1 /*
   2  * Copyright (c) 1999, 2014, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "memory/allocation.inline.hpp"
  36 #include "memory/filemap.hpp"
  37 #include "mutex_linux.inline.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_share_linux.hpp"
  40 #include "prims/jniFastGetField.hpp"
  41 #include "prims/jvm.h"
  42 #include "prims/jvm_misc.hpp"
  43 #include "runtime/arguments.hpp"
  44 #include "runtime/extendedPC.hpp"
  45 #include "runtime/globals.hpp"
  46 #include "runtime/interfaceSupport.hpp"
  47 #include "runtime/init.hpp"
  48 #include "runtime/java.hpp"
  49 #include "runtime/javaCalls.hpp"
  50 #include "runtime/mutexLocker.hpp"
  51 #include "runtime/objectMonitor.hpp"
  52 #include "runtime/orderAccess.inline.hpp"
  53 #include "runtime/osThread.hpp"
  54 #include "runtime/perfMemory.hpp"
  55 #include "runtime/sharedRuntime.hpp"
  56 #include "runtime/statSampler.hpp"
  57 #include "runtime/stubRoutines.hpp"
  58 #include "runtime/thread.inline.hpp"
  59 #include "runtime/threadCritical.hpp"
  60 #include "runtime/timer.hpp"
  61 #include "services/attachListener.hpp"
  62 #include "services/memTracker.hpp"
  63 #include "services/runtimeService.hpp"
  64 #include "utilities/decoder.hpp"
  65 #include "utilities/defaultStream.hpp"
  66 #include "utilities/events.hpp"
  67 #include "utilities/elfFile.hpp"
  68 #include "utilities/growableArray.hpp"
  69 #include "utilities/vmError.hpp"
  70 
  71 // put OS-includes here
  72 # include <sys/types.h>
  73 # include <sys/mman.h>
  74 # include <sys/stat.h>
  75 # include <sys/select.h>
  76 # include <pthread.h>
  77 # include <signal.h>
  78 # include <errno.h>
  79 # include <dlfcn.h>
  80 # include <stdio.h>
  81 # include <unistd.h>
  82 # include <sys/resource.h>
  83 # include <pthread.h>
  84 # include <sys/stat.h>
  85 # include <sys/time.h>
  86 # include <sys/times.h>
  87 # include <sys/utsname.h>
  88 # include <sys/socket.h>
  89 # include <sys/wait.h>
  90 # include <pwd.h>
  91 # include <poll.h>
  92 # include <semaphore.h>
  93 # include <fcntl.h>
  94 # include <string.h>
  95 # include <syscall.h>
  96 # include <sys/sysinfo.h>
  97 # include <gnu/libc-version.h>
  98 # include <sys/ipc.h>
  99 # include <sys/shm.h>
 100 # include <link.h>
 101 # include <stdint.h>
 102 # include <inttypes.h>
 103 # include <sys/ioctl.h>
 104 
 105 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
 106 
 107 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 108 // getrusage() is prepared to handle the associated failure.
 109 #ifndef RUSAGE_THREAD
 110 #define RUSAGE_THREAD   (1)               /* only the calling thread */
 111 #endif
 112 
 113 #define MAX_PATH    (2 * K)
 114 
 115 #define MAX_SECS 100000000
 116 
 117 // for timer info max values which include all bits
 118 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 119 
 120 #define LARGEPAGES_BIT (1 << 6)
 121 ////////////////////////////////////////////////////////////////////////////////
 122 // global variables
 123 julong os::Linux::_physical_memory = 0;
 124 
 125 address   os::Linux::_initial_thread_stack_bottom = NULL;
 126 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 127 
 128 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 129 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 130 Mutex* os::Linux::_createThread_lock = NULL;
 131 pthread_t os::Linux::_main_thread;
 132 int os::Linux::_page_size = -1;
 133 const int os::Linux::_vm_default_page_size = (8 * K);
 134 bool os::Linux::_is_floating_stack = false;
 135 bool os::Linux::_is_NPTL = false;
 136 bool os::Linux::_supports_fast_thread_cpu_time = false;
 137 const char * os::Linux::_glibc_version = NULL;
 138 const char * os::Linux::_libpthread_version = NULL;
 139 pthread_condattr_t os::Linux::_condattr[1];
 140 
 141 static jlong initial_time_count=0;
 142 
 143 static int clock_tics_per_sec = 100;
 144 
 145 // For diagnostics to print a message once. see run_periodic_checks
 146 static sigset_t check_signal_done;
 147 static bool check_signals = true;
 148 
 149 static pid_t _initial_pid = 0;
 150 
 151 /* Signal number used to suspend/resume a thread */
 152 
 153 /* do not use any signal number less than SIGSEGV, see 4355769 */
 154 static int SR_signum = SIGUSR2;
 155 sigset_t SR_sigset;
 156 
 157 /* Used to protect dlsym() calls */
 158 static pthread_mutex_t dl_mutex;
 159 
 160 // Declarations
 161 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 162 
 163 #ifdef JAVASE_EMBEDDED
 164 class MemNotifyThread: public Thread {
 165   friend class VMStructs;
 166  public:
 167   virtual void run();
 168 
 169  private:
 170   static MemNotifyThread* _memnotify_thread;
 171   int _fd;
 172 
 173  public:
 174 
 175   // Constructor
 176   MemNotifyThread(int fd);
 177 
 178   // Tester
 179   bool is_memnotify_thread() const { return true; }
 180 
 181   // Printing
 182   char* name() const { return (char*)"Linux MemNotify Thread"; }
 183 
 184   // Returns the single instance of the MemNotifyThread
 185   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 186 
 187   // Create and start the single instance of MemNotifyThread
 188   static void start();
 189 };
 190 #endif // JAVASE_EMBEDDED
 191 
 192 // utility functions
 193 
 194 static int SR_initialize();
 195 
 196 julong os::available_memory() {
 197   return Linux::available_memory();
 198 }
 199 
 200 julong os::Linux::available_memory() {
 201   // values in struct sysinfo are "unsigned long"
 202   struct sysinfo si;
 203   sysinfo(&si);
 204 
 205   return (julong)si.freeram * si.mem_unit;
 206 }
 207 
 208 julong os::physical_memory() {
 209   return Linux::physical_memory();
 210 }
 211 
 212 ////////////////////////////////////////////////////////////////////////////////
 213 // environment support
 214 
 215 bool os::getenv(const char* name, char* buf, int len) {
 216   const char* val = ::getenv(name);
 217   if (val != NULL && strlen(val) < (size_t)len) {
 218     strcpy(buf, val);
 219     return true;
 220   }
 221   if (len > 0) buf[0] = 0;  // return a null string
 222   return false;
 223 }
 224 
 225 
 226 // Return true if user is running as root.
 227 
 228 bool os::have_special_privileges() {
 229   static bool init = false;
 230   static bool privileges = false;
 231   if (!init) {
 232     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 233     init = true;
 234   }
 235   return privileges;
 236 }
 237 
 238 
 239 #ifndef SYS_gettid
 240 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 241 #ifdef __ia64__
 242 #define SYS_gettid 1105
 243 #elif __i386__
 244 #define SYS_gettid 224
 245 #elif __amd64__
 246 #define SYS_gettid 186
 247 #elif __sparc__
 248 #define SYS_gettid 143
 249 #else
 250 #error define gettid for the arch
 251 #endif
 252 #endif
 253 
 254 // Cpu architecture string
 255 #if   defined(ZERO)
 256 static char cpu_arch[] = ZERO_LIBARCH;
 257 #elif defined(IA64)
 258 static char cpu_arch[] = "ia64";
 259 #elif defined(IA32)
 260 static char cpu_arch[] = "i386";
 261 #elif defined(AMD64)
 262 static char cpu_arch[] = "amd64";
 263 #elif defined(ARM)
 264 static char cpu_arch[] = "arm";
 265 #elif defined(PPC32)
 266 static char cpu_arch[] = "ppc";
 267 #elif defined(PPC64)
 268 static char cpu_arch[] = "ppc64";
 269 #elif defined(SPARC)
 270 #  ifdef _LP64
 271 static char cpu_arch[] = "sparcv9";
 272 #  else
 273 static char cpu_arch[] = "sparc";
 274 #  endif
 275 #else
 276 #error Add appropriate cpu_arch setting
 277 #endif
 278 
 279 
 280 // pid_t gettid()
 281 //
 282 // Returns the kernel thread id of the currently running thread. Kernel
 283 // thread id is used to access /proc.
 284 //
 285 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 286 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 287 //
 288 pid_t os::Linux::gettid() {
 289   int rslt = syscall(SYS_gettid);
 290   if (rslt == -1) {
 291      // old kernel, no NPTL support
 292      return getpid();
 293   } else {
 294      return (pid_t)rslt;
 295   }
 296 }
 297 
 298 // Most versions of linux have a bug where the number of processors are
 299 // determined by looking at the /proc file system.  In a chroot environment,
 300 // the system call returns 1.  This causes the VM to act as if it is
 301 // a single processor and elide locking (see is_MP() call).
 302 static bool unsafe_chroot_detected = false;
 303 static const char *unstable_chroot_error = "/proc file system not found.\n"
 304                      "Java may be unstable running multithreaded in a chroot "
 305                      "environment on Linux when /proc filesystem is not mounted.";
 306 
 307 void os::Linux::initialize_system_info() {
 308   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 309   if (processor_count() == 1) {
 310     pid_t pid = os::Linux::gettid();
 311     char fname[32];
 312     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 313     FILE *fp = fopen(fname, "r");
 314     if (fp == NULL) {
 315       unsafe_chroot_detected = true;
 316     } else {
 317       fclose(fp);
 318     }
 319   }
 320   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 321   assert(processor_count() > 0, "linux error");
 322 }
 323 
 324 void os::init_system_properties_values() {
 325   // The next steps are taken in the product version:
 326   //
 327   // Obtain the JAVA_HOME value from the location of libjvm.so.
 328   // This library should be located at:
 329   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 330   //
 331   // If "/jre/lib/" appears at the right place in the path, then we
 332   // assume libjvm.so is installed in a JDK and we use this path.
 333   //
 334   // Otherwise exit with message: "Could not create the Java virtual machine."
 335   //
 336   // The following extra steps are taken in the debugging version:
 337   //
 338   // If "/jre/lib/" does NOT appear at the right place in the path
 339   // instead of exit check for $JAVA_HOME environment variable.
 340   //
 341   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 342   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 343   // it looks like libjvm.so is installed there
 344   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 345   //
 346   // Otherwise exit.
 347   //
 348   // Important note: if the location of libjvm.so changes this
 349   // code needs to be changed accordingly.
 350 
 351 // See ld(1):
 352 //      The linker uses the following search paths to locate required
 353 //      shared libraries:
 354 //        1: ...
 355 //        ...
 356 //        7: The default directories, normally /lib and /usr/lib.
 357 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 358 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 359 #else
 360 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 361 #endif
 362 
 363 // Base path of extensions installed on the system.
 364 #define SYS_EXT_DIR     "/usr/java/packages"
 365 #define EXTENSIONS_DIR  "/lib/ext"
 366 #define ENDORSED_DIR    "/lib/endorsed"
 367 
 368   // Buffer that fits several sprintfs.
 369   // Note that the space for the colon and the trailing null are provided
 370   // by the nulls included by the sizeof operator.
 371   const size_t bufsize =
 372     MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
 373          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
 374          (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
 375   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 376 
 377   // sysclasspath, java_home, dll_dir
 378   {
 379     char *pslash;
 380     os::jvm_path(buf, bufsize);
 381 
 382     // Found the full path to libjvm.so.
 383     // Now cut the path to <java_home>/jre if we can.
 384     *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
 385     pslash = strrchr(buf, '/');
 386     if (pslash != NULL) {
 387       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 388     }
 389     Arguments::set_dll_dir(buf);
 390 
 391     if (pslash != NULL) {
 392       pslash = strrchr(buf, '/');
 393       if (pslash != NULL) {
 394         *pslash = '\0';          // Get rid of /<arch>.
 395         pslash = strrchr(buf, '/');
 396         if (pslash != NULL) {
 397           *pslash = '\0';        // Get rid of /lib.
 398         }
 399       }
 400     }
 401     Arguments::set_java_home(buf);
 402     set_boot_path('/', ':');
 403   }
 404 
 405   // Where to look for native libraries.
 406   //
 407   // Note: Due to a legacy implementation, most of the library path
 408   // is set in the launcher. This was to accomodate linking restrictions
 409   // on legacy Linux implementations (which are no longer supported).
 410   // Eventually, all the library path setting will be done here.
 411   //
 412   // However, to prevent the proliferation of improperly built native
 413   // libraries, the new path component /usr/java/packages is added here.
 414   // Eventually, all the library path setting will be done here.
 415   {
 416     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 417     // should always exist (until the legacy problem cited above is
 418     // addressed).
 419     const char *v = ::getenv("LD_LIBRARY_PATH");
 420     const char *v_colon = ":";
 421     if (v == NULL) { v = ""; v_colon = ""; }
 422     // That's +1 for the colon and +1 for the trailing '\0'.
 423     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 424                                                      strlen(v) + 1 +
 425                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 426                                                      mtInternal);
 427     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 428     Arguments::set_library_path(ld_library_path);
 429     FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
 430   }
 431 
 432   // Extensions directories.
 433   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 434   Arguments::set_ext_dirs(buf);
 435 
 436   // Endorsed standards default directory.
 437   sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 438   Arguments::set_endorsed_dirs(buf);
 439 
 440   FREE_C_HEAP_ARRAY(char, buf, mtInternal);
 441 
 442 #undef DEFAULT_LIBPATH
 443 #undef SYS_EXT_DIR
 444 #undef EXTENSIONS_DIR
 445 #undef ENDORSED_DIR
 446 }
 447 
 448 ////////////////////////////////////////////////////////////////////////////////
 449 // breakpoint support
 450 
 451 void os::breakpoint() {
 452   BREAKPOINT;
 453 }
 454 
 455 extern "C" void breakpoint() {
 456   // use debugger to set breakpoint here
 457 }
 458 
 459 ////////////////////////////////////////////////////////////////////////////////
 460 // signal support
 461 
 462 debug_only(static bool signal_sets_initialized = false);
 463 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 464 
 465 bool os::Linux::is_sig_ignored(int sig) {
 466       struct sigaction oact;
 467       sigaction(sig, (struct sigaction*)NULL, &oact);
 468       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 469                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 470       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 471            return true;
 472       else
 473            return false;
 474 }
 475 
 476 void os::Linux::signal_sets_init() {
 477   // Should also have an assertion stating we are still single-threaded.
 478   assert(!signal_sets_initialized, "Already initialized");
 479   // Fill in signals that are necessarily unblocked for all threads in
 480   // the VM. Currently, we unblock the following signals:
 481   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 482   //                         by -Xrs (=ReduceSignalUsage));
 483   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 484   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 485   // the dispositions or masks wrt these signals.
 486   // Programs embedding the VM that want to use the above signals for their
 487   // own purposes must, at this time, use the "-Xrs" option to prevent
 488   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 489   // (See bug 4345157, and other related bugs).
 490   // In reality, though, unblocking these signals is really a nop, since
 491   // these signals are not blocked by default.
 492   sigemptyset(&unblocked_sigs);
 493   sigemptyset(&allowdebug_blocked_sigs);
 494   sigaddset(&unblocked_sigs, SIGILL);
 495   sigaddset(&unblocked_sigs, SIGSEGV);
 496   sigaddset(&unblocked_sigs, SIGBUS);
 497   sigaddset(&unblocked_sigs, SIGFPE);
 498 #if defined(PPC64)
 499   sigaddset(&unblocked_sigs, SIGTRAP);
 500 #endif
 501   sigaddset(&unblocked_sigs, SR_signum);
 502 
 503   if (!ReduceSignalUsage) {
 504    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 505       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 506       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 507    }
 508    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 509       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 510       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 511    }
 512    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 513       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 514       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 515    }
 516   }
 517   // Fill in signals that are blocked by all but the VM thread.
 518   sigemptyset(&vm_sigs);
 519   if (!ReduceSignalUsage)
 520     sigaddset(&vm_sigs, BREAK_SIGNAL);
 521   debug_only(signal_sets_initialized = true);
 522 
 523 }
 524 
 525 // These are signals that are unblocked while a thread is running Java.
 526 // (For some reason, they get blocked by default.)
 527 sigset_t* os::Linux::unblocked_signals() {
 528   assert(signal_sets_initialized, "Not initialized");
 529   return &unblocked_sigs;
 530 }
 531 
 532 // These are the signals that are blocked while a (non-VM) thread is
 533 // running Java. Only the VM thread handles these signals.
 534 sigset_t* os::Linux::vm_signals() {
 535   assert(signal_sets_initialized, "Not initialized");
 536   return &vm_sigs;
 537 }
 538 
 539 // These are signals that are blocked during cond_wait to allow debugger in
 540 sigset_t* os::Linux::allowdebug_blocked_signals() {
 541   assert(signal_sets_initialized, "Not initialized");
 542   return &allowdebug_blocked_sigs;
 543 }
 544 
 545 void os::Linux::hotspot_sigmask(Thread* thread) {
 546 
 547   //Save caller's signal mask before setting VM signal mask
 548   sigset_t caller_sigmask;
 549   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 550 
 551   OSThread* osthread = thread->osthread();
 552   osthread->set_caller_sigmask(caller_sigmask);
 553 
 554   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 555 
 556   if (!ReduceSignalUsage) {
 557     if (thread->is_VM_thread()) {
 558       // Only the VM thread handles BREAK_SIGNAL ...
 559       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 560     } else {
 561       // ... all other threads block BREAK_SIGNAL
 562       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 563     }
 564   }
 565 }
 566 
 567 //////////////////////////////////////////////////////////////////////////////
 568 // detecting pthread library
 569 
 570 void os::Linux::libpthread_init() {
 571   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 572   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 573   // generic name for earlier versions.
 574   // Define macros here so we can build HotSpot on old systems.
 575 # ifndef _CS_GNU_LIBC_VERSION
 576 # define _CS_GNU_LIBC_VERSION 2
 577 # endif
 578 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 579 # define _CS_GNU_LIBPTHREAD_VERSION 3
 580 # endif
 581 
 582   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 583   if (n > 0) {
 584      char *str = (char *)malloc(n, mtInternal);
 585      confstr(_CS_GNU_LIBC_VERSION, str, n);
 586      os::Linux::set_glibc_version(str);
 587   } else {
 588      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 589      static char _gnu_libc_version[32];
 590      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 591               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 592      os::Linux::set_glibc_version(_gnu_libc_version);
 593   }
 594 
 595   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 596   if (n > 0) {
 597      char *str = (char *)malloc(n, mtInternal);
 598      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 599      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 600      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 601      // is the case. LinuxThreads has a hard limit on max number of threads.
 602      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 603      // On the other hand, NPTL does not have such a limit, sysconf()
 604      // will return -1 and errno is not changed. Check if it is really NPTL.
 605      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 606          strstr(str, "NPTL") &&
 607          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 608        free(str);
 609        os::Linux::set_libpthread_version("linuxthreads");
 610      } else {
 611        os::Linux::set_libpthread_version(str);
 612      }
 613   } else {
 614     // glibc before 2.3.2 only has LinuxThreads.
 615     os::Linux::set_libpthread_version("linuxthreads");
 616   }
 617 
 618   if (strstr(libpthread_version(), "NPTL")) {
 619      os::Linux::set_is_NPTL();
 620   } else {
 621      os::Linux::set_is_LinuxThreads();
 622   }
 623 
 624   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 625   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 626   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 627      os::Linux::set_is_floating_stack();
 628   }
 629 }
 630 
 631 /////////////////////////////////////////////////////////////////////////////
 632 // thread stack
 633 
 634 // Force Linux kernel to expand current thread stack. If "bottom" is close
 635 // to the stack guard, caller should block all signals.
 636 //
 637 // MAP_GROWSDOWN:
 638 //   A special mmap() flag that is used to implement thread stacks. It tells
 639 //   kernel that the memory region should extend downwards when needed. This
 640 //   allows early versions of LinuxThreads to only mmap the first few pages
 641 //   when creating a new thread. Linux kernel will automatically expand thread
 642 //   stack as needed (on page faults).
 643 //
 644 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 645 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 646 //   region, it's hard to tell if the fault is due to a legitimate stack
 647 //   access or because of reading/writing non-exist memory (e.g. buffer
 648 //   overrun). As a rule, if the fault happens below current stack pointer,
 649 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 650 //   application (see Linux kernel fault.c).
 651 //
 652 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 653 //   stack overflow detection.
 654 //
 655 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 656 //   not use this flag. However, the stack of initial thread is not created
 657 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 658 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 659 //   and then attach the thread to JVM.
 660 //
 661 // To get around the problem and allow stack banging on Linux, we need to
 662 // manually expand thread stack after receiving the SIGSEGV.
 663 //
 664 // There are two ways to expand thread stack to address "bottom", we used
 665 // both of them in JVM before 1.5:
 666 //   1. adjust stack pointer first so that it is below "bottom", and then
 667 //      touch "bottom"
 668 //   2. mmap() the page in question
 669 //
 670 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 671 // if current sp is already near the lower end of page 101, and we need to
 672 // call mmap() to map page 100, it is possible that part of the mmap() frame
 673 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 674 // That will destroy the mmap() frame and cause VM to crash.
 675 //
 676 // The following code works by adjusting sp first, then accessing the "bottom"
 677 // page to force a page fault. Linux kernel will then automatically expand the
 678 // stack mapping.
 679 //
 680 // _expand_stack_to() assumes its frame size is less than page size, which
 681 // should always be true if the function is not inlined.
 682 
 683 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 684 #define NOINLINE
 685 #else
 686 #define NOINLINE __attribute__ ((noinline))
 687 #endif
 688 
 689 static void _expand_stack_to(address bottom) NOINLINE;
 690 
 691 static void _expand_stack_to(address bottom) {
 692   address sp;
 693   size_t size;
 694   volatile char *p;
 695 
 696   // Adjust bottom to point to the largest address within the same page, it
 697   // gives us a one-page buffer if alloca() allocates slightly more memory.
 698   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 699   bottom += os::Linux::page_size() - 1;
 700 
 701   // sp might be slightly above current stack pointer; if that's the case, we
 702   // will alloca() a little more space than necessary, which is OK. Don't use
 703   // os::current_stack_pointer(), as its result can be slightly below current
 704   // stack pointer, causing us to not alloca enough to reach "bottom".
 705   sp = (address)&sp;
 706 
 707   if (sp > bottom) {
 708     size = sp - bottom;
 709     p = (volatile char *)alloca(size);
 710     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 711     p[0] = '\0';
 712   }
 713 }
 714 
 715 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 716   assert(t!=NULL, "just checking");
 717   assert(t->osthread()->expanding_stack(), "expand should be set");
 718   assert(t->stack_base() != NULL, "stack_base was not initialized");
 719 
 720   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 721     sigset_t mask_all, old_sigset;
 722     sigfillset(&mask_all);
 723     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 724     _expand_stack_to(addr);
 725     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 726     return true;
 727   }
 728   return false;
 729 }
 730 
 731 //////////////////////////////////////////////////////////////////////////////
 732 // create new thread
 733 
 734 static address highest_vm_reserved_address();
 735 
 736 // check if it's safe to start a new thread
 737 static bool _thread_safety_check(Thread* thread) {
 738   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 739     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 740     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 741     //   allocated (MAP_FIXED) from high address space. Every thread stack
 742     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 743     //   it to other values if they rebuild LinuxThreads).
