1 /*
   2  * Copyright (c) 1999, 2011, 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/runtimeService.hpp"
  59 #include "thread_bsd.inline.hpp"
  60 #include "utilities/decoder.hpp"
  61 #include "utilities/defaultStream.hpp"
  62 #include "utilities/events.hpp"
  63 #include "utilities/growableArray.hpp"
  64 #include "utilities/vmError.hpp"
  65 #ifdef TARGET_ARCH_x86
  66 # include "assembler_x86.inline.hpp"
  67 # include "nativeInst_x86.hpp"
  68 #endif
  69 #ifdef TARGET_ARCH_sparc
  70 # include "assembler_sparc.inline.hpp"
  71 # include "nativeInst_sparc.hpp"
  72 #endif
  73 #ifdef TARGET_ARCH_zero
  74 # include "assembler_zero.inline.hpp"
  75 # include "nativeInst_zero.hpp"
  76 #endif
  77 #ifdef TARGET_ARCH_arm
  78 # include "assembler_arm.inline.hpp"
  79 # include "nativeInst_arm.hpp"
  80 #endif
  81 #ifdef TARGET_ARCH_ppc
  82 # include "assembler_ppc.inline.hpp"
  83 # include "nativeInst_ppc.hpp"
  84 #endif
  85 #ifdef COMPILER1
  86 #include "c1/c1_Runtime1.hpp"
  87 #endif
  88 #ifdef COMPILER2
  89 #include "opto/runtime.hpp"
  90 #endif
  91 
  92 // put OS-includes here
  93 # include <sys/types.h>
  94 # include <sys/mman.h>
  95 # include <sys/stat.h>
  96 # include <sys/select.h>
  97 # include <pthread.h>
  98 # include <signal.h>
  99 # include <errno.h>
 100 # include <dlfcn.h>
 101 # include <stdio.h>
 102 # include <unistd.h>
 103 # include <sys/resource.h>
 104 # include <pthread.h>
 105 # include <sys/stat.h>
 106 # include <sys/time.h>
 107 # include <sys/times.h>
 108 # include <sys/utsname.h>
 109 # include <sys/socket.h>
 110 # include <sys/wait.h>
 111 # include <time.h>
 112 # include <pwd.h>
 113 # include <poll.h>
 114 # include <semaphore.h>
 115 # include <fcntl.h>
 116 # include <string.h>
 117 #ifdef _ALLBSD_SOURCE
 118 # include <sys/param.h>
 119 # include <sys/sysctl.h>
 120 #else
 121 # include <syscall.h>
 122 # include <sys/sysinfo.h>
 123 # include <gnu/libc-version.h>
 124 #endif
 125 # include <sys/ipc.h>
 126 # include <sys/shm.h>
 127 #ifndef __APPLE__
 128 # include <link.h>
 129 #endif
 130 # include <stdint.h>
 131 # include <inttypes.h>
 132 # include <sys/ioctl.h>
 133 
 134 #if defined(__FreeBSD__) || defined(__NetBSD__)
 135 # include <elf.h>
 136 #endif
 137 
 138 #ifdef __APPLE__
 139 # include <mach/mach.h> // semaphore_* API
 140 # include <mach-o/dyld.h>
 141 # include <sys/proc_info.h>
 142 # include <objc/objc-auto.h>
 143 #endif
 144 
 145 #ifndef MAP_ANONYMOUS
 146 #define MAP_ANONYMOUS MAP_ANON
 147 #endif
 148 
 149 #define MAX_PATH    (2 * K)
 150 
 151 // for timer info max values which include all bits
 152 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 153 
 154 #define LARGEPAGES_BIT (1 << 6)
 155 ////////////////////////////////////////////////////////////////////////////////
 156 // global variables
 157 julong os::Bsd::_physical_memory = 0;
 158 
 159 #ifndef _ALLBSD_SOURCE
 160 address   os::Bsd::_initial_thread_stack_bottom = NULL;
 161 uintptr_t os::Bsd::_initial_thread_stack_size   = 0;
 162 #endif
 163 
 164 int (*os::Bsd::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 165 #ifndef _ALLBSD_SOURCE
 166 int (*os::Bsd::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 167 Mutex* os::Bsd::_createThread_lock = NULL;
 168 #endif
 169 pthread_t os::Bsd::_main_thread;
 170 int os::Bsd::_page_size = -1;
 171 #ifndef _ALLBSD_SOURCE
 172 bool os::Bsd::_is_floating_stack = false;
 173 bool os::Bsd::_is_NPTL = false;
 174 bool os::Bsd::_supports_fast_thread_cpu_time = false;
 175 const char * os::Bsd::_glibc_version = NULL;
 176 const char * os::Bsd::_libpthread_version = NULL;
 177 #endif
 178 
 179 static jlong initial_time_count=0;
 180 
 181 static int clock_tics_per_sec = 100;
 182 
 183 // For diagnostics to print a message once. see run_periodic_checks
 184 static sigset_t check_signal_done;
 185 static bool check_signals = true;;
 186 
 187 static pid_t _initial_pid = 0;
 188 
 189 /* Signal number used to suspend/resume a thread */
 190 
 191 /* do not use any signal number less than SIGSEGV, see 4355769 */
 192 static int SR_signum = SIGUSR2;
 193 sigset_t SR_sigset;
 194 
 195 
 196 ////////////////////////////////////////////////////////////////////////////////
 197 // utility functions
 198 
 199 static int SR_initialize();
 200 static int SR_finalize();
 201 
 202 julong os::available_memory() {
 203   return Bsd::available_memory();
 204 }
 205 
 206 julong os::Bsd::available_memory() {
 207 #ifdef _ALLBSD_SOURCE
 208   // XXXBSD: this is just a stopgap implementation
 209   return physical_memory() >> 2;
 210 #else
 211   // values in struct sysinfo are "unsigned long"
 212   struct sysinfo si;
 213   sysinfo(&si);
 214 
 215   return (julong)si.freeram * si.mem_unit;
 216 #endif
 217 }
 218 
 219 julong os::physical_memory() {
 220   return Bsd::physical_memory();
 221 }
 222 
 223 julong os::allocatable_physical_memory(julong size) {
 224 #ifdef _LP64
 225   return size;
 226 #else
 227   julong result = MIN2(size, (julong)3800*M);
 228    if (!is_allocatable(result)) {
 229      // See comments under solaris for alignment considerations
 230      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 231      result =  MIN2(size, reasonable_size);
 232    }
 233    return result;
 234 #endif // _LP64
 235 }
 236 
 237 ////////////////////////////////////////////////////////////////////////////////
 238 // environment support
 239 
 240 bool os::getenv(const char* name, char* buf, int len) {
 241   const char* val = ::getenv(name);
 242   if (val != NULL && strlen(val) < (size_t)len) {
 243     strcpy(buf, val);
 244     return true;
 245   }
 246   if (len > 0) buf[0] = 0;  // return a null string
 247   return false;
 248 }
 249 
 250 
 251 // Return true if user is running as root.
 252 
 253 bool os::have_special_privileges() {
 254   static bool init = false;
 255   static bool privileges = false;
 256   if (!init) {
 257     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 258     init = true;
 259   }
 260   return privileges;
 261 }
 262 
 263 
 264 #ifndef _ALLBSD_SOURCE
 265 #ifndef SYS_gettid
 266 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 267 #ifdef __ia64__
 268 #define SYS_gettid 1105
 269 #elif __i386__
 270 #define SYS_gettid 224
 271 #elif __amd64__
 272 #define SYS_gettid 186
 273 #elif __sparc__
 274 #define SYS_gettid 143
 275 #else
 276 #error define gettid for the arch
 277 #endif
 278 #endif
 279 #endif
 280 
 281 // Cpu architecture string
 282 #if   defined(ZERO)
 283 static char cpu_arch[] = ZERO_LIBARCH;
 284 #elif defined(IA64)
 285 static char cpu_arch[] = "ia64";
 286 #elif defined(IA32)
 287 static char cpu_arch[] = "i386";
 288 #elif defined(AMD64)
 289 static char cpu_arch[] = "amd64";
 290 #elif defined(ARM)
 291 static char cpu_arch[] = "arm";
 292 #elif defined(PPC)
 293 static char cpu_arch[] = "ppc";
 294 #elif defined(SPARC)
 295 #  ifdef _LP64
 296 static char cpu_arch[] = "sparcv9";
 297 #  else
 298 static char cpu_arch[] = "sparc";
 299 #  endif
 300 #else
 301 #error Add appropriate cpu_arch setting
 302 #endif
 303 
 304 
 305 #ifndef _ALLBSD_SOURCE
 306 // pid_t gettid()
 307 //
 308 // Returns the kernel thread id of the currently running thread. Kernel
 309 // thread id is used to access /proc.
 310 //
 311 // (Note that getpid() on BsdThreads returns kernel thread id too; but
 312 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 313 //
 314 pid_t os::Bsd::gettid() {
 315   int rslt = syscall(SYS_gettid);
 316   if (rslt == -1) {
 317      // old kernel, no NPTL support
 318      return getpid();
 319   } else {
 320      return (pid_t)rslt;
 321   }
 322 }
 323 
 324 // Most versions of bsd have a bug where the number of processors are
 325 // determined by looking at the /proc file system.  In a chroot environment,
 326 // the system call returns 1.  This causes the VM to act as if it is
 327 // a single processor and elide locking (see is_MP() call).
 328 static bool unsafe_chroot_detected = false;
 329 static const char *unstable_chroot_error = "/proc file system not found.\n"
 330                      "Java may be unstable running multithreaded in a chroot "
 331                      "environment on Bsd when /proc filesystem is not mounted.";
 332 #endif
 333 
 334 #ifdef _ALLBSD_SOURCE
 335 void os::Bsd::initialize_system_info() {
 336   int mib[2];
 337   size_t len;
 338   int cpu_val;
 339   u_long mem_val;
 340 
 341   /* get processors count via hw.ncpus sysctl */
 342   mib[0] = CTL_HW;
 343   mib[1] = HW_NCPU;
 344   len = sizeof(cpu_val);
 345   if (sysctl(mib, 2, &cpu_val, &len, NULL, 0) != -1 && cpu_val >= 1) {
 346        set_processor_count(cpu_val);
 347   }
 348   else {
 349        set_processor_count(1);   // fallback
 350   }
 351 
 352   /* get physical memory via hw.usermem sysctl (hw.usermem is used
 353    * instead of hw.physmem because we need size of allocatable memory
 354    */
 355   mib[0] = CTL_HW;
 356   mib[1] = HW_USERMEM;
 357   len = sizeof(mem_val);
 358   if (sysctl(mib, 2, &mem_val, &len, NULL, 0) != -1)
 359        _physical_memory = mem_val;
 360   else
 361        _physical_memory = 256*1024*1024;       // fallback (XXXBSD?)
 362 
 363 #ifdef __OpenBSD__
 364   {
 365        // limit _physical_memory memory view on OpenBSD since
 366        // datasize rlimit restricts us anyway.
 367        struct rlimit limits;
 368        getrlimit(RLIMIT_DATA, &limits);
 369        _physical_memory = MIN2(_physical_memory, (julong)limits.rlim_cur);
 370   }
 371 #endif
 372 }
 373 #else
 374 void os::Bsd::initialize_system_info() {
 375   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 376   if (processor_count() == 1) {
 377     pid_t pid = os::Bsd::gettid();
 378     char fname[32];
 379     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 380     FILE *fp = fopen(fname, "r");
 381     if (fp == NULL) {
 382       unsafe_chroot_detected = true;
 383     } else {
 384       fclose(fp);
 385     }
 386   }
 387   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 388   assert(processor_count() > 0, "bsd error");
 389 }
 390 #endif
 391 
 392 #ifdef __APPLE__
 393 static const char *get_home() {
 394   const char *home_dir = ::getenv("HOME");
 395   if ((home_dir == NULL) || (*home_dir == '\0')) {
 396     struct passwd *passwd_info = getpwuid(geteuid());
 397     if (passwd_info != NULL) {
 398       home_dir = passwd_info->pw_dir;
 399     }
 400   }
 401 
 402   return home_dir;
 403 }
 404 #endif
 405 
 406 void os::init_system_properties_values() {
 407 //  char arch[12];
 408 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 409 
 410   // The next steps are taken in the product version:
 411   //
 412   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 413   // This library should be located at:
 414   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 415   //
 416   // If "/jre/lib/" appears at the right place in the path, then we
 417   // assume libjvm[_g].so is installed in a JDK and we use this path.
 418   //
 419   // Otherwise exit with message: "Could not create the Java virtual machine."
 420   //
 421   // The following extra steps are taken in the debugging version:
 422   //
 423   // If "/jre/lib/" does NOT appear at the right place in the path
 424   // instead of exit check for $JAVA_HOME environment variable.
 425   //
 426   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 427   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 428   // it looks like libjvm[_g].so is installed there
 429   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 430   //
 431   // Otherwise exit.
 432   //
 433   // Important note: if the location of libjvm.so changes this
 434   // code needs to be changed accordingly.
 435 
 436   // The next few definitions allow the code to be verbatim:
 437 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 438 #define getenv(n) ::getenv(n)
 439 
 440 /*
 441  * See ld(1):
 442  *      The linker uses the following search paths to locate required
 443  *      shared libraries:
 444  *        1: ...
 445  *        ...
 446  *        7: The default directories, normally /lib and /usr/lib.
 447  */
 448 #ifndef DEFAULT_LIBPATH
 449 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 450 #endif
 451 
 452 #define EXTENSIONS_DIR  "/lib/ext"
 453 #define ENDORSED_DIR    "/lib/endorsed"
 454 #define REG_DIR         "/usr/java/packages"
 455 
 456 #ifdef __APPLE__
 457 #define SYS_EXTENSIONS_DIR   "/Library/Java/Extensions"
 458 #define SYS_EXTENSIONS_DIRS  SYS_EXTENSIONS_DIR ":/Network" SYS_EXTENSIONS_DIR ":/System" SYS_EXTENSIONS_DIR ":/usr/lib/java"
 459         const char *user_home_dir = get_home();
 460         // the null in SYS_EXTENSIONS_DIRS counts for the size of the colon after user_home_dir
 461         int system_ext_size = strlen(user_home_dir) + sizeof(SYS_EXTENSIONS_DIR) +
 462             sizeof(SYS_EXTENSIONS_DIRS);
 463 #endif
 464 
 465   {
 466     /* sysclasspath, java_home, dll_dir */
 467     {
 468         char *home_path;
 469         char *dll_path;
 470         char *pslash;
 471         char buf[MAXPATHLEN];
 472         os::jvm_path(buf, sizeof(buf));
 473 
 474         // Found the full path to libjvm.so.
 475         // Now cut the path to <java_home>/jre if we can.
 476         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 477         pslash = strrchr(buf, '/');
 478         if (pslash != NULL)
 479             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 480         dll_path = malloc(strlen(buf) + 1);
 481         if (dll_path == NULL)
 482             return;
 483         strcpy(dll_path, buf);
 484         Arguments::set_dll_dir(dll_path);
 485 
 486         if (pslash != NULL) {
 487             pslash = strrchr(buf, '/');
 488             if (pslash != NULL) {
 489                 *pslash = '\0';       /* get rid of /<arch> (/lib on macosx) */
 490 #ifndef __APPLE__
 491                 pslash = strrchr(buf, '/');
 492                 if (pslash != NULL)
 493                     *pslash = '\0';   /* get rid of /lib */
 494 #endif
 495             }
 496         }
 497 
 498         home_path = malloc(strlen(buf) + 1);
 499         if (home_path == NULL)
 500             return;
 501         strcpy(home_path, buf);
 502         Arguments::set_java_home(home_path);
 503 
 504         if (!set_boot_path('/', ':'))
 505             return;
 506     }
 507 
 508     /*
 509      * Where to look for native libraries
 510      *
 511      * Note: Due to a legacy implementation, most of the library path
 512      * is set in the launcher.  This was to accomodate linking restrictions
 513      * on legacy Bsd implementations (which are no longer supported).
 514      * Eventually, all the library path setting will be done here.
 515      *
 516      * However, to prevent the proliferation of improperly built native
 517      * libraries, the new path component /usr/java/packages is added here.
 518      * Eventually, all the library path setting will be done here.
 519      */
 520     {
 521         char *ld_library_path;
 522 
 523         /*
 524          * Construct the invariant part of ld_library_path. Note that the
 525          * space for the colon and the trailing null are provided by the
 526          * nulls included by the sizeof operator (so actually we allocate
 527          * a byte more than necessary).
 528          */
 529 #ifdef __APPLE__
 530         ld_library_path = (char *) malloc(system_ext_size);
 531         sprintf(ld_library_path, "%s" SYS_EXTENSIONS_DIR ":" SYS_EXTENSIONS_DIRS, user_home_dir);
 532 #else
 533         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 534             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 535         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 536 #endif
 537 
 538         /*
 539          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 540          * should always exist (until the legacy problem cited above is
 541          * addressed).
 542          */
 543 #ifdef __APPLE__
 544         // Prepend the default path with the JAVA_LIBRARY_PATH so that the app launcher code can specify a directory inside an app wrapper
 545         char *l = getenv("JAVA_LIBRARY_PATH");
 546         if (l != NULL) {
 547             char *t = ld_library_path;
 548             /* That's +1 for the colon and +1 for the trailing '\0' */
 549             ld_library_path = (char *) malloc(strlen(l) + 1 + strlen(t) + 1);
 550             sprintf(ld_library_path, "%s:%s", l, t);
 551             free(t);
 552         }
 553 
 554         char *v = getenv("DYLD_LIBRARY_PATH");
 555 #else
 556         char *v = getenv("LD_LIBRARY_PATH");
 557 #endif
 558         if (v != NULL) {
 559             char *t = ld_library_path;
 560             /* That's +1 for the colon and +1 for the trailing '\0' */
 561             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 562             sprintf(ld_library_path, "%s:%s", v, t);
 563             free(t);
 564         }
 565         Arguments::set_library_path(ld_library_path);
 566     }
 567 
 568     /*
 569      * Extensions directories.
 570      *
 571      * Note that the space for the colon and the trailing null are provided
 572      * by the nulls included by the sizeof operator (so actually one byte more
 573      * than necessary is allocated).
 574      */
 575     {
 576 #ifdef __APPLE__
 577         char *buf = malloc(strlen(Arguments::get_java_home()) +
 578             sizeof(EXTENSIONS_DIR) + system_ext_size);
 579         sprintf(buf, "%s" SYS_EXTENSIONS_DIR ":%s" EXTENSIONS_DIR ":"
 580             SYS_EXTENSIONS_DIRS, user_home_dir, Arguments::get_java_home());
 581 #else
 582         char *buf = malloc(strlen(Arguments::get_java_home()) +
 583             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 584         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 585             Arguments::get_java_home());
 586 #endif
 587 
 588         Arguments::set_ext_dirs(buf);
 589     }
 590 
 591     /* Endorsed standards default directory. */
 592     {
 593         char * buf;
 594         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 595         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 596         Arguments::set_endorsed_dirs(buf);
 597     }
 598   }
 599 
 600 #ifdef __APPLE__
 601 #undef SYS_EXTENSIONS_DIR
 602 #endif
 603 #undef malloc
 604 #undef getenv
 605 #undef EXTENSIONS_DIR
 606 #undef ENDORSED_DIR
 607 
 608   // Done
 609   return;
 610 }
 611 
 612 ////////////////////////////////////////////////////////////////////////////////
 613 // breakpoint support
 614 
 615 void os::breakpoint() {
 616   BREAKPOINT;
 617 }
 618 
 619 extern "C" void breakpoint() {
 620   // use debugger to set breakpoint here
 621 }
 622 
 623 ////////////////////////////////////////////////////////////////////////////////
 624 // signal support
 625 
 626 debug_only(static bool signal_sets_initialized = false);
 627 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 628 
 629 bool os::Bsd::is_sig_ignored(int sig) {
 630       struct sigaction oact;
 631       sigaction(sig, (struct sigaction*)NULL, &oact);
 632       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 633                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 634       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 635            return true;
 636       else
 637            return false;
 638 }
 639 
 640 void os::Bsd::signal_sets_init() {
 641   // Should also have an assertion stating we are still single-threaded.
 642   assert(!signal_sets_initialized, "Already initialized");
 643   // Fill in signals that are necessarily unblocked for all threads in
 644   // the VM. Currently, we unblock the following signals:
 645   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 646   //                         by -Xrs (=ReduceSignalUsage));
 647   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 648   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 649   // the dispositions or masks wrt these signals.
 650   // Programs embedding the VM that want to use the above signals for their
 651   // own purposes must, at this time, use the "-Xrs" option to prevent
 652   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 653   // (See bug 4345157, and other related bugs).
 654   // In reality, though, unblocking these signals is really a nop, since
 655   // these signals are not blocked by default.
 656   sigemptyset(&unblocked_sigs);
 657   sigemptyset(&allowdebug_blocked_sigs);
 658   sigaddset(&unblocked_sigs, SIGILL);
 659   sigaddset(&unblocked_sigs, SIGSEGV);
 660   sigaddset(&unblocked_sigs, SIGBUS);
 661   sigaddset(&unblocked_sigs, SIGFPE);
 662   sigaddset(&unblocked_sigs, SR_signum);
 663 
 664   if (!ReduceSignalUsage) {
 665    if (!os::Bsd::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 666       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 667       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 668    }
 669    if (!os::Bsd::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 670       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 671       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 672    }
 673    if (!os::Bsd::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 674       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 675       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 676    }
 677   }
 678   // Fill in signals that are blocked by all but the VM thread.
 679   sigemptyset(&vm_sigs);
 680   if (!ReduceSignalUsage)
 681     sigaddset(&vm_sigs, BREAK_SIGNAL);
 682   debug_only(signal_sets_initialized = true);
 683 
 684 }
 685 
 686 // These are signals that are unblocked while a thread is running Java.
 687 // (For some reason, they get blocked by default.)
 688 sigset_t* os::Bsd::unblocked_signals() {
 689   assert(signal_sets_initialized, "Not initialized");
 690   return &unblocked_sigs;
 691 }
 692 
 693 // These are the signals that are blocked while a (non-VM) thread is
 694 // running Java. Only the VM thread handles these signals.
 695 sigset_t* os::Bsd::vm_signals() {
 696   assert(signal_sets_initialized, "Not initialized");
 697   return &vm_sigs;
 698 }
 699 
 700 // These are signals that are blocked during cond_wait to allow debugger in
 701 sigset_t* os::Bsd::allowdebug_blocked_signals() {
 702   assert(signal_sets_initialized, "Not initialized");
 703   return &allowdebug_blocked_sigs;
 704 }
 705 
 706 void os::Bsd::hotspot_sigmask(Thread* thread) {
 707 
 708   //Save caller's signal mask before setting VM signal mask
 709   sigset_t caller_sigmask;
 710   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 711 
 712   OSThread* osthread = thread->osthread();
 713   osthread->set_caller_sigmask(caller_sigmask);
 714 
 715   pthread_sigmask(SIG_UNBLOCK, os::Bsd::unblocked_signals(), NULL);
 716 
 717   if (!ReduceSignalUsage) {
 718     if (thread->is_VM_thread()) {
 719       // Only the VM thread handles BREAK_SIGNAL ...
 720       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 721     } else {
 722       // ... all other threads block BREAK_SIGNAL
 723       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 724     }
 725   }
 726 }
 727 
 728 #ifndef _ALLBSD_SOURCE
 729 //////////////////////////////////////////////////////////////////////////////
 730 // detecting pthread library
 731 
 732 void os::Bsd::libpthread_init() {
 733   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 734   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 735   // generic name for earlier versions.
 736   // Define macros here so we can build HotSpot on old systems.
