1 /*
   2  * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_bsd.h"
  35 #include "memory/allocation.inline.hpp"
  36 #include "memory/filemap.hpp"
  37 #include "mutex_bsd.inline.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_share_bsd.hpp"
  40 #include "prims/jniFastGetField.hpp"
  41 #include "prims/jvm.h"
  42 #include "prims/jvm_misc.hpp"
  43 #include "runtime/arguments.hpp"
  44 #include "runtime/extendedPC.hpp"
  45 #include "runtime/globals.hpp"
  46 #include "runtime/interfaceSupport.hpp"
  47 #include "runtime/java.hpp"
  48 #include "runtime/javaCalls.hpp"
  49 #include "runtime/mutexLocker.hpp"
  50 #include "runtime/objectMonitor.hpp"
  51 #include "runtime/osThread.hpp"
  52 #include "runtime/perfMemory.hpp"
  53 #include "runtime/sharedRuntime.hpp"
  54 #include "runtime/statSampler.hpp"
  55 #include "runtime/stubRoutines.hpp"
  56 #include "runtime/thread.inline.hpp"
  57 #include "runtime/threadCritical.hpp"
  58 #include "runtime/timer.hpp"
  59 #include "services/attachListener.hpp"
  60 #include "services/memTracker.hpp"
  61 #include "services/runtimeService.hpp"
  62 #include "utilities/decoder.hpp"
  63 #include "utilities/defaultStream.hpp"
  64 #include "utilities/events.hpp"
  65 #include "utilities/growableArray.hpp"
  66 #include "utilities/vmError.hpp"
  67 
  68 // put OS-includes here
  69 # include <sys/types.h>
  70 # include <sys/mman.h>
  71 # include <sys/stat.h>
  72 # include <sys/select.h>
  73 # include <pthread.h>
  74 # include <signal.h>
  75 # include <errno.h>
  76 # include <dlfcn.h>
  77 # include <stdio.h>
  78 # include <unistd.h>
  79 # include <sys/resource.h>
  80 # include <pthread.h>
  81 # include <sys/stat.h>
  82 # include <sys/time.h>
  83 # include <sys/times.h>
  84 # include <sys/utsname.h>
  85 # include <sys/socket.h>
  86 # include <sys/wait.h>
  87 # include <time.h>
  88 # include <pwd.h>
  89 # include <poll.h>
  90 # include <semaphore.h>
  91 # include <fcntl.h>
  92 # include <string.h>
  93 # include <sys/param.h>
  94 # include <sys/sysctl.h>
  95 # include <sys/ipc.h>
  96 # include <sys/shm.h>
  97 #ifndef __APPLE__
  98 # include <link.h>
  99 #endif
 100 # include <stdint.h>
 101 # include <inttypes.h>
 102 # include <sys/ioctl.h>
 103 # include <sys/syscall.h>
 104 
 105 #if defined(__FreeBSD__) || defined(__NetBSD__)
 106 # include <elf.h>
 107 #endif
 108 
 109 #ifdef __APPLE__
 110 # include <mach/mach.h> // semaphore_* API
 111 # include <mach-o/dyld.h>
 112 # include <sys/proc_info.h>
 113 # include <objc/objc-auto.h>
 114 #endif
 115 
 116 #ifndef MAP_ANONYMOUS
 117 #define MAP_ANONYMOUS MAP_ANON
 118 #endif
 119 
 120 #define MAX_PATH    (2 * K)
 121 
 122 // for timer info max values which include all bits
 123 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 124 
 125 #define LARGEPAGES_BIT (1 << 6)
 126 ////////////////////////////////////////////////////////////////////////////////
 127 // global variables
 128 julong os::Bsd::_physical_memory = 0;
 129 
 130 
 131 int (*os::Bsd::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 132 pthread_t os::Bsd::_main_thread;
 133 int os::Bsd::_page_size = -1;
 134 
 135 static jlong initial_time_count=0;
 136 
 137 static int clock_tics_per_sec = 100;
 138 
 139 // For diagnostics to print a message once. see run_periodic_checks
 140 static sigset_t check_signal_done;
 141 static bool check_signals = true;
 142 
 143 static pid_t _initial_pid = 0;
 144 
 145 /* Signal number used to suspend/resume a thread */
 146 
 147 /* do not use any signal number less than SIGSEGV, see 4355769 */
 148 static int SR_signum = SIGUSR2;
 149 sigset_t SR_sigset;
 150 
 151 
 152 ////////////////////////////////////////////////////////////////////////////////
 153 // utility functions
 154 
 155 static int SR_initialize();
 156 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 157 
 158 julong os::available_memory() {
 159   return Bsd::available_memory();
 160 }
 161 
 162 // available here means free
 163 julong os::Bsd::available_memory() {
 164   uint64_t available = physical_memory() >> 2;
 165 #ifdef __APPLE__
 166   mach_msg_type_number_t count = HOST_VM_INFO64_COUNT;
 167   vm_statistics64_data_t vmstat;
 168   kern_return_t kerr = host_statistics64(mach_host_self(), HOST_VM_INFO64,
 169                                          (host_info64_t)&vmstat, &count);
 170   assert(kerr == KERN_SUCCESS,
 171          "host_statistics64 failed - check mach_host_self() and count");
 172   if (kerr == KERN_SUCCESS) {
 173     available = vmstat.free_count * os::vm_page_size();
 174   }
 175 #endif
 176   return available;
 177 }
 178 
 179 julong os::physical_memory() {
 180   return Bsd::physical_memory();
 181 }
 182 
 183 ////////////////////////////////////////////////////////////////////////////////
 184 // environment support
 185 
 186 bool os::getenv(const char* name, char* buf, int len) {
 187   const char* val = ::getenv(name);
 188   if (val != NULL && strlen(val) < (size_t)len) {
 189     strcpy(buf, val);
 190     return true;
 191   }
 192   if (len > 0) buf[0] = 0;  // return a null string
 193   return false;
 194 }
 195 
 196 
 197 // Return true if user is running as root.
 198 
 199 bool os::have_special_privileges() {
 200   static bool init = false;
 201   static bool privileges = false;
 202   if (!init) {
 203     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 204     init = true;
 205   }
 206   return privileges;
 207 }
 208 
 209 
 210 
 211 // Cpu architecture string
 212 #if   defined(ZERO)
 213 static char cpu_arch[] = ZERO_LIBARCH;
 214 #elif defined(IA64)
 215 static char cpu_arch[] = "ia64";
 216 #elif defined(IA32)
 217 static char cpu_arch[] = "i386";
 218 #elif defined(AMD64)
 219 static char cpu_arch[] = "amd64";
 220 #elif defined(ARM)
 221 static char cpu_arch[] = "arm";
 222 #elif defined(PPC)
 223 static char cpu_arch[] = "ppc";
 224 #elif defined(SPARC)
 225 #  ifdef _LP64
 226 static char cpu_arch[] = "sparcv9";
 227 #  else
 228 static char cpu_arch[] = "sparc";
 229 #  endif
 230 #else
 231 #error Add appropriate cpu_arch setting
 232 #endif
 233 
 234 // Compiler variant
 235 #ifdef COMPILER2
 236 #define COMPILER_VARIANT "server"
 237 #else
 238 #define COMPILER_VARIANT "client"
 239 #endif
 240 
 241 
 242 void os::Bsd::initialize_system_info() {
 243   int mib[2];
 244   size_t len;
 245   int cpu_val;
 246   julong mem_val;
 247 
 248   /* get processors count via hw.ncpus sysctl */
 249   mib[0] = CTL_HW;
 250   mib[1] = HW_NCPU;
 251   len = sizeof(cpu_val);
 252   if (sysctl(mib, 2, &cpu_val, &len, NULL, 0) != -1 && cpu_val >= 1) {
 253        assert(len == sizeof(cpu_val), "unexpected data size");
 254        set_processor_count(cpu_val);
 255   }
 256   else {
 257        set_processor_count(1);   // fallback
 258   }
 259 
 260   /* get physical memory via hw.memsize sysctl (hw.memsize is used
 261    * since it returns a 64 bit value)
 262    */
 263   mib[0] = CTL_HW;
 264 
 265 #if defined (HW_MEMSIZE) // Apple
 266   mib[1] = HW_MEMSIZE;
 267 #elif defined(HW_PHYSMEM) // Most of BSD
 268   mib[1] = HW_PHYSMEM;
 269 #elif defined(HW_REALMEM) // Old FreeBSD
 270   mib[1] = HW_REALMEM;
 271 #else
 272   #error No ways to get physmem
 273 #endif
 274 
 275   len = sizeof(mem_val);
 276   if (sysctl(mib, 2, &mem_val, &len, NULL, 0) != -1) {
 277        assert(len == sizeof(mem_val), "unexpected data size");
 278        _physical_memory = mem_val;
 279   } else {
 280        _physical_memory = 256*1024*1024;       // fallback (XXXBSD?)
 281   }
 282 
 283 #ifdef __OpenBSD__
 284   {
 285        // limit _physical_memory memory view on OpenBSD since
 286        // datasize rlimit restricts us anyway.
 287        struct rlimit limits;
 288        getrlimit(RLIMIT_DATA, &limits);
 289        _physical_memory = MIN2(_physical_memory, (julong)limits.rlim_cur);
 290   }
 291 #endif
 292 }
 293 
 294 #ifdef __APPLE__
 295 static const char *get_home() {
 296   const char *home_dir = ::getenv("HOME");
 297   if ((home_dir == NULL) || (*home_dir == '\0')) {
 298     struct passwd *passwd_info = getpwuid(geteuid());
 299     if (passwd_info != NULL) {
 300       home_dir = passwd_info->pw_dir;
 301     }
 302   }
 303 
 304   return home_dir;
 305 }
 306 #endif
 307 
 308 void os::init_system_properties_values() {
 309 //  char arch[12];
 310 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 311 
 312   // The next steps are taken in the product version:
 313   //
 314   // Obtain the JAVA_HOME value from the location of libjvm.so.
 315   // This library should be located at:
 316   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 317   //
 318   // If "/jre/lib/" appears at the right place in the path, then we
 319   // assume libjvm.so is installed in a JDK and we use this path.
 320   //
 321   // Otherwise exit with message: "Could not create the Java virtual machine."
 322   //
 323   // The following extra steps are taken in the debugging version:
 324   //
 325   // If "/jre/lib/" does NOT appear at the right place in the path
 326   // instead of exit check for $JAVA_HOME environment variable.
 327   //
 328   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 329   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 330   // it looks like libjvm.so is installed there
 331   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 332   //
 333   // Otherwise exit.
 334   //
 335   // Important note: if the location of libjvm.so changes this
 336   // code needs to be changed accordingly.
 337 
 338   // The next few definitions allow the code to be verbatim:
 339 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
 340 #define getenv(n) ::getenv(n)
 341 
 342 /*
 343  * See ld(1):
 344  *      The linker uses the following search paths to locate required
 345  *      shared libraries:
 346  *        1: ...
 347  *        ...
 348  *        7: The default directories, normally /lib and /usr/lib.
 349  */
 350 #ifndef DEFAULT_LIBPATH
 351 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 352 #endif
 353 
 354 #define EXTENSIONS_DIR  "/lib/ext"
 355 #define ENDORSED_DIR    "/lib/endorsed"
 356 #define REG_DIR         "/usr/java/packages"
 357 
 358 #ifdef __APPLE__
 359 #define SYS_EXTENSIONS_DIR   "/Library/Java/Extensions"
 360 #define SYS_EXTENSIONS_DIRS  SYS_EXTENSIONS_DIR ":/Network" SYS_EXTENSIONS_DIR ":/System" SYS_EXTENSIONS_DIR ":/usr/lib/java"
 361         const char *user_home_dir = get_home();
 362         // the null in SYS_EXTENSIONS_DIRS counts for the size of the colon after user_home_dir
 363         int system_ext_size = strlen(user_home_dir) + sizeof(SYS_EXTENSIONS_DIR) +
 364             sizeof(SYS_EXTENSIONS_DIRS);
 365 #endif
 366 
 367   {
 368     /* sysclasspath, java_home, dll_dir */
 369     {
 370         char *home_path;
 371         char *dll_path;
 372         char *pslash;
 373         char buf[MAXPATHLEN];
 374         os::jvm_path(buf, sizeof(buf));
 375 
 376         // Found the full path to libjvm.so.
 377         // Now cut the path to <java_home>/jre if we can.
 378         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 379         pslash = strrchr(buf, '/');
 380         if (pslash != NULL)
 381             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 382         dll_path = malloc(strlen(buf) + 1);
 383         if (dll_path == NULL)
 384             return;
 385         strcpy(dll_path, buf);
 386         Arguments::set_dll_dir(dll_path);
 387 
 388         if (pslash != NULL) {
 389             pslash = strrchr(buf, '/');
 390             if (pslash != NULL) {
 391                 *pslash = '\0';       /* get rid of /<arch> (/lib on macosx) */
 392 #ifndef __APPLE__
 393                 pslash = strrchr(buf, '/');
 394                 if (pslash != NULL)
 395                     *pslash = '\0';   /* get rid of /lib */
 396 #endif
 397             }
 398         }
 399 
 400         home_path = malloc(strlen(buf) + 1);
 401         if (home_path == NULL)
 402             return;
 403         strcpy(home_path, buf);
 404         Arguments::set_java_home(home_path);
 405 
 406         if (!set_boot_path('/', ':'))
 407             return;
 408     }
 409 
 410     /*
 411      * Where to look for native libraries
 412      *
 413      * Note: Due to a legacy implementation, most of the library path
 414      * is set in the launcher.  This was to accomodate linking restrictions
 415      * on legacy Bsd implementations (which are no longer supported).
 416      * Eventually, all the library path setting will be done here.
 417      *
 418      * However, to prevent the proliferation of improperly built native
 419      * libraries, the new path component /usr/java/packages is added here.
 420      * Eventually, all the library path setting will be done here.
 421      */
 422     {
 423         char *ld_library_path;
 424 
 425         /*
 426          * Construct the invariant part of ld_library_path. Note that the
 427          * space for the colon and the trailing null are provided by the
 428          * nulls included by the sizeof operator (so actually we allocate
 429          * a byte more than necessary).
 430          */
 431 #ifdef __APPLE__
 432         ld_library_path = (char *) malloc(system_ext_size);
 433         sprintf(ld_library_path, "%s" SYS_EXTENSIONS_DIR ":" SYS_EXTENSIONS_DIRS, user_home_dir);
 434 #else
 435         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 436             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 437         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 438 #endif
 439 
 440         /*
 441          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 442          * should always exist (until the legacy problem cited above is
 443          * addressed).
 444          */
 445 #ifdef __APPLE__
 446         // Prepend the default path with the JAVA_LIBRARY_PATH so that the app launcher code can specify a directory inside an app wrapper
 447         char *l = getenv("JAVA_LIBRARY_PATH");
 448         if (l != NULL) {
 449             char *t = ld_library_path;
 450             /* That's +1 for the colon and +1 for the trailing '\0' */
 451             ld_library_path = (char *) malloc(strlen(l) + 1 + strlen(t) + 1);
 452             sprintf(ld_library_path, "%s:%s", l, t);
 453             free(t);
 454         }
 455 
 456         char *v = getenv("DYLD_LIBRARY_PATH");
 457 #else
 458         char *v = getenv("LD_LIBRARY_PATH");
 459 #endif
 460         if (v != NULL) {
 461             char *t = ld_library_path;
 462             /* That's +1 for the colon and +1 for the trailing '\0' */
 463             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 464             sprintf(ld_library_path, "%s:%s", v, t);
 465             free(t);
 466         }
 467 
 468 #ifdef __APPLE__
 469         // Apple's Java6 has "." at the beginning of java.library.path.
 470         // OpenJDK on Windows has "." at the end of java.library.path.
 471         // OpenJDK on Linux and Solaris don't have "." in java.library.path
 472         // at all. To ease the transition from Apple's Java6 to OpenJDK7,
 473         // "." is appended to the end of java.library.path. Yes, this
 474         // could cause a change in behavior, but Apple's Java6 behavior
 475         // can be achieved by putting "." at the beginning of the
 476         // JAVA_LIBRARY_PATH environment variable.
 477         {
 478             char *t = ld_library_path;
 479             // that's +3 for appending ":." and the trailing '\0'
 480             ld_library_path = (char *) malloc(strlen(t) + 3);
 481             sprintf(ld_library_path, "%s:%s", t, ".");
 482             free(t);
 483         }
 484 #endif
 485 
 486         Arguments::set_library_path(ld_library_path);
 487     }
 488 
 489     /*
 490      * Extensions directories.
 491      *
 492      * Note that the space for the colon and the trailing null are provided
 493      * by the nulls included by the sizeof operator (so actually one byte more
 494      * than necessary is allocated).
 495      */
 496     {
 497 #ifdef __APPLE__
 498         char *buf = malloc(strlen(Arguments::get_java_home()) +
 499             sizeof(EXTENSIONS_DIR) + system_ext_size);
 500         sprintf(buf, "%s" SYS_EXTENSIONS_DIR ":%s" EXTENSIONS_DIR ":"
 501             SYS_EXTENSIONS_DIRS, user_home_dir, Arguments::get_java_home());
 502 #else
 503         char *buf = malloc(strlen(Arguments::get_java_home()) +
 504             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 505         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 506             Arguments::get_java_home());
 507 #endif
 508 
 509         Arguments::set_ext_dirs(buf);
 510     }
 511 
 512     /* Endorsed standards default directory. */
 513     {
 514         char * buf;
 515         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 516         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 517         Arguments::set_endorsed_dirs(buf);
 518     }
 519   }
 520 
 521 #ifdef __APPLE__
 522 #undef SYS_EXTENSIONS_DIR
 523 #endif
 524 #undef malloc
 525 #undef getenv
 526 #undef EXTENSIONS_DIR
 527 #undef ENDORSED_DIR
 528 
 529   // Done
 530   return;
 531 }
 532 
 533 ////////////////////////////////////////////////////////////////////////////////
 534 // breakpoint support
 535 
 536 void os::breakpoint() {
 537   BREAKPOINT;
 538 }
 539 
 540 extern "C" void breakpoint() {
 541   // use debugger to set breakpoint here
 542 }
 543 
 544 ////////////////////////////////////////////////////////////////////////////////
 545 // signal support
 546 
 547 debug_only(static bool signal_sets_initialized = false);
 548 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 549 
 550 bool os::Bsd::is_sig_ignored(int sig) {
 551       struct sigaction oact;
 552       sigaction(sig, (struct sigaction*)NULL, &oact);
 553       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 554                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 555       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 556            return true;
 557       else
 558            return false;
 559 }
 560 
 561 void os::Bsd::signal_sets_init() {
 562   // Should also have an assertion stating we are still single-threaded.
 563   assert(!signal_sets_initialized, "Already initialized");
 564   // Fill in signals that are necessarily unblocked for all threads in
 565   // the VM. Currently, we unblock the following signals:
 566   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 567   //                         by -Xrs (=ReduceSignalUsage));
 568   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 569   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 570   // the dispositions or masks wrt these signals.
 571   // Programs embedding the VM that want to use the above signals for their
 572   // own purposes must, at this time, use the "-Xrs" option to prevent
 573   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 574   // (See bug 4345157, and other related bugs).
 575   // In reality, though, unblocking these signals is really a nop, since
 576   // these signals are not blocked by default.
 577   sigemptyset(&unblocked_sigs);
 578   sigemptyset(&allowdebug_blocked_sigs);
 579   sigaddset(&unblocked_sigs, SIGILL);
 580   sigaddset(&unblocked_sigs, SIGSEGV);
 581   sigaddset(&unblocked_sigs, SIGBUS);
 582   sigaddset(&unblocked_sigs, SIGFPE);
 583   sigaddset(&unblocked_sigs, SR_signum);
 584 
 585   if (!ReduceSignalUsage) {
 586    if (!os::Bsd::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 587       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 588       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 589    }
 590    if (!os::Bsd::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 591       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 592       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 593    }
 594    if (!os::Bsd::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 595       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 596       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 597    }
 598   }
 599   // Fill in signals that are blocked by all but the VM thread.