 744     //
 745     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 746     // the memory region has already been mmap'ed. That means if we have too
 747     // many threads and/or very large heap, eventually thread stack will
 748     // collide with heap.
 749     //
 750     // Here we try to prevent heap/stack collision by comparing current
 751     // stack bottom with the highest address that has been mmap'ed by JVM
 752     // plus a safety margin for memory maps created by native code.
 753     //
 754     // This feature can be disabled by setting ThreadSafetyMargin to 0
 755     //
 756     if (ThreadSafetyMargin > 0) {
 757       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 758 
 759       // not safe if our stack extends below the safety margin
 760       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 761     } else {
 762       return true;
 763     }
 764   } else {
 765     // Floating stack LinuxThreads or NPTL:
 766     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 767     //   there's not enough space left, pthread_create() will fail. If we come
 768     //   here, that means enough space has been reserved for stack.
 769     return true;
 770   }
 771 }
 772 
 773 // Thread start routine for all newly created threads
 774 static void *java_start(Thread *thread) {
 775   // Try to randomize the cache line index of hot stack frames.
 776   // This helps when threads of the same stack traces evict each other's
 777   // cache lines. The threads can be either from the same JVM instance, or
 778   // from different JVM instances. The benefit is especially true for
 779   // processors with hyperthreading technology.
 780   static int counter = 0;
 781   int pid = os::current_process_id();
 782   alloca(((pid ^ counter++) & 7) * 128);
 783 
 784   ThreadLocalStorage::set_thread(thread);
 785 
 786   OSThread* osthread = thread->osthread();
 787   Monitor* sync = osthread->startThread_lock();
 788 
 789   // non floating stack LinuxThreads needs extra check, see above
 790   if (!_thread_safety_check(thread)) {
 791     // notify parent thread
 792     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 793     osthread->set_state(ZOMBIE);
 794     sync->notify_all();
 795     return NULL;
 796   }
 797 
 798   // thread_id is kernel thread id (similar to Solaris LWP id)
 799   osthread->set_thread_id(os::Linux::gettid());
 800 
 801   if (UseNUMA) {
 802     int lgrp_id = os::numa_get_group_id();
 803     if (lgrp_id != -1) {
 804       thread->set_lgrp_id(lgrp_id);
 805     }
 806   }
 807   // initialize signal mask for this thread
 808   os::Linux::hotspot_sigmask(thread);
 809 
 810   // initialize floating point control register
 811   os::Linux::init_thread_fpu_state();
 812 
 813   // handshaking with parent thread
 814   {
 815     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 816 
 817     // notify parent thread
 818     osthread->set_state(INITIALIZED);
 819     sync->notify_all();
 820 
 821     // wait until os::start_thread()
 822     while (osthread->get_state() == INITIALIZED) {
 823       sync->wait(Mutex::_no_safepoint_check_flag);
 824     }
 825   }
 826 
 827   // call one more level start routine
 828   thread->run();
 829 
 830   return 0;
 831 }
 832 
 833 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 834   assert(thread->osthread() == NULL, "caller responsible");
 835 
 836   // Allocate the OSThread object
 837   OSThread* osthread = new OSThread(NULL, NULL);
 838   if (osthread == NULL) {
 839     return false;
 840   }
 841 
 842   // set the correct thread state
 843   osthread->set_thread_type(thr_type);
 844 
 845   // Initial state is ALLOCATED but not INITIALIZED
 846   osthread->set_state(ALLOCATED);
 847 
 848   thread->set_osthread(osthread);
 849 
 850   // init thread attributes
 851   pthread_attr_t attr;
 852   pthread_attr_init(&attr);
 853   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 854 
 855   // stack size
 856   if (os::Linux::supports_variable_stack_size()) {
 857     // calculate stack size if it's not specified by caller
 858     if (stack_size == 0) {
 859       stack_size = os::Linux::default_stack_size(thr_type);
 860 
 861       switch (thr_type) {
 862       case os::java_thread:
 863         // Java threads use ThreadStackSize which default value can be
 864         // changed with the flag -Xss
 865         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 866         stack_size = JavaThread::stack_size_at_create();
 867         break;
 868       case os::compiler_thread:
 869         if (CompilerThreadStackSize > 0) {
 870           stack_size = (size_t)(CompilerThreadStackSize * K);
 871           break;
 872         } // else fall through:
 873           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 874       case os::vm_thread:
 875       case os::pgc_thread:
 876       case os::cgc_thread:
 877       case os::watcher_thread:
 878         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 879         break;
 880       }
 881     }
 882 
 883     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 884     pthread_attr_setstacksize(&attr, stack_size);
 885   } else {
 886     // let pthread_create() pick the default value.
 887   }
 888 
 889   // glibc guard page
 890   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 891 
 892   ThreadState state;
 893 
 894   {
 895     // Serialize thread creation if we are running with fixed stack LinuxThreads
 896     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 897     if (lock) {
 898       os::Linux::createThread_lock()->lock_without_safepoint_check();
 899     }
 900 
 901     pthread_t tid;
 902     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 903 
 904     pthread_attr_destroy(&attr);
 905 
 906     if (ret != 0) {
 907       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 908         perror("pthread_create()");
 909       }
 910       // Need to clean up stuff we've allocated so far
 911       thread->set_osthread(NULL);
 912       delete osthread;
 913       if (lock) os::Linux::createThread_lock()->unlock();
 914       return false;
 915     }
 916 
 917     // Store pthread info into the OSThread
 918     osthread->set_pthread_id(tid);
 919 
 920     // Wait until child thread is either initialized or aborted
 921     {
 922       Monitor* sync_with_child = osthread->startThread_lock();
 923       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 924       while ((state = osthread->get_state()) == ALLOCATED) {
 925         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 926       }
 927     }
 928 
 929     if (lock) {
 930       os::Linux::createThread_lock()->unlock();
 931     }
 932   }
 933 
 934   // Aborted due to thread limit being reached
 935   if (state == ZOMBIE) {
 936       thread->set_osthread(NULL);
 937       delete osthread;
 938       return false;
 939   }
 940 
 941   // The thread is returned suspended (in state INITIALIZED),
 942   // and is started higher up in the call chain
 943   assert(state == INITIALIZED, "race condition");
 944   return true;
 945 }
 946 
 947 /////////////////////////////////////////////////////////////////////////////
 948 // attach existing thread
 949 
 950 // bootstrap the main thread
 951 bool os::create_main_thread(JavaThread* thread) {
 952   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 953   return create_attached_thread(thread);
 954 }
 955 
 956 bool os::create_attached_thread(JavaThread* thread) {
 957 #ifdef ASSERT
 958     thread->verify_not_published();
 959 #endif
 960 
 961   // Allocate the OSThread object
 962   OSThread* osthread = new OSThread(NULL, NULL);
 963 
 964   if (osthread == NULL) {
 965     return false;
 966   }
 967 
 968   // Store pthread info into the OSThread
 969   osthread->set_thread_id(os::Linux::gettid());
 970   osthread->set_pthread_id(::pthread_self());
 971 
 972   // initialize floating point control register
 973   os::Linux::init_thread_fpu_state();
 974 
 975   // Initial thread state is RUNNABLE
 976   osthread->set_state(RUNNABLE);
 977 
 978   thread->set_osthread(osthread);
 979 
 980   if (UseNUMA) {
 981     int lgrp_id = os::numa_get_group_id();
 982     if (lgrp_id != -1) {
 983       thread->set_lgrp_id(lgrp_id);
 984     }
 985   }
 986 
 987   if (os::Linux::is_initial_thread()) {
 988     // If current thread is initial thread, its stack is mapped on demand,
 989     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 990     // the entire stack region to avoid SEGV in stack banging.
 991     // It is also useful to get around the heap-stack-gap problem on SuSE
 992     // kernel (see 4821821 for details). We first expand stack to the top
 993     // of yellow zone, then enable stack yellow zone (order is significant,
 994     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 995     // is no gap between the last two virtual memory regions.
 996 
 997     JavaThread *jt = (JavaThread *)thread;
 998     address addr = jt->stack_yellow_zone_base();
 999     assert(addr != NULL, "initialization problem?");
1000     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1001 
1002     osthread->set_expanding_stack();
1003     os::Linux::manually_expand_stack(jt, addr);
1004     osthread->clear_expanding_stack();
1005   }
1006 
1007   // initialize signal mask for this thread
1008   // and save the caller's signal mask
1009   os::Linux::hotspot_sigmask(thread);
1010 
1011   return true;
1012 }
1013 
1014 void os::pd_start_thread(Thread* thread) {
1015   OSThread * osthread = thread->osthread();
1016   assert(osthread->get_state() != INITIALIZED, "just checking");
1017   Monitor* sync_with_child = osthread->startThread_lock();
1018   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1019   sync_with_child->notify();
1020 }
1021 
1022 // Free Linux resources related to the OSThread
1023 void os::free_thread(OSThread* osthread) {
1024   assert(osthread != NULL, "osthread not set");
1025 
1026   if (Thread::current()->osthread() == osthread) {
1027     // Restore caller's signal mask
1028     sigset_t sigmask = osthread->caller_sigmask();
1029     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1030    }
1031 
1032   delete osthread;
1033 }
1034 
1035 //////////////////////////////////////////////////////////////////////////////
1036 // thread local storage
1037 
1038 // Restore the thread pointer if the destructor is called. This is in case
1039 // someone from JNI code sets up a destructor with pthread_key_create to run
1040 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1041 // will hang or crash. When detachCurrentThread is called the key will be set
1042 // to null and we will not be called again. If detachCurrentThread is never
1043 // called we could loop forever depending on the pthread implementation.
1044 static void restore_thread_pointer(void* p) {
1045   Thread* thread = (Thread*) p;
1046   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1047 }
1048 
1049 int os::allocate_thread_local_storage() {
1050   pthread_key_t key;
1051   int rslt = pthread_key_create(&key, restore_thread_pointer);
1052   assert(rslt == 0, "cannot allocate thread local storage");
1053   return (int)key;
1054 }
1055 
1056 // Note: This is currently not used by VM, as we don't destroy TLS key
1057 // on VM exit.
1058 void os::free_thread_local_storage(int index) {
1059   int rslt = pthread_key_delete((pthread_key_t)index);
1060   assert(rslt == 0, "invalid index");
1061 }
1062 
1063 void os::thread_local_storage_at_put(int index, void* value) {
1064   int rslt = pthread_setspecific((pthread_key_t)index, value);
1065   assert(rslt == 0, "pthread_setspecific failed");
1066 }
1067 
1068 extern "C" Thread* get_thread() {
1069   return ThreadLocalStorage::thread();
1070 }
1071 
1072 //////////////////////////////////////////////////////////////////////////////
1073 // initial thread
1074 
1075 // Check if current thread is the initial thread, similar to Solaris thr_main.
1076 bool os::Linux::is_initial_thread(void) {
1077   char dummy;
1078   // If called before init complete, thread stack bottom will be null.
1079   // Can be called if fatal error occurs before initialization.
1080   if (initial_thread_stack_bottom() == NULL) return false;
1081   assert(initial_thread_stack_bottom() != NULL &&
1082          initial_thread_stack_size()   != 0,
1083          "os::init did not locate initial thread's stack region");
1084   if ((address)&dummy >= initial_thread_stack_bottom() &&
1085       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1086        return true;
1087   else return false;
1088 }
1089 
1090 // Find the virtual memory area that contains addr
1091 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1092   FILE *fp = fopen("/proc/self/maps", "r");
1093   if (fp) {
1094     address low, high;
1095     while (!feof(fp)) {
1096       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1097         if (low <= addr && addr < high) {
1098            if (vma_low)  *vma_low  = low;
1099            if (vma_high) *vma_high = high;
1100            fclose (fp);
1101            return true;
1102         }
1103       }
1104       for (;;) {
1105         int ch = fgetc(fp);
1106         if (ch == EOF || ch == (int)'\n') break;
1107       }
1108     }
1109     fclose(fp);
1110   }
1111   return false;
1112 }
1113 
1114 // Locate initial thread stack. This special handling of initial thread stack
1115 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1116 // bogus value for initial thread.
1117 void os::Linux::capture_initial_stack(size_t max_size) {
1118   // stack size is the easy part, get it from RLIMIT_STACK
1119   size_t stack_size;
1120   struct rlimit rlim;
1121   getrlimit(RLIMIT_STACK, &rlim);
1122   stack_size = rlim.rlim_cur;
1123 
1124   // 6308388: a bug in ld.so will relocate its own .data section to the
1125   //   lower end of primordial stack; reduce ulimit -s value a little bit
1126   //   so we won't install guard page on ld.so's data section.
1127   stack_size -= 2 * page_size();
1128 
1129   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1130   //   7.1, in both cases we will get 2G in return value.
1131   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1132   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1133   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1134   //   in case other parts in glibc still assumes 2M max stack size.
1135   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1136   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1137   if (stack_size > 2 * K * K IA64_ONLY(*2))
1138       stack_size = 2 * K * K IA64_ONLY(*2);
1139   // Try to figure out where the stack base (top) is. This is harder.
1140   //
1141   // When an application is started, glibc saves the initial stack pointer in
1142   // a global variable "__libc_stack_end", which is then used by system
1143   // libraries. __libc_stack_end should be pretty close to stack top. The
1144   // variable is available since the very early days. However, because it is
1145   // a private interface, it could disappear in the future.
1146   //
1147   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1148   // to __libc_stack_end, it is very close to stack top, but isn't the real
1149   // stack top. Note that /proc may not exist if VM is running as a chroot
1150   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1151   // /proc/<pid>/stat could change in the future (though unlikely).
1152   //
1153   // We try __libc_stack_end first. If that doesn't work, look for
1154   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1155   // as a hint, which should work well in most cases.
1156 
1157   uintptr_t stack_start;
1158 
1159   // try __libc_stack_end first
1160   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1161   if (p && *p) {
1162     stack_start = *p;
1163   } else {
1164     // see if we can get the start_stack field from /proc/self/stat
1165     FILE *fp;
1166     int pid;
1167     char state;
1168     int ppid;
1169     int pgrp;
1170     int session;
1171     int nr;
1172     int tpgrp;
1173     unsigned long flags;
1174     unsigned long minflt;
1175     unsigned long cminflt;
1176     unsigned long majflt;
1177     unsigned long cmajflt;
1178     unsigned long utime;
1179     unsigned long stime;
1180     long cutime;
1181     long cstime;
1182     long prio;
1183     long nice;
1184     long junk;
1185     long it_real;
1186     uintptr_t start;
1187     uintptr_t vsize;
1188     intptr_t rss;
1189     uintptr_t rsslim;
1190     uintptr_t scodes;
1191     uintptr_t ecode;
1192     int i;
1193 
1194     // Figure what the primordial thread stack base is. Code is inspired
1195     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1196     // followed by command name surrounded by parentheses, state, etc.
1197     char stat[2048];
1198     int statlen;
1199 
1200     fp = fopen("/proc/self/stat", "r");
1201     if (fp) {
1202       statlen = fread(stat, 1, 2047, fp);
1203       stat[statlen] = '\0';
1204       fclose(fp);
1205 
1206       // Skip pid and the command string. Note that we could be dealing with
1207       // weird command names, e.g. user could decide to rename java launcher
1208       // to "java 1.4.2 :)", then the stat file would look like
1209       //                1234 (java 1.4.2 :)) R ... ...
1210       // We don't really need to know the command string, just find the last
1211       // occurrence of ")" and then start parsing from there. See bug 4726580.
1212       char * s = strrchr(stat, ')');
1213 
1214       i = 0;
1215       if (s) {
1216         // Skip blank chars
1217         do s++; while (isspace(*s));
1218 
1219 #define _UFM UINTX_FORMAT
1220 #define _DFM INTX_FORMAT
1221 
1222         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1223         /*              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 */
1224         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,
1225              &state,          /* 3  %c  */
1226              &ppid,           /* 4  %d  */
1227              &pgrp,           /* 5  %d  */
1228              &session,        /* 6  %d  */
1229              &nr,             /* 7  %d  */
1230              &tpgrp,          /* 8  %d  */
1231              &flags,          /* 9  %lu  */
1232              &minflt,         /* 10 %lu  */
1233              &cminflt,        /* 11 %lu  */
1234              &majflt,         /* 12 %lu  */
1235              &cmajflt,        /* 13 %lu  */
1236              &utime,          /* 14 %lu  */
1237              &stime,          /* 15 %lu  */
1238              &cutime,         /* 16 %ld  */
1239              &cstime,         /* 17 %ld  */
1240              &prio,           /* 18 %ld  */
1241              &nice,           /* 19 %ld  */
1242              &junk,           /* 20 %ld  */
1243              &it_real,        /* 21 %ld  */
1244              &start,          /* 22 UINTX_FORMAT */
1245              &vsize,          /* 23 UINTX_FORMAT */
1246              &rss,            /* 24 INTX_FORMAT  */
1247              &rsslim,         /* 25 UINTX_FORMAT */
1248              &scodes,         /* 26 UINTX_FORMAT */
1249              &ecode,          /* 27 UINTX_FORMAT */
1250              &stack_start);   /* 28 UINTX_FORMAT */
1251       }
1252 
1253 #undef _UFM
1254 #undef _DFM
1255 
1256       if (i != 28 - 2) {
1257          assert(false, "Bad conversion from /proc/self/stat");
1258          // product mode - assume we are the initial thread, good luck in the
1259          // embedded case.
1260          warning("Can't detect initial thread stack location - bad conversion");
1261          stack_start = (uintptr_t) &rlim;
1262       }
1263     } else {
1264       // For some reason we can't open /proc/self/stat (for example, running on
1265       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1266       // most cases, so don't abort:
1267       warning("Can't detect initial thread stack location - no /proc/self/stat");
1268       stack_start = (uintptr_t) &rlim;
1269     }
1270   }
1271 
1272   // Now we have a pointer (stack_start) very close to the stack top, the
1273   // next thing to do is to figure out the exact location of stack top. We
1274   // can find out the virtual memory area that contains stack_start by
1275   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1276   // and its upper limit is the real stack top. (again, this would fail if
1277   // running inside chroot, because /proc may not exist.)
1278 
1279   uintptr_t stack_top;
1280   address low, high;
1281   if (find_vma((address)stack_start, &low, &high)) {
1282     // success, "high" is the true stack top. (ignore "low", because initial
1283     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1284     stack_top = (uintptr_t)high;
1285   } else {
1286     // failed, likely because /proc/self/maps does not exist
1287     warning("Can't detect initial thread stack location - find_vma failed");
1288     // best effort: stack_start is normally within a few pages below the real
1289     // stack top, use it as stack top, and reduce stack size so we won't put
1290     // guard page outside stack.
1291     stack_top = stack_start;
1292     stack_size -= 16 * page_size();
1293   }
1294 
1295   // stack_top could be partially down the page so align it
1296   stack_top = align_size_up(stack_top, page_size());
1297 
1298   if (max_size && stack_size > max_size) {
1299      _initial_thread_stack_size = max_size;
1300   } else {
1301      _initial_thread_stack_size = stack_size;
1302   }
1303 
1304   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1305   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1306 }
1307 
1308 ////////////////////////////////////////////////////////////////////////////////
1309 // time support
1310 
1311 // Time since start-up in seconds to a fine granularity.
1312 // Used by VMSelfDestructTimer and the MemProfiler.
1313 double os::elapsedTime() {
1314 
1315   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1316 }
1317 
1318 jlong os::elapsed_counter() {
1319   return javaTimeNanos() - initial_time_count;
1320 }
1321 
1322 jlong os::elapsed_frequency() {
1323   return NANOSECS_PER_SEC; // nanosecond resolution
1324 }
1325 
1326 bool os::supports_vtime() { return true; }
1327 bool os::enable_vtime()   { return false; }
1328 bool os::vtime_enabled()  { return false; }
1329 
1330 double os::elapsedVTime() {
1331   struct rusage usage;
1332   int retval = getrusage(RUSAGE_THREAD, &usage);
1333   if (retval == 0) {
1334     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);
1335   } else {
1336     // better than nothing, but not much
1337     return elapsedTime();
1338   }
1339 }
1340 
1341 jlong os::javaTimeMillis() {
1342   timeval time;
1343   int status = gettimeofday(&time, NULL);
1344   assert(status != -1, "linux error");
1345   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1346 }
1347 
1348 #ifndef CLOCK_MONOTONIC
1349 #define CLOCK_MONOTONIC (1)
1350 #endif
1351 
1352 void os::Linux::clock_init() {
1353   // we do dlopen's in this particular order due to bug in linux
1354   // dynamical loader (see 6348968) leading to crash on exit
1355   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1356   if (handle == NULL) {
1357     handle = dlopen("librt.so", RTLD_LAZY);
1358   }
1359 
1360   if (handle) {
1361     int (*clock_getres_func)(clockid_t, struct timespec*) =
1362            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1363     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1364            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1365     if (clock_getres_func && clock_gettime_func) {
1366       // See if monotonic clock is supported by the kernel. Note that some
1367       // early implementations simply return kernel jiffies (updated every
1368       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1369       // for nano time (though the monotonic property is still nice to have).
1370       // It's fixed in newer kernels, however clock_getres() still returns
1371       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1372       // resolution for now. Hopefully as people move to new kernels, this
1373       // won't be a problem.
1374       struct timespec res;
1375       struct timespec tp;
1376       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1377           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1378         // yes, monotonic clock is supported
1379         _clock_gettime = clock_gettime_func;
1380         return;
1381       } else {
1382         // close librt if there is no monotonic clock
1383         dlclose(handle);
1384       }
1385     }
1386   }
1387   warning("No monotonic clock was available - timed services may " \
1388           "be adversely affected if the time-of-day clock changes");
1389 }
1390 
1391 #ifndef SYS_clock_getres
1392 
1393 #if defined(IA32) || defined(AMD64)
1394 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1395 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1396 #else
1397 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1398 #define sys_clock_getres(x,y)  -1
1399 #endif
1400 
1401 #else
1402 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1403 #endif
1404 
1405 void os::Linux::fast_thread_clock_init() {
1406   if (!UseLinuxPosixThreadCPUClocks) {
1407     return;
1408   }
1409   clockid_t clockid;
1410   struct timespec tp;
1411   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1412       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1413 
1414   // Switch to using fast clocks for thread cpu time if
1415   // the sys_clock_getres() returns 0 error code.
1416   // Note, that some kernels may support the current thread
1417   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1418   // returned by the pthread_getcpuclockid().
1419   // If the fast Posix clocks are supported then the sys_clock_getres()
1420   // must return at least tp.tv_sec == 0 which means a resolution
1421   // better than 1 sec. This is extra check for reliability.
