1 /*
   2  * Copyright (c) 1999, 2014, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "memory/allocation.inline.hpp"
  36 #include "memory/filemap.hpp"
  37 #include "mutex_linux.inline.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_share_linux.hpp"
  40 #include "prims/jniFastGetField.hpp"
  41 #include "prims/jvm.h"
  42 #include "prims/jvm_misc.hpp"
  43 #include "runtime/arguments.hpp"
  44 #include "runtime/extendedPC.hpp"
  45 #include "runtime/globals.hpp"
  46 #include "runtime/interfaceSupport.hpp"
  47 #include "runtime/init.hpp"
  48 #include "runtime/java.hpp"
  49 #include "runtime/javaCalls.hpp"
  50 #include "runtime/mutexLocker.hpp"
  51 #include "runtime/objectMonitor.hpp"
  52 #include "runtime/osThread.hpp"
  53 #include "runtime/perfMemory.hpp"
  54 #include "runtime/sharedRuntime.hpp"
  55 #include "runtime/statSampler.hpp"
  56 #include "runtime/stubRoutines.hpp"
  57 #include "runtime/thread.inline.hpp"
  58 #include "runtime/threadCritical.hpp"
  59 #include "runtime/timer.hpp"
  60 #include "services/attachListener.hpp"
  61 #include "services/memTracker.hpp"
  62 #include "services/runtimeService.hpp"
  63 #include "utilities/decoder.hpp"
  64 #include "utilities/defaultStream.hpp"
  65 #include "utilities/events.hpp"
  66 #include "utilities/elfFile.hpp"
  67 #include "utilities/growableArray.hpp"
  68 #include "utilities/vmError.hpp"
  69 
  70 // put OS-includes here
  71 # include <sys/types.h>
  72 # include <sys/mman.h>
  73 # include <sys/stat.h>
  74 # include <sys/select.h>
  75 # include <pthread.h>
  76 # include <signal.h>
  77 # include <errno.h>
  78 # include <dlfcn.h>
  79 # include <stdio.h>
  80 # include <unistd.h>
  81 # include <sys/resource.h>
  82 # include <pthread.h>
  83 # include <sys/stat.h>
  84 # include <sys/time.h>
  85 # include <sys/times.h>
  86 # include <sys/utsname.h>
  87 # include <sys/socket.h>
  88 # include <sys/wait.h>
  89 # include <pwd.h>
  90 # include <poll.h>
  91 # include <semaphore.h>
  92 # include <fcntl.h>
  93 # include <string.h>
  94 # include <syscall.h>
  95 # include <sys/sysinfo.h>
  96 # include <gnu/libc-version.h>
  97 # include <sys/ipc.h>
  98 # include <sys/shm.h>
  99 # include <link.h>
 100 # include <stdint.h>
 101 # include <inttypes.h>
 102 # include <sys/ioctl.h>
 103 
 104 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 105 // getrusage() is prepared to handle the associated failure.
 106 #ifndef RUSAGE_THREAD
 107 #define RUSAGE_THREAD   (1)               /* only the calling thread */
 108 #endif
 109 
 110 #define MAX_PATH    (2 * K)
 111 
 112 // for timer info max values which include all bits
 113 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 114 
 115 #define LARGEPAGES_BIT (1 << 6)
 116 ////////////////////////////////////////////////////////////////////////////////
 117 // global variables
 118 julong os::Linux::_physical_memory = 0;
 119 
 120 address   os::Linux::_initial_thread_stack_bottom = NULL;
 121 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 122 
 123 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 124 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 125 Mutex* os::Linux::_createThread_lock = NULL;
 126 pthread_t os::Linux::_main_thread;
 127 int os::Linux::_page_size = -1;
 128 const int os::Linux::_vm_default_page_size = (8 * K);
 129 bool os::Linux::_is_floating_stack = false;
 130 bool os::Linux::_is_NPTL = false;
 131 bool os::Linux::_supports_fast_thread_cpu_time = false;
 132 const char * os::Linux::_glibc_version = NULL;
 133 const char * os::Linux::_libpthread_version = NULL;
 134 pthread_condattr_t os::Linux::_condattr[1];
 135 
 136 static jlong initial_time_count=0;
 137 
 138 static int clock_tics_per_sec = 100;
 139 
 140 // For diagnostics to print a message once. see run_periodic_checks
 141 static sigset_t check_signal_done;
 142 static bool check_signals = true;;
 143 
 144 static pid_t _initial_pid = 0;
 145 
 146 /* Signal number used to suspend/resume a thread */
 147 
 148 /* do not use any signal number less than SIGSEGV, see 4355769 */
 149 static int SR_signum = SIGUSR2;
 150 sigset_t SR_sigset;
 151 
 152 /* Used to protect dlsym() calls */
 153 static pthread_mutex_t dl_mutex;
 154 
 155 // Declarations
 156 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 157 
 158 #ifdef JAVASE_EMBEDDED
 159 class MemNotifyThread: public Thread {
 160   friend class VMStructs;
 161  public:
 162   virtual void run();
 163 
 164  private:
 165   static MemNotifyThread* _memnotify_thread;
 166   int _fd;
 167 
 168  public:
 169 
 170   // Constructor
 171   MemNotifyThread(int fd);
 172 
 173   // Tester
 174   bool is_memnotify_thread() const { return true; }
 175 
 176   // Printing
 177   char* name() const { return (char*)"Linux MemNotify Thread"; }
 178 
 179   // Returns the single instance of the MemNotifyThread
 180   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 181 
 182   // Create and start the single instance of MemNotifyThread
 183   static void start();
 184 };
 185 #endif // JAVASE_EMBEDDED
 186 
 187 // utility functions
 188 
 189 static int SR_initialize();
 190 
 191 julong os::available_memory() {
 192   return Linux::available_memory();
 193 }
 194 
 195 julong os::Linux::available_memory() {
 196   // values in struct sysinfo are "unsigned long"
 197   struct sysinfo si;
 198   sysinfo(&si);
 199 
 200   return (julong)si.freeram * si.mem_unit;
 201 }
 202 
 203 julong os::physical_memory() {
 204   return Linux::physical_memory();
 205 }
 206 
 207 ////////////////////////////////////////////////////////////////////////////////
 208 // environment support
 209 
 210 bool os::getenv(const char* name, char* buf, int len) {
 211   const char* val = ::getenv(name);
 212   if (val != NULL && strlen(val) < (size_t)len) {
 213     strcpy(buf, val);
 214     return true;
 215   }
 216   if (len > 0) buf[0] = 0;  // return a null string
 217   return false;
 218 }
 219 
 220 
 221 // Return true if user is running as root.
