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_linux.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "prims/jniFastGetField.hpp"
  42 #include "prims/jvm.h"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.inline.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.hpp"
  49 #include "runtime/init.hpp"
  50 #include "runtime/java.hpp"
  51 #include "runtime/javaCalls.hpp"
  52 #include "runtime/mutexLocker.hpp"
  53 #include "runtime/objectMonitor.hpp"
  54 #include "runtime/orderAccess.inline.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/perfMemory.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/statSampler.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "runtime/thread.inline.hpp"
  61 #include "runtime/threadCritical.hpp"
  62 #include "runtime/timer.hpp"
  63 #include "services/attachListener.hpp"
  64 #include "services/memTracker.hpp"
  65 #include "services/runtimeService.hpp"
  66 #include "utilities/decoder.hpp"
  67 #include "utilities/defaultStream.hpp"
  68 #include "utilities/events.hpp"
  69 #include "utilities/elfFile.hpp"
  70 #include "utilities/growableArray.hpp"
  71 #include "utilities/vmError.hpp"
  72 
  73 // put OS-includes here
  74 # include <sys/types.h>
  75 # include <sys/mman.h>
  76 # include <sys/stat.h>
  77 # include <sys/select.h>
  78 # include <pthread.h>
  79 # include <signal.h>
  80 # include <errno.h>
  81 # include <dlfcn.h>
  82 # include <stdio.h>
  83 # include <unistd.h>
  84 # include <sys/resource.h>
  85 # include <pthread.h>
  86 # include <sys/stat.h>
  87 # include <sys/time.h>
  88 # include <sys/times.h>
  89 # include <sys/utsname.h>
  90 # include <sys/socket.h>
  91 # include <sys/wait.h>
  92 # include <pwd.h>
  93 # include <poll.h>
  94 # include <semaphore.h>
  95 # include <fcntl.h>
  96 # include <string.h>
  97 # include <syscall.h>
  98 # include <sys/sysinfo.h>
  99 # include <gnu/libc-version.h>
 100 # include <sys/ipc.h>
 101 # include <sys/shm.h>
 102 # include <link.h>
 103 # include <stdint.h>
 104 # include <inttypes.h>
 105 # include <sys/ioctl.h>
 106 
 107 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
 108 
 109 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 110 // getrusage() is prepared to handle the associated failure.
 111 #ifndef RUSAGE_THREAD
 112 #define RUSAGE_THREAD   (1)               /* only the calling thread */
 113 #endif
 114 
 115 #define MAX_PATH    (2 * K)
 116 
 117 #define MAX_SECS 100000000
 118 
 119 // for timer info max values which include all bits
 120 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 121 
 122 #define LARGEPAGES_BIT (1 << 6)
 123 ////////////////////////////////////////////////////////////////////////////////
 124 // global variables
 125 julong os::Linux::_physical_memory = 0;
 126 
 127 address   os::Linux::_initial_thread_stack_bottom = NULL;
 128 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 129 
 130 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 131 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 132 Mutex* os::Linux::_createThread_lock = NULL;
 133 pthread_t os::Linux::_main_thread;
 134 int os::Linux::_page_size = -1;
 135 const int os::Linux::_vm_default_page_size = (8 * K);
 136 bool os::Linux::_is_floating_stack = false;
 137 bool os::Linux::_is_NPTL = false;
 138 bool os::Linux::_supports_fast_thread_cpu_time = false;
 139 const char * os::Linux::_glibc_version = NULL;
 140 const char * os::Linux::_libpthread_version = NULL;
 141 pthread_condattr_t os::Linux::_condattr[1];
 142 
 143 static jlong initial_time_count=0;
 144 
 145 static int clock_tics_per_sec = 100;
 146 
 147 // For diagnostics to print a message once. see run_periodic_checks
 148 static sigset_t check_signal_done;
 149 static bool check_signals = true;
 150 
 151 static pid_t _initial_pid = 0;
 152 
 153 /* Signal number used to suspend/resume a thread */
 154 
 155 /* do not use any signal number less than SIGSEGV, see 4355769 */
 156 static int SR_signum = SIGUSR2;
 157 sigset_t SR_sigset;
 158 
 159 /* Used to protect dlsym() calls */
 160 static pthread_mutex_t dl_mutex;
 161 
 162 // Declarations
 163 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 164 
 165 #ifdef JAVASE_EMBEDDED
 166 class MemNotifyThread: public Thread {
 167   friend class VMStructs;
 168  public:
 169   virtual void run();
 170 
 171  private:
 172   static MemNotifyThread* _memnotify_thread;
 173   int _fd;
 174 
 175  public:
 176 
 177   // Constructor
 178   MemNotifyThread(int fd);
 179 
 180   // Tester
 181   bool is_memnotify_thread() const { return true; }
 182 
 183   // Printing
 184   char* name() const { return (char*)"Linux MemNotify Thread"; }
 185 
 186   // Returns the single instance of the MemNotifyThread
 187   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 188 
 189   // Create and start the single instance of MemNotifyThread
 190   static void start();
 191 };
 192 #endif // JAVASE_EMBEDDED
 193 
 194 // utility functions
 195 
 196 static int SR_initialize();
 197 
 198 julong os::available_memory() {
 199   return Linux::available_memory();
 200 }
 201 
 202 julong os::Linux::available_memory() {
 203   // values in struct sysinfo are "unsigned long"
 204   struct sysinfo si;
 205   sysinfo(&si);
 206 
 207   return (julong)si.freeram * si.mem_unit;
 208 }
 209 
 210 julong os::physical_memory() {
 211   return Linux::physical_memory();
 212 }
 213 
 214 ////////////////////////////////////////////////////////////////////////////////
 215 // environment support
 216 
 217 bool os::getenv(const char* name, char* buf, int len) {
 218   const char* val = ::getenv(name);
 219   if (val != NULL && strlen(val) < (size_t)len) {
 220     strcpy(buf, val);
 221     return true;
 222   }
 223   if (len > 0) buf[0] = 0;  // return a null string
 224   return false;
 225 }
 226 
 227 
 228 // Return true if user is running as root.
 229 
 230 bool os::have_special_privileges() {
 231   static bool init = false;
 232   static bool privileges = false;
 233   if (!init) {
 234     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 235     init = true;
 236   }
 237   return privileges;
 238 }
 239 
 240 
 241 #ifndef SYS_gettid
 242 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 243 #ifdef __ia64__
 244 #define SYS_gettid 1105
 245 #elif __i386__
 246 #define SYS_gettid 224
 247 #elif __amd64__
 248 #define SYS_gettid 186
 249 #elif __sparc__
 250 #define SYS_gettid 143
 251 #else
 252 #error define gettid for the arch
 253 #endif
 254 #endif
 255 
 256 // Cpu architecture string
 257 #if   defined(ZERO)
 258 static char cpu_arch[] = ZERO_LIBARCH;
 259 #elif defined(IA64)
 260 static char cpu_arch[] = "ia64";
 261 #elif defined(IA32)
 262 static char cpu_arch[] = "i386";
 263 #elif defined(AMD64)
 264 static char cpu_arch[] = "amd64";
 265 #elif defined(ARM)
 266 static char cpu_arch[] = "arm";
 267 #elif defined(PPC32)
 268 static char cpu_arch[] = "ppc";
 269 #elif defined(PPC64)
 270 static char cpu_arch[] = "ppc64";
 271 #elif defined(SPARC)
 272 #  ifdef _LP64
 273 static char cpu_arch[] = "sparcv9";
 274 #  else
 275 static char cpu_arch[] = "sparc";
 276 #  endif
 277 #else
 278 #error Add appropriate cpu_arch setting
 279 #endif
 280 
 281 
 282 // pid_t gettid()
 283 //
 284 // Returns the kernel thread id of the currently running thread. Kernel
 285 // thread id is used to access /proc.
 286 //
 287 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 288 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 289 //
 290 pid_t os::Linux::gettid() {
 291   int rslt = syscall(SYS_gettid);
 292   if (rslt == -1) {
 293      // old kernel, no NPTL support
 294      return getpid();
 295   } else {
 296      return (pid_t)rslt;
 297   }
 298 }
 299 
 300 // Most versions of linux have a bug where the number of processors are
 301 // determined by looking at the /proc file system.  In a chroot environment,
 302 // the system call returns 1.  This causes the VM to act as if it is
 303 // a single processor and elide locking (see is_MP() call).
 304 static bool unsafe_chroot_detected = false;
 305 static const char *unstable_chroot_error = "/proc file system not found.\n"
 306                      "Java may be unstable running multithreaded in a chroot "
 307                      "environment on Linux when /proc filesystem is not mounted.";
 308 
 309 void os::Linux::initialize_system_info() {
 310   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 311   if (processor_count() == 1) {
 312     pid_t pid = os::Linux::gettid();
 313     char fname[32];
 314     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 315     FILE *fp = fopen(fname, "r");
 316     if (fp == NULL) {
 317       unsafe_chroot_detected = true;
 318     } else {
 319       fclose(fp);
 320     }
 321   }
 322   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 323   assert(processor_count() > 0, "linux error");
 324 }
 325 
 326 void os::init_system_properties_values() {
 327   // The next steps are taken in the product version:
 328   //
 329   // Obtain the JAVA_HOME value from the location of libjvm.so.
 330   // This library should be located at:
 331   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 332   //
 333   // If "/jre/lib/" appears at the right place in the path, then we
 334   // assume libjvm.so is installed in a JDK and we use this path.
 335   //
 336   // Otherwise exit with message: "Could not create the Java virtual machine."
 337   //
 338   // The following extra steps are taken in the debugging version:
 339   //
 340   // If "/jre/lib/" does NOT appear at the right place in the path
 341   // instead of exit check for $JAVA_HOME environment variable.
 342   //
 343   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 344   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 345   // it looks like libjvm.so is installed there
 346   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 347   //
 348   // Otherwise exit.
 349   //
 350   // Important note: if the location of libjvm.so changes this
 351   // code needs to be changed accordingly.
 352 
 353 // See ld(1):
 354 //      The linker uses the following search paths to locate required
 355 //      shared libraries:
 356 //        1: ...
 357 //        ...
 358 //        7: The default directories, normally /lib and /usr/lib.
 359 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 360 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 361 #else
 362 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 363 #endif
 364 
 365 // Base path of extensions installed on the system.
 366 #define SYS_EXT_DIR     "/usr/java/packages"
 367 #define EXTENSIONS_DIR  "/lib/ext"
 368 #define ENDORSED_DIR    "/lib/endorsed"
 369 
 370   // Buffer that fits several sprintfs.
 371   // Note that the space for the colon and the trailing null are provided
 372   // by the nulls included by the sizeof operator.
 373   const size_t bufsize =
 374     MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
 375          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
 376          (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
 377   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 378 
 379   // sysclasspath, java_home, dll_dir
 380   {
 381     char *pslash;
 382     os::jvm_path(buf, bufsize);
 383 
 384     // Found the full path to libjvm.so.
 385     // Now cut the path to <java_home>/jre if we can.
 386     *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
 387     pslash = strrchr(buf, '/');
 388     if (pslash != NULL) {
 389       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 390     }
 391     Arguments::set_dll_dir(buf);
 392 
 393     if (pslash != NULL) {
 394       pslash = strrchr(buf, '/');
 395       if (pslash != NULL) {
 396         *pslash = '\0';          // Get rid of /<arch>.
 397         pslash = strrchr(buf, '/');
 398         if (pslash != NULL) {
 399           *pslash = '\0';        // Get rid of /lib.
 400         }
 401       }
 402     }
 403     Arguments::set_java_home(buf);
 404     set_boot_path('/', ':');
 405   }
 406 
 407   // Where to look for native libraries.
 408   //
 409   // Note: Due to a legacy implementation, most of the library path
 410   // is set in the launcher. This was to accomodate linking restrictions
 411   // on legacy Linux implementations (which are no longer supported).
 412   // Eventually, all the library path setting will be done here.
 413   //
 414   // However, to prevent the proliferation of improperly built native
 415   // libraries, the new path component /usr/java/packages is added here.
 416   // Eventually, all the library path setting will be done here.
 417   {
 418     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 419     // should always exist (until the legacy problem cited above is
 420     // addressed).
 421     const char *v = ::getenv("LD_LIBRARY_PATH");
 422     const char *v_colon = ":";
 423     if (v == NULL) { v = ""; v_colon = ""; }
 424     // That's +1 for the colon and +1 for the trailing '\0'.
 425     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 426                                                      strlen(v) + 1 +
 427                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 428                                                      mtInternal);
 429     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 430     Arguments::set_library_path(ld_library_path);
 431     FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
 432   }
 433 
 434   // Extensions directories.
 435   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 436   Arguments::set_ext_dirs(buf);
 437 
 438   // Endorsed standards default directory.
 439   sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 440   Arguments::set_endorsed_dirs(buf);
 441 
 442   FREE_C_HEAP_ARRAY(char, buf, mtInternal);
 443 
 444 #undef DEFAULT_LIBPATH
 445 #undef SYS_EXT_DIR
 446 #undef EXTENSIONS_DIR
 447 #undef ENDORSED_DIR
 448 }
 449 
 450 ////////////////////////////////////////////////////////////////////////////////
 451 // breakpoint support
 452 
 453 void os::breakpoint() {
 454   BREAKPOINT;
 455 }
 456 
 457 extern "C" void breakpoint() {
 458   // use debugger to set breakpoint here
 459 }
 460 
 461 ////////////////////////////////////////////////////////////////////////////////
 462 // signal support
 463 
 464 debug_only(static bool signal_sets_initialized = false);
 465 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 466 
 467 bool os::Linux::is_sig_ignored(int sig) {
 468       struct sigaction oact;
 469       sigaction(sig, (struct sigaction*)NULL, &oact);
 470       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 471                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 472       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 473            return true;
 474       else
 475            return false;
 476 }
 477 
 478 void os::Linux::signal_sets_init() {
 479   // Should also have an assertion stating we are still single-threaded.
 480   assert(!signal_sets_initialized, "Already initialized");
 481   // Fill in signals that are necessarily unblocked for all threads in
 482   // the VM. Currently, we unblock the following signals:
 483   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 484   //                         by -Xrs (=ReduceSignalUsage));
 485   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 486   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 487   // the dispositions or masks wrt these signals.
 488   // Programs embedding the VM that want to use the above signals for their
 489   // own purposes must, at this time, use the "-Xrs" option to prevent
 490   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 491   // (See bug 4345157, and other related bugs).
 492   // In reality, though, unblocking these signals is really a nop, since
 493   // these signals are not blocked by default.
 494   sigemptyset(&unblocked_sigs);
 495   sigemptyset(&allowdebug_blocked_sigs);
 496   sigaddset(&unblocked_sigs, SIGILL);
 497   sigaddset(&unblocked_sigs, SIGSEGV);
 498   sigaddset(&unblocked_sigs, SIGBUS);
 499   sigaddset(&unblocked_sigs, SIGFPE);
 500 #if defined(PPC64)
 501   sigaddset(&unblocked_sigs, SIGTRAP);
 502 #endif
 503   sigaddset(&unblocked_sigs, SR_signum);
 504 
 505   if (!ReduceSignalUsage) {
 506    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 507       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 508       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 509    }
 510    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 511       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 512       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 513    }
 514    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 515       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 516       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 517    }
 518   }
 519   // Fill in signals that are blocked by all but the VM thread.
 520   sigemptyset(&vm_sigs);
 521   if (!ReduceSignalUsage)
 522     sigaddset(&vm_sigs, BREAK_SIGNAL);
 523   debug_only(signal_sets_initialized = true);
 524 
 525 }
 526 
 527 // These are signals that are unblocked while a thread is running Java.
 528 // (For some reason, they get blocked by default.)
 529 sigset_t* os::Linux::unblocked_signals() {
 530   assert(signal_sets_initialized, "Not initialized");
 531   return &unblocked_sigs;
 532 }
 533 
 534 // These are the signals that are blocked while a (non-VM) thread is
 535 // running Java. Only the VM thread handles these signals.
 536 sigset_t* os::Linux::vm_signals() {
 537   assert(signal_sets_initialized, "Not initialized");
 538   return &vm_sigs;
 539 }
 540 
 541 // These are signals that are blocked during cond_wait to allow debugger in
 542 sigset_t* os::Linux::allowdebug_blocked_signals() {
 543   assert(signal_sets_initialized, "Not initialized");
 544   return &allowdebug_blocked_sigs;
 545 }
 546 
 547 void os::Linux::hotspot_sigmask(Thread* thread) {
 548 
 549   //Save caller's signal mask before setting VM signal mask
 550   sigset_t caller_sigmask;
 551   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 552 
 553   OSThread* osthread = thread->osthread();
 554   osthread->set_caller_sigmask(caller_sigmask);
 555 
 556   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 557 
 558   if (!ReduceSignalUsage) {
 559     if (thread->is_VM_thread()) {
 560       // Only the VM thread handles BREAK_SIGNAL ...
 561       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 562     } else {
 563       // ... all other threads block BREAK_SIGNAL
 564       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 565     }
 566   }
 567 }
 568 
 569 //////////////////////////////////////////////////////////////////////////////
 570 // detecting pthread library
 571 
 572 void os::Linux::libpthread_init() {
 573   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 574   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 575   // generic name for earlier versions.
 576   // Define macros here so we can build HotSpot on old systems.
 577 # ifndef _CS_GNU_LIBC_VERSION
 578 # define _CS_GNU_LIBC_VERSION 2
 579 # endif
 580 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 581 # define _CS_GNU_LIBPTHREAD_VERSION 3
 582 # endif
 583 
 584   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 585   if (n > 0) {
 586      char *str = (char *)malloc(n, mtInternal);
 587      confstr(_CS_GNU_LIBC_VERSION, str, n);
 588      os::Linux::set_glibc_version(str);
 589   } else {
 590      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 591      static char _gnu_libc_version[32];
 592      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 593               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 594      os::Linux::set_glibc_version(_gnu_libc_version);
 595   }
 596 
 597   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 598   if (n > 0) {
 599      char *str = (char *)malloc(n, mtInternal);
 600      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 601      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 602      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 603      // is the case. LinuxThreads has a hard limit on max number of threads.
 604      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 605      // On the other hand, NPTL does not have such a limit, sysconf()
 606      // will return -1 and errno is not changed. Check if it is really NPTL.
 607      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 608          strstr(str, "NPTL") &&
 609          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 610        free(str);
 611        os::Linux::set_libpthread_version("linuxthreads");
 612      } else {
 613        os::Linux::set_libpthread_version(str);
 614      }
 615   } else {
 616     // glibc before 2.3.2 only has LinuxThreads.
 617     os::Linux::set_libpthread_version("linuxthreads");
 618   }
 619 
 620   if (strstr(libpthread_version(), "NPTL")) {
 621      os::Linux::set_is_NPTL();
 622   } else {
 623      os::Linux::set_is_LinuxThreads();
 624   }
 625 
 626   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 627   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 628   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 629      os::Linux::set_is_floating_stack();
 630   }
 631 }
 632 
 633 /////////////////////////////////////////////////////////////////////////////
 634 // thread stack
 635 
 636 // Force Linux kernel to expand current thread stack. If "bottom" is close
 637 // to the stack guard, caller should block all signals.
 638 //
 639 // MAP_GROWSDOWN:
 640 //   A special mmap() flag that is used to implement thread stacks. It tells
 641 //   kernel that the memory region should extend downwards when needed. This
 642 //   allows early versions of LinuxThreads to only mmap the first few pages
 643 //   when creating a new thread. Linux kernel will automatically expand thread
 644 //   stack as needed (on page faults).
 645 //
 646 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 647 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 648 //   region, it's hard to tell if the fault is due to a legitimate stack
 649 //   access or because of reading/writing non-exist memory (e.g. buffer
 650 //   overrun). As a rule, if the fault happens below current stack pointer,
 651 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 652 //   application (see Linux kernel fault.c).
 653 //
 654 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 655 //   stack overflow detection.
 656 //
 657 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 658 //   not use this flag. However, the stack of initial thread is not created
 659 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 660 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 661 //   and then attach the thread to JVM.
 662 //
 663 // To get around the problem and allow stack banging on Linux, we need to
 664 // manually expand thread stack after receiving the SIGSEGV.
 665 //
 666 // There are two ways to expand thread stack to address "bottom", we used
 667 // both of them in JVM before 1.5:
 668 //   1. adjust stack pointer first so that it is below "bottom", and then
 669 //      touch "bottom"
 670 //   2. mmap() the page in question
 671 //
 672 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 673 // if current sp is already near the lower end of page 101, and we need to
 674 // call mmap() to map page 100, it is possible that part of the mmap() frame
 675 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 676 // That will destroy the mmap() frame and cause VM to crash.
 677 //
 678 // The following code works by adjusting sp first, then accessing the "bottom"
 679 // page to force a page fault. Linux kernel will then automatically expand the
 680 // stack mapping.
 681 //
 682 // _expand_stack_to() assumes its frame size is less than page size, which
 683 // should always be true if the function is not inlined.
 684 
 685 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 686 #define NOINLINE
 687 #else
 688 #define NOINLINE __attribute__ ((noinline))
 689 #endif
 690 
 691 static void _expand_stack_to(address bottom) NOINLINE;
 692 
 693 static void _expand_stack_to(address bottom) {
 694   address sp;
 695   size_t size;
 696   volatile char *p;
 697 
 698   // Adjust bottom to point to the largest address within the same page, it
 699   // gives us a one-page buffer if alloca() allocates slightly more memory.
 700   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 701   bottom += os::Linux::page_size() - 1;
 702 
 703   // sp might be slightly above current stack pointer; if that's the case, we
 704   // will alloca() a little more space than necessary, which is OK. Don't use
 705   // os::current_stack_pointer(), as its result can be slightly below current
 706   // stack pointer, causing us to not alloca enough to reach "bottom".
 707   sp = (address)&sp;
 708 
 709   if (sp > bottom) {
 710     size = sp - bottom;
 711     p = (volatile char *)alloca(size);
 712     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 713     p[0] = '\0';
 714   }
 715 }
 716 
 717 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 718   assert(t!=NULL, "just checking");
 719   assert(t->osthread()->expanding_stack(), "expand should be set");
 720   assert(t->stack_base() != NULL, "stack_base was not initialized");
 721 
 722   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 723     sigset_t mask_all, old_sigset;
 724     sigfillset(&mask_all);
 725     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 726     _expand_stack_to(addr);
 727     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 728     return true;
 729   }
 730   return false;
 731 }
 732 
 733 //////////////////////////////////////////////////////////////////////////////
 734 // create new thread
 735 
 736 static address highest_vm_reserved_address();
 737 
 738 // check if it's safe to start a new thread
 739 static bool _thread_safety_check(Thread* thread) {
 740   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 741     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 742     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 743     //   allocated (MAP_FIXED) from high address space. Every thread stack
 744     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 745     //   it to other values if they rebuild LinuxThreads).
 746     //
 747     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 748     // the memory region has already been mmap'ed. That means if we have too
 749     // many threads and/or very large heap, eventually thread stack will
 750     // collide with heap.
 751     //
 752     // Here we try to prevent heap/stack collision by comparing current
 753     // stack bottom with the highest address that has been mmap'ed by JVM
 754     // plus a safety margin for memory maps created by native code.
 755     //
 756     // This feature can be disabled by setting ThreadSafetyMargin to 0
 757     //
 758     if (ThreadSafetyMargin > 0) {
 759       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 760 
 761       // not safe if our stack extends below the safety margin
 762       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 763     } else {
 764       return true;
 765     }
 766   } else {
 767     // Floating stack LinuxThreads or NPTL:
 768     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 769     //   there's not enough space left, pthread_create() will fail. If we come
 770     //   here, that means enough space has been reserved for stack.
