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/orderAccess.inline.hpp"
  53 #include "runtime/osThread.hpp"
  54 #include "runtime/perfMemory.hpp"
  55 #include "runtime/sharedRuntime.hpp"
  56 #include "runtime/statSampler.hpp"
  57 #include "runtime/stubRoutines.hpp"
  58 #include "runtime/thread.inline.hpp"
  59 #include "runtime/threadCritical.hpp"
  60 #include "runtime/timer.hpp"
  61 #include "services/attachListener.hpp"
  62 #include "services/memTracker.hpp"
  63 #include "services/runtimeService.hpp"
  64 #include "utilities/decoder.hpp"
  65 #include "utilities/defaultStream.hpp"
  66 #include "utilities/events.hpp"
  67 #include "utilities/elfFile.hpp"
  68 #include "utilities/growableArray.hpp"
  69 #include "utilities/vmError.hpp"
  70 
  71 // put OS-includes here
  72 # include <sys/types.h>
  73 # include <sys/mman.h>
  74 # include <sys/stat.h>
  75 # include <sys/select.h>
  76 # include <pthread.h>
  77 # include <signal.h>
  78 # include <errno.h>
  79 # include <dlfcn.h>
  80 # include <stdio.h>
  81 # include <unistd.h>
  82 # include <sys/resource.h>
  83 # include <pthread.h>
  84 # include <sys/stat.h>
  85 # include <sys/time.h>
  86 # include <sys/times.h>
  87 # include <sys/utsname.h>
  88 # include <sys/socket.h>
  89 # include <sys/wait.h>
  90 # include <pwd.h>
  91 # include <poll.h>
  92 # include <semaphore.h>
  93 # include <fcntl.h>
  94 # include <string.h>
  95 # include <syscall.h>
  96 # include <sys/sysinfo.h>
  97 # include <gnu/libc-version.h>
  98 # include <sys/ipc.h>
  99 # include <sys/shm.h>
 100 # include <link.h>
 101 # include <stdint.h>
 102 # include <inttypes.h>
 103 # include <sys/ioctl.h>
 104 
 105 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
 106 
 107 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 108 // getrusage() is prepared to handle the associated failure.
 109 #ifndef RUSAGE_THREAD
 110 #define RUSAGE_THREAD   (1)               /* only the calling thread */
 111 #endif
 112 
 113 #define MAX_PATH    (2 * K)
 114 
 115 #define MAX_SECS 100000000
 116 
 117 // for timer info max values which include all bits
 118 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 119 
 120 #define LARGEPAGES_BIT (1 << 6)
 121 ////////////////////////////////////////////////////////////////////////////////
 122 // global variables
 123 julong os::Linux::_physical_memory = 0;
 124 
 125 address   os::Linux::_initial_thread_stack_bottom = NULL;
 126 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 127 
 128 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 129 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 130 Mutex* os::Linux::_createThread_lock = NULL;
 131 pthread_t os::Linux::_main_thread;
 132 int os::Linux::_page_size = -1;
 133 const int os::Linux::_vm_default_page_size = (8 * K);
 134 bool os::Linux::_is_floating_stack = false;
 135 bool os::Linux::_is_NPTL = false;
 136 bool os::Linux::_supports_fast_thread_cpu_time = false;
 137 const char * os::Linux::_glibc_version = NULL;
 138 const char * os::Linux::_libpthread_version = NULL;
 139 pthread_condattr_t os::Linux::_condattr[1];
 140 
 141 static jlong initial_time_count=0;
 142 
 143 static int clock_tics_per_sec = 100;
 144 
 145 // For diagnostics to print a message once. see run_periodic_checks
 146 static sigset_t check_signal_done;
 147 static bool check_signals = true;
 148 
 149 static pid_t _initial_pid = 0;
 150 
 151 /* Signal number used to suspend/resume a thread */
 152 
 153 /* do not use any signal number less than SIGSEGV, see 4355769 */
 154 static int SR_signum = SIGUSR2;
 155 sigset_t SR_sigset;
 156 
 157 /* Used to protect dlsym() calls */
 158 static pthread_mutex_t dl_mutex;
 159 
 160 // Declarations
 161 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 162 
 163 #ifdef JAVASE_EMBEDDED
 164 class MemNotifyThread: public Thread {
 165   friend class VMStructs;
 166  public:
 167   virtual void run();
 168 
 169  private:
 170   static MemNotifyThread* _memnotify_thread;
 171   int _fd;
 172 
 173  public:
 174 
 175   // Constructor
 176   MemNotifyThread(int fd);
 177 
 178   // Tester
 179   bool is_memnotify_thread() const { return true; }
 180 
 181   // Printing
 182   char* name() const { return (char*)"Linux MemNotify Thread"; }
 183 
 184   // Returns the single instance of the MemNotifyThread
 185   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 186 
 187   // Create and start the single instance of MemNotifyThread
 188   static void start();
 189 };
 190 #endif // JAVASE_EMBEDDED
 191 
 192 // utility functions
 193 
 194 static int SR_initialize();
 195 
 196 julong os::available_memory() {
 197   return Linux::available_memory();
 198 }
 199 
 200 julong os::Linux::available_memory() {
 201   // values in struct sysinfo are "unsigned long"
 202   struct sysinfo si;
 203   sysinfo(&si);
 204 
 205   return (julong)si.freeram * si.mem_unit;
 206 }
 207 
 208 julong os::physical_memory() {
 209   return Linux::physical_memory();
 210 }
 211 
 212 ////////////////////////////////////////////////////////////////////////////////
 213 // environment support
 214 
 215 bool os::getenv(const char* name, char* buf, int len) {
 216   const char* val = ::getenv(name);
 217   if (val != NULL && strlen(val) < (size_t)len) {
 218     strcpy(buf, val);
 219     return true;
 220   }
 221   if (len > 0) buf[0] = 0;  // return a null string
 222   return false;
 223 }
 224 
 225 
 226 // Return true if user is running as root.
 227 
 228 bool os::have_special_privileges() {
 229   static bool init = false;
 230   static bool privileges = false;
 231   if (!init) {
 232     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 233     init = true;
 234   }
 235   return privileges;
 236 }
 237 
 238 
 239 #ifndef SYS_gettid
 240 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 241 #ifdef __ia64__
 242 #define SYS_gettid 1105
 243 #elif __i386__
 244 #define SYS_gettid 224
 245 #elif __amd64__
 246 #define SYS_gettid 186
 247 #elif __sparc__
 248 #define SYS_gettid 143
 249 #else
 250 #error define gettid for the arch
 251 #endif
 252 #endif
 253 
 254 // Cpu architecture string
 255 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 256 
 257 // pid_t gettid()
 258 //
 259 // Returns the kernel thread id of the currently running thread. Kernel
 260 // thread id is used to access /proc.
 261 //
 262 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 263 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 264 //
 265 pid_t os::Linux::gettid() {
 266   int rslt = syscall(SYS_gettid);
 267   if (rslt == -1) {
 268      // old kernel, no NPTL support
 269      return getpid();
 270   } else {
 271      return (pid_t)rslt;
 272   }
 273 }
 274 
 275 // Most versions of linux have a bug where the number of processors are
 276 // determined by looking at the /proc file system.  In a chroot environment,
 277 // the system call returns 1.  This causes the VM to act as if it is
 278 // a single processor and elide locking (see is_MP() call).
 279 static bool unsafe_chroot_detected = false;
 280 static const char *unstable_chroot_error = "/proc file system not found.\n"
 281                      "Java may be unstable running multithreaded in a chroot "
 282                      "environment on Linux when /proc filesystem is not mounted.";
 283 
 284 void os::Linux::initialize_system_info() {
 285   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 286   if (processor_count() == 1) {
 287     pid_t pid = os::Linux::gettid();
 288     char fname[32];
 289     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 290     FILE *fp = fopen(fname, "r");
 291     if (fp == NULL) {
 292       unsafe_chroot_detected = true;
 293     } else {
 294       fclose(fp);
 295     }
 296   }
 297   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 298   assert(processor_count() > 0, "linux error");
 299 }
 300 
 301 void os::init_system_properties_values() {
 302   // The next steps are taken in the product version:
 303   //
 304   // Obtain the JAVA_HOME value from the location of libjvm.so.
 305   // This library should be located at:
 306   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 307   //
 308   // If "/jre/lib/" appears at the right place in the path, then we
 309   // assume libjvm.so is installed in a JDK and we use this path.
 310   //
 311   // Otherwise exit with message: "Could not create the Java virtual machine."
 312   //
 313   // The following extra steps are taken in the debugging version:
 314   //
 315   // If "/jre/lib/" does NOT appear at the right place in the path
 316   // instead of exit check for $JAVA_HOME environment variable.
 317   //
 318   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 319   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 320   // it looks like libjvm.so is installed there
 321   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 322   //
 323   // Otherwise exit.
 324   //
 325   // Important note: if the location of libjvm.so changes this
 326   // code needs to be changed accordingly.
 327 
 328 // See ld(1):
 329 //      The linker uses the following search paths to locate required
 330 //      shared libraries:
 331 //        1: ...
 332 //        ...
 333 //        7: The default directories, normally /lib and /usr/lib.
 334 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 335 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 336 #else
 337 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 338 #endif
 339 
 340 // Base path of extensions installed on the system.
 341 #define SYS_EXT_DIR     "/usr/java/packages"
 342 #define EXTENSIONS_DIR  "/lib/ext"
 343 #define ENDORSED_DIR    "/lib/endorsed"
 344 
 345   // Buffer that fits several sprintfs.
 346   // Note that the space for the colon and the trailing null are provided
 347   // by the nulls included by the sizeof operator.
 348   const size_t bufsize =
 349     MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
 350          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
 351          (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
 352   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 353 
 354   // sysclasspath, java_home, dll_dir
 355   {
 356     char *pslash;
 357     os::jvm_path(buf, bufsize);
 358 
 359     // Found the full path to libjvm.so.
 360     // Now cut the path to <java_home>/jre if we can.
 361     *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
 362     pslash = strrchr(buf, '/');
 363     if (pslash != NULL) {
 364       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 365     }
 366     Arguments::set_dll_dir(buf);
 367 
 368     if (pslash != NULL) {
 369       pslash = strrchr(buf, '/');
 370       if (pslash != NULL) {
 371         *pslash = '\0';          // Get rid of /<arch>.
 372         pslash = strrchr(buf, '/');
 373         if (pslash != NULL) {
 374           *pslash = '\0';        // Get rid of /lib.
 375         }
 376       }
 377     }
 378     Arguments::set_java_home(buf);
 379     set_boot_path('/', ':');
 380   }
 381 
 382   // Where to look for native libraries.
 383   //
 384   // Note: Due to a legacy implementation, most of the library path
 385   // is set in the launcher. This was to accomodate linking restrictions
 386   // on legacy Linux implementations (which are no longer supported).
 387   // Eventually, all the library path setting will be done here.
 388   //
 389   // However, to prevent the proliferation of improperly built native
 390   // libraries, the new path component /usr/java/packages is added here.
 391   // Eventually, all the library path setting will be done here.
 392   {
 393     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 394     // should always exist (until the legacy problem cited above is
 395     // addressed).
 396     const char *v = ::getenv("LD_LIBRARY_PATH");
 397     const char *v_colon = ":";
 398     if (v == NULL) { v = ""; v_colon = ""; }
 399     // That's +1 for the colon and +1 for the trailing '\0'.
 400     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 401                                                      strlen(v) + 1 +
 402                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 403                                                      mtInternal);
 404     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 405     Arguments::set_library_path(ld_library_path);
 406     FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
 407   }
 408 
 409   // Extensions directories.
 410   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 411   Arguments::set_ext_dirs(buf);
 412 
 413   // Endorsed standards default directory.
 414   sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 415   Arguments::set_endorsed_dirs(buf);
 416 
 417   FREE_C_HEAP_ARRAY(char, buf, mtInternal);
 418 
 419 #undef DEFAULT_LIBPATH
 420 #undef SYS_EXT_DIR
 421 #undef EXTENSIONS_DIR
 422 #undef ENDORSED_DIR
 423 }
 424 
 425 ////////////////////////////////////////////////////////////////////////////////
 426 // breakpoint support
 427 
 428 void os::breakpoint() {
 429   BREAKPOINT;
 430 }
 431 
 432 extern "C" void breakpoint() {
 433   // use debugger to set breakpoint here
 434 }
 435 
 436 ////////////////////////////////////////////////////////////////////////////////
 437 // signal support
 438 
 439 debug_only(static bool signal_sets_initialized = false);
 440 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 441 
 442 bool os::Linux::is_sig_ignored(int sig) {
 443       struct sigaction oact;
 444       sigaction(sig, (struct sigaction*)NULL, &oact);
 445       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 446                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 447       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 448            return true;
 449       else
 450            return false;
 451 }
 452 
 453 void os::Linux::signal_sets_init() {
 454   // Should also have an assertion stating we are still single-threaded.
 455   assert(!signal_sets_initialized, "Already initialized");
 456   // Fill in signals that are necessarily unblocked for all threads in
 457   // the VM. Currently, we unblock the following signals:
 458   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 459   //                         by -Xrs (=ReduceSignalUsage));
 460   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 461   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 462   // the dispositions or masks wrt these signals.
 463   // Programs embedding the VM that want to use the above signals for their
 464   // own purposes must, at this time, use the "-Xrs" option to prevent
 465   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 466   // (See bug 4345157, and other related bugs).
 467   // In reality, though, unblocking these signals is really a nop, since
 468   // these signals are not blocked by default.
 469   sigemptyset(&unblocked_sigs);
 470   sigemptyset(&allowdebug_blocked_sigs);
 471   sigaddset(&unblocked_sigs, SIGILL);
 472   sigaddset(&unblocked_sigs, SIGSEGV);
 473   sigaddset(&unblocked_sigs, SIGBUS);
 474   sigaddset(&unblocked_sigs, SIGFPE);
 475 #if defined(PPC64)
 476   sigaddset(&unblocked_sigs, SIGTRAP);
 477 #endif
 478   sigaddset(&unblocked_sigs, SR_signum);
 479 
 480   if (!ReduceSignalUsage) {
 481    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 482       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 483       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 484    }
 485    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 486       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 487       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 488    }
 489    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 490       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 491       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 492    }
 493   }
 494   // Fill in signals that are blocked by all but the VM thread.
 495   sigemptyset(&vm_sigs);
 496   if (!ReduceSignalUsage)
 497     sigaddset(&vm_sigs, BREAK_SIGNAL);
 498   debug_only(signal_sets_initialized = true);
 499 
 500 }
 501 
 502 // These are signals that are unblocked while a thread is running Java.
 503 // (For some reason, they get blocked by default.)
 504 sigset_t* os::Linux::unblocked_signals() {
 505   assert(signal_sets_initialized, "Not initialized");
 506   return &unblocked_sigs;
 507 }
 508 
 509 // These are the signals that are blocked while a (non-VM) thread is
 510 // running Java. Only the VM thread handles these signals.
 511 sigset_t* os::Linux::vm_signals() {
 512   assert(signal_sets_initialized, "Not initialized");
 513   return &vm_sigs;
 514 }
 515 
 516 // These are signals that are blocked during cond_wait to allow debugger in
 517 sigset_t* os::Linux::allowdebug_blocked_signals() {
 518   assert(signal_sets_initialized, "Not initialized");
 519   return &allowdebug_blocked_sigs;
 520 }
 521 
 522 void os::Linux::hotspot_sigmask(Thread* thread) {
 523 
 524   //Save caller's signal mask before setting VM signal mask
 525   sigset_t caller_sigmask;
 526   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 527 
 528   OSThread* osthread = thread->osthread();
 529   osthread->set_caller_sigmask(caller_sigmask);
 530 
 531   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 532 
 533   if (!ReduceSignalUsage) {
 534     if (thread->is_VM_thread()) {
 535       // Only the VM thread handles BREAK_SIGNAL ...
 536       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 537     } else {
 538       // ... all other threads block BREAK_SIGNAL
 539       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 540     }
 541   }
 542 }
 543 
 544 //////////////////////////////////////////////////////////////////////////////
 545 // detecting pthread library
 546 
 547 void os::Linux::libpthread_init() {
 548   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 549   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 550   // generic name for earlier versions.
 551   // Define macros here so we can build HotSpot on old systems.
 552 # ifndef _CS_GNU_LIBC_VERSION
 553 # define _CS_GNU_LIBC_VERSION 2
 554 # endif
 555 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 556 # define _CS_GNU_LIBPTHREAD_VERSION 3
 557 # endif
 558 
 559   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 560   if (n > 0) {
 561      char *str = (char *)malloc(n, mtInternal);
 562      confstr(_CS_GNU_LIBC_VERSION, str, n);
 563      os::Linux::set_glibc_version(str);
 564   } else {
 565      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 566      static char _gnu_libc_version[32];
 567      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 568               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 569      os::Linux::set_glibc_version(_gnu_libc_version);
 570   }
 571 
 572   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 573   if (n > 0) {
 574      char *str = (char *)malloc(n, mtInternal);
 575      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 576      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 577      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 578      // is the case. LinuxThreads has a hard limit on max number of threads.
 579      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 580      // On the other hand, NPTL does not have such a limit, sysconf()
 581      // will return -1 and errno is not changed. Check if it is really NPTL.
 582      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 583          strstr(str, "NPTL") &&
 584          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 585        free(str);
 586        os::Linux::set_libpthread_version("linuxthreads");
 587      } else {
 588        os::Linux::set_libpthread_version(str);
 589      }
 590   } else {
 591     // glibc before 2.3.2 only has LinuxThreads.
 592     os::Linux::set_libpthread_version("linuxthreads");
 593   }
 594 
 595   if (strstr(libpthread_version(), "NPTL")) {
 596      os::Linux::set_is_NPTL();
 597   } else {
 598      os::Linux::set_is_LinuxThreads();
 599   }
 600 
 601   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 602   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 603   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 604      os::Linux::set_is_floating_stack();
 605   }
 606 }
 607 
 608 /////////////////////////////////////////////////////////////////////////////
 609 // thread stack
 610 
 611 // Force Linux kernel to expand current thread stack. If "bottom" is close
 612 // to the stack guard, caller should block all signals.
 613 //
 614 // MAP_GROWSDOWN:
 615 //   A special mmap() flag that is used to implement thread stacks. It tells
 616 //   kernel that the memory region should extend downwards when needed. This
 617 //   allows early versions of LinuxThreads to only mmap the first few pages
 618 //   when creating a new thread. Linux kernel will automatically expand thread
 619 //   stack as needed (on page faults).
 620 //
 621 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 622 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 623 //   region, it's hard to tell if the fault is due to a legitimate stack
 624 //   access or because of reading/writing non-exist memory (e.g. buffer
 625 //   overrun). As a rule, if the fault happens below current stack pointer,
 626 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 627 //   application (see Linux kernel fault.c).
 628 //
 629 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 630 //   stack overflow detection.
 631 //
 632 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 633 //   not use this flag. However, the stack of initial thread is not created
 634 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 635 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 636 //   and then attach the thread to JVM.
 637 //
 638 // To get around the problem and allow stack banging on Linux, we need to
 639 // manually expand thread stack after receiving the SIGSEGV.
 640 //
 641 // There are two ways to expand thread stack to address "bottom", we used
 642 // both of them in JVM before 1.5:
 643 //   1. adjust stack pointer first so that it is below "bottom", and then
 644 //      touch "bottom"
 645 //   2. mmap() the page in question
 646 //
 647 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 648 // if current sp is already near the lower end of page 101, and we need to
 649 // call mmap() to map page 100, it is possible that part of the mmap() frame
 650 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 651 // That will destroy the mmap() frame and cause VM to crash.
 652 //
 653 // The following code works by adjusting sp first, then accessing the "bottom"
 654 // page to force a page fault. Linux kernel will then automatically expand the
 655 // stack mapping.
 656 //
 657 // _expand_stack_to() assumes its frame size is less than page size, which
 658 // should always be true if the function is not inlined.
 659 
 660 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 661 #define NOINLINE
 662 #else
 663 #define NOINLINE __attribute__ ((noinline))
 664 #endif
 665 
 666 static void _expand_stack_to(address bottom) NOINLINE;
 667 
 668 static void _expand_stack_to(address bottom) {
 669   address sp;
 670   size_t size;
 671   volatile char *p;
 672 
 673   // Adjust bottom to point to the largest address within the same page, it
 674   // gives us a one-page buffer if alloca() allocates slightly more memory.
 675   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 676   bottom += os::Linux::page_size() - 1;
 677 
 678   // sp might be slightly above current stack pointer; if that's the case, we
 679   // will alloca() a little more space than necessary, which is OK. Don't use
 680   // os::current_stack_pointer(), as its result can be slightly below current
 681   // stack pointer, causing us to not alloca enough to reach "bottom".
 682   sp = (address)&sp;
 683 
 684   if (sp > bottom) {
 685     size = sp - bottom;
 686     p = (volatile char *)alloca(size);
 687     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 688     p[0] = '\0';
 689   }
 690 }
 691 
 692 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 693   assert(t!=NULL, "just checking");
 694   assert(t->osthread()->expanding_stack(), "expand should be set");
 695   assert(t->stack_base() != NULL, "stack_base was not initialized");
 696 
 697   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 698     sigset_t mask_all, old_sigset;
 699     sigfillset(&mask_all);
 700     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 701     _expand_stack_to(addr);
 702     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 703     return true;
 704   }
 705   return false;
 706 }
 707 
 708 //////////////////////////////////////////////////////////////////////////////
 709 // create new thread
 710 
 711 static address highest_vm_reserved_address();
 712 
 713 // check if it's safe to start a new thread
 714 static bool _thread_safety_check(Thread* thread) {
 715   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 716     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 717     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 718     //   allocated (MAP_FIXED) from high address space. Every thread stack
 719     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 720     //   it to other values if they rebuild LinuxThreads).
 721     //
 722     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 723     // the memory region has already been mmap'ed. That means if we have too
 724     // many threads and/or very large heap, eventually thread stack will
 725     // collide with heap.
 726     //
 727     // Here we try to prevent heap/stack collision by comparing current
 728     // stack bottom with the highest address that has been mmap'ed by JVM
 729     // plus a safety margin for memory maps created by native code.
 730     //
 731     // This feature can be disabled by setting ThreadSafetyMargin to 0
 732     //
 733     if (ThreadSafetyMargin > 0) {
 734       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 735 
 736       // not safe if our stack extends below the safety margin
 737       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 738     } else {
 739       return true;
 740     }
 741   } else {
 742     // Floating stack LinuxThreads or NPTL:
 743     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 744     //   there's not enough space left, pthread_create() will fail. If we come
 745     //   here, that means enough space has been reserved for stack.
 746     return true;
 747   }
 748 }
 749 
 750 // Thread start routine for all newly created threads
 751 static void *java_start(Thread *thread) {
 752   // Try to randomize the cache line index of hot stack frames.
 753   // This helps when threads of the same stack traces evict each other's
 754   // cache lines. The threads can be either from the same JVM instance, or
 755   // from different JVM instances. The benefit is especially true for
 756   // processors with hyperthreading technology.