 737 # ifndef _CS_GNU_LIBC_VERSION
 738 # define _CS_GNU_LIBC_VERSION 2
 739 # endif
 740 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 741 # define _CS_GNU_LIBPTHREAD_VERSION 3
 742 # endif
 743 
 744   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 745   if (n > 0) {
 746      char *str = (char *)malloc(n);
 747      confstr(_CS_GNU_LIBC_VERSION, str, n);
 748      os::Bsd::set_glibc_version(str);
 749   } else {
 750      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 751      static char _gnu_libc_version[32];
 752      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 753               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 754      os::Bsd::set_glibc_version(_gnu_libc_version);
 755   }
 756 
 757   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 758   if (n > 0) {
 759      char *str = (char *)malloc(n);
 760      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 761      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 762      // us "NPTL-0.29" even we are running with BsdThreads. Check if this
 763      // is the case. BsdThreads has a hard limit on max number of threads.
 764      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 765      // On the other hand, NPTL does not have such a limit, sysconf()
 766      // will return -1 and errno is not changed. Check if it is really NPTL.
 767      if (strcmp(os::Bsd::glibc_version(), "glibc 2.3.2") == 0 &&
 768          strstr(str, "NPTL") &&
 769          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 770        free(str);
 771        os::Bsd::set_libpthread_version("bsdthreads");
 772      } else {
 773        os::Bsd::set_libpthread_version(str);
 774      }
 775   } else {
 776     // glibc before 2.3.2 only has BsdThreads.
 777     os::Bsd::set_libpthread_version("bsdthreads");
 778   }
 779 
 780   if (strstr(libpthread_version(), "NPTL")) {
 781      os::Bsd::set_is_NPTL();
 782   } else {
 783      os::Bsd::set_is_BsdThreads();
 784   }
 785 
 786   // BsdThreads have two flavors: floating-stack mode, which allows variable
 787   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 788   if (os::Bsd::is_NPTL() || os::Bsd::supports_variable_stack_size()) {
 789      os::Bsd::set_is_floating_stack();
 790   }
 791 }
 792 
 793 /////////////////////////////////////////////////////////////////////////////
 794 // thread stack
 795 
 796 // Force Bsd kernel to expand current thread stack. If "bottom" is close
 797 // to the stack guard, caller should block all signals.
 798 //
 799 // MAP_GROWSDOWN:
 800 //   A special mmap() flag that is used to implement thread stacks. It tells
 801 //   kernel that the memory region should extend downwards when needed. This
 802 //   allows early versions of BsdThreads to only mmap the first few pages
 803 //   when creating a new thread. Bsd kernel will automatically expand thread
 804 //   stack as needed (on page faults).
 805 //
 806 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 807 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 808 //   region, it's hard to tell if the fault is due to a legitimate stack
 809 //   access or because of reading/writing non-exist memory (e.g. buffer
 810 //   overrun). As a rule, if the fault happens below current stack pointer,
 811 //   Bsd kernel does not expand stack, instead a SIGSEGV is sent to the
 812 //   application (see Bsd kernel fault.c).
 813 //
 814 //   This Bsd feature can cause SIGSEGV when VM bangs thread stack for
 815 //   stack overflow detection.
 816 //
 817 //   Newer version of BsdThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 818 //   not use this flag. However, the stack of initial thread is not created
 819 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 820 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 821 //   and then attach the thread to JVM.
 822 //
 823 // To get around the problem and allow stack banging on Bsd, we need to
 824 // manually expand thread stack after receiving the SIGSEGV.
 825 //
 826 // There are two ways to expand thread stack to address "bottom", we used
 827 // both of them in JVM before 1.5:
 828 //   1. adjust stack pointer first so that it is below "bottom", and then
 829 //      touch "bottom"
 830 //   2. mmap() the page in question
 831 //
 832 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 833 // if current sp is already near the lower end of page 101, and we need to
 834 // call mmap() to map page 100, it is possible that part of the mmap() frame
 835 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 836 // That will destroy the mmap() frame and cause VM to crash.
 837 //
 838 // The following code works by adjusting sp first, then accessing the "bottom"
 839 // page to force a page fault. Bsd kernel will then automatically expand the
 840 // stack mapping.
 841 //
 842 // _expand_stack_to() assumes its frame size is less than page size, which
 843 // should always be true if the function is not inlined.
 844 
 845 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 846 #define NOINLINE
 847 #else
 848 #define NOINLINE __attribute__ ((noinline))
 849 #endif
 850 
 851 static void _expand_stack_to(address bottom) NOINLINE;
 852 
 853 static void _expand_stack_to(address bottom) {
 854   address sp;
 855   size_t size;
 856   volatile char *p;
 857 
 858   // Adjust bottom to point to the largest address within the same page, it
 859   // gives us a one-page buffer if alloca() allocates slightly more memory.
 860   bottom = (address)align_size_down((uintptr_t)bottom, os::Bsd::page_size());
 861   bottom += os::Bsd::page_size() - 1;
 862 
 863   // sp might be slightly above current stack pointer; if that's the case, we
 864   // will alloca() a little more space than necessary, which is OK. Don't use
 865   // os::current_stack_pointer(), as its result can be slightly below current
 866   // stack pointer, causing us to not alloca enough to reach "bottom".
 867   sp = (address)&sp;
 868 
 869   if (sp > bottom) {
 870     size = sp - bottom;
 871     p = (volatile char *)alloca(size);
 872     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 873     p[0] = '\0';
 874   }
 875 }
 876 
 877 bool os::Bsd::manually_expand_stack(JavaThread * t, address addr) {
 878   assert(t!=NULL, "just checking");
 879   assert(t->osthread()->expanding_stack(), "expand should be set");
 880   assert(t->stack_base() != NULL, "stack_base was not initialized");
 881 
 882   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 883     sigset_t mask_all, old_sigset;
 884     sigfillset(&mask_all);
 885     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 886     _expand_stack_to(addr);
 887     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 888     return true;
 889   }
 890   return false;
 891 }
 892 #endif
 893 
 894 //////////////////////////////////////////////////////////////////////////////
 895 // create new thread
 896 
 897 static address highest_vm_reserved_address();
 898 
 899 // check if it's safe to start a new thread
 900 static bool _thread_safety_check(Thread* thread) {
 901 #ifdef _ALLBSD_SOURCE
 902     return true;
 903 #else
 904   if (os::Bsd::is_BsdThreads() && !os::Bsd::is_floating_stack()) {
 905     // Fixed stack BsdThreads (SuSE Bsd/x86, and some versions of Redhat)
 906     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 907     //   allocated (MAP_FIXED) from high address space. Every thread stack
 908     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 909     //   it to other values if they rebuild BsdThreads).
 910     //
 911     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 912     // the memory region has already been mmap'ed. That means if we have too
 913     // many threads and/or very large heap, eventually thread stack will
 914     // collide with heap.
 915     //
 916     // Here we try to prevent heap/stack collision by comparing current
 917     // stack bottom with the highest address that has been mmap'ed by JVM
 918     // plus a safety margin for memory maps created by native code.
 919     //
 920     // This feature can be disabled by setting ThreadSafetyMargin to 0
 921     //
 922     if (ThreadSafetyMargin > 0) {
 923       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 924 
 925       // not safe if our stack extends below the safety margin
 926       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 927     } else {
 928       return true;
 929     }
 930   } else {
 931     // Floating stack BsdThreads or NPTL:
 932     //   Unlike fixed stack BsdThreads, thread stacks are not MAP_FIXED. When
 933     //   there's not enough space left, pthread_create() will fail. If we come
 934     //   here, that means enough space has been reserved for stack.
 935     return true;
 936   }
 937 #endif
 938 }
 939 
 940 #ifdef __APPLE__
 941 // library handle for calling objc_registerThreadWithCollector()
 942 // without static linking to the libobjc library
 943 #define OBJC_LIB "/usr/lib/libobjc.dylib"
 944 #define OBJC_GCREGISTER "objc_registerThreadWithCollector"
 945 typedef void (*objc_registerThreadWithCollector_t)();
 946 extern "C" objc_registerThreadWithCollector_t objc_registerThreadWithCollectorFunction;
 947 objc_registerThreadWithCollector_t objc_registerThreadWithCollectorFunction = NULL;
 948 #endif
 949 
 950 // Thread start routine for all newly created threads
 951 static void *java_start(Thread *thread) {
 952   // Try to randomize the cache line index of hot stack frames.
 953   // This helps when threads of the same stack traces evict each other's
 954   // cache lines. The threads can be either from the same JVM instance, or
 955   // from different JVM instances. The benefit is especially true for
 956   // processors with hyperthreading technology.
 957   static int counter = 0;
 958   int pid = os::current_process_id();
 959   alloca(((pid ^ counter++) & 7) * 128);
 960 
 961   ThreadLocalStorage::set_thread(thread);
 962 
 963   OSThread* osthread = thread->osthread();
 964   Monitor* sync = osthread->startThread_lock();
 965 
 966   // non floating stack BsdThreads needs extra check, see above
 967   if (!_thread_safety_check(thread)) {
 968     // notify parent thread
 969     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 970     osthread->set_state(ZOMBIE);
 971     sync->notify_all();
 972     return NULL;
 973   }
 974 
 975 #ifdef _ALLBSD_SOURCE
 976   // thread_id is pthread_id on BSD
 977   osthread->set_thread_id(::pthread_self());
 978 #else
 979   // thread_id is kernel thread id (similar to Solaris LWP id)
 980   osthread->set_thread_id(os::Bsd::gettid());
 981 
 982   if (UseNUMA) {
 983     int lgrp_id = os::numa_get_group_id();
 984     if (lgrp_id != -1) {
 985       thread->set_lgrp_id(lgrp_id);
 986     }
 987   }
 988 #endif
 989   // initialize signal mask for this thread
 990   os::Bsd::hotspot_sigmask(thread);
 991 
 992   // initialize floating point control register
 993   os::Bsd::init_thread_fpu_state();
 994 
 995 #ifdef __APPLE__
 996   // register thread with objc gc
 997   if (objc_registerThreadWithCollectorFunction != NULL) {
 998     objc_registerThreadWithCollectorFunction();
 999   }
1000 #endif
1001 
1002   // handshaking with parent thread
1003   {
1004     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
1005 
1006     // notify parent thread
1007     osthread->set_state(INITIALIZED);
1008     sync->notify_all();
1009 
1010     // wait until os::start_thread()
1011     while (osthread->get_state() == INITIALIZED) {
1012       sync->wait(Mutex::_no_safepoint_check_flag);
1013     }
1014   }
1015 
1016   // call one more level start routine
1017   thread->run();
1018 
1019   return 0;
1020 }
1021 
1022 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
1023   assert(thread->osthread() == NULL, "caller responsible");
1024 
1025   // Allocate the OSThread object
1026   OSThread* osthread = new OSThread(NULL, NULL);
1027   if (osthread == NULL) {
1028     return false;
1029   }
1030 
1031   // set the correct thread state
1032   osthread->set_thread_type(thr_type);
1033 
1034   // Initial state is ALLOCATED but not INITIALIZED
1035   osthread->set_state(ALLOCATED);
1036 
1037   thread->set_osthread(osthread);
1038 
1039   // init thread attributes
1040   pthread_attr_t attr;
1041   pthread_attr_init(&attr);
1042   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1043 
1044   // stack size
1045   if (os::Bsd::supports_variable_stack_size()) {
1046     // calculate stack size if it's not specified by caller
1047     if (stack_size == 0) {
1048       stack_size = os::Bsd::default_stack_size(thr_type);
1049 
1050       switch (thr_type) {
1051       case os::java_thread:
1052         // Java threads use ThreadStackSize which default value can be
1053         // changed with the flag -Xss
1054         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
1055         stack_size = JavaThread::stack_size_at_create();
1056         break;
1057       case os::compiler_thread:
1058         if (CompilerThreadStackSize > 0) {
1059           stack_size = (size_t)(CompilerThreadStackSize * K);
1060           break;
1061         } // else fall through:
1062           // use VMThreadStackSize if CompilerThreadStackSize is not defined
1063       case os::vm_thread:
1064       case os::pgc_thread:
1065       case os::cgc_thread:
1066       case os::watcher_thread:
1067         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1068         break;
1069       }
1070     }
1071 
1072     stack_size = MAX2(stack_size, os::Bsd::min_stack_allowed);
1073     pthread_attr_setstacksize(&attr, stack_size);
1074   } else {
1075     // let pthread_create() pick the default value.
1076   }
1077 
1078 #ifndef _ALLBSD_SOURCE
1079   // glibc guard page
1080   pthread_attr_setguardsize(&attr, os::Bsd::default_guard_size(thr_type));
1081 #endif
1082 
1083   ThreadState state;
1084 
1085   {
1086 
1087 #ifndef _ALLBSD_SOURCE
1088     // Serialize thread creation if we are running with fixed stack BsdThreads
1089     bool lock = os::Bsd::is_BsdThreads() && !os::Bsd::is_floating_stack();
1090     if (lock) {
1091       os::Bsd::createThread_lock()->lock_without_safepoint_check();
1092     }
1093 #endif
1094 
1095     pthread_t tid;
1096     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
1097 
1098     pthread_attr_destroy(&attr);
1099 
1100     if (ret != 0) {
1101       if (PrintMiscellaneous && (Verbose || WizardMode)) {
1102         perror("pthread_create()");
1103       }
1104       // Need to clean up stuff we've allocated so far
1105       thread->set_osthread(NULL);
1106       delete osthread;
1107 #ifndef _ALLBSD_SOURCE
1108       if (lock) os::Bsd::createThread_lock()->unlock();
1109 #endif
1110       return false;
1111     }
1112 
1113     // Store pthread info into the OSThread
1114     osthread->set_pthread_id(tid);
1115 
1116     // Wait until child thread is either initialized or aborted
1117     {
1118       Monitor* sync_with_child = osthread->startThread_lock();
1119       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1120       while ((state = osthread->get_state()) == ALLOCATED) {
1121         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
1122       }
1123     }
1124 
1125 #ifndef _ALLBSD_SOURCE
1126     if (lock) {
1127       os::Bsd::createThread_lock()->unlock();
1128     }
1129 #endif
1130   }
1131 
1132   // Aborted due to thread limit being reached
1133   if (state == ZOMBIE) {
1134       thread->set_osthread(NULL);
1135       delete osthread;
1136       return false;
1137   }
1138 
1139   // The thread is returned suspended (in state INITIALIZED),
1140   // and is started higher up in the call chain
1141   assert(state == INITIALIZED, "race condition");
1142   return true;
1143 }
1144 
1145 /////////////////////////////////////////////////////////////////////////////
1146 // attach existing thread
1147 
1148 // bootstrap the main thread
1149 bool os::create_main_thread(JavaThread* thread) {
1150   assert(os::Bsd::_main_thread == pthread_self(), "should be called inside main thread");
1151   return create_attached_thread(thread);
1152 }
1153 
1154 bool os::create_attached_thread(JavaThread* thread) {
1155 #ifdef ASSERT
1156     thread->verify_not_published();
1157 #endif
1158 
1159   // Allocate the OSThread object
1160   OSThread* osthread = new OSThread(NULL, NULL);
1161 
1162   if (osthread == NULL) {
1163     return false;
1164   }
1165 
1166   // Store pthread info into the OSThread
1167 #ifdef _ALLBSD_SOURCE
1168   osthread->set_thread_id(::pthread_self());
1169 #else
1170   osthread->set_thread_id(os::Bsd::gettid());
1171 #endif
1172   osthread->set_pthread_id(::pthread_self());
1173 
1174   // initialize floating point control register
1175   os::Bsd::init_thread_fpu_state();
1176 
1177   // Initial thread state is RUNNABLE
1178   osthread->set_state(RUNNABLE);
1179 
1180   thread->set_osthread(osthread);
1181 
1182 #ifndef _ALLBSD_SOURCE
1183   if (UseNUMA) {
1184     int lgrp_id = os::numa_get_group_id();
1185     if (lgrp_id != -1) {
1186       thread->set_lgrp_id(lgrp_id);
1187     }
1188   }
1189 
1190   if (os::Bsd::is_initial_thread()) {
1191     // If current thread is initial thread, its stack is mapped on demand,
1192     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1193     // the entire stack region to avoid SEGV in stack banging.
1194     // It is also useful to get around the heap-stack-gap problem on SuSE
1195     // kernel (see 4821821 for details). We first expand stack to the top
1196     // of yellow zone, then enable stack yellow zone (order is significant,
1197     // enabling yellow zone first will crash JVM on SuSE Bsd), so there
1198     // is no gap between the last two virtual memory regions.
1199 
1200     JavaThread *jt = (JavaThread *)thread;
1201     address addr = jt->stack_yellow_zone_base();
1202     assert(addr != NULL, "initialization problem?");
1203     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1204 
1205     osthread->set_expanding_stack();
1206     os::Bsd::manually_expand_stack(jt, addr);
1207     osthread->clear_expanding_stack();
1208   }
1209 #endif
1210 
1211   // initialize signal mask for this thread
1212   // and save the caller's signal mask
1213   os::Bsd::hotspot_sigmask(thread);
1214 
1215   return true;
1216 }
1217 
1218 void os::pd_start_thread(Thread* thread) {
1219   OSThread * osthread = thread->osthread();
1220   assert(osthread->get_state() != INITIALIZED, "just checking");
1221   Monitor* sync_with_child = osthread->startThread_lock();
1222   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1223   sync_with_child->notify();
1224 }
1225 
1226 // Free Bsd resources related to the OSThread
1227 void os::free_thread(OSThread* osthread) {
1228   assert(osthread != NULL, "osthread not set");
1229 
1230   if (Thread::current()->osthread() == osthread) {
1231     // Restore caller's signal mask
1232     sigset_t sigmask = osthread->caller_sigmask();
1233     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1234    }
1235 
1236   delete osthread;
1237 }
1238 
1239 //////////////////////////////////////////////////////////////////////////////
1240 // thread local storage
1241 
1242 int os::allocate_thread_local_storage() {
1243   pthread_key_t key;
1244   int rslt = pthread_key_create(&key, NULL);
1245   assert(rslt == 0, "cannot allocate thread local storage");
1246   return (int)key;
1247 }
1248 
1249 // Note: This is currently not used by VM, as we don't destroy TLS key
1250 // on VM exit.
1251 void os::free_thread_local_storage(int index) {
1252   int rslt = pthread_key_delete((pthread_key_t)index);
1253   assert(rslt == 0, "invalid index");
1254 }
1255 
1256 void os::thread_local_storage_at_put(int index, void* value) {
1257   int rslt = pthread_setspecific((pthread_key_t)index, value);
1258   assert(rslt == 0, "pthread_setspecific failed");
1259 }
1260 
1261 extern "C" Thread* get_thread() {
1262   return ThreadLocalStorage::thread();
1263 }
1264 
1265 //////////////////////////////////////////////////////////////////////////////
1266 // initial thread
1267 
1268 #ifndef _ALLBSD_SOURCE
1269 // Check if current thread is the initial thread, similar to Solaris thr_main.
1270 bool os::Bsd::is_initial_thread(void) {
1271   char dummy;
1272   // If called before init complete, thread stack bottom will be null.
1273   // Can be called if fatal error occurs before initialization.
1274   if (initial_thread_stack_bottom() == NULL) return false;
1275   assert(initial_thread_stack_bottom() != NULL &&
1276          initial_thread_stack_size()   != 0,
1277          "os::init did not locate initial thread's stack region");
1278   if ((address)&dummy >= initial_thread_stack_bottom() &&
1279       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1280        return true;
1281   else return false;
1282 }
1283 
1284 // Find the virtual memory area that contains addr
1285 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1286   FILE *fp = fopen("/proc/self/maps", "r");
1287   if (fp) {
1288     address low, high;
1289     while (!feof(fp)) {
1290       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1291         if (low <= addr && addr < high) {
1292            if (vma_low)  *vma_low  = low;
1293            if (vma_high) *vma_high = high;
1294            fclose (fp);
1295            return true;
1296         }
1297       }
1298       for (;;) {
1299         int ch = fgetc(fp);
1300         if (ch == EOF || ch == (int)'\n') break;
1301       }
1302     }
1303     fclose(fp);
1304   }
1305   return false;
1306 }
1307 
1308 // Locate initial thread stack. This special handling of initial thread stack
1309 // is needed because pthread_getattr_np() on most (all?) Bsd distros returns
1310 // bogus value for initial thread.
1311 void os::Bsd::capture_initial_stack(size_t max_size) {
1312   // stack size is the easy part, get it from RLIMIT_STACK
1313   size_t stack_size;
1314   struct rlimit rlim;
1315   getrlimit(RLIMIT_STACK, &rlim);
1316   stack_size = rlim.rlim_cur;
1317 
1318   // 6308388: a bug in ld.so will relocate its own .data section to the
1319   //   lower end of primordial stack; reduce ulimit -s value a little bit
1320   //   so we won't install guard page on ld.so's data section.
1321   stack_size -= 2 * page_size();
1322 
1323   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1324   //   7.1, in both cases we will get 2G in return value.
1325   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1326   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1327   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1328   //   in case other parts in glibc still assumes 2M max stack size.
1329   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1330 #ifndef IA64
1331   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1332 #else
1333   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1334   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1335 #endif
1336 
1337   // Try to figure out where the stack base (top) is. This is harder.
1338   //
1339   // When an application is started, glibc saves the initial stack pointer in
1340   // a global variable "__libc_stack_end", which is then used by system
1341   // libraries. __libc_stack_end should be pretty close to stack top. The
1342   // variable is available since the very early days. However, because it is
1343   // a private interface, it could disappear in the future.
1344   //
1345   // Bsd kernel saves start_stack information in /proc/<pid>/stat. Similar
1346   // to __libc_stack_end, it is very close to stack top, but isn't the real
1347   // stack top. Note that /proc may not exist if VM is running as a chroot
1348   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1349   // /proc/<pid>/stat could change in the future (though unlikely).
1350   //
1351   // We try __libc_stack_end first. If that doesn't work, look for
1352   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1353   // as a hint, which should work well in most cases.
1354 
1355   uintptr_t stack_start;
1356 
1357   // try __libc_stack_end first
1358   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1359   if (p && *p) {
1360     stack_start = *p;
1361   } else {
1362     // see if we can get the start_stack field from /proc/self/stat
1363     FILE *fp;
1364     int pid;
1365     char state;
1366     int ppid;
1367     int pgrp;
1368     int session;
1369     int nr;
1370     int tpgrp;
1371     unsigned long flags;
1372     unsigned long minflt;
1373     unsigned long cminflt;
1374     unsigned long majflt;
1375     unsigned long cmajflt;
1376     unsigned long utime;
1377     unsigned long stime;
1378     long cutime;
1379     long cstime;
1380     long prio;
1381     long nice;
1382     long junk;
1383     long it_real;
1384     uintptr_t start;
1385     uintptr_t vsize;
1386     intptr_t rss;
1387     uintptr_t rsslim;
1388     uintptr_t scodes;
1389     uintptr_t ecode;
1390     int i;
1391 
1392     // Figure what the primordial thread stack base is. Code is inspired
1393     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1394     // followed by command name surrounded by parentheses, state, etc.
1395     char stat[2048];
1396     int statlen;
1397 
1398     fp = fopen("/proc/self/stat", "r");
1399     if (fp) {
1400       statlen = fread(stat, 1, 2047, fp);
1401       stat[statlen] = '\0';
1402       fclose(fp);
1403 
1404       // Skip pid and the command string. Note that we could be dealing with
1405       // weird command names, e.g. user could decide to rename java launcher
1406       // to "java 1.4.2 :)", then the stat file would look like
1407       //                1234 (java 1.4.2 :)) R ... ...
1408       // We don't really need to know the command string, just find the last
1409       // occurrence of ")" and then start parsing from there. See bug 4726580.
