4288     } else if (UseSignalChaining) {
4289       // save the old handler in jvm
4290       save_preinstalled_handler(sig, oldAct);
4291       // libjsig also interposes the sigaction() call below and saves the
4292       // old sigaction on it own.
4293     } else {
4294       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4295                     "%#lx for signal %d.", (long)oldhand, sig));
4296     }
4297   }
4298 
4299   struct sigaction sigAct;
4300   sigfillset(&(sigAct.sa_mask));
4301   sigAct.sa_handler = SIG_DFL;
4302   if (!set_installed) {
4303     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4304   } else {
4305     sigAct.sa_sigaction = signalHandler;
4306     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4307   }
4308 #if __APPLE__
4309   // Needed for main thread as XNU (Mac OS X kernel) will only deliver SIGSEGV
4310   // (which starts as SIGBUS) on main thread with faulting address inside "stack+guard pages"
4311   // if the signal handler declares it will handle it on alternate stack.
4312   // Notice we only declare we will handle it on alt stack, but we are not
4313   // actually going to use real alt stack - this is just a workaround.
4314   // Please see ux_exception.c, method catch_mach_exception_raise for details
4315   // link http://www.opensource.apple.com/source/xnu/xnu-2050.18.24/bsd/uxkern/ux_exception.c
4316   if (sig == SIGSEGV) {
4317     sigAct.sa_flags |= SA_ONSTACK;
4318   }
4319 #endif
4320 
4321   // Save flags, which are set by ours
4322   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4323   sigflags[sig] = sigAct.sa_flags;
4324 
4325   int ret = sigaction(sig, &sigAct, &oldAct);
4326   assert(ret == 0, "check");
4327 
4328   void* oldhand2  = oldAct.sa_sigaction
4329                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4330                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4331   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4332 }
4333 
4334 // install signal handlers for signals that HotSpot needs to
4335 // handle in order to support Java-level exception handling.
4336 
4337 void os::Bsd::install_signal_handlers() {
4338   if (!signal_handlers_are_installed) {
4339     signal_handlers_are_installed = true;
4340 
4341     // signal-chaining
4342     typedef void (*signal_setting_t)();
4343     signal_setting_t begin_signal_setting = NULL;
4344     signal_setting_t end_signal_setting = NULL;
4345     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4346                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4347     if (begin_signal_setting != NULL) {
4348       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4349                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4350       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4351                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4352       libjsig_is_loaded = true;
4353       assert(UseSignalChaining, "should enable signal-chaining");
4354     }
4355     if (libjsig_is_loaded) {
4356       // Tell libjsig jvm is setting signal handlers
4357       (*begin_signal_setting)();
4358     }
4359 
4360     set_signal_handler(SIGSEGV, true);
4361     set_signal_handler(SIGPIPE, true);
4362     set_signal_handler(SIGBUS, true);
4363     set_signal_handler(SIGILL, true);
4364     set_signal_handler(SIGFPE, true);
4365     set_signal_handler(SIGXFSZ, true);
4366 
4367 #if defined(__APPLE__)
4368     // In Mac OS X 10.4, CrashReporter will write a crash log for all 'fatal' signals, including
4369     // signals caught and handled by the JVM. To work around this, we reset the mach task
4370     // signal handler that's placed on our process by CrashReporter. This disables
4371     // CrashReporter-based reporting.
4372     //
4373     // This work-around is not necessary for 10.5+, as CrashReporter no longer intercedes
4374     // on caught fatal signals.
4375     //
4376     // Additionally, gdb installs both standard BSD signal handlers, and mach exception
4377     // handlers. By replacing the existing task exception handler, we disable gdb's mach
4378     // exception handling, while leaving the standard BSD signal handlers functional.
4379     kern_return_t kr;
4380     kr = task_set_exception_ports(mach_task_self(),
4381         EXC_MASK_BAD_ACCESS | EXC_MASK_ARITHMETIC,
4382         MACH_PORT_NULL,
4383         EXCEPTION_STATE_IDENTITY,
4384         MACHINE_THREAD_STATE);
4385 
4386     assert(kr == KERN_SUCCESS, "could not set mach task signal handler");
4387 #endif
4388 
4389     if (libjsig_is_loaded) {
4390       // Tell libjsig jvm finishes setting signal handlers
4391       (*end_signal_setting)();
4392     }
4393 
4394     // We don't activate signal checker if libjsig is in place, we trust ourselves
4395     // and if UserSignalHandler is installed all bets are off
4396     if (CheckJNICalls) {
4397       if (libjsig_is_loaded) {
4398         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4399         check_signals = false;
4400       }
4401       if (AllowUserSignalHandlers) {
4402         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4403         check_signals = false;
4404       }
4405     }
4406   }
4407 }
4408 
4409 #ifndef _ALLBSD_SOURCE
4410 // This is the fastest way to get thread cpu time on Bsd.
4411 // Returns cpu time (user+sys) for any thread, not only for current.
4412 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4413 // It might work on 2.6.10+ with a special kernel/glibc patch.
4414 // For reference, please, see IEEE Std 1003.1-2004:
4415 //   http://www.unix.org/single_unix_specification
4416 
4417 jlong os::Bsd::fast_thread_cpu_time(clockid_t clockid) {
4418   struct timespec tp;
4419   int rc = os::Bsd::clock_gettime(clockid, &tp);
4420   assert(rc == 0, "clock_gettime is expected to return 0 code");
4421 
4422   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4423 }
4424 #endif
4425 
4426 /////
4427 // glibc on Bsd platform uses non-documented flag
4428 // to indicate, that some special sort of signal
4429 // trampoline is used.
4430 // We will never set this flag, and we should
4431 // ignore this flag in our diagnostic
4432 #ifdef SIGNIFICANT_SIGNAL_MASK
4433 #undef SIGNIFICANT_SIGNAL_MASK
4434 #endif
4435 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4436 
4437 static const char* get_signal_handler_name(address handler,
4438                                            char* buf, int buflen) {
4439   int offset;
4440   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4441   if (found) {
4442     // skip directory names
4443     const char *p1, *p2;
4444     p1 = buf;
4445     size_t len = strlen(os::file_separator());
4446     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4447     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4448   } else {
4449     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4450   }
4451   return buf;
4452 }
4453 
4454 static void print_signal_handler(outputStream* st, int sig,
4455                                  char* buf, size_t buflen) {
4456   struct sigaction sa;
4457 
4458   sigaction(sig, NULL, &sa);
4459 
4460   // See comment for SIGNIFICANT_SIGNAL_MASK define
4461   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4462 
4463   st->print("%s: ", os::exception_name(sig, buf, buflen));
4464 
4465   address handler = (sa.sa_flags & SA_SIGINFO)
4466     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4467     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4468 
4469   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4470     st->print("SIG_DFL");
4471   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4472     st->print("SIG_IGN");
4473   } else {
4474     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4475   }
4476 
4477   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4478 
4479   address rh = VMError::get_resetted_sighandler(sig);
4480   // May be, handler was resetted by VMError?
4481   if(rh != NULL) {
4482     handler = rh;
4483     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4484   }
4485 
4486   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
4487 
4488   // Check: is it our handler?
4489   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4490      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4491     // It is our signal handler
4492     // check for flags, reset system-used one!