 757   static int counter = 0;
 758   int pid = os::current_process_id();
 759   alloca(((pid ^ counter++) & 7) * 128);
 760 
 761   ThreadLocalStorage::set_thread(thread);
 762 
 763   OSThread* osthread = thread->osthread();
 764   Monitor* sync = osthread->startThread_lock();
 765 
 766   // non floating stack LinuxThreads needs extra check, see above
 767   if (!_thread_safety_check(thread)) {
 768     // notify parent thread
 769     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 770     osthread->set_state(ZOMBIE);
 771     sync->notify_all();
 772     return NULL;
 773   }
 774 
 775   // thread_id is kernel thread id (similar to Solaris LWP id)
 776   osthread->set_thread_id(os::Linux::gettid());
 777 
 778   if (UseNUMA) {
 779     int lgrp_id = os::numa_get_group_id();
 780     if (lgrp_id != -1) {
 781       thread->set_lgrp_id(lgrp_id);
 782     }
 783   }
 784   // initialize signal mask for this thread
 785   os::Linux::hotspot_sigmask(thread);
 786 
 787   // initialize floating point control register
 788   os::Linux::init_thread_fpu_state();
 789 
 790   // handshaking with parent thread
 791   {
 792     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 793 
 794     // notify parent thread
 795     osthread->set_state(INITIALIZED);
 796     sync->notify_all();
 797 
 798     // wait until os::start_thread()
 799     while (osthread->get_state() == INITIALIZED) {
 800       sync->wait(Mutex::_no_safepoint_check_flag);
 801     }
 802   }
 803 
 804   // call one more level start routine
 805   thread->run();
 806 
 807   return 0;
 808 }
 809 
 810 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 811   assert(thread->osthread() == NULL, "caller responsible");
 812 
 813   // Allocate the OSThread object
 814   OSThread* osthread = new OSThread(NULL, NULL);
 815   if (osthread == NULL) {
 816     return false;
 817   }
 818 
 819   // set the correct thread state
 820   osthread->set_thread_type(thr_type);
 821 
 822   // Initial state is ALLOCATED but not INITIALIZED
 823   osthread->set_state(ALLOCATED);
 824 
 825   thread->set_osthread(osthread);
 826 
 827   // init thread attributes
 828   pthread_attr_t attr;
 829   pthread_attr_init(&attr);
 830   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 831 
 832   // stack size
 833   if (os::Linux::supports_variable_stack_size()) {
 834     // calculate stack size if it's not specified by caller
 835     if (stack_size == 0) {
 836       stack_size = os::Linux::default_stack_size(thr_type);
 837 
 838       switch (thr_type) {
 839       case os::java_thread:
 840         // Java threads use ThreadStackSize which default value can be
 841         // changed with the flag -Xss
 842         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 843         stack_size = JavaThread::stack_size_at_create();
 844         break;
 845       case os::compiler_thread:
 846         if (CompilerThreadStackSize > 0) {
 847           stack_size = (size_t)(CompilerThreadStackSize * K);
 848           break;
 849         } // else fall through:
 850           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 851       case os::vm_thread:
 852       case os::pgc_thread:
 853       case os::cgc_thread:
 854       case os::watcher_thread:
 855         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 856         break;
 857       }
 858     }
 859 
 860     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 861     pthread_attr_setstacksize(&attr, stack_size);
 862   } else {
 863     // let pthread_create() pick the default value.
 864   }
 865 
 866   // glibc guard page
 867   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 868 
 869   ThreadState state;
 870 
 871   {
 872     // Serialize thread creation if we are running with fixed stack LinuxThreads
 873     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 874     if (lock) {
 875       os::Linux::createThread_lock()->lock_without_safepoint_check();
 876     }
 877 
 878     pthread_t tid;
 879     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 880 
 881     pthread_attr_destroy(&attr);
 882 
 883     if (ret != 0) {
 884       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 885         perror("pthread_create()");
 886       }
 887       // Need to clean up stuff we've allocated so far
 888       thread->set_osthread(NULL);
 889       delete osthread;
 890       if (lock) os::Linux::createThread_lock()->unlock();
 891       return false;
 892     }
 893 
 894     // Store pthread info into the OSThread
 895     osthread->set_pthread_id(tid);
 896 
 897     // Wait until child thread is either initialized or aborted
 898     {
 899       Monitor* sync_with_child = osthread->startThread_lock();
 900       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 901       while ((state = osthread->get_state()) == ALLOCATED) {
 902         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 903       }
 904     }
 905 
 906     if (lock) {
 907       os::Linux::createThread_lock()->unlock();
 908     }
 909   }
 910 
 911   // Aborted due to thread limit being reached
 912   if (state == ZOMBIE) {
 913       thread->set_osthread(NULL);
 914       delete osthread;
 915       return false;
 916   }
 917 
 918   // The thread is returned suspended (in state INITIALIZED),
 919   // and is started higher up in the call chain
 920   assert(state == INITIALIZED, "race condition");
 921   return true;
 922 }
 923 
 924 /////////////////////////////////////////////////////////////////////////////
 925 // attach existing thread
 926 
 927 // bootstrap the main thread
 928 bool os::create_main_thread(JavaThread* thread) {
 929   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 930   return create_attached_thread(thread);
 931 }
 932 
 933 bool os::create_attached_thread(JavaThread* thread) {
 934 #ifdef ASSERT
 935     thread->verify_not_published();
 936 #endif
 937 
 938   // Allocate the OSThread object
 939   OSThread* osthread = new OSThread(NULL, NULL);
 940 
 941   if (osthread == NULL) {
 942     return false;
 943   }
 944 
 945   // Store pthread info into the OSThread
 946   osthread->set_thread_id(os::Linux::gettid());
 947   osthread->set_pthread_id(::pthread_self());
 948 
 949   // initialize floating point control register
 950   os::Linux::init_thread_fpu_state();
 951 
 952   // Initial thread state is RUNNABLE
 953   osthread->set_state(RUNNABLE);
 954 
 955   thread->set_osthread(osthread);
 956 
 957   if (UseNUMA) {
 958     int lgrp_id = os::numa_get_group_id();
 959     if (lgrp_id != -1) {
 960       thread->set_lgrp_id(lgrp_id);
 961     }
 962   }
 963 
 964   if (os::Linux::is_initial_thread()) {
 965     // If current thread is initial thread, its stack is mapped on demand,
 966     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 967     // the entire stack region to avoid SEGV in stack banging.
 968     // It is also useful to get around the heap-stack-gap problem on SuSE
 969     // kernel (see 4821821 for details). We first expand stack to the top
 970     // of yellow zone, then enable stack yellow zone (order is significant,
 971     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 972     // is no gap between the last two virtual memory regions.
 973 
 974     JavaThread *jt = (JavaThread *)thread;
 975     address addr = jt->stack_yellow_zone_base();
 976     assert(addr != NULL, "initialization problem?");
 977     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 978 
 979     osthread->set_expanding_stack();
 980     os::Linux::manually_expand_stack(jt, addr);
 981     osthread->clear_expanding_stack();
 982   }
 983 
 984   // initialize signal mask for this thread
 985   // and save the caller's signal mask
 986   os::Linux::hotspot_sigmask(thread);
 987 
 988   return true;
 989 }
 990 
 991 void os::pd_start_thread(Thread* thread) {
 992   OSThread * osthread = thread->osthread();
 993   assert(osthread->get_state() != INITIALIZED, "just checking");
 994   Monitor* sync_with_child = osthread->startThread_lock();
 995   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 996   sync_with_child->notify();
 997 }
 998 
 999 // Free Linux resources related to the OSThread
1000 void os::free_thread(OSThread* osthread) {
1001   assert(osthread != NULL, "osthread not set");
1002 
1003   if (Thread::current()->osthread() == osthread) {
1004     // Restore caller's signal mask
1005     sigset_t sigmask = osthread->caller_sigmask();
1006     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1007    }
1008 
1009   delete osthread;
1010 }
1011 
1012 //////////////////////////////////////////////////////////////////////////////
1013 // thread local storage
1014 
1015 // Restore the thread pointer if the destructor is called. This is in case
1016 // someone from JNI code sets up a destructor with pthread_key_create to run
1017 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1018 // will hang or crash. When detachCurrentThread is called the key will be set
1019 // to null and we will not be called again. If detachCurrentThread is never
1020 // called we could loop forever depending on the pthread implementation.
1021 static void restore_thread_pointer(void* p) {
1022   Thread* thread = (Thread*) p;
1023   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1024 }
1025 
1026 int os::allocate_thread_local_storage() {
1027   pthread_key_t key;
1028   int rslt = pthread_key_create(&key, restore_thread_pointer);
1029   assert(rslt == 0, "cannot allocate thread local storage");
1030   return (int)key;
1031 }
1032 
1033 // Note: This is currently not used by VM, as we don't destroy TLS key
1034 // on VM exit.
1035 void os::free_thread_local_storage(int index) {
1036   int rslt = pthread_key_delete((pthread_key_t)index);
1037   assert(rslt == 0, "invalid index");
1038 }
1039 
1040 void os::thread_local_storage_at_put(int index, void* value) {
1041   int rslt = pthread_setspecific((pthread_key_t)index, value);
1042   assert(rslt == 0, "pthread_setspecific failed");
1043 }
1044 
1045 extern "C" Thread* get_thread() {
1046   return ThreadLocalStorage::thread();
1047 }
1048 
1049 //////////////////////////////////////////////////////////////////////////////
1050 // initial thread
1051 
1052 // Check if current thread is the initial thread, similar to Solaris thr_main.
1053 bool os::Linux::is_initial_thread(void) {
1054   char dummy;
1055   // If called before init complete, thread stack bottom will be null.
1056   // Can be called if fatal error occurs before initialization.
1057   if (initial_thread_stack_bottom() == NULL) return false;
1058   assert(initial_thread_stack_bottom() != NULL &&
1059          initial_thread_stack_size()   != 0,
1060          "os::init did not locate initial thread's stack region");
1061   if ((address)&dummy >= initial_thread_stack_bottom() &&
1062       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1063        return true;
1064   else return false;
1065 }
1066 
1067 // Find the virtual memory area that contains addr
1068 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1069   FILE *fp = fopen("/proc/self/maps", "r");
1070   if (fp) {
1071     address low, high;
1072     while (!feof(fp)) {
1073       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1074         if (low <= addr && addr < high) {
1075            if (vma_low)  *vma_low  = low;
1076            if (vma_high) *vma_high = high;
1077            fclose (fp);
1078            return true;
1079         }
1080       }
1081       for (;;) {
1082         int ch = fgetc(fp);
1083         if (ch == EOF || ch == (int)'\n') break;
1084       }
1085     }
1086     fclose(fp);
1087   }
1088   return false;
1089 }
1090 
1091 // Locate initial thread stack. This special handling of initial thread stack
1092 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1093 // bogus value for initial thread.
1094 void os::Linux::capture_initial_stack(size_t max_size) {
1095   // stack size is the easy part, get it from RLIMIT_STACK
1096   size_t stack_size;
1097   struct rlimit rlim;
1098   getrlimit(RLIMIT_STACK, &rlim);
1099   stack_size = rlim.rlim_cur;
1100 
1101   // 6308388: a bug in ld.so will relocate its own .data section to the
1102   //   lower end of primordial stack; reduce ulimit -s value a little bit
1103   //   so we won't install guard page on ld.so's data section.
1104   stack_size -= 2 * page_size();
1105 
1106   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1107   //   7.1, in both cases we will get 2G in return value.
1108   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1109   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1110   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1111   //   in case other parts in glibc still assumes 2M max stack size.
1112   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1113   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1114   if (stack_size > 2 * K * K IA64_ONLY(*2))
1115       stack_size = 2 * K * K IA64_ONLY(*2);
1116   // Try to figure out where the stack base (top) is. This is harder.
1117   //
1118   // When an application is started, glibc saves the initial stack pointer in
1119   // a global variable "__libc_stack_end", which is then used by system
1120   // libraries. __libc_stack_end should be pretty close to stack top. The
1121   // variable is available since the very early days. However, because it is
1122   // a private interface, it could disappear in the future.
1123   //
1124   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1125   // to __libc_stack_end, it is very close to stack top, but isn't the real
1126   // stack top. Note that /proc may not exist if VM is running as a chroot
1127   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1128   // /proc/<pid>/stat could change in the future (though unlikely).
1129   //
1130   // We try __libc_stack_end first. If that doesn't work, look for
1131   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1132   // as a hint, which should work well in most cases.
1133 
1134   uintptr_t stack_start;
1135 
1136   // try __libc_stack_end first
1137   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1138   if (p && *p) {
1139     stack_start = *p;
1140   } else {
1141     // see if we can get the start_stack field from /proc/self/stat
1142     FILE *fp;
1143     int pid;
1144     char state;
1145     int ppid;
1146     int pgrp;
1147     int session;
1148     int nr;
1149     int tpgrp;
1150     unsigned long flags;
1151     unsigned long minflt;
1152     unsigned long cminflt;
1153     unsigned long majflt;
1154     unsigned long cmajflt;
1155     unsigned long utime;
1156     unsigned long stime;
1157     long cutime;
1158     long cstime;
1159     long prio;
1160     long nice;
1161     long junk;
1162     long it_real;
1163     uintptr_t start;
1164     uintptr_t vsize;
1165     intptr_t rss;
1166     uintptr_t rsslim;
1167     uintptr_t scodes;
1168     uintptr_t ecode;
1169     int i;
1170 
1171     // Figure what the primordial thread stack base is. Code is inspired
1172     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1173     // followed by command name surrounded by parentheses, state, etc.
1174     char stat[2048];
1175     int statlen;
1176 
1177     fp = fopen("/proc/self/stat", "r");
1178     if (fp) {
1179       statlen = fread(stat, 1, 2047, fp);
1180       stat[statlen] = '\0';
1181       fclose(fp);
1182 
1183       // Skip pid and the command string. Note that we could be dealing with
1184       // weird command names, e.g. user could decide to rename java launcher
1185       // to "java 1.4.2 :)", then the stat file would look like
1186       //                1234 (java 1.4.2 :)) R ... ...
1187       // We don't really need to know the command string, just find the last
1188       // occurrence of ")" and then start parsing from there. See bug 4726580.
1189       char * s = strrchr(stat, ')');
1190 
1191       i = 0;
1192       if (s) {
1193         // Skip blank chars
1194         do s++; while (isspace(*s));
1195 
1196 #define _UFM UINTX_FORMAT
1197 #define _DFM INTX_FORMAT
1198 
1199         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1200         /*              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 */
1201         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,
1202              &state,          /* 3  %c  */
1203              &ppid,           /* 4  %d  */
1204              &pgrp,           /* 5  %d  */
1205              &session,        /* 6  %d  */
1206              &nr,             /* 7  %d  */
1207              &tpgrp,          /* 8  %d  */
1208              &flags,          /* 9  %lu  */
1209              &minflt,         /* 10 %lu  */
1210              &cminflt,        /* 11 %lu  */
1211              &majflt,         /* 12 %lu  */
1212              &cmajflt,        /* 13 %lu  */
1213              &utime,          /* 14 %lu  */
1214              &stime,          /* 15 %lu  */
1215              &cutime,         /* 16 %ld  */
1216              &cstime,         /* 17 %ld  */
1217              &prio,           /* 18 %ld  */
1218              &nice,           /* 19 %ld  */
1219              &junk,           /* 20 %ld  */
1220              &it_real,        /* 21 %ld  */
1221              &start,          /* 22 UINTX_FORMAT */
1222              &vsize,          /* 23 UINTX_FORMAT */
1223              &rss,            /* 24 INTX_FORMAT  */
1224              &rsslim,         /* 25 UINTX_FORMAT */
1225              &scodes,         /* 26 UINTX_FORMAT */
1226              &ecode,          /* 27 UINTX_FORMAT */
1227              &stack_start);   /* 28 UINTX_FORMAT */
1228       }
1229 
1230 #undef _UFM
1231 #undef _DFM
1232 
1233       if (i != 28 - 2) {
1234          assert(false, "Bad conversion from /proc/self/stat");
1235          // product mode - assume we are the initial thread, good luck in the
1236          // embedded case.
1237          warning("Can't detect initial thread stack location - bad conversion");
1238          stack_start = (uintptr_t) &rlim;
1239       }
1240     } else {
1241       // For some reason we can't open /proc/self/stat (for example, running on
1242       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1243       // most cases, so don't abort:
1244       warning("Can't detect initial thread stack location - no /proc/self/stat");
1245       stack_start = (uintptr_t) &rlim;
1246     }
1247   }
1248 
1249   // Now we have a pointer (stack_start) very close to the stack top, the
1250   // next thing to do is to figure out the exact location of stack top. We
1251   // can find out the virtual memory area that contains stack_start by
1252   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1253   // and its upper limit is the real stack top. (again, this would fail if
1254   // running inside chroot, because /proc may not exist.)
1255 
1256   uintptr_t stack_top;
1257   address low, high;
1258   if (find_vma((address)stack_start, &low, &high)) {
1259     // success, "high" is the true stack top. (ignore "low", because initial
1260     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1261     stack_top = (uintptr_t)high;
1262   } else {
1263     // failed, likely because /proc/self/maps does not exist
1264     warning("Can't detect initial thread stack location - find_vma failed");
1265     // best effort: stack_start is normally within a few pages below the real
1266     // stack top, use it as stack top, and reduce stack size so we won't put
1267     // guard page outside stack.
1268     stack_top = stack_start;
1269     stack_size -= 16 * page_size();
1270   }
1271 
1272   // stack_top could be partially down the page so align it
1273   stack_top = align_size_up(stack_top, page_size());
1274 
1275   if (max_size && stack_size > max_size) {
1276      _initial_thread_stack_size = max_size;
1277   } else {
1278      _initial_thread_stack_size = stack_size;
1279   }
1280 
1281   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1282   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1283 }
1284 
1285 ////////////////////////////////////////////////////////////////////////////////
1286 // time support
1287 
1288 // Time since start-up in seconds to a fine granularity.
1289 // Used by VMSelfDestructTimer and the MemProfiler.
1290 double os::elapsedTime() {
1291 
1292   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1293 }
1294 
1295 jlong os::elapsed_counter() {
1296   return javaTimeNanos() - initial_time_count;
1297 }
1298 
1299 jlong os::elapsed_frequency() {
1300   return NANOSECS_PER_SEC; // nanosecond resolution
1301 }
1302 
1303 bool os::supports_vtime() { return true; }
1304 bool os::enable_vtime()   { return false; }
1305 bool os::vtime_enabled()  { return false; }
1306 
1307 double os::elapsedVTime() {
1308   struct rusage usage;
1309   int retval = getrusage(RUSAGE_THREAD, &usage);
1310   if (retval == 0) {
1311     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);
1312   } else {
1313     // better than nothing, but not much
1314     return elapsedTime();
1315   }
1316 }
1317 
1318 jlong os::javaTimeMillis() {
1319   timeval time;
1320   int status = gettimeofday(&time, NULL);
1321   assert(status != -1, "linux error");
1322   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1323 }
1324 
1325 #ifndef CLOCK_MONOTONIC
1326 #define CLOCK_MONOTONIC (1)
1327 #endif
1328 
1329 void os::Linux::clock_init() {
1330   // we do dlopen's in this particular order due to bug in linux
1331   // dynamical loader (see 6348968) leading to crash on exit
1332   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1333   if (handle == NULL) {
1334     handle = dlopen("librt.so", RTLD_LAZY);
1335   }
1336 
1337   if (handle) {
1338     int (*clock_getres_func)(clockid_t, struct timespec*) =
1339            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1340     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1341            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1342     if (clock_getres_func && clock_gettime_func) {
1343       // See if monotonic clock is supported by the kernel. Note that some
1344       // early implementations simply return kernel jiffies (updated every
1345       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1346       // for nano time (though the monotonic property is still nice to have).
1347       // It's fixed in newer kernels, however clock_getres() still returns
1348       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1349       // resolution for now. Hopefully as people move to new kernels, this
1350       // won't be a problem.
1351       struct timespec res;
1352       struct timespec tp;
1353       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1354           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1355         // yes, monotonic clock is supported
1356         _clock_gettime = clock_gettime_func;
1357         return;
1358       } else {
1359         // close librt if there is no monotonic clock
1360         dlclose(handle);
1361       }
1362     }
1363   }
1364   warning("No monotonic clock was available - timed services may " \
1365           "be adversely affected if the time-of-day clock changes");
1366 }
1367 
1368 #ifndef SYS_clock_getres
1369 
1370 #if defined(IA32) || defined(AMD64)
1371 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1372 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1373 #else
1374 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1375 #define sys_clock_getres(x,y)  -1
1376 #endif
1377 
1378 #else
1379 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1380 #endif
1381 
1382 void os::Linux::fast_thread_clock_init() {
1383   if (!UseLinuxPosixThreadCPUClocks) {
1384     return;
1385   }
1386   clockid_t clockid;
1387   struct timespec tp;
1388   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1389       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1390 
1391   // Switch to using fast clocks for thread cpu time if
1392   // the sys_clock_getres() returns 0 error code.
1393   // Note, that some kernels may support the current thread
1394   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1395   // returned by the pthread_getcpuclockid().
1396   // If the fast Posix clocks are supported then the sys_clock_getres()
1397   // must return at least tp.tv_sec == 0 which means a resolution
1398   // better than 1 sec. This is extra check for reliability.
1399 
1400   if(pthread_getcpuclockid_func &&
1401      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1402      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1403 
1404     _supports_fast_thread_cpu_time = true;
1405     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1406   }
1407 }
1408 
1409 jlong os::javaTimeNanos() {
1410   if (Linux::supports_monotonic_clock()) {
1411     struct timespec tp;
1412     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1413     assert(status == 0, "gettime error");
1414     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1415     return result;
1416   } else {
1417     timeval time;
1418     int status = gettimeofday(&time, NULL);
1419     assert(status != -1, "linux error");
1420     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1421     return 1000 * usecs;
1422   }
1423 }
1424 
1425 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1426   if (Linux::supports_monotonic_clock()) {
1427     info_ptr->max_value = ALL_64_BITS;
1428 
1429     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1430     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1431     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1432   } else {
1433     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1434     info_ptr->max_value = ALL_64_BITS;
1435 
1436     // gettimeofday is a real time clock so it skips
1437     info_ptr->may_skip_backward = true;
1438     info_ptr->may_skip_forward = true;
1439   }
1440 
1441   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1442 }
1443 
1444 // Return the real, user, and system times in seconds from an
1445 // arbitrary fixed point in the past.
1446 bool os::getTimesSecs(double* process_real_time,
1447                       double* process_user_time,
1448                       double* process_system_time) {
1449   struct tms ticks;
1450   clock_t real_ticks = times(&ticks);
1451 
1452   if (real_ticks == (clock_t) (-1)) {
1453     return false;
1454   } else {
1455     double ticks_per_second = (double) clock_tics_per_sec;
1456     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1457     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1458     *process_real_time = ((double) real_ticks) / ticks_per_second;
1459 
1460     return true;
1461   }
1462 }
1463 
1464 
1465 char * os::local_time_string(char *buf, size_t buflen) {
1466   struct tm t;
1467   time_t long_time;
1468   time(&long_time);
1469   localtime_r(&long_time, &t);
1470   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1471                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1472                t.tm_hour, t.tm_min, t.tm_sec);
1473   return buf;
1474 }
1475 
1476 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1477   return localtime_r(clock, res);
1478 }
1479 
1480 ////////////////////////////////////////////////////////////////////////////////
1481 // runtime exit support
1482 
1483 // Note: os::shutdown() might be called very early during initialization, or
1484 // called from signal handler. Before adding something to os::shutdown(), make
1485 // sure it is async-safe and can handle partially initialized VM.
1486 void os::shutdown() {
1487 
1488   // allow PerfMemory to attempt cleanup of any persistent resources
1489   perfMemory_exit();
1490 
1491   // needs to remove object in file system
1492   AttachListener::abort();
1493 
1494   // flush buffered output, finish log files
1495   ostream_abort();
1496 
1497   // Check for abort hook
1498   abort_hook_t abort_hook = Arguments::abort_hook();
1499   if (abort_hook != NULL) {
1500     abort_hook();
1501   }
1502 
1503 }
1504 
1505 // Note: os::abort() might be called very early during initialization, or
1506 // called from signal handler. Before adding something to os::abort(), make
1507 // sure it is async-safe and can handle partially initialized VM.
1508 void os::abort(bool dump_core) {
1509   os::shutdown();
1510   if (dump_core) {
1511 #ifndef PRODUCT
1512     fdStream out(defaultStream::output_fd());
1513     out.print_raw("Current thread is ");
1514     char buf[16];
1515     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1516     out.print_raw_cr(buf);
1517     out.print_raw_cr("Dumping core ...");
1518 #endif
1519     ::abort(); // dump core
1520   }
1521 
1522   ::exit(1);
1523 }
1524 
1525 // Die immediately, no exit hook, no abort hook, no cleanup.