 600   sigemptyset(&vm_sigs);
 601   if (!ReduceSignalUsage)
 602     sigaddset(&vm_sigs, BREAK_SIGNAL);
 603   debug_only(signal_sets_initialized = true);
 604 
 605 }
 606 
 607 // These are signals that are unblocked while a thread is running Java.
 608 // (For some reason, they get blocked by default.)
 609 sigset_t* os::Bsd::unblocked_signals() {
 610   assert(signal_sets_initialized, "Not initialized");
 611   return &unblocked_sigs;
 612 }
 613 
 614 // These are the signals that are blocked while a (non-VM) thread is
 615 // running Java. Only the VM thread handles these signals.
 616 sigset_t* os::Bsd::vm_signals() {
 617   assert(signal_sets_initialized, "Not initialized");
 618   return &vm_sigs;
 619 }
 620 
 621 // These are signals that are blocked during cond_wait to allow debugger in
 622 sigset_t* os::Bsd::allowdebug_blocked_signals() {
 623   assert(signal_sets_initialized, "Not initialized");
 624   return &allowdebug_blocked_sigs;
 625 }
 626 
 627 void os::Bsd::hotspot_sigmask(Thread* thread) {
 628 
 629   //Save caller's signal mask before setting VM signal mask
 630   sigset_t caller_sigmask;
 631   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 632 
 633   OSThread* osthread = thread->osthread();
 634   osthread->set_caller_sigmask(caller_sigmask);
 635 
 636   pthread_sigmask(SIG_UNBLOCK, os::Bsd::unblocked_signals(), NULL);
 637 
 638   if (!ReduceSignalUsage) {
 639     if (thread->is_VM_thread()) {
 640       // Only the VM thread handles BREAK_SIGNAL ...
 641       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 642     } else {
 643       // ... all other threads block BREAK_SIGNAL
 644       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 645     }
 646   }
 647 }
 648 
 649 
 650 //////////////////////////////////////////////////////////////////////////////
 651 // create new thread
 652 
 653 // check if it's safe to start a new thread
 654 static bool _thread_safety_check(Thread* thread) {
 655   return true;
 656 }
 657 
 658 #ifdef __APPLE__
 659 // library handle for calling objc_registerThreadWithCollector()
 660 // without static linking to the libobjc library
 661 #define OBJC_LIB "/usr/lib/libobjc.dylib"
 662 #define OBJC_GCREGISTER "objc_registerThreadWithCollector"
 663 typedef void (*objc_registerThreadWithCollector_t)();
 664 extern "C" objc_registerThreadWithCollector_t objc_registerThreadWithCollectorFunction;
 665 objc_registerThreadWithCollector_t objc_registerThreadWithCollectorFunction = NULL;
 666 #endif
 667 
 668 #ifdef __APPLE__
 669 static uint64_t locate_unique_thread_id(mach_port_t mach_thread_port) {
 670   // Additional thread_id used to correlate threads in SA
 671   thread_identifier_info_data_t     m_ident_info;
 672   mach_msg_type_number_t            count = THREAD_IDENTIFIER_INFO_COUNT;
 673 
 674   thread_info(mach_thread_port, THREAD_IDENTIFIER_INFO,
 675               (thread_info_t) &m_ident_info, &count);
 676 
 677   return m_ident_info.thread_id;
 678 }
 679 #endif
 680 
 681 // Thread start routine for all newly created threads
 682 static void *java_start(Thread *thread) {
 683   // Try to randomize the cache line index of hot stack frames.
 684   // This helps when threads of the same stack traces evict each other's
 685   // cache lines. The threads can be either from the same JVM instance, or
 686   // from different JVM instances. The benefit is especially true for
 687   // processors with hyperthreading technology.
 688   static int counter = 0;
 689   int pid = os::current_process_id();
 690   alloca(((pid ^ counter++) & 7) * 128);
 691 
 692   ThreadLocalStorage::set_thread(thread);
 693 
 694   OSThread* osthread = thread->osthread();
 695   Monitor* sync = osthread->startThread_lock();
 696 
 697   // non floating stack BsdThreads needs extra check, see above
 698   if (!_thread_safety_check(thread)) {
 699     // notify parent thread
 700     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 701     osthread->set_state(ZOMBIE);
 702     sync->notify_all();
 703     return NULL;
 704   }
 705 
 706   osthread->set_thread_id(os::Bsd::gettid());
 707 
 708 #ifdef __APPLE__
 709   uint64_t unique_thread_id = locate_unique_thread_id(osthread->thread_id());
 710   guarantee(unique_thread_id != 0, "unique thread id was not found");
 711   osthread->set_unique_thread_id(unique_thread_id);
 712 #endif
 713   // initialize signal mask for this thread
 714   os::Bsd::hotspot_sigmask(thread);
 715 
 716   // initialize floating point control register
 717   os::Bsd::init_thread_fpu_state();
 718 
 719 #ifdef __APPLE__
 720   // register thread with objc gc
 721   if (objc_registerThreadWithCollectorFunction != NULL) {
 722     objc_registerThreadWithCollectorFunction();
 723   }
 724 #endif
 725 
 726   // handshaking with parent thread
 727   {
 728     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 729 
 730     // notify parent thread
 731     osthread->set_state(INITIALIZED);
 732     sync->notify_all();
 733 
 734     // wait until os::start_thread()
 735     while (osthread->get_state() == INITIALIZED) {
 736       sync->wait(Mutex::_no_safepoint_check_flag);
 737     }
 738   }
 739 
 740   // call one more level start routine
 741   thread->run();
 742 
 743   return 0;
 744 }
 745 
 746 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 747   assert(thread->osthread() == NULL, "caller responsible");
 748 
 749   // Allocate the OSThread object
 750   OSThread* osthread = new OSThread(NULL, NULL);
 751   if (osthread == NULL) {
 752     return false;
 753   }
 754 
 755   // set the correct thread state
 756   osthread->set_thread_type(thr_type);
 757 
 758   // Initial state is ALLOCATED but not INITIALIZED
 759   osthread->set_state(ALLOCATED);
 760 
 761   thread->set_osthread(osthread);
 762 
 763   // init thread attributes
 764   pthread_attr_t attr;
 765   pthread_attr_init(&attr);
 766   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 767 
 768   // stack size
 769   if (os::Bsd::supports_variable_stack_size()) {
 770     // calculate stack size if it's not specified by caller
 771     if (stack_size == 0) {
 772       stack_size = os::Bsd::default_stack_size(thr_type);
 773 
 774       switch (thr_type) {
 775       case os::java_thread:
 776         // Java threads use ThreadStackSize which default value can be
 777         // changed with the flag -Xss
 778         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 779         stack_size = JavaThread::stack_size_at_create();
 780         break;
 781       case os::compiler_thread:
 782         if (CompilerThreadStackSize > 0) {
 783           stack_size = (size_t)(CompilerThreadStackSize * K);
 784           break;
 785         } // else fall through:
 786           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 787       case os::vm_thread:
 788       case os::pgc_thread:
 789       case os::cgc_thread:
 790       case os::watcher_thread:
 791         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 792         break;
 793       }
 794     }
 795 
 796     stack_size = MAX2(stack_size, os::Bsd::min_stack_allowed);
 797     pthread_attr_setstacksize(&attr, stack_size);
 798   } else {
 799     // let pthread_create() pick the default value.
 800   }
 801 
 802   ThreadState state;
 803 
 804   {
 805     pthread_t tid;
 806     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 807 
 808     pthread_attr_destroy(&attr);
 809 
 810     if (ret != 0) {
 811       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 812         perror("pthread_create()");
 813       }
 814       // Need to clean up stuff we've allocated so far
 815       thread->set_osthread(NULL);
 816       delete osthread;
 817       return false;
 818     }
 819 
 820     // Store pthread info into the OSThread
 821     osthread->set_pthread_id(tid);
 822 
 823     // Wait until child thread is either initialized or aborted
 824     {
 825       Monitor* sync_with_child = osthread->startThread_lock();
 826       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 827       while ((state = osthread->get_state()) == ALLOCATED) {
 828         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 829       }
 830     }
 831 
 832   }
 833 
 834   // Aborted due to thread limit being reached
 835   if (state == ZOMBIE) {
 836       thread->set_osthread(NULL);
 837       delete osthread;
 838       return false;
 839   }
 840 
 841   // The thread is returned suspended (in state INITIALIZED),
 842   // and is started higher up in the call chain
 843   assert(state == INITIALIZED, "race condition");
 844   return true;
 845 }
 846 
 847 /////////////////////////////////////////////////////////////////////////////
 848 // attach existing thread
 849 
 850 // bootstrap the main thread
 851 bool os::create_main_thread(JavaThread* thread) {
 852   assert(os::Bsd::_main_thread == pthread_self(), "should be called inside main thread");
 853   return create_attached_thread(thread);
 854 }
 855 
 856 bool os::create_attached_thread(JavaThread* thread) {
 857 #ifdef ASSERT
 858     thread->verify_not_published();
 859 #endif
 860 
 861   // Allocate the OSThread object
 862   OSThread* osthread = new OSThread(NULL, NULL);
 863 
 864   if (osthread == NULL) {
 865     return false;
 866   }
 867 
 868   osthread->set_thread_id(os::Bsd::gettid());
 869 
 870   // Store pthread info into the OSThread
 871 #ifdef __APPLE__
 872   uint64_t unique_thread_id = locate_unique_thread_id(osthread->thread_id());
 873   guarantee(unique_thread_id != 0, "just checking");
 874   osthread->set_unique_thread_id(unique_thread_id);
 875 #endif
 876   osthread->set_pthread_id(::pthread_self());
 877 
 878   // initialize floating point control register
 879   os::Bsd::init_thread_fpu_state();
 880 
 881   // Initial thread state is RUNNABLE
 882   osthread->set_state(RUNNABLE);
 883 
 884   thread->set_osthread(osthread);
 885 
 886   // initialize signal mask for this thread
 887   // and save the caller's signal mask
 888   os::Bsd::hotspot_sigmask(thread);
 889 
 890   return true;
 891 }
 892 
 893 void os::pd_start_thread(Thread* thread) {
 894   OSThread * osthread = thread->osthread();
 895   assert(osthread->get_state() != INITIALIZED, "just checking");
 896   Monitor* sync_with_child = osthread->startThread_lock();
 897   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 898   sync_with_child->notify();
 899 }
 900 
 901 // Free Bsd resources related to the OSThread
 902 void os::free_thread(OSThread* osthread) {
 903   assert(osthread != NULL, "osthread not set");
 904 
 905   if (Thread::current()->osthread() == osthread) {
 906     // Restore caller's signal mask
 907     sigset_t sigmask = osthread->caller_sigmask();
 908     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 909    }
 910 
 911   delete osthread;
 912 }
 913 
 914 //////////////////////////////////////////////////////////////////////////////
 915 // thread local storage
 916 
 917 int os::allocate_thread_local_storage() {
 918   pthread_key_t key;
 919   int rslt = pthread_key_create(&key, NULL);
 920   assert(rslt == 0, "cannot allocate thread local storage");
 921   return (int)key;
 922 }
 923 
 924 // Note: This is currently not used by VM, as we don't destroy TLS key
 925 // on VM exit.
 926 void os::free_thread_local_storage(int index) {
 927   int rslt = pthread_key_delete((pthread_key_t)index);
 928   assert(rslt == 0, "invalid index");
 929 }
 930 
 931 void os::thread_local_storage_at_put(int index, void* value) {
 932   int rslt = pthread_setspecific((pthread_key_t)index, value);
 933   assert(rslt == 0, "pthread_setspecific failed");
 934 }
 935 
 936 extern "C" Thread* get_thread() {
 937   return ThreadLocalStorage::thread();
 938 }
 939 
 940 
 941 ////////////////////////////////////////////////////////////////////////////////
 942 // time support
 943 
 944 // Time since start-up in seconds to a fine granularity.
 945 // Used by VMSelfDestructTimer and the MemProfiler.
 946 double os::elapsedTime() {
 947 
 948   return ((double)os::elapsed_counter()) / os::elapsed_frequency();
 949 }
 950 
 951 jlong os::elapsed_counter() {
 952   return javaTimeNanos() - initial_time_count;
 953 }
 954 
 955 jlong os::elapsed_frequency() {
 956   return NANOSECS_PER_SEC; // nanosecond resolution
 957 }
 958 
 959 bool os::supports_vtime() { return true; }
 960 bool os::enable_vtime()   { return false; }
 961 bool os::vtime_enabled()  { return false; }
 962 
 963 double os::elapsedVTime() {
 964   // better than nothing, but not much
 965   return elapsedTime();
 966 }
 967 
 968 jlong os::javaTimeMillis() {
 969   timeval time;
 970   int status = gettimeofday(&time, NULL);
 971   assert(status != -1, "bsd error");
 972   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
 973 }
 974 
 975 #ifndef CLOCK_MONOTONIC
 976 #define CLOCK_MONOTONIC (1)
 977 #endif
 978 
 979 #ifdef __APPLE__
 980 void os::Bsd::clock_init() {
 981         // XXXDARWIN: Investigate replacement monotonic clock
 982 }
 983 #else
 984 void os::Bsd::clock_init() {
 985   struct timespec res;
 986   struct timespec tp;
 987   if (::clock_getres(CLOCK_MONOTONIC, &res) == 0 &&
 988       ::clock_gettime(CLOCK_MONOTONIC, &tp)  == 0) {
 989     // yes, monotonic clock is supported
 990     _clock_gettime = ::clock_gettime;
 991   }
 992 }
 993 #endif
 994 
 995 
 996 jlong os::javaTimeNanos() {
 997   if (Bsd::supports_monotonic_clock()) {
 998     struct timespec tp;
 999     int status = Bsd::clock_gettime(CLOCK_MONOTONIC, &tp);
1000     assert(status == 0, "gettime error");
1001     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1002     return result;
1003   } else {
1004     timeval time;
1005     int status = gettimeofday(&time, NULL);
1006     assert(status != -1, "bsd error");
1007     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1008     return 1000 * usecs;
1009   }
1010 }
1011 
1012 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1013   if (Bsd::supports_monotonic_clock()) {
1014     info_ptr->max_value = ALL_64_BITS;
1015 
1016     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1017     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1018     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1019   } else {
1020     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1021     info_ptr->max_value = ALL_64_BITS;
1022 
1023     // gettimeofday is a real time clock so it skips
1024     info_ptr->may_skip_backward = true;
1025     info_ptr->may_skip_forward = true;
1026   }
1027 
1028   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1029 }
1030 
1031 // Return the real, user, and system times in seconds from an
1032 // arbitrary fixed point in the past.
1033 bool os::getTimesSecs(double* process_real_time,
1034                       double* process_user_time,
1035                       double* process_system_time) {
1036   struct tms ticks;
1037   clock_t real_ticks = times(&ticks);
1038 
1039   if (real_ticks == (clock_t) (-1)) {
1040     return false;
1041   } else {
1042     double ticks_per_second = (double) clock_tics_per_sec;
1043     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1044     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1045     *process_real_time = ((double) real_ticks) / ticks_per_second;
1046 
1047     return true;
1048   }
1049 }
1050 
1051 
1052 char * os::local_time_string(char *buf, size_t buflen) {
1053   struct tm t;
1054   time_t long_time;
1055   time(&long_time);
1056   localtime_r(&long_time, &t);
1057   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1058                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1059                t.tm_hour, t.tm_min, t.tm_sec);
1060   return buf;
1061 }
1062 
1063 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1064   return localtime_r(clock, res);
1065 }
1066 
1067 ////////////////////////////////////////////////////////////////////////////////
1068 // runtime exit support
1069 
1070 // Note: os::shutdown() might be called very early during initialization, or
1071 // called from signal handler. Before adding something to os::shutdown(), make
1072 // sure it is async-safe and can handle partially initialized VM.
1073 void os::shutdown() {
1074 
1075   // allow PerfMemory to attempt cleanup of any persistent resources
1076   perfMemory_exit();
1077 
1078   // needs to remove object in file system
1079   AttachListener::abort();
1080 
1081   // flush buffered output, finish log files
1082   ostream_abort();
1083 
1084   // Check for abort hook
1085   abort_hook_t abort_hook = Arguments::abort_hook();
1086   if (abort_hook != NULL) {
1087     abort_hook();
1088   }
1089 
1090 }
1091 
1092 // Note: os::abort() might be called very early during initialization, or
1093 // called from signal handler. Before adding something to os::abort(), make
1094 // sure it is async-safe and can handle partially initialized VM.
1095 void os::abort(bool dump_core) {
1096   os::shutdown();
1097   if (dump_core) {
1098 #ifndef PRODUCT
1099     fdStream out(defaultStream::output_fd());
1100     out.print_raw("Current thread is ");
1101     char buf[16];
1102     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1103     out.print_raw_cr(buf);
1104     out.print_raw_cr("Dumping core ...");
1105 #endif
1106     ::abort(); // dump core
1107   }
1108 
1109   ::exit(1);
1110 }
1111 
1112 // Die immediately, no exit hook, no abort hook, no cleanup.
1113 void os::die() {
1114   // _exit() on BsdThreads only kills current thread
1115   ::abort();
1116 }
1117 
1118 // unused on bsd for now.
1119 void os::set_error_file(const char *logfile) {}
1120 
1121 
1122 // This method is a copy of JDK's sysGetLastErrorString
1123 // from src/solaris/hpi/src/system_md.c
1124 
1125 size_t os::lasterror(char *buf, size_t len) {
1126 
1127   if (errno == 0)  return 0;
1128 
1129   const char *s = ::strerror(errno);
1130   size_t n = ::strlen(s);
1131   if (n >= len) {
1132     n = len - 1;
1133   }
1134   ::strncpy(buf, s, n);
1135   buf[n] = '\0';
1136   return n;
1137 }
1138 
1139 // Information of current thread in variety of formats
1140 pid_t os::Bsd::gettid() {
1141   int retval = -1;
1142 
1143 #ifdef __APPLE__ //XNU kernel
1144   // despite the fact mach port is actually not a thread id use it
1145   // instead of syscall(SYS_thread_selfid) as it certainly fits to u4
1146   retval = ::pthread_mach_thread_np(::pthread_self());
1147   guarantee(retval != 0, "just checking");
1148   return retval;
1149 
1150 #elif __FreeBSD__
1151   retval = syscall(SYS_thr_self);
1152 #elif __OpenBSD__
1153   retval = syscall(SYS_getthrid);
1154 #elif __NetBSD__
1155   retval = (pid_t) syscall(SYS__lwp_self);
1156 #endif
1157 
1158   if (retval == -1) {
1159     return getpid();
1160   }
1161 }
1162 
1163 intx os::current_thread_id() {
1164 #ifdef __APPLE__
1165   return (intx)::pthread_mach_thread_np(::pthread_self());
1166 #else
1167   return (intx)::pthread_self();
1168 #endif
1169 }
1170 
1171 int os::current_process_id() {
1172 
1173   // Under the old bsd thread library, bsd gives each thread
1174   // its own process id. Because of this each thread will return
1175   // a different pid if this method were to return the result
1176   // of getpid(2). Bsd provides no api that returns the pid
1177   // of the launcher thread for the vm. This implementation
1178   // returns a unique pid, the pid of the launcher thread
1179   // that starts the vm 'process'.
1180 
1181   // Under the NPTL, getpid() returns the same pid as the
1182   // launcher thread rather than a unique pid per thread.
1183   // Use gettid() if you want the old pre NPTL behaviour.
