1 /*
   2  * Copyright (c) 1999, 2014, 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 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 105 // getrusage() is prepared to handle the associated failure.
 106 #ifndef RUSAGE_THREAD
 107 #define RUSAGE_THREAD   (1)               /* only the calling thread */
 108 #endif
 109 
 110 #define MAX_PATH    (2 * K)
 111 
 112 #define MAX_SECS 100000000
 113 
 114 // for timer info max values which include all bits
 115 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 116 
 117 #define LARGEPAGES_BIT (1 << 6)
 118 ////////////////////////////////////////////////////////////////////////////////
 119 // global variables
 120 julong os::Linux::_physical_memory = 0;
 121 
 122 address   os::Linux::_initial_thread_stack_bottom = NULL;
 123 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 124 
 125 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 126 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 127 Mutex* os::Linux::_createThread_lock = NULL;
 128 pthread_t os::Linux::_main_thread;
 129 int os::Linux::_page_size = -1;
 130 const int os::Linux::_vm_default_page_size = (8 * K);
 131 bool os::Linux::_is_floating_stack = false;
 132 bool os::Linux::_is_NPTL = false;
 133 bool os::Linux::_supports_fast_thread_cpu_time = false;
 134 const char * os::Linux::_glibc_version = NULL;
 135 const char * os::Linux::_libpthread_version = NULL;
 136 pthread_condattr_t os::Linux::_condattr[1];
 137 
 138 static jlong initial_time_count=0;
 139 
 140 static int clock_tics_per_sec = 100;
 141 
 142 // For diagnostics to print a message once. see run_periodic_checks
 143 static sigset_t check_signal_done;
 144 static bool check_signals = true;
 145 
 146 static pid_t _initial_pid = 0;
 147 
 148 /* Signal number used to suspend/resume a thread */
 149 
 150 /* do not use any signal number less than SIGSEGV, see 4355769 */
 151 static int SR_signum = SIGUSR2;
 152 sigset_t SR_sigset;
 153 
 154 /* Used to protect dlsym() calls */
 155 static pthread_mutex_t dl_mutex;
 156 
 157 // Declarations
 158 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 159 
 160 #ifdef JAVASE_EMBEDDED
 161 class MemNotifyThread: public Thread {
 162   friend class VMStructs;
 163  public:
 164   virtual void run();
 165 
 166  private:
 167   static MemNotifyThread* _memnotify_thread;
 168   int _fd;
 169 
 170  public:
 171 
 172   // Constructor
 173   MemNotifyThread(int fd);
 174 
 175   // Tester
 176   bool is_memnotify_thread() const { return true; }
 177 
 178   // Printing
 179   char* name() const { return (char*)"Linux MemNotify Thread"; }
 180 
 181   // Returns the single instance of the MemNotifyThread
 182   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 183 
 184   // Create and start the single instance of MemNotifyThread
 185   static void start();
 186 };
 187 #endif // JAVASE_EMBEDDED
 188 
 189 // utility functions
 190 
 191 static int SR_initialize();
 192 
 193 julong os::available_memory() {
 194   return Linux::available_memory();
 195 }
 196 
 197 julong os::Linux::available_memory() {
 198   // values in struct sysinfo are "unsigned long"
 199   struct sysinfo si;
 200   sysinfo(&si);
 201 
 202   return (julong)si.freeram * si.mem_unit;
 203 }
 204 
 205 julong os::physical_memory() {
 206   return Linux::physical_memory();
 207 }
 208 
 209 ////////////////////////////////////////////////////////////////////////////////
 210 // environment support
 211 
 212 bool os::getenv(const char* name, char* buf, int len) {
 213   const char* val = ::getenv(name);
 214   if (val != NULL && strlen(val) < (size_t)len) {
 215     strcpy(buf, val);
 216     return true;
 217   }
 218   if (len > 0) buf[0] = 0;  // return a null string
 219   return false;
 220 }
 221 
 222 
 223 // Return true if user is running as root.
 224 
 225 bool os::have_special_privileges() {
 226   static bool init = false;
 227   static bool privileges = false;
 228   if (!init) {
 229     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 230     init = true;
 231   }
 232   return privileges;
 233 }
 234 
 235 
 236 #ifndef SYS_gettid
 237 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 238 #ifdef __ia64__
 239 #define SYS_gettid 1105
 240 #elif __i386__
 241 #define SYS_gettid 224
 242 #elif __amd64__
 243 #define SYS_gettid 186
 244 #elif __sparc__
 245 #define SYS_gettid 143
 246 #else
 247 #error define gettid for the arch
 248 #endif
 249 #endif
 250 
 251 // Cpu architecture string
 252 #if   defined(ZERO)
 253 static char cpu_arch[] = ZERO_LIBARCH;
 254 #elif defined(IA64)
 255 static char cpu_arch[] = "ia64";
 256 #elif defined(IA32)
 257 static char cpu_arch[] = "i386";
 258 #elif defined(AMD64)
 259 static char cpu_arch[] = "amd64";
 260 #elif defined(ARM)
 261 static char cpu_arch[] = "arm";
 262 #elif defined(PPC32)
 263 static char cpu_arch[] = "ppc";
 264 #elif defined(PPC64)
 265 static char cpu_arch[] = "ppc64";
 266 #elif defined(SPARC)
 267 #  ifdef _LP64
 268 static char cpu_arch[] = "sparcv9";
 269 #  else
 270 static char cpu_arch[] = "sparc";
 271 #  endif
 272 #else
 273 #error Add appropriate cpu_arch setting
 274 #endif
 275 
 276 
 277 // pid_t gettid()
 278 //
 279 // Returns the kernel thread id of the currently running thread. Kernel
 280 // thread id is used to access /proc.
 281 //
 282 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 283 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 284 //
 285 pid_t os::Linux::gettid() {
 286   int rslt = syscall(SYS_gettid);
 287   if (rslt == -1) {
 288      // old kernel, no NPTL support
 289      return getpid();
 290   } else {
 291      return (pid_t)rslt;
 292   }
 293 }
 294 
 295 // Most versions of linux have a bug where the number of processors are
 296 // determined by looking at the /proc file system.  In a chroot environment,
 297 // the system call returns 1.  This causes the VM to act as if it is
 298 // a single processor and elide locking (see is_MP() call).
 299 static bool unsafe_chroot_detected = false;
 300 static const char *unstable_chroot_error = "/proc file system not found.\n"
 301                      "Java may be unstable running multithreaded in a chroot "
 302                      "environment on Linux when /proc filesystem is not mounted.";
 303 
 304 void os::Linux::initialize_system_info() {
 305   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 306   if (processor_count() == 1) {
 307     pid_t pid = os::Linux::gettid();
 308     char fname[32];
 309     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 310     FILE *fp = fopen(fname, "r");
 311     if (fp == NULL) {
 312       unsafe_chroot_detected = true;
 313     } else {
 314       fclose(fp);
 315     }
 316   }
 317   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 318   assert(processor_count() > 0, "linux error");
 319 }
 320 
 321 void os::init_system_properties_values() {
 322   // The next steps are taken in the product version:
 323   //
 324   // Obtain the JAVA_HOME value from the location of libjvm.so.
 325   // This library should be located at:
 326   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 327   //
 328   // If "/jre/lib/" appears at the right place in the path, then we
 329   // assume libjvm.so is installed in a JDK and we use this path.
 330   //
 331   // Otherwise exit with message: "Could not create the Java virtual machine."
 332   //
 333   // The following extra steps are taken in the debugging version:
 334   //
 335   // If "/jre/lib/" does NOT appear at the right place in the path
 336   // instead of exit check for $JAVA_HOME environment variable.
 337   //
 338   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 339   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 340   // it looks like libjvm.so is installed there
 341   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 342   //
 343   // Otherwise exit.
 344   //
 345   // Important note: if the location of libjvm.so changes this
 346   // code needs to be changed accordingly.
 347 
 348 // See ld(1):
 349 //      The linker uses the following search paths to locate required
 350 //      shared libraries:
 351 //        1: ...
 352 //        ...
 353 //        7: The default directories, normally /lib and /usr/lib.
 354 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 355 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 356 #else
 357 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 358 #endif
 359 
 360 // Base path of extensions installed on the system.
 361 #define SYS_EXT_DIR     "/usr/java/packages"
 362 #define EXTENSIONS_DIR  "/lib/ext"
 363 #define ENDORSED_DIR    "/lib/endorsed"
 364 
 365   // Buffer that fits several sprintfs.
 366   // Note that the space for the colon and the trailing null are provided
 367   // by the nulls included by the sizeof operator.
 368   const size_t bufsize =
 369     MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
 370          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
 371          (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
 372   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 373 
 374   // sysclasspath, java_home, dll_dir
 375   {
 376     char *pslash;
 377     os::jvm_path(buf, bufsize);
 378 
 379     // Found the full path to libjvm.so.
 380     // Now cut the path to <java_home>/jre if we can.
 381     *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
 382     pslash = strrchr(buf, '/');
 383     if (pslash != NULL) {
 384       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 385     }
 386     Arguments::set_dll_dir(buf);
 387 
 388     if (pslash != NULL) {
 389       pslash = strrchr(buf, '/');
 390       if (pslash != NULL) {
 391         *pslash = '\0';          // Get rid of /<arch>.
 392         pslash = strrchr(buf, '/');
 393         if (pslash != NULL) {
 394           *pslash = '\0';        // Get rid of /lib.
 395         }
 396       }
 397     }
 398     Arguments::set_java_home(buf);
 399     set_boot_path('/', ':');
 400   }
 401 
 402   // Where to look for native libraries.
 403   //
 404   // Note: Due to a legacy implementation, most of the library path
 405   // is set in the launcher. This was to accomodate linking restrictions
 406   // on legacy Linux implementations (which are no longer supported).
 407   // Eventually, all the library path setting will be done here.
 408   //
 409   // However, to prevent the proliferation of improperly built native
 410   // libraries, the new path component /usr/java/packages is added here.
 411   // Eventually, all the library path setting will be done here.
 412   {
 413     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 414     // should always exist (until the legacy problem cited above is
 415     // addressed).
 416     const char *v = ::getenv("LD_LIBRARY_PATH");
 417     const char *v_colon = ":";
 418     if (v == NULL) { v = ""; v_colon = ""; }
 419     // That's +1 for the colon and +1 for the trailing '\0'.
 420     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 421                                                      strlen(v) + 1 +
 422                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 423                                                      mtInternal);
 424     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 425     Arguments::set_library_path(ld_library_path);
 426     FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
 427   }
 428 
 429   // Extensions directories.
 430   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 431   Arguments::set_ext_dirs(buf);
 432 
 433   // Endorsed standards default directory.
 434   sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 435   Arguments::set_endorsed_dirs(buf);
 436 
 437   FREE_C_HEAP_ARRAY(char, buf, mtInternal);
 438 
 439 #undef DEFAULT_LIBPATH
 440 #undef SYS_EXT_DIR
 441 #undef EXTENSIONS_DIR
 442 #undef ENDORSED_DIR
 443 }
 444 
 445 ////////////////////////////////////////////////////////////////////////////////
 446 // breakpoint support
 447 
 448 void os::breakpoint() {
 449   BREAKPOINT;
 450 }
 451 
 452 extern "C" void breakpoint() {
 453   // use debugger to set breakpoint here
 454 }
 455 
 456 ////////////////////////////////////////////////////////////////////////////////
 457 // signal support
 458 
 459 debug_only(static bool signal_sets_initialized = false);
 460 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 461 
 462 bool os::Linux::is_sig_ignored(int sig) {
 463       struct sigaction oact;
 464       sigaction(sig, (struct sigaction*)NULL, &oact);
 465       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 466                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 467       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 468            return true;
 469       else
 470            return false;
 471 }
 472 
 473 void os::Linux::signal_sets_init() {
 474   // Should also have an assertion stating we are still single-threaded.
 475   assert(!signal_sets_initialized, "Already initialized");
 476   // Fill in signals that are necessarily unblocked for all threads in
 477   // the VM. Currently, we unblock the following signals:
 478   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 479   //                         by -Xrs (=ReduceSignalUsage));
 480   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 481   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 482   // the dispositions or masks wrt these signals.
 483   // Programs embedding the VM that want to use the above signals for their
 484   // own purposes must, at this time, use the "-Xrs" option to prevent
 485   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 486   // (See bug 4345157, and other related bugs).
 487   // In reality, though, unblocking these signals is really a nop, since
 488   // these signals are not blocked by default.
 489   sigemptyset(&unblocked_sigs);
 490   sigemptyset(&allowdebug_blocked_sigs);
 491   sigaddset(&unblocked_sigs, SIGILL);
 492   sigaddset(&unblocked_sigs, SIGSEGV);
 493   sigaddset(&unblocked_sigs, SIGBUS);
 494   sigaddset(&unblocked_sigs, SIGFPE);
 495 #if defined(PPC64)
 496   sigaddset(&unblocked_sigs, SIGTRAP);
 497 #endif
 498   sigaddset(&unblocked_sigs, SR_signum);
 499 
 500   if (!ReduceSignalUsage) {
 501    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 502       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 503       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 504    }
 505    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 506       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 507       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 508    }
 509    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 510       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 511       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 512    }
 513   }
 514   // Fill in signals that are blocked by all but the VM thread.
 515   sigemptyset(&vm_sigs);
 516   if (!ReduceSignalUsage)
 517     sigaddset(&vm_sigs, BREAK_SIGNAL);
 518   debug_only(signal_sets_initialized = true);
 519 
 520 }
 521 
 522 // These are signals that are unblocked while a thread is running Java.
 523 // (For some reason, they get blocked by default.)
 524 sigset_t* os::Linux::unblocked_signals() {
 525   assert(signal_sets_initialized, "Not initialized");
 526   return &unblocked_sigs;
 527 }
 528 
 529 // These are the signals that are blocked while a (non-VM) thread is
 530 // running Java. Only the VM thread handles these signals.
 531 sigset_t* os::Linux::vm_signals() {
 532   assert(signal_sets_initialized, "Not initialized");
 533   return &vm_sigs;
 534 }
 535 
 536 // These are signals that are blocked during cond_wait to allow debugger in
 537 sigset_t* os::Linux::allowdebug_blocked_signals() {
 538   assert(signal_sets_initialized, "Not initialized");
 539   return &allowdebug_blocked_sigs;
 540 }
 541 
 542 void os::Linux::hotspot_sigmask(Thread* thread) {
 543 
 544   //Save caller's signal mask before setting VM signal mask
 545   sigset_t caller_sigmask;
 546   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 547 
 548   OSThread* osthread = thread->osthread();
 549   osthread->set_caller_sigmask(caller_sigmask);
 550 
 551   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 552 
 553   if (!ReduceSignalUsage) {
 554     if (thread->is_VM_thread()) {
 555       // Only the VM thread handles BREAK_SIGNAL ...
 556       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 557     } else {
 558       // ... all other threads block BREAK_SIGNAL
 559       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 560     }
 561   }
 562 }
 563 
 564 //////////////////////////////////////////////////////////////////////////////
 565 // detecting pthread library
 566 
 567 void os::Linux::libpthread_init() {
 568   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 569   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 570   // generic name for earlier versions.
 571   // Define macros here so we can build HotSpot on old systems.
 572 # ifndef _CS_GNU_LIBC_VERSION
 573 # define _CS_GNU_LIBC_VERSION 2
 574 # endif
 575 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 576 # define _CS_GNU_LIBPTHREAD_VERSION 3
 577 # endif
 578 
 579   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 580   if (n > 0) {
 581      char *str = (char *)malloc(n, mtInternal);
 582      confstr(_CS_GNU_LIBC_VERSION, str, n);
 583      os::Linux::set_glibc_version(str);
 584   } else {
 585      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 586      static char _gnu_libc_version[32];
 587      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 588               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 589      os::Linux::set_glibc_version(_gnu_libc_version);
 590   }
 591 
 592   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 593   if (n > 0) {
 594      char *str = (char *)malloc(n, mtInternal);
 595      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 596      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 597      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 598      // is the case. LinuxThreads has a hard limit on max number of threads.
 599      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 600      // On the other hand, NPTL does not have such a limit, sysconf()
 601      // will return -1 and errno is not changed. Check if it is really NPTL.
 602      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 603          strstr(str, "NPTL") &&
 604          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 605        free(str);
 606        os::Linux::set_libpthread_version("linuxthreads");
 607      } else {
 608        os::Linux::set_libpthread_version(str);
 609      }
 610   } else {
 611     // glibc before 2.3.2 only has LinuxThreads.
 612     os::Linux::set_libpthread_version("linuxthreads");
 613   }
 614 
 615   if (strstr(libpthread_version(), "NPTL")) {
 616      os::Linux::set_is_NPTL();
 617   } else {
 618      os::Linux::set_is_LinuxThreads();
 619   }
 620 
 621   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 622   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 623   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 624      os::Linux::set_is_floating_stack();
 625   }
 626 }
 627 
 628 /////////////////////////////////////////////////////////////////////////////
 629 // thread stack
 630 
 631 // Force Linux kernel to expand current thread stack. If "bottom" is close
 632 // to the stack guard, caller should block all signals.
 633 //
 634 // MAP_GROWSDOWN:
 635 //   A special mmap() flag that is used to implement thread stacks. It tells
 636 //   kernel that the memory region should extend downwards when needed. This
 637 //   allows early versions of LinuxThreads to only mmap the first few pages
 638 //   when creating a new thread. Linux kernel will automatically expand thread
 639 //   stack as needed (on page faults).
 640 //
 641 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 642 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 643 //   region, it's hard to tell if the fault is due to a legitimate stack
 644 //   access or because of reading/writing non-exist memory (e.g. buffer
 645 //   overrun). As a rule, if the fault happens below current stack pointer,
 646 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 647 //   application (see Linux kernel fault.c).
 648 //
 649 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 650 //   stack overflow detection.
 651 //
 652 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 653 //   not use this flag. However, the stack of initial thread is not created
 654 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 655 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 656 //   and then attach the thread to JVM.
 657 //
 658 // To get around the problem and allow stack banging on Linux, we need to
 659 // manually expand thread stack after receiving the SIGSEGV.
 660 //
 661 // There are two ways to expand thread stack to address "bottom", we used
 662 // both of them in JVM before 1.5:
 663 //   1. adjust stack pointer first so that it is below "bottom", and then
 664 //      touch "bottom"
 665 //   2. mmap() the page in question
 666 //
 667 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 668 // if current sp is already near the lower end of page 101, and we need to
 669 // call mmap() to map page 100, it is possible that part of the mmap() frame
 670 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 671 // That will destroy the mmap() frame and cause VM to crash.
 672 //
 673 // The following code works by adjusting sp first, then accessing the "bottom"
 674 // page to force a page fault. Linux kernel will then automatically expand the
 675 // stack mapping.
 676 //
 677 // _expand_stack_to() assumes its frame size is less than page size, which
 678 // should always be true if the function is not inlined.
 679 
 680 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 681 #define NOINLINE
 682 #else
 683 #define NOINLINE __attribute__ ((noinline))
 684 #endif
 685 
 686 static void _expand_stack_to(address bottom) NOINLINE;
 687 
 688 static void _expand_stack_to(address bottom) {
 689   address sp;
 690   size_t size;
 691   volatile char *p;
 692 
 693   // Adjust bottom to point to the largest address within the same page, it
 694   // gives us a one-page buffer if alloca() allocates slightly more memory.
 695   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 696   bottom += os::Linux::page_size() - 1;
 697 
 698   // sp might be slightly above current stack pointer; if that's the case, we
 699   // will alloca() a little more space than necessary, which is OK. Don't use
 700   // os::current_stack_pointer(), as its result can be slightly below current
 701   // stack pointer, causing us to not alloca enough to reach "bottom".
 702   sp = (address)&sp;
 703 
 704   if (sp > bottom) {
 705     size = sp - bottom;
 706     p = (volatile char *)alloca(size);
 707     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 708     p[0] = '\0';
 709   }
 710 }
 711 
 712 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 713   assert(t!=NULL, "just checking");
 714   assert(t->osthread()->expanding_stack(), "expand should be set");
 715   assert(t->stack_base() != NULL, "stack_base was not initialized");
 716 
 717   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 718     sigset_t mask_all, old_sigset;
 719     sigfillset(&mask_all);
 720     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 721     _expand_stack_to(addr);
 722     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 723     return true;
 724   }
 725   return false;
 726 }
 727 
 728 //////////////////////////////////////////////////////////////////////////////
 729 // create new thread
 730 
 731 static address highest_vm_reserved_address();
 732 
 733 // check if it's safe to start a new thread
 734 static bool _thread_safety_check(Thread* thread) {
 735   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 736     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 737     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 738     //   allocated (MAP_FIXED) from high address space. Every thread stack
 739     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 740     //   it to other values if they rebuild LinuxThreads).
 741     //
 742     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 743     // the memory region has already been mmap'ed. That means if we have too
 744     // many threads and/or very large heap, eventually thread stack will
 745     // collide with heap.
 746     //
 747     // Here we try to prevent heap/stack collision by comparing current
 748     // stack bottom with the highest address that has been mmap'ed by JVM
 749     // plus a safety margin for memory maps created by native code.
 750     //
 751     // This feature can be disabled by setting ThreadSafetyMargin to 0
 752     //
 753     if (ThreadSafetyMargin > 0) {
 754       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 755 
 756       // not safe if our stack extends below the safety margin
 757       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 758     } else {
 759       return true;
 760     }
 761   } else {
 762     // Floating stack LinuxThreads or NPTL:
 763     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 764     //   there's not enough space left, pthread_create() will fail. If we come
 765     //   here, that means enough space has been reserved for stack.
 766     return true;
 767   }
 768 }
 769 
 770 // Thread start routine for all newly created threads
 771 static void *java_start(Thread *thread) {
 772   // Try to randomize the cache line index of hot stack frames.
 773   // This helps when threads of the same stack traces evict each other's
 774   // cache lines. The threads can be either from the same JVM instance, or
 775   // from different JVM instances. The benefit is especially true for
 776   // processors with hyperthreading technology.