4493     if((int)sa.sa_flags != os::Bsd::get_our_sigflags(sig)) {
4494       st->print(
4495                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4496                 os::Bsd::get_our_sigflags(sig));
4497     }
4498   }
4499   st->cr();
4500 }
4501 
4502 
4503 #define DO_SIGNAL_CHECK(sig) \
4504   if (!sigismember(&check_signal_done, sig)) \
4505     os::Bsd::check_signal_handler(sig)
4506 
4507 // This method is a periodic task to check for misbehaving JNI applications
4508 // under CheckJNI, we can add any periodic checks here
4509 
4510 void os::run_periodic_checks() {
4511 
4512   if (check_signals == false) return;
4513 
4514   // SEGV and BUS if overridden could potentially prevent
4515   // generation of hs*.log in the event of a crash, debugging
4516   // such a case can be very challenging, so we absolutely
4517   // check the following for a good measure:
4518   DO_SIGNAL_CHECK(SIGSEGV);
4519   DO_SIGNAL_CHECK(SIGILL);
4520   DO_SIGNAL_CHECK(SIGFPE);
4521   DO_SIGNAL_CHECK(SIGBUS);
4522   DO_SIGNAL_CHECK(SIGPIPE);
4523   DO_SIGNAL_CHECK(SIGXFSZ);
4524 
4525 
4526   // ReduceSignalUsage allows the user to override these handlers
4527   // see comments at the very top and jvm_solaris.h
4528   if (!ReduceSignalUsage) {
4529     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4530     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4531     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4532     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4533   }
4534 
4535   DO_SIGNAL_CHECK(SR_signum);
4536   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4537 }
4538 
4539 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4540 
4541 static os_sigaction_t os_sigaction = NULL;
4542 
4543 void os::Bsd::check_signal_handler(int sig) {
4544   char buf[O_BUFLEN];
4545   address jvmHandler = NULL;
4546 
4547 
4548   struct sigaction act;
4549   if (os_sigaction == NULL) {
4550     // only trust the default sigaction, in case it has been interposed
4551     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4552     if (os_sigaction == NULL) return;
4553   }
4554 
4555   os_sigaction(sig, (struct sigaction*)NULL, &act);
4556 
4557 
4558   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4559 
4560   address thisHandler = (act.sa_flags & SA_SIGINFO)
4561     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4562     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4563 
4564 
4565   switch(sig) {
4566   case SIGSEGV:
4567   case SIGBUS:
4568   case SIGFPE:
4569   case SIGPIPE:
4570   case SIGILL:
4571   case SIGXFSZ:
4572     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4573     break;
4574 
4575   case SHUTDOWN1_SIGNAL:
4576   case SHUTDOWN2_SIGNAL:
4577   case SHUTDOWN3_SIGNAL:
4578   case BREAK_SIGNAL:
4579     jvmHandler = (address)user_handler();
4580     break;
4581 
4582   case INTERRUPT_SIGNAL:
4583     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4584     break;
4585 
4586   default:
4587     if (sig == SR_signum) {
4588       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4589     } else {
4590       return;
4591     }
4592     break;
4593   }
4594 
4595   if (thisHandler != jvmHandler) {
4596     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4597     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4598     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4599     // No need to check this sig any longer
4600     sigaddset(&check_signal_done, sig);
4601   } else if(os::Bsd::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Bsd::get_our_sigflags(sig)) {
4602     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4603     tty->print("expected:" PTR32_FORMAT, os::Bsd::get_our_sigflags(sig));
4604     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4605     // No need to check this sig any longer
4606     sigaddset(&check_signal_done, sig);
4607   }
4608 
4609   // Dump all the signal
4610   if (sigismember(&check_signal_done, sig)) {
4611     print_signal_handlers(tty, buf, O_BUFLEN);
4612   }
4613 }
4614 
4615 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4616 
4617 extern bool signal_name(int signo, char* buf, size_t len);
4618 
4619 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4620   if (0 < exception_code && exception_code <= SIGRTMAX) {
4621     // signal
4622     if (!signal_name(exception_code, buf, size)) {
4623       jio_snprintf(buf, size, "SIG%d", exception_code);
4624     }
4625     return buf;
4626   } else {
4627     return NULL;
4628   }
4629 }
4630 
4631 // this is called _before_ the most of global arguments have been parsed
4632 void os::init(void) {
4633   char dummy;   /* used to get a guess on initial stack address */
4634 //  first_hrtime = gethrtime();
4635 
4636   // With BsdThreads the JavaMain thread pid (primordial thread)
4637   // is different than the pid of the java launcher thread.
4638   // So, on Bsd, the launcher thread pid is passed to the VM
4639   // via the sun.java.launcher.pid property.
4640   // Use this property instead of getpid() if it was correctly passed.
4641   // See bug 6351349.
4642   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4643 
4644   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4645 
4646   clock_tics_per_sec = CLK_TCK;
4647 
4648   init_random(1234567);
4649 
4650   ThreadCritical::initialize();
4651 
4652   Bsd::set_page_size(getpagesize());
4653   if (Bsd::page_size() == -1) {
4654     fatal(err_msg("os_bsd.cpp: os::init: sysconf failed (%s)",
4655                   strerror(errno)));
4656   }
4657   init_page_sizes((size_t) Bsd::page_size());
4658 
4659   Bsd::initialize_system_info();
4660 
4661   // main_thread points to the aboriginal thread
4662   Bsd::_main_thread = pthread_self();
4663 
4664   Bsd::clock_init();
4665   initial_time_count = os::elapsed_counter();
4666 
4667 #ifdef __APPLE__
4668   // XXXDARWIN
4669   // Work around the unaligned VM callbacks in hotspot's
4670   // sharedRuntime. The callbacks don't use SSE2 instructions, and work on
4671   // Linux, Solaris, and FreeBSD. On Mac OS X, dyld (rightly so) enforces
4672   // alignment when doing symbol lookup. To work around this, we force early
4673   // binding of all symbols now, thus binding when alignment is known-good.
4674   _dyld_bind_fully_image_containing_address((const void *) &os::init);
4675 #endif
4676 }
4677 
4678 // To install functions for atexit system call
4679 extern "C" {
4680   static void perfMemory_exit_helper() {
4681     perfMemory_exit();
4682   }
4683 }
4684 
4685 // this is called _after_ the global arguments have been parsed
4686 jint os::init_2(void)
4687 {
4688 #ifndef _ALLBSD_SOURCE
4689   Bsd::fast_thread_clock_init();
4690 #endif
4691 
4692   // Allocate a single page and mark it as readable for safepoint polling
4693   address polling_page = (address) ::mmap(NULL, Bsd::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4694   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4695 
4696   os::set_polling_page( polling_page );
4697 
4698 #ifndef PRODUCT
4699   if(Verbose && PrintMiscellaneous)
4700     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4701 #endif
4702 
4703   if (!UseMembar) {
4704     address mem_serialize_page = (address) ::mmap(NULL, Bsd::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4705     guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4706     os::set_memory_serialize_page( mem_serialize_page );
4707 
4708 #ifndef PRODUCT
4709     if(Verbose && PrintMiscellaneous)
4710       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4711 #endif
4712   }
4713 
4714   os::large_page_init();
4715 
4716   // initialize suspend/resume support - must do this before signal_sets_init()
4717   if (SR_initialize() != 0) {
4718     perror("SR_initialize failed");
4719     return JNI_ERR;
4720   }
4721 
4722   Bsd::signal_sets_init();
4723   Bsd::install_signal_handlers();
4724 
4725   // Check minimum allowable stack size for thread creation and to initialize
4726   // the java system classes, including StackOverflowError - depends on page
4727   // size.  Add a page for compiler2 recursion in main thread.
4728   // Add in 2*BytesPerWord times page size to account for VM stack during
4729   // class initialization depending on 32 or 64 bit VM.