1410       char * s = strrchr(stat, ')');
1411 
1412       i = 0;
1413       if (s) {
1414         // Skip blank chars
1415         do s++; while (isspace(*s));
1416 
1417 #define _UFM UINTX_FORMAT
1418 #define _DFM INTX_FORMAT
1419 
1420         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1421         /*              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 */
1422         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,
1423              &state,          /* 3  %c  */
1424              &ppid,           /* 4  %d  */
1425              &pgrp,           /* 5  %d  */
1426              &session,        /* 6  %d  */
1427              &nr,             /* 7  %d  */
1428              &tpgrp,          /* 8  %d  */
1429              &flags,          /* 9  %lu  */
1430              &minflt,         /* 10 %lu  */
1431              &cminflt,        /* 11 %lu  */
1432              &majflt,         /* 12 %lu  */
1433              &cmajflt,        /* 13 %lu  */
1434              &utime,          /* 14 %lu  */
1435              &stime,          /* 15 %lu  */
1436              &cutime,         /* 16 %ld  */
1437              &cstime,         /* 17 %ld  */
1438              &prio,           /* 18 %ld  */
1439              &nice,           /* 19 %ld  */
1440              &junk,           /* 20 %ld  */
1441              &it_real,        /* 21 %ld  */
1442              &start,          /* 22 UINTX_FORMAT */
1443              &vsize,          /* 23 UINTX_FORMAT */
1444              &rss,            /* 24 INTX_FORMAT  */
1445              &rsslim,         /* 25 UINTX_FORMAT */
1446              &scodes,         /* 26 UINTX_FORMAT */
1447              &ecode,          /* 27 UINTX_FORMAT */
1448              &stack_start);   /* 28 UINTX_FORMAT */
1449       }
1450 
1451 #undef _UFM
1452 #undef _DFM
1453 
1454       if (i != 28 - 2) {
1455          assert(false, "Bad conversion from /proc/self/stat");
1456          // product mode - assume we are the initial thread, good luck in the
1457          // embedded case.
1458          warning("Can't detect initial thread stack location - bad conversion");
1459          stack_start = (uintptr_t) &rlim;
1460       }
1461     } else {
1462       // For some reason we can't open /proc/self/stat (for example, running on
1463       // FreeBSD with a Bsd emulator, or inside chroot), this should work for
1464       // most cases, so don't abort:
1465       warning("Can't detect initial thread stack location - no /proc/self/stat");
1466       stack_start = (uintptr_t) &rlim;
1467     }
1468   }
1469 
1470   // Now we have a pointer (stack_start) very close to the stack top, the
1471   // next thing to do is to figure out the exact location of stack top. We
1472   // can find out the virtual memory area that contains stack_start by
1473   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1474   // and its upper limit is the real stack top. (again, this would fail if
1475   // running inside chroot, because /proc may not exist.)
1476 
1477   uintptr_t stack_top;
1478   address low, high;
1479   if (find_vma((address)stack_start, &low, &high)) {
1480     // success, "high" is the true stack top. (ignore "low", because initial
1481     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1482     stack_top = (uintptr_t)high;
1483   } else {
1484     // failed, likely because /proc/self/maps does not exist
1485     warning("Can't detect initial thread stack location - find_vma failed");
1486     // best effort: stack_start is normally within a few pages below the real
1487     // stack top, use it as stack top, and reduce stack size so we won't put
1488     // guard page outside stack.
1489     stack_top = stack_start;
1490     stack_size -= 16 * page_size();
1491   }
1492 
1493   // stack_top could be partially down the page so align it
1494   stack_top = align_size_up(stack_top, page_size());
1495 
1496   if (max_size && stack_size > max_size) {
1497      _initial_thread_stack_size = max_size;
1498   } else {
1499      _initial_thread_stack_size = stack_size;
1500   }
1501 
1502   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1503   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1504 }
1505 #endif
1506 
1507 ////////////////////////////////////////////////////////////////////////////////
1508 // time support
1509 
1510 // Time since start-up in seconds to a fine granularity.
1511 // Used by VMSelfDestructTimer and the MemProfiler.
1512 double os::elapsedTime() {
1513 
1514   return (double)(os::elapsed_counter()) * 0.000001;
1515 }
1516 
1517 jlong os::elapsed_counter() {
1518   timeval time;
1519   int status = gettimeofday(&time, NULL);
1520   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1521 }
1522 
1523 jlong os::elapsed_frequency() {
1524   return (1000 * 1000);
1525 }
1526 
1527 // XXX: For now, code this as if BSD does not support vtime.
1528 bool os::supports_vtime() { return false; }
1529 bool os::enable_vtime()   { return false; }
1530 bool os::vtime_enabled()  { return false; }
1531 double os::elapsedVTime() {
1532   // better than nothing, but not much
1533   return elapsedTime();
1534 }
1535 
1536 jlong os::javaTimeMillis() {
1537   timeval time;
1538   int status = gettimeofday(&time, NULL);
1539   assert(status != -1, "bsd error");
1540   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1541 }
1542 
1543 #ifndef CLOCK_MONOTONIC
1544 #define CLOCK_MONOTONIC (1)
1545 #endif
1546 
1547 #ifdef __APPLE__
1548 void os::Bsd::clock_init() {
1549         // XXXDARWIN: Investigate replacement monotonic clock
1550 }
1551 #elif defined(_ALLBSD_SOURCE)
1552 void os::Bsd::clock_init() {
1553   struct timespec res;
1554   struct timespec tp;
1555   if (::clock_getres(CLOCK_MONOTONIC, &res) == 0 &&
1556       ::clock_gettime(CLOCK_MONOTONIC, &tp)  == 0) {
1557     // yes, monotonic clock is supported
1558     _clock_gettime = ::clock_gettime;
1559   }
1560 }
1561 #else
1562 void os::Bsd::clock_init() {
1563   // we do dlopen's in this particular order due to bug in bsd
1564   // dynamical loader (see 6348968) leading to crash on exit
1565   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1566   if (handle == NULL) {
1567     handle = dlopen("librt.so", RTLD_LAZY);
1568   }
1569 
1570   if (handle) {
1571     int (*clock_getres_func)(clockid_t, struct timespec*) =
1572            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1573     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1574            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1575     if (clock_getres_func && clock_gettime_func) {
1576       // See if monotonic clock is supported by the kernel. Note that some
1577       // early implementations simply return kernel jiffies (updated every
1578       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1579       // for nano time (though the monotonic property is still nice to have).
1580       // It's fixed in newer kernels, however clock_getres() still returns
1581       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1582       // resolution for now. Hopefully as people move to new kernels, this
1583       // won't be a problem.
1584       struct timespec res;
1585       struct timespec tp;
1586       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1587           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1588         // yes, monotonic clock is supported
1589         _clock_gettime = clock_gettime_func;
1590       } else {
1591         // close librt if there is no monotonic clock
1592         dlclose(handle);
1593       }
1594     }
1595   }
1596 }
1597 #endif
1598 
1599 #ifndef _ALLBSD_SOURCE
1600 #ifndef SYS_clock_getres
1601 
1602 #if defined(IA32) || defined(AMD64)
1603 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1604 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1605 #else
1606 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1607 #define sys_clock_getres(x,y)  -1
1608 #endif
1609 
1610 #else
1611 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1612 #endif
1613 
1614 void os::Bsd::fast_thread_clock_init() {
1615   if (!UseBsdPosixThreadCPUClocks) {
1616     return;
1617   }
1618   clockid_t clockid;
1619   struct timespec tp;
1620   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1621       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1622 
1623   // Switch to using fast clocks for thread cpu time if
1624   // the sys_clock_getres() returns 0 error code.
1625   // Note, that some kernels may support the current thread
1626   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1627   // returned by the pthread_getcpuclockid().
1628   // If the fast Posix clocks are supported then the sys_clock_getres()
1629   // must return at least tp.tv_sec == 0 which means a resolution
1630   // better than 1 sec. This is extra check for reliability.
1631 
1632   if(pthread_getcpuclockid_func &&
1633      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1634      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1635 
1636     _supports_fast_thread_cpu_time = true;
1637     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1638   }
1639 }
1640 #endif
1641 
1642 jlong os::javaTimeNanos() {
1643   if (Bsd::supports_monotonic_clock()) {
1644     struct timespec tp;
1645     int status = Bsd::clock_gettime(CLOCK_MONOTONIC, &tp);
1646     assert(status == 0, "gettime error");
1647     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1648     return result;
1649   } else {
1650     timeval time;
1651     int status = gettimeofday(&time, NULL);
1652     assert(status != -1, "bsd error");
1653     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1654     return 1000 * usecs;
1655   }
1656 }
1657 
1658 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1659   if (Bsd::supports_monotonic_clock()) {
1660     info_ptr->max_value = ALL_64_BITS;
1661 
1662     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1663     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1664     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1665   } else {
1666     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1667     info_ptr->max_value = ALL_64_BITS;
1668 
1669     // gettimeofday is a real time clock so it skips
1670     info_ptr->may_skip_backward = true;
1671     info_ptr->may_skip_forward = true;
1672   }
1673 
1674   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1675 }
1676 
1677 // Return the real, user, and system times in seconds from an
1678 // arbitrary fixed point in the past.
1679 bool os::getTimesSecs(double* process_real_time,
1680                       double* process_user_time,
1681                       double* process_system_time) {
1682   struct tms ticks;
1683   clock_t real_ticks = times(&ticks);
1684 
1685   if (real_ticks == (clock_t) (-1)) {
1686     return false;
1687   } else {
1688     double ticks_per_second = (double) clock_tics_per_sec;
1689     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1690     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1691     *process_real_time = ((double) real_ticks) / ticks_per_second;
1692 
1693     return true;
1694   }
1695 }
1696 
1697 
1698 char * os::local_time_string(char *buf, size_t buflen) {
1699   struct tm t;
1700   time_t long_time;
1701   time(&long_time);
1702   localtime_r(&long_time, &t);
1703   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1704                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1705                t.tm_hour, t.tm_min, t.tm_sec);
1706   return buf;
1707 }
1708 
1709 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1710   return localtime_r(clock, res);
1711 }
1712 
1713 ////////////////////////////////////////////////////////////////////////////////
1714 // runtime exit support
1715 
1716 // Note: os::shutdown() might be called very early during initialization, or
1717 // called from signal handler. Before adding something to os::shutdown(), make
1718 // sure it is async-safe and can handle partially initialized VM.
1719 void os::shutdown() {
1720 
1721   // allow PerfMemory to attempt cleanup of any persistent resources
1722   perfMemory_exit();
1723 
1724   // needs to remove object in file system
1725   AttachListener::abort();
1726 
1727   // flush buffered output, finish log files
1728   ostream_abort();
1729 
1730   // Check for abort hook
1731   abort_hook_t abort_hook = Arguments::abort_hook();
1732   if (abort_hook != NULL) {
1733     abort_hook();
1734   }
1735 
1736 }
1737 
1738 // Note: os::abort() might be called very early during initialization, or
1739 // called from signal handler. Before adding something to os::abort(), make
1740 // sure it is async-safe and can handle partially initialized VM.
1741 void os::abort(bool dump_core) {
1742   os::shutdown();
1743   if (dump_core) {
1744 #ifndef PRODUCT
1745     fdStream out(defaultStream::output_fd());
1746     out.print_raw("Current thread is ");
1747     char buf[16];
1748     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1749     out.print_raw_cr(buf);
1750     out.print_raw_cr("Dumping core ...");
1751 #endif
1752     ::abort(); // dump core
1753   }
1754 
1755   ::exit(1);
1756 }
1757 
1758 // Die immediately, no exit hook, no abort hook, no cleanup.
1759 void os::die() {
1760   // _exit() on BsdThreads only kills current thread
1761   ::abort();
1762 }
1763 
1764 // unused on bsd for now.
1765 void os::set_error_file(const char *logfile) {}
1766 
1767 
1768 // This method is a copy of JDK's sysGetLastErrorString
1769 // from src/solaris/hpi/src/system_md.c
1770 
1771 size_t os::lasterror(char *buf, size_t len) {
1772 
1773   if (errno == 0)  return 0;
1774 
1775   const char *s = ::strerror(errno);
1776   size_t n = ::strlen(s);
1777   if (n >= len) {
1778     n = len - 1;
1779   }
1780   ::strncpy(buf, s, n);
1781   buf[n] = '\0';
1782   return n;
1783 }
1784 
1785 intx os::current_thread_id() { return (intx)pthread_self(); }
1786 int os::current_process_id() {
1787 
1788   // Under the old bsd thread library, bsd gives each thread
1789   // its own process id. Because of this each thread will return
1790   // a different pid if this method were to return the result
1791   // of getpid(2). Bsd provides no api that returns the pid
1792   // of the launcher thread for the vm. This implementation
1793   // returns a unique pid, the pid of the launcher thread
1794   // that starts the vm 'process'.
1795 
1796   // Under the NPTL, getpid() returns the same pid as the
1797   // launcher thread rather than a unique pid per thread.
1798   // Use gettid() if you want the old pre NPTL behaviour.
1799 
1800   // if you are looking for the result of a call to getpid() that
1801   // returns a unique pid for the calling thread, then look at the
1802   // OSThread::thread_id() method in osThread_bsd.hpp file
1803 
1804   return (int)(_initial_pid ? _initial_pid : getpid());
1805 }
1806 
1807 // DLL functions
1808 
1809 #define JNI_LIB_PREFIX "lib"
1810 #ifdef __APPLE__
1811 #define JNI_LIB_SUFFIX ".dylib"
1812 #else
1813 #define JNI_LIB_SUFFIX ".so"
1814 #endif
1815 
1816 const char* os::dll_file_extension() { return JNI_LIB_SUFFIX; }
1817 
1818 // This must be hard coded because it's the system's temporary
1819 // directory not the java application's temp directory, ala java.io.tmpdir.
1820 #ifdef __APPLE__
1821 // macosx has a secure per-user temporary directory
1822 char temp_path_storage[PATH_MAX];
1823 const char* os::get_temp_directory() {
1824   static char *temp_path = NULL;
1825   if (temp_path == NULL) {
1826     int pathSize = confstr(_CS_DARWIN_USER_TEMP_DIR, temp_path_storage, PATH_MAX);
1827     if (pathSize == 0 || pathSize > PATH_MAX) {
1828       strlcpy(temp_path_storage, "/tmp/", sizeof(temp_path_storage));
1829     }
1830     temp_path = temp_path_storage;
1831   }
1832   return temp_path;
1833 }
1834 #else /* __APPLE__ */
1835 const char* os::get_temp_directory() { return "/tmp"; }
1836 #endif /* __APPLE__ */
1837 
1838 static bool file_exists(const char* filename) {
1839   struct stat statbuf;
1840   if (filename == NULL || strlen(filename) == 0) {
1841     return false;
1842   }
1843   return os::stat(filename, &statbuf) == 0;
1844 }
1845 
1846 void os::dll_build_name(char* buffer, size_t buflen,
1847                         const char* pname, const char* fname) {
1848   // Copied from libhpi
1849   const size_t pnamelen = pname ? strlen(pname) : 0;
1850 
1851   // Quietly truncate on buffer overflow.  Should be an error.
1852   if (pnamelen + strlen(fname) + strlen(JNI_LIB_PREFIX) + strlen(JNI_LIB_SUFFIX) + 2 > buflen) {
1853       *buffer = '\0';
1854       return;
1855   }
1856 
1857   if (pnamelen == 0) {
1858     snprintf(buffer, buflen, JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX, fname);
1859   } else if (strchr(pname, *os::path_separator()) != NULL) {
1860     int n;
1861     char** pelements = split_path(pname, &n);
1862     for (int i = 0 ; i < n ; i++) {
1863       // Really shouldn't be NULL, but check can't hurt
1864       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1865         continue; // skip the empty path values
1866       }
1867       snprintf(buffer, buflen, "%s/" JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX,
1868           pelements[i], fname);
1869       if (file_exists(buffer)) {
1870         break;
1871       }
1872     }
1873     // release the storage
1874     for (int i = 0 ; i < n ; i++) {
1875       if (pelements[i] != NULL) {
1876         FREE_C_HEAP_ARRAY(char, pelements[i]);
1877       }
1878     }
1879     if (pelements != NULL) {
1880       FREE_C_HEAP_ARRAY(char*, pelements);
1881     }
1882   } else {
1883     snprintf(buffer, buflen, "%s/" JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX, pname, fname);
1884   }
1885 }
1886 
1887 const char* os::get_current_directory(char *buf, int buflen) {
1888   return getcwd(buf, buflen);
1889 }
1890 
1891 // check if addr is inside libjvm[_g].so
1892 bool os::address_is_in_vm(address addr) {
1893   static address libjvm_base_addr;
1894   Dl_info dlinfo;
1895 
1896   if (libjvm_base_addr == NULL) {
1897     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1898     libjvm_base_addr = (address)dlinfo.dli_fbase;
1899     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1900   }
1901 
1902   if (dladdr((void *)addr, &dlinfo)) {
1903     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1904   }
1905 
1906   return false;
1907 }
1908 
1909 bool os::dll_address_to_function_name(address addr, char *buf,
1910                                       int buflen, int *offset) {
1911   Dl_info dlinfo;
1912 
1913   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1914     if (buf != NULL) {
1915       if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1916         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1917       }
1918     }
1919     if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1920     return true;
1921   } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1922     if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1923        dlinfo.dli_fname, buf, buflen, offset) == Decoder::no_error) {
1924        return true;
1925     }
1926   }
1927 
1928   if (buf != NULL) buf[0] = '\0';
1929   if (offset != NULL) *offset = -1;
1930   return false;
1931 }
1932 
1933 #ifdef _ALLBSD_SOURCE
1934 // ported from solaris version
1935 bool os::dll_address_to_library_name(address addr, char* buf,
1936                                      int buflen, int* offset) {
1937   Dl_info dlinfo;
1938 
1939   if (dladdr((void*)addr, &dlinfo)){
1940      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1941      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1942      return true;
1943   } else {
1944      if (buf) buf[0] = '\0';
1945      if (offset) *offset = -1;
1946      return false;
1947   }
1948 }
1949 #else
1950 struct _address_to_library_name {
1951   address addr;          // input : memory address
1952   size_t  buflen;        //         size of fname
1953   char*   fname;         // output: library name
1954   address base;          //         library base addr
1955 };
1956 
1957 static int address_to_library_name_callback(struct dl_phdr_info *info,
1958                                             size_t size, void *data) {
1959   int i;
1960   bool found = false;
1961   address libbase = NULL;
1962   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1963 
1964   // iterate through all loadable segments
1965   for (i = 0; i < info->dlpi_phnum; i++) {
1966     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1967     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1968       // base address of a library is the lowest address of its loaded
1969       // segments.
1970       if (libbase == NULL || libbase > segbase) {
1971         libbase = segbase;
1972       }
1973       // see if 'addr' is within current segment
1974       if (segbase <= d->addr &&
1975           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1976         found = true;
1977       }
1978     }
1979   }
1980 
1981   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1982   // so dll_address_to_library_name() can fall through to use dladdr() which
1983   // can figure out executable name from argv[0].
1984   if (found && info->dlpi_name && info->dlpi_name[0]) {
1985     d->base = libbase;
1986     if (d->fname) {
1987       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1988     }
1989     return 1;
1990   }
1991   return 0;
1992 }
1993 
1994 bool os::dll_address_to_library_name(address addr, char* buf,
1995                                      int buflen, int* offset) {
1996   Dl_info dlinfo;
1997   struct _address_to_library_name data;
1998 
1999   // There is a bug in old glibc dladdr() implementation that it could resolve
2000   // to wrong library name if the .so file has a base address != NULL. Here
2001   // we iterate through the program headers of all loaded libraries to find
2002   // out which library 'addr' really belongs to. This workaround can be
2003   // removed once the minimum requirement for glibc is moved to 2.3.x.
2004   data.addr = addr;
2005   data.fname = buf;
2006   data.buflen = buflen;
2007   data.base = NULL;
2008   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
2009 
2010   if (rslt) {
2011      // buf already contains library name
2012      if (offset) *offset = addr - data.base;
2013      return true;
2014   } else if (dladdr((void*)addr, &dlinfo)){
2015      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
2016      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
2017      return true;
2018   } else {
2019      if (buf) buf[0] = '\0';
2020      if (offset) *offset = -1;
2021      return false;
2022   }
2023 }
2024 #endif
2025 
2026   // Loads .dll/.so and
2027   // in case of error it checks if .dll/.so was built for the
2028   // same architecture as Hotspot is running on
2029 
2030 #ifdef __APPLE__
2031 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
2032   void * result= ::dlopen(filename, RTLD_LAZY);
2033   if (result != NULL) {
2034     // Successful loading
2035     return result;
2036   }
2037 
2038   // Read system error message into ebuf
2039   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2040   ebuf[ebuflen-1]='\0';
2041 
2042   return NULL;
2043 }
2044 #else
2045 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
2046 {
2047   void * result= ::dlopen(filename, RTLD_LAZY);
2048   if (result != NULL) {
2049     // Successful loading
2050     return result;
2051   }
2052 
2053   Elf32_Ehdr elf_head;
2054 
2055   // Read system error message into ebuf
2056   // It may or may not be overwritten below
2057   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2058   ebuf[ebuflen-1]='\0';
2059   int diag_msg_max_length=ebuflen-strlen(ebuf);
2060   char* diag_msg_buf=ebuf+strlen(ebuf);
2061 
2062   if (diag_msg_max_length==0) {
2063     // No more space in ebuf for additional diagnostics message
2064     return NULL;
2065   }
2066 
2067 
2068   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
2069 
2070   if (file_descriptor < 0) {
2071     // Can't open library, report dlerror() message
2072     return NULL;
2073   }
2074 
2075   bool failed_to_read_elf_head=
2076     (sizeof(elf_head)!=
2077         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
2078 
2079   ::close(file_descriptor);
2080   if (failed_to_read_elf_head) {
2081     // file i/o error - report dlerror() msg
2082     return NULL;
2083   }
2084 
2085   typedef struct {
2086     Elf32_Half  code;         // Actual value as defined in elf.h
2087     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
2088     char        elf_class;    // 32 or 64 bit
2089     char        endianess;    // MSB or LSB
2090     char*       name;         // String representation
2091   } arch_t;
2092 
2093   #ifndef EM_486
2094   #define EM_486          6               /* Intel 80486 */
2095   #endif
2096 
2097   #ifndef EM_MIPS_RS3_LE
2098   #define EM_MIPS_RS3_LE  10              /* MIPS */
2099   #endif
2100 
2101   #ifndef EM_PPC64
2102   #define EM_PPC64        21              /* PowerPC64 */
2103   #endif
2104 
2105   #ifndef EM_S390
2106   #define EM_S390         22              /* IBM System/390 */
2107   #endif
2108 
2109   #ifndef EM_IA_64
2110   #define EM_IA_64        50              /* HP/Intel IA-64 */
2111   #endif
2112 
2113   #ifndef EM_X86_64
2114   #define EM_X86_64       62              /* AMD x86-64 */
2115   #endif
2116 
2117   static const arch_t arch_array[]={
2118     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2119     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2120     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2121     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2122     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2123     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2124     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2125     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2126     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
2127     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
2128     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
2129     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
2130     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
2131     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
2132     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
2133     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
2134   };
2135 
2136   #if  (defined IA32)
2137     static  Elf32_Half running_arch_code=EM_386;
2138   #elif   (defined AMD64)
2139     static  Elf32_Half running_arch_code=EM_X86_64;
2140   #elif  (defined IA64)
2141     static  Elf32_Half running_arch_code=EM_IA_64;
2142   #elif  (defined __sparc) && (defined _LP64)
2143     static  Elf32_Half running_arch_code=EM_SPARCV9;
2144   #elif  (defined __sparc) && (!defined _LP64)
2145     static  Elf32_Half running_arch_code=EM_SPARC;
2146   #elif  (defined __powerpc64__)
2147     static  Elf32_Half running_arch_code=EM_PPC64;
2148   #elif  (defined __powerpc__)
2149     static  Elf32_Half running_arch_code=EM_PPC;
2150   #elif  (defined ARM)
2151     static  Elf32_Half running_arch_code=EM_ARM;
2152   #elif  (defined S390)
2153     static  Elf32_Half running_arch_code=EM_S390;
2154   #elif  (defined ALPHA)
2155     static  Elf32_Half running_arch_code=EM_ALPHA;
2156   #elif  (defined MIPSEL)
2157     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
2158   #elif  (defined PARISC)
2159     static  Elf32_Half running_arch_code=EM_PARISC;
2160   #elif  (defined MIPS)
2161     static  Elf32_Half running_arch_code=EM_MIPS;
2162   #elif  (defined M68K)
2163     static  Elf32_Half running_arch_code=EM_68K;
2164   #else
2165     #error Method os::dll_load requires that one of following is defined:\
2166          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2167   #endif
2168 
2169   // Identify compatability class for VM's architecture and library's architecture
2170   // Obtain string descriptions for architectures
2171 
2172   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2173   int running_arch_index=-1;
2174 
2175   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2176     if (running_arch_code == arch_array[i].code) {
2177       running_arch_index    = i;
2178     }
2179     if (lib_arch.code == arch_array[i].code) {
2180       lib_arch.compat_class = arch_array[i].compat_class;
2181       lib_arch.name         = arch_array[i].name;
2182     }
2183   }
2184 
2185   assert(running_arch_index != -1,
2186     "Didn't find running architecture code (running_arch_code) in arch_array");
2187   if (running_arch_index == -1) {
2188     // Even though running architecture detection failed
2189     // we may still continue with reporting dlerror() message
2190     return NULL;
2191   }
2192 
2193   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2194     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2195     return NULL;
2196   }
2197 
2198 #ifndef S390
2199   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2200     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2201     return NULL;
2202   }
2203 #endif // !S390
2204 
2205   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2206     if ( lib_arch.name!=NULL ) {
2207       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2208         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2209         lib_arch.name, arch_array[running_arch_index].name);
2210     } else {
2211       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2212       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2213         lib_arch.code,
2214         arch_array[running_arch_index].name);
2215     }
2216   }
2217 
2218   return NULL;
2219 }
2220 #endif /* !__APPLE__ */
2221 
2222 // XXX: Do we need a lock around this as per Linux?