1422 
1423   if(pthread_getcpuclockid_func &&
1424      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1425      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1426 
1427     _supports_fast_thread_cpu_time = true;
1428     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1429   }
1430 }
1431 
1432 jlong os::javaTimeNanos() {
1433   if (os::supports_monotonic_clock()) {
1434     struct timespec tp;
1435     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1436     assert(status == 0, "gettime error");
1437     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1438     return result;
1439   } else {
1440     timeval time;
1441     int status = gettimeofday(&time, NULL);
1442     assert(status != -1, "linux error");
1443     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1444     return 1000 * usecs;
1445   }
1446 }
1447 
1448 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1449   if (os::supports_monotonic_clock()) {
1450     info_ptr->max_value = ALL_64_BITS;
1451 
1452     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1453     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1454     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1455   } else {
1456     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1457     info_ptr->max_value = ALL_64_BITS;
1458 
1459     // gettimeofday is a real time clock so it skips
1460     info_ptr->may_skip_backward = true;
1461     info_ptr->may_skip_forward = true;
1462   }
1463 
1464   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1465 }
1466 
1467 // Return the real, user, and system times in seconds from an
1468 // arbitrary fixed point in the past.
1469 bool os::getTimesSecs(double* process_real_time,
1470                       double* process_user_time,
1471                       double* process_system_time) {
1472   struct tms ticks;
1473   clock_t real_ticks = times(&ticks);
1474 
1475   if (real_ticks == (clock_t) (-1)) {
1476     return false;
1477   } else {
1478     double ticks_per_second = (double) clock_tics_per_sec;
1479     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1480     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1481     *process_real_time = ((double) real_ticks) / ticks_per_second;
1482 
1483     return true;
1484   }
1485 }
1486 
1487 
1488 char * os::local_time_string(char *buf, size_t buflen) {
1489   struct tm t;
1490   time_t long_time;
1491   time(&long_time);
1492   localtime_r(&long_time, &t);
1493   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1494                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1495                t.tm_hour, t.tm_min, t.tm_sec);
1496   return buf;
1497 }
1498 
1499 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1500   return localtime_r(clock, res);
1501 }
1502 
1503 ////////////////////////////////////////////////////////////////////////////////
1504 // runtime exit support
1505 
1506 // Note: os::shutdown() might be called very early during initialization, or
1507 // called from signal handler. Before adding something to os::shutdown(), make
1508 // sure it is async-safe and can handle partially initialized VM.
1509 void os::shutdown() {
1510 
1511   // allow PerfMemory to attempt cleanup of any persistent resources
1512   perfMemory_exit();
1513 
1514   // needs to remove object in file system
1515   AttachListener::abort();
1516 
1517   // flush buffered output, finish log files
1518   ostream_abort();
1519 
1520   // Check for abort hook
1521   abort_hook_t abort_hook = Arguments::abort_hook();
1522   if (abort_hook != NULL) {
1523     abort_hook();
1524   }
1525 
1526 }
1527 
1528 // Note: os::abort() might be called very early during initialization, or
1529 // called from signal handler. Before adding something to os::abort(), make
1530 // sure it is async-safe and can handle partially initialized VM.
1531 void os::abort(bool dump_core) {
1532   os::shutdown();
1533   if (dump_core) {
1534 #ifndef PRODUCT
1535     fdStream out(defaultStream::output_fd());
1536     out.print_raw("Current thread is ");
1537     char buf[16];
1538     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1539     out.print_raw_cr(buf);
1540     out.print_raw_cr("Dumping core ...");
1541 #endif
1542     ::abort(); // dump core
1543   }
1544 
1545   ::exit(1);
1546 }
1547 
1548 // Die immediately, no exit hook, no abort hook, no cleanup.
1549 void os::die() {
1550   // _exit() on LinuxThreads only kills current thread
1551   ::abort();
1552 }
1553 
1554 // unused on linux for now.
1555 void os::set_error_file(const char *logfile) {}
1556 
1557 
1558 // This method is a copy of JDK's sysGetLastErrorString
1559 // from src/solaris/hpi/src/system_md.c
1560 
1561 size_t os::lasterror(char *buf, size_t len) {
1562 
1563   if (errno == 0)  return 0;
1564 
1565   const char *s = ::strerror(errno);
1566   size_t n = ::strlen(s);
1567   if (n >= len) {
1568     n = len - 1;
1569   }
1570   ::strncpy(buf, s, n);
1571   buf[n] = '\0';
1572   return n;
1573 }
1574 
1575 intx os::current_thread_id() { return (intx)pthread_self(); }
1576 int os::current_process_id() {
1577 
1578   // Under the old linux thread library, linux gives each thread
1579   // its own process id. Because of this each thread will return
1580   // a different pid if this method were to return the result
1581   // of getpid(2). Linux provides no api that returns the pid
1582   // of the launcher thread for the vm. This implementation
1583   // returns a unique pid, the pid of the launcher thread
1584   // that starts the vm 'process'.
1585 
1586   // Under the NPTL, getpid() returns the same pid as the
1587   // launcher thread rather than a unique pid per thread.
1588   // Use gettid() if you want the old pre NPTL behaviour.
1589 
1590   // if you are looking for the result of a call to getpid() that
1591   // returns a unique pid for the calling thread, then look at the
1592   // OSThread::thread_id() method in osThread_linux.hpp file
1593 
1594   return (int)(_initial_pid ? _initial_pid : getpid());
1595 }
1596 
1597 // DLL functions
1598 
1599 const char* os::dll_file_extension() { return ".so"; }
1600 
1601 // This must be hard coded because it's the system's temporary
1602 // directory not the java application's temp directory, ala java.io.tmpdir.
1603 const char* os::get_temp_directory() { return "/tmp"; }
1604 
1605 static bool file_exists(const char* filename) {
1606   struct stat statbuf;
1607   if (filename == NULL || strlen(filename) == 0) {
1608     return false;
1609   }
1610   return os::stat(filename, &statbuf) == 0;
1611 }
1612 
1613 bool os::dll_build_name(char* buffer, size_t buflen,
1614                         const char* pname, const char* fname) {
1615   bool retval = false;
1616   // Copied from libhpi
1617   const size_t pnamelen = pname ? strlen(pname) : 0;
1618 
1619   // Return error on buffer overflow.
1620   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1621     return retval;
1622   }
1623 
1624   if (pnamelen == 0) {
1625     snprintf(buffer, buflen, "lib%s.so", fname);
1626     retval = true;
1627   } else if (strchr(pname, *os::path_separator()) != NULL) {
1628     int n;
1629     char** pelements = split_path(pname, &n);
1630     if (pelements == NULL) {
1631       return false;
1632     }
1633     for (int i = 0 ; i < n ; i++) {
1634       // Really shouldn't be NULL, but check can't hurt
1635       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1636         continue; // skip the empty path values
1637       }
1638       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1639       if (file_exists(buffer)) {
1640         retval = true;
1641         break;
1642       }
1643     }
1644     // release the storage
1645     for (int i = 0 ; i < n ; i++) {
1646       if (pelements[i] != NULL) {
1647         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1648       }
1649     }
1650     if (pelements != NULL) {
1651       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1652     }
1653   } else {
1654     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1655     retval = true;
1656   }
1657   return retval;
1658 }
1659 
1660 // check if addr is inside libjvm.so
1661 bool os::address_is_in_vm(address addr) {
1662   static address libjvm_base_addr;
1663   Dl_info dlinfo;
1664 
1665   if (libjvm_base_addr == NULL) {
1666     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1667       libjvm_base_addr = (address)dlinfo.dli_fbase;
1668     }
1669     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1670   }
1671 
1672   if (dladdr((void *)addr, &dlinfo) != 0) {
1673     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1674   }
1675 
1676   return false;
1677 }
1678 
1679 bool os::dll_address_to_function_name(address addr, char *buf,
1680                                       int buflen, int *offset) {
1681   // buf is not optional, but offset is optional
1682   assert(buf != NULL, "sanity check");
1683 
1684   Dl_info dlinfo;
1685 
1686   if (dladdr((void*)addr, &dlinfo) != 0) {
1687     // see if we have a matching symbol
1688     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1689       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1690         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1691       }
1692       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1693       return true;
1694     }
1695     // no matching symbol so try for just file info
1696     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1697       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1698                           buf, buflen, offset, dlinfo.dli_fname)) {
1699         return true;
1700       }
1701     }
1702   }
1703 
1704   buf[0] = '\0';
1705   if (offset != NULL) *offset = -1;
1706   return false;
1707 }
1708 
1709 struct _address_to_library_name {
1710   address addr;          // input : memory address
1711   size_t  buflen;        //         size of fname
1712   char*   fname;         // output: library name
1713   address base;          //         library base addr
1714 };
1715 
1716 static int address_to_library_name_callback(struct dl_phdr_info *info,
1717                                             size_t size, void *data) {
1718   int i;
1719   bool found = false;
1720   address libbase = NULL;
1721   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1722 
1723   // iterate through all loadable segments
1724   for (i = 0; i < info->dlpi_phnum; i++) {
1725     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1726     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1727       // base address of a library is the lowest address of its loaded
1728       // segments.
1729       if (libbase == NULL || libbase > segbase) {
1730         libbase = segbase;
1731       }
1732       // see if 'addr' is within current segment
1733       if (segbase <= d->addr &&
1734           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1735         found = true;
1736       }
1737     }
1738   }
1739 
1740   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1741   // so dll_address_to_library_name() can fall through to use dladdr() which
1742   // can figure out executable name from argv[0].
1743   if (found && info->dlpi_name && info->dlpi_name[0]) {
1744     d->base = libbase;
1745     if (d->fname) {
1746       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1747     }
1748     return 1;
1749   }
1750   return 0;
1751 }
1752 
1753 bool os::dll_address_to_library_name(address addr, char* buf,
1754                                      int buflen, int* offset) {
1755   // buf is not optional, but offset is optional
1756   assert(buf != NULL, "sanity check");
1757 
1758   Dl_info dlinfo;
1759   struct _address_to_library_name data;
1760 
1761   // There is a bug in old glibc dladdr() implementation that it could resolve
1762   // to wrong library name if the .so file has a base address != NULL. Here
1763   // we iterate through the program headers of all loaded libraries to find
1764   // out which library 'addr' really belongs to. This workaround can be
1765   // removed once the minimum requirement for glibc is moved to 2.3.x.
1766   data.addr = addr;
1767   data.fname = buf;
1768   data.buflen = buflen;
1769   data.base = NULL;
1770   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1771 
1772   if (rslt) {
1773      // buf already contains library name
1774      if (offset) *offset = addr - data.base;
1775      return true;
1776   }
1777   if (dladdr((void*)addr, &dlinfo) != 0) {
1778     if (dlinfo.dli_fname != NULL) {
1779       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1780     }
1781     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1782       *offset = addr - (address)dlinfo.dli_fbase;
1783     }
1784     return true;
1785   }
1786 
1787   buf[0] = '\0';
1788   if (offset) *offset = -1;
1789   return false;
1790 }
1791 
1792   // Loads .dll/.so and
1793   // in case of error it checks if .dll/.so was built for the
1794   // same architecture as Hotspot is running on
1795 
1796 
1797 // Remember the stack's state. The Linux dynamic linker will change
1798 // the stack to 'executable' at most once, so we must safepoint only once.
1799 bool os::Linux::_stack_is_executable = false;
1800 
1801 // VM operation that loads a library.  This is necessary if stack protection
1802 // of the Java stacks can be lost during loading the library.  If we
1803 // do not stop the Java threads, they can stack overflow before the stacks
1804 // are protected again.
1805 class VM_LinuxDllLoad: public VM_Operation {
1806  private:
1807   const char *_filename;
1808   char *_ebuf;
1809   int _ebuflen;
1810   void *_lib;
1811  public:
1812   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1813     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1814   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1815   void doit() {
1816     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1817     os::Linux::_stack_is_executable = true;
1818   }
1819   void* loaded_library() { return _lib; }
1820 };
1821 
1822 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1823 {
1824   void * result = NULL;
1825   bool load_attempted = false;
1826 
1827   // Check whether the library to load might change execution rights
1828   // of the stack. If they are changed, the protection of the stack
1829   // guard pages will be lost. We need a safepoint to fix this.
1830   //
1831   // See Linux man page execstack(8) for more info.
1832   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1833     ElfFile ef(filename);
1834     if (!ef.specifies_noexecstack()) {
1835       if (!is_init_completed()) {
1836         os::Linux::_stack_is_executable = true;
1837         // This is OK - No Java threads have been created yet, and hence no
1838         // stack guard pages to fix.
1839         //
1840         // This should happen only when you are building JDK7 using a very
1841         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1842         //
1843         // Dynamic loader will make all stacks executable after
1844         // this function returns, and will not do that again.
1845         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1846       } else {
1847         warning("You have loaded library %s which might have disabled stack guard. "
1848                 "The VM will try to fix the stack guard now.\n"
1849                 "It's highly recommended that you fix the library with "
1850                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1851                 filename);
1852 
1853         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1854         JavaThread *jt = JavaThread::current();
1855         if (jt->thread_state() != _thread_in_native) {
1856           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1857           // that requires ExecStack. Cannot enter safe point. Let's give up.
1858           warning("Unable to fix stack guard. Giving up.");
1859         } else {
1860           if (!LoadExecStackDllInVMThread) {
1861             // This is for the case where the DLL has an static
1862             // constructor function that executes JNI code. We cannot
1863             // load such DLLs in the VMThread.
1864             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1865           }
1866 
1867           ThreadInVMfromNative tiv(jt);
1868           debug_only(VMNativeEntryWrapper vew;)
1869 
1870           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1871           VMThread::execute(&op);
1872           if (LoadExecStackDllInVMThread) {
1873             result = op.loaded_library();
1874           }
1875           load_attempted = true;
1876         }
1877       }
1878     }
1879   }
1880 
1881   if (!load_attempted) {
1882     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1883   }
1884 
1885   if (result != NULL) {
1886     // Successful loading
1887     return result;
1888   }
1889 
1890   Elf32_Ehdr elf_head;
1891   int diag_msg_max_length=ebuflen-strlen(ebuf);
1892   char* diag_msg_buf=ebuf+strlen(ebuf);
1893 
1894   if (diag_msg_max_length==0) {
1895     // No more space in ebuf for additional diagnostics message
1896     return NULL;
1897   }
1898 
1899 
1900   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1901 
1902   if (file_descriptor < 0) {
1903     // Can't open library, report dlerror() message
1904     return NULL;
1905   }
1906 
1907   bool failed_to_read_elf_head=
1908     (sizeof(elf_head)!=
1909         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1910 
1911   ::close(file_descriptor);
1912   if (failed_to_read_elf_head) {
1913     // file i/o error - report dlerror() msg
1914     return NULL;
1915   }
1916 
1917   typedef struct {
1918     Elf32_Half  code;         // Actual value as defined in elf.h
1919     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1920     char        elf_class;    // 32 or 64 bit
1921     char        endianess;    // MSB or LSB
1922     char*       name;         // String representation
1923   } arch_t;
1924 
1925   #ifndef EM_486
1926   #define EM_486          6               /* Intel 80486 */
1927   #endif
1928 
1929   static const arch_t arch_array[]={
1930     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1931     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1932     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1933     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1934     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1935     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1936     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1937     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1938 #if defined(VM_LITTLE_ENDIAN)
1939     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1940 #else
1941     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1942 #endif
1943     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1944     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1945     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1946     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1947     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1948     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1949     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1950   };
1951 
1952   #if  (defined IA32)
1953     static  Elf32_Half running_arch_code=EM_386;
1954   #elif   (defined AMD64)
1955     static  Elf32_Half running_arch_code=EM_X86_64;
1956   #elif  (defined IA64)
1957     static  Elf32_Half running_arch_code=EM_IA_64;
1958   #elif  (defined __sparc) && (defined _LP64)
1959     static  Elf32_Half running_arch_code=EM_SPARCV9;
1960   #elif  (defined __sparc) && (!defined _LP64)
1961     static  Elf32_Half running_arch_code=EM_SPARC;
1962   #elif  (defined __powerpc64__)
1963     static  Elf32_Half running_arch_code=EM_PPC64;
1964   #elif  (defined __powerpc__)
1965     static  Elf32_Half running_arch_code=EM_PPC;
1966   #elif  (defined ARM)
1967     static  Elf32_Half running_arch_code=EM_ARM;
1968   #elif  (defined S390)
1969     static  Elf32_Half running_arch_code=EM_S390;
1970   #elif  (defined ALPHA)
1971     static  Elf32_Half running_arch_code=EM_ALPHA;
1972   #elif  (defined MIPSEL)
1973     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1974   #elif  (defined PARISC)
1975     static  Elf32_Half running_arch_code=EM_PARISC;
1976   #elif  (defined MIPS)
1977     static  Elf32_Half running_arch_code=EM_MIPS;
1978   #elif  (defined M68K)
1979     static  Elf32_Half running_arch_code=EM_68K;
1980   #else
1981     #error Method os::dll_load requires that one of following is defined:\
1982          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1983   #endif
1984 
1985   // Identify compatability class for VM's architecture and library's architecture
1986   // Obtain string descriptions for architectures
1987 
1988   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1989   int running_arch_index=-1;
1990 
1991   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1992     if (running_arch_code == arch_array[i].code) {
1993       running_arch_index    = i;
1994     }
1995     if (lib_arch.code == arch_array[i].code) {
1996       lib_arch.compat_class = arch_array[i].compat_class;
1997       lib_arch.name         = arch_array[i].name;
1998     }
1999   }
2000 
2001   assert(running_arch_index != -1,
2002     "Didn't find running architecture code (running_arch_code) in arch_array");
2003   if (running_arch_index == -1) {
2004     // Even though running architecture detection failed
2005     // we may still continue with reporting dlerror() message
2006     return NULL;
2007   }
2008 
2009   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2010     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2011     return NULL;
2012   }
2013 
2014 #ifndef S390
2015   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2016     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2017     return NULL;
2018   }
2019 #endif // !S390
2020 
2021   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2022     if ( lib_arch.name!=NULL ) {
2023       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2024         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2025         lib_arch.name, arch_array[running_arch_index].name);
2026     } else {
2027       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2028       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2029         lib_arch.code,
2030         arch_array[running_arch_index].name);
2031     }
2032   }
2033 
2034   return NULL;
2035 }
2036 
2037 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2038   void * result = ::dlopen(filename, RTLD_LAZY);
2039   if (result == NULL) {
2040     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2041     ebuf[ebuflen-1] = '\0';
2042   }
2043   return result;
2044 }
2045 
2046 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2047   void * result = NULL;
2048   if (LoadExecStackDllInVMThread) {
2049     result = dlopen_helper(filename, ebuf, ebuflen);
2050   }
2051 
2052   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2053   // library that requires an executable stack, or which does not have this
2054   // stack attribute set, dlopen changes the stack attribute to executable. The
2055   // read protection of the guard pages gets lost.
2056   //
2057   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2058   // may have been queued at the same time.
2059 
2060   if (!_stack_is_executable) {
2061     JavaThread *jt = Threads::first();
2062 
2063     while (jt) {
2064       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2065           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2066         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2067                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2068           warning("Attempt to reguard stack yellow zone failed.");
2069         }
2070       }
2071       jt = jt->next();
2072     }
2073   }
2074 
2075   return result;
2076 }
2077 
2078 /*
2079  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2080  * chances are you might want to run the generated bits against glibc-2.0
2081  * libdl.so, so always use locking for any version of glibc.
2082  */
2083 void* os::dll_lookup(void* handle, const char* name) {
2084   pthread_mutex_lock(&dl_mutex);
2085   void* res = dlsym(handle, name);
2086   pthread_mutex_unlock(&dl_mutex);
2087   return res;
2088 }
2089 
2090 void* os::get_default_process_handle() {
2091   return (void*)::dlopen(NULL, RTLD_LAZY);
2092 }
2093 
2094 static bool _print_ascii_file(const char* filename, outputStream* st) {
2095   int fd = ::open(filename, O_RDONLY);
2096   if (fd == -1) {
2097      return false;
2098   }
2099 
2100   char buf[32];
2101   int bytes;
2102   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2103     st->print_raw(buf, bytes);
2104   }
2105 
2106   ::close(fd);
2107 
2108   return true;
2109 }
2110 
2111 void os::print_dll_info(outputStream *st) {
2112    st->print_cr("Dynamic libraries:");
2113 
2114    char fname[32];
2115    pid_t pid = os::Linux::gettid();
2116 
2117    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2118 
2119    if (!_print_ascii_file(fname, st)) {
2120      st->print("Can not get library information for pid = %d\n", pid);
2121    }
2122 }
2123 
2124 void os::print_os_info_brief(outputStream* st) {
2125   os::Linux::print_distro_info(st);
2126 
2127   os::Posix::print_uname_info(st);
2128 
2129   os::Linux::print_libversion_info(st);
2130 
2131 }
2132 
2133 void os::print_os_info(outputStream* st) {
2134   st->print("OS:");
2135 
2136   os::Linux::print_distro_info(st);
2137 
2138   os::Posix::print_uname_info(st);
2139 
2140   // Print warning if unsafe chroot environment detected
2141   if (unsafe_chroot_detected) {
2142     st->print("WARNING!! ");
2143     st->print_cr("%s", unstable_chroot_error);
2144   }
2145 
2146   os::Linux::print_libversion_info(st);
2147 
2148   os::Posix::print_rlimit_info(st);
2149 
2150   os::Posix::print_load_average(st);
2151 
2152   os::Linux::print_full_memory_info(st);
2153 }
2154 
2155 // Try to identify popular distros.
2156 // Most Linux distributions have a /etc/XXX-release file, which contains
2157 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2158 // file that also contains the OS version string. Some have more than one
2159 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2160 // /etc/redhat-release.), so the order is important.
2161 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2162 // their own specific XXX-release file as well as a redhat-release file.
2163 // Because of this the XXX-release file needs to be searched for before the
2164 // redhat-release file.
2165 // Since Red Hat has a lsb-release file that is not very descriptive the
2166 // search for redhat-release needs to be before lsb-release.
2167 // Since the lsb-release file is the new standard it needs to be searched
2168 // before the older style release files.
2169 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2170 // next to last resort.  The os-release file is a new standard that contains
2171 // distribution information and the system-release file seems to be an old
2172 // standard that has been replaced by the lsb-release and os-release files.
2173 // Searching for the debian_version file is the last resort.  It contains
2174 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2175 // "Debian " is printed before the contents of the debian_version file.