1526 void os::die() {
1527   // _exit() on LinuxThreads only kills current thread
1528   ::abort();
1529 }
1530 
1531 
1532 // This method is a copy of JDK's sysGetLastErrorString
1533 // from src/solaris/hpi/src/system_md.c
1534 
1535 size_t os::lasterror(char *buf, size_t len) {
1536 
1537   if (errno == 0)  return 0;
1538 
1539   const char *s = ::strerror(errno);
1540   size_t n = ::strlen(s);
1541   if (n >= len) {
1542     n = len - 1;
1543   }
1544   ::strncpy(buf, s, n);
1545   buf[n] = '\0';
1546   return n;
1547 }
1548 
1549 intx os::current_thread_id() { return (intx)pthread_self(); }
1550 int os::current_process_id() {
1551 
1552   // Under the old linux thread library, linux gives each thread
1553   // its own process id. Because of this each thread will return
1554   // a different pid if this method were to return the result
1555   // of getpid(2). Linux provides no api that returns the pid
1556   // of the launcher thread for the vm. This implementation
1557   // returns a unique pid, the pid of the launcher thread
1558   // that starts the vm 'process'.
1559 
1560   // Under the NPTL, getpid() returns the same pid as the
1561   // launcher thread rather than a unique pid per thread.
1562   // Use gettid() if you want the old pre NPTL behaviour.
1563 
1564   // if you are looking for the result of a call to getpid() that
1565   // returns a unique pid for the calling thread, then look at the
1566   // OSThread::thread_id() method in osThread_linux.hpp file
1567 
1568   return (int)(_initial_pid ? _initial_pid : getpid());
1569 }
1570 
1571 // DLL functions
1572 
1573 const char* os::dll_file_extension() { return ".so"; }
1574 
1575 // This must be hard coded because it's the system's temporary
1576 // directory not the java application's temp directory, ala java.io.tmpdir.
1577 const char* os::get_temp_directory() { return "/tmp"; }
1578 
1579 static bool file_exists(const char* filename) {
1580   struct stat statbuf;
1581   if (filename == NULL || strlen(filename) == 0) {
1582     return false;
1583   }
1584   return os::stat(filename, &statbuf) == 0;
1585 }
1586 
1587 bool os::dll_build_name(char* buffer, size_t buflen,
1588                         const char* pname, const char* fname) {
1589   bool retval = false;
1590   // Copied from libhpi
1591   const size_t pnamelen = pname ? strlen(pname) : 0;
1592 
1593   // Return error on buffer overflow.
1594   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1595     return retval;
1596   }
1597 
1598   if (pnamelen == 0) {
1599     snprintf(buffer, buflen, "lib%s.so", fname);
1600     retval = true;
1601   } else if (strchr(pname, *os::path_separator()) != NULL) {
1602     int n;
1603     char** pelements = split_path(pname, &n);
1604     if (pelements == NULL) {
1605       return false;
1606     }
1607     for (int i = 0 ; i < n ; i++) {
1608       // Really shouldn't be NULL, but check can't hurt
1609       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1610         continue; // skip the empty path values
1611       }
1612       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1613       if (file_exists(buffer)) {
1614         retval = true;
1615         break;
1616       }
1617     }
1618     // release the storage
1619     for (int i = 0 ; i < n ; i++) {
1620       if (pelements[i] != NULL) {
1621         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1622       }
1623     }
1624     if (pelements != NULL) {
1625       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1626     }
1627   } else {
1628     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1629     retval = true;
1630   }
1631   return retval;
1632 }
1633 
1634 // check if addr is inside libjvm.so
1635 bool os::address_is_in_vm(address addr) {
1636   static address libjvm_base_addr;
1637   Dl_info dlinfo;
1638 
1639   if (libjvm_base_addr == NULL) {
1640     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1641       libjvm_base_addr = (address)dlinfo.dli_fbase;
1642     }
1643     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1644   }
1645 
1646   if (dladdr((void *)addr, &dlinfo) != 0) {
1647     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1648   }
1649 
1650   return false;
1651 }
1652 
1653 bool os::dll_address_to_function_name(address addr, char *buf,
1654                                       int buflen, int *offset) {
1655   // buf is not optional, but offset is optional
1656   assert(buf != NULL, "sanity check");
1657 
1658   Dl_info dlinfo;
1659 
1660   if (dladdr((void*)addr, &dlinfo) != 0) {
1661     // see if we have a matching symbol
1662     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1663       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1664         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1665       }
1666       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1667       return true;
1668     }
1669     // no matching symbol so try for just file info
1670     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1671       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1672                           buf, buflen, offset, dlinfo.dli_fname)) {
1673         return true;
1674       }
1675     }
1676   }
1677 
1678   buf[0] = '\0';
1679   if (offset != NULL) *offset = -1;
1680   return false;
1681 }
1682 
1683 struct _address_to_library_name {
1684   address addr;          // input : memory address
1685   size_t  buflen;        //         size of fname
1686   char*   fname;         // output: library name
1687   address base;          //         library base addr
1688 };
1689 
1690 static int address_to_library_name_callback(struct dl_phdr_info *info,
1691                                             size_t size, void *data) {
1692   int i;
1693   bool found = false;
1694   address libbase = NULL;
1695   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1696 
1697   // iterate through all loadable segments
1698   for (i = 0; i < info->dlpi_phnum; i++) {
1699     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1700     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1701       // base address of a library is the lowest address of its loaded
1702       // segments.
1703       if (libbase == NULL || libbase > segbase) {
1704         libbase = segbase;
1705       }
1706       // see if 'addr' is within current segment
1707       if (segbase <= d->addr &&
1708           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1709         found = true;
1710       }
1711     }
1712   }
1713 
1714   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1715   // so dll_address_to_library_name() can fall through to use dladdr() which
1716   // can figure out executable name from argv[0].
1717   if (found && info->dlpi_name && info->dlpi_name[0]) {
1718     d->base = libbase;
1719     if (d->fname) {
1720       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1721     }
1722     return 1;
1723   }
1724   return 0;
1725 }
1726 
1727 bool os::dll_address_to_library_name(address addr, char* buf,
1728                                      int buflen, int* offset) {
1729   // buf is not optional, but offset is optional
1730   assert(buf != NULL, "sanity check");
1731 
1732   Dl_info dlinfo;
1733   struct _address_to_library_name data;
1734 
1735   // There is a bug in old glibc dladdr() implementation that it could resolve
1736   // to wrong library name if the .so file has a base address != NULL. Here
1737   // we iterate through the program headers of all loaded libraries to find
1738   // out which library 'addr' really belongs to. This workaround can be
1739   // removed once the minimum requirement for glibc is moved to 2.3.x.
1740   data.addr = addr;
1741   data.fname = buf;
1742   data.buflen = buflen;
1743   data.base = NULL;
1744   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1745 
1746   if (rslt) {
1747      // buf already contains library name
1748      if (offset) *offset = addr - data.base;
1749      return true;
1750   }
1751   if (dladdr((void*)addr, &dlinfo) != 0) {
1752     if (dlinfo.dli_fname != NULL) {
1753       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1754     }
1755     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1756       *offset = addr - (address)dlinfo.dli_fbase;
1757     }
1758     return true;
1759   }
1760 
1761   buf[0] = '\0';
1762   if (offset) *offset = -1;
1763   return false;
1764 }
1765 
1766   // Loads .dll/.so and
1767   // in case of error it checks if .dll/.so was built for the
1768   // same architecture as Hotspot is running on
1769 
1770 
1771 // Remember the stack's state. The Linux dynamic linker will change
1772 // the stack to 'executable' at most once, so we must safepoint only once.
1773 bool os::Linux::_stack_is_executable = false;
1774 
1775 // VM operation that loads a library.  This is necessary if stack protection
1776 // of the Java stacks can be lost during loading the library.  If we
1777 // do not stop the Java threads, they can stack overflow before the stacks
1778 // are protected again.
1779 class VM_LinuxDllLoad: public VM_Operation {
1780  private:
1781   const char *_filename;
1782   char *_ebuf;
1783   int _ebuflen;
1784   void *_lib;
1785  public:
1786   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1787     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1788   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1789   void doit() {
1790     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1791     os::Linux::_stack_is_executable = true;
1792   }
1793   void* loaded_library() { return _lib; }
1794 };
1795 
1796 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1797 {
1798   void * result = NULL;
1799   bool load_attempted = false;
1800 
1801   // Check whether the library to load might change execution rights
1802   // of the stack. If they are changed, the protection of the stack
1803   // guard pages will be lost. We need a safepoint to fix this.
1804   //
1805   // See Linux man page execstack(8) for more info.
1806   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1807     ElfFile ef(filename);
1808     if (!ef.specifies_noexecstack()) {
1809       if (!is_init_completed()) {
1810         os::Linux::_stack_is_executable = true;
1811         // This is OK - No Java threads have been created yet, and hence no
1812         // stack guard pages to fix.
1813         //
1814         // This should happen only when you are building JDK7 using a very
1815         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1816         //
1817         // Dynamic loader will make all stacks executable after
1818         // this function returns, and will not do that again.
1819         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1820       } else {
1821         warning("You have loaded library %s which might have disabled stack guard. "
1822                 "The VM will try to fix the stack guard now.\n"
1823                 "It's highly recommended that you fix the library with "
1824                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1825                 filename);
1826 
1827         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1828         JavaThread *jt = JavaThread::current();
1829         if (jt->thread_state() != _thread_in_native) {
1830           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1831           // that requires ExecStack. Cannot enter safe point. Let's give up.
1832           warning("Unable to fix stack guard. Giving up.");
1833         } else {
1834           if (!LoadExecStackDllInVMThread) {
1835             // This is for the case where the DLL has an static
1836             // constructor function that executes JNI code. We cannot
1837             // load such DLLs in the VMThread.
1838             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1839           }
1840 
1841           ThreadInVMfromNative tiv(jt);
1842           debug_only(VMNativeEntryWrapper vew;)
1843 
1844           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1845           VMThread::execute(&op);
1846           if (LoadExecStackDllInVMThread) {
1847             result = op.loaded_library();
1848           }
1849           load_attempted = true;
1850         }
1851       }
1852     }
1853   }
1854 
1855   if (!load_attempted) {
1856     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1857   }
1858 
1859   if (result != NULL) {
1860     // Successful loading
1861     return result;
1862   }
1863 
1864   Elf32_Ehdr elf_head;
1865   int diag_msg_max_length=ebuflen-strlen(ebuf);
1866   char* diag_msg_buf=ebuf+strlen(ebuf);
1867 
1868   if (diag_msg_max_length==0) {
1869     // No more space in ebuf for additional diagnostics message
1870     return NULL;
1871   }
1872 
1873 
1874   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1875 
1876   if (file_descriptor < 0) {
1877     // Can't open library, report dlerror() message
1878     return NULL;
1879   }
1880 
1881   bool failed_to_read_elf_head=
1882     (sizeof(elf_head)!=
1883         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1884 
1885   ::close(file_descriptor);
1886   if (failed_to_read_elf_head) {
1887     // file i/o error - report dlerror() msg
1888     return NULL;
1889   }
1890 
1891   typedef struct {
1892     Elf32_Half  code;         // Actual value as defined in elf.h
1893     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1894     char        elf_class;    // 32 or 64 bit
1895     char        endianess;    // MSB or LSB
1896     char*       name;         // String representation
1897   } arch_t;
1898 
1899   #ifndef EM_486
1900   #define EM_486          6               /* Intel 80486 */
1901   #endif
1902 
1903   static const arch_t arch_array[]={
1904     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1905     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1906     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1907     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1908     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1909     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1910     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1911     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1912 #if defined(VM_LITTLE_ENDIAN)
1913     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1914 #else
1915     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1916 #endif
1917     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1918     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1919     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1920     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1921     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1922     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1923     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1924   };
1925 
1926   #if  (defined IA32)
1927     static  Elf32_Half running_arch_code=EM_386;
1928   #elif   (defined AMD64)
1929     static  Elf32_Half running_arch_code=EM_X86_64;
1930   #elif  (defined IA64)
1931     static  Elf32_Half running_arch_code=EM_IA_64;
1932   #elif  (defined __sparc) && (defined _LP64)
1933     static  Elf32_Half running_arch_code=EM_SPARCV9;
1934   #elif  (defined __sparc) && (!defined _LP64)
1935     static  Elf32_Half running_arch_code=EM_SPARC;
1936   #elif  (defined __powerpc64__)
1937     static  Elf32_Half running_arch_code=EM_PPC64;
1938   #elif  (defined __powerpc__)
1939     static  Elf32_Half running_arch_code=EM_PPC;
1940   #elif  (defined ARM)
1941     static  Elf32_Half running_arch_code=EM_ARM;
1942   #elif  (defined S390)
1943     static  Elf32_Half running_arch_code=EM_S390;
1944   #elif  (defined ALPHA)
1945     static  Elf32_Half running_arch_code=EM_ALPHA;
1946   #elif  (defined MIPSEL)
1947     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1948   #elif  (defined PARISC)
1949     static  Elf32_Half running_arch_code=EM_PARISC;
1950   #elif  (defined MIPS)
1951     static  Elf32_Half running_arch_code=EM_MIPS;
1952   #elif  (defined M68K)
1953     static  Elf32_Half running_arch_code=EM_68K;
1954   #else
1955     #error Method os::dll_load requires that one of following is defined:\
1956          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1957   #endif
1958 
1959   // Identify compatability class for VM's architecture and library's architecture
1960   // Obtain string descriptions for architectures
1961 
1962   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1963   int running_arch_index=-1;
1964 
1965   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1966     if (running_arch_code == arch_array[i].code) {
1967       running_arch_index    = i;
1968     }
1969     if (lib_arch.code == arch_array[i].code) {
1970       lib_arch.compat_class = arch_array[i].compat_class;
1971       lib_arch.name         = arch_array[i].name;
1972     }
1973   }
1974 
1975   assert(running_arch_index != -1,
1976     "Didn't find running architecture code (running_arch_code) in arch_array");
1977   if (running_arch_index == -1) {
1978     // Even though running architecture detection failed
1979     // we may still continue with reporting dlerror() message
1980     return NULL;
1981   }
1982 
1983   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1984     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1985     return NULL;
1986   }
1987 
1988 #ifndef S390
1989   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1990     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1991     return NULL;
1992   }
1993 #endif // !S390
1994 
1995   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1996     if ( lib_arch.name!=NULL ) {
1997       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1998         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1999         lib_arch.name, arch_array[running_arch_index].name);
2000     } else {
2001       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2002       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2003         lib_arch.code,
2004         arch_array[running_arch_index].name);
2005     }
2006   }
2007 
2008   return NULL;
2009 }
2010 
2011 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2012   void * result = ::dlopen(filename, RTLD_LAZY);
2013   if (result == NULL) {
2014     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2015     ebuf[ebuflen-1] = '\0';
2016   }
2017   return result;
2018 }
2019 
2020 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2021   void * result = NULL;
2022   if (LoadExecStackDllInVMThread) {
2023     result = dlopen_helper(filename, ebuf, ebuflen);
2024   }
2025 
2026   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2027   // library that requires an executable stack, or which does not have this
2028   // stack attribute set, dlopen changes the stack attribute to executable. The
2029   // read protection of the guard pages gets lost.
2030   //
2031   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2032   // may have been queued at the same time.
2033 
2034   if (!_stack_is_executable) {
2035     JavaThread *jt = Threads::first();
2036 
2037     while (jt) {
2038       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2039           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2040         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2041                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2042           warning("Attempt to reguard stack yellow zone failed.");
2043         }
2044       }
2045       jt = jt->next();
2046     }
2047   }
2048 
2049   return result;
2050 }
2051 
2052 /*
2053  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2054  * chances are you might want to run the generated bits against glibc-2.0
2055  * libdl.so, so always use locking for any version of glibc.
2056  */
2057 void* os::dll_lookup(void* handle, const char* name) {
2058   pthread_mutex_lock(&dl_mutex);
2059   void* res = dlsym(handle, name);
2060   pthread_mutex_unlock(&dl_mutex);
2061   return res;
2062 }
2063 
2064 void* os::get_default_process_handle() {
2065   return (void*)::dlopen(NULL, RTLD_LAZY);
2066 }
2067 
2068 static bool _print_ascii_file(const char* filename, outputStream* st) {
2069   int fd = ::open(filename, O_RDONLY);
2070   if (fd == -1) {
2071      return false;
2072   }
2073 
2074   char buf[32];
2075   int bytes;
2076   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2077     st->print_raw(buf, bytes);
2078   }
2079 
2080   ::close(fd);
2081 
2082   return true;
2083 }
2084 
2085 void os::print_dll_info(outputStream *st) {
2086    st->print_cr("Dynamic libraries:");
2087 
2088    char fname[32];
2089    pid_t pid = os::Linux::gettid();
2090 
2091    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2092 
2093    if (!_print_ascii_file(fname, st)) {
2094      st->print("Can not get library information for pid = %d\n", pid);
2095    }
2096 }
2097 
2098 void os::print_os_info_brief(outputStream* st) {
2099   os::Linux::print_distro_info(st);
2100 
2101   os::Posix::print_uname_info(st);
2102 
2103   os::Linux::print_libversion_info(st);
2104 
2105 }
2106 
2107 void os::print_os_info(outputStream* st) {
2108   st->print("OS:");
2109 
2110   os::Linux::print_distro_info(st);
2111 
2112   os::Posix::print_uname_info(st);
2113 
2114   // Print warning if unsafe chroot environment detected
2115   if (unsafe_chroot_detected) {
2116     st->print("WARNING!! ");
2117     st->print_cr("%s", unstable_chroot_error);
2118   }
2119 
2120   os::Linux::print_libversion_info(st);
2121 
2122   os::Posix::print_rlimit_info(st);
2123 
2124   os::Posix::print_load_average(st);
2125 
2126   os::Linux::print_full_memory_info(st);
2127 }
2128 
2129 // Try to identify popular distros.
2130 // Most Linux distributions have a /etc/XXX-release file, which contains
2131 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2132 // file that also contains the OS version string. Some have more than one
2133 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2134 // /etc/redhat-release.), so the order is important.
2135 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2136 // their own specific XXX-release file as well as a redhat-release file.
2137 // Because of this the XXX-release file needs to be searched for before the
2138 // redhat-release file.
2139 // Since Red Hat has a lsb-release file that is not very descriptive the
2140 // search for redhat-release needs to be before lsb-release.
2141 // Since the lsb-release file is the new standard it needs to be searched
2142 // before the older style release files.
2143 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2144 // next to last resort.  The os-release file is a new standard that contains
2145 // distribution information and the system-release file seems to be an old
2146 // standard that has been replaced by the lsb-release and os-release files.
2147 // Searching for the debian_version file is the last resort.  It contains
2148 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2149 // "Debian " is printed before the contents of the debian_version file.
2150 void os::Linux::print_distro_info(outputStream* st) {
2151    if (!_print_ascii_file("/etc/oracle-release", st) &&
2152        !_print_ascii_file("/etc/mandriva-release", st) &&
2153        !_print_ascii_file("/etc/mandrake-release", st) &&
2154        !_print_ascii_file("/etc/sun-release", st) &&
2155        !_print_ascii_file("/etc/redhat-release", st) &&
2156        !_print_ascii_file("/etc/lsb-release", st) &&
2157        !_print_ascii_file("/etc/SuSE-release", st) &&
2158        !_print_ascii_file("/etc/turbolinux-release", st) &&
2159        !_print_ascii_file("/etc/gentoo-release", st) &&
2160        !_print_ascii_file("/etc/ltib-release", st) &&
2161        !_print_ascii_file("/etc/angstrom-version", st) &&
2162        !_print_ascii_file("/etc/system-release", st) &&
2163        !_print_ascii_file("/etc/os-release", st)) {
2164 
2165        if (file_exists("/etc/debian_version")) {
2166          st->print("Debian ");
2167          _print_ascii_file("/etc/debian_version", st);
2168        } else {
2169          st->print("Linux");
2170        }
2171    }
2172    st->cr();
2173 }
2174 
2175 void os::Linux::print_libversion_info(outputStream* st) {
2176   // libc, pthread
2177   st->print("libc:");
2178   st->print("%s ", os::Linux::glibc_version());
2179   st->print("%s ", os::Linux::libpthread_version());
2180   if (os::Linux::is_LinuxThreads()) {
2181      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2182   }
2183   st->cr();
2184 }
2185 
2186 void os::Linux::print_full_memory_info(outputStream* st) {
2187    st->print("\n/proc/meminfo:\n");
2188    _print_ascii_file("/proc/meminfo", st);
2189    st->cr();
2190 }
2191 
2192 void os::print_memory_info(outputStream* st) {
2193 
2194   st->print("Memory:");
2195   st->print(" %dk page", os::vm_page_size()>>10);
2196 
2197   // values in struct sysinfo are "unsigned long"
2198   struct sysinfo si;
2199   sysinfo(&si);
2200 
2201   st->print(", physical " UINT64_FORMAT "k",
2202             os::physical_memory() >> 10);
2203   st->print("(" UINT64_FORMAT "k free)",
2204             os::available_memory() >> 10);
2205   st->print(", swap " UINT64_FORMAT "k",
2206             ((jlong)si.totalswap * si.mem_unit) >> 10);
2207   st->print("(" UINT64_FORMAT "k free)",
2208             ((jlong)si.freeswap * si.mem_unit) >> 10);
2209   st->cr();
2210 }
2211 
2212 void os::pd_print_cpu_info(outputStream* st) {
2213   st->print("\n/proc/cpuinfo:\n");
2214   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2215     st->print("  <Not Available>");
2216   }
2217   st->cr();
2218 }
2219 
2220 void os::print_siginfo(outputStream* st, void* siginfo) {
2221   const siginfo_t* si = (const siginfo_t*)siginfo;
2222 
2223   os::Posix::print_siginfo_brief(st, si);
2224 #if INCLUDE_CDS
2225   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2226       UseSharedSpaces) {
2227     FileMapInfo* mapinfo = FileMapInfo::current_info();
2228     if (mapinfo->is_in_shared_space(si->si_addr)) {
2229       st->print("\n\nError accessing class data sharing archive."   \
2230                 " Mapped file inaccessible during execution, "      \
2231                 " possible disk/network problem.");
2232     }
2233   }
2234 #endif
2235   st->cr();
2236 }
2237 
2238 
2239 static void print_signal_handler(outputStream* st, int sig,
2240                                  char* buf, size_t buflen);
2241 
2242 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2243   st->print_cr("Signal Handlers:");
2244   print_signal_handler(st, SIGSEGV, buf, buflen);
2245   print_signal_handler(st, SIGBUS , buf, buflen);
2246   print_signal_handler(st, SIGFPE , buf, buflen);
2247   print_signal_handler(st, SIGPIPE, buf, buflen);
2248   print_signal_handler(st, SIGXFSZ, buf, buflen);
2249   print_signal_handler(st, SIGILL , buf, buflen);
2250   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2251   print_signal_handler(st, SR_signum, buf, buflen);
2252   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2253   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2254   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2255   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2256 #if defined(PPC64)
2257   print_signal_handler(st, SIGTRAP, buf, buflen);
2258 #endif
2259 }
2260 
2261 static char saved_jvm_path[MAXPATHLEN] = {0};
2262 
2263 // Find the full path to the current module, libjvm.so
2264 void os::jvm_path(char *buf, jint buflen) {
2265   // Error checking.
2266   if (buflen < MAXPATHLEN) {
2267     assert(false, "must use a large-enough buffer");
2268     buf[0] = '\0';
2269     return;
2270   }
2271   // Lazy resolve the path to current module.
2272   if (saved_jvm_path[0] != 0) {
2273     strcpy(buf, saved_jvm_path);
2274     return;
2275   }
2276 
2277   char dli_fname[MAXPATHLEN];
2278   bool ret = dll_address_to_library_name(
2279                 CAST_FROM_FN_PTR(address, os::jvm_path),
2280                 dli_fname, sizeof(dli_fname), NULL);
2281   assert(ret, "cannot locate libjvm");
2282   char *rp = NULL;
2283   if (ret && dli_fname[0] != '\0') {
2284     rp = realpath(dli_fname, buf);
2285   }
2286   if (rp == NULL)
2287     return;
2288 
2289   if (Arguments::created_by_gamma_launcher()) {
2290     // Support for the gamma launcher.  Typical value for buf is
2291     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2292     // the right place in the string, then assume we are installed in a JDK and
2293     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2294     // up the path so it looks like libjvm.so is installed there (append a
2295     // fake suffix hotspot/libjvm.so).