 771     return true;
 772   }
 773 }
 774 
 775 // Thread start routine for all newly created threads
 776 static void *java_start(Thread *thread) {
 777   // Try to randomize the cache line index of hot stack frames.
 778   // This helps when threads of the same stack traces evict each other's
 779   // cache lines. The threads can be either from the same JVM instance, or
 780   // from different JVM instances. The benefit is especially true for
 781   // processors with hyperthreading technology.
 782   static int counter = 0;
 783   int pid = os::current_process_id();
 784   alloca(((pid ^ counter++) & 7) * 128);
 785 
 786   ThreadLocalStorage::set_thread(thread);
 787 
 788   OSThread* osthread = thread->osthread();
 789   Monitor* sync = osthread->startThread_lock();
 790 
 791   // non floating stack LinuxThreads needs extra check, see above
 792   if (!_thread_safety_check(thread)) {
 793     // notify parent thread
 794     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 795     osthread->set_state(ZOMBIE);
 796     sync->notify_all();
 797     return NULL;
 798   }
 799 
 800   // thread_id is kernel thread id (similar to Solaris LWP id)
 801   osthread->set_thread_id(os::Linux::gettid());
 802 
 803   if (UseNUMA) {
 804     int lgrp_id = os::numa_get_group_id();
 805     if (lgrp_id != -1) {
 806       thread->set_lgrp_id(lgrp_id);
 807     }
 808   }
 809   // initialize signal mask for this thread
 810   os::Linux::hotspot_sigmask(thread);
 811 
 812   // initialize floating point control register
 813   os::Linux::init_thread_fpu_state();
 814 
 815   // handshaking with parent thread
 816   {
 817     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 818 
 819     // notify parent thread
 820     osthread->set_state(INITIALIZED);
 821     sync->notify_all();
 822 
 823     // wait until os::start_thread()
 824     while (osthread->get_state() == INITIALIZED) {
 825       sync->wait(Mutex::_no_safepoint_check_flag);
 826     }
 827   }
 828 
 829   // call one more level start routine
 830   thread->run();
 831 
 832   return 0;
 833 }
 834 
 835 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 836   assert(thread->osthread() == NULL, "caller responsible");
 837 
 838   // Allocate the OSThread object
 839   OSThread* osthread = new OSThread(NULL, NULL);
 840   if (osthread == NULL) {
 841     return false;
 842   }
 843 
 844   // set the correct thread state
 845   osthread->set_thread_type(thr_type);
 846 
 847   // Initial state is ALLOCATED but not INITIALIZED
 848   osthread->set_state(ALLOCATED);
 849 
 850   thread->set_osthread(osthread);
 851 
 852   // init thread attributes
 853   pthread_attr_t attr;
 854   pthread_attr_init(&attr);
 855   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 856 
 857   // stack size
 858   if (os::Linux::supports_variable_stack_size()) {
 859     // calculate stack size if it's not specified by caller
 860     if (stack_size == 0) {
 861       stack_size = os::Linux::default_stack_size(thr_type);
 862 
 863       switch (thr_type) {
 864       case os::java_thread:
 865         // Java threads use ThreadStackSize which default value can be
 866         // changed with the flag -Xss
 867         assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 868         stack_size = JavaThread::stack_size_at_create();
 869         break;
 870       case os::compiler_thread:
 871         if (CompilerThreadStackSize > 0) {
 872           stack_size = (size_t)(CompilerThreadStackSize * K);
 873           break;
 874         } // else fall through:
 875           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 876       case os::vm_thread:
 877       case os::pgc_thread:
 878       case os::cgc_thread:
 879       case os::watcher_thread:
 880         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 881         break;
 882       }
 883     }
 884 
 885     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 886     pthread_attr_setstacksize(&attr, stack_size);
 887   } else {
 888     // let pthread_create() pick the default value.
 889   }
 890 
 891   // glibc guard page
 892   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 893 
 894   ThreadState state;
 895 
 896   {
 897     // Serialize thread creation if we are running with fixed stack LinuxThreads
 898     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 899     if (lock) {
 900       os::Linux::createThread_lock()->lock_without_safepoint_check();
 901     }
 902 
 903     pthread_t tid;
 904     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 905 
 906     pthread_attr_destroy(&attr);
 907 
 908     if (ret != 0) {
 909       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 910         perror("pthread_create()");
 911       }
 912       // Need to clean up stuff we've allocated so far
 913       thread->set_osthread(NULL);
 914       delete osthread;
 915       if (lock) os::Linux::createThread_lock()->unlock();
 916       return false;
 917     }
 918 
 919     // Store pthread info into the OSThread
 920     osthread->set_pthread_id(tid);
 921 
 922     // Wait until child thread is either initialized or aborted
 923     {
 924       Monitor* sync_with_child = osthread->startThread_lock();
 925       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 926       while ((state = osthread->get_state()) == ALLOCATED) {
 927         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 928       }
 929     }
 930 
 931     if (lock) {
 932       os::Linux::createThread_lock()->unlock();
 933     }
 934   }
 935 
 936   // Aborted due to thread limit being reached
 937   if (state == ZOMBIE) {
 938       thread->set_osthread(NULL);
 939       delete osthread;
 940       return false;
 941   }
 942 
 943   // The thread is returned suspended (in state INITIALIZED),
 944   // and is started higher up in the call chain
 945   assert(state == INITIALIZED, "race condition");
 946   return true;
 947 }
 948 
 949 /////////////////////////////////////////////////////////////////////////////
 950 // attach existing thread
 951 
 952 // bootstrap the main thread
 953 bool os::create_main_thread(JavaThread* thread) {
 954   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 955   return create_attached_thread(thread);
 956 }
 957 
 958 bool os::create_attached_thread(JavaThread* thread) {
 959 #ifdef ASSERT
 960     thread->verify_not_published();
 961 #endif
 962 
 963   // Allocate the OSThread object
 964   OSThread* osthread = new OSThread(NULL, NULL);
 965 
 966   if (osthread == NULL) {
 967     return false;
 968   }
 969 
 970   // Store pthread info into the OSThread
 971   osthread->set_thread_id(os::Linux::gettid());
 972   osthread->set_pthread_id(::pthread_self());
 973 
 974   // initialize floating point control register
 975   os::Linux::init_thread_fpu_state();
 976 
 977   // Initial thread state is RUNNABLE
 978   osthread->set_state(RUNNABLE);
 979 
 980   thread->set_osthread(osthread);
 981 
 982   if (UseNUMA) {
 983     int lgrp_id = os::numa_get_group_id();
 984     if (lgrp_id != -1) {
 985       thread->set_lgrp_id(lgrp_id);
 986     }
 987   }
 988 
 989   if (os::Linux::is_initial_thread()) {
 990     // If current thread is initial thread, its stack is mapped on demand,
 991     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 992     // the entire stack region to avoid SEGV in stack banging.
 993     // It is also useful to get around the heap-stack-gap problem on SuSE
 994     // kernel (see 4821821 for details). We first expand stack to the top
 995     // of yellow zone, then enable stack yellow zone (order is significant,
 996     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 997     // is no gap between the last two virtual memory regions.
 998 
 999     JavaThread *jt = (JavaThread *)thread;
1000     address addr = jt->stack_yellow_zone_base();
1001     assert(addr != NULL, "initialization problem?");
1002     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1003 
1004     osthread->set_expanding_stack();
1005     os::Linux::manually_expand_stack(jt, addr);
1006     osthread->clear_expanding_stack();
1007   }
1008 
1009   // initialize signal mask for this thread
1010   // and save the caller's signal mask
1011   os::Linux::hotspot_sigmask(thread);
1012 
1013   return true;
1014 }
1015 
1016 void os::pd_start_thread(Thread* thread) {
1017   OSThread * osthread = thread->osthread();
1018   assert(osthread->get_state() != INITIALIZED, "just checking");
1019   Monitor* sync_with_child = osthread->startThread_lock();
1020   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1021   sync_with_child->notify();
1022 }
1023 
1024 // Free Linux resources related to the OSThread
1025 void os::free_thread(OSThread* osthread) {
1026   assert(osthread != NULL, "osthread not set");
1027 
1028   if (Thread::current()->osthread() == osthread) {
1029     // Restore caller's signal mask
1030     sigset_t sigmask = osthread->caller_sigmask();
1031     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1032    }
1033 
1034   delete osthread;
1035 }
1036 
1037 //////////////////////////////////////////////////////////////////////////////
1038 // thread local storage
1039 
1040 // Restore the thread pointer if the destructor is called. This is in case
1041 // someone from JNI code sets up a destructor with pthread_key_create to run
1042 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1043 // will hang or crash. When detachCurrentThread is called the key will be set
1044 // to null and we will not be called again. If detachCurrentThread is never
1045 // called we could loop forever depending on the pthread implementation.
1046 static void restore_thread_pointer(void* p) {
1047   Thread* thread = (Thread*) p;
1048   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1049 }
1050 
1051 int os::allocate_thread_local_storage() {
1052   pthread_key_t key;
1053   int rslt = pthread_key_create(&key, restore_thread_pointer);
1054   assert(rslt == 0, "cannot allocate thread local storage");
1055   return (int)key;
1056 }
1057 
1058 // Note: This is currently not used by VM, as we don't destroy TLS key
1059 // on VM exit.
1060 void os::free_thread_local_storage(int index) {
1061   int rslt = pthread_key_delete((pthread_key_t)index);
1062   assert(rslt == 0, "invalid index");
1063 }
1064 
1065 void os::thread_local_storage_at_put(int index, void* value) {
1066   int rslt = pthread_setspecific((pthread_key_t)index, value);
1067   assert(rslt == 0, "pthread_setspecific failed");
1068 }
1069 
1070 extern "C" Thread* get_thread() {
1071   return ThreadLocalStorage::thread();
1072 }
1073 
1074 //////////////////////////////////////////////////////////////////////////////
1075 // initial thread
1076 
1077 // Check if current thread is the initial thread, similar to Solaris thr_main.
1078 bool os::Linux::is_initial_thread(void) {
1079   char dummy;
1080   // If called before init complete, thread stack bottom will be null.
1081   // Can be called if fatal error occurs before initialization.
1082   if (initial_thread_stack_bottom() == NULL) return false;
1083   assert(initial_thread_stack_bottom() != NULL &&
1084          initial_thread_stack_size()   != 0,
1085          "os::init did not locate initial thread's stack region");
1086   if ((address)&dummy >= initial_thread_stack_bottom() &&
1087       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1088        return true;
1089   else return false;
1090 }
1091 
1092 // Find the virtual memory area that contains addr
1093 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1094   FILE *fp = fopen("/proc/self/maps", "r");
1095   if (fp) {
1096     address low, high;
1097     while (!feof(fp)) {
1098       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1099         if (low <= addr && addr < high) {
1100            if (vma_low)  *vma_low  = low;
1101            if (vma_high) *vma_high = high;
1102            fclose(fp);
1103            return true;
1104         }
1105       }
1106       for (;;) {
1107         int ch = fgetc(fp);
1108         if (ch == EOF || ch == (int)'\n') break;
1109       }
1110     }
1111     fclose(fp);
1112   }
1113   return false;
1114 }
1115 
1116 // Locate initial thread stack. This special handling of initial thread stack
1117 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1118 // bogus value for initial thread.
1119 void os::Linux::capture_initial_stack(size_t max_size) {
1120   // stack size is the easy part, get it from RLIMIT_STACK
1121   size_t stack_size;
1122   struct rlimit rlim;
1123   getrlimit(RLIMIT_STACK, &rlim);
1124   stack_size = rlim.rlim_cur;
1125 
1126   // 6308388: a bug in ld.so will relocate its own .data section to the
1127   //   lower end of primordial stack; reduce ulimit -s value a little bit
1128   //   so we won't install guard page on ld.so's data section.
1129   stack_size -= 2 * page_size();
1130 
1131   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1132   //   7.1, in both cases we will get 2G in return value.
1133   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1134   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1135   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1136   //   in case other parts in glibc still assumes 2M max stack size.
1137   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1138   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1139   if (stack_size > 2 * K * K IA64_ONLY(*2))
1140       stack_size = 2 * K * K IA64_ONLY(*2);
1141   // Try to figure out where the stack base (top) is. This is harder.
1142   //
1143   // When an application is started, glibc saves the initial stack pointer in
1144   // a global variable "__libc_stack_end", which is then used by system
1145   // libraries. __libc_stack_end should be pretty close to stack top. The
1146   // variable is available since the very early days. However, because it is
1147   // a private interface, it could disappear in the future.
1148   //
1149   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1150   // to __libc_stack_end, it is very close to stack top, but isn't the real
1151   // stack top. Note that /proc may not exist if VM is running as a chroot
1152   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1153   // /proc/<pid>/stat could change in the future (though unlikely).
1154   //
1155   // We try __libc_stack_end first. If that doesn't work, look for
1156   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1157   // as a hint, which should work well in most cases.
1158 
1159   uintptr_t stack_start;
1160 
1161   // try __libc_stack_end first
1162   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1163   if (p && *p) {
1164     stack_start = *p;
1165   } else {
1166     // see if we can get the start_stack field from /proc/self/stat
1167     FILE *fp;
1168     int pid;
1169     char state;
1170     int ppid;
1171     int pgrp;
1172     int session;
1173     int nr;
1174     int tpgrp;
1175     unsigned long flags;
1176     unsigned long minflt;
1177     unsigned long cminflt;
1178     unsigned long majflt;
1179     unsigned long cmajflt;
1180     unsigned long utime;
1181     unsigned long stime;
1182     long cutime;
1183     long cstime;
1184     long prio;
1185     long nice;
1186     long junk;
1187     long it_real;
1188     uintptr_t start;
1189     uintptr_t vsize;
1190     intptr_t rss;
1191     uintptr_t rsslim;
1192     uintptr_t scodes;
1193     uintptr_t ecode;
1194     int i;
1195 
1196     // Figure what the primordial thread stack base is. Code is inspired
1197     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1198     // followed by command name surrounded by parentheses, state, etc.
1199     char stat[2048];
1200     int statlen;
1201 
1202     fp = fopen("/proc/self/stat", "r");
1203     if (fp) {
1204       statlen = fread(stat, 1, 2047, fp);
1205       stat[statlen] = '\0';
1206       fclose(fp);
1207 
1208       // Skip pid and the command string. Note that we could be dealing with
1209       // weird command names, e.g. user could decide to rename java launcher
1210       // to "java 1.4.2 :)", then the stat file would look like
1211       //                1234 (java 1.4.2 :)) R ... ...
1212       // We don't really need to know the command string, just find the last
1213       // occurrence of ")" and then start parsing from there. See bug 4726580.
1214       char * s = strrchr(stat, ')');
1215 
1216       i = 0;
1217       if (s) {
1218         // Skip blank chars
1219         do s++; while (isspace(*s));
1220 
1221 #define _UFM UINTX_FORMAT
1222 #define _DFM INTX_FORMAT
1223 
1224         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1225         /*              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 */
1226         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,
1227              &state,          /* 3  %c  */
1228              &ppid,           /* 4  %d  */
1229              &pgrp,           /* 5  %d  */
1230              &session,        /* 6  %d  */
1231              &nr,             /* 7  %d  */
1232              &tpgrp,          /* 8  %d  */
1233              &flags,          /* 9  %lu  */
1234              &minflt,         /* 10 %lu  */
1235              &cminflt,        /* 11 %lu  */
1236              &majflt,         /* 12 %lu  */
1237              &cmajflt,        /* 13 %lu  */
1238              &utime,          /* 14 %lu  */
1239              &stime,          /* 15 %lu  */
1240              &cutime,         /* 16 %ld  */
1241              &cstime,         /* 17 %ld  */
1242              &prio,           /* 18 %ld  */
1243              &nice,           /* 19 %ld  */
1244              &junk,           /* 20 %ld  */
1245              &it_real,        /* 21 %ld  */
1246              &start,          /* 22 UINTX_FORMAT */
1247              &vsize,          /* 23 UINTX_FORMAT */
1248              &rss,            /* 24 INTX_FORMAT  */
1249              &rsslim,         /* 25 UINTX_FORMAT */
1250              &scodes,         /* 26 UINTX_FORMAT */
1251              &ecode,          /* 27 UINTX_FORMAT */
1252              &stack_start);   /* 28 UINTX_FORMAT */
1253       }
1254 
1255 #undef _UFM
1256 #undef _DFM
1257 
1258       if (i != 28 - 2) {
1259          assert(false, "Bad conversion from /proc/self/stat");
1260          // product mode - assume we are the initial thread, good luck in the
1261          // embedded case.
1262          warning("Can't detect initial thread stack location - bad conversion");
1263          stack_start = (uintptr_t) &rlim;
1264       }
1265     } else {
1266       // For some reason we can't open /proc/self/stat (for example, running on
1267       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1268       // most cases, so don't abort:
1269       warning("Can't detect initial thread stack location - no /proc/self/stat");
1270       stack_start = (uintptr_t) &rlim;
1271     }
1272   }
1273 
1274   // Now we have a pointer (stack_start) very close to the stack top, the
1275   // next thing to do is to figure out the exact location of stack top. We
1276   // can find out the virtual memory area that contains stack_start by
1277   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1278   // and its upper limit is the real stack top. (again, this would fail if
1279   // running inside chroot, because /proc may not exist.)
1280 
1281   uintptr_t stack_top;
1282   address low, high;
1283   if (find_vma((address)stack_start, &low, &high)) {
1284     // success, "high" is the true stack top. (ignore "low", because initial
1285     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1286     stack_top = (uintptr_t)high;
1287   } else {
1288     // failed, likely because /proc/self/maps does not exist
1289     warning("Can't detect initial thread stack location - find_vma failed");
1290     // best effort: stack_start is normally within a few pages below the real
1291     // stack top, use it as stack top, and reduce stack size so we won't put
1292     // guard page outside stack.
1293     stack_top = stack_start;
1294     stack_size -= 16 * page_size();
1295   }
1296 
1297   // stack_top could be partially down the page so align it
1298   stack_top = align_size_up(stack_top, page_size());
1299 
1300   if (max_size && stack_size > max_size) {
1301      _initial_thread_stack_size = max_size;
1302   } else {
1303      _initial_thread_stack_size = stack_size;
1304   }
1305 
1306   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1307   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1308 }
1309 
1310 ////////////////////////////////////////////////////////////////////////////////
1311 // time support
1312 
1313 // Time since start-up in seconds to a fine granularity.
1314 // Used by VMSelfDestructTimer and the MemProfiler.
1315 double os::elapsedTime() {
1316 
1317   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1318 }
1319 
1320 jlong os::elapsed_counter() {
1321   return javaTimeNanos() - initial_time_count;
1322 }
1323 
1324 jlong os::elapsed_frequency() {
1325   return NANOSECS_PER_SEC; // nanosecond resolution
1326 }
1327 
1328 bool os::supports_vtime() { return true; }
1329 bool os::enable_vtime()   { return false; }
1330 bool os::vtime_enabled()  { return false; }
1331 
1332 double os::elapsedVTime() {
1333   struct rusage usage;
1334   int retval = getrusage(RUSAGE_THREAD, &usage);
1335   if (retval == 0) {
1336     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);
1337   } else {
1338     // better than nothing, but not much
1339     return elapsedTime();
1340   }
1341 }
1342 
1343 jlong os::javaTimeMillis() {
1344   timeval time;
1345   int status = gettimeofday(&time, NULL);
1346   assert(status != -1, "linux error");
1347   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1348 }
1349 
1350 #ifndef CLOCK_MONOTONIC
1351 #define CLOCK_MONOTONIC (1)
1352 #endif
1353 
1354 void os::Linux::clock_init() {
1355   // we do dlopen's in this particular order due to bug in linux
1356   // dynamical loader (see 6348968) leading to crash on exit
1357   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1358   if (handle == NULL) {
1359     handle = dlopen("librt.so", RTLD_LAZY);
1360   }
1361 
1362   if (handle) {
1363     int (*clock_getres_func)(clockid_t, struct timespec*) =
1364            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1365     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1366            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1367     if (clock_getres_func && clock_gettime_func) {
1368       // See if monotonic clock is supported by the kernel. Note that some
1369       // early implementations simply return kernel jiffies (updated every
1370       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1371       // for nano time (though the monotonic property is still nice to have).
1372       // It's fixed in newer kernels, however clock_getres() still returns
1373       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1374       // resolution for now. Hopefully as people move to new kernels, this
1375       // won't be a problem.
1376       struct timespec res;
1377       struct timespec tp;
1378       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1379           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1380         // yes, monotonic clock is supported
1381         _clock_gettime = clock_gettime_func;
1382         return;
1383       } else {
1384         // close librt if there is no monotonic clock
1385         dlclose(handle);
1386       }
1387     }
1388   }
1389   warning("No monotonic clock was available - timed services may " \
1390           "be adversely affected if the time-of-day clock changes");
1391 }
1392 
1393 #ifndef SYS_clock_getres
1394 
1395 #if defined(IA32) || defined(AMD64)
1396 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1397 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1398 #else
1399 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1400 #define sys_clock_getres(x,y)  -1
1401 #endif
1402 
1403 #else
1404 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1405 #endif
1406 
1407 void os::Linux::fast_thread_clock_init() {
1408   if (!UseLinuxPosixThreadCPUClocks) {
1409     return;
1410   }
1411   clockid_t clockid;
1412   struct timespec tp;
1413   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1414       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1415 
1416   // Switch to using fast clocks for thread cpu time if
1417   // the sys_clock_getres() returns 0 error code.
1418   // Note, that some kernels may support the current thread
1419   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1420   // returned by the pthread_getcpuclockid().
1421   // If the fast Posix clocks are supported then the sys_clock_getres()
1422   // must return at least tp.tv_sec == 0 which means a resolution
1423   // better than 1 sec. This is extra check for reliability.
1424 
1425   if (pthread_getcpuclockid_func &&
1426      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1427      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1428 
1429     _supports_fast_thread_cpu_time = true;
1430     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1431   }
1432 }
1433 
1434 jlong os::javaTimeNanos() {
1435   if (os::supports_monotonic_clock()) {
1436     struct timespec tp;
1437     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1438     assert(status == 0, "gettime error");
1439     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1440     return result;
1441   } else {
1442     timeval time;
1443     int status = gettimeofday(&time, NULL);
1444     assert(status != -1, "linux error");
1445     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1446     return 1000 * usecs;
1447   }
1448 }
1449 
1450 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1451   if (os::supports_monotonic_clock()) {
1452     info_ptr->max_value = ALL_64_BITS;
1453 
1454     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1455     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1456     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1457   } else {
1458     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1459     info_ptr->max_value = ALL_64_BITS;
1460 
1461     // gettimeofday is a real time clock so it skips
1462     info_ptr->may_skip_backward = true;
1463     info_ptr->may_skip_forward = true;
1464   }
1465 
1466   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1467 }
1468 
1469 // Return the real, user, and system times in seconds from an
1470 // arbitrary fixed point in the past.
