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_linux.h"
  35 #include "memory/allocation.inline.hpp"
  36 #include "memory/filemap.hpp"
  37 #include "mutex_linux.inline.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_share_linux.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/init.hpp"
  48 #include "runtime/java.hpp"
  49 #include "runtime/javaCalls.hpp"
  50 #include "runtime/mutexLocker.hpp"
  51 #include "runtime/objectMonitor.hpp"
  52 #include "runtime/osThread.hpp"
  53 #include "runtime/perfMemory.hpp"
  54 #include "runtime/sharedRuntime.hpp"
  55 #include "runtime/statSampler.hpp"
  56 #include "runtime/stubRoutines.hpp"
  57 #include "runtime/thread.inline.hpp"
  58 #include "runtime/threadCritical.hpp"
  59 #include "runtime/timer.hpp"
  60 #include "services/attachListener.hpp"
  61 #include "services/memTracker.hpp"
  62 #include "services/runtimeService.hpp"
  63 #include "utilities/decoder.hpp"
  64 #include "utilities/defaultStream.hpp"
  65 #include "utilities/events.hpp"
  66 #include "utilities/elfFile.hpp"
  67 #include "utilities/growableArray.hpp"
  68 #include "utilities/vmError.hpp"
  69 
  70 // put OS-includes here
  71 # include <sys/types.h>
  72 # include <sys/mman.h>
  73 # include <sys/stat.h>
  74 # include <sys/select.h>
  75 # include <pthread.h>
  76 # include <signal.h>
  77 # include <errno.h>
  78 # include <dlfcn.h>
  79 # include <stdio.h>
  80 # include <unistd.h>
  81 # include <sys/resource.h>
  82 # include <pthread.h>
  83 # include <sys/stat.h>
  84 # include <sys/time.h>
  85 # include <sys/times.h>
  86 # include <sys/utsname.h>
  87 # include <sys/socket.h>
  88 # include <sys/wait.h>
  89 # include <pwd.h>
  90 # include <poll.h>
  91 # include <semaphore.h>
  92 # include <fcntl.h>
  93 # include <string.h>
  94 # include <syscall.h>
  95 # include <sys/sysinfo.h>
  96 # include <gnu/libc-version.h>
  97 # include <sys/ipc.h>
  98 # include <sys/shm.h>
  99 # include <link.h>
 100 # include <stdint.h>
 101 # include <inttypes.h>
 102 # include <sys/ioctl.h>
 103 
 104 #define MAX_PATH    (2 * K)
 105 
 106 // for timer info max values which include all bits
 107 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 108 
 109 #define LARGEPAGES_BIT (1 << 6)
 110 ////////////////////////////////////////////////////////////////////////////////
 111 // global variables
 112 julong os::Linux::_physical_memory = 0;
 113 
 114 address   os::Linux::_initial_thread_stack_bottom = NULL;
 115 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 116 
 117 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 118 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 119 Mutex* os::Linux::_createThread_lock = NULL;
 120 pthread_t os::Linux::_main_thread;
 121 int os::Linux::_page_size = -1;
 122 bool os::Linux::_is_floating_stack = false;
 123 bool os::Linux::_is_NPTL = false;
 124 bool os::Linux::_supports_fast_thread_cpu_time = false;
 125 const char * os::Linux::_glibc_version = NULL;
 126 const char * os::Linux::_libpthread_version = NULL;
 127 
 128 static jlong initial_time_count=0;
 129 
 130 static int clock_tics_per_sec = 100;
 131 
 132 // For diagnostics to print a message once. see run_periodic_checks
 133 static sigset_t check_signal_done;
 134 static bool check_signals = true;;
 135 
 136 static pid_t _initial_pid = 0;
 137 
 138 /* Signal number used to suspend/resume a thread */
 139 
 140 /* do not use any signal number less than SIGSEGV, see 4355769 */
 141 static int SR_signum = SIGUSR2;
 142 sigset_t SR_sigset;
 143 
 144 /* Used to protect dlsym() calls */
 145 static pthread_mutex_t dl_mutex;
 146 
 147 #ifdef JAVASE_EMBEDDED
 148 class MemNotifyThread: public Thread {
 149   friend class VMStructs;
 150  public:
 151   virtual void run();
 152 
 153  private:
 154   static MemNotifyThread* _memnotify_thread;
 155   int _fd;
 156 
 157  public:
 158 
 159   // Constructor
 160   MemNotifyThread(int fd);
 161 
 162   // Tester
 163   bool is_memnotify_thread() const { return true; }
 164 
 165   // Printing
 166   char* name() const { return (char*)"Linux MemNotify Thread"; }
 167 
 168   // Returns the single instance of the MemNotifyThread
 169   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 170 
 171   // Create and start the single instance of MemNotifyThread
 172   static void start();
 173 };
 174 #endif // JAVASE_EMBEDDED
 175 
 176 // utility functions
 177 
 178 static int SR_initialize();
 179 
 180 julong os::available_memory() {
 181   return Linux::available_memory();
 182 }
 183 
 184 julong os::Linux::available_memory() {
 185   // values in struct sysinfo are "unsigned long"
 186   struct sysinfo si;
 187   sysinfo(&si);
 188 
 189   return (julong)si.freeram * si.mem_unit;
 190 }
 191 
 192 julong os::physical_memory() {
 193   return Linux::physical_memory();
 194 }
 195 
 196 ////////////////////////////////////////////////////////////////////////////////
 197 // environment support
 198 
 199 bool os::getenv(const char* name, char* buf, int len) {
 200   const char* val = ::getenv(name);
 201   if (val != NULL && strlen(val) < (size_t)len) {
 202     strcpy(buf, val);
 203     return true;
 204   }
 205   if (len > 0) buf[0] = 0;  // return a null string
 206   return false;
 207 }
 208 
 209 
 210 // Return true if user is running as root.
 211 
 212 bool os::have_special_privileges() {
 213   static bool init = false;
 214   static bool privileges = false;
 215   if (!init) {
 216     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 217     init = true;
 218   }
 219   return privileges;
 220 }
 221 
 222 
 223 #ifndef SYS_gettid
 224 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 225 #ifdef __ia64__
 226 #define SYS_gettid 1105
 227 #elif __i386__
 228 #define SYS_gettid 224
 229 #elif __amd64__
 230 #define SYS_gettid 186
 231 #elif __sparc__
 232 #define SYS_gettid 143
 233 #else
 234 #error define gettid for the arch
 235 #endif
 236 #endif
 237 
 238 // Cpu architecture string
 239 #if   defined(ZERO)
 240 static char cpu_arch[] = ZERO_LIBARCH;
 241 #elif defined(IA64)
 242 static char cpu_arch[] = "ia64";
 243 #elif defined(IA32)
 244 static char cpu_arch[] = "i386";
 245 #elif defined(AMD64)
 246 static char cpu_arch[] = "amd64";
 247 #elif defined(ARM)
 248 static char cpu_arch[] = "arm";
 249 #elif defined(PPC)
 250 static char cpu_arch[] = "ppc";
 251 #elif defined(SPARC)
 252 #  ifdef _LP64
 253 static char cpu_arch[] = "sparcv9";
 254 #  else
 255 static char cpu_arch[] = "sparc";
 256 #  endif
 257 #else
 258 #error Add appropriate cpu_arch setting
 259 #endif
 260 
 261 
 262 // pid_t gettid()
 263 //
 264 // Returns the kernel thread id of the currently running thread. Kernel
 265 // thread id is used to access /proc.
 266 //
 267 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 268 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 269 //
 270 pid_t os::Linux::gettid() {
 271   int rslt = syscall(SYS_gettid);
 272   if (rslt == -1) {
 273      // old kernel, no NPTL support
 274      return getpid();
 275   } else {
 276      return (pid_t)rslt;
 277   }
 278 }
 279 
 280 // Most versions of linux have a bug where the number of processors are
 281 // determined by looking at the /proc file system.  In a chroot environment,
 282 // the system call returns 1.  This causes the VM to act as if it is
 283 // a single processor and elide locking (see is_MP() call).
 284 static bool unsafe_chroot_detected = false;
 285 static const char *unstable_chroot_error = "/proc file system not found.\n"
 286                      "Java may be unstable running multithreaded in a chroot "
 287                      "environment on Linux when /proc filesystem is not mounted.";
 288 
 289 void os::Linux::initialize_system_info() {
 290   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 291   if (processor_count() == 1) {
 292     pid_t pid = os::Linux::gettid();
 293     char fname[32];
 294     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 295     FILE *fp = fopen(fname, "r");
 296     if (fp == NULL) {
 297       unsafe_chroot_detected = true;
 298     } else {
 299       fclose(fp);
 300     }
 301   }
 302   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 303   assert(processor_count() > 0, "linux error");
 304 }
 305 
 306 void os::init_system_properties_values() {
 307 //  char arch[12];
 308 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 309 
 310   // The next steps are taken in the product version:
 311   //
 312   // Obtain the JAVA_HOME value from the location of libjvm.so.
 313   // This library should be located at:
 314   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 315   //
 316   // If "/jre/lib/" appears at the right place in the path, then we
 317   // assume libjvm.so is installed in a JDK and we use this path.
 318   //
 319   // Otherwise exit with message: "Could not create the Java virtual machine."
 320   //
 321   // The following extra steps are taken in the debugging version:
 322   //
 323   // If "/jre/lib/" does NOT appear at the right place in the path
 324   // instead of exit check for $JAVA_HOME environment variable.
 325   //
 326   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 327   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 328   // it looks like libjvm.so is installed there
 329   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 330   //
 331   // Otherwise exit.
 332   //
 333   // Important note: if the location of libjvm.so changes this
 334   // code needs to be changed accordingly.
 335 
 336   // The next few definitions allow the code to be verbatim:
 337 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
 338 #define getenv(n) ::getenv(n)
 339 
 340 /*
 341  * See ld(1):
 342  *      The linker uses the following search paths to locate required
 343  *      shared libraries:
 344  *        1: ...
 345  *        ...
 346  *        7: The default directories, normally /lib and /usr/lib.
 347  */
 348 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 349 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 350 #else
 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   {
 359     /* sysclasspath, java_home, dll_dir */
 360     {
 361         char *home_path;
 362         char *dll_path;
 363         char *pslash;
 364         char buf[MAXPATHLEN];
 365         os::jvm_path(buf, sizeof(buf));
 366 
 367         // Found the full path to libjvm.so.
 368         // Now cut the path to <java_home>/jre if we can.
 369         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 370         pslash = strrchr(buf, '/');
 371         if (pslash != NULL)
 372             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 373         dll_path = malloc(strlen(buf) + 1);
 374         if (dll_path == NULL)
 375             return;
 376         strcpy(dll_path, buf);
 377         Arguments::set_dll_dir(dll_path);
 378 
 379         if (pslash != NULL) {
 380             pslash = strrchr(buf, '/');
 381             if (pslash != NULL) {
 382                 *pslash = '\0';       /* get rid of /<arch> */
 383                 pslash = strrchr(buf, '/');
 384                 if (pslash != NULL)
 385                     *pslash = '\0';   /* get rid of /lib */
 386             }
 387         }
 388 
 389         home_path = malloc(strlen(buf) + 1);
 390         if (home_path == NULL)
 391             return;
 392         strcpy(home_path, buf);
 393         Arguments::set_java_home(home_path);
 394 
 395         if (!set_boot_path('/', ':'))
 396             return;
 397     }
 398 
 399     /*
 400      * Where to look for native libraries
 401      *
 402      * Note: Due to a legacy implementation, most of the library path
 403      * is set in the launcher.  This was to accomodate linking restrictions
 404      * on legacy Linux implementations (which are no longer supported).
 405      * Eventually, all the library path setting will be done here.
 406      *
 407      * However, to prevent the proliferation of improperly built native
 408      * libraries, the new path component /usr/java/packages is added here.
 409      * Eventually, all the library path setting will be done here.
 410      */
 411     {
 412         char *ld_library_path;
 413 
 414         /*
 415          * Construct the invariant part of ld_library_path. Note that the
 416          * space for the colon and the trailing null are provided by the
 417          * nulls included by the sizeof operator (so actually we allocate
 418          * a byte more than necessary).
 419          */
 420         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 421             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 422         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 423 
 424         /*
 425          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 426          * should always exist (until the legacy problem cited above is
 427          * addressed).
 428          */
 429         char *v = getenv("LD_LIBRARY_PATH");
 430         if (v != NULL) {
 431             char *t = ld_library_path;
 432             /* That's +1 for the colon and +1 for the trailing '\0' */
 433             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 434             sprintf(ld_library_path, "%s:%s", v, t);
 435         }
 436         Arguments::set_library_path(ld_library_path);
 437     }
 438 
 439     /*
 440      * Extensions directories.
 441      *
 442      * Note that the space for the colon and the trailing null are provided
 443      * by the nulls included by the sizeof operator (so actually one byte more
 444      * than necessary is allocated).
 445      */
 446     {
 447         char *buf = malloc(strlen(Arguments::get_java_home()) +
 448             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 449         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 450             Arguments::get_java_home());
 451         Arguments::set_ext_dirs(buf);
 452     }
 453 
 454     /* Endorsed standards default directory. */
 455     {
 456         char * buf;
 457         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 458         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 459         Arguments::set_endorsed_dirs(buf);
 460     }
 461   }
 462 
 463 #undef malloc
 464 #undef getenv
 465 #undef EXTENSIONS_DIR
 466 #undef ENDORSED_DIR
 467 
 468   // Done
 469   return;
 470 }
 471 
 472 ////////////////////////////////////////////////////////////////////////////////
 473 // breakpoint support
 474 
 475 void os::breakpoint() {
 476   BREAKPOINT;
 477 }
 478 
 479 extern "C" void breakpoint() {
 480   // use debugger to set breakpoint here
 481 }
 482 
 483 ////////////////////////////////////////////////////////////////////////////////
 484 // signal support
 485 
 486 debug_only(static bool signal_sets_initialized = false);
 487 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 488 
 489 bool os::Linux::is_sig_ignored(int sig) {
 490       struct sigaction oact;
 491       sigaction(sig, (struct sigaction*)NULL, &oact);
 492       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 493                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 494       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 495            return true;
 496       else
 497            return false;
 498 }
 499 
 500 void os::Linux::signal_sets_init() {
 501   // Should also have an assertion stating we are still single-threaded.
 502   assert(!signal_sets_initialized, "Already initialized");
 503   // Fill in signals that are necessarily unblocked for all threads in
 504   // the VM. Currently, we unblock the following signals:
 505   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 506   //                         by -Xrs (=ReduceSignalUsage));
 507   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 508   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 509   // the dispositions or masks wrt these signals.
 510   // Programs embedding the VM that want to use the above signals for their
 511   // own purposes must, at this time, use the "-Xrs" option to prevent
 512   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 513   // (See bug 4345157, and other related bugs).
 514   // In reality, though, unblocking these signals is really a nop, since
 515   // these signals are not blocked by default.
 516   sigemptyset(&unblocked_sigs);
 517   sigemptyset(&allowdebug_blocked_sigs);
 518   sigaddset(&unblocked_sigs, SIGILL);
 519   sigaddset(&unblocked_sigs, SIGSEGV);
 520   sigaddset(&unblocked_sigs, SIGBUS);
 521   sigaddset(&unblocked_sigs, SIGFPE);
 522   sigaddset(&unblocked_sigs, SR_signum);
 523 
 524   if (!ReduceSignalUsage) {
 525    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 526       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 527       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 528    }
 529    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 530       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 531       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 532    }
 533    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 534       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 535       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 536    }
 537   }
 538   // Fill in signals that are blocked by all but the VM thread.
 539   sigemptyset(&vm_sigs);
 540   if (!ReduceSignalUsage)
 541     sigaddset(&vm_sigs, BREAK_SIGNAL);
 542   debug_only(signal_sets_initialized = true);
 543 
 544 }
 545 
 546 // These are signals that are unblocked while a thread is running Java.
 547 // (For some reason, they get blocked by default.)
 548 sigset_t* os::Linux::unblocked_signals() {
 549   assert(signal_sets_initialized, "Not initialized");
 550   return &unblocked_sigs;
 551 }
 552 
 553 // These are the signals that are blocked while a (non-VM) thread is
 554 // running Java. Only the VM thread handles these signals.
 555 sigset_t* os::Linux::vm_signals() {
 556   assert(signal_sets_initialized, "Not initialized");
 557   return &vm_sigs;
 558 }
 559 
 560 // These are signals that are blocked during cond_wait to allow debugger in
 561 sigset_t* os::Linux::allowdebug_blocked_signals() {
 562   assert(signal_sets_initialized, "Not initialized");
 563   return &allowdebug_blocked_sigs;
 564 }
 565 
 566 void os::Linux::hotspot_sigmask(Thread* thread) {
 567 
 568   //Save caller's signal mask before setting VM signal mask
 569   sigset_t caller_sigmask;
 570   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 571 
 572   OSThread* osthread = thread->osthread();
 573   osthread->set_caller_sigmask(caller_sigmask);
 574 
 575   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 576 
 577   if (!ReduceSignalUsage) {
 578     if (thread->is_VM_thread()) {
 579       // Only the VM thread handles BREAK_SIGNAL ...
 580       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 581     } else {
 582       // ... all other threads block BREAK_SIGNAL
 583       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 584     }
 585   }
 586 }
 587 
 588 //////////////////////////////////////////////////////////////////////////////
 589 // detecting pthread library
 590 
 591 void os::Linux::libpthread_init() {
 592   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 593   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 594   // generic name for earlier versions.
 595   // Define macros here so we can build HotSpot on old systems.
 596 # ifndef _CS_GNU_LIBC_VERSION
 597 # define _CS_GNU_LIBC_VERSION 2
 598 # endif
 599 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 600 # define _CS_GNU_LIBPTHREAD_VERSION 3
 601 # endif
 602 
 603   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 604   if (n > 0) {
 605      char *str = (char *)malloc(n, mtInternal);
 606      confstr(_CS_GNU_LIBC_VERSION, str, n);
 607      os::Linux::set_glibc_version(str);
 608   } else {
 609      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 610      static char _gnu_libc_version[32];
 611      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 612               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 613      os::Linux::set_glibc_version(_gnu_libc_version);
 614   }
 615 
 616   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 617   if (n > 0) {
 618      char *str = (char *)malloc(n, mtInternal);
 619      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 620      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 621      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 622      // is the case. LinuxThreads has a hard limit on max number of threads.
 623      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 624      // On the other hand, NPTL does not have such a limit, sysconf()
 625      // will return -1 and errno is not changed. Check if it is really NPTL.
 626      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 627          strstr(str, "NPTL") &&
 628          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 629        free(str);
 630        os::Linux::set_libpthread_version("linuxthreads");
 631      } else {
 632        os::Linux::set_libpthread_version(str);
 633      }
 634   } else {
 635     // glibc before 2.3.2 only has LinuxThreads.
 636     os::Linux::set_libpthread_version("linuxthreads");
 637   }
 638 
 639   if (strstr(libpthread_version(), "NPTL")) {
 640      os::Linux::set_is_NPTL();
 641   } else {
 642      os::Linux::set_is_LinuxThreads();
 643   }
 644 
 645   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 646   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 647   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 648      os::Linux::set_is_floating_stack();
 649   }
 650 }
 651 
 652 /////////////////////////////////////////////////////////////////////////////
 653 // thread stack
 654 
 655 // Force Linux kernel to expand current thread stack. If "bottom" is close
 656 // to the stack guard, caller should block all signals.
 657 //
 658 // MAP_GROWSDOWN:
 659 //   A special mmap() flag that is used to implement thread stacks. It tells
 660 //   kernel that the memory region should extend downwards when needed. This
 661 //   allows early versions of LinuxThreads to only mmap the first few pages
 662 //   when creating a new thread. Linux kernel will automatically expand thread
 663 //   stack as needed (on page faults).
 664 //
 665 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 666 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 667 //   region, it's hard to tell if the fault is due to a legitimate stack
 668 //   access or because of reading/writing non-exist memory (e.g. buffer
 669 //   overrun). As a rule, if the fault happens below current stack pointer,
 670 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 671 //   application (see Linux kernel fault.c).
 672 //
 673 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 674 //   stack overflow detection.
 675 //
 676 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 677 //   not use this flag. However, the stack of initial thread is not created
 678 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 679 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 680 //   and then attach the thread to JVM.
 681 //
 682 // To get around the problem and allow stack banging on Linux, we need to
 683 // manually expand thread stack after receiving the SIGSEGV.
 684 //
 685 // There are two ways to expand thread stack to address "bottom", we used
 686 // both of them in JVM before 1.5:
 687 //   1. adjust stack pointer first so that it is below "bottom", and then
 688 //      touch "bottom"
 689 //   2. mmap() the page in question
 690 //
 691 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 692 // if current sp is already near the lower end of page 101, and we need to
 693 // call mmap() to map page 100, it is possible that part of the mmap() frame
 694 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 695 // That will destroy the mmap() frame and cause VM to crash.
 696 //
 697 // The following code works by adjusting sp first, then accessing the "bottom"
 698 // page to force a page fault. Linux kernel will then automatically expand the
 699 // stack mapping.
 700 //
 701 // _expand_stack_to() assumes its frame size is less than page size, which
 702 // should always be true if the function is not inlined.
 703 
 704 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 705 #define NOINLINE
 706 #else
 707 #define NOINLINE __attribute__ ((noinline))
 708 #endif
 709 
 710 static void _expand_stack_to(address bottom) NOINLINE;
 711 
 712 static void _expand_stack_to(address bottom) {
 713   address sp;
 714   size_t size;
 715   volatile char *p;
 716 
 717   // Adjust bottom to point to the largest address within the same page, it
 718   // gives us a one-page buffer if alloca() allocates slightly more memory.