2296     const char *p = buf + strlen(buf) - 1;
2297     for (int count = 0; p > buf && count < 5; ++count) {
2298       for (--p; p > buf && *p != '/'; --p)
2299         /* empty */ ;
2300     }
2301 
2302     if (strncmp(p, "/jre/lib/", 9) != 0) {
2303       // Look for JAVA_HOME in the environment.
2304       char* java_home_var = ::getenv("JAVA_HOME");
2305       if (java_home_var != NULL && java_home_var[0] != 0) {
2306         char* jrelib_p;
2307         int len;
2308 
2309         // Check the current module name "libjvm.so".
2310         p = strrchr(buf, '/');
2311         assert(strstr(p, "/libjvm") == p, "invalid library name");
2312 
2313         rp = realpath(java_home_var, buf);
2314         if (rp == NULL)
2315           return;
2316 
2317         // determine if this is a legacy image or modules image
2318         // modules image doesn't have "jre" subdirectory
2319         len = strlen(buf);
2320         assert(len < buflen, "Ran out of buffer room");
2321         jrelib_p = buf + len;
2322         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2323         if (0 != access(buf, F_OK)) {
2324           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2325         }
2326 
2327         if (0 == access(buf, F_OK)) {
2328           // Use current module name "libjvm.so"
2329           len = strlen(buf);
2330           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2331         } else {
2332           // Go back to path of .so
2333           rp = realpath(dli_fname, buf);
2334           if (rp == NULL)
2335             return;
2336         }
2337       }
2338     }
2339   }
2340 
2341   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2342 }
2343 
2344 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2345   // no prefix required, not even "_"
2346 }
2347 
2348 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2349   // no suffix required
2350 }
2351 
2352 ////////////////////////////////////////////////////////////////////////////////
2353 // sun.misc.Signal support
2354 
2355 static volatile jint sigint_count = 0;
2356 
2357 static void
2358 UserHandler(int sig, void *siginfo, void *context) {
2359   // 4511530 - sem_post is serialized and handled by the manager thread. When
2360   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2361   // don't want to flood the manager thread with sem_post requests.
2362   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2363       return;
2364 
2365   // Ctrl-C is pressed during error reporting, likely because the error
2366   // handler fails to abort. Let VM die immediately.
2367   if (sig == SIGINT && is_error_reported()) {
2368      os::die();
2369   }
2370 
2371   os::signal_notify(sig);
2372 }
2373 
2374 void* os::user_handler() {
2375   return CAST_FROM_FN_PTR(void*, UserHandler);
2376 }
2377 
2378 class Semaphore : public StackObj {
2379   public:
2380     Semaphore();
2381     ~Semaphore();
2382     void signal();
2383     void wait();
2384     bool trywait();
2385     bool timedwait(unsigned int sec, int nsec);
2386   private:
2387     sem_t _semaphore;
2388 };
2389 
2390 Semaphore::Semaphore() {
2391   sem_init(&_semaphore, 0, 0);
2392 }
2393 
2394 Semaphore::~Semaphore() {
2395   sem_destroy(&_semaphore);
2396 }
2397 
2398 void Semaphore::signal() {
2399   sem_post(&_semaphore);
2400 }
2401 
2402 void Semaphore::wait() {
2403   sem_wait(&_semaphore);
2404 }
2405 
2406 bool Semaphore::trywait() {
2407   return sem_trywait(&_semaphore) == 0;
2408 }
2409 
2410 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2411 
2412   struct timespec ts;
2413   // Semaphore's are always associated with CLOCK_REALTIME
2414   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2415   // see unpackTime for discussion on overflow checking
2416   if (sec >= MAX_SECS) {
2417     ts.tv_sec += MAX_SECS;
2418     ts.tv_nsec = 0;
2419   } else {
2420     ts.tv_sec += sec;
2421     ts.tv_nsec += nsec;
2422     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2423       ts.tv_nsec -= NANOSECS_PER_SEC;
2424       ++ts.tv_sec; // note: this must be <= max_secs
2425     }
2426   }
2427 
2428   while (1) {
2429     int result = sem_timedwait(&_semaphore, &ts);
2430     if (result == 0) {
2431       return true;
2432     } else if (errno == EINTR) {
2433       continue;
2434     } else if (errno == ETIMEDOUT) {
2435       return false;
2436     } else {
2437       return false;
2438     }
2439   }
2440 }
2441 
2442 extern "C" {
2443   typedef void (*sa_handler_t)(int);
2444   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2445 }
2446 
2447 void* os::signal(int signal_number, void* handler) {
2448   struct sigaction sigAct, oldSigAct;
2449 
2450   sigfillset(&(sigAct.sa_mask));
2451   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2452   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2453 
2454   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2455     // -1 means registration failed
2456     return (void *)-1;
2457   }
2458 
2459   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2460 }
2461 
2462 void os::signal_raise(int signal_number) {
2463   ::raise(signal_number);
2464 }
2465 
2466 /*
2467  * The following code is moved from os.cpp for making this
2468  * code platform specific, which it is by its very nature.
2469  */
2470 
2471 // Will be modified when max signal is changed to be dynamic
2472 int os::sigexitnum_pd() {
2473   return NSIG;
2474 }
2475 
2476 // a counter for each possible signal value
2477 static volatile jint pending_signals[NSIG+1] = { 0 };
2478 
2479 // Linux(POSIX) specific hand shaking semaphore.
2480 static sem_t sig_sem;
2481 static Semaphore sr_semaphore;
2482 
2483 void os::signal_init_pd() {
2484   // Initialize signal structures
2485   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2486 
2487   // Initialize signal semaphore
2488   ::sem_init(&sig_sem, 0, 0);
2489 }
2490 
2491 void os::signal_notify(int sig) {
2492   Atomic::inc(&pending_signals[sig]);
2493   ::sem_post(&sig_sem);
2494 }
2495 
2496 static int check_pending_signals(bool wait) {
2497   Atomic::store(0, &sigint_count);
2498   for (;;) {
2499     for (int i = 0; i < NSIG + 1; i++) {
2500       jint n = pending_signals[i];
2501       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2502         return i;
2503       }
2504     }
2505     if (!wait) {
2506       return -1;
2507     }
2508     JavaThread *thread = JavaThread::current();
2509     ThreadBlockInVM tbivm(thread);
2510 
2511     bool threadIsSuspended;
2512     do {
2513       thread->set_suspend_equivalent();
2514       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2515       ::sem_wait(&sig_sem);
2516 
2517       // were we externally suspended while we were waiting?
2518       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2519       if (threadIsSuspended) {
2520         //
2521         // The semaphore has been incremented, but while we were waiting
2522         // another thread suspended us. We don't want to continue running
2523         // while suspended because that would surprise the thread that
2524         // suspended us.
2525         //
2526         ::sem_post(&sig_sem);
2527 
2528         thread->java_suspend_self();
2529       }
2530     } while (threadIsSuspended);
2531   }
2532 }
2533 
2534 int os::signal_lookup() {
2535   return check_pending_signals(false);
2536 }
2537 
2538 int os::signal_wait() {
2539   return check_pending_signals(true);
2540 }
2541 
2542 ////////////////////////////////////////////////////////////////////////////////
2543 // Virtual Memory
2544 
2545 int os::vm_page_size() {
2546   // Seems redundant as all get out
2547   assert(os::Linux::page_size() != -1, "must call os::init");
2548   return os::Linux::page_size();
2549 }
2550 
2551 // Solaris allocates memory by pages.
2552 int os::vm_allocation_granularity() {
2553   assert(os::Linux::page_size() != -1, "must call os::init");
2554   return os::Linux::page_size();
2555 }
2556 
2557 // Rationale behind this function:
2558 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2559 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2560 //  samples for JITted code. Here we create private executable mapping over the code cache
2561 //  and then we can use standard (well, almost, as mapping can change) way to provide
2562 //  info for the reporting script by storing timestamp and location of symbol
2563 void linux_wrap_code(char* base, size_t size) {
2564   static volatile jint cnt = 0;
2565 
2566   if (!UseOprofile) {
2567     return;
2568   }
2569 
2570   char buf[PATH_MAX+1];
2571   int num = Atomic::add(1, &cnt);
2572 
2573   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2574            os::get_temp_directory(), os::current_process_id(), num);
2575   unlink(buf);
2576 
2577   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2578 
2579   if (fd != -1) {
2580     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2581     if (rv != (off_t)-1) {
2582       if (::write(fd, "", 1) == 1) {
2583         mmap(base, size,
2584              PROT_READ|PROT_WRITE|PROT_EXEC,
2585              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2586       }
2587     }
2588     ::close(fd);
2589     unlink(buf);
2590   }
2591 }
2592 
2593 static bool recoverable_mmap_error(int err) {
2594   // See if the error is one we can let the caller handle. This
2595   // list of errno values comes from JBS-6843484. I can't find a
2596   // Linux man page that documents this specific set of errno
2597   // values so while this list currently matches Solaris, it may
2598   // change as we gain experience with this failure mode.
2599   switch (err) {
2600   case EBADF:
2601   case EINVAL:
2602   case ENOTSUP:
2603     // let the caller deal with these errors
2604     return true;
2605 
2606   default:
2607     // Any remaining errors on this OS can cause our reserved mapping
2608     // to be lost. That can cause confusion where different data
2609     // structures think they have the same memory mapped. The worst
2610     // scenario is if both the VM and a library think they have the
2611     // same memory mapped.
2612     return false;
2613   }
2614 }
2615 
2616 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2617                                     int err) {
2618   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2619           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2620           strerror(err), err);
2621 }
2622 
2623 static void warn_fail_commit_memory(char* addr, size_t size,
2624                                     size_t alignment_hint, bool exec,
2625                                     int err) {
2626   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2627           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2628           alignment_hint, exec, strerror(err), err);
2629 }
2630 
2631 // NOTE: Linux kernel does not really reserve the pages for us.
2632 //       All it does is to check if there are enough free pages
2633 //       left at the time of mmap(). This could be a potential
2634 //       problem.
2635 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2636   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2637   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2638                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2639   if (res != (uintptr_t) MAP_FAILED) {
2640     if (UseNUMAInterleaving) {
2641       numa_make_global(addr, size);
2642     }
2643     return 0;
2644   }
2645 
2646   int err = errno;  // save errno from mmap() call above
2647 
2648   if (!recoverable_mmap_error(err)) {
2649     warn_fail_commit_memory(addr, size, exec, err);
2650     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2651   }
2652 
2653   return err;
2654 }
2655 
2656 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2657   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2658 }
2659 
2660 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2661                                   const char* mesg) {
2662   assert(mesg != NULL, "mesg must be specified");
2663   int err = os::Linux::commit_memory_impl(addr, size, exec);
2664   if (err != 0) {
2665     // the caller wants all commit errors to exit with the specified mesg:
2666     warn_fail_commit_memory(addr, size, exec, err);
2667     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2668   }
2669 }
2670 
2671 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2672 #ifndef MAP_HUGETLB
2673 #define MAP_HUGETLB 0x40000
2674 #endif
2675 
2676 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2677 #ifndef MADV_HUGEPAGE
2678 #define MADV_HUGEPAGE 14
2679 #endif
2680 
2681 int os::Linux::commit_memory_impl(char* addr, size_t size,
2682                                   size_t alignment_hint, bool exec) {
2683   int err = os::Linux::commit_memory_impl(addr, size, exec);
2684   if (err == 0) {
2685     realign_memory(addr, size, alignment_hint);
2686   }
2687   return err;
2688 }
2689 
2690 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2691                           bool exec) {
2692   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2693 }
2694 
2695 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2696                                   size_t alignment_hint, bool exec,
2697                                   const char* mesg) {
2698   assert(mesg != NULL, "mesg must be specified");
2699   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2700   if (err != 0) {
2701     // the caller wants all commit errors to exit with the specified mesg:
2702     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2703     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2704   }
2705 }
2706 
2707 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2708   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2709     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2710     // be supported or the memory may already be backed by huge pages.
2711     ::madvise(addr, bytes, MADV_HUGEPAGE);
2712   }
2713 }
2714 
2715 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2716   // This method works by doing an mmap over an existing mmaping and effectively discarding
2717   // the existing pages. However it won't work for SHM-based large pages that cannot be
2718   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2719   // small pages on top of the SHM segment. This method always works for small pages, so we
2720   // allow that in any case.
2721   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2722     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2723   }
2724 }
2725 
2726 void os::numa_make_global(char *addr, size_t bytes) {
2727   Linux::numa_interleave_memory(addr, bytes);
2728 }
2729 
2730 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2731 // bind policy to MPOL_PREFERRED for the current thread.
2732 #define USE_MPOL_PREFERRED 0
2733 
2734 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2735   // To make NUMA and large pages more robust when both enabled, we need to ease
2736   // the requirements on where the memory should be allocated. MPOL_BIND is the
2737   // default policy and it will force memory to be allocated on the specified
2738   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2739   // the specified node, but will not force it. Using this policy will prevent
2740   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2741   // free large pages.
2742   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2743   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2744 }
2745 
2746 bool os::numa_topology_changed()   { return false; }
2747 
2748 size_t os::numa_get_groups_num() {
2749   int max_node = Linux::numa_max_node();
2750   return max_node > 0 ? max_node + 1 : 1;
2751 }
2752 
2753 int os::numa_get_group_id() {
2754   int cpu_id = Linux::sched_getcpu();
2755   if (cpu_id != -1) {
2756     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2757     if (lgrp_id != -1) {
2758       return lgrp_id;
2759     }
2760   }
2761   return 0;
2762 }
2763 
2764 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2765   for (size_t i = 0; i < size; i++) {
2766     ids[i] = i;
2767   }
2768   return size;
2769 }
2770 
2771 bool os::get_page_info(char *start, page_info* info) {
2772   return false;
2773 }
2774 
2775 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2776   return end;
2777 }
2778 
2779 
2780 int os::Linux::sched_getcpu_syscall(void) {
2781   unsigned int cpu;
2782   int retval = -1;
2783 
2784 #if defined(IA32)
2785 # ifndef SYS_getcpu
2786 # define SYS_getcpu 318
2787 # endif
2788   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2789 #elif defined(AMD64)
2790 // Unfortunately we have to bring all these macros here from vsyscall.h
2791 // to be able to compile on old linuxes.
2792 # define __NR_vgetcpu 2
2793 # define VSYSCALL_START (-10UL << 20)
2794 # define VSYSCALL_SIZE 1024
2795 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2796   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2797   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2798   retval = vgetcpu(&cpu, NULL, NULL);
2799 #endif
2800 
2801   return (retval == -1) ? retval : cpu;
2802 }
2803 
2804 // Something to do with the numa-aware allocator needs these symbols
2805 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2806 extern "C" JNIEXPORT void numa_error(char *where) { }
2807 extern "C" JNIEXPORT int fork1() { return fork(); }
2808 
2809 
2810 // If we are running with libnuma version > 2, then we should
2811 // be trying to use symbols with versions 1.1
2812 // If we are running with earlier version, which did not have symbol versions,
2813 // we should use the base version.
2814 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2815   void *f = dlvsym(handle, name, "libnuma_1.1");
2816   if (f == NULL) {
2817     f = dlsym(handle, name);
2818   }
2819   return f;
2820 }
2821 
2822 bool os::Linux::libnuma_init() {
2823   // sched_getcpu() should be in libc.
2824   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2825                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2826 
2827   // If it's not, try a direct syscall.
2828   if (sched_getcpu() == -1)
2829     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2830 
2831   if (sched_getcpu() != -1) { // Does it work?
2832     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2833     if (handle != NULL) {
2834       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2835                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2836       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2837                                        libnuma_dlsym(handle, "numa_max_node")));
2838       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2839                                         libnuma_dlsym(handle, "numa_available")));
2840       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2841                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2842       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2843                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2844       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2845                                             libnuma_dlsym(handle, "numa_set_bind_policy")));
2846 
2847 
2848       if (numa_available() != -1) {
2849         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2850         // Create a cpu -> node mapping
2851         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2852         rebuild_cpu_to_node_map();
2853         return true;
2854       }
2855     }
2856   }
2857   return false;
2858 }
2859 
2860 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2861 // The table is later used in get_node_by_cpu().
2862 void os::Linux::rebuild_cpu_to_node_map() {
2863   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2864                               // in libnuma (possible values are starting from 16,
2865                               // and continuing up with every other power of 2, but less
2866                               // than the maximum number of CPUs supported by kernel), and
2867                               // is a subject to change (in libnuma version 2 the requirements
2868                               // are more reasonable) we'll just hardcode the number they use
2869                               // in the library.
2870   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2871 
2872   size_t cpu_num = os::active_processor_count();
2873   size_t cpu_map_size = NCPUS / BitsPerCLong;
2874   size_t cpu_map_valid_size =
2875     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2876 
2877   cpu_to_node()->clear();
2878   cpu_to_node()->at_grow(cpu_num - 1);
2879   size_t node_num = numa_get_groups_num();
2880 
2881   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2882   for (size_t i = 0; i < node_num; i++) {
2883     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2884       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2885         if (cpu_map[j] != 0) {
2886           for (size_t k = 0; k < BitsPerCLong; k++) {
2887             if (cpu_map[j] & (1UL << k)) {
2888               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2889             }
2890           }
2891         }
2892       }
2893     }
2894   }
2895   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2896 }
2897 
2898 int os::Linux::get_node_by_cpu(int cpu_id) {
2899   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2900     return cpu_to_node()->at(cpu_id);
2901   }
2902   return -1;
2903 }
2904 
2905 GrowableArray<int>* os::Linux::_cpu_to_node;
2906 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2907 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2908 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2909 os::Linux::numa_available_func_t os::Linux::_numa_available;
2910 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2911 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2912 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2913 unsigned long* os::Linux::_numa_all_nodes;
2914 
2915 bool os::pd_uncommit_memory(char* addr, size_t size) {
2916   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2917                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2918   return res  != (uintptr_t) MAP_FAILED;
2919 }
2920 
2921 static
2922 address get_stack_commited_bottom(address bottom, size_t size) {
2923   address nbot = bottom;
2924   address ntop = bottom + size;
2925 
2926   size_t page_sz = os::vm_page_size();
2927   unsigned pages = size / page_sz;
2928 
2929   unsigned char vec[1];
2930   unsigned imin = 1, imax = pages + 1, imid;
2931   int mincore_return_value = 0;
2932 
2933   assert(imin <= imax, "Unexpected page size");
2934 
2935   while (imin < imax) {
2936     imid = (imax + imin) / 2;
2937     nbot = ntop - (imid * page_sz);
2938 
2939     // Use a trick with mincore to check whether the page is mapped or not.
2940     // mincore sets vec to 1 if page resides in memory and to 0 if page
2941     // is swapped output but if page we are asking for is unmapped
2942     // it returns -1,ENOMEM
2943     mincore_return_value = mincore(nbot, page_sz, vec);
2944 
2945     if (mincore_return_value == -1) {
2946       // Page is not mapped go up
2947       // to find first mapped page
2948       if (errno != EAGAIN) {
2949         assert(errno == ENOMEM, "Unexpected mincore errno");
2950         imax = imid;
2951       }
2952     } else {
2953       // Page is mapped go down
2954       // to find first not mapped page
2955       imin = imid + 1;
2956     }
2957   }
2958 
2959   nbot = nbot + page_sz;
2960 
2961   // Adjust stack bottom one page up if last checked page is not mapped
2962   if (mincore_return_value == -1) {
2963     nbot = nbot + page_sz;
2964   }
2965 
2966   return nbot;
2967 }
2968 
2969 
2970 // Linux uses a growable mapping for the stack, and if the mapping for
2971 // the stack guard pages is not removed when we detach a thread the
2972 // stack cannot grow beyond the pages where the stack guard was
2973 // mapped.  If at some point later in the process the stack expands to
2974 // that point, the Linux kernel cannot expand the stack any further
2975 // because the guard pages are in the way, and a segfault occurs.
2976 //
2977 // However, it's essential not to split the stack region by unmapping
2978 // a region (leaving a hole) that's already part of the stack mapping,
2979 // so if the stack mapping has already grown beyond the guard pages at
2980 // the time we create them, we have to truncate the stack mapping.
2981 // So, we need to know the extent of the stack mapping when
2982 // create_stack_guard_pages() is called.
2983 
2984 // We only need this for stacks that are growable: at the time of
2985 // writing thread stacks don't use growable mappings (i.e. those
2986 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2987 // only applies to the main thread.
2988 
2989 // If the (growable) stack mapping already extends beyond the point
2990 // where we're going to put our guard pages, truncate the mapping at
2991 // that point by munmap()ping it.  This ensures that when we later
2992 // munmap() the guard pages we don't leave a hole in the stack
2993 // mapping. This only affects the main/initial thread
2994 
2995 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2996 
2997   if (os::Linux::is_initial_thread()) {
2998     // As we manually grow stack up to bottom inside create_attached_thread(),
2999     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3000     // we don't need to do anything special.
3001     // Check it first, before calling heavy function.
3002     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3003     unsigned char vec[1];
3004 
3005     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3006       // Fallback to slow path on all errors, including EAGAIN
3007       stack_extent = (uintptr_t) get_stack_commited_bottom(
3008                                     os::Linux::initial_thread_stack_bottom(),
3009                                     (size_t)addr - stack_extent);
3010     }
3011 
3012     if (stack_extent < (uintptr_t)addr) {
3013       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3014     }
3015   }
3016 
3017   return os::commit_memory(addr, size, !ExecMem);
3018 }
3019 
3020 // If this is a growable mapping, remove the guard pages entirely by
3021 // munmap()ping them.  If not, just call uncommit_memory(). This only
3022 // affects the main/initial thread, but guard against future OS changes
3023 // It's safe to always unmap guard pages for initial thread because we
3024 // always place it right after end of the mapped region
3025 
3026 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3027   uintptr_t stack_extent, stack_base;
3028 
3029   if (os::Linux::is_initial_thread()) {
3030     return ::munmap(addr, size) == 0;
3031   }
3032 
3033   return os::uncommit_memory(addr, size);
3034 }
3035 
3036 static address _highest_vm_reserved_address = NULL;
3037 
3038 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3039 // at 'requested_addr'. If there are existing memory mappings at the same
3040 // location, however, they will be overwritten. If 'fixed' is false,
3041 // 'requested_addr' is only treated as a hint, the return value may or
3042 // may not start from the requested address. Unlike Linux mmap(), this
3043 // function returns NULL to indicate failure.
3044 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3045   char * addr;
3046   int flags;
3047 
3048   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3049   if (fixed) {
3050     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3051     flags |= MAP_FIXED;
3052   }
3053 
3054   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3055   // touch an uncommitted page. Otherwise, the read/write might
3056   // succeed if we have enough swap space to back the physical page.
3057   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3058                        flags, -1, 0);
3059 
3060   if (addr != MAP_FAILED) {
3061     // anon_mmap() should only get called during VM initialization,
3062     // don't need lock (actually we can skip locking even it can be called
3063     // from multiple threads, because _highest_vm_reserved_address is just a
3064     // hint about the upper limit of non-stack memory regions.)
3065     if ((address)addr + bytes > _highest_vm_reserved_address) {
3066       _highest_vm_reserved_address = (address)addr + bytes;
3067     }
3068   }
3069 
3070   return addr == MAP_FAILED ? NULL : addr;
3071 }
3072 
3073 // Don't update _highest_vm_reserved_address, because there might be memory
3074 // regions above addr + size. If so, releasing a memory region only creates
3075 // a hole in the address space, it doesn't help prevent heap-stack collision.