1184 
1185   // if you are looking for the result of a call to getpid() that
1186   // returns a unique pid for the calling thread, then look at the
1187   // OSThread::thread_id() method in osThread_bsd.hpp file
1188 
1189   return (int)(_initial_pid ? _initial_pid : getpid());
1190 }
1191 
1192 // DLL functions
1193 
1194 #define JNI_LIB_PREFIX "lib"
1195 #ifdef __APPLE__
1196 #define JNI_LIB_SUFFIX ".dylib"
1197 #else
1198 #define JNI_LIB_SUFFIX ".so"
1199 #endif
1200 
1201 const char* os::dll_file_extension() { return JNI_LIB_SUFFIX; }
1202 
1203 // This must be hard coded because it's the system's temporary
1204 // directory not the java application's temp directory, ala java.io.tmpdir.
1205 #ifdef __APPLE__
1206 // macosx has a secure per-user temporary directory
1207 char temp_path_storage[PATH_MAX];
1208 const char* os::get_temp_directory() {
1209   static char *temp_path = NULL;
1210   if (temp_path == NULL) {
1211     int pathSize = confstr(_CS_DARWIN_USER_TEMP_DIR, temp_path_storage, PATH_MAX);
1212     if (pathSize == 0 || pathSize > PATH_MAX) {
1213       strlcpy(temp_path_storage, "/tmp/", sizeof(temp_path_storage));
1214     }
1215     temp_path = temp_path_storage;
1216   }
1217   return temp_path;
1218 }
1219 #else /* __APPLE__ */
1220 const char* os::get_temp_directory() { return "/tmp"; }
1221 #endif /* __APPLE__ */
1222 
1223 static bool file_exists(const char* filename) {
1224   struct stat statbuf;
1225   if (filename == NULL || strlen(filename) == 0) {
1226     return false;
1227   }
1228   return os::stat(filename, &statbuf) == 0;
1229 }
1230 
1231 bool os::dll_build_name(char* buffer, size_t buflen,
1232                         const char* pname, const char* fname) {
1233   bool retval = false;
1234   // Copied from libhpi
1235   const size_t pnamelen = pname ? strlen(pname) : 0;
1236 
1237   // Return error on buffer overflow.
1238   if (pnamelen + strlen(fname) + strlen(JNI_LIB_PREFIX) + strlen(JNI_LIB_SUFFIX) + 2 > buflen) {
1239     return retval;
1240   }
1241 
1242   if (pnamelen == 0) {
1243     snprintf(buffer, buflen, JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX, fname);
1244     retval = true;
1245   } else if (strchr(pname, *os::path_separator()) != NULL) {
1246     int n;
1247     char** pelements = split_path(pname, &n);
1248     if (pelements == NULL) {
1249       return false;
1250     }
1251     for (int i = 0 ; i < n ; i++) {
1252       // Really shouldn't be NULL, but check can't hurt
1253       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1254         continue; // skip the empty path values
1255       }
1256       snprintf(buffer, buflen, "%s/" JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX,
1257           pelements[i], fname);
1258       if (file_exists(buffer)) {
1259         retval = true;
1260         break;
1261       }
1262     }
1263     // release the storage
1264     for (int i = 0 ; i < n ; i++) {
1265       if (pelements[i] != NULL) {
1266         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1267       }
1268     }
1269     if (pelements != NULL) {
1270       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1271     }
1272   } else {
1273     snprintf(buffer, buflen, "%s/" JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX, pname, fname);
1274     retval = true;
1275   }
1276   return retval;
1277 }
1278 
1279 // check if addr is inside libjvm.so
1280 bool os::address_is_in_vm(address addr) {
1281   static address libjvm_base_addr;
1282   Dl_info dlinfo;
1283 
1284   if (libjvm_base_addr == NULL) {
1285     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1286       libjvm_base_addr = (address)dlinfo.dli_fbase;
1287     }
1288     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1289   }
1290 
1291   if (dladdr((void *)addr, &dlinfo) != 0) {
1292     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1293   }
1294 
1295   return false;
1296 }
1297 
1298 
1299 #define MACH_MAXSYMLEN 256
1300 
1301 bool os::dll_address_to_function_name(address addr, char *buf,
1302                                       int buflen, int *offset) {
1303   // buf is not optional, but offset is optional
1304   assert(buf != NULL, "sanity check");
1305 
1306   Dl_info dlinfo;
1307   char localbuf[MACH_MAXSYMLEN];
1308 
1309   if (dladdr((void*)addr, &dlinfo) != 0) {
1310     // see if we have a matching symbol
1311     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1312       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1313         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1314       }
1315       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1316       return true;
1317     }
1318     // no matching symbol so try for just file info
1319     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1320       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1321                           buf, buflen, offset, dlinfo.dli_fname)) {
1322          return true;
1323       }
1324     }
1325 
1326     // Handle non-dynamic manually:
1327     if (dlinfo.dli_fbase != NULL &&
1328         Decoder::decode(addr, localbuf, MACH_MAXSYMLEN, offset,
1329                         dlinfo.dli_fbase)) {
1330       if (!Decoder::demangle(localbuf, buf, buflen)) {
1331         jio_snprintf(buf, buflen, "%s", localbuf);
1332       }
1333       return true;
1334     }
1335   }
1336   buf[0] = '\0';
1337   if (offset != NULL) *offset = -1;
1338   return false;
1339 }
1340 
1341 // ported from solaris version
1342 bool os::dll_address_to_library_name(address addr, char* buf,
1343                                      int buflen, int* offset) {
1344   // buf is not optional, but offset is optional
1345   assert(buf != NULL, "sanity check");
1346 
1347   Dl_info dlinfo;
1348 
1349   if (dladdr((void*)addr, &dlinfo) != 0) {
1350     if (dlinfo.dli_fname != NULL) {
1351       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1352     }
1353     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1354       *offset = addr - (address)dlinfo.dli_fbase;
1355     }
1356     return true;
1357   }
1358 
1359   buf[0] = '\0';
1360   if (offset) *offset = -1;
1361   return false;
1362 }
1363 
1364 // Loads .dll/.so and
1365 // in case of error it checks if .dll/.so was built for the
1366 // same architecture as Hotspot is running on
1367 
1368 #ifdef __APPLE__
1369 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1370   void * result= ::dlopen(filename, RTLD_LAZY);
1371   if (result != NULL) {
1372     // Successful loading
1373     return result;
1374   }
1375 
1376   // Read system error message into ebuf
1377   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1378   ebuf[ebuflen-1]='\0';
1379 
1380   return NULL;
1381 }
1382 #else
1383 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1384 {
1385   void * result= ::dlopen(filename, RTLD_LAZY);
1386   if (result != NULL) {
1387     // Successful loading
1388     return result;
1389   }
1390 
1391   Elf32_Ehdr elf_head;
1392 
1393   // Read system error message into ebuf
1394   // It may or may not be overwritten below
1395   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1396   ebuf[ebuflen-1]='\0';
1397   int diag_msg_max_length=ebuflen-strlen(ebuf);
1398   char* diag_msg_buf=ebuf+strlen(ebuf);
1399 
1400   if (diag_msg_max_length==0) {
1401     // No more space in ebuf for additional diagnostics message
1402     return NULL;
1403   }
1404 
1405 
1406   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1407 
1408   if (file_descriptor < 0) {
1409     // Can't open library, report dlerror() message
1410     return NULL;
1411   }
1412 
1413   bool failed_to_read_elf_head=
1414     (sizeof(elf_head)!=
1415         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1416 
1417   ::close(file_descriptor);
1418   if (failed_to_read_elf_head) {
1419     // file i/o error - report dlerror() msg
1420     return NULL;
1421   }
1422 
1423   typedef struct {
1424     Elf32_Half  code;         // Actual value as defined in elf.h
1425     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1426     char        elf_class;    // 32 or 64 bit
1427     char        endianess;    // MSB or LSB
1428     char*       name;         // String representation
1429   } arch_t;
1430 
1431   #ifndef EM_486
1432   #define EM_486          6               /* Intel 80486 */
1433   #endif
1434 
1435   #ifndef EM_MIPS_RS3_LE
1436   #define EM_MIPS_RS3_LE  10              /* MIPS */
1437   #endif
1438 
1439   #ifndef EM_PPC64
1440   #define EM_PPC64        21              /* PowerPC64 */
1441   #endif
1442 
1443   #ifndef EM_S390
1444   #define EM_S390         22              /* IBM System/390 */
1445   #endif
1446 
1447   #ifndef EM_IA_64
1448   #define EM_IA_64        50              /* HP/Intel IA-64 */
1449   #endif
1450 
1451   #ifndef EM_X86_64
1452   #define EM_X86_64       62              /* AMD x86-64 */
1453   #endif
1454 
1455   static const arch_t arch_array[]={
1456     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1457     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1458     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1459     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1460     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1461     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1462     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1463     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1464     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1465     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1466     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1467     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1468     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1469     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1470     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1471     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1472   };
1473 
1474   #if  (defined IA32)
1475     static  Elf32_Half running_arch_code=EM_386;
1476   #elif   (defined AMD64)
1477     static  Elf32_Half running_arch_code=EM_X86_64;
1478   #elif  (defined IA64)
1479     static  Elf32_Half running_arch_code=EM_IA_64;
1480   #elif  (defined __sparc) && (defined _LP64)
1481     static  Elf32_Half running_arch_code=EM_SPARCV9;
1482   #elif  (defined __sparc) && (!defined _LP64)
1483     static  Elf32_Half running_arch_code=EM_SPARC;
1484   #elif  (defined __powerpc64__)
1485     static  Elf32_Half running_arch_code=EM_PPC64;
1486   #elif  (defined __powerpc__)
1487     static  Elf32_Half running_arch_code=EM_PPC;
1488   #elif  (defined ARM)
1489     static  Elf32_Half running_arch_code=EM_ARM;
1490   #elif  (defined S390)
1491     static  Elf32_Half running_arch_code=EM_S390;
1492   #elif  (defined ALPHA)
1493     static  Elf32_Half running_arch_code=EM_ALPHA;
1494   #elif  (defined MIPSEL)
1495     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1496   #elif  (defined PARISC)
1497     static  Elf32_Half running_arch_code=EM_PARISC;
1498   #elif  (defined MIPS)
1499     static  Elf32_Half running_arch_code=EM_MIPS;
1500   #elif  (defined M68K)
1501     static  Elf32_Half running_arch_code=EM_68K;
1502   #else
1503     #error Method os::dll_load requires that one of following is defined:\
1504          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1505   #endif
1506 
1507   // Identify compatability class for VM's architecture and library's architecture
1508   // Obtain string descriptions for architectures
1509 
1510   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1511   int running_arch_index=-1;
1512 
1513   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1514     if (running_arch_code == arch_array[i].code) {
1515       running_arch_index    = i;
1516     }
1517     if (lib_arch.code == arch_array[i].code) {
1518       lib_arch.compat_class = arch_array[i].compat_class;
1519       lib_arch.name         = arch_array[i].name;
1520     }
1521   }
1522 
1523   assert(running_arch_index != -1,
1524     "Didn't find running architecture code (running_arch_code) in arch_array");
1525   if (running_arch_index == -1) {
1526     // Even though running architecture detection failed
1527     // we may still continue with reporting dlerror() message
1528     return NULL;
1529   }
1530 
1531   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1532     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1533     return NULL;
1534   }
1535 
1536 #ifndef S390
1537   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1538     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1539     return NULL;
1540   }
1541 #endif // !S390
1542 
1543   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1544     if ( lib_arch.name!=NULL ) {
1545       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1546         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1547         lib_arch.name, arch_array[running_arch_index].name);
1548     } else {
1549       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1550       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1551         lib_arch.code,
1552         arch_array[running_arch_index].name);
1553     }
1554   }
1555 
1556   return NULL;
1557 }
1558 #endif /* !__APPLE__ */
1559 
1560 void* os::get_default_process_handle() {
1561 #ifdef __APPLE__
1562   // MacOS X needs to use RTLD_FIRST instead of RTLD_LAZY
1563   // to avoid finding unexpected symbols on second (or later)
1564   // loads of a library.
1565   return (void*)::dlopen(NULL, RTLD_FIRST);
1566 #else
1567   return (void*)::dlopen(NULL, RTLD_LAZY);
1568 #endif
1569 }
1570 
1571 // XXX: Do we need a lock around this as per Linux?
1572 void* os::dll_lookup(void* handle, const char* name) {
1573   return dlsym(handle, name);
1574 }
1575 
1576 
1577 static bool _print_ascii_file(const char* filename, outputStream* st) {
1578   int fd = ::open(filename, O_RDONLY);
1579   if (fd == -1) {
1580      return false;
1581   }
1582 
1583   char buf[32];
1584   int bytes;
1585   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1586     st->print_raw(buf, bytes);
1587   }
1588 
1589   ::close(fd);
1590 
1591   return true;
1592 }
1593 
1594 void os::print_dll_info(outputStream *st) {
1595   st->print_cr("Dynamic libraries:");
1596 #ifdef RTLD_DI_LINKMAP
1597   Dl_info dli;
1598   void *handle;
1599   Link_map *map;
1600   Link_map *p;
1601 
1602   if (dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli) == 0 ||
1603       dli.dli_fname == NULL) {
1604     st->print_cr("Error: Cannot print dynamic libraries.");
1605     return;
1606   }
1607   handle = dlopen(dli.dli_fname, RTLD_LAZY);
1608   if (handle == NULL) {
1609     st->print_cr("Error: Cannot print dynamic libraries.");
1610     return;
1611   }
1612   dlinfo(handle, RTLD_DI_LINKMAP, &map);
1613   if (map == NULL) {
1614     st->print_cr("Error: Cannot print dynamic libraries.");
1615     return;
1616   }
1617 
1618   while (map->l_prev != NULL)
1619     map = map->l_prev;
1620 
1621   while (map != NULL) {
1622     st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
1623     map = map->l_next;
1624   }
1625 
1626   dlclose(handle);
1627 #elif defined(__APPLE__)
1628   uint32_t count;
1629   uint32_t i;
1630 
1631   count = _dyld_image_count();
1632   for (i = 1; i < count; i++) {
1633     const char *name = _dyld_get_image_name(i);
1634     intptr_t slide = _dyld_get_image_vmaddr_slide(i);
1635     st->print_cr(PTR_FORMAT " \t%s", slide, name);
1636   }
1637 #else
1638   st->print_cr("Error: Cannot print dynamic libraries.");
1639 #endif
1640 }
1641 
1642 void os::print_os_info_brief(outputStream* st) {
1643   st->print("Bsd");
1644 
1645   os::Posix::print_uname_info(st);
1646 }
1647 
1648 void os::print_os_info(outputStream* st) {
1649   st->print("OS:");
1650   st->print("Bsd");
1651 
1652   os::Posix::print_uname_info(st);
1653 
1654   os::Posix::print_rlimit_info(st);
1655 
1656   os::Posix::print_load_average(st);
1657 }
1658 
1659 void os::pd_print_cpu_info(outputStream* st) {
1660   // Nothing to do for now.
1661 }
1662 
1663 void os::print_memory_info(outputStream* st) {
1664 
1665   st->print("Memory:");
1666   st->print(" %dk page", os::vm_page_size()>>10);
1667 
1668   st->print(", physical " UINT64_FORMAT "k",
1669             os::physical_memory() >> 10);
1670   st->print("(" UINT64_FORMAT "k free)",
1671             os::available_memory() >> 10);
1672   st->cr();
1673 
1674   // meminfo
1675   st->print("\n/proc/meminfo:\n");
1676   _print_ascii_file("/proc/meminfo", st);
1677   st->cr();
1678 }
1679 
1680 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
1681 // but they're the same for all the bsd arch that we support
1682 // and they're the same for solaris but there's no common place to put this.
1683 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
1684                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
1685                           "ILL_COPROC", "ILL_BADSTK" };
1686 
1687 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
1688                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
1689                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
1690 
1691 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
1692 
1693 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
1694 
1695 void os::print_siginfo(outputStream* st, void* siginfo) {
1696   st->print("siginfo:");
1697 
1698   const int buflen = 100;
1699   char buf[buflen];
1700   siginfo_t *si = (siginfo_t*)siginfo;
1701   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
1702   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
1703     st->print("si_errno=%s", buf);
1704   } else {
1705     st->print("si_errno=%d", si->si_errno);
1706   }
1707   const int c = si->si_code;
1708   assert(c > 0, "unexpected si_code");
1709   switch (si->si_signo) {
1710   case SIGILL:
1711     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
1712     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1713     break;
1714   case SIGFPE:
1715     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
1716     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1717     break;
1718   case SIGSEGV:
1719     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
1720     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1721     break;
1722   case SIGBUS:
1723     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
1724     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1725     break;
1726   default:
1727     st->print(", si_code=%d", si->si_code);
1728     // no si_addr
1729   }
1730 
1731   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
1732       UseSharedSpaces) {
1733     FileMapInfo* mapinfo = FileMapInfo::current_info();
1734     if (mapinfo->is_in_shared_space(si->si_addr)) {
1735       st->print("\n\nError accessing class data sharing archive."   \
1736                 " Mapped file inaccessible during execution, "      \
1737                 " possible disk/network problem.");
1738     }
1739   }
1740   st->cr();
1741 }
1742 
1743 
1744 static void print_signal_handler(outputStream* st, int sig,
1745                                  char* buf, size_t buflen);
1746 
1747 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1748   st->print_cr("Signal Handlers:");
1749   print_signal_handler(st, SIGSEGV, buf, buflen);
1750   print_signal_handler(st, SIGBUS , buf, buflen);
1751   print_signal_handler(st, SIGFPE , buf, buflen);
1752   print_signal_handler(st, SIGPIPE, buf, buflen);
1753   print_signal_handler(st, SIGXFSZ, buf, buflen);
1754   print_signal_handler(st, SIGILL , buf, buflen);
1755   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
1756   print_signal_handler(st, SR_signum, buf, buflen);
1757   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
1758   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
1759   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
1760   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
1761 }
1762 
1763 static char saved_jvm_path[MAXPATHLEN] = {0};
1764 
1765 // Find the full path to the current module, libjvm
1766 void os::jvm_path(char *buf, jint buflen) {
1767   // Error checking.
1768   if (buflen < MAXPATHLEN) {
1769     assert(false, "must use a large-enough buffer");
1770     buf[0] = '\0';
1771     return;
1772   }
1773   // Lazy resolve the path to current module.
1774   if (saved_jvm_path[0] != 0) {
1775     strcpy(buf, saved_jvm_path);
1776     return;
1777   }
1778 
1779   char dli_fname[MAXPATHLEN];
1780   bool ret = dll_address_to_library_name(
1781                 CAST_FROM_FN_PTR(address, os::jvm_path),
1782                 dli_fname, sizeof(dli_fname), NULL);
1783   assert(ret, "cannot locate libjvm");
1784   char *rp = NULL;
1785   if (ret && dli_fname[0] != '\0') {
1786     rp = realpath(dli_fname, buf);
1787   }
1788   if (rp == NULL)
1789     return;
1790 
1791   if (Arguments::sun_java_launcher_is_altjvm()) {
1792     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
1793     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so"
1794     // or "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.dylib". If "/jre/lib/"
1795     // appears at the right place in the string, then assume we are
1796     // installed in a JDK and we're done. Otherwise, check for a
1797     // JAVA_HOME environment variable and construct a path to the JVM
1798     // being overridden.
1799 
1800     const char *p = buf + strlen(buf) - 1;
1801     for (int count = 0; p > buf && count < 5; ++count) {
1802       for (--p; p > buf && *p != '/'; --p)
1803         /* empty */ ;
1804     }
1805 
1806     if (strncmp(p, "/jre/lib/", 9) != 0) {
1807       // Look for JAVA_HOME in the environment.