 719   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 720   bottom += os::Linux::page_size() - 1;
 721 
 722   // sp might be slightly above current stack pointer; if that's the case, we
 723   // will alloca() a little more space than necessary, which is OK. Don't use
 724   // os::current_stack_pointer(), as its result can be slightly below current
 725   // stack pointer, causing us to not alloca enough to reach "bottom".
 726   sp = (address)&sp;
 727 
 728   if (sp > bottom) {
 729     size = sp - bottom;
 730     p = (volatile char *)alloca(size);
 731     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 732     p[0] = '\0';
 733   }
 734 }
 735 
 736 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 737   assert(t!=NULL, "just checking");
 738   assert(t->osthread()->expanding_stack(), "expand should be set");
 739   assert(t->stack_base() != NULL, "stack_base was not initialized");
 740 
 741   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 742     sigset_t mask_all, old_sigset;
 743     sigfillset(&mask_all);
 744     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 745     _expand_stack_to(addr);
 746     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 747     return true;
 748   }
 749   return false;
 750 }
 751 
 752 //////////////////////////////////////////////////////////////////////////////
 753 // create new thread
 754 
 755 static address highest_vm_reserved_address();
 756 
 757 // check if it's safe to start a new thread
 758 static bool _thread_safety_check(Thread* thread) {
 759   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 760     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 761     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 762     //   allocated (MAP_FIXED) from high address space. Every thread stack
 763     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 764     //   it to other values if they rebuild LinuxThreads).
 765     //
 766     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 767     // the memory region has already been mmap'ed. That means if we have too
 768     // many threads and/or very large heap, eventually thread stack will
 769     // collide with heap.
 770     //
 771     // Here we try to prevent heap/stack collision by comparing current
 772     // stack bottom with the highest address that has been mmap'ed by JVM
 773     // plus a safety margin for memory maps created by native code.
 774     //
 775     // This feature can be disabled by setting ThreadSafetyMargin to 0
 776     //
 777     if (ThreadSafetyMargin > 0) {
 778       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 779 
 780       // not safe if our stack extends below the safety margin
 781       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 782     } else {
 783       return true;
 784     }
 785   } else {
 786     // Floating stack LinuxThreads or NPTL:
 787     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 788     //   there's not enough space left, pthread_create() will fail. If we come
 789     //   here, that means enough space has been reserved for stack.
 790     return true;
 791   }
 792 }
 793 
 794 // Thread start routine for all newly created threads
 795 static void *java_start(Thread *thread) {
 796   // Try to randomize the cache line index of hot stack frames.
 797   // This helps when threads of the same stack traces evict each other's
 798   // cache lines. The threads can be either from the same JVM instance, or
 799   // from different JVM instances. The benefit is especially true for
 800   // processors with hyperthreading technology.
 801   static int counter = 0;
 802   int pid = os::current_process_id();
 803   alloca(((pid ^ counter++) & 7) * 128);
 804 
 805   ThreadLocalStorage::set_thread(thread);
 806 
 807   OSThread* osthread = thread->osthread();
 808   Monitor* sync = osthread->startThread_lock();
 809 
 810   // non floating stack LinuxThreads needs extra check, see above
 811   if (!_thread_safety_check(thread)) {
 812     // notify parent thread
 813     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 814     osthread->set_state(ZOMBIE);
 815     sync->notify_all();
 816     return NULL;
 817   }
 818 
 819   // thread_id is kernel thread id (similar to Solaris LWP id)
 820   osthread->set_thread_id(os::Linux::gettid());
 821 
 822   if (UseNUMA) {
 823     int lgrp_id = os::numa_get_group_id();
 824     if (lgrp_id != -1) {
 825       thread->set_lgrp_id(lgrp_id);
 826     }
 827   }
 828   // initialize signal mask for this thread
 829   os::Linux::hotspot_sigmask(thread);
 830 
 831   // initialize floating point control register
 832   os::Linux::init_thread_fpu_state();
 833 
 834   // handshaking with parent thread
 835   {
 836     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 837 
 838     // notify parent thread
 839     osthread->set_state(INITIALIZED);
 840     sync->notify_all();
 841 
 842     // wait until os::start_thread()
 843     while (osthread->get_state() == INITIALIZED) {
 844       sync->wait(Mutex::_no_safepoint_check_flag);
 845     }
 846   }
 847 
 848   // call one more level start routine
 849   thread->run();
 850 
 851   return 0;
 852 }
 853 
 854 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 855   assert(thread->osthread() == NULL, "caller responsible");
 856 
 857   // Allocate the OSThread object
 858   OSThread* osthread = new OSThread(NULL, NULL);
 859   if (osthread == NULL) {
 860     return false;
 861   }
 862 
 863   // set the correct thread state
 864   osthread->set_thread_type(thr_type);
 865 
 866   // Initial state is ALLOCATED but not INITIALIZED
 867   osthread->set_state(ALLOCATED);
 868 
 869   thread->set_osthread(osthread);
 870 
 871   // init thread attributes
 872   pthread_attr_t attr;
 873   pthread_attr_init(&attr);
 874   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 875 
 876   // stack size
 877   if (os::Linux::supports_variable_stack_size()) {
 878     // calculate stack size if it's not specified by caller
 879     if (stack_size == 0) {
 880       stack_size = os::Linux::default_stack_size(thr_type);
 881 
 882       switch (thr_type) {
 883       case os::java_thread:
 884         // Java threads use ThreadStackSize which default value can be
 885         // changed with the flag -Xss
 886         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 887         stack_size = JavaThread::stack_size_at_create();
 888         break;
 889       case os::compiler_thread:
 890         if (CompilerThreadStackSize > 0) {
 891           stack_size = (size_t)(CompilerThreadStackSize * K);
 892           break;
 893         } // else fall through:
 894           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 895       case os::vm_thread:
 896       case os::pgc_thread:
 897       case os::cgc_thread:
 898       case os::watcher_thread:
 899         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 900         break;
 901       }
 902     }
 903 
 904     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 905     pthread_attr_setstacksize(&attr, stack_size);
 906   } else {
 907     // let pthread_create() pick the default value.
 908   }
 909 
 910   // glibc guard page
 911   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 912 
 913   ThreadState state;
 914 
 915   {
 916     // Serialize thread creation if we are running with fixed stack LinuxThreads
 917     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 918     if (lock) {
 919       os::Linux::createThread_lock()->lock_without_safepoint_check();
 920     }
 921 
 922     pthread_t tid;
 923     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 924 
 925     pthread_attr_destroy(&attr);
 926 
 927     if (ret != 0) {
 928       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 929         perror("pthread_create()");
 930       }
 931       // Need to clean up stuff we've allocated so far
 932       thread->set_osthread(NULL);
 933       delete osthread;
 934       if (lock) os::Linux::createThread_lock()->unlock();
 935       return false;
 936     }
 937 
 938     // Store pthread info into the OSThread
 939     osthread->set_pthread_id(tid);
 940 
 941     // Wait until child thread is either initialized or aborted
 942     {
 943       Monitor* sync_with_child = osthread->startThread_lock();
 944       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 945       while ((state = osthread->get_state()) == ALLOCATED) {
 946         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 947       }
 948     }
 949 
 950     if (lock) {
 951       os::Linux::createThread_lock()->unlock();
 952     }
 953   }
 954 
 955   // Aborted due to thread limit being reached
 956   if (state == ZOMBIE) {
 957       thread->set_osthread(NULL);
 958       delete osthread;
 959       return false;
 960   }
 961 
 962   // The thread is returned suspended (in state INITIALIZED),
 963   // and is started higher up in the call chain
 964   assert(state == INITIALIZED, "race condition");
 965   return true;
 966 }
 967 
 968 /////////////////////////////////////////////////////////////////////////////
 969 // attach existing thread
 970 
 971 // bootstrap the main thread
 972 bool os::create_main_thread(JavaThread* thread) {
 973   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 974   return create_attached_thread(thread);
 975 }
 976 
 977 bool os::create_attached_thread(JavaThread* thread) {
 978 #ifdef ASSERT
 979     thread->verify_not_published();
 980 #endif
 981 
 982   // Allocate the OSThread object
 983   OSThread* osthread = new OSThread(NULL, NULL);
 984 
 985   if (osthread == NULL) {
 986     return false;
 987   }
 988 
 989   // Store pthread info into the OSThread
 990   osthread->set_thread_id(os::Linux::gettid());
 991   osthread->set_pthread_id(::pthread_self());
 992 
 993   // initialize floating point control register
 994   os::Linux::init_thread_fpu_state();
 995 
 996   // Initial thread state is RUNNABLE
 997   osthread->set_state(RUNNABLE);
 998 
 999   thread->set_osthread(osthread);
1000 
1001   if (UseNUMA) {
1002     int lgrp_id = os::numa_get_group_id();
1003     if (lgrp_id != -1) {
1004       thread->set_lgrp_id(lgrp_id);
1005     }
1006   }
1007 
1008   if (os::Linux::is_initial_thread()) {
1009     // If current thread is initial thread, its stack is mapped on demand,
1010     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1011     // the entire stack region to avoid SEGV in stack banging.
1012     // It is also useful to get around the heap-stack-gap problem on SuSE
1013     // kernel (see 4821821 for details). We first expand stack to the top
1014     // of yellow zone, then enable stack yellow zone (order is significant,
1015     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1016     // is no gap between the last two virtual memory regions.
1017 
1018     JavaThread *jt = (JavaThread *)thread;
1019     address addr = jt->stack_yellow_zone_base();
1020     assert(addr != NULL, "initialization problem?");
1021     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1022 
1023     osthread->set_expanding_stack();
1024     os::Linux::manually_expand_stack(jt, addr);
1025     osthread->clear_expanding_stack();
1026   }
1027 
1028   // initialize signal mask for this thread
1029   // and save the caller's signal mask
1030   os::Linux::hotspot_sigmask(thread);
1031 
1032   return true;
1033 }
1034 
1035 void os::pd_start_thread(Thread* thread) {
1036   OSThread * osthread = thread->osthread();
1037   assert(osthread->get_state() != INITIALIZED, "just checking");
1038   Monitor* sync_with_child = osthread->startThread_lock();
1039   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1040   sync_with_child->notify();
1041 }
1042 
1043 // Free Linux resources related to the OSThread
1044 void os::free_thread(OSThread* osthread) {
1045   assert(osthread != NULL, "osthread not set");
1046 
1047   if (Thread::current()->osthread() == osthread) {
1048     // Restore caller's signal mask
1049     sigset_t sigmask = osthread->caller_sigmask();
1050     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1051    }
1052 
1053   delete osthread;
1054 }
1055 
1056 //////////////////////////////////////////////////////////////////////////////
1057 // thread local storage
1058 
1059 int os::allocate_thread_local_storage() {
1060   pthread_key_t key;
1061   int rslt = pthread_key_create(&key, NULL);
1062   assert(rslt == 0, "cannot allocate thread local storage");
1063   return (int)key;
1064 }
1065 
1066 // Note: This is currently not used by VM, as we don't destroy TLS key
1067 // on VM exit.
1068 void os::free_thread_local_storage(int index) {
1069   int rslt = pthread_key_delete((pthread_key_t)index);
1070   assert(rslt == 0, "invalid index");
1071 }
1072 
1073 void os::thread_local_storage_at_put(int index, void* value) {
1074   int rslt = pthread_setspecific((pthread_key_t)index, value);
1075   assert(rslt == 0, "pthread_setspecific failed");
1076 }
1077 
1078 extern "C" Thread* get_thread() {
1079   return ThreadLocalStorage::thread();
1080 }
1081 
1082 //////////////////////////////////////////////////////////////////////////////
1083 // initial thread
1084 
1085 // Check if current thread is the initial thread, similar to Solaris thr_main.
1086 bool os::Linux::is_initial_thread(void) {
1087   char dummy;
1088   // If called before init complete, thread stack bottom will be null.
1089   // Can be called if fatal error occurs before initialization.
1090   if (initial_thread_stack_bottom() == NULL) return false;
1091   assert(initial_thread_stack_bottom() != NULL &&
1092          initial_thread_stack_size()   != 0,
1093          "os::init did not locate initial thread's stack region");
1094   if ((address)&dummy >= initial_thread_stack_bottom() &&
1095       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1096        return true;
1097   else return false;
1098 }
1099 
1100 // Find the virtual memory area that contains addr
1101 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1102   FILE *fp = fopen("/proc/self/maps", "r");
1103   if (fp) {
1104     address low, high;
1105     while (!feof(fp)) {
1106       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1107         if (low <= addr && addr < high) {
1108            if (vma_low)  *vma_low  = low;
1109            if (vma_high) *vma_high = high;
1110            fclose (fp);
1111            return true;
1112         }
1113       }
1114       for (;;) {
1115         int ch = fgetc(fp);
1116         if (ch == EOF || ch == (int)'\n') break;
1117       }
1118     }
1119     fclose(fp);
1120   }
1121   return false;
1122 }
1123 
1124 // Locate initial thread stack. This special handling of initial thread stack
1125 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1126 // bogus value for initial thread.
1127 void os::Linux::capture_initial_stack(size_t max_size) {
1128   // stack size is the easy part, get it from RLIMIT_STACK
1129   size_t stack_size;
1130   struct rlimit rlim;
1131   getrlimit(RLIMIT_STACK, &rlim);
1132   stack_size = rlim.rlim_cur;
1133 
1134   // 6308388: a bug in ld.so will relocate its own .data section to the
1135   //   lower end of primordial stack; reduce ulimit -s value a little bit
1136   //   so we won't install guard page on ld.so's data section.
1137   stack_size -= 2 * page_size();
1138 
1139   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1140   //   7.1, in both cases we will get 2G in return value.
1141   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1142   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1143   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1144   //   in case other parts in glibc still assumes 2M max stack size.
1145   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1146   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1147   if (stack_size > 2 * K * K IA64_ONLY(*2))
1148       stack_size = 2 * K * K IA64_ONLY(*2);
1149   // Try to figure out where the stack base (top) is. This is harder.
1150   //
1151   // When an application is started, glibc saves the initial stack pointer in
1152   // a global variable "__libc_stack_end", which is then used by system
1153   // libraries. __libc_stack_end should be pretty close to stack top. The
1154   // variable is available since the very early days. However, because it is
1155   // a private interface, it could disappear in the future.
1156   //
1157   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1158   // to __libc_stack_end, it is very close to stack top, but isn't the real
1159   // stack top. Note that /proc may not exist if VM is running as a chroot
1160   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1161   // /proc/<pid>/stat could change in the future (though unlikely).
1162   //
1163   // We try __libc_stack_end first. If that doesn't work, look for
1164   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1165   // as a hint, which should work well in most cases.
1166 
1167   uintptr_t stack_start;
1168 
1169   // try __libc_stack_end first
1170   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1171   if (p && *p) {
1172     stack_start = *p;
1173   } else {
1174     // see if we can get the start_stack field from /proc/self/stat
1175     FILE *fp;
1176     int pid;
1177     char state;
1178     int ppid;
1179     int pgrp;
1180     int session;
1181     int nr;
1182     int tpgrp;
1183     unsigned long flags;
1184     unsigned long minflt;
1185     unsigned long cminflt;
1186     unsigned long majflt;
1187     unsigned long cmajflt;
1188     unsigned long utime;
1189     unsigned long stime;
1190     long cutime;
1191     long cstime;
1192     long prio;
1193     long nice;
1194     long junk;
1195     long it_real;
1196     uintptr_t start;
1197     uintptr_t vsize;
1198     intptr_t rss;
1199     uintptr_t rsslim;
1200     uintptr_t scodes;
1201     uintptr_t ecode;
1202     int i;
1203 
1204     // Figure what the primordial thread stack base is. Code is inspired
1205     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1206     // followed by command name surrounded by parentheses, state, etc.
1207     char stat[2048];
1208     int statlen;
1209 
1210     fp = fopen("/proc/self/stat", "r");
1211     if (fp) {
1212       statlen = fread(stat, 1, 2047, fp);
1213       stat[statlen] = '\0';
1214       fclose(fp);
1215 
1216       // Skip pid and the command string. Note that we could be dealing with
1217       // weird command names, e.g. user could decide to rename java launcher
1218       // to "java 1.4.2 :)", then the stat file would look like
1219       //                1234 (java 1.4.2 :)) R ... ...
1220       // We don't really need to know the command string, just find the last
1221       // occurrence of ")" and then start parsing from there. See bug 4726580.
1222       char * s = strrchr(stat, ')');
1223 
1224       i = 0;
1225       if (s) {
1226         // Skip blank chars
1227         do s++; while (isspace(*s));
1228 
1229 #define _UFM UINTX_FORMAT
1230 #define _DFM INTX_FORMAT
1231 
1232         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1233         /*              3  4  5  6  7  8   9   0   1   2   3   4   5   6   7   8   9   0   1    2    3    4    5    6    7    8 */
1234         i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1235              &state,          /* 3  %c  */
1236              &ppid,           /* 4  %d  */
1237              &pgrp,           /* 5  %d  */
1238              &session,        /* 6  %d  */
1239              &nr,             /* 7  %d  */
1240              &tpgrp,          /* 8  %d  */
1241              &flags,          /* 9  %lu  */
1242              &minflt,         /* 10 %lu  */
1243              &cminflt,        /* 11 %lu  */
1244              &majflt,         /* 12 %lu  */
1245              &cmajflt,        /* 13 %lu  */
1246              &utime,          /* 14 %lu  */
1247              &stime,          /* 15 %lu  */
1248              &cutime,         /* 16 %ld  */
1249              &cstime,         /* 17 %ld  */
1250              &prio,           /* 18 %ld  */
1251              &nice,           /* 19 %ld  */
1252              &junk,           /* 20 %ld  */
1253              &it_real,        /* 21 %ld  */
1254              &start,          /* 22 UINTX_FORMAT */
1255              &vsize,          /* 23 UINTX_FORMAT */
1256              &rss,            /* 24 INTX_FORMAT  */
1257              &rsslim,         /* 25 UINTX_FORMAT */
1258              &scodes,         /* 26 UINTX_FORMAT */
1259              &ecode,          /* 27 UINTX_FORMAT */
1260              &stack_start);   /* 28 UINTX_FORMAT */
1261       }
1262 
1263 #undef _UFM
1264 #undef _DFM
1265 
1266       if (i != 28 - 2) {
1267          assert(false, "Bad conversion from /proc/self/stat");
1268          // product mode - assume we are the initial thread, good luck in the
1269          // embedded case.
1270          warning("Can't detect initial thread stack location - bad conversion");
1271          stack_start = (uintptr_t) &rlim;
1272       }
1273     } else {
1274       // For some reason we can't open /proc/self/stat (for example, running on
1275       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1276       // most cases, so don't abort:
1277       warning("Can't detect initial thread stack location - no /proc/self/stat");
1278       stack_start = (uintptr_t) &rlim;
1279     }
1280   }
1281 
1282   // Now we have a pointer (stack_start) very close to the stack top, the
1283   // next thing to do is to figure out the exact location of stack top. We
1284   // can find out the virtual memory area that contains stack_start by
1285   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1286   // and its upper limit is the real stack top. (again, this would fail if
1287   // running inside chroot, because /proc may not exist.)
1288 
1289   uintptr_t stack_top;
1290   address low, high;
1291   if (find_vma((address)stack_start, &low, &high)) {
1292     // success, "high" is the true stack top. (ignore "low", because initial
1293     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1294     stack_top = (uintptr_t)high;
1295   } else {
1296     // failed, likely because /proc/self/maps does not exist
1297     warning("Can't detect initial thread stack location - find_vma failed");
1298     // best effort: stack_start is normally within a few pages below the real
1299     // stack top, use it as stack top, and reduce stack size so we won't put
1300     // guard page outside stack.
1301     stack_top = stack_start;
1302     stack_size -= 16 * page_size();
1303   }
1304 
1305   // stack_top could be partially down the page so align it
1306   stack_top = align_size_up(stack_top, page_size());
1307 
1308   if (max_size && stack_size > max_size) {
1309      _initial_thread_stack_size = max_size;
1310   } else {
1311      _initial_thread_stack_size = stack_size;
1312   }
1313 
1314   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1315   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1316 }
1317 
1318 ////////////////////////////////////////////////////////////////////////////////
1319 // time support
1320 
1321 // Time since start-up in seconds to a fine granularity.
1322 // Used by VMSelfDestructTimer and the MemProfiler.
1323 double os::elapsedTime() {
1324 
1325   return (double)(os::elapsed_counter()) * 0.000001;
1326 }
1327 
1328 jlong os::elapsed_counter() {
1329   timeval time;
1330   int status = gettimeofday(&time, NULL);
1331   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1332 }
1333 
1334 jlong os::elapsed_frequency() {
1335   return (1000 * 1000);
1336 }
1337 
1338 // For now, we say that linux does not support vtime.  I have no idea
1339 // whether it can actually be made to (DLD, 9/13/05).