2176 void os::Linux::print_distro_info(outputStream* st) {
2177    if (!_print_ascii_file("/etc/oracle-release", st) &&
2178        !_print_ascii_file("/etc/mandriva-release", st) &&
2179        !_print_ascii_file("/etc/mandrake-release", st) &&
2180        !_print_ascii_file("/etc/sun-release", st) &&
2181        !_print_ascii_file("/etc/redhat-release", st) &&
2182        !_print_ascii_file("/etc/lsb-release", st) &&
2183        !_print_ascii_file("/etc/SuSE-release", st) &&
2184        !_print_ascii_file("/etc/turbolinux-release", st) &&
2185        !_print_ascii_file("/etc/gentoo-release", st) &&
2186        !_print_ascii_file("/etc/ltib-release", st) &&
2187        !_print_ascii_file("/etc/angstrom-version", st) &&
2188        !_print_ascii_file("/etc/system-release", st) &&
2189        !_print_ascii_file("/etc/os-release", st)) {
2190 
2191        if (file_exists("/etc/debian_version")) {
2192          st->print("Debian ");
2193          _print_ascii_file("/etc/debian_version", st);
2194        } else {
2195          st->print("Linux");
2196        }
2197    }
2198    st->cr();
2199 }
2200 
2201 void os::Linux::print_libversion_info(outputStream* st) {
2202   // libc, pthread
2203   st->print("libc:");
2204   st->print("%s ", os::Linux::glibc_version());
2205   st->print("%s ", os::Linux::libpthread_version());
2206   if (os::Linux::is_LinuxThreads()) {
2207      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2208   }
2209   st->cr();
2210 }
2211 
2212 void os::Linux::print_full_memory_info(outputStream* st) {
2213    st->print("\n/proc/meminfo:\n");
2214    _print_ascii_file("/proc/meminfo", st);
2215    st->cr();
2216 }
2217 
2218 void os::print_memory_info(outputStream* st) {
2219 
2220   st->print("Memory:");
2221   st->print(" %dk page", os::vm_page_size()>>10);
2222 
2223   // values in struct sysinfo are "unsigned long"
2224   struct sysinfo si;
2225   sysinfo(&si);
2226 
2227   st->print(", physical " UINT64_FORMAT "k",
2228             os::physical_memory() >> 10);
2229   st->print("(" UINT64_FORMAT "k free)",
2230             os::available_memory() >> 10);
2231   st->print(", swap " UINT64_FORMAT "k",
2232             ((jlong)si.totalswap * si.mem_unit) >> 10);
2233   st->print("(" UINT64_FORMAT "k free)",
2234             ((jlong)si.freeswap * si.mem_unit) >> 10);
2235   st->cr();
2236 }
2237 
2238 void os::pd_print_cpu_info(outputStream* st) {
2239   st->print("\n/proc/cpuinfo:\n");
2240   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2241     st->print("  <Not Available>");
2242   }
2243   st->cr();
2244 }
2245 
2246 void os::print_siginfo(outputStream* st, void* siginfo) {
2247   const siginfo_t* si = (const siginfo_t*)siginfo;
2248 
2249   os::Posix::print_siginfo_brief(st, si);
2250 
2251   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2252       UseSharedSpaces) {
2253     FileMapInfo* mapinfo = FileMapInfo::current_info();
2254     if (mapinfo->is_in_shared_space(si->si_addr)) {
2255       st->print("\n\nError accessing class data sharing archive."   \
2256                 " Mapped file inaccessible during execution, "      \
2257                 " possible disk/network problem.");
2258     }
2259   }
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         jrelib_p = buf + len;
2347         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2348         if (0 != access(buf, F_OK)) {
2349           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2350         }
2351 
2352         if (0 == access(buf, F_OK)) {
2353           // Use current module name "libjvm.so"
2354           len = strlen(buf);
2355           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2356         } else {
2357           // Go back to path of .so
2358           rp = realpath(dli_fname, buf);
2359           if (rp == NULL)
2360             return;
2361         }
2362       }
2363     }
2364   }
2365 
2366   strcpy(saved_jvm_path, buf);
2367 }
2368 
2369 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2370   // no prefix required, not even "_"
2371 }
2372 
2373 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2374   // no suffix required
2375 }
2376 
2377 ////////////////////////////////////////////////////////////////////////////////
2378 // sun.misc.Signal support
2379 
2380 static volatile jint sigint_count = 0;
2381 
2382 static void
2383 UserHandler(int sig, void *siginfo, void *context) {
2384   // 4511530 - sem_post is serialized and handled by the manager thread. When
2385   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2386   // don't want to flood the manager thread with sem_post requests.
2387   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2388       return;
2389 
2390   // Ctrl-C is pressed during error reporting, likely because the error
2391   // handler fails to abort. Let VM die immediately.
2392   if (sig == SIGINT && is_error_reported()) {
2393      os::die();
2394   }
2395 
2396   os::signal_notify(sig);
2397 }
2398 
2399 void* os::user_handler() {
2400   return CAST_FROM_FN_PTR(void*, UserHandler);
2401 }
2402 
2403 class Semaphore : public StackObj {
2404   public:
2405     Semaphore();
2406     ~Semaphore();
2407     void signal();
2408     void wait();
2409     bool trywait();
2410     bool timedwait(unsigned int sec, int nsec);
2411   private:
2412     sem_t _semaphore;
2413 };
2414 
2415 Semaphore::Semaphore() {
2416   sem_init(&_semaphore, 0, 0);
2417 }
2418 
2419 Semaphore::~Semaphore() {
2420   sem_destroy(&_semaphore);
2421 }
2422 
2423 void Semaphore::signal() {
2424   sem_post(&_semaphore);
2425 }
2426 
2427 void Semaphore::wait() {
2428   sem_wait(&_semaphore);
2429 }
2430 
2431 bool Semaphore::trywait() {
2432   return sem_trywait(&_semaphore) == 0;
2433 }
2434 
2435 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2436 
2437   struct timespec ts;
2438   // Semaphore's are always associated with CLOCK_REALTIME
2439   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2440   // see unpackTime for discussion on overflow checking
2441   if (sec >= MAX_SECS) {
2442     ts.tv_sec += MAX_SECS;
2443     ts.tv_nsec = 0;
2444   } else {
2445     ts.tv_sec += sec;
2446     ts.tv_nsec += nsec;
2447     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2448       ts.tv_nsec -= NANOSECS_PER_SEC;
2449       ++ts.tv_sec; // note: this must be <= max_secs
2450     }
2451   }
2452 
2453   while (1) {
2454     int result = sem_timedwait(&_semaphore, &ts);
2455     if (result == 0) {
2456       return true;
2457     } else if (errno == EINTR) {
2458       continue;
2459     } else if (errno == ETIMEDOUT) {
2460       return false;
2461     } else {
2462       return false;
2463     }
2464   }
2465 }
2466 
2467 extern "C" {
2468   typedef void (*sa_handler_t)(int);
2469   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2470 }
2471 
2472 void* os::signal(int signal_number, void* handler) {
2473   struct sigaction sigAct, oldSigAct;
2474 
2475   sigfillset(&(sigAct.sa_mask));
2476   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2477   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2478 
2479   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2480     // -1 means registration failed
2481     return (void *)-1;
2482   }
2483 
2484   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2485 }
2486 
2487 void os::signal_raise(int signal_number) {
2488   ::raise(signal_number);
2489 }
2490 
2491 /*
2492  * The following code is moved from os.cpp for making this
2493  * code platform specific, which it is by its very nature.
2494  */
2495 
2496 // Will be modified when max signal is changed to be dynamic
2497 int os::sigexitnum_pd() {
2498   return NSIG;
2499 }
2500 
2501 // a counter for each possible signal value
2502 static volatile jint pending_signals[NSIG+1] = { 0 };
2503 
2504 // Linux(POSIX) specific hand shaking semaphore.
2505 static sem_t sig_sem;
2506 static Semaphore sr_semaphore;
2507 
2508 void os::signal_init_pd() {
2509   // Initialize signal structures
2510   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2511 
2512   // Initialize signal semaphore
2513   ::sem_init(&sig_sem, 0, 0);
2514 }
2515 
2516 void os::signal_notify(int sig) {
2517   Atomic::inc(&pending_signals[sig]);
2518   ::sem_post(&sig_sem);
2519 }
2520 
2521 static int check_pending_signals(bool wait) {
2522   Atomic::store(0, &sigint_count);
2523   for (;;) {
2524     for (int i = 0; i < NSIG + 1; i++) {
2525       jint n = pending_signals[i];
2526       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2527         return i;
2528       }
2529     }
2530     if (!wait) {
2531       return -1;
2532     }
2533     JavaThread *thread = JavaThread::current();
2534     ThreadBlockInVM tbivm(thread);
2535 
2536     bool threadIsSuspended;
2537     do {
2538       thread->set_suspend_equivalent();
2539       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2540       ::sem_wait(&sig_sem);
2541 
2542       // were we externally suspended while we were waiting?
2543       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2544       if (threadIsSuspended) {
2545         //
2546         // The semaphore has been incremented, but while we were waiting
2547         // another thread suspended us. We don't want to continue running
2548         // while suspended because that would surprise the thread that
2549         // suspended us.
2550         //
2551         ::sem_post(&sig_sem);
2552 
2553         thread->java_suspend_self();
2554       }
2555     } while (threadIsSuspended);
2556   }
2557 }
2558 
2559 int os::signal_lookup() {
2560   return check_pending_signals(false);
2561 }
2562 
2563 int os::signal_wait() {
2564   return check_pending_signals(true);
2565 }
2566 
2567 ////////////////////////////////////////////////////////////////////////////////
2568 // Virtual Memory
2569 
2570 int os::vm_page_size() {
2571   // Seems redundant as all get out
2572   assert(os::Linux::page_size() != -1, "must call os::init");
2573   return os::Linux::page_size();
2574 }
2575 
2576 // Solaris allocates memory by pages.
2577 int os::vm_allocation_granularity() {
2578   assert(os::Linux::page_size() != -1, "must call os::init");
2579   return os::Linux::page_size();
2580 }
2581 
2582 // Rationale behind this function:
2583 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2584 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2585 //  samples for JITted code. Here we create private executable mapping over the code cache
2586 //  and then we can use standard (well, almost, as mapping can change) way to provide
2587 //  info for the reporting script by storing timestamp and location of symbol
2588 void linux_wrap_code(char* base, size_t size) {
2589   static volatile jint cnt = 0;
2590 
2591   if (!UseOprofile) {
2592     return;
2593   }
2594 
2595   char buf[PATH_MAX+1];
2596   int num = Atomic::add(1, &cnt);
2597 
2598   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2599            os::get_temp_directory(), os::current_process_id(), num);
2600   unlink(buf);
2601 
2602   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2603 
2604   if (fd != -1) {
2605     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2606     if (rv != (off_t)-1) {
2607       if (::write(fd, "", 1) == 1) {
2608         mmap(base, size,
2609              PROT_READ|PROT_WRITE|PROT_EXEC,
2610              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2611       }
2612     }
2613     ::close(fd);
2614     unlink(buf);
2615   }
2616 }
2617 
2618 static bool recoverable_mmap_error(int err) {
2619   // See if the error is one we can let the caller handle. This
2620   // list of errno values comes from JBS-6843484. I can't find a
2621   // Linux man page that documents this specific set of errno
2622   // values so while this list currently matches Solaris, it may
2623   // change as we gain experience with this failure mode.
2624   switch (err) {
2625   case EBADF:
2626   case EINVAL:
2627   case ENOTSUP:
2628     // let the caller deal with these errors
2629     return true;
2630 
2631   default:
2632     // Any remaining errors on this OS can cause our reserved mapping
2633     // to be lost. That can cause confusion where different data
2634     // structures think they have the same memory mapped. The worst
2635     // scenario is if both the VM and a library think they have the
2636     // same memory mapped.
2637     return false;
2638   }
2639 }
2640 
2641 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2642                                     int err) {
2643   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2644           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2645           strerror(err), err);
2646 }
2647 
2648 static void warn_fail_commit_memory(char* addr, size_t size,
2649                                     size_t alignment_hint, bool exec,
2650                                     int err) {
2651   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2652           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2653           alignment_hint, exec, strerror(err), err);
2654 }
2655 
2656 // NOTE: Linux kernel does not really reserve the pages for us.
2657 //       All it does is to check if there are enough free pages
2658 //       left at the time of mmap(). This could be a potential
2659 //       problem.
2660 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2661   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2662   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2663                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2664   if (res != (uintptr_t) MAP_FAILED) {
2665     if (UseNUMAInterleaving) {
2666       numa_make_global(addr, size);
2667     }
2668     return 0;
2669   }
2670 
2671   int err = errno;  // save errno from mmap() call above
2672 
2673   if (!recoverable_mmap_error(err)) {
2674     warn_fail_commit_memory(addr, size, exec, err);
2675     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2676   }
2677 
2678   return err;
2679 }
2680 
2681 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2682   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2683 }
2684 
2685 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2686                                   const char* mesg) {
2687   assert(mesg != NULL, "mesg must be specified");
2688   int err = os::Linux::commit_memory_impl(addr, size, exec);
2689   if (err != 0) {
2690     // the caller wants all commit errors to exit with the specified mesg:
2691     warn_fail_commit_memory(addr, size, exec, err);
2692     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2693   }
2694 }
2695 
2696 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2697 #ifndef MAP_HUGETLB
2698 #define MAP_HUGETLB 0x40000
2699 #endif
2700 
2701 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2702 #ifndef MADV_HUGEPAGE
2703 #define MADV_HUGEPAGE 14
2704 #endif
2705 
2706 int os::Linux::commit_memory_impl(char* addr, size_t size,
2707                                   size_t alignment_hint, bool exec) {
2708   int err = os::Linux::commit_memory_impl(addr, size, exec);
2709   if (err == 0) {
2710     realign_memory(addr, size, alignment_hint);
2711   }
2712   return err;
2713 }
2714 
2715 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2716                           bool exec) {
2717   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2718 }
2719 
2720 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2721                                   size_t alignment_hint, bool exec,
2722                                   const char* mesg) {
2723   assert(mesg != NULL, "mesg must be specified");
2724   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2725   if (err != 0) {
2726     // the caller wants all commit errors to exit with the specified mesg:
2727     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2728     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2729   }
2730 }
2731 
2732 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2733   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2734     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2735     // be supported or the memory may already be backed by huge pages.
2736     ::madvise(addr, bytes, MADV_HUGEPAGE);
2737   }
2738 }
2739 
2740 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2741   // This method works by doing an mmap over an existing mmaping and effectively discarding
2742   // the existing pages. However it won't work for SHM-based large pages that cannot be
2743   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2744   // small pages on top of the SHM segment. This method always works for small pages, so we
2745   // allow that in any case.
2746   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2747     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2748   }
2749 }
2750 
2751 void os::numa_make_global(char *addr, size_t bytes) {
2752   Linux::numa_interleave_memory(addr, bytes);
2753 }
2754 
2755 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2756 // bind policy to MPOL_PREFERRED for the current thread.
2757 #define USE_MPOL_PREFERRED 0
2758 
2759 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2760   // To make NUMA and large pages more robust when both enabled, we need to ease
2761   // the requirements on where the memory should be allocated. MPOL_BIND is the
2762   // default policy and it will force memory to be allocated on the specified
2763   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2764   // the specified node, but will not force it. Using this policy will prevent
2765   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2766   // free large pages.
2767   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2768   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2769 }
2770 
2771 bool os::numa_topology_changed()   { return false; }
2772 
2773 size_t os::numa_get_groups_num() {
2774   int max_node = Linux::numa_max_node();
2775   return max_node > 0 ? max_node + 1 : 1;
2776 }
2777 
2778 int os::numa_get_group_id() {
2779   int cpu_id = Linux::sched_getcpu();
2780   if (cpu_id != -1) {
2781     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2782     if (lgrp_id != -1) {
2783       return lgrp_id;
2784     }
2785   }
2786   return 0;
2787 }
2788 
2789 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2790   for (size_t i = 0; i < size; i++) {
2791     ids[i] = i;
2792   }
2793   return size;
2794 }
2795 
2796 bool os::get_page_info(char *start, page_info* info) {
2797   return false;
2798 }
2799 
2800 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2801   return end;
2802 }
2803 
2804 
2805 int os::Linux::sched_getcpu_syscall(void) {
2806   unsigned int cpu;
2807   int retval = -1;
2808 
2809 #if defined(IA32)
2810 # ifndef SYS_getcpu
2811 # define SYS_getcpu 318
2812 # endif
2813   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2814 #elif defined(AMD64)
2815 // Unfortunately we have to bring all these macros here from vsyscall.h
2816 // to be able to compile on old linuxes.
2817 # define __NR_vgetcpu 2
2818 # define VSYSCALL_START (-10UL << 20)
2819 # define VSYSCALL_SIZE 1024
2820 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2821   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2822   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2823   retval = vgetcpu(&cpu, NULL, NULL);
2824 #endif
2825 
2826   return (retval == -1) ? retval : cpu;
2827 }
2828 
2829 // Something to do with the numa-aware allocator needs these symbols
2830 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2831 extern "C" JNIEXPORT void numa_error(char *where) { }
2832 extern "C" JNIEXPORT int fork1() { return fork(); }
2833 
2834 
2835 // If we are running with libnuma version > 2, then we should
2836 // be trying to use symbols with versions 1.1
2837 // If we are running with earlier version, which did not have symbol versions,
2838 // we should use the base version.
2839 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2840   void *f = dlvsym(handle, name, "libnuma_1.1");
2841   if (f == NULL) {
2842     f = dlsym(handle, name);
2843   }
2844   return f;
2845 }
2846 
2847 bool os::Linux::libnuma_init() {
2848   // sched_getcpu() should be in libc.
2849   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2850                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2851 
2852   // If it's not, try a direct syscall.
2853   if (sched_getcpu() == -1)
2854     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2855 
2856   if (sched_getcpu() != -1) { // Does it work?
2857     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2858     if (handle != NULL) {
2859       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2860                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2861       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2862                                        libnuma_dlsym(handle, "numa_max_node")));
2863       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2864                                         libnuma_dlsym(handle, "numa_available")));
2865       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2866                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2867       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2868                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2869       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2870                                             libnuma_dlsym(handle, "numa_set_bind_policy")));
2871 
2872 
2873       if (numa_available() != -1) {
2874         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2875         // Create a cpu -> node mapping
2876         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2877         rebuild_cpu_to_node_map();
2878         return true;
2879       }
2880     }
2881   }
2882   return false;
2883 }
2884 
2885 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2886 // The table is later used in get_node_by_cpu().
2887 void os::Linux::rebuild_cpu_to_node_map() {
2888   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2889                               // in libnuma (possible values are starting from 16,
2890                               // and continuing up with every other power of 2, but less
2891                               // than the maximum number of CPUs supported by kernel), and
2892                               // is a subject to change (in libnuma version 2 the requirements
2893                               // are more reasonable) we'll just hardcode the number they use
2894                               // in the library.
2895   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2896 
2897   size_t cpu_num = os::active_processor_count();
2898   size_t cpu_map_size = NCPUS / BitsPerCLong;
2899   size_t cpu_map_valid_size =
2900     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2901 
2902   cpu_to_node()->clear();
2903   cpu_to_node()->at_grow(cpu_num - 1);
2904   size_t node_num = numa_get_groups_num();
2905 
2906   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2907   for (size_t i = 0; i < node_num; i++) {
2908     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2909       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2910         if (cpu_map[j] != 0) {
2911           for (size_t k = 0; k < BitsPerCLong; k++) {
2912             if (cpu_map[j] & (1UL << k)) {
2913               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2914             }
2915           }
2916         }
2917       }
2918     }
2919   }
2920   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2921 }
2922 
2923 int os::Linux::get_node_by_cpu(int cpu_id) {
2924   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2925     return cpu_to_node()->at(cpu_id);
2926   }
2927   return -1;
2928 }
2929 
2930 GrowableArray<int>* os::Linux::_cpu_to_node;
2931 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2932 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2933 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2934 os::Linux::numa_available_func_t os::Linux::_numa_available;
2935 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2936 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2937 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2938 unsigned long* os::Linux::_numa_all_nodes;
2939 
2940 bool os::pd_uncommit_memory(char* addr, size_t size) {
2941   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2942                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2943   return res  != (uintptr_t) MAP_FAILED;
2944 }
2945 
2946 static
2947 address get_stack_commited_bottom(address bottom, size_t size) {
2948   address nbot = bottom;
2949   address ntop = bottom + size;
2950 
2951   size_t page_sz = os::vm_page_size();
2952   unsigned pages = size / page_sz;
2953 
2954   unsigned char vec[1];
2955   unsigned imin = 1, imax = pages + 1, imid;
2956   int mincore_return_value = 0;
2957 
2958   assert(imin <= imax, "Unexpected page size");
2959 
2960   while (imin < imax) {
2961     imid = (imax + imin) / 2;
2962     nbot = ntop - (imid * page_sz);
2963 
2964     // Use a trick with mincore to check whether the page is mapped or not.
2965     // mincore sets vec to 1 if page resides in memory and to 0 if page
2966     // is swapped output but if page we are asking for is unmapped
2967     // it returns -1,ENOMEM
2968     mincore_return_value = mincore(nbot, page_sz, vec);
2969 
2970     if (mincore_return_value == -1) {
2971       // Page is not mapped go up
2972       // to find first mapped page
2973       if (errno != EAGAIN) {
2974         assert(errno == ENOMEM, "Unexpected mincore errno");
2975         imax = imid;
2976       }
2977     } else {
2978       // Page is mapped go down
2979       // to find first not mapped page
2980       imin = imid + 1;
2981     }
2982   }
2983 
2984   nbot = nbot + page_sz;
2985 
2986   // Adjust stack bottom one page up if last checked page is not mapped
2987   if (mincore_return_value == -1) {
2988     nbot = nbot + page_sz;
2989   }
2990 
2991   return nbot;
2992 }
2993 
2994 
2995 // Linux uses a growable mapping for the stack, and if the mapping for
2996 // the stack guard pages is not removed when we detach a thread the
2997 // stack cannot grow beyond the pages where the stack guard was
2998 // mapped.  If at some point later in the process the stack expands to
2999 // that point, the Linux kernel cannot expand the stack any further
3000 // because the guard pages are in the way, and a segfault occurs.
3001 //
3002 // However, it's essential not to split the stack region by unmapping
3003 // a region (leaving a hole) that's already part of the stack mapping,
3004 // so if the stack mapping has already grown beyond the guard pages at
3005 // the time we create them, we have to truncate the stack mapping.
3006 // So, we need to know the extent of the stack mapping when
3007 // create_stack_guard_pages() is called.
3008 
3009 // We only need this for stacks that are growable: at the time of
3010 // writing thread stacks don't use growable mappings (i.e. those
3011 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3012 // only applies to the main thread.
3013 
3014 // If the (growable) stack mapping already extends beyond the point
3015 // where we're going to put our guard pages, truncate the mapping at
3016 // that point by munmap()ping it.  This ensures that when we later
3017 // munmap() the guard pages we don't leave a hole in the stack
3018 // mapping. This only affects the main/initial thread
3019 
3020 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3021 
3022   if (os::Linux::is_initial_thread()) {
3023     // As we manually grow stack up to bottom inside create_attached_thread(),
3024     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3025     // we don't need to do anything special.
3026     // Check it first, before calling heavy function.
3027     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3028     unsigned char vec[1];
3029 
3030     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3031       // Fallback to slow path on all errors, including EAGAIN
3032       stack_extent = (uintptr_t) get_stack_commited_bottom(
3033                                     os::Linux::initial_thread_stack_bottom(),
3034                                     (size_t)addr - stack_extent);
3035     }
3036 
3037     if (stack_extent < (uintptr_t)addr) {
3038       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3039     }
3040   }
3041 
3042   return os::commit_memory(addr, size, !ExecMem);
3043 }
3044 
3045 // If this is a growable mapping, remove the guard pages entirely by
3046 // munmap()ping them.  If not, just call uncommit_memory(). This only
3047 // affects the main/initial thread, but guard against future OS changes
3048 // It's safe to always unmap guard pages for initial thread because we
3049 // always place it right after end of the mapped region
3050 
3051 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3052   uintptr_t stack_extent, stack_base;
3053 
3054   if (os::Linux::is_initial_thread()) {
3055     return ::munmap(addr, size) == 0;
3056   }
3057 
3058   return os::uncommit_memory(addr, size);
3059 }
3060 
3061 static address _highest_vm_reserved_address = NULL;
3062 
3063 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3064 // at 'requested_addr'. If there are existing memory mappings at the same
3065 // location, however, they will be overwritten. If 'fixed' is false,
3066 // 'requested_addr' is only treated as a hint, the return value may or
3067 // may not start from the requested address. Unlike Linux mmap(), this
3068 // function returns NULL to indicate failure.