1471 bool os::getTimesSecs(double* process_real_time,
1472                       double* process_user_time,
1473                       double* process_system_time) {
1474   struct tms ticks;
1475   clock_t real_ticks = times(&ticks);
1476 
1477   if (real_ticks == (clock_t) (-1)) {
1478     return false;
1479   } else {
1480     double ticks_per_second = (double) clock_tics_per_sec;
1481     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1482     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1483     *process_real_time = ((double) real_ticks) / ticks_per_second;
1484 
1485     return true;
1486   }
1487 }
1488 
1489 
1490 char * os::local_time_string(char *buf, size_t buflen) {
1491   struct tm t;
1492   time_t long_time;
1493   time(&long_time);
1494   localtime_r(&long_time, &t);
1495   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1496                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1497                t.tm_hour, t.tm_min, t.tm_sec);
1498   return buf;
1499 }
1500 
1501 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1502   return localtime_r(clock, res);
1503 }
1504 
1505 ////////////////////////////////////////////////////////////////////////////////
1506 // runtime exit support
1507 
1508 // Note: os::shutdown() might be called very early during initialization, or
1509 // called from signal handler. Before adding something to os::shutdown(), make
1510 // sure it is async-safe and can handle partially initialized VM.
1511 void os::shutdown() {
1512 
1513   // allow PerfMemory to attempt cleanup of any persistent resources
1514   perfMemory_exit();
1515 
1516   // needs to remove object in file system
1517   AttachListener::abort();
1518 
1519   // flush buffered output, finish log files
1520   ostream_abort();
1521 
1522   // Check for abort hook
1523   abort_hook_t abort_hook = Arguments::abort_hook();
1524   if (abort_hook != NULL) {
1525     abort_hook();
1526   }
1527 
1528 }
1529 
1530 // Note: os::abort() might be called very early during initialization, or
1531 // called from signal handler. Before adding something to os::abort(), make
1532 // sure it is async-safe and can handle partially initialized VM.
1533 void os::abort(bool dump_core) {
1534   os::shutdown();
1535   if (dump_core) {
1536 #ifndef PRODUCT
1537     fdStream out(defaultStream::output_fd());
1538     out.print_raw("Current thread is ");
1539     char buf[16];
1540     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1541     out.print_raw_cr(buf);
1542     out.print_raw_cr("Dumping core ...");
1543 #endif
1544     ::abort(); // dump core
1545   }
1546 
1547   ::exit(1);
1548 }
1549 
1550 // Die immediately, no exit hook, no abort hook, no cleanup.
1551 void os::die() {
1552   // _exit() on LinuxThreads only kills current thread
1553   ::abort();
1554 }
1555 
1556 
1557 // This method is a copy of JDK's sysGetLastErrorString
1558 // from src/solaris/hpi/src/system_md.c
1559 
1560 size_t os::lasterror(char *buf, size_t len) {
1561 
1562   if (errno == 0)  return 0;
1563 
1564   const char *s = ::strerror(errno);
1565   size_t n = ::strlen(s);
1566   if (n >= len) {
1567     n = len - 1;
1568   }
1569   ::strncpy(buf, s, n);
1570   buf[n] = '\0';
1571   return n;
1572 }
1573 
1574 intx os::current_thread_id() { return (intx)pthread_self(); }
1575 int os::current_process_id() {
1576 
1577   // Under the old linux thread library, linux gives each thread
1578   // its own process id. Because of this each thread will return
1579   // a different pid if this method were to return the result
1580   // of getpid(2). Linux provides no api that returns the pid
1581   // of the launcher thread for the vm. This implementation
1582   // returns a unique pid, the pid of the launcher thread
1583   // that starts the vm 'process'.
1584 
1585   // Under the NPTL, getpid() returns the same pid as the
1586   // launcher thread rather than a unique pid per thread.
1587   // Use gettid() if you want the old pre NPTL behaviour.
1588 
1589   // if you are looking for the result of a call to getpid() that
1590   // returns a unique pid for the calling thread, then look at the
1591   // OSThread::thread_id() method in osThread_linux.hpp file
1592 
1593   return (int)(_initial_pid ? _initial_pid : getpid());
1594 }
1595 
1596 // DLL functions
1597 
1598 const char* os::dll_file_extension() { return ".so"; }
1599 
1600 // This must be hard coded because it's the system's temporary
1601 // directory not the java application's temp directory, ala java.io.tmpdir.
1602 const char* os::get_temp_directory() { return "/tmp"; }
1603 
1604 static bool file_exists(const char* filename) {
1605   struct stat statbuf;
1606   if (filename == NULL || strlen(filename) == 0) {
1607     return false;
1608   }
1609   return os::stat(filename, &statbuf) == 0;
1610 }
1611 
1612 bool os::dll_build_name(char* buffer, size_t buflen,
1613                         const char* pname, const char* fname) {
1614   bool retval = false;
1615   // Copied from libhpi
1616   const size_t pnamelen = pname ? strlen(pname) : 0;
1617 
1618   // Return error on buffer overflow.
1619   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1620     return retval;
1621   }
1622 
1623   if (pnamelen == 0) {
1624     snprintf(buffer, buflen, "lib%s.so", fname);
1625     retval = true;
1626   } else if (strchr(pname, *os::path_separator()) != NULL) {
1627     int n;
1628     char** pelements = split_path(pname, &n);
1629     if (pelements == NULL) {
1630       return false;
1631     }
1632     for (int i = 0; i < n; i++) {
1633       // Really shouldn't be NULL, but check can't hurt
1634       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1635         continue; // skip the empty path values
1636       }
1637       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1638       if (file_exists(buffer)) {
1639         retval = true;
1640         break;
1641       }
1642     }
1643     // release the storage
1644     for (int i = 0; i < n; i++) {
1645       if (pelements[i] != NULL) {
1646         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1647       }
1648     }
1649     if (pelements != NULL) {
1650       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1651     }
1652   } else {
1653     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1654     retval = true;
1655   }
1656   return retval;
1657 }
1658 
1659 // check if addr is inside libjvm.so
1660 bool os::address_is_in_vm(address addr) {
1661   static address libjvm_base_addr;
1662   Dl_info dlinfo;
1663 
1664   if (libjvm_base_addr == NULL) {
1665     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1666       libjvm_base_addr = (address)dlinfo.dli_fbase;
1667     }
1668     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1669   }
1670 
1671   if (dladdr((void *)addr, &dlinfo) != 0) {
1672     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1673   }
1674 
1675   return false;
1676 }
1677 
1678 bool os::dll_address_to_function_name(address addr, char *buf,
1679                                       int buflen, int *offset) {
1680   // buf is not optional, but offset is optional
1681   assert(buf != NULL, "sanity check");
1682 
1683   Dl_info dlinfo;
1684 
1685   if (dladdr((void*)addr, &dlinfo) != 0) {
1686     // see if we have a matching symbol
1687     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1688       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1689         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1690       }
1691       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1692       return true;
1693     }
1694     // no matching symbol so try for just file info
1695     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1696       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1697                           buf, buflen, offset, dlinfo.dli_fname)) {
1698         return true;
1699       }
1700     }
1701   }
1702 
1703   buf[0] = '\0';
1704   if (offset != NULL) *offset = -1;
1705   return false;
1706 }
1707 
1708 struct _address_to_library_name {
1709   address addr;          // input : memory address
1710   size_t  buflen;        //         size of fname
1711   char*   fname;         // output: library name
1712   address base;          //         library base addr
1713 };
1714 
1715 static int address_to_library_name_callback(struct dl_phdr_info *info,
1716                                             size_t size, void *data) {
1717   int i;
1718   bool found = false;
1719   address libbase = NULL;
1720   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1721 
1722   // iterate through all loadable segments
1723   for (i = 0; i < info->dlpi_phnum; i++) {
1724     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1725     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1726       // base address of a library is the lowest address of its loaded
1727       // segments.
1728       if (libbase == NULL || libbase > segbase) {
1729         libbase = segbase;
1730       }
1731       // see if 'addr' is within current segment
1732       if (segbase <= d->addr &&
1733           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1734         found = true;
1735       }
1736     }
1737   }
1738 
1739   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1740   // so dll_address_to_library_name() can fall through to use dladdr() which
1741   // can figure out executable name from argv[0].
1742   if (found && info->dlpi_name && info->dlpi_name[0]) {
1743     d->base = libbase;
1744     if (d->fname) {
1745       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1746     }
1747     return 1;
1748   }
1749   return 0;
1750 }
1751 
1752 bool os::dll_address_to_library_name(address addr, char* buf,
1753                                      int buflen, int* offset) {
1754   // buf is not optional, but offset is optional
1755   assert(buf != NULL, "sanity check");
1756 
1757   Dl_info dlinfo;
1758   struct _address_to_library_name data;
1759 
1760   // There is a bug in old glibc dladdr() implementation that it could resolve
1761   // to wrong library name if the .so file has a base address != NULL. Here
1762   // we iterate through the program headers of all loaded libraries to find
1763   // out which library 'addr' really belongs to. This workaround can be
1764   // removed once the minimum requirement for glibc is moved to 2.3.x.
1765   data.addr = addr;
1766   data.fname = buf;
1767   data.buflen = buflen;
1768   data.base = NULL;
1769   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1770 
1771   if (rslt) {
1772      // buf already contains library name
1773      if (offset) *offset = addr - data.base;
1774      return true;
1775   }
1776   if (dladdr((void*)addr, &dlinfo) != 0) {
1777     if (dlinfo.dli_fname != NULL) {
1778       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1779     }
1780     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1781       *offset = addr - (address)dlinfo.dli_fbase;
1782     }
1783     return true;
1784   }
1785 
1786   buf[0] = '\0';
1787   if (offset) *offset = -1;
1788   return false;
1789 }
1790 
1791   // Loads .dll/.so and
1792   // in case of error it checks if .dll/.so was built for the
1793   // same architecture as Hotspot is running on
1794 
1795 
1796 // Remember the stack's state. The Linux dynamic linker will change
1797 // the stack to 'executable' at most once, so we must safepoint only once.
1798 bool os::Linux::_stack_is_executable = false;
1799 
1800 // VM operation that loads a library.  This is necessary if stack protection
1801 // of the Java stacks can be lost during loading the library.  If we
1802 // do not stop the Java threads, they can stack overflow before the stacks
1803 // are protected again.
1804 class VM_LinuxDllLoad: public VM_Operation {
1805  private:
1806   const char *_filename;
1807   char *_ebuf;
1808   int _ebuflen;
1809   void *_lib;
1810  public:
1811   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1812     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1813   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1814   void doit() {
1815     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1816     os::Linux::_stack_is_executable = true;
1817   }
1818   void* loaded_library() { return _lib; }
1819 };
1820 
1821 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1822 {
1823   void * result = NULL;
1824   bool load_attempted = false;
1825 
1826   // Check whether the library to load might change execution rights
1827   // of the stack. If they are changed, the protection of the stack
1828   // guard pages will be lost. We need a safepoint to fix this.
1829   //
1830   // See Linux man page execstack(8) for more info.
1831   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1832     ElfFile ef(filename);
1833     if (!ef.specifies_noexecstack()) {
1834       if (!is_init_completed()) {
1835         os::Linux::_stack_is_executable = true;
1836         // This is OK - No Java threads have been created yet, and hence no
1837         // stack guard pages to fix.
1838         //
1839         // This should happen only when you are building JDK7 using a very
1840         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1841         //
1842         // Dynamic loader will make all stacks executable after
1843         // this function returns, and will not do that again.
1844         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1845       } else {
1846         warning("You have loaded library %s which might have disabled stack guard. "
1847                 "The VM will try to fix the stack guard now.\n"
1848                 "It's highly recommended that you fix the library with "
1849                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1850                 filename);
1851 
1852         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1853         JavaThread *jt = JavaThread::current();
1854         if (jt->thread_state() != _thread_in_native) {
1855           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1856           // that requires ExecStack. Cannot enter safe point. Let's give up.
1857           warning("Unable to fix stack guard. Giving up.");
1858         } else {
1859           if (!LoadExecStackDllInVMThread) {
1860             // This is for the case where the DLL has an static
1861             // constructor function that executes JNI code. We cannot
1862             // load such DLLs in the VMThread.
1863             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1864           }
1865 
1866           ThreadInVMfromNative tiv(jt);
1867           debug_only(VMNativeEntryWrapper vew;)
1868 
1869           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1870           VMThread::execute(&op);
1871           if (LoadExecStackDllInVMThread) {
1872             result = op.loaded_library();
1873           }
1874           load_attempted = true;
1875         }
1876       }
1877     }
1878   }
1879 
1880   if (!load_attempted) {
1881     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1882   }
1883 
1884   if (result != NULL) {
1885     // Successful loading
1886     return result;
1887   }
1888 
1889   Elf32_Ehdr elf_head;
1890   int diag_msg_max_length=ebuflen-strlen(ebuf);
1891   char* diag_msg_buf=ebuf+strlen(ebuf);
1892 
1893   if (diag_msg_max_length==0) {
1894     // No more space in ebuf for additional diagnostics message
1895     return NULL;
1896   }
1897 
1898 
1899   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1900 
1901   if (file_descriptor < 0) {
1902     // Can't open library, report dlerror() message
1903     return NULL;
1904   }
1905 
1906   bool failed_to_read_elf_head=
1907     (sizeof(elf_head)!=
1908         (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1909 
1910   ::close(file_descriptor);
1911   if (failed_to_read_elf_head) {
1912     // file i/o error - report dlerror() msg
1913     return NULL;
1914   }
1915 
1916   typedef struct {
1917     Elf32_Half  code;         // Actual value as defined in elf.h
1918     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1919     char        elf_class;    // 32 or 64 bit
1920     char        endianess;    // MSB or LSB
1921     char*       name;         // String representation
1922   } arch_t;
1923 
1924   #ifndef EM_486
1925   #define EM_486          6               /* Intel 80486 */
1926   #endif
1927 
1928   static const arch_t arch_array[]={
1929     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1930     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1931     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1932     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1933     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1934     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1935     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1936     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1937 #if defined(VM_LITTLE_ENDIAN)
1938     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1939 #else
1940     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1941 #endif
1942     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1943     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1944     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1945     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1946     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1947     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1948     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1949   };
1950 
1951   #if  (defined IA32)
1952     static  Elf32_Half running_arch_code=EM_386;
1953   #elif   (defined AMD64)
1954     static  Elf32_Half running_arch_code=EM_X86_64;
1955   #elif  (defined IA64)
1956     static  Elf32_Half running_arch_code=EM_IA_64;
1957   #elif  (defined __sparc) && (defined _LP64)
1958     static  Elf32_Half running_arch_code=EM_SPARCV9;
1959   #elif  (defined __sparc) && (!defined _LP64)
1960     static  Elf32_Half running_arch_code=EM_SPARC;
1961   #elif  (defined __powerpc64__)
1962     static  Elf32_Half running_arch_code=EM_PPC64;
1963   #elif  (defined __powerpc__)
1964     static  Elf32_Half running_arch_code=EM_PPC;
1965   #elif  (defined ARM)
1966     static  Elf32_Half running_arch_code=EM_ARM;
1967   #elif  (defined S390)
1968     static  Elf32_Half running_arch_code=EM_S390;
1969   #elif  (defined ALPHA)
1970     static  Elf32_Half running_arch_code=EM_ALPHA;
1971   #elif  (defined MIPSEL)
1972     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1973   #elif  (defined PARISC)
1974     static  Elf32_Half running_arch_code=EM_PARISC;
1975   #elif  (defined MIPS)
1976     static  Elf32_Half running_arch_code=EM_MIPS;
1977   #elif  (defined M68K)
1978     static  Elf32_Half running_arch_code=EM_68K;
1979   #else
1980     #error Method os::dll_load requires that one of following is defined:\
1981          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1982   #endif
1983 
1984   // Identify compatability class for VM's architecture and library's architecture
1985   // Obtain string descriptions for architectures
1986 
1987   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1988   int running_arch_index=-1;
1989 
1990   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1991     if (running_arch_code == arch_array[i].code) {
1992       running_arch_index    = i;
1993     }
1994     if (lib_arch.code == arch_array[i].code) {
1995       lib_arch.compat_class = arch_array[i].compat_class;
1996       lib_arch.name         = arch_array[i].name;
1997     }
1998   }
1999 
2000   assert(running_arch_index != -1,
2001     "Didn't find running architecture code (running_arch_code) in arch_array");
2002   if (running_arch_index == -1) {
2003     // Even though running architecture detection failed
2004     // we may still continue with reporting dlerror() message
2005     return NULL;
2006   }
2007 
2008   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2009     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2010     return NULL;
2011   }
2012 
2013 #ifndef S390
2014   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2015     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2016     return NULL;
2017   }
2018 #endif // !S390
2019 
2020   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2021     if (lib_arch.name!=NULL) {
2022       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2023         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2024         lib_arch.name, arch_array[running_arch_index].name);
2025     } else {
2026       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2027       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2028         lib_arch.code,
2029         arch_array[running_arch_index].name);
2030     }
2031   }
2032 
2033   return NULL;
2034 }
2035 
2036 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2037   void * result = ::dlopen(filename, RTLD_LAZY);
2038   if (result == NULL) {
2039     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2040     ebuf[ebuflen-1] = '\0';
2041   }
2042   return result;
2043 }
2044 
2045 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2046   void * result = NULL;
2047   if (LoadExecStackDllInVMThread) {
2048     result = dlopen_helper(filename, ebuf, ebuflen);
2049   }
2050 
2051   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2052   // library that requires an executable stack, or which does not have this
2053   // stack attribute set, dlopen changes the stack attribute to executable. The
2054   // read protection of the guard pages gets lost.
2055   //
2056   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2057   // may have been queued at the same time.
2058 
2059   if (!_stack_is_executable) {
2060     JavaThread *jt = Threads::first();
2061 
2062     while (jt) {
2063       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2064           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2065         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2066                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2067           warning("Attempt to reguard stack yellow zone failed.");
2068         }
2069       }
2070       jt = jt->next();
2071     }
2072   }
2073 
2074   return result;
2075 }
2076 
2077 /*
2078  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2079  * chances are you might want to run the generated bits against glibc-2.0
2080  * libdl.so, so always use locking for any version of glibc.
2081  */
2082 void* os::dll_lookup(void* handle, const char* name) {
2083   pthread_mutex_lock(&dl_mutex);
2084   void* res = dlsym(handle, name);
2085   pthread_mutex_unlock(&dl_mutex);
2086   return res;
2087 }
2088 
2089 void* os::get_default_process_handle() {
2090   return (void*)::dlopen(NULL, RTLD_LAZY);
2091 }
2092 
2093 static bool _print_ascii_file(const char* filename, outputStream* st) {
2094   int fd = ::open(filename, O_RDONLY);
2095   if (fd == -1) {
2096      return false;
2097   }
2098 
2099   char buf[32];
2100   int bytes;
2101   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2102     st->print_raw(buf, bytes);
2103   }
2104 
2105   ::close(fd);
2106 
2107   return true;
2108 }
2109 
2110 void os::print_dll_info(outputStream *st) {
2111    st->print_cr("Dynamic libraries:");
2112 
2113    char fname[32];
2114    pid_t pid = os::Linux::gettid();
2115 
2116    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2117 
2118    if (!_print_ascii_file(fname, st)) {
2119      st->print("Can not get library information for pid = %d\n", pid);
2120    }
2121 }
2122 
2123 void os::print_os_info_brief(outputStream* st) {
2124   os::Linux::print_distro_info(st);
2125 
2126   os::Posix::print_uname_info(st);
2127 
2128   os::Linux::print_libversion_info(st);
2129 
2130 }
2131 
2132 void os::print_os_info(outputStream* st) {
2133   st->print("OS:");
2134 
2135   os::Linux::print_distro_info(st);
2136 
2137   os::Posix::print_uname_info(st);
2138 
2139   // Print warning if unsafe chroot environment detected
2140   if (unsafe_chroot_detected) {
2141     st->print("WARNING!! ");
2142     st->print_cr("%s", unstable_chroot_error);
2143   }
2144 
2145   os::Linux::print_libversion_info(st);
2146 
2147   os::Posix::print_rlimit_info(st);
2148 
2149   os::Posix::print_load_average(st);
2150 
2151   os::Linux::print_full_memory_info(st);
2152 }
2153 
2154 // Try to identify popular distros.
2155 // Most Linux distributions have a /etc/XXX-release file, which contains
2156 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2157 // file that also contains the OS version string. Some have more than one
2158 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2159 // /etc/redhat-release.), so the order is important.
2160 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2161 // their own specific XXX-release file as well as a redhat-release file.
2162 // Because of this the XXX-release file needs to be searched for before the
2163 // redhat-release file.
2164 // Since Red Hat has a lsb-release file that is not very descriptive the
2165 // search for redhat-release needs to be before lsb-release.
2166 // Since the lsb-release file is the new standard it needs to be searched
2167 // before the older style release files.
2168 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2169 // next to last resort.  The os-release file is a new standard that contains
2170 // distribution information and the system-release file seems to be an old
2171 // standard that has been replaced by the lsb-release and os-release files.
2172 // Searching for the debian_version file is the last resort.  It contains
2173 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2174 // "Debian " is printed before the contents of the debian_version file.
2175 void os::Linux::print_distro_info(outputStream* st) {
2176    if (!_print_ascii_file("/etc/oracle-release", st) &&
2177        !_print_ascii_file("/etc/mandriva-release", st) &&
2178        !_print_ascii_file("/etc/mandrake-release", st) &&
2179        !_print_ascii_file("/etc/sun-release", st) &&
2180        !_print_ascii_file("/etc/redhat-release", st) &&
2181        !_print_ascii_file("/etc/lsb-release", st) &&
2182        !_print_ascii_file("/etc/SuSE-release", st) &&
2183        !_print_ascii_file("/etc/turbolinux-release", st) &&
2184        !_print_ascii_file("/etc/gentoo-release", st) &&
2185        !_print_ascii_file("/etc/ltib-release", st) &&
2186        !_print_ascii_file("/etc/angstrom-version", st) &&
2187        !_print_ascii_file("/etc/system-release", st) &&
2188        !_print_ascii_file("/etc/os-release", st)) {
2189 
2190        if (file_exists("/etc/debian_version")) {
2191          st->print("Debian ");
2192          _print_ascii_file("/etc/debian_version", st);
2193        } else {
2194          st->print("Linux");
2195        }
2196    }
2197    st->cr();
2198 }
2199 
2200 void os::Linux::print_libversion_info(outputStream* st) {
2201   // libc, pthread
2202   st->print("libc:");
2203   st->print("%s ", os::Linux::glibc_version());
2204   st->print("%s ", os::Linux::libpthread_version());
2205   if (os::Linux::is_LinuxThreads()) {
2206      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2207   }
2208   st->cr();
2209 }
2210 
2211 void os::Linux::print_full_memory_info(outputStream* st) {
2212    st->print("\n/proc/meminfo:\n");
2213    _print_ascii_file("/proc/meminfo", st);
2214    st->cr();
2215 }
2216 
2217 void os::print_memory_info(outputStream* st) {
2218 
2219   st->print("Memory:");
2220   st->print(" %dk page", os::vm_page_size()>>10);
2221 
2222   // values in struct sysinfo are "unsigned long"
2223   struct sysinfo si;
2224   sysinfo(&si);
2225 
2226   st->print(", physical " UINT64_FORMAT "k",
2227             os::physical_memory() >> 10);
2228   st->print("(" UINT64_FORMAT "k free)",
2229             os::available_memory() >> 10);
2230   st->print(", swap " UINT64_FORMAT "k",
2231             ((jlong)si.totalswap * si.mem_unit) >> 10);
2232   st->print("(" UINT64_FORMAT "k free)",
2233             ((jlong)si.freeswap * si.mem_unit) >> 10);
2234   st->cr();
2235 }
2236 
2237 void os::pd_print_cpu_info(outputStream* st) {
2238   st->print("\n/proc/cpuinfo:\n");
2239   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2240     st->print("  <Not Available>");
2241   }
2242   st->cr();
2243 }
2244 
2245 void os::print_siginfo(outputStream* st, void* siginfo) {
2246   const siginfo_t* si = (const siginfo_t*)siginfo;
2247 
2248   os::Posix::print_siginfo_brief(st, si);
2249 #if INCLUDE_CDS
2250   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2251       UseSharedSpaces) {
2252     FileMapInfo* mapinfo = FileMapInfo::current_info();
2253     if (mapinfo->is_in_shared_space(si->si_addr)) {
2254       st->print("\n\nError accessing class data sharing archive."   \
2255                 " Mapped file inaccessible during execution, "      \
2256                 " possible disk/network problem.");
2257     }
2258   }
2259 #endif
2260   st->cr();
2261 }
2262 
2263 
2264 static void print_signal_handler(outputStream* st, int sig,
2265                                  char* buf, size_t buflen);
2266 
2267 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2268   st->print_cr("Signal Handlers:");
2269   print_signal_handler(st, SIGSEGV, buf, buflen);
2270   print_signal_handler(st, SIGBUS , buf, buflen);
2271   print_signal_handler(st, SIGFPE , buf, buflen);
2272   print_signal_handler(st, SIGPIPE, buf, buflen);
2273   print_signal_handler(st, SIGXFSZ, buf, buflen);
2274   print_signal_handler(st, SIGILL , buf, buflen);
2275   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2276   print_signal_handler(st, SR_signum, buf, buflen);
2277   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2278   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2279   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2280   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2281 #if defined(PPC64)
2282   print_signal_handler(st, SIGTRAP, buf, buflen);
2283 #endif
2284 }
2285 
2286 static char saved_jvm_path[MAXPATHLEN] = {0};
2287 
2288 // Find the full path to the current module, libjvm.so
2289 void os::jvm_path(char *buf, jint buflen) {
2290   // Error checking.