1340 
1341 bool os::supports_vtime() { return false; }
1342 bool os::enable_vtime()   { return false; }
1343 bool os::vtime_enabled()  { return false; }
1344 double os::elapsedVTime() {
1345   // better than nothing, but not much
1346   return elapsedTime();
1347 }
1348 
1349 jlong os::javaTimeMillis() {
1350   timeval time;
1351   int status = gettimeofday(&time, NULL);
1352   assert(status != -1, "linux error");
1353   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1354 }
1355 
1356 #ifndef CLOCK_MONOTONIC
1357 #define CLOCK_MONOTONIC (1)
1358 #endif
1359 
1360 void os::Linux::clock_init() {
1361   // we do dlopen's in this particular order due to bug in linux
1362   // dynamical loader (see 6348968) leading to crash on exit
1363   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1364   if (handle == NULL) {
1365     handle = dlopen("librt.so", RTLD_LAZY);
1366   }
1367 
1368   if (handle) {
1369     int (*clock_getres_func)(clockid_t, struct timespec*) =
1370            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1371     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1372            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1373     if (clock_getres_func && clock_gettime_func) {
1374       // See if monotonic clock is supported by the kernel. Note that some
1375       // early implementations simply return kernel jiffies (updated every
1376       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1377       // for nano time (though the monotonic property is still nice to have).
1378       // It's fixed in newer kernels, however clock_getres() still returns
1379       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1380       // resolution for now. Hopefully as people move to new kernels, this
1381       // won't be a problem.
1382       struct timespec res;
1383       struct timespec tp;
1384       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1385           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1386         // yes, monotonic clock is supported
1387         _clock_gettime = clock_gettime_func;
1388       } else {
1389         // close librt if there is no monotonic clock
1390         dlclose(handle);
1391       }
1392     }
1393   }
1394 }
1395 
1396 #ifndef SYS_clock_getres
1397 
1398 #if defined(IA32) || defined(AMD64)
1399 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1400 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1401 #else
1402 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1403 #define sys_clock_getres(x,y)  -1
1404 #endif
1405 
1406 #else
1407 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1408 #endif
1409 
1410 void os::Linux::fast_thread_clock_init() {
1411   if (!UseLinuxPosixThreadCPUClocks) {
1412     return;
1413   }
1414   clockid_t clockid;
1415   struct timespec tp;
1416   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1417       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1418 
1419   // Switch to using fast clocks for thread cpu time if
1420   // the sys_clock_getres() returns 0 error code.
1421   // Note, that some kernels may support the current thread
1422   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1423   // returned by the pthread_getcpuclockid().
1424   // If the fast Posix clocks are supported then the sys_clock_getres()
1425   // must return at least tp.tv_sec == 0 which means a resolution
1426   // better than 1 sec. This is extra check for reliability.
1427 
1428   if(pthread_getcpuclockid_func &&
1429      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1430      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1431 
1432     _supports_fast_thread_cpu_time = true;
1433     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1434   }
1435 }
1436 
1437 jlong os::javaTimeNanos() {
1438   if (Linux::supports_monotonic_clock()) {
1439     struct timespec tp;
1440     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1441     assert(status == 0, "gettime error");
1442     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1443     return result;
1444   } else {
1445     timeval time;
1446     int status = gettimeofday(&time, NULL);
1447     assert(status != -1, "linux error");
1448     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1449     return 1000 * usecs;
1450   }
1451 }
1452 
1453 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1454   if (Linux::supports_monotonic_clock()) {
1455     info_ptr->max_value = ALL_64_BITS;
1456 
1457     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1458     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1459     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1460   } else {
1461     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1462     info_ptr->max_value = ALL_64_BITS;
1463 
1464     // gettimeofday is a real time clock so it skips
1465     info_ptr->may_skip_backward = true;
1466     info_ptr->may_skip_forward = true;
1467   }
1468 
1469   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1470 }
1471 
1472 // Return the real, user, and system times in seconds from an
1473 // arbitrary fixed point in the past.
1474 bool os::getTimesSecs(double* process_real_time,
1475                       double* process_user_time,
1476                       double* process_system_time) {
1477   struct tms ticks;
1478   clock_t real_ticks = times(&ticks);
1479 
1480   if (real_ticks == (clock_t) (-1)) {
1481     return false;
1482   } else {
1483     double ticks_per_second = (double) clock_tics_per_sec;
1484     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1485     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1486     *process_real_time = ((double) real_ticks) / ticks_per_second;
1487 
1488     return true;
1489   }
1490 }
1491 
1492 
1493 char * os::local_time_string(char *buf, size_t buflen) {
1494   struct tm t;
1495   time_t long_time;
1496   time(&long_time);
1497   localtime_r(&long_time, &t);
1498   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1499                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1500                t.tm_hour, t.tm_min, t.tm_sec);
1501   return buf;
1502 }
1503 
1504 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1505   return localtime_r(clock, res);
1506 }
1507 
1508 ////////////////////////////////////////////////////////////////////////////////
1509 // runtime exit support
1510 
1511 // Note: os::shutdown() might be called very early during initialization, or
1512 // called from signal handler. Before adding something to os::shutdown(), make
1513 // sure it is async-safe and can handle partially initialized VM.
1514 void os::shutdown() {
1515 
1516   // allow PerfMemory to attempt cleanup of any persistent resources
1517   perfMemory_exit();
1518 
1519   // needs to remove object in file system
1520   AttachListener::abort();
1521 
1522   // flush buffered output, finish log files
1523   ostream_abort();
1524 
1525   // Check for abort hook
1526   abort_hook_t abort_hook = Arguments::abort_hook();
1527   if (abort_hook != NULL) {
1528     abort_hook();
1529   }
1530 
1531 }
1532 
1533 // Note: os::abort() might be called very early during initialization, or
1534 // called from signal handler. Before adding something to os::abort(), make
1535 // sure it is async-safe and can handle partially initialized VM.
1536 void os::abort(bool dump_core) {
1537   os::shutdown();
1538   if (dump_core) {
1539 #ifndef PRODUCT
1540     fdStream out(defaultStream::output_fd());
1541     out.print_raw("Current thread is ");
1542     char buf[16];
1543     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1544     out.print_raw_cr(buf);
1545     out.print_raw_cr("Dumping core ...");
1546 #endif
1547     ::abort(); // dump core
1548   }
1549 
1550   ::exit(1);
1551 }
1552 
1553 // Die immediately, no exit hook, no abort hook, no cleanup.
1554 void os::die() {
1555   // _exit() on LinuxThreads only kills current thread
1556   ::abort();
1557 }
1558 
1559 // unused on linux for now.
1560 void os::set_error_file(const char *logfile) {}
1561 
1562 
1563 // This method is a copy of JDK's sysGetLastErrorString
1564 // from src/solaris/hpi/src/system_md.c
1565 
1566 size_t os::lasterror(char *buf, size_t len) {
1567 
1568   if (errno == 0)  return 0;
1569 
1570   const char *s = ::strerror(errno);
1571   size_t n = ::strlen(s);
1572   if (n >= len) {
1573     n = len - 1;
1574   }
1575   ::strncpy(buf, s, n);
1576   buf[n] = '\0';
1577   return n;
1578 }
1579 
1580 intx os::current_thread_id() { return (intx)pthread_self(); }
1581 int os::current_process_id() {
1582 
1583   // Under the old linux thread library, linux gives each thread
1584   // its own process id. Because of this each thread will return
1585   // a different pid if this method were to return the result
1586   // of getpid(2). Linux provides no api that returns the pid
1587   // of the launcher thread for the vm. This implementation
1588   // returns a unique pid, the pid of the launcher thread
1589   // that starts the vm 'process'.
1590 
1591   // Under the NPTL, getpid() returns the same pid as the
1592   // launcher thread rather than a unique pid per thread.
1593   // Use gettid() if you want the old pre NPTL behaviour.
1594 
1595   // if you are looking for the result of a call to getpid() that
1596   // returns a unique pid for the calling thread, then look at the
1597   // OSThread::thread_id() method in osThread_linux.hpp file
1598 
1599   return (int)(_initial_pid ? _initial_pid : getpid());
1600 }
1601 
1602 // DLL functions
1603 
1604 const char* os::dll_file_extension() { return ".so"; }
1605 
1606 // This must be hard coded because it's the system's temporary
1607 // directory not the java application's temp directory, ala java.io.tmpdir.
1608 const char* os::get_temp_directory() { return "/tmp"; }
1609 
1610 static bool file_exists(const char* filename) {
1611   struct stat statbuf;
1612   if (filename == NULL || strlen(filename) == 0) {
1613     return false;
1614   }
1615   return os::stat(filename, &statbuf) == 0;
1616 }
1617 
1618 bool os::dll_build_name(char* buffer, size_t buflen,
1619                         const char* pname, const char* fname) {
1620   bool retval = false;
1621   // Copied from libhpi
1622   const size_t pnamelen = pname ? strlen(pname) : 0;
1623 
1624   // Return error on buffer overflow.
1625   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1626     return retval;
1627   }
1628 
1629   if (pnamelen == 0) {
1630     snprintf(buffer, buflen, "lib%s.so", fname);
1631     retval = true;
1632   } else if (strchr(pname, *os::path_separator()) != NULL) {
1633     int n;
1634     char** pelements = split_path(pname, &n);
1635     if (pelements == NULL) {
1636       return false;
1637     }
1638     for (int i = 0 ; i < n ; i++) {
1639       // Really shouldn't be NULL, but check can't hurt
1640       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1641         continue; // skip the empty path values
1642       }
1643       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1644       if (file_exists(buffer)) {
1645         retval = true;
1646         break;
1647       }
1648     }
1649     // release the storage
1650     for (int i = 0 ; i < n ; i++) {
1651       if (pelements[i] != NULL) {
1652         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1653       }
1654     }
1655     if (pelements != NULL) {
1656       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1657     }
1658   } else {
1659     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1660     retval = true;
1661   }
1662   return retval;
1663 }
1664 
1665 const char* os::get_current_directory(char *buf, int buflen) {
1666   return getcwd(buf, buflen);
1667 }
1668 
1669 // check if addr is inside libjvm.so
1670 bool os::address_is_in_vm(address addr) {
1671   static address libjvm_base_addr;
1672   Dl_info dlinfo;
1673 
1674   if (libjvm_base_addr == NULL) {
1675     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1676     libjvm_base_addr = (address)dlinfo.dli_fbase;
1677     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1678   }
1679 
1680   if (dladdr((void *)addr, &dlinfo)) {
1681     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1682   }
1683 
1684   return false;
1685 }
1686 
1687 bool os::dll_address_to_function_name(address addr, char *buf,
1688                                       int buflen, int *offset) {
1689   Dl_info dlinfo;
1690 
1691   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1692     if (buf != NULL) {
1693       if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1694         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1695       }
1696     }
1697     if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1698     return true;
1699   } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1700     if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1701         buf, buflen, offset, dlinfo.dli_fname)) {
1702        return true;
1703     }
1704   }
1705 
1706   if (buf != NULL) buf[0] = '\0';
1707   if (offset != NULL) *offset = -1;
1708   return false;
1709 }
1710 
1711 struct _address_to_library_name {
1712   address addr;          // input : memory address
1713   size_t  buflen;        //         size of fname
1714   char*   fname;         // output: library name
1715   address base;          //         library base addr
1716 };
1717 
1718 static int address_to_library_name_callback(struct dl_phdr_info *info,
1719                                             size_t size, void *data) {
1720   int i;
1721   bool found = false;
1722   address libbase = NULL;
1723   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1724 
1725   // iterate through all loadable segments
1726   for (i = 0; i < info->dlpi_phnum; i++) {
1727     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1728     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1729       // base address of a library is the lowest address of its loaded
1730       // segments.
1731       if (libbase == NULL || libbase > segbase) {
1732         libbase = segbase;
1733       }
1734       // see if 'addr' is within current segment
1735       if (segbase <= d->addr &&
1736           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1737         found = true;
1738       }
1739     }
1740   }
1741 
1742   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1743   // so dll_address_to_library_name() can fall through to use dladdr() which
1744   // can figure out executable name from argv[0].
1745   if (found && info->dlpi_name && info->dlpi_name[0]) {
1746     d->base = libbase;
1747     if (d->fname) {
1748       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1749     }
1750     return 1;
1751   }
1752   return 0;
1753 }
1754 
1755 bool os::dll_address_to_library_name(address addr, char* buf,
1756                                      int buflen, int* offset) {
1757   Dl_info dlinfo;
1758   struct _address_to_library_name data;
1759 
1760   // There is a bug in old glibc dladdr() implementation that it could resolve
1761   // to wrong library name if the .so file has a base address != NULL. Here
1762   // we iterate through the program headers of all loaded libraries to find
1763   // out which library 'addr' really belongs to. This workaround can be
1764   // removed once the minimum requirement for glibc is moved to 2.3.x.
1765   data.addr = addr;
1766   data.fname = buf;
1767   data.buflen = buflen;
1768   data.base = NULL;
1769   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1770 
1771   if (rslt) {
1772      // buf already contains library name
1773      if (offset) *offset = addr - data.base;
1774      return true;
1775   } else if (dladdr((void*)addr, &dlinfo)){
1776      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1777      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1778      return true;
1779   } else {
1780      if (buf) buf[0] = '\0';
1781      if (offset) *offset = -1;
1782      return false;
1783   }
1784 }
1785 
1786   // Loads .dll/.so and
1787   // in case of error it checks if .dll/.so was built for the
1788   // same architecture as Hotspot is running on
1789 
1790 
1791 // Remember the stack's state. The Linux dynamic linker will change
1792 // the stack to 'executable' at most once, so we must safepoint only once.
1793 bool os::Linux::_stack_is_executable = false;
1794 
1795 // VM operation that loads a library.  This is necessary if stack protection
1796 // of the Java stacks can be lost during loading the library.  If we
1797 // do not stop the Java threads, they can stack overflow before the stacks
1798 // are protected again.
1799 class VM_LinuxDllLoad: public VM_Operation {
1800  private:
1801   const char *_filename;
1802   char *_ebuf;
1803   int _ebuflen;
1804   void *_lib;
1805  public:
1806   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1807     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1808   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1809   void doit() {
1810     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1811     os::Linux::_stack_is_executable = true;
1812   }
1813   void* loaded_library() { return _lib; }
1814 };
1815 
1816 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1817 {
1818   void * result = NULL;
1819   bool load_attempted = false;
1820 
1821   // Check whether the library to load might change execution rights
1822   // of the stack. If they are changed, the protection of the stack
1823   // guard pages will be lost. We need a safepoint to fix this.
1824   //
1825   // See Linux man page execstack(8) for more info.
1826   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1827     ElfFile ef(filename);
1828     if (!ef.specifies_noexecstack()) {
1829       if (!is_init_completed()) {
1830         os::Linux::_stack_is_executable = true;
1831         // This is OK - No Java threads have been created yet, and hence no
1832         // stack guard pages to fix.
1833         //
1834         // This should happen only when you are building JDK7 using a very
1835         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1836         //
1837         // Dynamic loader will make all stacks executable after
1838         // this function returns, and will not do that again.
1839         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1840       } else {
1841         warning("You have loaded library %s which might have disabled stack guard. "
1842                 "The VM will try to fix the stack guard now.\n"
1843                 "It's highly recommended that you fix the library with "
1844                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1845                 filename);
1846 
1847         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1848         JavaThread *jt = JavaThread::current();
1849         if (jt->thread_state() != _thread_in_native) {
1850           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1851           // that requires ExecStack. Cannot enter safe point. Let's give up.
1852           warning("Unable to fix stack guard. Giving up.");
1853         } else {
1854           if (!LoadExecStackDllInVMThread) {
1855             // This is for the case where the DLL has an static
1856             // constructor function that executes JNI code. We cannot
1857             // load such DLLs in the VMThread.
1858             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1859           }
1860 
1861           ThreadInVMfromNative tiv(jt);
1862           debug_only(VMNativeEntryWrapper vew;)
1863 
1864           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1865           VMThread::execute(&op);
1866           if (LoadExecStackDllInVMThread) {
1867             result = op.loaded_library();
1868           }
1869           load_attempted = true;
1870         }
1871       }
1872     }
1873   }
1874 
1875   if (!load_attempted) {
1876     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1877   }
1878 
1879   if (result != NULL) {
1880     // Successful loading
1881     return result;
1882   }
1883 
1884   Elf32_Ehdr elf_head;
1885   int diag_msg_max_length=ebuflen-strlen(ebuf);
1886   char* diag_msg_buf=ebuf+strlen(ebuf);
1887 
1888   if (diag_msg_max_length==0) {
1889     // No more space in ebuf for additional diagnostics message
1890     return NULL;
1891   }
1892 
1893 
1894   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1895 
1896   if (file_descriptor < 0) {
1897     // Can't open library, report dlerror() message
1898     return NULL;
1899   }
1900 
1901   bool failed_to_read_elf_head=
1902     (sizeof(elf_head)!=
1903         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1904 
1905   ::close(file_descriptor);
1906   if (failed_to_read_elf_head) {
1907     // file i/o error - report dlerror() msg
1908     return NULL;
1909   }
1910 
1911   typedef struct {
1912     Elf32_Half  code;         // Actual value as defined in elf.h
1913     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1914     char        elf_class;    // 32 or 64 bit
1915     char        endianess;    // MSB or LSB
1916     char*       name;         // String representation
1917   } arch_t;
1918 
1919   #ifndef EM_486
1920   #define EM_486          6               /* Intel 80486 */
1921   #endif
1922 
1923   static const arch_t arch_array[]={
1924     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1925     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1926     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1927     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1928     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1929     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1930     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1931     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1932     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1933     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1934     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1935     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1936     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1937     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1938     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1939     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1940   };
1941 
1942   #if  (defined IA32)
1943     static  Elf32_Half running_arch_code=EM_386;
1944   #elif   (defined AMD64)
1945     static  Elf32_Half running_arch_code=EM_X86_64;
1946   #elif  (defined IA64)
1947     static  Elf32_Half running_arch_code=EM_IA_64;
1948   #elif  (defined __sparc) && (defined _LP64)
1949     static  Elf32_Half running_arch_code=EM_SPARCV9;
1950   #elif  (defined __sparc) && (!defined _LP64)
1951     static  Elf32_Half running_arch_code=EM_SPARC;
1952   #elif  (defined __powerpc64__)
1953     static  Elf32_Half running_arch_code=EM_PPC64;
1954   #elif  (defined __powerpc__)
1955     static  Elf32_Half running_arch_code=EM_PPC;
1956   #elif  (defined ARM)
1957     static  Elf32_Half running_arch_code=EM_ARM;
1958   #elif  (defined S390)
1959     static  Elf32_Half running_arch_code=EM_S390;
1960   #elif  (defined ALPHA)
1961     static  Elf32_Half running_arch_code=EM_ALPHA;
1962   #elif  (defined MIPSEL)
1963     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1964   #elif  (defined PARISC)
1965     static  Elf32_Half running_arch_code=EM_PARISC;
1966   #elif  (defined MIPS)
1967     static  Elf32_Half running_arch_code=EM_MIPS;
1968   #elif  (defined M68K)
1969     static  Elf32_Half running_arch_code=EM_68K;
1970   #else
1971     #error Method os::dll_load requires that one of following is defined:\
1972          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1973   #endif
1974 
1975   // Identify compatability class for VM's architecture and library's architecture
1976   // Obtain string descriptions for architectures
1977 
1978   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1979   int running_arch_index=-1;
1980 
1981   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1982     if (running_arch_code == arch_array[i].code) {
1983       running_arch_index    = i;
1984     }
1985     if (lib_arch.code == arch_array[i].code) {
1986       lib_arch.compat_class = arch_array[i].compat_class;
1987       lib_arch.name         = arch_array[i].name;
1988     }
1989   }
1990 
1991   assert(running_arch_index != -1,
1992     "Didn't find running architecture code (running_arch_code) in arch_array");
1993   if (running_arch_index == -1) {
1994     // Even though running architecture detection failed
1995     // we may still continue with reporting dlerror() message
1996     return NULL;
1997   }
1998 
1999   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2000     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2001     return NULL;
2002   }
2003 
2004 #ifndef S390
2005   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2006     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2007     return NULL;
2008   }
2009 #endif // !S390
2010 
2011   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2012     if ( lib_arch.name!=NULL ) {
2013       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2014         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2015         lib_arch.name, arch_array[running_arch_index].name);
2016     } else {
2017       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2018       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2019         lib_arch.code,
2020         arch_array[running_arch_index].name);
2021     }
2022   }
2023 
2024   return NULL;
2025 }
2026 
2027 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2028   void * result = ::dlopen(filename, RTLD_LAZY);
2029   if (result == NULL) {
2030     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2031     ebuf[ebuflen-1] = '\0';
2032   }
2033   return result;
2034 }
2035 
2036 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2037   void * result = NULL;
2038   if (LoadExecStackDllInVMThread) {
2039     result = dlopen_helper(filename, ebuf, ebuflen);
2040   }
2041 
2042   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2043   // library that requires an executable stack, or which does not have this
2044   // stack attribute set, dlopen changes the stack attribute to executable. The
2045   // read protection of the guard pages gets lost.
2046   //
2047   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2048   // may have been queued at the same time.
2049 
2050   if (!_stack_is_executable) {
2051     JavaThread *jt = Threads::first();
2052 
2053     while (jt) {
2054       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2055           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2056         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2057                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2058           warning("Attempt to reguard stack yellow zone failed.");
2059         }
2060       }
2061       jt = jt->next();
2062     }
2063   }
2064 
2065   return result;
2066 }
2067 
2068 /*
2069  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2070  * chances are you might want to run the generated bits against glibc-2.0
2071  * libdl.so, so always use locking for any version of glibc.