3069 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3070   char * addr;
3071   int flags;
3072 
3073   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3074   if (fixed) {
3075     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3076     flags |= MAP_FIXED;
3077   }
3078 
3079   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3080   // touch an uncommitted page. Otherwise, the read/write might
3081   // succeed if we have enough swap space to back the physical page.
3082   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3083                        flags, -1, 0);
3084 
3085   if (addr != MAP_FAILED) {
3086     // anon_mmap() should only get called during VM initialization,
3087     // don't need lock (actually we can skip locking even it can be called
3088     // from multiple threads, because _highest_vm_reserved_address is just a
3089     // hint about the upper limit of non-stack memory regions.)
3090     if ((address)addr + bytes > _highest_vm_reserved_address) {
3091       _highest_vm_reserved_address = (address)addr + bytes;
3092     }
3093   }
3094 
3095   return addr == MAP_FAILED ? NULL : addr;
3096 }
3097 
3098 // Don't update _highest_vm_reserved_address, because there might be memory
3099 // regions above addr + size. If so, releasing a memory region only creates
3100 // a hole in the address space, it doesn't help prevent heap-stack collision.
3101 //
3102 static int anon_munmap(char * addr, size_t size) {
3103   return ::munmap(addr, size) == 0;
3104 }
3105 
3106 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3107                          size_t alignment_hint) {
3108   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3109 }
3110 
3111 bool os::pd_release_memory(char* addr, size_t size) {
3112   return anon_munmap(addr, size);
3113 }
3114 
3115 static address highest_vm_reserved_address() {
3116   return _highest_vm_reserved_address;
3117 }
3118 
3119 static bool linux_mprotect(char* addr, size_t size, int prot) {
3120   // Linux wants the mprotect address argument to be page aligned.
3121   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3122 
3123   // According to SUSv3, mprotect() should only be used with mappings
3124   // established by mmap(), and mmap() always maps whole pages. Unaligned
3125   // 'addr' likely indicates problem in the VM (e.g. trying to change
3126   // protection of malloc'ed or statically allocated memory). Check the
3127   // caller if you hit this assert.
3128   assert(addr == bottom, "sanity check");
3129 
3130   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3131   return ::mprotect(bottom, size, prot) == 0;
3132 }
3133 
3134 // Set protections specified
3135 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3136                         bool is_committed) {
3137   unsigned int p = 0;
3138   switch (prot) {
3139   case MEM_PROT_NONE: p = PROT_NONE; break;
3140   case MEM_PROT_READ: p = PROT_READ; break;
3141   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3142   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3143   default:
3144     ShouldNotReachHere();
3145   }
3146   // is_committed is unused.
3147   return linux_mprotect(addr, bytes, p);
3148 }
3149 
3150 bool os::guard_memory(char* addr, size_t size) {
3151   return linux_mprotect(addr, size, PROT_NONE);
3152 }
3153 
3154 bool os::unguard_memory(char* addr, size_t size) {
3155   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3156 }
3157 
3158 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3159   bool result = false;
3160   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3161                  MAP_ANONYMOUS|MAP_PRIVATE,
3162                  -1, 0);
3163   if (p != MAP_FAILED) {
3164     void *aligned_p = align_ptr_up(p, page_size);
3165 
3166     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3167 
3168     munmap(p, page_size * 2);
3169   }
3170 
3171   if (warn && !result) {
3172     warning("TransparentHugePages is not supported by the operating system.");
3173   }
3174 
3175   return result;
3176 }
3177 
3178 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3179   bool result = false;
3180   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3181                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3182                  -1, 0);
3183 
3184   if (p != MAP_FAILED) {
3185     // We don't know if this really is a huge page or not.
3186     FILE *fp = fopen("/proc/self/maps", "r");
3187     if (fp) {
3188       while (!feof(fp)) {
3189         char chars[257];
3190         long x = 0;
3191         if (fgets(chars, sizeof(chars), fp)) {
3192           if (sscanf(chars, "%lx-%*x", &x) == 1
3193               && x == (long)p) {
3194             if (strstr (chars, "hugepage")) {
3195               result = true;
3196               break;
3197             }
3198           }
3199         }
3200       }
3201       fclose(fp);
3202     }
3203     munmap(p, page_size);
3204   }
3205 
3206   if (warn && !result) {
3207     warning("HugeTLBFS is not supported by the operating system.");
3208   }
3209 
3210   return result;
3211 }
3212 
3213 /*
3214 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3215 *
3216 * From the coredump_filter documentation:
3217 *
3218 * - (bit 0) anonymous private memory
3219 * - (bit 1) anonymous shared memory
3220 * - (bit 2) file-backed private memory
3221 * - (bit 3) file-backed shared memory
3222 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3223 *           effective only if the bit 2 is cleared)
3224 * - (bit 5) hugetlb private memory
3225 * - (bit 6) hugetlb shared memory
3226 */
3227 static void set_coredump_filter(void) {
3228   FILE *f;
3229   long cdm;
3230 
3231   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3232     return;
3233   }
3234 
3235   if (fscanf(f, "%lx", &cdm) != 1) {
3236     fclose(f);
3237     return;
3238   }
3239 
3240   rewind(f);
3241 
3242   if ((cdm & LARGEPAGES_BIT) == 0) {
3243     cdm |= LARGEPAGES_BIT;
3244     fprintf(f, "%#lx", cdm);
3245   }
3246 
3247   fclose(f);
3248 }
3249 
3250 // Large page support
3251 
3252 static size_t _large_page_size = 0;
3253 
3254 size_t os::Linux::find_large_page_size() {
3255   size_t large_page_size = 0;
3256 
3257   // large_page_size on Linux is used to round up heap size. x86 uses either
3258   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3259   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3260   // page as large as 256M.
3261   //
3262   // Here we try to figure out page size by parsing /proc/meminfo and looking
3263   // for a line with the following format:
3264   //    Hugepagesize:     2048 kB
3265   //
3266   // If we can't determine the value (e.g. /proc is not mounted, or the text
3267   // format has been changed), we'll use the largest page size supported by
3268   // the processor.
3269 
3270 #ifndef ZERO
3271   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3272                      ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3273 #endif // ZERO
3274 
3275   FILE *fp = fopen("/proc/meminfo", "r");
3276   if (fp) {
3277     while (!feof(fp)) {
3278       int x = 0;
3279       char buf[16];
3280       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3281         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3282           large_page_size = x * K;
3283           break;
3284         }
3285       } else {
3286         // skip to next line
3287         for (;;) {
3288           int ch = fgetc(fp);
3289           if (ch == EOF || ch == (int)'\n') break;
3290         }
3291       }
3292     }
3293     fclose(fp);
3294   }
3295 
3296   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3297     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3298         SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3299         proper_unit_for_byte_size(large_page_size));
3300   }
3301 
3302   return large_page_size;
3303 }
3304 
3305 size_t os::Linux::setup_large_page_size() {
3306   _large_page_size = Linux::find_large_page_size();
3307   const size_t default_page_size = (size_t)Linux::page_size();
3308   if (_large_page_size > default_page_size) {
3309     _page_sizes[0] = _large_page_size;
3310     _page_sizes[1] = default_page_size;
3311     _page_sizes[2] = 0;
3312   }
3313 
3314   return _large_page_size;
3315 }
3316 
3317 bool os::Linux::setup_large_page_type(size_t page_size) {
3318   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3319       FLAG_IS_DEFAULT(UseSHM) &&
3320       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3321 
3322     // The type of large pages has not been specified by the user.
3323 
3324     // Try UseHugeTLBFS and then UseSHM.
3325     UseHugeTLBFS = UseSHM = true;
3326 
3327     // Don't try UseTransparentHugePages since there are known
3328     // performance issues with it turned on. This might change in the future.
3329     UseTransparentHugePages = false;
3330   }
3331 
3332   if (UseTransparentHugePages) {
3333     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3334     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3335       UseHugeTLBFS = false;
3336       UseSHM = false;
3337       return true;
3338     }
3339     UseTransparentHugePages = false;
3340   }
3341 
3342   if (UseHugeTLBFS) {
3343     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3344     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3345       UseSHM = false;
3346       return true;
3347     }
3348     UseHugeTLBFS = false;
3349   }
3350 
3351   return UseSHM;
3352 }
3353 
3354 void os::large_page_init() {
3355   if (!UseLargePages &&
3356       !UseTransparentHugePages &&
3357       !UseHugeTLBFS &&
3358       !UseSHM) {
3359     // Not using large pages.
3360     return;
3361   }
3362 
3363   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3364     // The user explicitly turned off large pages.
3365     // Ignore the rest of the large pages flags.
3366     UseTransparentHugePages = false;
3367     UseHugeTLBFS = false;
3368     UseSHM = false;
3369     return;
3370   }
3371 
3372   size_t large_page_size = Linux::setup_large_page_size();
3373   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3374 
3375   set_coredump_filter();
3376 }
3377 
3378 #ifndef SHM_HUGETLB
3379 #define SHM_HUGETLB 04000
3380 #endif
3381 
3382 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3383   // "exec" is passed in but not used.  Creating the shared image for
3384   // the code cache doesn't have an SHM_X executable permission to check.
3385   assert(UseLargePages && UseSHM, "only for SHM large pages");
3386   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3387 
3388   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3389     return NULL; // Fallback to small pages.
3390   }
3391 
3392   key_t key = IPC_PRIVATE;
3393   char *addr;
3394 
3395   bool warn_on_failure = UseLargePages &&
3396                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3397                          !FLAG_IS_DEFAULT(UseSHM) ||
3398                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3399                         );
3400   char msg[128];
3401 
3402   // Create a large shared memory region to attach to based on size.
3403   // Currently, size is the total size of the heap
3404   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3405   if (shmid == -1) {
3406      // Possible reasons for shmget failure:
3407      // 1. shmmax is too small for Java heap.
3408      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3409      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3410      // 2. not enough large page memory.
3411      //    > check available large pages: cat /proc/meminfo
3412      //    > increase amount of large pages:
3413      //          echo new_value > /proc/sys/vm/nr_hugepages
3414      //      Note 1: different Linux may use different name for this property,
3415      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3416      //      Note 2: it's possible there's enough physical memory available but
3417      //            they are so fragmented after a long run that they can't
3418      //            coalesce into large pages. Try to reserve large pages when
3419      //            the system is still "fresh".
3420      if (warn_on_failure) {
3421        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3422        warning("%s", msg);
3423      }
3424      return NULL;
3425   }
3426 
3427   // attach to the region
3428   addr = (char*)shmat(shmid, req_addr, 0);
3429   int err = errno;
3430 
3431   // Remove shmid. If shmat() is successful, the actual shared memory segment
3432   // will be deleted when it's detached by shmdt() or when the process
3433   // terminates. If shmat() is not successful this will remove the shared
3434   // segment immediately.
3435   shmctl(shmid, IPC_RMID, NULL);
3436 
3437   if ((intptr_t)addr == -1) {
3438      if (warn_on_failure) {
3439        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3440        warning("%s", msg);
3441      }
3442      return NULL;
3443   }
3444 
3445   return addr;
3446 }
3447 
3448 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3449   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3450 
3451   bool warn_on_failure = UseLargePages &&
3452       (!FLAG_IS_DEFAULT(UseLargePages) ||
3453        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3454        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3455 
3456   if (warn_on_failure) {
3457     char msg[128];
3458     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3459         PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3460     warning("%s", msg);
3461   }
3462 }
3463 
3464 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3465   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3466   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3467   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3468 
3469   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3470   char* addr = (char*)::mmap(req_addr, bytes, prot,
3471                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3472                              -1, 0);
3473 
3474   if (addr == MAP_FAILED) {
3475     warn_on_large_pages_failure(req_addr, bytes, errno);
3476     return NULL;
3477   }
3478 
3479   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3480 
3481   return addr;
3482 }
3483 
3484 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3485   size_t large_page_size = os::large_page_size();
3486 
3487   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3488 
3489   // Allocate small pages.
3490 
3491   char* start;
3492   if (req_addr != NULL) {
3493     assert(is_ptr_aligned(req_addr, alignment), "Must be");
3494     assert(is_size_aligned(bytes, alignment), "Must be");
3495     start = os::reserve_memory(bytes, req_addr);
3496     assert(start == NULL || start == req_addr, "Must be");
3497   } else {
3498     start = os::reserve_memory_aligned(bytes, alignment);
3499   }
3500 
3501   if (start == NULL) {
3502     return NULL;
3503   }
3504 
3505   assert(is_ptr_aligned(start, alignment), "Must be");
3506 
3507   // os::reserve_memory_special will record this memory area.
3508   // Need to release it here to prevent overlapping reservations.
3509   MemTracker::record_virtual_memory_release((address)start, bytes);
3510 
3511   char* end = start + bytes;
3512 
3513   // Find the regions of the allocated chunk that can be promoted to large pages.
3514   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3515   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3516 
3517   size_t lp_bytes = lp_end - lp_start;
3518 
3519   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3520 
3521   if (lp_bytes == 0) {
3522     // The mapped region doesn't even span the start and the end of a large page.
3523     // Fall back to allocate a non-special area.
3524     ::munmap(start, end - start);
3525     return NULL;
3526   }
3527 
3528   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3529 
3530 
3531   void* result;
3532 
3533   if (start != lp_start) {
3534     result = ::mmap(start, lp_start - start, prot,
3535                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3536                     -1, 0);
3537     if (result == MAP_FAILED) {
3538       ::munmap(lp_start, end - lp_start);
3539       return NULL;
3540     }
3541   }
3542 
3543   result = ::mmap(lp_start, lp_bytes, prot,
3544                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3545                   -1, 0);
3546   if (result == MAP_FAILED) {
3547     warn_on_large_pages_failure(req_addr, bytes, errno);
3548     // If the mmap above fails, the large pages region will be unmapped and we
3549     // have regions before and after with small pages. Release these regions.
3550     //
3551     // |  mapped  |  unmapped  |  mapped  |
3552     // ^          ^            ^          ^
3553     // start      lp_start     lp_end     end
3554     //
3555     ::munmap(start, lp_start - start);
3556     ::munmap(lp_end, end - lp_end);
3557     return NULL;
3558   }
3559 
3560   if (lp_end != end) {
3561       result = ::mmap(lp_end, end - lp_end, prot,
3562                       MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3563                       -1, 0);
3564     if (result == MAP_FAILED) {
3565       ::munmap(start, lp_end - start);
3566       return NULL;
3567     }
3568   }
3569 
3570   return start;
3571 }
3572 
3573 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3574   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3575   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3576   assert(is_power_of_2(alignment), "Must be");
3577   assert(is_power_of_2(os::large_page_size()), "Must be");
3578   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3579 
3580   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3581     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3582   } else {
3583     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3584   }
3585 }
3586 
3587 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3588   assert(UseLargePages, "only for large pages");
3589 
3590   char* addr;
3591   if (UseSHM) {
3592     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3593   } else {
3594     assert(UseHugeTLBFS, "must be");
3595     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3596   }
3597 
3598   if (addr != NULL) {
3599     if (UseNUMAInterleaving) {
3600       numa_make_global(addr, bytes);
3601     }
3602 
3603     // The memory is committed
3604     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3605   }
3606 
3607   return addr;
3608 }
3609 
3610 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3611   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3612   return shmdt(base) == 0;
3613 }
3614 
3615 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3616   return pd_release_memory(base, bytes);
3617 }
3618 
3619 bool os::release_memory_special(char* base, size_t bytes) {
3620   assert(UseLargePages, "only for large pages");
3621 
3622   MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3623 
3624   bool res;
3625   if (UseSHM) {
3626     res = os::Linux::release_memory_special_shm(base, bytes);
3627   } else {
3628     assert(UseHugeTLBFS, "must be");
3629     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3630   }
3631 
3632   if (res) {
3633     tkr.record((address)base, bytes);
3634   } else {
3635     tkr.discard();
3636   }
3637 
3638   return res;
3639 }
3640 
3641 size_t os::large_page_size() {
3642   return _large_page_size;
3643 }
3644 
3645 // With SysV SHM the entire memory region must be allocated as shared
3646 // memory.
3647 // HugeTLBFS allows application to commit large page memory on demand.
3648 // However, when committing memory with HugeTLBFS fails, the region
3649 // that was supposed to be committed will lose the old reservation
3650 // and allow other threads to steal that memory region. Because of this
3651 // behavior we can't commit HugeTLBFS memory.
3652 bool os::can_commit_large_page_memory() {
3653   return UseTransparentHugePages;
3654 }
3655 
3656 bool os::can_execute_large_page_memory() {
3657   return UseTransparentHugePages || UseHugeTLBFS;
3658 }
3659 
3660 // Reserve memory at an arbitrary address, only if that area is
3661 // available (and not reserved for something else).
3662 
3663 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3664   const int max_tries = 10;
3665   char* base[max_tries];
3666   size_t size[max_tries];
3667   const size_t gap = 0x000000;
3668 
3669   // Assert only that the size is a multiple of the page size, since
3670   // that's all that mmap requires, and since that's all we really know
3671   // about at this low abstraction level.  If we need higher alignment,
3672   // we can either pass an alignment to this method or verify alignment
3673   // in one of the methods further up the call chain.  See bug 5044738.
3674   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3675 
3676   // Repeatedly allocate blocks until the block is allocated at the
3677   // right spot. Give up after max_tries. Note that reserve_memory() will
3678   // automatically update _highest_vm_reserved_address if the call is
3679   // successful. The variable tracks the highest memory address every reserved
3680   // by JVM. It is used to detect heap-stack collision if running with
3681   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3682   // space than needed, it could confuse the collision detecting code. To
3683   // solve the problem, save current _highest_vm_reserved_address and
3684   // calculate the correct value before return.
3685   address old_highest = _highest_vm_reserved_address;
3686 
3687   // Linux mmap allows caller to pass an address as hint; give it a try first,
3688   // if kernel honors the hint then we can return immediately.
3689   char * addr = anon_mmap(requested_addr, bytes, false);
3690   if (addr == requested_addr) {
3691      return requested_addr;
3692   }
3693 
3694   if (addr != NULL) {
3695      // mmap() is successful but it fails to reserve at the requested address
3696      anon_munmap(addr, bytes);
3697   }
3698 
3699   int i;
3700   for (i = 0; i < max_tries; ++i) {
3701     base[i] = reserve_memory(bytes);
3702 
3703     if (base[i] != NULL) {
3704       // Is this the block we wanted?
3705       if (base[i] == requested_addr) {
3706         size[i] = bytes;
3707         break;
3708       }
3709 
3710       // Does this overlap the block we wanted? Give back the overlapped
3711       // parts and try again.
3712 
3713       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3714       if (top_overlap >= 0 && top_overlap < bytes) {
3715         unmap_memory(base[i], top_overlap);
3716         base[i] += top_overlap;
3717         size[i] = bytes - top_overlap;
3718       } else {
3719         size_t bottom_overlap = base[i] + bytes - requested_addr;
3720         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3721           unmap_memory(requested_addr, bottom_overlap);
3722           size[i] = bytes - bottom_overlap;
3723         } else {
3724           size[i] = bytes;
3725         }
3726       }
3727     }
3728   }
3729 
3730   // Give back the unused reserved pieces.
3731 
3732   for (int j = 0; j < i; ++j) {
3733     if (base[j] != NULL) {
3734       unmap_memory(base[j], size[j]);
3735     }
3736   }
3737 
3738   if (i < max_tries) {
3739     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3740     return requested_addr;
3741   } else {
3742     _highest_vm_reserved_address = old_highest;
3743     return NULL;
3744   }
3745 }
3746 
3747 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3748   return ::read(fd, buf, nBytes);
3749 }
3750 
3751 //
3752 // Short sleep, direct OS call.
3753 //
3754 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3755 // sched_yield(2) will actually give up the CPU:
3756 //
3757 //   * Alone on this pariticular CPU, keeps running.
3758 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3759 //     (pre 2.6.39).
3760 //
3761 // So calling this with 0 is an alternative.
3762 //
3763 void os::naked_short_sleep(jlong ms) {
3764   struct timespec req;
3765 
3766   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3767   req.tv_sec = 0;
3768   if (ms > 0) {
3769     req.tv_nsec = (ms % 1000) * 1000000;
3770   }
3771   else {
3772     req.tv_nsec = 1;
3773   }
3774 
3775   nanosleep(&req, NULL);
3776 
3777   return;
3778 }
3779 
3780 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3781 void os::infinite_sleep() {
3782   while (true) {    // sleep forever ...
3783     ::sleep(100);   // ... 100 seconds at a time
3784   }
3785 }
3786 
3787 // Used to convert frequent JVM_Yield() to nops
3788 bool os::dont_yield() {
3789   return DontYieldALot;
3790 }
3791 
3792 void os::yield() {
3793   sched_yield();
3794 }
3795 
3796 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3797 
3798 void os::yield_all() {
3799   // Yields to all threads, including threads with lower priorities
3800   // Threads on Linux are all with same priority. The Solaris style
3801   // os::yield_all() with nanosleep(1ms) is not necessary.
3802   sched_yield();
3803 }
3804 
3805 ////////////////////////////////////////////////////////////////////////////////
3806 // thread priority support
3807 
3808 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3809 // only supports dynamic priority, static priority must be zero. For real-time
3810 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3811 // However, for large multi-threaded applications, SCHED_RR is not only slower
3812 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3813 // of 5 runs - Sep 2005).
3814 //
3815 // The following code actually changes the niceness of kernel-thread/LWP. It
3816 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3817 // not the entire user process, and user level threads are 1:1 mapped to kernel
3818 // threads. It has always been the case, but could change in the future. For
3819 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3820 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3821 
3822 int os::java_to_os_priority[CriticalPriority + 1] = {
3823   19,              // 0 Entry should never be used
3824 
3825    4,              // 1 MinPriority
3826    3,              // 2
3827    2,              // 3
3828 
3829    1,              // 4
3830    0,              // 5 NormPriority
3831   -1,              // 6
3832 
3833   -2,              // 7
3834   -3,              // 8
3835   -4,              // 9 NearMaxPriority
3836 
3837   -5,              // 10 MaxPriority
3838 
3839   -5               // 11 CriticalPriority
3840 };
3841 
3842 static int prio_init() {
3843   if (ThreadPriorityPolicy == 1) {
3844     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3845     // if effective uid is not root. Perhaps, a more elegant way of doing
3846     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3847     if (geteuid() != 0) {
3848       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3849         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3850       }
3851       ThreadPriorityPolicy = 0;
3852     }
3853   }
3854   if (UseCriticalJavaThreadPriority) {
3855     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3856   }
3857   return 0;
3858 }
3859 
3860 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3861   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3862 
3863   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3864   return (ret == 0) ? OS_OK : OS_ERR;
3865 }
3866 
3867 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3868   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3869     *priority_ptr = java_to_os_priority[NormPriority];
3870     return OS_OK;
3871   }
3872 
3873   errno = 0;
3874   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3875   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3876 }
3877 
3878 // Hint to the underlying OS that a task switch would not be good.