3076 //
3077 static int anon_munmap(char * addr, size_t size) {
3078   return ::munmap(addr, size) == 0;
3079 }
3080 
3081 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3082                          size_t alignment_hint) {
3083   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3084 }
3085 
3086 bool os::pd_release_memory(char* addr, size_t size) {
3087   return anon_munmap(addr, size);
3088 }
3089 
3090 static address highest_vm_reserved_address() {
3091   return _highest_vm_reserved_address;
3092 }
3093 
3094 static bool linux_mprotect(char* addr, size_t size, int prot) {
3095   // Linux wants the mprotect address argument to be page aligned.
3096   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3097 
3098   // According to SUSv3, mprotect() should only be used with mappings
3099   // established by mmap(), and mmap() always maps whole pages. Unaligned
3100   // 'addr' likely indicates problem in the VM (e.g. trying to change
3101   // protection of malloc'ed or statically allocated memory). Check the
3102   // caller if you hit this assert.
3103   assert(addr == bottom, "sanity check");
3104 
3105   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3106   return ::mprotect(bottom, size, prot) == 0;
3107 }
3108 
3109 // Set protections specified
3110 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3111                         bool is_committed) {
3112   unsigned int p = 0;
3113   switch (prot) {
3114   case MEM_PROT_NONE: p = PROT_NONE; break;
3115   case MEM_PROT_READ: p = PROT_READ; break;
3116   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3117   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3118   default:
3119     ShouldNotReachHere();
3120   }
3121   // is_committed is unused.
3122   return linux_mprotect(addr, bytes, p);
3123 }
3124 
3125 bool os::guard_memory(char* addr, size_t size) {
3126   return linux_mprotect(addr, size, PROT_NONE);
3127 }
3128 
3129 bool os::unguard_memory(char* addr, size_t size) {
3130   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3131 }
3132 
3133 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3134   bool result = false;
3135   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3136                  MAP_ANONYMOUS|MAP_PRIVATE,
3137                  -1, 0);
3138   if (p != MAP_FAILED) {
3139     void *aligned_p = align_ptr_up(p, page_size);
3140 
3141     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3142 
3143     munmap(p, page_size * 2);
3144   }
3145 
3146   if (warn && !result) {
3147     warning("TransparentHugePages is not supported by the operating system.");
3148   }
3149 
3150   return result;
3151 }
3152 
3153 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3154   bool result = false;
3155   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3156                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3157                  -1, 0);
3158 
3159   if (p != MAP_FAILED) {
3160     // We don't know if this really is a huge page or not.
3161     FILE *fp = fopen("/proc/self/maps", "r");
3162     if (fp) {
3163       while (!feof(fp)) {
3164         char chars[257];
3165         long x = 0;
3166         if (fgets(chars, sizeof(chars), fp)) {
3167           if (sscanf(chars, "%lx-%*x", &x) == 1
3168               && x == (long)p) {
3169             if (strstr (chars, "hugepage")) {
3170               result = true;
3171               break;
3172             }
3173           }
3174         }
3175       }
3176       fclose(fp);
3177     }
3178     munmap(p, page_size);
3179   }
3180 
3181   if (warn && !result) {
3182     warning("HugeTLBFS is not supported by the operating system.");
3183   }
3184 
3185   return result;
3186 }
3187 
3188 /*
3189 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3190 *
3191 * From the coredump_filter documentation:
3192 *
3193 * - (bit 0) anonymous private memory
3194 * - (bit 1) anonymous shared memory
3195 * - (bit 2) file-backed private memory
3196 * - (bit 3) file-backed shared memory
3197 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3198 *           effective only if the bit 2 is cleared)
3199 * - (bit 5) hugetlb private memory
3200 * - (bit 6) hugetlb shared memory
3201 */
3202 static void set_coredump_filter(void) {
3203   FILE *f;
3204   long cdm;
3205 
3206   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3207     return;
3208   }
3209 
3210   if (fscanf(f, "%lx", &cdm) != 1) {
3211     fclose(f);
3212     return;
3213   }
3214 
3215   rewind(f);
3216 
3217   if ((cdm & LARGEPAGES_BIT) == 0) {
3218     cdm |= LARGEPAGES_BIT;
3219     fprintf(f, "%#lx", cdm);
3220   }
3221 
3222   fclose(f);
3223 }
3224 
3225 // Large page support
3226 
3227 static size_t _large_page_size = 0;
3228 
3229 size_t os::Linux::find_large_page_size() {
3230   size_t large_page_size = 0;
3231 
3232   // large_page_size on Linux is used to round up heap size. x86 uses either
3233   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3234   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3235   // page as large as 256M.
3236   //
3237   // Here we try to figure out page size by parsing /proc/meminfo and looking
3238   // for a line with the following format:
3239   //    Hugepagesize:     2048 kB
3240   //
3241   // If we can't determine the value (e.g. /proc is not mounted, or the text
3242   // format has been changed), we'll use the largest page size supported by
3243   // the processor.
3244 
3245 #ifndef ZERO
3246   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3247                      ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3248 #endif // ZERO
3249 
3250   FILE *fp = fopen("/proc/meminfo", "r");
3251   if (fp) {
3252     while (!feof(fp)) {
3253       int x = 0;
3254       char buf[16];
3255       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3256         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3257           large_page_size = x * K;
3258           break;
3259         }
3260       } else {
3261         // skip to next line
3262         for (;;) {
3263           int ch = fgetc(fp);
3264           if (ch == EOF || ch == (int)'\n') break;
3265         }
3266       }
3267     }
3268     fclose(fp);
3269   }
3270 
3271   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3272     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3273         SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3274         proper_unit_for_byte_size(large_page_size));
3275   }
3276 
3277   return large_page_size;
3278 }
3279 
3280 size_t os::Linux::setup_large_page_size() {
3281   _large_page_size = Linux::find_large_page_size();
3282   const size_t default_page_size = (size_t)Linux::page_size();
3283   if (_large_page_size > default_page_size) {
3284     _page_sizes[0] = _large_page_size;
3285     _page_sizes[1] = default_page_size;
3286     _page_sizes[2] = 0;
3287   }
3288 
3289   return _large_page_size;
3290 }
3291 
3292 bool os::Linux::setup_large_page_type(size_t page_size) {
3293   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3294       FLAG_IS_DEFAULT(UseSHM) &&
3295       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3296 
3297     // The type of large pages has not been specified by the user.
3298 
3299     // Try UseHugeTLBFS and then UseSHM.
3300     UseHugeTLBFS = UseSHM = true;
3301 
3302     // Don't try UseTransparentHugePages since there are known
3303     // performance issues with it turned on. This might change in the future.
3304     UseTransparentHugePages = false;
3305   }
3306 
3307   if (UseTransparentHugePages) {
3308     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3309     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3310       UseHugeTLBFS = false;
3311       UseSHM = false;
3312       return true;
3313     }
3314     UseTransparentHugePages = false;
3315   }
3316 
3317   if (UseHugeTLBFS) {
3318     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3319     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3320       UseSHM = false;
3321       return true;
3322     }
3323     UseHugeTLBFS = false;
3324   }
3325 
3326   return UseSHM;
3327 }
3328 
3329 void os::large_page_init() {
3330   if (!UseLargePages &&
3331       !UseTransparentHugePages &&
3332       !UseHugeTLBFS &&
3333       !UseSHM) {
3334     // Not using large pages.
3335     return;
3336   }
3337 
3338   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3339     // The user explicitly turned off large pages.
3340     // Ignore the rest of the large pages flags.
3341     UseTransparentHugePages = false;
3342     UseHugeTLBFS = false;
3343     UseSHM = false;
3344     return;
3345   }
3346 
3347   size_t large_page_size = Linux::setup_large_page_size();
3348   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3349 
3350   set_coredump_filter();
3351 }
3352 
3353 #ifndef SHM_HUGETLB
3354 #define SHM_HUGETLB 04000
3355 #endif
3356 
3357 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3358   // "exec" is passed in but not used.  Creating the shared image for
3359   // the code cache doesn't have an SHM_X executable permission to check.
3360   assert(UseLargePages && UseSHM, "only for SHM large pages");
3361   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3362 
3363   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3364     return NULL; // Fallback to small pages.
3365   }
3366 
3367   key_t key = IPC_PRIVATE;
3368   char *addr;
3369 
3370   bool warn_on_failure = UseLargePages &&
3371                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3372                          !FLAG_IS_DEFAULT(UseSHM) ||
3373                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3374                         );
3375   char msg[128];
3376 
3377   // Create a large shared memory region to attach to based on size.
3378   // Currently, size is the total size of the heap
3379   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3380   if (shmid == -1) {
3381      // Possible reasons for shmget failure:
3382      // 1. shmmax is too small for Java heap.
3383      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3384      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3385      // 2. not enough large page memory.
3386      //    > check available large pages: cat /proc/meminfo
3387      //    > increase amount of large pages:
3388      //          echo new_value > /proc/sys/vm/nr_hugepages
3389      //      Note 1: different Linux may use different name for this property,
3390      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3391      //      Note 2: it's possible there's enough physical memory available but
3392      //            they are so fragmented after a long run that they can't
3393      //            coalesce into large pages. Try to reserve large pages when
3394      //            the system is still "fresh".
3395      if (warn_on_failure) {
3396        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3397        warning("%s", msg);
3398      }
3399      return NULL;
3400   }
3401 
3402   // attach to the region
3403   addr = (char*)shmat(shmid, req_addr, 0);
3404   int err = errno;
3405 
3406   // Remove shmid. If shmat() is successful, the actual shared memory segment
3407   // will be deleted when it's detached by shmdt() or when the process
3408   // terminates. If shmat() is not successful this will remove the shared
3409   // segment immediately.
3410   shmctl(shmid, IPC_RMID, NULL);
3411 
3412   if ((intptr_t)addr == -1) {
3413      if (warn_on_failure) {
3414        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3415        warning("%s", msg);
3416      }
3417      return NULL;
3418   }
3419 
3420   return addr;
3421 }
3422 
3423 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3424   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3425 
3426   bool warn_on_failure = UseLargePages &&
3427       (!FLAG_IS_DEFAULT(UseLargePages) ||
3428        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3429        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3430 
3431   if (warn_on_failure) {
3432     char msg[128];
3433     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3434         PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3435     warning("%s", msg);
3436   }
3437 }
3438 
3439 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3440   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3441   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3442   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3443 
3444   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3445   char* addr = (char*)::mmap(req_addr, bytes, prot,
3446                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3447                              -1, 0);
3448 
3449   if (addr == MAP_FAILED) {
3450     warn_on_large_pages_failure(req_addr, bytes, errno);
3451     return NULL;
3452   }
3453 
3454   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3455 
3456   return addr;
3457 }
3458 
3459 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3460   size_t large_page_size = os::large_page_size();
3461 
3462   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3463 
3464   // Allocate small pages.
3465 
3466   char* start;
3467   if (req_addr != NULL) {
3468     assert(is_ptr_aligned(req_addr, alignment), "Must be");
3469     assert(is_size_aligned(bytes, alignment), "Must be");
3470     start = os::reserve_memory(bytes, req_addr);
3471     assert(start == NULL || start == req_addr, "Must be");
3472   } else {
3473     start = os::reserve_memory_aligned(bytes, alignment);
3474   }
3475 
3476   if (start == NULL) {
3477     return NULL;
3478   }
3479 
3480   assert(is_ptr_aligned(start, alignment), "Must be");
3481 
3482   if (MemTracker::tracking_level() > NMT_minimal) {
3483     // os::reserve_memory_special will record this memory area.
3484     // Need to release it here to prevent overlapping reservations.
3485     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3486     tkr.record((address)start, bytes);
3487   }
3488 
3489   char* end = start + bytes;
3490 
3491   // Find the regions of the allocated chunk that can be promoted to large pages.
3492   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3493   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3494 
3495   size_t lp_bytes = lp_end - lp_start;
3496 
3497   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3498 
3499   if (lp_bytes == 0) {
3500     // The mapped region doesn't even span the start and the end of a large page.
3501     // Fall back to allocate a non-special area.
3502     ::munmap(start, end - start);
3503     return NULL;
3504   }
3505 
3506   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3507 
3508 
3509   void* result;
3510 
3511   if (start != lp_start) {
3512     result = ::mmap(start, lp_start - start, prot,
3513                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3514                     -1, 0);
3515     if (result == MAP_FAILED) {
3516       ::munmap(lp_start, end - lp_start);
3517       return NULL;
3518     }
3519   }
3520 
3521   result = ::mmap(lp_start, lp_bytes, prot,
3522                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3523                   -1, 0);
3524   if (result == MAP_FAILED) {
3525     warn_on_large_pages_failure(req_addr, bytes, errno);
3526     // If the mmap above fails, the large pages region will be unmapped and we
3527     // have regions before and after with small pages. Release these regions.
3528     //
3529     // |  mapped  |  unmapped  |  mapped  |
3530     // ^          ^            ^          ^
3531     // start      lp_start     lp_end     end
3532     //
3533     ::munmap(start, lp_start - start);
3534     ::munmap(lp_end, end - lp_end);
3535     return NULL;
3536   }
3537 
3538   if (lp_end != end) {
3539       result = ::mmap(lp_end, end - lp_end, prot,
3540                       MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3541                       -1, 0);
3542     if (result == MAP_FAILED) {
3543       ::munmap(start, lp_end - start);
3544       return NULL;
3545     }
3546   }
3547 
3548   return start;
3549 }
3550 
3551 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3552   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3553   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3554   assert(is_power_of_2(alignment), "Must be");
3555   assert(is_power_of_2(os::large_page_size()), "Must be");
3556   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3557 
3558   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3559     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3560   } else {
3561     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3562   }
3563 }
3564 
3565 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3566   assert(UseLargePages, "only for large pages");
3567 
3568   char* addr;
3569   if (UseSHM) {
3570     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3571   } else {
3572     assert(UseHugeTLBFS, "must be");
3573     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3574   }
3575 
3576   if (addr != NULL) {
3577     if (UseNUMAInterleaving) {
3578       numa_make_global(addr, bytes);
3579     }
3580 
3581     // The memory is committed
3582     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3583   }
3584 
3585   return addr;
3586 }
3587 
3588 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3589   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3590   return shmdt(base) == 0;
3591 }
3592 
3593 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3594   return pd_release_memory(base, bytes);
3595 }
3596 
3597 bool os::release_memory_special(char* base, size_t bytes) {
3598   bool res;
3599   if (MemTracker::tracking_level() > NMT_minimal) {
3600     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3601     res = os::Linux::release_memory_special_impl(base, bytes);
3602     if (res) {
3603       tkr.record((address)base, bytes);
3604     }
3605 
3606   } else {
3607     res = os::Linux::release_memory_special_impl(base, bytes);
3608   }
3609   return res;
3610 }
3611 
3612 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3613   assert(UseLargePages, "only for large pages");
3614   bool res;
3615 
3616   if (UseSHM) {
3617     res = os::Linux::release_memory_special_shm(base, bytes);
3618   } else {
3619     assert(UseHugeTLBFS, "must be");
3620     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3621   }
3622   return res;
3623 }
3624 
3625 size_t os::large_page_size() {
3626   return _large_page_size;
3627 }
3628 
3629 // With SysV SHM the entire memory region must be allocated as shared
3630 // memory.
3631 // HugeTLBFS allows application to commit large page memory on demand.
3632 // However, when committing memory with HugeTLBFS fails, the region
3633 // that was supposed to be committed will lose the old reservation
3634 // and allow other threads to steal that memory region. Because of this
3635 // behavior we can't commit HugeTLBFS memory.
3636 bool os::can_commit_large_page_memory() {
3637   return UseTransparentHugePages;
3638 }
3639 
3640 bool os::can_execute_large_page_memory() {
3641   return UseTransparentHugePages || UseHugeTLBFS;
3642 }
3643 
3644 // Reserve memory at an arbitrary address, only if that area is
3645 // available (and not reserved for something else).
3646 
3647 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3648   const int max_tries = 10;
3649   char* base[max_tries];
3650   size_t size[max_tries];
3651   const size_t gap = 0x000000;
3652 
3653   // Assert only that the size is a multiple of the page size, since
3654   // that's all that mmap requires, and since that's all we really know
3655   // about at this low abstraction level.  If we need higher alignment,
3656   // we can either pass an alignment to this method or verify alignment
3657   // in one of the methods further up the call chain.  See bug 5044738.
3658   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3659 
3660   // Repeatedly allocate blocks until the block is allocated at the
3661   // right spot. Give up after max_tries. Note that reserve_memory() will
3662   // automatically update _highest_vm_reserved_address if the call is
3663   // successful. The variable tracks the highest memory address every reserved
3664   // by JVM. It is used to detect heap-stack collision if running with
3665   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3666   // space than needed, it could confuse the collision detecting code. To
3667   // solve the problem, save current _highest_vm_reserved_address and
3668   // calculate the correct value before return.
3669   address old_highest = _highest_vm_reserved_address;
3670 
3671   // Linux mmap allows caller to pass an address as hint; give it a try first,
3672   // if kernel honors the hint then we can return immediately.
3673   char * addr = anon_mmap(requested_addr, bytes, false);
3674   if (addr == requested_addr) {
3675      return requested_addr;
3676   }
3677 
3678   if (addr != NULL) {
3679      // mmap() is successful but it fails to reserve at the requested address
3680      anon_munmap(addr, bytes);
3681   }
3682 
3683   int i;
3684   for (i = 0; i < max_tries; ++i) {
3685     base[i] = reserve_memory(bytes);
3686 
3687     if (base[i] != NULL) {
3688       // Is this the block we wanted?
3689       if (base[i] == requested_addr) {
3690         size[i] = bytes;
3691         break;
3692       }
3693 
3694       // Does this overlap the block we wanted? Give back the overlapped
3695       // parts and try again.
3696 
3697       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3698       if (top_overlap >= 0 && top_overlap < bytes) {
3699         unmap_memory(base[i], top_overlap);
3700         base[i] += top_overlap;
3701         size[i] = bytes - top_overlap;
3702       } else {
3703         size_t bottom_overlap = base[i] + bytes - requested_addr;
3704         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3705           unmap_memory(requested_addr, bottom_overlap);
3706           size[i] = bytes - bottom_overlap;
3707         } else {
3708           size[i] = bytes;
3709         }
3710       }
3711     }
3712   }
3713 
3714   // Give back the unused reserved pieces.
3715 
3716   for (int j = 0; j < i; ++j) {
3717     if (base[j] != NULL) {
3718       unmap_memory(base[j], size[j]);
3719     }
3720   }
3721 
3722   if (i < max_tries) {
3723     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3724     return requested_addr;
3725   } else {
3726     _highest_vm_reserved_address = old_highest;
3727     return NULL;
3728   }
3729 }
3730 
3731 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3732   return ::read(fd, buf, nBytes);
3733 }
3734 
3735 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3736 // Solaris uses poll(), linux uses park().
3737 // Poll() is likely a better choice, assuming that Thread.interrupt()
3738 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3739 // SIGSEGV, see 4355769.
3740 
3741 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3742   assert(thread == Thread::current(),  "thread consistency check");
3743 
3744   ParkEvent * const slp = thread->_SleepEvent ;
3745   slp->reset() ;
3746   OrderAccess::fence() ;
3747 
3748   if (interruptible) {
3749     jlong prevtime = javaTimeNanos();
3750 
3751     for (;;) {
3752       if (os::is_interrupted(thread, true)) {
3753         return OS_INTRPT;
3754       }
3755 
3756       jlong newtime = javaTimeNanos();
3757 
3758       if (newtime - prevtime < 0) {
3759         // time moving backwards, should only happen if no monotonic clock
3760         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3761         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3762       } else {
3763         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3764       }
3765 
3766       if(millis <= 0) {
3767         return OS_OK;
3768       }
3769 
3770       prevtime = newtime;
3771 
3772       {
3773         assert(thread->is_Java_thread(), "sanity check");
3774         JavaThread *jt = (JavaThread *) thread;
3775         ThreadBlockInVM tbivm(jt);
3776         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3777 
3778         jt->set_suspend_equivalent();
3779         // cleared by handle_special_suspend_equivalent_condition() or
3780         // java_suspend_self() via check_and_wait_while_suspended()
3781 
3782         slp->park(millis);
3783 
3784         // were we externally suspended while we were waiting?
3785         jt->check_and_wait_while_suspended();
3786       }
3787     }
3788   } else {
3789     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3790     jlong prevtime = javaTimeNanos();
3791 
3792     for (;;) {
3793       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3794       // the 1st iteration ...
3795       jlong newtime = javaTimeNanos();
3796 
3797       if (newtime - prevtime < 0) {
3798         // time moving backwards, should only happen if no monotonic clock
3799         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3800         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3801       } else {
3802         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3803       }
3804 
3805       if(millis <= 0) break ;
3806 
3807       prevtime = newtime;
3808       slp->park(millis);
3809     }
3810     return OS_OK ;
3811   }
3812 }
3813 
3814 //
3815 // Short sleep, direct OS call.
3816 //
3817 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3818 // sched_yield(2) will actually give up the CPU:
3819 //
3820 //   * Alone on this pariticular CPU, keeps running.
3821 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3822 //     (pre 2.6.39).
3823 //
3824 // So calling this with 0 is an alternative.
3825 //
3826 void os::naked_short_sleep(jlong ms) {
3827   struct timespec req;
3828 
3829   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3830   req.tv_sec = 0;
3831   if (ms > 0) {
3832     req.tv_nsec = (ms % 1000) * 1000000;
3833   }
3834   else {
3835     req.tv_nsec = 1;
3836   }
3837 
3838   nanosleep(&req, NULL);
3839 
3840   return;
3841 }
3842 
3843 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3844 void os::infinite_sleep() {
3845   while (true) {    // sleep forever ...
3846     ::sleep(100);   // ... 100 seconds at a time
3847   }
3848 }
3849 
3850 // Used to convert frequent JVM_Yield() to nops
3851 bool os::dont_yield() {
3852   return DontYieldALot;
3853 }
3854 
3855 void os::yield() {
3856   sched_yield();
3857 }
3858 
3859 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3860 
3861 void os::yield_all(int attempts) {
3862   // Yields to all threads, including threads with lower priorities
3863   // Threads on Linux are all with same priority. The Solaris style
3864   // os::yield_all() with nanosleep(1ms) is not necessary.
3865   sched_yield();
3866 }
3867 
3868 // Called from the tight loops to possibly influence time-sharing heuristics
3869 void os::loop_breaker(int attempts) {
3870   os::yield_all(attempts);
3871 }
3872 
3873 ////////////////////////////////////////////////////////////////////////////////
3874 // thread priority support
3875 
3876 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3877 // only supports dynamic priority, static priority must be zero. For real-time
3878 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3879 // However, for large multi-threaded applications, SCHED_RR is not only slower
3880 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3881 // of 5 runs - Sep 2005).
3882 //
3883 // The following code actually changes the niceness of kernel-thread/LWP. It
3884 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3885 // not the entire user process, and user level threads are 1:1 mapped to kernel
3886 // threads. It has always been the case, but could change in the future. For
3887 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3888 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3889 
3890 int os::java_to_os_priority[CriticalPriority + 1] = {
3891   19,              // 0 Entry should never be used
3892 
3893    4,              // 1 MinPriority
3894    3,              // 2
3895    2,              // 3
3896 
3897    1,              // 4
3898    0,              // 5 NormPriority
3899   -1,              // 6
3900 
3901   -2,              // 7
3902   -3,              // 8
3903   -4,              // 9 NearMaxPriority
3904 
3905   -5,              // 10 MaxPriority
3906 
3907   -5               // 11 CriticalPriority
3908 };
3909 
3910 static int prio_init() {
3911   if (ThreadPriorityPolicy == 1) {
3912     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3913     // if effective uid is not root. Perhaps, a more elegant way of doing
3914     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3915     if (geteuid() != 0) {
3916       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3917         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3918       }
3919       ThreadPriorityPolicy = 0;
3920     }
3921   }
3922   if (UseCriticalJavaThreadPriority) {
3923     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3924   }
3925   return 0;
3926 }
3927 
3928 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3929   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3930 
3931   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3932   return (ret == 0) ? OS_OK : OS_ERR;
3933 }
3934 
3935 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3936   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3937     *priority_ptr = java_to_os_priority[NormPriority];
3938     return OS_OK;
3939   }
3940 
3941   errno = 0;
3942   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3943   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3944 }
3945 
3946 // Hint to the underlying OS that a task switch would not be good.