2072  */
2073 void* os::dll_lookup(void* handle, const char* name) {
2074   pthread_mutex_lock(&dl_mutex);
2075   void* res = dlsym(handle, name);
2076   pthread_mutex_unlock(&dl_mutex);
2077   return res;
2078 }
2079 
2080 
2081 static bool _print_ascii_file(const char* filename, outputStream* st) {
2082   int fd = ::open(filename, O_RDONLY);
2083   if (fd == -1) {
2084      return false;
2085   }
2086 
2087   char buf[32];
2088   int bytes;
2089   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2090     st->print_raw(buf, bytes);
2091   }
2092 
2093   ::close(fd);
2094 
2095   return true;
2096 }
2097 
2098 void os::print_dll_info(outputStream *st) {
2099    st->print_cr("Dynamic libraries:");
2100 
2101    char fname[32];
2102    pid_t pid = os::Linux::gettid();
2103 
2104    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2105 
2106    if (!_print_ascii_file(fname, st)) {
2107      st->print("Can not get library information for pid = %d\n", pid);
2108    }
2109 }
2110 
2111 void os::print_os_info_brief(outputStream* st) {
2112   os::Linux::print_distro_info(st);
2113 
2114   os::Posix::print_uname_info(st);
2115 
2116   os::Linux::print_libversion_info(st);
2117 
2118 }
2119 
2120 void os::print_os_info(outputStream* st) {
2121   st->print("OS:");
2122 
2123   os::Linux::print_distro_info(st);
2124 
2125   os::Posix::print_uname_info(st);
2126 
2127   // Print warning if unsafe chroot environment detected
2128   if (unsafe_chroot_detected) {
2129     st->print("WARNING!! ");
2130     st->print_cr(unstable_chroot_error);
2131   }
2132 
2133   os::Linux::print_libversion_info(st);
2134 
2135   os::Posix::print_rlimit_info(st);
2136 
2137   os::Posix::print_load_average(st);
2138 
2139   os::Linux::print_full_memory_info(st);
2140 }
2141 
2142 // Try to identify popular distros.
2143 // Most Linux distributions have /etc/XXX-release file, which contains
2144 // the OS version string. Some have more than one /etc/XXX-release file
2145 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2146 // so the order is important.
2147 void os::Linux::print_distro_info(outputStream* st) {
2148   if (!_print_ascii_file("/etc/mandrake-release", st) &&
2149       !_print_ascii_file("/etc/sun-release", st) &&
2150       !_print_ascii_file("/etc/redhat-release", st) &&
2151       !_print_ascii_file("/etc/SuSE-release", st) &&
2152       !_print_ascii_file("/etc/turbolinux-release", st) &&
2153       !_print_ascii_file("/etc/gentoo-release", st) &&
2154       !_print_ascii_file("/etc/debian_version", st) &&
2155       !_print_ascii_file("/etc/ltib-release", st) &&
2156       !_print_ascii_file("/etc/angstrom-version", st)) {
2157       st->print("Linux");
2158   }
2159   st->cr();
2160 }
2161 
2162 void os::Linux::print_libversion_info(outputStream* st) {
2163   // libc, pthread
2164   st->print("libc:");
2165   st->print(os::Linux::glibc_version()); st->print(" ");
2166   st->print(os::Linux::libpthread_version()); st->print(" ");
2167   if (os::Linux::is_LinuxThreads()) {
2168      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2169   }
2170   st->cr();
2171 }
2172 
2173 void os::Linux::print_full_memory_info(outputStream* st) {
2174    st->print("\n/proc/meminfo:\n");
2175    _print_ascii_file("/proc/meminfo", st);
2176    st->cr();
2177 }
2178 
2179 void os::print_memory_info(outputStream* st) {
2180 
2181   st->print("Memory:");
2182   st->print(" %dk page", os::vm_page_size()>>10);
2183 
2184   // values in struct sysinfo are "unsigned long"
2185   struct sysinfo si;
2186   sysinfo(&si);
2187 
2188   st->print(", physical " UINT64_FORMAT "k",
2189             os::physical_memory() >> 10);
2190   st->print("(" UINT64_FORMAT "k free)",
2191             os::available_memory() >> 10);
2192   st->print(", swap " UINT64_FORMAT "k",
2193             ((jlong)si.totalswap * si.mem_unit) >> 10);
2194   st->print("(" UINT64_FORMAT "k free)",
2195             ((jlong)si.freeswap * si.mem_unit) >> 10);
2196   st->cr();
2197 }
2198 
2199 void os::pd_print_cpu_info(outputStream* st) {
2200   st->print("\n/proc/cpuinfo:\n");
2201   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2202     st->print("  <Not Available>");
2203   }
2204   st->cr();
2205 }
2206 
2207 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2208 // but they're the same for all the linux arch that we support
2209 // and they're the same for solaris but there's no common place to put this.
2210 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2211                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2212                           "ILL_COPROC", "ILL_BADSTK" };
2213 
2214 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2215                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2216                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2217 
2218 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2219 
2220 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2221 
2222 void os::print_siginfo(outputStream* st, void* siginfo) {
2223   st->print("siginfo:");
2224 
2225   const int buflen = 100;
2226   char buf[buflen];
2227   siginfo_t *si = (siginfo_t*)siginfo;
2228   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2229   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2230     st->print("si_errno=%s", buf);
2231   } else {
2232     st->print("si_errno=%d", si->si_errno);
2233   }
2234   const int c = si->si_code;
2235   assert(c > 0, "unexpected si_code");
2236   switch (si->si_signo) {
2237   case SIGILL:
2238     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2239     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2240     break;
2241   case SIGFPE:
2242     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2243     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2244     break;
2245   case SIGSEGV:
2246     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2247     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2248     break;
2249   case SIGBUS:
2250     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2251     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2252     break;
2253   default:
2254     st->print(", si_code=%d", si->si_code);
2255     // no si_addr
2256   }
2257 
2258   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2259       UseSharedSpaces) {
2260     FileMapInfo* mapinfo = FileMapInfo::current_info();
2261     if (mapinfo->is_in_shared_space(si->si_addr)) {
2262       st->print("\n\nError accessing class data sharing archive."   \
2263                 " Mapped file inaccessible during execution, "      \
2264                 " possible disk/network problem.");
2265     }
2266   }
2267   st->cr();
2268 }
2269 
2270 
2271 static void print_signal_handler(outputStream* st, int sig,
2272                                  char* buf, size_t buflen);
2273 
2274 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2275   st->print_cr("Signal Handlers:");
2276   print_signal_handler(st, SIGSEGV, buf, buflen);
2277   print_signal_handler(st, SIGBUS , buf, buflen);
2278   print_signal_handler(st, SIGFPE , buf, buflen);
2279   print_signal_handler(st, SIGPIPE, buf, buflen);
2280   print_signal_handler(st, SIGXFSZ, buf, buflen);
2281   print_signal_handler(st, SIGILL , buf, buflen);
2282   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2283   print_signal_handler(st, SR_signum, buf, buflen);
2284   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2285   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2286   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2287   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2288 }
2289 
2290 static char saved_jvm_path[MAXPATHLEN] = {0};
2291 
2292 // Find the full path to the current module, libjvm.so
2293 void os::jvm_path(char *buf, jint buflen) {
2294   // Error checking.
2295   if (buflen < MAXPATHLEN) {
2296     assert(false, "must use a large-enough buffer");
2297     buf[0] = '\0';
2298     return;
2299   }
2300   // Lazy resolve the path to current module.
2301   if (saved_jvm_path[0] != 0) {
2302     strcpy(buf, saved_jvm_path);
2303     return;
2304   }
2305 
2306   char dli_fname[MAXPATHLEN];
2307   bool ret = dll_address_to_library_name(
2308                 CAST_FROM_FN_PTR(address, os::jvm_path),
2309                 dli_fname, sizeof(dli_fname), NULL);
2310   assert(ret != 0, "cannot locate libjvm");
2311   char *rp = realpath(dli_fname, buf);
2312   if (rp == NULL)
2313     return;
2314 
2315   if (Arguments::created_by_gamma_launcher()) {
2316     // Support for the gamma launcher.  Typical value for buf is
2317     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2318     // the right place in the string, then assume we are installed in a JDK and
2319     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2320     // up the path so it looks like libjvm.so is installed there (append a
2321     // fake suffix hotspot/libjvm.so).
2322     const char *p = buf + strlen(buf) - 1;
2323     for (int count = 0; p > buf && count < 5; ++count) {
2324       for (--p; p > buf && *p != '/'; --p)
2325         /* empty */ ;
2326     }
2327 
2328     if (strncmp(p, "/jre/lib/", 9) != 0) {
2329       // Look for JAVA_HOME in the environment.
2330       char* java_home_var = ::getenv("JAVA_HOME");
2331       if (java_home_var != NULL && java_home_var[0] != 0) {
2332         char* jrelib_p;
2333         int len;
2334 
2335         // Check the current module name "libjvm.so".
2336         p = strrchr(buf, '/');
2337         assert(strstr(p, "/libjvm") == p, "invalid library name");
2338 
2339         rp = realpath(java_home_var, buf);
2340         if (rp == NULL)
2341           return;
2342 
2343         // determine if this is a legacy image or modules image
2344         // modules image doesn't have "jre" subdirectory
2345         len = strlen(buf);
2346         jrelib_p = buf + len;
2347         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2348         if (0 != access(buf, F_OK)) {
2349           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2350         }
2351 
2352         if (0 == access(buf, F_OK)) {
2353           // Use current module name "libjvm.so"
2354           len = strlen(buf);
2355           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2356         } else {
2357           // Go back to path of .so
2358           rp = realpath(dli_fname, buf);
2359           if (rp == NULL)
2360             return;
2361         }
2362       }
2363     }
2364   }
2365 
2366   strcpy(saved_jvm_path, buf);
2367 }
2368 
2369 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2370   // no prefix required, not even "_"
2371 }
2372 
2373 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2374   // no suffix required
2375 }
2376 
2377 ////////////////////////////////////////////////////////////////////////////////
2378 // sun.misc.Signal support
2379 
2380 static volatile jint sigint_count = 0;
2381 
2382 static void
2383 UserHandler(int sig, void *siginfo, void *context) {
2384   // 4511530 - sem_post is serialized and handled by the manager thread. When
2385   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2386   // don't want to flood the manager thread with sem_post requests.
2387   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2388       return;
2389 
2390   // Ctrl-C is pressed during error reporting, likely because the error
2391   // handler fails to abort. Let VM die immediately.
2392   if (sig == SIGINT && is_error_reported()) {
2393      os::die();
2394   }
2395 
2396   os::signal_notify(sig);
2397 }
2398 
2399 void* os::user_handler() {
2400   return CAST_FROM_FN_PTR(void*, UserHandler);
2401 }
2402 
2403 extern "C" {
2404   typedef void (*sa_handler_t)(int);
2405   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2406 }
2407 
2408 void* os::signal(int signal_number, void* handler) {
2409   struct sigaction sigAct, oldSigAct;
2410 
2411   sigfillset(&(sigAct.sa_mask));
2412   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2413   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2414 
2415   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2416     // -1 means registration failed
2417     return (void *)-1;
2418   }
2419 
2420   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2421 }
2422 
2423 void os::signal_raise(int signal_number) {
2424   ::raise(signal_number);
2425 }
2426 
2427 /*
2428  * The following code is moved from os.cpp for making this
2429  * code platform specific, which it is by its very nature.
2430  */
2431 
2432 // Will be modified when max signal is changed to be dynamic
2433 int os::sigexitnum_pd() {
2434   return NSIG;
2435 }
2436 
2437 // a counter for each possible signal value
2438 static volatile jint pending_signals[NSIG+1] = { 0 };
2439 
2440 // Linux(POSIX) specific hand shaking semaphore.
2441 static sem_t sig_sem;
2442 
2443 void os::signal_init_pd() {
2444   // Initialize signal structures
2445   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2446 
2447   // Initialize signal semaphore
2448   ::sem_init(&sig_sem, 0, 0);
2449 }
2450 
2451 void os::signal_notify(int sig) {
2452   Atomic::inc(&pending_signals[sig]);
2453   ::sem_post(&sig_sem);
2454 }
2455 
2456 static int check_pending_signals(bool wait) {
2457   Atomic::store(0, &sigint_count);
2458   for (;;) {
2459     for (int i = 0; i < NSIG + 1; i++) {
2460       jint n = pending_signals[i];
2461       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2462         return i;
2463       }
2464     }
2465     if (!wait) {
2466       return -1;
2467     }
2468     JavaThread *thread = JavaThread::current();
2469     ThreadBlockInVM tbivm(thread);
2470 
2471     bool threadIsSuspended;
2472     do {
2473       thread->set_suspend_equivalent();
2474       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2475       ::sem_wait(&sig_sem);
2476 
2477       // were we externally suspended while we were waiting?
2478       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2479       if (threadIsSuspended) {
2480         //
2481         // The semaphore has been incremented, but while we were waiting
2482         // another thread suspended us. We don't want to continue running
2483         // while suspended because that would surprise the thread that
2484         // suspended us.
2485         //
2486         ::sem_post(&sig_sem);
2487 
2488         thread->java_suspend_self();
2489       }
2490     } while (threadIsSuspended);
2491   }
2492 }
2493 
2494 int os::signal_lookup() {
2495   return check_pending_signals(false);
2496 }
2497 
2498 int os::signal_wait() {
2499   return check_pending_signals(true);
2500 }
2501 
2502 ////////////////////////////////////////////////////////////////////////////////
2503 // Virtual Memory
2504 
2505 int os::vm_page_size() {
2506   // Seems redundant as all get out
2507   assert(os::Linux::page_size() != -1, "must call os::init");
2508   return os::Linux::page_size();
2509 }
2510 
2511 // Solaris allocates memory by pages.
2512 int os::vm_allocation_granularity() {
2513   assert(os::Linux::page_size() != -1, "must call os::init");
2514   return os::Linux::page_size();
2515 }
2516 
2517 // Rationale behind this function:
2518 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2519 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2520 //  samples for JITted code. Here we create private executable mapping over the code cache
2521 //  and then we can use standard (well, almost, as mapping can change) way to provide
2522 //  info for the reporting script by storing timestamp and location of symbol
2523 void linux_wrap_code(char* base, size_t size) {
2524   static volatile jint cnt = 0;
2525 
2526   if (!UseOprofile) {
2527     return;
2528   }
2529 
2530   char buf[PATH_MAX+1];
2531   int num = Atomic::add(1, &cnt);
2532 
2533   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2534            os::get_temp_directory(), os::current_process_id(), num);
2535   unlink(buf);
2536 
2537   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2538 
2539   if (fd != -1) {
2540     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2541     if (rv != (off_t)-1) {
2542       if (::write(fd, "", 1) == 1) {
2543         mmap(base, size,
2544              PROT_READ|PROT_WRITE|PROT_EXEC,
2545              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2546       }
2547     }
2548     ::close(fd);
2549     unlink(buf);
2550   }
2551 }
2552 
2553 // NOTE: Linux kernel does not really reserve the pages for us.
2554 //       All it does is to check if there are enough free pages
2555 //       left at the time of mmap(). This could be a potential
2556 //       problem.
2557 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2558   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2559   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2560                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2561   if (res != (uintptr_t) MAP_FAILED) {
2562     if (UseNUMAInterleaving) {
2563       numa_make_global(addr, size);
2564     }
2565     return true;
2566   }
2567   return false;
2568 }
2569 
2570 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2571 #ifndef MAP_HUGETLB
2572 #define MAP_HUGETLB 0x40000
2573 #endif
2574 
2575 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2576 #ifndef MADV_HUGEPAGE
2577 #define MADV_HUGEPAGE 14
2578 #endif
2579 
2580 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2581                        bool exec) {
2582   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2583     int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2584     uintptr_t res =
2585       (uintptr_t) ::mmap(addr, size, prot,
2586                          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2587                          -1, 0);
2588     if (res != (uintptr_t) MAP_FAILED) {
2589       if (UseNUMAInterleaving) {
2590         numa_make_global(addr, size);
2591       }
2592       return true;
2593     }
2594     // Fall through and try to use small pages
2595   }
2596 
2597   if (commit_memory(addr, size, exec)) {
2598     realign_memory(addr, size, alignment_hint);
2599     return true;
2600   }
2601   return false;
2602 }
2603 
2604 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2605   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2606     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2607     // be supported or the memory may already be backed by huge pages.
2608     ::madvise(addr, bytes, MADV_HUGEPAGE);
2609   }
2610 }
2611 
2612 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2613   // This method works by doing an mmap over an existing mmaping and effectively discarding
2614   // the existing pages. However it won't work for SHM-based large pages that cannot be
2615   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2616   // small pages on top of the SHM segment. This method always works for small pages, so we
2617   // allow that in any case.
2618   if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2619     commit_memory(addr, bytes, alignment_hint, false);
2620   }
2621 }
2622 
2623 void os::numa_make_global(char *addr, size_t bytes) {
2624   Linux::numa_interleave_memory(addr, bytes);
2625 }
2626 
2627 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2628   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2629 }
2630 
2631 bool os::numa_topology_changed()   { return false; }
2632 
2633 size_t os::numa_get_groups_num() {
2634   int max_node = Linux::numa_max_node();
2635   return max_node > 0 ? max_node + 1 : 1;
2636 }
2637 
2638 int os::numa_get_group_id() {
2639   int cpu_id = Linux::sched_getcpu();
2640   if (cpu_id != -1) {
2641     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2642     if (lgrp_id != -1) {
2643       return lgrp_id;
2644     }
2645   }
2646   return 0;
2647 }
2648 
2649 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2650   for (size_t i = 0; i < size; i++) {
2651     ids[i] = i;
2652   }
2653   return size;
2654 }
2655 
2656 bool os::get_page_info(char *start, page_info* info) {
2657   return false;
2658 }
2659 
2660 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2661   return end;
2662 }
2663 
2664 
2665 int os::Linux::sched_getcpu_syscall(void) {
2666   unsigned int cpu;
2667   int retval = -1;
2668 
2669 #if defined(IA32)
2670 # ifndef SYS_getcpu
2671 # define SYS_getcpu 318
2672 # endif
2673   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2674 #elif defined(AMD64)
2675 // Unfortunately we have to bring all these macros here from vsyscall.h
2676 // to be able to compile on old linuxes.
2677 # define __NR_vgetcpu 2
2678 # define VSYSCALL_START (-10UL << 20)
2679 # define VSYSCALL_SIZE 1024
2680 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2681   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2682   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2683   retval = vgetcpu(&cpu, NULL, NULL);
2684 #endif
2685 
2686   return (retval == -1) ? retval : cpu;
2687 }
2688 
2689 // Something to do with the numa-aware allocator needs these symbols
2690 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2691 extern "C" JNIEXPORT void numa_error(char *where) { }
2692 extern "C" JNIEXPORT int fork1() { return fork(); }
2693 
2694 
2695 // If we are running with libnuma version > 2, then we should
2696 // be trying to use symbols with versions 1.1
2697 // If we are running with earlier version, which did not have symbol versions,
2698 // we should use the base version.
2699 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2700   void *f = dlvsym(handle, name, "libnuma_1.1");
2701   if (f == NULL) {
2702     f = dlsym(handle, name);
2703   }
2704   return f;
2705 }
2706 
2707 bool os::Linux::libnuma_init() {
2708   // sched_getcpu() should be in libc.
2709   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2710                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2711 
2712   // If it's not, try a direct syscall.
2713   if (sched_getcpu() == -1)
2714     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2715 
2716   if (sched_getcpu() != -1) { // Does it work?
2717     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2718     if (handle != NULL) {
2719       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2720                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2721       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2722                                        libnuma_dlsym(handle, "numa_max_node")));
2723       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2724                                         libnuma_dlsym(handle, "numa_available")));
2725       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2726                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2727       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2728                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2729 
2730 
2731       if (numa_available() != -1) {
2732         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2733         // Create a cpu -> node mapping
2734         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2735         rebuild_cpu_to_node_map();
2736         return true;
2737       }
2738     }
2739   }
2740   return false;
2741 }
2742 
2743 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2744 // The table is later used in get_node_by_cpu().
2745 void os::Linux::rebuild_cpu_to_node_map() {
2746   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2747                               // in libnuma (possible values are starting from 16,
2748                               // and continuing up with every other power of 2, but less
2749                               // than the maximum number of CPUs supported by kernel), and
2750                               // is a subject to change (in libnuma version 2 the requirements
2751                               // are more reasonable) we'll just hardcode the number they use
2752                               // in the library.
2753   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2754 
2755   size_t cpu_num = os::active_processor_count();
2756   size_t cpu_map_size = NCPUS / BitsPerCLong;
2757   size_t cpu_map_valid_size =
2758     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2759 
2760   cpu_to_node()->clear();
2761   cpu_to_node()->at_grow(cpu_num - 1);
2762   size_t node_num = numa_get_groups_num();
2763 
2764   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2765   for (size_t i = 0; i < node_num; i++) {
2766     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2767       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2768         if (cpu_map[j] != 0) {
2769           for (size_t k = 0; k < BitsPerCLong; k++) {
2770             if (cpu_map[j] & (1UL << k)) {
2771               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2772             }
2773           }
2774         }
2775       }
2776     }
2777   }
2778   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2779 }
2780 
2781 int os::Linux::get_node_by_cpu(int cpu_id) {
2782   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2783     return cpu_to_node()->at(cpu_id);
2784   }
2785   return -1;
2786 }
2787 
2788 GrowableArray<int>* os::Linux::_cpu_to_node;
2789 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2790 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2791 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2792 os::Linux::numa_available_func_t os::Linux::_numa_available;
2793 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2794 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2795 unsigned long* os::Linux::_numa_all_nodes;
2796 
2797 bool os::pd_uncommit_memory(char* addr, size_t size) {
2798   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2799                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2800   return res  != (uintptr_t) MAP_FAILED;
2801 }
2802 
2803 // Linux uses a growable mapping for the stack, and if the mapping for
2804 // the stack guard pages is not removed when we detach a thread the
2805 // stack cannot grow beyond the pages where the stack guard was
2806 // mapped.  If at some point later in the process the stack expands to
2807 // that point, the Linux kernel cannot expand the stack any further
2808 // because the guard pages are in the way, and a segfault occurs.