2223 void* os::dll_lookup(void* handle, const char* name) {
2224   return dlsym(handle, name);
2225 }
2226 
2227 
2228 static bool _print_ascii_file(const char* filename, outputStream* st) {
2229   int fd = ::open(filename, O_RDONLY);
2230   if (fd == -1) {
2231      return false;
2232   }
2233 
2234   char buf[32];
2235   int bytes;
2236   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2237     st->print_raw(buf, bytes);
2238   }
2239 
2240   ::close(fd);
2241 
2242   return true;
2243 }
2244 
2245 void os::print_dll_info(outputStream *st) {
2246    st->print_cr("Dynamic libraries:");
2247 #ifdef _ALLBSD_SOURCE
2248 #ifdef RTLD_DI_LINKMAP
2249     Dl_info dli;
2250     void *handle;
2251     Link_map *map;
2252     Link_map *p;
2253 
2254     if (!dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli)) {
2255         st->print_cr("Error: Cannot print dynamic libraries.");
2256         return;
2257     }
2258     handle = dlopen(dli.dli_fname, RTLD_LAZY);
2259     if (handle == NULL) {
2260         st->print_cr("Error: Cannot print dynamic libraries.");
2261         return;
2262     }
2263     dlinfo(handle, RTLD_DI_LINKMAP, &map);
2264     if (map == NULL) {
2265         st->print_cr("Error: Cannot print dynamic libraries.");
2266         return;
2267     }
2268 
2269     while (map->l_prev != NULL)
2270         map = map->l_prev;
2271 
2272     while (map != NULL) {
2273         st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
2274         map = map->l_next;
2275     }
2276 
2277     dlclose(handle);
2278 #elif defined(__APPLE__)
2279     uint32_t count;
2280     uint32_t i;
2281 
2282     count = _dyld_image_count();
2283     for (i = 1; i < count; i++) {
2284         const char *name = _dyld_get_image_name(i);
2285         intptr_t slide = _dyld_get_image_vmaddr_slide(i);
2286         st->print_cr(PTR_FORMAT " \t%s", slide, name);
2287     }
2288 #else
2289    st->print_cr("Error: Cannot print dynamic libraries.");
2290 #endif
2291 #else
2292    char fname[32];
2293    pid_t pid = os::Bsd::gettid();
2294 
2295    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2296 
2297    if (!_print_ascii_file(fname, st)) {
2298      st->print("Can not get library information for pid = %d\n", pid);
2299    }
2300 #endif
2301 }
2302 
2303 
2304 void os::print_os_info(outputStream* st) {
2305   st->print("OS:");
2306 
2307   // Try to identify popular distros.
2308   // Most Bsd distributions have /etc/XXX-release file, which contains
2309   // the OS version string. Some have more than one /etc/XXX-release file
2310   // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2311   // so the order is important.
2312   if (!_print_ascii_file("/etc/mandrake-release", st) &&
2313       !_print_ascii_file("/etc/sun-release", st) &&
2314       !_print_ascii_file("/etc/redhat-release", st) &&
2315       !_print_ascii_file("/etc/SuSE-release", st) &&
2316       !_print_ascii_file("/etc/turbobsd-release", st) &&
2317       !_print_ascii_file("/etc/gentoo-release", st) &&
2318       !_print_ascii_file("/etc/debian_version", st) &&
2319       !_print_ascii_file("/etc/ltib-release", st) &&
2320       !_print_ascii_file("/etc/angstrom-version", st)) {
2321       st->print("Bsd");
2322   }
2323   st->cr();
2324 
2325   // kernel
2326   st->print("uname:");
2327   struct utsname name;
2328   uname(&name);
2329   st->print(name.sysname); st->print(" ");
2330   st->print(name.release); st->print(" ");
2331   st->print(name.version); st->print(" ");
2332   st->print(name.machine);
2333   st->cr();
2334 
2335 #ifndef _ALLBSD_SOURCE
2336   // Print warning if unsafe chroot environment detected
2337   if (unsafe_chroot_detected) {
2338     st->print("WARNING!! ");
2339     st->print_cr(unstable_chroot_error);
2340   }
2341 
2342   // libc, pthread
2343   st->print("libc:");
2344   st->print(os::Bsd::glibc_version()); st->print(" ");
2345   st->print(os::Bsd::libpthread_version()); st->print(" ");
2346   if (os::Bsd::is_BsdThreads()) {
2347      st->print("(%s stack)", os::Bsd::is_floating_stack() ? "floating" : "fixed");
2348   }
2349   st->cr();
2350 #endif
2351 
2352   // rlimit
2353   st->print("rlimit:");
2354   struct rlimit rlim;
2355 
2356   st->print(" STACK ");
2357   getrlimit(RLIMIT_STACK, &rlim);
2358   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2359   else st->print("%uk", rlim.rlim_cur >> 10);
2360 
2361   st->print(", CORE ");
2362   getrlimit(RLIMIT_CORE, &rlim);
2363   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2364   else st->print("%uk", rlim.rlim_cur >> 10);
2365 
2366   st->print(", NPROC ");
2367   getrlimit(RLIMIT_NPROC, &rlim);
2368   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2369   else st->print("%d", rlim.rlim_cur);
2370 
2371   st->print(", NOFILE ");
2372   getrlimit(RLIMIT_NOFILE, &rlim);
2373   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2374   else st->print("%d", rlim.rlim_cur);
2375 
2376 #ifndef _ALLBSD_SOURCE
2377   st->print(", AS ");
2378   getrlimit(RLIMIT_AS, &rlim);
2379   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2380   else st->print("%uk", rlim.rlim_cur >> 10);
2381   st->cr();
2382 
2383   // load average
2384   st->print("load average:");
2385   double loadavg[3];
2386   os::loadavg(loadavg, 3);
2387   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2388   st->cr();
2389 #endif
2390 }
2391 
2392 void os::pd_print_cpu_info(outputStream* st) {
2393   // Nothing to do for now.
2394 }
2395 
2396 void os::print_memory_info(outputStream* st) {
2397 
2398   st->print("Memory:");
2399   st->print(" %dk page", os::vm_page_size()>>10);
2400 
2401 #ifndef _ALLBSD_SOURCE
2402   // values in struct sysinfo are "unsigned long"
2403   struct sysinfo si;
2404   sysinfo(&si);
2405 #endif
2406 
2407   st->print(", physical " UINT64_FORMAT "k",
2408             os::physical_memory() >> 10);
2409   st->print("(" UINT64_FORMAT "k free)",
2410             os::available_memory() >> 10);
2411 #ifndef _ALLBSD_SOURCE
2412   st->print(", swap " UINT64_FORMAT "k",
2413             ((jlong)si.totalswap * si.mem_unit) >> 10);
2414   st->print("(" UINT64_FORMAT "k free)",
2415             ((jlong)si.freeswap * si.mem_unit) >> 10);
2416 #endif
2417   st->cr();
2418 
2419   // meminfo
2420   st->print("\n/proc/meminfo:\n");
2421   _print_ascii_file("/proc/meminfo", st);
2422   st->cr();
2423 }
2424 
2425 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2426 // but they're the same for all the bsd arch that we support
2427 // and they're the same for solaris but there's no common place to put this.
2428 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2429                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2430                           "ILL_COPROC", "ILL_BADSTK" };
2431 
2432 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2433                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2434                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2435 
2436 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2437 
2438 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2439 
2440 void os::print_siginfo(outputStream* st, void* siginfo) {
2441   st->print("siginfo:");
2442 
2443   const int buflen = 100;
2444   char buf[buflen];
2445   siginfo_t *si = (siginfo_t*)siginfo;
2446   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2447   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2448     st->print("si_errno=%s", buf);
2449   } else {
2450     st->print("si_errno=%d", si->si_errno);
2451   }
2452   const int c = si->si_code;
2453   assert(c > 0, "unexpected si_code");
2454   switch (si->si_signo) {
2455   case SIGILL:
2456     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2457     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2458     break;
2459   case SIGFPE:
2460     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2461     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2462     break;
2463   case SIGSEGV:
2464     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2465     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2466     break;
2467   case SIGBUS:
2468     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2469     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2470     break;
2471   default:
2472     st->print(", si_code=%d", si->si_code);
2473     // no si_addr
2474   }
2475 
2476   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2477       UseSharedSpaces) {
2478     FileMapInfo* mapinfo = FileMapInfo::current_info();
2479     if (mapinfo->is_in_shared_space(si->si_addr)) {
2480       st->print("\n\nError accessing class data sharing archive."   \
2481                 " Mapped file inaccessible during execution, "      \
2482                 " possible disk/network problem.");
2483     }
2484   }
2485   st->cr();
2486 }
2487 
2488 
2489 static void print_signal_handler(outputStream* st, int sig,
2490                                  char* buf, size_t buflen);
2491 
2492 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2493   st->print_cr("Signal Handlers:");
2494   print_signal_handler(st, SIGSEGV, buf, buflen);
2495   print_signal_handler(st, SIGBUS , buf, buflen);
2496   print_signal_handler(st, SIGFPE , buf, buflen);
2497   print_signal_handler(st, SIGPIPE, buf, buflen);
2498   print_signal_handler(st, SIGXFSZ, buf, buflen);
2499   print_signal_handler(st, SIGILL , buf, buflen);
2500   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2501   print_signal_handler(st, SR_signum, buf, buflen);
2502   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2503   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2504   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2505   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2506 }
2507 
2508 static char saved_jvm_path[MAXPATHLEN] = {0};
2509 
2510 // Find the full path to the current module, libjvm.so or libjvm_g.so
2511 void os::jvm_path(char *buf, jint buflen) {
2512   // Error checking.
2513   if (buflen < MAXPATHLEN) {
2514     assert(false, "must use a large-enough buffer");
2515     buf[0] = '\0';
2516     return;
2517   }
2518   // Lazy resolve the path to current module.
2519   if (saved_jvm_path[0] != 0) {
2520     strcpy(buf, saved_jvm_path);
2521     return;
2522   }
2523 
2524   char dli_fname[MAXPATHLEN];
2525   bool ret = dll_address_to_library_name(
2526                 CAST_FROM_FN_PTR(address, os::jvm_path),
2527                 dli_fname, sizeof(dli_fname), NULL);
2528   assert(ret != 0, "cannot locate libjvm");
2529   char *rp = realpath(dli_fname, buf);
2530   if (rp == NULL)
2531     return;
2532 
2533   if (Arguments::created_by_gamma_launcher()) {
2534     // Support for the gamma launcher.  Typical value for buf is
2535     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2536     // the right place in the string, then assume we are installed in a JDK and
2537     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2538     // up the path so it looks like libjvm.so is installed there (append a
2539     // fake suffix hotspot/libjvm.so).
2540     const char *p = buf + strlen(buf) - 1;
2541     for (int count = 0; p > buf && count < 5; ++count) {
2542       for (--p; p > buf && *p != '/'; --p)
2543         /* empty */ ;
2544     }
2545 
2546     if (strncmp(p, "/jre/lib/", 9) != 0) {
2547       // Look for JAVA_HOME in the environment.
2548       char* java_home_var = ::getenv("JAVA_HOME");
2549       if (java_home_var != NULL && java_home_var[0] != 0) {
2550         char* jrelib_p;
2551         int len;
2552 
2553         // Check the current module name "libjvm.so" or "libjvm_g.so".
2554         p = strrchr(buf, '/');
2555         assert(strstr(p, "/libjvm") == p, "invalid library name");
2556         p = strstr(p, "_g") ? "_g" : "";
2557 
2558         rp = realpath(java_home_var, buf);
2559         if (rp == NULL)
2560           return;
2561 
2562         // determine if this is a legacy image or modules image
2563         // modules image doesn't have "jre" subdirectory
2564         len = strlen(buf);
2565         jrelib_p = buf + len;
2566         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2567         if (0 != access(buf, F_OK)) {
2568           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2569         }
2570 
2571         if (0 == access(buf, F_OK)) {
2572           // Use current module name "libjvm[_g].so" instead of
2573           // "libjvm"debug_only("_g")".so" since for fastdebug version
2574           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2575           len = strlen(buf);
2576           snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2577         } else {
2578           // Go back to path of .so
2579           rp = realpath(dli_fname, buf);
2580           if (rp == NULL)
2581             return;
2582         }
2583       }
2584     }
2585   }
2586 
2587   strcpy(saved_jvm_path, buf);
2588 }
2589 
2590 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2591   // no prefix required, not even "_"
2592 }
2593 
2594 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2595   // no suffix required
2596 }
2597 
2598 ////////////////////////////////////////////////////////////////////////////////
2599 // sun.misc.Signal support
2600 
2601 static volatile jint sigint_count = 0;
2602 
2603 static void
2604 UserHandler(int sig, void *siginfo, void *context) {
2605   // 4511530 - sem_post is serialized and handled by the manager thread. When
2606   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2607   // don't want to flood the manager thread with sem_post requests.
2608   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2609       return;
2610 
2611   // Ctrl-C is pressed during error reporting, likely because the error
2612   // handler fails to abort. Let VM die immediately.
2613   if (sig == SIGINT && is_error_reported()) {
2614      os::die();
2615   }
2616 
2617   os::signal_notify(sig);
2618 }
2619 
2620 void* os::user_handler() {
2621   return CAST_FROM_FN_PTR(void*, UserHandler);
2622 }
2623 
2624 extern "C" {
2625   typedef void (*sa_handler_t)(int);
2626   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2627 }
2628 
2629 void* os::signal(int signal_number, void* handler) {
2630   struct sigaction sigAct, oldSigAct;
2631 
2632   sigfillset(&(sigAct.sa_mask));
2633   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2634   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2635 
2636   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2637     // -1 means registration failed
2638     return (void *)-1;
2639   }
2640 
2641   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2642 }
2643 
2644 void os::signal_raise(int signal_number) {
2645   ::raise(signal_number);
2646 }
2647 
2648 /*
2649  * The following code is moved from os.cpp for making this
2650  * code platform specific, which it is by its very nature.
2651  */
2652 
2653 // Will be modified when max signal is changed to be dynamic
2654 int os::sigexitnum_pd() {
2655   return NSIG;
2656 }
2657 
2658 // a counter for each possible signal value
2659 static volatile jint pending_signals[NSIG+1] = { 0 };
2660 
2661 // Bsd(POSIX) specific hand shaking semaphore.
2662 #ifdef __APPLE__
2663 static semaphore_t sig_sem;
2664 #define SEM_INIT(sem, value)    semaphore_create(mach_task_self(), &sem, SYNC_POLICY_FIFO, value)
2665 #define SEM_WAIT(sem)           semaphore_wait(sem);
2666 #define SEM_POST(sem)           semaphore_signal(sem);
2667 #else
2668 static sem_t sig_sem;
2669 #define SEM_INIT(sem, value)    sem_init(&sem, 0, value)
2670 #define SEM_WAIT(sem)           sem_wait(&sem);
2671 #define SEM_POST(sem)           sem_post(&sem);
2672 #endif
2673 
2674 void os::signal_init_pd() {
2675   // Initialize signal structures
2676   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2677 
2678   // Initialize signal semaphore
2679   ::SEM_INIT(sig_sem, 0);
2680 }
2681 
2682 void os::signal_notify(int sig) {
2683   Atomic::inc(&pending_signals[sig]);
2684   ::SEM_POST(sig_sem);
2685 }
2686 
2687 static int check_pending_signals(bool wait) {
2688   Atomic::store(0, &sigint_count);
2689   for (;;) {
2690     for (int i = 0; i < NSIG + 1; i++) {
2691       jint n = pending_signals[i];
2692       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2693         return i;
2694       }
2695     }
2696     if (!wait) {
2697       return -1;
2698     }
2699     JavaThread *thread = JavaThread::current();
2700     ThreadBlockInVM tbivm(thread);
2701 
2702     bool threadIsSuspended;
2703     do {
2704       thread->set_suspend_equivalent();
2705       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2706       ::SEM_WAIT(sig_sem);
2707 
2708       // were we externally suspended while we were waiting?
2709       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2710       if (threadIsSuspended) {
2711         //
2712         // The semaphore has been incremented, but while we were waiting
2713         // another thread suspended us. We don't want to continue running
2714         // while suspended because that would surprise the thread that
2715         // suspended us.
2716         //
2717         ::SEM_POST(sig_sem);
2718 
2719         thread->java_suspend_self();
2720       }
2721     } while (threadIsSuspended);
2722   }
2723 }
2724 
2725 int os::signal_lookup() {
2726   return check_pending_signals(false);
2727 }
2728 
2729 int os::signal_wait() {
2730   return check_pending_signals(true);
2731 }
2732 
2733 ////////////////////////////////////////////////////////////////////////////////
2734 // Virtual Memory
2735 
2736 int os::vm_page_size() {
2737   // Seems redundant as all get out
2738   assert(os::Bsd::page_size() != -1, "must call os::init");
2739   return os::Bsd::page_size();
2740 }
2741 
2742 // Solaris allocates memory by pages.
2743 int os::vm_allocation_granularity() {
2744   assert(os::Bsd::page_size() != -1, "must call os::init");
2745   return os::Bsd::page_size();
2746 }
2747 
2748 // Rationale behind this function:
2749 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2750 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2751 //  samples for JITted code. Here we create private executable mapping over the code cache
2752 //  and then we can use standard (well, almost, as mapping can change) way to provide
2753 //  info for the reporting script by storing timestamp and location of symbol
2754 void bsd_wrap_code(char* base, size_t size) {
2755   static volatile jint cnt = 0;
2756 
2757   if (!UseOprofile) {
2758     return;
2759   }
2760 
2761   char buf[PATH_MAX + 1];
2762   int num = Atomic::add(1, &cnt);
2763 
2764   snprintf(buf, PATH_MAX + 1, "%s/hs-vm-%d-%d",
2765            os::get_temp_directory(), os::current_process_id(), num);
2766   unlink(buf);
2767 
2768   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2769 
2770   if (fd != -1) {
2771     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2772     if (rv != (off_t)-1) {
2773       if (::write(fd, "", 1) == 1) {
2774         mmap(base, size,
2775              PROT_READ|PROT_WRITE|PROT_EXEC,
2776              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2777       }
2778     }
2779     ::close(fd);
2780     unlink(buf);
2781   }
2782 }
2783 
2784 // NOTE: Bsd kernel does not really reserve the pages for us.
2785 //       All it does is to check if there are enough free pages
2786 //       left at the time of mmap(). This could be a potential
2787 //       problem.
2788 bool os::commit_memory(char* addr, size_t size, bool exec) {
2789   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2790 #ifdef __OpenBSD__
2791   // XXX: Work-around mmap/MAP_FIXED bug temporarily on OpenBSD
2792   return ::mprotect(addr, size, prot) == 0;
2793 #else
2794   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2795                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2796   return res != (uintptr_t) MAP_FAILED;
2797 #endif
2798 }
2799 
2800 #ifndef _ALLBSD_SOURCE
2801 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2802 #ifndef MAP_HUGETLB
2803 #define MAP_HUGETLB 0x40000
2804 #endif
2805 
2806 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2807 #ifndef MADV_HUGEPAGE
2808 #define MADV_HUGEPAGE 14
2809 #endif
2810 #endif
2811 
2812 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2813                        bool exec) {
2814 #ifndef _ALLBSD_SOURCE
2815   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2816     int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2817     uintptr_t res =
2818       (uintptr_t) ::mmap(addr, size, prot,
2819                          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2820                          -1, 0);
2821     return res != (uintptr_t) MAP_FAILED;
2822   }
2823 #endif
2824 
2825   return commit_memory(addr, size, exec);
2826 }
2827 
2828 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2829 #ifndef _ALLBSD_SOURCE
2830   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2831     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2832     // be supported or the memory may already be backed by huge pages.
2833     ::madvise(addr, bytes, MADV_HUGEPAGE);
2834   }
2835 #endif
2836 }
2837 
2838 void os::free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2839   ::madvise(addr, bytes, MADV_DONTNEED);
2840 }
2841 
2842 void os::numa_make_global(char *addr, size_t bytes) {
2843 }
2844 
2845 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2846 }
2847 
2848 bool os::numa_topology_changed()   { return false; }
2849 
2850 size_t os::numa_get_groups_num() {
2851   return 1;
2852 }
2853 
2854 int os::numa_get_group_id() {
2855   return 0;
2856 }
2857 
2858 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2859   if (size > 0) {
2860     ids[0] = 0;
2861     return 1;
2862   }
2863   return 0;
2864 }
2865 
2866 bool os::get_page_info(char *start, page_info* info) {
2867   return false;
2868 }
2869 
2870 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2871   return end;
2872 }
2873 
2874 #ifndef _ALLBSD_SOURCE
2875 // Something to do with the numa-aware allocator needs these symbols
2876 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2877 extern "C" JNIEXPORT void numa_error(char *where) { }
2878 extern "C" JNIEXPORT int fork1() { return fork(); }
2879 
2880 
2881 // If we are running with libnuma version > 2, then we should
2882 // be trying to use symbols with versions 1.1
2883 // If we are running with earlier version, which did not have symbol versions,
2884 // we should use the base version.
2885 void* os::Bsd::libnuma_dlsym(void* handle, const char *name) {
2886   void *f = dlvsym(handle, name, "libnuma_1.1");
2887   if (f == NULL) {
2888     f = dlsym(handle, name);
2889   }
2890   return f;
2891 }
2892 
2893 bool os::Bsd::libnuma_init() {
2894   // sched_getcpu() should be in libc.
2895   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2896                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2897 
2898   if (sched_getcpu() != -1) { // Does it work?
2899     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2900     if (handle != NULL) {
2901       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2902                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2903       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2904                                        libnuma_dlsym(handle, "numa_max_node")));
2905       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2906                                         libnuma_dlsym(handle, "numa_available")));
2907       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2908                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2909       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2910                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2911 
2912 
2913       if (numa_available() != -1) {
2914         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2915         // Create a cpu -> node mapping
2916         _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2917         rebuild_cpu_to_node_map();
2918         return true;
2919       }
2920     }
2921   }
2922   return false;
2923 }
2924 
2925 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2926 // The table is later used in get_node_by_cpu().