 222 
 223 bool os::have_special_privileges() {
 224   static bool init = false;
 225   static bool privileges = false;
 226   if (!init) {
 227     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 228     init = true;
 229   }
 230   return privileges;
 231 }
 232 
 233 
 234 #ifndef SYS_gettid
 235 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 236 #ifdef __ia64__
 237 #define SYS_gettid 1105
 238 #elif __i386__
 239 #define SYS_gettid 224
 240 #elif __amd64__
 241 #define SYS_gettid 186
 242 #elif __sparc__
 243 #define SYS_gettid 143
 244 #else
 245 #error define gettid for the arch
 246 #endif
 247 #endif
 248 
 249 // Cpu architecture string
 250 #if   defined(ZERO)
 251 static char cpu_arch[] = ZERO_LIBARCH;
 252 #elif defined(IA64)
 253 static char cpu_arch[] = "ia64";
 254 #elif defined(IA32)
 255 static char cpu_arch[] = "i386";
 256 #elif defined(AMD64)
 257 static char cpu_arch[] = "amd64";
 258 #elif defined(ARM)
 259 static char cpu_arch[] = "arm";
 260 #elif defined(PPC)
 261 static char cpu_arch[] = "ppc";
 262 #elif defined(SPARC)
 263 #  ifdef _LP64
 264 static char cpu_arch[] = "sparcv9";
 265 #  else
 266 static char cpu_arch[] = "sparc";
 267 #  endif
 268 #else
 269 #error Add appropriate cpu_arch setting
 270 #endif
 271 
 272 
 273 // pid_t gettid()
 274 //
 275 // Returns the kernel thread id of the currently running thread. Kernel
 276 // thread id is used to access /proc.
 277 //
 278 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 279 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 280 //
 281 pid_t os::Linux::gettid() {
 282   int rslt = syscall(SYS_gettid);
 283   if (rslt == -1) {
 284      // old kernel, no NPTL support
 285      return getpid();
 286   } else {
 287      return (pid_t)rslt;
 288   }
 289 }
 290 
 291 // Most versions of linux have a bug where the number of processors are
 292 // determined by looking at the /proc file system.  In a chroot environment,
 293 // the system call returns 1.  This causes the VM to act as if it is
 294 // a single processor and elide locking (see is_MP() call).
 295 static bool unsafe_chroot_detected = false;
 296 static const char *unstable_chroot_error = "/proc file system not found.\n"
 297                      "Java may be unstable running multithreaded in a chroot "
 298                      "environment on Linux when /proc filesystem is not mounted.";
 299 
 300 void os::Linux::initialize_system_info() {
 301   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 302   if (processor_count() == 1) {
 303     pid_t pid = os::Linux::gettid();
 304     char fname[32];
 305     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 306     FILE *fp = fopen(fname, "r");
 307     if (fp == NULL) {
 308       unsafe_chroot_detected = true;
 309     } else {
 310       fclose(fp);
 311     }
 312   }
 313   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 314   assert(processor_count() > 0, "linux error");
 315 }
 316 
 317 void os::init_system_properties_values() {
 318 //  char arch[12];
 319 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 320 
 321   // The next steps are taken in the product version:
 322   //
 323   // Obtain the JAVA_HOME value from the location of libjvm.so.
 324   // This library should be located at:
 325   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 326   //
 327   // If "/jre/lib/" appears at the right place in the path, then we
 328   // assume libjvm.so is installed in a JDK and we use this path.
 329   //
 330   // Otherwise exit with message: "Could not create the Java virtual machine."
 331   //
 332   // The following extra steps are taken in the debugging version:
 333   //
 334   // If "/jre/lib/" does NOT appear at the right place in the path
 335   // instead of exit check for $JAVA_HOME environment variable.
 336   //
 337   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 338   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 339   // it looks like libjvm.so is installed there
 340   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 341   //
 342   // Otherwise exit.
 343   //
 344   // Important note: if the location of libjvm.so changes this
 345   // code needs to be changed accordingly.
 346 
 347   // The next few definitions allow the code to be verbatim:
 348 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
 349 #define getenv(n) ::getenv(n)
 350 
 351 /*
 352  * See ld(1):
 353  *      The linker uses the following search paths to locate required
 354  *      shared libraries:
 355  *        1: ...
 356  *        ...
 357  *        7: The default directories, normally /lib and /usr/lib.
 358  */
 359 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 360 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 361 #else
 362 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 363 #endif
 364 
 365 #define EXTENSIONS_DIR  "/lib/ext"
 366 #define ENDORSED_DIR    "/lib/endorsed"
 367 #define REG_DIR         "/usr/java/packages"
 368 
 369   {
 370     /* sysclasspath, java_home, dll_dir */
 371     {
 372         char *home_path;
 373         char *dll_path;
 374         char *pslash;
 375         char buf[MAXPATHLEN];
 376         os::jvm_path(buf, sizeof(buf));
 377 
 378         // Found the full path to libjvm.so.
 379         // Now cut the path to <java_home>/jre if we can.