4730   os::Bsd::min_stack_allowed = MAX2(os::Bsd::min_stack_allowed,
4731             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4732                     2*BytesPerWord COMPILER2_PRESENT(+1)) * Bsd::page_size());
4733 
4734   size_t threadStackSizeInBytes = ThreadStackSize * K;
4735   if (threadStackSizeInBytes != 0 &&
4736       threadStackSizeInBytes < os::Bsd::min_stack_allowed) {
4737         tty->print_cr("\nThe stack size specified is too small, "
4738                       "Specify at least %dk",
4739                       os::Bsd::min_stack_allowed/ K);
4740         return JNI_ERR;
4741   }
4742 
4743   // Make the stack size a multiple of the page size so that
4744   // the yellow/red zones can be guarded.
4745   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4746         vm_page_size()));
4747 
4748 #ifndef _ALLBSD_SOURCE
4749   Bsd::capture_initial_stack(JavaThread::stack_size_at_create());
4750 
4751   Bsd::libpthread_init();
4752   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4753      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4754           Bsd::glibc_version(), Bsd::libpthread_version(),
4755           Bsd::is_floating_stack() ? "floating stack" : "fixed stack");
4756   }
4757 
4758   if (UseNUMA) {
4759     if (!Bsd::libnuma_init()) {
4760       UseNUMA = false;
4761     } else {
4762       if ((Bsd::numa_max_node() < 1)) {
4763         // There's only one node(they start from 0), disable NUMA.
4764         UseNUMA = false;
4765       }
4766     }
4767     // With SHM large pages we cannot uncommit a page, so there's not way
4768     // we can make the adaptive lgrp chunk resizing work. If the user specified
4769     // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4770     // disable adaptive resizing.
4771     if (UseNUMA && UseLargePages && UseSHM) {
4772       if (!FLAG_IS_DEFAULT(UseNUMA)) {
4773         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4774           UseLargePages = false;
4775         } else {
4776           warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4777           UseAdaptiveSizePolicy = false;
4778           UseAdaptiveNUMAChunkSizing = false;
4779         }
4780       } else {
4781         UseNUMA = false;
4782       }
4783     }
4784     if (!UseNUMA && ForceNUMA) {
4785       UseNUMA = true;
4786     }
4787   }
4788 #endif
4789 
4790   if (MaxFDLimit) {
4791     // set the number of file descriptors to max. print out error
4792     // if getrlimit/setrlimit fails but continue regardless.
4793     struct rlimit nbr_files;
4794     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4795     if (status != 0) {
4796       if (PrintMiscellaneous && (Verbose || WizardMode))
4797         perror("os::init_2 getrlimit failed");
4798     } else {
4799       nbr_files.rlim_cur = nbr_files.rlim_max;
4800 
4801 #ifdef __APPLE__
4802       // Darwin returns RLIM_INFINITY for rlim_max, but fails with EINVAL if
4803       // you attempt to use RLIM_INFINITY. As per setrlimit(2), OPEN_MAX must
4804       // be used instead
4805       nbr_files.rlim_cur = MIN(OPEN_MAX, nbr_files.rlim_cur);
4806 #endif
4807 
4808       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4809       if (status != 0) {
4810         if (PrintMiscellaneous && (Verbose || WizardMode))
4811           perror("os::init_2 setrlimit failed");
4812       }
4813     }
4814   }
4815 
4816 #ifndef _ALLBSD_SOURCE
4817   // Initialize lock used to serialize thread creation (see os::create_thread)
4818   Bsd::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4819 #endif
4820 
4821   // at-exit methods are called in the reverse order of their registration.
4822   // atexit functions are called on return from main or as a result of a
4823   // call to exit(3C). There can be only 32 of these functions registered
4824   // and atexit() does not set errno.
4825 
4826   if (PerfAllowAtExitRegistration) {
4827     // only register atexit functions if PerfAllowAtExitRegistration is set.
4828     // atexit functions can be delayed until process exit time, which
4829     // can be problematic for embedded VM situations. Embedded VMs should
4830     // call DestroyJavaVM() to assure that VM resources are released.
4831 
4832     // note: perfMemory_exit_helper atexit function may be removed in
4833     // the future if the appropriate cleanup code can be added to the
4834     // VM_Exit VMOperation's doit method.
4835     if (atexit(perfMemory_exit_helper) != 0) {
4836       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4837     }
4838   }
4839 
4840   // initialize thread priority policy
4841   prio_init();
4842 
4843 #ifdef __APPLE__
4844   // dynamically link to objective c gc registration
4845   void *handleLibObjc = dlopen(OBJC_LIB, RTLD_LAZY);
4846   if (handleLibObjc != NULL) {
4847     objc_registerThreadWithCollectorFunction = (objc_registerThreadWithCollector_t) dlsym(handleLibObjc, OBJC_GCREGISTER);
4848   }
4849 #endif
4850 
4851   return JNI_OK;
4852 }
4853 
4854 // this is called at the end of vm_initialization
4855 void os::init_3(void) { }
4856 
4857 // Mark the polling page as unreadable
4858 void os::make_polling_page_unreadable(void) {
4859   if( !guard_memory((char*)_polling_page, Bsd::page_size()) )
4860     fatal("Could not disable polling page");
4861 };
4862 
4863 // Mark the polling page as readable
4864 void os::make_polling_page_readable(void) {
4865   if( !bsd_mprotect((char *)_polling_page, Bsd::page_size(), PROT_READ)) {
4866     fatal("Could not enable polling page");
4867   }
4868 };
4869 
4870 int os::active_processor_count() {
4871 #ifdef _ALLBSD_SOURCE
4872   return _processor_count;
4873 #else
4874   // Bsd doesn't yet have a (official) notion of processor sets,
4875   // so just return the number of online processors.
4876   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4877   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4878   return online_cpus;
4879 #endif
4880 }
4881 
4882 void os::set_native_thread_name(const char *name) {
4883 #if defined(__APPLE__) && MAC_OS_X_VERSION_MIN_REQUIRED > MAC_OS_X_VERSION_10_5
4884   // This is only supported in Snow Leopard and beyond
4885   if (name != NULL) {
4886     // Add a "Java: " prefix to the name
4887     char buf[MAXTHREADNAMESIZE];
4888     snprintf(buf, sizeof(buf), "Java: %s", name);
4889     pthread_setname_np(buf);
4890   }
4891 #endif
4892 }
4893 
4894 bool os::distribute_processes(uint length, uint* distribution) {
4895   // Not yet implemented.
4896   return false;
4897 }
4898 
4899 bool os::bind_to_processor(uint processor_id) {
4900   // Not yet implemented.
4901   return false;
4902 }
4903 
4904 void os::SuspendedThreadTask::internal_do_task() {
4905   if (do_suspend(_thread->osthread())) {
4906     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4907     do_task(context);
4908     do_resume(_thread->osthread());
4909   }
4910 }
4911 
4912 ///
4913 class PcFetcher : public os::SuspendedThreadTask {
4914 public:
4915   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4916   ExtendedPC result();
4917 protected:
4918   void do_task(const os::SuspendedThreadTaskContext& context);
4919 private:
4920   ExtendedPC _epc;
4921 };
4922 
4923 ExtendedPC PcFetcher::result() {
4924   guarantee(is_done(), "task is not done yet.");
4925   return _epc;
4926 }
4927 
4928 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4929   Thread* thread = context.thread();
4930   OSThread* osthread = thread->osthread();
4931   if (osthread->ucontext() != NULL) {
4932     _epc = os::Bsd::ucontext_get_pc((ucontext_t *) context.ucontext());
4933   } else {
4934     // NULL context is unexpected, double-check this is the VMThread
4935     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4936   }
4937 }
4938 
4939 // Suspends the target using the signal mechanism and then grabs the PC before
4940 // resuming the target. Used by the flat-profiler only
4941 ExtendedPC os::get_thread_pc(Thread* thread) {
4942   // Make sure that it is called by the watcher for the VMThread
4943   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4944   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4945 
4946   PcFetcher fetcher(thread);
4947   fetcher.run();
4948   return fetcher.result();
4949 }
4950 
4951 int os::Bsd::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4952 {
4953 #ifdef _ALLBSD_SOURCE
4954   return pthread_cond_timedwait(_cond, _mutex, _abstime);
4955 #else
4956    if (is_NPTL()) {
4957       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4958    } else {
4959 #ifndef IA64
4960       // 6292965: BsdThreads pthread_cond_timedwait() resets FPU control
4961       // word back to default 64bit precision if condvar is signaled. Java
4962       // wants 53bit precision.  Save and restore current value.