2291   if (buflen < MAXPATHLEN) {
2292     assert(false, "must use a large-enough buffer");
2293     buf[0] = '\0';
2294     return;
2295   }
2296   // Lazy resolve the path to current module.
2297   if (saved_jvm_path[0] != 0) {
2298     strcpy(buf, saved_jvm_path);
2299     return;
2300   }
2301 
2302   char dli_fname[MAXPATHLEN];
2303   bool ret = dll_address_to_library_name(
2304                 CAST_FROM_FN_PTR(address, os::jvm_path),
2305                 dli_fname, sizeof(dli_fname), NULL);
2306   assert(ret, "cannot locate libjvm");
2307   char *rp = NULL;
2308   if (ret && dli_fname[0] != '\0') {
2309     rp = realpath(dli_fname, buf);
2310   }
2311   if (rp == NULL)
2312     return;
2313 
2314   if (Arguments::sun_java_launcher_is_altjvm()) {
2315     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2316     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2317     // If "/jre/lib/" appears at the right place in the string, then
2318     // assume we are installed in a JDK and we're done. Otherwise, check
2319     // for a JAVA_HOME environment variable and fix up the path so it
2320     // looks like libjvm.so is installed there (append a fake suffix
2321     // hotspot/libjvm.so).
2322     const char *p = buf + strlen(buf) - 1;
2323     for (int count = 0; p > buf && count < 5; ++count) {
2324       for (--p; p > buf && *p != '/'; --p)
2325         /* empty */ ;
2326     }
2327 
2328     if (strncmp(p, "/jre/lib/", 9) != 0) {
2329       // Look for JAVA_HOME in the environment.
2330       char* java_home_var = ::getenv("JAVA_HOME");
2331       if (java_home_var != NULL && java_home_var[0] != 0) {
2332         char* jrelib_p;
2333         int len;
2334 
2335         // Check the current module name "libjvm.so".
2336         p = strrchr(buf, '/');
2337         assert(strstr(p, "/libjvm") == p, "invalid library name");
2338 
2339         rp = realpath(java_home_var, buf);
2340         if (rp == NULL)
2341           return;
2342 
2343         // determine if this is a legacy image or modules image
2344         // modules image doesn't have "jre" subdirectory
2345         len = strlen(buf);
2346         assert(len < buflen, "Ran out of buffer room");
2347         jrelib_p = buf + len;
2348         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2349         if (0 != access(buf, F_OK)) {
2350           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2351         }
2352 
2353         if (0 == access(buf, F_OK)) {
2354           // Use current module name "libjvm.so"
2355           len = strlen(buf);
2356           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2357         } else {
2358           // Go back to path of .so
2359           rp = realpath(dli_fname, buf);
2360           if (rp == NULL)
2361             return;
2362         }
2363       }
2364     }
2365   }
2366 
2367   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2368 }
2369 
2370 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2371   // no prefix required, not even "_"
2372 }
2373 
2374 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2375   // no suffix required
2376 }
2377 
2378 ////////////////////////////////////////////////////////////////////////////////
2379 // sun.misc.Signal support
2380 
2381 static volatile jint sigint_count = 0;
2382 
2383 static void
2384 UserHandler(int sig, void *siginfo, void *context) {
2385   // 4511530 - sem_post is serialized and handled by the manager thread. When
2386   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2387   // don't want to flood the manager thread with sem_post requests.
2388   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2389       return;
2390 
2391   // Ctrl-C is pressed during error reporting, likely because the error
2392   // handler fails to abort. Let VM die immediately.
2393   if (sig == SIGINT && is_error_reported()) {
2394      os::die();
2395   }
2396 
2397   os::signal_notify(sig);
2398 }
2399 
2400 void* os::user_handler() {
2401   return CAST_FROM_FN_PTR(void*, UserHandler);
2402 }
2403 
2404 class Semaphore : public StackObj {
2405   public:
2406     Semaphore();
2407     ~Semaphore();
2408     void signal();
2409     void wait();
2410     bool trywait();
2411     bool timedwait(unsigned int sec, int nsec);
2412   private:
2413     sem_t _semaphore;
2414 };
2415 
2416 Semaphore::Semaphore() {
2417   sem_init(&_semaphore, 0, 0);
2418 }
2419 
2420 Semaphore::~Semaphore() {
2421   sem_destroy(&_semaphore);
2422 }
2423 
2424 void Semaphore::signal() {
2425   sem_post(&_semaphore);
2426 }
2427 
2428 void Semaphore::wait() {
2429   sem_wait(&_semaphore);
2430 }
2431 
2432 bool Semaphore::trywait() {
2433   return sem_trywait(&_semaphore) == 0;
2434 }
2435 
2436 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2437 
2438   struct timespec ts;
2439   // Semaphore's are always associated with CLOCK_REALTIME
2440   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2441   // see unpackTime for discussion on overflow checking
2442   if (sec >= MAX_SECS) {
2443     ts.tv_sec += MAX_SECS;
2444     ts.tv_nsec = 0;
2445   } else {
2446     ts.tv_sec += sec;
2447     ts.tv_nsec += nsec;
2448     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2449       ts.tv_nsec -= NANOSECS_PER_SEC;
2450       ++ts.tv_sec; // note: this must be <= max_secs
2451     }
2452   }
2453 
2454   while (1) {
2455     int result = sem_timedwait(&_semaphore, &ts);
2456     if (result == 0) {
2457       return true;
2458     } else if (errno == EINTR) {
2459       continue;
2460     } else if (errno == ETIMEDOUT) {
2461       return false;
2462     } else {
2463       return false;
2464     }
2465   }
2466 }
2467 
2468 extern "C" {
2469   typedef void (*sa_handler_t)(int);
2470   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2471 }
2472 
2473 void* os::signal(int signal_number, void* handler) {
2474   struct sigaction sigAct, oldSigAct;
2475 
2476   sigfillset(&(sigAct.sa_mask));
2477   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2478   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2479 
2480   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2481     // -1 means registration failed
2482     return (void *)-1;
2483   }
2484 
2485   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2486 }
2487 
2488 void os::signal_raise(int signal_number) {
2489   ::raise(signal_number);
2490 }
2491 
2492 /*
2493  * The following code is moved from os.cpp for making this
2494  * code platform specific, which it is by its very nature.
2495  */
2496 
2497 // Will be modified when max signal is changed to be dynamic
2498 int os::sigexitnum_pd() {
2499   return NSIG;
2500 }
2501 
2502 // a counter for each possible signal value
2503 static volatile jint pending_signals[NSIG+1] = { 0 };
2504 
2505 // Linux(POSIX) specific hand shaking semaphore.
2506 static sem_t sig_sem;
2507 static Semaphore sr_semaphore;
2508 
2509 void os::signal_init_pd() {
2510   // Initialize signal structures
2511   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2512 
2513   // Initialize signal semaphore
2514   ::sem_init(&sig_sem, 0, 0);
2515 }
2516 
2517 void os::signal_notify(int sig) {
2518   Atomic::inc(&pending_signals[sig]);
2519   ::sem_post(&sig_sem);
2520 }
2521 
2522 static int check_pending_signals(bool wait) {
2523   Atomic::store(0, &sigint_count);
2524   for (;;) {
2525     for (int i = 0; i < NSIG + 1; i++) {
2526       jint n = pending_signals[i];
2527       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2528         return i;
2529       }
2530     }
2531     if (!wait) {
2532       return -1;
2533     }
2534     JavaThread *thread = JavaThread::current();
2535     ThreadBlockInVM tbivm(thread);
2536 
2537     bool threadIsSuspended;
2538     do {
2539       thread->set_suspend_equivalent();
2540       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2541       ::sem_wait(&sig_sem);
2542 
2543       // were we externally suspended while we were waiting?
2544       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2545       if (threadIsSuspended) {
2546         //
2547         // The semaphore has been incremented, but while we were waiting
2548         // another thread suspended us. We don't want to continue running
2549         // while suspended because that would surprise the thread that
2550         // suspended us.
2551         //
2552         ::sem_post(&sig_sem);
2553 
2554         thread->java_suspend_self();
2555       }
2556     } while (threadIsSuspended);
2557   }
2558 }
2559 
2560 int os::signal_lookup() {
2561   return check_pending_signals(false);
2562 }
2563 
2564 int os::signal_wait() {
2565   return check_pending_signals(true);
2566 }
2567 
2568 ////////////////////////////////////////////////////////////////////////////////
2569 // Virtual Memory
2570 
2571 int os::vm_page_size() {
2572   // Seems redundant as all get out
2573   assert(os::Linux::page_size() != -1, "must call os::init");
2574   return os::Linux::page_size();
2575 }
2576 
2577 // Solaris allocates memory by pages.
2578 int os::vm_allocation_granularity() {
2579   assert(os::Linux::page_size() != -1, "must call os::init");
2580   return os::Linux::page_size();
2581 }
2582 
2583 // Rationale behind this function:
2584 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2585 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2586 //  samples for JITted code. Here we create private executable mapping over the code cache
2587 //  and then we can use standard (well, almost, as mapping can change) way to provide
2588 //  info for the reporting script by storing timestamp and location of symbol
2589 void linux_wrap_code(char* base, size_t size) {
2590   static volatile jint cnt = 0;
2591 
2592   if (!UseOprofile) {
2593     return;
2594   }
2595 
2596   char buf[PATH_MAX+1];
2597   int num = Atomic::add(1, &cnt);
2598 
2599   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2600            os::get_temp_directory(), os::current_process_id(), num);
2601   unlink(buf);
2602 
2603   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2604 
2605   if (fd != -1) {
2606     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2607     if (rv != (off_t)-1) {
2608       if (::write(fd, "", 1) == 1) {
2609         mmap(base, size,
2610              PROT_READ|PROT_WRITE|PROT_EXEC,
2611              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2612       }
2613     }
2614     ::close(fd);
2615     unlink(buf);
2616   }
2617 }
2618 
2619 static bool recoverable_mmap_error(int err) {
2620   // See if the error is one we can let the caller handle. This
2621   // list of errno values comes from JBS-6843484. I can't find a
2622   // Linux man page that documents this specific set of errno
2623   // values so while this list currently matches Solaris, it may
2624   // change as we gain experience with this failure mode.
2625   switch (err) {
2626   case EBADF:
2627   case EINVAL:
2628   case ENOTSUP:
2629     // let the caller deal with these errors
2630     return true;
2631 
2632   default:
2633     // Any remaining errors on this OS can cause our reserved mapping
2634     // to be lost. That can cause confusion where different data
2635     // structures think they have the same memory mapped. The worst
2636     // scenario is if both the VM and a library think they have the
2637     // same memory mapped.
2638     return false;
2639   }
2640 }
2641 
2642 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2643                                     int err) {
2644   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2645           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2646           strerror(err), err);
2647 }
2648 
2649 static void warn_fail_commit_memory(char* addr, size_t size,
2650                                     size_t alignment_hint, bool exec,
2651                                     int err) {
2652   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2653           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2654           alignment_hint, exec, strerror(err), err);
2655 }
2656 
2657 // NOTE: Linux kernel does not really reserve the pages for us.
2658 //       All it does is to check if there are enough free pages
2659 //       left at the time of mmap(). This could be a potential
2660 //       problem.
2661 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2662   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2663   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2664                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2665   if (res != (uintptr_t) MAP_FAILED) {
2666     if (UseNUMAInterleaving) {
2667       numa_make_global(addr, size);
2668     }
2669     return 0;
2670   }
2671 
2672   int err = errno;  // save errno from mmap() call above
2673 
2674   if (!recoverable_mmap_error(err)) {
2675     warn_fail_commit_memory(addr, size, exec, err);
2676     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2677   }
2678 
2679   return err;
2680 }
2681 
2682 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2683   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2684 }
2685 
2686 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2687                                   const char* mesg) {
2688   assert(mesg != NULL, "mesg must be specified");
2689   int err = os::Linux::commit_memory_impl(addr, size, exec);
2690   if (err != 0) {
2691     // the caller wants all commit errors to exit with the specified mesg:
2692     warn_fail_commit_memory(addr, size, exec, err);
2693     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2694   }
2695 }
2696 
2697 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2698 #ifndef MAP_HUGETLB
2699 #define MAP_HUGETLB 0x40000
2700 #endif
2701 
2702 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2703 #ifndef MADV_HUGEPAGE
2704 #define MADV_HUGEPAGE 14
2705 #endif
2706 
2707 int os::Linux::commit_memory_impl(char* addr, size_t size,
2708                                   size_t alignment_hint, bool exec) {
2709   int err = os::Linux::commit_memory_impl(addr, size, exec);
2710   if (err == 0) {
2711     realign_memory(addr, size, alignment_hint);
2712   }
2713   return err;
2714 }
2715 
2716 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2717                           bool exec) {
2718   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2719 }
2720 
2721 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2722                                   size_t alignment_hint, bool exec,
2723                                   const char* mesg) {
2724   assert(mesg != NULL, "mesg must be specified");
2725   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2726   if (err != 0) {
2727     // the caller wants all commit errors to exit with the specified mesg:
2728     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2729     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2730   }
2731 }
2732 
2733 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2734   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2735     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2736     // be supported or the memory may already be backed by huge pages.
2737     ::madvise(addr, bytes, MADV_HUGEPAGE);
2738   }
2739 }
2740 
2741 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2742   // This method works by doing an mmap over an existing mmaping and effectively discarding
2743   // the existing pages. However it won't work for SHM-based large pages that cannot be
2744   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2745   // small pages on top of the SHM segment. This method always works for small pages, so we
2746   // allow that in any case.
2747   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2748     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2749   }
2750 }
2751 
2752 void os::numa_make_global(char *addr, size_t bytes) {
2753   Linux::numa_interleave_memory(addr, bytes);
2754 }
2755 
2756 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2757 // bind policy to MPOL_PREFERRED for the current thread.
2758 #define USE_MPOL_PREFERRED 0
2759 
2760 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2761   // To make NUMA and large pages more robust when both enabled, we need to ease
2762   // the requirements on where the memory should be allocated. MPOL_BIND is the
2763   // default policy and it will force memory to be allocated on the specified
2764   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2765   // the specified node, but will not force it. Using this policy will prevent
2766   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2767   // free large pages.
2768   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2769   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2770 }
2771 
2772 bool os::numa_topology_changed()   { return false; }
2773 
2774 size_t os::numa_get_groups_num() {
2775   int max_node = Linux::numa_max_node();
2776   return max_node > 0 ? max_node + 1 : 1;
2777 }
2778 
2779 int os::numa_get_group_id() {
2780   int cpu_id = Linux::sched_getcpu();
2781   if (cpu_id != -1) {
2782     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2783     if (lgrp_id != -1) {
2784       return lgrp_id;
2785     }
2786   }
2787   return 0;
2788 }
2789 
2790 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2791   for (size_t i = 0; i < size; i++) {
2792     ids[i] = i;
2793   }
2794   return size;
2795 }
2796 
2797 bool os::get_page_info(char *start, page_info* info) {
2798   return false;
2799 }
2800 
2801 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2802   return end;
2803 }
2804 
2805 
2806 int os::Linux::sched_getcpu_syscall(void) {
2807   unsigned int cpu;
2808   int retval = -1;
2809 
2810 #if defined(IA32)
2811 # ifndef SYS_getcpu
2812 # define SYS_getcpu 318
2813 # endif
2814   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2815 #elif defined(AMD64)
2816 // Unfortunately we have to bring all these macros here from vsyscall.h
2817 // to be able to compile on old linuxes.
2818 # define __NR_vgetcpu 2
2819 # define VSYSCALL_START (-10UL << 20)
2820 # define VSYSCALL_SIZE 1024
2821 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2822   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2823   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2824   retval = vgetcpu(&cpu, NULL, NULL);
2825 #endif
2826 
2827   return (retval == -1) ? retval : cpu;
2828 }
2829 
2830 // Something to do with the numa-aware allocator needs these symbols
2831 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2832 extern "C" JNIEXPORT void numa_error(char *where) { }
2833 extern "C" JNIEXPORT int fork1() { return fork(); }
2834 
2835 
2836 // If we are running with libnuma version > 2, then we should
2837 // be trying to use symbols with versions 1.1
2838 // If we are running with earlier version, which did not have symbol versions,
2839 // we should use the base version.
2840 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2841   void *f = dlvsym(handle, name, "libnuma_1.1");
2842   if (f == NULL) {
2843     f = dlsym(handle, name);
2844   }
2845   return f;
2846 }
2847 
2848 bool os::Linux::libnuma_init() {
2849   // sched_getcpu() should be in libc.
2850   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2851                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2852 
2853   // If it's not, try a direct syscall.
2854   if (sched_getcpu() == -1)
2855     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2856 
2857   if (sched_getcpu() != -1) { // Does it work?
2858     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2859     if (handle != NULL) {
2860       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2861                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2862       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2863                                        libnuma_dlsym(handle, "numa_max_node")));
2864       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2865                                         libnuma_dlsym(handle, "numa_available")));
2866       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2867                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2868       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2869                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2870       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2871                                             libnuma_dlsym(handle, "numa_set_bind_policy")));
2872 
2873 
2874       if (numa_available() != -1) {
2875         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2876         // Create a cpu -> node mapping
2877         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2878         rebuild_cpu_to_node_map();
2879         return true;
2880       }
2881     }
2882   }
2883   return false;
2884 }
2885 
2886 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2887 // The table is later used in get_node_by_cpu().
2888 void os::Linux::rebuild_cpu_to_node_map() {
2889   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2890                               // in libnuma (possible values are starting from 16,
2891                               // and continuing up with every other power of 2, but less
2892                               // than the maximum number of CPUs supported by kernel), and
2893                               // is a subject to change (in libnuma version 2 the requirements
2894                               // are more reasonable) we'll just hardcode the number they use
2895                               // in the library.
2896   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2897 
2898   size_t cpu_num = os::active_processor_count();
2899   size_t cpu_map_size = NCPUS / BitsPerCLong;
2900   size_t cpu_map_valid_size =
2901     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2902 
2903   cpu_to_node()->clear();
2904   cpu_to_node()->at_grow(cpu_num - 1);
2905   size_t node_num = numa_get_groups_num();
2906 
2907   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2908   for (size_t i = 0; i < node_num; i++) {
2909     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2910       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2911         if (cpu_map[j] != 0) {
2912           for (size_t k = 0; k < BitsPerCLong; k++) {
2913             if (cpu_map[j] & (1UL << k)) {
2914               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2915             }
2916           }
2917         }
2918       }
2919     }
2920   }
2921   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2922 }
2923 
2924 int os::Linux::get_node_by_cpu(int cpu_id) {
2925   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2926     return cpu_to_node()->at(cpu_id);
2927   }
2928   return -1;
2929 }
2930 
2931 GrowableArray<int>* os::Linux::_cpu_to_node;
2932 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2933 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2934 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2935 os::Linux::numa_available_func_t os::Linux::_numa_available;
2936 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2937 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2938 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2939 unsigned long* os::Linux::_numa_all_nodes;
2940 
2941 bool os::pd_uncommit_memory(char* addr, size_t size) {
2942   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2943                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2944   return res  != (uintptr_t) MAP_FAILED;
2945 }
2946 
2947 static
2948 address get_stack_commited_bottom(address bottom, size_t size) {
2949   address nbot = bottom;
2950   address ntop = bottom + size;
2951 
2952   size_t page_sz = os::vm_page_size();
2953   unsigned pages = size / page_sz;
2954 
2955   unsigned char vec[1];
2956   unsigned imin = 1, imax = pages + 1, imid;
2957   int mincore_return_value = 0;
2958 
2959   assert(imin <= imax, "Unexpected page size");
2960 
2961   while (imin < imax) {
2962     imid = (imax + imin) / 2;
2963     nbot = ntop - (imid * page_sz);
2964 
2965     // Use a trick with mincore to check whether the page is mapped or not.
2966     // mincore sets vec to 1 if page resides in memory and to 0 if page
2967     // is swapped output but if page we are asking for is unmapped
2968     // it returns -1,ENOMEM
2969     mincore_return_value = mincore(nbot, page_sz, vec);
2970 
2971     if (mincore_return_value == -1) {
2972       // Page is not mapped go up
2973       // to find first mapped page
2974       if (errno != EAGAIN) {
2975         assert(errno == ENOMEM, "Unexpected mincore errno");
2976         imax = imid;
2977       }
2978     } else {
2979       // Page is mapped go down
2980       // to find first not mapped page
2981       imin = imid + 1;
2982     }
2983   }
2984 
2985   nbot = nbot + page_sz;
2986 
2987   // Adjust stack bottom one page up if last checked page is not mapped
2988   if (mincore_return_value == -1) {
2989     nbot = nbot + page_sz;
2990   }
2991 
2992   return nbot;
2993 }
2994 
2995 
2996 // Linux uses a growable mapping for the stack, and if the mapping for
2997 // the stack guard pages is not removed when we detach a thread the
2998 // stack cannot grow beyond the pages where the stack guard was
2999 // mapped.  If at some point later in the process the stack expands to
3000 // that point, the Linux kernel cannot expand the stack any further
3001 // because the guard pages are in the way, and a segfault occurs.
3002 //
3003 // However, it's essential not to split the stack region by unmapping
3004 // a region (leaving a hole) that's already part of the stack mapping,
3005 // so if the stack mapping has already grown beyond the guard pages at
3006 // the time we create them, we have to truncate the stack mapping.
3007 // So, we need to know the extent of the stack mapping when
3008 // create_stack_guard_pages() is called.
3009 
3010 // We only need this for stacks that are growable: at the time of
3011 // writing thread stacks don't use growable mappings (i.e. those
3012 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3013 // only applies to the main thread.