 380         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 381         pslash = strrchr(buf, '/');
 382         if (pslash != NULL)
 383             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 384         dll_path = malloc(strlen(buf) + 1);
 385         if (dll_path == NULL)
 386             return;
 387         strcpy(dll_path, buf);
 388         Arguments::set_dll_dir(dll_path);
 389 
 390         if (pslash != NULL) {
 391             pslash = strrchr(buf, '/');
 392             if (pslash != NULL) {
 393                 *pslash = '\0';       /* get rid of /<arch> */
 394                 pslash = strrchr(buf, '/');
 395                 if (pslash != NULL)
 396                     *pslash = '\0';   /* get rid of /lib */
 397             }
 398         }
 399 
 400         home_path = malloc(strlen(buf) + 1);
 401         if (home_path == NULL)
 402             return;
 403         strcpy(home_path, buf);
 404         Arguments::set_java_home(home_path);
 405 
 406         if (!set_boot_path('/', ':'))
 407             return;
 408     }
 409 
 410     /*
 411      * Where to look for native libraries
 412      *
 413      * Note: Due to a legacy implementation, most of the library path
 414      * is set in the launcher.  This was to accomodate linking restrictions
 415      * on legacy Linux implementations (which are no longer supported).
 416      * Eventually, all the library path setting will be done here.
 417      *
 418      * However, to prevent the proliferation of improperly built native
 419      * libraries, the new path component /usr/java/packages is added here.
 420      * Eventually, all the library path setting will be done here.
 421      */
 422     {
 423         char *ld_library_path;
 424 
 425         /*
 426          * Construct the invariant part of ld_library_path. Note that the
 427          * space for the colon and the trailing null are provided by the
 428          * nulls included by the sizeof operator (so actually we allocate
 429          * a byte more than necessary).
 430          */
 431         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 432             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 433         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 434 
 435         /*
 436          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 437          * should always exist (until the legacy problem cited above is
 438          * addressed).
 439          */
 440         char *v = getenv("LD_LIBRARY_PATH");
 441         if (v != NULL) {
 442             char *t = ld_library_path;
 443             /* That's +1 for the colon and +1 for the trailing '\0' */
 444             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 445             sprintf(ld_library_path, "%s:%s", v, t);
 446         }
 447         Arguments::set_library_path(ld_library_path);
 448     }
 449 
 450     /*
 451      * Extensions directories.
 452      *
 453      * Note that the space for the colon and the trailing null are provided
 454      * by the nulls included by the sizeof operator (so actually one byte more
 455      * than necessary is allocated).
 456      */
 457     {
 458         char *buf = malloc(strlen(Arguments::get_java_home()) +
 459             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 460         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 461             Arguments::get_java_home());
 462         Arguments::set_ext_dirs(buf);
 463     }
 464 
 465     /* Endorsed standards default directory. */
 466     {
 467         char * buf;
 468         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 469         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 470         Arguments::set_endorsed_dirs(buf);
 471     }
 472   }
 473 
 474 #undef malloc
 475 #undef getenv
 476 #undef EXTENSIONS_DIR
 477 #undef ENDORSED_DIR
 478 
 479   // Done
 480   return;
 481 }
 482 
 483 ////////////////////////////////////////////////////////////////////////////////
 484 // breakpoint support
 485 
 486 void os::breakpoint() {
 487   BREAKPOINT;
 488 }
 489 
 490 extern "C" void breakpoint() {
 491   // use debugger to set breakpoint here
 492 }
 493 
 494 ////////////////////////////////////////////////////////////////////////////////
 495 // signal support
 496 
 497 debug_only(static bool signal_sets_initialized = false);
 498 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 499 
 500 bool os::Linux::is_sig_ignored(int sig) {
 501       struct sigaction oact;
 502       sigaction(sig, (struct sigaction*)NULL, &oact);
 503       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 504                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 505       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 506            return true;
 507       else
 508            return false;
 509 }
 510 
 511 void os::Linux::signal_sets_init() {
 512   // Should also have an assertion stating we are still single-threaded.
 513   assert(!signal_sets_initialized, "Already initialized");
 514   // Fill in signals that are necessarily unblocked for all threads in
 515   // the VM. Currently, we unblock the following signals:
 516   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 517   //                         by -Xrs (=ReduceSignalUsage));
 518   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 519   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 520   // the dispositions or masks wrt these signals.
 521   // Programs embedding the VM that want to use the above signals for their
 522   // own purposes must, at this time, use the "-Xrs" option to prevent
 523   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 524   // (See bug 4345157, and other related bugs).
 525   // In reality, though, unblocking these signals is really a nop, since
 526   // these signals are not blocked by default.
 527   sigemptyset(&unblocked_sigs);
 528   sigemptyset(&allowdebug_blocked_sigs);
 529   sigaddset(&unblocked_sigs, SIGILL);
 530   sigaddset(&unblocked_sigs, SIGSEGV);
 531   sigaddset(&unblocked_sigs, SIGBUS);
 532   sigaddset(&unblocked_sigs, SIGFPE);
 533   sigaddset(&unblocked_sigs, SR_signum);
 534 
 535   if (!ReduceSignalUsage) {
 536    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 537       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 538       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 539    }
 540    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 541       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 542       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 543    }
 544    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 545       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 546       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 547    }
 548   }
 549   // Fill in signals that are blocked by all but the VM thread.
 550   sigemptyset(&vm_sigs);
 551   if (!ReduceSignalUsage)
 552     sigaddset(&vm_sigs, BREAK_SIGNAL);
 553   debug_only(signal_sets_initialized = true);
 554 
 555 }
 556 
 557 // These are signals that are unblocked while a thread is running Java.
 558 // (For some reason, they get blocked by default.)
 559 sigset_t* os::Linux::unblocked_signals() {
 560   assert(signal_sets_initialized, "Not initialized");
 561   return &unblocked_sigs;
 562 }
 563 
 564 // These are the signals that are blocked while a (non-VM) thread is
 565 // running Java. Only the VM thread handles these signals.