 777   static int counter = 0;
 778   int pid = os::current_process_id();
 779   alloca(((pid ^ counter++) & 7) * 128);
 780 
 781   ThreadLocalStorage::set_thread(thread);
 782 
 783   OSThread* osthread = thread->osthread();
 784   Monitor* sync = osthread->startThread_lock();
 785 
 786   // non floating stack LinuxThreads needs extra check, see above
 787   if (!_thread_safety_check(thread)) {
 788     // notify parent thread
 789     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 790     osthread->set_state(ZOMBIE);
 791     sync->notify_all();
 792     return NULL;
 793   }
 794 
 795   // thread_id is kernel thread id (similar to Solaris LWP id)
 796   osthread->set_thread_id(os::Linux::gettid());
 797 
 798   if (UseNUMA) {
 799     int lgrp_id = os::numa_get_group_id();
 800     if (lgrp_id != -1) {
 801       thread->set_lgrp_id(lgrp_id);
 802     }
 803   }
 804   // initialize signal mask for this thread
 805   os::Linux::hotspot_sigmask(thread);
 806 
 807   // initialize floating point control register
 808   os::Linux::init_thread_fpu_state();
 809 
 810   // handshaking with parent thread
 811   {
 812     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 813 
 814     // notify parent thread
 815     osthread->set_state(INITIALIZED);
 816     sync->notify_all();
 817 
 818     // wait until os::start_thread()
 819     while (osthread->get_state() == INITIALIZED) {
 820       sync->wait(Mutex::_no_safepoint_check_flag);
 821     }
 822   }
 823 
 824   // call one more level start routine
 825   thread->run();
 826 
 827   return 0;
 828 }
 829 
 830 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 831   assert(thread->osthread() == NULL, "caller responsible");
 832 
 833   // Allocate the OSThread object
 834   OSThread* osthread = new OSThread(NULL, NULL);
 835   if (osthread == NULL) {
 836     return false;
 837   }
 838 
 839   // set the correct thread state
 840   osthread->set_thread_type(thr_type);
 841 
 842   // Initial state is ALLOCATED but not INITIALIZED
 843   osthread->set_state(ALLOCATED);
 844 
 845   thread->set_osthread(osthread);
 846 
 847   // init thread attributes
 848   pthread_attr_t attr;
 849   pthread_attr_init(&attr);
 850   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 851 
 852   // stack size
 853   if (os::Linux::supports_variable_stack_size()) {
 854     // calculate stack size if it's not specified by caller
 855     if (stack_size == 0) {
 856       stack_size = os::Linux::default_stack_size(thr_type);
 857 
 858       switch (thr_type) {
 859       case os::java_thread:
 860         // Java threads use ThreadStackSize which default value can be
 861         // changed with the flag -Xss
 862         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 863         stack_size = JavaThread::stack_size_at_create();
 864         break;
 865       case os::compiler_thread:
 866         if (CompilerThreadStackSize > 0) {
 867           stack_size = (size_t)(CompilerThreadStackSize * K);
 868           break;
 869         } // else fall through:
 870           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 871       case os::vm_thread:
 872       case os::pgc_thread:
 873       case os::cgc_thread:
 874       case os::watcher_thread:
 875         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 876         break;
 877       }
 878     }
 879 
 880     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 881     pthread_attr_setstacksize(&attr, stack_size);
 882   } else {
 883     // let pthread_create() pick the default value.
 884   }
 885 
 886   // glibc guard page
 887   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 888 
 889   ThreadState state;
 890 
 891   {
 892     // Serialize thread creation if we are running with fixed stack LinuxThreads
 893     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 894     if (lock) {
 895       os::Linux::createThread_lock()->lock_without_safepoint_check();
 896     }
 897 
 898     pthread_t tid;
 899     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 900 
 901     pthread_attr_destroy(&attr);
 902 
 903     if (ret != 0) {
 904       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 905         perror("pthread_create()");
 906       }
 907       // Need to clean up stuff we've allocated so far
 908       thread->set_osthread(NULL);
 909       delete osthread;
 910       if (lock) os::Linux::createThread_lock()->unlock();
 911       return false;
 912     }
 913 
 914     // Store pthread info into the OSThread
 915     osthread->set_pthread_id(tid);
 916 
 917     // Wait until child thread is either initialized or aborted
 918     {
 919       Monitor* sync_with_child = osthread->startThread_lock();
 920       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 921       while ((state = osthread->get_state()) == ALLOCATED) {
 922         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 923       }
 924     }
 925 
 926     if (lock) {
 927       os::Linux::createThread_lock()->unlock();
 928     }
 929   }
 930 
 931   // Aborted due to thread limit being reached
 932   if (state == ZOMBIE) {
 933       thread->set_osthread(NULL);
 934       delete osthread;
 935       return false;
 936   }
 937 
 938   // The thread is returned suspended (in state INITIALIZED),
 939   // and is started higher up in the call chain
 940   assert(state == INITIALIZED, "race condition");
 941   return true;
 942 }
 943 
 944 /////////////////////////////////////////////////////////////////////////////
 945 // attach existing thread
 946 
 947 // bootstrap the main thread
 948 bool os::create_main_thread(JavaThread* thread) {
 949   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 950   return create_attached_thread(thread);
 951 }
 952 
 953 bool os::create_attached_thread(JavaThread* thread) {
 954 #ifdef ASSERT
 955     thread->verify_not_published();
 956 #endif
 957 
 958   // Allocate the OSThread object
 959   OSThread* osthread = new OSThread(NULL, NULL);
 960 
 961   if (osthread == NULL) {
 962     return false;
 963   }
 964 
 965   // Store pthread info into the OSThread
 966   osthread->set_thread_id(os::Linux::gettid());
 967   osthread->set_pthread_id(::pthread_self());
 968 
 969   // initialize floating point control register
 970   os::Linux::init_thread_fpu_state();
 971 
 972   // Initial thread state is RUNNABLE
 973   osthread->set_state(RUNNABLE);
 974 
 975   thread->set_osthread(osthread);
 976 
 977   if (UseNUMA) {
 978     int lgrp_id = os::numa_get_group_id();
 979     if (lgrp_id != -1) {
 980       thread->set_lgrp_id(lgrp_id);
 981     }
 982   }
 983 
 984   if (os::Linux::is_initial_thread()) {
 985     // If current thread is initial thread, its stack is mapped on demand,
 986     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 987     // the entire stack region to avoid SEGV in stack banging.
 988     // It is also useful to get around the heap-stack-gap problem on SuSE
 989     // kernel (see 4821821 for details). We first expand stack to the top
 990     // of yellow zone, then enable stack yellow zone (order is significant,
 991     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 992     // is no gap between the last two virtual memory regions.
 993 
 994     JavaThread *jt = (JavaThread *)thread;
 995     address addr = jt->stack_yellow_zone_base();
 996     assert(addr != NULL, "initialization problem?");
 997     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 998 
 999     osthread->set_expanding_stack();
1000     os::Linux::manually_expand_stack(jt, addr);
1001     osthread->clear_expanding_stack();
1002   }
1003 
1004   // initialize signal mask for this thread
1005   // and save the caller's signal mask
1006   os::Linux::hotspot_sigmask(thread);
1007 
1008   return true;
1009 }
1010 
1011 void os::pd_start_thread(Thread* thread) {
1012   OSThread * osthread = thread->osthread();
1013   assert(osthread->get_state() != INITIALIZED, "just checking");
1014   Monitor* sync_with_child = osthread->startThread_lock();
1015   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1016   sync_with_child->notify();
1017 }
1018 
1019 // Free Linux resources related to the OSThread
1020 void os::free_thread(OSThread* osthread) {
1021   assert(osthread != NULL, "osthread not set");
1022 
1023   if (Thread::current()->osthread() == osthread) {
1024     // Restore caller's signal mask
1025     sigset_t sigmask = osthread->caller_sigmask();
1026     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1027    }
1028 
1029   delete osthread;
1030 }
1031 
1032 //////////////////////////////////////////////////////////////////////////////
1033 // thread local storage
1034 
1035 // Restore the thread pointer if the destructor is called. This is in case
1036 // someone from JNI code sets up a destructor with pthread_key_create to run
1037 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1038 // will hang or crash. When detachCurrentThread is called the key will be set
1039 // to null and we will not be called again. If detachCurrentThread is never
1040 // called we could loop forever depending on the pthread implementation.
1041 static void restore_thread_pointer(void* p) {
1042   Thread* thread = (Thread*) p;
1043   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1044 }
1045 
1046 int os::allocate_thread_local_storage() {
1047   pthread_key_t key;
1048   int rslt = pthread_key_create(&key, restore_thread_pointer);
1049   assert(rslt == 0, "cannot allocate thread local storage");
1050   return (int)key;
1051 }
1052 
1053 // Note: This is currently not used by VM, as we don't destroy TLS key
1054 // on VM exit.
1055 void os::free_thread_local_storage(int index) {
1056   int rslt = pthread_key_delete((pthread_key_t)index);
1057   assert(rslt == 0, "invalid index");
1058 }
1059 
1060 void os::thread_local_storage_at_put(int index, void* value) {
1061   int rslt = pthread_setspecific((pthread_key_t)index, value);
1062   assert(rslt == 0, "pthread_setspecific failed");
1063 }
1064 
1065 extern "C" Thread* get_thread() {
1066   return ThreadLocalStorage::thread();
1067 }
1068 
1069 //////////////////////////////////////////////////////////////////////////////
1070 // initial thread
1071 
1072 // Check if current thread is the initial thread, similar to Solaris thr_main.
1073 bool os::Linux::is_initial_thread(void) {
1074   char dummy;
1075   // If called before init complete, thread stack bottom will be null.
1076   // Can be called if fatal error occurs before initialization.
1077   if (initial_thread_stack_bottom() == NULL) return false;
1078   assert(initial_thread_stack_bottom() != NULL &&
1079          initial_thread_stack_size()   != 0,
1080          "os::init did not locate initial thread's stack region");
1081   if ((address)&dummy >= initial_thread_stack_bottom() &&
1082       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1083        return true;
1084   else return false;
1085 }
1086 
1087 // Find the virtual memory area that contains addr
1088 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1089   FILE *fp = fopen("/proc/self/maps", "r");
1090   if (fp) {
1091     address low, high;
1092     while (!feof(fp)) {
1093       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1094         if (low <= addr && addr < high) {
1095            if (vma_low)  *vma_low  = low;
1096            if (vma_high) *vma_high = high;
1097            fclose (fp);
1098            return true;
1099         }
1100       }
1101       for (;;) {
1102         int ch = fgetc(fp);
1103         if (ch == EOF || ch == (int)'\n') break;
1104       }
1105     }
1106     fclose(fp);
1107   }
1108   return false;
1109 }
1110 
1111 // Locate initial thread stack. This special handling of initial thread stack
1112 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1113 // bogus value for initial thread.
1114 void os::Linux::capture_initial_stack(size_t max_size) {
1115   // stack size is the easy part, get it from RLIMIT_STACK
1116   size_t stack_size;
1117   struct rlimit rlim;
1118   getrlimit(RLIMIT_STACK, &rlim);
1119   stack_size = rlim.rlim_cur;
1120 
1121   // 6308388: a bug in ld.so will relocate its own .data section to the
1122   //   lower end of primordial stack; reduce ulimit -s value a little bit
1123   //   so we won't install guard page on ld.so's data section.
1124   stack_size -= 2 * page_size();
1125 
1126   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1127   //   7.1, in both cases we will get 2G in return value.
1128   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1129   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1130   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1131   //   in case other parts in glibc still assumes 2M max stack size.
1132   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1133   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1134   if (stack_size > 2 * K * K IA64_ONLY(*2))
1135       stack_size = 2 * K * K IA64_ONLY(*2);
1136   // Try to figure out where the stack base (top) is. This is harder.
1137   //
1138   // When an application is started, glibc saves the initial stack pointer in
1139   // a global variable "__libc_stack_end", which is then used by system
1140   // libraries. __libc_stack_end should be pretty close to stack top. The
1141   // variable is available since the very early days. However, because it is
1142   // a private interface, it could disappear in the future.
1143   //
1144   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1145   // to __libc_stack_end, it is very close to stack top, but isn't the real
1146   // stack top. Note that /proc may not exist if VM is running as a chroot
1147   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1148   // /proc/<pid>/stat could change in the future (though unlikely).
1149   //
1150   // We try __libc_stack_end first. If that doesn't work, look for
1151   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1152   // as a hint, which should work well in most cases.
1153 
1154   uintptr_t stack_start;
1155 
1156   // try __libc_stack_end first
1157   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1158   if (p && *p) {
1159     stack_start = *p;
1160   } else {
1161     // see if we can get the start_stack field from /proc/self/stat
1162     FILE *fp;
1163     int pid;
1164     char state;
1165     int ppid;
1166     int pgrp;
1167     int session;
1168     int nr;
1169     int tpgrp;
1170     unsigned long flags;
1171     unsigned long minflt;
1172     unsigned long cminflt;
1173     unsigned long majflt;
1174     unsigned long cmajflt;
1175     unsigned long utime;
1176     unsigned long stime;
1177     long cutime;
1178     long cstime;
1179     long prio;
1180     long nice;
1181     long junk;
1182     long it_real;
1183     uintptr_t start;
1184     uintptr_t vsize;
1185     intptr_t rss;
1186     uintptr_t rsslim;
1187     uintptr_t scodes;
1188     uintptr_t ecode;
1189     int i;
1190 
1191     // Figure what the primordial thread stack base is. Code is inspired
1192     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1193     // followed by command name surrounded by parentheses, state, etc.
1194     char stat[2048];
1195     int statlen;
1196 
1197     fp = fopen("/proc/self/stat", "r");
1198     if (fp) {
1199       statlen = fread(stat, 1, 2047, fp);
1200       stat[statlen] = '\0';
1201       fclose(fp);
1202 
1203       // Skip pid and the command string. Note that we could be dealing with
1204       // weird command names, e.g. user could decide to rename java launcher
1205       // to "java 1.4.2 :)", then the stat file would look like
1206       //                1234 (java 1.4.2 :)) R ... ...
1207       // We don't really need to know the command string, just find the last
1208       // occurrence of ")" and then start parsing from there. See bug 4726580.
1209       char * s = strrchr(stat, ')');
1210 
1211       i = 0;
1212       if (s) {
1213         // Skip blank chars
1214         do s++; while (isspace(*s));
1215 
1216 #define _UFM UINTX_FORMAT
1217 #define _DFM INTX_FORMAT
1218 
1219         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1220         /*              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 */
1221         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,
1222              &state,          /* 3  %c  */
1223              &ppid,           /* 4  %d  */
1224              &pgrp,           /* 5  %d  */
1225              &session,        /* 6  %d  */
1226              &nr,             /* 7  %d  */
1227              &tpgrp,          /* 8  %d  */
1228              &flags,          /* 9  %lu  */
1229              &minflt,         /* 10 %lu  */
1230              &cminflt,        /* 11 %lu  */
1231              &majflt,         /* 12 %lu  */
1232              &cmajflt,        /* 13 %lu  */
1233              &utime,          /* 14 %lu  */
1234              &stime,          /* 15 %lu  */
1235              &cutime,         /* 16 %ld  */
1236              &cstime,         /* 17 %ld  */
1237              &prio,           /* 18 %ld  */
1238              &nice,           /* 19 %ld  */
1239              &junk,           /* 20 %ld  */
1240              &it_real,        /* 21 %ld  */
1241              &start,          /* 22 UINTX_FORMAT */
1242              &vsize,          /* 23 UINTX_FORMAT */
1243              &rss,            /* 24 INTX_FORMAT  */
1244              &rsslim,         /* 25 UINTX_FORMAT */
1245              &scodes,         /* 26 UINTX_FORMAT */
1246              &ecode,          /* 27 UINTX_FORMAT */
1247              &stack_start);   /* 28 UINTX_FORMAT */
1248       }
1249 
1250 #undef _UFM
1251 #undef _DFM
1252 
1253       if (i != 28 - 2) {
1254          assert(false, "Bad conversion from /proc/self/stat");
1255          // product mode - assume we are the initial thread, good luck in the
1256          // embedded case.
1257          warning("Can't detect initial thread stack location - bad conversion");
1258          stack_start = (uintptr_t) &rlim;
1259       }
1260     } else {
1261       // For some reason we can't open /proc/self/stat (for example, running on
1262       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1263       // most cases, so don't abort:
1264       warning("Can't detect initial thread stack location - no /proc/self/stat");
1265       stack_start = (uintptr_t) &rlim;
1266     }
1267   }
1268 
1269   // Now we have a pointer (stack_start) very close to the stack top, the
1270   // next thing to do is to figure out the exact location of stack top. We
1271   // can find out the virtual memory area that contains stack_start by
1272   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1273   // and its upper limit is the real stack top. (again, this would fail if
1274   // running inside chroot, because /proc may not exist.)
1275 
1276   uintptr_t stack_top;
1277   address low, high;
1278   if (find_vma((address)stack_start, &low, &high)) {
1279     // success, "high" is the true stack top. (ignore "low", because initial
1280     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1281     stack_top = (uintptr_t)high;
1282   } else {
1283     // failed, likely because /proc/self/maps does not exist
1284     warning("Can't detect initial thread stack location - find_vma failed");
1285     // best effort: stack_start is normally within a few pages below the real
1286     // stack top, use it as stack top, and reduce stack size so we won't put
1287     // guard page outside stack.
1288     stack_top = stack_start;
1289     stack_size -= 16 * page_size();
1290   }
1291 
1292   // stack_top could be partially down the page so align it
1293   stack_top = align_size_up(stack_top, page_size());
1294 
1295   if (max_size && stack_size > max_size) {
1296      _initial_thread_stack_size = max_size;
1297   } else {
1298      _initial_thread_stack_size = stack_size;
1299   }
1300 
1301   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1302   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1303 }
1304 
1305 ////////////////////////////////////////////////////////////////////////////////
1306 // time support
1307 
1308 // Time since start-up in seconds to a fine granularity.
1309 // Used by VMSelfDestructTimer and the MemProfiler.
1310 double os::elapsedTime() {
1311 
1312   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1313 }
1314 
1315 jlong os::elapsed_counter() {
1316   return javaTimeNanos() - initial_time_count;
1317 }
1318 
1319 jlong os::elapsed_frequency() {
1320   return NANOSECS_PER_SEC; // nanosecond resolution
1321 }
1322 
1323 bool os::supports_vtime() { return true; }
1324 bool os::enable_vtime()   { return false; }
1325 bool os::vtime_enabled()  { return false; }
1326 
1327 double os::elapsedVTime() {
1328   struct rusage usage;
1329   int retval = getrusage(RUSAGE_THREAD, &usage);
1330   if (retval == 0) {
1331     return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1332   } else {
1333     // better than nothing, but not much
1334     return elapsedTime();
1335   }
1336 }
1337 
1338 jlong os::javaTimeMillis() {
1339   timeval time;
1340   int status = gettimeofday(&time, NULL);
1341   assert(status != -1, "linux error");
1342   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1343 }
1344 
1345 #ifndef CLOCK_MONOTONIC
1346 #define CLOCK_MONOTONIC (1)
1347 #endif
1348 
1349 void os::Linux::clock_init() {
1350   // we do dlopen's in this particular order due to bug in linux
1351   // dynamical loader (see 6348968) leading to crash on exit
1352   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1353   if (handle == NULL) {
1354     handle = dlopen("librt.so", RTLD_LAZY);
1355   }
1356 
1357   if (handle) {
1358     int (*clock_getres_func)(clockid_t, struct timespec*) =
1359            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1360     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1361            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1362     if (clock_getres_func && clock_gettime_func) {
1363       // See if monotonic clock is supported by the kernel. Note that some
1364       // early implementations simply return kernel jiffies (updated every
1365       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1366       // for nano time (though the monotonic property is still nice to have).
1367       // It's fixed in newer kernels, however clock_getres() still returns
1368       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1369       // resolution for now. Hopefully as people move to new kernels, this
1370       // won't be a problem.
1371       struct timespec res;
1372       struct timespec tp;
1373       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1374           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1375         // yes, monotonic clock is supported
1376         _clock_gettime = clock_gettime_func;
1377         return;
1378       } else {
1379         // close librt if there is no monotonic clock
1380         dlclose(handle);
1381       }
1382     }
1383   }
1384   warning("No monotonic clock was available - timed services may " \
1385           "be adversely affected if the time-of-day clock changes");
1386 }
1387 
1388 #ifndef SYS_clock_getres
1389 
1390 #if defined(IA32) || defined(AMD64)
1391 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1392 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1393 #else
1394 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1395 #define sys_clock_getres(x,y)  -1
1396 #endif
1397 
1398 #else
1399 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1400 #endif
1401 
1402 void os::Linux::fast_thread_clock_init() {
1403   if (!UseLinuxPosixThreadCPUClocks) {
1404     return;
1405   }
1406   clockid_t clockid;
1407   struct timespec tp;
1408   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1409       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1410 
1411   // Switch to using fast clocks for thread cpu time if
1412   // the sys_clock_getres() returns 0 error code.
1413   // Note, that some kernels may support the current thread
1414   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1415   // returned by the pthread_getcpuclockid().
1416   // If the fast Posix clocks are supported then the sys_clock_getres()
1417   // must return at least tp.tv_sec == 0 which means a resolution
1418   // better than 1 sec. This is extra check for reliability.
1419 
1420   if(pthread_getcpuclockid_func &&
1421      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1422      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1423 
1424     _supports_fast_thread_cpu_time = true;
1425     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1426   }
1427 }
1428 
1429 jlong os::javaTimeNanos() {
1430   if (os::supports_monotonic_clock()) {
1431     struct timespec tp;
1432     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1433     assert(status == 0, "gettime error");
1434     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1435     return result;
1436   } else {
1437     timeval time;
1438     int status = gettimeofday(&time, NULL);
1439     assert(status != -1, "linux error");
1440     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1441     return 1000 * usecs;
1442   }
1443 }
1444 
1445 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1446   if (os::supports_monotonic_clock()) {
1447     info_ptr->max_value = ALL_64_BITS;
1448 
1449     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1450     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1451     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1452   } else {
1453     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1454     info_ptr->max_value = ALL_64_BITS;
1455 
1456     // gettimeofday is a real time clock so it skips
1457     info_ptr->may_skip_backward = true;
1458     info_ptr->may_skip_forward = true;
1459   }
1460 
1461   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1462 }
1463 
1464 // Return the real, user, and system times in seconds from an
1465 // arbitrary fixed point in the past.
1466 bool os::getTimesSecs(double* process_real_time,
1467                       double* process_user_time,
1468                       double* process_system_time) {
1469   struct tms ticks;
1470   clock_t real_ticks = times(&ticks);
1471 
1472   if (real_ticks == (clock_t) (-1)) {
1473     return false;
1474   } else {
1475     double ticks_per_second = (double) clock_tics_per_sec;
1476     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1477     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1478     *process_real_time = ((double) real_ticks) / ticks_per_second;
1479 
1480     return true;
1481   }
1482 }
1483 
1484 
1485 char * os::local_time_string(char *buf, size_t buflen) {
1486   struct tm t;
1487   time_t long_time;
1488   time(&long_time);
1489   localtime_r(&long_time, &t);
1490   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1491                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1492                t.tm_hour, t.tm_min, t.tm_sec);
1493   return buf;
1494 }
1495 
1496 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1497   return localtime_r(clock, res);
1498 }
1499 
1500 ////////////////////////////////////////////////////////////////////////////////
1501 // runtime exit support
1502 
1503 // Note: os::shutdown() might be called very early during initialization, or
1504 // called from signal handler. Before adding something to os::shutdown(), make
1505 // sure it is async-safe and can handle partially initialized VM.
1506 void os::shutdown() {
1507 
1508   // allow PerfMemory to attempt cleanup of any persistent resources
1509   perfMemory_exit();
1510 
1511   // needs to remove object in file system
1512   AttachListener::abort();
1513 
1514   // flush buffered output, finish log files
1515   ostream_abort();
1516 
1517   // Check for abort hook
1518   abort_hook_t abort_hook = Arguments::abort_hook();
1519   if (abort_hook != NULL) {
1520     abort_hook();
1521   }
1522 
1523 }
1524 
1525 // Note: os::abort() might be called very early during initialization, or
1526 // called from signal handler. Before adding something to os::abort(), make
1527 // sure it is async-safe and can handle partially initialized VM.