4963       int fpu = get_fpu_control_word();
4964 #endif // IA64
4965       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4966 #ifndef IA64
4967       set_fpu_control_word(fpu);
4968 #endif // IA64
4969       return status;
4970    }
4971 #endif
4972 }
4973 
4974 ////////////////////////////////////////////////////////////////////////////////
4975 // debug support
4976 
4977 static address same_page(address x, address y) {
4978   int page_bits = -os::vm_page_size();
4979   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4980     return x;
4981   else if (x > y)
4982     return (address)(intptr_t(y) | ~page_bits) + 1;
4983   else
4984     return (address)(intptr_t(y) & page_bits);
4985 }
4986 
4987 bool os::find(address addr, outputStream* st) {
4988   Dl_info dlinfo;
4989   memset(&dlinfo, 0, sizeof(dlinfo));
4990   if (dladdr(addr, &dlinfo)) {
4991     st->print(PTR_FORMAT ": ", addr);
4992     if (dlinfo.dli_sname != NULL) {
4993       st->print("%s+%#x", dlinfo.dli_sname,
4994                  addr - (intptr_t)dlinfo.dli_saddr);
4995     } else if (dlinfo.dli_fname) {
4996       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4997     } else {
4998       st->print("<absolute address>");
4999     }
5000     if (dlinfo.dli_fname) {
5001       st->print(" in %s", dlinfo.dli_fname);
5002     }
5003     if (dlinfo.dli_fbase) {
5004       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5005     }
5006     st->cr();
5007 
5008     if (Verbose) {
5009       // decode some bytes around the PC
5010       address begin = same_page(addr-40, addr);
5011       address end   = same_page(addr+40, addr);
5012       address       lowest = (address) dlinfo.dli_sname;
5013       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5014       if (begin < lowest)  begin = lowest;
5015       Dl_info dlinfo2;
5016       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
5017           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5018         end = (address) dlinfo2.dli_saddr;
5019       Disassembler::decode(begin, end, st);
5020     }
5021     return true;
5022   }
5023   return false;
5024 }
5025 
5026 ////////////////////////////////////////////////////////////////////////////////
5027 // misc
5028 
5029 // This does not do anything on Bsd. This is basically a hook for being
5030 // able to use structured exception handling (thread-local exception filters)
5031 // on, e.g., Win32.
5032 void
5033 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5034                          JavaCallArguments* args, Thread* thread) {
5035   f(value, method, args, thread);
5036 }
5037 
5038 void os::print_statistics() {
5039 }
5040 
5041 int os::message_box(const char* title, const char* message) {
5042   int i;
5043   fdStream err(defaultStream::error_fd());
5044   for (i = 0; i < 78; i++) err.print_raw("=");
5045   err.cr();
5046   err.print_raw_cr(title);
5047   for (i = 0; i < 78; i++) err.print_raw("-");
5048   err.cr();
5049   err.print_raw_cr(message);
5050   for (i = 0; i < 78; i++) err.print_raw("=");
5051   err.cr();
5052 
5053   char buf[16];
5054   // Prevent process from exiting upon "read error" without consuming all CPU
5055   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5056 
5057   return buf[0] == 'y' || buf[0] == 'Y';
5058 }
5059 
5060 int os::stat(const char *path, struct stat *sbuf) {
5061   char pathbuf[MAX_PATH];
5062   if (strlen(path) > MAX_PATH - 1) {
5063     errno = ENAMETOOLONG;
5064     return -1;
5065   }
5066   os::native_path(strcpy(pathbuf, path));
5067   return ::stat(pathbuf, sbuf);
5068 }
5069 
5070 bool os::check_heap(bool force) {
5071   return true;
5072 }
5073 
5074 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5075   return ::vsnprintf(buf, count, format, args);
5076 }
5077 
5078 // Is a (classpath) directory empty?
5079 bool os::dir_is_empty(const char* path) {
5080   DIR *dir = NULL;
5081   struct dirent *ptr;
5082 
5083   dir = opendir(path);
5084   if (dir == NULL) return true;
5085 
5086   /* Scan the directory */
5087   bool result = true;
5088   char buf[sizeof(struct dirent) + MAX_PATH];
5089   while (result && (ptr = ::readdir(dir)) != NULL) {
5090     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5091       result = false;
5092     }
5093   }
5094   closedir(dir);
5095   return result;
5096 }
5097 
5098 // This code originates from JDK's sysOpen and open64_w
5099 // from src/solaris/hpi/src/system_md.c
5100 
5101 #ifndef O_DELETE
5102 #define O_DELETE 0x10000
5103 #endif
5104 
5105 // Open a file. Unlink the file immediately after open returns
5106 // if the specified oflag has the O_DELETE flag set.
5107 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5108 
5109 int os::open(const char *path, int oflag, int mode) {
5110 
5111   if (strlen(path) > MAX_PATH - 1) {
5112     errno = ENAMETOOLONG;
5113     return -1;
5114   }
5115   int fd;
5116   int o_delete = (oflag & O_DELETE);
5117   oflag = oflag & ~O_DELETE;
5118 
5119   fd = ::open(path, oflag, mode);
5120   if (fd == -1) return -1;
5121 
5122   //If the open succeeded, the file might still be a directory
5123   {
5124     struct stat buf;
5125     int ret = ::fstat(fd, &buf);
5126     int st_mode = buf.st_mode;
5127 
5128     if (ret != -1) {
5129       if ((st_mode & S_IFMT) == S_IFDIR) {
5130         errno = EISDIR;
5131         ::close(fd);
5132         return -1;
5133       }
5134     } else {
5135       ::close(fd);
5136       return -1;
5137     }
5138   }
5139 
5140     /*
5141      * All file descriptors that are opened in the JVM and not
5142      * specifically destined for a subprocess should have the
5143      * close-on-exec flag set.  If we don't set it, then careless 3rd
5144      * party native code might fork and exec without closing all
5145      * appropriate file descriptors (e.g. as we do in closeDescriptors in
5146      * UNIXProcess.c), and this in turn might:
5147      *
5148      * - cause end-of-file to fail to be detected on some file
5149      *   descriptors, resulting in mysterious hangs, or
5150      *
5151      * - might cause an fopen in the subprocess to fail on a system
5152      *   suffering from bug 1085341.