 566 sigset_t* os::Linux::vm_signals() {
 567   assert(signal_sets_initialized, "Not initialized");
 568   return &vm_sigs;
 569 }
 570 
 571 // These are signals that are blocked during cond_wait to allow debugger in
 572 sigset_t* os::Linux::allowdebug_blocked_signals() {
 573   assert(signal_sets_initialized, "Not initialized");
 574   return &allowdebug_blocked_sigs;
 575 }
 576 
 577 void os::Linux::hotspot_sigmask(Thread* thread) {
 578 
 579   //Save caller's signal mask before setting VM signal mask
 580   sigset_t caller_sigmask;
 581   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 582 
 583   OSThread* osthread = thread->osthread();
 584   osthread->set_caller_sigmask(caller_sigmask);
 585 
 586   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 587 
 588   if (!ReduceSignalUsage) {
 589     if (thread->is_VM_thread()) {
 590       // Only the VM thread handles BREAK_SIGNAL ...
 591       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 592     } else {
 593       // ... all other threads block BREAK_SIGNAL
 594       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 595     }
 596   }
 597 }
 598 
 599 //////////////////////////////////////////////////////////////////////////////
 600 // detecting pthread library
 601 
 602 void os::Linux::libpthread_init() {
 603   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 604   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 605   // generic name for earlier versions.
 606   // Define macros here so we can build HotSpot on old systems.
 607 # ifndef _CS_GNU_LIBC_VERSION
 608 # define _CS_GNU_LIBC_VERSION 2
 609 # endif
 610 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 611 # define _CS_GNU_LIBPTHREAD_VERSION 3
 612 # endif
 613 
 614   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 615   if (n > 0) {
 616      char *str = (char *)malloc(n, mtInternal);
 617      confstr(_CS_GNU_LIBC_VERSION, str, n);
 618      os::Linux::set_glibc_version(str);
 619   } else {
 620      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 621      static char _gnu_libc_version[32];
 622      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 623               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 624      os::Linux::set_glibc_version(_gnu_libc_version);
 625   }
 626 
 627   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 628   if (n > 0) {
 629      char *str = (char *)malloc(n, mtInternal);
 630      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 631      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 632      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 633      // is the case. LinuxThreads has a hard limit on max number of threads.
 634      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 635      // On the other hand, NPTL does not have such a limit, sysconf()
 636      // will return -1 and errno is not changed. Check if it is really NPTL.
 637      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 638          strstr(str, "NPTL") &&
 639          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 640        free(str);
 641        os::Linux::set_libpthread_version("linuxthreads");
 642      } else {
 643        os::Linux::set_libpthread_version(str);
 644      }
 645   } else {
 646     // glibc before 2.3.2 only has LinuxThreads.
 647     os::Linux::set_libpthread_version("linuxthreads");
 648   }
 649 
 650   if (strstr(libpthread_version(), "NPTL")) {
 651      os::Linux::set_is_NPTL();
 652   } else {
 653      os::Linux::set_is_LinuxThreads();
 654   }
 655 
 656   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 657   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 658   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 659      os::Linux::set_is_floating_stack();
 660   }
 661 }
 662 
 663 /////////////////////////////////////////////////////////////////////////////
 664 // thread stack
 665 
 666 // Force Linux kernel to expand current thread stack. If "bottom" is close
 667 // to the stack guard, caller should block all signals.
 668 //
 669 // MAP_GROWSDOWN:
 670 //   A special mmap() flag that is used to implement thread stacks. It tells
 671 //   kernel that the memory region should extend downwards when needed. This
 672 //   allows early versions of LinuxThreads to only mmap the first few pages
 673 //   when creating a new thread. Linux kernel will automatically expand thread
 674 //   stack as needed (on page faults).
 675 //
 676 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 677 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 678 //   region, it's hard to tell if the fault is due to a legitimate stack
 679 //   access or because of reading/writing non-exist memory (e.g. buffer
 680 //   overrun). As a rule, if the fault happens below current stack pointer,
 681 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 682 //   application (see Linux kernel fault.c).
 683 //
 684 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 685 //   stack overflow detection.
 686 //
 687 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 688 //   not use this flag. However, the stack of initial thread is not created
 689 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 690 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 691 //   and then attach the thread to JVM.
 692 //
 693 // To get around the problem and allow stack banging on Linux, we need to
 694 // manually expand thread stack after receiving the SIGSEGV.
 695 //
 696 // There are two ways to expand thread stack to address "bottom", we used
 697 // both of them in JVM before 1.5:
 698 //   1. adjust stack pointer first so that it is below "bottom", and then
 699 //      touch "bottom"
 700 //   2. mmap() the page in question
 701 //
 702 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 703 // if current sp is already near the lower end of page 101, and we need to
 704 // call mmap() to map page 100, it is possible that part of the mmap() frame
 705 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 706 // That will destroy the mmap() frame and cause VM to crash.
 707 //
 708 // The following code works by adjusting sp first, then accessing the "bottom"
 709 // page to force a page fault. Linux kernel will then automatically expand the
 710 // stack mapping.
 711 //
 712 // _expand_stack_to() assumes its frame size is less than page size, which
 713 // should always be true if the function is not inlined.
 714 
 715 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 716 #define NOINLINE
 717 #else
 718 #define NOINLINE __attribute__ ((noinline))
 719 #endif
 720 
 721 static void _expand_stack_to(address bottom) NOINLINE;
 722 
 723 static void _expand_stack_to(address bottom) {
 724   address sp;
 725   size_t size;
 726   volatile char *p;
 727 
 728   // Adjust bottom to point to the largest address within the same page, it
 729   // gives us a one-page buffer if alloca() allocates slightly more memory.
 730   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 731   bottom += os::Linux::page_size() - 1;
 732 
 733   // sp might be slightly above current stack pointer; if that's the case, we
 734   // will alloca() a little more space than necessary, which is OK. Don't use
 735   // os::current_stack_pointer(), as its result can be slightly below current
 736   // stack pointer, causing us to not alloca enough to reach "bottom".