1528 void os::abort(bool dump_core) {
1529   os::shutdown();
1530   if (dump_core) {
1531 #ifndef PRODUCT
1532     fdStream out(defaultStream::output_fd());
1533     out.print_raw("Current thread is ");
1534     char buf[16];
1535     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1536     out.print_raw_cr(buf);
1537     out.print_raw_cr("Dumping core ...");
1538 #endif
1539     ::abort(); // dump core
1540   }
1541 
1542   ::exit(1);
1543 }
1544 
1545 // Die immediately, no exit hook, no abort hook, no cleanup.
1546 void os::die() {
1547   // _exit() on LinuxThreads only kills current thread
1548   ::abort();
1549 }
1550 
1551 // unused on linux for now.
1552 void os::set_error_file(const char *logfile) {}
1553 
1554 
1555 // This method is a copy of JDK's sysGetLastErrorString
1556 // from src/solaris/hpi/src/system_md.c
1557 
1558 size_t os::lasterror(char *buf, size_t len) {
1559 
1560   if (errno == 0)  return 0;
1561 
1562   const char *s = ::strerror(errno);
1563   size_t n = ::strlen(s);
1564   if (n >= len) {
1565     n = len - 1;
1566   }
1567   ::strncpy(buf, s, n);
1568   buf[n] = '\0';
1569   return n;
1570 }
1571 
1572 intx os::current_thread_id() { return (intx)pthread_self(); }
1573 int os::current_process_id() {
1574 
1575   // Under the old linux thread library, linux gives each thread
1576   // its own process id. Because of this each thread will return
1577   // a different pid if this method were to return the result
1578   // of getpid(2). Linux provides no api that returns the pid
1579   // of the launcher thread for the vm. This implementation
1580   // returns a unique pid, the pid of the launcher thread
1581   // that starts the vm 'process'.
1582 
1583   // Under the NPTL, getpid() returns the same pid as the
1584   // launcher thread rather than a unique pid per thread.
1585   // Use gettid() if you want the old pre NPTL behaviour.
1586 
1587   // if you are looking for the result of a call to getpid() that
1588   // returns a unique pid for the calling thread, then look at the
1589   // OSThread::thread_id() method in osThread_linux.hpp file
1590 
1591   return (int)(_initial_pid ? _initial_pid : getpid());
1592 }
1593 
1594 // DLL functions
1595 
1596 const char* os::dll_file_extension() { return ".so"; }
1597 
1598 // This must be hard coded because it's the system's temporary
1599 // directory not the java application's temp directory, ala java.io.tmpdir.
1600 const char* os::get_temp_directory() { return "/tmp"; }
1601 
1602 static bool file_exists(const char* filename) {
1603   struct stat statbuf;
1604   if (filename == NULL || strlen(filename) == 0) {
1605     return false;
1606   }
1607   return os::stat(filename, &statbuf) == 0;
1608 }
1609 
1610 bool os::dll_build_name(char* buffer, size_t buflen,
1611                         const char* pname, const char* fname) {
1612   bool retval = false;
1613   // Copied from libhpi
1614   const size_t pnamelen = pname ? strlen(pname) : 0;
1615 
1616   // Return error on buffer overflow.
1617   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1618     return retval;
1619   }
1620 
1621   if (pnamelen == 0) {
1622     snprintf(buffer, buflen, "lib%s.so", fname);
1623     retval = true;
1624   } else if (strchr(pname, *os::path_separator()) != NULL) {
1625     int n;
1626     char** pelements = split_path(pname, &n);
1627     if (pelements == NULL) {
1628       return false;
1629     }
1630     for (int i = 0 ; i < n ; i++) {
1631       // Really shouldn't be NULL, but check can't hurt
1632       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1633         continue; // skip the empty path values
1634       }
1635       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1636       if (file_exists(buffer)) {
1637         retval = true;
1638         break;
1639       }
1640     }
1641     // release the storage
1642     for (int i = 0 ; i < n ; i++) {
1643       if (pelements[i] != NULL) {
1644         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1645       }
1646     }
1647     if (pelements != NULL) {
1648       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1649     }
1650   } else {
1651     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1652     retval = true;
1653   }
1654   return retval;
1655 }
1656 
1657 // check if addr is inside libjvm.so
1658 bool os::address_is_in_vm(address addr) {
1659   static address libjvm_base_addr;
1660   Dl_info dlinfo;
1661 
1662   if (libjvm_base_addr == NULL) {
1663     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1664       libjvm_base_addr = (address)dlinfo.dli_fbase;
1665     }
1666     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1667   }
1668 
1669   if (dladdr((void *)addr, &dlinfo) != 0) {
1670     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1671   }
1672 
1673   return false;
1674 }
1675 
1676 bool os::dll_address_to_function_name(address addr, char *buf,
1677                                       int buflen, int *offset) {
1678   // buf is not optional, but offset is optional
1679   assert(buf != NULL, "sanity check");
1680 
1681   Dl_info dlinfo;
1682 
1683   if (dladdr((void*)addr, &dlinfo) != 0) {
1684     // see if we have a matching symbol
1685     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1686       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1687         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1688       }
1689       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1690       return true;
1691     }
1692     // no matching symbol so try for just file info
1693     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1694       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1695                           buf, buflen, offset, dlinfo.dli_fname)) {
1696         return true;
1697       }
1698     }
1699   }
1700 
1701   buf[0] = '\0';
1702   if (offset != NULL) *offset = -1;
1703   return false;
1704 }
1705 
1706 struct _address_to_library_name {
1707   address addr;          // input : memory address
1708   size_t  buflen;        //         size of fname
1709   char*   fname;         // output: library name
1710   address base;          //         library base addr
1711 };
1712 
1713 static int address_to_library_name_callback(struct dl_phdr_info *info,
1714                                             size_t size, void *data) {
1715   int i;
1716   bool found = false;
1717   address libbase = NULL;
1718   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1719 
1720   // iterate through all loadable segments
1721   for (i = 0; i < info->dlpi_phnum; i++) {
1722     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1723     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1724       // base address of a library is the lowest address of its loaded
1725       // segments.
1726       if (libbase == NULL || libbase > segbase) {
1727         libbase = segbase;
1728       }
1729       // see if 'addr' is within current segment
1730       if (segbase <= d->addr &&
1731           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1732         found = true;
1733       }
1734     }
1735   }
1736 
1737   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1738   // so dll_address_to_library_name() can fall through to use dladdr() which
1739   // can figure out executable name from argv[0].
1740   if (found && info->dlpi_name && info->dlpi_name[0]) {
1741     d->base = libbase;
1742     if (d->fname) {
1743       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1744     }
1745     return 1;
1746   }
1747   return 0;
1748 }
1749 
1750 bool os::dll_address_to_library_name(address addr, char* buf,
1751                                      int buflen, int* offset) {
1752   // buf is not optional, but offset is optional
1753   assert(buf != NULL, "sanity check");
1754 
1755   Dl_info dlinfo;
1756   struct _address_to_library_name data;
1757 
1758   // There is a bug in old glibc dladdr() implementation that it could resolve
1759   // to wrong library name if the .so file has a base address != NULL. Here
1760   // we iterate through the program headers of all loaded libraries to find
1761   // out which library 'addr' really belongs to. This workaround can be
1762   // removed once the minimum requirement for glibc is moved to 2.3.x.
1763   data.addr = addr;
1764   data.fname = buf;
1765   data.buflen = buflen;
1766   data.base = NULL;
1767   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1768 
1769   if (rslt) {
1770      // buf already contains library name
1771      if (offset) *offset = addr - data.base;
1772      return true;
1773   }
1774   if (dladdr((void*)addr, &dlinfo) != 0) {
1775     if (dlinfo.dli_fname != NULL) {
1776       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1777     }
1778     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1779       *offset = addr - (address)dlinfo.dli_fbase;
1780     }
1781     return true;
1782   }
1783 
1784   buf[0] = '\0';
1785   if (offset) *offset = -1;
1786   return false;
1787 }
1788 
1789   // Loads .dll/.so and
1790   // in case of error it checks if .dll/.so was built for the
1791   // same architecture as Hotspot is running on
1792 
1793 
1794 // Remember the stack's state. The Linux dynamic linker will change
1795 // the stack to 'executable' at most once, so we must safepoint only once.
1796 bool os::Linux::_stack_is_executable = false;
1797 
1798 // VM operation that loads a library.  This is necessary if stack protection
1799 // of the Java stacks can be lost during loading the library.  If we
1800 // do not stop the Java threads, they can stack overflow before the stacks
1801 // are protected again.
1802 class VM_LinuxDllLoad: public VM_Operation {
1803  private:
1804   const char *_filename;
1805   char *_ebuf;
1806   int _ebuflen;
1807   void *_lib;
1808  public:
1809   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1810     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1811   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1812   void doit() {
1813     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1814     os::Linux::_stack_is_executable = true;
1815   }
1816   void* loaded_library() { return _lib; }
1817 };
1818 
1819 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1820 {
1821   void * result = NULL;
1822   bool load_attempted = false;
1823 
1824   // Check whether the library to load might change execution rights
1825   // of the stack. If they are changed, the protection of the stack
1826   // guard pages will be lost. We need a safepoint to fix this.
1827   //
1828   // See Linux man page execstack(8) for more info.
1829   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1830     ElfFile ef(filename);
1831     if (!ef.specifies_noexecstack()) {
1832       if (!is_init_completed()) {
1833         os::Linux::_stack_is_executable = true;
1834         // This is OK - No Java threads have been created yet, and hence no
1835         // stack guard pages to fix.
1836         //
1837         // This should happen only when you are building JDK7 using a very
1838         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1839         //
1840         // Dynamic loader will make all stacks executable after
1841         // this function returns, and will not do that again.
1842         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1843       } else {
1844         warning("You have loaded library %s which might have disabled stack guard. "
1845                 "The VM will try to fix the stack guard now.\n"
1846                 "It's highly recommended that you fix the library with "
1847                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1848                 filename);
1849 
1850         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1851         JavaThread *jt = JavaThread::current();
1852         if (jt->thread_state() != _thread_in_native) {
1853           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1854           // that requires ExecStack. Cannot enter safe point. Let's give up.
1855           warning("Unable to fix stack guard. Giving up.");
1856         } else {
1857           if (!LoadExecStackDllInVMThread) {
1858             // This is for the case where the DLL has an static
1859             // constructor function that executes JNI code. We cannot
1860             // load such DLLs in the VMThread.
1861             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1862           }
1863 
1864           ThreadInVMfromNative tiv(jt);
1865           debug_only(VMNativeEntryWrapper vew;)
1866 
1867           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1868           VMThread::execute(&op);
1869           if (LoadExecStackDllInVMThread) {
1870             result = op.loaded_library();
1871           }
1872           load_attempted = true;
1873         }
1874       }
1875     }
1876   }
1877 
1878   if (!load_attempted) {
1879     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1880   }
1881 
1882   if (result != NULL) {
1883     // Successful loading
1884     return result;
1885   }
1886 
1887   Elf32_Ehdr elf_head;
1888   int diag_msg_max_length=ebuflen-strlen(ebuf);
1889   char* diag_msg_buf=ebuf+strlen(ebuf);
1890 
1891   if (diag_msg_max_length==0) {
1892     // No more space in ebuf for additional diagnostics message
1893     return NULL;
1894   }
1895 
1896 
1897   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1898 
1899   if (file_descriptor < 0) {
1900     // Can't open library, report dlerror() message
1901     return NULL;
1902   }
1903 
1904   bool failed_to_read_elf_head=
1905     (sizeof(elf_head)!=
1906         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1907 
1908   ::close(file_descriptor);
1909   if (failed_to_read_elf_head) {
1910     // file i/o error - report dlerror() msg
1911     return NULL;
1912   }
1913 
1914   typedef struct {
1915     Elf32_Half  code;         // Actual value as defined in elf.h
1916     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1917     char        elf_class;    // 32 or 64 bit
1918     char        endianess;    // MSB or LSB
1919     char*       name;         // String representation
1920   } arch_t;
1921 
1922   #ifndef EM_486
1923   #define EM_486          6               /* Intel 80486 */
1924   #endif
1925 
1926   static const arch_t arch_array[]={
1927     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1928     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1929     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1930     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1931     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1932     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1933     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1934     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1935 #if defined(VM_LITTLE_ENDIAN)
1936     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1937 #else
1938     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1939 #endif
1940     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1941     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1942     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1943     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1944     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1945     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1946     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1947   };
1948 
1949   #if  (defined IA32)
1950     static  Elf32_Half running_arch_code=EM_386;
1951   #elif   (defined AMD64)
1952     static  Elf32_Half running_arch_code=EM_X86_64;
1953   #elif  (defined IA64)
1954     static  Elf32_Half running_arch_code=EM_IA_64;
1955   #elif  (defined __sparc) && (defined _LP64)
1956     static  Elf32_Half running_arch_code=EM_SPARCV9;
1957   #elif  (defined __sparc) && (!defined _LP64)
1958     static  Elf32_Half running_arch_code=EM_SPARC;
1959   #elif  (defined __powerpc64__)
1960     static  Elf32_Half running_arch_code=EM_PPC64;
1961   #elif  (defined __powerpc__)
1962     static  Elf32_Half running_arch_code=EM_PPC;
1963   #elif  (defined ARM)
1964     static  Elf32_Half running_arch_code=EM_ARM;
1965   #elif  (defined S390)
1966     static  Elf32_Half running_arch_code=EM_S390;
1967   #elif  (defined ALPHA)
1968     static  Elf32_Half running_arch_code=EM_ALPHA;
1969   #elif  (defined MIPSEL)
1970     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1971   #elif  (defined PARISC)
1972     static  Elf32_Half running_arch_code=EM_PARISC;
1973   #elif  (defined MIPS)
1974     static  Elf32_Half running_arch_code=EM_MIPS;
1975   #elif  (defined M68K)
1976     static  Elf32_Half running_arch_code=EM_68K;
1977   #else
1978     #error Method os::dll_load requires that one of following is defined:\
1979          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1980   #endif
1981 
1982   // Identify compatability class for VM's architecture and library's architecture
1983   // Obtain string descriptions for architectures
1984 
1985   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1986   int running_arch_index=-1;
1987 
1988   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1989     if (running_arch_code == arch_array[i].code) {
1990       running_arch_index    = i;
1991     }
1992     if (lib_arch.code == arch_array[i].code) {
1993       lib_arch.compat_class = arch_array[i].compat_class;
1994       lib_arch.name         = arch_array[i].name;
1995     }
1996   }
1997 
1998   assert(running_arch_index != -1,
1999     "Didn't find running architecture code (running_arch_code) in arch_array");
2000   if (running_arch_index == -1) {
2001     // Even though running architecture detection failed
2002     // we may still continue with reporting dlerror() message
2003     return NULL;
2004   }
2005 
2006   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2007     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2008     return NULL;
2009   }
2010 
2011 #ifndef S390
2012   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2013     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2014     return NULL;
2015   }
2016 #endif // !S390
2017 
2018   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2019     if ( lib_arch.name!=NULL ) {
2020       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2021         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2022         lib_arch.name, arch_array[running_arch_index].name);
2023     } else {
2024       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2025       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2026         lib_arch.code,
2027         arch_array[running_arch_index].name);
2028     }
2029   }
2030 
2031   return NULL;
2032 }
2033 
2034 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2035   void * result = ::dlopen(filename, RTLD_LAZY);
2036   if (result == NULL) {
2037     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2038     ebuf[ebuflen-1] = '\0';
2039   }
2040   return result;
2041 }
2042 
2043 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2044   void * result = NULL;
2045   if (LoadExecStackDllInVMThread) {
2046     result = dlopen_helper(filename, ebuf, ebuflen);
2047   }
2048 
2049   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2050   // library that requires an executable stack, or which does not have this
2051   // stack attribute set, dlopen changes the stack attribute to executable. The
2052   // read protection of the guard pages gets lost.
2053   //
2054   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2055   // may have been queued at the same time.
2056 
2057   if (!_stack_is_executable) {
2058     JavaThread *jt = Threads::first();
2059 
2060     while (jt) {
2061       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2062           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2063         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2064                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2065           warning("Attempt to reguard stack yellow zone failed.");
2066         }
2067       }
2068       jt = jt->next();
2069     }
2070   }
2071 
2072   return result;
2073 }
2074 
2075 /*
2076  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2077  * chances are you might want to run the generated bits against glibc-2.0
2078  * libdl.so, so always use locking for any version of glibc.
2079  */
2080 void* os::dll_lookup(void* handle, const char* name) {
2081   pthread_mutex_lock(&dl_mutex);
2082   void* res = dlsym(handle, name);
2083   pthread_mutex_unlock(&dl_mutex);
2084   return res;
2085 }
2086 
2087 void* os::get_default_process_handle() {
2088   return (void*)::dlopen(NULL, RTLD_LAZY);
2089 }
2090 
2091 static bool _print_ascii_file(const char* filename, outputStream* st) {
2092   int fd = ::open(filename, O_RDONLY);
2093   if (fd == -1) {
2094      return false;
2095   }
2096 
2097   char buf[32];
2098   int bytes;
2099   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2100     st->print_raw(buf, bytes);
2101   }
2102 
2103   ::close(fd);
2104 
2105   return true;
2106 }
2107 
2108 void os::print_dll_info(outputStream *st) {
2109    st->print_cr("Dynamic libraries:");
2110 
2111    char fname[32];
2112    pid_t pid = os::Linux::gettid();
2113 
2114    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2115 
2116    if (!_print_ascii_file(fname, st)) {
2117      st->print("Can not get library information for pid = %d\n", pid);
2118    }
2119 }
2120 
2121 void os::print_os_info_brief(outputStream* st) {
2122   os::Linux::print_distro_info(st);
2123 
2124   os::Posix::print_uname_info(st);
2125 
2126   os::Linux::print_libversion_info(st);
2127 
2128 }
2129 
2130 void os::print_os_info(outputStream* st) {
2131   st->print("OS:");
2132 
2133   os::Linux::print_distro_info(st);
2134 
2135   os::Posix::print_uname_info(st);
2136 
2137   // Print warning if unsafe chroot environment detected
2138   if (unsafe_chroot_detected) {
2139     st->print("WARNING!! ");
2140     st->print_cr(unstable_chroot_error);
2141   }
2142 
2143   os::Linux::print_libversion_info(st);
2144 
2145   os::Posix::print_rlimit_info(st);
2146 
2147   os::Posix::print_load_average(st);
2148 
2149   os::Linux::print_full_memory_info(st);
2150 }
2151 
2152 // Try to identify popular distros.
2153 // Most Linux distributions have a /etc/XXX-release file, which contains
2154 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2155 // file that also contains the OS version string. Some have more than one
2156 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2157 // /etc/redhat-release.), so the order is important.
2158 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2159 // their own specific XXX-release file as well as a redhat-release file.
2160 // Because of this the XXX-release file needs to be searched for before the
2161 // redhat-release file.
2162 // Since Red Hat has a lsb-release file that is not very descriptive the
2163 // search for redhat-release needs to be before lsb-release.
2164 // Since the lsb-release file is the new standard it needs to be searched
2165 // before the older style release files.
2166 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2167 // next to last resort.  The os-release file is a new standard that contains
2168 // distribution information and the system-release file seems to be an old
2169 // standard that has been replaced by the lsb-release and os-release files.
2170 // Searching for the debian_version file is the last resort.  It contains
2171 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2172 // "Debian " is printed before the contents of the debian_version file.
2173 void os::Linux::print_distro_info(outputStream* st) {
2174    if (!_print_ascii_file("/etc/oracle-release", st) &&
2175        !_print_ascii_file("/etc/mandriva-release", st) &&
2176        !_print_ascii_file("/etc/mandrake-release", st) &&
2177        !_print_ascii_file("/etc/sun-release", st) &&
2178        !_print_ascii_file("/etc/redhat-release", st) &&
2179        !_print_ascii_file("/etc/lsb-release", st) &&
2180        !_print_ascii_file("/etc/SuSE-release", st) &&
2181        !_print_ascii_file("/etc/turbolinux-release", st) &&
2182        !_print_ascii_file("/etc/gentoo-release", st) &&
2183        !_print_ascii_file("/etc/ltib-release", st) &&
2184        !_print_ascii_file("/etc/angstrom-version", st) &&
2185        !_print_ascii_file("/etc/system-release", st) &&
2186        !_print_ascii_file("/etc/os-release", st)) {
2187 
2188        if (file_exists("/etc/debian_version")) {
2189          st->print("Debian ");
2190          _print_ascii_file("/etc/debian_version", st);
2191        } else {
2192          st->print("Linux");
2193        }
2194    }
2195    st->cr();
2196 }
2197 
2198 void os::Linux::print_libversion_info(outputStream* st) {
2199   // libc, pthread
2200   st->print("libc:");
2201   st->print(os::Linux::glibc_version()); st->print(" ");
2202   st->print(os::Linux::libpthread_version()); st->print(" ");
2203   if (os::Linux::is_LinuxThreads()) {
2204      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2205   }
2206   st->cr();
2207 }
2208 
2209 void os::Linux::print_full_memory_info(outputStream* st) {
2210    st->print("\n/proc/meminfo:\n");
2211    _print_ascii_file("/proc/meminfo", st);
2212    st->cr();
2213 }
2214 
2215 void os::print_memory_info(outputStream* st) {
2216 
2217   st->print("Memory:");
2218   st->print(" %dk page", os::vm_page_size()>>10);
2219 
2220   // values in struct sysinfo are "unsigned long"
2221   struct sysinfo si;
2222   sysinfo(&si);
2223 
2224   st->print(", physical " UINT64_FORMAT "k",
2225             os::physical_memory() >> 10);
2226   st->print("(" UINT64_FORMAT "k free)",
2227             os::available_memory() >> 10);
2228   st->print(", swap " UINT64_FORMAT "k",
2229             ((jlong)si.totalswap * si.mem_unit) >> 10);
2230   st->print("(" UINT64_FORMAT "k free)",
2231             ((jlong)si.freeswap * si.mem_unit) >> 10);
2232   st->cr();
2233 }
2234 
2235 void os::pd_print_cpu_info(outputStream* st) {
2236   st->print("\n/proc/cpuinfo:\n");
2237   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2238     st->print("  <Not Available>");
2239   }
2240   st->cr();
2241 }
2242 
2243 void os::print_siginfo(outputStream* st, void* siginfo) {
2244   const siginfo_t* si = (const siginfo_t*)siginfo;
2245 
2246   os::Posix::print_siginfo_brief(st, si);
2247 
2248   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2249       UseSharedSpaces) {
2250     FileMapInfo* mapinfo = FileMapInfo::current_info();
2251     if (mapinfo->is_in_shared_space(si->si_addr)) {
2252       st->print("\n\nError accessing class data sharing archive."   \
2253                 " Mapped file inaccessible during execution, "      \
2254                 " possible disk/network problem.");
2255     }
2256   }
2257   st->cr();
2258 }
2259 
2260 
2261 static void print_signal_handler(outputStream* st, int sig,
2262                                  char* buf, size_t buflen);
2263 
2264 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2265   st->print_cr("Signal Handlers:");
2266   print_signal_handler(st, SIGSEGV, buf, buflen);
2267   print_signal_handler(st, SIGBUS , buf, buflen);
2268   print_signal_handler(st, SIGFPE , buf, buflen);
2269   print_signal_handler(st, SIGPIPE, buf, buflen);
2270   print_signal_handler(st, SIGXFSZ, buf, buflen);
2271   print_signal_handler(st, SIGILL , buf, buflen);
2272   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2273   print_signal_handler(st, SR_signum, buf, buflen);
2274   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2275   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2276   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2277   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2278 #if defined(PPC64)
2279   print_signal_handler(st, SIGTRAP, buf, buflen);
2280 #endif
2281 }
2282 
2283 static char saved_jvm_path[MAXPATHLEN] = {0};
2284 
2285 // Find the full path to the current module, libjvm.so
2286 void os::jvm_path(char *buf, jint buflen) {
2287   // Error checking.