2809 //
2810 // However, it's essential not to split the stack region by unmapping
2811 // a region (leaving a hole) that's already part of the stack mapping,
2812 // so if the stack mapping has already grown beyond the guard pages at
2813 // the time we create them, we have to truncate the stack mapping.
2814 // So, we need to know the extent of the stack mapping when
2815 // create_stack_guard_pages() is called.
2816 
2817 // Find the bounds of the stack mapping.  Return true for success.
2818 //
2819 // We only need this for stacks that are growable: at the time of
2820 // writing thread stacks don't use growable mappings (i.e. those
2821 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2822 // only applies to the main thread.
2823 
2824 static
2825 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2826 
2827   char buf[128];
2828   int fd, sz;
2829 
2830   if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2831     return false;
2832   }
2833 
2834   const char kw[] = "[stack]";
2835   const int kwlen = sizeof(kw)-1;
2836 
2837   // Address part of /proc/self/maps couldn't be more than 128 bytes
2838   while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2839      if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2840         // Extract addresses
2841         if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2842            uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2843            if (sp >= *bottom && sp <= *top) {
2844               ::close(fd);
2845               return true;
2846            }
2847         }
2848      }
2849   }
2850 
2851  ::close(fd);
2852   return false;
2853 }
2854 
2855 
2856 // If the (growable) stack mapping already extends beyond the point
2857 // where we're going to put our guard pages, truncate the mapping at
2858 // that point by munmap()ping it.  This ensures that when we later
2859 // munmap() the guard pages we don't leave a hole in the stack
2860 // mapping. This only affects the main/initial thread, but guard
2861 // against future OS changes
2862 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2863   uintptr_t stack_extent, stack_base;
2864   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2865   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2866       assert(os::Linux::is_initial_thread(),
2867            "growable stack in non-initial thread");
2868     if (stack_extent < (uintptr_t)addr)
2869       ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2870   }
2871 
2872   return os::commit_memory(addr, size);
2873 }
2874 
2875 // If this is a growable mapping, remove the guard pages entirely by
2876 // munmap()ping them.  If not, just call uncommit_memory(). This only
2877 // affects the main/initial thread, but guard against future OS changes
2878 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2879   uintptr_t stack_extent, stack_base;
2880   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2881   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2882       assert(os::Linux::is_initial_thread(),
2883            "growable stack in non-initial thread");
2884 
2885     return ::munmap(addr, size) == 0;
2886   }
2887 
2888   return os::uncommit_memory(addr, size);
2889 }
2890 
2891 static address _highest_vm_reserved_address = NULL;
2892 
2893 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2894 // at 'requested_addr'. If there are existing memory mappings at the same
2895 // location, however, they will be overwritten. If 'fixed' is false,
2896 // 'requested_addr' is only treated as a hint, the return value may or
2897 // may not start from the requested address. Unlike Linux mmap(), this
2898 // function returns NULL to indicate failure.
2899 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2900   char * addr;
2901   int flags;
2902 
2903   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2904   if (fixed) {
2905     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2906     flags |= MAP_FIXED;
2907   }
2908 
2909   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2910   // to PROT_EXEC if executable when we commit the page.
2911   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2912                        flags, -1, 0);
2913 
2914   if (addr != MAP_FAILED) {
2915     // anon_mmap() should only get called during VM initialization,
2916     // don't need lock (actually we can skip locking even it can be called
2917     // from multiple threads, because _highest_vm_reserved_address is just a
2918     // hint about the upper limit of non-stack memory regions.)
2919     if ((address)addr + bytes > _highest_vm_reserved_address) {
2920       _highest_vm_reserved_address = (address)addr + bytes;
2921     }
2922   }
2923 
2924   return addr == MAP_FAILED ? NULL : addr;
2925 }
2926 
2927 // Don't update _highest_vm_reserved_address, because there might be memory
2928 // regions above addr + size. If so, releasing a memory region only creates
2929 // a hole in the address space, it doesn't help prevent heap-stack collision.
2930 //
2931 static int anon_munmap(char * addr, size_t size) {
2932   return ::munmap(addr, size) == 0;
2933 }
2934 
2935 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2936                          size_t alignment_hint) {
2937   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2938 }
2939 
2940 bool os::pd_release_memory(char* addr, size_t size) {
2941   return anon_munmap(addr, size);
2942 }
2943 
2944 static address highest_vm_reserved_address() {
2945   return _highest_vm_reserved_address;
2946 }
2947 
2948 static bool linux_mprotect(char* addr, size_t size, int prot) {
2949   // Linux wants the mprotect address argument to be page aligned.
2950   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2951 
2952   // According to SUSv3, mprotect() should only be used with mappings
2953   // established by mmap(), and mmap() always maps whole pages. Unaligned
2954   // 'addr' likely indicates problem in the VM (e.g. trying to change
2955   // protection of malloc'ed or statically allocated memory). Check the
2956   // caller if you hit this assert.
2957   assert(addr == bottom, "sanity check");
2958 
2959   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2960   return ::mprotect(bottom, size, prot) == 0;
2961 }
2962 
2963 // Set protections specified
2964 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2965                         bool is_committed) {
2966   unsigned int p = 0;
2967   switch (prot) {
2968   case MEM_PROT_NONE: p = PROT_NONE; break;
2969   case MEM_PROT_READ: p = PROT_READ; break;
2970   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2971   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2972   default:
2973     ShouldNotReachHere();
2974   }
2975   // is_committed is unused.
2976   return linux_mprotect(addr, bytes, p);
2977 }
2978 
2979 bool os::guard_memory(char* addr, size_t size) {
2980   return linux_mprotect(addr, size, PROT_NONE);
2981 }
2982 
2983 bool os::unguard_memory(char* addr, size_t size) {
2984   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2985 }
2986 
2987 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2988   bool result = false;
2989   void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2990                   MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2991                   -1, 0);
2992 
2993   if (p != (void *) -1) {
2994     // We don't know if this really is a huge page or not.
2995     FILE *fp = fopen("/proc/self/maps", "r");
2996     if (fp) {
2997       while (!feof(fp)) {
2998         char chars[257];
2999         long x = 0;
3000         if (fgets(chars, sizeof(chars), fp)) {
3001           if (sscanf(chars, "%lx-%*x", &x) == 1
3002               && x == (long)p) {
3003             if (strstr (chars, "hugepage")) {
3004               result = true;
3005               break;
3006             }
3007           }
3008         }
3009       }
3010       fclose(fp);
3011     }
3012     munmap (p, page_size);
3013     if (result)
3014       return true;
3015   }
3016 
3017   if (warn) {
3018     warning("HugeTLBFS is not supported by the operating system.");
3019   }
3020 
3021   return result;
3022 }
3023 
3024 /*
3025 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3026 *
3027 * From the coredump_filter documentation:
3028 *
3029 * - (bit 0) anonymous private memory
3030 * - (bit 1) anonymous shared memory
3031 * - (bit 2) file-backed private memory
3032 * - (bit 3) file-backed shared memory
3033 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3034 *           effective only if the bit 2 is cleared)
3035 * - (bit 5) hugetlb private memory
3036 * - (bit 6) hugetlb shared memory
3037 */
3038 static void set_coredump_filter(void) {
3039   FILE *f;
3040   long cdm;
3041 
3042   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3043     return;
3044   }
3045 
3046   if (fscanf(f, "%lx", &cdm) != 1) {
3047     fclose(f);
3048     return;
3049   }
3050 
3051   rewind(f);
3052 
3053   if ((cdm & LARGEPAGES_BIT) == 0) {
3054     cdm |= LARGEPAGES_BIT;
3055     fprintf(f, "%#lx", cdm);
3056   }
3057 
3058   fclose(f);
3059 }
3060 
3061 // Large page support
3062 
3063 static size_t _large_page_size = 0;
3064 
3065 void os::large_page_init() {
3066   if (!UseLargePages) {
3067     UseHugeTLBFS = false;
3068     UseSHM = false;
3069     return;
3070   }
3071 
3072   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3073     // If UseLargePages is specified on the command line try both methods,
3074     // if it's default, then try only HugeTLBFS.
3075     if (FLAG_IS_DEFAULT(UseLargePages)) {
3076       UseHugeTLBFS = true;
3077     } else {
3078       UseHugeTLBFS = UseSHM = true;
3079     }
3080   }
3081 
3082   if (LargePageSizeInBytes) {
3083     _large_page_size = LargePageSizeInBytes;
3084   } else {
3085     // large_page_size on Linux is used to round up heap size. x86 uses either
3086     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3087     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3088     // page as large as 256M.
3089     //
3090     // Here we try to figure out page size by parsing /proc/meminfo and looking
3091     // for a line with the following format:
3092     //    Hugepagesize:     2048 kB
3093     //
3094     // If we can't determine the value (e.g. /proc is not mounted, or the text
3095     // format has been changed), we'll use the largest page size supported by
3096     // the processor.
3097 
3098 #ifndef ZERO
3099     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3100                        ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3101 #endif // ZERO
3102 
3103     FILE *fp = fopen("/proc/meminfo", "r");
3104     if (fp) {
3105       while (!feof(fp)) {
3106         int x = 0;
3107         char buf[16];
3108         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3109           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3110             _large_page_size = x * K;
3111             break;
3112           }
3113         } else {
3114           // skip to next line
3115           for (;;) {
3116             int ch = fgetc(fp);
3117             if (ch == EOF || ch == (int)'\n') break;
3118           }
3119         }
3120       }
3121       fclose(fp);
3122     }
3123   }
3124 
3125   // print a warning if any large page related flag is specified on command line
3126   bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3127 
3128   const size_t default_page_size = (size_t)Linux::page_size();
3129   if (_large_page_size > default_page_size) {
3130     _page_sizes[0] = _large_page_size;
3131     _page_sizes[1] = default_page_size;
3132     _page_sizes[2] = 0;
3133   }
3134   UseHugeTLBFS = UseHugeTLBFS &&
3135                  Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3136 
3137   if (UseHugeTLBFS)
3138     UseSHM = false;
3139 
3140   UseLargePages = UseHugeTLBFS || UseSHM;
3141 
3142   set_coredump_filter();
3143 }
3144 
3145 #ifndef SHM_HUGETLB
3146 #define SHM_HUGETLB 04000
3147 #endif
3148 
3149 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3150   // "exec" is passed in but not used.  Creating the shared image for
3151   // the code cache doesn't have an SHM_X executable permission to check.
3152   assert(UseLargePages && UseSHM, "only for SHM large pages");
3153 
3154   key_t key = IPC_PRIVATE;
3155   char *addr;
3156 
3157   bool warn_on_failure = UseLargePages &&
3158                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3159                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3160                         );
3161   char msg[128];
3162 
3163   // Create a large shared memory region to attach to based on size.
3164   // Currently, size is the total size of the heap
3165   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3166   if (shmid == -1) {
3167      // Possible reasons for shmget failure:
3168      // 1. shmmax is too small for Java heap.
3169      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3170      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3171      // 2. not enough large page memory.
3172      //    > check available large pages: cat /proc/meminfo
3173      //    > increase amount of large pages:
3174      //          echo new_value > /proc/sys/vm/nr_hugepages
3175      //      Note 1: different Linux may use different name for this property,
3176      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3177      //      Note 2: it's possible there's enough physical memory available but
3178      //            they are so fragmented after a long run that they can't
3179      //            coalesce into large pages. Try to reserve large pages when
3180      //            the system is still "fresh".
3181      if (warn_on_failure) {
3182        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3183        warning(msg);
3184      }
3185      return NULL;
3186   }
3187 
3188   // attach to the region
3189   addr = (char*)shmat(shmid, req_addr, 0);
3190   int err = errno;
3191 
3192   // Remove shmid. If shmat() is successful, the actual shared memory segment
3193   // will be deleted when it's detached by shmdt() or when the process
3194   // terminates. If shmat() is not successful this will remove the shared
3195   // segment immediately.
3196   shmctl(shmid, IPC_RMID, NULL);
3197 
3198   if ((intptr_t)addr == -1) {
3199      if (warn_on_failure) {
3200        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3201        warning(msg);
3202      }
3203      return NULL;
3204   }
3205 
3206   if ((addr != NULL) && UseNUMAInterleaving) {
3207     numa_make_global(addr, bytes);
3208   }
3209 
3210   // The memory is committed
3211   address pc = CALLER_PC;
3212   MemTracker::record_virtual_memory_reserve((address)addr, bytes, pc);
3213   MemTracker::record_virtual_memory_commit((address)addr, bytes, pc);
3214 
3215   return addr;
3216 }
3217 
3218 bool os::release_memory_special(char* base, size_t bytes) {
3219   // detaching the SHM segment will also delete it, see reserve_memory_special()
3220   int rslt = shmdt(base);
3221   if (rslt == 0) {
3222     MemTracker::record_virtual_memory_uncommit((address)base, bytes);
3223     MemTracker::record_virtual_memory_release((address)base, bytes);
3224     return true;
3225   } else {
3226    return false;
3227   }
3228 }
3229 
3230 size_t os::large_page_size() {
3231   return _large_page_size;
3232 }
3233 
3234 // HugeTLBFS allows application to commit large page memory on demand;
3235 // with SysV SHM the entire memory region must be allocated as shared
3236 // memory.
3237 bool os::can_commit_large_page_memory() {
3238   return UseHugeTLBFS;
3239 }
3240 
3241 bool os::can_execute_large_page_memory() {
3242   return UseHugeTLBFS;
3243 }
3244 
3245 // Reserve memory at an arbitrary address, only if that area is
3246 // available (and not reserved for something else).
3247 
3248 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3249   const int max_tries = 10;
3250   char* base[max_tries];
3251   size_t size[max_tries];
3252   const size_t gap = 0x000000;
3253 
3254   // Assert only that the size is a multiple of the page size, since
3255   // that's all that mmap requires, and since that's all we really know
3256   // about at this low abstraction level.  If we need higher alignment,
3257   // we can either pass an alignment to this method or verify alignment
3258   // in one of the methods further up the call chain.  See bug 5044738.
3259   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3260 
3261   // Repeatedly allocate blocks until the block is allocated at the
3262   // right spot. Give up after max_tries. Note that reserve_memory() will
3263   // automatically update _highest_vm_reserved_address if the call is
3264   // successful. The variable tracks the highest memory address every reserved
3265   // by JVM. It is used to detect heap-stack collision if running with
3266   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3267   // space than needed, it could confuse the collision detecting code. To
3268   // solve the problem, save current _highest_vm_reserved_address and
3269   // calculate the correct value before return.
3270   address old_highest = _highest_vm_reserved_address;
3271 
3272   // Linux mmap allows caller to pass an address as hint; give it a try first,
3273   // if kernel honors the hint then we can return immediately.
3274   char * addr = anon_mmap(requested_addr, bytes, false);
3275   if (addr == requested_addr) {
3276      return requested_addr;
3277   }
3278 
3279   if (addr != NULL) {
3280      // mmap() is successful but it fails to reserve at the requested address
3281      anon_munmap(addr, bytes);
3282   }
3283 
3284   int i;
3285   for (i = 0; i < max_tries; ++i) {
3286     base[i] = reserve_memory(bytes);
3287 
3288     if (base[i] != NULL) {
3289       // Is this the block we wanted?
3290       if (base[i] == requested_addr) {
3291         size[i] = bytes;
3292         break;
3293       }
3294 
3295       // Does this overlap the block we wanted? Give back the overlapped
3296       // parts and try again.
3297 
3298       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3299       if (top_overlap >= 0 && top_overlap < bytes) {
3300         unmap_memory(base[i], top_overlap);
3301         base[i] += top_overlap;
3302         size[i] = bytes - top_overlap;
3303       } else {
3304         size_t bottom_overlap = base[i] + bytes - requested_addr;
3305         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3306           unmap_memory(requested_addr, bottom_overlap);
3307           size[i] = bytes - bottom_overlap;
3308         } else {
3309           size[i] = bytes;
3310         }
3311       }
3312     }
3313   }
3314 
3315   // Give back the unused reserved pieces.
3316 
3317   for (int j = 0; j < i; ++j) {
3318     if (base[j] != NULL) {
3319       unmap_memory(base[j], size[j]);
3320     }
3321   }
3322 
3323   if (i < max_tries) {
3324     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3325     return requested_addr;
3326   } else {
3327     _highest_vm_reserved_address = old_highest;
3328     return NULL;
3329   }
3330 }
3331 
3332 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3333   return ::read(fd, buf, nBytes);
3334 }
3335 
3336 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3337 // Solaris uses poll(), linux uses park().
3338 // Poll() is likely a better choice, assuming that Thread.interrupt()
3339 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3340 // SIGSEGV, see 4355769.
3341 
3342 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3343   assert(thread == Thread::current(),  "thread consistency check");
3344 
3345   ParkEvent * const slp = thread->_SleepEvent ;
3346   slp->reset() ;
3347   OrderAccess::fence() ;
3348 
3349   if (interruptible) {
3350     jlong prevtime = javaTimeNanos();
3351 
3352     for (;;) {
3353       if (os::is_interrupted(thread, true)) {
3354         return OS_INTRPT;
3355       }
3356 
3357       jlong newtime = javaTimeNanos();
3358 
3359       if (newtime - prevtime < 0) {
3360         // time moving backwards, should only happen if no monotonic clock
3361         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3362         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3363       } else {
3364         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3365       }
3366 
3367       if(millis <= 0) {
3368         return OS_OK;
3369       }
3370 
3371       prevtime = newtime;
3372 
3373       {
3374         assert(thread->is_Java_thread(), "sanity check");
3375         JavaThread *jt = (JavaThread *) thread;
3376         ThreadBlockInVM tbivm(jt);
3377         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3378 
3379         jt->set_suspend_equivalent();
3380         // cleared by handle_special_suspend_equivalent_condition() or
3381         // java_suspend_self() via check_and_wait_while_suspended()
3382 
3383         slp->park(millis);
3384 
3385         // were we externally suspended while we were waiting?
3386         jt->check_and_wait_while_suspended();
3387       }
3388     }
3389   } else {
3390     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3391     jlong prevtime = javaTimeNanos();
3392 
3393     for (;;) {
3394       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3395       // the 1st iteration ...
3396       jlong newtime = javaTimeNanos();
3397 
3398       if (newtime - prevtime < 0) {
3399         // time moving backwards, should only happen if no monotonic clock
3400         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3401         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3402       } else {
3403         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3404       }
3405 
3406       if(millis <= 0) break ;
3407 
3408       prevtime = newtime;
3409       slp->park(millis);
3410     }
3411     return OS_OK ;
3412   }
3413 }
3414 
3415 int os::naked_sleep() {
3416   // %% make the sleep time an integer flag. for now use 1 millisec.
3417   return os::sleep(Thread::current(), 1, false);
3418 }
3419 
3420 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3421 void os::infinite_sleep() {
3422   while (true) {    // sleep forever ...
3423     ::sleep(100);   // ... 100 seconds at a time
3424   }
3425 }
3426 
3427 // Used to convert frequent JVM_Yield() to nops
3428 bool os::dont_yield() {
3429   return DontYieldALot;
3430 }
3431 
3432 void os::yield() {
3433   sched_yield();
3434 }
3435 
3436 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3437 
3438 void os::yield_all(int attempts) {
3439   // Yields to all threads, including threads with lower priorities
3440   // Threads on Linux are all with same priority. The Solaris style
3441   // os::yield_all() with nanosleep(1ms) is not necessary.
3442   sched_yield();
3443 }
3444 
3445 // Called from the tight loops to possibly influence time-sharing heuristics
3446 void os::loop_breaker(int attempts) {
3447   os::yield_all(attempts);
3448 }
3449 
3450 ////////////////////////////////////////////////////////////////////////////////
3451 // thread priority support
3452 
3453 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3454 // only supports dynamic priority, static priority must be zero. For real-time
3455 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3456 // However, for large multi-threaded applications, SCHED_RR is not only slower
3457 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3458 // of 5 runs - Sep 2005).
3459 //
3460 // The following code actually changes the niceness of kernel-thread/LWP. It
3461 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3462 // not the entire user process, and user level threads are 1:1 mapped to kernel
3463 // threads. It has always been the case, but could change in the future. For
3464 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3465 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3466 
3467 int os::java_to_os_priority[CriticalPriority + 1] = {
3468   19,              // 0 Entry should never be used
3469 
3470    4,              // 1 MinPriority
3471    3,              // 2
3472    2,              // 3
3473 
3474    1,              // 4
3475    0,              // 5 NormPriority
3476   -1,              // 6
3477 
3478   -2,              // 7
3479   -3,              // 8
3480   -4,              // 9 NearMaxPriority
3481 
3482   -5,              // 10 MaxPriority
3483 
3484   -5               // 11 CriticalPriority
3485 };
3486 
3487 static int prio_init() {
3488   if (ThreadPriorityPolicy == 1) {
3489     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3490     // if effective uid is not root. Perhaps, a more elegant way of doing
3491     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3492     if (geteuid() != 0) {
3493       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3494         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3495       }
3496       ThreadPriorityPolicy = 0;
3497     }
3498   }
3499   if (UseCriticalJavaThreadPriority) {
3500     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3501   }
3502   return 0;
3503 }
3504 
3505 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3506   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3507 
3508   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3509   return (ret == 0) ? OS_OK : OS_ERR;
3510 }
3511 
3512 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3513   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3514     *priority_ptr = java_to_os_priority[NormPriority];
3515     return OS_OK;
3516   }
3517 
3518   errno = 0;
3519   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3520   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3521 }
3522 
3523 // Hint to the underlying OS that a task switch would not be good.