5153      *
5154      * (Yes, the default setting of the close-on-exec flag is a Unix
5155      * design flaw)
5156      *
5157      * See:
5158      * 1085341: 32-bit stdio routines should support file descriptors >255
5159      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5160      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5161      */
5162 #ifdef FD_CLOEXEC
5163     {
5164         int flags = ::fcntl(fd, F_GETFD);
5165         if (flags != -1)
5166             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5167     }
5168 #endif
5169 
5170   if (o_delete != 0) {
5171     ::unlink(path);
5172   }
5173   return fd;
5174 }
5175 
5176 
5177 // create binary file, rewriting existing file if required
5178 int os::create_binary_file(const char* path, bool rewrite_existing) {
5179   int oflags = O_WRONLY | O_CREAT;
5180   if (!rewrite_existing) {
5181     oflags |= O_EXCL;
5182   }
5183   return ::open(path, oflags, S_IREAD | S_IWRITE);
5184 }
5185 
5186 // return current position of file pointer
5187 jlong os::current_file_offset(int fd) {
5188   return (jlong)::lseek(fd, (off_t)0, SEEK_CUR);
5189 }
5190 
5191 // move file pointer to the specified offset
5192 jlong os::seek_to_file_offset(int fd, jlong offset) {
5193   return (jlong)::lseek(fd, (off_t)offset, SEEK_SET);
5194 }
5195 
5196 // This code originates from JDK's sysAvailable
5197 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5198 
5199 int os::available(int fd, jlong *bytes) {
5200   jlong cur, end;
5201   int mode;
5202   struct stat buf;
5203 
5204   if (::fstat(fd, &buf) >= 0) {
5205     mode = buf.st_mode;
5206     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5207       /*
5208       * XXX: is the following call interruptible? If so, this might
5209       * need to go through the INTERRUPT_IO() wrapper as for other
5210       * blocking, interruptible calls in this file.
5211       */
5212       int n;
5213       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5214         *bytes = n;
5215         return 1;
5216       }
5217     }
5218   }
5219   if ((cur = ::lseek(fd, 0L, SEEK_CUR)) == -1) {
5220     return 0;
5221   } else if ((end = ::lseek(fd, 0L, SEEK_END)) == -1) {
5222     return 0;
5223   } else if (::lseek(fd, cur, SEEK_SET) == -1) {
5224     return 0;
5225   }
5226   *bytes = end - cur;
5227   return 1;
5228 }
5229 
5230 int os::socket_available(int fd, jint *pbytes) {
5231    if (fd < 0)
5232      return OS_OK;
5233 
5234    int ret;
5235 
5236    RESTARTABLE(::ioctl(fd, FIONREAD, pbytes), ret);
5237 
5238    //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5239    // is expected to return 0 on failure and 1 on success to the jdk.
5240 
5241    return (ret == OS_ERR) ? 0 : 1;
5242 }
5243 
5244 // Map a block of memory.
5245 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5246                      char *addr, size_t bytes, bool read_only,
5247                      bool allow_exec) {
5248   int prot;
5249   int flags;
5250 
5251   if (read_only) {
5252     prot = PROT_READ;
5253     flags = MAP_SHARED;
5254   } else {
5255     prot = PROT_READ | PROT_WRITE;
5256     flags = MAP_PRIVATE;
5257   }
5258 
5259   if (allow_exec) {
5260     prot |= PROT_EXEC;
5261   }
5262 
5263   if (addr != NULL) {
5264     flags |= MAP_FIXED;
5265   }
5266 
5267   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5268                                      fd, file_offset);
5269   if (mapped_address == MAP_FAILED) {
5270     return NULL;
5271   }
5272   return mapped_address;
5273 }
5274 
5275 
5276 // Remap a block of memory.
5277 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5278                        char *addr, size_t bytes, bool read_only,
5279                        bool allow_exec) {
5280   // same as map_memory() on this OS
5281   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5282                         allow_exec);
5283 }
5284 
5285 
5286 // Unmap a block of memory.
5287 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5288   return munmap(addr, bytes) == 0;
5289 }
5290 
5291 #ifndef _ALLBSD_SOURCE
5292 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5293 
5294 static clockid_t thread_cpu_clockid(Thread* thread) {
5295   pthread_t tid = thread->osthread()->pthread_id();
5296   clockid_t clockid;
5297 
5298   // Get thread clockid
5299   int rc = os::Bsd::pthread_getcpuclockid(tid, &clockid);
5300   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5301   return clockid;
5302 }
5303 #endif
5304 
5305 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5306 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5307 // of a thread.
5308 //
5309 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5310 // the fast estimate available on the platform.
5311 
5312 jlong os::current_thread_cpu_time() {
5313 #ifdef __APPLE__
5314   return os::thread_cpu_time(Thread::current(), true /* user + sys */);
5315 #elif !defined(_ALLBSD_SOURCE)
5316   if (os::Bsd::supports_fast_thread_cpu_time()) {
5317     return os::Bsd::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5318   } else {
5319     // return user + sys since the cost is the same
5320     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5321   }
5322 #endif
5323 }
5324 
5325 jlong os::thread_cpu_time(Thread* thread) {
5326 #ifndef _ALLBSD_SOURCE
5327   // consistent with what current_thread_cpu_time() returns
5328   if (os::Bsd::supports_fast_thread_cpu_time()) {
5329     return os::Bsd::fast_thread_cpu_time(thread_cpu_clockid(thread));
5330   } else {
5331     return slow_thread_cpu_time(thread, true /* user + sys */);
5332   }
5333 #endif
5334 }
5335 
5336 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5337 #ifdef __APPLE__
5338   return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5339 #elif !defined(_ALLBSD_SOURCE)
5340   if (user_sys_cpu_time && os::Bsd::supports_fast_thread_cpu_time()) {
5341     return os::Bsd::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5342   } else {
5343     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5344   }
5345 #endif
5346 }
5347 
5348 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5349 #ifdef __APPLE__
5350   struct thread_basic_info tinfo;
5351   mach_msg_type_number_t tcount = THREAD_INFO_MAX;
5352   kern_return_t kr;
5353   thread_t mach_thread;
5354 
5355   mach_thread = thread->osthread()->thread_id();
5356   kr = thread_info(mach_thread, THREAD_BASIC_INFO, (thread_info_t)&tinfo, &tcount);
5357   if (kr != KERN_SUCCESS)
5358     return -1;
5359 
5360   if (user_sys_cpu_time) {
5361     jlong nanos;
5362     nanos = ((jlong) tinfo.system_time.seconds + tinfo.user_time.seconds) * (jlong)1000000000;
5363     nanos += ((jlong) tinfo.system_time.microseconds + (jlong) tinfo.user_time.microseconds) * (jlong)1000;
5364     return nanos;
5365   } else {
5366     return ((jlong)tinfo.user_time.seconds * 1000000000) + ((jlong)tinfo.user_time.microseconds * (jlong)1000);
5367   }
5368 #elif !defined(_ALLBSD_SOURCE)
5369   if (user_sys_cpu_time && os::Bsd::supports_fast_thread_cpu_time()) {
5370     return os::Bsd::fast_thread_cpu_time(thread_cpu_clockid(thread));
5371   } else {
5372     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5373   }
5374 #endif
5375 }
5376 
5377 #ifndef _ALLBSD_SOURCE
5378 //
5379 //  -1 on error.
5380 //
5381 
5382 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5383   static bool proc_pid_cpu_avail = true;
5384   static bool proc_task_unchecked = true;
5385   static const char *proc_stat_path = "/proc/%d/stat";
5386   pid_t  tid = thread->osthread()->thread_id();
5387   int i;
5388   char *s;
5389   char stat[2048];
5390   int statlen;
5391   char proc_name[64];
5392   int count;
5393   long sys_time, user_time;
5394   char string[64];
5395   char cdummy;
5396   int idummy;
5397   long ldummy;
5398   FILE *fp;
5399 
5400   // We first try accessing /proc/<pid>/cpu since this is faster to
5401   // process.  If this file is not present (bsd kernels 2.5 and above)
5402   // then we open /proc/<pid>/stat.
5403   if ( proc_pid_cpu_avail ) {
5404     sprintf(proc_name, "/proc/%d/cpu", tid);
5405     fp =  fopen(proc_name, "r");
5406     if ( fp != NULL ) {
5407       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
5408       fclose(fp);
5409       if ( count != 3 ) return -1;
5410 
5411       if (user_sys_cpu_time) {
5412         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5413       } else {
5414         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5415       }
5416     }
5417     else proc_pid_cpu_avail = false;
5418   }
5419 
5420   // The /proc/<tid>/stat aggregates per-process usage on
5421   // new Bsd kernels 2.6+ where NPTL is supported.