3014 
3015 // If the (growable) stack mapping already extends beyond the point
3016 // where we're going to put our guard pages, truncate the mapping at
3017 // that point by munmap()ping it.  This ensures that when we later
3018 // munmap() the guard pages we don't leave a hole in the stack
3019 // mapping. This only affects the main/initial thread
3020 
3021 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3022 
3023   if (os::Linux::is_initial_thread()) {
3024     // As we manually grow stack up to bottom inside create_attached_thread(),
3025     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3026     // we don't need to do anything special.
3027     // Check it first, before calling heavy function.
3028     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3029     unsigned char vec[1];
3030 
3031     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3032       // Fallback to slow path on all errors, including EAGAIN
3033       stack_extent = (uintptr_t) get_stack_commited_bottom(
3034                                     os::Linux::initial_thread_stack_bottom(),
3035                                     (size_t)addr - stack_extent);
3036     }
3037 
3038     if (stack_extent < (uintptr_t)addr) {
3039       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3040     }
3041   }
3042 
3043   return os::commit_memory(addr, size, !ExecMem);
3044 }
3045 
3046 // If this is a growable mapping, remove the guard pages entirely by
3047 // munmap()ping them.  If not, just call uncommit_memory(). This only
3048 // affects the main/initial thread, but guard against future OS changes
3049 // It's safe to always unmap guard pages for initial thread because we
3050 // always place it right after end of the mapped region
3051 
3052 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3053   uintptr_t stack_extent, stack_base;
3054 
3055   if (os::Linux::is_initial_thread()) {
3056     return ::munmap(addr, size) == 0;
3057   }
3058 
3059   return os::uncommit_memory(addr, size);
3060 }
3061 
3062 static address _highest_vm_reserved_address = NULL;
3063 
3064 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3065 // at 'requested_addr'. If there are existing memory mappings at the same
3066 // location, however, they will be overwritten. If 'fixed' is false,
3067 // 'requested_addr' is only treated as a hint, the return value may or
3068 // may not start from the requested address. Unlike Linux mmap(), this
3069 // function returns NULL to indicate failure.
3070 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3071   char * addr;
3072   int flags;
3073 
3074   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3075   if (fixed) {
3076     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3077     flags |= MAP_FIXED;
3078   }
3079 
3080   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3081   // touch an uncommitted page. Otherwise, the read/write might
3082   // succeed if we have enough swap space to back the physical page.
3083   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3084                        flags, -1, 0);
3085 
3086   if (addr != MAP_FAILED) {
3087     // anon_mmap() should only get called during VM initialization,
3088     // don't need lock (actually we can skip locking even it can be called
3089     // from multiple threads, because _highest_vm_reserved_address is just a
3090     // hint about the upper limit of non-stack memory regions.)
3091     if ((address)addr + bytes > _highest_vm_reserved_address) {
3092       _highest_vm_reserved_address = (address)addr + bytes;
3093     }
3094   }
3095 
3096   return addr == MAP_FAILED ? NULL : addr;
3097 }
3098 
3099 // Don't update _highest_vm_reserved_address, because there might be memory
3100 // regions above addr + size. If so, releasing a memory region only creates
3101 // a hole in the address space, it doesn't help prevent heap-stack collision.
3102 //
3103 static int anon_munmap(char * addr, size_t size) {
3104   return ::munmap(addr, size) == 0;
3105 }
3106 
3107 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3108                          size_t alignment_hint) {
3109   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3110 }
3111 
3112 bool os::pd_release_memory(char* addr, size_t size) {
3113   return anon_munmap(addr, size);
3114 }
3115 
3116 static address highest_vm_reserved_address() {
3117   return _highest_vm_reserved_address;
3118 }
3119 
3120 static bool linux_mprotect(char* addr, size_t size, int prot) {
3121   // Linux wants the mprotect address argument to be page aligned.
3122   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3123 
3124   // According to SUSv3, mprotect() should only be used with mappings
3125   // established by mmap(), and mmap() always maps whole pages. Unaligned
3126   // 'addr' likely indicates problem in the VM (e.g. trying to change
3127   // protection of malloc'ed or statically allocated memory). Check the
3128   // caller if you hit this assert.
3129   assert(addr == bottom, "sanity check");
3130 
3131   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3132   return ::mprotect(bottom, size, prot) == 0;
3133 }
3134 
3135 // Set protections specified
3136 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3137                         bool is_committed) {
3138   unsigned int p = 0;
3139   switch (prot) {
3140   case MEM_PROT_NONE: p = PROT_NONE; break;
3141   case MEM_PROT_READ: p = PROT_READ; break;
3142   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3143   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3144   default:
3145     ShouldNotReachHere();
3146   }
3147   // is_committed is unused.
3148   return linux_mprotect(addr, bytes, p);
3149 }
3150 
3151 bool os::guard_memory(char* addr, size_t size) {
3152   return linux_mprotect(addr, size, PROT_NONE);
3153 }
3154 
3155 bool os::unguard_memory(char* addr, size_t size) {
3156   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3157 }
3158 
3159 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3160   bool result = false;
3161   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3162                  MAP_ANONYMOUS|MAP_PRIVATE,
3163                  -1, 0);
3164   if (p != MAP_FAILED) {
3165     void *aligned_p = align_ptr_up(p, page_size);
3166 
3167     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3168 
3169     munmap(p, page_size * 2);
3170   }
3171 
3172   if (warn && !result) {
3173     warning("TransparentHugePages is not supported by the operating system.");
3174   }
3175 
3176   return result;
3177 }
3178 
3179 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3180   bool result = false;
3181   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3182                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3183                  -1, 0);
3184 
3185   if (p != MAP_FAILED) {
3186     // We don't know if this really is a huge page or not.
3187     FILE *fp = fopen("/proc/self/maps", "r");
3188     if (fp) {
3189       while (!feof(fp)) {
3190         char chars[257];
3191         long x = 0;
3192         if (fgets(chars, sizeof(chars), fp)) {
3193           if (sscanf(chars, "%lx-%*x", &x) == 1
3194               && x == (long)p) {
3195             if (strstr (chars, "hugepage")) {
3196               result = true;
3197               break;
3198             }
3199           }
3200         }
3201       }
3202       fclose(fp);
3203     }
3204     munmap(p, page_size);
3205   }
3206 
3207   if (warn && !result) {
3208     warning("HugeTLBFS is not supported by the operating system.");
3209   }
3210 
3211   return result;
3212 }
3213 
3214 /*
3215 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3216 *
3217 * From the coredump_filter documentation:
3218 *
3219 * - (bit 0) anonymous private memory
3220 * - (bit 1) anonymous shared memory
3221 * - (bit 2) file-backed private memory
3222 * - (bit 3) file-backed shared memory
3223 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3224 *           effective only if the bit 2 is cleared)
3225 * - (bit 5) hugetlb private memory
3226 * - (bit 6) hugetlb shared memory
3227 */
3228 static void set_coredump_filter(void) {
3229   FILE *f;
3230   long cdm;
3231 
3232   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3233     return;
3234   }
3235 
3236   if (fscanf(f, "%lx", &cdm) != 1) {
3237     fclose(f);
3238     return;
3239   }
3240 
3241   rewind(f);
3242 
3243   if ((cdm & LARGEPAGES_BIT) == 0) {
3244     cdm |= LARGEPAGES_BIT;
3245     fprintf(f, "%#lx", cdm);
3246   }
3247 
3248   fclose(f);
3249 }
3250 
3251 // Large page support
3252 
3253 static size_t _large_page_size = 0;
3254 
3255 size_t os::Linux::find_large_page_size() {
3256   size_t large_page_size = 0;
3257 
3258   // large_page_size on Linux is used to round up heap size. x86 uses either
3259   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3260   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3261   // page as large as 256M.
3262   //
3263   // Here we try to figure out page size by parsing /proc/meminfo and looking
3264   // for a line with the following format:
3265   //    Hugepagesize:     2048 kB
3266   //
3267   // If we can't determine the value (e.g. /proc is not mounted, or the text
3268   // format has been changed), we'll use the largest page size supported by
3269   // the processor.
3270 
3271 #ifndef ZERO
3272   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3273                      ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3274 #endif // ZERO
3275 
3276   FILE *fp = fopen("/proc/meminfo", "r");
3277   if (fp) {
3278     while (!feof(fp)) {
3279       int x = 0;
3280       char buf[16];
3281       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3282         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3283           large_page_size = x * K;
3284           break;
3285         }
3286       } else {
3287         // skip to next line
3288         for (;;) {
3289           int ch = fgetc(fp);
3290           if (ch == EOF || ch == (int)'\n') break;
3291         }
3292       }
3293     }
3294     fclose(fp);
3295   }
3296 
3297   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3298     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3299         SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3300         proper_unit_for_byte_size(large_page_size));
3301   }
3302 
3303   return large_page_size;
3304 }
3305 
3306 size_t os::Linux::setup_large_page_size() {
3307   _large_page_size = Linux::find_large_page_size();
3308   const size_t default_page_size = (size_t)Linux::page_size();
3309   if (_large_page_size > default_page_size) {
3310     _page_sizes[0] = _large_page_size;
3311     _page_sizes[1] = default_page_size;
3312     _page_sizes[2] = 0;
3313   }
3314 
3315   return _large_page_size;
3316 }
3317 
3318 bool os::Linux::setup_large_page_type(size_t page_size) {
3319   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3320       FLAG_IS_DEFAULT(UseSHM) &&
3321       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3322 
3323     // The type of large pages has not been specified by the user.
3324 
3325     // Try UseHugeTLBFS and then UseSHM.
3326     UseHugeTLBFS = UseSHM = true;
3327 
3328     // Don't try UseTransparentHugePages since there are known
3329     // performance issues with it turned on. This might change in the future.
3330     UseTransparentHugePages = false;
3331   }
3332 
3333   if (UseTransparentHugePages) {
3334     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3335     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3336       UseHugeTLBFS = false;
3337       UseSHM = false;
3338       return true;
3339     }
3340     UseTransparentHugePages = false;
3341   }
3342 
3343   if (UseHugeTLBFS) {
3344     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3345     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3346       UseSHM = false;
3347       return true;
3348     }
3349     UseHugeTLBFS = false;
3350   }
3351 
3352   return UseSHM;
3353 }
3354 
3355 void os::large_page_init() {
3356   if (!UseLargePages &&
3357       !UseTransparentHugePages &&
3358       !UseHugeTLBFS &&
3359       !UseSHM) {
3360     // Not using large pages.
3361     return;
3362   }
3363 
3364   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3365     // The user explicitly turned off large pages.
3366     // Ignore the rest of the large pages flags.
3367     UseTransparentHugePages = false;
3368     UseHugeTLBFS = false;
3369     UseSHM = false;
3370     return;
3371   }
3372 
3373   size_t large_page_size = Linux::setup_large_page_size();
3374   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3375 
3376   set_coredump_filter();
3377 }
3378 
3379 #ifndef SHM_HUGETLB
3380 #define SHM_HUGETLB 04000
3381 #endif
3382 
3383 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3384   // "exec" is passed in but not used.  Creating the shared image for
3385   // the code cache doesn't have an SHM_X executable permission to check.
3386   assert(UseLargePages && UseSHM, "only for SHM large pages");
3387   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3388 
3389   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3390     return NULL; // Fallback to small pages.
3391   }
3392 
3393   key_t key = IPC_PRIVATE;
3394   char *addr;
3395 
3396   bool warn_on_failure = UseLargePages &&
3397                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3398                          !FLAG_IS_DEFAULT(UseSHM) ||
3399                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3400                         );
3401   char msg[128];
3402 
3403   // Create a large shared memory region to attach to based on size.
3404   // Currently, size is the total size of the heap
3405   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3406   if (shmid == -1) {
3407      // Possible reasons for shmget failure:
3408      // 1. shmmax is too small for Java heap.
3409      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3410      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3411      // 2. not enough large page memory.
3412      //    > check available large pages: cat /proc/meminfo
3413      //    > increase amount of large pages:
3414      //          echo new_value > /proc/sys/vm/nr_hugepages
3415      //      Note 1: different Linux may use different name for this property,
3416      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3417      //      Note 2: it's possible there's enough physical memory available but
3418      //            they are so fragmented after a long run that they can't
3419      //            coalesce into large pages. Try to reserve large pages when
3420      //            the system is still "fresh".
3421      if (warn_on_failure) {
3422        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3423        warning("%s", msg);
3424      }
3425      return NULL;
3426   }
3427 
3428   // attach to the region
3429   addr = (char*)shmat(shmid, req_addr, 0);
3430   int err = errno;
3431 
3432   // Remove shmid. If shmat() is successful, the actual shared memory segment
3433   // will be deleted when it's detached by shmdt() or when the process
3434   // terminates. If shmat() is not successful this will remove the shared
3435   // segment immediately.
3436   shmctl(shmid, IPC_RMID, NULL);
3437 
3438   if ((intptr_t)addr == -1) {
3439      if (warn_on_failure) {
3440        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3441        warning("%s", msg);
3442      }
3443      return NULL;
3444   }
3445 
3446   return addr;
3447 }
3448 
3449 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3450   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3451 
3452   bool warn_on_failure = UseLargePages &&
3453       (!FLAG_IS_DEFAULT(UseLargePages) ||
3454        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3455        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3456 
3457   if (warn_on_failure) {
3458     char msg[128];
3459     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3460         PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3461     warning("%s", msg);
3462   }
3463 }
3464 
3465 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3466   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3467   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3468   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3469 
3470   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3471   char* addr = (char*)::mmap(req_addr, bytes, prot,
3472                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3473                              -1, 0);
3474 
3475   if (addr == MAP_FAILED) {
3476     warn_on_large_pages_failure(req_addr, bytes, errno);
3477     return NULL;
3478   }
3479 
3480   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3481 
3482   return addr;
3483 }
3484 
3485 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3486   size_t large_page_size = os::large_page_size();
3487 
3488   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3489 
3490   // Allocate small pages.
3491 
3492   char* start;
3493   if (req_addr != NULL) {
3494     assert(is_ptr_aligned(req_addr, alignment), "Must be");
3495     assert(is_size_aligned(bytes, alignment), "Must be");
3496     start = os::reserve_memory(bytes, req_addr);
3497     assert(start == NULL || start == req_addr, "Must be");
3498   } else {
3499     start = os::reserve_memory_aligned(bytes, alignment);
3500   }
3501 
3502   if (start == NULL) {
3503     return NULL;
3504   }
3505 
3506   assert(is_ptr_aligned(start, alignment), "Must be");
3507 
3508   if (MemTracker::tracking_level() > NMT_minimal) {
3509     // os::reserve_memory_special will record this memory area.
3510     // Need to release it here to prevent overlapping reservations.
3511     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3512     tkr.record((address)start, bytes);
3513   }
3514 
3515   char* end = start + bytes;
3516 
3517   // Find the regions of the allocated chunk that can be promoted to large pages.
3518   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3519   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3520 
3521   size_t lp_bytes = lp_end - lp_start;
3522 
3523   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3524 
3525   if (lp_bytes == 0) {
3526     // The mapped region doesn't even span the start and the end of a large page.
3527     // Fall back to allocate a non-special area.
3528     ::munmap(start, end - start);
3529     return NULL;
3530   }
3531 
3532   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3533 
3534 
3535   void* result;
3536 
3537   if (start != lp_start) {
3538     result = ::mmap(start, lp_start - start, prot,
3539                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3540                     -1, 0);
3541     if (result == MAP_FAILED) {
3542       ::munmap(lp_start, end - lp_start);
3543       return NULL;
3544     }
3545   }
3546 
3547   result = ::mmap(lp_start, lp_bytes, prot,
3548                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3549                   -1, 0);
3550   if (result == MAP_FAILED) {
3551     warn_on_large_pages_failure(req_addr, bytes, errno);
3552     // If the mmap above fails, the large pages region will be unmapped and we
3553     // have regions before and after with small pages. Release these regions.
3554     //
3555     // |  mapped  |  unmapped  |  mapped  |
3556     // ^          ^            ^          ^
3557     // start      lp_start     lp_end     end
3558     //
3559     ::munmap(start, lp_start - start);
3560     ::munmap(lp_end, end - lp_end);
3561     return NULL;
3562   }
3563 
3564   if (lp_end != end) {
3565       result = ::mmap(lp_end, end - lp_end, prot,
3566                       MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3567                       -1, 0);
3568     if (result == MAP_FAILED) {
3569       ::munmap(start, lp_end - start);
3570       return NULL;
3571     }
3572   }
3573 
3574   return start;
3575 }
3576 
3577 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3578   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3579   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3580   assert(is_power_of_2(alignment), "Must be");
3581   assert(is_power_of_2(os::large_page_size()), "Must be");
3582   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3583 
3584   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3585     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3586   } else {
3587     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3588   }
3589 }
3590 
3591 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3592   assert(UseLargePages, "only for large pages");
3593 
3594   char* addr;
3595   if (UseSHM) {
3596     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3597   } else {
3598     assert(UseHugeTLBFS, "must be");
3599     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3600   }
3601 
3602   if (addr != NULL) {
3603     if (UseNUMAInterleaving) {
3604       numa_make_global(addr, bytes);
3605     }
3606 
3607     // The memory is committed
3608     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3609   }
3610 
3611   return addr;
3612 }
3613 
3614 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3615   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3616   return shmdt(base) == 0;
3617 }
3618 
3619 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3620   return pd_release_memory(base, bytes);
3621 }
3622 
3623 bool os::release_memory_special(char* base, size_t bytes) {
3624   bool res;
3625   if (MemTracker::tracking_level() > NMT_minimal) {
3626     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3627     res = os::Linux::release_memory_special_impl(base, bytes);
3628     if (res) {
3629       tkr.record((address)base, bytes);
3630     }
3631 
3632   } else {
3633     res = os::Linux::release_memory_special_impl(base, bytes);
3634   }
3635   return res;
3636 }
3637 
3638 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3639   assert(UseLargePages, "only for large pages");
3640   bool res;
3641 
3642   if (UseSHM) {
3643     res = os::Linux::release_memory_special_shm(base, bytes);
3644   } else {
3645     assert(UseHugeTLBFS, "must be");
3646     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3647   }
3648   return res;
3649 }
3650 
3651 size_t os::large_page_size() {
3652   return _large_page_size;
3653 }
3654 
3655 // With SysV SHM the entire memory region must be allocated as shared
3656 // memory.
3657 // HugeTLBFS allows application to commit large page memory on demand.
3658 // However, when committing memory with HugeTLBFS fails, the region
3659 // that was supposed to be committed will lose the old reservation
3660 // and allow other threads to steal that memory region. Because of this
3661 // behavior we can't commit HugeTLBFS memory.
3662 bool os::can_commit_large_page_memory() {
3663   return UseTransparentHugePages;
3664 }
3665 
3666 bool os::can_execute_large_page_memory() {
3667   return UseTransparentHugePages || UseHugeTLBFS;
3668 }
3669 
3670 // Reserve memory at an arbitrary address, only if that area is
3671 // available (and not reserved for something else).
3672 
3673 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3674   const int max_tries = 10;
3675   char* base[max_tries];
3676   size_t size[max_tries];
3677   const size_t gap = 0x000000;
3678 
3679   // Assert only that the size is a multiple of the page size, since
3680   // that's all that mmap requires, and since that's all we really know
3681   // about at this low abstraction level.  If we need higher alignment,
3682   // we can either pass an alignment to this method or verify alignment
3683   // in one of the methods further up the call chain.  See bug 5044738.
3684   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3685 
3686   // Repeatedly allocate blocks until the block is allocated at the
3687   // right spot. Give up after max_tries. Note that reserve_memory() will
3688   // automatically update _highest_vm_reserved_address if the call is
3689   // successful. The variable tracks the highest memory address every reserved
3690   // by JVM. It is used to detect heap-stack collision if running with
3691   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3692   // space than needed, it could confuse the collision detecting code. To
3693   // solve the problem, save current _highest_vm_reserved_address and
3694   // calculate the correct value before return.
3695   address old_highest = _highest_vm_reserved_address;
3696 
3697   // Linux mmap allows caller to pass an address as hint; give it a try first,
3698   // if kernel honors the hint then we can return immediately.
3699   char * addr = anon_mmap(requested_addr, bytes, false);
3700   if (addr == requested_addr) {
3701      return requested_addr;
3702   }
3703 
3704   if (addr != NULL) {
3705      // mmap() is successful but it fails to reserve at the requested address
3706      anon_munmap(addr, bytes);
3707   }
3708 
3709   int i;
3710   for (i = 0; i < max_tries; ++i) {
3711     base[i] = reserve_memory(bytes);
3712 
3713     if (base[i] != NULL) {
3714       // Is this the block we wanted?
3715       if (base[i] == requested_addr) {
3716         size[i] = bytes;
3717         break;
3718       }
3719 
3720       // Does this overlap the block we wanted? Give back the overlapped
3721       // parts and try again.
3722 
3723       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3724       if (top_overlap >= 0 && top_overlap < bytes) {
3725         unmap_memory(base[i], top_overlap);
3726         base[i] += top_overlap;
3727         size[i] = bytes - top_overlap;
3728       } else {
3729         size_t bottom_overlap = base[i] + bytes - requested_addr;
3730         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3731           unmap_memory(requested_addr, bottom_overlap);
3732           size[i] = bytes - bottom_overlap;
3733         } else {
3734           size[i] = bytes;
3735         }
3736       }
3737     }
3738   }
3739 
3740   // Give back the unused reserved pieces.
3741 
3742   for (int j = 0; j < i; ++j) {
3743     if (base[j] != NULL) {
3744       unmap_memory(base[j], size[j]);
3745     }
3746   }
3747 
3748   if (i < max_tries) {
3749     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3750     return requested_addr;
3751   } else {
3752     _highest_vm_reserved_address = old_highest;
3753     return NULL;
3754   }
3755 }
3756 
3757 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3758   return ::read(fd, buf, nBytes);
3759 }
3760 
3761 //
3762 // Short sleep, direct OS call.
3763 //
3764 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3765 // sched_yield(2) will actually give up the CPU:
3766 //
3767 //   * Alone on this pariticular CPU, keeps running.
3768 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3769 //     (pre 2.6.39).
3770 //
3771 // So calling this with 0 is an alternative.
3772 //
3773 void os::naked_short_sleep(jlong ms) {
3774   struct timespec req;
3775 
3776   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3777   req.tv_sec = 0;
3778   if (ms > 0) {
3779     req.tv_nsec = (ms % 1000) * 1000000;
3780   }
3781   else {
3782     req.tv_nsec = 1;
3783   }
3784 
3785   nanosleep(&req, NULL);
3786 
3787   return;
3788 }
3789 
3790 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3791 void os::infinite_sleep() {
3792   while (true) {    // sleep forever ...
3793     ::sleep(100);   // ... 100 seconds at a time
3794   }
3795 }
3796 
3797 // Used to convert frequent JVM_Yield() to nops
3798 bool os::dont_yield() {
3799   return DontYieldALot;
3800 }
3801 
3802 void os::naked_yield() {
3803   sched_yield();
3804 }
3805 
3806 ////////////////////////////////////////////////////////////////////////////////
3807 // thread priority support
3808 
3809 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3810 // only supports dynamic priority, static priority must be zero. For real-time
3811 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3812 // However, for large multi-threaded applications, SCHED_RR is not only slower
3813 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3814 // of 5 runs - Sep 2005).