 737   sp = (address)&sp;
 738 
 739   if (sp > bottom) {
 740     size = sp - bottom;
 741     p = (volatile char *)alloca(size);
 742     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 743     p[0] = '\0';
 744   }
 745 }
 746 
 747 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 748   assert(t!=NULL, "just checking");
 749   assert(t->osthread()->expanding_stack(), "expand should be set");
 750   assert(t->stack_base() != NULL, "stack_base was not initialized");
 751 
 752   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 753     sigset_t mask_all, old_sigset;
 754     sigfillset(&mask_all);
 755     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 756     _expand_stack_to(addr);
 757     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 758     return true;
 759   }
 760   return false;
 761 }
 762 
 763 //////////////////////////////////////////////////////////////////////////////
 764 // create new thread
 765 
 766 static address highest_vm_reserved_address();
 767 
 768 // check if it's safe to start a new thread
 769 static bool _thread_safety_check(Thread* thread) {
 770   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 771     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 772     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 773     //   allocated (MAP_FIXED) from high address space. Every thread stack
 774     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 775     //   it to other values if they rebuild LinuxThreads).
 776     //
 777     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 778     // the memory region has already been mmap'ed. That means if we have too
 779     // many threads and/or very large heap, eventually thread stack will
 780     // collide with heap.
 781     //
 782     // Here we try to prevent heap/stack collision by comparing current
 783     // stack bottom with the highest address that has been mmap'ed by JVM
 784     // plus a safety margin for memory maps created by native code.
 785     //
 786     // This feature can be disabled by setting ThreadSafetyMargin to 0
 787     //
 788     if (ThreadSafetyMargin > 0) {
 789       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 790 
 791       // not safe if our stack extends below the safety margin
 792       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 793     } else {
 794       return true;
 795     }
 796   } else {
 797     // Floating stack LinuxThreads or NPTL:
 798     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 799     //   there's not enough space left, pthread_create() will fail. If we come
 800     //   here, that means enough space has been reserved for stack.
 801     return true;
 802   }
 803 }
 804 
 805 // Thread start routine for all newly created threads
 806 static void *java_start(Thread *thread) {
 807   // Try to randomize the cache line index of hot stack frames.
 808   // This helps when threads of the same stack traces evict each other's
 809   // cache lines. The threads can be either from the same JVM instance, or
 810   // from different JVM instances. The benefit is especially true for
 811   // processors with hyperthreading technology.
 812   static int counter = 0;
 813   int pid = os::current_process_id();
 814   alloca(((pid ^ counter++) & 7) * 128);
 815 
 816   ThreadLocalStorage::set_thread(thread);
 817 
 818   OSThread* osthread = thread->osthread();
 819   Monitor* sync = osthread->startThread_lock();
 820 
 821   // non floating stack LinuxThreads needs extra check, see above
 822   if (!_thread_safety_check(thread)) {
 823     // notify parent thread
 824     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 825     osthread->set_state(ZOMBIE);
 826     sync->notify_all();
 827     return NULL;
 828   }
 829 
 830   // thread_id is kernel thread id (similar to Solaris LWP id)
 831   osthread->set_thread_id(os::Linux::gettid());
 832 
 833   if (UseNUMA) {
 834     int lgrp_id = os::numa_get_group_id();
 835     if (lgrp_id != -1) {
 836       thread->set_lgrp_id(lgrp_id);
 837     }
 838   }
 839   // initialize signal mask for this thread
 840   os::Linux::hotspot_sigmask(thread);
 841 
 842   // initialize floating point control register
 843   os::Linux::init_thread_fpu_state();
 844 
 845   // handshaking with parent thread
 846   {
 847     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 848 
 849     // notify parent thread
 850     osthread->set_state(INITIALIZED);
 851     sync->notify_all();
 852 
 853     // wait until os::start_thread()
 854     while (osthread->get_state() == INITIALIZED) {
 855       sync->wait(Mutex::_no_safepoint_check_flag);
 856     }
 857   }
 858 
 859   // call one more level start routine
 860   thread->run();
 861 
 862   return 0;
 863 }
 864 
 865 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 866   assert(thread->osthread() == NULL, "caller responsible");
 867 
 868   // Allocate the OSThread object
 869   OSThread* osthread = new OSThread(NULL, NULL);
 870   if (osthread == NULL) {
 871     return false;
 872   }
 873 
 874   // set the correct thread state
 875   osthread->set_thread_type(thr_type);
 876 
 877   // Initial state is ALLOCATED but not INITIALIZED
 878   osthread->set_state(ALLOCATED);
 879 
 880   thread->set_osthread(osthread);
 881 
 882   // init thread attributes
 883   pthread_attr_t attr;
 884   pthread_attr_init(&attr);
 885   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 886 
 887   // stack size
 888   if (os::Linux::supports_variable_stack_size()) {
 889     // calculate stack size if it's not specified by caller
 890     if (stack_size == 0) {
 891       stack_size = os::Linux::default_stack_size(thr_type);
 892 
 893       switch (thr_type) {
 894       case os::java_thread:
 895         // Java threads use ThreadStackSize which default value can be
 896         // changed with the flag -Xss
 897         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 898         stack_size = JavaThread::stack_size_at_create();
 899         break;
 900       case os::compiler_thread:
 901         if (CompilerThreadStackSize > 0) {
 902           stack_size = (size_t)(CompilerThreadStackSize * K);
 903           break;
 904         } // else fall through:
 905           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 906       case os::vm_thread:
 907       case os::pgc_thread:
 908       case os::cgc_thread:
 909       case os::watcher_thread:
 910         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 911         break;
 912       }
 913     }
 914 
 915     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 916     pthread_attr_setstacksize(&attr, stack_size);
 917   } else {
 918     // let pthread_create() pick the default value.