1808       char* java_home_var = ::getenv("JAVA_HOME");
1809       if (java_home_var != NULL && java_home_var[0] != 0) {
1810         char* jrelib_p;
1811         int len;
1812 
1813         // Check the current module name "libjvm"
1814         p = strrchr(buf, '/');
1815         assert(strstr(p, "/libjvm") == p, "invalid library name");
1816 
1817         rp = realpath(java_home_var, buf);
1818         if (rp == NULL)
1819           return;
1820 
1821         // determine if this is a legacy image or modules image
1822         // modules image doesn't have "jre" subdirectory
1823         len = strlen(buf);
1824         jrelib_p = buf + len;
1825 
1826         // Add the appropriate library subdir
1827         snprintf(jrelib_p, buflen-len, "/jre/lib");
1828         if (0 != access(buf, F_OK)) {
1829           snprintf(jrelib_p, buflen-len, "/lib");
1830         }
1831 
1832         // Add the appropriate client or server subdir
1833         len = strlen(buf);
1834         jrelib_p = buf + len;
1835         snprintf(jrelib_p, buflen-len, "/%s", COMPILER_VARIANT);
1836         if (0 != access(buf, F_OK)) {
1837           snprintf(jrelib_p, buflen-len, "%s", "");
1838         }
1839 
1840         // If the path exists within JAVA_HOME, add the JVM library name
1841         // to complete the path to JVM being overridden.  Otherwise fallback
1842         // to the path to the current library.
1843         if (0 == access(buf, F_OK)) {
1844           // Use current module name "libjvm"
1845           len = strlen(buf);
1846           snprintf(buf + len, buflen-len, "/libjvm%s", JNI_LIB_SUFFIX);
1847         } else {
1848           // Fall back to path of current library
1849           rp = realpath(dli_fname, buf);
1850           if (rp == NULL)
1851             return;
1852         }
1853       }
1854     }
1855   }
1856 
1857   strcpy(saved_jvm_path, buf);
1858 }
1859 
1860 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
1861   // no prefix required, not even "_"
1862 }
1863 
1864 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
1865   // no suffix required
1866 }
1867 
1868 ////////////////////////////////////////////////////////////////////////////////
1869 // sun.misc.Signal support
1870 
1871 static volatile jint sigint_count = 0;
1872 
1873 static void
1874 UserHandler(int sig, void *siginfo, void *context) {
1875   // 4511530 - sem_post is serialized and handled by the manager thread. When
1876   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
1877   // don't want to flood the manager thread with sem_post requests.
1878   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
1879       return;
1880 
1881   // Ctrl-C is pressed during error reporting, likely because the error
1882   // handler fails to abort. Let VM die immediately.
1883   if (sig == SIGINT && is_error_reported()) {
1884      os::die();
1885   }
1886 
1887   os::signal_notify(sig);
1888 }
1889 
1890 void* os::user_handler() {
1891   return CAST_FROM_FN_PTR(void*, UserHandler);
1892 }
1893 
1894 extern "C" {
1895   typedef void (*sa_handler_t)(int);
1896   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
1897 }
1898 
1899 void* os::signal(int signal_number, void* handler) {
1900   struct sigaction sigAct, oldSigAct;
1901 
1902   sigfillset(&(sigAct.sa_mask));
1903   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
1904   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
1905 
1906   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
1907     // -1 means registration failed
1908     return (void *)-1;
1909   }
1910 
1911   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
1912 }
1913 
1914 void os::signal_raise(int signal_number) {
1915   ::raise(signal_number);
1916 }
1917 
1918 /*
1919  * The following code is moved from os.cpp for making this
1920  * code platform specific, which it is by its very nature.
1921  */
1922 
1923 // Will be modified when max signal is changed to be dynamic
1924 int os::sigexitnum_pd() {
1925   return NSIG;
1926 }
1927 
1928 // a counter for each possible signal value
1929 static volatile jint pending_signals[NSIG+1] = { 0 };
1930 
1931 // Bsd(POSIX) specific hand shaking semaphore.
1932 #ifdef __APPLE__
1933 typedef semaphore_t os_semaphore_t;
1934 #define SEM_INIT(sem, value)    semaphore_create(mach_task_self(), &sem, SYNC_POLICY_FIFO, value)
1935 #define SEM_WAIT(sem)           semaphore_wait(sem)
1936 #define SEM_POST(sem)           semaphore_signal(sem)
1937 #define SEM_DESTROY(sem)        semaphore_destroy(mach_task_self(), sem)
1938 #else
1939 typedef sem_t os_semaphore_t;
1940 #define SEM_INIT(sem, value)    sem_init(&sem, 0, value)
1941 #define SEM_WAIT(sem)           sem_wait(&sem)
1942 #define SEM_POST(sem)           sem_post(&sem)
1943 #define SEM_DESTROY(sem)        sem_destroy(&sem)
1944 #endif
1945 
1946 class Semaphore : public StackObj {
1947   public:
1948     Semaphore();
1949     ~Semaphore();
1950     void signal();
1951     void wait();
1952     bool trywait();
1953     bool timedwait(unsigned int sec, int nsec);
1954   private:
1955     jlong currenttime() const;
1956     os_semaphore_t _semaphore;
1957 };
1958 
1959 Semaphore::Semaphore() : _semaphore(0) {
1960   SEM_INIT(_semaphore, 0);
1961 }
1962 
1963 Semaphore::~Semaphore() {
1964   SEM_DESTROY(_semaphore);
1965 }
1966 
1967 void Semaphore::signal() {
1968   SEM_POST(_semaphore);
1969 }
1970 
1971 void Semaphore::wait() {
1972   SEM_WAIT(_semaphore);
1973 }
1974 
1975 jlong Semaphore::currenttime() const {
1976     struct timeval tv;
1977     gettimeofday(&tv, NULL);
1978     return (tv.tv_sec * NANOSECS_PER_SEC) + (tv.tv_usec * 1000);
1979 }
1980 
1981 #ifdef __APPLE__
1982 bool Semaphore::trywait() {
1983   return timedwait(0, 0);
1984 }
1985 
1986 bool Semaphore::timedwait(unsigned int sec, int nsec) {
1987   kern_return_t kr = KERN_ABORTED;
1988   mach_timespec_t waitspec;
1989   waitspec.tv_sec = sec;
1990   waitspec.tv_nsec = nsec;
1991 
1992   jlong starttime = currenttime();
1993 
1994   kr = semaphore_timedwait(_semaphore, waitspec);
1995   while (kr == KERN_ABORTED) {
1996     jlong totalwait = (sec * NANOSECS_PER_SEC) + nsec;
1997 
1998     jlong current = currenttime();
1999     jlong passedtime = current - starttime;
2000 
2001     if (passedtime >= totalwait) {
2002       waitspec.tv_sec = 0;
2003       waitspec.tv_nsec = 0;
2004     } else {
2005       jlong waittime = totalwait - (current - starttime);
2006       waitspec.tv_sec = waittime / NANOSECS_PER_SEC;
2007       waitspec.tv_nsec = waittime % NANOSECS_PER_SEC;
2008     }
2009 
2010     kr = semaphore_timedwait(_semaphore, waitspec);
2011   }
2012 
2013   return kr == KERN_SUCCESS;
2014 }
2015 
2016 #else
2017 
2018 bool Semaphore::trywait() {
2019   return sem_trywait(&_semaphore) == 0;
2020 }
2021 
2022 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2023   struct timespec ts;
2024   unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2025 
2026   while (1) {
2027     int result = sem_timedwait(&_semaphore, &ts);
2028     if (result == 0) {
2029       return true;
2030     } else if (errno == EINTR) {
2031       continue;
2032     } else if (errno == ETIMEDOUT) {
2033       return false;
2034     } else {
2035       return false;
2036     }
2037   }
2038 }
2039 
2040 #endif // __APPLE__
2041 
2042 static os_semaphore_t sig_sem;
2043 static Semaphore sr_semaphore;
2044 
2045 void os::signal_init_pd() {
2046   // Initialize signal structures
2047   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2048 
2049   // Initialize signal semaphore
2050   ::SEM_INIT(sig_sem, 0);
2051 }
2052 
2053 void os::signal_notify(int sig) {
2054   Atomic::inc(&pending_signals[sig]);
2055   ::SEM_POST(sig_sem);
2056 }
2057 
2058 static int check_pending_signals(bool wait) {
2059   Atomic::store(0, &sigint_count);
2060   for (;;) {
2061     for (int i = 0; i < NSIG + 1; i++) {
2062       jint n = pending_signals[i];
2063       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2064         return i;
2065       }
2066     }
2067     if (!wait) {
2068       return -1;
2069     }
2070     JavaThread *thread = JavaThread::current();
2071     ThreadBlockInVM tbivm(thread);
2072 
2073     bool threadIsSuspended;
2074     do {
2075       thread->set_suspend_equivalent();
2076       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2077       ::SEM_WAIT(sig_sem);
2078 
2079       // were we externally suspended while we were waiting?
2080       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2081       if (threadIsSuspended) {
2082         //
2083         // The semaphore has been incremented, but while we were waiting
2084         // another thread suspended us. We don't want to continue running
2085         // while suspended because that would surprise the thread that
2086         // suspended us.
2087         //
2088         ::SEM_POST(sig_sem);
2089 
2090         thread->java_suspend_self();
2091       }
2092     } while (threadIsSuspended);
2093   }
2094 }
2095 
2096 int os::signal_lookup() {
2097   return check_pending_signals(false);
2098 }
2099 
2100 int os::signal_wait() {
2101   return check_pending_signals(true);
2102 }
2103 
2104 ////////////////////////////////////////////////////////////////////////////////
2105 // Virtual Memory
2106 
2107 int os::vm_page_size() {
2108   // Seems redundant as all get out
2109   assert(os::Bsd::page_size() != -1, "must call os::init");
2110   return os::Bsd::page_size();
2111 }
2112 
2113 // Solaris allocates memory by pages.
2114 int os::vm_allocation_granularity() {
2115   assert(os::Bsd::page_size() != -1, "must call os::init");
2116   return os::Bsd::page_size();
2117 }
2118 
2119 // Rationale behind this function:
2120 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2121 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2122 //  samples for JITted code. Here we create private executable mapping over the code cache
2123 //  and then we can use standard (well, almost, as mapping can change) way to provide
2124 //  info for the reporting script by storing timestamp and location of symbol
2125 void bsd_wrap_code(char* base, size_t size) {
2126   static volatile jint cnt = 0;
2127 
2128   if (!UseOprofile) {
2129     return;
2130   }
2131 
2132   char buf[PATH_MAX + 1];
2133   int num = Atomic::add(1, &cnt);
2134 
2135   snprintf(buf, PATH_MAX + 1, "%s/hs-vm-%d-%d",
2136            os::get_temp_directory(), os::current_process_id(), num);
2137   unlink(buf);
2138 
2139   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2140 
2141   if (fd != -1) {
2142     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2143     if (rv != (off_t)-1) {
2144       if (::write(fd, "", 1) == 1) {
2145         mmap(base, size,
2146              PROT_READ|PROT_WRITE|PROT_EXEC,
2147              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2148       }
2149     }
2150     ::close(fd);
2151     unlink(buf);
2152   }
2153 }
2154 
2155 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2156                                     int err) {
2157   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2158           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2159           strerror(err), err);
2160 }
2161 
2162 // NOTE: Bsd kernel does not really reserve the pages for us.
2163 //       All it does is to check if there are enough free pages
2164 //       left at the time of mmap(). This could be a potential
2165 //       problem.
2166 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2167   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2168 #ifdef __OpenBSD__
2169   // XXX: Work-around mmap/MAP_FIXED bug temporarily on OpenBSD
2170   if (::mprotect(addr, size, prot) == 0) {
2171     return true;
2172   }
2173 #else
2174   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2175                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2176   if (res != (uintptr_t) MAP_FAILED) {
2177     return true;
2178   }
2179 #endif
2180 
2181   // Warn about any commit errors we see in non-product builds just
2182   // in case mmap() doesn't work as described on the man page.
2183   NOT_PRODUCT(warn_fail_commit_memory(addr, size, exec, errno);)
2184 
2185   return false;
2186 }
2187 
2188 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2189                        bool exec) {
2190   // alignment_hint is ignored on this OS
2191   return pd_commit_memory(addr, size, exec);
2192 }
2193 
2194 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2195                                   const char* mesg) {
2196   assert(mesg != NULL, "mesg must be specified");
2197   if (!pd_commit_memory(addr, size, exec)) {
2198     // add extra info in product mode for vm_exit_out_of_memory():
2199     PRODUCT_ONLY(warn_fail_commit_memory(addr, size, exec, errno);)
2200     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2201   }
2202 }
2203 
2204 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2205                                   size_t alignment_hint, bool exec,
2206                                   const char* mesg) {
2207   // alignment_hint is ignored on this OS
2208   pd_commit_memory_or_exit(addr, size, exec, mesg);
2209 }
2210 
2211 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2212 }
2213 
2214 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2215   ::madvise(addr, bytes, MADV_DONTNEED);
2216 }
2217 
2218 void os::numa_make_global(char *addr, size_t bytes) {
2219 }
2220 
2221 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2222 }
2223 
2224 bool os::numa_topology_changed()   { return false; }
2225 
2226 size_t os::numa_get_groups_num() {
2227   return 1;
2228 }
2229 
2230 int os::numa_get_group_id() {
2231   return 0;
2232 }
2233 
2234 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2235   if (size > 0) {
2236     ids[0] = 0;
2237     return 1;
2238   }
2239   return 0;
2240 }
2241 
2242 bool os::get_page_info(char *start, page_info* info) {
2243   return false;
2244 }
2245 
2246 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2247   return end;
2248 }
2249 
2250 
2251 bool os::pd_uncommit_memory(char* addr, size_t size) {
2252 #ifdef __OpenBSD__
2253   // XXX: Work-around mmap/MAP_FIXED bug temporarily on OpenBSD
2254   return ::mprotect(addr, size, PROT_NONE) == 0;
2255 #else
2256   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2257                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2258   return res  != (uintptr_t) MAP_FAILED;
2259 #endif
2260 }
2261 
2262 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2263   return os::commit_memory(addr, size, !ExecMem);
2264 }
2265 
2266 // If this is a growable mapping, remove the guard pages entirely by
2267 // munmap()ping them.  If not, just call uncommit_memory().
2268 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2269   return os::uncommit_memory(addr, size);
2270 }
2271 
2272 static address _highest_vm_reserved_address = NULL;
2273 
2274 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2275 // at 'requested_addr'. If there are existing memory mappings at the same
2276 // location, however, they will be overwritten. If 'fixed' is false,
2277 // 'requested_addr' is only treated as a hint, the return value may or
2278 // may not start from the requested address. Unlike Bsd mmap(), this
2279 // function returns NULL to indicate failure.
2280 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2281   char * addr;
2282   int flags;
2283 
2284   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2285   if (fixed) {
2286     assert((uintptr_t)requested_addr % os::Bsd::page_size() == 0, "unaligned address");
2287     flags |= MAP_FIXED;
2288   }
2289 
2290   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
2291   // touch an uncommitted page. Otherwise, the read/write might
2292   // succeed if we have enough swap space to back the physical page.
2293   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
2294                        flags, -1, 0);
2295 
2296   if (addr != MAP_FAILED) {
2297     // anon_mmap() should only get called during VM initialization,
2298     // don't need lock (actually we can skip locking even it can be called
2299     // from multiple threads, because _highest_vm_reserved_address is just a
2300     // hint about the upper limit of non-stack memory regions.)
2301     if ((address)addr + bytes > _highest_vm_reserved_address) {
2302       _highest_vm_reserved_address = (address)addr + bytes;
2303     }
2304   }
2305 
2306   return addr == MAP_FAILED ? NULL : addr;
2307 }
2308 
2309 // Don't update _highest_vm_reserved_address, because there might be memory
2310 // regions above addr + size. If so, releasing a memory region only creates
2311 // a hole in the address space, it doesn't help prevent heap-stack collision.
2312 //
2313 static int anon_munmap(char * addr, size_t size) {
2314   return ::munmap(addr, size) == 0;
2315 }
2316 
2317 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2318                          size_t alignment_hint) {
2319   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2320 }
2321 
2322 bool os::pd_release_memory(char* addr, size_t size) {
2323   return anon_munmap(addr, size);
2324 }
2325 
2326 static bool bsd_mprotect(char* addr, size_t size, int prot) {
2327   // Bsd wants the mprotect address argument to be page aligned.
2328   char* bottom = (char*)align_size_down((intptr_t)addr, os::Bsd::page_size());
2329 
2330   // According to SUSv3, mprotect() should only be used with mappings
2331   // established by mmap(), and mmap() always maps whole pages. Unaligned
2332   // 'addr' likely indicates problem in the VM (e.g. trying to change
2333   // protection of malloc'ed or statically allocated memory). Check the
2334   // caller if you hit this assert.
2335   assert(addr == bottom, "sanity check");
2336 
2337   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Bsd::page_size());
2338   return ::mprotect(bottom, size, prot) == 0;
2339 }
2340 
2341 // Set protections specified
2342 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2343                         bool is_committed) {
2344   unsigned int p = 0;
2345   switch (prot) {
2346   case MEM_PROT_NONE: p = PROT_NONE; break;
2347   case MEM_PROT_READ: p = PROT_READ; break;
2348   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2349   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2350   default:
2351     ShouldNotReachHere();
2352   }
2353   // is_committed is unused.
2354   return bsd_mprotect(addr, bytes, p);
2355 }
2356 
2357 bool os::guard_memory(char* addr, size_t size) {
2358   return bsd_mprotect(addr, size, PROT_NONE);
2359 }
2360 
2361 bool os::unguard_memory(char* addr, size_t size) {
2362   return bsd_mprotect(addr, size, PROT_READ|PROT_WRITE);
2363 }
2364 
2365 bool os::Bsd::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2366   return false;
2367 }
2368 
2369 // Large page support
2370 
2371 static size_t _large_page_size = 0;
2372 
2373 void os::large_page_init() {
2374 }
2375 
2376 
2377 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
2378   fatal("This code is not used or maintained.");
2379 
2380   // "exec" is passed in but not used.  Creating the shared image for
2381   // the code cache doesn't have an SHM_X executable permission to check.
2382   assert(UseLargePages && UseSHM, "only for SHM large pages");
2383 
2384   key_t key = IPC_PRIVATE;
2385   char *addr;
2386 
2387   bool warn_on_failure = UseLargePages &&
2388                         (!FLAG_IS_DEFAULT(UseLargePages) ||
2389                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2390                         );
2391   char msg[128];
2392 
2393   // Create a large shared memory region to attach to based on size.
2394   // Currently, size is the total size of the heap
2395   int shmid = shmget(key, bytes, IPC_CREAT|SHM_R|SHM_W);
2396   if (shmid == -1) {
2397      // Possible reasons for shmget failure:
2398      // 1. shmmax is too small for Java heap.
2399      //    > check shmmax value: cat /proc/sys/kernel/shmmax
2400      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2401      // 2. not enough large page memory.
2402      //    > check available large pages: cat /proc/meminfo
2403      //    > increase amount of large pages:
2404      //          echo new_value > /proc/sys/vm/nr_hugepages
2405      //      Note 1: different Bsd may use different name for this property,
2406      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
2407      //      Note 2: it's possible there's enough physical memory available but
2408      //            they are so fragmented after a long run that they can't
2409      //            coalesce into large pages. Try to reserve large pages when
2410      //            the system is still "fresh".
2411      if (warn_on_failure) {
2412        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2413        warning(msg);
2414      }
2415      return NULL;
2416   }
2417 
2418   // attach to the region
2419   addr = (char*)shmat(shmid, req_addr, 0);
2420   int err = errno;
2421 
2422   // Remove shmid. If shmat() is successful, the actual shared memory segment
2423   // will be deleted when it's detached by shmdt() or when the process
2424   // terminates. If shmat() is not successful this will remove the shared
2425   // segment immediately.