3524 // Void return because it's a hint and can fail.
3525 void os::hint_no_preempt() {}
3526 
3527 ////////////////////////////////////////////////////////////////////////////////
3528 // suspend/resume support
3529 
3530 //  the low-level signal-based suspend/resume support is a remnant from the
3531 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3532 //  within hotspot. Now there is a single use-case for this:
3533 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3534 //      that runs in the watcher thread.
3535 //  The remaining code is greatly simplified from the more general suspension
3536 //  code that used to be used.
3537 //
3538 //  The protocol is quite simple:
3539 //  - suspend:
3540 //      - sends a signal to the target thread
3541 //      - polls the suspend state of the osthread using a yield loop
3542 //      - target thread signal handler (SR_handler) sets suspend state
3543 //        and blocks in sigsuspend until continued
3544 //  - resume:
3545 //      - sets target osthread state to continue
3546 //      - sends signal to end the sigsuspend loop in the SR_handler
3547 //
3548 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3549 //
3550 
3551 static void resume_clear_context(OSThread *osthread) {
3552   osthread->set_ucontext(NULL);
3553   osthread->set_siginfo(NULL);
3554 
3555   // notify the suspend action is completed, we have now resumed
3556   osthread->sr.clear_suspended();
3557 }
3558 
3559 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3560   osthread->set_ucontext(context);
3561   osthread->set_siginfo(siginfo);
3562 }
3563 
3564 //
3565 // Handler function invoked when a thread's execution is suspended or
3566 // resumed. We have to be careful that only async-safe functions are
3567 // called here (Note: most pthread functions are not async safe and
3568 // should be avoided.)
3569 //
3570 // Note: sigwait() is a more natural fit than sigsuspend() from an
3571 // interface point of view, but sigwait() prevents the signal hander
3572 // from being run. libpthread would get very confused by not having
3573 // its signal handlers run and prevents sigwait()'s use with the
3574 // mutex granting granting signal.
3575 //
3576 // Currently only ever called on the VMThread
3577 //
3578 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3579   // Save and restore errno to avoid confusing native code with EINTR
3580   // after sigsuspend.
3581   int old_errno = errno;
3582 
3583   Thread* thread = Thread::current();
3584   OSThread* osthread = thread->osthread();
3585   assert(thread->is_VM_thread(), "Must be VMThread");
3586   // read current suspend action
3587   int action = osthread->sr.suspend_action();
3588   if (action == os::Linux::SuspendResume::SR_SUSPEND) {
3589     suspend_save_context(osthread, siginfo, context);
3590 
3591     // Notify the suspend action is about to be completed. do_suspend()
3592     // waits until SR_SUSPENDED is set and then returns. We will wait
3593     // here for a resume signal and that completes the suspend-other
3594     // action. do_suspend/do_resume is always called as a pair from
3595     // the same thread - so there are no races
3596 
3597     // notify the caller
3598     osthread->sr.set_suspended();
3599 
3600     sigset_t suspend_set;  // signals for sigsuspend()
3601 
3602     // get current set of blocked signals and unblock resume signal
3603     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3604     sigdelset(&suspend_set, SR_signum);
3605 
3606     // wait here until we are resumed
3607     do {
3608       sigsuspend(&suspend_set);
3609       // ignore all returns until we get a resume signal
3610     } while (osthread->sr.suspend_action() != os::Linux::SuspendResume::SR_CONTINUE);
3611 
3612     resume_clear_context(osthread);
3613 
3614   } else {
3615     assert(action == os::Linux::SuspendResume::SR_CONTINUE, "unexpected sr action");
3616     // nothing special to do - just leave the handler
3617   }
3618 
3619   errno = old_errno;
3620 }
3621 
3622 
3623 static int SR_initialize() {
3624   struct sigaction act;
3625   char *s;
3626   /* Get signal number to use for suspend/resume */
3627   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3628     int sig = ::strtol(s, 0, 10);
3629     if (sig > 0 || sig < _NSIG) {
3630         SR_signum = sig;
3631     }
3632   }
3633 
3634   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3635         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3636 
3637   sigemptyset(&SR_sigset);
3638   sigaddset(&SR_sigset, SR_signum);
3639 
3640   /* Set up signal handler for suspend/resume */
3641   act.sa_flags = SA_RESTART|SA_SIGINFO;
3642   act.sa_handler = (void (*)(int)) SR_handler;
3643 
3644   // SR_signum is blocked by default.
3645   // 4528190 - We also need to block pthread restart signal (32 on all
3646   // supported Linux platforms). Note that LinuxThreads need to block
3647   // this signal for all threads to work properly. So we don't have
3648   // to use hard-coded signal number when setting up the mask.
3649   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3650 
3651   if (sigaction(SR_signum, &act, 0) == -1) {
3652     return -1;
3653   }
3654 
3655   // Save signal flag
3656   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3657   return 0;
3658 }
3659 
3660 
3661 // returns true on success and false on error - really an error is fatal
3662 // but this seems the normal response to library errors
3663 static bool do_suspend(OSThread* osthread) {
3664   // mark as suspended and send signal
3665   osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_SUSPEND);
3666   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3667   assert_status(status == 0, status, "pthread_kill");
3668 
3669   // check status and wait until notified of suspension
3670   if (status == 0) {
3671     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3672       os::yield_all(i);
3673     }
3674     osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3675     return true;
3676   }
3677   else {
3678     osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3679     return false;
3680   }
3681 }
3682 
3683 static void do_resume(OSThread* osthread) {
3684   assert(osthread->sr.is_suspended(), "thread should be suspended");
3685   osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_CONTINUE);
3686 
3687   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3688   assert_status(status == 0, status, "pthread_kill");
3689   // check status and wait unit notified of resumption
3690   if (status == 0) {
3691     for (int i = 0; osthread->sr.is_suspended(); i++) {
3692       os::yield_all(i);
3693     }
3694   }
3695   osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3696 }
3697 
3698 ////////////////////////////////////////////////////////////////////////////////
3699 // interrupt support
3700 
3701 void os::interrupt(Thread* thread) {
3702   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3703     "possibility of dangling Thread pointer");
3704 
3705   OSThread* osthread = thread->osthread();
3706 
3707   if (!osthread->interrupted()) {
3708     osthread->set_interrupted(true);
3709     // More than one thread can get here with the same value of osthread,
3710     // resulting in multiple notifications.  We do, however, want the store
3711     // to interrupted() to be visible to other threads before we execute unpark().
3712     OrderAccess::fence();
3713     ParkEvent * const slp = thread->_SleepEvent ;
3714     if (slp != NULL) slp->unpark() ;
3715   }
3716 
3717   // For JSR166. Unpark even if interrupt status already was set
3718   if (thread->is_Java_thread())
3719     ((JavaThread*)thread)->parker()->unpark();
3720 
3721   ParkEvent * ev = thread->_ParkEvent ;
3722   if (ev != NULL) ev->unpark() ;
3723 
3724 }
3725 
3726 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3727   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3728     "possibility of dangling Thread pointer");
3729 
3730   OSThread* osthread = thread->osthread();
3731 
3732   bool interrupted = osthread->interrupted();
3733 
3734   if (interrupted && clear_interrupted) {
3735     osthread->set_interrupted(false);
3736     // consider thread->_SleepEvent->reset() ... optional optimization
3737   }
3738 
3739   return interrupted;
3740 }
3741 
3742 ///////////////////////////////////////////////////////////////////////////////////
3743 // signal handling (except suspend/resume)
3744 
3745 // This routine may be used by user applications as a "hook" to catch signals.
3746 // The user-defined signal handler must pass unrecognized signals to this
3747 // routine, and if it returns true (non-zero), then the signal handler must
3748 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3749 // routine will never retun false (zero), but instead will execute a VM panic
3750 // routine kill the process.
3751 //
3752 // If this routine returns false, it is OK to call it again.  This allows
3753 // the user-defined signal handler to perform checks either before or after
3754 // the VM performs its own checks.  Naturally, the user code would be making
3755 // a serious error if it tried to handle an exception (such as a null check
3756 // or breakpoint) that the VM was generating for its own correct operation.
3757 //
3758 // This routine may recognize any of the following kinds of signals:
3759 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3760 // It should be consulted by handlers for any of those signals.
3761 //
3762 // The caller of this routine must pass in the three arguments supplied
3763 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3764 // field of the structure passed to sigaction().  This routine assumes that
3765 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3766 //
3767 // Note that the VM will print warnings if it detects conflicting signal
3768 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3769 //
3770 extern "C" JNIEXPORT int
3771 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3772                         void* ucontext, int abort_if_unrecognized);
3773 
3774 void signalHandler(int sig, siginfo_t* info, void* uc) {
3775   assert(info != NULL && uc != NULL, "it must be old kernel");
3776   int orig_errno = errno;  // Preserve errno value over signal handler.
3777   JVM_handle_linux_signal(sig, info, uc, true);
3778   errno = orig_errno;
3779 }
3780 
3781 
3782 // This boolean allows users to forward their own non-matching signals
3783 // to JVM_handle_linux_signal, harmlessly.
3784 bool os::Linux::signal_handlers_are_installed = false;
3785 
3786 // For signal-chaining
3787 struct sigaction os::Linux::sigact[MAXSIGNUM];
3788 unsigned int os::Linux::sigs = 0;
3789 bool os::Linux::libjsig_is_loaded = false;
3790 typedef struct sigaction *(*get_signal_t)(int);
3791 get_signal_t os::Linux::get_signal_action = NULL;
3792 
3793 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3794   struct sigaction *actp = NULL;
3795 
3796   if (libjsig_is_loaded) {
3797     // Retrieve the old signal handler from libjsig
3798     actp = (*get_signal_action)(sig);
3799   }
3800   if (actp == NULL) {
3801     // Retrieve the preinstalled signal handler from jvm
3802     actp = get_preinstalled_handler(sig);
3803   }
3804 
3805   return actp;
3806 }
3807 
3808 static bool call_chained_handler(struct sigaction *actp, int sig,
3809                                  siginfo_t *siginfo, void *context) {
3810   // Call the old signal handler
3811   if (actp->sa_handler == SIG_DFL) {
3812     // It's more reasonable to let jvm treat it as an unexpected exception
3813     // instead of taking the default action.
3814     return false;
3815   } else if (actp->sa_handler != SIG_IGN) {
3816     if ((actp->sa_flags & SA_NODEFER) == 0) {
3817       // automaticlly block the signal
3818       sigaddset(&(actp->sa_mask), sig);
3819     }
3820 
3821     sa_handler_t hand;
3822     sa_sigaction_t sa;
3823     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3824     // retrieve the chained handler
3825     if (siginfo_flag_set) {
3826       sa = actp->sa_sigaction;
3827     } else {
3828       hand = actp->sa_handler;
3829     }
3830 
3831     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3832       actp->sa_handler = SIG_DFL;
3833     }
3834 
3835     // try to honor the signal mask
3836     sigset_t oset;
3837     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3838 
3839     // call into the chained handler
3840     if (siginfo_flag_set) {
3841       (*sa)(sig, siginfo, context);
3842     } else {
3843       (*hand)(sig);
3844     }
3845 
3846     // restore the signal mask
3847     pthread_sigmask(SIG_SETMASK, &oset, 0);
3848   }
3849   // Tell jvm's signal handler the signal is taken care of.
3850   return true;
3851 }
3852 
3853 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3854   bool chained = false;
3855   // signal-chaining
3856   if (UseSignalChaining) {
3857     struct sigaction *actp = get_chained_signal_action(sig);
3858     if (actp != NULL) {
3859       chained = call_chained_handler(actp, sig, siginfo, context);
3860     }
3861   }
3862   return chained;
3863 }
3864 
3865 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3866   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3867     return &sigact[sig];
3868   }
3869   return NULL;
3870 }
3871 
3872 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3873   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3874   sigact[sig] = oldAct;
3875   sigs |= (unsigned int)1 << sig;
3876 }
3877 
3878 // for diagnostic
3879 int os::Linux::sigflags[MAXSIGNUM];
3880 
3881 int os::Linux::get_our_sigflags(int sig) {
3882   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3883   return sigflags[sig];
3884 }
3885 
3886 void os::Linux::set_our_sigflags(int sig, int flags) {
3887   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3888   sigflags[sig] = flags;
3889 }
3890 
3891 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3892   // Check for overwrite.
3893   struct sigaction oldAct;
3894   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3895 
3896   void* oldhand = oldAct.sa_sigaction
3897                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3898                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3899   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3900       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3901       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3902     if (AllowUserSignalHandlers || !set_installed) {
3903       // Do not overwrite; user takes responsibility to forward to us.
3904       return;
3905     } else if (UseSignalChaining) {
3906       // save the old handler in jvm
3907       save_preinstalled_handler(sig, oldAct);
3908       // libjsig also interposes the sigaction() call below and saves the
3909       // old sigaction on it own.
3910     } else {
3911       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3912                     "%#lx for signal %d.", (long)oldhand, sig));
3913     }
3914   }
3915 
3916   struct sigaction sigAct;
3917   sigfillset(&(sigAct.sa_mask));
3918   sigAct.sa_handler = SIG_DFL;
3919   if (!set_installed) {
3920     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3921   } else {
3922     sigAct.sa_sigaction = signalHandler;
3923     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3924   }
3925   // Save flags, which are set by ours
3926   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3927   sigflags[sig] = sigAct.sa_flags;
3928 
3929   int ret = sigaction(sig, &sigAct, &oldAct);
3930   assert(ret == 0, "check");
3931 
3932   void* oldhand2  = oldAct.sa_sigaction
3933                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3934                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3935   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3936 }
3937 
3938 // install signal handlers for signals that HotSpot needs to
3939 // handle in order to support Java-level exception handling.
3940 
3941 void os::Linux::install_signal_handlers() {
3942   if (!signal_handlers_are_installed) {
3943     signal_handlers_are_installed = true;
3944 
3945     // signal-chaining
3946     typedef void (*signal_setting_t)();
3947     signal_setting_t begin_signal_setting = NULL;
3948     signal_setting_t end_signal_setting = NULL;
3949     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3950                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3951     if (begin_signal_setting != NULL) {
3952       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3953                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3954       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3955                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3956       libjsig_is_loaded = true;
3957       assert(UseSignalChaining, "should enable signal-chaining");
3958     }
3959     if (libjsig_is_loaded) {
3960       // Tell libjsig jvm is setting signal handlers
3961       (*begin_signal_setting)();
3962     }
3963 
3964     set_signal_handler(SIGSEGV, true);
3965     set_signal_handler(SIGPIPE, true);
3966     set_signal_handler(SIGBUS, true);
3967     set_signal_handler(SIGILL, true);
3968     set_signal_handler(SIGFPE, true);
3969     set_signal_handler(SIGXFSZ, true);
3970 
3971     if (libjsig_is_loaded) {
3972       // Tell libjsig jvm finishes setting signal handlers
3973       (*end_signal_setting)();
3974     }
3975 
3976     // We don't activate signal checker if libjsig is in place, we trust ourselves
3977     // and if UserSignalHandler is installed all bets are off.
3978     // Log that signal checking is off only if -verbose:jni is specified.
3979     if (CheckJNICalls) {
3980       if (libjsig_is_loaded) {
3981         if (PrintJNIResolving) {
3982           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3983         }
3984         check_signals = false;
3985       }
3986       if (AllowUserSignalHandlers) {
3987         if (PrintJNIResolving) {
3988           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3989         }
3990         check_signals = false;
3991       }
3992     }
3993   }
3994 }
3995 
3996 // This is the fastest way to get thread cpu time on Linux.
3997 // Returns cpu time (user+sys) for any thread, not only for current.
3998 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3999 // It might work on 2.6.10+ with a special kernel/glibc patch.
4000 // For reference, please, see IEEE Std 1003.1-2004:
4001 //   http://www.unix.org/single_unix_specification
4002 
4003 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4004   struct timespec tp;
4005   int rc = os::Linux::clock_gettime(clockid, &tp);
4006   assert(rc == 0, "clock_gettime is expected to return 0 code");
4007 
4008   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4009 }
4010 
4011 /////
4012 // glibc on Linux platform uses non-documented flag
4013 // to indicate, that some special sort of signal
4014 // trampoline is used.
4015 // We will never set this flag, and we should
4016 // ignore this flag in our diagnostic
4017 #ifdef SIGNIFICANT_SIGNAL_MASK
4018 #undef SIGNIFICANT_SIGNAL_MASK
4019 #endif
4020 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4021 
4022 static const char* get_signal_handler_name(address handler,
4023                                            char* buf, int buflen) {
4024   int offset;
4025   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4026   if (found) {
4027     // skip directory names
4028     const char *p1, *p2;
4029     p1 = buf;
4030     size_t len = strlen(os::file_separator());
4031     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4032     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4033   } else {
4034     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4035   }
4036   return buf;
4037 }
4038 
4039 static void print_signal_handler(outputStream* st, int sig,
4040                                  char* buf, size_t buflen) {
4041   struct sigaction sa;
4042 
4043   sigaction(sig, NULL, &sa);
4044 
4045   // See comment for SIGNIFICANT_SIGNAL_MASK define
4046   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4047 
4048   st->print("%s: ", os::exception_name(sig, buf, buflen));
4049 
4050   address handler = (sa.sa_flags & SA_SIGINFO)
4051     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4052     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4053 
4054   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4055     st->print("SIG_DFL");
4056   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4057     st->print("SIG_IGN");
4058   } else {
4059     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4060   }
4061 
4062   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4063 
4064   address rh = VMError::get_resetted_sighandler(sig);
4065   // May be, handler was resetted by VMError?
4066   if(rh != NULL) {
4067     handler = rh;
4068     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4069   }
4070 
4071   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
4072 
4073   // Check: is it our handler?
4074   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4075      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4076     // It is our signal handler
4077     // check for flags, reset system-used one!
4078     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4079       st->print(
4080                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4081                 os::Linux::get_our_sigflags(sig));
4082     }
4083   }
4084   st->cr();
4085 }
4086 
4087 
4088 #define DO_SIGNAL_CHECK(sig) \
4089   if (!sigismember(&check_signal_done, sig)) \
4090     os::Linux::check_signal_handler(sig)
4091 
4092 // This method is a periodic task to check for misbehaving JNI applications
4093 // under CheckJNI, we can add any periodic checks here
4094 
4095 void os::run_periodic_checks() {
4096 
4097   if (check_signals == false) return;
4098 
4099   // SEGV and BUS if overridden could potentially prevent
4100   // generation of hs*.log in the event of a crash, debugging
4101   // such a case can be very challenging, so we absolutely
4102   // check the following for a good measure:
4103   DO_SIGNAL_CHECK(SIGSEGV);
4104   DO_SIGNAL_CHECK(SIGILL);
4105   DO_SIGNAL_CHECK(SIGFPE);
4106   DO_SIGNAL_CHECK(SIGBUS);
4107   DO_SIGNAL_CHECK(SIGPIPE);
4108   DO_SIGNAL_CHECK(SIGXFSZ);
4109 
4110 
4111   // ReduceSignalUsage allows the user to override these handlers
4112   // see comments at the very top and jvm_solaris.h
4113   if (!ReduceSignalUsage) {
4114     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4115     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4116     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4117     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4118   }
4119 
4120   DO_SIGNAL_CHECK(SR_signum);
4121   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4122 }
4123 
4124 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4125 
4126 static os_sigaction_t os_sigaction = NULL;
4127 
4128 void os::Linux::check_signal_handler(int sig) {
4129   char buf[O_BUFLEN];
4130   address jvmHandler = NULL;
4131 
4132 
4133   struct sigaction act;
4134   if (os_sigaction == NULL) {
4135     // only trust the default sigaction, in case it has been interposed
4136     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4137     if (os_sigaction == NULL) return;
4138   }
4139 
4140   os_sigaction(sig, (struct sigaction*)NULL, &act);
4141 
4142 
4143   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4144 
4145   address thisHandler = (act.sa_flags & SA_SIGINFO)
4146     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4147     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4148 
4149 
4150   switch(sig) {
4151   case SIGSEGV:
4152   case SIGBUS:
4153   case SIGFPE:
4154   case SIGPIPE:
4155   case SIGILL:
4156   case SIGXFSZ:
4157     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4158     break;
4159 
4160   case SHUTDOWN1_SIGNAL:
4161   case SHUTDOWN2_SIGNAL:
4162   case SHUTDOWN3_SIGNAL:
4163   case BREAK_SIGNAL:
4164     jvmHandler = (address)user_handler();
4165     break;
4166 
4167   case INTERRUPT_SIGNAL:
4168     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4169     break;
4170 
4171   default:
4172     if (sig == SR_signum) {
4173       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4174     } else {
4175       return;
4176     }
4177     break;
4178   }
4179 
4180   if (thisHandler != jvmHandler) {
4181     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4182     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4183     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4184     // No need to check this sig any longer
4185     sigaddset(&check_signal_done, sig);
4186   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4187     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4188     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4189     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4190     // No need to check this sig any longer
4191     sigaddset(&check_signal_done, sig);
4192   }
4193 
4194   // Dump all the signal
4195   if (sigismember(&check_signal_done, sig)) {
4196     print_signal_handlers(tty, buf, O_BUFLEN);
4197   }
4198 }
4199 
4200 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4201 
4202 extern bool signal_name(int signo, char* buf, size_t len);
4203 
4204 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4205   if (0 < exception_code && exception_code <= SIGRTMAX) {
4206     // signal
4207     if (!signal_name(exception_code, buf, size)) {
4208       jio_snprintf(buf, size, "SIG%d", exception_code);
4209     }
4210     return buf;
4211   } else {
4212     return NULL;
4213   }
4214 }
4215 
4216 // this is called _before_ the most of global arguments have been parsed
4217 void os::init(void) {
4218   char dummy;   /* used to get a guess on initial stack address */
4219 //  first_hrtime = gethrtime();
4220 
4221   // With LinuxThreads the JavaMain thread pid (primordial thread)
4222   // is different than the pid of the java launcher thread.