 919   }
 920 
 921   // glibc guard page
 922   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 923 
 924   ThreadState state;
 925 
 926   {
 927     // Serialize thread creation if we are running with fixed stack LinuxThreads
 928     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 929     if (lock) {
 930       os::Linux::createThread_lock()->lock_without_safepoint_check();
 931     }
 932 
 933     pthread_t tid;
 934     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 935 
 936     pthread_attr_destroy(&attr);
 937 
 938     if (ret != 0) {
 939       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 940         perror("pthread_create()");
 941       }
 942       // Need to clean up stuff we've allocated so far
 943       thread->set_osthread(NULL);
 944       delete osthread;
 945       if (lock) os::Linux::createThread_lock()->unlock();
 946       return false;
 947     }
 948 
 949     // Store pthread info into the OSThread
 950     osthread->set_pthread_id(tid);
 951 
 952     // Wait until child thread is either initialized or aborted
 953     {
 954       Monitor* sync_with_child = osthread->startThread_lock();
 955       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 956       while ((state = osthread->get_state()) == ALLOCATED) {
 957         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 958       }
 959     }
 960 
 961     if (lock) {
 962       os::Linux::createThread_lock()->unlock();
 963     }
 964   }
 965 
 966   // Aborted due to thread limit being reached
 967   if (state == ZOMBIE) {
 968       thread->set_osthread(NULL);
 969       delete osthread;
 970       return false;
 971   }
 972 
 973   // The thread is returned suspended (in state INITIALIZED),
 974   // and is started higher up in the call chain
 975   assert(state == INITIALIZED, "race condition");
 976   return true;
 977 }
 978 
 979 /////////////////////////////////////////////////////////////////////////////
 980 // attach existing thread
 981 
 982 // bootstrap the main thread
 983 bool os::create_main_thread(JavaThread* thread) {
 984   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 985   return create_attached_thread(thread);
 986 }
 987 
 988 bool os::create_attached_thread(JavaThread* thread) {
 989 #ifdef ASSERT
 990     thread->verify_not_published();
 991 #endif
 992 
 993   // Allocate the OSThread object
 994   OSThread* osthread = new OSThread(NULL, NULL);
 995 
 996   if (osthread == NULL) {
 997     return false;
 998   }
 999 
1000   // Store pthread info into the OSThread
1001   osthread->set_thread_id(os::Linux::gettid());
1002   osthread->set_pthread_id(::pthread_self());
1003 
1004   // initialize floating point control register
1005   os::Linux::init_thread_fpu_state();
1006 
1007   // Initial thread state is RUNNABLE
1008   osthread->set_state(RUNNABLE);
1009 
1010   thread->set_osthread(osthread);
1011 
1012   if (UseNUMA) {
1013     int lgrp_id = os::numa_get_group_id();
1014     if (lgrp_id != -1) {
1015       thread->set_lgrp_id(lgrp_id);
1016     }
1017   }
1018 
1019   if (os::Linux::is_initial_thread()) {
1020     // If current thread is initial thread, its stack is mapped on demand,
1021     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1022     // the entire stack region to avoid SEGV in stack banging.
1023     // It is also useful to get around the heap-stack-gap problem on SuSE
1024     // kernel (see 4821821 for details). We first expand stack to the top
1025     // of yellow zone, then enable stack yellow zone (order is significant,
1026     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1027     // is no gap between the last two virtual memory regions.
1028 
1029     JavaThread *jt = (JavaThread *)thread;
1030     address addr = jt->stack_yellow_zone_base();
1031     assert(addr != NULL, "initialization problem?");
1032     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1033 
1034     osthread->set_expanding_stack();
1035     os::Linux::manually_expand_stack(jt, addr);
1036     osthread->clear_expanding_stack();
1037   }
1038 
1039   // initialize signal mask for this thread
1040   // and save the caller's signal mask
1041   os::Linux::hotspot_sigmask(thread);
1042 
1043   return true;
1044 }
1045 
1046 void os::pd_start_thread(Thread* thread) {
1047   OSThread * osthread = thread->osthread();
1048   assert(osthread->get_state() != INITIALIZED, "just checking");
1049   Monitor* sync_with_child = osthread->startThread_lock();
1050   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1051   sync_with_child->notify();
1052 }
1053 
1054 // Free Linux resources related to the OSThread
1055 void os::free_thread(OSThread* osthread) {
1056   assert(osthread != NULL, "osthread not set");
1057 
1058   if (Thread::current()->osthread() == osthread) {
1059     // Restore caller's signal mask
1060     sigset_t sigmask = osthread->caller_sigmask();
1061     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1062    }
1063 
1064   delete osthread;
1065 }
1066 
1067 //////////////////////////////////////////////////////////////////////////////
1068 // thread local storage
1069 
1070 int os::allocate_thread_local_storage() {
1071   pthread_key_t key;
1072   int rslt = pthread_key_create(&key, NULL);
1073   assert(rslt == 0, "cannot allocate thread local storage");
1074   return (int)key;
1075 }
1076 
1077 // Note: This is currently not used by VM, as we don't destroy TLS key
1078 // on VM exit.
1079 void os::free_thread_local_storage(int index) {
1080   int rslt = pthread_key_delete((pthread_key_t)index);
1081   assert(rslt == 0, "invalid index");
1082 }
1083 
1084 void os::thread_local_storage_at_put(int index, void* value) {
1085   int rslt = pthread_setspecific((pthread_key_t)index, value);
1086   assert(rslt == 0, "pthread_setspecific failed");
1087 }
1088 
1089 extern "C" Thread* get_thread() {
1090   return ThreadLocalStorage::thread();
1091 }
1092 
1093 //////////////////////////////////////////////////////////////////////////////
1094 // initial thread
1095 
1096 // Check if current thread is the initial thread, similar to Solaris thr_main.
1097 bool os::Linux::is_initial_thread(void) {
1098   char dummy;
1099   // If called before init complete, thread stack bottom will be null.
1100   // Can be called if fatal error occurs before initialization.
1101   if (initial_thread_stack_bottom() == NULL) return false;
1102   assert(initial_thread_stack_bottom() != NULL &&
1103          initial_thread_stack_size()   != 0,
1104          "os::init did not locate initial thread's stack region");
1105   if ((address)&dummy >= initial_thread_stack_bottom() &&
1106       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1107        return true;
1108   else return false;
1109 }
1110 
1111 // Find the virtual memory area that contains addr
1112 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1113   FILE *fp = fopen("/proc/self/maps", "r");
1114   if (fp) {
1115     address low, high;
1116     while (!feof(fp)) {
1117       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1118         if (low <= addr && addr < high) {
1119            if (vma_low)  *vma_low  = low;
1120            if (vma_high) *vma_high = high;
1121            fclose (fp);
1122            return true;
1123         }
1124       }
1125       for (;;) {
1126         int ch = fgetc(fp);
1127         if (ch == EOF || ch == (int)'\n') break;
1128       }
1129     }
1130     fclose(fp);
1131   }
1132   return false;
1133 }
1134 
1135 // Locate initial thread stack. This special handling of initial thread stack
1136 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1137 // bogus value for initial thread.