2288   if (buflen < MAXPATHLEN) {
2289     assert(false, "must use a large-enough buffer");
2290     buf[0] = '\0';
2291     return;
2292   }
2293   // Lazy resolve the path to current module.
2294   if (saved_jvm_path[0] != 0) {
2295     strcpy(buf, saved_jvm_path);
2296     return;
2297   }
2298 
2299   char dli_fname[MAXPATHLEN];
2300   bool ret = dll_address_to_library_name(
2301                 CAST_FROM_FN_PTR(address, os::jvm_path),
2302                 dli_fname, sizeof(dli_fname), NULL);
2303   assert(ret, "cannot locate libjvm");
2304   char *rp = NULL;
2305   if (ret && dli_fname[0] != '\0') {
2306     rp = realpath(dli_fname, buf);
2307   }
2308   if (rp == NULL)
2309     return;
2310 
2311   if (Arguments::sun_java_launcher_is_altjvm()) {
2312     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2313     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2314     // If "/jre/lib/" appears at the right place in the string, then
2315     // assume we are installed in a JDK and we're done. Otherwise, check
2316     // for a JAVA_HOME environment variable and fix up the path so it
2317     // looks like libjvm.so is installed there (append a fake suffix
2318     // hotspot/libjvm.so).
2319     const char *p = buf + strlen(buf) - 1;
2320     for (int count = 0; p > buf && count < 5; ++count) {
2321       for (--p; p > buf && *p != '/'; --p)
2322         /* empty */ ;
2323     }
2324 
2325     if (strncmp(p, "/jre/lib/", 9) != 0) {
2326       // Look for JAVA_HOME in the environment.
2327       char* java_home_var = ::getenv("JAVA_HOME");
2328       if (java_home_var != NULL && java_home_var[0] != 0) {
2329         char* jrelib_p;
2330         int len;
2331 
2332         // Check the current module name "libjvm.so".
2333         p = strrchr(buf, '/');
2334         assert(strstr(p, "/libjvm") == p, "invalid library name");
2335 
2336         rp = realpath(java_home_var, buf);
2337         if (rp == NULL)
2338           return;
2339 
2340         // determine if this is a legacy image or modules image
2341         // modules image doesn't have "jre" subdirectory
2342         len = strlen(buf);
2343         jrelib_p = buf + len;
2344         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2345         if (0 != access(buf, F_OK)) {
2346           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2347         }
2348 
2349         if (0 == access(buf, F_OK)) {
2350           // Use current module name "libjvm.so"
2351           len = strlen(buf);
2352           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2353         } else {
2354           // Go back to path of .so
2355           rp = realpath(dli_fname, buf);
2356           if (rp == NULL)
2357             return;
2358         }
2359       }
2360     }
2361   }
2362 
2363   strcpy(saved_jvm_path, buf);
2364 }
2365 
2366 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2367   // no prefix required, not even "_"
2368 }
2369 
2370 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2371   // no suffix required
2372 }
2373 
2374 ////////////////////////////////////////////////////////////////////////////////
2375 // sun.misc.Signal support
2376 
2377 static volatile jint sigint_count = 0;
2378 
2379 static void
2380 UserHandler(int sig, void *siginfo, void *context) {
2381   // 4511530 - sem_post is serialized and handled by the manager thread. When
2382   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2383   // don't want to flood the manager thread with sem_post requests.
2384   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2385       return;
2386 
2387   // Ctrl-C is pressed during error reporting, likely because the error
2388   // handler fails to abort. Let VM die immediately.
2389   if (sig == SIGINT && is_error_reported()) {
2390      os::die();
2391   }
2392 
2393   os::signal_notify(sig);
2394 }
2395 
2396 void* os::user_handler() {
2397   return CAST_FROM_FN_PTR(void*, UserHandler);
2398 }
2399 
2400 class Semaphore : public StackObj {
2401   public:
2402     Semaphore();
2403     ~Semaphore();
2404     void signal();
2405     void wait();
2406     bool trywait();
2407     bool timedwait(unsigned int sec, int nsec);
2408   private:
2409     sem_t _semaphore;
2410 };
2411 
2412 Semaphore::Semaphore() {
2413   sem_init(&_semaphore, 0, 0);
2414 }
2415 
2416 Semaphore::~Semaphore() {
2417   sem_destroy(&_semaphore);
2418 }
2419 
2420 void Semaphore::signal() {
2421   sem_post(&_semaphore);
2422 }
2423 
2424 void Semaphore::wait() {
2425   sem_wait(&_semaphore);
2426 }
2427 
2428 bool Semaphore::trywait() {
2429   return sem_trywait(&_semaphore) == 0;
2430 }
2431 
2432 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2433 
2434   struct timespec ts;
2435   // Semaphore's are always associated with CLOCK_REALTIME
2436   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2437   // see unpackTime for discussion on overflow checking
2438   if (sec >= MAX_SECS) {
2439     ts.tv_sec += MAX_SECS;
2440     ts.tv_nsec = 0;
2441   } else {
2442     ts.tv_sec += sec;
2443     ts.tv_nsec += nsec;
2444     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2445       ts.tv_nsec -= NANOSECS_PER_SEC;
2446       ++ts.tv_sec; // note: this must be <= max_secs
2447     }
2448   }
2449 
2450   while (1) {
2451     int result = sem_timedwait(&_semaphore, &ts);
2452     if (result == 0) {
2453       return true;
2454     } else if (errno == EINTR) {
2455       continue;
2456     } else if (errno == ETIMEDOUT) {
2457       return false;
2458     } else {
2459       return false;
2460     }
2461   }
2462 }
2463 
2464 extern "C" {
2465   typedef void (*sa_handler_t)(int);
2466   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2467 }
2468 
2469 void* os::signal(int signal_number, void* handler) {
2470   struct sigaction sigAct, oldSigAct;
2471 
2472   sigfillset(&(sigAct.sa_mask));
2473   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2474   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2475 
2476   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2477     // -1 means registration failed
2478     return (void *)-1;
2479   }
2480 
2481   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2482 }
2483 
2484 void os::signal_raise(int signal_number) {
2485   ::raise(signal_number);
2486 }
2487 
2488 /*
2489  * The following code is moved from os.cpp for making this
2490  * code platform specific, which it is by its very nature.
2491  */
2492 
2493 // Will be modified when max signal is changed to be dynamic
2494 int os::sigexitnum_pd() {
2495   return NSIG;
2496 }
2497 
2498 // a counter for each possible signal value
2499 static volatile jint pending_signals[NSIG+1] = { 0 };
2500 
2501 // Linux(POSIX) specific hand shaking semaphore.
2502 static sem_t sig_sem;
2503 static Semaphore sr_semaphore;
2504 
2505 void os::signal_init_pd() {
2506   // Initialize signal structures
2507   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2508 
2509   // Initialize signal semaphore
2510   ::sem_init(&sig_sem, 0, 0);
2511 }
2512 
2513 void os::signal_notify(int sig) {
2514   Atomic::inc(&pending_signals[sig]);
2515   ::sem_post(&sig_sem);
2516 }
2517 
2518 static int check_pending_signals(bool wait) {
2519   Atomic::store(0, &sigint_count);
2520   for (;;) {
2521     for (int i = 0; i < NSIG + 1; i++) {
2522       jint n = pending_signals[i];
2523       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2524         return i;
2525       }
2526     }
2527     if (!wait) {
2528       return -1;
2529     }
2530     JavaThread *thread = JavaThread::current();
2531     ThreadBlockInVM tbivm(thread);
2532 
2533     bool threadIsSuspended;
2534     do {
2535       thread->set_suspend_equivalent();
2536       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2537       ::sem_wait(&sig_sem);
2538 
2539       // were we externally suspended while we were waiting?
2540       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2541       if (threadIsSuspended) {
2542         //
2543         // The semaphore has been incremented, but while we were waiting
2544         // another thread suspended us. We don't want to continue running
2545         // while suspended because that would surprise the thread that
2546         // suspended us.
2547         //
2548         ::sem_post(&sig_sem);
2549 
2550         thread->java_suspend_self();
2551       }
2552     } while (threadIsSuspended);
2553   }
2554 }
2555 
2556 int os::signal_lookup() {
2557   return check_pending_signals(false);
2558 }
2559 
2560 int os::signal_wait() {
2561   return check_pending_signals(true);
2562 }
2563 
2564 ////////////////////////////////////////////////////////////////////////////////
2565 // Virtual Memory
2566 
2567 int os::vm_page_size() {
2568   // Seems redundant as all get out
2569   assert(os::Linux::page_size() != -1, "must call os::init");
2570   return os::Linux::page_size();
2571 }
2572 
2573 // Solaris allocates memory by pages.
2574 int os::vm_allocation_granularity() {
2575   assert(os::Linux::page_size() != -1, "must call os::init");
2576   return os::Linux::page_size();
2577 }
2578 
2579 // Rationale behind this function:
2580 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2581 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2582 //  samples for JITted code. Here we create private executable mapping over the code cache
2583 //  and then we can use standard (well, almost, as mapping can change) way to provide
2584 //  info for the reporting script by storing timestamp and location of symbol
2585 void linux_wrap_code(char* base, size_t size) {
2586   static volatile jint cnt = 0;
2587 
2588   if (!UseOprofile) {
2589     return;
2590   }
2591 
2592   char buf[PATH_MAX+1];
2593   int num = Atomic::add(1, &cnt);
2594 
2595   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2596            os::get_temp_directory(), os::current_process_id(), num);
2597   unlink(buf);
2598 
2599   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2600 
2601   if (fd != -1) {
2602     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2603     if (rv != (off_t)-1) {
2604       if (::write(fd, "", 1) == 1) {
2605         mmap(base, size,
2606              PROT_READ|PROT_WRITE|PROT_EXEC,
2607              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2608       }
2609     }
2610     ::close(fd);
2611     unlink(buf);
2612   }
2613 }
2614 
2615 static bool recoverable_mmap_error(int err) {
2616   // See if the error is one we can let the caller handle. This
2617   // list of errno values comes from JBS-6843484. I can't find a
2618   // Linux man page that documents this specific set of errno
2619   // values so while this list currently matches Solaris, it may
2620   // change as we gain experience with this failure mode.
2621   switch (err) {
2622   case EBADF:
2623   case EINVAL:
2624   case ENOTSUP:
2625     // let the caller deal with these errors
2626     return true;
2627 
2628   default:
2629     // Any remaining errors on this OS can cause our reserved mapping
2630     // to be lost. That can cause confusion where different data
2631     // structures think they have the same memory mapped. The worst
2632     // scenario is if both the VM and a library think they have the
2633     // same memory mapped.
2634     return false;
2635   }
2636 }
2637 
2638 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2639                                     int err) {
2640   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2641           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2642           strerror(err), err);
2643 }
2644 
2645 static void warn_fail_commit_memory(char* addr, size_t size,
2646                                     size_t alignment_hint, bool exec,
2647                                     int err) {
2648   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2649           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2650           alignment_hint, exec, strerror(err), err);
2651 }
2652 
2653 // NOTE: Linux kernel does not really reserve the pages for us.
2654 //       All it does is to check if there are enough free pages
2655 //       left at the time of mmap(). This could be a potential
2656 //       problem.
2657 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2658   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2659   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2660                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2661   if (res != (uintptr_t) MAP_FAILED) {
2662     if (UseNUMAInterleaving) {
2663       numa_make_global(addr, size);
2664     }
2665     return 0;
2666   }
2667 
2668   int err = errno;  // save errno from mmap() call above
2669 
2670   if (!recoverable_mmap_error(err)) {
2671     warn_fail_commit_memory(addr, size, exec, err);
2672     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2673   }
2674 
2675   return err;
2676 }
2677 
2678 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2679   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2680 }
2681 
2682 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2683                                   const char* mesg) {
2684   assert(mesg != NULL, "mesg must be specified");
2685   int err = os::Linux::commit_memory_impl(addr, size, exec);
2686   if (err != 0) {
2687     // the caller wants all commit errors to exit with the specified mesg:
2688     warn_fail_commit_memory(addr, size, exec, err);
2689     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2690   }
2691 }
2692 
2693 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2694 #ifndef MAP_HUGETLB
2695 #define MAP_HUGETLB 0x40000
2696 #endif
2697 
2698 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2699 #ifndef MADV_HUGEPAGE
2700 #define MADV_HUGEPAGE 14
2701 #endif
2702 
2703 int os::Linux::commit_memory_impl(char* addr, size_t size,
2704                                   size_t alignment_hint, bool exec) {
2705   int err = os::Linux::commit_memory_impl(addr, size, exec);
2706   if (err == 0) {
2707     realign_memory(addr, size, alignment_hint);
2708   }
2709   return err;
2710 }
2711 
2712 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2713                           bool exec) {
2714   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2715 }
2716 
2717 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2718                                   size_t alignment_hint, bool exec,
2719                                   const char* mesg) {
2720   assert(mesg != NULL, "mesg must be specified");
2721   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2722   if (err != 0) {
2723     // the caller wants all commit errors to exit with the specified mesg:
2724     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2725     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2726   }
2727 }
2728 
2729 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2730   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2731     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2732     // be supported or the memory may already be backed by huge pages.
2733     ::madvise(addr, bytes, MADV_HUGEPAGE);
2734   }
2735 }
2736 
2737 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2738   // This method works by doing an mmap over an existing mmaping and effectively discarding
2739   // the existing pages. However it won't work for SHM-based large pages that cannot be
2740   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2741   // small pages on top of the SHM segment. This method always works for small pages, so we
2742   // allow that in any case.
2743   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2744     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2745   }
2746 }
2747 
2748 void os::numa_make_global(char *addr, size_t bytes) {
2749   Linux::numa_interleave_memory(addr, bytes);
2750 }
2751 
2752 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2753 // bind policy to MPOL_PREFERRED for the current thread.
2754 #define USE_MPOL_PREFERRED 0
2755 
2756 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2757   // To make NUMA and large pages more robust when both enabled, we need to ease
2758   // the requirements on where the memory should be allocated. MPOL_BIND is the
2759   // default policy and it will force memory to be allocated on the specified
2760   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2761   // the specified node, but will not force it. Using this policy will prevent
2762   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2763   // free large pages.
2764   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2765   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2766 }
2767 
2768 bool os::numa_topology_changed()   { return false; }
2769 
2770 size_t os::numa_get_groups_num() {
2771   int max_node = Linux::numa_max_node();
2772   return max_node > 0 ? max_node + 1 : 1;
2773 }
2774 
2775 int os::numa_get_group_id() {
2776   int cpu_id = Linux::sched_getcpu();
2777   if (cpu_id != -1) {
2778     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2779     if (lgrp_id != -1) {
2780       return lgrp_id;
2781     }
2782   }
2783   return 0;
2784 }
2785 
2786 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2787   for (size_t i = 0; i < size; i++) {
2788     ids[i] = i;
2789   }
2790   return size;
2791 }
2792 
2793 bool os::get_page_info(char *start, page_info* info) {
2794   return false;
2795 }
2796 
2797 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2798   return end;
2799 }
2800 
2801 
2802 int os::Linux::sched_getcpu_syscall(void) {
2803   unsigned int cpu;
2804   int retval = -1;
2805 
2806 #if defined(IA32)
2807 # ifndef SYS_getcpu
2808 # define SYS_getcpu 318
2809 # endif
2810   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2811 #elif defined(AMD64)
2812 // Unfortunately we have to bring all these macros here from vsyscall.h
2813 // to be able to compile on old linuxes.
2814 # define __NR_vgetcpu 2
2815 # define VSYSCALL_START (-10UL << 20)
2816 # define VSYSCALL_SIZE 1024
2817 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2818   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2819   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2820   retval = vgetcpu(&cpu, NULL, NULL);
2821 #endif
2822 
2823   return (retval == -1) ? retval : cpu;
2824 }
2825 
2826 // Something to do with the numa-aware allocator needs these symbols
2827 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2828 extern "C" JNIEXPORT void numa_error(char *where) { }
2829 extern "C" JNIEXPORT int fork1() { return fork(); }
2830 
2831 
2832 // If we are running with libnuma version > 2, then we should
2833 // be trying to use symbols with versions 1.1
2834 // If we are running with earlier version, which did not have symbol versions,
2835 // we should use the base version.
2836 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2837   void *f = dlvsym(handle, name, "libnuma_1.1");
2838   if (f == NULL) {
2839     f = dlsym(handle, name);
2840   }
2841   return f;
2842 }
2843 
2844 bool os::Linux::libnuma_init() {
2845   // sched_getcpu() should be in libc.
2846   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2847                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2848 
2849   // If it's not, try a direct syscall.
2850   if (sched_getcpu() == -1)
2851     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2852 
2853   if (sched_getcpu() != -1) { // Does it work?
2854     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2855     if (handle != NULL) {
2856       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2857                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2858       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2859                                        libnuma_dlsym(handle, "numa_max_node")));
2860       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2861                                         libnuma_dlsym(handle, "numa_available")));
2862       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2863                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2864       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2865                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2866       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2867                                             libnuma_dlsym(handle, "numa_set_bind_policy")));
2868 
2869 
2870       if (numa_available() != -1) {
2871         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2872         // Create a cpu -> node mapping
2873         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2874         rebuild_cpu_to_node_map();
2875         return true;
2876       }
2877     }
2878   }
2879   return false;
2880 }
2881 
2882 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2883 // The table is later used in get_node_by_cpu().
2884 void os::Linux::rebuild_cpu_to_node_map() {
2885   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2886                               // in libnuma (possible values are starting from 16,
2887                               // and continuing up with every other power of 2, but less
2888                               // than the maximum number of CPUs supported by kernel), and
2889                               // is a subject to change (in libnuma version 2 the requirements
2890                               // are more reasonable) we'll just hardcode the number they use
2891                               // in the library.
2892   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2893 
2894   size_t cpu_num = os::active_processor_count();
2895   size_t cpu_map_size = NCPUS / BitsPerCLong;
2896   size_t cpu_map_valid_size =
2897     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2898 
2899   cpu_to_node()->clear();
2900   cpu_to_node()->at_grow(cpu_num - 1);
2901   size_t node_num = numa_get_groups_num();
2902 
2903   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2904   for (size_t i = 0; i < node_num; i++) {
2905     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2906       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2907         if (cpu_map[j] != 0) {
2908           for (size_t k = 0; k < BitsPerCLong; k++) {
2909             if (cpu_map[j] & (1UL << k)) {
2910               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2911             }
2912           }
2913         }
2914       }
2915     }
2916   }
2917   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2918 }
2919 
2920 int os::Linux::get_node_by_cpu(int cpu_id) {
2921   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2922     return cpu_to_node()->at(cpu_id);
2923   }
2924   return -1;
2925 }
2926 
2927 GrowableArray<int>* os::Linux::_cpu_to_node;
2928 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2929 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2930 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2931 os::Linux::numa_available_func_t os::Linux::_numa_available;
2932 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2933 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2934 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2935 unsigned long* os::Linux::_numa_all_nodes;
2936 
2937 bool os::pd_uncommit_memory(char* addr, size_t size) {
2938   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2939                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2940   return res  != (uintptr_t) MAP_FAILED;
2941 }
2942 
2943 static
2944 address get_stack_commited_bottom(address bottom, size_t size) {
2945   address nbot = bottom;
2946   address ntop = bottom + size;
2947 
2948   size_t page_sz = os::vm_page_size();
2949   unsigned pages = size / page_sz;
2950 
2951   unsigned char vec[1];
2952   unsigned imin = 1, imax = pages + 1, imid;
2953   int mincore_return_value = 0;
2954 
2955   assert(imin <= imax, "Unexpected page size");
2956 
2957   while (imin < imax) {
2958     imid = (imax + imin) / 2;
2959     nbot = ntop - (imid * page_sz);
2960 
2961     // Use a trick with mincore to check whether the page is mapped or not.
2962     // mincore sets vec to 1 if page resides in memory and to 0 if page
2963     // is swapped output but if page we are asking for is unmapped
2964     // it returns -1,ENOMEM
2965     mincore_return_value = mincore(nbot, page_sz, vec);
2966 
2967     if (mincore_return_value == -1) {
2968       // Page is not mapped go up
2969       // to find first mapped page
2970       if (errno != EAGAIN) {
2971         assert(errno == ENOMEM, "Unexpected mincore errno");
2972         imax = imid;
2973       }
2974     } else {
2975       // Page is mapped go down
2976       // to find first not mapped page
2977       imin = imid + 1;
2978     }
2979   }
2980 
2981   nbot = nbot + page_sz;
2982 
2983   // Adjust stack bottom one page up if last checked page is not mapped
2984   if (mincore_return_value == -1) {
2985     nbot = nbot + page_sz;
2986   }
2987 
2988   return nbot;
2989 }
2990 
2991 
2992 // Linux uses a growable mapping for the stack, and if the mapping for
2993 // the stack guard pages is not removed when we detach a thread the
2994 // stack cannot grow beyond the pages where the stack guard was
2995 // mapped.  If at some point later in the process the stack expands to
2996 // that point, the Linux kernel cannot expand the stack any further
2997 // because the guard pages are in the way, and a segfault occurs.
2998 //
2999 // However, it's essential not to split the stack region by unmapping
3000 // a region (leaving a hole) that's already part of the stack mapping,
3001 // so if the stack mapping has already grown beyond the guard pages at
3002 // the time we create them, we have to truncate the stack mapping.
3003 // So, we need to know the extent of the stack mapping when
3004 // create_stack_guard_pages() is called.
3005 
3006 // We only need this for stacks that are growable: at the time of
3007 // writing thread stacks don't use growable mappings (i.e. those
3008 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3009 // only applies to the main thread.
3010 
3011 // If the (growable) stack mapping already extends beyond the point
3012 // where we're going to put our guard pages, truncate the mapping at
3013 // that point by munmap()ping it.  This ensures that when we later
3014 // munmap() the guard pages we don't leave a hole in the stack
3015 // mapping. This only affects the main/initial thread
3016 
3017 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3018 
3019   if (os::Linux::is_initial_thread()) {
3020     // As we manually grow stack up to bottom inside create_attached_thread(),
3021     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3022     // we don't need to do anything special.
3023     // Check it first, before calling heavy function.