2927 void os::Bsd::rebuild_cpu_to_node_map() {
2928   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2929                               // in libnuma (possible values are starting from 16,
2930                               // and continuing up with every other power of 2, but less
2931                               // than the maximum number of CPUs supported by kernel), and
2932                               // is a subject to change (in libnuma version 2 the requirements
2933                               // are more reasonable) we'll just hardcode the number they use
2934                               // in the library.
2935   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2936 
2937   size_t cpu_num = os::active_processor_count();
2938   size_t cpu_map_size = NCPUS / BitsPerCLong;
2939   size_t cpu_map_valid_size =
2940     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2941 
2942   cpu_to_node()->clear();
2943   cpu_to_node()->at_grow(cpu_num - 1);
2944   size_t node_num = numa_get_groups_num();
2945 
2946   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2947   for (size_t i = 0; i < node_num; i++) {
2948     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2949       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2950         if (cpu_map[j] != 0) {
2951           for (size_t k = 0; k < BitsPerCLong; k++) {
2952             if (cpu_map[j] & (1UL << k)) {
2953               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2954             }
2955           }
2956         }
2957       }
2958     }
2959   }
2960   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2961 }
2962 
2963 int os::Bsd::get_node_by_cpu(int cpu_id) {
2964   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2965     return cpu_to_node()->at(cpu_id);
2966   }
2967   return -1;
2968 }
2969 
2970 GrowableArray<int>* os::Bsd::_cpu_to_node;
2971 os::Bsd::sched_getcpu_func_t os::Bsd::_sched_getcpu;
2972 os::Bsd::numa_node_to_cpus_func_t os::Bsd::_numa_node_to_cpus;
2973 os::Bsd::numa_max_node_func_t os::Bsd::_numa_max_node;
2974 os::Bsd::numa_available_func_t os::Bsd::_numa_available;
2975 os::Bsd::numa_tonode_memory_func_t os::Bsd::_numa_tonode_memory;
2976 os::Bsd::numa_interleave_memory_func_t os::Bsd::_numa_interleave_memory;
2977 unsigned long* os::Bsd::_numa_all_nodes;
2978 #endif
2979 
2980 bool os::uncommit_memory(char* addr, size_t size) {
2981 #ifdef __OpenBSD__
2982   // XXX: Work-around mmap/MAP_FIXED bug temporarily on OpenBSD
2983   return ::mprotect(addr, size, PROT_NONE) == 0;
2984 #else
2985   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2986                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2987   return res  != (uintptr_t) MAP_FAILED;
2988 #endif
2989 }
2990 
2991 bool os::create_stack_guard_pages(char* addr, size_t size) {
2992   return os::commit_memory(addr, size);
2993 }
2994 
2995 // If this is a growable mapping, remove the guard pages entirely by
2996 // munmap()ping them.  If not, just call uncommit_memory().
2997 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2998   return os::uncommit_memory(addr, size);
2999 }
3000 
3001 static address _highest_vm_reserved_address = NULL;
3002 
3003 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3004 // at 'requested_addr'. If there are existing memory mappings at the same
3005 // location, however, they will be overwritten. If 'fixed' is false,
3006 // 'requested_addr' is only treated as a hint, the return value may or
3007 // may not start from the requested address. Unlike Bsd mmap(), this
3008 // function returns NULL to indicate failure.
3009 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3010   char * addr;
3011   int flags;
3012 
3013   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3014   if (fixed) {
3015     assert((uintptr_t)requested_addr % os::Bsd::page_size() == 0, "unaligned address");
3016     flags |= MAP_FIXED;
3017   }
3018 
3019   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
3020   // to PROT_EXEC if executable when we commit the page.
3021   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
3022                        flags, -1, 0);
3023 
3024   if (addr != MAP_FAILED) {
3025     // anon_mmap() should only get called during VM initialization,
3026     // don't need lock (actually we can skip locking even it can be called
3027     // from multiple threads, because _highest_vm_reserved_address is just a
3028     // hint about the upper limit of non-stack memory regions.)
3029     if ((address)addr + bytes > _highest_vm_reserved_address) {
3030       _highest_vm_reserved_address = (address)addr + bytes;
3031     }
3032   }
3033 
3034   return addr == MAP_FAILED ? NULL : addr;
3035 }
3036 
3037 // Don't update _highest_vm_reserved_address, because there might be memory
3038 // regions above addr + size. If so, releasing a memory region only creates
3039 // a hole in the address space, it doesn't help prevent heap-stack collision.
3040 //
3041 static int anon_munmap(char * addr, size_t size) {
3042   return ::munmap(addr, size) == 0;
3043 }
3044 
3045 char* os::reserve_memory(size_t bytes, char* requested_addr,
3046                          size_t alignment_hint) {
3047   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3048 }
3049 
3050 bool os::release_memory(char* addr, size_t size) {
3051   return anon_munmap(addr, size);
3052 }
3053 
3054 static address highest_vm_reserved_address() {
3055   return _highest_vm_reserved_address;
3056 }
3057 
3058 static bool bsd_mprotect(char* addr, size_t size, int prot) {
3059   // Bsd wants the mprotect address argument to be page aligned.
3060   char* bottom = (char*)align_size_down((intptr_t)addr, os::Bsd::page_size());
3061 
3062   // According to SUSv3, mprotect() should only be used with mappings
3063   // established by mmap(), and mmap() always maps whole pages. Unaligned
3064   // 'addr' likely indicates problem in the VM (e.g. trying to change
3065   // protection of malloc'ed or statically allocated memory). Check the
3066   // caller if you hit this assert.
3067   assert(addr == bottom, "sanity check");
3068 
3069   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Bsd::page_size());
3070   return ::mprotect(bottom, size, prot) == 0;
3071 }
3072 
3073 // Set protections specified
3074 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3075                         bool is_committed) {
3076   unsigned int p = 0;
3077   switch (prot) {
3078   case MEM_PROT_NONE: p = PROT_NONE; break;
3079   case MEM_PROT_READ: p = PROT_READ; break;
3080   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3081   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3082   default:
3083     ShouldNotReachHere();
3084   }
3085   // is_committed is unused.
3086   return bsd_mprotect(addr, bytes, p);
3087 }
3088 
3089 bool os::guard_memory(char* addr, size_t size) {
3090   return bsd_mprotect(addr, size, PROT_NONE);
3091 }
3092 
3093 bool os::unguard_memory(char* addr, size_t size) {
3094   return bsd_mprotect(addr, size, PROT_READ|PROT_WRITE);
3095 }
3096 
3097 bool os::Bsd::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3098   bool result = false;
3099 #ifndef _ALLBSD_SOURCE
3100   void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
3101                   MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3102                   -1, 0);
3103 
3104   if (p != (void *) -1) {
3105     // We don't know if this really is a huge page or not.
3106     FILE *fp = fopen("/proc/self/maps", "r");
3107     if (fp) {
3108       while (!feof(fp)) {
3109         char chars[257];
3110         long x = 0;
3111         if (fgets(chars, sizeof(chars), fp)) {
3112           if (sscanf(chars, "%lx-%*x", &x) == 1
3113               && x == (long)p) {
3114             if (strstr (chars, "hugepage")) {
3115               result = true;
3116               break;
3117             }
3118           }
3119         }
3120       }
3121       fclose(fp);
3122     }
3123     munmap (p, page_size);
3124     if (result)
3125       return true;
3126   }
3127 
3128   if (warn) {
3129     warning("HugeTLBFS is not supported by the operating system.");
3130   }
3131 #endif
3132 
3133   return result;
3134 }
3135 
3136 /*
3137 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3138 *
3139 * From the coredump_filter documentation:
3140 *
3141 * - (bit 0) anonymous private memory
3142 * - (bit 1) anonymous shared memory
3143 * - (bit 2) file-backed private memory
3144 * - (bit 3) file-backed shared memory
3145 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3146 *           effective only if the bit 2 is cleared)
3147 * - (bit 5) hugetlb private memory
3148 * - (bit 6) hugetlb shared memory
3149 */
3150 static void set_coredump_filter(void) {
3151   FILE *f;
3152   long cdm;
3153 
3154   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3155     return;
3156   }
3157 
3158   if (fscanf(f, "%lx", &cdm) != 1) {
3159     fclose(f);
3160     return;
3161   }
3162 
3163   rewind(f);
3164 
3165   if ((cdm & LARGEPAGES_BIT) == 0) {
3166     cdm |= LARGEPAGES_BIT;
3167     fprintf(f, "%#lx", cdm);
3168   }
3169 
3170   fclose(f);
3171 }
3172 
3173 // Large page support
3174 
3175 static size_t _large_page_size = 0;
3176 
3177 void os::large_page_init() {
3178 #ifndef _ALLBSD_SOURCE
3179   if (!UseLargePages) {
3180     UseHugeTLBFS = false;
3181     UseSHM = false;
3182     return;
3183   }
3184 
3185   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3186     // If UseLargePages is specified on the command line try both methods,
3187     // if it's default, then try only HugeTLBFS.
3188     if (FLAG_IS_DEFAULT(UseLargePages)) {
3189       UseHugeTLBFS = true;
3190     } else {
3191       UseHugeTLBFS = UseSHM = true;
3192     }
3193   }
3194 
3195   if (LargePageSizeInBytes) {
3196     _large_page_size = LargePageSizeInBytes;
3197   } else {
3198     // large_page_size on Bsd is used to round up heap size. x86 uses either
3199     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3200     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3201     // page as large as 256M.
3202     //
3203     // Here we try to figure out page size by parsing /proc/meminfo and looking
3204     // for a line with the following format:
3205     //    Hugepagesize:     2048 kB
3206     //
3207     // If we can't determine the value (e.g. /proc is not mounted, or the text
3208     // format has been changed), we'll use the largest page size supported by
3209     // the processor.
3210 
3211 #ifndef ZERO
3212     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3213                        ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3214 #endif // ZERO
3215 
3216     FILE *fp = fopen("/proc/meminfo", "r");
3217     if (fp) {
3218       while (!feof(fp)) {
3219         int x = 0;
3220         char buf[16];
3221         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3222           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3223             _large_page_size = x * K;
3224             break;
3225           }
3226         } else {
3227           // skip to next line
3228           for (;;) {
3229             int ch = fgetc(fp);
3230             if (ch == EOF || ch == (int)'\n') break;
3231           }
3232         }
3233       }
3234       fclose(fp);
3235     }
3236   }
3237 
3238   // print a warning if any large page related flag is specified on command line
3239   bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3240 
3241   const size_t default_page_size = (size_t)Bsd::page_size();
3242   if (_large_page_size > default_page_size) {
3243     _page_sizes[0] = _large_page_size;
3244     _page_sizes[1] = default_page_size;
3245     _page_sizes[2] = 0;
3246   }
3247   UseHugeTLBFS = UseHugeTLBFS &&
3248                  Bsd::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3249 
3250   if (UseHugeTLBFS)
3251     UseSHM = false;
3252 
3253   UseLargePages = UseHugeTLBFS || UseSHM;
3254 
3255   set_coredump_filter();
3256 #endif
3257 }
3258 
3259 #ifndef _ALLBSD_SOURCE
3260 #ifndef SHM_HUGETLB
3261 #define SHM_HUGETLB 04000
3262 #endif
3263 #endif
3264 
3265 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3266   // "exec" is passed in but not used.  Creating the shared image for
3267   // the code cache doesn't have an SHM_X executable permission to check.
3268   assert(UseLargePages && UseSHM, "only for SHM large pages");
3269 
3270   key_t key = IPC_PRIVATE;
3271   char *addr;
3272 
3273   bool warn_on_failure = UseLargePages &&
3274                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3275                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3276                         );
3277   char msg[128];
3278 
3279   // Create a large shared memory region to attach to based on size.
3280   // Currently, size is the total size of the heap
3281 #ifndef _ALLBSD_SOURCE
3282   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3283 #else
3284   int shmid = shmget(key, bytes, IPC_CREAT|SHM_R|SHM_W);
3285 #endif
3286   if (shmid == -1) {
3287      // Possible reasons for shmget failure:
3288      // 1. shmmax is too small for Java heap.
3289      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3290      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3291      // 2. not enough large page memory.
3292      //    > check available large pages: cat /proc/meminfo
3293      //    > increase amount of large pages:
3294      //          echo new_value > /proc/sys/vm/nr_hugepages
3295      //      Note 1: different Bsd may use different name for this property,
3296      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3297      //      Note 2: it's possible there's enough physical memory available but
3298      //            they are so fragmented after a long run that they can't
3299      //            coalesce into large pages. Try to reserve large pages when
3300      //            the system is still "fresh".
3301      if (warn_on_failure) {
3302        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3303        warning(msg);
3304      }
3305      return NULL;
3306   }
3307 
3308   // attach to the region
3309   addr = (char*)shmat(shmid, req_addr, 0);
3310   int err = errno;
3311 
3312   // Remove shmid. If shmat() is successful, the actual shared memory segment
3313   // will be deleted when it's detached by shmdt() or when the process
3314   // terminates. If shmat() is not successful this will remove the shared
3315   // segment immediately.
3316   shmctl(shmid, IPC_RMID, NULL);
3317 
3318   if ((intptr_t)addr == -1) {
3319      if (warn_on_failure) {
3320        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3321        warning(msg);
3322      }
3323      return NULL;
3324   }
3325 
3326   return addr;
3327 }
3328 
3329 bool os::release_memory_special(char* base, size_t bytes) {
3330   // detaching the SHM segment will also delete it, see reserve_memory_special()
3331   int rslt = shmdt(base);
3332   return rslt == 0;
3333 }
3334 
3335 size_t os::large_page_size() {
3336   return _large_page_size;
3337 }
3338 
3339 // HugeTLBFS allows application to commit large page memory on demand;
3340 // with SysV SHM the entire memory region must be allocated as shared
3341 // memory.
3342 bool os::can_commit_large_page_memory() {
3343   return UseHugeTLBFS;
3344 }
3345 
3346 bool os::can_execute_large_page_memory() {
3347   return UseHugeTLBFS;
3348 }
3349 
3350 // Reserve memory at an arbitrary address, only if that area is
3351 // available (and not reserved for something else).
3352 
3353 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3354   const int max_tries = 10;
3355   char* base[max_tries];
3356   size_t size[max_tries];
3357   const size_t gap = 0x000000;
3358 
3359   // Assert only that the size is a multiple of the page size, since
3360   // that's all that mmap requires, and since that's all we really know
3361   // about at this low abstraction level.  If we need higher alignment,
3362   // we can either pass an alignment to this method or verify alignment
3363   // in one of the methods further up the call chain.  See bug 5044738.
3364   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3365 
3366   // Repeatedly allocate blocks until the block is allocated at the
3367   // right spot. Give up after max_tries. Note that reserve_memory() will
3368   // automatically update _highest_vm_reserved_address if the call is
3369   // successful. The variable tracks the highest memory address every reserved
3370   // by JVM. It is used to detect heap-stack collision if running with
3371   // fixed-stack BsdThreads. Because here we may attempt to reserve more
3372   // space than needed, it could confuse the collision detecting code. To
3373   // solve the problem, save current _highest_vm_reserved_address and
3374   // calculate the correct value before return.
3375   address old_highest = _highest_vm_reserved_address;
3376 
3377   // Bsd mmap allows caller to pass an address as hint; give it a try first,
3378   // if kernel honors the hint then we can return immediately.
3379   char * addr = anon_mmap(requested_addr, bytes, false);
3380   if (addr == requested_addr) {
3381      return requested_addr;
3382   }
3383 
3384   if (addr != NULL) {
3385      // mmap() is successful but it fails to reserve at the requested address
3386      anon_munmap(addr, bytes);
3387   }
3388 
3389   int i;
3390   for (i = 0; i < max_tries; ++i) {
3391     base[i] = reserve_memory(bytes);
3392 
3393     if (base[i] != NULL) {
3394       // Is this the block we wanted?
3395       if (base[i] == requested_addr) {
3396         size[i] = bytes;
3397         break;
3398       }
3399 
3400       // Does this overlap the block we wanted? Give back the overlapped
3401       // parts and try again.
3402 
3403       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3404       if (top_overlap >= 0 && top_overlap < bytes) {
3405         unmap_memory(base[i], top_overlap);
3406         base[i] += top_overlap;
3407         size[i] = bytes - top_overlap;
3408       } else {
3409         size_t bottom_overlap = base[i] + bytes - requested_addr;
3410         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3411           unmap_memory(requested_addr, bottom_overlap);
3412           size[i] = bytes - bottom_overlap;
3413         } else {
3414           size[i] = bytes;
3415         }
3416       }
3417     }
3418   }
3419 
3420   // Give back the unused reserved pieces.
3421 
3422   for (int j = 0; j < i; ++j) {
3423     if (base[j] != NULL) {
3424       unmap_memory(base[j], size[j]);
3425     }
3426   }
3427 
3428   if (i < max_tries) {
3429     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3430     return requested_addr;
3431   } else {
3432     _highest_vm_reserved_address = old_highest;
3433     return NULL;
3434   }
3435 }
3436 
3437 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3438   RESTARTABLE_RETURN_INT(::read(fd, buf, nBytes));
3439 }
3440 
3441 // TODO-FIXME: reconcile Solaris' os::sleep with the bsd variation.
3442 // Solaris uses poll(), bsd uses park().
3443 // Poll() is likely a better choice, assuming that Thread.interrupt()
3444 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3445 // SIGSEGV, see 4355769.
3446 
3447 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3448   assert(thread == Thread::current(),  "thread consistency check");
3449 
3450   ParkEvent * const slp = thread->_SleepEvent ;
3451   slp->reset() ;
3452   OrderAccess::fence() ;
3453 
3454   if (interruptible) {
3455     jlong prevtime = javaTimeNanos();
3456 
3457     for (;;) {
3458       if (os::is_interrupted(thread, true)) {
3459         return OS_INTRPT;
3460       }
3461 
3462       jlong newtime = javaTimeNanos();
3463 
3464       if (newtime - prevtime < 0) {
3465         // time moving backwards, should only happen if no monotonic clock
3466         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3467         assert(!Bsd::supports_monotonic_clock(), "time moving backwards");
3468       } else {
3469         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3470       }
3471 
3472       if(millis <= 0) {
3473         return OS_OK;
3474       }
3475 
3476       prevtime = newtime;
3477 
3478       {
3479         assert(thread->is_Java_thread(), "sanity check");
3480         JavaThread *jt = (JavaThread *) thread;
3481         ThreadBlockInVM tbivm(jt);
3482         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3483 
3484         jt->set_suspend_equivalent();
3485         // cleared by handle_special_suspend_equivalent_condition() or
3486         // java_suspend_self() via check_and_wait_while_suspended()
3487 
3488         slp->park(millis);
3489 
3490         // were we externally suspended while we were waiting?
3491         jt->check_and_wait_while_suspended();
3492       }
3493     }
3494   } else {
3495     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3496     jlong prevtime = javaTimeNanos();
3497 
3498     for (;;) {
3499       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3500       // the 1st iteration ...
3501       jlong newtime = javaTimeNanos();
3502 
3503       if (newtime - prevtime < 0) {
3504         // time moving backwards, should only happen if no monotonic clock
3505         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3506         assert(!Bsd::supports_monotonic_clock(), "time moving backwards");
3507       } else {
3508         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3509       }
3510 
3511       if(millis <= 0) break ;
3512 
3513       prevtime = newtime;
3514       slp->park(millis);
3515     }
3516     return OS_OK ;
3517   }
3518 }
3519 
3520 int os::naked_sleep() {
3521   // %% make the sleep time an integer flag. for now use 1 millisec.
3522   return os::sleep(Thread::current(), 1, false);
3523 }
3524 
3525 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3526 void os::infinite_sleep() {
3527   while (true) {    // sleep forever ...
3528     ::sleep(100);   // ... 100 seconds at a time
3529   }
3530 }
3531 
3532 // Used to convert frequent JVM_Yield() to nops
3533 bool os::dont_yield() {
3534   return DontYieldALot;
3535 }
3536 
3537 void os::yield() {
3538   sched_yield();
3539 }
3540 
3541 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3542 
3543 void os::yield_all(int attempts) {
3544   // Yields to all threads, including threads with lower priorities
3545   // Threads on Bsd are all with same priority. The Solaris style
3546   // os::yield_all() with nanosleep(1ms) is not necessary.
3547   sched_yield();
3548 }
3549 
3550 // Called from the tight loops to possibly influence time-sharing heuristics
3551 void os::loop_breaker(int attempts) {
3552   os::yield_all(attempts);
3553 }
3554 
3555 ////////////////////////////////////////////////////////////////////////////////
3556 // thread priority support
3557 
3558 // Note: Normal Bsd applications are run with SCHED_OTHER policy. SCHED_OTHER
3559 // only supports dynamic priority, static priority must be zero. For real-time
3560 // applications, Bsd supports SCHED_RR which allows static priority (1-99).
3561 // However, for large multi-threaded applications, SCHED_RR is not only slower
3562 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3563 // of 5 runs - Sep 2005).
3564 //
3565 // The following code actually changes the niceness of kernel-thread/LWP. It
3566 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3567 // not the entire user process, and user level threads are 1:1 mapped to kernel
3568 // threads. It has always been the case, but could change in the future. For
3569 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3570 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3571 
3572 #if defined(_ALLBSD_SOURCE) && !defined(__APPLE__)
3573 int os::java_to_os_priority[CriticalPriority + 1] = {
3574   19,              // 0 Entry should never be used
3575 
3576    0,              // 1 MinPriority
3577    3,              // 2
3578    6,              // 3
3579 
3580   10,              // 4
3581   15,              // 5 NormPriority
3582   18,              // 6
3583 
3584   21,              // 7
3585   25,              // 8
3586   28,              // 9 NearMaxPriority
3587 
3588   31,              // 10 MaxPriority
3589 
3590   31               // 11 CriticalPriority
3591 };
3592 #elif defined(__APPLE__)
3593 /* Using Mach high-level priority assignments */
3594 int os::java_to_os_priority[CriticalPriority + 1] = {
3595    0,              // 0 Entry should never be used (MINPRI_USER)
3596 
3597   27,              // 1 MinPriority
3598   28,              // 2
3599   29,              // 3
3600 
3601   30,              // 4
3602   31,              // 5 NormPriority (BASEPRI_DEFAULT)
3603   32,              // 6
3604 
3605   33,              // 7
3606   34,              // 8
3607   35,              // 9 NearMaxPriority
3608 
3609   36,              // 10 MaxPriority
3610 
3611   36               // 11 CriticalPriority
3612 };
3613 #else
3614 int os::java_to_os_priority[CriticalPriority + 1] = {
3615   19,              // 0 Entry should never be used
3616 
3617    4,              // 1 MinPriority
3618    3,              // 2
3619    2,              // 3
3620 
3621    1,              // 4
3622    0,              // 5 NormPriority
3623   -1,              // 6
3624 
3625   -2,              // 7
3626   -3,              // 8
3627   -4,              // 9 NearMaxPriority
3628 
3629   -5,              // 10 MaxPriority
3630 
3631   -5               // 11 CriticalPriority
3632 };
3633 #endif
3634 
3635 static int prio_init() {
3636   if (ThreadPriorityPolicy == 1) {
3637     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3638     // if effective uid is not root. Perhaps, a more elegant way of doing
3639     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3640     if (geteuid() != 0) {
3641       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3642         warning("-XX:ThreadPriorityPolicy requires root privilege on Bsd");
3643       }
3644       ThreadPriorityPolicy = 0;
3645     }
3646   }
3647   return 0;
3648 }
3649 
3650 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3651   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3652 
3653 #ifdef __OpenBSD__
3654   // OpenBSD pthread_setprio starves low priority threads
3655   return OS_OK;
3656 #elif defined(__FreeBSD__)
3657   int ret = pthread_setprio(thread->osthread()->pthread_id(), newpri);
3658 #elif defined(__APPLE__) || defined(__NetBSD__)
3659   struct sched_param sp;
3660   int policy;
3661   pthread_t self = pthread_self();
3662 
3663   if (pthread_getschedparam(self, &policy, &sp) != 0)
3664     return OS_ERR;
3665 
3666   sp.sched_priority = newpri;
3667   if (pthread_setschedparam(self, policy, &sp) != 0)
3668     return OS_ERR;
3669 
3670   return OS_OK;
3671 #else
3672   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3673   return (ret == 0) ? OS_OK : OS_ERR;
3674 #endif
3675 }
3676 
3677 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3678   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3679     *priority_ptr = java_to_os_priority[NormPriority];
3680     return OS_OK;
3681   }
3682 
3683   errno = 0;
3684 #if defined(__OpenBSD__) || defined(__FreeBSD__)
3685   *priority_ptr = pthread_getprio(thread->osthread()->pthread_id());
3686 #elif defined(__APPLE__) || defined(__NetBSD__)
3687   int policy;
3688   struct sched_param sp;
3689 
3690   pthread_getschedparam(pthread_self(), &policy, &sp);
3691   *priority_ptr = sp.sched_priority;
3692 #else
3693   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3694 #endif
3695   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3696 }
3697 
3698 // Hint to the underlying OS that a task switch would not be good.