3947 // Void return because it's a hint and can fail.
3948 void os::hint_no_preempt() {}
3949 
3950 ////////////////////////////////////////////////////////////////////////////////
3951 // suspend/resume support
3952 
3953 //  the low-level signal-based suspend/resume support is a remnant from the
3954 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3955 //  within hotspot. Now there is a single use-case for this:
3956 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3957 //      that runs in the watcher thread.
3958 //  The remaining code is greatly simplified from the more general suspension
3959 //  code that used to be used.
3960 //
3961 //  The protocol is quite simple:
3962 //  - suspend:
3963 //      - sends a signal to the target thread
3964 //      - polls the suspend state of the osthread using a yield loop
3965 //      - target thread signal handler (SR_handler) sets suspend state
3966 //        and blocks in sigsuspend until continued
3967 //  - resume:
3968 //      - sets target osthread state to continue
3969 //      - sends signal to end the sigsuspend loop in the SR_handler
3970 //
3971 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3972 //
3973 
3974 static void resume_clear_context(OSThread *osthread) {
3975   osthread->set_ucontext(NULL);
3976   osthread->set_siginfo(NULL);
3977 }
3978 
3979 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3980   osthread->set_ucontext(context);
3981   osthread->set_siginfo(siginfo);
3982 }
3983 
3984 //
3985 // Handler function invoked when a thread's execution is suspended or
3986 // resumed. We have to be careful that only async-safe functions are
3987 // called here (Note: most pthread functions are not async safe and
3988 // should be avoided.)
3989 //
3990 // Note: sigwait() is a more natural fit than sigsuspend() from an
3991 // interface point of view, but sigwait() prevents the signal hander
3992 // from being run. libpthread would get very confused by not having
3993 // its signal handlers run and prevents sigwait()'s use with the
3994 // mutex granting granting signal.
3995 //
3996 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3997 //
3998 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3999   // Save and restore errno to avoid confusing native code with EINTR
4000   // after sigsuspend.
4001   int old_errno = errno;
4002 
4003   Thread* thread = Thread::current();
4004   OSThread* osthread = thread->osthread();
4005   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4006 
4007   os::SuspendResume::State current = osthread->sr.state();
4008   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4009     suspend_save_context(osthread, siginfo, context);
4010 
4011     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4012     os::SuspendResume::State state = osthread->sr.suspended();
4013     if (state == os::SuspendResume::SR_SUSPENDED) {
4014       sigset_t suspend_set;  // signals for sigsuspend()
4015 
4016       // get current set of blocked signals and unblock resume signal
4017       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4018       sigdelset(&suspend_set, SR_signum);
4019 
4020       sr_semaphore.signal();
4021       // wait here until we are resumed
4022       while (1) {
4023         sigsuspend(&suspend_set);
4024 
4025         os::SuspendResume::State result = osthread->sr.running();
4026         if (result == os::SuspendResume::SR_RUNNING) {
4027           sr_semaphore.signal();
4028           break;
4029         }
4030       }
4031 
4032     } else if (state == os::SuspendResume::SR_RUNNING) {
4033       // request was cancelled, continue
4034     } else {
4035       ShouldNotReachHere();
4036     }
4037 
4038     resume_clear_context(osthread);
4039   } else if (current == os::SuspendResume::SR_RUNNING) {
4040     // request was cancelled, continue
4041   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4042     // ignore
4043   } else {
4044     // ignore
4045   }
4046 
4047   errno = old_errno;
4048 }
4049 
4050 
4051 static int SR_initialize() {
4052   struct sigaction act;
4053   char *s;
4054   /* Get signal number to use for suspend/resume */
4055   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4056     int sig = ::strtol(s, 0, 10);
4057     if (sig > 0 || sig < _NSIG) {
4058         SR_signum = sig;
4059     }
4060   }
4061 
4062   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4063         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4064 
4065   sigemptyset(&SR_sigset);
4066   sigaddset(&SR_sigset, SR_signum);
4067 
4068   /* Set up signal handler for suspend/resume */
4069   act.sa_flags = SA_RESTART|SA_SIGINFO;
4070   act.sa_handler = (void (*)(int)) SR_handler;
4071 
4072   // SR_signum is blocked by default.
4073   // 4528190 - We also need to block pthread restart signal (32 on all
4074   // supported Linux platforms). Note that LinuxThreads need to block
4075   // this signal for all threads to work properly. So we don't have
4076   // to use hard-coded signal number when setting up the mask.
4077   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4078 
4079   if (sigaction(SR_signum, &act, 0) == -1) {
4080     return -1;
4081   }
4082 
4083   // Save signal flag
4084   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4085   return 0;
4086 }
4087 
4088 static int sr_notify(OSThread* osthread) {
4089   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4090   assert_status(status == 0, status, "pthread_kill");
4091   return status;
4092 }
4093 
4094 // "Randomly" selected value for how long we want to spin
4095 // before bailing out on suspending a thread, also how often
4096 // we send a signal to a thread we want to resume
4097 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4098 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4099 
4100 // returns true on success and false on error - really an error is fatal
4101 // but this seems the normal response to library errors
4102 static bool do_suspend(OSThread* osthread) {
4103   assert(osthread->sr.is_running(), "thread should be running");
4104   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4105 
4106   // mark as suspended and send signal
4107   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4108     // failed to switch, state wasn't running?
4109     ShouldNotReachHere();
4110     return false;
4111   }
4112 
4113   if (sr_notify(osthread) != 0) {
4114     ShouldNotReachHere();
4115   }
4116 
4117   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4118   while (true) {
4119     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4120       break;
4121     } else {
4122       // timeout
4123       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4124       if (cancelled == os::SuspendResume::SR_RUNNING) {
4125         return false;
4126       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4127         // make sure that we consume the signal on the semaphore as well
4128         sr_semaphore.wait();
4129         break;
4130       } else {
4131         ShouldNotReachHere();
4132         return false;
4133       }
4134     }
4135   }
4136 
4137   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4138   return true;
4139 }
4140 
4141 static void do_resume(OSThread* osthread) {
4142   assert(osthread->sr.is_suspended(), "thread should be suspended");
4143   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4144 
4145   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4146     // failed to switch to WAKEUP_REQUEST
4147     ShouldNotReachHere();
4148     return;
4149   }
4150 
4151   while (true) {
4152     if (sr_notify(osthread) == 0) {
4153       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4154         if (osthread->sr.is_running()) {
4155           return;
4156         }
4157       }
4158     } else {
4159       ShouldNotReachHere();
4160     }
4161   }
4162 
4163   guarantee(osthread->sr.is_running(), "Must be running!");
4164 }
4165 
4166 ////////////////////////////////////////////////////////////////////////////////
4167 // interrupt support
4168 
4169 void os::interrupt(Thread* thread) {
4170   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4171     "possibility of dangling Thread pointer");
4172 
4173   OSThread* osthread = thread->osthread();
4174 
4175   if (!osthread->interrupted()) {
4176     osthread->set_interrupted(true);
4177     // More than one thread can get here with the same value of osthread,
4178     // resulting in multiple notifications.  We do, however, want the store
4179     // to interrupted() to be visible to other threads before we execute unpark().
4180     OrderAccess::fence();
4181     ParkEvent * const slp = thread->_SleepEvent ;
4182     if (slp != NULL) slp->unpark() ;
4183   }
4184 
4185   // For JSR166. Unpark even if interrupt status already was set
4186   if (thread->is_Java_thread())
4187     ((JavaThread*)thread)->parker()->unpark();
4188 
4189   ParkEvent * ev = thread->_ParkEvent ;
4190   if (ev != NULL) ev->unpark() ;
4191 
4192 }
4193 
4194 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4195   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4196     "possibility of dangling Thread pointer");
4197 
4198   OSThread* osthread = thread->osthread();
4199 
4200   bool interrupted = osthread->interrupted();
4201 
4202   if (interrupted && clear_interrupted) {
4203     osthread->set_interrupted(false);
4204     // consider thread->_SleepEvent->reset() ... optional optimization
4205   }
4206 
4207   return interrupted;
4208 }
4209 
4210 ///////////////////////////////////////////////////////////////////////////////////
4211 // signal handling (except suspend/resume)
4212 
4213 // This routine may be used by user applications as a "hook" to catch signals.
4214 // The user-defined signal handler must pass unrecognized signals to this
4215 // routine, and if it returns true (non-zero), then the signal handler must
4216 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4217 // routine will never retun false (zero), but instead will execute a VM panic
4218 // routine kill the process.
4219 //
4220 // If this routine returns false, it is OK to call it again.  This allows
4221 // the user-defined signal handler to perform checks either before or after
4222 // the VM performs its own checks.  Naturally, the user code would be making
4223 // a serious error if it tried to handle an exception (such as a null check
4224 // or breakpoint) that the VM was generating for its own correct operation.
4225 //
4226 // This routine may recognize any of the following kinds of signals:
4227 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4228 // It should be consulted by handlers for any of those signals.
4229 //
4230 // The caller of this routine must pass in the three arguments supplied
4231 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4232 // field of the structure passed to sigaction().  This routine assumes that
4233 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4234 //
4235 // Note that the VM will print warnings if it detects conflicting signal
4236 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4237 //
4238 extern "C" JNIEXPORT int
4239 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4240                         void* ucontext, int abort_if_unrecognized);
4241 
4242 void signalHandler(int sig, siginfo_t* info, void* uc) {
4243   assert(info != NULL && uc != NULL, "it must be old kernel");
4244   int orig_errno = errno;  // Preserve errno value over signal handler.
4245   JVM_handle_linux_signal(sig, info, uc, true);
4246   errno = orig_errno;
4247 }
4248 
4249 
4250 // This boolean allows users to forward their own non-matching signals
4251 // to JVM_handle_linux_signal, harmlessly.
4252 bool os::Linux::signal_handlers_are_installed = false;
4253 
4254 // For signal-chaining
4255 struct sigaction os::Linux::sigact[MAXSIGNUM];
4256 unsigned int os::Linux::sigs = 0;
4257 bool os::Linux::libjsig_is_loaded = false;
4258 typedef struct sigaction *(*get_signal_t)(int);
4259 get_signal_t os::Linux::get_signal_action = NULL;
4260 
4261 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4262   struct sigaction *actp = NULL;
4263 
4264   if (libjsig_is_loaded) {
4265     // Retrieve the old signal handler from libjsig
4266     actp = (*get_signal_action)(sig);
4267   }
4268   if (actp == NULL) {
4269     // Retrieve the preinstalled signal handler from jvm
4270     actp = get_preinstalled_handler(sig);
4271   }
4272 
4273   return actp;
4274 }
4275 
4276 static bool call_chained_handler(struct sigaction *actp, int sig,
4277                                  siginfo_t *siginfo, void *context) {
4278   // Call the old signal handler
4279   if (actp->sa_handler == SIG_DFL) {
4280     // It's more reasonable to let jvm treat it as an unexpected exception
4281     // instead of taking the default action.
4282     return false;
4283   } else if (actp->sa_handler != SIG_IGN) {
4284     if ((actp->sa_flags & SA_NODEFER) == 0) {
4285       // automaticlly block the signal
4286       sigaddset(&(actp->sa_mask), sig);
4287     }
4288 
4289     sa_handler_t hand;
4290     sa_sigaction_t sa;
4291     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4292     // retrieve the chained handler
4293     if (siginfo_flag_set) {
4294       sa = actp->sa_sigaction;
4295     } else {
4296       hand = actp->sa_handler;
4297     }
4298 
4299     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4300       actp->sa_handler = SIG_DFL;
4301     }
4302 
4303     // try to honor the signal mask
4304     sigset_t oset;
4305     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4306 
4307     // call into the chained handler
4308     if (siginfo_flag_set) {
4309       (*sa)(sig, siginfo, context);
4310     } else {
4311       (*hand)(sig);
4312     }
4313 
4314     // restore the signal mask
4315     pthread_sigmask(SIG_SETMASK, &oset, 0);
4316   }
4317   // Tell jvm's signal handler the signal is taken care of.
4318   return true;
4319 }
4320 
4321 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4322   bool chained = false;
4323   // signal-chaining
4324   if (UseSignalChaining) {
4325     struct sigaction *actp = get_chained_signal_action(sig);
4326     if (actp != NULL) {
4327       chained = call_chained_handler(actp, sig, siginfo, context);
4328     }
4329   }
4330   return chained;
4331 }
4332 
4333 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4334   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4335     return &sigact[sig];
4336   }
4337   return NULL;
4338 }
4339 
4340 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4341   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4342   sigact[sig] = oldAct;
4343   sigs |= (unsigned int)1 << sig;
4344 }
4345 
4346 // for diagnostic
4347 int os::Linux::sigflags[MAXSIGNUM];
4348 
4349 int os::Linux::get_our_sigflags(int sig) {
4350   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4351   return sigflags[sig];
4352 }
4353 
4354 void os::Linux::set_our_sigflags(int sig, int flags) {
4355   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4356   sigflags[sig] = flags;
4357 }
4358 
4359 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4360   // Check for overwrite.
4361   struct sigaction oldAct;
4362   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4363 
4364   void* oldhand = oldAct.sa_sigaction
4365                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4366                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4367   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4368       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4369       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4370     if (AllowUserSignalHandlers || !set_installed) {
4371       // Do not overwrite; user takes responsibility to forward to us.
4372       return;
4373     } else if (UseSignalChaining) {
4374       // save the old handler in jvm
4375       save_preinstalled_handler(sig, oldAct);
4376       // libjsig also interposes the sigaction() call below and saves the
4377       // old sigaction on it own.
4378     } else {
4379       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4380                     "%#lx for signal %d.", (long)oldhand, sig));
4381     }
4382   }
4383 
4384   struct sigaction sigAct;
4385   sigfillset(&(sigAct.sa_mask));
4386   sigAct.sa_handler = SIG_DFL;
4387   if (!set_installed) {
4388     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4389   } else {
4390     sigAct.sa_sigaction = signalHandler;
4391     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4392   }
4393   // Save flags, which are set by ours
4394   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4395   sigflags[sig] = sigAct.sa_flags;
4396 
4397   int ret = sigaction(sig, &sigAct, &oldAct);
4398   assert(ret == 0, "check");
4399 
4400   void* oldhand2  = oldAct.sa_sigaction
4401                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4402                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4403   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4404 }
4405 
4406 // install signal handlers for signals that HotSpot needs to
4407 // handle in order to support Java-level exception handling.
4408 
4409 void os::Linux::install_signal_handlers() {
4410   if (!signal_handlers_are_installed) {
4411     signal_handlers_are_installed = true;
4412 
4413     // signal-chaining
4414     typedef void (*signal_setting_t)();
4415     signal_setting_t begin_signal_setting = NULL;
4416     signal_setting_t end_signal_setting = NULL;
4417     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4418                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4419     if (begin_signal_setting != NULL) {
4420       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4421                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4422       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4423                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4424       libjsig_is_loaded = true;
4425       assert(UseSignalChaining, "should enable signal-chaining");
4426     }
4427     if (libjsig_is_loaded) {
4428       // Tell libjsig jvm is setting signal handlers
4429       (*begin_signal_setting)();
4430     }
4431 
4432     set_signal_handler(SIGSEGV, true);
4433     set_signal_handler(SIGPIPE, true);
4434     set_signal_handler(SIGBUS, true);
4435     set_signal_handler(SIGILL, true);
4436     set_signal_handler(SIGFPE, true);
4437 #if defined(PPC64)
4438     set_signal_handler(SIGTRAP, true);
4439 #endif
4440     set_signal_handler(SIGXFSZ, true);
4441 
4442     if (libjsig_is_loaded) {
4443       // Tell libjsig jvm finishes setting signal handlers
4444       (*end_signal_setting)();
4445     }
4446 
4447     // We don't activate signal checker if libjsig is in place, we trust ourselves
4448     // and if UserSignalHandler is installed all bets are off.
4449     // Log that signal checking is off only if -verbose:jni is specified.
4450     if (CheckJNICalls) {
4451       if (libjsig_is_loaded) {
4452         if (PrintJNIResolving) {
4453           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4454         }
4455         check_signals = false;
4456       }
4457       if (AllowUserSignalHandlers) {
4458         if (PrintJNIResolving) {
4459           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4460         }
4461         check_signals = false;
4462       }
4463     }
4464   }
4465 }
4466 
4467 // This is the fastest way to get thread cpu time on Linux.
4468 // Returns cpu time (user+sys) for any thread, not only for current.
4469 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4470 // It might work on 2.6.10+ with a special kernel/glibc patch.
4471 // For reference, please, see IEEE Std 1003.1-2004:
4472 //   http://www.unix.org/single_unix_specification
4473 
4474 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4475   struct timespec tp;
4476   int rc = os::Linux::clock_gettime(clockid, &tp);
4477   assert(rc == 0, "clock_gettime is expected to return 0 code");
4478 
4479   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4480 }
4481 
4482 /////
4483 // glibc on Linux platform uses non-documented flag
4484 // to indicate, that some special sort of signal
4485 // trampoline is used.
4486 // We will never set this flag, and we should
4487 // ignore this flag in our diagnostic
4488 #ifdef SIGNIFICANT_SIGNAL_MASK
4489 #undef SIGNIFICANT_SIGNAL_MASK
4490 #endif
4491 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4492 
4493 static const char* get_signal_handler_name(address handler,
4494                                            char* buf, int buflen) {
4495   int offset;
4496   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4497   if (found) {
4498     // skip directory names
4499     const char *p1, *p2;
4500     p1 = buf;
4501     size_t len = strlen(os::file_separator());
4502     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4503     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4504   } else {
4505     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4506   }
4507   return buf;
4508 }
4509 
4510 static void print_signal_handler(outputStream* st, int sig,
4511                                  char* buf, size_t buflen) {
4512   struct sigaction sa;
4513 
4514   sigaction(sig, NULL, &sa);
4515 
4516   // See comment for SIGNIFICANT_SIGNAL_MASK define
4517   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4518 
4519   st->print("%s: ", os::exception_name(sig, buf, buflen));
4520 
4521   address handler = (sa.sa_flags & SA_SIGINFO)
4522     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4523     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4524 
4525   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4526     st->print("SIG_DFL");
4527   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4528     st->print("SIG_IGN");
4529   } else {
4530     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4531   }
4532 
4533   st->print(", sa_mask[0]=");
4534   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4535 
4536   address rh = VMError::get_resetted_sighandler(sig);
4537   // May be, handler was resetted by VMError?
4538   if(rh != NULL) {
4539     handler = rh;
4540     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4541   }
4542 
4543   st->print(", sa_flags=");
4544   os::Posix::print_sa_flags(st, sa.sa_flags);
4545 
4546   // Check: is it our handler?
4547   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4548      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4549     // It is our signal handler
4550     // check for flags, reset system-used one!
4551     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4552       st->print(
4553                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4554                 os::Linux::get_our_sigflags(sig));
4555     }
4556   }
4557   st->cr();
4558 }
4559 
4560 
4561 #define DO_SIGNAL_CHECK(sig) \
4562   if (!sigismember(&check_signal_done, sig)) \
4563     os::Linux::check_signal_handler(sig)
4564 
4565 // This method is a periodic task to check for misbehaving JNI applications
4566 // under CheckJNI, we can add any periodic checks here
4567 
4568 void os::run_periodic_checks() {
4569 
4570   if (check_signals == false) return;
4571 
4572   // SEGV and BUS if overridden could potentially prevent
4573   // generation of hs*.log in the event of a crash, debugging
4574   // such a case can be very challenging, so we absolutely
4575   // check the following for a good measure:
4576   DO_SIGNAL_CHECK(SIGSEGV);
4577   DO_SIGNAL_CHECK(SIGILL);
4578   DO_SIGNAL_CHECK(SIGFPE);
4579   DO_SIGNAL_CHECK(SIGBUS);
4580   DO_SIGNAL_CHECK(SIGPIPE);
4581   DO_SIGNAL_CHECK(SIGXFSZ);
4582 #if defined(PPC64)
4583   DO_SIGNAL_CHECK(SIGTRAP);
4584 #endif
4585 
4586   // ReduceSignalUsage allows the user to override these handlers
4587   // see comments at the very top and jvm_solaris.h
4588   if (!ReduceSignalUsage) {
4589     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4590     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4591     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4592     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4593   }
4594 
4595   DO_SIGNAL_CHECK(SR_signum);
4596   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4597 }
4598 
4599 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4600 
4601 static os_sigaction_t os_sigaction = NULL;
4602 
4603 void os::Linux::check_signal_handler(int sig) {
4604   char buf[O_BUFLEN];
4605   address jvmHandler = NULL;
4606 
4607 
4608   struct sigaction act;
4609   if (os_sigaction == NULL) {
4610     // only trust the default sigaction, in case it has been interposed
4611     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4612     if (os_sigaction == NULL) return;
4613   }
4614 
4615   os_sigaction(sig, (struct sigaction*)NULL, &act);
4616 
4617 
4618   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4619 
4620   address thisHandler = (act.sa_flags & SA_SIGINFO)
4621     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4622     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4623 
4624 
4625   switch(sig) {
4626   case SIGSEGV:
4627   case SIGBUS:
4628   case SIGFPE:
4629   case SIGPIPE:
4630   case SIGILL:
4631   case SIGXFSZ:
4632     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4633     break;
4634 
4635   case SHUTDOWN1_SIGNAL:
4636   case SHUTDOWN2_SIGNAL:
4637   case SHUTDOWN3_SIGNAL:
4638   case BREAK_SIGNAL:
4639     jvmHandler = (address)user_handler();
4640     break;
4641 
4642   case INTERRUPT_SIGNAL:
4643     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4644     break;
4645 
4646   default:
4647     if (sig == SR_signum) {
4648       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4649     } else {
4650       return;
4651     }
4652     break;
4653   }
4654 
4655   if (thisHandler != jvmHandler) {
4656     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4657     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4658     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4659     // No need to check this sig any longer
4660     sigaddset(&check_signal_done, sig);
4661   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4662     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4663     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4664     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4665     // No need to check this sig any longer
4666     sigaddset(&check_signal_done, sig);
4667   }
4668 
4669   // Dump all the signal
4670   if (sigismember(&check_signal_done, sig)) {
4671     print_signal_handlers(tty, buf, O_BUFLEN);
4672   }
4673 }
4674 
4675 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4676 
4677 extern bool signal_name(int signo, char* buf, size_t len);
4678 
4679 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4680   if (0 < exception_code && exception_code <= SIGRTMAX) {
4681     // signal
4682     if (!signal_name(exception_code, buf, size)) {
4683       jio_snprintf(buf, size, "SIG%d", exception_code);
4684     }
4685     return buf;
4686   } else {
4687     return NULL;
4688   }
4689 }
4690 
4691 // this is called _before_ the most of global arguments have been parsed
4692 void os::init(void) {
4693   char dummy;   /* used to get a guess on initial stack address */
4694 //  first_hrtime = gethrtime();
4695 
4696   // With LinuxThreads the JavaMain thread pid (primordial thread)
4697   // is different than the pid of the java launcher thread.
4698   // So, on Linux, the launcher thread pid is passed to the VM
4699   // via the sun.java.launcher.pid property.
4700   // Use this property instead of getpid() if it was correctly passed.
4701   // See bug 6351349.