2426   shmctl(shmid, IPC_RMID, NULL);
2427 
2428   if ((intptr_t)addr == -1) {
2429      if (warn_on_failure) {
2430        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2431        warning(msg);
2432      }
2433      return NULL;
2434   }
2435 
2436   // The memory is committed
2437   MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
2438 
2439   return addr;
2440 }
2441 
2442 bool os::release_memory_special(char* base, size_t bytes) {
2443   MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
2444   // detaching the SHM segment will also delete it, see reserve_memory_special()
2445   int rslt = shmdt(base);
2446   if (rslt == 0) {
2447     tkr.record((address)base, bytes);
2448     return true;
2449   } else {
2450     tkr.discard();
2451     return false;
2452   }
2453 
2454 }
2455 
2456 size_t os::large_page_size() {
2457   return _large_page_size;
2458 }
2459 
2460 // HugeTLBFS allows application to commit large page memory on demand;
2461 // with SysV SHM the entire memory region must be allocated as shared
2462 // memory.
2463 bool os::can_commit_large_page_memory() {
2464   return UseHugeTLBFS;
2465 }
2466 
2467 bool os::can_execute_large_page_memory() {
2468   return UseHugeTLBFS;
2469 }
2470 
2471 // Reserve memory at an arbitrary address, only if that area is
2472 // available (and not reserved for something else).
2473 
2474 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2475   const int max_tries = 10;
2476   char* base[max_tries];
2477   size_t size[max_tries];
2478   const size_t gap = 0x000000;
2479 
2480   // Assert only that the size is a multiple of the page size, since
2481   // that's all that mmap requires, and since that's all we really know
2482   // about at this low abstraction level.  If we need higher alignment,
2483   // we can either pass an alignment to this method or verify alignment
2484   // in one of the methods further up the call chain.  See bug 5044738.
2485   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2486 
2487   // Repeatedly allocate blocks until the block is allocated at the
2488   // right spot. Give up after max_tries. Note that reserve_memory() will
2489   // automatically update _highest_vm_reserved_address if the call is
2490   // successful. The variable tracks the highest memory address every reserved
2491   // by JVM. It is used to detect heap-stack collision if running with
2492   // fixed-stack BsdThreads. Because here we may attempt to reserve more
2493   // space than needed, it could confuse the collision detecting code. To
2494   // solve the problem, save current _highest_vm_reserved_address and
2495   // calculate the correct value before return.
2496   address old_highest = _highest_vm_reserved_address;
2497 
2498   // Bsd mmap allows caller to pass an address as hint; give it a try first,
2499   // if kernel honors the hint then we can return immediately.
2500   char * addr = anon_mmap(requested_addr, bytes, false);
2501   if (addr == requested_addr) {
2502      return requested_addr;
2503   }
2504 
2505   if (addr != NULL) {
2506      // mmap() is successful but it fails to reserve at the requested address
2507      anon_munmap(addr, bytes);
2508   }
2509 
2510   int i;
2511   for (i = 0; i < max_tries; ++i) {
2512     base[i] = reserve_memory(bytes);
2513 
2514     if (base[i] != NULL) {
2515       // Is this the block we wanted?
2516       if (base[i] == requested_addr) {
2517         size[i] = bytes;
2518         break;
2519       }
2520 
2521       // Does this overlap the block we wanted? Give back the overlapped
2522       // parts and try again.
2523 
2524       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2525       if (top_overlap >= 0 && top_overlap < bytes) {
2526         unmap_memory(base[i], top_overlap);
2527         base[i] += top_overlap;
2528         size[i] = bytes - top_overlap;
2529       } else {
2530         size_t bottom_overlap = base[i] + bytes - requested_addr;
2531         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2532           unmap_memory(requested_addr, bottom_overlap);
2533           size[i] = bytes - bottom_overlap;
2534         } else {
2535           size[i] = bytes;
2536         }
2537       }
2538     }
2539   }
2540 
2541   // Give back the unused reserved pieces.
2542 
2543   for (int j = 0; j < i; ++j) {
2544     if (base[j] != NULL) {
2545       unmap_memory(base[j], size[j]);
2546     }
2547   }
2548 
2549   if (i < max_tries) {
2550     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2551     return requested_addr;
2552   } else {
2553     _highest_vm_reserved_address = old_highest;
2554     return NULL;
2555   }
2556 }
2557 
2558 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2559   RESTARTABLE_RETURN_INT(::read(fd, buf, nBytes));
2560 }
2561 
2562 // TODO-FIXME: reconcile Solaris' os::sleep with the bsd variation.
2563 // Solaris uses poll(), bsd uses park().
2564 // Poll() is likely a better choice, assuming that Thread.interrupt()
2565 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2566 // SIGSEGV, see 4355769.
2567 
2568 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2569   assert(thread == Thread::current(),  "thread consistency check");
2570 
2571   ParkEvent * const slp = thread->_SleepEvent ;
2572   slp->reset() ;
2573   OrderAccess::fence() ;
2574 
2575   if (interruptible) {
2576     jlong prevtime = javaTimeNanos();
2577 
2578     for (;;) {
2579       if (os::is_interrupted(thread, true)) {
2580         return OS_INTRPT;
2581       }
2582 
2583       jlong newtime = javaTimeNanos();
2584 
2585       if (newtime - prevtime < 0) {
2586         // time moving backwards, should only happen if no monotonic clock
2587         // not a guarantee() because JVM should not abort on kernel/glibc bugs
2588         assert(!Bsd::supports_monotonic_clock(), "time moving backwards");
2589       } else {
2590         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
2591       }
2592 
2593       if(millis <= 0) {
2594         return OS_OK;
2595       }
2596 
2597       prevtime = newtime;
2598 
2599       {
2600         assert(thread->is_Java_thread(), "sanity check");
2601         JavaThread *jt = (JavaThread *) thread;
2602         ThreadBlockInVM tbivm(jt);
2603         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
2604 
2605         jt->set_suspend_equivalent();
2606         // cleared by handle_special_suspend_equivalent_condition() or
2607         // java_suspend_self() via check_and_wait_while_suspended()
2608 
2609         slp->park(millis);
2610 
2611         // were we externally suspended while we were waiting?
2612         jt->check_and_wait_while_suspended();
2613       }
2614     }
2615   } else {
2616     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2617     jlong prevtime = javaTimeNanos();
2618 
2619     for (;;) {
2620       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
2621       // the 1st iteration ...
2622       jlong newtime = javaTimeNanos();
2623 
2624       if (newtime - prevtime < 0) {
2625         // time moving backwards, should only happen if no monotonic clock
2626         // not a guarantee() because JVM should not abort on kernel/glibc bugs
2627         assert(!Bsd::supports_monotonic_clock(), "time moving backwards");
2628       } else {
2629         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
2630       }
2631 
2632       if(millis <= 0) break ;
2633 
2634       prevtime = newtime;
2635       slp->park(millis);
2636     }
2637     return OS_OK ;
2638   }
2639 }
2640 
2641 void os::naked_short_sleep(jlong ms) {
2642   struct timespec req;
2643 
2644   assert(ms < 1000, "Un-interruptable sleep, short time use only");
2645   req.tv_sec = 0;
2646   if (ms > 0) {
2647     req.tv_nsec = (ms % 1000) * 1000000;
2648   }
2649   else {
2650     req.tv_nsec = 1;
2651   }
2652 
2653   nanosleep(&req, NULL);
2654 
2655   return;
2656 }
2657 
2658 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2659 void os::infinite_sleep() {
2660   while (true) {    // sleep forever ...
2661     ::sleep(100);   // ... 100 seconds at a time
2662   }
2663 }
2664 
2665 // Used to convert frequent JVM_Yield() to nops
2666 bool os::dont_yield() {
2667   return DontYieldALot;
2668 }
2669 
2670 void os::yield() {
2671   sched_yield();
2672 }
2673 
2674 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
2675 
2676 void os::yield_all(int attempts) {
2677   // Yields to all threads, including threads with lower priorities
2678   // Threads on Bsd are all with same priority. The Solaris style
2679   // os::yield_all() with nanosleep(1ms) is not necessary.
2680   sched_yield();
2681 }
2682 
2683 // Called from the tight loops to possibly influence time-sharing heuristics
2684 void os::loop_breaker(int attempts) {
2685   os::yield_all(attempts);
2686 }
2687 
2688 ////////////////////////////////////////////////////////////////////////////////
2689 // thread priority support
2690 
2691 // Note: Normal Bsd applications are run with SCHED_OTHER policy. SCHED_OTHER
2692 // only supports dynamic priority, static priority must be zero. For real-time
2693 // applications, Bsd supports SCHED_RR which allows static priority (1-99).
2694 // However, for large multi-threaded applications, SCHED_RR is not only slower
2695 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
2696 // of 5 runs - Sep 2005).
2697 //
2698 // The following code actually changes the niceness of kernel-thread/LWP. It
2699 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
2700 // not the entire user process, and user level threads are 1:1 mapped to kernel
2701 // threads. It has always been the case, but could change in the future. For
2702 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
2703 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
2704 
2705 #if !defined(__APPLE__)
2706 int os::java_to_os_priority[CriticalPriority + 1] = {
2707   19,              // 0 Entry should never be used
2708 
2709    0,              // 1 MinPriority
2710    3,              // 2
2711    6,              // 3
2712 
2713   10,              // 4
2714   15,              // 5 NormPriority
2715   18,              // 6
2716 
2717   21,              // 7
2718   25,              // 8
2719   28,              // 9 NearMaxPriority
2720 
2721   31,              // 10 MaxPriority
2722 
2723   31               // 11 CriticalPriority
2724 };
2725 #else
2726 /* Using Mach high-level priority assignments */
2727 int os::java_to_os_priority[CriticalPriority + 1] = {
2728    0,              // 0 Entry should never be used (MINPRI_USER)
2729 
2730   27,              // 1 MinPriority
2731   28,              // 2
2732   29,              // 3
2733 
2734   30,              // 4
2735   31,              // 5 NormPriority (BASEPRI_DEFAULT)
2736   32,              // 6
2737 
2738   33,              // 7
2739   34,              // 8
2740   35,              // 9 NearMaxPriority
2741 
2742   36,              // 10 MaxPriority
2743 
2744   36               // 11 CriticalPriority
2745 };
2746 #endif
2747 
2748 static int prio_init() {
2749   if (ThreadPriorityPolicy == 1) {
2750     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
2751     // if effective uid is not root. Perhaps, a more elegant way of doing
2752     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
2753     if (geteuid() != 0) {
2754       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
2755         warning("-XX:ThreadPriorityPolicy requires root privilege on Bsd");
2756       }
2757       ThreadPriorityPolicy = 0;
2758     }
2759   }
2760   if (UseCriticalJavaThreadPriority) {
2761     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
2762   }
2763   return 0;
2764 }
2765 
2766 OSReturn os::set_native_priority(Thread* thread, int newpri) {
2767   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
2768 
2769 #ifdef __OpenBSD__
2770   // OpenBSD pthread_setprio starves low priority threads
2771   return OS_OK;
2772 #elif defined(__FreeBSD__)
2773   int ret = pthread_setprio(thread->osthread()->pthread_id(), newpri);
2774 #elif defined(__APPLE__) || defined(__NetBSD__)
2775   struct sched_param sp;
2776   int policy;
2777   pthread_t self = pthread_self();
2778 
2779   if (pthread_getschedparam(self, &policy, &sp) != 0)
2780     return OS_ERR;
2781 
2782   sp.sched_priority = newpri;
2783   if (pthread_setschedparam(self, policy, &sp) != 0)
2784     return OS_ERR;
2785 
2786   return OS_OK;
2787 #else
2788   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
2789   return (ret == 0) ? OS_OK : OS_ERR;
2790 #endif
2791 }
2792 
2793 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
2794   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
2795     *priority_ptr = java_to_os_priority[NormPriority];
2796     return OS_OK;
2797   }
2798 
2799   errno = 0;
2800 #if defined(__OpenBSD__) || defined(__FreeBSD__)
2801   *priority_ptr = pthread_getprio(thread->osthread()->pthread_id());
2802 #elif defined(__APPLE__) || defined(__NetBSD__)
2803   int policy;
2804   struct sched_param sp;
2805 
2806   pthread_getschedparam(pthread_self(), &policy, &sp);
2807   *priority_ptr = sp.sched_priority;
2808 #else
2809   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
2810 #endif
2811   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
2812 }
2813 
2814 // Hint to the underlying OS that a task switch would not be good.
2815 // Void return because it's a hint and can fail.
2816 void os::hint_no_preempt() {}
2817 
2818 ////////////////////////////////////////////////////////////////////////////////
2819 // suspend/resume support
2820 
2821 //  the low-level signal-based suspend/resume support is a remnant from the
2822 //  old VM-suspension that used to be for java-suspension, safepoints etc,
2823 //  within hotspot. Now there is a single use-case for this:
2824 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
2825 //      that runs in the watcher thread.
2826 //  The remaining code is greatly simplified from the more general suspension
2827 //  code that used to be used.
2828 //
2829 //  The protocol is quite simple:
2830 //  - suspend:
2831 //      - sends a signal to the target thread
2832 //      - polls the suspend state of the osthread using a yield loop
2833 //      - target thread signal handler (SR_handler) sets suspend state
2834 //        and blocks in sigsuspend until continued
2835 //  - resume:
2836 //      - sets target osthread state to continue
2837 //      - sends signal to end the sigsuspend loop in the SR_handler
2838 //
2839 //  Note that the SR_lock plays no role in this suspend/resume protocol.
2840 //
2841 
2842 static void resume_clear_context(OSThread *osthread) {
2843   osthread->set_ucontext(NULL);
2844   osthread->set_siginfo(NULL);
2845 }
2846 
2847 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
2848   osthread->set_ucontext(context);
2849   osthread->set_siginfo(siginfo);
2850 }
2851 
2852 //
2853 // Handler function invoked when a thread's execution is suspended or
2854 // resumed. We have to be careful that only async-safe functions are
2855 // called here (Note: most pthread functions are not async safe and
2856 // should be avoided.)
2857 //
2858 // Note: sigwait() is a more natural fit than sigsuspend() from an
2859 // interface point of view, but sigwait() prevents the signal hander
2860 // from being run. libpthread would get very confused by not having
2861 // its signal handlers run and prevents sigwait()'s use with the
2862 // mutex granting granting signal.
2863 //
2864 // Currently only ever called on the VMThread or JavaThread
2865 //
2866 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
2867   // Save and restore errno to avoid confusing native code with EINTR
2868   // after sigsuspend.
2869   int old_errno = errno;
2870 
2871   Thread* thread = Thread::current();
2872   OSThread* osthread = thread->osthread();
2873   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
2874 
2875   os::SuspendResume::State current = osthread->sr.state();
2876   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
2877     suspend_save_context(osthread, siginfo, context);
2878 
2879     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
2880     os::SuspendResume::State state = osthread->sr.suspended();
2881     if (state == os::SuspendResume::SR_SUSPENDED) {
2882       sigset_t suspend_set;  // signals for sigsuspend()
2883 
2884       // get current set of blocked signals and unblock resume signal
2885       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
2886       sigdelset(&suspend_set, SR_signum);
2887 
2888       sr_semaphore.signal();
2889       // wait here until we are resumed
2890       while (1) {
2891         sigsuspend(&suspend_set);
2892 
2893         os::SuspendResume::State result = osthread->sr.running();
2894         if (result == os::SuspendResume::SR_RUNNING) {
2895           sr_semaphore.signal();
2896           break;
2897         } else if (result != os::SuspendResume::SR_SUSPENDED) {
2898           ShouldNotReachHere();
2899         }
2900       }
2901 
2902     } else if (state == os::SuspendResume::SR_RUNNING) {
2903       // request was cancelled, continue
2904     } else {
2905       ShouldNotReachHere();
2906     }
2907 
2908     resume_clear_context(osthread);
2909   } else if (current == os::SuspendResume::SR_RUNNING) {
2910     // request was cancelled, continue
2911   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
2912     // ignore
2913   } else {
2914     // ignore
2915   }
2916 
2917   errno = old_errno;
2918 }
2919 
2920 
2921 static int SR_initialize() {
2922   struct sigaction act;
2923   char *s;
2924   /* Get signal number to use for suspend/resume */
2925   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
2926     int sig = ::strtol(s, 0, 10);
2927     if (sig > 0 || sig < NSIG) {
2928         SR_signum = sig;
2929     }
2930   }
2931 
2932   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
2933         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
2934 
2935   sigemptyset(&SR_sigset);
2936   sigaddset(&SR_sigset, SR_signum);
2937 
2938   /* Set up signal handler for suspend/resume */
2939   act.sa_flags = SA_RESTART|SA_SIGINFO;
2940   act.sa_handler = (void (*)(int)) SR_handler;
2941 
2942   // SR_signum is blocked by default.
2943   // 4528190 - We also need to block pthread restart signal (32 on all
2944   // supported Bsd platforms). Note that BsdThreads need to block
2945   // this signal for all threads to work properly. So we don't have
2946   // to use hard-coded signal number when setting up the mask.
2947   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
2948 
2949   if (sigaction(SR_signum, &act, 0) == -1) {
2950     return -1;
2951   }
2952 
2953   // Save signal flag
2954   os::Bsd::set_our_sigflags(SR_signum, act.sa_flags);
2955   return 0;
2956 }
2957 
2958 static int sr_notify(OSThread* osthread) {
2959   int status = pthread_kill(osthread->pthread_id(), SR_signum);
2960   assert_status(status == 0, status, "pthread_kill");
2961   return status;
2962 }
2963 
2964 // "Randomly" selected value for how long we want to spin
2965 // before bailing out on suspending a thread, also how often
2966 // we send a signal to a thread we want to resume
2967 static const int RANDOMLY_LARGE_INTEGER = 1000000;
2968 static const int RANDOMLY_LARGE_INTEGER2 = 100;
2969 
2970 // returns true on success and false on error - really an error is fatal
2971 // but this seems the normal response to library errors
2972 static bool do_suspend(OSThread* osthread) {
2973   assert(osthread->sr.is_running(), "thread should be running");
2974   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
2975 
2976   // mark as suspended and send signal
2977   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
2978     // failed to switch, state wasn't running?
2979     ShouldNotReachHere();
2980     return false;
2981   }
2982 
2983   if (sr_notify(osthread) != 0) {
2984     ShouldNotReachHere();
2985   }
2986 
2987   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
2988   while (true) {
2989     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
2990       break;
2991     } else {
2992       // timeout
2993       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
2994       if (cancelled == os::SuspendResume::SR_RUNNING) {
2995         return false;
2996       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
2997         // make sure that we consume the signal on the semaphore as well
2998         sr_semaphore.wait();
2999         break;
3000       } else {
3001         ShouldNotReachHere();
3002         return false;
3003       }
3004     }
3005   }
3006 
3007   guarantee(osthread->sr.is_suspended(), "Must be suspended");
3008   return true;
3009 }
3010 
3011 static void do_resume(OSThread* osthread) {
3012   assert(osthread->sr.is_suspended(), "thread should be suspended");
3013   assert(!sr_semaphore.trywait(), "invalid semaphore state");
3014 
3015   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3016     // failed to switch to WAKEUP_REQUEST
3017     ShouldNotReachHere();
3018     return;
3019   }
3020 
3021   while (true) {
3022     if (sr_notify(osthread) == 0) {
3023       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3024         if (osthread->sr.is_running()) {
3025           return;
3026         }
3027       }
3028     } else {
3029       ShouldNotReachHere();
3030     }
3031   }
3032 
3033   guarantee(osthread->sr.is_running(), "Must be running!");
3034 }
3035 
3036 ////////////////////////////////////////////////////////////////////////////////
3037 // interrupt support
3038 
3039 void os::interrupt(Thread* thread) {
3040   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3041     "possibility of dangling Thread pointer");
3042 
3043   OSThread* osthread = thread->osthread();
3044 
3045   if (!osthread->interrupted()) {
3046     osthread->set_interrupted(true);
3047     // More than one thread can get here with the same value of osthread,
3048     // resulting in multiple notifications.  We do, however, want the store
3049     // to interrupted() to be visible to other threads before we execute unpark().