3699 // Void return because it's a hint and can fail.
3700 void os::hint_no_preempt() {}
3701 
3702 ////////////////////////////////////////////////////////////////////////////////
3703 // suspend/resume support
3704 
3705 //  the low-level signal-based suspend/resume support is a remnant from the
3706 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3707 //  within hotspot. Now there is a single use-case for this:
3708 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3709 //      that runs in the watcher thread.
3710 //  The remaining code is greatly simplified from the more general suspension
3711 //  code that used to be used.
3712 //
3713 //  The protocol is quite simple:
3714 //  - suspend:
3715 //      - sends a signal to the target thread
3716 //      - polls the suspend state of the osthread using a yield loop
3717 //      - target thread signal handler (SR_handler) sets suspend state
3718 //        and blocks in sigsuspend until continued
3719 //  - resume:
3720 //      - sets target osthread state to continue
3721 //      - sends signal to end the sigsuspend loop in the SR_handler
3722 //
3723 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3724 //
3725 
3726 static void resume_clear_context(OSThread *osthread) {
3727   osthread->set_ucontext(NULL);
3728   osthread->set_siginfo(NULL);
3729 
3730   // notify the suspend action is completed, we have now resumed
3731   osthread->sr.clear_suspended();
3732 }
3733 
3734 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3735   osthread->set_ucontext(context);
3736   osthread->set_siginfo(siginfo);
3737 }
3738 
3739 //
3740 // Handler function invoked when a thread's execution is suspended or
3741 // resumed. We have to be careful that only async-safe functions are
3742 // called here (Note: most pthread functions are not async safe and
3743 // should be avoided.)
3744 //
3745 // Note: sigwait() is a more natural fit than sigsuspend() from an
3746 // interface point of view, but sigwait() prevents the signal hander
3747 // from being run. libpthread would get very confused by not having
3748 // its signal handlers run and prevents sigwait()'s use with the
3749 // mutex granting granting signal.
3750 //
3751 // Currently only ever called on the VMThread
3752 //
3753 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3754   // Save and restore errno to avoid confusing native code with EINTR
3755   // after sigsuspend.
3756   int old_errno = errno;
3757 
3758   Thread* thread = Thread::current();
3759   OSThread* osthread = thread->osthread();
3760   assert(thread->is_VM_thread(), "Must be VMThread");
3761   // read current suspend action
3762   int action = osthread->sr.suspend_action();
3763   if (action == SR_SUSPEND) {
3764     suspend_save_context(osthread, siginfo, context);
3765 
3766     // Notify the suspend action is about to be completed. do_suspend()
3767     // waits until SR_SUSPENDED is set and then returns. We will wait
3768     // here for a resume signal and that completes the suspend-other
3769     // action. do_suspend/do_resume is always called as a pair from
3770     // the same thread - so there are no races
3771 
3772     // notify the caller
3773     osthread->sr.set_suspended();
3774 
3775     sigset_t suspend_set;  // signals for sigsuspend()
3776 
3777     // get current set of blocked signals and unblock resume signal
3778     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3779     sigdelset(&suspend_set, SR_signum);
3780 
3781     // wait here until we are resumed
3782     do {
3783       sigsuspend(&suspend_set);
3784       // ignore all returns until we get a resume signal
3785     } while (osthread->sr.suspend_action() != SR_CONTINUE);
3786 
3787     resume_clear_context(osthread);
3788 
3789   } else {
3790     assert(action == SR_CONTINUE, "unexpected sr action");
3791     // nothing special to do - just leave the handler
3792   }
3793 
3794   errno = old_errno;
3795 }
3796 
3797 
3798 static int SR_initialize() {
3799   struct sigaction act;
3800   char *s;
3801   /* Get signal number to use for suspend/resume */
3802   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3803     int sig = ::strtol(s, 0, 10);
3804     if (sig > 0 || sig < NSIG) {
3805         SR_signum = sig;
3806     }
3807   }
3808 
3809   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3810         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3811 
3812   sigemptyset(&SR_sigset);
3813   sigaddset(&SR_sigset, SR_signum);
3814 
3815   /* Set up signal handler for suspend/resume */
3816   act.sa_flags = SA_RESTART|SA_SIGINFO;
3817   act.sa_handler = (void (*)(int)) SR_handler;
3818 
3819   // SR_signum is blocked by default.
3820   // 4528190 - We also need to block pthread restart signal (32 on all
3821   // supported Bsd platforms). Note that BsdThreads need to block
3822   // this signal for all threads to work properly. So we don't have
3823   // to use hard-coded signal number when setting up the mask.
3824   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3825 
3826   if (sigaction(SR_signum, &act, 0) == -1) {
3827     return -1;
3828   }
3829 
3830   // Save signal flag
3831   os::Bsd::set_our_sigflags(SR_signum, act.sa_flags);
3832   return 0;
3833 }
3834 
3835 static int SR_finalize() {
3836   return 0;
3837 }
3838 
3839 
3840 // returns true on success and false on error - really an error is fatal
3841 // but this seems the normal response to library errors
3842 static bool do_suspend(OSThread* osthread) {
3843   // mark as suspended and send signal
3844   osthread->sr.set_suspend_action(SR_SUSPEND);
3845   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3846   assert_status(status == 0, status, "pthread_kill");
3847 
3848   // check status and wait until notified of suspension
3849   if (status == 0) {
3850     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3851       os::yield_all(i);
3852     }
3853     osthread->sr.set_suspend_action(SR_NONE);
3854     return true;
3855   }
3856   else {
3857     osthread->sr.set_suspend_action(SR_NONE);
3858     return false;
3859   }
3860 }
3861 
3862 static void do_resume(OSThread* osthread) {
3863   assert(osthread->sr.is_suspended(), "thread should be suspended");
3864   osthread->sr.set_suspend_action(SR_CONTINUE);
3865 
3866   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3867   assert_status(status == 0, status, "pthread_kill");
3868   // check status and wait unit notified of resumption
3869   if (status == 0) {
3870     for (int i = 0; osthread->sr.is_suspended(); i++) {
3871       os::yield_all(i);
3872     }
3873   }
3874   osthread->sr.set_suspend_action(SR_NONE);
3875 }
3876 
3877 ////////////////////////////////////////////////////////////////////////////////
3878 // interrupt support
3879 
3880 void os::interrupt(Thread* thread) {
3881   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3882     "possibility of dangling Thread pointer");
3883 
3884   OSThread* osthread = thread->osthread();
3885 
3886   if (!osthread->interrupted()) {
3887     osthread->set_interrupted(true);
3888     // More than one thread can get here with the same value of osthread,
3889     // resulting in multiple notifications.  We do, however, want the store
3890     // to interrupted() to be visible to other threads before we execute unpark().
3891     OrderAccess::fence();
3892     ParkEvent * const slp = thread->_SleepEvent ;
3893     if (slp != NULL) slp->unpark() ;
3894   }
3895 
3896   // For JSR166. Unpark even if interrupt status already was set
3897   if (thread->is_Java_thread())
3898     ((JavaThread*)thread)->parker()->unpark();
3899 
3900   ParkEvent * ev = thread->_ParkEvent ;
3901   if (ev != NULL) ev->unpark() ;
3902 
3903 }
3904 
3905 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3906   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3907     "possibility of dangling Thread pointer");
3908 
3909   OSThread* osthread = thread->osthread();
3910 
3911   bool interrupted = osthread->interrupted();
3912 
3913   if (interrupted && clear_interrupted) {
3914     osthread->set_interrupted(false);
3915     // consider thread->_SleepEvent->reset() ... optional optimization
3916   }
3917 
3918   return interrupted;
3919 }
3920 
3921 ///////////////////////////////////////////////////////////////////////////////////
3922 // signal handling (except suspend/resume)
3923 
3924 // This routine may be used by user applications as a "hook" to catch signals.
3925 // The user-defined signal handler must pass unrecognized signals to this
3926 // routine, and if it returns true (non-zero), then the signal handler must
3927 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3928 // routine will never retun false (zero), but instead will execute a VM panic
3929 // routine kill the process.
3930 //
3931 // If this routine returns false, it is OK to call it again.  This allows
3932 // the user-defined signal handler to perform checks either before or after
3933 // the VM performs its own checks.  Naturally, the user code would be making
3934 // a serious error if it tried to handle an exception (such as a null check
3935 // or breakpoint) that the VM was generating for its own correct operation.
3936 //
3937 // This routine may recognize any of the following kinds of signals:
3938 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3939 // It should be consulted by handlers for any of those signals.
3940 //
3941 // The caller of this routine must pass in the three arguments supplied
3942 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3943 // field of the structure passed to sigaction().  This routine assumes that
3944 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3945 //
3946 // Note that the VM will print warnings if it detects conflicting signal
3947 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3948 //
3949 extern "C" JNIEXPORT int
3950 JVM_handle_bsd_signal(int signo, siginfo_t* siginfo,
3951                         void* ucontext, int abort_if_unrecognized);
3952 
3953 void signalHandler(int sig, siginfo_t* info, void* uc) {
3954   assert(info != NULL && uc != NULL, "it must be old kernel");
3955   JVM_handle_bsd_signal(sig, info, uc, true);
3956 }
3957 
3958 
3959 // This boolean allows users to forward their own non-matching signals
3960 // to JVM_handle_bsd_signal, harmlessly.
3961 bool os::Bsd::signal_handlers_are_installed = false;
3962 
3963 // For signal-chaining
3964 struct sigaction os::Bsd::sigact[MAXSIGNUM];
3965 unsigned int os::Bsd::sigs = 0;
3966 bool os::Bsd::libjsig_is_loaded = false;
3967 typedef struct sigaction *(*get_signal_t)(int);
3968 get_signal_t os::Bsd::get_signal_action = NULL;
3969 
3970 struct sigaction* os::Bsd::get_chained_signal_action(int sig) {
3971   struct sigaction *actp = NULL;
3972 
3973   if (libjsig_is_loaded) {
3974     // Retrieve the old signal handler from libjsig
3975     actp = (*get_signal_action)(sig);
3976   }
3977   if (actp == NULL) {
3978     // Retrieve the preinstalled signal handler from jvm
3979     actp = get_preinstalled_handler(sig);
3980   }
3981 
3982   return actp;
3983 }
3984 
3985 static bool call_chained_handler(struct sigaction *actp, int sig,
3986                                  siginfo_t *siginfo, void *context) {
3987   // Call the old signal handler
3988   if (actp->sa_handler == SIG_DFL) {
3989     // It's more reasonable to let jvm treat it as an unexpected exception
3990     // instead of taking the default action.
3991     return false;
3992   } else if (actp->sa_handler != SIG_IGN) {
3993     if ((actp->sa_flags & SA_NODEFER) == 0) {
3994       // automaticlly block the signal
3995       sigaddset(&(actp->sa_mask), sig);
3996     }
3997 
3998     sa_handler_t hand;
3999     sa_sigaction_t sa;
4000     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4001     // retrieve the chained handler
4002     if (siginfo_flag_set) {
4003       sa = actp->sa_sigaction;
4004     } else {
4005       hand = actp->sa_handler;
4006     }
4007 
4008     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4009       actp->sa_handler = SIG_DFL;
4010     }
4011 
4012     // try to honor the signal mask
4013     sigset_t oset;
4014     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4015 
4016     // call into the chained handler
4017     if (siginfo_flag_set) {
4018       (*sa)(sig, siginfo, context);
4019     } else {
4020       (*hand)(sig);
4021     }
4022 
4023     // restore the signal mask
4024     pthread_sigmask(SIG_SETMASK, &oset, 0);
4025   }
4026   // Tell jvm's signal handler the signal is taken care of.
4027   return true;
4028 }
4029 
4030 bool os::Bsd::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4031   bool chained = false;
4032   // signal-chaining
4033   if (UseSignalChaining) {
4034     struct sigaction *actp = get_chained_signal_action(sig);
4035     if (actp != NULL) {
4036       chained = call_chained_handler(actp, sig, siginfo, context);
4037     }
4038   }
4039   return chained;
4040 }
4041 
4042 struct sigaction* os::Bsd::get_preinstalled_handler(int sig) {
4043   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4044     return &sigact[sig];
4045   }
4046   return NULL;
4047 }
4048 
4049 void os::Bsd::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4050   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4051   sigact[sig] = oldAct;
4052   sigs |= (unsigned int)1 << sig;
4053 }
4054 
4055 // for diagnostic
4056 int os::Bsd::sigflags[MAXSIGNUM];
4057 
4058 int os::Bsd::get_our_sigflags(int sig) {
4059   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4060   return sigflags[sig];
4061 }
4062 
4063 void os::Bsd::set_our_sigflags(int sig, int flags) {
4064   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4065   sigflags[sig] = flags;
4066 }
4067 
4068 void os::Bsd::set_signal_handler(int sig, bool set_installed) {
4069   // Check for overwrite.
4070   struct sigaction oldAct;
4071   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4072 
4073   void* oldhand = oldAct.sa_sigaction
4074                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4075                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4076   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4077       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4078       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4079     if (AllowUserSignalHandlers || !set_installed) {
4080       // Do not overwrite; user takes responsibility to forward to us.
4081       return;
4082     } else if (UseSignalChaining) {
4083       // save the old handler in jvm
4084       save_preinstalled_handler(sig, oldAct);
4085       // libjsig also interposes the sigaction() call below and saves the
4086       // old sigaction on it own.
4087     } else {
4088       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4089                     "%#lx for signal %d.", (long)oldhand, sig));
4090     }
4091   }
4092 
4093   struct sigaction sigAct;
4094   sigfillset(&(sigAct.sa_mask));
4095   sigAct.sa_handler = SIG_DFL;
4096   if (!set_installed) {
4097     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4098   } else {
4099     sigAct.sa_sigaction = signalHandler;
4100     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4101   }
4102   // Save flags, which are set by ours
4103   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4104   sigflags[sig] = sigAct.sa_flags;
4105 
4106   int ret = sigaction(sig, &sigAct, &oldAct);
4107   assert(ret == 0, "check");
4108 
4109   void* oldhand2  = oldAct.sa_sigaction
4110                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4111                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4112   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4113 }
4114 
4115 // install signal handlers for signals that HotSpot needs to
4116 // handle in order to support Java-level exception handling.
4117 
4118 void os::Bsd::install_signal_handlers() {
4119   if (!signal_handlers_are_installed) {
4120     signal_handlers_are_installed = true;
4121 
4122     // signal-chaining
4123     typedef void (*signal_setting_t)();
4124     signal_setting_t begin_signal_setting = NULL;
4125     signal_setting_t end_signal_setting = NULL;
4126     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4127                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4128     if (begin_signal_setting != NULL) {
4129       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4130                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4131       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4132                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4133       libjsig_is_loaded = true;
4134       assert(UseSignalChaining, "should enable signal-chaining");
4135     }
4136     if (libjsig_is_loaded) {
4137       // Tell libjsig jvm is setting signal handlers
4138       (*begin_signal_setting)();
4139     }
4140 
4141     set_signal_handler(SIGSEGV, true);
4142     set_signal_handler(SIGPIPE, true);
4143     set_signal_handler(SIGBUS, true);
4144     set_signal_handler(SIGILL, true);
4145     set_signal_handler(SIGFPE, true);
4146     set_signal_handler(SIGXFSZ, true);
4147 
4148 #if defined(__APPLE__)
4149     // In Mac OS X 10.4, CrashReporter will write a crash log for all 'fatal' signals, including
4150     // signals caught and handled by the JVM. To work around this, we reset the mach task
4151     // signal handler that's placed on our process by CrashReporter. This disables
4152     // CrashReporter-based reporting.
4153     //
4154     // This work-around is not necessary for 10.5+, as CrashReporter no longer intercedes
4155     // on caught fatal signals.
4156     //
4157     // Additionally, gdb installs both standard BSD signal handlers, and mach exception
4158     // handlers. By replacing the existing task exception handler, we disable gdb's mach
4159     // exception handling, while leaving the standard BSD signal handlers functional.
4160     kern_return_t kr;
4161     kr = task_set_exception_ports(mach_task_self(),
4162         EXC_MASK_BAD_ACCESS | EXC_MASK_ARITHMETIC,
4163         MACH_PORT_NULL,
4164         EXCEPTION_STATE_IDENTITY,
4165         MACHINE_THREAD_STATE);
4166 
4167     assert(kr == KERN_SUCCESS, "could not set mach task signal handler");
4168 #endif
4169 
4170     if (libjsig_is_loaded) {
4171       // Tell libjsig jvm finishes setting signal handlers
4172       (*end_signal_setting)();
4173     }
4174 
4175     // We don't activate signal checker if libjsig is in place, we trust ourselves
4176     // and if UserSignalHandler is installed all bets are off
4177     if (CheckJNICalls) {
4178       if (libjsig_is_loaded) {
4179         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4180         check_signals = false;
4181       }
4182       if (AllowUserSignalHandlers) {
4183         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4184         check_signals = false;
4185       }
4186     }
4187   }
4188 }
4189 
4190 #ifndef _ALLBSD_SOURCE
4191 // This is the fastest way to get thread cpu time on Bsd.
4192 // Returns cpu time (user+sys) for any thread, not only for current.
4193 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4194 // It might work on 2.6.10+ with a special kernel/glibc patch.
4195 // For reference, please, see IEEE Std 1003.1-2004:
4196 //   http://www.unix.org/single_unix_specification
4197 
4198 jlong os::Bsd::fast_thread_cpu_time(clockid_t clockid) {
4199   struct timespec tp;
4200   int rc = os::Bsd::clock_gettime(clockid, &tp);
4201   assert(rc == 0, "clock_gettime is expected to return 0 code");
4202 
4203   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4204 }
4205 #endif
4206 
4207 /////
4208 // glibc on Bsd platform uses non-documented flag
4209 // to indicate, that some special sort of signal
4210 // trampoline is used.
4211 // We will never set this flag, and we should
4212 // ignore this flag in our diagnostic
4213 #ifdef SIGNIFICANT_SIGNAL_MASK
4214 #undef SIGNIFICANT_SIGNAL_MASK
4215 #endif
4216 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4217 
4218 static const char* get_signal_handler_name(address handler,
4219                                            char* buf, int buflen) {
4220   int offset;
4221   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4222   if (found) {
4223     // skip directory names
4224     const char *p1, *p2;
4225     p1 = buf;
4226     size_t len = strlen(os::file_separator());
4227     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4228     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4229   } else {
4230     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4231   }
4232   return buf;
4233 }
4234 
4235 static void print_signal_handler(outputStream* st, int sig,
4236                                  char* buf, size_t buflen) {
4237   struct sigaction sa;
4238 
4239   sigaction(sig, NULL, &sa);
4240 
4241   // See comment for SIGNIFICANT_SIGNAL_MASK define
4242   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4243 
4244   st->print("%s: ", os::exception_name(sig, buf, buflen));
4245 
4246   address handler = (sa.sa_flags & SA_SIGINFO)
4247     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4248     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4249 
4250   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4251     st->print("SIG_DFL");
4252   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4253     st->print("SIG_IGN");
4254   } else {
4255     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4256   }
4257 
4258   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4259 
4260   address rh = VMError::get_resetted_sighandler(sig);
4261   // May be, handler was resetted by VMError?
4262   if(rh != NULL) {
4263     handler = rh;
4264     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4265   }
4266 
4267   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
4268 
4269   // Check: is it our handler?
4270   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4271      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4272     // It is our signal handler
4273     // check for flags, reset system-used one!
4274     if((int)sa.sa_flags != os::Bsd::get_our_sigflags(sig)) {
4275       st->print(
4276                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4277                 os::Bsd::get_our_sigflags(sig));
4278     }
4279   }
4280   st->cr();
4281 }
4282 
4283 
4284 #define DO_SIGNAL_CHECK(sig) \
4285   if (!sigismember(&check_signal_done, sig)) \
4286     os::Bsd::check_signal_handler(sig)
4287 
4288 // This method is a periodic task to check for misbehaving JNI applications
4289 // under CheckJNI, we can add any periodic checks here
4290 
4291 void os::run_periodic_checks() {
4292 
4293   if (check_signals == false) return;
4294 
4295   // SEGV and BUS if overridden could potentially prevent
4296   // generation of hs*.log in the event of a crash, debugging
4297   // such a case can be very challenging, so we absolutely
4298   // check the following for a good measure:
4299   DO_SIGNAL_CHECK(SIGSEGV);
4300   DO_SIGNAL_CHECK(SIGILL);
4301   DO_SIGNAL_CHECK(SIGFPE);
4302   DO_SIGNAL_CHECK(SIGBUS);
4303   DO_SIGNAL_CHECK(SIGPIPE);
4304   DO_SIGNAL_CHECK(SIGXFSZ);
4305 
4306 
4307   // ReduceSignalUsage allows the user to override these handlers
4308   // see comments at the very top and jvm_solaris.h
4309   if (!ReduceSignalUsage) {
4310     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4311     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4312     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4313     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4314   }
4315 
4316   DO_SIGNAL_CHECK(SR_signum);
4317   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4318 }
4319 
4320 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4321 
4322 static os_sigaction_t os_sigaction = NULL;
4323 
4324 void os::Bsd::check_signal_handler(int sig) {
4325   char buf[O_BUFLEN];
4326   address jvmHandler = NULL;
4327 
4328 
4329   struct sigaction act;
4330   if (os_sigaction == NULL) {
4331     // only trust the default sigaction, in case it has been interposed
4332     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4333     if (os_sigaction == NULL) return;
4334   }
4335 
4336   os_sigaction(sig, (struct sigaction*)NULL, &act);
4337 
4338 
4339   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4340 
4341   address thisHandler = (act.sa_flags & SA_SIGINFO)
4342     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4343     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4344 
4345 
4346   switch(sig) {
4347   case SIGSEGV:
4348   case SIGBUS:
4349   case SIGFPE:
4350   case SIGPIPE:
4351   case SIGILL:
4352   case SIGXFSZ:
4353     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4354     break;
4355 
4356   case SHUTDOWN1_SIGNAL:
4357   case SHUTDOWN2_SIGNAL:
4358   case SHUTDOWN3_SIGNAL:
4359   case BREAK_SIGNAL:
4360     jvmHandler = (address)user_handler();
4361     break;
4362 
4363   case INTERRUPT_SIGNAL:
4364     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4365     break;
4366 
4367   default:
4368     if (sig == SR_signum) {
4369       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4370     } else {
4371       return;
4372     }
4373     break;
4374   }
4375 
4376   if (thisHandler != jvmHandler) {
4377     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4378     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4379     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4380     // No need to check this sig any longer
4381     sigaddset(&check_signal_done, sig);
4382   } else if(os::Bsd::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Bsd::get_our_sigflags(sig)) {
4383     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4384     tty->print("expected:" PTR32_FORMAT, os::Bsd::get_our_sigflags(sig));
4385     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4386     // No need to check this sig any longer
4387     sigaddset(&check_signal_done, sig);
4388   }
4389 
4390   // Dump all the signal
4391   if (sigismember(&check_signal_done, sig)) {
4392     print_signal_handlers(tty, buf, O_BUFLEN);
4393   }
4394 }
4395 
4396 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4397 
4398 extern bool signal_name(int signo, char* buf, size_t len);
4399 
4400 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4401   if (0 < exception_code && exception_code <= SIGRTMAX) {
4402     // signal
4403     if (!signal_name(exception_code, buf, size)) {
4404       jio_snprintf(buf, size, "SIG%d", exception_code);
4405     }
4406     return buf;
4407   } else {
4408     return NULL;
4409   }
4410 }
4411 
4412 // this is called _before_ the most of global arguments have been parsed
4413 void os::init(void) {
4414   char dummy;   /* used to get a guess on initial stack address */
4415 //  first_hrtime = gethrtime();
4416 
4417   // With BsdThreads the JavaMain thread pid (primordial thread)
4418   // is different than the pid of the java launcher thread.