5422   // The /proc/self/task/<tid>/stat still has the per-thread usage.
5423   // See bug 6328462.
5424   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
5425   // and possibly in some other cases, so we check its availability.
5426   if (proc_task_unchecked && os::Bsd::is_NPTL()) {
5427     // This is executed only once
5428     proc_task_unchecked = false;
5429     fp = fopen("/proc/self/task", "r");
5430     if (fp != NULL) {
5431       proc_stat_path = "/proc/self/task/%d/stat";
5432       fclose(fp);
5433     }
5434   }
5435 
5436   sprintf(proc_name, proc_stat_path, tid);
5437   fp = fopen(proc_name, "r");
5438   if ( fp == NULL ) return -1;
5439   statlen = fread(stat, 1, 2047, fp);
5440   stat[statlen] = '\0';
5441   fclose(fp);
5442 
5443   // Skip pid and the command string. Note that we could be dealing with
5444   // weird command names, e.g. user could decide to rename java launcher
5445   // to "java 1.4.2 :)", then the stat file would look like
5446   //                1234 (java 1.4.2 :)) R ... ...
5447   // We don't really need to know the command string, just find the last
5448   // occurrence of ")" and then start parsing from there. See bug 4726580.
5449   s = strrchr(stat, ')');
5450   i = 0;
5451   if (s == NULL ) return -1;
5452 
5453   // Skip blank chars
5454   do s++; while (isspace(*s));
5455 
5456   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5457                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5458                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5459                  &user_time, &sys_time);
5460   if ( count != 13 ) return -1;
5461   if (user_sys_cpu_time) {
5462     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5463   } else {
5464     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5465   }
5466 }
5467 #endif
5468 
5469 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5470   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5471   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5472   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5473   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5474 }
5475 
5476 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5477   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5478   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5479   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5480   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5481 }
5482 
5483 bool os::is_thread_cpu_time_supported() {
5484 #ifdef __APPLE__
5485   return true;
5486 #elif defined(_ALLBSD_SOURCE)
5487   return false;
5488 #else
5489   return true;
5490 #endif
5491 }
5492 
5493 // System loadavg support.  Returns -1 if load average cannot be obtained.
5494 // Bsd doesn't yet have a (official) notion of processor sets,
5495 // so just return the system wide load average.
5496 int os::loadavg(double loadavg[], int nelem) {
5497   return ::getloadavg(loadavg, nelem);
5498 }
5499 
5500 void os::pause() {
5501   char filename[MAX_PATH];
5502   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5503     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5504   } else {
5505     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5506   }
5507 
5508   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5509   if (fd != -1) {
5510     struct stat buf;
5511     ::close(fd);
5512     while (::stat(filename, &buf) == 0) {
5513       (void)::poll(NULL, 0, 100);
5514     }
5515   } else {
5516     jio_fprintf(stderr,
5517       "Could not open pause file '%s', continuing immediately.\n", filename);
5518   }
5519 }
5520 
5521 
5522 // Refer to the comments in os_solaris.cpp park-unpark.
5523 //
5524 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5525 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5526 // For specifics regarding the bug see GLIBC BUGID 261237 :
5527 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5528 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5529 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5530 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5531 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5532 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5533 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5534 // of libpthread avoids the problem, but isn't practical.
5535 //
5536 // Possible remedies:
5537 //
5538 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5539 //      This is palliative and probabilistic, however.  If the thread is preempted
5540 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5541 //      than the minimum period may have passed, and the abstime may be stale (in the
5542 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5543 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5544 //
5545 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5546 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5547 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5548 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5549 //      thread.
5550 //
5551 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5552 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5553 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5554 //      This also works well.  In fact it avoids kernel-level scalability impediments
5555 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5556 //      timers in a graceful fashion.
5557 //
5558 // 4.   When the abstime value is in the past it appears that control returns
5559 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5560 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5561 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5562 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5563 //      It may be possible to avoid reinitialization by checking the return
5564 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5565 //      condvar we must establish the invariant that cond_signal() is only called
5566 //      within critical sections protected by the adjunct mutex.  This prevents
5567 //      cond_signal() from "seeing" a condvar that's in the midst of being
5568 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5569 //      desirable signal-after-unlock optimization that avoids futile context switching.
5570 //
5571 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5572 //      structure when a condvar is used or initialized.  cond_destroy()  would
5573 //      release the helper structure.  Our reinitialize-after-timedwait fix
5574 //      put excessive stress on malloc/free and locks protecting the c-heap.
5575 //
5576 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5577 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5578 // and only enabling the work-around for vulnerable environments.
5579 
5580 // utility to compute the abstime argument to timedwait:
5581 // millis is the relative timeout time
5582 // abstime will be the absolute timeout time
5583 // TODO: replace compute_abstime() with unpackTime()
5584 
5585 static struct timespec* compute_abstime(struct timespec* abstime, jlong millis) {
5586   if (millis < 0)  millis = 0;
5587   struct timeval now;
5588   int status = gettimeofday(&now, NULL);
5589   assert(status == 0, "gettimeofday");
5590   jlong seconds = millis / 1000;
5591   millis %= 1000;
5592   if (seconds > 50000000) { // see man cond_timedwait(3T)
5593     seconds = 50000000;
5594   }
5595   abstime->tv_sec = now.tv_sec  + seconds;
5596   long       usec = now.tv_usec + millis * 1000;
5597   if (usec >= 1000000) {
5598     abstime->tv_sec += 1;
5599     usec -= 1000000;
5600   }
5601   abstime->tv_nsec = usec * 1000;
5602   return abstime;
5603 }
5604 
5605 
5606 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5607 // Conceptually TryPark() should be equivalent to park(0).
5608 
5609 int os::PlatformEvent::TryPark() {
5610   for (;;) {
5611     const int v = _Event ;
5612     guarantee ((v == 0) || (v == 1), "invariant") ;
5613     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5614   }
5615 }
5616 
5617 void os::PlatformEvent::park() {       // AKA "down()"
5618   // Invariant: Only the thread associated with the Event/PlatformEvent
5619   // may call park().
5620   // TODO: assert that _Assoc != NULL or _Assoc == Self
5621   int v ;
5622   for (;;) {
5623       v = _Event ;
5624       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5625   }
5626   guarantee (v >= 0, "invariant") ;
5627   if (v == 0) {
5628      // Do this the hard way by blocking ...
5629      int status = pthread_mutex_lock(_mutex);
5630      assert_status(status == 0, status, "mutex_lock");
5631      guarantee (_nParked == 0, "invariant") ;
5632      ++ _nParked ;
5633      while (_Event < 0) {
5634         status = pthread_cond_wait(_cond, _mutex);
5635         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5636         // Treat this the same as if the wait was interrupted
5637         if (status == ETIMEDOUT) { status = EINTR; }
5638         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5639      }
5640      -- _nParked ;
5641 
5642     _Event = 0 ;
5643      status = pthread_mutex_unlock(_mutex);
5644      assert_status(status == 0, status, "mutex_unlock");
5645     // Paranoia to ensure our locked and lock-free paths interact
5646     // correctly with each other.
5647     OrderAccess::fence();
5648   }
5649   guarantee (_Event >= 0, "invariant") ;
5650 }
5651 
5652 int os::PlatformEvent::park(jlong millis) {
5653   guarantee (_nParked == 0, "invariant") ;
5654 
5655   int v ;
5656   for (;;) {
5657       v = _Event ;
5658       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5659   }
5660   guarantee (v >= 0, "invariant") ;
5661   if (v != 0) return OS_OK ;
5662 
5663   // We do this the hard way, by blocking the thread.