3024     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3025     unsigned char vec[1];
3026 
3027     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3028       // Fallback to slow path on all errors, including EAGAIN
3029       stack_extent = (uintptr_t) get_stack_commited_bottom(
3030                                     os::Linux::initial_thread_stack_bottom(),
3031                                     (size_t)addr - stack_extent);
3032     }
3033 
3034     if (stack_extent < (uintptr_t)addr) {
3035       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3036     }
3037   }
3038 
3039   return os::commit_memory(addr, size, !ExecMem);
3040 }
3041 
3042 // If this is a growable mapping, remove the guard pages entirely by
3043 // munmap()ping them.  If not, just call uncommit_memory(). This only
3044 // affects the main/initial thread, but guard against future OS changes
3045 // It's safe to always unmap guard pages for initial thread because we
3046 // always place it right after end of the mapped region
3047 
3048 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3049   uintptr_t stack_extent, stack_base;
3050 
3051   if (os::Linux::is_initial_thread()) {
3052     return ::munmap(addr, size) == 0;
3053   }
3054 
3055   return os::uncommit_memory(addr, size);
3056 }
3057 
3058 static address _highest_vm_reserved_address = NULL;
3059 
3060 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3061 // at 'requested_addr'. If there are existing memory mappings at the same
3062 // location, however, they will be overwritten. If 'fixed' is false,
3063 // 'requested_addr' is only treated as a hint, the return value may or
3064 // may not start from the requested address. Unlike Linux mmap(), this
3065 // function returns NULL to indicate failure.
3066 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3067   char * addr;
3068   int flags;
3069 
3070   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3071   if (fixed) {
3072     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3073     flags |= MAP_FIXED;
3074   }
3075 
3076   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3077   // touch an uncommitted page. Otherwise, the read/write might
3078   // succeed if we have enough swap space to back the physical page.
3079   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3080                        flags, -1, 0);
3081 
3082   if (addr != MAP_FAILED) {
3083     // anon_mmap() should only get called during VM initialization,
3084     // don't need lock (actually we can skip locking even it can be called
3085     // from multiple threads, because _highest_vm_reserved_address is just a
3086     // hint about the upper limit of non-stack memory regions.)
3087     if ((address)addr + bytes > _highest_vm_reserved_address) {
3088       _highest_vm_reserved_address = (address)addr + bytes;
3089     }
3090   }
3091 
3092   return addr == MAP_FAILED ? NULL : addr;
3093 }
3094 
3095 // Don't update _highest_vm_reserved_address, because there might be memory
3096 // regions above addr + size. If so, releasing a memory region only creates
3097 // a hole in the address space, it doesn't help prevent heap-stack collision.
3098 //
3099 static int anon_munmap(char * addr, size_t size) {
3100   return ::munmap(addr, size) == 0;
3101 }
3102 
3103 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3104                          size_t alignment_hint) {
3105   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3106 }
3107 
3108 bool os::pd_release_memory(char* addr, size_t size) {
3109   return anon_munmap(addr, size);
3110 }
3111 
3112 static address highest_vm_reserved_address() {
3113   return _highest_vm_reserved_address;
3114 }
3115 
3116 static bool linux_mprotect(char* addr, size_t size, int prot) {
3117   // Linux wants the mprotect address argument to be page aligned.
3118   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3119 
3120   // According to SUSv3, mprotect() should only be used with mappings
3121   // established by mmap(), and mmap() always maps whole pages. Unaligned
3122   // 'addr' likely indicates problem in the VM (e.g. trying to change
3123   // protection of malloc'ed or statically allocated memory). Check the
3124   // caller if you hit this assert.
3125   assert(addr == bottom, "sanity check");
3126 
3127   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3128   return ::mprotect(bottom, size, prot) == 0;
3129 }
3130 
3131 // Set protections specified
3132 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3133                         bool is_committed) {
3134   unsigned int p = 0;
3135   switch (prot) {
3136   case MEM_PROT_NONE: p = PROT_NONE; break;
3137   case MEM_PROT_READ: p = PROT_READ; break;
3138   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3139   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3140   default:
3141     ShouldNotReachHere();
3142   }
3143   // is_committed is unused.
3144   return linux_mprotect(addr, bytes, p);
3145 }
3146 
3147 bool os::guard_memory(char* addr, size_t size) {
3148   return linux_mprotect(addr, size, PROT_NONE);
3149 }
3150 
3151 bool os::unguard_memory(char* addr, size_t size) {
3152   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3153 }
3154 
3155 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3156   bool result = false;
3157   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3158                  MAP_ANONYMOUS|MAP_PRIVATE,
3159                  -1, 0);
3160   if (p != MAP_FAILED) {
3161     void *aligned_p = align_ptr_up(p, page_size);
3162 
3163     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3164 
3165     munmap(p, page_size * 2);
3166   }
3167 
3168   if (warn && !result) {
3169     warning("TransparentHugePages is not supported by the operating system.");
3170   }
3171 
3172   return result;
3173 }
3174 
3175 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3176   bool result = false;
3177   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3178                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3179                  -1, 0);
3180 
3181   if (p != MAP_FAILED) {
3182     // We don't know if this really is a huge page or not.
3183     FILE *fp = fopen("/proc/self/maps", "r");
3184     if (fp) {
3185       while (!feof(fp)) {
3186         char chars[257];
3187         long x = 0;
3188         if (fgets(chars, sizeof(chars), fp)) {
3189           if (sscanf(chars, "%lx-%*x", &x) == 1
3190               && x == (long)p) {
3191             if (strstr (chars, "hugepage")) {
3192               result = true;
3193               break;
3194             }
3195           }
3196         }
3197       }
3198       fclose(fp);
3199     }
3200     munmap(p, page_size);
3201   }
3202 
3203   if (warn && !result) {
3204     warning("HugeTLBFS is not supported by the operating system.");
3205   }
3206 
3207   return result;
3208 }
3209 
3210 /*
3211 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3212 *
3213 * From the coredump_filter documentation:
3214 *
3215 * - (bit 0) anonymous private memory
3216 * - (bit 1) anonymous shared memory
3217 * - (bit 2) file-backed private memory
3218 * - (bit 3) file-backed shared memory
3219 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3220 *           effective only if the bit 2 is cleared)
3221 * - (bit 5) hugetlb private memory
3222 * - (bit 6) hugetlb shared memory
3223 */
3224 static void set_coredump_filter(void) {
3225   FILE *f;
3226   long cdm;
3227 
3228   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3229     return;
3230   }
3231 
3232   if (fscanf(f, "%lx", &cdm) != 1) {
3233     fclose(f);
3234     return;
3235   }
3236 
3237   rewind(f);
3238 
3239   if ((cdm & LARGEPAGES_BIT) == 0) {
3240     cdm |= LARGEPAGES_BIT;
3241     fprintf(f, "%#lx", cdm);
3242   }
3243 
3244   fclose(f);
3245 }
3246 
3247 // Large page support
3248 
3249 static size_t _large_page_size = 0;
3250 
3251 size_t os::Linux::find_large_page_size() {
3252   size_t large_page_size = 0;
3253 
3254   // large_page_size on Linux is used to round up heap size. x86 uses either
3255   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3256   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3257   // page as large as 256M.
3258   //
3259   // Here we try to figure out page size by parsing /proc/meminfo and looking
3260   // for a line with the following format:
3261   //    Hugepagesize:     2048 kB
3262   //
3263   // If we can't determine the value (e.g. /proc is not mounted, or the text
3264   // format has been changed), we'll use the largest page size supported by
3265   // the processor.
3266 
3267 #ifndef ZERO
3268   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3269                      ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3270 #endif // ZERO
3271 
3272   FILE *fp = fopen("/proc/meminfo", "r");
3273   if (fp) {
3274     while (!feof(fp)) {
3275       int x = 0;
3276       char buf[16];
3277       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3278         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3279           large_page_size = x * K;
3280           break;
3281         }
3282       } else {
3283         // skip to next line
3284         for (;;) {
3285           int ch = fgetc(fp);
3286           if (ch == EOF || ch == (int)'\n') break;
3287         }
3288       }
3289     }
3290     fclose(fp);
3291   }
3292 
3293   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3294     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3295         SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3296         proper_unit_for_byte_size(large_page_size));
3297   }
3298 
3299   return large_page_size;
3300 }
3301 
3302 size_t os::Linux::setup_large_page_size() {
3303   _large_page_size = Linux::find_large_page_size();
3304   const size_t default_page_size = (size_t)Linux::page_size();
3305   if (_large_page_size > default_page_size) {
3306     _page_sizes[0] = _large_page_size;
3307     _page_sizes[1] = default_page_size;
3308     _page_sizes[2] = 0;
3309   }
3310 
3311   return _large_page_size;
3312 }
3313 
3314 bool os::Linux::setup_large_page_type(size_t page_size) {
3315   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3316       FLAG_IS_DEFAULT(UseSHM) &&
3317       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3318 
3319     // The type of large pages has not been specified by the user.
3320 
3321     // Try UseHugeTLBFS and then UseSHM.
3322     UseHugeTLBFS = UseSHM = true;
3323 
3324     // Don't try UseTransparentHugePages since there are known
3325     // performance issues with it turned on. This might change in the future.
3326     UseTransparentHugePages = false;
3327   }
3328 
3329   if (UseTransparentHugePages) {
3330     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3331     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3332       UseHugeTLBFS = false;
3333       UseSHM = false;
3334       return true;
3335     }
3336     UseTransparentHugePages = false;
3337   }
3338 
3339   if (UseHugeTLBFS) {
3340     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3341     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3342       UseSHM = false;
3343       return true;
3344     }
3345     UseHugeTLBFS = false;
3346   }
3347 
3348   return UseSHM;
3349 }
3350 
3351 void os::large_page_init() {
3352   if (!UseLargePages &&
3353       !UseTransparentHugePages &&
3354       !UseHugeTLBFS &&
3355       !UseSHM) {
3356     // Not using large pages.
3357     return;
3358   }
3359 
3360   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3361     // The user explicitly turned off large pages.
3362     // Ignore the rest of the large pages flags.
3363     UseTransparentHugePages = false;
3364     UseHugeTLBFS = false;
3365     UseSHM = false;
3366     return;
3367   }
3368 
3369   size_t large_page_size = Linux::setup_large_page_size();
3370   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3371 
3372   set_coredump_filter();
3373 }
3374 
3375 #ifndef SHM_HUGETLB
3376 #define SHM_HUGETLB 04000
3377 #endif
3378 
3379 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3380   // "exec" is passed in but not used.  Creating the shared image for
3381   // the code cache doesn't have an SHM_X executable permission to check.
3382   assert(UseLargePages && UseSHM, "only for SHM large pages");
3383   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3384 
3385   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3386     return NULL; // Fallback to small pages.
3387   }
3388 
3389   key_t key = IPC_PRIVATE;
3390   char *addr;
3391 
3392   bool warn_on_failure = UseLargePages &&
3393                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3394                          !FLAG_IS_DEFAULT(UseSHM) ||
3395                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3396                         );
3397   char msg[128];
3398 
3399   // Create a large shared memory region to attach to based on size.
3400   // Currently, size is the total size of the heap
3401   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3402   if (shmid == -1) {
3403      // Possible reasons for shmget failure:
3404      // 1. shmmax is too small for Java heap.
3405      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3406      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3407      // 2. not enough large page memory.
3408      //    > check available large pages: cat /proc/meminfo
3409      //    > increase amount of large pages:
3410      //          echo new_value > /proc/sys/vm/nr_hugepages
3411      //      Note 1: different Linux may use different name for this property,
3412      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3413      //      Note 2: it's possible there's enough physical memory available but
3414      //            they are so fragmented after a long run that they can't
3415      //            coalesce into large pages. Try to reserve large pages when
3416      //            the system is still "fresh".
3417      if (warn_on_failure) {
3418        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3419        warning(msg);
3420      }
3421      return NULL;
3422   }
3423 
3424   // attach to the region
3425   addr = (char*)shmat(shmid, req_addr, 0);
3426   int err = errno;
3427 
3428   // Remove shmid. If shmat() is successful, the actual shared memory segment
3429   // will be deleted when it's detached by shmdt() or when the process
3430   // terminates. If shmat() is not successful this will remove the shared
3431   // segment immediately.
3432   shmctl(shmid, IPC_RMID, NULL);
3433 
3434   if ((intptr_t)addr == -1) {
3435      if (warn_on_failure) {
3436        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3437        warning(msg);
3438      }
3439      return NULL;
3440   }
3441 
3442   return addr;
3443 }
3444 
3445 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3446   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3447 
3448   bool warn_on_failure = UseLargePages &&
3449       (!FLAG_IS_DEFAULT(UseLargePages) ||
3450        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3451        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3452 
3453   if (warn_on_failure) {
3454     char msg[128];
3455     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3456         PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3457     warning(msg);
3458   }
3459 }
3460 
3461 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3462   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3463   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3464   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3465 
3466   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3467   char* addr = (char*)::mmap(req_addr, bytes, prot,
3468                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3469                              -1, 0);
3470 
3471   if (addr == MAP_FAILED) {
3472     warn_on_large_pages_failure(req_addr, bytes, errno);
3473     return NULL;
3474   }
3475 
3476   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3477 
3478   return addr;
3479 }
3480 
3481 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3482   size_t large_page_size = os::large_page_size();
3483 
3484   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3485 
3486   // Allocate small pages.
3487 
3488   char* start;
3489   if (req_addr != NULL) {
3490     assert(is_ptr_aligned(req_addr, alignment), "Must be");
3491     assert(is_size_aligned(bytes, alignment), "Must be");
3492     start = os::reserve_memory(bytes, req_addr);
3493     assert(start == NULL || start == req_addr, "Must be");
3494   } else {
3495     start = os::reserve_memory_aligned(bytes, alignment);
3496   }
3497 
3498   if (start == NULL) {
3499     return NULL;
3500   }
3501 
3502   assert(is_ptr_aligned(start, alignment), "Must be");
3503 
3504   // os::reserve_memory_special will record this memory area.
3505   // Need to release it here to prevent overlapping reservations.
3506   MemTracker::record_virtual_memory_release((address)start, bytes);
3507 
3508   char* end = start + bytes;
3509 
3510   // Find the regions of the allocated chunk that can be promoted to large pages.
3511   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3512   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3513 
3514   size_t lp_bytes = lp_end - lp_start;
3515 
3516   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3517 
3518   if (lp_bytes == 0) {
3519     // The mapped region doesn't even span the start and the end of a large page.
3520     // Fall back to allocate a non-special area.
3521     ::munmap(start, end - start);
3522     return NULL;
3523   }
3524 
3525   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3526 
3527 
3528   void* result;
3529 
3530   if (start != lp_start) {
3531     result = ::mmap(start, lp_start - start, prot,
3532                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3533                     -1, 0);
3534     if (result == MAP_FAILED) {
3535       ::munmap(lp_start, end - lp_start);
3536       return NULL;
3537     }
3538   }
3539 
3540   result = ::mmap(lp_start, lp_bytes, prot,
3541                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3542                   -1, 0);
3543   if (result == MAP_FAILED) {
3544     warn_on_large_pages_failure(req_addr, bytes, errno);
3545     // If the mmap above fails, the large pages region will be unmapped and we
3546     // have regions before and after with small pages. Release these regions.
3547     //
3548     // |  mapped  |  unmapped  |  mapped  |
3549     // ^          ^            ^          ^
3550     // start      lp_start     lp_end     end
3551     //
3552     ::munmap(start, lp_start - start);
3553     ::munmap(lp_end, end - lp_end);
3554     return NULL;
3555   }
3556 
3557   if (lp_end != end) {
3558       result = ::mmap(lp_end, end - lp_end, prot,
3559                       MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3560                       -1, 0);
3561     if (result == MAP_FAILED) {
3562       ::munmap(start, lp_end - start);
3563       return NULL;
3564     }
3565   }
3566 
3567   return start;
3568 }
3569 
3570 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3571   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3572   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3573   assert(is_power_of_2(alignment), "Must be");
3574   assert(is_power_of_2(os::large_page_size()), "Must be");
3575   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3576 
3577   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3578     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3579   } else {
3580     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3581   }
3582 }
3583 
3584 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3585   assert(UseLargePages, "only for large pages");
3586 
3587   char* addr;
3588   if (UseSHM) {
3589     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3590   } else {
3591     assert(UseHugeTLBFS, "must be");
3592     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3593   }
3594 
3595   if (addr != NULL) {
3596     if (UseNUMAInterleaving) {
3597       numa_make_global(addr, bytes);
3598     }
3599 
3600     // The memory is committed
3601     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3602   }
3603 
3604   return addr;
3605 }
3606 
3607 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3608   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3609   return shmdt(base) == 0;
3610 }
3611 
3612 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3613   return pd_release_memory(base, bytes);
3614 }
3615 
3616 bool os::release_memory_special(char* base, size_t bytes) {
3617   assert(UseLargePages, "only for large pages");
3618 
3619   MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3620 
3621   bool res;
3622   if (UseSHM) {
3623     res = os::Linux::release_memory_special_shm(base, bytes);
3624   } else {
3625     assert(UseHugeTLBFS, "must be");
3626     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3627   }
3628 
3629   if (res) {
3630     tkr.record((address)base, bytes);
3631   } else {
3632     tkr.discard();
3633   }
3634 
3635   return res;
3636 }
3637 
3638 size_t os::large_page_size() {
3639   return _large_page_size;
3640 }
3641 
3642 // With SysV SHM the entire memory region must be allocated as shared
3643 // memory.
3644 // HugeTLBFS allows application to commit large page memory on demand.
3645 // However, when committing memory with HugeTLBFS fails, the region
3646 // that was supposed to be committed will lose the old reservation
3647 // and allow other threads to steal that memory region. Because of this
3648 // behavior we can't commit HugeTLBFS memory.
3649 bool os::can_commit_large_page_memory() {
3650   return UseTransparentHugePages;
3651 }
3652 
3653 bool os::can_execute_large_page_memory() {
3654   return UseTransparentHugePages || UseHugeTLBFS;
3655 }
3656 
3657 // Reserve memory at an arbitrary address, only if that area is
3658 // available (and not reserved for something else).
3659 
3660 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3661   const int max_tries = 10;
3662   char* base[max_tries];
3663   size_t size[max_tries];
3664   const size_t gap = 0x000000;
3665 
3666   // Assert only that the size is a multiple of the page size, since
3667   // that's all that mmap requires, and since that's all we really know
3668   // about at this low abstraction level.  If we need higher alignment,
3669   // we can either pass an alignment to this method or verify alignment
3670   // in one of the methods further up the call chain.  See bug 5044738.
3671   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3672 
3673   // Repeatedly allocate blocks until the block is allocated at the
3674   // right spot. Give up after max_tries. Note that reserve_memory() will
3675   // automatically update _highest_vm_reserved_address if the call is
3676   // successful. The variable tracks the highest memory address every reserved
3677   // by JVM. It is used to detect heap-stack collision if running with
3678   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3679   // space than needed, it could confuse the collision detecting code. To
3680   // solve the problem, save current _highest_vm_reserved_address and
3681   // calculate the correct value before return.
3682   address old_highest = _highest_vm_reserved_address;
3683 
3684   // Linux mmap allows caller to pass an address as hint; give it a try first,
3685   // if kernel honors the hint then we can return immediately.
3686   char * addr = anon_mmap(requested_addr, bytes, false);
3687   if (addr == requested_addr) {
3688      return requested_addr;
3689   }
3690 
3691   if (addr != NULL) {
3692      // mmap() is successful but it fails to reserve at the requested address
3693      anon_munmap(addr, bytes);
3694   }
3695 
3696   int i;
3697   for (i = 0; i < max_tries; ++i) {
3698     base[i] = reserve_memory(bytes);
3699 
3700     if (base[i] != NULL) {
3701       // Is this the block we wanted?
3702       if (base[i] == requested_addr) {
3703         size[i] = bytes;
3704         break;
3705       }
3706 
3707       // Does this overlap the block we wanted? Give back the overlapped
3708       // parts and try again.
3709 
3710       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3711       if (top_overlap >= 0 && top_overlap < bytes) {
3712         unmap_memory(base[i], top_overlap);
3713         base[i] += top_overlap;
3714         size[i] = bytes - top_overlap;
3715       } else {
3716         size_t bottom_overlap = base[i] + bytes - requested_addr;
3717         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3718           unmap_memory(requested_addr, bottom_overlap);
3719           size[i] = bytes - bottom_overlap;
3720         } else {
3721           size[i] = bytes;
3722         }
3723       }
3724     }
3725   }
3726 
3727   // Give back the unused reserved pieces.
3728 
3729   for (int j = 0; j < i; ++j) {
3730     if (base[j] != NULL) {
3731       unmap_memory(base[j], size[j]);
3732     }
3733   }
3734 
3735   if (i < max_tries) {
3736     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3737     return requested_addr;
3738   } else {
3739     _highest_vm_reserved_address = old_highest;
3740     return NULL;
3741   }
3742 }
3743 
3744 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3745   return ::read(fd, buf, nBytes);
3746 }
3747 
3748 //
3749 // Short sleep, direct OS call.
3750 //
3751 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3752 // sched_yield(2) will actually give up the CPU:
3753 //
3754 //   * Alone on this pariticular CPU, keeps running.
3755 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3756 //     (pre 2.6.39).
3757 //
3758 // So calling this with 0 is an alternative.
3759 //
3760 void os::naked_short_sleep(jlong ms) {
3761   struct timespec req;
3762 
3763   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3764   req.tv_sec = 0;
3765   if (ms > 0) {
3766     req.tv_nsec = (ms % 1000) * 1000000;
3767   }
3768   else {
3769     req.tv_nsec = 1;
3770   }
3771 
3772   nanosleep(&req, NULL);
3773 
3774   return;
3775 }
3776 
3777 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3778 void os::infinite_sleep() {
3779   while (true) {    // sleep forever ...
3780     ::sleep(100);   // ... 100 seconds at a time
3781   }
3782 }
3783 
3784 // Used to convert frequent JVM_Yield() to nops
3785 bool os::dont_yield() {
3786   return DontYieldALot;
3787 }
3788 
3789 void os::yield() {
3790   sched_yield();
3791 }
3792 
3793 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3794 
3795 void os::yield_all() {
3796   // Yields to all threads, including threads with lower priorities
3797   // Threads on Linux are all with same priority. The Solaris style
3798   // os::yield_all() with nanosleep(1ms) is not necessary.
3799   sched_yield();
3800 }
3801 
3802 ////////////////////////////////////////////////////////////////////////////////
3803 // thread priority support
3804 
3805 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3806 // only supports dynamic priority, static priority must be zero. For real-time
3807 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3808 // However, for large multi-threaded applications, SCHED_RR is not only slower
3809 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3810 // of 5 runs - Sep 2005).
3811 //
3812 // The following code actually changes the niceness of kernel-thread/LWP. It
3813 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3814 // not the entire user process, and user level threads are 1:1 mapped to kernel
3815 // threads. It has always been the case, but could change in the future. For
3816 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3817 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3818 
3819 int os::java_to_os_priority[CriticalPriority + 1] = {
3820   19,              // 0 Entry should never be used
3821 
3822    4,              // 1 MinPriority
3823    3,              // 2
3824    2,              // 3
3825 
3826    1,              // 4
3827    0,              // 5 NormPriority
3828   -1,              // 6
3829 
3830   -2,              // 7
3831   -3,              // 8
3832   -4,              // 9 NearMaxPriority
3833 
3834   -5,              // 10 MaxPriority
3835 
3836   -5               // 11 CriticalPriority
3837 };
3838 
3839 static int prio_init() {
3840   if (ThreadPriorityPolicy == 1) {
3841     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3842     // if effective uid is not root. Perhaps, a more elegant way of doing
3843     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3844     if (geteuid() != 0) {
3845       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3846         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3847       }
3848       ThreadPriorityPolicy = 0;
3849     }
3850   }
3851   if (UseCriticalJavaThreadPriority) {
3852     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3853   }
3854   return 0;
3855 }
3856 
3857 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3858   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3859 
3860   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3861   return (ret == 0) ? OS_OK : OS_ERR;
3862 }
3863 
3864 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3865   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3866     *priority_ptr = java_to_os_priority[NormPriority];
3867     return OS_OK;
3868   }
3869 
3870   errno = 0;
3871   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3872   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3873 }
3874 
3875 // Hint to the underlying OS that a task switch would not be good.