4419   // So, on Bsd, the launcher thread pid is passed to the VM
4420   // via the sun.java.launcher.pid property.
4421   // Use this property instead of getpid() if it was correctly passed.
4422   // See bug 6351349.
4423   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4424 
4425   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4426 
4427   clock_tics_per_sec = CLK_TCK;
4428 
4429   init_random(1234567);
4430 
4431   ThreadCritical::initialize();
4432 
4433   Bsd::set_page_size(getpagesize());
4434   if (Bsd::page_size() == -1) {
4435     fatal(err_msg("os_bsd.cpp: os::init: sysconf failed (%s)",
4436                   strerror(errno)));
4437   }
4438   init_page_sizes((size_t) Bsd::page_size());
4439 
4440   Bsd::initialize_system_info();
4441 
4442   // main_thread points to the aboriginal thread
4443   Bsd::_main_thread = pthread_self();
4444 
4445   Bsd::clock_init();
4446   initial_time_count = os::elapsed_counter();
4447 
4448 #ifdef __APPLE__
4449   // XXXDARWIN
4450   // Work around the unaligned VM callbacks in hotspot's
4451   // sharedRuntime. The callbacks don't use SSE2 instructions, and work on
4452   // Linux, Solaris, and FreeBSD. On Mac OS X, dyld (rightly so) enforces
4453   // alignment when doing symbol lookup. To work around this, we force early
4454   // binding of all symbols now, thus binding when alignment is known-good.
4455   _dyld_bind_fully_image_containing_address((const void *) &os::init);
4456 #endif
4457 }
4458 
4459 // To install functions for atexit system call
4460 extern "C" {
4461   static void perfMemory_exit_helper() {
4462     perfMemory_exit();
4463   }
4464 }
4465 
4466 // this is called _after_ the global arguments have been parsed
4467 jint os::init_2(void)
4468 {
4469 #ifndef _ALLBSD_SOURCE
4470   Bsd::fast_thread_clock_init();
4471 #endif
4472 
4473   // Allocate a single page and mark it as readable for safepoint polling
4474   address polling_page = (address) ::mmap(NULL, Bsd::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4475   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4476 
4477   os::set_polling_page( polling_page );
4478 
4479 #ifndef PRODUCT
4480   if(Verbose && PrintMiscellaneous)
4481     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4482 #endif
4483 
4484   if (!UseMembar) {
4485     address mem_serialize_page = (address) ::mmap(NULL, Bsd::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4486     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4487     os::set_memory_serialize_page( mem_serialize_page );
4488 
4489 #ifndef PRODUCT
4490     if(Verbose && PrintMiscellaneous)
4491       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4492 #endif
4493   }
4494 
4495   os::large_page_init();
4496 
4497   // initialize suspend/resume support - must do this before signal_sets_init()
4498   if (SR_initialize() != 0) {
4499     perror("SR_initialize failed");
4500     return JNI_ERR;
4501   }
4502 
4503   Bsd::signal_sets_init();
4504   Bsd::install_signal_handlers();
4505 
4506   // Check minimum allowable stack size for thread creation and to initialize
4507   // the java system classes, including StackOverflowError - depends on page
4508   // size.  Add a page for compiler2 recursion in main thread.
4509   // Add in 2*BytesPerWord times page size to account for VM stack during
4510   // class initialization depending on 32 or 64 bit VM.
4511   os::Bsd::min_stack_allowed = MAX2(os::Bsd::min_stack_allowed,
4512             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4513                     2*BytesPerWord COMPILER2_PRESENT(+1)) * Bsd::page_size());
4514 
4515   size_t threadStackSizeInBytes = ThreadStackSize * K;
4516   if (threadStackSizeInBytes != 0 &&
4517       threadStackSizeInBytes < os::Bsd::min_stack_allowed) {
4518         tty->print_cr("\nThe stack size specified is too small, "
4519                       "Specify at least %dk",
4520                       os::Bsd::min_stack_allowed/ K);
4521         return JNI_ERR;
4522   }
4523 
4524   // Make the stack size a multiple of the page size so that
4525   // the yellow/red zones can be guarded.
4526   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4527         vm_page_size()));
4528 
4529 #ifndef _ALLBSD_SOURCE
4530   Bsd::capture_initial_stack(JavaThread::stack_size_at_create());
4531 
4532   Bsd::libpthread_init();
4533   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4534      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4535           Bsd::glibc_version(), Bsd::libpthread_version(),
4536           Bsd::is_floating_stack() ? "floating stack" : "fixed stack");
4537   }
4538 
4539   if (UseNUMA) {
4540     if (!Bsd::libnuma_init()) {
4541       UseNUMA = false;
4542     } else {
4543       if ((Bsd::numa_max_node() < 1)) {
4544         // There's only one node(they start from 0), disable NUMA.
4545         UseNUMA = false;
4546       }
4547     }
4548     // With SHM large pages we cannot uncommit a page, so there's not way
4549     // we can make the adaptive lgrp chunk resizing work. If the user specified
4550     // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4551     // disable adaptive resizing.
4552     if (UseNUMA && UseLargePages && UseSHM) {
4553       if (!FLAG_IS_DEFAULT(UseNUMA)) {
4554         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4555           UseLargePages = false;
4556         } else {
4557           warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4558           UseAdaptiveSizePolicy = false;
4559           UseAdaptiveNUMAChunkSizing = false;
4560         }
4561       } else {
4562         UseNUMA = false;
4563       }
4564     }
4565     if (!UseNUMA && ForceNUMA) {
4566       UseNUMA = true;
4567     }
4568   }
4569 #endif
4570 
4571   if (MaxFDLimit) {
4572     // set the number of file descriptors to max. print out error
4573     // if getrlimit/setrlimit fails but continue regardless.
4574     struct rlimit nbr_files;
4575     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4576     if (status != 0) {
4577       if (PrintMiscellaneous && (Verbose || WizardMode))
4578         perror("os::init_2 getrlimit failed");
4579     } else {
4580       nbr_files.rlim_cur = nbr_files.rlim_max;
4581 
4582 #ifdef __APPLE__
4583       // Darwin returns RLIM_INFINITY for rlim_max, but fails with EINVAL if
4584       // you attempt to use RLIM_INFINITY. As per setrlimit(2), OPEN_MAX must
4585       // be used instead
4586       nbr_files.rlim_cur = MIN(OPEN_MAX, nbr_files.rlim_cur);
4587 #endif
4588 
4589       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4590       if (status != 0) {
4591         if (PrintMiscellaneous && (Verbose || WizardMode))
4592           perror("os::init_2 setrlimit failed");
4593       }
4594     }
4595   }
4596 
4597 #ifndef _ALLBSD_SOURCE
4598   // Initialize lock used to serialize thread creation (see os::create_thread)
4599   Bsd::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4600 #endif
4601 
4602   // at-exit methods are called in the reverse order of their registration.
4603   // atexit functions are called on return from main or as a result of a
4604   // call to exit(3C). There can be only 32 of these functions registered
4605   // and atexit() does not set errno.
4606 
4607   if (PerfAllowAtExitRegistration) {
4608     // only register atexit functions if PerfAllowAtExitRegistration is set.
4609     // atexit functions can be delayed until process exit time, which
4610     // can be problematic for embedded VM situations. Embedded VMs should
4611     // call DestroyJavaVM() to assure that VM resources are released.
4612 
4613     // note: perfMemory_exit_helper atexit function may be removed in
4614     // the future if the appropriate cleanup code can be added to the
4615     // VM_Exit VMOperation's doit method.
4616     if (atexit(perfMemory_exit_helper) != 0) {
4617       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4618     }
4619   }
4620 
4621   // initialize thread priority policy
4622   prio_init();
4623 
4624 #ifdef __APPLE__
4625   // dynamically link to objective c gc registration
4626   void *handleLibObjc = dlopen(OBJC_LIB, RTLD_LAZY);
4627   if (handleLibObjc != NULL) {
4628     objc_registerThreadWithCollectorFunction = (objc_registerThreadWithCollector_t) dlsym(handleLibObjc, OBJC_GCREGISTER);
4629   }
4630 #endif
4631 
4632   return JNI_OK;
4633 }
4634 
4635 // this is called at the end of vm_initialization
4636 void os::init_3(void) { }
4637 
4638 // Mark the polling page as unreadable
4639 void os::make_polling_page_unreadable(void) {
4640   if( !guard_memory((char*)_polling_page, Bsd::page_size()) )
4641     fatal("Could not disable polling page");
4642 };
4643 
4644 // Mark the polling page as readable
4645 void os::make_polling_page_readable(void) {
4646   if( !bsd_mprotect((char *)_polling_page, Bsd::page_size(), PROT_READ)) {
4647     fatal("Could not enable polling page");
4648   }
4649 };
4650 
4651 int os::active_processor_count() {
4652 #ifdef _ALLBSD_SOURCE
4653   return _processor_count;
4654 #else
4655   // Bsd doesn't yet have a (official) notion of processor sets,
4656   // so just return the number of online processors.
4657   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4658   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4659   return online_cpus;
4660 #endif
4661 }
4662 
4663 void os::set_native_thread_name(const char *name) {
4664 #if defined(__APPLE__) && MAC_OS_X_VERSION_MIN_REQUIRED > MAC_OS_X_VERSION_10_5
4665   // This is only supported in Snow Leopard and beyond
4666   if (name != NULL) {
4667     // Add a "Java: " prefix to the name
4668     char buf[MAXTHREADNAMESIZE];
4669     snprintf(buf, sizeof(buf), "Java: %s", name);
4670     pthread_setname_np(buf);
4671   }
4672 #endif
4673 }
4674 
4675 bool os::distribute_processes(uint length, uint* distribution) {
4676   // Not yet implemented.
4677   return false;
4678 }
4679 
4680 bool os::bind_to_processor(uint processor_id) {
4681   // Not yet implemented.
4682   return false;
4683 }
4684 
4685 ///
4686 
4687 // Suspends the target using the signal mechanism and then grabs the PC before
4688 // resuming the target. Used by the flat-profiler only
4689 ExtendedPC os::get_thread_pc(Thread* thread) {
4690   // Make sure that it is called by the watcher for the VMThread
4691   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4692   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4693 
4694   ExtendedPC epc;
4695 
4696   OSThread* osthread = thread->osthread();
4697   if (do_suspend(osthread)) {
4698     if (osthread->ucontext() != NULL) {
4699       epc = os::Bsd::ucontext_get_pc(osthread->ucontext());
4700     } else {
4701       // NULL context is unexpected, double-check this is the VMThread
4702       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4703     }
4704     do_resume(osthread);
4705   }
4706   // failure means pthread_kill failed for some reason - arguably this is
4707   // a fatal problem, but such problems are ignored elsewhere
4708 
4709   return epc;
4710 }
4711 
4712 int os::Bsd::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4713 {
4714 #ifdef _ALLBSD_SOURCE
4715   return pthread_cond_timedwait(_cond, _mutex, _abstime);
4716 #else
4717    if (is_NPTL()) {
4718       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4719    } else {
4720 #ifndef IA64
4721       // 6292965: BsdThreads pthread_cond_timedwait() resets FPU control
4722       // word back to default 64bit precision if condvar is signaled. Java
4723       // wants 53bit precision.  Save and restore current value.
4724       int fpu = get_fpu_control_word();
4725 #endif // IA64
4726       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4727 #ifndef IA64
4728       set_fpu_control_word(fpu);
4729 #endif // IA64
4730       return status;
4731    }
4732 #endif
4733 }
4734 
4735 ////////////////////////////////////////////////////////////////////////////////
4736 // debug support
4737 
4738 static address same_page(address x, address y) {
4739   int page_bits = -os::vm_page_size();
4740   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4741     return x;
4742   else if (x > y)
4743     return (address)(intptr_t(y) | ~page_bits) + 1;
4744   else
4745     return (address)(intptr_t(y) & page_bits);
4746 }
4747 
4748 bool os::find(address addr, outputStream* st) {
4749   Dl_info dlinfo;
4750   memset(&dlinfo, 0, sizeof(dlinfo));
4751   if (dladdr(addr, &dlinfo)) {
4752     st->print(PTR_FORMAT ": ", addr);
4753     if (dlinfo.dli_sname != NULL) {
4754       st->print("%s+%#x", dlinfo.dli_sname,
4755                  addr - (intptr_t)dlinfo.dli_saddr);
4756     } else if (dlinfo.dli_fname) {
4757       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4758     } else {
4759       st->print("<absolute address>");
4760     }
4761     if (dlinfo.dli_fname) {
4762       st->print(" in %s", dlinfo.dli_fname);
4763     }
4764     if (dlinfo.dli_fbase) {
4765       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4766     }
4767     st->cr();
4768 
4769     if (Verbose) {
4770       // decode some bytes around the PC
4771       address begin = same_page(addr-40, addr);
4772       address end   = same_page(addr+40, addr);
4773       address       lowest = (address) dlinfo.dli_sname;
4774       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4775       if (begin < lowest)  begin = lowest;
4776       Dl_info dlinfo2;
4777       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4778           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4779         end = (address) dlinfo2.dli_saddr;
4780       Disassembler::decode(begin, end, st);
4781     }
4782     return true;
4783   }
4784   return false;
4785 }
4786 
4787 ////////////////////////////////////////////////////////////////////////////////
4788 // misc
4789 
4790 // This does not do anything on Bsd. This is basically a hook for being
4791 // able to use structured exception handling (thread-local exception filters)
4792 // on, e.g., Win32.
4793 void
4794 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4795                          JavaCallArguments* args, Thread* thread) {
4796   f(value, method, args, thread);
4797 }
4798 
4799 void os::print_statistics() {
4800 }
4801 
4802 int os::message_box(const char* title, const char* message) {
4803   int i;
4804   fdStream err(defaultStream::error_fd());
4805   for (i = 0; i < 78; i++) err.print_raw("=");
4806   err.cr();
4807   err.print_raw_cr(title);
4808   for (i = 0; i < 78; i++) err.print_raw("-");
4809   err.cr();
4810   err.print_raw_cr(message);
4811   for (i = 0; i < 78; i++) err.print_raw("=");
4812   err.cr();
4813 
4814   char buf[16];
4815   // Prevent process from exiting upon "read error" without consuming all CPU
4816   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4817 
4818   return buf[0] == 'y' || buf[0] == 'Y';
4819 }
4820 
4821 int os::stat(const char *path, struct stat *sbuf) {
4822   char pathbuf[MAX_PATH];
4823   if (strlen(path) > MAX_PATH - 1) {
4824     errno = ENAMETOOLONG;
4825     return -1;
4826   }
4827   os::native_path(strcpy(pathbuf, path));
4828   return ::stat(pathbuf, sbuf);
4829 }
4830 
4831 bool os::check_heap(bool force) {
4832   return true;
4833 }
4834 
4835 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4836   return ::vsnprintf(buf, count, format, args);
4837 }
4838 
4839 // Is a (classpath) directory empty?
4840 bool os::dir_is_empty(const char* path) {
4841   DIR *dir = NULL;
4842   struct dirent *ptr;
4843 
4844   dir = opendir(path);
4845   if (dir == NULL) return true;
4846 
4847   /* Scan the directory */
4848   bool result = true;
4849   char buf[sizeof(struct dirent) + MAX_PATH];
4850   while (result && (ptr = ::readdir(dir)) != NULL) {
4851     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4852       result = false;
4853     }
4854   }
4855   closedir(dir);
4856   return result;
4857 }
4858 
4859 // This code originates from JDK's sysOpen and open64_w
4860 // from src/solaris/hpi/src/system_md.c
4861 
4862 #ifndef O_DELETE
4863 #define O_DELETE 0x10000
4864 #endif
4865 
4866 // Open a file. Unlink the file immediately after open returns
4867 // if the specified oflag has the O_DELETE flag set.
4868 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4869 
4870 int os::open(const char *path, int oflag, int mode) {
4871 
4872   if (strlen(path) > MAX_PATH - 1) {
4873     errno = ENAMETOOLONG;
4874     return -1;
4875   }
4876   int fd;
4877   int o_delete = (oflag & O_DELETE);
4878   oflag = oflag & ~O_DELETE;
4879 
4880   fd = ::open(path, oflag, mode);
4881   if (fd == -1) return -1;
4882 
4883   //If the open succeeded, the file might still be a directory
4884   {
4885     struct stat buf;
4886     int ret = ::fstat(fd, &buf);
4887     int st_mode = buf.st_mode;
4888 
4889     if (ret != -1) {
4890       if ((st_mode & S_IFMT) == S_IFDIR) {
4891         errno = EISDIR;
4892         ::close(fd);
4893         return -1;
4894       }
4895     } else {
4896       ::close(fd);
4897       return -1;
4898     }
4899   }
4900 
4901     /*
4902      * All file descriptors that are opened in the JVM and not
4903      * specifically destined for a subprocess should have the
4904      * close-on-exec flag set.  If we don't set it, then careless 3rd
4905      * party native code might fork and exec without closing all
4906      * appropriate file descriptors (e.g. as we do in closeDescriptors in
4907      * UNIXProcess.c), and this in turn might:
4908      *
4909      * - cause end-of-file to fail to be detected on some file
4910      *   descriptors, resulting in mysterious hangs, or
4911      *
4912      * - might cause an fopen in the subprocess to fail on a system
4913      *   suffering from bug 1085341.
4914      *
4915      * (Yes, the default setting of the close-on-exec flag is a Unix
4916      * design flaw)
4917      *
4918      * See:
4919      * 1085341: 32-bit stdio routines should support file descriptors >255
4920      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4921      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4922      */
4923 #ifdef FD_CLOEXEC
4924     {
4925         int flags = ::fcntl(fd, F_GETFD);
4926         if (flags != -1)
4927             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4928     }
4929 #endif
4930 
4931   if (o_delete != 0) {
4932     ::unlink(path);
4933   }
4934   return fd;
4935 }
4936 
4937 
4938 // create binary file, rewriting existing file if required
4939 int os::create_binary_file(const char* path, bool rewrite_existing) {
4940   int oflags = O_WRONLY | O_CREAT;
4941   if (!rewrite_existing) {
4942     oflags |= O_EXCL;
4943   }
4944   return ::open(path, oflags, S_IREAD | S_IWRITE);
4945 }
4946 
4947 // return current position of file pointer
4948 jlong os::current_file_offset(int fd) {
4949   return (jlong)::lseek(fd, (off_t)0, SEEK_CUR);
4950 }
4951 
4952 // move file pointer to the specified offset
4953 jlong os::seek_to_file_offset(int fd, jlong offset) {
4954   return (jlong)::lseek(fd, (off_t)offset, SEEK_SET);
4955 }
4956 
4957 // This code originates from JDK's sysAvailable
4958 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4959 
4960 int os::available(int fd, jlong *bytes) {
4961   jlong cur, end;
4962   int mode;
4963   struct stat buf;
4964 
4965   if (::fstat(fd, &buf) >= 0) {
4966     mode = buf.st_mode;
4967     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4968       /*
4969       * XXX: is the following call interruptible? If so, this might
4970       * need to go through the INTERRUPT_IO() wrapper as for other
4971       * blocking, interruptible calls in this file.
4972       */
4973       int n;
4974       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4975         *bytes = n;
4976         return 1;
4977       }
4978     }
4979   }
4980   if ((cur = ::lseek(fd, 0L, SEEK_CUR)) == -1) {
4981     return 0;
4982   } else if ((end = ::lseek(fd, 0L, SEEK_END)) == -1) {
4983     return 0;
4984   } else if (::lseek(fd, cur, SEEK_SET) == -1) {
4985     return 0;
4986   }
4987   *bytes = end - cur;
4988   return 1;
4989 }
4990 
4991 int os::socket_available(int fd, jint *pbytes) {
4992    if (fd < 0)
4993      return OS_OK;
4994 
4995    int ret;
4996 
4997    RESTARTABLE(::ioctl(fd, FIONREAD, pbytes), ret);
4998 
4999    //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5000    // is expected to return 0 on failure and 1 on success to the jdk.
5001 
5002    return (ret == OS_ERR) ? 0 : 1;
5003 }
5004 
5005 // Map a block of memory.
5006 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
5007                      char *addr, size_t bytes, bool read_only,
5008                      bool allow_exec) {
5009   int prot;
5010   int flags;
5011 
5012   if (read_only) {
5013     prot = PROT_READ;
5014     flags = MAP_SHARED;
5015   } else {
5016     prot = PROT_READ | PROT_WRITE;
5017     flags = MAP_PRIVATE;
5018   }
5019 
5020   if (allow_exec) {
5021     prot |= PROT_EXEC;
5022   }
5023 
5024   if (addr != NULL) {
5025     flags |= MAP_FIXED;
5026   }
5027 
5028   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5029                                      fd, file_offset);
5030   if (mapped_address == MAP_FAILED) {
5031     return NULL;
5032   }
5033   return mapped_address;
5034 }
5035 
5036 
5037 // Remap a block of memory.
5038 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
5039                        char *addr, size_t bytes, bool read_only,
5040                        bool allow_exec) {
5041   // same as map_memory() on this OS
5042   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5043                         allow_exec);
5044 }
5045 
5046 
5047 // Unmap a block of memory.
5048 bool os::unmap_memory(char* addr, size_t bytes) {
5049   return munmap(addr, bytes) == 0;
5050 }
5051 
5052 #ifndef _ALLBSD_SOURCE
5053 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5054 
5055 static clockid_t thread_cpu_clockid(Thread* thread) {
5056   pthread_t tid = thread->osthread()->pthread_id();
5057   clockid_t clockid;
5058 
5059   // Get thread clockid
5060   int rc = os::Bsd::pthread_getcpuclockid(tid, &clockid);
5061   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5062   return clockid;
5063 }
5064 #endif
5065 
5066 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5067 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5068 // of a thread.
5069 //
5070 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5071 // the fast estimate available on the platform.