5664   // Consider enforcing a minimum timeout value.
5665   struct timespec abst;
5666   compute_abstime(&abst, millis);
5667 
5668   int ret = OS_TIMEOUT;
5669   int status = pthread_mutex_lock(_mutex);
5670   assert_status(status == 0, status, "mutex_lock");
5671   guarantee (_nParked == 0, "invariant") ;
5672   ++_nParked ;
5673 
5674   // Object.wait(timo) will return because of
5675   // (a) notification
5676   // (b) timeout
5677   // (c) thread.interrupt
5678   //
5679   // Thread.interrupt and object.notify{All} both call Event::set.
5680   // That is, we treat thread.interrupt as a special case of notification.
5681   // The underlying Solaris implementation, cond_timedwait, admits
5682   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5683   // JVM from making those visible to Java code.  As such, we must
5684   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5685   //
5686   // TODO: properly differentiate simultaneous notify+interrupt.
5687   // In that case, we should propagate the notify to another waiter.
5688 
5689   while (_Event < 0) {
5690     status = os::Bsd::safe_cond_timedwait(_cond, _mutex, &abst);
5691     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5692       pthread_cond_destroy (_cond);
5693       pthread_cond_init (_cond, NULL) ;
5694     }
5695     assert_status(status == 0 || status == EINTR ||
5696                   status == ETIMEDOUT,
5697                   status, "cond_timedwait");
5698     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5699     if (status == ETIMEDOUT) break ;
5700     // We consume and ignore EINTR and spurious wakeups.
5701   }
5702   --_nParked ;
5703   if (_Event >= 0) {
5704      ret = OS_OK;
5705   }
5706   _Event = 0 ;
5707   status = pthread_mutex_unlock(_mutex);
5708   assert_status(status == 0, status, "mutex_unlock");
5709   assert (_nParked == 0, "invariant") ;
5710   // Paranoia to ensure our locked and lock-free paths interact
5711   // correctly with each other.
5712   OrderAccess::fence();
5713   return ret;
5714 }
5715 
5716 void os::PlatformEvent::unpark() {
5717   // Transitions for _Event:
5718   //    0 :=> 1
5719   //    1 :=> 1
5720   //   -1 :=> either 0 or 1; must signal target thread
5721   //          That is, we can safely transition _Event from -1 to either
5722   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
5723   //          unpark() calls.
5724   // See also: "Semaphores in Plan 9" by Mullender & Cox
5725   //
5726   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5727   // that it will take two back-to-back park() calls for the owning
5728   // thread to block. This has the benefit of forcing a spurious return
5729   // from the first park() call after an unpark() call which will help
5730   // shake out uses of park() and unpark() without condition variables.
5731 
5732   if (Atomic::xchg(1, &_Event) >= 0) return;
5733 
5734   // Wait for the thread associated with the event to vacate
5735   int status = pthread_mutex_lock(_mutex);
5736   assert_status(status == 0, status, "mutex_lock");
5737   int AnyWaiters = _nParked;
5738   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5739   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5740     AnyWaiters = 0;
5741     pthread_cond_signal(_cond);
5742   }
5743   status = pthread_mutex_unlock(_mutex);
5744   assert_status(status == 0, status, "mutex_unlock");
5745   if (AnyWaiters != 0) {
5746     status = pthread_cond_signal(_cond);
5747     assert_status(status == 0, status, "cond_signal");
5748   }
5749 
5750   // Note that we signal() _after dropping the lock for "immortal" Events.
5751   // This is safe and avoids a common class of  futile wakeups.  In rare
5752   // circumstances this can cause a thread to return prematurely from
5753   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5754   // simply re-test the condition and re-park itself.
5755 }
5756 
5757 
5758 // JSR166
5759 // -------------------------------------------------------
5760 
5761 /*
5762  * The solaris and bsd implementations of park/unpark are fairly
5763  * conservative for now, but can be improved. They currently use a
5764  * mutex/condvar pair, plus a a count.
5765  * Park decrements count if > 0, else does a condvar wait.  Unpark
5766  * sets count to 1 and signals condvar.  Only one thread ever waits
5767  * on the condvar. Contention seen when trying to park implies that someone
5768  * is unparking you, so don't wait. And spurious returns are fine, so there
5769  * is no need to track notifications.
5770  */
5771 
5772 #define MAX_SECS 100000000
5773 /*
5774  * This code is common to bsd and solaris and will be moved to a
5775  * common place in dolphin.
5776  *
5777  * The passed in time value is either a relative time in nanoseconds
5778  * or an absolute time in milliseconds. Either way it has to be unpacked
5779  * into suitable seconds and nanoseconds components and stored in the
5780  * given timespec structure.
5781  * Given time is a 64-bit value and the time_t used in the timespec is only
5782  * a signed-32-bit value (except on 64-bit Bsd) we have to watch for
5783  * overflow if times way in the future are given. Further on Solaris versions
5784  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5785  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5786  * As it will be 28 years before "now + 100000000" will overflow we can
5787  * ignore overflow and just impose a hard-limit on seconds using the value
5788  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5789  * years from "now".
5790  */
5791 
5792 static void unpackTime(struct timespec* absTime, bool isAbsolute, jlong time) {
5793   assert (time > 0, "convertTime");
5794 
5795   struct timeval now;
5796   int status = gettimeofday(&now, NULL);
5797   assert(status == 0, "gettimeofday");
5798 
5799   time_t max_secs = now.tv_sec + MAX_SECS;
5800 
5801   if (isAbsolute) {
5802     jlong secs = time / 1000;
5803     if (secs > max_secs) {
5804       absTime->tv_sec = max_secs;
5805     }
5806     else {
5807       absTime->tv_sec = secs;
5808     }
5809     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5810   }
5811   else {
5812     jlong secs = time / NANOSECS_PER_SEC;
5813     if (secs >= MAX_SECS) {
5814       absTime->tv_sec = max_secs;
5815       absTime->tv_nsec = 0;
5816     }
5817     else {
5818       absTime->tv_sec = now.tv_sec + secs;
5819       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5820       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5821         absTime->tv_nsec -= NANOSECS_PER_SEC;
5822         ++absTime->tv_sec; // note: this must be <= max_secs
5823       }
5824     }
5825   }
5826   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5827   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5828   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5829   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5830 }
5831 
5832 void Parker::park(bool isAbsolute, jlong time) {
5833   // Ideally we'd do something useful while spinning, such
5834   // as calling unpackTime().
5835 
5836   // Optional fast-path check:
5837   // Return immediately if a permit is available.
5838   // We depend on Atomic::xchg() having full barrier semantics
5839   // since we are doing a lock-free update to _counter.
5840   if (Atomic::xchg(0, &_counter) > 0) return;
5841 
5842   Thread* thread = Thread::current();
5843   assert(thread->is_Java_thread(), "Must be JavaThread");
5844   JavaThread *jt = (JavaThread *)thread;
5845 
5846   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5847   // Check interrupt before trying to wait
5848   if (Thread::is_interrupted(thread, false)) {
5849     return;
5850   }
5851 
5852   // Next, demultiplex/decode time arguments
5853   struct timespec absTime;
5854   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5855     return;
5856   }
5857   if (time > 0) {
5858     unpackTime(&absTime, isAbsolute, time);
5859   }
5860 
5861 
5862   // Enter safepoint region
5863   // Beware of deadlocks such as 6317397.
5864   // The per-thread Parker:: mutex is a classic leaf-lock.
