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