3879 // Void return because it's a hint and can fail.
3880 void os::hint_no_preempt() {}
3881 
3882 ////////////////////////////////////////////////////////////////////////////////
3883 // suspend/resume support
3884 
3885 //  the low-level signal-based suspend/resume support is a remnant from the
3886 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3887 //  within hotspot. Now there is a single use-case for this:
3888 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3889 //      that runs in the watcher thread.
3890 //  The remaining code is greatly simplified from the more general suspension
3891 //  code that used to be used.
3892 //
3893 //  The protocol is quite simple:
3894 //  - suspend:
3895 //      - sends a signal to the target thread
3896 //      - polls the suspend state of the osthread using a yield loop
3897 //      - target thread signal handler (SR_handler) sets suspend state
3898 //        and blocks in sigsuspend until continued
3899 //  - resume:
3900 //      - sets target osthread state to continue
3901 //      - sends signal to end the sigsuspend loop in the SR_handler
3902 //
3903 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3904 //
3905 
3906 static void resume_clear_context(OSThread *osthread) {
3907   osthread->set_ucontext(NULL);
3908   osthread->set_siginfo(NULL);
3909 }
3910 
3911 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3912   osthread->set_ucontext(context);
3913   osthread->set_siginfo(siginfo);
3914 }
3915 
3916 //
3917 // Handler function invoked when a thread's execution is suspended or
3918 // resumed. We have to be careful that only async-safe functions are
3919 // called here (Note: most pthread functions are not async safe and
3920 // should be avoided.)
3921 //
3922 // Note: sigwait() is a more natural fit than sigsuspend() from an
3923 // interface point of view, but sigwait() prevents the signal hander
3924 // from being run. libpthread would get very confused by not having
3925 // its signal handlers run and prevents sigwait()'s use with the
3926 // mutex granting granting signal.
3927 //
3928 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3929 //
3930 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3931   // Save and restore errno to avoid confusing native code with EINTR
3932   // after sigsuspend.
3933   int old_errno = errno;
3934 
3935   Thread* thread = Thread::current();
3936   OSThread* osthread = thread->osthread();
3937   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3938 
3939   os::SuspendResume::State current = osthread->sr.state();
3940   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3941     suspend_save_context(osthread, siginfo, context);
3942 
3943     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3944     os::SuspendResume::State state = osthread->sr.suspended();
3945     if (state == os::SuspendResume::SR_SUSPENDED) {
3946       sigset_t suspend_set;  // signals for sigsuspend()
3947 
3948       // get current set of blocked signals and unblock resume signal
3949       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3950       sigdelset(&suspend_set, SR_signum);
3951 
3952       sr_semaphore.signal();
3953       // wait here until we are resumed
3954       while (1) {
3955         sigsuspend(&suspend_set);
3956 
3957         os::SuspendResume::State result = osthread->sr.running();
3958         if (result == os::SuspendResume::SR_RUNNING) {
3959           sr_semaphore.signal();
3960           break;
3961         }
3962       }
3963 
3964     } else if (state == os::SuspendResume::SR_RUNNING) {
3965       // request was cancelled, continue
3966     } else {
3967       ShouldNotReachHere();
3968     }
3969 
3970     resume_clear_context(osthread);
3971   } else if (current == os::SuspendResume::SR_RUNNING) {
3972     // request was cancelled, continue
3973   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3974     // ignore
3975   } else {
3976     // ignore
3977   }
3978 
3979   errno = old_errno;
3980 }
3981 
3982 
3983 static int SR_initialize() {
3984   struct sigaction act;
3985   char *s;
3986   /* Get signal number to use for suspend/resume */
3987   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3988     int sig = ::strtol(s, 0, 10);
3989     if (sig > 0 || sig < _NSIG) {
3990         SR_signum = sig;
3991     }
3992   }
3993 
3994   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3995         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3996 
3997   sigemptyset(&SR_sigset);
3998   sigaddset(&SR_sigset, SR_signum);
3999 
4000   /* Set up signal handler for suspend/resume */
4001   act.sa_flags = SA_RESTART|SA_SIGINFO;
4002   act.sa_handler = (void (*)(int)) SR_handler;
4003 
4004   // SR_signum is blocked by default.
4005   // 4528190 - We also need to block pthread restart signal (32 on all
4006   // supported Linux platforms). Note that LinuxThreads need to block
4007   // this signal for all threads to work properly. So we don't have
4008   // to use hard-coded signal number when setting up the mask.
4009   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4010 
4011   if (sigaction(SR_signum, &act, 0) == -1) {
4012     return -1;
4013   }
4014 
4015   // Save signal flag
4016   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4017   return 0;
4018 }
4019 
4020 static int sr_notify(OSThread* osthread) {
4021   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4022   assert_status(status == 0, status, "pthread_kill");
4023   return status;
4024 }
4025 
4026 // "Randomly" selected value for how long we want to spin
4027 // before bailing out on suspending a thread, also how often
4028 // we send a signal to a thread we want to resume
4029 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4030 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4031 
4032 // returns true on success and false on error - really an error is fatal
4033 // but this seems the normal response to library errors
4034 static bool do_suspend(OSThread* osthread) {
4035   assert(osthread->sr.is_running(), "thread should be running");
4036   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4037 
4038   // mark as suspended and send signal
4039   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4040     // failed to switch, state wasn't running?
4041     ShouldNotReachHere();
4042     return false;
4043   }
4044 
4045   if (sr_notify(osthread) != 0) {
4046     ShouldNotReachHere();
4047   }
4048 
4049   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4050   while (true) {
4051     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4052       break;
4053     } else {
4054       // timeout
4055       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4056       if (cancelled == os::SuspendResume::SR_RUNNING) {
4057         return false;
4058       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4059         // make sure that we consume the signal on the semaphore as well
4060         sr_semaphore.wait();
4061         break;
4062       } else {
4063         ShouldNotReachHere();
4064         return false;
4065       }
4066     }
4067   }
4068 
4069   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4070   return true;
4071 }
4072 
4073 static void do_resume(OSThread* osthread) {
4074   assert(osthread->sr.is_suspended(), "thread should be suspended");
4075   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4076 
4077   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4078     // failed to switch to WAKEUP_REQUEST
4079     ShouldNotReachHere();
4080     return;
4081   }
4082 
4083   while (true) {
4084     if (sr_notify(osthread) == 0) {
4085       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4086         if (osthread->sr.is_running()) {
4087           return;
4088         }
4089       }
4090     } else {
4091       ShouldNotReachHere();
4092     }
4093   }
4094 
4095   guarantee(osthread->sr.is_running(), "Must be running!");
4096 }
4097 
4098 ///////////////////////////////////////////////////////////////////////////////////
4099 // signal handling (except suspend/resume)
4100 
4101 // This routine may be used by user applications as a "hook" to catch signals.
4102 // The user-defined signal handler must pass unrecognized signals to this
4103 // routine, and if it returns true (non-zero), then the signal handler must
4104 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4105 // routine will never retun false (zero), but instead will execute a VM panic
4106 // routine kill the process.
4107 //
4108 // If this routine returns false, it is OK to call it again.  This allows
4109 // the user-defined signal handler to perform checks either before or after
4110 // the VM performs its own checks.  Naturally, the user code would be making
4111 // a serious error if it tried to handle an exception (such as a null check
4112 // or breakpoint) that the VM was generating for its own correct operation.
4113 //
4114 // This routine may recognize any of the following kinds of signals:
4115 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4116 // It should be consulted by handlers for any of those signals.
4117 //
4118 // The caller of this routine must pass in the three arguments supplied
4119 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4120 // field of the structure passed to sigaction().  This routine assumes that
4121 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4122 //
4123 // Note that the VM will print warnings if it detects conflicting signal
4124 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4125 //
4126 extern "C" JNIEXPORT int
4127 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4128                         void* ucontext, int abort_if_unrecognized);
4129 
4130 void signalHandler(int sig, siginfo_t* info, void* uc) {
4131   assert(info != NULL && uc != NULL, "it must be old kernel");
4132   int orig_errno = errno;  // Preserve errno value over signal handler.
4133   JVM_handle_linux_signal(sig, info, uc, true);
4134   errno = orig_errno;
4135 }
4136 
4137 
4138 // This boolean allows users to forward their own non-matching signals
4139 // to JVM_handle_linux_signal, harmlessly.
4140 bool os::Linux::signal_handlers_are_installed = false;
4141 
4142 // For signal-chaining
4143 struct sigaction os::Linux::sigact[MAXSIGNUM];
4144 unsigned int os::Linux::sigs = 0;
4145 bool os::Linux::libjsig_is_loaded = false;
4146 typedef struct sigaction *(*get_signal_t)(int);
4147 get_signal_t os::Linux::get_signal_action = NULL;
4148 
4149 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4150   struct sigaction *actp = NULL;
4151 
4152   if (libjsig_is_loaded) {
4153     // Retrieve the old signal handler from libjsig
4154     actp = (*get_signal_action)(sig);
4155   }
4156   if (actp == NULL) {
4157     // Retrieve the preinstalled signal handler from jvm
4158     actp = get_preinstalled_handler(sig);
4159   }
4160 
4161   return actp;
4162 }
4163 
4164 static bool call_chained_handler(struct sigaction *actp, int sig,
4165                                  siginfo_t *siginfo, void *context) {
4166   // Call the old signal handler
4167   if (actp->sa_handler == SIG_DFL) {
4168     // It's more reasonable to let jvm treat it as an unexpected exception
4169     // instead of taking the default action.
4170     return false;
4171   } else if (actp->sa_handler != SIG_IGN) {
4172     if ((actp->sa_flags & SA_NODEFER) == 0) {
4173       // automaticlly block the signal
4174       sigaddset(&(actp->sa_mask), sig);
4175     }
4176 
4177     sa_handler_t hand;
4178     sa_sigaction_t sa;
4179     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4180     // retrieve the chained handler
4181     if (siginfo_flag_set) {
4182       sa = actp->sa_sigaction;
4183     } else {
4184       hand = actp->sa_handler;
4185     }
4186 
4187     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4188       actp->sa_handler = SIG_DFL;
4189     }
4190 
4191     // try to honor the signal mask
4192     sigset_t oset;
4193     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4194 
4195     // call into the chained handler
4196     if (siginfo_flag_set) {
4197       (*sa)(sig, siginfo, context);
4198     } else {
4199       (*hand)(sig);
4200     }
4201 
4202     // restore the signal mask
4203     pthread_sigmask(SIG_SETMASK, &oset, 0);
4204   }
4205   // Tell jvm's signal handler the signal is taken care of.
4206   return true;
4207 }
4208 
4209 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4210   bool chained = false;
4211   // signal-chaining
4212   if (UseSignalChaining) {
4213     struct sigaction *actp = get_chained_signal_action(sig);
4214     if (actp != NULL) {
4215       chained = call_chained_handler(actp, sig, siginfo, context);
4216     }
4217   }
4218   return chained;
4219 }
4220 
4221 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4222   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4223     return &sigact[sig];
4224   }
4225   return NULL;
4226 }
4227 
4228 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4229   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4230   sigact[sig] = oldAct;
4231   sigs |= (unsigned int)1 << sig;
4232 }
4233 
4234 // for diagnostic
4235 int os::Linux::sigflags[MAXSIGNUM];
4236 
4237 int os::Linux::get_our_sigflags(int sig) {
4238   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4239   return sigflags[sig];
4240 }
4241 
4242 void os::Linux::set_our_sigflags(int sig, int flags) {
4243   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4244   sigflags[sig] = flags;
4245 }
4246 
4247 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4248   // Check for overwrite.
4249   struct sigaction oldAct;
4250   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4251 
4252   void* oldhand = oldAct.sa_sigaction
4253                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4254                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4255   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4256       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4257       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4258     if (AllowUserSignalHandlers || !set_installed) {
4259       // Do not overwrite; user takes responsibility to forward to us.
4260       return;
4261     } else if (UseSignalChaining) {
4262       // save the old handler in jvm
4263       save_preinstalled_handler(sig, oldAct);
4264       // libjsig also interposes the sigaction() call below and saves the
4265       // old sigaction on it own.
4266     } else {
4267       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4268                     "%#lx for signal %d.", (long)oldhand, sig));
4269     }
4270   }
4271 
4272   struct sigaction sigAct;
4273   sigfillset(&(sigAct.sa_mask));
4274   sigAct.sa_handler = SIG_DFL;
4275   if (!set_installed) {
4276     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4277   } else {
4278     sigAct.sa_sigaction = signalHandler;
4279     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4280   }
4281   // Save flags, which are set by ours
4282   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4283   sigflags[sig] = sigAct.sa_flags;
4284 
4285   int ret = sigaction(sig, &sigAct, &oldAct);
4286   assert(ret == 0, "check");
4287 
4288   void* oldhand2  = oldAct.sa_sigaction
4289                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4290                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4291   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4292 }
4293 
4294 // install signal handlers for signals that HotSpot needs to
4295 // handle in order to support Java-level exception handling.
4296 
4297 void os::Linux::install_signal_handlers() {
4298   if (!signal_handlers_are_installed) {
4299     signal_handlers_are_installed = true;
4300 
4301     // signal-chaining
4302     typedef void (*signal_setting_t)();
4303     signal_setting_t begin_signal_setting = NULL;
4304     signal_setting_t end_signal_setting = NULL;
4305     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4306                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4307     if (begin_signal_setting != NULL) {
4308       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4309                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4310       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4311                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4312       libjsig_is_loaded = true;
4313       assert(UseSignalChaining, "should enable signal-chaining");
4314     }
4315     if (libjsig_is_loaded) {
4316       // Tell libjsig jvm is setting signal handlers
4317       (*begin_signal_setting)();
4318     }
4319 
4320     set_signal_handler(SIGSEGV, true);
4321     set_signal_handler(SIGPIPE, true);
4322     set_signal_handler(SIGBUS, true);
4323     set_signal_handler(SIGILL, true);
4324     set_signal_handler(SIGFPE, true);
4325 #if defined(PPC64)
4326     set_signal_handler(SIGTRAP, true);
4327 #endif
4328     set_signal_handler(SIGXFSZ, true);
4329 
4330     if (libjsig_is_loaded) {
4331       // Tell libjsig jvm finishes setting signal handlers
4332       (*end_signal_setting)();
4333     }
4334 
4335     // We don't activate signal checker if libjsig is in place, we trust ourselves
4336     // and if UserSignalHandler is installed all bets are off.
4337     // Log that signal checking is off only if -verbose:jni is specified.
4338     if (CheckJNICalls) {
4339       if (libjsig_is_loaded) {
4340         if (PrintJNIResolving) {
4341           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4342         }
4343         check_signals = false;
4344       }
4345       if (AllowUserSignalHandlers) {
4346         if (PrintJNIResolving) {
4347           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4348         }
4349         check_signals = false;
4350       }
4351     }
4352   }
4353 }
4354 
4355 // This is the fastest way to get thread cpu time on Linux.
4356 // Returns cpu time (user+sys) for any thread, not only for current.
4357 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4358 // It might work on 2.6.10+ with a special kernel/glibc patch.
4359 // For reference, please, see IEEE Std 1003.1-2004:
4360 //   http://www.unix.org/single_unix_specification
4361 
4362 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4363   struct timespec tp;
4364   int rc = os::Linux::clock_gettime(clockid, &tp);
4365   assert(rc == 0, "clock_gettime is expected to return 0 code");
4366 
4367   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4368 }
4369 
4370 /////
4371 // glibc on Linux platform uses non-documented flag
4372 // to indicate, that some special sort of signal
4373 // trampoline is used.
4374 // We will never set this flag, and we should
4375 // ignore this flag in our diagnostic
4376 #ifdef SIGNIFICANT_SIGNAL_MASK
4377 #undef SIGNIFICANT_SIGNAL_MASK
4378 #endif
4379 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4380 
4381 static const char* get_signal_handler_name(address handler,
4382                                            char* buf, int buflen) {
4383   int offset;
4384   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4385   if (found) {
4386     // skip directory names
4387     const char *p1, *p2;
4388     p1 = buf;
4389     size_t len = strlen(os::file_separator());
4390     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4391     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4392   } else {
4393     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4394   }
4395   return buf;
4396 }
4397 
4398 static void print_signal_handler(outputStream* st, int sig,
4399                                  char* buf, size_t buflen) {
4400   struct sigaction sa;
4401 
4402   sigaction(sig, NULL, &sa);
4403 
4404   // See comment for SIGNIFICANT_SIGNAL_MASK define
4405   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4406 
4407   st->print("%s: ", os::exception_name(sig, buf, buflen));
4408 
4409   address handler = (sa.sa_flags & SA_SIGINFO)
4410     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4411     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4412 
4413   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4414     st->print("SIG_DFL");
4415   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4416     st->print("SIG_IGN");
4417   } else {
4418     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4419   }
4420 
4421   st->print(", sa_mask[0]=");
4422   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4423 
4424   address rh = VMError::get_resetted_sighandler(sig);
4425   // May be, handler was resetted by VMError?
4426   if(rh != NULL) {
4427     handler = rh;
4428     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4429   }
4430 
4431   st->print(", sa_flags=");
4432   os::Posix::print_sa_flags(st, sa.sa_flags);
4433 
4434   // Check: is it our handler?
4435   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4436      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4437     // It is our signal handler
4438     // check for flags, reset system-used one!
4439     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4440       st->print(
4441                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4442                 os::Linux::get_our_sigflags(sig));
4443     }
4444   }
4445   st->cr();
4446 }
4447 
4448 
4449 #define DO_SIGNAL_CHECK(sig) \
4450   if (!sigismember(&check_signal_done, sig)) \
4451     os::Linux::check_signal_handler(sig)
4452 
4453 // This method is a periodic task to check for misbehaving JNI applications
4454 // under CheckJNI, we can add any periodic checks here
4455 
4456 void os::run_periodic_checks() {
4457 
4458   if (check_signals == false) return;
4459 
4460   // SEGV and BUS if overridden could potentially prevent
4461   // generation of hs*.log in the event of a crash, debugging
4462   // such a case can be very challenging, so we absolutely
4463   // check the following for a good measure:
4464   DO_SIGNAL_CHECK(SIGSEGV);
4465   DO_SIGNAL_CHECK(SIGILL);
4466   DO_SIGNAL_CHECK(SIGFPE);
4467   DO_SIGNAL_CHECK(SIGBUS);
4468   DO_SIGNAL_CHECK(SIGPIPE);
4469   DO_SIGNAL_CHECK(SIGXFSZ);
4470 #if defined(PPC64)
4471   DO_SIGNAL_CHECK(SIGTRAP);
4472 #endif
4473 
4474   // ReduceSignalUsage allows the user to override these handlers
4475   // see comments at the very top and jvm_solaris.h
4476   if (!ReduceSignalUsage) {
4477     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4478     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4479     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4480     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4481   }
4482 
4483   DO_SIGNAL_CHECK(SR_signum);
4484   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4485 }
4486 
4487 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4488 
4489 static os_sigaction_t os_sigaction = NULL;
4490 
4491 void os::Linux::check_signal_handler(int sig) {
4492   char buf[O_BUFLEN];
4493   address jvmHandler = NULL;
4494 
4495 
4496   struct sigaction act;
4497   if (os_sigaction == NULL) {
4498     // only trust the default sigaction, in case it has been interposed
4499     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4500     if (os_sigaction == NULL) return;
4501   }
4502 
4503   os_sigaction(sig, (struct sigaction*)NULL, &act);
4504 
4505 
4506   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4507 
4508   address thisHandler = (act.sa_flags & SA_SIGINFO)
4509     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4510     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4511 
4512 
4513   switch(sig) {
4514   case SIGSEGV:
4515   case SIGBUS:
4516   case SIGFPE:
4517   case SIGPIPE:
4518   case SIGILL:
4519   case SIGXFSZ:
4520     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4521     break;
4522 
4523   case SHUTDOWN1_SIGNAL:
4524   case SHUTDOWN2_SIGNAL:
4525   case SHUTDOWN3_SIGNAL:
4526   case BREAK_SIGNAL:
4527     jvmHandler = (address)user_handler();
4528     break;
4529 
4530   case INTERRUPT_SIGNAL:
4531     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4532     break;
4533 
4534   default:
4535     if (sig == SR_signum) {
4536       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4537     } else {
4538       return;
4539     }
4540     break;
4541   }
4542 
4543   if (thisHandler != jvmHandler) {
4544     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4545     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4546     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4547     // No need to check this sig any longer
4548     sigaddset(&check_signal_done, sig);
4549     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4550     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4551       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4552                     exception_name(sig, buf, O_BUFLEN));
4553     }
4554   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4555     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4556     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4557     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4558     // No need to check this sig any longer
4559     sigaddset(&check_signal_done, sig);
4560   }
4561 
4562   // Dump all the signal
4563   if (sigismember(&check_signal_done, sig)) {
4564     print_signal_handlers(tty, buf, O_BUFLEN);
4565   }
4566 }
4567 
4568 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4569 
4570 extern bool signal_name(int signo, char* buf, size_t len);
4571 
4572 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4573   if (0 < exception_code && exception_code <= SIGRTMAX) {
4574     // signal
4575     if (!signal_name(exception_code, buf, size)) {
4576       jio_snprintf(buf, size, "SIG%d", exception_code);
4577     }
4578     return buf;
4579   } else {
4580     return NULL;
4581   }
4582 }
4583 
4584 // this is called _before_ the most of global arguments have been parsed
4585 void os::init(void) {
4586   char dummy;   /* used to get a guess on initial stack address */
4587 //  first_hrtime = gethrtime();
4588 
4589   // With LinuxThreads the JavaMain thread pid (primordial thread)
4590   // is different than the pid of the java launcher thread.
4591   // So, on Linux, the launcher thread pid is passed to the VM
4592   // via the sun.java.launcher.pid property.
4593   // Use this property instead of getpid() if it was correctly passed.
4594   // See bug 6351349.
4595   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4596 
4597   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4598 
4599   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4600 
4601   init_random(1234567);
4602 
4603   ThreadCritical::initialize();
4604 
4605   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4606   if (Linux::page_size() == -1) {
4607     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4608                   strerror(errno)));
4609   }
4610   init_page_sizes((size_t) Linux::page_size());
4611 
4612   Linux::initialize_system_info();
4613 
4614   // main_thread points to the aboriginal thread
4615   Linux::_main_thread = pthread_self();
4616 
4617   Linux::clock_init();
4618   initial_time_count = javaTimeNanos();
4619 
4620   // pthread_condattr initialization for monotonic clock
4621   int status;
4622   pthread_condattr_t* _condattr = os::Linux::condAttr();
4623   if ((status = pthread_condattr_init(_condattr)) != 0) {
4624     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4625   }
4626   // Only set the clock if CLOCK_MONOTONIC is available
4627   if (os::supports_monotonic_clock()) {
4628     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4629       if (status == EINVAL) {
4630         warning("Unable to use monotonic clock with relative timed-waits" \
4631                 " - changes to the time-of-day clock may have adverse affects");
4632       } else {
4633         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4634       }
4635     }
4636   }
4637   // else it defaults to CLOCK_REALTIME
4638 
4639   pthread_mutex_init(&dl_mutex, NULL);
4640 
4641   // If the pagesize of the VM is greater than 8K determine the appropriate
4642   // number of initial guard pages.  The user can change this with the
4643   // command line arguments, if needed.