3876 // Void return because it's a hint and can fail.
3877 void os::hint_no_preempt() {}
3878 
3879 ////////////////////////////////////////////////////////////////////////////////
3880 // suspend/resume support
3881 
3882 //  the low-level signal-based suspend/resume support is a remnant from the
3883 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3884 //  within hotspot. Now there is a single use-case for this:
3885 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3886 //      that runs in the watcher thread.
3887 //  The remaining code is greatly simplified from the more general suspension
3888 //  code that used to be used.
3889 //
3890 //  The protocol is quite simple:
3891 //  - suspend:
3892 //      - sends a signal to the target thread
3893 //      - polls the suspend state of the osthread using a yield loop
3894 //      - target thread signal handler (SR_handler) sets suspend state
3895 //        and blocks in sigsuspend until continued
3896 //  - resume:
3897 //      - sets target osthread state to continue
3898 //      - sends signal to end the sigsuspend loop in the SR_handler
3899 //
3900 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3901 //
3902 
3903 static void resume_clear_context(OSThread *osthread) {
3904   osthread->set_ucontext(NULL);
3905   osthread->set_siginfo(NULL);
3906 }
3907 
3908 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3909   osthread->set_ucontext(context);
3910   osthread->set_siginfo(siginfo);
3911 }
3912 
3913 //
3914 // Handler function invoked when a thread's execution is suspended or
3915 // resumed. We have to be careful that only async-safe functions are
3916 // called here (Note: most pthread functions are not async safe and
3917 // should be avoided.)
3918 //
3919 // Note: sigwait() is a more natural fit than sigsuspend() from an
3920 // interface point of view, but sigwait() prevents the signal hander
3921 // from being run. libpthread would get very confused by not having
3922 // its signal handlers run and prevents sigwait()'s use with the
3923 // mutex granting granting signal.
3924 //
3925 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3926 //
3927 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3928   // Save and restore errno to avoid confusing native code with EINTR
3929   // after sigsuspend.
3930   int old_errno = errno;
3931 
3932   Thread* thread = Thread::current();
3933   OSThread* osthread = thread->osthread();
3934   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3935 
3936   os::SuspendResume::State current = osthread->sr.state();
3937   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3938     suspend_save_context(osthread, siginfo, context);
3939 
3940     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3941     os::SuspendResume::State state = osthread->sr.suspended();
3942     if (state == os::SuspendResume::SR_SUSPENDED) {
3943       sigset_t suspend_set;  // signals for sigsuspend()
3944 
3945       // get current set of blocked signals and unblock resume signal
3946       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3947       sigdelset(&suspend_set, SR_signum);
3948 
3949       sr_semaphore.signal();
3950       // wait here until we are resumed
3951       while (1) {
3952         sigsuspend(&suspend_set);
3953 
3954         os::SuspendResume::State result = osthread->sr.running();
3955         if (result == os::SuspendResume::SR_RUNNING) {
3956           sr_semaphore.signal();
3957           break;
3958         }
3959       }
3960 
3961     } else if (state == os::SuspendResume::SR_RUNNING) {
3962       // request was cancelled, continue
3963     } else {
3964       ShouldNotReachHere();
3965     }
3966 
3967     resume_clear_context(osthread);
3968   } else if (current == os::SuspendResume::SR_RUNNING) {
3969     // request was cancelled, continue
3970   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3971     // ignore
3972   } else {
3973     // ignore
3974   }
3975 
3976   errno = old_errno;
3977 }
3978 
3979 
3980 static int SR_initialize() {
3981   struct sigaction act;
3982   char *s;
3983   /* Get signal number to use for suspend/resume */
3984   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3985     int sig = ::strtol(s, 0, 10);
3986     if (sig > 0 || sig < _NSIG) {
3987         SR_signum = sig;
3988     }
3989   }
3990 
3991   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3992         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3993 
3994   sigemptyset(&SR_sigset);
3995   sigaddset(&SR_sigset, SR_signum);
3996 
3997   /* Set up signal handler for suspend/resume */
3998   act.sa_flags = SA_RESTART|SA_SIGINFO;
3999   act.sa_handler = (void (*)(int)) SR_handler;
4000 
4001   // SR_signum is blocked by default.
4002   // 4528190 - We also need to block pthread restart signal (32 on all
4003   // supported Linux platforms). Note that LinuxThreads need to block
4004   // this signal for all threads to work properly. So we don't have
4005   // to use hard-coded signal number when setting up the mask.
4006   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4007 
4008   if (sigaction(SR_signum, &act, 0) == -1) {
4009     return -1;
4010   }
4011 
4012   // Save signal flag
4013   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4014   return 0;
4015 }
4016 
4017 static int sr_notify(OSThread* osthread) {
4018   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4019   assert_status(status == 0, status, "pthread_kill");
4020   return status;
4021 }
4022 
4023 // "Randomly" selected value for how long we want to spin
4024 // before bailing out on suspending a thread, also how often
4025 // we send a signal to a thread we want to resume
4026 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4027 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4028 
4029 // returns true on success and false on error - really an error is fatal
4030 // but this seems the normal response to library errors
4031 static bool do_suspend(OSThread* osthread) {
4032   assert(osthread->sr.is_running(), "thread should be running");
4033   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4034 
4035   // mark as suspended and send signal
4036   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4037     // failed to switch, state wasn't running?
4038     ShouldNotReachHere();
4039     return false;
4040   }
4041 
4042   if (sr_notify(osthread) != 0) {
4043     ShouldNotReachHere();
4044   }
4045 
4046   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4047   while (true) {
4048     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4049       break;
4050     } else {
4051       // timeout
4052       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4053       if (cancelled == os::SuspendResume::SR_RUNNING) {
4054         return false;
4055       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4056         // make sure that we consume the signal on the semaphore as well
4057         sr_semaphore.wait();
4058         break;
4059       } else {
4060         ShouldNotReachHere();
4061         return false;
4062       }
4063     }
4064   }
4065 
4066   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4067   return true;
4068 }
4069 
4070 static void do_resume(OSThread* osthread) {
4071   assert(osthread->sr.is_suspended(), "thread should be suspended");
4072   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4073 
4074   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4075     // failed to switch to WAKEUP_REQUEST
4076     ShouldNotReachHere();
4077     return;
4078   }
4079 
4080   while (true) {
4081     if (sr_notify(osthread) == 0) {
4082       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4083         if (osthread->sr.is_running()) {
4084           return;
4085         }
4086       }
4087     } else {
4088       ShouldNotReachHere();
4089     }
4090   }
4091 
4092   guarantee(osthread->sr.is_running(), "Must be running!");
4093 }
4094 
4095 ///////////////////////////////////////////////////////////////////////////////////
4096 // signal handling (except suspend/resume)
4097 
4098 // This routine may be used by user applications as a "hook" to catch signals.
4099 // The user-defined signal handler must pass unrecognized signals to this
4100 // routine, and if it returns true (non-zero), then the signal handler must
4101 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4102 // routine will never retun false (zero), but instead will execute a VM panic
4103 // routine kill the process.
4104 //
4105 // If this routine returns false, it is OK to call it again.  This allows
4106 // the user-defined signal handler to perform checks either before or after
4107 // the VM performs its own checks.  Naturally, the user code would be making
4108 // a serious error if it tried to handle an exception (such as a null check
4109 // or breakpoint) that the VM was generating for its own correct operation.
4110 //
4111 // This routine may recognize any of the following kinds of signals:
4112 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4113 // It should be consulted by handlers for any of those signals.
4114 //
4115 // The caller of this routine must pass in the three arguments supplied
4116 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4117 // field of the structure passed to sigaction().  This routine assumes that
4118 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4119 //
4120 // Note that the VM will print warnings if it detects conflicting signal
4121 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4122 //
4123 extern "C" JNIEXPORT int
4124 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4125                         void* ucontext, int abort_if_unrecognized);
4126 
4127 void signalHandler(int sig, siginfo_t* info, void* uc) {
4128   assert(info != NULL && uc != NULL, "it must be old kernel");
4129   int orig_errno = errno;  // Preserve errno value over signal handler.
4130   JVM_handle_linux_signal(sig, info, uc, true);
4131   errno = orig_errno;
4132 }
4133 
4134 
4135 // This boolean allows users to forward their own non-matching signals
4136 // to JVM_handle_linux_signal, harmlessly.
4137 bool os::Linux::signal_handlers_are_installed = false;
4138 
4139 // For signal-chaining
4140 struct sigaction os::Linux::sigact[MAXSIGNUM];
4141 unsigned int os::Linux::sigs = 0;
4142 bool os::Linux::libjsig_is_loaded = false;
4143 typedef struct sigaction *(*get_signal_t)(int);
4144 get_signal_t os::Linux::get_signal_action = NULL;
4145 
4146 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4147   struct sigaction *actp = NULL;
4148 
4149   if (libjsig_is_loaded) {
4150     // Retrieve the old signal handler from libjsig
4151     actp = (*get_signal_action)(sig);
4152   }
4153   if (actp == NULL) {
4154     // Retrieve the preinstalled signal handler from jvm
4155     actp = get_preinstalled_handler(sig);
4156   }
4157 
4158   return actp;
4159 }
4160 
4161 static bool call_chained_handler(struct sigaction *actp, int sig,
4162                                  siginfo_t *siginfo, void *context) {
4163   // Call the old signal handler
4164   if (actp->sa_handler == SIG_DFL) {
4165     // It's more reasonable to let jvm treat it as an unexpected exception
4166     // instead of taking the default action.
4167     return false;
4168   } else if (actp->sa_handler != SIG_IGN) {
4169     if ((actp->sa_flags & SA_NODEFER) == 0) {
4170       // automaticlly block the signal
4171       sigaddset(&(actp->sa_mask), sig);
4172     }
4173 
4174     sa_handler_t hand;
4175     sa_sigaction_t sa;
4176     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4177     // retrieve the chained handler
4178     if (siginfo_flag_set) {
4179       sa = actp->sa_sigaction;
4180     } else {
4181       hand = actp->sa_handler;
4182     }
4183 
4184     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4185       actp->sa_handler = SIG_DFL;
4186     }
4187 
4188     // try to honor the signal mask
4189     sigset_t oset;
4190     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4191 
4192     // call into the chained handler
4193     if (siginfo_flag_set) {
4194       (*sa)(sig, siginfo, context);
4195     } else {
4196       (*hand)(sig);
4197     }
4198 
4199     // restore the signal mask
4200     pthread_sigmask(SIG_SETMASK, &oset, 0);
4201   }
4202   // Tell jvm's signal handler the signal is taken care of.
4203   return true;
4204 }
4205 
4206 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4207   bool chained = false;
4208   // signal-chaining
4209   if (UseSignalChaining) {
4210     struct sigaction *actp = get_chained_signal_action(sig);
4211     if (actp != NULL) {
4212       chained = call_chained_handler(actp, sig, siginfo, context);
4213     }
4214   }
4215   return chained;
4216 }
4217 
4218 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4219   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4220     return &sigact[sig];
4221   }
4222   return NULL;
4223 }
4224 
4225 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4226   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4227   sigact[sig] = oldAct;
4228   sigs |= (unsigned int)1 << sig;
4229 }
4230 
4231 // for diagnostic
4232 int os::Linux::sigflags[MAXSIGNUM];
4233 
4234 int os::Linux::get_our_sigflags(int sig) {
4235   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4236   return sigflags[sig];
4237 }
4238 
4239 void os::Linux::set_our_sigflags(int sig, int flags) {
4240   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4241   sigflags[sig] = flags;
4242 }
4243 
4244 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4245   // Check for overwrite.
4246   struct sigaction oldAct;
4247   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4248 
4249   void* oldhand = oldAct.sa_sigaction
4250                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4251                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4252   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4253       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4254       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4255     if (AllowUserSignalHandlers || !set_installed) {
4256       // Do not overwrite; user takes responsibility to forward to us.
4257       return;
4258     } else if (UseSignalChaining) {
4259       // save the old handler in jvm
4260       save_preinstalled_handler(sig, oldAct);
4261       // libjsig also interposes the sigaction() call below and saves the
4262       // old sigaction on it own.
4263     } else {
4264       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4265                     "%#lx for signal %d.", (long)oldhand, sig));
4266     }
4267   }
4268 
4269   struct sigaction sigAct;
4270   sigfillset(&(sigAct.sa_mask));
4271   sigAct.sa_handler = SIG_DFL;
4272   if (!set_installed) {
4273     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4274   } else {
4275     sigAct.sa_sigaction = signalHandler;
4276     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4277   }
4278   // Save flags, which are set by ours
4279   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4280   sigflags[sig] = sigAct.sa_flags;
4281 
4282   int ret = sigaction(sig, &sigAct, &oldAct);
4283   assert(ret == 0, "check");
4284 
4285   void* oldhand2  = oldAct.sa_sigaction
4286                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4287                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4288   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4289 }
4290 
4291 // install signal handlers for signals that HotSpot needs to
4292 // handle in order to support Java-level exception handling.
4293 
4294 void os::Linux::install_signal_handlers() {
4295   if (!signal_handlers_are_installed) {
4296     signal_handlers_are_installed = true;
4297 
4298     // signal-chaining
4299     typedef void (*signal_setting_t)();
4300     signal_setting_t begin_signal_setting = NULL;
4301     signal_setting_t end_signal_setting = NULL;
4302     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4303                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4304     if (begin_signal_setting != NULL) {
4305       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4306                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4307       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4308                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4309       libjsig_is_loaded = true;
4310       assert(UseSignalChaining, "should enable signal-chaining");
4311     }
4312     if (libjsig_is_loaded) {
4313       // Tell libjsig jvm is setting signal handlers
4314       (*begin_signal_setting)();
4315     }
4316 
4317     set_signal_handler(SIGSEGV, true);
4318     set_signal_handler(SIGPIPE, true);
4319     set_signal_handler(SIGBUS, true);
4320     set_signal_handler(SIGILL, true);
4321     set_signal_handler(SIGFPE, true);
4322 #if defined(PPC64)
4323     set_signal_handler(SIGTRAP, true);
4324 #endif
4325     set_signal_handler(SIGXFSZ, true);
4326 
4327     if (libjsig_is_loaded) {
4328       // Tell libjsig jvm finishes setting signal handlers
4329       (*end_signal_setting)();
4330     }
4331 
4332     // We don't activate signal checker if libjsig is in place, we trust ourselves
4333     // and if UserSignalHandler is installed all bets are off.
4334     // Log that signal checking is off only if -verbose:jni is specified.
4335     if (CheckJNICalls) {
4336       if (libjsig_is_loaded) {
4337         if (PrintJNIResolving) {
4338           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4339         }
4340         check_signals = false;
4341       }
4342       if (AllowUserSignalHandlers) {
4343         if (PrintJNIResolving) {
4344           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4345         }
4346         check_signals = false;
4347       }
4348     }
4349   }
4350 }
4351 
4352 // This is the fastest way to get thread cpu time on Linux.
4353 // Returns cpu time (user+sys) for any thread, not only for current.
4354 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4355 // It might work on 2.6.10+ with a special kernel/glibc patch.
4356 // For reference, please, see IEEE Std 1003.1-2004:
4357 //   http://www.unix.org/single_unix_specification
4358 
4359 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4360   struct timespec tp;
4361   int rc = os::Linux::clock_gettime(clockid, &tp);
4362   assert(rc == 0, "clock_gettime is expected to return 0 code");
4363 
4364   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4365 }
4366 
4367 /////
4368 // glibc on Linux platform uses non-documented flag
4369 // to indicate, that some special sort of signal
4370 // trampoline is used.
4371 // We will never set this flag, and we should
4372 // ignore this flag in our diagnostic
4373 #ifdef SIGNIFICANT_SIGNAL_MASK
4374 #undef SIGNIFICANT_SIGNAL_MASK
4375 #endif
4376 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4377 
4378 static const char* get_signal_handler_name(address handler,
4379                                            char* buf, int buflen) {
4380   int offset;
4381   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4382   if (found) {
4383     // skip directory names
4384     const char *p1, *p2;
4385     p1 = buf;
4386     size_t len = strlen(os::file_separator());
4387     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4388     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4389   } else {
4390     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4391   }
4392   return buf;
4393 }
4394 
4395 static void print_signal_handler(outputStream* st, int sig,
4396                                  char* buf, size_t buflen) {
4397   struct sigaction sa;
4398 
4399   sigaction(sig, NULL, &sa);
4400 
4401   // See comment for SIGNIFICANT_SIGNAL_MASK define
4402   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4403 
4404   st->print("%s: ", os::exception_name(sig, buf, buflen));
4405 
4406   address handler = (sa.sa_flags & SA_SIGINFO)
4407     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4408     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4409 
4410   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4411     st->print("SIG_DFL");
4412   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4413     st->print("SIG_IGN");
4414   } else {
4415     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4416   }
4417 
4418   st->print(", sa_mask[0]=");
4419   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4420 
4421   address rh = VMError::get_resetted_sighandler(sig);
4422   // May be, handler was resetted by VMError?
4423   if(rh != NULL) {
4424     handler = rh;
4425     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4426   }
4427 
4428   st->print(", sa_flags=");
4429   os::Posix::print_sa_flags(st, sa.sa_flags);
4430 
4431   // Check: is it our handler?
4432   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4433      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4434     // It is our signal handler
4435     // check for flags, reset system-used one!
4436     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4437       st->print(
4438                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4439                 os::Linux::get_our_sigflags(sig));
4440     }
4441   }
4442   st->cr();
4443 }
4444 
4445 
4446 #define DO_SIGNAL_CHECK(sig) \
4447   if (!sigismember(&check_signal_done, sig)) \
4448     os::Linux::check_signal_handler(sig)
4449 
4450 // This method is a periodic task to check for misbehaving JNI applications
4451 // under CheckJNI, we can add any periodic checks here
4452 
4453 void os::run_periodic_checks() {
4454 
4455   if (check_signals == false) return;
4456 
4457   // SEGV and BUS if overridden could potentially prevent
4458   // generation of hs*.log in the event of a crash, debugging
4459   // such a case can be very challenging, so we absolutely
4460   // check the following for a good measure:
4461   DO_SIGNAL_CHECK(SIGSEGV);
4462   DO_SIGNAL_CHECK(SIGILL);
4463   DO_SIGNAL_CHECK(SIGFPE);
4464   DO_SIGNAL_CHECK(SIGBUS);
4465   DO_SIGNAL_CHECK(SIGPIPE);
4466   DO_SIGNAL_CHECK(SIGXFSZ);
4467 #if defined(PPC64)
4468   DO_SIGNAL_CHECK(SIGTRAP);
4469 #endif
4470 
4471   // ReduceSignalUsage allows the user to override these handlers
4472   // see comments at the very top and jvm_solaris.h
4473   if (!ReduceSignalUsage) {
4474     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4475     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4476     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4477     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4478   }
4479 
4480   DO_SIGNAL_CHECK(SR_signum);
4481   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4482 }
4483 
4484 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4485 
4486 static os_sigaction_t os_sigaction = NULL;
4487 
4488 void os::Linux::check_signal_handler(int sig) {
4489   char buf[O_BUFLEN];
4490   address jvmHandler = NULL;
4491 
4492 
4493   struct sigaction act;
4494   if (os_sigaction == NULL) {
4495     // only trust the default sigaction, in case it has been interposed
4496     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4497     if (os_sigaction == NULL) return;
4498   }
4499 
4500   os_sigaction(sig, (struct sigaction*)NULL, &act);
4501 
4502 
4503   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4504 
4505   address thisHandler = (act.sa_flags & SA_SIGINFO)
4506     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4507     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4508 
4509 
4510   switch(sig) {
4511   case SIGSEGV:
4512   case SIGBUS:
4513   case SIGFPE:
4514   case SIGPIPE:
4515   case SIGILL:
4516   case SIGXFSZ:
4517     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4518     break;
4519 
4520   case SHUTDOWN1_SIGNAL:
4521   case SHUTDOWN2_SIGNAL:
4522   case SHUTDOWN3_SIGNAL:
4523   case BREAK_SIGNAL:
4524     jvmHandler = (address)user_handler();
4525     break;
4526 
4527   case INTERRUPT_SIGNAL:
4528     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4529     break;
4530 
4531   default:
4532     if (sig == SR_signum) {
4533       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4534     } else {
4535       return;
4536     }
4537     break;
4538   }
4539 
4540   if (thisHandler != jvmHandler) {
4541     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4542     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4543     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4544     // No need to check this sig any longer
4545     sigaddset(&check_signal_done, sig);
4546     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4547     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4548       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4549                     exception_name(sig, buf, O_BUFLEN));
4550     }
4551   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4552     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4553     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4554     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4555     // No need to check this sig any longer
4556     sigaddset(&check_signal_done, sig);
4557   }
4558 
4559   // Dump all the signal
4560   if (sigismember(&check_signal_done, sig)) {
4561     print_signal_handlers(tty, buf, O_BUFLEN);
4562   }
4563 }
4564 
4565 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4566 
4567 extern bool signal_name(int signo, char* buf, size_t len);
4568 
4569 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4570   if (0 < exception_code && exception_code <= SIGRTMAX) {
4571     // signal
4572     if (!signal_name(exception_code, buf, size)) {
4573       jio_snprintf(buf, size, "SIG%d", exception_code);
4574     }
4575     return buf;
4576   } else {
4577     return NULL;
4578   }
4579 }
4580 
4581 // this is called _before_ the most of global arguments have been parsed
4582 void os::init(void) {
4583   char dummy;   /* used to get a guess on initial stack address */
4584 //  first_hrtime = gethrtime();
4585 
4586   // With LinuxThreads the JavaMain thread pid (primordial thread)
4587   // is different than the pid of the java launcher thread.
4588   // So, on Linux, the launcher thread pid is passed to the VM
4589   // via the sun.java.launcher.pid property.
4590   // Use this property instead of getpid() if it was correctly passed.
4591   // See bug 6351349.
4592   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4593 
4594   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4595 
4596   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4597 
4598   init_random(1234567);
4599 
4600   ThreadCritical::initialize();
4601 
4602   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4603   if (Linux::page_size() == -1) {
4604     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4605                   strerror(errno)));
4606   }
4607   init_page_sizes((size_t) Linux::page_size());
4608 
4609   Linux::initialize_system_info();
4610 
4611   // main_thread points to the aboriginal thread
4612   Linux::_main_thread = pthread_self();
4613 
4614   Linux::clock_init();
4615   initial_time_count = javaTimeNanos();
4616 
4617   // pthread_condattr initialization for monotonic clock
4618   int status;
4619   pthread_condattr_t* _condattr = os::Linux::condAttr();
4620   if ((status = pthread_condattr_init(_condattr)) != 0) {
4621     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4622   }
4623   // Only set the clock if CLOCK_MONOTONIC is available
4624   if (os::supports_monotonic_clock()) {
4625     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4626       if (status == EINVAL) {
4627         warning("Unable to use monotonic clock with relative timed-waits" \
4628                 " - changes to the time-of-day clock may have adverse affects");
4629       } else {
4630         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4631       }
4632     }
4633   }
4634   // else it defaults to CLOCK_REALTIME
4635 
4636   pthread_mutex_init(&dl_mutex, NULL);
4637 
4638   // If the pagesize of the VM is greater than 8K determine the appropriate
4639   // number of initial guard pages.  The user can change this with the
4640   // command line arguments, if needed.