3815 //
3816 // The following code actually changes the niceness of kernel-thread/LWP. It
3817 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3818 // not the entire user process, and user level threads are 1:1 mapped to kernel
3819 // threads. It has always been the case, but could change in the future. For
3820 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3821 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3822 
3823 int os::java_to_os_priority[CriticalPriority + 1] = {
3824   19,              // 0 Entry should never be used
3825 
3826    4,              // 1 MinPriority
3827    3,              // 2
3828    2,              // 3
3829 
3830    1,              // 4
3831    0,              // 5 NormPriority
3832   -1,              // 6
3833 
3834   -2,              // 7
3835   -3,              // 8
3836   -4,              // 9 NearMaxPriority
3837 
3838   -5,              // 10 MaxPriority
3839 
3840   -5               // 11 CriticalPriority
3841 };
3842 
3843 static int prio_init() {
3844   if (ThreadPriorityPolicy == 1) {
3845     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3846     // if effective uid is not root. Perhaps, a more elegant way of doing
3847     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3848     if (geteuid() != 0) {
3849       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3850         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3851       }
3852       ThreadPriorityPolicy = 0;
3853     }
3854   }
3855   if (UseCriticalJavaThreadPriority) {
3856     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3857   }
3858   return 0;
3859 }
3860 
3861 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3862   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3863 
3864   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3865   return (ret == 0) ? OS_OK : OS_ERR;
3866 }
3867 
3868 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3869   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3870     *priority_ptr = java_to_os_priority[NormPriority];
3871     return OS_OK;
3872   }
3873 
3874   errno = 0;
3875   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3876   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3877 }
3878 
3879 // Hint to the underlying OS that a task switch would not be good.
3880 // Void return because it's a hint and can fail.
3881 void os::hint_no_preempt() {}
3882 
3883 ////////////////////////////////////////////////////////////////////////////////
3884 // suspend/resume support
3885 
3886 //  the low-level signal-based suspend/resume support is a remnant from the
3887 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3888 //  within hotspot. Now there is a single use-case for this:
3889 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3890 //      that runs in the watcher thread.
3891 //  The remaining code is greatly simplified from the more general suspension
3892 //  code that used to be used.
3893 //
3894 //  The protocol is quite simple:
3895 //  - suspend:
3896 //      - sends a signal to the target thread
3897 //      - polls the suspend state of the osthread using a yield loop
3898 //      - target thread signal handler (SR_handler) sets suspend state
3899 //        and blocks in sigsuspend until continued
3900 //  - resume:
3901 //      - sets target osthread state to continue
3902 //      - sends signal to end the sigsuspend loop in the SR_handler
3903 //
3904 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3905 //
3906 
3907 static void resume_clear_context(OSThread *osthread) {
3908   osthread->set_ucontext(NULL);
3909   osthread->set_siginfo(NULL);
3910 }
3911 
3912 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3913   osthread->set_ucontext(context);
3914   osthread->set_siginfo(siginfo);
3915 }
3916 
3917 //
3918 // Handler function invoked when a thread's execution is suspended or
3919 // resumed. We have to be careful that only async-safe functions are
3920 // called here (Note: most pthread functions are not async safe and
3921 // should be avoided.)
3922 //
3923 // Note: sigwait() is a more natural fit than sigsuspend() from an
3924 // interface point of view, but sigwait() prevents the signal hander
3925 // from being run. libpthread would get very confused by not having
3926 // its signal handlers run and prevents sigwait()'s use with the
3927 // mutex granting granting signal.
3928 //
3929 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3930 //
3931 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3932   // Save and restore errno to avoid confusing native code with EINTR
3933   // after sigsuspend.
3934   int old_errno = errno;
3935 
3936   Thread* thread = Thread::current();
3937   OSThread* osthread = thread->osthread();
3938   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3939 
3940   os::SuspendResume::State current = osthread->sr.state();
3941   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3942     suspend_save_context(osthread, siginfo, context);
3943 
3944     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3945     os::SuspendResume::State state = osthread->sr.suspended();
3946     if (state == os::SuspendResume::SR_SUSPENDED) {
3947       sigset_t suspend_set;  // signals for sigsuspend()
3948 
3949       // get current set of blocked signals and unblock resume signal
3950       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3951       sigdelset(&suspend_set, SR_signum);
3952 
3953       sr_semaphore.signal();
3954       // wait here until we are resumed
3955       while (1) {
3956         sigsuspend(&suspend_set);
3957 
3958         os::SuspendResume::State result = osthread->sr.running();
3959         if (result == os::SuspendResume::SR_RUNNING) {
3960           sr_semaphore.signal();
3961           break;
3962         }
3963       }
3964 
3965     } else if (state == os::SuspendResume::SR_RUNNING) {
3966       // request was cancelled, continue
3967     } else {
3968       ShouldNotReachHere();
3969     }
3970 
3971     resume_clear_context(osthread);
3972   } else if (current == os::SuspendResume::SR_RUNNING) {
3973     // request was cancelled, continue
3974   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3975     // ignore
3976   } else {
3977     // ignore
3978   }
3979 
3980   errno = old_errno;
3981 }
3982 
3983 
3984 static int SR_initialize() {
3985   struct sigaction act;
3986   char *s;
3987   /* Get signal number to use for suspend/resume */
3988   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3989     int sig = ::strtol(s, 0, 10);
3990     if (sig > 0 || sig < _NSIG) {
3991         SR_signum = sig;
3992     }
3993   }
3994 
3995   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3996         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3997 
3998   sigemptyset(&SR_sigset);
3999   sigaddset(&SR_sigset, SR_signum);
4000 
4001   /* Set up signal handler for suspend/resume */
4002   act.sa_flags = SA_RESTART|SA_SIGINFO;
4003   act.sa_handler = (void (*)(int)) SR_handler;
4004 
4005   // SR_signum is blocked by default.
4006   // 4528190 - We also need to block pthread restart signal (32 on all
4007   // supported Linux platforms). Note that LinuxThreads need to block
4008   // this signal for all threads to work properly. So we don't have
4009   // to use hard-coded signal number when setting up the mask.
4010   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4011 
4012   if (sigaction(SR_signum, &act, 0) == -1) {
4013     return -1;
4014   }
4015 
4016   // Save signal flag
4017   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4018   return 0;
4019 }
4020 
4021 static int sr_notify(OSThread* osthread) {
4022   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4023   assert_status(status == 0, status, "pthread_kill");
4024   return status;
4025 }
4026 
4027 // "Randomly" selected value for how long we want to spin
4028 // before bailing out on suspending a thread, also how often
4029 // we send a signal to a thread we want to resume
4030 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4031 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4032 
4033 // returns true on success and false on error - really an error is fatal
4034 // but this seems the normal response to library errors
4035 static bool do_suspend(OSThread* osthread) {
4036   assert(osthread->sr.is_running(), "thread should be running");
4037   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4038 
4039   // mark as suspended and send signal
4040   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4041     // failed to switch, state wasn't running?
4042     ShouldNotReachHere();
4043     return false;
4044   }
4045 
4046   if (sr_notify(osthread) != 0) {
4047     ShouldNotReachHere();
4048   }
4049 
4050   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4051   while (true) {
4052     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4053       break;
4054     } else {
4055       // timeout
4056       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4057       if (cancelled == os::SuspendResume::SR_RUNNING) {
4058         return false;
4059       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4060         // make sure that we consume the signal on the semaphore as well
4061         sr_semaphore.wait();
4062         break;
4063       } else {
4064         ShouldNotReachHere();
4065         return false;
4066       }
4067     }
4068   }
4069 
4070   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4071   return true;
4072 }
4073 
4074 static void do_resume(OSThread* osthread) {
4075   assert(osthread->sr.is_suspended(), "thread should be suspended");
4076   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4077 
4078   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4079     // failed to switch to WAKEUP_REQUEST
4080     ShouldNotReachHere();
4081     return;
4082   }
4083 
4084   while (true) {
4085     if (sr_notify(osthread) == 0) {
4086       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4087         if (osthread->sr.is_running()) {
4088           return;
4089         }
4090       }
4091     } else {
4092       ShouldNotReachHere();
4093     }
4094   }
4095 
4096   guarantee(osthread->sr.is_running(), "Must be running!");
4097 }
4098 
4099 ///////////////////////////////////////////////////////////////////////////////////
4100 // signal handling (except suspend/resume)
4101 
4102 // This routine may be used by user applications as a "hook" to catch signals.
4103 // The user-defined signal handler must pass unrecognized signals to this
4104 // routine, and if it returns true (non-zero), then the signal handler must
4105 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4106 // routine will never retun false (zero), but instead will execute a VM panic
4107 // routine kill the process.
4108 //
4109 // If this routine returns false, it is OK to call it again.  This allows
4110 // the user-defined signal handler to perform checks either before or after
4111 // the VM performs its own checks.  Naturally, the user code would be making
4112 // a serious error if it tried to handle an exception (such as a null check
4113 // or breakpoint) that the VM was generating for its own correct operation.
4114 //
4115 // This routine may recognize any of the following kinds of signals:
4116 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4117 // It should be consulted by handlers for any of those signals.
4118 //
4119 // The caller of this routine must pass in the three arguments supplied
4120 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4121 // field of the structure passed to sigaction().  This routine assumes that
4122 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4123 //
4124 // Note that the VM will print warnings if it detects conflicting signal
4125 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4126 //
4127 extern "C" JNIEXPORT int
4128 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4129                         void* ucontext, int abort_if_unrecognized);
4130 
4131 void signalHandler(int sig, siginfo_t* info, void* uc) {
4132   assert(info != NULL && uc != NULL, "it must be old kernel");
4133   int orig_errno = errno;  // Preserve errno value over signal handler.
4134   JVM_handle_linux_signal(sig, info, uc, true);
4135   errno = orig_errno;
4136 }
4137 
4138 
4139 // This boolean allows users to forward their own non-matching signals
4140 // to JVM_handle_linux_signal, harmlessly.
4141 bool os::Linux::signal_handlers_are_installed = false;
4142 
4143 // For signal-chaining
4144 struct sigaction os::Linux::sigact[MAXSIGNUM];
4145 unsigned int os::Linux::sigs = 0;
4146 bool os::Linux::libjsig_is_loaded = false;
4147 typedef struct sigaction *(*get_signal_t)(int);
4148 get_signal_t os::Linux::get_signal_action = NULL;
4149 
4150 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4151   struct sigaction *actp = NULL;
4152 
4153   if (libjsig_is_loaded) {
4154     // Retrieve the old signal handler from libjsig
4155     actp = (*get_signal_action)(sig);
4156   }
4157   if (actp == NULL) {
4158     // Retrieve the preinstalled signal handler from jvm
4159     actp = get_preinstalled_handler(sig);
4160   }
4161 
4162   return actp;
4163 }
4164 
4165 static bool call_chained_handler(struct sigaction *actp, int sig,
4166                                  siginfo_t *siginfo, void *context) {
4167   // Call the old signal handler
4168   if (actp->sa_handler == SIG_DFL) {
4169     // It's more reasonable to let jvm treat it as an unexpected exception
4170     // instead of taking the default action.
4171     return false;
4172   } else if (actp->sa_handler != SIG_IGN) {
4173     if ((actp->sa_flags & SA_NODEFER) == 0) {
4174       // automaticlly block the signal
4175       sigaddset(&(actp->sa_mask), sig);
4176     }
4177 
4178     sa_handler_t hand;
4179     sa_sigaction_t sa;
4180     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4181     // retrieve the chained handler
4182     if (siginfo_flag_set) {
4183       sa = actp->sa_sigaction;
4184     } else {
4185       hand = actp->sa_handler;
4186     }
4187 
4188     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4189       actp->sa_handler = SIG_DFL;
4190     }
4191 
4192     // try to honor the signal mask
4193     sigset_t oset;
4194     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4195 
4196     // call into the chained handler
4197     if (siginfo_flag_set) {
4198       (*sa)(sig, siginfo, context);
4199     } else {
4200       (*hand)(sig);
4201     }
4202 
4203     // restore the signal mask
4204     pthread_sigmask(SIG_SETMASK, &oset, 0);
4205   }
4206   // Tell jvm's signal handler the signal is taken care of.
4207   return true;
4208 }
4209 
4210 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4211   bool chained = false;
4212   // signal-chaining
4213   if (UseSignalChaining) {
4214     struct sigaction *actp = get_chained_signal_action(sig);
4215     if (actp != NULL) {
4216       chained = call_chained_handler(actp, sig, siginfo, context);
4217     }
4218   }
4219   return chained;
4220 }
4221 
4222 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4223   if ((((unsigned int)1 << sig) & sigs) != 0) {
4224     return &sigact[sig];
4225   }
4226   return NULL;
4227 }
4228 
4229 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4230   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4231   sigact[sig] = oldAct;
4232   sigs |= (unsigned int)1 << sig;
4233 }
4234 
4235 // for diagnostic
4236 int os::Linux::sigflags[MAXSIGNUM];
4237 
4238 int os::Linux::get_our_sigflags(int sig) {
4239   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4240   return sigflags[sig];
4241 }
4242 
4243 void os::Linux::set_our_sigflags(int sig, int flags) {
4244   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4245   sigflags[sig] = flags;
4246 }
4247 
4248 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4249   // Check for overwrite.
4250   struct sigaction oldAct;
4251   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4252 
4253   void* oldhand = oldAct.sa_sigaction
4254                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4255                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4256   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4257       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4258       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4259     if (AllowUserSignalHandlers || !set_installed) {
4260       // Do not overwrite; user takes responsibility to forward to us.
4261       return;
4262     } else if (UseSignalChaining) {
4263       // save the old handler in jvm
4264       save_preinstalled_handler(sig, oldAct);
4265       // libjsig also interposes the sigaction() call below and saves the
4266       // old sigaction on it own.
4267     } else {
4268       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4269                     "%#lx for signal %d.", (long)oldhand, sig));
4270     }
4271   }
4272 
4273   struct sigaction sigAct;
4274   sigfillset(&(sigAct.sa_mask));
4275   sigAct.sa_handler = SIG_DFL;
4276   if (!set_installed) {
4277     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4278   } else {
4279     sigAct.sa_sigaction = signalHandler;
4280     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4281   }
4282   // Save flags, which are set by ours
4283   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4284   sigflags[sig] = sigAct.sa_flags;
4285 
4286   int ret = sigaction(sig, &sigAct, &oldAct);
4287   assert(ret == 0, "check");
4288 
4289   void* oldhand2  = oldAct.sa_sigaction
4290                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4291                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4292   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4293 }
4294 
4295 // install signal handlers for signals that HotSpot needs to
4296 // handle in order to support Java-level exception handling.
4297 
4298 void os::Linux::install_signal_handlers() {
4299   if (!signal_handlers_are_installed) {
4300     signal_handlers_are_installed = true;
4301 
4302     // signal-chaining
4303     typedef void (*signal_setting_t)();
4304     signal_setting_t begin_signal_setting = NULL;
4305     signal_setting_t end_signal_setting = NULL;
4306     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4307                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4308     if (begin_signal_setting != NULL) {
4309       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4310                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4311       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4312                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4313       libjsig_is_loaded = true;
4314       assert(UseSignalChaining, "should enable signal-chaining");
4315     }
4316     if (libjsig_is_loaded) {
4317       // Tell libjsig jvm is setting signal handlers
4318       (*begin_signal_setting)();
4319     }
4320 
4321     set_signal_handler(SIGSEGV, true);
4322     set_signal_handler(SIGPIPE, true);
4323     set_signal_handler(SIGBUS, true);
4324     set_signal_handler(SIGILL, true);
4325     set_signal_handler(SIGFPE, true);
4326 #if defined(PPC64)
4327     set_signal_handler(SIGTRAP, true);
4328 #endif
4329     set_signal_handler(SIGXFSZ, true);
4330 
4331     if (libjsig_is_loaded) {
4332       // Tell libjsig jvm finishes setting signal handlers
4333       (*end_signal_setting)();
4334     }
4335 
4336     // We don't activate signal checker if libjsig is in place, we trust ourselves
4337     // and if UserSignalHandler is installed all bets are off.
4338     // Log that signal checking is off only if -verbose:jni is specified.
4339     if (CheckJNICalls) {
4340       if (libjsig_is_loaded) {
4341         if (PrintJNIResolving) {
4342           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4343         }
4344         check_signals = false;
4345       }
4346       if (AllowUserSignalHandlers) {
4347         if (PrintJNIResolving) {
4348           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4349         }
4350         check_signals = false;
4351       }
4352     }
4353   }
4354 }
4355 
4356 // This is the fastest way to get thread cpu time on Linux.
4357 // Returns cpu time (user+sys) for any thread, not only for current.
4358 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4359 // It might work on 2.6.10+ with a special kernel/glibc patch.
4360 // For reference, please, see IEEE Std 1003.1-2004:
4361 //   http://www.unix.org/single_unix_specification
4362 
4363 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4364   struct timespec tp;
4365   int rc = os::Linux::clock_gettime(clockid, &tp);
4366   assert(rc == 0, "clock_gettime is expected to return 0 code");
4367 
4368   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4369 }
4370 
4371 /////
4372 // glibc on Linux platform uses non-documented flag
4373 // to indicate, that some special sort of signal
4374 // trampoline is used.
4375 // We will never set this flag, and we should
4376 // ignore this flag in our diagnostic
4377 #ifdef SIGNIFICANT_SIGNAL_MASK
4378 #undef SIGNIFICANT_SIGNAL_MASK
4379 #endif
4380 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4381 
4382 static const char* get_signal_handler_name(address handler,
4383                                            char* buf, int buflen) {
4384   int offset;
4385   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4386   if (found) {
4387     // skip directory names
4388     const char *p1, *p2;
4389     p1 = buf;
4390     size_t len = strlen(os::file_separator());
4391     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4392     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4393   } else {
4394     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4395   }
4396   return buf;
4397 }
4398 
4399 static void print_signal_handler(outputStream* st, int sig,
4400                                  char* buf, size_t buflen) {
4401   struct sigaction sa;
4402 
4403   sigaction(sig, NULL, &sa);
4404 
4405   // See comment for SIGNIFICANT_SIGNAL_MASK define
4406   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4407 
4408   st->print("%s: ", os::exception_name(sig, buf, buflen));
4409 
4410   address handler = (sa.sa_flags & SA_SIGINFO)
4411     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4412     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4413 
4414   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4415     st->print("SIG_DFL");
4416   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4417     st->print("SIG_IGN");
4418   } else {
4419     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4420   }
4421 
4422   st->print(", sa_mask[0]=");
4423   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4424 
4425   address rh = VMError::get_resetted_sighandler(sig);
4426   // May be, handler was resetted by VMError?
4427   if (rh != NULL) {
4428     handler = rh;
4429     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4430   }
4431 
4432   st->print(", sa_flags=");
4433   os::Posix::print_sa_flags(st, sa.sa_flags);
4434 
4435   // Check: is it our handler?
4436   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4437      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4438     // It is our signal handler
4439     // check for flags, reset system-used one!
4440     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4441       st->print(
4442                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4443                 os::Linux::get_our_sigflags(sig));
4444     }
4445   }
4446   st->cr();
4447 }
4448 
4449 
4450 #define DO_SIGNAL_CHECK(sig) \
4451   if (!sigismember(&check_signal_done, sig)) \
4452     os::Linux::check_signal_handler(sig)
4453 
4454 // This method is a periodic task to check for misbehaving JNI applications
4455 // under CheckJNI, we can add any periodic checks here
4456 
4457 void os::run_periodic_checks() {
4458 
4459   if (check_signals == false) return;
4460 
4461   // SEGV and BUS if overridden could potentially prevent
4462   // generation of hs*.log in the event of a crash, debugging
4463   // such a case can be very challenging, so we absolutely
4464   // check the following for a good measure:
4465   DO_SIGNAL_CHECK(SIGSEGV);
4466   DO_SIGNAL_CHECK(SIGILL);
4467   DO_SIGNAL_CHECK(SIGFPE);
4468   DO_SIGNAL_CHECK(SIGBUS);
4469   DO_SIGNAL_CHECK(SIGPIPE);
4470   DO_SIGNAL_CHECK(SIGXFSZ);
4471 #if defined(PPC64)
4472   DO_SIGNAL_CHECK(SIGTRAP);
4473 #endif
4474 
4475   // ReduceSignalUsage allows the user to override these handlers
4476   // see comments at the very top and jvm_solaris.h
4477   if (!ReduceSignalUsage) {
4478     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4479     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4480     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4481     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4482   }
4483 
4484   DO_SIGNAL_CHECK(SR_signum);
4485   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4486 }
4487 
4488 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4489 
4490 static os_sigaction_t os_sigaction = NULL;
4491 
4492 void os::Linux::check_signal_handler(int sig) {
4493   char buf[O_BUFLEN];
4494   address jvmHandler = NULL;
4495 
4496 
4497   struct sigaction act;
4498   if (os_sigaction == NULL) {
4499     // only trust the default sigaction, in case it has been interposed
4500     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4501     if (os_sigaction == NULL) return;
4502   }
4503 
4504   os_sigaction(sig, (struct sigaction*)NULL, &act);
4505 
4506 
4507   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4508 
4509   address thisHandler = (act.sa_flags & SA_SIGINFO)
4510     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4511     : CAST_FROM_FN_PTR(address, act.sa_handler);
4512 
4513 
4514   switch (sig) {
4515   case SIGSEGV:
4516   case SIGBUS:
4517   case SIGFPE:
4518   case SIGPIPE:
4519   case SIGILL:
4520   case SIGXFSZ:
4521     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4522     break;
4523 
4524   case SHUTDOWN1_SIGNAL:
4525   case SHUTDOWN2_SIGNAL:
4526   case SHUTDOWN3_SIGNAL:
4527   case BREAK_SIGNAL:
4528     jvmHandler = (address)user_handler();
4529     break;
4530 
4531   case INTERRUPT_SIGNAL:
4532     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4533     break;
4534 
4535   default:
4536     if (sig == SR_signum) {
4537       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4538     } else {
4539       return;
4540     }
4541     break;
4542   }
4543 
4544   if (thisHandler != jvmHandler) {
4545     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4546     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4547     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4548     // No need to check this sig any longer
4549     sigaddset(&check_signal_done, sig);
4550     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4551     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4552       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4553                     exception_name(sig, buf, O_BUFLEN));
4554     }
4555   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4556     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4557     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4558     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4559     // No need to check this sig any longer
4560     sigaddset(&check_signal_done, sig);
4561   }
4562 
4563   // Dump all the signal
4564   if (sigismember(&check_signal_done, sig)) {
4565     print_signal_handlers(tty, buf, O_BUFLEN);
4566   }
4567 }
4568 
4569 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4570 
4571 extern bool signal_name(int signo, char* buf, size_t len);
4572 
4573 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4574   if (0 < exception_code && exception_code <= SIGRTMAX) {
4575     // signal
4576     if (!signal_name(exception_code, buf, size)) {
4577       jio_snprintf(buf, size, "SIG%d", exception_code);
4578     }
4579     return buf;
4580   } else {
4581     return NULL;
4582   }
4583 }
4584 
4585 // this is called _before_ the most of global arguments have been parsed
4586 void os::init(void) {
4587   char dummy;   /* used to get a guess on initial stack address */
4588 //  first_hrtime = gethrtime();
4589 
4590   // With LinuxThreads the JavaMain thread pid (primordial thread)
4591   // is different than the pid of the java launcher thread.
4592   // So, on Linux, the launcher thread pid is passed to the VM
4593   // via the sun.java.launcher.pid property.
4594   // Use this property instead of getpid() if it was correctly passed.
4595   // See bug 6351349.