5072 
5073 jlong os::current_thread_cpu_time() {
5074 #ifdef __APPLE__
5075   return os::thread_cpu_time(Thread::current(), true /* user + sys */);
5076 #elif !defined(_ALLBSD_SOURCE)
5077   if (os::Bsd::supports_fast_thread_cpu_time()) {
5078     return os::Bsd::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5079   } else {
5080     // return user + sys since the cost is the same
5081     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5082   }
5083 #endif
5084 }
5085 
5086 jlong os::thread_cpu_time(Thread* thread) {
5087 #ifndef _ALLBSD_SOURCE
5088   // consistent with what current_thread_cpu_time() returns
5089   if (os::Bsd::supports_fast_thread_cpu_time()) {
5090     return os::Bsd::fast_thread_cpu_time(thread_cpu_clockid(thread));
5091   } else {
5092     return slow_thread_cpu_time(thread, true /* user + sys */);
5093   }
5094 #endif
5095 }
5096 
5097 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5098 #ifdef __APPLE__
5099   return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5100 #elif !defined(_ALLBSD_SOURCE)
5101   if (user_sys_cpu_time && os::Bsd::supports_fast_thread_cpu_time()) {
5102     return os::Bsd::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5103   } else {
5104     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5105   }
5106 #endif
5107 }
5108 
5109 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5110 #ifdef __APPLE__
5111   struct thread_basic_info tinfo;
5112   mach_msg_type_number_t tcount = THREAD_INFO_MAX;
5113   kern_return_t kr;
5114   mach_port_t mach_thread;
5115 
5116   mach_thread = pthread_mach_thread_np(thread->osthread()->thread_id());
5117   kr = thread_info(mach_thread, THREAD_BASIC_INFO, (thread_info_t)&tinfo, &tcount);
5118   if (kr != KERN_SUCCESS)
5119     return -1;
5120 
5121   if (user_sys_cpu_time) {
5122     jlong nanos;
5123     nanos = ((jlong) tinfo.system_time.seconds + tinfo.user_time.seconds) * (jlong)1000000000;
5124     nanos += ((jlong) tinfo.system_time.microseconds + (jlong) tinfo.user_time.microseconds) * (jlong)1000;
5125     return nanos;
5126   } else {
5127     return ((jlong)tinfo.user_time.seconds * 1000000000) + ((jlong)tinfo.user_time.microseconds * (jlong)1000);
5128   }
5129 #elif !defined(_ALLBSD_SOURCE)
5130   if (user_sys_cpu_time && os::Bsd::supports_fast_thread_cpu_time()) {
5131     return os::Bsd::fast_thread_cpu_time(thread_cpu_clockid(thread));
5132   } else {
5133     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5134   }
5135 #endif
5136 }
5137 
5138 #ifndef _ALLBSD_SOURCE
5139 //
5140 //  -1 on error.
5141 //
5142 
5143 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5144   static bool proc_pid_cpu_avail = true;
5145   static bool proc_task_unchecked = true;
5146   static const char *proc_stat_path = "/proc/%d/stat";
5147   pid_t  tid = thread->osthread()->thread_id();
5148   int i;
5149   char *s;
5150   char stat[2048];
5151   int statlen;
5152   char proc_name[64];
5153   int count;
5154   long sys_time, user_time;
5155   char string[64];
5156   char cdummy;
5157   int idummy;
5158   long ldummy;
5159   FILE *fp;
5160 
5161   // We first try accessing /proc/<pid>/cpu since this is faster to
5162   // process.  If this file is not present (bsd kernels 2.5 and above)
5163   // then we open /proc/<pid>/stat.
5164   if ( proc_pid_cpu_avail ) {
5165     sprintf(proc_name, "/proc/%d/cpu", tid);
5166     fp =  fopen(proc_name, "r");
5167     if ( fp != NULL ) {
5168       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
5169       fclose(fp);
5170       if ( count != 3 ) return -1;
5171 
5172       if (user_sys_cpu_time) {
5173         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5174       } else {
5175         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5176       }
5177     }
5178     else proc_pid_cpu_avail = false;
5179   }
5180 
5181   // The /proc/<tid>/stat aggregates per-process usage on
5182   // new Bsd kernels 2.6+ where NPTL is supported.
5183   // The /proc/self/task/<tid>/stat still has the per-thread usage.
5184   // See bug 6328462.
5185   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
5186   // and possibly in some other cases, so we check its availability.
5187   if (proc_task_unchecked && os::Bsd::is_NPTL()) {
5188     // This is executed only once
5189     proc_task_unchecked = false;
5190     fp = fopen("/proc/self/task", "r");
5191     if (fp != NULL) {
5192       proc_stat_path = "/proc/self/task/%d/stat";
5193       fclose(fp);
5194     }
5195   }
5196 
5197   sprintf(proc_name, proc_stat_path, tid);
5198   fp = fopen(proc_name, "r");
5199   if ( fp == NULL ) return -1;
5200   statlen = fread(stat, 1, 2047, fp);
5201   stat[statlen] = '\0';
5202   fclose(fp);
5203 
5204   // Skip pid and the command string. Note that we could be dealing with
5205   // weird command names, e.g. user could decide to rename java launcher
5206   // to "java 1.4.2 :)", then the stat file would look like
5207   //                1234 (java 1.4.2 :)) R ... ...
5208   // We don't really need to know the command string, just find the last
5209   // occurrence of ")" and then start parsing from there. See bug 4726580.
5210   s = strrchr(stat, ')');
5211   i = 0;
5212   if (s == NULL ) return -1;
5213 
5214   // Skip blank chars
5215   do s++; while (isspace(*s));
5216 
5217   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5218                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5219                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5220                  &user_time, &sys_time);
5221   if ( count != 13 ) return -1;
5222   if (user_sys_cpu_time) {
5223     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5224   } else {
5225     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5226   }
5227 }
5228 #endif
5229 
5230 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5231   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5232   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5233   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5234   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5235 }
5236 
5237 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5238   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5239   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5240   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5241   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5242 }
5243 
5244 bool os::is_thread_cpu_time_supported() {
5245 #ifdef __APPLE__
5246   return true;
5247 #elif defined(_ALLBSD_SOURCE)
5248   return false;
5249 #else
5250   return true;
5251 #endif
5252 }
5253 
5254 // System loadavg support.  Returns -1 if load average cannot be obtained.
5255 // Bsd doesn't yet have a (official) notion of processor sets,
5256 // so just return the system wide load average.
5257 int os::loadavg(double loadavg[], int nelem) {
5258   return ::getloadavg(loadavg, nelem);
5259 }
5260 
5261 void os::pause() {
5262   char filename[MAX_PATH];
5263   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5264     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5265   } else {
5266     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5267   }
5268 
5269   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5270   if (fd != -1) {
5271     struct stat buf;
5272     ::close(fd);
5273     while (::stat(filename, &buf) == 0) {
5274       (void)::poll(NULL, 0, 100);
5275     }
5276   } else {
5277     jio_fprintf(stderr,
5278       "Could not open pause file '%s', continuing immediately.\n", filename);
5279   }
5280 }
5281 
5282 
5283 // Refer to the comments in os_solaris.cpp park-unpark.
5284 //
5285 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5286 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5287 // For specifics regarding the bug see GLIBC BUGID 261237 :
5288 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5289 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5290 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5291 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5292 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5293 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5294 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5295 // of libpthread avoids the problem, but isn't practical.
5296 //
5297 // Possible remedies:
5298 //
5299 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5300 //      This is palliative and probabilistic, however.  If the thread is preempted
5301 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5302 //      than the minimum period may have passed, and the abstime may be stale (in the
5303 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5304 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5305 //
5306 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5307 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5308 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5309 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5310 //      thread.
5311 //
5312 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5313 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5314 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5315 //      This also works well.  In fact it avoids kernel-level scalability impediments
5316 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5317 //      timers in a graceful fashion.
5318 //
5319 // 4.   When the abstime value is in the past it appears that control returns
5320 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5321 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5322 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5323 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5324 //      It may be possible to avoid reinitialization by checking the return
5325 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5326 //      condvar we must establish the invariant that cond_signal() is only called
5327 //      within critical sections protected by the adjunct mutex.  This prevents
5328 //      cond_signal() from "seeing" a condvar that's in the midst of being
5329 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5330 //      desirable signal-after-unlock optimization that avoids futile context switching.
5331 //
5332 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5333 //      structure when a condvar is used or initialized.  cond_destroy()  would
5334 //      release the helper structure.  Our reinitialize-after-timedwait fix
5335 //      put excessive stress on malloc/free and locks protecting the c-heap.
5336 //
5337 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5338 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5339 // and only enabling the work-around for vulnerable environments.
5340 
5341 // utility to compute the abstime argument to timedwait:
5342 // millis is the relative timeout time
5343 // abstime will be the absolute timeout time
5344 // TODO: replace compute_abstime() with unpackTime()
5345 
5346 static struct timespec* compute_abstime(struct timespec* abstime, jlong millis) {
5347   if (millis < 0)  millis = 0;
5348   struct timeval now;
5349   int status = gettimeofday(&now, NULL);
5350   assert(status == 0, "gettimeofday");
5351   jlong seconds = millis / 1000;
5352   millis %= 1000;
5353   if (seconds > 50000000) { // see man cond_timedwait(3T)
5354     seconds = 50000000;
5355   }
5356   abstime->tv_sec = now.tv_sec  + seconds;
5357   long       usec = now.tv_usec + millis * 1000;
5358   if (usec >= 1000000) {
5359     abstime->tv_sec += 1;
5360     usec -= 1000000;
5361   }
5362   abstime->tv_nsec = usec * 1000;
5363   return abstime;
5364 }
5365 
5366 
5367 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5368 // Conceptually TryPark() should be equivalent to park(0).
5369 
5370 int os::PlatformEvent::TryPark() {
5371   for (;;) {
5372     const int v = _Event ;
5373     guarantee ((v == 0) || (v == 1), "invariant") ;
5374     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5375   }
5376 }
5377 
5378 void os::PlatformEvent::park() {       // AKA "down()"
5379   // Invariant: Only the thread associated with the Event/PlatformEvent
5380   // may call park().
5381   // TODO: assert that _Assoc != NULL or _Assoc == Self
5382   int v ;
5383   for (;;) {
5384       v = _Event ;
5385       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5386   }
5387   guarantee (v >= 0, "invariant") ;
5388   if (v == 0) {
5389      // Do this the hard way by blocking ...
5390      int status = pthread_mutex_lock(_mutex);
5391      assert_status(status == 0, status, "mutex_lock");
5392      guarantee (_nParked == 0, "invariant") ;
5393      ++ _nParked ;
5394      while (_Event < 0) {
5395         status = pthread_cond_wait(_cond, _mutex);
5396         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5397         // Treat this the same as if the wait was interrupted
5398         if (status == ETIMEDOUT) { status = EINTR; }
5399         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5400      }
5401      -- _nParked ;
5402 
5403     // In theory we could move the ST of 0 into _Event past the unlock(),
5404     // but then we'd need a MEMBAR after the ST.
5405     _Event = 0 ;
5406      status = pthread_mutex_unlock(_mutex);
5407      assert_status(status == 0, status, "mutex_unlock");
5408   }
5409   guarantee (_Event >= 0, "invariant") ;
5410 }
5411 
5412 int os::PlatformEvent::park(jlong millis) {
5413   guarantee (_nParked == 0, "invariant") ;
5414 
5415   int v ;
5416   for (;;) {
5417       v = _Event ;
5418       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5419   }
5420   guarantee (v >= 0, "invariant") ;
5421   if (v != 0) return OS_OK ;
5422 
5423   // We do this the hard way, by blocking the thread.
5424   // Consider enforcing a minimum timeout value.
5425   struct timespec abst;
5426   compute_abstime(&abst, millis);
5427 
5428   int ret = OS_TIMEOUT;
5429   int status = pthread_mutex_lock(_mutex);
5430   assert_status(status == 0, status, "mutex_lock");
5431   guarantee (_nParked == 0, "invariant") ;
5432   ++_nParked ;
5433 
5434   // Object.wait(timo) will return because of
5435   // (a) notification
5436   // (b) timeout
5437   // (c) thread.interrupt
5438   //
5439   // Thread.interrupt and object.notify{All} both call Event::set.
5440   // That is, we treat thread.interrupt as a special case of notification.
5441   // The underlying Solaris implementation, cond_timedwait, admits
5442   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5443   // JVM from making those visible to Java code.  As such, we must
5444   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5445   //
5446   // TODO: properly differentiate simultaneous notify+interrupt.
5447   // In that case, we should propagate the notify to another waiter.
5448 
5449   while (_Event < 0) {
5450     status = os::Bsd::safe_cond_timedwait(_cond, _mutex, &abst);
5451     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5452       pthread_cond_destroy (_cond);
5453       pthread_cond_init (_cond, NULL) ;
5454     }
5455     assert_status(status == 0 || status == EINTR ||
5456                   status == ETIMEDOUT,
5457                   status, "cond_timedwait");
5458     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5459     if (status == ETIMEDOUT) break ;
5460     // We consume and ignore EINTR and spurious wakeups.
5461   }
5462   --_nParked ;
5463   if (_Event >= 0) {
5464      ret = OS_OK;
5465   }
5466   _Event = 0 ;
5467   status = pthread_mutex_unlock(_mutex);
5468   assert_status(status == 0, status, "mutex_unlock");
5469   assert (_nParked == 0, "invariant") ;
5470   return ret;
5471 }
5472 
5473 void os::PlatformEvent::unpark() {
5474   int v, AnyWaiters ;
5475   for (;;) {
5476       v = _Event ;
5477       if (v > 0) {
5478          // The LD of _Event could have reordered or be satisfied
5479          // by a read-aside from this processor's write buffer.
5480          // To avoid problems execute a barrier and then
5481          // ratify the value.
5482          OrderAccess::fence() ;
5483          if (_Event == v) return ;
5484          continue ;
5485       }
5486       if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5487   }
5488   if (v < 0) {
5489      // Wait for the thread associated with the event to vacate
5490      int status = pthread_mutex_lock(_mutex);
5491      assert_status(status == 0, status, "mutex_lock");
5492      AnyWaiters = _nParked ;
5493      assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5494      if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5495         AnyWaiters = 0 ;
5496         pthread_cond_signal (_cond);
5497      }
5498      status = pthread_mutex_unlock(_mutex);
5499      assert_status(status == 0, status, "mutex_unlock");
5500      if (AnyWaiters != 0) {
5501         status = pthread_cond_signal(_cond);
5502         assert_status(status == 0, status, "cond_signal");
5503      }
5504   }
5505 
5506   // Note that we signal() _after dropping the lock for "immortal" Events.
5507   // This is safe and avoids a common class of  futile wakeups.  In rare
5508   // circumstances this can cause a thread to return prematurely from
5509   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5510   // simply re-test the condition and re-park itself.
5511 }
5512 
5513 
5514 // JSR166
5515 // -------------------------------------------------------
5516 
5517 /*
5518  * The solaris and bsd implementations of park/unpark are fairly
5519  * conservative for now, but can be improved. They currently use a
5520  * mutex/condvar pair, plus a a count.
5521  * Park decrements count if > 0, else does a condvar wait.  Unpark
5522  * sets count to 1 and signals condvar.  Only one thread ever waits
5523  * on the condvar. Contention seen when trying to park implies that someone
5524  * is unparking you, so don't wait. And spurious returns are fine, so there
5525  * is no need to track notifications.
5526  */
5527 
5528 #define MAX_SECS 100000000
5529 /*
5530  * This code is common to bsd and solaris and will be moved to a
5531  * common place in dolphin.
5532  *
5533  * The passed in time value is either a relative time in nanoseconds
5534  * or an absolute time in milliseconds. Either way it has to be unpacked
5535  * into suitable seconds and nanoseconds components and stored in the
5536  * given timespec structure.
5537  * Given time is a 64-bit value and the time_t used in the timespec is only
5538  * a signed-32-bit value (except on 64-bit Bsd) we have to watch for
5539  * overflow if times way in the future are given. Further on Solaris versions
5540  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5541  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5542  * As it will be 28 years before "now + 100000000" will overflow we can
5543  * ignore overflow and just impose a hard-limit on seconds using the value
5544  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5545  * years from "now".
5546  */
5547 
5548 static void unpackTime(struct timespec* absTime, bool isAbsolute, jlong time) {
5549   assert (time > 0, "convertTime");
5550 
5551   struct timeval now;
5552   int status = gettimeofday(&now, NULL);
5553   assert(status == 0, "gettimeofday");
5554 
5555   time_t max_secs = now.tv_sec + MAX_SECS;
5556 
5557   if (isAbsolute) {
5558     jlong secs = time / 1000;
5559     if (secs > max_secs) {
5560       absTime->tv_sec = max_secs;
5561     }
5562     else {
5563       absTime->tv_sec = secs;
5564     }
5565     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5566   }
5567   else {
5568     jlong secs = time / NANOSECS_PER_SEC;
5569     if (secs >= MAX_SECS) {
5570       absTime->tv_sec = max_secs;
5571       absTime->tv_nsec = 0;
5572     }
5573     else {
5574       absTime->tv_sec = now.tv_sec + secs;
5575       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5576       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5577         absTime->tv_nsec -= NANOSECS_PER_SEC;
5578         ++absTime->tv_sec; // note: this must be <= max_secs
5579       }
5580     }
5581   }
5582   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5583   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5584   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5585   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5586 }
5587 
5588 void Parker::park(bool isAbsolute, jlong time) {
5589   // Optional fast-path check:
5590   // Return immediately if a permit is available.
5591   if (_counter > 0) {
5592       _counter = 0 ;
5593       OrderAccess::fence();
5594       return ;
5595   }
5596 
5597   Thread* thread = Thread::current();
5598   assert(thread->is_Java_thread(), "Must be JavaThread");
5599   JavaThread *jt = (JavaThread *)thread;
5600 
5601   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5602   // Check interrupt before trying to wait
5603   if (Thread::is_interrupted(thread, false)) {
5604     return;
5605   }
5606 
5607   // Next, demultiplex/decode time arguments
5608   struct timespec absTime;
5609   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5610     return;
5611   }
5612   if (time > 0) {
5613     unpackTime(&absTime, isAbsolute, time);
5614   }
5615 
5616 
5617   // Enter safepoint region
5618   // Beware of deadlocks such as 6317397.
5619   // The per-thread Parker:: mutex is a classic leaf-lock.
5620   // In particular a thread must never block on the Threads_lock while
5621   // holding the Parker:: mutex.  If safepoints are pending both the
5622   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5623   ThreadBlockInVM tbivm(jt);
5624 
5625   // Don't wait if cannot get lock since interference arises from
5626   // unblocking.  Also. check interrupt before trying wait
5627   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5628     return;
5629   }
5630 
5631   int status ;
5632   if (_counter > 0)  { // no wait needed
5633     _counter = 0;
5634     status = pthread_mutex_unlock(_mutex);
5635     assert (status == 0, "invariant") ;
5636     OrderAccess::fence();
5637     return;
5638   }
5639 
5640 #ifdef ASSERT
5641   // Don't catch signals while blocked; let the running threads have the signals.
5642   // (This allows a debugger to break into the running thread.)
5643   sigset_t oldsigs;
5644   sigset_t* allowdebug_blocked = os::Bsd::allowdebug_blocked_signals();
5645   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5646 #endif
5647 
5648   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5649   jt->set_suspend_equivalent();
5650   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5651 
5652   if (time == 0) {
5653     status = pthread_cond_wait (_cond, _mutex) ;
5654   } else {
5655     status = os::Bsd::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5656     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5657       pthread_cond_destroy (_cond) ;
5658       pthread_cond_init    (_cond, NULL);
5659     }
5660   }
5661   assert_status(status == 0 || status == EINTR ||
5662                 status == ETIMEDOUT,
5663                 status, "cond_timedwait");
5664 
5665 #ifdef ASSERT
5666   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5667 #endif
5668 
5669   _counter = 0 ;
5670   status = pthread_mutex_unlock(_mutex) ;
5671   assert_status(status == 0, status, "invariant") ;
5672   // If externally suspended while waiting, re-suspend
5673   if (jt->handle_special_suspend_equivalent_condition()) {
5674     jt->java_suspend_self();
5675   }
5676 
5677   OrderAccess::fence();
5678 }
5679 
5680 void Parker::unpark() {
5681   int s, status ;
5682   status = pthread_mutex_lock(_mutex);
5683   assert (status == 0, "invariant") ;
5684   s = _counter;
5685   _counter = 1;
5686   if (s < 1) {
5687      if (WorkAroundNPTLTimedWaitHang) {
5688         status = pthread_cond_signal (_cond) ;
5689         assert (status == 0, "invariant") ;
5690         status = pthread_mutex_unlock(_mutex);
5691         assert (status == 0, "invariant") ;
5692      } else {
5693         status = pthread_mutex_unlock(_mutex);
5694         assert (status == 0, "invariant") ;
5695         status = pthread_cond_signal (_cond) ;
5696         assert (status == 0, "invariant") ;
5697      }
5698   } else {
5699     pthread_mutex_unlock(_mutex);
5700     assert (status == 0, "invariant") ;
5701   }
5702 }
5703 
5704 
5705 /* Darwin has no "environ" in a dynamic library. */
5706 #ifdef __APPLE__
5707 #include <crt_externs.h>
5708 #define environ (*_NSGetEnviron())
5709 #else
5710 extern char** environ;
5711 #endif
5712 
5713 // Run the specified command in a separate process. Return its exit value,
5714 // or -1 on failure (e.g. can't fork a new process).
5715 // Unlike system(), this function can be called from signal handler. It
5716 // doesn't block SIGINT et al.
5717 int os::fork_and_exec(char* cmd) {
5718   const char * argv[4] = {"sh", "-c", cmd, NULL};
5719 
5720   // fork() in BsdThreads/NPTL is not async-safe. It needs to run
5721   // pthread_atfork handlers and reset pthread library. All we need is a
5722   // separate process to execve. Make a direct syscall to fork process.
5723   // On IA64 there's no fork syscall, we have to use fork() and hope for
5724   // the best...
5725   pid_t pid = fork();
5726 
5727   if (pid < 0) {
5728     // fork failed
5729     return -1;
5730 
5731   } else if (pid == 0) {
5732     // child process
5733 
5734     // execve() in BsdThreads will call pthread_kill_other_threads_np()
5735     // first to kill every thread on the thread list. Because this list is
5736     // not reset by fork() (see notes above), execve() will instead kill
5737     // every thread in the parent process. We know this is the only thread
5738     // in the new process, so make a system call directly.
5739     // IA64 should use normal execve() from glibc to match the glibc fork()
5740     // above.
5741     execve("/bin/sh", (char* const*)argv, environ);
5742 
5743     // execve failed
5744     _exit(-1);
5745 
5746   } else  {
5747     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5748     // care about the actual exit code, for now.
5749 
5750     int status;
5751 
5752     // Wait for the child process to exit.  This returns immediately if
5753     // the child has already exited. */
5754     while (waitpid(pid, &status, 0) < 0) {
5755         switch (errno) {
5756         case ECHILD: return 0;
5757         case EINTR: break;
5758         default: return -1;
5759         }
5760     }
5761 
5762     if (WIFEXITED(status)) {
5763        // The child exited normally; get its exit code.
5764        return WEXITSTATUS(status);
5765     } else if (WIFSIGNALED(status)) {
5766        // The child exited because of a signal
5767        // The best value to return is 0x80 + signal number,
5768        // because that is what all Unix shells do, and because
5769        // it allows callers to distinguish between process exit and
5770        // process death by signal.
5771        return 0x80 + WTERMSIG(status);
5772     } else {
5773        // Unknown exit code; pass it through
5774        return status;
5775     }
5776   }
5777 }
5778 
5779 // is_headless_jre()
5780 //
5781 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5782 // in order to report if we are running in a headless jre
5783 //
5784 // Since JDK8 xawt/libmawt.so was moved into the same directory
5785 // as libawt.so, and renamed libawt_xawt.so
5786 //
5787 bool os::is_headless_jre() {
5788     struct stat statbuf;
5789     char buf[MAXPATHLEN];
5790     char libmawtpath[MAXPATHLEN];
5791     const char *xawtstr  = "/xawt/libmawt" JNI_LIB_SUFFIX;
5792     const char *new_xawtstr = "/libawt_xawt" JNI_LIB_SUFFIX;
5793     char *p;
5794 
5795     // Get path to libjvm.so
5796     os::jvm_path(buf, sizeof(buf));
5797 
5798     // Get rid of libjvm.so
5799     p = strrchr(buf, '/');
5800     if (p == NULL) return false;
5801     else *p = '\0';
5802 
5803     // Get rid of client or server
5804     p = strrchr(buf, '/');
5805     if (p == NULL) return false;
5806     else *p = '\0';
5807 
5808     // check xawt/libmawt.so
5809     strcpy(libmawtpath, buf);
5810     strcat(libmawtpath, xawtstr);
5811     if (::stat(libmawtpath, &statbuf) == 0) return false;
5812 
5813     // check libawt_xawt.so
5814     strcpy(libmawtpath, buf);
5815     strcat(libmawtpath, new_xawtstr);
5816     if (::stat(libmawtpath, &statbuf) == 0) return false;
5817 
5818     return true;
5819 }
--- EOF ---