5865   // In particular a thread must never block on the Threads_lock while
5866   // holding the Parker:: mutex.  If safepoints are pending both the
5867   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5868   ThreadBlockInVM tbivm(jt);
5869 
5870   // Don't wait if cannot get lock since interference arises from
5871   // unblocking.  Also. check interrupt before trying wait
5872   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5873     return;
5874   }
5875 
5876   int status ;
5877   if (_counter > 0)  { // no wait needed
5878     _counter = 0;
5879     status = pthread_mutex_unlock(_mutex);
5880     assert (status == 0, "invariant") ;
5881     // Paranoia to ensure our locked and lock-free paths interact
5882     // correctly with each other and Java-level accesses.
5883     OrderAccess::fence();
5884     return;
5885   }
5886 
5887 #ifdef ASSERT
5888   // Don't catch signals while blocked; let the running threads have the signals.
5889   // (This allows a debugger to break into the running thread.)
5890   sigset_t oldsigs;
5891   sigset_t* allowdebug_blocked = os::Bsd::allowdebug_blocked_signals();
5892   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5893 #endif
5894 
5895   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5896   jt->set_suspend_equivalent();
5897   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5898 
5899   if (time == 0) {
5900     status = pthread_cond_wait (_cond, _mutex) ;
5901   } else {
5902     status = os::Bsd::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5903     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5904       pthread_cond_destroy (_cond) ;
5905       pthread_cond_init    (_cond, NULL);
5906     }
5907   }
5908   assert_status(status == 0 || status == EINTR ||
5909                 status == ETIMEDOUT,
5910                 status, "cond_timedwait");
5911 
5912 #ifdef ASSERT
5913   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5914 #endif
5915 
5916   _counter = 0 ;
5917   status = pthread_mutex_unlock(_mutex) ;
5918   assert_status(status == 0, status, "invariant") ;
5919   // Paranoia to ensure our locked and lock-free paths interact
5920   // correctly with each other and Java-level accesses.
5921   OrderAccess::fence();
5922 
5923   // If externally suspended while waiting, re-suspend
5924   if (jt->handle_special_suspend_equivalent_condition()) {
5925     jt->java_suspend_self();
5926   }
5927 }
5928 
5929 void Parker::unpark() {
5930   int s, status ;
5931   status = pthread_mutex_lock(_mutex);
5932   assert (status == 0, "invariant") ;
5933   s = _counter;
5934   _counter = 1;
5935   if (s < 1) {
5936      if (WorkAroundNPTLTimedWaitHang) {
5937         status = pthread_cond_signal (_cond) ;
5938         assert (status == 0, "invariant") ;
5939         status = pthread_mutex_unlock(_mutex);
5940         assert (status == 0, "invariant") ;
5941      } else {
5942         status = pthread_mutex_unlock(_mutex);
5943         assert (status == 0, "invariant") ;
5944         status = pthread_cond_signal (_cond) ;
5945         assert (status == 0, "invariant") ;
5946      }
5947   } else {
5948     pthread_mutex_unlock(_mutex);
5949     assert (status == 0, "invariant") ;
5950   }
5951 }
5952 
5953 
5954 /* Darwin has no "environ" in a dynamic library. */
5955 #ifdef __APPLE__
5956 #include <crt_externs.h>
5957 #define environ (*_NSGetEnviron())
5958 #else
5959 extern char** environ;
5960 #endif
5961 
5962 // Run the specified command in a separate process. Return its exit value,
5963 // or -1 on failure (e.g. can't fork a new process).
5964 // Unlike system(), this function can be called from signal handler. It
5965 // doesn't block SIGINT et al.
5966 int os::fork_and_exec(char* cmd) {
5967   const char * argv[4] = {"sh", "-c", cmd, NULL};
5968 
5969   // fork() in BsdThreads/NPTL is not async-safe. It needs to run
5970   // pthread_atfork handlers and reset pthread library. All we need is a
5971   // separate process to execve. Make a direct syscall to fork process.
5972   // On IA64 there's no fork syscall, we have to use fork() and hope for
5973   // the best...
5974   pid_t pid = fork();
5975 
5976   if (pid < 0) {
5977     // fork failed
5978     return -1;
5979 
5980   } else if (pid == 0) {
5981     // child process
5982 
5983     // execve() in BsdThreads will call pthread_kill_other_threads_np()
5984     // first to kill every thread on the thread list. Because this list is
5985     // not reset by fork() (see notes above), execve() will instead kill
5986     // every thread in the parent process. We know this is the only thread
5987     // in the new process, so make a system call directly.
5988     // IA64 should use normal execve() from glibc to match the glibc fork()
5989     // above.
5990     execve("/bin/sh", (char* const*)argv, environ);
5991 
5992     // execve failed
5993     _exit(-1);
5994 
5995   } else  {
5996     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5997     // care about the actual exit code, for now.
5998 
5999     int status;
6000 
6001     // Wait for the child process to exit.  This returns immediately if
6002     // the child has already exited. */
6003     while (waitpid(pid, &status, 0) < 0) {
6004         switch (errno) {
6005         case ECHILD: return 0;
6006         case EINTR: break;
6007         default: return -1;
6008         }
6009     }
6010 
6011     if (WIFEXITED(status)) {
6012        // The child exited normally; get its exit code.
6013        return WEXITSTATUS(status);
6014     } else if (WIFSIGNALED(status)) {
6015        // The child exited because of a signal
6016        // The best value to return is 0x80 + signal number,
6017        // because that is what all Unix shells do, and because
6018        // it allows callers to distinguish between process exit and
6019        // process death by signal.
6020        return 0x80 + WTERMSIG(status);
6021     } else {
6022        // Unknown exit code; pass it through
6023        return status;
6024     }
6025   }
6026 }
6027 
6028 // is_headless_jre()
6029 //
6030 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6031 // in order to report if we are running in a headless jre
6032 //
6033 // Since JDK8 xawt/libmawt.so was moved into the same directory
6034 // as libawt.so, and renamed libawt_xawt.so
6035 //
6036 bool os::is_headless_jre() {
6037     struct stat statbuf;
6038     char buf[MAXPATHLEN];
6039     char libmawtpath[MAXPATHLEN];
6040     const char *xawtstr  = "/xawt/libmawt" JNI_LIB_SUFFIX;
6041     const char *new_xawtstr = "/libawt_xawt" JNI_LIB_SUFFIX;
6042     char *p;
6043 
6044     // Get path to libjvm.so
6045     os::jvm_path(buf, sizeof(buf));
6046 
6047     // Get rid of libjvm.so
6048     p = strrchr(buf, '/');
6049     if (p == NULL) return false;
6050     else *p = '\0';
6051 
6052     // Get rid of client or server
6053     p = strrchr(buf, '/');
6054     if (p == NULL) return false;
6055     else *p = '\0';
6056 
6057     // check xawt/libmawt.so
6058     strcpy(libmawtpath, buf);
6059     strcat(libmawtpath, xawtstr);
6060     if (::stat(libmawtpath, &statbuf) == 0) return false;
6061 
6062     // check libawt_xawt.so
6063     strcpy(libmawtpath, buf);
6064     strcat(libmawtpath, new_xawtstr);
6065     if (::stat(libmawtpath, &statbuf) == 0) return false;
6066 
6067     return true;
6068 }
6069 
6070 // Get the default path to the core file
6071 // Returns the length of the string
6072 int os::get_core_path(char* buffer, size_t bufferSize) {
6073   int n = jio_snprintf(buffer, bufferSize, "/cores");
6074 
6075   // Truncate if theoretical string was longer than bufferSize
6076   n = MIN2(n, (int)bufferSize);
6077 
6078   return n;
6079 }
6080 
6081 #ifndef PRODUCT
6082 void TestReserveMemorySpecial_test() {
6083   // No tests available for this platform
6084 }
6085 #endif
6086