4596   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4597 
4598   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4599 
4600   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4601 
4602   init_random(1234567);
4603 
4604   ThreadCritical::initialize();
4605 
4606   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4607   if (Linux::page_size() == -1) {
4608     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4609                   strerror(errno)));
4610   }
4611   init_page_sizes((size_t) Linux::page_size());
4612 
4613   Linux::initialize_system_info();
4614 
4615   // main_thread points to the aboriginal thread
4616   Linux::_main_thread = pthread_self();
4617 
4618   Linux::clock_init();
4619   initial_time_count = javaTimeNanos();
4620 
4621   // pthread_condattr initialization for monotonic clock
4622   int status;
4623   pthread_condattr_t* _condattr = os::Linux::condAttr();
4624   if ((status = pthread_condattr_init(_condattr)) != 0) {
4625     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4626   }
4627   // Only set the clock if CLOCK_MONOTONIC is available
4628   if (os::supports_monotonic_clock()) {
4629     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4630       if (status == EINVAL) {
4631         warning("Unable to use monotonic clock with relative timed-waits" \
4632                 " - changes to the time-of-day clock may have adverse affects");
4633       } else {
4634         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4635       }
4636     }
4637   }
4638   // else it defaults to CLOCK_REALTIME
4639 
4640   pthread_mutex_init(&dl_mutex, NULL);
4641 
4642   // If the pagesize of the VM is greater than 8K determine the appropriate
4643   // number of initial guard pages.  The user can change this with the
4644   // command line arguments, if needed.
4645   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4646     StackYellowPages = 1;
4647     StackRedPages = 1;
4648     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4649   }
4650 }
4651 
4652 // To install functions for atexit system call
4653 extern "C" {
4654   static void perfMemory_exit_helper() {
4655     perfMemory_exit();
4656   }
4657 }
4658 
4659 // this is called _after_ the global arguments have been parsed
4660 jint os::init_2(void)
4661 {
4662   Linux::fast_thread_clock_init();
4663 
4664   // Allocate a single page and mark it as readable for safepoint polling
4665   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4666   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4667 
4668   os::set_polling_page(polling_page);
4669 
4670 #ifndef PRODUCT
4671   if (Verbose && PrintMiscellaneous)
4672     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4673 #endif
4674 
4675   if (!UseMembar) {
4676     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4677     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4678     os::set_memory_serialize_page(mem_serialize_page);
4679 
4680 #ifndef PRODUCT
4681     if (Verbose && PrintMiscellaneous)
4682       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4683 #endif
4684   }
4685 
4686   // initialize suspend/resume support - must do this before signal_sets_init()
4687   if (SR_initialize() != 0) {
4688     perror("SR_initialize failed");
4689     return JNI_ERR;
4690   }
4691 
4692   Linux::signal_sets_init();
4693   Linux::install_signal_handlers();
4694 
4695   // Check minimum allowable stack size for thread creation and to initialize
4696   // the java system classes, including StackOverflowError - depends on page
4697   // size.  Add a page for compiler2 recursion in main thread.
4698   // Add in 2*BytesPerWord times page size to account for VM stack during
4699   // class initialization depending on 32 or 64 bit VM.
4700   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4701             (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4702                     (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4703 
4704   size_t threadStackSizeInBytes = ThreadStackSize * K;
4705   if (threadStackSizeInBytes != 0 &&
4706       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4707         tty->print_cr("\nThe stack size specified is too small, "
4708                       "Specify at least %dk",
4709                       os::Linux::min_stack_allowed/ K);
4710         return JNI_ERR;
4711   }
4712 
4713   // Make the stack size a multiple of the page size so that
4714   // the yellow/red zones can be guarded.
4715   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4716         vm_page_size()));
4717 
4718   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4719 
4720 #if defined(IA32)
4721   workaround_expand_exec_shield_cs_limit();
4722 #endif
4723 
4724   Linux::libpthread_init();
4725   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4726      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4727           Linux::glibc_version(), Linux::libpthread_version(),
4728           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4729   }
4730 
4731   if (UseNUMA) {
4732     if (!Linux::libnuma_init()) {
4733       UseNUMA = false;
4734     } else {
4735       if ((Linux::numa_max_node() < 1)) {
4736         // There's only one node(they start from 0), disable NUMA.
4737         UseNUMA = false;
4738       }
4739     }
4740     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4741     // we can make the adaptive lgrp chunk resizing work. If the user specified
4742     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4743     // disable adaptive resizing.
4744     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4745       if (FLAG_IS_DEFAULT(UseNUMA)) {
4746         UseNUMA = false;
4747       } else {
4748         if (FLAG_IS_DEFAULT(UseLargePages) &&
4749             FLAG_IS_DEFAULT(UseSHM) &&
4750             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4751           UseLargePages = false;
4752         } else {
4753           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4754           UseAdaptiveSizePolicy = false;
4755           UseAdaptiveNUMAChunkSizing = false;
4756         }
4757       }
4758     }
4759     if (!UseNUMA && ForceNUMA) {
4760       UseNUMA = true;
4761     }
4762   }
4763 
4764   if (MaxFDLimit) {
4765     // set the number of file descriptors to max. print out error
4766     // if getrlimit/setrlimit fails but continue regardless.
4767     struct rlimit nbr_files;
4768     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4769     if (status != 0) {
4770       if (PrintMiscellaneous && (Verbose || WizardMode))
4771         perror("os::init_2 getrlimit failed");
4772     } else {
4773       nbr_files.rlim_cur = nbr_files.rlim_max;
4774       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4775       if (status != 0) {
4776         if (PrintMiscellaneous && (Verbose || WizardMode))
4777           perror("os::init_2 setrlimit failed");
4778       }
4779     }
4780   }
4781 
4782   // Initialize lock used to serialize thread creation (see os::create_thread)
4783   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4784 
4785   // at-exit methods are called in the reverse order of their registration.
4786   // atexit functions are called on return from main or as a result of a
4787   // call to exit(3C). There can be only 32 of these functions registered
4788   // and atexit() does not set errno.
4789 
4790   if (PerfAllowAtExitRegistration) {
4791     // only register atexit functions if PerfAllowAtExitRegistration is set.
4792     // atexit functions can be delayed until process exit time, which
4793     // can be problematic for embedded VM situations. Embedded VMs should
4794     // call DestroyJavaVM() to assure that VM resources are released.
4795 
4796     // note: perfMemory_exit_helper atexit function may be removed in
4797     // the future if the appropriate cleanup code can be added to the
4798     // VM_Exit VMOperation's doit method.
4799     if (atexit(perfMemory_exit_helper) != 0) {
4800       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4801     }
4802   }
4803 
4804   // initialize thread priority policy
4805   prio_init();
4806 
4807   return JNI_OK;
4808 }
4809 
4810 // this is called at the end of vm_initialization
4811 void os::init_3(void) {
4812 #ifdef JAVASE_EMBEDDED
4813   // Start the MemNotifyThread
4814   if (LowMemoryProtection) {
4815     MemNotifyThread::start();
4816   }
4817   return;
4818 #endif
4819 }
4820 
4821 // Mark the polling page as unreadable
4822 void os::make_polling_page_unreadable(void) {
4823   if (!guard_memory((char*)_polling_page, Linux::page_size()))
4824     fatal("Could not disable polling page");
4825 };
4826 
4827 // Mark the polling page as readable
4828 void os::make_polling_page_readable(void) {
4829   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4830     fatal("Could not enable polling page");
4831   }
4832 };
4833 
4834 int os::active_processor_count() {
4835   // Linux doesn't yet have a (official) notion of processor sets,
4836   // so just return the number of online processors.
4837   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4838   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4839   return online_cpus;
4840 }
4841 
4842 void os::set_native_thread_name(const char *name) {
4843   // Not yet implemented.
4844   return;
4845 }
4846 
4847 bool os::distribute_processes(uint length, uint* distribution) {
4848   // Not yet implemented.
4849   return false;
4850 }
4851 
4852 bool os::bind_to_processor(uint processor_id) {
4853   // Not yet implemented.
4854   return false;
4855 }
4856 
4857 ///
4858 
4859 void os::SuspendedThreadTask::internal_do_task() {
4860   if (do_suspend(_thread->osthread())) {
4861     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4862     do_task(context);
4863     do_resume(_thread->osthread());
4864   }
4865 }
4866 
4867 class PcFetcher : public os::SuspendedThreadTask {
4868 public:
4869   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4870   ExtendedPC result();
4871 protected:
4872   void do_task(const os::SuspendedThreadTaskContext& context);
4873 private:
4874   ExtendedPC _epc;
4875 };
4876 
4877 ExtendedPC PcFetcher::result() {
4878   guarantee(is_done(), "task is not done yet.");
4879   return _epc;
4880 }
4881 
4882 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4883   Thread* thread = context.thread();
4884   OSThread* osthread = thread->osthread();
4885   if (osthread->ucontext() != NULL) {
4886     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4887   } else {
4888     // NULL context is unexpected, double-check this is the VMThread
4889     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4890   }
4891 }
4892 
4893 // Suspends the target using the signal mechanism and then grabs the PC before
4894 // resuming the target. Used by the flat-profiler only
4895 ExtendedPC os::get_thread_pc(Thread* thread) {
4896   // Make sure that it is called by the watcher for the VMThread
4897   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4898   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4899 
4900   PcFetcher fetcher(thread);
4901   fetcher.run();
4902   return fetcher.result();
4903 }
4904 
4905 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4906 {
4907    if (is_NPTL()) {
4908       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4909    } else {
4910       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4911       // word back to default 64bit precision if condvar is signaled. Java
4912       // wants 53bit precision.  Save and restore current value.
4913       int fpu = get_fpu_control_word();
4914       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4915       set_fpu_control_word(fpu);
4916       return status;
4917    }
4918 }
4919 
4920 ////////////////////////////////////////////////////////////////////////////////
4921 // debug support
4922 
4923 bool os::find(address addr, outputStream* st) {
4924   Dl_info dlinfo;
4925   memset(&dlinfo, 0, sizeof(dlinfo));
4926   if (dladdr(addr, &dlinfo) != 0) {
4927     st->print(PTR_FORMAT ": ", addr);
4928     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4929       st->print("%s+%#x", dlinfo.dli_sname,
4930                  addr - (intptr_t)dlinfo.dli_saddr);
4931     } else if (dlinfo.dli_fbase != NULL) {
4932       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4933     } else {
4934       st->print("<absolute address>");
4935     }
4936     if (dlinfo.dli_fname != NULL) {
4937       st->print(" in %s", dlinfo.dli_fname);
4938     }
4939     if (dlinfo.dli_fbase != NULL) {
4940       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4941     }
4942     st->cr();
4943 
4944     if (Verbose) {
4945       // decode some bytes around the PC
4946       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4947       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4948       address       lowest = (address) dlinfo.dli_sname;
4949       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4950       if (begin < lowest)  begin = lowest;
4951       Dl_info dlinfo2;
4952       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4953           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4954         end = (address) dlinfo2.dli_saddr;
4955       Disassembler::decode(begin, end, st);
4956     }
4957     return true;
4958   }
4959   return false;
4960 }
4961 
4962 ////////////////////////////////////////////////////////////////////////////////
4963 // misc
4964 
4965 // This does not do anything on Linux. This is basically a hook for being
4966 // able to use structured exception handling (thread-local exception filters)
4967 // on, e.g., Win32.
4968 void
4969 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4970                          JavaCallArguments* args, Thread* thread) {
4971   f(value, method, args, thread);
4972 }
4973 
4974 void os::print_statistics() {
4975 }
4976 
4977 int os::message_box(const char* title, const char* message) {
4978   int i;
4979   fdStream err(defaultStream::error_fd());
4980   for (i = 0; i < 78; i++) err.print_raw("=");
4981   err.cr();
4982   err.print_raw_cr(title);
4983   for (i = 0; i < 78; i++) err.print_raw("-");
4984   err.cr();
4985   err.print_raw_cr(message);
4986   for (i = 0; i < 78; i++) err.print_raw("=");
4987   err.cr();
4988 
4989   char buf[16];
4990   // Prevent process from exiting upon "read error" without consuming all CPU
4991   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4992 
4993   return buf[0] == 'y' || buf[0] == 'Y';
4994 }
4995 
4996 int os::stat(const char *path, struct stat *sbuf) {
4997   char pathbuf[MAX_PATH];
4998   if (strlen(path) > MAX_PATH - 1) {
4999     errno = ENAMETOOLONG;
5000     return -1;
5001   }
5002   os::native_path(strcpy(pathbuf, path));
5003   return ::stat(pathbuf, sbuf);
5004 }
5005 
5006 bool os::check_heap(bool force) {
5007   return true;
5008 }
5009 
5010 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5011   return ::vsnprintf(buf, count, format, args);
5012 }
5013 
5014 // Is a (classpath) directory empty?
5015 bool os::dir_is_empty(const char* path) {
5016   DIR *dir = NULL;
5017   struct dirent *ptr;
5018 
5019   dir = opendir(path);
5020   if (dir == NULL) return true;
5021 
5022   /* Scan the directory */
5023   bool result = true;
5024   char buf[sizeof(struct dirent) + MAX_PATH];
5025   while (result && (ptr = ::readdir(dir)) != NULL) {
5026     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5027       result = false;
5028     }
5029   }
5030   closedir(dir);
5031   return result;
5032 }
5033 
5034 // This code originates from JDK's sysOpen and open64_w
5035 // from src/solaris/hpi/src/system_md.c
5036 
5037 #ifndef O_DELETE
5038 #define O_DELETE 0x10000
5039 #endif
5040 
5041 // Open a file. Unlink the file immediately after open returns
5042 // if the specified oflag has the O_DELETE flag set.
5043 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5044 
5045 int os::open(const char *path, int oflag, int mode) {
5046 
5047   if (strlen(path) > MAX_PATH - 1) {
5048     errno = ENAMETOOLONG;
5049     return -1;
5050   }
5051   int fd;
5052   int o_delete = (oflag & O_DELETE);
5053   oflag = oflag & ~O_DELETE;
5054 
5055   fd = ::open64(path, oflag, mode);
5056   if (fd == -1) return -1;
5057 
5058   //If the open succeeded, the file might still be a directory
5059   {
5060     struct stat64 buf64;
5061     int ret = ::fstat64(fd, &buf64);
5062     int st_mode = buf64.st_mode;
5063 
5064     if (ret != -1) {
5065       if ((st_mode & S_IFMT) == S_IFDIR) {
5066         errno = EISDIR;
5067         ::close(fd);
5068         return -1;
5069       }
5070     } else {
5071       ::close(fd);
5072       return -1;
5073     }
5074   }
5075 
5076     /*
5077      * All file descriptors that are opened in the JVM and not
5078      * specifically destined for a subprocess should have the
5079      * close-on-exec flag set.  If we don't set it, then careless 3rd
5080      * party native code might fork and exec without closing all
5081      * appropriate file descriptors (e.g. as we do in closeDescriptors in
5082      * UNIXProcess.c), and this in turn might:
5083      *
5084      * - cause end-of-file to fail to be detected on some file
5085      *   descriptors, resulting in mysterious hangs, or
5086      *
5087      * - might cause an fopen in the subprocess to fail on a system
5088      *   suffering from bug 1085341.
5089      *
5090      * (Yes, the default setting of the close-on-exec flag is a Unix
5091      * design flaw)
5092      *
5093      * See:
5094      * 1085341: 32-bit stdio routines should support file descriptors >255
5095      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5096      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5097      */
5098 #ifdef FD_CLOEXEC
5099     {
5100         int flags = ::fcntl(fd, F_GETFD);
5101         if (flags != -1)
5102             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5103     }
5104 #endif
5105 
5106   if (o_delete != 0) {
5107     ::unlink(path);
5108   }
5109   return fd;
5110 }
5111 
5112 
5113 // create binary file, rewriting existing file if required
5114 int os::create_binary_file(const char* path, bool rewrite_existing) {
5115   int oflags = O_WRONLY | O_CREAT;
5116   if (!rewrite_existing) {
5117     oflags |= O_EXCL;
5118   }
5119   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5120 }
5121 
5122 // return current position of file pointer
5123 jlong os::current_file_offset(int fd) {
5124   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5125 }
5126 
5127 // move file pointer to the specified offset
5128 jlong os::seek_to_file_offset(int fd, jlong offset) {
5129   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5130 }
5131 
5132 // This code originates from JDK's sysAvailable
5133 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5134 
5135 int os::available(int fd, jlong *bytes) {
5136   jlong cur, end;
5137   int mode;
5138   struct stat64 buf64;
5139 
5140   if (::fstat64(fd, &buf64) >= 0) {
5141     mode = buf64.st_mode;
5142     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5143       /*
5144       * XXX: is the following call interruptible? If so, this might
5145       * need to go through the INTERRUPT_IO() wrapper as for other
5146       * blocking, interruptible calls in this file.
5147       */
5148       int n;
5149       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5150         *bytes = n;
5151         return 1;
5152       }
5153     }
5154   }
5155   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5156     return 0;
5157   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5158     return 0;
5159   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5160     return 0;
5161   }
5162   *bytes = end - cur;
5163   return 1;
5164 }
5165 
5166 int os::socket_available(int fd, jint *pbytes) {
5167   // Linux doc says EINTR not returned, unlike Solaris
5168   int ret = ::ioctl(fd, FIONREAD, pbytes);
5169 
5170   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5171   // is expected to return 0 on failure and 1 on success to the jdk.
5172   return (ret < 0) ? 0 : 1;
5173 }
5174 
5175 // Map a block of memory.
5176 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5177                      char *addr, size_t bytes, bool read_only,
5178                      bool allow_exec) {
5179   int prot;
5180   int flags = MAP_PRIVATE;
5181 
5182   if (read_only) {
5183     prot = PROT_READ;
5184   } else {
5185     prot = PROT_READ | PROT_WRITE;
5186   }
5187 
5188   if (allow_exec) {
5189     prot |= PROT_EXEC;
5190   }
5191 
5192   if (addr != NULL) {
5193     flags |= MAP_FIXED;
5194   }
5195 
5196   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5197                                      fd, file_offset);
5198   if (mapped_address == MAP_FAILED) {
5199     return NULL;
5200   }
5201   return mapped_address;
5202 }
5203 
5204 
5205 // Remap a block of memory.
5206 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5207                        char *addr, size_t bytes, bool read_only,
5208                        bool allow_exec) {
5209   // same as map_memory() on this OS
5210   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5211                         allow_exec);
5212 }
5213 
5214 
5215 // Unmap a block of memory.
5216 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5217   return munmap(addr, bytes) == 0;
5218 }
5219 
5220 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5221 
5222 static clockid_t thread_cpu_clockid(Thread* thread) {
5223   pthread_t tid = thread->osthread()->pthread_id();
5224   clockid_t clockid;
5225 
5226   // Get thread clockid
5227   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5228   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5229   return clockid;
5230 }
5231 
5232 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5233 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5234 // of a thread.
5235 //
5236 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5237 // the fast estimate available on the platform.
5238 
5239 jlong os::current_thread_cpu_time() {
5240   if (os::Linux::supports_fast_thread_cpu_time()) {
5241     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5242   } else {
5243     // return user + sys since the cost is the same
5244     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5245   }
5246 }
5247 
5248 jlong os::thread_cpu_time(Thread* thread) {
5249   // consistent with what current_thread_cpu_time() returns
5250   if (os::Linux::supports_fast_thread_cpu_time()) {
5251     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5252   } else {
5253     return slow_thread_cpu_time(thread, true /* user + sys */);
5254   }
5255 }
5256 
5257 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5258   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5259     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5260   } else {
5261     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5262   }
5263 }
5264 
5265 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5266   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5267     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5268   } else {
5269     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5270   }
5271 }
5272 
5273 //
5274 //  -1 on error.
5275 //
5276 
5277 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5278   pid_t  tid = thread->osthread()->thread_id();
5279   char *s;
5280   char stat[2048];
5281   int statlen;
5282   char proc_name[64];
5283   int count;
5284   long sys_time, user_time;
5285   char cdummy;
5286   int idummy;
5287   long ldummy;
5288   FILE *fp;
5289 
5290   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5291   fp = fopen(proc_name, "r");
5292   if (fp == NULL) return -1;
5293   statlen = fread(stat, 1, 2047, fp);
5294   stat[statlen] = '\0';
5295   fclose(fp);
5296 
5297   // Skip pid and the command string. Note that we could be dealing with
5298   // weird command names, e.g. user could decide to rename java launcher
5299   // to "java 1.4.2 :)", then the stat file would look like
5300   //                1234 (java 1.4.2 :)) R ... ...
5301   // We don't really need to know the command string, just find the last
5302   // occurrence of ")" and then start parsing from there. See bug 4726580.
5303   s = strrchr(stat, ')');
5304   if (s == NULL) return -1;
5305 
5306   // Skip blank chars
5307   do s++; while (isspace(*s));
5308 
5309   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5310                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5311                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5312                  &user_time, &sys_time);
5313   if (count != 13) return -1;
5314   if (user_sys_cpu_time) {
5315     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5316   } else {
5317     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5318   }
5319 }
5320 
5321 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5322   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5323   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5324   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5325   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5326 }
5327 
5328 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5329   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5330   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5331   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5332   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5333 }
5334 
5335 bool os::is_thread_cpu_time_supported() {
5336   return true;
5337 }
5338 
5339 // System loadavg support.  Returns -1 if load average cannot be obtained.
5340 // Linux doesn't yet have a (official) notion of processor sets,
5341 // so just return the system wide load average.
5342 int os::loadavg(double loadavg[], int nelem) {
5343   return ::getloadavg(loadavg, nelem);
5344 }
5345 
5346 void os::pause() {
5347   char filename[MAX_PATH];
5348   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5349     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5350   } else {
5351     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5352   }
5353 
5354   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5355   if (fd != -1) {
5356     struct stat buf;
5357     ::close(fd);
5358     while (::stat(filename, &buf) == 0) {
5359       (void)::poll(NULL, 0, 100);
5360     }
5361   } else {
5362     jio_fprintf(stderr,
5363       "Could not open pause file '%s', continuing immediately.\n", filename);
5364   }
5365 }
5366 
5367 
5368 // Refer to the comments in os_solaris.cpp park-unpark.
5369 //
5370 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5371 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5372 // For specifics regarding the bug see GLIBC BUGID 261237 :
5373 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5374 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5375 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5376 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5377 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5378 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5379 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5380 // of libpthread avoids the problem, but isn't practical.
5381 //
5382 // Possible remedies:
5383 //
5384 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5385 //      This is palliative and probabilistic, however.  If the thread is preempted
5386 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5387 //      than the minimum period may have passed, and the abstime may be stale (in the
5388 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5389 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5390 //
5391 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5392 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5393 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5394 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5395 //      thread.
5396 //
5397 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5398 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5399 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5400 //      This also works well.  In fact it avoids kernel-level scalability impediments
5401 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5402 //      timers in a graceful fashion.
5403 //
5404 // 4.   When the abstime value is in the past it appears that control returns
5405 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5406 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5407 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5408 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5409 //      It may be possible to avoid reinitialization by checking the return
5410 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5411 //      condvar we must establish the invariant that cond_signal() is only called
5412 //      within critical sections protected by the adjunct mutex.  This prevents
5413 //      cond_signal() from "seeing" a condvar that's in the midst of being
5414 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5415 //      desirable signal-after-unlock optimization that avoids futile context switching.
5416 //
5417 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5418 //      structure when a condvar is used or initialized.  cond_destroy()  would
5419 //      release the helper structure.  Our reinitialize-after-timedwait fix
5420 //      put excessive stress on malloc/free and locks protecting the c-heap.
5421 //
5422 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5423 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5424 // and only enabling the work-around for vulnerable environments.