1138 void os::Linux::capture_initial_stack(size_t max_size) {
1139   // stack size is the easy part, get it from RLIMIT_STACK
1140   size_t stack_size;
1141   struct rlimit rlim;
1142   getrlimit(RLIMIT_STACK, &rlim);
1143   stack_size = rlim.rlim_cur;
1144 
1145   // 6308388: a bug in ld.so will relocate its own .data section to the
1146   //   lower end of primordial stack; reduce ulimit -s value a little bit
1147   //   so we won't install guard page on ld.so's data section.
1148   stack_size -= 2 * page_size();
1149 
1150   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1151   //   7.1, in both cases we will get 2G in return value.
1152   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1153   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1154   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1155   //   in case other parts in glibc still assumes 2M max stack size.
1156   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1157   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1158   if (stack_size > 2 * K * K IA64_ONLY(*2))
1159       stack_size = 2 * K * K IA64_ONLY(*2);
1160   // Try to figure out where the stack base (top) is. This is harder.
1161   //
1162   // When an application is started, glibc saves the initial stack pointer in
1163   // a global variable "__libc_stack_end", which is then used by system
1164   // libraries. __libc_stack_end should be pretty close to stack top. The
1165   // variable is available since the very early days. However, because it is
1166   // a private interface, it could disappear in the future.
1167   //
1168   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1169   // to __libc_stack_end, it is very close to stack top, but isn't the real
1170   // stack top. Note that /proc may not exist if VM is running as a chroot
1171   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1172   // /proc/<pid>/stat could change in the future (though unlikely).
1173   //
1174   // We try __libc_stack_end first. If that doesn't work, look for
1175   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1176   // as a hint, which should work well in most cases.
1177 
1178   uintptr_t stack_start;
1179 
1180   // try __libc_stack_end first
1181   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1182   if (p && *p) {
1183     stack_start = *p;
1184   } else {
1185     // see if we can get the start_stack field from /proc/self/stat
1186     FILE *fp;
1187     int pid;
1188     char state;
1189     int ppid;
1190     int pgrp;
1191     int session;
1192     int nr;
1193     int tpgrp;
1194     unsigned long flags;
1195     unsigned long minflt;
1196     unsigned long cminflt;
1197     unsigned long majflt;
1198     unsigned long cmajflt;
1199     unsigned long utime;
1200     unsigned long stime;
1201     long cutime;
1202     long cstime;
1203     long prio;
1204     long nice;
1205     long junk;
1206     long it_real;
1207     uintptr_t start;
1208     uintptr_t vsize;
1209     intptr_t rss;
1210     uintptr_t rsslim;
1211     uintptr_t scodes;
1212     uintptr_t ecode;
1213     int i;
1214 
1215     // Figure what the primordial thread stack base is. Code is inspired
1216     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1217     // followed by command name surrounded by parentheses, state, etc.
1218     char stat[2048];
1219     int statlen;
1220 
1221     fp = fopen("/proc/self/stat", "r");
1222     if (fp) {
1223       statlen = fread(stat, 1, 2047, fp);
1224       stat[statlen] = '\0';
1225       fclose(fp);
1226 
1227       // Skip pid and the command string. Note that we could be dealing with
1228       // weird command names, e.g. user could decide to rename java launcher
1229       // to "java 1.4.2 :)", then the stat file would look like
1230       //                1234 (java 1.4.2 :)) R ... ...
1231       // We don't really need to know the command string, just find the last
1232       // occurrence of ")" and then start parsing from there. See bug 4726580.
1233       char * s = strrchr(stat, ')');
1234 
1235       i = 0;
1236       if (s) {
1237         // Skip blank chars
1238         do s++; while (isspace(*s));
1239 
1240 #define _UFM UINTX_FORMAT
1241 #define _DFM INTX_FORMAT
1242 
1243         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1244         /*              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 */
1245         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,
1246              &state,          /* 3  %c  */
1247              &ppid,           /* 4  %d  */
1248              &pgrp,           /* 5  %d  */
1249              &session,        /* 6  %d  */
1250              &nr,             /* 7  %d  */
1251              &tpgrp,          /* 8  %d  */
1252              &flags,          /* 9  %lu  */
1253              &minflt,         /* 10 %lu  */
1254              &cminflt,        /* 11 %lu  */
1255              &majflt,         /* 12 %lu  */
1256              &cmajflt,        /* 13 %lu  */
1257              &utime,          /* 14 %lu  */
1258              &stime,          /* 15 %lu  */
1259              &cutime,         /* 16 %ld  */
1260              &cstime,         /* 17 %ld  */
1261              &prio,           /* 18 %ld  */
1262              &nice,           /* 19 %ld  */
1263              &junk,           /* 20 %ld  */
1264              &it_real,        /* 21 %ld  */
1265              &start,          /* 22 UINTX_FORMAT */
1266              &vsize,          /* 23 UINTX_FORMAT */
1267              &rss,            /* 24 INTX_FORMAT  */
1268              &rsslim,         /* 25 UINTX_FORMAT */
1269              &scodes,         /* 26 UINTX_FORMAT */
1270              &ecode,          /* 27 UINTX_FORMAT */
1271              &stack_start);   /* 28 UINTX_FORMAT */
1272       }
1273 
1274 #undef _UFM
1275 #undef _DFM
1276 
1277       if (i != 28 - 2) {
1278          assert(false, "Bad conversion from /proc/self/stat");
1279          // product mode - assume we are the initial thread, good luck in the
1280          // embedded case.