4223   // So, on Linux, the launcher thread pid is passed to the VM
4224   // via the sun.java.launcher.pid property.
4225   // Use this property instead of getpid() if it was correctly passed.
4226   // See bug 6351349.
4227   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4228 
4229   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4230 
4231   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4232 
4233   init_random(1234567);
4234 
4235   ThreadCritical::initialize();
4236 
4237   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4238   if (Linux::page_size() == -1) {
4239     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4240                   strerror(errno)));
4241   }
4242   init_page_sizes((size_t) Linux::page_size());
4243 
4244   Linux::initialize_system_info();
4245 
4246   // main_thread points to the aboriginal thread
4247   Linux::_main_thread = pthread_self();
4248 
4249   Linux::clock_init();
4250   initial_time_count = os::elapsed_counter();
4251   pthread_mutex_init(&dl_mutex, NULL);
4252 }
4253 
4254 // To install functions for atexit system call
4255 extern "C" {
4256   static void perfMemory_exit_helper() {
4257     perfMemory_exit();
4258   }
4259 }
4260 
4261 // this is called _after_ the global arguments have been parsed
4262 jint os::init_2(void)
4263 {
4264   Linux::fast_thread_clock_init();
4265 
4266   // Allocate a single page and mark it as readable for safepoint polling
4267   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4268   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4269 
4270   os::set_polling_page( polling_page );
4271 
4272 #ifndef PRODUCT
4273   if(Verbose && PrintMiscellaneous)
4274     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4275 #endif
4276 
4277   if (!UseMembar) {
4278     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4279     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4280     os::set_memory_serialize_page( mem_serialize_page );
4281 
4282 #ifndef PRODUCT
4283     if(Verbose && PrintMiscellaneous)
4284       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4285 #endif
4286   }
4287 
4288   os::large_page_init();
4289 
4290   // initialize suspend/resume support - must do this before signal_sets_init()
4291   if (SR_initialize() != 0) {
4292     perror("SR_initialize failed");
4293     return JNI_ERR;
4294   }
4295 
4296   Linux::signal_sets_init();
4297   Linux::install_signal_handlers();
4298 
4299   // Check minimum allowable stack size for thread creation and to initialize
4300   // the java system classes, including StackOverflowError - depends on page
4301   // size.  Add a page for compiler2 recursion in main thread.
4302   // Add in 2*BytesPerWord times page size to account for VM stack during
4303   // class initialization depending on 32 or 64 bit VM.
4304   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4305             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4306                     2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4307 
4308   size_t threadStackSizeInBytes = ThreadStackSize * K;
4309   if (threadStackSizeInBytes != 0 &&
4310       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4311         tty->print_cr("\nThe stack size specified is too small, "
4312                       "Specify at least %dk",
4313                       os::Linux::min_stack_allowed/ K);
4314         return JNI_ERR;
4315   }
4316 
4317   // Make the stack size a multiple of the page size so that
4318   // the yellow/red zones can be guarded.
4319   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4320         vm_page_size()));
4321 
4322   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4323 
4324   Linux::libpthread_init();
4325   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4326      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4327           Linux::glibc_version(), Linux::libpthread_version(),
4328           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4329   }
4330 
4331   if (UseNUMA) {
4332     if (!Linux::libnuma_init()) {
4333       UseNUMA = false;
4334     } else {
4335       if ((Linux::numa_max_node() < 1)) {
4336         // There's only one node(they start from 0), disable NUMA.
4337         UseNUMA = false;
4338       }
4339     }
4340     // With SHM large pages we cannot uncommit a page, so there's not way
4341     // we can make the adaptive lgrp chunk resizing work. If the user specified
4342     // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4343     // disable adaptive resizing.
4344     if (UseNUMA && UseLargePages && UseSHM) {
4345       if (!FLAG_IS_DEFAULT(UseNUMA)) {
4346         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4347           UseLargePages = false;
4348         } else {
4349           warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4350           UseAdaptiveSizePolicy = false;
4351           UseAdaptiveNUMAChunkSizing = false;
4352         }
4353       } else {
4354         UseNUMA = false;
4355       }
4356     }
4357     if (!UseNUMA && ForceNUMA) {
4358       UseNUMA = true;
4359     }
4360   }
4361 
4362   if (MaxFDLimit) {
4363     // set the number of file descriptors to max. print out error
4364     // if getrlimit/setrlimit fails but continue regardless.
4365     struct rlimit nbr_files;
4366     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4367     if (status != 0) {
4368       if (PrintMiscellaneous && (Verbose || WizardMode))
4369         perror("os::init_2 getrlimit failed");
4370     } else {
4371       nbr_files.rlim_cur = nbr_files.rlim_max;
4372       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4373       if (status != 0) {
4374         if (PrintMiscellaneous && (Verbose || WizardMode))
4375           perror("os::init_2 setrlimit failed");
4376       }
4377     }
4378   }
4379 
4380   // Initialize lock used to serialize thread creation (see os::create_thread)
4381   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4382 
4383   // at-exit methods are called in the reverse order of their registration.
4384   // atexit functions are called on return from main or as a result of a
4385   // call to exit(3C). There can be only 32 of these functions registered
4386   // and atexit() does not set errno.
4387 
4388   if (PerfAllowAtExitRegistration) {
4389     // only register atexit functions if PerfAllowAtExitRegistration is set.
4390     // atexit functions can be delayed until process exit time, which
4391     // can be problematic for embedded VM situations. Embedded VMs should
4392     // call DestroyJavaVM() to assure that VM resources are released.
4393 
4394     // note: perfMemory_exit_helper atexit function may be removed in
4395     // the future if the appropriate cleanup code can be added to the
4396     // VM_Exit VMOperation's doit method.
4397     if (atexit(perfMemory_exit_helper) != 0) {
4398       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4399     }
4400   }
4401 
4402   // initialize thread priority policy
4403   prio_init();
4404 
4405   return JNI_OK;
4406 }
4407 
4408 // this is called at the end of vm_initialization
4409 void os::init_3(void)
4410 {
4411 #ifdef JAVASE_EMBEDDED
4412   // Start the MemNotifyThread
4413   if (LowMemoryProtection) {
4414     MemNotifyThread::start();
4415   }
4416   return;
4417 #endif
4418 }
4419 
4420 // Mark the polling page as unreadable
4421 void os::make_polling_page_unreadable(void) {
4422   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4423     fatal("Could not disable polling page");
4424 };
4425 
4426 // Mark the polling page as readable
4427 void os::make_polling_page_readable(void) {
4428   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4429     fatal("Could not enable polling page");
4430   }
4431 };
4432 
4433 int os::active_processor_count() {
4434   // Linux doesn't yet have a (official) notion of processor sets,
4435   // so just return the number of online processors.
4436   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4437   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4438   return online_cpus;
4439 }
4440 
4441 void os::set_native_thread_name(const char *name) {
4442   // Not yet implemented.
4443   return;
4444 }
4445 
4446 bool os::distribute_processes(uint length, uint* distribution) {
4447   // Not yet implemented.
4448   return false;
4449 }
4450 
4451 bool os::bind_to_processor(uint processor_id) {
4452   // Not yet implemented.
4453   return false;
4454 }
4455 
4456 ///
4457 
4458 // Suspends the target using the signal mechanism and then grabs the PC before
4459 // resuming the target. Used by the flat-profiler only
4460 ExtendedPC os::get_thread_pc(Thread* thread) {
4461   // Make sure that it is called by the watcher for the VMThread
4462   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4463   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4464 
4465   ExtendedPC epc;
4466 
4467   OSThread* osthread = thread->osthread();
4468   if (do_suspend(osthread)) {
4469     if (osthread->ucontext() != NULL) {
4470       epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4471     } else {
4472       // NULL context is unexpected, double-check this is the VMThread
4473       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4474     }
4475     do_resume(osthread);
4476   }
4477   // failure means pthread_kill failed for some reason - arguably this is
4478   // a fatal problem, but such problems are ignored elsewhere
4479 
4480   return epc;
4481 }
4482 
4483 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4484 {
4485    if (is_NPTL()) {
4486       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4487    } else {
4488       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4489       // word back to default 64bit precision if condvar is signaled. Java
4490       // wants 53bit precision.  Save and restore current value.
4491       int fpu = get_fpu_control_word();
4492       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4493       set_fpu_control_word(fpu);
4494       return status;
4495    }
4496 }
4497 
4498 ////////////////////////////////////////////////////////////////////////////////
4499 // debug support
4500 
4501 bool os::find(address addr, outputStream* st) {
4502   Dl_info dlinfo;
4503   memset(&dlinfo, 0, sizeof(dlinfo));
4504   if (dladdr(addr, &dlinfo)) {
4505     st->print(PTR_FORMAT ": ", addr);
4506     if (dlinfo.dli_sname != NULL) {
4507       st->print("%s+%#x", dlinfo.dli_sname,
4508                  addr - (intptr_t)dlinfo.dli_saddr);
4509     } else if (dlinfo.dli_fname) {
4510       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4511     } else {
4512       st->print("<absolute address>");
4513     }
4514     if (dlinfo.dli_fname) {
4515       st->print(" in %s", dlinfo.dli_fname);
4516     }
4517     if (dlinfo.dli_fbase) {
4518       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4519     }
4520     st->cr();
4521 
4522     if (Verbose) {
4523       // decode some bytes around the PC
4524       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4525       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4526       address       lowest = (address) dlinfo.dli_sname;
4527       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4528       if (begin < lowest)  begin = lowest;
4529       Dl_info dlinfo2;
4530       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4531           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4532         end = (address) dlinfo2.dli_saddr;
4533       Disassembler::decode(begin, end, st);
4534     }
4535     return true;
4536   }
4537   return false;
4538 }
4539 
4540 ////////////////////////////////////////////////////////////////////////////////
4541 // misc
4542 
4543 // This does not do anything on Linux. This is basically a hook for being
4544 // able to use structured exception handling (thread-local exception filters)
4545 // on, e.g., Win32.
4546 void
4547 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4548                          JavaCallArguments* args, Thread* thread) {
4549   f(value, method, args, thread);
4550 }
4551 
4552 void os::print_statistics() {
4553 }
4554 
4555 int os::message_box(const char* title, const char* message) {
4556   int i;
4557   fdStream err(defaultStream::error_fd());
4558   for (i = 0; i < 78; i++) err.print_raw("=");
4559   err.cr();
4560   err.print_raw_cr(title);
4561   for (i = 0; i < 78; i++) err.print_raw("-");
4562   err.cr();
4563   err.print_raw_cr(message);
4564   for (i = 0; i < 78; i++) err.print_raw("=");
4565   err.cr();
4566 
4567   char buf[16];
4568   // Prevent process from exiting upon "read error" without consuming all CPU
4569   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4570 
4571   return buf[0] == 'y' || buf[0] == 'Y';
4572 }
4573 
4574 int os::stat(const char *path, struct stat *sbuf) {
4575   char pathbuf[MAX_PATH];
4576   if (strlen(path) > MAX_PATH - 1) {
4577     errno = ENAMETOOLONG;
4578     return -1;
4579   }
4580   os::native_path(strcpy(pathbuf, path));
4581   return ::stat(pathbuf, sbuf);
4582 }
4583 
4584 bool os::check_heap(bool force) {
4585   return true;
4586 }
4587 
4588 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4589   return ::vsnprintf(buf, count, format, args);
4590 }
4591 
4592 // Is a (classpath) directory empty?
4593 bool os::dir_is_empty(const char* path) {
4594   DIR *dir = NULL;
4595   struct dirent *ptr;
4596 
4597   dir = opendir(path);
4598   if (dir == NULL) return true;
4599 
4600   /* Scan the directory */
4601   bool result = true;
4602   char buf[sizeof(struct dirent) + MAX_PATH];
4603   while (result && (ptr = ::readdir(dir)) != NULL) {
4604     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4605       result = false;
4606     }
4607   }
4608   closedir(dir);
4609   return result;
4610 }
4611 
4612 // This code originates from JDK's sysOpen and open64_w
4613 // from src/solaris/hpi/src/system_md.c
4614 
4615 #ifndef O_DELETE
4616 #define O_DELETE 0x10000
4617 #endif
4618 
4619 // Open a file. Unlink the file immediately after open returns
4620 // if the specified oflag has the O_DELETE flag set.
4621 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4622 
4623 int os::open(const char *path, int oflag, int mode) {
4624 
4625   if (strlen(path) > MAX_PATH - 1) {
4626     errno = ENAMETOOLONG;
4627     return -1;
4628   }
4629   int fd;
4630   int o_delete = (oflag & O_DELETE);
4631   oflag = oflag & ~O_DELETE;
4632 
4633   fd = ::open64(path, oflag, mode);
4634   if (fd == -1) return -1;
4635 
4636   //If the open succeeded, the file might still be a directory
4637   {
4638     struct stat64 buf64;
4639     int ret = ::fstat64(fd, &buf64);
4640     int st_mode = buf64.st_mode;
4641 
4642     if (ret != -1) {
4643       if ((st_mode & S_IFMT) == S_IFDIR) {
4644         errno = EISDIR;
4645         ::close(fd);
4646         return -1;
4647       }
4648     } else {
4649       ::close(fd);
4650       return -1;
4651     }
4652   }
4653 
4654     /*
4655      * All file descriptors that are opened in the JVM and not
4656      * specifically destined for a subprocess should have the
4657      * close-on-exec flag set.  If we don't set it, then careless 3rd
4658      * party native code might fork and exec without closing all
4659      * appropriate file descriptors (e.g. as we do in closeDescriptors in
4660      * UNIXProcess.c), and this in turn might:
4661      *
4662      * - cause end-of-file to fail to be detected on some file
4663      *   descriptors, resulting in mysterious hangs, or
4664      *
4665      * - might cause an fopen in the subprocess to fail on a system
4666      *   suffering from bug 1085341.
4667      *
4668      * (Yes, the default setting of the close-on-exec flag is a Unix
4669      * design flaw)
4670      *
4671      * See:
4672      * 1085341: 32-bit stdio routines should support file descriptors >255
4673      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4674      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4675      */
4676 #ifdef FD_CLOEXEC
4677     {
4678         int flags = ::fcntl(fd, F_GETFD);
4679         if (flags != -1)
4680             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4681     }
4682 #endif
4683 
4684   if (o_delete != 0) {
4685     ::unlink(path);
4686   }
4687   return fd;
4688 }
4689 
4690 
4691 // create binary file, rewriting existing file if required
4692 int os::create_binary_file(const char* path, bool rewrite_existing) {
4693   int oflags = O_WRONLY | O_CREAT;
4694   if (!rewrite_existing) {
4695     oflags |= O_EXCL;
4696   }
4697   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4698 }
4699 
4700 // return current position of file pointer
4701 jlong os::current_file_offset(int fd) {
4702   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4703 }
4704 
4705 // move file pointer to the specified offset
4706 jlong os::seek_to_file_offset(int fd, jlong offset) {
4707   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4708 }
4709 
4710 // This code originates from JDK's sysAvailable
4711 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4712 
4713 int os::available(int fd, jlong *bytes) {
4714   jlong cur, end;
4715   int mode;
4716   struct stat64 buf64;
4717 
4718   if (::fstat64(fd, &buf64) >= 0) {
4719     mode = buf64.st_mode;
4720     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4721       /*
4722       * XXX: is the following call interruptible? If so, this might
4723       * need to go through the INTERRUPT_IO() wrapper as for other
4724       * blocking, interruptible calls in this file.
4725       */
4726       int n;
4727       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4728         *bytes = n;
4729         return 1;
4730       }
4731     }
4732   }
4733   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4734     return 0;
4735   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4736     return 0;
4737   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4738     return 0;
4739   }
4740   *bytes = end - cur;
4741   return 1;
4742 }
4743 
4744 int os::socket_available(int fd, jint *pbytes) {
4745   // Linux doc says EINTR not returned, unlike Solaris
4746   int ret = ::ioctl(fd, FIONREAD, pbytes);
4747 
4748   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4749   // is expected to return 0 on failure and 1 on success to the jdk.
4750   return (ret < 0) ? 0 : 1;
4751 }
4752 
4753 // Map a block of memory.
4754 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4755                      char *addr, size_t bytes, bool read_only,
4756                      bool allow_exec) {
4757   int prot;
4758   int flags = MAP_PRIVATE;
4759 
4760   if (read_only) {
4761     prot = PROT_READ;
4762   } else {
4763     prot = PROT_READ | PROT_WRITE;
4764   }
4765 
4766   if (allow_exec) {
4767     prot |= PROT_EXEC;
4768   }
4769 
4770   if (addr != NULL) {
4771     flags |= MAP_FIXED;
4772   }
4773 
4774   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4775                                      fd, file_offset);
4776   if (mapped_address == MAP_FAILED) {
4777     return NULL;
4778   }
4779   return mapped_address;
4780 }
4781 
4782 
4783 // Remap a block of memory.
4784 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4785                        char *addr, size_t bytes, bool read_only,
4786                        bool allow_exec) {
4787   // same as map_memory() on this OS
4788   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4789                         allow_exec);
4790 }
4791 
4792 
4793 // Unmap a block of memory.
4794 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4795   return munmap(addr, bytes) == 0;
4796 }
4797 
4798 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4799 
4800 static clockid_t thread_cpu_clockid(Thread* thread) {
4801   pthread_t tid = thread->osthread()->pthread_id();
4802   clockid_t clockid;
4803 
4804   // Get thread clockid
4805   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4806   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4807   return clockid;
4808 }
4809 
4810 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4811 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4812 // of a thread.
4813 //
4814 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4815 // the fast estimate available on the platform.
4816 
4817 jlong os::current_thread_cpu_time() {
4818   if (os::Linux::supports_fast_thread_cpu_time()) {
4819     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4820   } else {
4821     // return user + sys since the cost is the same
4822     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4823   }
4824 }
4825 
4826 jlong os::thread_cpu_time(Thread* thread) {
4827   // consistent with what current_thread_cpu_time() returns
4828   if (os::Linux::supports_fast_thread_cpu_time()) {
4829     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4830   } else {
4831     return slow_thread_cpu_time(thread, true /* user + sys */);
4832   }
4833 }
4834 
4835 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4836   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4837     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4838   } else {
4839     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4840   }
4841 }
4842 
4843 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4844   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4845     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4846   } else {
4847     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4848   }
4849 }
4850 
4851 //
4852 //  -1 on error.
4853 //
4854 
4855 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4856   static bool proc_task_unchecked = true;
4857   static const char *proc_stat_path = "/proc/%d/stat";
4858   pid_t  tid = thread->osthread()->thread_id();
4859   char *s;
4860   char stat[2048];
4861   int statlen;
4862   char proc_name[64];
4863   int count;
4864   long sys_time, user_time;
4865   char cdummy;
4866   int idummy;
4867   long ldummy;
4868   FILE *fp;
4869 
4870   // The /proc/<tid>/stat aggregates per-process usage on
4871   // new Linux kernels 2.6+ where NPTL is supported.
4872   // The /proc/self/task/<tid>/stat still has the per-thread usage.
4873   // See bug 6328462.
4874   // There possibly can be cases where there is no directory
4875   // /proc/self/task, so we check its availability.
4876   if (proc_task_unchecked && os::Linux::is_NPTL()) {
4877     // This is executed only once
4878     proc_task_unchecked = false;
4879     fp = fopen("/proc/self/task", "r");
4880     if (fp != NULL) {
4881       proc_stat_path = "/proc/self/task/%d/stat";
4882       fclose(fp);
4883     }
4884   }
4885 
4886   sprintf(proc_name, proc_stat_path, tid);
4887   fp = fopen(proc_name, "r");
4888   if ( fp == NULL ) return -1;
4889   statlen = fread(stat, 1, 2047, fp);
4890   stat[statlen] = '\0';
4891   fclose(fp);
4892 
4893   // Skip pid and the command string. Note that we could be dealing with
4894   // weird command names, e.g. user could decide to rename java launcher
4895   // to "java 1.4.2 :)", then the stat file would look like
4896   //                1234 (java 1.4.2 :)) R ... ...
4897   // We don't really need to know the command string, just find the last
4898   // occurrence of ")" and then start parsing from there. See bug 4726580.
4899   s = strrchr(stat, ')');
4900   if (s == NULL ) return -1;
4901 
4902   // Skip blank chars
4903   do s++; while (isspace(*s));
4904 
4905   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4906                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4907                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4908                  &user_time, &sys_time);
4909   if ( count != 13 ) return -1;
4910   if (user_sys_cpu_time) {
4911     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4912   } else {
4913     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4914   }
4915 }
4916 
4917 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4918   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4919   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4920   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4921   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4922 }
4923 
4924 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4925   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4926   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4927   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4928   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4929 }
4930 
4931 bool os::is_thread_cpu_time_supported() {
4932   return true;
4933 }
4934 
4935 // System loadavg support.  Returns -1 if load average cannot be obtained.