4644   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4645     StackYellowPages = 1;
4646     StackRedPages = 1;
4647     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4648   }
4649 }
4650 
4651 // To install functions for atexit system call
4652 extern "C" {
4653   static void perfMemory_exit_helper() {
4654     perfMemory_exit();
4655   }
4656 }
4657 
4658 // this is called _after_ the global arguments have been parsed
4659 jint os::init_2(void)
4660 {
4661   Linux::fast_thread_clock_init();
4662 
4663   // Allocate a single page and mark it as readable for safepoint polling
4664   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4665   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4666 
4667   os::set_polling_page( polling_page );
4668 
4669 #ifndef PRODUCT
4670   if(Verbose && PrintMiscellaneous)
4671     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4672 #endif
4673 
4674   if (!UseMembar) {
4675     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4676     guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4677     os::set_memory_serialize_page( mem_serialize_page );
4678 
4679 #ifndef PRODUCT
4680     if(Verbose && PrintMiscellaneous)
4681       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4682 #endif
4683   }
4684 
4685   // initialize suspend/resume support - must do this before signal_sets_init()
4686   if (SR_initialize() != 0) {
4687     perror("SR_initialize failed");
4688     return JNI_ERR;
4689   }
4690 
4691   Linux::signal_sets_init();
4692   Linux::install_signal_handlers();
4693 
4694   // Check minimum allowable stack size for thread creation and to initialize
4695   // the java system classes, including StackOverflowError - depends on page
4696   // size.  Add a page for compiler2 recursion in main thread.
4697   // Add in 2*BytesPerWord times page size to account for VM stack during
4698   // class initialization depending on 32 or 64 bit VM.
4699   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4700             (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4701                     (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4702 
4703   size_t threadStackSizeInBytes = ThreadStackSize * K;
4704   if (threadStackSizeInBytes != 0 &&
4705       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4706         tty->print_cr("\nThe stack size specified is too small, "
4707                       "Specify at least %dk",
4708                       os::Linux::min_stack_allowed/ K);
4709         return JNI_ERR;
4710   }
4711 
4712   // Make the stack size a multiple of the page size so that
4713   // the yellow/red zones can be guarded.
4714   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4715         vm_page_size()));
4716 
4717   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4718 
4719 #if defined(IA32)
4720   workaround_expand_exec_shield_cs_limit();
4721 #endif
4722 
4723   Linux::libpthread_init();
4724   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4725      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4726           Linux::glibc_version(), Linux::libpthread_version(),
4727           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4728   }
4729 
4730   if (UseNUMA) {
4731     if (!Linux::libnuma_init()) {
4732       UseNUMA = false;
4733     } else {
4734       if ((Linux::numa_max_node() < 1)) {
4735         // There's only one node(they start from 0), disable NUMA.
4736         UseNUMA = false;
4737       }
4738     }
4739     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4740     // we can make the adaptive lgrp chunk resizing work. If the user specified
4741     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4742     // disable adaptive resizing.
4743     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4744       if (FLAG_IS_DEFAULT(UseNUMA)) {
4745         UseNUMA = false;
4746       } else {
4747         if (FLAG_IS_DEFAULT(UseLargePages) &&
4748             FLAG_IS_DEFAULT(UseSHM) &&
4749             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4750           UseLargePages = false;
4751         } else {
4752           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4753           UseAdaptiveSizePolicy = false;
4754           UseAdaptiveNUMAChunkSizing = false;
4755         }
4756       }
4757     }
4758     if (!UseNUMA && ForceNUMA) {
4759       UseNUMA = true;
4760     }
4761   }
4762 
4763   if (MaxFDLimit) {
4764     // set the number of file descriptors to max. print out error
4765     // if getrlimit/setrlimit fails but continue regardless.
4766     struct rlimit nbr_files;
4767     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4768     if (status != 0) {
4769       if (PrintMiscellaneous && (Verbose || WizardMode))
4770         perror("os::init_2 getrlimit failed");
4771     } else {
4772       nbr_files.rlim_cur = nbr_files.rlim_max;
4773       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4774       if (status != 0) {
4775         if (PrintMiscellaneous && (Verbose || WizardMode))
4776           perror("os::init_2 setrlimit failed");
4777       }
4778     }
4779   }
4780 
4781   // Initialize lock used to serialize thread creation (see os::create_thread)
4782   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4783 
4784   // at-exit methods are called in the reverse order of their registration.
4785   // atexit functions are called on return from main or as a result of a
4786   // call to exit(3C). There can be only 32 of these functions registered
4787   // and atexit() does not set errno.
4788 
4789   if (PerfAllowAtExitRegistration) {
4790     // only register atexit functions if PerfAllowAtExitRegistration is set.
4791     // atexit functions can be delayed until process exit time, which
4792     // can be problematic for embedded VM situations. Embedded VMs should
4793     // call DestroyJavaVM() to assure that VM resources are released.
4794 
4795     // note: perfMemory_exit_helper atexit function may be removed in
4796     // the future if the appropriate cleanup code can be added to the
4797     // VM_Exit VMOperation's doit method.
4798     if (atexit(perfMemory_exit_helper) != 0) {
4799       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4800     }
4801   }
4802 
4803   // initialize thread priority policy
4804   prio_init();
4805 
4806   return JNI_OK;
4807 }
4808 
4809 // this is called at the end of vm_initialization
4810 void os::init_3(void) {
4811 #ifdef JAVASE_EMBEDDED
4812   // Start the MemNotifyThread
4813   if (LowMemoryProtection) {
4814     MemNotifyThread::start();
4815   }
4816   return;
4817 #endif
4818 }
4819 
4820 // Mark the polling page as unreadable
4821 void os::make_polling_page_unreadable(void) {
4822   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4823     fatal("Could not disable polling page");
4824 };
4825 
4826 // Mark the polling page as readable
4827 void os::make_polling_page_readable(void) {
4828   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4829     fatal("Could not enable polling page");
4830   }
4831 };
4832 
4833 int os::active_processor_count() {
4834   // Linux doesn't yet have a (official) notion of processor sets,
4835   // so just return the number of online processors.
4836   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4837   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4838   return online_cpus;
4839 }
4840 
4841 void os::set_native_thread_name(const char *name) {
4842   // Not yet implemented.
4843   return;
4844 }
4845 
4846 bool os::distribute_processes(uint length, uint* distribution) {
4847   // Not yet implemented.
4848   return false;
4849 }
4850 
4851 bool os::bind_to_processor(uint processor_id) {
4852   // Not yet implemented.
4853   return false;
4854 }
4855 
4856 ///
4857 
4858 void os::SuspendedThreadTask::internal_do_task() {
4859   if (do_suspend(_thread->osthread())) {
4860     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4861     do_task(context);
4862     do_resume(_thread->osthread());
4863   }
4864 }
4865 
4866 class PcFetcher : public os::SuspendedThreadTask {
4867 public:
4868   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4869   ExtendedPC result();
4870 protected:
4871   void do_task(const os::SuspendedThreadTaskContext& context);
4872 private:
4873   ExtendedPC _epc;
4874 };
4875 
4876 ExtendedPC PcFetcher::result() {
4877   guarantee(is_done(), "task is not done yet.");
4878   return _epc;
4879 }
4880 
4881 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4882   Thread* thread = context.thread();
4883   OSThread* osthread = thread->osthread();
4884   if (osthread->ucontext() != NULL) {
4885     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4886   } else {
4887     // NULL context is unexpected, double-check this is the VMThread
4888     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4889   }
4890 }
4891 
4892 // Suspends the target using the signal mechanism and then grabs the PC before
4893 // resuming the target. Used by the flat-profiler only
4894 ExtendedPC os::get_thread_pc(Thread* thread) {
4895   // Make sure that it is called by the watcher for the VMThread
4896   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4897   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4898 
4899   PcFetcher fetcher(thread);
4900   fetcher.run();
4901   return fetcher.result();
4902 }
4903 
4904 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4905 {
4906    if (is_NPTL()) {
4907       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4908    } else {
4909       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4910       // word back to default 64bit precision if condvar is signaled. Java
4911       // wants 53bit precision.  Save and restore current value.
4912       int fpu = get_fpu_control_word();
4913       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4914       set_fpu_control_word(fpu);
4915       return status;
4916    }
4917 }
4918 
4919 ////////////////////////////////////////////////////////////////////////////////
4920 // debug support
4921 
4922 bool os::find(address addr, outputStream* st) {
4923   Dl_info dlinfo;
4924   memset(&dlinfo, 0, sizeof(dlinfo));
4925   if (dladdr(addr, &dlinfo) != 0) {
4926     st->print(PTR_FORMAT ": ", addr);
4927     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4928       st->print("%s+%#x", dlinfo.dli_sname,
4929                  addr - (intptr_t)dlinfo.dli_saddr);
4930     } else if (dlinfo.dli_fbase != NULL) {
4931       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4932     } else {
4933       st->print("<absolute address>");
4934     }
4935     if (dlinfo.dli_fname != NULL) {
4936       st->print(" in %s", dlinfo.dli_fname);
4937     }
4938     if (dlinfo.dli_fbase != NULL) {
4939       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4940     }
4941     st->cr();
4942 
4943     if (Verbose) {
4944       // decode some bytes around the PC
4945       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4946       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4947       address       lowest = (address) dlinfo.dli_sname;
4948       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4949       if (begin < lowest)  begin = lowest;
4950       Dl_info dlinfo2;
4951       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4952           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4953         end = (address) dlinfo2.dli_saddr;
4954       Disassembler::decode(begin, end, st);
4955     }
4956     return true;
4957   }
4958   return false;
4959 }
4960 
4961 ////////////////////////////////////////////////////////////////////////////////
4962 // misc
4963 
4964 // This does not do anything on Linux. This is basically a hook for being
4965 // able to use structured exception handling (thread-local exception filters)
4966 // on, e.g., Win32.
4967 void
4968 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4969                          JavaCallArguments* args, Thread* thread) {
4970   f(value, method, args, thread);
4971 }
4972 
4973 void os::print_statistics() {
4974 }
4975 
4976 int os::message_box(const char* title, const char* message) {
4977   int i;
4978   fdStream err(defaultStream::error_fd());
4979   for (i = 0; i < 78; i++) err.print_raw("=");
4980   err.cr();
4981   err.print_raw_cr(title);
4982   for (i = 0; i < 78; i++) err.print_raw("-");
4983   err.cr();
4984   err.print_raw_cr(message);
4985   for (i = 0; i < 78; i++) err.print_raw("=");
4986   err.cr();
4987 
4988   char buf[16];
4989   // Prevent process from exiting upon "read error" without consuming all CPU
4990   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4991 
4992   return buf[0] == 'y' || buf[0] == 'Y';
4993 }
4994 
4995 int os::stat(const char *path, struct stat *sbuf) {
4996   char pathbuf[MAX_PATH];
4997   if (strlen(path) > MAX_PATH - 1) {
4998     errno = ENAMETOOLONG;
4999     return -1;
5000   }
5001   os::native_path(strcpy(pathbuf, path));
5002   return ::stat(pathbuf, sbuf);
5003 }
5004 
5005 bool os::check_heap(bool force) {
5006   return true;
5007 }
5008 
5009 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5010   return ::vsnprintf(buf, count, format, args);
5011 }
5012 
5013 // Is a (classpath) directory empty?
5014 bool os::dir_is_empty(const char* path) {
5015   DIR *dir = NULL;
5016   struct dirent *ptr;
5017 
5018   dir = opendir(path);
5019   if (dir == NULL) return true;
5020 
5021   /* Scan the directory */
5022   bool result = true;
5023   char buf[sizeof(struct dirent) + MAX_PATH];
5024   while (result && (ptr = ::readdir(dir)) != NULL) {
5025     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5026       result = false;
5027     }
5028   }
5029   closedir(dir);
5030   return result;
5031 }
5032 
5033 // This code originates from JDK's sysOpen and open64_w
5034 // from src/solaris/hpi/src/system_md.c
5035 
5036 #ifndef O_DELETE
5037 #define O_DELETE 0x10000
5038 #endif
5039 
5040 // Open a file. Unlink the file immediately after open returns
5041 // if the specified oflag has the O_DELETE flag set.
5042 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5043 
5044 int os::open(const char *path, int oflag, int mode) {
5045 
5046   if (strlen(path) > MAX_PATH - 1) {
5047     errno = ENAMETOOLONG;
5048     return -1;
5049   }
5050   int fd;
5051   int o_delete = (oflag & O_DELETE);
5052   oflag = oflag & ~O_DELETE;
5053 
5054   fd = ::open64(path, oflag, mode);
5055   if (fd == -1) return -1;
5056 
5057   //If the open succeeded, the file might still be a directory
5058   {
5059     struct stat64 buf64;
5060     int ret = ::fstat64(fd, &buf64);
5061     int st_mode = buf64.st_mode;
5062 
5063     if (ret != -1) {
5064       if ((st_mode & S_IFMT) == S_IFDIR) {
5065         errno = EISDIR;
5066         ::close(fd);
5067         return -1;
5068       }
5069     } else {
5070       ::close(fd);
5071       return -1;
5072     }
5073   }
5074 
5075     /*
5076      * All file descriptors that are opened in the JVM and not
5077      * specifically destined for a subprocess should have the
5078      * close-on-exec flag set.  If we don't set it, then careless 3rd
5079      * party native code might fork and exec without closing all
5080      * appropriate file descriptors (e.g. as we do in closeDescriptors in
5081      * UNIXProcess.c), and this in turn might:
5082      *
5083      * - cause end-of-file to fail to be detected on some file
5084      *   descriptors, resulting in mysterious hangs, or
5085      *
5086      * - might cause an fopen in the subprocess to fail on a system
5087      *   suffering from bug 1085341.
5088      *
5089      * (Yes, the default setting of the close-on-exec flag is a Unix
5090      * design flaw)
5091      *
5092      * See:
5093      * 1085341: 32-bit stdio routines should support file descriptors >255
5094      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5095      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5096      */
5097 #ifdef FD_CLOEXEC
5098     {
5099         int flags = ::fcntl(fd, F_GETFD);
5100         if (flags != -1)
5101             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5102     }
5103 #endif
5104 
5105   if (o_delete != 0) {
5106     ::unlink(path);
5107   }
5108   return fd;
5109 }
5110 
5111 
5112 // create binary file, rewriting existing file if required
5113 int os::create_binary_file(const char* path, bool rewrite_existing) {
5114   int oflags = O_WRONLY | O_CREAT;
5115   if (!rewrite_existing) {
5116     oflags |= O_EXCL;
5117   }
5118   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5119 }
5120 
5121 // return current position of file pointer
5122 jlong os::current_file_offset(int fd) {
5123   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5124 }
5125 
5126 // move file pointer to the specified offset
5127 jlong os::seek_to_file_offset(int fd, jlong offset) {
5128   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5129 }
5130 
5131 // This code originates from JDK's sysAvailable
5132 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5133 
5134 int os::available(int fd, jlong *bytes) {
5135   jlong cur, end;
5136   int mode;
5137   struct stat64 buf64;
5138 
5139   if (::fstat64(fd, &buf64) >= 0) {
5140     mode = buf64.st_mode;
5141     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5142       /*
5143       * XXX: is the following call interruptible? If so, this might
5144       * need to go through the INTERRUPT_IO() wrapper as for other
5145       * blocking, interruptible calls in this file.
5146       */
5147       int n;
5148       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5149         *bytes = n;
5150         return 1;
5151       }
5152     }
5153   }
5154   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5155     return 0;
5156   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5157     return 0;
5158   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5159     return 0;
5160   }
5161   *bytes = end - cur;
5162   return 1;
5163 }
5164 
5165 int os::socket_available(int fd, jint *pbytes) {
5166   // Linux doc says EINTR not returned, unlike Solaris
5167   int ret = ::ioctl(fd, FIONREAD, pbytes);
5168 
5169   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5170   // is expected to return 0 on failure and 1 on success to the jdk.
5171   return (ret < 0) ? 0 : 1;
5172 }
5173 
5174 // Map a block of memory.
5175 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5176                      char *addr, size_t bytes, bool read_only,
5177                      bool allow_exec) {
5178   int prot;
5179   int flags = MAP_PRIVATE;
5180 
5181   if (read_only) {
5182     prot = PROT_READ;
5183   } else {
5184     prot = PROT_READ | PROT_WRITE;
5185   }
5186 
5187   if (allow_exec) {
5188     prot |= PROT_EXEC;
5189   }
5190 
5191   if (addr != NULL) {
5192     flags |= MAP_FIXED;
5193   }
5194 
5195   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5196                                      fd, file_offset);
5197   if (mapped_address == MAP_FAILED) {
5198     return NULL;
5199   }
5200   return mapped_address;
5201 }
5202 
5203 
5204 // Remap a block of memory.
5205 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5206                        char *addr, size_t bytes, bool read_only,
5207                        bool allow_exec) {
5208   // same as map_memory() on this OS
5209   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5210                         allow_exec);
5211 }
5212 
5213 
5214 // Unmap a block of memory.
5215 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5216   return munmap(addr, bytes) == 0;
5217 }
5218 
5219 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5220 
5221 static clockid_t thread_cpu_clockid(Thread* thread) {
5222   pthread_t tid = thread->osthread()->pthread_id();
5223   clockid_t clockid;
5224 
5225   // Get thread clockid
5226   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5227   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5228   return clockid;
5229 }
5230 
5231 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5232 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5233 // of a thread.
5234 //
5235 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5236 // the fast estimate available on the platform.
5237 
5238 jlong os::current_thread_cpu_time() {
5239   if (os::Linux::supports_fast_thread_cpu_time()) {
5240     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5241   } else {
5242     // return user + sys since the cost is the same
5243     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5244   }
5245 }
5246 
5247 jlong os::thread_cpu_time(Thread* thread) {
5248   // consistent with what current_thread_cpu_time() returns
5249   if (os::Linux::supports_fast_thread_cpu_time()) {
5250     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5251   } else {
5252     return slow_thread_cpu_time(thread, true /* user + sys */);
5253   }
5254 }
5255 
5256 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5257   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5258     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5259   } else {
5260     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5261   }
5262 }
5263 
5264 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5265   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5266     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5267   } else {
5268     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5269   }
5270 }
5271 
5272 //
5273 //  -1 on error.
5274 //
5275 
5276 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5277   pid_t  tid = thread->osthread()->thread_id();
5278   char *s;
5279   char stat[2048];
5280   int statlen;
5281   char proc_name[64];
5282   int count;
5283   long sys_time, user_time;
5284   char cdummy;
5285   int idummy;
5286   long ldummy;
5287   FILE *fp;
5288 
5289   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5290   fp = fopen(proc_name, "r");
5291   if ( fp == NULL ) return -1;
5292   statlen = fread(stat, 1, 2047, fp);
5293   stat[statlen] = '\0';
5294   fclose(fp);
5295 
5296   // Skip pid and the command string. Note that we could be dealing with
5297   // weird command names, e.g. user could decide to rename java launcher
5298   // to "java 1.4.2 :)", then the stat file would look like
5299   //                1234 (java 1.4.2 :)) R ... ...
5300   // We don't really need to know the command string, just find the last
5301   // occurrence of ")" and then start parsing from there. See bug 4726580.
5302   s = strrchr(stat, ')');
5303   if (s == NULL ) return -1;
5304 
5305   // Skip blank chars
5306   do s++; while (isspace(*s));
5307 
5308   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5309                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5310                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5311                  &user_time, &sys_time);
5312   if ( count != 13 ) return -1;
5313   if (user_sys_cpu_time) {
5314     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5315   } else {
5316     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5317   }
5318 }
5319 
5320 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5321   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5322   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5323   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5324   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5325 }
5326 
5327 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5328   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5329   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5330   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5331   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5332 }
5333 
5334 bool os::is_thread_cpu_time_supported() {
5335   return true;
5336 }
5337 
5338 // System loadavg support.  Returns -1 if load average cannot be obtained.
5339 // Linux doesn't yet have a (official) notion of processor sets,
5340 // so just return the system wide load average.
5341 int os::loadavg(double loadavg[], int nelem) {
5342   return ::getloadavg(loadavg, nelem);
5343 }
5344 
5345 void os::pause() {
5346   char filename[MAX_PATH];
5347   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5348     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5349   } else {
5350     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5351   }
5352 
5353   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5354   if (fd != -1) {
5355     struct stat buf;
5356     ::close(fd);
5357     while (::stat(filename, &buf) == 0) {
5358       (void)::poll(NULL, 0, 100);
5359     }
5360   } else {
5361     jio_fprintf(stderr,
5362       "Could not open pause file '%s', continuing immediately.\n", filename);
5363   }
5364 }
5365 
5366 
5367 // Refer to the comments in os_solaris.cpp park-unpark.
5368 //
5369 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5370 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5371 // For specifics regarding the bug see GLIBC BUGID 261237 :
5372 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5373 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5374 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5375 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5376 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5377 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5378 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5379 // of libpthread avoids the problem, but isn't practical.
5380 //
5381 // Possible remedies:
5382 //
5383 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5384 //      This is palliative and probabilistic, however.  If the thread is preempted
5385 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5386 //      than the minimum period may have passed, and the abstime may be stale (in the
5387 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5388 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5389 //
5390 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5391 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5392 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5393 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5394 //      thread.
5395 //
5396 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5397 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5398 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5399 //      This also works well.  In fact it avoids kernel-level scalability impediments
5400 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5401 //      timers in a graceful fashion.
5402 //
5403 // 4.   When the abstime value is in the past it appears that control returns
5404 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5405 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5406 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5407 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5408 //      It may be possible to avoid reinitialization by checking the return
5409 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5410 //      condvar we must establish the invariant that cond_signal() is only called
5411 //      within critical sections protected by the adjunct mutex.  This prevents
5412 //      cond_signal() from "seeing" a condvar that's in the midst of being
5413 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5414 //      desirable signal-after-unlock optimization that avoids futile context switching.
5415 //
5416 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5417 //      structure when a condvar is used or initialized.  cond_destroy()  would
5418 //      release the helper structure.  Our reinitialize-after-timedwait fix
5419 //      put excessive stress on malloc/free and locks protecting the c-heap.
5420 //
5421 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5422 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5423 // and only enabling the work-around for vulnerable environments.
5424 
5425 // utility to compute the abstime argument to timedwait:
5426 // millis is the relative timeout time
5427 // abstime will be the absolute timeout time
5428 // TODO: replace compute_abstime() with unpackTime()
5429 
5430 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5431   if (millis < 0)  millis = 0;
5432 
5433   jlong seconds = millis / 1000;
5434   millis %= 1000;
5435   if (seconds > 50000000) { // see man cond_timedwait(3T)
5436     seconds = 50000000;
5437   }
5438 
5439   if (os::supports_monotonic_clock()) {
5440     struct timespec now;
5441     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5442     assert_status(status == 0, status, "clock_gettime");
5443     abstime->tv_sec = now.tv_sec  + seconds;
5444     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5445     if (nanos >= NANOSECS_PER_SEC) {
5446       abstime->tv_sec += 1;
5447       nanos -= NANOSECS_PER_SEC;
5448     }
5449     abstime->tv_nsec = nanos;
5450   } else {
5451     struct timeval now;
5452     int status = gettimeofday(&now, NULL);
5453     assert(status == 0, "gettimeofday");
5454     abstime->tv_sec = now.tv_sec  + seconds;
5455     long usec = now.tv_usec + millis * 1000;
5456     if (usec >= 1000000) {
5457       abstime->tv_sec += 1;
5458       usec -= 1000000;
5459     }
5460     abstime->tv_nsec = usec * 1000;
5461   }
5462   return abstime;
5463 }
5464 
5465 
5466 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5467 // Conceptually TryPark() should be equivalent to park(0).