5425 
5426 // utility to compute the abstime argument to timedwait:
5427 // millis is the relative timeout time
5428 // abstime will be the absolute timeout time
5429 // TODO: replace compute_abstime() with unpackTime()
5430 
5431 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5432   if (millis < 0)  millis = 0;
5433 
5434   jlong seconds = millis / 1000;
5435   millis %= 1000;
5436   if (seconds > 50000000) { // see man cond_timedwait(3T)
5437     seconds = 50000000;
5438   }
5439 
5440   if (os::supports_monotonic_clock()) {
5441     struct timespec now;
5442     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5443     assert_status(status == 0, status, "clock_gettime");
5444     abstime->tv_sec = now.tv_sec  + seconds;
5445     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5446     if (nanos >= NANOSECS_PER_SEC) {
5447       abstime->tv_sec += 1;
5448       nanos -= NANOSECS_PER_SEC;
5449     }
5450     abstime->tv_nsec = nanos;
5451   } else {
5452     struct timeval now;
5453     int status = gettimeofday(&now, NULL);
5454     assert(status == 0, "gettimeofday");
5455     abstime->tv_sec = now.tv_sec  + seconds;
5456     long usec = now.tv_usec + millis * 1000;
5457     if (usec >= 1000000) {
5458       abstime->tv_sec += 1;
5459       usec -= 1000000;
5460     }
5461     abstime->tv_nsec = usec * 1000;
5462   }
5463   return abstime;
5464 }
5465 
5466 void os::PlatformEvent::park() {       // AKA "down()"
5467   // Invariant: Only the thread associated with the Event/PlatformEvent
5468   // may call park().
5469   // TODO: assert that _Assoc != NULL or _Assoc == Self
5470   assert(_nParked == 0, "invariant");
5471 
5472   int v;
5473   for (;;) {
5474       v = _Event;
5475       if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5476   }
5477   guarantee(v >= 0, "invariant");
5478   if (v == 0) {
5479      // Do this the hard way by blocking ...
5480      int status = pthread_mutex_lock(_mutex);
5481      assert_status(status == 0, status, "mutex_lock");
5482      guarantee(_nParked == 0, "invariant");
5483      ++_nParked;
5484      while (_Event < 0) {
5485         status = pthread_cond_wait(_cond, _mutex);
5486         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5487         // Treat this the same as if the wait was interrupted
5488         if (status == ETIME) { status = EINTR; }
5489         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5490      }
5491      --_nParked;
5492 
5493     _Event = 0;
5494      status = pthread_mutex_unlock(_mutex);
5495      assert_status(status == 0, status, "mutex_unlock");
5496     // Paranoia to ensure our locked and lock-free paths interact
5497     // correctly with each other.
5498     OrderAccess::fence();
5499   }
5500   guarantee(_Event >= 0, "invariant");
5501 }
5502 
5503 int os::PlatformEvent::park(jlong millis) {
5504   guarantee(_nParked == 0, "invariant");
5505 
5506   int v;
5507   for (;;) {
5508       v = _Event;
5509       if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5510   }
5511   guarantee(v >= 0, "invariant");
5512   if (v != 0) return OS_OK;
5513 
5514   // We do this the hard way, by blocking the thread.
5515   // Consider enforcing a minimum timeout value.
5516   struct timespec abst;
5517   compute_abstime(&abst, millis);
5518 
5519   int ret = OS_TIMEOUT;
5520   int status = pthread_mutex_lock(_mutex);
5521   assert_status(status == 0, status, "mutex_lock");
5522   guarantee(_nParked == 0, "invariant");
5523   ++_nParked;
5524 
5525   // Object.wait(timo) will return because of
5526   // (a) notification
5527   // (b) timeout
5528   // (c) thread.interrupt
5529   //
5530   // Thread.interrupt and object.notify{All} both call Event::set.
5531   // That is, we treat thread.interrupt as a special case of notification.
5532   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5533   // We assume all ETIME returns are valid.
5534   //
5535   // TODO: properly differentiate simultaneous notify+interrupt.
5536   // In that case, we should propagate the notify to another waiter.
5537 
5538   while (_Event < 0) {
5539     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5540     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5541       pthread_cond_destroy(_cond);
5542       pthread_cond_init(_cond, os::Linux::condAttr());
5543     }
5544     assert_status(status == 0 || status == EINTR ||
5545                   status == ETIME || status == ETIMEDOUT,
5546                   status, "cond_timedwait");
5547     if (!FilterSpuriousWakeups) break;                 // previous semantics
5548     if (status == ETIME || status == ETIMEDOUT) break;
5549     // We consume and ignore EINTR and spurious wakeups.
5550   }
5551   --_nParked;
5552   if (_Event >= 0) {
5553      ret = OS_OK;
5554   }
5555   _Event = 0;
5556   status = pthread_mutex_unlock(_mutex);
5557   assert_status(status == 0, status, "mutex_unlock");
5558   assert(_nParked == 0, "invariant");
5559   // Paranoia to ensure our locked and lock-free paths interact
5560   // correctly with each other.
5561   OrderAccess::fence();
5562   return ret;
5563 }
5564 
5565 void os::PlatformEvent::unpark() {
5566   // Transitions for _Event:
5567   //    0 :=> 1
5568   //    1 :=> 1
5569   //   -1 :=> either 0 or 1; must signal target thread
5570   //          That is, we can safely transition _Event from -1 to either
5571   //          0 or 1.
5572   // See also: "Semaphores in Plan 9" by Mullender & Cox
5573   //
5574   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5575   // that it will take two back-to-back park() calls for the owning
5576   // thread to block. This has the benefit of forcing a spurious return
5577   // from the first park() call after an unpark() call which will help
5578   // shake out uses of park() and unpark() without condition variables.
5579 
5580   if (Atomic::xchg(1, &_Event) >= 0) return;
5581 
5582   // Wait for the thread associated with the event to vacate
5583   int status = pthread_mutex_lock(_mutex);
5584   assert_status(status == 0, status, "mutex_lock");
5585   int AnyWaiters = _nParked;
5586   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5587   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5588     AnyWaiters = 0;
5589     pthread_cond_signal(_cond);
5590   }
5591   status = pthread_mutex_unlock(_mutex);
5592   assert_status(status == 0, status, "mutex_unlock");
5593   if (AnyWaiters != 0) {
5594     status = pthread_cond_signal(_cond);
5595     assert_status(status == 0, status, "cond_signal");
5596   }
5597 
5598   // Note that we signal() _after dropping the lock for "immortal" Events.
5599   // This is safe and avoids a common class of  futile wakeups.  In rare
5600   // circumstances this can cause a thread to return prematurely from
5601   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5602   // simply re-test the condition and re-park itself.
5603 }
5604 
5605 
5606 // JSR166
5607 // -------------------------------------------------------
5608 
5609 /*
5610  * The solaris and linux implementations of park/unpark are fairly
5611  * conservative for now, but can be improved. They currently use a
5612  * mutex/condvar pair, plus a a count.
5613  * Park decrements count if > 0, else does a condvar wait.  Unpark
5614  * sets count to 1 and signals condvar.  Only one thread ever waits
5615  * on the condvar. Contention seen when trying to park implies that someone
5616  * is unparking you, so don't wait. And spurious returns are fine, so there
5617  * is no need to track notifications.
5618  */
5619 
5620 /*
5621  * This code is common to linux and solaris and will be moved to a
5622  * common place in dolphin.
5623  *
5624  * The passed in time value is either a relative time in nanoseconds
5625  * or an absolute time in milliseconds. Either way it has to be unpacked
5626  * into suitable seconds and nanoseconds components and stored in the
5627  * given timespec structure.
5628  * Given time is a 64-bit value and the time_t used in the timespec is only
5629  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5630  * overflow if times way in the future are given. Further on Solaris versions
5631  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5632  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5633  * As it will be 28 years before "now + 100000000" will overflow we can
5634  * ignore overflow and just impose a hard-limit on seconds using the value
5635  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5636  * years from "now".
5637  */
5638 
5639 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5640   assert(time > 0, "convertTime");
5641   time_t max_secs = 0;
5642 
5643   if (!os::supports_monotonic_clock() || isAbsolute) {
5644     struct timeval now;
5645     int status = gettimeofday(&now, NULL);
5646     assert(status == 0, "gettimeofday");
5647 
5648     max_secs = now.tv_sec + MAX_SECS;
5649 
5650     if (isAbsolute) {
5651       jlong secs = time / 1000;
5652       if (secs > max_secs) {
5653         absTime->tv_sec = max_secs;
5654       } else {
5655         absTime->tv_sec = secs;
5656       }
5657       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5658     } else {
5659       jlong secs = time / NANOSECS_PER_SEC;
5660       if (secs >= MAX_SECS) {
5661         absTime->tv_sec = max_secs;
5662         absTime->tv_nsec = 0;
5663       } else {
5664         absTime->tv_sec = now.tv_sec + secs;
5665         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5666         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5667           absTime->tv_nsec -= NANOSECS_PER_SEC;
5668           ++absTime->tv_sec; // note: this must be <= max_secs
5669         }
5670       }
5671     }
5672   } else {
5673     // must be relative using monotonic clock
5674     struct timespec now;
5675     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5676     assert_status(status == 0, status, "clock_gettime");
5677     max_secs = now.tv_sec + MAX_SECS;
5678     jlong secs = time / NANOSECS_PER_SEC;
5679     if (secs >= MAX_SECS) {
5680       absTime->tv_sec = max_secs;
5681       absTime->tv_nsec = 0;
5682     } else {
5683       absTime->tv_sec = now.tv_sec + secs;
5684       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5685       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5686         absTime->tv_nsec -= NANOSECS_PER_SEC;
5687         ++absTime->tv_sec; // note: this must be <= max_secs
5688       }
5689     }
5690   }
5691   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5692   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5693   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5694   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5695 }
5696 
5697 void Parker::park(bool isAbsolute, jlong time) {
5698   // Ideally we'd do something useful while spinning, such
5699   // as calling unpackTime().
5700 
5701   // Optional fast-path check:
5702   // Return immediately if a permit is available.
5703   // We depend on Atomic::xchg() having full barrier semantics
5704   // since we are doing a lock-free update to _counter.
5705   if (Atomic::xchg(0, &_counter) > 0) return;
5706 
5707   Thread* thread = Thread::current();
5708   assert(thread->is_Java_thread(), "Must be JavaThread");
5709   JavaThread *jt = (JavaThread *)thread;
5710 
5711   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5712   // Check interrupt before trying to wait
5713   if (Thread::is_interrupted(thread, false)) {
5714     return;
5715   }
5716 
5717   // Next, demultiplex/decode time arguments
5718   timespec absTime;
5719   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5720     return;
5721   }
5722   if (time > 0) {
5723     unpackTime(&absTime, isAbsolute, time);
5724   }
5725 
5726 
5727   // Enter safepoint region
5728   // Beware of deadlocks such as 6317397.
5729   // The per-thread Parker:: mutex is a classic leaf-lock.
5730   // In particular a thread must never block on the Threads_lock while
5731   // holding the Parker:: mutex.  If safepoints are pending both the
5732   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5733   ThreadBlockInVM tbivm(jt);
5734 
5735   // Don't wait if cannot get lock since interference arises from
5736   // unblocking.  Also. check interrupt before trying wait
5737   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5738     return;
5739   }
5740 
5741   int status;
5742   if (_counter > 0)  { // no wait needed
5743     _counter = 0;
5744     status = pthread_mutex_unlock(_mutex);
5745     assert(status == 0, "invariant");
5746     // Paranoia to ensure our locked and lock-free paths interact
5747     // correctly with each other and Java-level accesses.
5748     OrderAccess::fence();
5749     return;
5750   }
5751 
5752 #ifdef ASSERT
5753   // Don't catch signals while blocked; let the running threads have the signals.
5754   // (This allows a debugger to break into the running thread.)
5755   sigset_t oldsigs;
5756   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5757   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5758 #endif
5759 
5760   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5761   jt->set_suspend_equivalent();
5762   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5763 
5764   assert(_cur_index == -1, "invariant");
5765   if (time == 0) {
5766     _cur_index = REL_INDEX; // arbitrary choice when not timed
5767     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5768   } else {
5769     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5770     status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5771     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5772       pthread_cond_destroy(&_cond[_cur_index]);
5773       pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5774     }
5775   }
5776   _cur_index = -1;
5777   assert_status(status == 0 || status == EINTR ||
5778                 status == ETIME || status == ETIMEDOUT,
5779                 status, "cond_timedwait");
5780 
5781 #ifdef ASSERT
5782   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5783 #endif
5784 
5785   _counter = 0;
5786   status = pthread_mutex_unlock(_mutex);
5787   assert_status(status == 0, status, "invariant");
5788   // Paranoia to ensure our locked and lock-free paths interact
5789   // correctly with each other and Java-level accesses.
5790   OrderAccess::fence();
5791 
5792   // If externally suspended while waiting, re-suspend
5793   if (jt->handle_special_suspend_equivalent_condition()) {
5794     jt->java_suspend_self();
5795   }
5796 }
5797 
5798 void Parker::unpark() {
5799   int status = pthread_mutex_lock(_mutex);
5800   assert(status == 0, "invariant");
5801   const int s = _counter;
5802   _counter = 1;
5803   if (s < 1) {
5804     // thread might be parked
5805     if (_cur_index != -1) {
5806       // thread is definitely parked
5807       if (WorkAroundNPTLTimedWaitHang) {
5808         status = pthread_cond_signal (&_cond[_cur_index]);
5809         assert(status == 0, "invariant");
5810         status = pthread_mutex_unlock(_mutex);
5811         assert(status == 0, "invariant");
5812       } else {
5813         status = pthread_mutex_unlock(_mutex);
5814         assert(status == 0, "invariant");
5815         status = pthread_cond_signal (&_cond[_cur_index]);
5816         assert(status == 0, "invariant");
5817       }
5818     } else {
5819       pthread_mutex_unlock(_mutex);
5820       assert(status == 0, "invariant");
5821     }
5822   } else {
5823     pthread_mutex_unlock(_mutex);
5824     assert(status == 0, "invariant");
5825   }
5826 }
5827 
5828 
5829 extern char** environ;
5830 
5831 #ifndef __NR_fork
5832 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5833 #endif
5834 
5835 #ifndef __NR_execve
5836 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5837 #endif
5838 
5839 // Run the specified command in a separate process. Return its exit value,
5840 // or -1 on failure (e.g. can't fork a new process).
5841 // Unlike system(), this function can be called from signal handler. It
5842 // doesn't block SIGINT et al.
5843 int os::fork_and_exec(char* cmd) {
5844   const char * argv[4] = {"sh", "-c", cmd, NULL};
5845 
5846   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5847   // pthread_atfork handlers and reset pthread library. All we need is a
5848   // separate process to execve. Make a direct syscall to fork process.
5849   // On IA64 there's no fork syscall, we have to use fork() and hope for
5850   // the best...
5851   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5852               IA64_ONLY(fork();)
5853 
5854   if (pid < 0) {
5855     // fork failed
5856     return -1;
5857 
5858   } else if (pid == 0) {
5859     // child process
5860 
5861     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5862     // first to kill every thread on the thread list. Because this list is
5863     // not reset by fork() (see notes above), execve() will instead kill
5864     // every thread in the parent process. We know this is the only thread
5865     // in the new process, so make a system call directly.
5866     // IA64 should use normal execve() from glibc to match the glibc fork()
5867     // above.
5868     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5869     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5870 
5871     // execve failed
5872     _exit(-1);
5873 
5874   } else  {
5875     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5876     // care about the actual exit code, for now.
5877 
5878     int status;
5879 
5880     // Wait for the child process to exit.  This returns immediately if
5881     // the child has already exited. */
5882     while (waitpid(pid, &status, 0) < 0) {
5883         switch (errno) {
5884         case ECHILD: return 0;
5885         case EINTR: break;
5886         default: return -1;
5887         }
5888     }
5889 
5890     if (WIFEXITED(status)) {
5891        // The child exited normally; get its exit code.
5892        return WEXITSTATUS(status);
5893     } else if (WIFSIGNALED(status)) {
5894        // The child exited because of a signal
5895        // The best value to return is 0x80 + signal number,
5896        // because that is what all Unix shells do, and because
5897        // it allows callers to distinguish between process exit and
5898        // process death by signal.
5899        return 0x80 + WTERMSIG(status);
5900     } else {
5901        // Unknown exit code; pass it through
5902        return status;
5903     }
5904   }
5905 }
5906 
5907 // is_headless_jre()
5908 //
5909 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5910 // in order to report if we are running in a headless jre
5911 //
5912 // Since JDK8 xawt/libmawt.so was moved into the same directory
5913 // as libawt.so, and renamed libawt_xawt.so
5914 //
5915 bool os::is_headless_jre() {
5916     struct stat statbuf;
5917     char buf[MAXPATHLEN];
5918     char libmawtpath[MAXPATHLEN];
5919     const char *xawtstr  = "/xawt/libmawt.so";
5920     const char *new_xawtstr = "/libawt_xawt.so";
5921     char *p;
5922 
5923     // Get path to libjvm.so
5924     os::jvm_path(buf, sizeof(buf));
5925 
5926     // Get rid of libjvm.so
5927     p = strrchr(buf, '/');
5928     if (p == NULL) return false;
5929     else *p = '\0';
5930 
5931     // Get rid of client or server
5932     p = strrchr(buf, '/');
5933     if (p == NULL) return false;
5934     else *p = '\0';
5935 
5936     // check xawt/libmawt.so
5937     strcpy(libmawtpath, buf);
5938     strcat(libmawtpath, xawtstr);
5939     if (::stat(libmawtpath, &statbuf) == 0) return false;
5940 
5941     // check libawt_xawt.so
5942     strcpy(libmawtpath, buf);
5943     strcat(libmawtpath, new_xawtstr);
5944     if (::stat(libmawtpath, &statbuf) == 0) return false;
5945 
5946     return true;
5947 }
5948 
5949 // Get the default path to the core file
5950 // Returns the length of the string
5951 int os::get_core_path(char* buffer, size_t bufferSize) {
5952   const char* p = get_current_directory(buffer, bufferSize);
5953 
5954   if (p == NULL) {
5955     assert(p != NULL, "failed to get current directory");
5956     return 0;
5957   }
5958 
5959   return strlen(buffer);
5960 }
5961 
5962 #ifdef JAVASE_EMBEDDED
5963 //
5964 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5965 //
5966 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5967 
5968 // ctor
5969 //
5970 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5971   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5972   _fd = fd;
5973 
5974   if (os::create_thread(this, os::os_thread)) {
5975     _memnotify_thread = this;
5976     os::set_priority(this, NearMaxPriority);
5977     os::start_thread(this);
5978   }
5979 }
5980 
5981 // Where all the work gets done
5982 //
5983 void MemNotifyThread::run() {
5984   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5985 
5986   // Set up the select arguments
5987   fd_set rfds;
5988   if (_fd != -1) {
5989     FD_ZERO(&rfds);
5990     FD_SET(_fd, &rfds);
5991   }
5992 
5993   // Now wait for the mem_notify device to wake up
5994   while (1) {
5995     // Wait for the mem_notify device to signal us..
5996     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5997     if (rc == -1) {
5998       perror("select!\n");
5999       break;
6000     } else if (rc) {
6001       //ssize_t free_before = os::available_memory();
6002       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6003 
6004       // The kernel is telling us there is not much memory left...
6005       // try to do something about that
6006 
6007       // If we are not already in a GC, try one.
6008       if (!Universe::heap()->is_gc_active()) {
6009         Universe::heap()->collect(GCCause::_allocation_failure);
6010 
6011         //ssize_t free_after = os::available_memory();
6012         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6013         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6014       }
6015       // We might want to do something like the following if we find the GC's are not helping...
6016       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6017     }
6018   }
6019 }
6020 
6021 //
6022 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6023 //
6024 void MemNotifyThread::start() {
6025   int    fd;
6026   fd = open ("/dev/mem_notify", O_RDONLY, 0);
6027   if (fd < 0) {
6028       return;
6029   }
6030 
6031   if (memnotify_thread() == NULL) {
6032     new MemNotifyThread(fd);
6033   }
6034 }
6035 
6036 #endif // JAVASE_EMBEDDED
6037 
6038 
6039 /////////////// Unit tests ///////////////
6040 
6041 #ifndef PRODUCT
6042 
6043 #define test_log(...) \
6044   do {\
6045     if (VerboseInternalVMTests) { \
6046       tty->print_cr(__VA_ARGS__); \
6047       tty->flush(); \
6048     }\
6049   } while (false)
6050 
6051 class TestReserveMemorySpecial : AllStatic {
6052  public:
6053   static void small_page_write(void* addr, size_t size) {
6054     size_t page_size = os::vm_page_size();
6055 
6056     char* end = (char*)addr + size;
6057     for (char* p = (char*)addr; p < end; p += page_size) {
6058       *p = 1;
6059     }
6060   }
6061 
6062   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6063     if (!UseHugeTLBFS) {
6064       return;
6065     }
6066 
6067     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6068 
6069     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6070 
6071     if (addr != NULL) {
6072       small_page_write(addr, size);
6073 
6074       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6075     }
6076   }
6077 
6078   static void test_reserve_memory_special_huge_tlbfs_only() {
6079     if (!UseHugeTLBFS) {
6080       return;
6081     }
6082 
6083     size_t lp = os::large_page_size();
6084 
6085     for (size_t size = lp; size <= lp * 10; size += lp) {
6086       test_reserve_memory_special_huge_tlbfs_only(size);
6087     }
6088   }
6089 
6090   static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6091     if (!UseHugeTLBFS) {
6092         return;
6093     }
6094 
6095     test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6096         size, alignment);
6097 
6098     assert(size >= os::large_page_size(), "Incorrect input to test");
6099 
6100     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6101 
6102     if (addr != NULL) {
6103       small_page_write(addr, size);
6104 
6105       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6106     }
6107   }
6108 
6109   static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6110     size_t lp = os::large_page_size();
6111     size_t ag = os::vm_allocation_granularity();
6112 
6113     for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6114       test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6115     }
6116   }
6117 
6118   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6119     size_t lp = os::large_page_size();
6120     size_t ag = os::vm_allocation_granularity();
6121 
6122     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6123     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6124     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6125     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6126     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6127     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6128     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6129     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6130     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6131   }
6132 
6133   static void test_reserve_memory_special_huge_tlbfs() {
6134     if (!UseHugeTLBFS) {
6135       return;
6136     }
6137 
6138     test_reserve_memory_special_huge_tlbfs_only();
6139     test_reserve_memory_special_huge_tlbfs_mixed();
6140   }
6141 
6142   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6143     if (!UseSHM) {
6144       return;
6145     }
6146 
6147     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6148 
6149     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6150 
6151     if (addr != NULL) {
6152       assert(is_ptr_aligned(addr, alignment), "Check");
6153       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6154 
6155       small_page_write(addr, size);
6156 
6157       os::Linux::release_memory_special_shm(addr, size);
6158     }
6159   }
6160 
6161   static void test_reserve_memory_special_shm() {
6162     size_t lp = os::large_page_size();
6163     size_t ag = os::vm_allocation_granularity();
6164 
6165     for (size_t size = ag; size < lp * 3; size += ag) {
6166       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6167         test_reserve_memory_special_shm(size, alignment);
6168       }
6169     }
6170   }
6171 
6172   static void test() {
6173     test_reserve_memory_special_huge_tlbfs();
6174     test_reserve_memory_special_shm();
6175   }
6176 };
6177 
6178 void TestReserveMemorySpecial_test() {
6179   TestReserveMemorySpecial::test();
6180 }
6181 
6182 #endif