4936 // Linux doesn't yet have a (official) notion of processor sets,
4937 // so just return the system wide load average.
4938 int os::loadavg(double loadavg[], int nelem) {
4939   return ::getloadavg(loadavg, nelem);
4940 }
4941 
4942 void os::pause() {
4943   char filename[MAX_PATH];
4944   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4945     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4946   } else {
4947     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4948   }
4949 
4950   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4951   if (fd != -1) {
4952     struct stat buf;
4953     ::close(fd);
4954     while (::stat(filename, &buf) == 0) {
4955       (void)::poll(NULL, 0, 100);
4956     }
4957   } else {
4958     jio_fprintf(stderr,
4959       "Could not open pause file '%s', continuing immediately.\n", filename);
4960   }
4961 }
4962 
4963 
4964 // Refer to the comments in os_solaris.cpp park-unpark.
4965 //
4966 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4967 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4968 // For specifics regarding the bug see GLIBC BUGID 261237 :
4969 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4970 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4971 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4972 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4973 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4974 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4975 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4976 // of libpthread avoids the problem, but isn't practical.
4977 //
4978 // Possible remedies:
4979 //
4980 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4981 //      This is palliative and probabilistic, however.  If the thread is preempted
4982 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4983 //      than the minimum period may have passed, and the abstime may be stale (in the
4984 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
4985 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4986 //
4987 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4988 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
4989 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4990 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
4991 //      thread.
4992 //
4993 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4994 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
4995 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4996 //      This also works well.  In fact it avoids kernel-level scalability impediments
4997 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4998 //      timers in a graceful fashion.
4999 //
5000 // 4.   When the abstime value is in the past it appears that control returns
5001 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5002 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5003 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5004 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5005 //      It may be possible to avoid reinitialization by checking the return
5006 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5007 //      condvar we must establish the invariant that cond_signal() is only called
5008 //      within critical sections protected by the adjunct mutex.  This prevents
5009 //      cond_signal() from "seeing" a condvar that's in the midst of being
5010 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5011 //      desirable signal-after-unlock optimization that avoids futile context switching.
5012 //
5013 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5014 //      structure when a condvar is used or initialized.  cond_destroy()  would
5015 //      release the helper structure.  Our reinitialize-after-timedwait fix
5016 //      put excessive stress on malloc/free and locks protecting the c-heap.
5017 //
5018 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5019 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5020 // and only enabling the work-around for vulnerable environments.
5021 
5022 // utility to compute the abstime argument to timedwait:
5023 // millis is the relative timeout time
5024 // abstime will be the absolute timeout time
5025 // TODO: replace compute_abstime() with unpackTime()
5026 
5027 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5028   if (millis < 0)  millis = 0;
5029   struct timeval now;
5030   int status = gettimeofday(&now, NULL);
5031   assert(status == 0, "gettimeofday");
5032   jlong seconds = millis / 1000;
5033   millis %= 1000;
5034   if (seconds > 50000000) { // see man cond_timedwait(3T)
5035     seconds = 50000000;
5036   }
5037   abstime->tv_sec = now.tv_sec  + seconds;
5038   long       usec = now.tv_usec + millis * 1000;
5039   if (usec >= 1000000) {
5040     abstime->tv_sec += 1;
5041     usec -= 1000000;
5042   }
5043   abstime->tv_nsec = usec * 1000;
5044   return abstime;
5045 }
5046 
5047 
5048 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5049 // Conceptually TryPark() should be equivalent to park(0).
5050 
5051 int os::PlatformEvent::TryPark() {
5052   for (;;) {
5053     const int v = _Event ;
5054     guarantee ((v == 0) || (v == 1), "invariant") ;
5055     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5056   }
5057 }
5058 
5059 void os::PlatformEvent::park() {       // AKA "down()"
5060   // Invariant: Only the thread associated with the Event/PlatformEvent
5061   // may call park().
5062   // TODO: assert that _Assoc != NULL or _Assoc == Self
5063   int v ;
5064   for (;;) {
5065       v = _Event ;
5066       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5067   }
5068   guarantee (v >= 0, "invariant") ;
5069   if (v == 0) {
5070      // Do this the hard way by blocking ...
5071      int status = pthread_mutex_lock(_mutex);
5072      assert_status(status == 0, status, "mutex_lock");
5073      guarantee (_nParked == 0, "invariant") ;
5074      ++ _nParked ;
5075      while (_Event < 0) {
5076         status = pthread_cond_wait(_cond, _mutex);
5077         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5078         // Treat this the same as if the wait was interrupted
5079         if (status == ETIME) { status = EINTR; }
5080         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5081      }
5082      -- _nParked ;
5083 
5084     _Event = 0 ;
5085      status = pthread_mutex_unlock(_mutex);
5086      assert_status(status == 0, status, "mutex_unlock");
5087     // Paranoia to ensure our locked and lock-free paths interact
5088     // correctly with each other.
5089     OrderAccess::fence();
5090   }
5091   guarantee (_Event >= 0, "invariant") ;
5092 }
5093 
5094 int os::PlatformEvent::park(jlong millis) {
5095   guarantee (_nParked == 0, "invariant") ;
5096 
5097   int v ;
5098   for (;;) {
5099       v = _Event ;
5100       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5101   }
5102   guarantee (v >= 0, "invariant") ;
5103   if (v != 0) return OS_OK ;
5104 
5105   // We do this the hard way, by blocking the thread.
5106   // Consider enforcing a minimum timeout value.
5107   struct timespec abst;
5108   compute_abstime(&abst, millis);
5109 
5110   int ret = OS_TIMEOUT;
5111   int status = pthread_mutex_lock(_mutex);
5112   assert_status(status == 0, status, "mutex_lock");
5113   guarantee (_nParked == 0, "invariant") ;
5114   ++_nParked ;
5115 
5116   // Object.wait(timo) will return because of
5117   // (a) notification
5118   // (b) timeout
5119   // (c) thread.interrupt
5120   //
5121   // Thread.interrupt and object.notify{All} both call Event::set.
5122   // That is, we treat thread.interrupt as a special case of notification.
5123   // The underlying Solaris implementation, cond_timedwait, admits
5124   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5125   // JVM from making those visible to Java code.  As such, we must
5126   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5127   //
5128   // TODO: properly differentiate simultaneous notify+interrupt.
5129   // In that case, we should propagate the notify to another waiter.
5130 
5131   while (_Event < 0) {
5132     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5133     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5134       pthread_cond_destroy (_cond);
5135       pthread_cond_init (_cond, NULL) ;
5136     }
5137     assert_status(status == 0 || status == EINTR ||
5138                   status == ETIME || status == ETIMEDOUT,
5139                   status, "cond_timedwait");
5140     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5141     if (status == ETIME || status == ETIMEDOUT) break ;
5142     // We consume and ignore EINTR and spurious wakeups.
5143   }
5144   --_nParked ;
5145   if (_Event >= 0) {
5146      ret = OS_OK;
5147   }
5148   _Event = 0 ;
5149   status = pthread_mutex_unlock(_mutex);
5150   assert_status(status == 0, status, "mutex_unlock");
5151   assert (_nParked == 0, "invariant") ;
5152   // Paranoia to ensure our locked and lock-free paths interact
5153   // correctly with each other.
5154   OrderAccess::fence();
5155   return ret;
5156 }
5157 
5158 void os::PlatformEvent::unpark() {
5159   // Transitions for _Event:
5160   //    0 :=> 1
5161   //    1 :=> 1
5162   //   -1 :=> either 0 or 1; must signal target thread
5163   //          That is, we can safely transition _Event from -1 to either
5164   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
5165   //          unpark() calls.
5166   // See also: "Semaphores in Plan 9" by Mullender & Cox
5167   //
5168   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5169   // that it will take two back-to-back park() calls for the owning
5170   // thread to block. This has the benefit of forcing a spurious return
5171   // from the first park() call after an unpark() call which will help
5172   // shake out uses of park() and unpark() without condition variables.
5173 
5174   if (Atomic::xchg(1, &_Event) >= 0) return;
5175 
5176   // Wait for the thread associated with the event to vacate
5177   int status = pthread_mutex_lock(_mutex);
5178   assert_status(status == 0, status, "mutex_lock");
5179   int AnyWaiters = _nParked;
5180   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5181   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5182     AnyWaiters = 0;
5183     pthread_cond_signal(_cond);
5184   }
5185   status = pthread_mutex_unlock(_mutex);
5186   assert_status(status == 0, status, "mutex_unlock");
5187   if (AnyWaiters != 0) {
5188     status = pthread_cond_signal(_cond);
5189     assert_status(status == 0, status, "cond_signal");
5190   }
5191 
5192   // Note that we signal() _after dropping the lock for "immortal" Events.
5193   // This is safe and avoids a common class of  futile wakeups.  In rare
5194   // circumstances this can cause a thread to return prematurely from
5195   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5196   // simply re-test the condition and re-park itself.
5197 }
5198 
5199 
5200 // JSR166
5201 // -------------------------------------------------------
5202 
5203 /*
5204  * The solaris and linux implementations of park/unpark are fairly
5205  * conservative for now, but can be improved. They currently use a
5206  * mutex/condvar pair, plus a a count.
5207  * Park decrements count if > 0, else does a condvar wait.  Unpark
5208  * sets count to 1 and signals condvar.  Only one thread ever waits
5209  * on the condvar. Contention seen when trying to park implies that someone
5210  * is unparking you, so don't wait. And spurious returns are fine, so there
5211  * is no need to track notifications.
5212  */
5213 
5214 #define MAX_SECS 100000000
5215 /*
5216  * This code is common to linux and solaris and will be moved to a
5217  * common place in dolphin.
5218  *
5219  * The passed in time value is either a relative time in nanoseconds
5220  * or an absolute time in milliseconds. Either way it has to be unpacked
5221  * into suitable seconds and nanoseconds components and stored in the
5222  * given timespec structure.
5223  * Given time is a 64-bit value and the time_t used in the timespec is only
5224  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5225  * overflow if times way in the future are given. Further on Solaris versions
5226  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5227  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5228  * As it will be 28 years before "now + 100000000" will overflow we can
5229  * ignore overflow and just impose a hard-limit on seconds using the value
5230  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5231  * years from "now".
5232  */
5233 
5234 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5235   assert (time > 0, "convertTime");
5236 
5237   struct timeval now;
5238   int status = gettimeofday(&now, NULL);
5239   assert(status == 0, "gettimeofday");
5240 
5241   time_t max_secs = now.tv_sec + MAX_SECS;
5242 
5243   if (isAbsolute) {
5244     jlong secs = time / 1000;
5245     if (secs > max_secs) {
5246       absTime->tv_sec = max_secs;
5247     }
5248     else {
5249       absTime->tv_sec = secs;
5250     }
5251     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5252   }
5253   else {
5254     jlong secs = time / NANOSECS_PER_SEC;
5255     if (secs >= MAX_SECS) {
5256       absTime->tv_sec = max_secs;
5257       absTime->tv_nsec = 0;
5258     }
5259     else {
5260       absTime->tv_sec = now.tv_sec + secs;
5261       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5262       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5263         absTime->tv_nsec -= NANOSECS_PER_SEC;
5264         ++absTime->tv_sec; // note: this must be <= max_secs
5265       }
5266     }
5267   }
5268   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5269   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5270   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5271   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5272 }
5273 
5274 void Parker::park(bool isAbsolute, jlong time) {
5275   // Ideally we'd do something useful while spinning, such
5276   // as calling unpackTime().
5277 
5278   // Optional fast-path check:
5279   // Return immediately if a permit is available.
5280   // We depend on Atomic::xchg() having full barrier semantics
5281   // since we are doing a lock-free update to _counter.
5282   if (Atomic::xchg(0, &_counter) > 0) return;
5283 
5284   Thread* thread = Thread::current();
5285   assert(thread->is_Java_thread(), "Must be JavaThread");
5286   JavaThread *jt = (JavaThread *)thread;
5287 
5288   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5289   // Check interrupt before trying to wait
5290   if (Thread::is_interrupted(thread, false)) {
5291     return;
5292   }
5293 
5294   // Next, demultiplex/decode time arguments
5295   timespec absTime;
5296   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5297     return;
5298   }
5299   if (time > 0) {
5300     unpackTime(&absTime, isAbsolute, time);
5301   }
5302 
5303 
5304   // Enter safepoint region
5305   // Beware of deadlocks such as 6317397.
5306   // The per-thread Parker:: mutex is a classic leaf-lock.
5307   // In particular a thread must never block on the Threads_lock while
5308   // holding the Parker:: mutex.  If safepoints are pending both the
5309   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5310   ThreadBlockInVM tbivm(jt);
5311 
5312   // Don't wait if cannot get lock since interference arises from
5313   // unblocking.  Also. check interrupt before trying wait
5314   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5315     return;
5316   }
5317 
5318   int status ;
5319   if (_counter > 0)  { // no wait needed
5320     _counter = 0;
5321     status = pthread_mutex_unlock(_mutex);
5322     assert (status == 0, "invariant") ;
5323     // Paranoia to ensure our locked and lock-free paths interact
5324     // correctly with each other and Java-level accesses.
5325     OrderAccess::fence();
5326     return;
5327   }
5328 
5329 #ifdef ASSERT
5330   // Don't catch signals while blocked; let the running threads have the signals.
5331   // (This allows a debugger to break into the running thread.)
5332   sigset_t oldsigs;
5333   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5334   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5335 #endif
5336 
5337   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5338   jt->set_suspend_equivalent();
5339   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5340 
5341   if (time == 0) {
5342     status = pthread_cond_wait (_cond, _mutex) ;
5343   } else {
5344     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5345     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5346       pthread_cond_destroy (_cond) ;
5347       pthread_cond_init    (_cond, NULL);
5348     }
5349   }
5350   assert_status(status == 0 || status == EINTR ||
5351                 status == ETIME || status == ETIMEDOUT,
5352                 status, "cond_timedwait");
5353 
5354 #ifdef ASSERT
5355   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5356 #endif
5357 
5358   _counter = 0 ;
5359   status = pthread_mutex_unlock(_mutex) ;
5360   assert_status(status == 0, status, "invariant") ;
5361   // Paranoia to ensure our locked and lock-free paths interact
5362   // correctly with each other and Java-level accesses.
5363   OrderAccess::fence();
5364 
5365   // If externally suspended while waiting, re-suspend
5366   if (jt->handle_special_suspend_equivalent_condition()) {
5367     jt->java_suspend_self();
5368   }
5369 }
5370 
5371 void Parker::unpark() {
5372   int s, status ;
5373   status = pthread_mutex_lock(_mutex);
5374   assert (status == 0, "invariant") ;
5375   s = _counter;
5376   _counter = 1;
5377   if (s < 1) {
5378      if (WorkAroundNPTLTimedWaitHang) {
5379         status = pthread_cond_signal (_cond) ;
5380         assert (status == 0, "invariant") ;
5381         status = pthread_mutex_unlock(_mutex);
5382         assert (status == 0, "invariant") ;
5383      } else {
5384         status = pthread_mutex_unlock(_mutex);
5385         assert (status == 0, "invariant") ;
5386         status = pthread_cond_signal (_cond) ;
5387         assert (status == 0, "invariant") ;
5388      }
5389   } else {
5390     pthread_mutex_unlock(_mutex);
5391     assert (status == 0, "invariant") ;
5392   }
5393 }
5394 
5395 
5396 extern char** environ;
5397 
5398 #ifndef __NR_fork
5399 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5400 #endif
5401 
5402 #ifndef __NR_execve
5403 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5404 #endif
5405 
5406 // Run the specified command in a separate process. Return its exit value,
5407 // or -1 on failure (e.g. can't fork a new process).
5408 // Unlike system(), this function can be called from signal handler. It
5409 // doesn't block SIGINT et al.
5410 int os::fork_and_exec(char* cmd) {
5411   const char * argv[4] = {"sh", "-c", cmd, NULL};
5412 
5413   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5414   // pthread_atfork handlers and reset pthread library. All we need is a
5415   // separate process to execve. Make a direct syscall to fork process.
5416   // On IA64 there's no fork syscall, we have to use fork() and hope for
5417   // the best...
5418   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5419               IA64_ONLY(fork();)
5420 
5421   if (pid < 0) {
5422     // fork failed
5423     return -1;
5424 
5425   } else if (pid == 0) {
5426     // child process
5427 
5428     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5429     // first to kill every thread on the thread list. Because this list is
5430     // not reset by fork() (see notes above), execve() will instead kill
5431     // every thread in the parent process. We know this is the only thread
5432     // in the new process, so make a system call directly.
5433     // IA64 should use normal execve() from glibc to match the glibc fork()
5434     // above.
5435     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5436     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5437 
5438     // execve failed
5439     _exit(-1);
5440 
5441   } else  {
5442     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5443     // care about the actual exit code, for now.
5444 
5445     int status;
5446 
5447     // Wait for the child process to exit.  This returns immediately if
5448     // the child has already exited. */
5449     while (waitpid(pid, &status, 0) < 0) {
5450         switch (errno) {
5451         case ECHILD: return 0;
5452         case EINTR: break;
5453         default: return -1;
5454         }
5455     }
5456 
5457     if (WIFEXITED(status)) {
5458        // The child exited normally; get its exit code.
5459        return WEXITSTATUS(status);
5460     } else if (WIFSIGNALED(status)) {
5461        // The child exited because of a signal
5462        // The best value to return is 0x80 + signal number,
5463        // because that is what all Unix shells do, and because
5464        // it allows callers to distinguish between process exit and
5465        // process death by signal.
5466        return 0x80 + WTERMSIG(status);
5467     } else {
5468        // Unknown exit code; pass it through
5469        return status;
5470     }
5471   }
5472 }
5473 
5474 // is_headless_jre()
5475 //
5476 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5477 // in order to report if we are running in a headless jre
5478 //
5479 // Since JDK8 xawt/libmawt.so was moved into the same directory
5480 // as libawt.so, and renamed libawt_xawt.so
5481 //
5482 bool os::is_headless_jre() {
5483     struct stat statbuf;
5484     char buf[MAXPATHLEN];
5485     char libmawtpath[MAXPATHLEN];
5486     const char *xawtstr  = "/xawt/libmawt.so";
5487     const char *new_xawtstr = "/libawt_xawt.so";
5488     char *p;
5489 
5490     // Get path to libjvm.so
5491     os::jvm_path(buf, sizeof(buf));
5492 
5493     // Get rid of libjvm.so
5494     p = strrchr(buf, '/');
5495     if (p == NULL) return false;
5496     else *p = '\0';
5497 
5498     // Get rid of client or server
5499     p = strrchr(buf, '/');
5500     if (p == NULL) return false;
5501     else *p = '\0';
5502 
5503     // check xawt/libmawt.so
5504     strcpy(libmawtpath, buf);
5505     strcat(libmawtpath, xawtstr);
5506     if (::stat(libmawtpath, &statbuf) == 0) return false;
5507 
5508     // check libawt_xawt.so
5509     strcpy(libmawtpath, buf);
5510     strcat(libmawtpath, new_xawtstr);
5511     if (::stat(libmawtpath, &statbuf) == 0) return false;
5512 
5513     return true;
5514 }
5515 
5516 // Get the default path to the core file
5517 // Returns the length of the string
5518 int os::get_core_path(char* buffer, size_t bufferSize) {
5519   const char* p = get_current_directory(buffer, bufferSize);
5520 
5521   if (p == NULL) {
5522     assert(p != NULL, "failed to get current directory");
5523     return 0;
5524   }
5525 
5526   return strlen(buffer);
5527 }
5528 
5529 #ifdef JAVASE_EMBEDDED
5530 //
5531 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5532 //
5533 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5534 
5535 // ctor
5536 //
5537 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5538   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5539   _fd = fd;
5540 
5541   if (os::create_thread(this, os::os_thread)) {
5542     _memnotify_thread = this;
5543     os::set_priority(this, NearMaxPriority);
5544     os::start_thread(this);
5545   }
5546 }
5547 
5548 // Where all the work gets done
5549 //
5550 void MemNotifyThread::run() {
5551   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5552 
5553   // Set up the select arguments
5554   fd_set rfds;
5555   if (_fd != -1) {
5556     FD_ZERO(&rfds);
5557     FD_SET(_fd, &rfds);
5558   }
5559 
5560   // Now wait for the mem_notify device to wake up
5561   while (1) {
5562     // Wait for the mem_notify device to signal us..
5563     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5564     if (rc == -1) {
5565       perror("select!\n");
5566       break;
5567     } else if (rc) {
5568       //ssize_t free_before = os::available_memory();
5569       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5570 
5571       // The kernel is telling us there is not much memory left...
5572       // try to do something about that
5573 
5574       // If we are not already in a GC, try one.
5575       if (!Universe::heap()->is_gc_active()) {
5576         Universe::heap()->collect(GCCause::_allocation_failure);
5577 
5578         //ssize_t free_after = os::available_memory();
5579         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5580         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5581       }
5582       // We might want to do something like the following if we find the GC's are not helping...
5583       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5584     }
5585   }
5586 }
5587 
5588 //
5589 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5590 //
5591 void MemNotifyThread::start() {
5592   int    fd;
5593   fd = open ("/dev/mem_notify", O_RDONLY, 0);
5594   if (fd < 0) {
5595       return;
5596   }
5597 
5598   if (memnotify_thread() == NULL) {
5599     new MemNotifyThread(fd);
5600   }
5601 }
5602 #endif // JAVASE_EMBEDDED