5468 
5469 int os::PlatformEvent::TryPark() {
5470   for (;;) {
5471     const int v = _Event ;
5472     guarantee ((v == 0) || (v == 1), "invariant") ;
5473     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5474   }
5475 }
5476 
5477 void os::PlatformEvent::park() {       // AKA "down()"
5478   // Invariant: Only the thread associated with the Event/PlatformEvent
5479   // may call park().
5480   // TODO: assert that _Assoc != NULL or _Assoc == Self
5481   int v ;
5482   for (;;) {
5483       v = _Event ;
5484       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5485   }
5486   guarantee (v >= 0, "invariant") ;
5487   if (v == 0) {
5488      // Do this the hard way by blocking ...
5489      int status = pthread_mutex_lock(_mutex);
5490      assert_status(status == 0, status, "mutex_lock");
5491      guarantee (_nParked == 0, "invariant") ;
5492      ++ _nParked ;
5493      while (_Event < 0) {
5494         status = pthread_cond_wait(_cond, _mutex);
5495         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5496         // Treat this the same as if the wait was interrupted
5497         if (status == ETIME) { status = EINTR; }
5498         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5499      }
5500      -- _nParked ;
5501 
5502     _Event = 0 ;
5503      status = pthread_mutex_unlock(_mutex);
5504      assert_status(status == 0, status, "mutex_unlock");
5505     // Paranoia to ensure our locked and lock-free paths interact
5506     // correctly with each other.
5507     OrderAccess::fence();
5508   }
5509   guarantee (_Event >= 0, "invariant") ;
5510 }
5511 
5512 int os::PlatformEvent::park(jlong millis) {
5513   guarantee (_nParked == 0, "invariant") ;
5514 
5515   int v ;
5516   for (;;) {
5517       v = _Event ;
5518       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5519   }
5520   guarantee (v >= 0, "invariant") ;
5521   if (v != 0) return OS_OK ;
5522 
5523   // We do this the hard way, by blocking the thread.
5524   // Consider enforcing a minimum timeout value.
5525   struct timespec abst;
5526   compute_abstime(&abst, millis);
5527 
5528   int ret = OS_TIMEOUT;
5529   int status = pthread_mutex_lock(_mutex);
5530   assert_status(status == 0, status, "mutex_lock");
5531   guarantee (_nParked == 0, "invariant") ;
5532   ++_nParked ;
5533 
5534   // Object.wait(timo) will return because of
5535   // (a) notification
5536   // (b) timeout
5537   // (c) thread.interrupt
5538   //
5539   // Thread.interrupt and object.notify{All} both call Event::set.
5540   // That is, we treat thread.interrupt as a special case of notification.
5541   // The underlying Solaris implementation, cond_timedwait, admits
5542   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5543   // JVM from making those visible to Java code.  As such, we must
5544   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5545   //
5546   // TODO: properly differentiate simultaneous notify+interrupt.
5547   // In that case, we should propagate the notify to another waiter.
5548 
5549   while (_Event < 0) {
5550     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5551     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5552       pthread_cond_destroy (_cond);
5553       pthread_cond_init (_cond, os::Linux::condAttr()) ;
5554     }
5555     assert_status(status == 0 || status == EINTR ||
5556                   status == ETIME || status == ETIMEDOUT,
5557                   status, "cond_timedwait");
5558     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5559     if (status == ETIME || status == ETIMEDOUT) break ;
5560     // We consume and ignore EINTR and spurious wakeups.
5561   }
5562   --_nParked ;
5563   if (_Event >= 0) {
5564      ret = OS_OK;
5565   }
5566   _Event = 0 ;
5567   status = pthread_mutex_unlock(_mutex);
5568   assert_status(status == 0, status, "mutex_unlock");
5569   assert (_nParked == 0, "invariant") ;
5570   // Paranoia to ensure our locked and lock-free paths interact
5571   // correctly with each other.
5572   OrderAccess::fence();
5573   return ret;
5574 }
5575 
5576 void os::PlatformEvent::unpark() {
5577   // Transitions for _Event:
5578   //    0 :=> 1
5579   //    1 :=> 1
5580   //   -1 :=> either 0 or 1; must signal target thread
5581   //          That is, we can safely transition _Event from -1 to either
5582   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
5583   //          unpark() calls.
5584   // See also: "Semaphores in Plan 9" by Mullender & Cox
5585   //
5586   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5587   // that it will take two back-to-back park() calls for the owning
5588   // thread to block. This has the benefit of forcing a spurious return
5589   // from the first park() call after an unpark() call which will help
5590   // shake out uses of park() and unpark() without condition variables.
5591 
5592   if (Atomic::xchg(1, &_Event) >= 0) return;
5593 
5594   // Wait for the thread associated with the event to vacate
5595   int status = pthread_mutex_lock(_mutex);
5596   assert_status(status == 0, status, "mutex_lock");
5597   int AnyWaiters = _nParked;
5598   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5599   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5600     AnyWaiters = 0;
5601     pthread_cond_signal(_cond);
5602   }
5603   status = pthread_mutex_unlock(_mutex);
5604   assert_status(status == 0, status, "mutex_unlock");
5605   if (AnyWaiters != 0) {
5606     status = pthread_cond_signal(_cond);
5607     assert_status(status == 0, status, "cond_signal");
5608   }
5609 
5610   // Note that we signal() _after dropping the lock for "immortal" Events.
5611   // This is safe and avoids a common class of  futile wakeups.  In rare
5612   // circumstances this can cause a thread to return prematurely from
5613   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5614   // simply re-test the condition and re-park itself.
5615 }
5616 
5617 
5618 // JSR166
5619 // -------------------------------------------------------
5620 
5621 /*
5622  * The solaris and linux implementations of park/unpark are fairly
5623  * conservative for now, but can be improved. They currently use a
5624  * mutex/condvar pair, plus a a count.
5625  * Park decrements count if > 0, else does a condvar wait.  Unpark
5626  * sets count to 1 and signals condvar.  Only one thread ever waits
5627  * on the condvar. Contention seen when trying to park implies that someone
5628  * is unparking you, so don't wait. And spurious returns are fine, so there
5629  * is no need to track notifications.
5630  */
5631 
5632 /*
5633  * This code is common to linux and solaris and will be moved to a
5634  * common place in dolphin.
5635  *
5636  * The passed in time value is either a relative time in nanoseconds
5637  * or an absolute time in milliseconds. Either way it has to be unpacked
5638  * into suitable seconds and nanoseconds components and stored in the
5639  * given timespec structure.
5640  * Given time is a 64-bit value and the time_t used in the timespec is only
5641  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5642  * overflow if times way in the future are given. Further on Solaris versions
5643  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5644  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5645  * As it will be 28 years before "now + 100000000" will overflow we can
5646  * ignore overflow and just impose a hard-limit on seconds using the value
5647  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5648  * years from "now".
5649  */
5650 
5651 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5652   assert (time > 0, "convertTime");
5653   time_t max_secs = 0;
5654 
5655   if (!os::supports_monotonic_clock() || isAbsolute) {
5656     struct timeval now;
5657     int status = gettimeofday(&now, NULL);
5658     assert(status == 0, "gettimeofday");
5659 
5660     max_secs = now.tv_sec + MAX_SECS;
5661 
5662     if (isAbsolute) {
5663       jlong secs = time / 1000;
5664       if (secs > max_secs) {
5665         absTime->tv_sec = max_secs;
5666       } else {
5667         absTime->tv_sec = secs;
5668       }
5669       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5670     } else {
5671       jlong secs = time / NANOSECS_PER_SEC;
5672       if (secs >= MAX_SECS) {
5673         absTime->tv_sec = max_secs;
5674         absTime->tv_nsec = 0;
5675       } else {
5676         absTime->tv_sec = now.tv_sec + secs;
5677         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5678         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5679           absTime->tv_nsec -= NANOSECS_PER_SEC;
5680           ++absTime->tv_sec; // note: this must be <= max_secs
5681         }
5682       }
5683     }
5684   } else {
5685     // must be relative using monotonic clock
5686     struct timespec now;
5687     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5688     assert_status(status == 0, status, "clock_gettime");
5689     max_secs = now.tv_sec + MAX_SECS;
5690     jlong secs = time / NANOSECS_PER_SEC;
5691     if (secs >= MAX_SECS) {
5692       absTime->tv_sec = max_secs;
5693       absTime->tv_nsec = 0;
5694     } else {
5695       absTime->tv_sec = now.tv_sec + secs;
5696       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5697       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5698         absTime->tv_nsec -= NANOSECS_PER_SEC;
5699         ++absTime->tv_sec; // note: this must be <= max_secs
5700       }
5701     }
5702   }
5703   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5704   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5705   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5706   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5707 }
5708 
5709 void Parker::park(bool isAbsolute, jlong time) {
5710   // Ideally we'd do something useful while spinning, such
5711   // as calling unpackTime().
5712 
5713   // Optional fast-path check:
5714   // Return immediately if a permit is available.
5715   // We depend on Atomic::xchg() having full barrier semantics
5716   // since we are doing a lock-free update to _counter.
5717   if (Atomic::xchg(0, &_counter) > 0) return;
5718 
5719   Thread* thread = Thread::current();
5720   assert(thread->is_Java_thread(), "Must be JavaThread");
5721   JavaThread *jt = (JavaThread *)thread;
5722 
5723   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5724   // Check interrupt before trying to wait
5725   if (Thread::is_interrupted(thread, false)) {
5726     return;
5727   }
5728 
5729   // Next, demultiplex/decode time arguments
5730   timespec absTime;
5731   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5732     return;
5733   }
5734   if (time > 0) {
5735     unpackTime(&absTime, isAbsolute, time);
5736   }
5737 
5738 
5739   // Enter safepoint region
5740   // Beware of deadlocks such as 6317397.
5741   // The per-thread Parker:: mutex is a classic leaf-lock.
5742   // In particular a thread must never block on the Threads_lock while
5743   // holding the Parker:: mutex.  If safepoints are pending both the
5744   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5745   ThreadBlockInVM tbivm(jt);
5746 
5747   // Don't wait if cannot get lock since interference arises from
5748   // unblocking.  Also. check interrupt before trying wait
5749   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5750     return;
5751   }
5752 
5753   int status ;
5754   if (_counter > 0)  { // no wait needed
5755     _counter = 0;
5756     status = pthread_mutex_unlock(_mutex);
5757     assert (status == 0, "invariant") ;
5758     // Paranoia to ensure our locked and lock-free paths interact
5759     // correctly with each other and Java-level accesses.
5760     OrderAccess::fence();
5761     return;
5762   }
5763 
5764 #ifdef ASSERT
5765   // Don't catch signals while blocked; let the running threads have the signals.
5766   // (This allows a debugger to break into the running thread.)
5767   sigset_t oldsigs;
5768   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5769   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5770 #endif
5771 
5772   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5773   jt->set_suspend_equivalent();
5774   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5775 
5776   assert(_cur_index == -1, "invariant");
5777   if (time == 0) {
5778     _cur_index = REL_INDEX; // arbitrary choice when not timed
5779     status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5780   } else {
5781     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5782     status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5783     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5784       pthread_cond_destroy (&_cond[_cur_index]) ;
5785       pthread_cond_init    (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5786     }
5787   }
5788   _cur_index = -1;
5789   assert_status(status == 0 || status == EINTR ||
5790                 status == ETIME || status == ETIMEDOUT,
5791                 status, "cond_timedwait");
5792 
5793 #ifdef ASSERT
5794   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5795 #endif
5796 
5797   _counter = 0 ;
5798   status = pthread_mutex_unlock(_mutex) ;
5799   assert_status(status == 0, status, "invariant") ;
5800   // Paranoia to ensure our locked and lock-free paths interact
5801   // correctly with each other and Java-level accesses.
5802   OrderAccess::fence();
5803 
5804   // If externally suspended while waiting, re-suspend
5805   if (jt->handle_special_suspend_equivalent_condition()) {
5806     jt->java_suspend_self();
5807   }
5808 }
5809 
5810 void Parker::unpark() {
5811   int s, status ;
5812   status = pthread_mutex_lock(_mutex);
5813   assert (status == 0, "invariant") ;
5814   s = _counter;
5815   _counter = 1;
5816   if (s < 1) {
5817     // thread might be parked
5818     if (_cur_index != -1) {
5819       // thread is definitely parked
5820       if (WorkAroundNPTLTimedWaitHang) {
5821         status = pthread_cond_signal (&_cond[_cur_index]);
5822         assert (status == 0, "invariant");
5823         status = pthread_mutex_unlock(_mutex);
5824         assert (status == 0, "invariant");
5825       } else {
5826         status = pthread_mutex_unlock(_mutex);
5827         assert (status == 0, "invariant");
5828         status = pthread_cond_signal (&_cond[_cur_index]);
5829         assert (status == 0, "invariant");
5830       }
5831     } else {
5832       pthread_mutex_unlock(_mutex);
5833       assert (status == 0, "invariant") ;
5834     }
5835   } else {
5836     pthread_mutex_unlock(_mutex);
5837     assert (status == 0, "invariant") ;
5838   }
5839 }
5840 
5841 
5842 extern char** environ;
5843 
5844 #ifndef __NR_fork
5845 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5846 #endif
5847 
5848 #ifndef __NR_execve
5849 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5850 #endif
5851 
5852 // Run the specified command in a separate process. Return its exit value,
5853 // or -1 on failure (e.g. can't fork a new process).
5854 // Unlike system(), this function can be called from signal handler. It
5855 // doesn't block SIGINT et al.
5856 int os::fork_and_exec(char* cmd) {
5857   const char * argv[4] = {"sh", "-c", cmd, NULL};
5858 
5859   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5860   // pthread_atfork handlers and reset pthread library. All we need is a
5861   // separate process to execve. Make a direct syscall to fork process.
5862   // On IA64 there's no fork syscall, we have to use fork() and hope for
5863   // the best...
5864   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5865               IA64_ONLY(fork();)
5866 
5867   if (pid < 0) {
5868     // fork failed
5869     return -1;
5870 
5871   } else if (pid == 0) {
5872     // child process
5873 
5874     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5875     // first to kill every thread on the thread list. Because this list is
5876     // not reset by fork() (see notes above), execve() will instead kill
5877     // every thread in the parent process. We know this is the only thread
5878     // in the new process, so make a system call directly.
5879     // IA64 should use normal execve() from glibc to match the glibc fork()
5880     // above.
5881     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5882     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5883 
5884     // execve failed
5885     _exit(-1);
5886 
5887   } else  {
5888     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5889     // care about the actual exit code, for now.
5890 
5891     int status;
5892 
5893     // Wait for the child process to exit.  This returns immediately if
5894     // the child has already exited. */
5895     while (waitpid(pid, &status, 0) < 0) {
5896         switch (errno) {
5897         case ECHILD: return 0;
5898         case EINTR: break;
5899         default: return -1;
5900         }
5901     }
5902 
5903     if (WIFEXITED(status)) {
5904        // The child exited normally; get its exit code.
5905        return WEXITSTATUS(status);
5906     } else if (WIFSIGNALED(status)) {
5907        // The child exited because of a signal
5908        // The best value to return is 0x80 + signal number,
5909        // because that is what all Unix shells do, and because
5910        // it allows callers to distinguish between process exit and
5911        // process death by signal.
5912        return 0x80 + WTERMSIG(status);
5913     } else {
5914        // Unknown exit code; pass it through
5915        return status;
5916     }
5917   }
5918 }
5919 
5920 // is_headless_jre()
5921 //
5922 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5923 // in order to report if we are running in a headless jre
5924 //
5925 // Since JDK8 xawt/libmawt.so was moved into the same directory
5926 // as libawt.so, and renamed libawt_xawt.so
5927 //
5928 bool os::is_headless_jre() {
5929     struct stat statbuf;
5930     char buf[MAXPATHLEN];
5931     char libmawtpath[MAXPATHLEN];
5932     const char *xawtstr  = "/xawt/libmawt.so";
5933     const char *new_xawtstr = "/libawt_xawt.so";
5934     char *p;
5935 
5936     // Get path to libjvm.so
5937     os::jvm_path(buf, sizeof(buf));
5938 
5939     // Get rid of libjvm.so
5940     p = strrchr(buf, '/');
5941     if (p == NULL) return false;
5942     else *p = '\0';
5943 
5944     // Get rid of client or server
5945     p = strrchr(buf, '/');
5946     if (p == NULL) return false;
5947     else *p = '\0';
5948 
5949     // check xawt/libmawt.so
5950     strcpy(libmawtpath, buf);
5951     strcat(libmawtpath, xawtstr);
5952     if (::stat(libmawtpath, &statbuf) == 0) return false;
5953 
5954     // check libawt_xawt.so
5955     strcpy(libmawtpath, buf);
5956     strcat(libmawtpath, new_xawtstr);
5957     if (::stat(libmawtpath, &statbuf) == 0) return false;
5958 
5959     return true;
5960 }
5961 
5962 // Get the default path to the core file
5963 // Returns the length of the string
5964 int os::get_core_path(char* buffer, size_t bufferSize) {
5965   const char* p = get_current_directory(buffer, bufferSize);
5966 
5967   if (p == NULL) {
5968     assert(p != NULL, "failed to get current directory");
5969     return 0;
5970   }
5971 
5972   return strlen(buffer);
5973 }
5974 
5975 #ifdef JAVASE_EMBEDDED
5976 //
5977 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5978 //
5979 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5980 
5981 // ctor
5982 //
5983 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5984   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5985   _fd = fd;
5986 
5987   if (os::create_thread(this, os::os_thread)) {
5988     _memnotify_thread = this;
5989     os::set_priority(this, NearMaxPriority);
5990     os::start_thread(this);
5991   }
5992 }
5993 
5994 // Where all the work gets done
5995 //
5996 void MemNotifyThread::run() {
5997   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5998 
5999   // Set up the select arguments
6000   fd_set rfds;
6001   if (_fd != -1) {
6002     FD_ZERO(&rfds);
6003     FD_SET(_fd, &rfds);
6004   }
6005 
6006   // Now wait for the mem_notify device to wake up
6007   while (1) {
6008     // Wait for the mem_notify device to signal us..
6009     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6010     if (rc == -1) {
6011       perror("select!\n");
6012       break;
6013     } else if (rc) {
6014       //ssize_t free_before = os::available_memory();
6015       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6016 
6017       // The kernel is telling us there is not much memory left...
6018       // try to do something about that
6019 
6020       // If we are not already in a GC, try one.
6021       if (!Universe::heap()->is_gc_active()) {
6022         Universe::heap()->collect(GCCause::_allocation_failure);
6023 
6024         //ssize_t free_after = os::available_memory();
6025         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6026         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6027       }
6028       // We might want to do something like the following if we find the GC's are not helping...
6029       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6030     }
6031   }
6032 }
6033 
6034 //
6035 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6036 //
6037 void MemNotifyThread::start() {
6038   int    fd;
6039   fd = open ("/dev/mem_notify", O_RDONLY, 0);
6040   if (fd < 0) {
6041       return;
6042   }
6043 
6044   if (memnotify_thread() == NULL) {
6045     new MemNotifyThread(fd);
6046   }
6047 }
6048 
6049 #endif // JAVASE_EMBEDDED
6050 
6051 
6052 /////////////// Unit tests ///////////////
6053 
6054 #ifndef PRODUCT
6055 
6056 #define test_log(...) \
6057   do {\
6058     if (VerboseInternalVMTests) { \
6059       tty->print_cr(__VA_ARGS__); \
6060       tty->flush(); \
6061     }\
6062   } while (false)
6063 
6064 class TestReserveMemorySpecial : AllStatic {
6065  public:
6066   static void small_page_write(void* addr, size_t size) {
6067     size_t page_size = os::vm_page_size();
6068 
6069     char* end = (char*)addr + size;
6070     for (char* p = (char*)addr; p < end; p += page_size) {
6071       *p = 1;
6072     }
6073   }
6074 
6075   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6076     if (!UseHugeTLBFS) {
6077       return;
6078     }
6079 
6080     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6081 
6082     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6083 
6084     if (addr != NULL) {
6085       small_page_write(addr, size);
6086 
6087       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6088     }
6089   }
6090 
6091   static void test_reserve_memory_special_huge_tlbfs_only() {
6092     if (!UseHugeTLBFS) {
6093       return;
6094     }
6095 
6096     size_t lp = os::large_page_size();
6097 
6098     for (size_t size = lp; size <= lp * 10; size += lp) {
6099       test_reserve_memory_special_huge_tlbfs_only(size);
6100     }
6101   }
6102 
6103   static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6104     if (!UseHugeTLBFS) {
6105         return;
6106     }
6107 
6108     test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6109         size, alignment);
6110 
6111     assert(size >= os::large_page_size(), "Incorrect input to test");
6112 
6113     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6114 
6115     if (addr != NULL) {
6116       small_page_write(addr, size);
6117 
6118       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6119     }
6120   }
6121 
6122   static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6123     size_t lp = os::large_page_size();
6124     size_t ag = os::vm_allocation_granularity();
6125 
6126     for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6127       test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6128     }
6129   }
6130 
6131   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6132     size_t lp = os::large_page_size();
6133     size_t ag = os::vm_allocation_granularity();
6134 
6135     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6136     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6137     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6138     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6139     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6140     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6141     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6142     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6143     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6144   }
6145 
6146   static void test_reserve_memory_special_huge_tlbfs() {
6147     if (!UseHugeTLBFS) {
6148       return;
6149     }
6150 
6151     test_reserve_memory_special_huge_tlbfs_only();
6152     test_reserve_memory_special_huge_tlbfs_mixed();
6153   }
6154 
6155   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6156     if (!UseSHM) {
6157       return;
6158     }
6159 
6160     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6161 
6162     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6163 
6164     if (addr != NULL) {
6165       assert(is_ptr_aligned(addr, alignment), "Check");
6166       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6167 
6168       small_page_write(addr, size);
6169 
6170       os::Linux::release_memory_special_shm(addr, size);
6171     }
6172   }
6173 
6174   static void test_reserve_memory_special_shm() {
6175     size_t lp = os::large_page_size();
6176     size_t ag = os::vm_allocation_granularity();
6177 
6178     for (size_t size = ag; size < lp * 3; size += ag) {
6179       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6180         test_reserve_memory_special_shm(size, alignment);
6181       }
6182     }
6183   }
6184 
6185   static void test() {
6186     test_reserve_memory_special_huge_tlbfs();
6187     test_reserve_memory_special_shm();
6188   }
6189 };
6190 
6191 void TestReserveMemorySpecial_test() {
6192   TestReserveMemorySpecial::test();
6193 }
6194 
6195 #endif
--- EOF ---