4702   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4703 
4704   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4705 
4706   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4707 
4708   init_random(1234567);
4709 
4710   ThreadCritical::initialize();
4711 
4712   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4713   if (Linux::page_size() == -1) {
4714     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4715                   strerror(errno)));
4716   }
4717   init_page_sizes((size_t) Linux::page_size());
4718 
4719   Linux::initialize_system_info();
4720 
4721   // main_thread points to the aboriginal thread
4722   Linux::_main_thread = pthread_self();
4723 
4724   Linux::clock_init();
4725   initial_time_count = javaTimeNanos();
4726 
4727   // pthread_condattr initialization for monotonic clock
4728   int status;
4729   pthread_condattr_t* _condattr = os::Linux::condAttr();
4730   if ((status = pthread_condattr_init(_condattr)) != 0) {
4731     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4732   }
4733   // Only set the clock if CLOCK_MONOTONIC is available
4734   if (Linux::supports_monotonic_clock()) {
4735     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4736       if (status == EINVAL) {
4737         warning("Unable to use monotonic clock with relative timed-waits" \
4738                 " - changes to the time-of-day clock may have adverse affects");
4739       } else {
4740         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4741       }
4742     }
4743   }
4744   // else it defaults to CLOCK_REALTIME
4745 
4746   pthread_mutex_init(&dl_mutex, NULL);
4747 
4748   // If the pagesize of the VM is greater than 8K determine the appropriate
4749   // number of initial guard pages.  The user can change this with the
4750   // command line arguments, if needed.
4751   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4752     StackYellowPages = 1;
4753     StackRedPages = 1;
4754     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4755   }
4756 }
4757 
4758 // To install functions for atexit system call
4759 extern "C" {
4760   static void perfMemory_exit_helper() {
4761     perfMemory_exit();
4762   }
4763 }
4764 
4765 // this is called _after_ the global arguments have been parsed
4766 jint os::init_2(void)
4767 {
4768   Linux::fast_thread_clock_init();
4769 
4770   // Allocate a single page and mark it as readable for safepoint polling
4771   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4772   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4773 
4774   os::set_polling_page( polling_page );
4775 
4776 #ifndef PRODUCT
4777   if(Verbose && PrintMiscellaneous)
4778     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4779 #endif
4780 
4781   if (!UseMembar) {
4782     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4783     guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4784     os::set_memory_serialize_page( mem_serialize_page );
4785 
4786 #ifndef PRODUCT
4787     if(Verbose && PrintMiscellaneous)
4788       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4789 #endif
4790   }
4791 
4792   // initialize suspend/resume support - must do this before signal_sets_init()
4793   if (SR_initialize() != 0) {
4794     perror("SR_initialize failed");
4795     return JNI_ERR;
4796   }
4797 
4798   Linux::signal_sets_init();
4799   Linux::install_signal_handlers();
4800 
4801   // Check minimum allowable stack size for thread creation and to initialize
4802   // the java system classes, including StackOverflowError - depends on page
4803   // size.  Add a page for compiler2 recursion in main thread.
4804   // Add in 2*BytesPerWord times page size to account for VM stack during
4805   // class initialization depending on 32 or 64 bit VM.
4806   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4807             (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4808                     (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4809 
4810   size_t threadStackSizeInBytes = ThreadStackSize * K;
4811   if (threadStackSizeInBytes != 0 &&
4812       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4813         tty->print_cr("\nThe stack size specified is too small, "
4814                       "Specify at least %dk",
4815                       os::Linux::min_stack_allowed/ K);
4816         return JNI_ERR;
4817   }
4818 
4819   // Make the stack size a multiple of the page size so that
4820   // the yellow/red zones can be guarded.
4821   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4822         vm_page_size()));
4823 
4824   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4825 
4826 #if defined(IA32)
4827   workaround_expand_exec_shield_cs_limit();
4828 #endif
4829 
4830   Linux::libpthread_init();
4831   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4832      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4833           Linux::glibc_version(), Linux::libpthread_version(),
4834           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4835   }
4836 
4837   if (UseNUMA) {
4838     if (!Linux::libnuma_init()) {
4839       UseNUMA = false;
4840     } else {
4841       if ((Linux::numa_max_node() < 1)) {
4842         // There's only one node(they start from 0), disable NUMA.
4843         UseNUMA = false;
4844       }
4845     }
4846     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4847     // we can make the adaptive lgrp chunk resizing work. If the user specified
4848     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4849     // disable adaptive resizing.
4850     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4851       if (FLAG_IS_DEFAULT(UseNUMA)) {
4852         UseNUMA = false;
4853       } else {
4854         if (FLAG_IS_DEFAULT(UseLargePages) &&
4855             FLAG_IS_DEFAULT(UseSHM) &&
4856             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4857           UseLargePages = false;
4858         } else {
4859           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4860           UseAdaptiveSizePolicy = false;
4861           UseAdaptiveNUMAChunkSizing = false;
4862         }
4863       }
4864     }
4865     if (!UseNUMA && ForceNUMA) {
4866       UseNUMA = true;
4867     }
4868   }
4869 
4870   if (MaxFDLimit) {
4871     // set the number of file descriptors to max. print out error
4872     // if getrlimit/setrlimit fails but continue regardless.
4873     struct rlimit nbr_files;
4874     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4875     if (status != 0) {
4876       if (PrintMiscellaneous && (Verbose || WizardMode))
4877         perror("os::init_2 getrlimit failed");
4878     } else {
4879       nbr_files.rlim_cur = nbr_files.rlim_max;
4880       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4881       if (status != 0) {
4882         if (PrintMiscellaneous && (Verbose || WizardMode))
4883           perror("os::init_2 setrlimit failed");
4884       }
4885     }
4886   }
4887 
4888   // Initialize lock used to serialize thread creation (see os::create_thread)
4889   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4890 
4891   // at-exit methods are called in the reverse order of their registration.
4892   // atexit functions are called on return from main or as a result of a
4893   // call to exit(3C). There can be only 32 of these functions registered
4894   // and atexit() does not set errno.
4895 
4896   if (PerfAllowAtExitRegistration) {
4897     // only register atexit functions if PerfAllowAtExitRegistration is set.
4898     // atexit functions can be delayed until process exit time, which
4899     // can be problematic for embedded VM situations. Embedded VMs should
4900     // call DestroyJavaVM() to assure that VM resources are released.
4901 
4902     // note: perfMemory_exit_helper atexit function may be removed in
4903     // the future if the appropriate cleanup code can be added to the
4904     // VM_Exit VMOperation's doit method.
4905     if (atexit(perfMemory_exit_helper) != 0) {
4906       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4907     }
4908   }
4909 
4910   // initialize thread priority policy
4911   prio_init();
4912 
4913   return JNI_OK;
4914 }
4915 
4916 // this is called at the end of vm_initialization
4917 void os::init_3(void) {
4918 #ifdef JAVASE_EMBEDDED
4919   // Start the MemNotifyThread
4920   if (LowMemoryProtection) {
4921     MemNotifyThread::start();
4922   }
4923   return;
4924 #endif
4925 }
4926 
4927 // Mark the polling page as unreadable
4928 void os::make_polling_page_unreadable(void) {
4929   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4930     fatal("Could not disable polling page");
4931 };
4932 
4933 // Mark the polling page as readable
4934 void os::make_polling_page_readable(void) {
4935   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4936     fatal("Could not enable polling page");
4937   }
4938 };
4939 
4940 int os::active_processor_count() {
4941   // Linux doesn't yet have a (official) notion of processor sets,
4942   // so just return the number of online processors.
4943   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4944   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4945   return online_cpus;
4946 }
4947 
4948 void os::set_native_thread_name(const char *name) {
4949   // Not yet implemented.
4950   return;
4951 }
4952 
4953 bool os::distribute_processes(uint length, uint* distribution) {
4954   // Not yet implemented.
4955   return false;
4956 }
4957 
4958 bool os::bind_to_processor(uint processor_id) {
4959   // Not yet implemented.
4960   return false;
4961 }
4962 
4963 ///
4964 
4965 void os::SuspendedThreadTask::internal_do_task() {
4966   if (do_suspend(_thread->osthread())) {
4967     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4968     do_task(context);
4969     do_resume(_thread->osthread());
4970   }
4971 }
4972 
4973 class PcFetcher : public os::SuspendedThreadTask {
4974 public:
4975   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4976   ExtendedPC result();
4977 protected:
4978   void do_task(const os::SuspendedThreadTaskContext& context);
4979 private:
4980   ExtendedPC _epc;
4981 };
4982 
4983 ExtendedPC PcFetcher::result() {
4984   guarantee(is_done(), "task is not done yet.");
4985   return _epc;
4986 }
4987 
4988 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4989   Thread* thread = context.thread();
4990   OSThread* osthread = thread->osthread();
4991   if (osthread->ucontext() != NULL) {
4992     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4993   } else {
4994     // NULL context is unexpected, double-check this is the VMThread
4995     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4996   }
4997 }
4998 
4999 // Suspends the target using the signal mechanism and then grabs the PC before
5000 // resuming the target. Used by the flat-profiler only
5001 ExtendedPC os::get_thread_pc(Thread* thread) {
5002   // Make sure that it is called by the watcher for the VMThread
5003   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5004   assert(thread->is_VM_thread(), "Can only be called for VMThread");
5005 
5006   PcFetcher fetcher(thread);
5007   fetcher.run();
5008   return fetcher.result();
5009 }
5010 
5011 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5012 {
5013    if (is_NPTL()) {
5014       return pthread_cond_timedwait(_cond, _mutex, _abstime);
5015    } else {
5016       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5017       // word back to default 64bit precision if condvar is signaled. Java
5018       // wants 53bit precision.  Save and restore current value.
5019       int fpu = get_fpu_control_word();
5020       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5021       set_fpu_control_word(fpu);
5022       return status;
5023    }
5024 }
5025 
5026 ////////////////////////////////////////////////////////////////////////////////
5027 // debug support
5028 
5029 bool os::find(address addr, outputStream* st) {
5030   Dl_info dlinfo;
5031   memset(&dlinfo, 0, sizeof(dlinfo));
5032   if (dladdr(addr, &dlinfo) != 0) {
5033     st->print(PTR_FORMAT ": ", addr);
5034     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5035       st->print("%s+%#x", dlinfo.dli_sname,
5036                  addr - (intptr_t)dlinfo.dli_saddr);
5037     } else if (dlinfo.dli_fbase != NULL) {
5038       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5039     } else {
5040       st->print("<absolute address>");
5041     }
5042     if (dlinfo.dli_fname != NULL) {
5043       st->print(" in %s", dlinfo.dli_fname);
5044     }
5045     if (dlinfo.dli_fbase != NULL) {
5046       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5047     }
5048     st->cr();
5049 
5050     if (Verbose) {
5051       // decode some bytes around the PC
5052       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5053       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5054       address       lowest = (address) dlinfo.dli_sname;
5055       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5056       if (begin < lowest)  begin = lowest;
5057       Dl_info dlinfo2;
5058       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5059           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5060         end = (address) dlinfo2.dli_saddr;
5061       Disassembler::decode(begin, end, st);
5062     }
5063     return true;
5064   }
5065   return false;
5066 }
5067 
5068 ////////////////////////////////////////////////////////////////////////////////
5069 // misc
5070 
5071 // This does not do anything on Linux. This is basically a hook for being
5072 // able to use structured exception handling (thread-local exception filters)
5073 // on, e.g., Win32.
5074 void
5075 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5076                          JavaCallArguments* args, Thread* thread) {
5077   f(value, method, args, thread);
5078 }
5079 
5080 void os::print_statistics() {
5081 }
5082 
5083 int os::message_box(const char* title, const char* message) {
5084   int i;
5085   fdStream err(defaultStream::error_fd());
5086   for (i = 0; i < 78; i++) err.print_raw("=");
5087   err.cr();
5088   err.print_raw_cr(title);
5089   for (i = 0; i < 78; i++) err.print_raw("-");
5090   err.cr();
5091   err.print_raw_cr(message);
5092   for (i = 0; i < 78; i++) err.print_raw("=");
5093   err.cr();
5094 
5095   char buf[16];
5096   // Prevent process from exiting upon "read error" without consuming all CPU
5097   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5098 
5099   return buf[0] == 'y' || buf[0] == 'Y';
5100 }
5101 
5102 int os::stat(const char *path, struct stat *sbuf) {
5103   char pathbuf[MAX_PATH];
5104   if (strlen(path) > MAX_PATH - 1) {
5105     errno = ENAMETOOLONG;
5106     return -1;
5107   }
5108   os::native_path(strcpy(pathbuf, path));
5109   return ::stat(pathbuf, sbuf);
5110 }
5111 
5112 bool os::check_heap(bool force) {
5113   return true;
5114 }
5115 
5116 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5117   return ::vsnprintf(buf, count, format, args);
5118 }
5119 
5120 // Is a (classpath) directory empty?
5121 bool os::dir_is_empty(const char* path) {
5122   DIR *dir = NULL;
5123   struct dirent *ptr;
5124 
5125   dir = opendir(path);
5126   if (dir == NULL) return true;
5127 
5128   /* Scan the directory */
5129   bool result = true;
5130   char buf[sizeof(struct dirent) + MAX_PATH];
5131   while (result && (ptr = ::readdir(dir)) != NULL) {
5132     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5133       result = false;
5134     }
5135   }
5136   closedir(dir);
5137   return result;
5138 }
5139 
5140 // This code originates from JDK's sysOpen and open64_w
5141 // from src/solaris/hpi/src/system_md.c
5142 
5143 #ifndef O_DELETE
5144 #define O_DELETE 0x10000
5145 #endif
5146 
5147 // Open a file. Unlink the file immediately after open returns
5148 // if the specified oflag has the O_DELETE flag set.
5149 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5150 
5151 int os::open(const char *path, int oflag, int mode) {
5152 
5153   if (strlen(path) > MAX_PATH - 1) {
5154     errno = ENAMETOOLONG;
5155     return -1;
5156   }
5157   int fd;
5158   int o_delete = (oflag & O_DELETE);
5159   oflag = oflag & ~O_DELETE;
5160 
5161   fd = ::open64(path, oflag, mode);
5162   if (fd == -1) return -1;
5163 
5164   //If the open succeeded, the file might still be a directory
5165   {
5166     struct stat64 buf64;
5167     int ret = ::fstat64(fd, &buf64);
5168     int st_mode = buf64.st_mode;
5169 
5170     if (ret != -1) {
5171       if ((st_mode & S_IFMT) == S_IFDIR) {
5172         errno = EISDIR;
5173         ::close(fd);
5174         return -1;
5175       }
5176     } else {
5177       ::close(fd);
5178       return -1;
5179     }
5180   }
5181 
5182     /*
5183      * All file descriptors that are opened in the JVM and not
5184      * specifically destined for a subprocess should have the
5185      * close-on-exec flag set.  If we don't set it, then careless 3rd
5186      * party native code might fork and exec without closing all
5187      * appropriate file descriptors (e.g. as we do in closeDescriptors in
5188      * UNIXProcess.c), and this in turn might:
5189      *
5190      * - cause end-of-file to fail to be detected on some file
5191      *   descriptors, resulting in mysterious hangs, or
5192      *
5193      * - might cause an fopen in the subprocess to fail on a system
5194      *   suffering from bug 1085341.
5195      *
5196      * (Yes, the default setting of the close-on-exec flag is a Unix
5197      * design flaw)
5198      *
5199      * See:
5200      * 1085341: 32-bit stdio routines should support file descriptors >255
5201      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5202      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5203      */
5204 #ifdef FD_CLOEXEC
5205     {
5206         int flags = ::fcntl(fd, F_GETFD);
5207         if (flags != -1)
5208             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5209     }
5210 #endif
5211 
5212   if (o_delete != 0) {
5213     ::unlink(path);
5214   }
5215   return fd;
5216 }
5217 
5218 
5219 // create binary file, rewriting existing file if required
5220 int os::create_binary_file(const char* path, bool rewrite_existing) {
5221   int oflags = O_WRONLY | O_CREAT;
5222   if (!rewrite_existing) {
5223     oflags |= O_EXCL;
5224   }
5225   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5226 }
5227 
5228 // return current position of file pointer
5229 jlong os::current_file_offset(int fd) {
5230   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5231 }
5232 
5233 // move file pointer to the specified offset
5234 jlong os::seek_to_file_offset(int fd, jlong offset) {
5235   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5236 }
5237 
5238 // This code originates from JDK's sysAvailable
5239 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5240 
5241 int os::available(int fd, jlong *bytes) {
5242   jlong cur, end;
5243   int mode;
5244   struct stat64 buf64;
5245 
5246   if (::fstat64(fd, &buf64) >= 0) {
5247     mode = buf64.st_mode;
5248     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5249       /*
5250       * XXX: is the following call interruptible? If so, this might
5251       * need to go through the INTERRUPT_IO() wrapper as for other
5252       * blocking, interruptible calls in this file.
5253       */
5254       int n;
5255       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5256         *bytes = n;
5257         return 1;
5258       }
5259     }
5260   }
5261   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5262     return 0;
5263   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5264     return 0;
5265   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5266     return 0;
5267   }
5268   *bytes = end - cur;
5269   return 1;
5270 }
5271 
5272 int os::socket_available(int fd, jint *pbytes) {
5273   // Linux doc says EINTR not returned, unlike Solaris
5274   int ret = ::ioctl(fd, FIONREAD, pbytes);
5275 
5276   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5277   // is expected to return 0 on failure and 1 on success to the jdk.
5278   return (ret < 0) ? 0 : 1;
5279 }
5280 
5281 // Map a block of memory.
5282 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5283                      char *addr, size_t bytes, bool read_only,
5284                      bool allow_exec) {
5285   int prot;
5286   int flags = MAP_PRIVATE;
5287 
5288   if (read_only) {
5289     prot = PROT_READ;
5290   } else {
5291     prot = PROT_READ | PROT_WRITE;
5292   }
5293 
5294   if (allow_exec) {
5295     prot |= PROT_EXEC;
5296   }
5297 
5298   if (addr != NULL) {
5299     flags |= MAP_FIXED;
5300   }
5301 
5302   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5303                                      fd, file_offset);
5304   if (mapped_address == MAP_FAILED) {
5305     return NULL;
5306   }
5307   return mapped_address;
5308 }
5309 
5310 
5311 // Remap a block of memory.
5312 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5313                        char *addr, size_t bytes, bool read_only,
5314                        bool allow_exec) {
5315   // same as map_memory() on this OS
5316   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5317                         allow_exec);
5318 }
5319 
5320 
5321 // Unmap a block of memory.
5322 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5323   return munmap(addr, bytes) == 0;
5324 }
5325 
5326 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5327 
5328 static clockid_t thread_cpu_clockid(Thread* thread) {
5329   pthread_t tid = thread->osthread()->pthread_id();
5330   clockid_t clockid;
5331 
5332   // Get thread clockid
5333   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5334   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5335   return clockid;
5336 }
5337 
5338 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5339 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5340 // of a thread.
5341 //
5342 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5343 // the fast estimate available on the platform.
5344 
5345 jlong os::current_thread_cpu_time() {
5346   if (os::Linux::supports_fast_thread_cpu_time()) {
5347     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5348   } else {
5349     // return user + sys since the cost is the same
5350     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5351   }
5352 }
5353 
5354 jlong os::thread_cpu_time(Thread* thread) {
5355   // consistent with what current_thread_cpu_time() returns
5356   if (os::Linux::supports_fast_thread_cpu_time()) {
5357     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5358   } else {
5359     return slow_thread_cpu_time(thread, true /* user + sys */);
5360   }
5361 }
5362 
5363 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5364   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5365     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5366   } else {
5367     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5368   }
5369 }
5370 
5371 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5372   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5373     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5374   } else {
5375     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5376   }
5377 }
5378 
5379 //
5380 //  -1 on error.
5381 //
5382 
5383 PRAGMA_DIAG_PUSH
5384 PRAGMA_FORMAT_NONLITERAL_IGNORED
5385 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5386   static bool proc_task_unchecked = true;
5387   static const char *proc_stat_path = "/proc/%d/stat";
5388   pid_t  tid = thread->osthread()->thread_id();
5389   char *s;
5390   char stat[2048];
5391   int statlen;
5392   char proc_name[64];
5393   int count;
5394   long sys_time, user_time;
5395   char cdummy;
5396   int idummy;
5397   long ldummy;
5398   FILE *fp;
5399 
5400   // The /proc/<tid>/stat aggregates per-process usage on
5401   // new Linux kernels 2.6+ where NPTL is supported.
5402   // The /proc/self/task/<tid>/stat still has the per-thread usage.
5403   // See bug 6328462.
5404   // There possibly can be cases where there is no directory
5405   // /proc/self/task, so we check its availability.
5406   if (proc_task_unchecked && os::Linux::is_NPTL()) {
5407     // This is executed only once
5408     proc_task_unchecked = false;
5409     fp = fopen("/proc/self/task", "r");
5410     if (fp != NULL) {
5411       proc_stat_path = "/proc/self/task/%d/stat";
5412       fclose(fp);
5413     }
5414   }
5415 
5416   sprintf(proc_name, proc_stat_path, tid);
5417   fp = fopen(proc_name, "r");
5418   if ( fp == NULL ) return -1;
5419   statlen = fread(stat, 1, 2047, fp);
5420   stat[statlen] = '\0';
5421   fclose(fp);
5422 
5423   // Skip pid and the command string. Note that we could be dealing with
5424   // weird command names, e.g. user could decide to rename java launcher
5425   // to "java 1.4.2 :)", then the stat file would look like
5426   //                1234 (java 1.4.2 :)) R ... ...
5427   // We don't really need to know the command string, just find the last
5428   // occurrence of ")" and then start parsing from there. See bug 4726580.
5429   s = strrchr(stat, ')');
5430   if (s == NULL ) return -1;
5431 
5432   // Skip blank chars
5433   do s++; while (isspace(*s));
5434 
5435   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5436                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5437                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5438                  &user_time, &sys_time);
5439   if ( count != 13 ) return -1;
5440   if (user_sys_cpu_time) {
5441     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5442   } else {
5443     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5444   }
5445 }
5446 PRAGMA_DIAG_POP
5447 
5448 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5449   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5450   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5451   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5452   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5453 }
5454 
5455 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5456   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5457   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5458   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5459   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5460 }
5461 
5462 bool os::is_thread_cpu_time_supported() {
5463   return true;
5464 }
5465 
5466 // System loadavg support.  Returns -1 if load average cannot be obtained.
5467 // Linux doesn't yet have a (official) notion of processor sets,
5468 // so just return the system wide load average.
5469 int os::loadavg(double loadavg[], int nelem) {
5470   return ::getloadavg(loadavg, nelem);
5471 }
5472 
5473 void os::pause() {
5474   char filename[MAX_PATH];
5475   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5476     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5477   } else {
5478     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5479   }
5480 
5481   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5482   if (fd != -1) {
5483     struct stat buf;
5484     ::close(fd);
5485     while (::stat(filename, &buf) == 0) {
5486       (void)::poll(NULL, 0, 100);
5487     }
5488   } else {
5489     jio_fprintf(stderr,
5490       "Could not open pause file '%s', continuing immediately.\n", filename);
5491   }
5492 }
5493 
5494 
5495 // Refer to the comments in os_solaris.cpp park-unpark.
5496 //
5497 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5498 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5499 // For specifics regarding the bug see GLIBC BUGID 261237 :
5500 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5501 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5502 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5503 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5504 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5505 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5506 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5507 // of libpthread avoids the problem, but isn't practical.
5508 //
5509 // Possible remedies:
5510 //
5511 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5512 //      This is palliative and probabilistic, however.  If the thread is preempted
5513 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5514 //      than the minimum period may have passed, and the abstime may be stale (in the
5515 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5516 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5517 //
5518 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5519 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5520 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5521 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5522 //      thread.
5523 //
5524 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5525 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5526 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5527 //      This also works well.  In fact it avoids kernel-level scalability impediments
5528 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5529 //      timers in a graceful fashion.
5530 //
5531 // 4.   When the abstime value is in the past it appears that control returns
5532 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5533 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5534 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5535 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5536 //      It may be possible to avoid reinitialization by checking the return
5537 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5538 //      condvar we must establish the invariant that cond_signal() is only called
5539 //      within critical sections protected by the adjunct mutex.  This prevents
5540 //      cond_signal() from "seeing" a condvar that's in the midst of being
5541 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5542 //      desirable signal-after-unlock optimization that avoids futile context switching.
5543 //
5544 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5545 //      structure when a condvar is used or initialized.  cond_destroy()  would
5546 //      release the helper structure.  Our reinitialize-after-timedwait fix
5547 //      put excessive stress on malloc/free and locks protecting the c-heap.
5548 //
5549 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5550 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5551 // and only enabling the work-around for vulnerable environments.