4641   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4642     StackYellowPages = 1;
4643     StackRedPages = 1;
4644     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4645   }
4646 }
4647 
4648 // To install functions for atexit system call
4649 extern "C" {
4650   static void perfMemory_exit_helper() {
4651     perfMemory_exit();
4652   }
4653 }
4654 
4655 // this is called _after_ the global arguments have been parsed
4656 jint os::init_2(void)
4657 {
4658   Linux::fast_thread_clock_init();
4659 
4660   // Allocate a single page and mark it as readable for safepoint polling
4661   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4662   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4663 
4664   os::set_polling_page( polling_page );
4665 
4666 #ifndef PRODUCT
4667   if(Verbose && PrintMiscellaneous)
4668     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4669 #endif
4670 
4671   if (!UseMembar) {
4672     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4673     guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4674     os::set_memory_serialize_page( mem_serialize_page );
4675 
4676 #ifndef PRODUCT
4677     if(Verbose && PrintMiscellaneous)
4678       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4679 #endif
4680   }
4681 
4682   // initialize suspend/resume support - must do this before signal_sets_init()
4683   if (SR_initialize() != 0) {
4684     perror("SR_initialize failed");
4685     return JNI_ERR;
4686   }
4687 
4688   Linux::signal_sets_init();
4689   Linux::install_signal_handlers();
4690 
4691   // Check minimum allowable stack size for thread creation and to initialize
4692   // the java system classes, including StackOverflowError - depends on page
4693   // size.  Add a page for compiler2 recursion in main thread.
4694   // Add in 2*BytesPerWord times page size to account for VM stack during
4695   // class initialization depending on 32 or 64 bit VM.
4696   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4697             (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4698                     (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4699 
4700   size_t threadStackSizeInBytes = ThreadStackSize * K;
4701   if (threadStackSizeInBytes != 0 &&
4702       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4703         tty->print_cr("\nThe stack size specified is too small, "
4704                       "Specify at least %dk",
4705                       os::Linux::min_stack_allowed/ K);
4706         return JNI_ERR;
4707   }
4708 
4709   // Make the stack size a multiple of the page size so that
4710   // the yellow/red zones can be guarded.
4711   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4712         vm_page_size()));
4713 
4714   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4715 
4716 #if defined(IA32)
4717   workaround_expand_exec_shield_cs_limit();
4718 #endif
4719 
4720   Linux::libpthread_init();
4721   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4722      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4723           Linux::glibc_version(), Linux::libpthread_version(),
4724           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4725   }
4726 
4727   if (UseNUMA) {
4728     if (!Linux::libnuma_init()) {
4729       UseNUMA = false;
4730     } else {
4731       if ((Linux::numa_max_node() < 1)) {
4732         // There's only one node(they start from 0), disable NUMA.
4733         UseNUMA = false;
4734       }
4735     }
4736     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4737     // we can make the adaptive lgrp chunk resizing work. If the user specified
4738     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4739     // disable adaptive resizing.
4740     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4741       if (FLAG_IS_DEFAULT(UseNUMA)) {
4742         UseNUMA = false;
4743       } else {
4744         if (FLAG_IS_DEFAULT(UseLargePages) &&
4745             FLAG_IS_DEFAULT(UseSHM) &&
4746             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4747           UseLargePages = false;
4748         } else {
4749           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4750           UseAdaptiveSizePolicy = false;
4751           UseAdaptiveNUMAChunkSizing = false;
4752         }
4753       }
4754     }
4755     if (!UseNUMA && ForceNUMA) {
4756       UseNUMA = true;
4757     }
4758   }
4759 
4760   if (MaxFDLimit) {
4761     // set the number of file descriptors to max. print out error
4762     // if getrlimit/setrlimit fails but continue regardless.
4763     struct rlimit nbr_files;
4764     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4765     if (status != 0) {
4766       if (PrintMiscellaneous && (Verbose || WizardMode))
4767         perror("os::init_2 getrlimit failed");
4768     } else {
4769       nbr_files.rlim_cur = nbr_files.rlim_max;
4770       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4771       if (status != 0) {
4772         if (PrintMiscellaneous && (Verbose || WizardMode))
4773           perror("os::init_2 setrlimit failed");
4774       }
4775     }
4776   }
4777 
4778   // Initialize lock used to serialize thread creation (see os::create_thread)
4779   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4780 
4781   // at-exit methods are called in the reverse order of their registration.
4782   // atexit functions are called on return from main or as a result of a
4783   // call to exit(3C). There can be only 32 of these functions registered
4784   // and atexit() does not set errno.
4785 
4786   if (PerfAllowAtExitRegistration) {
4787     // only register atexit functions if PerfAllowAtExitRegistration is set.
4788     // atexit functions can be delayed until process exit time, which
4789     // can be problematic for embedded VM situations. Embedded VMs should
4790     // call DestroyJavaVM() to assure that VM resources are released.
4791 
4792     // note: perfMemory_exit_helper atexit function may be removed in
4793     // the future if the appropriate cleanup code can be added to the
4794     // VM_Exit VMOperation's doit method.
4795     if (atexit(perfMemory_exit_helper) != 0) {
4796       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4797     }
4798   }
4799 
4800   // initialize thread priority policy
4801   prio_init();
4802 
4803   return JNI_OK;
4804 }
4805 
4806 // this is called at the end of vm_initialization
4807 void os::init_3(void) {
4808 #ifdef JAVASE_EMBEDDED
4809   // Start the MemNotifyThread
4810   if (LowMemoryProtection) {
4811     MemNotifyThread::start();
4812   }
4813   return;
4814 #endif
4815 }
4816 
4817 // Mark the polling page as unreadable
4818 void os::make_polling_page_unreadable(void) {
4819   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4820     fatal("Could not disable polling page");
4821 };
4822 
4823 // Mark the polling page as readable
4824 void os::make_polling_page_readable(void) {
4825   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4826     fatal("Could not enable polling page");
4827   }
4828 };
4829 
4830 int os::active_processor_count() {
4831   // Linux doesn't yet have a (official) notion of processor sets,
4832   // so just return the number of online processors.
4833   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4834   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4835   return online_cpus;
4836 }
4837 
4838 void os::set_native_thread_name(const char *name) {
4839   // Not yet implemented.
4840   return;
4841 }
4842 
4843 bool os::distribute_processes(uint length, uint* distribution) {
4844   // Not yet implemented.
4845   return false;
4846 }
4847 
4848 bool os::bind_to_processor(uint processor_id) {
4849   // Not yet implemented.
4850   return false;
4851 }
4852 
4853 ///
4854 
4855 void os::SuspendedThreadTask::internal_do_task() {
4856   if (do_suspend(_thread->osthread())) {
4857     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4858     do_task(context);
4859     do_resume(_thread->osthread());
4860   }
4861 }
4862 
4863 class PcFetcher : public os::SuspendedThreadTask {
4864 public:
4865   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4866   ExtendedPC result();
4867 protected:
4868   void do_task(const os::SuspendedThreadTaskContext& context);
4869 private:
4870   ExtendedPC _epc;
4871 };
4872 
4873 ExtendedPC PcFetcher::result() {
4874   guarantee(is_done(), "task is not done yet.");
4875   return _epc;
4876 }
4877 
4878 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4879   Thread* thread = context.thread();
4880   OSThread* osthread = thread->osthread();
4881   if (osthread->ucontext() != NULL) {
4882     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4883   } else {
4884     // NULL context is unexpected, double-check this is the VMThread
4885     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4886   }
4887 }
4888 
4889 // Suspends the target using the signal mechanism and then grabs the PC before
4890 // resuming the target. Used by the flat-profiler only
4891 ExtendedPC os::get_thread_pc(Thread* thread) {
4892   // Make sure that it is called by the watcher for the VMThread
4893   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4894   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4895 
4896   PcFetcher fetcher(thread);
4897   fetcher.run();
4898   return fetcher.result();
4899 }
4900 
4901 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4902 {
4903    if (is_NPTL()) {
4904       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4905    } else {
4906       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4907       // word back to default 64bit precision if condvar is signaled. Java
4908       // wants 53bit precision.  Save and restore current value.
4909       int fpu = get_fpu_control_word();
4910       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4911       set_fpu_control_word(fpu);
4912       return status;
4913    }
4914 }
4915 
4916 ////////////////////////////////////////////////////////////////////////////////
4917 // debug support
4918 
4919 bool os::find(address addr, outputStream* st) {
4920   Dl_info dlinfo;
4921   memset(&dlinfo, 0, sizeof(dlinfo));
4922   if (dladdr(addr, &dlinfo) != 0) {
4923     st->print(PTR_FORMAT ": ", addr);
4924     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4925       st->print("%s+%#x", dlinfo.dli_sname,
4926                  addr - (intptr_t)dlinfo.dli_saddr);
4927     } else if (dlinfo.dli_fbase != NULL) {
4928       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4929     } else {
4930       st->print("<absolute address>");
4931     }
4932     if (dlinfo.dli_fname != NULL) {
4933       st->print(" in %s", dlinfo.dli_fname);
4934     }
4935     if (dlinfo.dli_fbase != NULL) {
4936       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4937     }
4938     st->cr();
4939 
4940     if (Verbose) {
4941       // decode some bytes around the PC
4942       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4943       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4944       address       lowest = (address) dlinfo.dli_sname;
4945       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4946       if (begin < lowest)  begin = lowest;
4947       Dl_info dlinfo2;
4948       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4949           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4950         end = (address) dlinfo2.dli_saddr;
4951       Disassembler::decode(begin, end, st);
4952     }
4953     return true;
4954   }
4955   return false;
4956 }
4957 
4958 ////////////////////////////////////////////////////////////////////////////////
4959 // misc
4960 
4961 // This does not do anything on Linux. This is basically a hook for being
4962 // able to use structured exception handling (thread-local exception filters)
4963 // on, e.g., Win32.
4964 void
4965 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4966                          JavaCallArguments* args, Thread* thread) {
4967   f(value, method, args, thread);
4968 }
4969 
4970 void os::print_statistics() {
4971 }
4972 
4973 int os::message_box(const char* title, const char* message) {
4974   int i;
4975   fdStream err(defaultStream::error_fd());
4976   for (i = 0; i < 78; i++) err.print_raw("=");
4977   err.cr();
4978   err.print_raw_cr(title);
4979   for (i = 0; i < 78; i++) err.print_raw("-");
4980   err.cr();
4981   err.print_raw_cr(message);
4982   for (i = 0; i < 78; i++) err.print_raw("=");
4983   err.cr();
4984 
4985   char buf[16];
4986   // Prevent process from exiting upon "read error" without consuming all CPU
4987   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4988 
4989   return buf[0] == 'y' || buf[0] == 'Y';
4990 }
4991 
4992 int os::stat(const char *path, struct stat *sbuf) {
4993   char pathbuf[MAX_PATH];
4994   if (strlen(path) > MAX_PATH - 1) {
4995     errno = ENAMETOOLONG;
4996     return -1;
4997   }
4998   os::native_path(strcpy(pathbuf, path));
4999   return ::stat(pathbuf, sbuf);
5000 }
5001 
5002 bool os::check_heap(bool force) {
5003   return true;
5004 }
5005 
5006 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5007   return ::vsnprintf(buf, count, format, args);
5008 }
5009 
5010 // Is a (classpath) directory empty?
5011 bool os::dir_is_empty(const char* path) {
5012   DIR *dir = NULL;
5013   struct dirent *ptr;
5014 
5015   dir = opendir(path);
5016   if (dir == NULL) return true;
5017 
5018   /* Scan the directory */
5019   bool result = true;
5020   char buf[sizeof(struct dirent) + MAX_PATH];
5021   while (result && (ptr = ::readdir(dir)) != NULL) {
5022     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5023       result = false;
5024     }
5025   }
5026   closedir(dir);
5027   return result;
5028 }
5029 
5030 // This code originates from JDK's sysOpen and open64_w
5031 // from src/solaris/hpi/src/system_md.c
5032 
5033 #ifndef O_DELETE
5034 #define O_DELETE 0x10000
5035 #endif
5036 
5037 // Open a file. Unlink the file immediately after open returns
5038 // if the specified oflag has the O_DELETE flag set.
5039 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5040 
5041 int os::open(const char *path, int oflag, int mode) {
5042 
5043   if (strlen(path) > MAX_PATH - 1) {
5044     errno = ENAMETOOLONG;
5045     return -1;
5046   }
5047   int fd;
5048   int o_delete = (oflag & O_DELETE);
5049   oflag = oflag & ~O_DELETE;
5050 
5051   fd = ::open64(path, oflag, mode);
5052   if (fd == -1) return -1;
5053 
5054   //If the open succeeded, the file might still be a directory
5055   {
5056     struct stat64 buf64;
5057     int ret = ::fstat64(fd, &buf64);
5058     int st_mode = buf64.st_mode;
5059 
5060     if (ret != -1) {
5061       if ((st_mode & S_IFMT) == S_IFDIR) {
5062         errno = EISDIR;
5063         ::close(fd);
5064         return -1;
5065       }
5066     } else {
5067       ::close(fd);
5068       return -1;
5069     }
5070   }
5071 
5072     /*
5073      * All file descriptors that are opened in the JVM and not
5074      * specifically destined for a subprocess should have the
5075      * close-on-exec flag set.  If we don't set it, then careless 3rd
5076      * party native code might fork and exec without closing all
5077      * appropriate file descriptors (e.g. as we do in closeDescriptors in
5078      * UNIXProcess.c), and this in turn might:
5079      *
5080      * - cause end-of-file to fail to be detected on some file
5081      *   descriptors, resulting in mysterious hangs, or
5082      *
5083      * - might cause an fopen in the subprocess to fail on a system
5084      *   suffering from bug 1085341.
5085      *
5086      * (Yes, the default setting of the close-on-exec flag is a Unix
5087      * design flaw)
5088      *
5089      * See:
5090      * 1085341: 32-bit stdio routines should support file descriptors >255
5091      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5092      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5093      */
5094 #ifdef FD_CLOEXEC
5095     {
5096         int flags = ::fcntl(fd, F_GETFD);
5097         if (flags != -1)
5098             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5099     }
5100 #endif
5101 
5102   if (o_delete != 0) {
5103     ::unlink(path);
5104   }
5105   return fd;
5106 }
5107 
5108 
5109 // create binary file, rewriting existing file if required
5110 int os::create_binary_file(const char* path, bool rewrite_existing) {
5111   int oflags = O_WRONLY | O_CREAT;
5112   if (!rewrite_existing) {
5113     oflags |= O_EXCL;
5114   }
5115   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5116 }
5117 
5118 // return current position of file pointer
5119 jlong os::current_file_offset(int fd) {
5120   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5121 }
5122 
5123 // move file pointer to the specified offset
5124 jlong os::seek_to_file_offset(int fd, jlong offset) {
5125   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5126 }
5127 
5128 // This code originates from JDK's sysAvailable
5129 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5130 
5131 int os::available(int fd, jlong *bytes) {
5132   jlong cur, end;
5133   int mode;
5134   struct stat64 buf64;
5135 
5136   if (::fstat64(fd, &buf64) >= 0) {
5137     mode = buf64.st_mode;
5138     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5139       /*
5140       * XXX: is the following call interruptible? If so, this might
5141       * need to go through the INTERRUPT_IO() wrapper as for other
5142       * blocking, interruptible calls in this file.
5143       */
5144       int n;
5145       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5146         *bytes = n;
5147         return 1;
5148       }
5149     }
5150   }
5151   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5152     return 0;
5153   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5154     return 0;
5155   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5156     return 0;
5157   }
5158   *bytes = end - cur;
5159   return 1;
5160 }
5161 
5162 int os::socket_available(int fd, jint *pbytes) {
5163   // Linux doc says EINTR not returned, unlike Solaris
5164   int ret = ::ioctl(fd, FIONREAD, pbytes);
5165 
5166   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5167   // is expected to return 0 on failure and 1 on success to the jdk.
5168   return (ret < 0) ? 0 : 1;
5169 }
5170 
5171 // Map a block of memory.
5172 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5173                      char *addr, size_t bytes, bool read_only,
5174                      bool allow_exec) {
5175   int prot;
5176   int flags = MAP_PRIVATE;
5177 
5178   if (read_only) {
5179     prot = PROT_READ;
5180   } else {
5181     prot = PROT_READ | PROT_WRITE;
5182   }
5183 
5184   if (allow_exec) {
5185     prot |= PROT_EXEC;
5186   }
5187 
5188   if (addr != NULL) {
5189     flags |= MAP_FIXED;
5190   }
5191 
5192   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5193                                      fd, file_offset);
5194   if (mapped_address == MAP_FAILED) {
5195     return NULL;
5196   }
5197   return mapped_address;
5198 }
5199 
5200 
5201 // Remap a block of memory.
5202 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5203                        char *addr, size_t bytes, bool read_only,
5204                        bool allow_exec) {
5205   // same as map_memory() on this OS
5206   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5207                         allow_exec);
5208 }
5209 
5210 
5211 // Unmap a block of memory.
5212 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5213   return munmap(addr, bytes) == 0;
5214 }
5215 
5216 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5217 
5218 static clockid_t thread_cpu_clockid(Thread* thread) {
5219   pthread_t tid = thread->osthread()->pthread_id();
5220   clockid_t clockid;
5221 
5222   // Get thread clockid
5223   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5224   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5225   return clockid;
5226 }
5227 
5228 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5229 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5230 // of a thread.
5231 //
5232 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5233 // the fast estimate available on the platform.
5234 
5235 jlong os::current_thread_cpu_time() {
5236   if (os::Linux::supports_fast_thread_cpu_time()) {
5237     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5238   } else {
5239     // return user + sys since the cost is the same
5240     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5241   }
5242 }
5243 
5244 jlong os::thread_cpu_time(Thread* thread) {
5245   // consistent with what current_thread_cpu_time() returns
5246   if (os::Linux::supports_fast_thread_cpu_time()) {
5247     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5248   } else {
5249     return slow_thread_cpu_time(thread, true /* user + sys */);
5250   }
5251 }
5252 
5253 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5254   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5255     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5256   } else {
5257     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5258   }
5259 }
5260 
5261 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5262   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5263     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5264   } else {
5265     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5266   }
5267 }
5268 
5269 //
5270 //  -1 on error.
5271 //
5272 
5273 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5274   pid_t  tid = thread->osthread()->thread_id();
5275   char *s;
5276   char stat[2048];
5277   int statlen;
5278   char proc_name[64];
5279   int count;
5280   long sys_time, user_time;
5281   char cdummy;
5282   int idummy;
5283   long ldummy;
5284   FILE *fp;
5285 
5286   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5287   fp = fopen(proc_name, "r");
5288   if ( fp == NULL ) return -1;
5289   statlen = fread(stat, 1, 2047, fp);
5290   stat[statlen] = '\0';
5291   fclose(fp);
5292 
5293   // Skip pid and the command string. Note that we could be dealing with
5294   // weird command names, e.g. user could decide to rename java launcher
5295   // to "java 1.4.2 :)", then the stat file would look like
5296   //                1234 (java 1.4.2 :)) R ... ...
5297   // We don't really need to know the command string, just find the last
5298   // occurrence of ")" and then start parsing from there. See bug 4726580.
5299   s = strrchr(stat, ')');
5300   if (s == NULL ) return -1;
5301 
5302   // Skip blank chars
5303   do s++; while (isspace(*s));
5304 
5305   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5306                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5307                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5308                  &user_time, &sys_time);
5309   if ( count != 13 ) return -1;
5310   if (user_sys_cpu_time) {
5311     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5312   } else {
5313     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5314   }
5315 }
5316 
5317 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5318   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5319   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5320   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5321   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5322 }
5323 
5324 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5325   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5326   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5327   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5328   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5329 }
5330 
5331 bool os::is_thread_cpu_time_supported() {
5332   return true;
5333 }
5334 
5335 // System loadavg support.  Returns -1 if load average cannot be obtained.
5336 // Linux doesn't yet have a (official) notion of processor sets,
5337 // so just return the system wide load average.
5338 int os::loadavg(double loadavg[], int nelem) {
5339   return ::getloadavg(loadavg, nelem);
5340 }
5341 
5342 void os::pause() {
5343   char filename[MAX_PATH];
5344   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5345     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5346   } else {
5347     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5348   }
5349 
5350   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5351   if (fd != -1) {
5352     struct stat buf;
5353     ::close(fd);
5354     while (::stat(filename, &buf) == 0) {
5355       (void)::poll(NULL, 0, 100);
5356     }
5357   } else {
5358     jio_fprintf(stderr,
5359       "Could not open pause file '%s', continuing immediately.\n", filename);
5360   }
5361 }
5362 
5363 
5364 // Refer to the comments in os_solaris.cpp park-unpark.
5365 //
5366 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5367 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5368 // For specifics regarding the bug see GLIBC BUGID 261237 :
5369 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5370 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5371 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5372 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5373 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5374 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5375 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5376 // of libpthread avoids the problem, but isn't practical.
5377 //
5378 // Possible remedies:
5379 //
5380 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5381 //      This is palliative and probabilistic, however.  If the thread is preempted
5382 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5383 //      than the minimum period may have passed, and the abstime may be stale (in the
5384 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5385 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5386 //
5387 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5388 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5389 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5390 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5391 //      thread.
5392 //
5393 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5394 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5395 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5396 //      This also works well.  In fact it avoids kernel-level scalability impediments
5397 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5398 //      timers in a graceful fashion.
5399 //
5400 // 4.   When the abstime value is in the past it appears that control returns
5401 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5402 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5403 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5404 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5405 //      It may be possible to avoid reinitialization by checking the return
5406 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5407 //      condvar we must establish the invariant that cond_signal() is only called
5408 //      within critical sections protected by the adjunct mutex.  This prevents
5409 //      cond_signal() from "seeing" a condvar that's in the midst of being
5410 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5411 //      desirable signal-after-unlock optimization that avoids futile context switching.
5412 //
5413 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5414 //      structure when a condvar is used or initialized.  cond_destroy()  would
5415 //      release the helper structure.  Our reinitialize-after-timedwait fix
5416 //      put excessive stress on malloc/free and locks protecting the c-heap.
5417 //
5418 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5419 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5420 // and only enabling the work-around for vulnerable environments.
5421 
5422 // utility to compute the abstime argument to timedwait:
5423 // millis is the relative timeout time
5424 // abstime will be the absolute timeout time
5425 // TODO: replace compute_abstime() with unpackTime()
5426 
5427 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5428   if (millis < 0)  millis = 0;
5429 
5430   jlong seconds = millis / 1000;
5431   millis %= 1000;
5432   if (seconds > 50000000) { // see man cond_timedwait(3T)
5433     seconds = 50000000;
5434   }
5435 
5436   if (os::supports_monotonic_clock()) {
5437     struct timespec now;
5438     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5439     assert_status(status == 0, status, "clock_gettime");
5440     abstime->tv_sec = now.tv_sec  + seconds;
5441     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5442     if (nanos >= NANOSECS_PER_SEC) {
5443       abstime->tv_sec += 1;
5444       nanos -= NANOSECS_PER_SEC;
5445     }
5446     abstime->tv_nsec = nanos;
5447   } else {
5448     struct timeval now;
5449     int status = gettimeofday(&now, NULL);
5450     assert(status == 0, "gettimeofday");
5451     abstime->tv_sec = now.tv_sec  + seconds;
5452     long usec = now.tv_usec + millis * 1000;
5453     if (usec >= 1000000) {
5454       abstime->tv_sec += 1;
5455       usec -= 1000000;
5456     }
5457     abstime->tv_nsec = usec * 1000;
5458   }
5459   return abstime;
5460 }
5461 
5462 
5463 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5464 // Conceptually TryPark() should be equivalent to park(0).