3050     OrderAccess::fence();
3051     ParkEvent * const slp = thread->_SleepEvent ;
3052     if (slp != NULL) slp->unpark() ;
3053   }
3054 
3055   // For JSR166. Unpark even if interrupt status already was set
3056   if (thread->is_Java_thread())
3057     ((JavaThread*)thread)->parker()->unpark();
3058 
3059   ParkEvent * ev = thread->_ParkEvent ;
3060   if (ev != NULL) ev->unpark() ;
3061 
3062 }
3063 
3064 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3065   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3066     "possibility of dangling Thread pointer");
3067 
3068   OSThread* osthread = thread->osthread();
3069 
3070   bool interrupted = osthread->interrupted();
3071 
3072   if (interrupted && clear_interrupted) {
3073     osthread->set_interrupted(false);
3074     // consider thread->_SleepEvent->reset() ... optional optimization
3075   }
3076 
3077   return interrupted;
3078 }
3079 
3080 ///////////////////////////////////////////////////////////////////////////////////
3081 // signal handling (except suspend/resume)
3082 
3083 // This routine may be used by user applications as a "hook" to catch signals.
3084 // The user-defined signal handler must pass unrecognized signals to this
3085 // routine, and if it returns true (non-zero), then the signal handler must
3086 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3087 // routine will never retun false (zero), but instead will execute a VM panic
3088 // routine kill the process.
3089 //
3090 // If this routine returns false, it is OK to call it again.  This allows
3091 // the user-defined signal handler to perform checks either before or after
3092 // the VM performs its own checks.  Naturally, the user code would be making
3093 // a serious error if it tried to handle an exception (such as a null check
3094 // or breakpoint) that the VM was generating for its own correct operation.
3095 //
3096 // This routine may recognize any of the following kinds of signals:
3097 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3098 // It should be consulted by handlers for any of those signals.
3099 //
3100 // The caller of this routine must pass in the three arguments supplied
3101 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3102 // field of the structure passed to sigaction().  This routine assumes that
3103 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3104 //
3105 // Note that the VM will print warnings if it detects conflicting signal
3106 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3107 //
3108 extern "C" JNIEXPORT int
3109 JVM_handle_bsd_signal(int signo, siginfo_t* siginfo,
3110                         void* ucontext, int abort_if_unrecognized);
3111 
3112 void signalHandler(int sig, siginfo_t* info, void* uc) {
3113   assert(info != NULL && uc != NULL, "it must be old kernel");
3114   int orig_errno = errno;  // Preserve errno value over signal handler.
3115   JVM_handle_bsd_signal(sig, info, uc, true);
3116   errno = orig_errno;
3117 }
3118 
3119 
3120 // This boolean allows users to forward their own non-matching signals
3121 // to JVM_handle_bsd_signal, harmlessly.
3122 bool os::Bsd::signal_handlers_are_installed = false;
3123 
3124 // For signal-chaining
3125 struct sigaction os::Bsd::sigact[MAXSIGNUM];
3126 unsigned int os::Bsd::sigs = 0;
3127 bool os::Bsd::libjsig_is_loaded = false;
3128 typedef struct sigaction *(*get_signal_t)(int);
3129 get_signal_t os::Bsd::get_signal_action = NULL;
3130 
3131 struct sigaction* os::Bsd::get_chained_signal_action(int sig) {
3132   struct sigaction *actp = NULL;
3133 
3134   if (libjsig_is_loaded) {
3135     // Retrieve the old signal handler from libjsig
3136     actp = (*get_signal_action)(sig);
3137   }
3138   if (actp == NULL) {
3139     // Retrieve the preinstalled signal handler from jvm
3140     actp = get_preinstalled_handler(sig);
3141   }
3142 
3143   return actp;
3144 }
3145 
3146 static bool call_chained_handler(struct sigaction *actp, int sig,
3147                                  siginfo_t *siginfo, void *context) {
3148   // Call the old signal handler
3149   if (actp->sa_handler == SIG_DFL) {
3150     // It's more reasonable to let jvm treat it as an unexpected exception
3151     // instead of taking the default action.
3152     return false;
3153   } else if (actp->sa_handler != SIG_IGN) {
3154     if ((actp->sa_flags & SA_NODEFER) == 0) {
3155       // automaticlly block the signal
3156       sigaddset(&(actp->sa_mask), sig);
3157     }
3158 
3159     sa_handler_t hand;
3160     sa_sigaction_t sa;
3161     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3162     // retrieve the chained handler
3163     if (siginfo_flag_set) {
3164       sa = actp->sa_sigaction;
3165     } else {
3166       hand = actp->sa_handler;
3167     }
3168 
3169     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3170       actp->sa_handler = SIG_DFL;
3171     }
3172 
3173     // try to honor the signal mask
3174     sigset_t oset;
3175     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3176 
3177     // call into the chained handler
3178     if (siginfo_flag_set) {
3179       (*sa)(sig, siginfo, context);
3180     } else {
3181       (*hand)(sig);
3182     }
3183 
3184     // restore the signal mask
3185     pthread_sigmask(SIG_SETMASK, &oset, 0);
3186   }
3187   // Tell jvm's signal handler the signal is taken care of.
3188   return true;
3189 }
3190 
3191 bool os::Bsd::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3192   bool chained = false;
3193   // signal-chaining
3194   if (UseSignalChaining) {
3195     struct sigaction *actp = get_chained_signal_action(sig);
3196     if (actp != NULL) {
3197       chained = call_chained_handler(actp, sig, siginfo, context);
3198     }
3199   }
3200   return chained;
3201 }
3202 
3203 struct sigaction* os::Bsd::get_preinstalled_handler(int sig) {
3204   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3205     return &sigact[sig];
3206   }
3207   return NULL;
3208 }
3209 
3210 void os::Bsd::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3211   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3212   sigact[sig] = oldAct;
3213   sigs |= (unsigned int)1 << sig;
3214 }
3215 
3216 // for diagnostic
3217 int os::Bsd::sigflags[MAXSIGNUM];
3218 
3219 int os::Bsd::get_our_sigflags(int sig) {
3220   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3221   return sigflags[sig];
3222 }
3223 
3224 void os::Bsd::set_our_sigflags(int sig, int flags) {
3225   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3226   sigflags[sig] = flags;
3227 }
3228 
3229 void os::Bsd::set_signal_handler(int sig, bool set_installed) {
3230   // Check for overwrite.
3231   struct sigaction oldAct;
3232   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3233 
3234   void* oldhand = oldAct.sa_sigaction
3235                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3236                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3237   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3238       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3239       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3240     if (AllowUserSignalHandlers || !set_installed) {
3241       // Do not overwrite; user takes responsibility to forward to us.
3242       return;
3243     } else if (UseSignalChaining) {
3244       // save the old handler in jvm
3245       save_preinstalled_handler(sig, oldAct);
3246       // libjsig also interposes the sigaction() call below and saves the
3247       // old sigaction on it own.
3248     } else {
3249       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3250                     "%#lx for signal %d.", (long)oldhand, sig));
3251     }
3252   }
3253 
3254   struct sigaction sigAct;
3255   sigfillset(&(sigAct.sa_mask));
3256   sigAct.sa_handler = SIG_DFL;
3257   if (!set_installed) {
3258     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3259   } else {
3260     sigAct.sa_sigaction = signalHandler;
3261     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3262   }
3263 #if __APPLE__
3264   // Needed for main thread as XNU (Mac OS X kernel) will only deliver SIGSEGV
3265   // (which starts as SIGBUS) on main thread with faulting address inside "stack+guard pages"
3266   // if the signal handler declares it will handle it on alternate stack.
3267   // Notice we only declare we will handle it on alt stack, but we are not
3268   // actually going to use real alt stack - this is just a workaround.
3269   // Please see ux_exception.c, method catch_mach_exception_raise for details
3270   // link http://www.opensource.apple.com/source/xnu/xnu-2050.18.24/bsd/uxkern/ux_exception.c
3271   if (sig == SIGSEGV) {
3272     sigAct.sa_flags |= SA_ONSTACK;
3273   }
3274 #endif
3275 
3276   // Save flags, which are set by ours
3277   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3278   sigflags[sig] = sigAct.sa_flags;
3279 
3280   int ret = sigaction(sig, &sigAct, &oldAct);
3281   assert(ret == 0, "check");
3282 
3283   void* oldhand2  = oldAct.sa_sigaction
3284                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3285                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3286   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3287 }
3288 
3289 // install signal handlers for signals that HotSpot needs to
3290 // handle in order to support Java-level exception handling.
3291 
3292 void os::Bsd::install_signal_handlers() {
3293   if (!signal_handlers_are_installed) {
3294     signal_handlers_are_installed = true;
3295 
3296     // signal-chaining
3297     typedef void (*signal_setting_t)();
3298     signal_setting_t begin_signal_setting = NULL;
3299     signal_setting_t end_signal_setting = NULL;
3300     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3301                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3302     if (begin_signal_setting != NULL) {
3303       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3304                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3305       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3306                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3307       libjsig_is_loaded = true;
3308       assert(UseSignalChaining, "should enable signal-chaining");
3309     }
3310     if (libjsig_is_loaded) {
3311       // Tell libjsig jvm is setting signal handlers
3312       (*begin_signal_setting)();
3313     }
3314 
3315     set_signal_handler(SIGSEGV, true);
3316     set_signal_handler(SIGPIPE, true);
3317     set_signal_handler(SIGBUS, true);
3318     set_signal_handler(SIGILL, true);
3319     set_signal_handler(SIGFPE, true);
3320     set_signal_handler(SIGXFSZ, true);
3321 
3322 #if defined(__APPLE__)
3323     // In Mac OS X 10.4, CrashReporter will write a crash log for all 'fatal' signals, including
3324     // signals caught and handled by the JVM. To work around this, we reset the mach task
3325     // signal handler that's placed on our process by CrashReporter. This disables
3326     // CrashReporter-based reporting.
3327     //
3328     // This work-around is not necessary for 10.5+, as CrashReporter no longer intercedes
3329     // on caught fatal signals.
3330     //
3331     // Additionally, gdb installs both standard BSD signal handlers, and mach exception
3332     // handlers. By replacing the existing task exception handler, we disable gdb's mach
3333     // exception handling, while leaving the standard BSD signal handlers functional.
3334     kern_return_t kr;
3335     kr = task_set_exception_ports(mach_task_self(),
3336         EXC_MASK_BAD_ACCESS | EXC_MASK_ARITHMETIC,
3337         MACH_PORT_NULL,
3338         EXCEPTION_STATE_IDENTITY,
3339         MACHINE_THREAD_STATE);
3340 
3341     assert(kr == KERN_SUCCESS, "could not set mach task signal handler");
3342 #endif
3343 
3344     if (libjsig_is_loaded) {
3345       // Tell libjsig jvm finishes setting signal handlers
3346       (*end_signal_setting)();
3347     }
3348 
3349     // We don't activate signal checker if libjsig is in place, we trust ourselves
3350     // and if UserSignalHandler is installed all bets are off
3351     if (CheckJNICalls) {
3352       if (libjsig_is_loaded) {
3353         if (PrintJNIResolving) {
3354           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3355         }
3356         check_signals = false;
3357       }
3358       if (AllowUserSignalHandlers) {
3359         if (PrintJNIResolving) {
3360           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3361         }
3362         check_signals = false;
3363       }
3364     }
3365   }
3366 }
3367 
3368 
3369 /////
3370 // glibc on Bsd platform uses non-documented flag
3371 // to indicate, that some special sort of signal
3372 // trampoline is used.
3373 // We will never set this flag, and we should
3374 // ignore this flag in our diagnostic
3375 #ifdef SIGNIFICANT_SIGNAL_MASK
3376 #undef SIGNIFICANT_SIGNAL_MASK
3377 #endif
3378 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3379 
3380 static const char* get_signal_handler_name(address handler,
3381                                            char* buf, int buflen) {
3382   int offset;
3383   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3384   if (found) {
3385     // skip directory names
3386     const char *p1, *p2;
3387     p1 = buf;
3388     size_t len = strlen(os::file_separator());
3389     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3390     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3391   } else {
3392     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3393   }
3394   return buf;
3395 }
3396 
3397 static void print_signal_handler(outputStream* st, int sig,
3398                                  char* buf, size_t buflen) {
3399   struct sigaction sa;
3400 
3401   sigaction(sig, NULL, &sa);
3402 
3403   // See comment for SIGNIFICANT_SIGNAL_MASK define
3404   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3405 
3406   st->print("%s: ", os::exception_name(sig, buf, buflen));
3407 
3408   address handler = (sa.sa_flags & SA_SIGINFO)
3409     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3410     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3411 
3412   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3413     st->print("SIG_DFL");
3414   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3415     st->print("SIG_IGN");
3416   } else {
3417     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3418   }
3419 
3420   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3421 
3422   address rh = VMError::get_resetted_sighandler(sig);
3423   // May be, handler was resetted by VMError?
3424   if(rh != NULL) {
3425     handler = rh;
3426     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3427   }
3428 
3429   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
3430 
3431   // Check: is it our handler?
3432   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3433      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3434     // It is our signal handler
3435     // check for flags, reset system-used one!
3436     if((int)sa.sa_flags != os::Bsd::get_our_sigflags(sig)) {
3437       st->print(
3438                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3439                 os::Bsd::get_our_sigflags(sig));
3440     }
3441   }
3442   st->cr();
3443 }
3444 
3445 
3446 #define DO_SIGNAL_CHECK(sig) \
3447   if (!sigismember(&check_signal_done, sig)) \
3448     os::Bsd::check_signal_handler(sig)
3449 
3450 // This method is a periodic task to check for misbehaving JNI applications
3451 // under CheckJNI, we can add any periodic checks here
3452 
3453 void os::run_periodic_checks() {
3454 
3455   if (check_signals == false) return;
3456 
3457   // SEGV and BUS if overridden could potentially prevent
3458   // generation of hs*.log in the event of a crash, debugging
3459   // such a case can be very challenging, so we absolutely
3460   // check the following for a good measure:
3461   DO_SIGNAL_CHECK(SIGSEGV);
3462   DO_SIGNAL_CHECK(SIGILL);
3463   DO_SIGNAL_CHECK(SIGFPE);
3464   DO_SIGNAL_CHECK(SIGBUS);
3465   DO_SIGNAL_CHECK(SIGPIPE);
3466   DO_SIGNAL_CHECK(SIGXFSZ);
3467 
3468 
3469   // ReduceSignalUsage allows the user to override these handlers
3470   // see comments at the very top and jvm_solaris.h
3471   if (!ReduceSignalUsage) {
3472     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3473     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3474     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3475     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3476   }
3477 
3478   DO_SIGNAL_CHECK(SR_signum);
3479   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3480 }
3481 
3482 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3483 
3484 static os_sigaction_t os_sigaction = NULL;
3485 
3486 void os::Bsd::check_signal_handler(int sig) {
3487   char buf[O_BUFLEN];
3488   address jvmHandler = NULL;
3489 
3490 
3491   struct sigaction act;
3492   if (os_sigaction == NULL) {
3493     // only trust the default sigaction, in case it has been interposed
3494     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3495     if (os_sigaction == NULL) return;
3496   }
3497 
3498   os_sigaction(sig, (struct sigaction*)NULL, &act);
3499 
3500 
3501   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3502 
3503   address thisHandler = (act.sa_flags & SA_SIGINFO)
3504     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3505     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3506 
3507 
3508   switch(sig) {
3509   case SIGSEGV:
3510   case SIGBUS:
3511   case SIGFPE:
3512   case SIGPIPE:
3513   case SIGILL:
3514   case SIGXFSZ:
3515     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3516     break;
3517 
3518   case SHUTDOWN1_SIGNAL:
3519   case SHUTDOWN2_SIGNAL:
3520   case SHUTDOWN3_SIGNAL:
3521   case BREAK_SIGNAL:
3522     jvmHandler = (address)user_handler();
3523     break;
3524 
3525   case INTERRUPT_SIGNAL:
3526     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3527     break;
3528 
3529   default:
3530     if (sig == SR_signum) {
3531       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3532     } else {
3533       return;
3534     }
3535     break;
3536   }
3537 
3538   if (thisHandler != jvmHandler) {
3539     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3540     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3541     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3542     // No need to check this sig any longer
3543     sigaddset(&check_signal_done, sig);
3544   } else if(os::Bsd::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Bsd::get_our_sigflags(sig)) {
3545     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3546     tty->print("expected:" PTR32_FORMAT, os::Bsd::get_our_sigflags(sig));
3547     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
3548     // No need to check this sig any longer
3549     sigaddset(&check_signal_done, sig);
3550   }
3551 
3552   // Dump all the signal
3553   if (sigismember(&check_signal_done, sig)) {
3554     print_signal_handlers(tty, buf, O_BUFLEN);
3555   }
3556 }
3557 
3558 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3559 
3560 extern bool signal_name(int signo, char* buf, size_t len);
3561 
3562 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3563   if (0 < exception_code && exception_code <= SIGRTMAX) {
3564     // signal
3565     if (!signal_name(exception_code, buf, size)) {
3566       jio_snprintf(buf, size, "SIG%d", exception_code);
3567     }
3568     return buf;
3569   } else {
3570     return NULL;
3571   }
3572 }
3573 
3574 // this is called _before_ the most of global arguments have been parsed
3575 void os::init(void) {
3576   char dummy;   /* used to get a guess on initial stack address */
3577 //  first_hrtime = gethrtime();
3578 
3579   // With BsdThreads the JavaMain thread pid (primordial thread)
3580   // is different than the pid of the java launcher thread.
3581   // So, on Bsd, the launcher thread pid is passed to the VM
3582   // via the sun.java.launcher.pid property.
3583   // Use this property instead of getpid() if it was correctly passed.
3584   // See bug 6351349.
3585   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3586 
3587   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3588 
3589   clock_tics_per_sec = CLK_TCK;
3590 
3591   init_random(1234567);
3592 
3593   ThreadCritical::initialize();
3594 
3595   Bsd::set_page_size(getpagesize());
3596   if (Bsd::page_size() == -1) {
3597     fatal(err_msg("os_bsd.cpp: os::init: sysconf failed (%s)",
3598                   strerror(errno)));
3599   }
3600   init_page_sizes((size_t) Bsd::page_size());
3601 
3602   Bsd::initialize_system_info();
3603 
3604   // main_thread points to the aboriginal thread
3605   Bsd::_main_thread = pthread_self();
3606 
3607   Bsd::clock_init();
3608   initial_time_count = javaTimeNanos();
3609 
3610 #ifdef __APPLE__
3611   // XXXDARWIN
3612   // Work around the unaligned VM callbacks in hotspot's
3613   // sharedRuntime. The callbacks don't use SSE2 instructions, and work on
3614   // Linux, Solaris, and FreeBSD. On Mac OS X, dyld (rightly so) enforces
3615   // alignment when doing symbol lookup. To work around this, we force early
3616   // binding of all symbols now, thus binding when alignment is known-good.
3617   _dyld_bind_fully_image_containing_address((const void *) &os::init);
3618 #endif
3619 }
3620 
3621 // To install functions for atexit system call
3622 extern "C" {
3623   static void perfMemory_exit_helper() {
3624     perfMemory_exit();
3625   }
3626 }
3627 
3628 // this is called _after_ the global arguments have been parsed
3629 jint os::init_2(void)
3630 {
3631   // Allocate a single page and mark it as readable for safepoint polling
3632   address polling_page = (address) ::mmap(NULL, Bsd::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3633   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3634 
3635   os::set_polling_page( polling_page );
3636 
3637 #ifndef PRODUCT
3638   if(Verbose && PrintMiscellaneous)
3639     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3640 #endif
3641 
3642   if (!UseMembar) {
3643     address mem_serialize_page = (address) ::mmap(NULL, Bsd::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3644     guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
3645     os::set_memory_serialize_page( mem_serialize_page );
3646 
3647 #ifndef PRODUCT
3648     if(Verbose && PrintMiscellaneous)
3649       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3650 #endif
3651   }
3652 
3653   // initialize suspend/resume support - must do this before signal_sets_init()
3654   if (SR_initialize() != 0) {
3655     perror("SR_initialize failed");
3656     return JNI_ERR;
3657   }
3658 
3659   Bsd::signal_sets_init();
3660   Bsd::install_signal_handlers();
3661 
3662   // Check minimum allowable stack size for thread creation and to initialize
3663   // the java system classes, including StackOverflowError - depends on page
3664   // size.  Add a page for compiler2 recursion in main thread.