5552 
5553 // utility to compute the abstime argument to timedwait:
5554 // millis is the relative timeout time
5555 // abstime will be the absolute timeout time
5556 // TODO: replace compute_abstime() with unpackTime()
5557 
5558 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5559   if (millis < 0)  millis = 0;
5560 
5561   jlong seconds = millis / 1000;
5562   millis %= 1000;
5563   if (seconds > 50000000) { // see man cond_timedwait(3T)
5564     seconds = 50000000;
5565   }
5566 
5567   if (os::Linux::supports_monotonic_clock()) {
5568     struct timespec now;
5569     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5570     assert_status(status == 0, status, "clock_gettime");
5571     abstime->tv_sec = now.tv_sec  + seconds;
5572     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5573     if (nanos >= NANOSECS_PER_SEC) {
5574       abstime->tv_sec += 1;
5575       nanos -= NANOSECS_PER_SEC;
5576     }
5577     abstime->tv_nsec = nanos;
5578   } else {
5579     struct timeval now;
5580     int status = gettimeofday(&now, NULL);
5581     assert(status == 0, "gettimeofday");
5582     abstime->tv_sec = now.tv_sec  + seconds;
5583     long usec = now.tv_usec + millis * 1000;
5584     if (usec >= 1000000) {
5585       abstime->tv_sec += 1;
5586       usec -= 1000000;
5587     }
5588     abstime->tv_nsec = usec * 1000;
5589   }
5590   return abstime;
5591 }
5592 
5593 
5594 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5595 // Conceptually TryPark() should be equivalent to park(0).
5596 
5597 int os::PlatformEvent::TryPark() {
5598   for (;;) {
5599     const int v = _Event ;
5600     guarantee ((v == 0) || (v == 1), "invariant") ;
5601     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5602   }
5603 }
5604 
5605 void os::PlatformEvent::park() {       // AKA "down()"
5606   // Invariant: Only the thread associated with the Event/PlatformEvent
5607   // may call park().
5608   // TODO: assert that _Assoc != NULL or _Assoc == Self
5609   int v ;
5610   for (;;) {
5611       v = _Event ;
5612       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5613   }
5614   guarantee (v >= 0, "invariant") ;
5615   if (v == 0) {
5616      // Do this the hard way by blocking ...
5617      int status = pthread_mutex_lock(_mutex);
5618      assert_status(status == 0, status, "mutex_lock");
5619      guarantee (_nParked == 0, "invariant") ;
5620      ++ _nParked ;
5621      while (_Event < 0) {
5622         status = pthread_cond_wait(_cond, _mutex);
5623         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5624         // Treat this the same as if the wait was interrupted
5625         if (status == ETIME) { status = EINTR; }
5626         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5627      }
5628      -- _nParked ;
5629 
5630     _Event = 0 ;
5631      status = pthread_mutex_unlock(_mutex);
5632      assert_status(status == 0, status, "mutex_unlock");
5633     // Paranoia to ensure our locked and lock-free paths interact
5634     // correctly with each other.
5635     OrderAccess::fence();
5636   }
5637   guarantee (_Event >= 0, "invariant") ;
5638 }
5639 
5640 int os::PlatformEvent::park(jlong millis) {
5641   guarantee (_nParked == 0, "invariant") ;
5642 
5643   int v ;
5644   for (;;) {
5645       v = _Event ;
5646       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5647   }
5648   guarantee (v >= 0, "invariant") ;
5649   if (v != 0) return OS_OK ;
5650 
5651   // We do this the hard way, by blocking the thread.
5652   // Consider enforcing a minimum timeout value.
5653   struct timespec abst;
5654   compute_abstime(&abst, millis);
5655 
5656   int ret = OS_TIMEOUT;
5657   int status = pthread_mutex_lock(_mutex);
5658   assert_status(status == 0, status, "mutex_lock");
5659   guarantee (_nParked == 0, "invariant") ;
5660   ++_nParked ;
5661 
5662   // Object.wait(timo) will return because of
5663   // (a) notification
5664   // (b) timeout
5665   // (c) thread.interrupt
5666   //
5667   // Thread.interrupt and object.notify{All} both call Event::set.
5668   // That is, we treat thread.interrupt as a special case of notification.
5669   // The underlying Solaris implementation, cond_timedwait, admits
5670   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5671   // JVM from making those visible to Java code.  As such, we must
5672   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5673   //
5674   // TODO: properly differentiate simultaneous notify+interrupt.
5675   // In that case, we should propagate the notify to another waiter.
5676 
5677   while (_Event < 0) {
5678     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5679     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5680       pthread_cond_destroy (_cond);
5681       pthread_cond_init (_cond, os::Linux::condAttr()) ;
5682     }
5683     assert_status(status == 0 || status == EINTR ||
5684                   status == ETIME || status == ETIMEDOUT,
5685                   status, "cond_timedwait");
5686     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5687     if (status == ETIME || status == ETIMEDOUT) break ;
5688     // We consume and ignore EINTR and spurious wakeups.
5689   }
5690   --_nParked ;
5691   if (_Event >= 0) {
5692      ret = OS_OK;
5693   }
5694   _Event = 0 ;
5695   status = pthread_mutex_unlock(_mutex);
5696   assert_status(status == 0, status, "mutex_unlock");
5697   assert (_nParked == 0, "invariant") ;
5698   // Paranoia to ensure our locked and lock-free paths interact
5699   // correctly with each other.
5700   OrderAccess::fence();
5701   return ret;
5702 }
5703 
5704 void os::PlatformEvent::unpark() {
5705   // Transitions for _Event:
5706   //    0 :=> 1
5707   //    1 :=> 1
5708   //   -1 :=> either 0 or 1; must signal target thread
5709   //          That is, we can safely transition _Event from -1 to either
5710   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
5711   //          unpark() calls.
5712   // See also: "Semaphores in Plan 9" by Mullender & Cox
5713   //
5714   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5715   // that it will take two back-to-back park() calls for the owning
5716   // thread to block. This has the benefit of forcing a spurious return
5717   // from the first park() call after an unpark() call which will help
5718   // shake out uses of park() and unpark() without condition variables.
5719 
5720   if (Atomic::xchg(1, &_Event) >= 0) return;
5721 
5722   // Wait for the thread associated with the event to vacate
5723   int status = pthread_mutex_lock(_mutex);
5724   assert_status(status == 0, status, "mutex_lock");
5725   int AnyWaiters = _nParked;
5726   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5727   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5728     AnyWaiters = 0;
5729     pthread_cond_signal(_cond);
5730   }
5731   status = pthread_mutex_unlock(_mutex);
5732   assert_status(status == 0, status, "mutex_unlock");
5733   if (AnyWaiters != 0) {
5734     status = pthread_cond_signal(_cond);
5735     assert_status(status == 0, status, "cond_signal");
5736   }
5737 
5738   // Note that we signal() _after dropping the lock for "immortal" Events.
5739   // This is safe and avoids a common class of  futile wakeups.  In rare
5740   // circumstances this can cause a thread to return prematurely from
5741   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5742   // simply re-test the condition and re-park itself.
5743 }
5744 
5745 
5746 // JSR166
5747 // -------------------------------------------------------
5748 
5749 /*
5750  * The solaris and linux implementations of park/unpark are fairly
5751  * conservative for now, but can be improved. They currently use a
5752  * mutex/condvar pair, plus a a count.
5753  * Park decrements count if > 0, else does a condvar wait.  Unpark
5754  * sets count to 1 and signals condvar.  Only one thread ever waits
5755  * on the condvar. Contention seen when trying to park implies that someone
5756  * is unparking you, so don't wait. And spurious returns are fine, so there
5757  * is no need to track notifications.
5758  */
5759 
5760 /*
5761  * This code is common to linux and solaris and will be moved to a
5762  * common place in dolphin.
5763  *
5764  * The passed in time value is either a relative time in nanoseconds
5765  * or an absolute time in milliseconds. Either way it has to be unpacked
5766  * into suitable seconds and nanoseconds components and stored in the
5767  * given timespec structure.
5768  * Given time is a 64-bit value and the time_t used in the timespec is only
5769  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5770  * overflow if times way in the future are given. Further on Solaris versions
5771  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5772  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5773  * As it will be 28 years before "now + 100000000" will overflow we can
5774  * ignore overflow and just impose a hard-limit on seconds using the value
5775  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5776  * years from "now".
5777  */
5778 
5779 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5780   assert (time > 0, "convertTime");
5781   time_t max_secs = 0;
5782 
5783   if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5784     struct timeval now;
5785     int status = gettimeofday(&now, NULL);
5786     assert(status == 0, "gettimeofday");
5787 
5788     max_secs = now.tv_sec + MAX_SECS;
5789 
5790     if (isAbsolute) {
5791       jlong secs = time / 1000;
5792       if (secs > max_secs) {
5793         absTime->tv_sec = max_secs;
5794       } else {
5795         absTime->tv_sec = secs;
5796       }
5797       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5798     } else {
5799       jlong secs = time / NANOSECS_PER_SEC;
5800       if (secs >= MAX_SECS) {
5801         absTime->tv_sec = max_secs;
5802         absTime->tv_nsec = 0;
5803       } else {
5804         absTime->tv_sec = now.tv_sec + secs;
5805         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5806         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5807           absTime->tv_nsec -= NANOSECS_PER_SEC;
5808           ++absTime->tv_sec; // note: this must be <= max_secs
5809         }
5810       }
5811     }
5812   } else {
5813     // must be relative using monotonic clock
5814     struct timespec now;
5815     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5816     assert_status(status == 0, status, "clock_gettime");
5817     max_secs = now.tv_sec + MAX_SECS;
5818     jlong secs = time / NANOSECS_PER_SEC;
5819     if (secs >= MAX_SECS) {
5820       absTime->tv_sec = max_secs;
5821       absTime->tv_nsec = 0;
5822     } else {
5823       absTime->tv_sec = now.tv_sec + secs;
5824       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5825       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5826         absTime->tv_nsec -= NANOSECS_PER_SEC;
5827         ++absTime->tv_sec; // note: this must be <= max_secs
5828       }
5829     }
5830   }
5831   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5832   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5833   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5834   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5835 }
5836 
5837 void Parker::park(bool isAbsolute, jlong time) {
5838   // Ideally we'd do something useful while spinning, such
5839   // as calling unpackTime().
5840 
5841   // Optional fast-path check:
5842   // Return immediately if a permit is available.
5843   // We depend on Atomic::xchg() having full barrier semantics
5844   // since we are doing a lock-free update to _counter.
5845   if (Atomic::xchg(0, &_counter) > 0) return;
5846 
5847   Thread* thread = Thread::current();
5848   assert(thread->is_Java_thread(), "Must be JavaThread");
5849   JavaThread *jt = (JavaThread *)thread;
5850 
5851   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5852   // Check interrupt before trying to wait
5853   if (Thread::is_interrupted(thread, false)) {
5854     return;
5855   }
5856 
5857   // Next, demultiplex/decode time arguments
5858   timespec absTime;
5859   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5860     return;
5861   }
5862   if (time > 0) {
5863     unpackTime(&absTime, isAbsolute, time);
5864   }
5865 
5866 
5867   // Enter safepoint region
5868   // Beware of deadlocks such as 6317397.
5869   // The per-thread Parker:: mutex is a classic leaf-lock.
5870   // In particular a thread must never block on the Threads_lock while
5871   // holding the Parker:: mutex.  If safepoints are pending both the
5872   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5873   ThreadBlockInVM tbivm(jt);
5874 
5875   // Don't wait if cannot get lock since interference arises from
5876   // unblocking.  Also. check interrupt before trying wait
5877   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5878     return;
5879   }
5880 
5881   int status ;
5882   if (_counter > 0)  { // no wait needed
5883     _counter = 0;
5884     status = pthread_mutex_unlock(_mutex);
5885     assert (status == 0, "invariant") ;
5886     // Paranoia to ensure our locked and lock-free paths interact
5887     // correctly with each other and Java-level accesses.
5888     OrderAccess::fence();
5889     return;
5890   }
5891 
5892 #ifdef ASSERT
5893   // Don't catch signals while blocked; let the running threads have the signals.
5894   // (This allows a debugger to break into the running thread.)
5895   sigset_t oldsigs;
5896   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5897   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5898 #endif
5899 
5900   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5901   jt->set_suspend_equivalent();
5902   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5903 
5904   assert(_cur_index == -1, "invariant");
5905   if (time == 0) {
5906     _cur_index = REL_INDEX; // arbitrary choice when not timed
5907     status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5908   } else {
5909     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5910     status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5911     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5912       pthread_cond_destroy (&_cond[_cur_index]) ;
5913       pthread_cond_init    (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5914     }
5915   }
5916   _cur_index = -1;
5917   assert_status(status == 0 || status == EINTR ||
5918                 status == ETIME || status == ETIMEDOUT,
5919                 status, "cond_timedwait");
5920 
5921 #ifdef ASSERT
5922   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5923 #endif
5924 
5925   _counter = 0 ;
5926   status = pthread_mutex_unlock(_mutex) ;
5927   assert_status(status == 0, status, "invariant") ;
5928   // Paranoia to ensure our locked and lock-free paths interact
5929   // correctly with each other and Java-level accesses.
5930   OrderAccess::fence();
5931 
5932   // If externally suspended while waiting, re-suspend
5933   if (jt->handle_special_suspend_equivalent_condition()) {
5934     jt->java_suspend_self();
5935   }
5936 }
5937 
5938 void Parker::unpark() {
5939   int s, status ;
5940   status = pthread_mutex_lock(_mutex);
5941   assert (status == 0, "invariant") ;
5942   s = _counter;
5943   _counter = 1;
5944   if (s < 1) {
5945     // thread might be parked
5946     if (_cur_index != -1) {
5947       // thread is definitely parked
5948       if (WorkAroundNPTLTimedWaitHang) {
5949         status = pthread_cond_signal (&_cond[_cur_index]);
5950         assert (status == 0, "invariant");
5951         status = pthread_mutex_unlock(_mutex);
5952         assert (status == 0, "invariant");
5953       } else {
5954         status = pthread_mutex_unlock(_mutex);
5955         assert (status == 0, "invariant");
5956         status = pthread_cond_signal (&_cond[_cur_index]);
5957         assert (status == 0, "invariant");
5958       }
5959     } else {
5960       pthread_mutex_unlock(_mutex);
5961       assert (status == 0, "invariant") ;
5962     }
5963   } else {
5964     pthread_mutex_unlock(_mutex);
5965     assert (status == 0, "invariant") ;
5966   }
5967 }
5968 
5969 
5970 extern char** environ;
5971 
5972 #ifndef __NR_fork
5973 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5974 #endif
5975 
5976 #ifndef __NR_execve
5977 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5978 #endif
5979 
5980 // Run the specified command in a separate process. Return its exit value,
5981 // or -1 on failure (e.g. can't fork a new process).
5982 // Unlike system(), this function can be called from signal handler. It
5983 // doesn't block SIGINT et al.
5984 int os::fork_and_exec(char* cmd) {
5985   const char * argv[4] = {"sh", "-c", cmd, NULL};
5986 
5987   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5988   // pthread_atfork handlers and reset pthread library. All we need is a
5989   // separate process to execve. Make a direct syscall to fork process.
5990   // On IA64 there's no fork syscall, we have to use fork() and hope for
5991   // the best...
5992   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5993               IA64_ONLY(fork();)
5994 
5995   if (pid < 0) {
5996     // fork failed
5997     return -1;
5998 
5999   } else if (pid == 0) {
6000     // child process
6001 
6002     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
6003     // first to kill every thread on the thread list. Because this list is
6004     // not reset by fork() (see notes above), execve() will instead kill
6005     // every thread in the parent process. We know this is the only thread
6006     // in the new process, so make a system call directly.
6007     // IA64 should use normal execve() from glibc to match the glibc fork()
6008     // above.
6009     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
6010     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
6011 
6012     // execve failed
6013     _exit(-1);
6014 
6015   } else  {
6016     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6017     // care about the actual exit code, for now.
6018 
6019     int status;
6020 
6021     // Wait for the child process to exit.  This returns immediately if
6022     // the child has already exited. */
6023     while (waitpid(pid, &status, 0) < 0) {
6024         switch (errno) {
6025         case ECHILD: return 0;
6026         case EINTR: break;
6027         default: return -1;
6028         }
6029     }
6030 
6031     if (WIFEXITED(status)) {
6032        // The child exited normally; get its exit code.
6033        return WEXITSTATUS(status);
6034     } else if (WIFSIGNALED(status)) {
6035        // The child exited because of a signal
6036        // The best value to return is 0x80 + signal number,
6037        // because that is what all Unix shells do, and because
6038        // it allows callers to distinguish between process exit and
6039        // process death by signal.
6040        return 0x80 + WTERMSIG(status);
6041     } else {
6042        // Unknown exit code; pass it through
6043        return status;
6044     }
6045   }
6046 }
6047 
6048 // is_headless_jre()
6049 //
6050 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6051 // in order to report if we are running in a headless jre
6052 //
6053 // Since JDK8 xawt/libmawt.so was moved into the same directory
6054 // as libawt.so, and renamed libawt_xawt.so
6055 //
6056 bool os::is_headless_jre() {
6057     struct stat statbuf;
6058     char buf[MAXPATHLEN];
6059     char libmawtpath[MAXPATHLEN];
6060     const char *xawtstr  = "/xawt/libmawt.so";
6061     const char *new_xawtstr = "/libawt_xawt.so";
6062     char *p;
6063 
6064     // Get path to libjvm.so
6065     os::jvm_path(buf, sizeof(buf));
6066 
6067     // Get rid of libjvm.so
6068     p = strrchr(buf, '/');
6069     if (p == NULL) return false;
6070     else *p = '\0';
6071 
6072     // Get rid of client or server
6073     p = strrchr(buf, '/');
6074     if (p == NULL) return false;
6075     else *p = '\0';
6076 
6077     // check xawt/libmawt.so
6078     strcpy(libmawtpath, buf);
6079     strcat(libmawtpath, xawtstr);
6080     if (::stat(libmawtpath, &statbuf) == 0) return false;
6081 
6082     // check libawt_xawt.so
6083     strcpy(libmawtpath, buf);
6084     strcat(libmawtpath, new_xawtstr);
6085     if (::stat(libmawtpath, &statbuf) == 0) return false;
6086 
6087     return true;
6088 }
6089 
6090 // Get the default path to the core file
6091 // Returns the length of the string
6092 int os::get_core_path(char* buffer, size_t bufferSize) {
6093   const char* p = get_current_directory(buffer, bufferSize);
6094 
6095   if (p == NULL) {
6096     assert(p != NULL, "failed to get current directory");
6097     return 0;
6098   }
6099 
6100   return strlen(buffer);
6101 }
6102 
6103 #ifdef JAVASE_EMBEDDED
6104 //
6105 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6106 //
6107 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6108 
6109 // ctor
6110 //
6111 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6112   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6113   _fd = fd;
6114 
6115   if (os::create_thread(this, os::os_thread)) {
6116     _memnotify_thread = this;
6117     os::set_priority(this, NearMaxPriority);
6118     os::start_thread(this);
6119   }
6120 }
6121 
6122 // Where all the work gets done
6123 //
6124 void MemNotifyThread::run() {
6125   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6126 
6127   // Set up the select arguments
6128   fd_set rfds;
6129   if (_fd != -1) {
6130     FD_ZERO(&rfds);
6131     FD_SET(_fd, &rfds);
6132   }
6133 
6134   // Now wait for the mem_notify device to wake up
6135   while (1) {
6136     // Wait for the mem_notify device to signal us..
6137     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6138     if (rc == -1) {
6139       perror("select!\n");
6140       break;
6141     } else if (rc) {
6142       //ssize_t free_before = os::available_memory();
6143       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6144 
6145       // The kernel is telling us there is not much memory left...
6146       // try to do something about that
6147 
6148       // If we are not already in a GC, try one.
6149       if (!Universe::heap()->is_gc_active()) {
6150         Universe::heap()->collect(GCCause::_allocation_failure);
6151 
6152         //ssize_t free_after = os::available_memory();
6153         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6154         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6155       }
6156       // We might want to do something like the following if we find the GC's are not helping...
6157       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6158     }
6159   }
6160 }
6161 
6162 //
6163 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6164 //
6165 void MemNotifyThread::start() {
6166   int    fd;
6167   fd = open ("/dev/mem_notify", O_RDONLY, 0);
6168   if (fd < 0) {
6169       return;
6170   }
6171 
6172   if (memnotify_thread() == NULL) {
6173     new MemNotifyThread(fd);
6174   }
6175 }
6176 
6177 #endif // JAVASE_EMBEDDED
6178 
6179 
6180 /////////////// Unit tests ///////////////
6181 
6182 #ifndef PRODUCT
6183 
6184 #define test_log(...) \
6185   do {\
6186     if (VerboseInternalVMTests) { \
6187       tty->print_cr(__VA_ARGS__); \
6188       tty->flush(); \
6189     }\
6190   } while (false)
6191 
6192 class TestReserveMemorySpecial : AllStatic {
6193  public:
6194   static void small_page_write(void* addr, size_t size) {
6195     size_t page_size = os::vm_page_size();
6196 
6197     char* end = (char*)addr + size;
6198     for (char* p = (char*)addr; p < end; p += page_size) {
6199       *p = 1;
6200     }
6201   }
6202 
6203   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6204     if (!UseHugeTLBFS) {
6205       return;
6206     }
6207 
6208     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6209 
6210     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6211 
6212     if (addr != NULL) {
6213       small_page_write(addr, size);
6214 
6215       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6216     }
6217   }
6218 
6219   static void test_reserve_memory_special_huge_tlbfs_only() {
6220     if (!UseHugeTLBFS) {
6221       return;
6222     }
6223 
6224     size_t lp = os::large_page_size();
6225 
6226     for (size_t size = lp; size <= lp * 10; size += lp) {
6227       test_reserve_memory_special_huge_tlbfs_only(size);
6228     }
6229   }
6230 
6231   static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6232     if (!UseHugeTLBFS) {
6233         return;
6234     }
6235 
6236     test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6237         size, alignment);
6238 
6239     assert(size >= os::large_page_size(), "Incorrect input to test");
6240 
6241     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6242 
6243     if (addr != NULL) {
6244       small_page_write(addr, size);
6245 
6246       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6247     }
6248   }
6249 
6250   static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6251     size_t lp = os::large_page_size();
6252     size_t ag = os::vm_allocation_granularity();
6253 
6254     for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6255       test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6256     }
6257   }
6258 
6259   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6260     size_t lp = os::large_page_size();
6261     size_t ag = os::vm_allocation_granularity();
6262 
6263     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6264     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6265     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6266     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6267     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6268     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6269     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6270     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6271     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6272   }
6273 
6274   static void test_reserve_memory_special_huge_tlbfs() {
6275     if (!UseHugeTLBFS) {
6276       return;
6277     }
6278 
6279     test_reserve_memory_special_huge_tlbfs_only();
6280     test_reserve_memory_special_huge_tlbfs_mixed();
6281   }
6282 
6283   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6284     if (!UseSHM) {
6285       return;
6286     }
6287 
6288     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6289 
6290     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6291 
6292     if (addr != NULL) {
6293       assert(is_ptr_aligned(addr, alignment), "Check");
6294       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6295 
6296       small_page_write(addr, size);
6297 
6298       os::Linux::release_memory_special_shm(addr, size);
6299     }
6300   }
6301 
6302   static void test_reserve_memory_special_shm() {
6303     size_t lp = os::large_page_size();
6304     size_t ag = os::vm_allocation_granularity();
6305 
6306     for (size_t size = ag; size < lp * 3; size += ag) {
6307       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6308         test_reserve_memory_special_shm(size, alignment);
6309       }
6310     }
6311   }
6312 
6313   static void test() {
6314     test_reserve_memory_special_huge_tlbfs();
6315     test_reserve_memory_special_shm();
6316   }
6317 };
6318 
6319 void TestReserveMemorySpecial_test() {
6320   TestReserveMemorySpecial::test();
6321 }
6322 
6323 #endif