5465 
5466 int os::PlatformEvent::TryPark() {
5467   for (;;) {
5468     const int v = _Event ;
5469     guarantee ((v == 0) || (v == 1), "invariant") ;
5470     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5471   }
5472 }
5473 
5474 void os::PlatformEvent::park() {       // AKA "down()"
5475   // Invariant: Only the thread associated with the Event/PlatformEvent
5476   // may call park().
5477   // TODO: assert that _Assoc != NULL or _Assoc == Self
5478   int v ;
5479   for (;;) {
5480       v = _Event ;
5481       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5482   }
5483   guarantee (v >= 0, "invariant") ;
5484   if (v == 0) {
5485      // Do this the hard way by blocking ...
5486      int status = pthread_mutex_lock(_mutex);
5487      assert_status(status == 0, status, "mutex_lock");
5488      guarantee (_nParked == 0, "invariant") ;
5489      ++ _nParked ;
5490      while (_Event < 0) {
5491         status = pthread_cond_wait(_cond, _mutex);
5492         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5493         // Treat this the same as if the wait was interrupted
5494         if (status == ETIME) { status = EINTR; }
5495         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5496      }
5497      -- _nParked ;
5498 
5499     _Event = 0 ;
5500      status = pthread_mutex_unlock(_mutex);
5501      assert_status(status == 0, status, "mutex_unlock");
5502     // Paranoia to ensure our locked and lock-free paths interact
5503     // correctly with each other.
5504     OrderAccess::fence();
5505   }
5506   guarantee (_Event >= 0, "invariant") ;
5507 }
5508 
5509 int os::PlatformEvent::park(jlong millis) {
5510   guarantee (_nParked == 0, "invariant") ;
5511 
5512   int v ;
5513   for (;;) {
5514       v = _Event ;
5515       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5516   }
5517   guarantee (v >= 0, "invariant") ;
5518   if (v != 0) return OS_OK ;
5519 
5520   // We do this the hard way, by blocking the thread.
5521   // Consider enforcing a minimum timeout value.
5522   struct timespec abst;
5523   compute_abstime(&abst, millis);
5524 
5525   int ret = OS_TIMEOUT;
5526   int status = pthread_mutex_lock(_mutex);
5527   assert_status(status == 0, status, "mutex_lock");
5528   guarantee (_nParked == 0, "invariant") ;
5529   ++_nParked ;
5530 
5531   // Object.wait(timo) will return because of
5532   // (a) notification
5533   // (b) timeout
5534   // (c) thread.interrupt
5535   //
5536   // Thread.interrupt and object.notify{All} both call Event::set.
5537   // That is, we treat thread.interrupt as a special case of notification.
5538   // The underlying Solaris implementation, cond_timedwait, admits
5539   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5540   // JVM from making those visible to Java code.  As such, we must
5541   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5542   //
5543   // TODO: properly differentiate simultaneous notify+interrupt.
5544   // In that case, we should propagate the notify to another waiter.
5545 
5546   while (_Event < 0) {
5547     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5548     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5549       pthread_cond_destroy (_cond);
5550       pthread_cond_init (_cond, os::Linux::condAttr()) ;
5551     }
5552     assert_status(status == 0 || status == EINTR ||
5553                   status == ETIME || status == ETIMEDOUT,
5554                   status, "cond_timedwait");
5555     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5556     if (status == ETIME || status == ETIMEDOUT) break ;
5557     // We consume and ignore EINTR and spurious wakeups.
5558   }
5559   --_nParked ;
5560   if (_Event >= 0) {
5561      ret = OS_OK;
5562   }
5563   _Event = 0 ;
5564   status = pthread_mutex_unlock(_mutex);
5565   assert_status(status == 0, status, "mutex_unlock");
5566   assert (_nParked == 0, "invariant") ;
5567   // Paranoia to ensure our locked and lock-free paths interact
5568   // correctly with each other.
5569   OrderAccess::fence();
5570   return ret;
5571 }
5572 
5573 void os::PlatformEvent::unpark() {
5574   // Transitions for _Event:
5575   //    0 :=> 1
5576   //    1 :=> 1
5577   //   -1 :=> either 0 or 1; must signal target thread
5578   //          That is, we can safely transition _Event from -1 to either
5579   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
5580   //          unpark() calls.
5581   // See also: "Semaphores in Plan 9" by Mullender & Cox
5582   //
5583   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5584   // that it will take two back-to-back park() calls for the owning
5585   // thread to block. This has the benefit of forcing a spurious return
5586   // from the first park() call after an unpark() call which will help
5587   // shake out uses of park() and unpark() without condition variables.
5588 
5589   if (Atomic::xchg(1, &_Event) >= 0) return;
5590 
5591   // Wait for the thread associated with the event to vacate
5592   int status = pthread_mutex_lock(_mutex);
5593   assert_status(status == 0, status, "mutex_lock");
5594   int AnyWaiters = _nParked;
5595   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5596   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5597     AnyWaiters = 0;
5598     pthread_cond_signal(_cond);
5599   }
5600   status = pthread_mutex_unlock(_mutex);
5601   assert_status(status == 0, status, "mutex_unlock");
5602   if (AnyWaiters != 0) {
5603     status = pthread_cond_signal(_cond);
5604     assert_status(status == 0, status, "cond_signal");
5605   }
5606 
5607   // Note that we signal() _after dropping the lock for "immortal" Events.
5608   // This is safe and avoids a common class of  futile wakeups.  In rare
5609   // circumstances this can cause a thread to return prematurely from
5610   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5611   // simply re-test the condition and re-park itself.
5612 }
5613 
5614 
5615 // JSR166
5616 // -------------------------------------------------------
5617 
5618 /*
5619  * The solaris and linux implementations of park/unpark are fairly
5620  * conservative for now, but can be improved. They currently use a
5621  * mutex/condvar pair, plus a a count.
5622  * Park decrements count if > 0, else does a condvar wait.  Unpark
5623  * sets count to 1 and signals condvar.  Only one thread ever waits
5624  * on the condvar. Contention seen when trying to park implies that someone
5625  * is unparking you, so don't wait. And spurious returns are fine, so there
5626  * is no need to track notifications.
5627  */
5628 
5629 /*
5630  * This code is common to linux and solaris and will be moved to a
5631  * common place in dolphin.
5632  *
5633  * The passed in time value is either a relative time in nanoseconds
5634  * or an absolute time in milliseconds. Either way it has to be unpacked
5635  * into suitable seconds and nanoseconds components and stored in the
5636  * given timespec structure.
5637  * Given time is a 64-bit value and the time_t used in the timespec is only
5638  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5639  * overflow if times way in the future are given. Further on Solaris versions
5640  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5641  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5642  * As it will be 28 years before "now + 100000000" will overflow we can
5643  * ignore overflow and just impose a hard-limit on seconds using the value
5644  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5645  * years from "now".
5646  */
5647 
5648 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5649   assert (time > 0, "convertTime");
5650   time_t max_secs = 0;
5651 
5652   if (!os::supports_monotonic_clock() || isAbsolute) {
5653     struct timeval now;
5654     int status = gettimeofday(&now, NULL);
5655     assert(status == 0, "gettimeofday");
5656 
5657     max_secs = now.tv_sec + MAX_SECS;
5658 
5659     if (isAbsolute) {
5660       jlong secs = time / 1000;
5661       if (secs > max_secs) {
5662         absTime->tv_sec = max_secs;
5663       } else {
5664         absTime->tv_sec = secs;
5665       }
5666       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5667     } else {
5668       jlong secs = time / NANOSECS_PER_SEC;
5669       if (secs >= MAX_SECS) {
5670         absTime->tv_sec = max_secs;
5671         absTime->tv_nsec = 0;
5672       } else {
5673         absTime->tv_sec = now.tv_sec + secs;
5674         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5675         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5676           absTime->tv_nsec -= NANOSECS_PER_SEC;
5677           ++absTime->tv_sec; // note: this must be <= max_secs
5678         }
5679       }
5680     }
5681   } else {
5682     // must be relative using monotonic clock
5683     struct timespec now;
5684     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5685     assert_status(status == 0, status, "clock_gettime");
5686     max_secs = now.tv_sec + MAX_SECS;
5687     jlong secs = time / NANOSECS_PER_SEC;
5688     if (secs >= MAX_SECS) {
5689       absTime->tv_sec = max_secs;
5690       absTime->tv_nsec = 0;
5691     } else {
5692       absTime->tv_sec = now.tv_sec + secs;
5693       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5694       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5695         absTime->tv_nsec -= NANOSECS_PER_SEC;
5696         ++absTime->tv_sec; // note: this must be <= max_secs
5697       }
5698     }
5699   }
5700   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5701   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5702   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5703   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5704 }
5705 
5706 void Parker::park(bool isAbsolute, jlong time) {
5707   // Ideally we'd do something useful while spinning, such
5708   // as calling unpackTime().
5709 
5710   // Optional fast-path check:
5711   // Return immediately if a permit is available.
5712   // We depend on Atomic::xchg() having full barrier semantics
5713   // since we are doing a lock-free update to _counter.
5714   if (Atomic::xchg(0, &_counter) > 0) return;
5715 
5716   Thread* thread = Thread::current();
5717   assert(thread->is_Java_thread(), "Must be JavaThread");
5718   JavaThread *jt = (JavaThread *)thread;
5719 
5720   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5721   // Check interrupt before trying to wait
5722   if (Thread::is_interrupted(thread, false)) {
5723     return;
5724   }
5725 
5726   // Next, demultiplex/decode time arguments
5727   timespec absTime;
5728   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5729     return;
5730   }
5731   if (time > 0) {
5732     unpackTime(&absTime, isAbsolute, time);
5733   }
5734 
5735 
5736   // Enter safepoint region
5737   // Beware of deadlocks such as 6317397.
5738   // The per-thread Parker:: mutex is a classic leaf-lock.
5739   // In particular a thread must never block on the Threads_lock while
5740   // holding the Parker:: mutex.  If safepoints are pending both the
5741   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5742   ThreadBlockInVM tbivm(jt);
5743 
5744   // Don't wait if cannot get lock since interference arises from
5745   // unblocking.  Also. check interrupt before trying wait
5746   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5747     return;
5748   }
5749 
5750   int status ;
5751   if (_counter > 0)  { // no wait needed
5752     _counter = 0;
5753     status = pthread_mutex_unlock(_mutex);
5754     assert (status == 0, "invariant") ;
5755     // Paranoia to ensure our locked and lock-free paths interact
5756     // correctly with each other and Java-level accesses.
5757     OrderAccess::fence();
5758     return;
5759   }
5760 
5761 #ifdef ASSERT
5762   // Don't catch signals while blocked; let the running threads have the signals.
5763   // (This allows a debugger to break into the running thread.)
5764   sigset_t oldsigs;
5765   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5766   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5767 #endif
5768 
5769   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5770   jt->set_suspend_equivalent();
5771   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5772 
5773   assert(_cur_index == -1, "invariant");
5774   if (time == 0) {
5775     _cur_index = REL_INDEX; // arbitrary choice when not timed
5776     status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5777   } else {
5778     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5779     status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5780     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5781       pthread_cond_destroy (&_cond[_cur_index]) ;
5782       pthread_cond_init    (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5783     }
5784   }
5785   _cur_index = -1;
5786   assert_status(status == 0 || status == EINTR ||
5787                 status == ETIME || status == ETIMEDOUT,
5788                 status, "cond_timedwait");
5789 
5790 #ifdef ASSERT
5791   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5792 #endif
5793 
5794   _counter = 0 ;
5795   status = pthread_mutex_unlock(_mutex) ;
5796   assert_status(status == 0, status, "invariant") ;
5797   // Paranoia to ensure our locked and lock-free paths interact
5798   // correctly with each other and Java-level accesses.
5799   OrderAccess::fence();
5800 
5801   // If externally suspended while waiting, re-suspend
5802   if (jt->handle_special_suspend_equivalent_condition()) {
5803     jt->java_suspend_self();
5804   }
5805 }
5806 
5807 void Parker::unpark() {
5808   int s, status ;
5809   status = pthread_mutex_lock(_mutex);
5810   assert (status == 0, "invariant") ;
5811   s = _counter;
5812   _counter = 1;
5813   if (s < 1) {
5814     // thread might be parked
5815     if (_cur_index != -1) {
5816       // thread is definitely parked
5817       if (WorkAroundNPTLTimedWaitHang) {
5818         status = pthread_cond_signal (&_cond[_cur_index]);
5819         assert (status == 0, "invariant");
5820         status = pthread_mutex_unlock(_mutex);
5821         assert (status == 0, "invariant");
5822       } else {
5823         status = pthread_mutex_unlock(_mutex);
5824         assert (status == 0, "invariant");
5825         status = pthread_cond_signal (&_cond[_cur_index]);
5826         assert (status == 0, "invariant");
5827       }
5828     } else {
5829       pthread_mutex_unlock(_mutex);
5830       assert (status == 0, "invariant") ;
5831     }
5832   } else {
5833     pthread_mutex_unlock(_mutex);
5834     assert (status == 0, "invariant") ;
5835   }
5836 }
5837 
5838 
5839 extern char** environ;
5840 
5841 #ifndef __NR_fork
5842 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5843 #endif
5844 
5845 #ifndef __NR_execve
5846 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5847 #endif
5848 
5849 // Run the specified command in a separate process. Return its exit value,
5850 // or -1 on failure (e.g. can't fork a new process).
5851 // Unlike system(), this function can be called from signal handler. It
5852 // doesn't block SIGINT et al.
5853 int os::fork_and_exec(char* cmd) {
5854   const char * argv[4] = {"sh", "-c", cmd, NULL};
5855 
5856   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5857   // pthread_atfork handlers and reset pthread library. All we need is a
5858   // separate process to execve. Make a direct syscall to fork process.
5859   // On IA64 there's no fork syscall, we have to use fork() and hope for
5860   // the best...
5861   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5862               IA64_ONLY(fork();)
5863 
5864   if (pid < 0) {
5865     // fork failed
5866     return -1;
5867 
5868   } else if (pid == 0) {
5869     // child process
5870 
5871     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5872     // first to kill every thread on the thread list. Because this list is
5873     // not reset by fork() (see notes above), execve() will instead kill
5874     // every thread in the parent process. We know this is the only thread
5875     // in the new process, so make a system call directly.
5876     // IA64 should use normal execve() from glibc to match the glibc fork()
5877     // above.
5878     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5879     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5880 
5881     // execve failed
5882     _exit(-1);
5883 
5884   } else  {
5885     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5886     // care about the actual exit code, for now.
5887 
5888     int status;
5889 
5890     // Wait for the child process to exit.  This returns immediately if
5891     // the child has already exited. */
5892     while (waitpid(pid, &status, 0) < 0) {
5893         switch (errno) {
5894         case ECHILD: return 0;
5895         case EINTR: break;
5896         default: return -1;
5897         }
5898     }
5899 
5900     if (WIFEXITED(status)) {
5901        // The child exited normally; get its exit code.
5902        return WEXITSTATUS(status);
5903     } else if (WIFSIGNALED(status)) {
5904        // The child exited because of a signal
5905        // The best value to return is 0x80 + signal number,
5906        // because that is what all Unix shells do, and because
5907        // it allows callers to distinguish between process exit and
5908        // process death by signal.
5909        return 0x80 + WTERMSIG(status);
5910     } else {
5911        // Unknown exit code; pass it through
5912        return status;
5913     }
5914   }
5915 }
5916 
5917 // is_headless_jre()
5918 //
5919 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5920 // in order to report if we are running in a headless jre
5921 //
5922 // Since JDK8 xawt/libmawt.so was moved into the same directory
5923 // as libawt.so, and renamed libawt_xawt.so
5924 //
5925 bool os::is_headless_jre() {
5926     struct stat statbuf;
5927     char buf[MAXPATHLEN];
5928     char libmawtpath[MAXPATHLEN];
5929     const char *xawtstr  = "/xawt/libmawt.so";
5930     const char *new_xawtstr = "/libawt_xawt.so";
5931     char *p;
5932 
5933     // Get path to libjvm.so
5934     os::jvm_path(buf, sizeof(buf));
5935 
5936     // Get rid of libjvm.so
5937     p = strrchr(buf, '/');
5938     if (p == NULL) return false;
5939     else *p = '\0';
5940 
5941     // Get rid of client or server
5942     p = strrchr(buf, '/');
5943     if (p == NULL) return false;
5944     else *p = '\0';
5945 
5946     // check xawt/libmawt.so
5947     strcpy(libmawtpath, buf);
5948     strcat(libmawtpath, xawtstr);
5949     if (::stat(libmawtpath, &statbuf) == 0) return false;
5950 
5951     // check libawt_xawt.so
5952     strcpy(libmawtpath, buf);
5953     strcat(libmawtpath, new_xawtstr);
5954     if (::stat(libmawtpath, &statbuf) == 0) return false;
5955 
5956     return true;
5957 }
5958 
5959 // Get the default path to the core file
5960 // Returns the length of the string
5961 int os::get_core_path(char* buffer, size_t bufferSize) {
5962   const char* p = get_current_directory(buffer, bufferSize);
5963 
5964   if (p == NULL) {
5965     assert(p != NULL, "failed to get current directory");
5966     return 0;
5967   }
5968 
5969   return strlen(buffer);
5970 }
5971 
5972 #ifdef JAVASE_EMBEDDED
5973 //
5974 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5975 //
5976 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5977 
5978 // ctor
5979 //
5980 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5981   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5982   _fd = fd;
5983 
5984   if (os::create_thread(this, os::os_thread)) {
5985     _memnotify_thread = this;
5986     os::set_priority(this, NearMaxPriority);
5987     os::start_thread(this);
5988   }
5989 }
5990 
5991 // Where all the work gets done
5992 //
5993 void MemNotifyThread::run() {
5994   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5995 
5996   // Set up the select arguments
5997   fd_set rfds;
5998   if (_fd != -1) {
5999     FD_ZERO(&rfds);
6000     FD_SET(_fd, &rfds);
6001   }
6002 
6003   // Now wait for the mem_notify device to wake up
6004   while (1) {
6005     // Wait for the mem_notify device to signal us..
6006     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6007     if (rc == -1) {
6008       perror("select!\n");
6009       break;
6010     } else if (rc) {
6011       //ssize_t free_before = os::available_memory();
6012       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6013 
6014       // The kernel is telling us there is not much memory left...
6015       // try to do something about that
6016 
6017       // If we are not already in a GC, try one.
6018       if (!Universe::heap()->is_gc_active()) {
6019         Universe::heap()->collect(GCCause::_allocation_failure);
6020 
6021         //ssize_t free_after = os::available_memory();
6022         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6023         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6024       }
6025       // We might want to do something like the following if we find the GC's are not helping...
6026       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6027     }
6028   }
6029 }
6030 
6031 //
6032 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6033 //
6034 void MemNotifyThread::start() {
6035   int    fd;
6036   fd = open ("/dev/mem_notify", O_RDONLY, 0);
6037   if (fd < 0) {
6038       return;
6039   }
6040 
6041   if (memnotify_thread() == NULL) {
6042     new MemNotifyThread(fd);
6043   }
6044 }
6045 
6046 #endif // JAVASE_EMBEDDED
6047 
6048 
6049 /////////////// Unit tests ///////////////
6050 
6051 #ifndef PRODUCT
6052 
6053 #define test_log(...) \
6054   do {\
6055     if (VerboseInternalVMTests) { \
6056       tty->print_cr(__VA_ARGS__); \
6057       tty->flush(); \
6058     }\
6059   } while (false)
6060 
6061 class TestReserveMemorySpecial : AllStatic {
6062  public:
6063   static void small_page_write(void* addr, size_t size) {
6064     size_t page_size = os::vm_page_size();
6065 
6066     char* end = (char*)addr + size;
6067     for (char* p = (char*)addr; p < end; p += page_size) {
6068       *p = 1;
6069     }
6070   }
6071 
6072   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6073     if (!UseHugeTLBFS) {
6074       return;
6075     }
6076 
6077     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6078 
6079     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6080 
6081     if (addr != NULL) {
6082       small_page_write(addr, size);
6083 
6084       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6085     }
6086   }
6087 
6088   static void test_reserve_memory_special_huge_tlbfs_only() {
6089     if (!UseHugeTLBFS) {
6090       return;
6091     }
6092 
6093     size_t lp = os::large_page_size();
6094 
6095     for (size_t size = lp; size <= lp * 10; size += lp) {
6096       test_reserve_memory_special_huge_tlbfs_only(size);
6097     }
6098   }
6099 
6100   static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6101     if (!UseHugeTLBFS) {
6102         return;
6103     }
6104 
6105     test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6106         size, alignment);
6107 
6108     assert(size >= os::large_page_size(), "Incorrect input to test");
6109 
6110     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6111 
6112     if (addr != NULL) {
6113       small_page_write(addr, size);
6114 
6115       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6116     }
6117   }
6118 
6119   static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6120     size_t lp = os::large_page_size();
6121     size_t ag = os::vm_allocation_granularity();
6122 
6123     for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6124       test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6125     }
6126   }
6127 
6128   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6129     size_t lp = os::large_page_size();
6130     size_t ag = os::vm_allocation_granularity();
6131 
6132     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6133     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6134     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6135     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6136     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6137     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6138     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6139     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6140     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6141   }
6142 
6143   static void test_reserve_memory_special_huge_tlbfs() {
6144     if (!UseHugeTLBFS) {
6145       return;
6146     }
6147 
6148     test_reserve_memory_special_huge_tlbfs_only();
6149     test_reserve_memory_special_huge_tlbfs_mixed();
6150   }
6151 
6152   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6153     if (!UseSHM) {
6154       return;
6155     }
6156 
6157     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6158 
6159     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6160 
6161     if (addr != NULL) {
6162       assert(is_ptr_aligned(addr, alignment), "Check");
6163       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6164 
6165       small_page_write(addr, size);
6166 
6167       os::Linux::release_memory_special_shm(addr, size);
6168     }
6169   }
6170 
6171   static void test_reserve_memory_special_shm() {
6172     size_t lp = os::large_page_size();
6173     size_t ag = os::vm_allocation_granularity();
6174 
6175     for (size_t size = ag; size < lp * 3; size += ag) {
6176       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6177         test_reserve_memory_special_shm(size, alignment);
6178       }
6179     }
6180   }
6181 
6182   static void test() {
6183     test_reserve_memory_special_huge_tlbfs();
6184     test_reserve_memory_special_shm();
6185   }
6186 };
6187 
6188 void TestReserveMemorySpecial_test() {
6189   TestReserveMemorySpecial::test();
6190 }
6191 
6192 #endif