3665   // Add in 2*BytesPerWord times page size to account for VM stack during
3666   // class initialization depending on 32 or 64 bit VM.
3667   os::Bsd::min_stack_allowed = MAX2(os::Bsd::min_stack_allowed,
3668             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
3669                     2*BytesPerWord COMPILER2_PRESENT(+1)) * Bsd::page_size());
3670 
3671   size_t threadStackSizeInBytes = ThreadStackSize * K;
3672   if (threadStackSizeInBytes != 0 &&
3673       threadStackSizeInBytes < os::Bsd::min_stack_allowed) {
3674         tty->print_cr("\nThe stack size specified is too small, "
3675                       "Specify at least %dk",
3676                       os::Bsd::min_stack_allowed/ K);
3677         return JNI_ERR;
3678   }
3679 
3680   // Make the stack size a multiple of the page size so that
3681   // the yellow/red zones can be guarded.
3682   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3683         vm_page_size()));
3684 
3685   if (MaxFDLimit) {
3686     // set the number of file descriptors to max. print out error
3687     // if getrlimit/setrlimit fails but continue regardless.
3688     struct rlimit nbr_files;
3689     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3690     if (status != 0) {
3691       if (PrintMiscellaneous && (Verbose || WizardMode))
3692         perror("os::init_2 getrlimit failed");
3693     } else {
3694       nbr_files.rlim_cur = nbr_files.rlim_max;
3695 
3696 #ifdef __APPLE__
3697       // Darwin returns RLIM_INFINITY for rlim_max, but fails with EINVAL if
3698       // you attempt to use RLIM_INFINITY. As per setrlimit(2), OPEN_MAX must
3699       // be used instead
3700       nbr_files.rlim_cur = MIN(OPEN_MAX, nbr_files.rlim_cur);
3701 #endif
3702 
3703       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3704       if (status != 0) {
3705         if (PrintMiscellaneous && (Verbose || WizardMode))
3706           perror("os::init_2 setrlimit failed");
3707       }
3708     }
3709   }
3710 
3711   // at-exit methods are called in the reverse order of their registration.
3712   // atexit functions are called on return from main or as a result of a
3713   // call to exit(3C). There can be only 32 of these functions registered
3714   // and atexit() does not set errno.
3715 
3716   if (PerfAllowAtExitRegistration) {
3717     // only register atexit functions if PerfAllowAtExitRegistration is set.
3718     // atexit functions can be delayed until process exit time, which
3719     // can be problematic for embedded VM situations. Embedded VMs should
3720     // call DestroyJavaVM() to assure that VM resources are released.
3721 
3722     // note: perfMemory_exit_helper atexit function may be removed in
3723     // the future if the appropriate cleanup code can be added to the
3724     // VM_Exit VMOperation's doit method.
3725     if (atexit(perfMemory_exit_helper) != 0) {
3726       warning("os::init2 atexit(perfMemory_exit_helper) failed");
3727     }
3728   }
3729 
3730   // initialize thread priority policy
3731   prio_init();
3732 
3733 #ifdef __APPLE__
3734   // dynamically link to objective c gc registration
3735   void *handleLibObjc = dlopen(OBJC_LIB, RTLD_LAZY);
3736   if (handleLibObjc != NULL) {
3737     objc_registerThreadWithCollectorFunction = (objc_registerThreadWithCollector_t) dlsym(handleLibObjc, OBJC_GCREGISTER);
3738   }
3739 #endif
3740 
3741   return JNI_OK;
3742 }
3743 
3744 // this is called at the end of vm_initialization
3745 void os::init_3(void) { }
3746 
3747 // Mark the polling page as unreadable
3748 void os::make_polling_page_unreadable(void) {
3749   if( !guard_memory((char*)_polling_page, Bsd::page_size()) )
3750     fatal("Could not disable polling page");
3751 };
3752 
3753 // Mark the polling page as readable
3754 void os::make_polling_page_readable(void) {
3755   if( !bsd_mprotect((char *)_polling_page, Bsd::page_size(), PROT_READ)) {
3756     fatal("Could not enable polling page");
3757   }
3758 };
3759 
3760 int os::active_processor_count() {
3761   return _processor_count;
3762 }
3763 
3764 void os::set_native_thread_name(const char *name) {
3765 #if defined(__APPLE__) && MAC_OS_X_VERSION_MIN_REQUIRED > MAC_OS_X_VERSION_10_5
3766   // This is only supported in Snow Leopard and beyond
3767   if (name != NULL) {
3768     // Add a "Java: " prefix to the name
3769     char buf[MAXTHREADNAMESIZE];
3770     snprintf(buf, sizeof(buf), "Java: %s", name);
3771     pthread_setname_np(buf);
3772   }
3773 #endif
3774 }
3775 
3776 bool os::distribute_processes(uint length, uint* distribution) {
3777   // Not yet implemented.
3778   return false;
3779 }
3780 
3781 bool os::bind_to_processor(uint processor_id) {
3782   // Not yet implemented.
3783   return false;
3784 }
3785 
3786 void os::SuspendedThreadTask::internal_do_task() {
3787   if (do_suspend(_thread->osthread())) {
3788     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
3789     do_task(context);
3790     do_resume(_thread->osthread());
3791   }
3792 }
3793 
3794 ///
3795 class PcFetcher : public os::SuspendedThreadTask {
3796 public:
3797   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
3798   ExtendedPC result();
3799 protected:
3800   void do_task(const os::SuspendedThreadTaskContext& context);
3801 private:
3802   ExtendedPC _epc;
3803 };
3804 
3805 ExtendedPC PcFetcher::result() {
3806   guarantee(is_done(), "task is not done yet.");
3807   return _epc;
3808 }
3809 
3810 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
3811   Thread* thread = context.thread();
3812   OSThread* osthread = thread->osthread();
3813   if (osthread->ucontext() != NULL) {
3814     _epc = os::Bsd::ucontext_get_pc((ucontext_t *) context.ucontext());
3815   } else {
3816     // NULL context is unexpected, double-check this is the VMThread
3817     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3818   }
3819 }
3820 
3821 // Suspends the target using the signal mechanism and then grabs the PC before
3822 // resuming the target. Used by the flat-profiler only
3823 ExtendedPC os::get_thread_pc(Thread* thread) {
3824   // Make sure that it is called by the watcher for the VMThread
3825   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
3826   assert(thread->is_VM_thread(), "Can only be called for VMThread");
3827 
3828   PcFetcher fetcher(thread);
3829   fetcher.run();
3830   return fetcher.result();
3831 }
3832 
3833 int os::Bsd::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
3834 {
3835   return pthread_cond_timedwait(_cond, _mutex, _abstime);
3836 }
3837 
3838 ////////////////////////////////////////////////////////////////////////////////
3839 // debug support
3840 
3841 bool os::find(address addr, outputStream* st) {
3842   Dl_info dlinfo;
3843   memset(&dlinfo, 0, sizeof(dlinfo));
3844   if (dladdr(addr, &dlinfo) != 0) {
3845     st->print(PTR_FORMAT ": ", addr);
3846     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
3847       st->print("%s+%#x", dlinfo.dli_sname,
3848                  addr - (intptr_t)dlinfo.dli_saddr);
3849     } else if (dlinfo.dli_fbase != NULL) {
3850       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
3851     } else {
3852       st->print("<absolute address>");
3853     }
3854     if (dlinfo.dli_fname != NULL) {
3855       st->print(" in %s", dlinfo.dli_fname);
3856     }
3857     if (dlinfo.dli_fbase != NULL) {
3858       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
3859     }
3860     st->cr();
3861 
3862     if (Verbose) {
3863       // decode some bytes around the PC
3864       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
3865       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
3866       address       lowest = (address) dlinfo.dli_sname;
3867       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
3868       if (begin < lowest)  begin = lowest;
3869       Dl_info dlinfo2;
3870       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
3871           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
3872         end = (address) dlinfo2.dli_saddr;
3873       Disassembler::decode(begin, end, st);
3874     }
3875     return true;
3876   }
3877   return false;
3878 }
3879 
3880 ////////////////////////////////////////////////////////////////////////////////
3881 // misc
3882 
3883 // This does not do anything on Bsd. This is basically a hook for being
3884 // able to use structured exception handling (thread-local exception filters)
3885 // on, e.g., Win32.
3886 void
3887 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
3888                          JavaCallArguments* args, Thread* thread) {
3889   f(value, method, args, thread);
3890 }
3891 
3892 void os::print_statistics() {
3893 }
3894 
3895 int os::message_box(const char* title, const char* message) {
3896   int i;
3897   fdStream err(defaultStream::error_fd());
3898   for (i = 0; i < 78; i++) err.print_raw("=");
3899   err.cr();
3900   err.print_raw_cr(title);
3901   for (i = 0; i < 78; i++) err.print_raw("-");
3902   err.cr();
3903   err.print_raw_cr(message);
3904   for (i = 0; i < 78; i++) err.print_raw("=");
3905   err.cr();
3906 
3907   char buf[16];
3908   // Prevent process from exiting upon "read error" without consuming all CPU
3909   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3910 
3911   return buf[0] == 'y' || buf[0] == 'Y';
3912 }
3913 
3914 int os::stat(const char *path, struct stat *sbuf) {
3915   char pathbuf[MAX_PATH];
3916   if (strlen(path) > MAX_PATH - 1) {
3917     errno = ENAMETOOLONG;
3918     return -1;
3919   }
3920   os::native_path(strcpy(pathbuf, path));
3921   return ::stat(pathbuf, sbuf);
3922 }
3923 
3924 bool os::check_heap(bool force) {
3925   return true;
3926 }
3927 
3928 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
3929   return ::vsnprintf(buf, count, format, args);
3930 }
3931 
3932 // Is a (classpath) directory empty?
3933 bool os::dir_is_empty(const char* path) {
3934   DIR *dir = NULL;
3935   struct dirent *ptr;
3936 
3937   dir = opendir(path);
3938   if (dir == NULL) return true;
3939 
3940   /* Scan the directory */
3941   bool result = true;
3942   char buf[sizeof(struct dirent) + MAX_PATH];
3943   while (result && (ptr = ::readdir(dir)) != NULL) {
3944     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
3945       result = false;
3946     }
3947   }
3948   closedir(dir);
3949   return result;
3950 }
3951 
3952 // This code originates from JDK's sysOpen and open64_w
3953 // from src/solaris/hpi/src/system_md.c
3954 
3955 #ifndef O_DELETE
3956 #define O_DELETE 0x10000
3957 #endif
3958 
3959 // Open a file. Unlink the file immediately after open returns
3960 // if the specified oflag has the O_DELETE flag set.
3961 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
3962 
3963 int os::open(const char *path, int oflag, int mode) {
3964 
3965   if (strlen(path) > MAX_PATH - 1) {
3966     errno = ENAMETOOLONG;
3967     return -1;
3968   }
3969   int fd;
3970   int o_delete = (oflag & O_DELETE);
3971   oflag = oflag & ~O_DELETE;
3972 
3973   fd = ::open(path, oflag, mode);
3974   if (fd == -1) return -1;
3975 
3976   //If the open succeeded, the file might still be a directory
3977   {
3978     struct stat buf;
3979     int ret = ::fstat(fd, &buf);
3980     int st_mode = buf.st_mode;
3981 
3982     if (ret != -1) {
3983       if ((st_mode & S_IFMT) == S_IFDIR) {
3984         errno = EISDIR;
3985         ::close(fd);
3986         return -1;
3987       }
3988     } else {
3989       ::close(fd);
3990       return -1;
3991     }
3992   }
3993 
3994     /*
3995      * All file descriptors that are opened in the JVM and not
3996      * specifically destined for a subprocess should have the
3997      * close-on-exec flag set.  If we don't set it, then careless 3rd
3998      * party native code might fork and exec without closing all
3999      * appropriate file descriptors (e.g. as we do in closeDescriptors in
4000      * UNIXProcess.c), and this in turn might:
4001      *
4002      * - cause end-of-file to fail to be detected on some file
4003      *   descriptors, resulting in mysterious hangs, or
4004      *
4005      * - might cause an fopen in the subprocess to fail on a system
4006      *   suffering from bug 1085341.
4007      *
4008      * (Yes, the default setting of the close-on-exec flag is a Unix
4009      * design flaw)
4010      *
4011      * See:
4012      * 1085341: 32-bit stdio routines should support file descriptors >255
4013      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4014      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4015      */
4016 #ifdef FD_CLOEXEC
4017     {
4018         int flags = ::fcntl(fd, F_GETFD);
4019         if (flags != -1)
4020             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4021     }
4022 #endif
4023 
4024   if (o_delete != 0) {
4025     ::unlink(path);
4026   }
4027   return fd;
4028 }
4029 
4030 
4031 // create binary file, rewriting existing file if required
4032 int os::create_binary_file(const char* path, bool rewrite_existing) {
4033   int oflags = O_WRONLY | O_CREAT;
4034   if (!rewrite_existing) {
4035     oflags |= O_EXCL;
4036   }
4037   return ::open(path, oflags, S_IREAD | S_IWRITE);
4038 }
4039 
4040 // return current position of file pointer
4041 jlong os::current_file_offset(int fd) {
4042   return (jlong)::lseek(fd, (off_t)0, SEEK_CUR);
4043 }
4044 
4045 // move file pointer to the specified offset
4046 jlong os::seek_to_file_offset(int fd, jlong offset) {
4047   return (jlong)::lseek(fd, (off_t)offset, SEEK_SET);
4048 }
4049 
4050 // This code originates from JDK's sysAvailable
4051 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4052 
4053 int os::available(int fd, jlong *bytes) {
4054   jlong cur, end;
4055   int mode;
4056   struct stat buf;
4057 
4058   if (::fstat(fd, &buf) >= 0) {
4059     mode = buf.st_mode;
4060     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4061       /*
4062       * XXX: is the following call interruptible? If so, this might
4063       * need to go through the INTERRUPT_IO() wrapper as for other
4064       * blocking, interruptible calls in this file.
4065       */
4066       int n;
4067       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4068         *bytes = n;
4069         return 1;
4070       }
4071     }
4072   }
4073   if ((cur = ::lseek(fd, 0L, SEEK_CUR)) == -1) {
4074     return 0;
4075   } else if ((end = ::lseek(fd, 0L, SEEK_END)) == -1) {
4076     return 0;
4077   } else if (::lseek(fd, cur, SEEK_SET) == -1) {
4078     return 0;
4079   }
4080   *bytes = end - cur;
4081   return 1;
4082 }
4083 
4084 int os::socket_available(int fd, jint *pbytes) {
4085    if (fd < 0)
4086      return OS_OK;
4087 
4088    int ret;
4089 
4090    RESTARTABLE(::ioctl(fd, FIONREAD, pbytes), ret);
4091 
4092    //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4093    // is expected to return 0 on failure and 1 on success to the jdk.
4094 
4095    return (ret == OS_ERR) ? 0 : 1;
4096 }
4097 
4098 // Map a block of memory.
4099 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4100                      char *addr, size_t bytes, bool read_only,
4101                      bool allow_exec) {
4102   int prot;
4103   int flags;
4104 
4105   if (read_only) {
4106     prot = PROT_READ;
4107     flags = MAP_SHARED;
4108   } else {
4109     prot = PROT_READ | PROT_WRITE;
4110     flags = MAP_PRIVATE;
4111   }
4112 
4113   if (allow_exec) {
4114     prot |= PROT_EXEC;
4115   }
4116 
4117   if (addr != NULL) {
4118     flags |= MAP_FIXED;
4119   }
4120 
4121   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4122                                      fd, file_offset);
4123   if (mapped_address == MAP_FAILED) {
4124     return NULL;
4125   }
4126   return mapped_address;
4127 }
4128 
4129 
4130 // Remap a block of memory.
4131 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4132                        char *addr, size_t bytes, bool read_only,
4133                        bool allow_exec) {
4134   // same as map_memory() on this OS
4135   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4136                         allow_exec);
4137 }
4138 
4139 
4140 // Unmap a block of memory.
4141 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4142   return munmap(addr, bytes) == 0;
4143 }
4144 
4145 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4146 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4147 // of a thread.
4148 //
4149 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4150 // the fast estimate available on the platform.
4151 
4152 jlong os::current_thread_cpu_time() {
4153 #ifdef __APPLE__
4154   return os::thread_cpu_time(Thread::current(), true /* user + sys */);
4155 #else
4156   Unimplemented();
4157   return 0;
4158 #endif
4159 }
4160 
4161 jlong os::thread_cpu_time(Thread* thread) {
4162 #ifdef __APPLE__
4163   return os::thread_cpu_time(thread, true /* user + sys */);
4164 #else
4165   Unimplemented();
4166   return 0;
4167 #endif
4168 }
4169 
4170 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4171 #ifdef __APPLE__
4172   return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
4173 #else
4174   Unimplemented();
4175   return 0;
4176 #endif
4177 }
4178 
4179 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4180 #ifdef __APPLE__
4181   struct thread_basic_info tinfo;
4182   mach_msg_type_number_t tcount = THREAD_INFO_MAX;
4183   kern_return_t kr;
4184   thread_t mach_thread;
4185 
4186   mach_thread = thread->osthread()->thread_id();
4187   kr = thread_info(mach_thread, THREAD_BASIC_INFO, (thread_info_t)&tinfo, &tcount);
4188   if (kr != KERN_SUCCESS)
4189     return -1;
4190 
4191   if (user_sys_cpu_time) {
4192     jlong nanos;
4193     nanos = ((jlong) tinfo.system_time.seconds + tinfo.user_time.seconds) * (jlong)1000000000;
4194     nanos += ((jlong) tinfo.system_time.microseconds + (jlong) tinfo.user_time.microseconds) * (jlong)1000;
4195     return nanos;
4196   } else {
4197     return ((jlong)tinfo.user_time.seconds * 1000000000) + ((jlong)tinfo.user_time.microseconds * (jlong)1000);
4198   }
4199 #else
4200   Unimplemented();
4201   return 0;
4202 #endif
4203 }
4204 
4205 
4206 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4207   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4208   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4209   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4210   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4211 }
4212 
4213 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4214   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4215   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4216   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4217   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4218 }
4219 
4220 bool os::is_thread_cpu_time_supported() {
4221 #ifdef __APPLE__
4222   return true;
4223 #else
4224   return false;
4225 #endif
4226 }
4227 
4228 // System loadavg support.  Returns -1 if load average cannot be obtained.
4229 // Bsd doesn't yet have a (official) notion of processor sets,
4230 // so just return the system wide load average.
4231 int os::loadavg(double loadavg[], int nelem) {
4232   return ::getloadavg(loadavg, nelem);
4233 }
4234 
4235 void os::pause() {
4236   char filename[MAX_PATH];
4237   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4238     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4239   } else {
4240     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4241   }
4242 
4243   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4244   if (fd != -1) {
4245     struct stat buf;
4246     ::close(fd);
4247     while (::stat(filename, &buf) == 0) {
4248       (void)::poll(NULL, 0, 100);
4249     }
4250   } else {
4251     jio_fprintf(stderr,
4252       "Could not open pause file '%s', continuing immediately.\n", filename);
4253   }
4254 }
4255 
4256 
4257 // Refer to the comments in os_solaris.cpp park-unpark.
4258 //
4259 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4260 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4261 // For specifics regarding the bug see GLIBC BUGID 261237 :
4262 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4263 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4264 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4265 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4266 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4267 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4268 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4269 // of libpthread avoids the problem, but isn't practical.
4270 //
4271 // Possible remedies:
4272 //
4273 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4274 //      This is palliative and probabilistic, however.  If the thread is preempted
4275 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4276 //      than the minimum period may have passed, and the abstime may be stale (in the
4277 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
4278 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4279 //
4280 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4281 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
4282 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4283 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
4284 //      thread.
4285 //
4286 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4287 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
4288 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4289 //      This also works well.  In fact it avoids kernel-level scalability impediments
4290 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4291 //      timers in a graceful fashion.
4292 //
4293 // 4.   When the abstime value is in the past it appears that control returns
4294 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4295 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
4296 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
4297 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
4298 //      It may be possible to avoid reinitialization by checking the return
4299 //      value from pthread_cond_timedwait().  In addition to reinitializing the
4300 //      condvar we must establish the invariant that cond_signal() is only called
4301 //      within critical sections protected by the adjunct mutex.  This prevents
4302 //      cond_signal() from "seeing" a condvar that's in the midst of being
4303 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
4304 //      desirable signal-after-unlock optimization that avoids futile context switching.
4305 //
4306 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
4307 //      structure when a condvar is used or initialized.  cond_destroy()  would
4308 //      release the helper structure.  Our reinitialize-after-timedwait fix
4309 //      put excessive stress on malloc/free and locks protecting the c-heap.
4310 //
4311 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
4312 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4313 // and only enabling the work-around for vulnerable environments.
4314 
4315 // utility to compute the abstime argument to timedwait:
4316 // millis is the relative timeout time
4317 // abstime will be the absolute timeout time
4318 // TODO: replace compute_abstime() with unpackTime()
4319 
4320 static struct timespec* compute_abstime(struct timespec* abstime, jlong millis) {
4321   if (millis < 0)  millis = 0;
4322   struct timeval now;
4323   int status = gettimeofday(&now, NULL);
4324   assert(status == 0, "gettimeofday");
4325   jlong seconds = millis / 1000;
4326   millis %= 1000;
4327   if (seconds > 50000000) { // see man cond_timedwait(3T)
4328     seconds = 50000000;
4329   }
4330   abstime->tv_sec = now.tv_sec  + seconds;
4331   long       usec = now.tv_usec + millis * 1000;
4332   if (usec >= 1000000) {
4333     abstime->tv_sec += 1;
4334     usec -= 1000000;
4335   }
4336   abstime->tv_nsec = usec * 1000;
4337   return abstime;
4338 }
4339 
4340 
4341 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4342 // Conceptually TryPark() should be equivalent to park(0).
4343 
4344 int os::PlatformEvent::TryPark() {
4345   for (;;) {
4346     const int v = _Event ;
4347     guarantee ((v == 0) || (v == 1), "invariant") ;
4348     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
4349   }
4350 }
4351 
4352 void os::PlatformEvent::park() {       // AKA "down()"
4353   // Invariant: Only the thread associated with the Event/PlatformEvent
4354   // may call park().
4355   // TODO: assert that _Assoc != NULL or _Assoc == Self
4356   int v ;
4357   for (;;) {
4358       v = _Event ;
4359       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4360   }
4361   guarantee (v >= 0, "invariant") ;
4362   if (v == 0) {
4363      // Do this the hard way by blocking ...
4364      int status = pthread_mutex_lock(_mutex);
4365      assert_status(status == 0, status, "mutex_lock");
4366      guarantee (_nParked == 0, "invariant") ;
4367      ++ _nParked ;
4368      while (_Event < 0) {
4369         status = pthread_cond_wait(_cond, _mutex);
4370         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4371         // Treat this the same as if the wait was interrupted
4372         if (status == ETIMEDOUT) { status = EINTR; }
4373         assert_status(status == 0 || status == EINTR, status, "cond_wait");
4374      }
4375      -- _nParked ;
4376 
4377     _Event = 0 ;
4378      status = pthread_mutex_unlock(_mutex);
4379      assert_status(status == 0, status, "mutex_unlock");
4380     // Paranoia to ensure our locked and lock-free paths interact
4381     // correctly with each other.
4382     OrderAccess::fence();
4383   }
4384   guarantee (_Event >= 0, "invariant") ;
4385 }
4386 
4387 int os::PlatformEvent::park(jlong millis) {
4388   guarantee (_nParked == 0, "invariant") ;
4389 
4390   int v ;
4391   for (;;) {
4392       v = _Event ;
4393       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4394   }
4395   guarantee (v >= 0, "invariant") ;
4396   if (v != 0) return OS_OK ;
4397 
4398   // We do this the hard way, by blocking the thread.
4399   // Consider enforcing a minimum timeout value.
4400   struct timespec abst;
4401   compute_abstime(&abst, millis);
4402 
4403   int ret = OS_TIMEOUT;
4404   int status = pthread_mutex_lock(_mutex);
4405   assert_status(status == 0, status, "mutex_lock");
4406   guarantee (_nParked == 0, "invariant") ;
4407   ++_nParked ;
4408 
4409   // Object.wait(timo) will return because of
4410   // (a) notification
4411   // (b) timeout
4412   // (c) thread.interrupt
4413   //
4414   // Thread.interrupt and object.notify{All} both call Event::set.
4415   // That is, we treat thread.interrupt as a special case of notification.
4416   // The underlying Solaris implementation, cond_timedwait, admits
4417   // spurious/premature wakeups, but the JLS/JVM spec prevents the
4418   // JVM from making those visible to Java code.  As such, we must
4419   // filter out spurious wakeups.  We assume all ETIME returns are valid.
4420   //
4421   // TODO: properly differentiate simultaneous notify+interrupt.
4422   // In that case, we should propagate the notify to another waiter.
4423 
4424   while (_Event < 0) {
4425     status = os::Bsd::safe_cond_timedwait(_cond, _mutex, &abst);
4426     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4427       pthread_cond_destroy (_cond);
4428       pthread_cond_init (_cond, NULL) ;
4429     }
4430     assert_status(status == 0 || status == EINTR ||
4431                   status == ETIMEDOUT,
4432                   status, "cond_timedwait");
4433     if (!FilterSpuriousWakeups) break ;                 // previous semantics
4434     if (status == ETIMEDOUT) break ;
4435     // We consume and ignore EINTR and spurious wakeups.
4436   }
4437   --_nParked ;
4438   if (_Event >= 0) {
4439      ret = OS_OK;
4440   }
4441   _Event = 0 ;
4442   status = pthread_mutex_unlock(_mutex);
4443   assert_status(status == 0, status, "mutex_unlock");
4444   assert (_nParked == 0, "invariant") ;
4445   // Paranoia to ensure our locked and lock-free paths interact
4446   // correctly with each other.
4447   OrderAccess::fence();
4448   return ret;
4449 }
4450 
4451 void os::PlatformEvent::unpark() {
4452   // Transitions for _Event:
4453   //    0 :=> 1
4454   //    1 :=> 1
4455   //   -1 :=> either 0 or 1; must signal target thread
4456   //          That is, we can safely transition _Event from -1 to either
4457   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
4458   //          unpark() calls.
4459   // See also: "Semaphores in Plan 9" by Mullender & Cox
4460   //
4461   // Note: Forcing a transition from "-1" to "1" on an unpark() means
4462   // that it will take two back-to-back park() calls for the owning
4463   // thread to block. This has the benefit of forcing a spurious return
4464   // from the first park() call after an unpark() call which will help
4465   // shake out uses of park() and unpark() without condition variables.
4466 
4467   if (Atomic::xchg(1, &_Event) >= 0) return;
4468 
4469   // Wait for the thread associated with the event to vacate
4470   int status = pthread_mutex_lock(_mutex);
4471   assert_status(status == 0, status, "mutex_lock");
4472   int AnyWaiters = _nParked;
4473   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
4474   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4475     AnyWaiters = 0;
4476     pthread_cond_signal(_cond);
4477   }
4478   status = pthread_mutex_unlock(_mutex);
4479   assert_status(status == 0, status, "mutex_unlock");
4480   if (AnyWaiters != 0) {
4481     status = pthread_cond_signal(_cond);
4482     assert_status(status == 0, status, "cond_signal");
4483   }
4484 
4485   // Note that we signal() _after dropping the lock for "immortal" Events.
4486   // This is safe and avoids a common class of  futile wakeups.  In rare
4487   // circumstances this can cause a thread to return prematurely from
4488   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4489   // simply re-test the condition and re-park itself.
4490 }
4491 
4492 
4493 // JSR166
4494 // -------------------------------------------------------
4495 
4496 /*
4497  * The solaris and bsd implementations of park/unpark are fairly
4498  * conservative for now, but can be improved. They currently use a
4499  * mutex/condvar pair, plus a a count.
4500  * Park decrements count if > 0, else does a condvar wait.  Unpark
4501  * sets count to 1 and signals condvar.  Only one thread ever waits
4502  * on the condvar. Contention seen when trying to park implies that someone
4503  * is unparking you, so don't wait. And spurious returns are fine, so there
4504  * is no need to track notifications.
4505  */
4506 
4507 #define MAX_SECS 100000000
4508 /*
4509  * This code is common to bsd and solaris and will be moved to a
4510  * common place in dolphin.
4511  *
4512  * The passed in time value is either a relative time in nanoseconds
4513  * or an absolute time in milliseconds. Either way it has to be unpacked
4514  * into suitable seconds and nanoseconds components and stored in the
4515  * given timespec structure.
4516  * Given time is a 64-bit value and the time_t used in the timespec is only
4517  * a signed-32-bit value (except on 64-bit Bsd) we have to watch for
4518  * overflow if times way in the future are given. Further on Solaris versions
4519  * prior to 10 there is a restriction (see cond_timedwait) that the specified
4520  * number of seconds, in abstime, is less than current_time  + 100,000,000.
4521  * As it will be 28 years before "now + 100000000" will overflow we can
4522  * ignore overflow and just impose a hard-limit on seconds using the value
4523  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4524  * years from "now".
4525  */
4526 
4527 static void unpackTime(struct timespec* absTime, bool isAbsolute, jlong time) {
4528   assert (time > 0, "convertTime");
4529 
4530   struct timeval now;
4531   int status = gettimeofday(&now, NULL);
4532   assert(status == 0, "gettimeofday");
4533 
4534   time_t max_secs = now.tv_sec + MAX_SECS;
4535 
4536   if (isAbsolute) {
4537     jlong secs = time / 1000;
4538     if (secs > max_secs) {
4539       absTime->tv_sec = max_secs;
4540     }
4541     else {
4542       absTime->tv_sec = secs;
4543     }
4544     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4545   }
4546   else {
4547     jlong secs = time / NANOSECS_PER_SEC;
4548     if (secs >= MAX_SECS) {
4549       absTime->tv_sec = max_secs;
4550       absTime->tv_nsec = 0;
4551     }
4552     else {
4553       absTime->tv_sec = now.tv_sec + secs;
4554       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4555       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4556         absTime->tv_nsec -= NANOSECS_PER_SEC;
4557         ++absTime->tv_sec; // note: this must be <= max_secs
4558       }
4559     }
4560   }
4561   assert(absTime->tv_sec >= 0, "tv_sec < 0");
4562   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4563   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4564   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4565 }
4566 
4567 void Parker::park(bool isAbsolute, jlong time) {
4568   // Ideally we'd do something useful while spinning, such
4569   // as calling unpackTime().
4570 
4571   // Optional fast-path check:
4572   // Return immediately if a permit is available.
4573   // We depend on Atomic::xchg() having full barrier semantics
4574   // since we are doing a lock-free update to _counter.
4575   if (Atomic::xchg(0, &_counter) > 0) return;
4576 
4577   Thread* thread = Thread::current();
4578   assert(thread->is_Java_thread(), "Must be JavaThread");
4579   JavaThread *jt = (JavaThread *)thread;
4580 
4581   // Optional optimization -- avoid state transitions if there's an interrupt pending.
4582   // Check interrupt before trying to wait
4583   if (Thread::is_interrupted(thread, false)) {
4584     return;
4585   }
4586 
4587   // Next, demultiplex/decode time arguments
4588   struct timespec absTime;
4589   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
4590     return;
4591   }
4592   if (time > 0) {
4593     unpackTime(&absTime, isAbsolute, time);
4594   }
4595 
4596 
4597   // Enter safepoint region
4598   // Beware of deadlocks such as 6317397.
4599   // The per-thread Parker:: mutex is a classic leaf-lock.
4600   // In particular a thread must never block on the Threads_lock while
4601   // holding the Parker:: mutex.  If safepoints are pending both the
4602   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4603   ThreadBlockInVM tbivm(jt);
4604 
4605   // Don't wait if cannot get lock since interference arises from
4606   // unblocking.  Also. check interrupt before trying wait
4607   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4608     return;
4609   }
4610 
4611   int status ;
4612   if (_counter > 0)  { // no wait needed
4613     _counter = 0;
4614     status = pthread_mutex_unlock(_mutex);
4615     assert (status == 0, "invariant") ;
4616     // Paranoia to ensure our locked and lock-free paths interact
4617     // correctly with each other and Java-level accesses.
4618     OrderAccess::fence();
4619     return;
4620   }
4621 
4622 #ifdef ASSERT
4623   // Don't catch signals while blocked; let the running threads have the signals.
4624   // (This allows a debugger to break into the running thread.)
4625   sigset_t oldsigs;
4626   sigset_t* allowdebug_blocked = os::Bsd::allowdebug_blocked_signals();
4627   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4628 #endif
4629 
4630   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4631   jt->set_suspend_equivalent();
4632   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4633 
4634   if (time == 0) {
4635     status = pthread_cond_wait (_cond, _mutex) ;
4636   } else {
4637     status = os::Bsd::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4638     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4639       pthread_cond_destroy (_cond) ;
4640       pthread_cond_init    (_cond, NULL);
4641     }
4642   }
4643   assert_status(status == 0 || status == EINTR ||
4644                 status == ETIMEDOUT,
4645                 status, "cond_timedwait");
4646 
4647 #ifdef ASSERT
4648   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4649 #endif
4650 
4651   _counter = 0 ;
4652   status = pthread_mutex_unlock(_mutex) ;
4653   assert_status(status == 0, status, "invariant") ;
4654   // Paranoia to ensure our locked and lock-free paths interact
4655   // correctly with each other and Java-level accesses.
4656   OrderAccess::fence();
4657 
4658   // If externally suspended while waiting, re-suspend
4659   if (jt->handle_special_suspend_equivalent_condition()) {
4660     jt->java_suspend_self();
4661   }
4662 }
4663 
4664 void Parker::unpark() {
4665   int s, status ;
4666   status = pthread_mutex_lock(_mutex);
4667   assert (status == 0, "invariant") ;
4668   s = _counter;
4669   _counter = 1;
4670   if (s < 1) {
4671      if (WorkAroundNPTLTimedWaitHang) {
4672         status = pthread_cond_signal (_cond) ;
4673         assert (status == 0, "invariant") ;
4674         status = pthread_mutex_unlock(_mutex);
4675         assert (status == 0, "invariant") ;
4676      } else {
4677         status = pthread_mutex_unlock(_mutex);
4678         assert (status == 0, "invariant") ;
4679         status = pthread_cond_signal (_cond) ;
4680         assert (status == 0, "invariant") ;
4681      }
4682   } else {
4683     pthread_mutex_unlock(_mutex);
4684     assert (status == 0, "invariant") ;
4685   }
4686 }
4687 
4688 
4689 /* Darwin has no "environ" in a dynamic library. */
4690 #ifdef __APPLE__
4691 #include <crt_externs.h>
4692 #define environ (*_NSGetEnviron())
4693 #else
4694 extern char** environ;
4695 #endif
4696 
4697 // Run the specified command in a separate process. Return its exit value,
4698 // or -1 on failure (e.g. can't fork a new process).
4699 // Unlike system(), this function can be called from signal handler. It
4700 // doesn't block SIGINT et al.
4701 int os::fork_and_exec(char* cmd) {
4702   const char * argv[4] = {"sh", "-c", cmd, NULL};
4703 
4704   // fork() in BsdThreads/NPTL is not async-safe. It needs to run
4705   // pthread_atfork handlers and reset pthread library. All we need is a
4706   // separate process to execve. Make a direct syscall to fork process.
4707   // On IA64 there's no fork syscall, we have to use fork() and hope for
4708   // the best...
4709   pid_t pid = fork();
4710 
4711   if (pid < 0) {
4712     // fork failed
4713     return -1;
4714 
4715   } else if (pid == 0) {
4716     // child process
4717 
4718     // execve() in BsdThreads will call pthread_kill_other_threads_np()
4719     // first to kill every thread on the thread list. Because this list is
4720     // not reset by fork() (see notes above), execve() will instead kill
4721     // every thread in the parent process. We know this is the only thread
4722     // in the new process, so make a system call directly.
4723     // IA64 should use normal execve() from glibc to match the glibc fork()
4724     // above.
4725     execve("/bin/sh", (char* const*)argv, environ);
4726 
4727     // execve failed
4728     _exit(-1);
4729 
4730   } else  {
4731     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4732     // care about the actual exit code, for now.
4733 
4734     int status;
4735 
4736     // Wait for the child process to exit.  This returns immediately if
4737     // the child has already exited. */
4738     while (waitpid(pid, &status, 0) < 0) {
4739         switch (errno) {
4740         case ECHILD: return 0;
4741         case EINTR: break;
4742         default: return -1;
4743         }
4744     }
4745 
4746     if (WIFEXITED(status)) {
4747        // The child exited normally; get its exit code.
4748        return WEXITSTATUS(status);
4749     } else if (WIFSIGNALED(status)) {
4750        // The child exited because of a signal
4751        // The best value to return is 0x80 + signal number,
4752        // because that is what all Unix shells do, and because
4753        // it allows callers to distinguish between process exit and
4754        // process death by signal.
4755        return 0x80 + WTERMSIG(status);
4756     } else {
4757        // Unknown exit code; pass it through
4758        return status;
4759     }
4760   }
4761 }
4762 
4763 // is_headless_jre()
4764 //
4765 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
4766 // in order to report if we are running in a headless jre
4767 //
4768 // Since JDK8 xawt/libmawt.so was moved into the same directory
4769 // as libawt.so, and renamed libawt_xawt.so
4770 //
4771 bool os::is_headless_jre() {
4772 #ifdef __APPLE__
4773     // We no longer build headless-only on Mac OS X
4774     return false;
4775 #else
4776     struct stat statbuf;
4777     char buf[MAXPATHLEN];
4778     char libmawtpath[MAXPATHLEN];
4779     const char *xawtstr  = "/xawt/libmawt" JNI_LIB_SUFFIX;
4780     const char *new_xawtstr = "/libawt_xawt" JNI_LIB_SUFFIX;
4781     char *p;
4782 
4783     // Get path to libjvm.so
4784     os::jvm_path(buf, sizeof(buf));
4785 
4786     // Get rid of libjvm.so
4787     p = strrchr(buf, '/');
4788     if (p == NULL) return false;
4789     else *p = '\0';
4790 
4791     // Get rid of client or server
4792     p = strrchr(buf, '/');
4793     if (p == NULL) return false;
4794     else *p = '\0';
4795 
4796     // check xawt/libmawt.so
4797     strcpy(libmawtpath, buf);
4798     strcat(libmawtpath, xawtstr);
4799     if (::stat(libmawtpath, &statbuf) == 0) return false;
4800 
4801     // check libawt_xawt.so
4802     strcpy(libmawtpath, buf);
4803     strcat(libmawtpath, new_xawtstr);
4804     if (::stat(libmawtpath, &statbuf) == 0) return false;
4805 
4806     return true;
4807 #endif
4808 }
4809 
4810 // Get the default path to the core file
4811 // Returns the length of the string
4812 int os::get_core_path(char* buffer, size_t bufferSize) {
4813   int n = jio_snprintf(buffer, bufferSize, "/cores");
4814 
4815   // Truncate if theoretical string was longer than bufferSize
4816   n = MIN2(n, (int)bufferSize);
4817 
4818   return n;
4819 }
4820 
4821 #ifndef PRODUCT
4822 void TestReserveMemorySpecial_test() {
4823   // No tests available for this platform
4824 }
4825 #endif