1281          warning("Can't detect initial thread stack location - bad conversion");
1282          stack_start = (uintptr_t) &rlim;
1283       }
1284     } else {
1285       // For some reason we can't open /proc/self/stat (for example, running on
1286       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1287       // most cases, so don't abort:
1288       warning("Can't detect initial thread stack location - no /proc/self/stat");
1289       stack_start = (uintptr_t) &rlim;
1290     }
1291   }
1292 
1293   // Now we have a pointer (stack_start) very close to the stack top, the
1294   // next thing to do is to figure out the exact location of stack top. We
1295   // can find out the virtual memory area that contains stack_start by
1296   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1297   // and its upper limit is the real stack top. (again, this would fail if
1298   // running inside chroot, because /proc may not exist.)
1299 
1300   uintptr_t stack_top;
1301   address low, high;
1302   if (find_vma((address)stack_start, &low, &high)) {
1303     // success, "high" is the true stack top. (ignore "low", because initial
1304     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1305     stack_top = (uintptr_t)high;
1306   } else {
1307     // failed, likely because /proc/self/maps does not exist
1308     warning("Can't detect initial thread stack location - find_vma failed");
1309     // best effort: stack_start is normally within a few pages below the real
1310     // stack top, use it as stack top, and reduce stack size so we won't put
1311     // guard page outside stack.
1312     stack_top = stack_start;
1313     stack_size -= 16 * page_size();
1314   }
1315 
1316   // stack_top could be partially down the page so align it
1317   stack_top = align_size_up(stack_top, page_size());
1318 
1319   if (max_size && stack_size > max_size) {
1320      _initial_thread_stack_size = max_size;
1321   } else {
1322      _initial_thread_stack_size = stack_size;
1323   }
1324 
1325   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1326   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1327 }
1328 
1329 ////////////////////////////////////////////////////////////////////////////////
1330 // time support
1331 
1332 // Time since start-up in seconds to a fine granularity.
1333 // Used by VMSelfDestructTimer and the MemProfiler.
1334 double os::elapsedTime() {
1335 
1336   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1337 }
1338 
1339 jlong os::elapsed_counter() {
1340   return javaTimeNanos() - initial_time_count;
1341 }
1342 
1343 jlong os::elapsed_frequency() {
1344   return NANOSECS_PER_SEC; // nanosecond resolution
1345 }
1346 
1347 bool os::supports_vtime() { return true; }
1348 bool os::enable_vtime()   { return false; }
1349 bool os::vtime_enabled()  { return false; }
1350 
1351 double os::elapsedVTime() {
1352   struct rusage usage;
1353   int retval = getrusage(RUSAGE_THREAD, &usage);
1354   if (retval == 0) {
1355     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);
1356   } else {
1357     // better than nothing, but not much
1358     return elapsedTime();
1359   }
1360 }
1361 
1362 jlong os::javaTimeMillis() {
1363   timeval time;
1364   int status = gettimeofday(&time, NULL);
1365   assert(status != -1, "linux error");
1366   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1367 }
1368 
1369 #ifndef CLOCK_MONOTONIC
1370 #define CLOCK_MONOTONIC (1)
1371 #endif
1372 
1373 void os::Linux::clock_init() {
1374   // we do dlopen's in this particular order due to bug in linux
1375   // dynamical loader (see 6348968) leading to crash on exit
1376   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1377   if (handle == NULL) {
1378     handle = dlopen("librt.so", RTLD_LAZY);
1379   }
1380 
1381   if (handle) {
1382     int (*clock_getres_func)(clockid_t, struct timespec*) =
1383            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1384     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1385            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1386     if (clock_getres_func && clock_gettime_func) {
1387       // See if monotonic clock is supported by the kernel. Note that some
1388       // early implementations simply return kernel jiffies (updated every
1389       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1390       // for nano time (though the monotonic property is still nice to have).
1391       // It's fixed in newer kernels, however clock_getres() still returns
1392       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1393       // resolution for now. Hopefully as people move to new kernels, this
1394       // won't be a problem.
1395       struct timespec res;
1396       struct timespec tp;
1397       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1398           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1399         // yes, monotonic clock is supported
1400         _clock_gettime = clock_gettime_func;
1401         return;
1402       } else {
1403         // close librt if there is no monotonic clock
1404         dlclose(handle);
1405       }
1406     }
1407   }
1408   warning("No monotonic clock was available - timed services may " \
1409           "be adversely affected if the time-of-day clock changes");
1410 }
1411 
1412 #ifndef SYS_clock_getres
1413 
1414 #if defined(IA32) || defined(AMD64)
1415 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1416 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1417 #else
1418 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1419 #define sys_clock_getres(x,y)  -1
1420 #endif
1421 
1422 #else
1423 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1424 #endif
1425 
1426 void os::Linux::fast_thread_clock_init() {
1427   if (!UseLinuxPosixThreadCPUClocks) {
1428     return;
1429   }
1430   clockid_t clockid;
1431   struct timespec tp;
1432   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1433       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1434 
1435   // Switch to using fast clocks for thread cpu time if
1436   // the sys_clock_getres() returns 0 error code.
1437   // Note, that some kernels may support the current thread
1438   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1439   // returned by the pthread_getcpuclockid().
1440   // If the fast Posix clocks are supported then the sys_clock_getres()
1441   // must return at least tp.tv_sec == 0 which means a resolution
1442   // better than 1 sec. This is extra check for reliability.
1443 
1444   if(pthread_getcpuclockid_func &&
1445      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1446      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1447 
1448     _supports_fast_thread_cpu_time = true;
1449     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1450   }
1451 }
1452 
1453 jlong os::javaTimeNanos() {
1454   if (Linux::supports_monotonic_clock()) {
1455     struct timespec tp;
1456     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1457     assert(status == 0, "gettime error");
1458     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1459     return result;
1460   } else {
1461     timeval time;
1462     int status = gettimeofday(&time, NULL);
1463     assert(status != -1, "linux error");
1464     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1465     return 1000 * usecs;
1466   }
1467 }
1468 
1469 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1470   if (Linux::supports_monotonic_clock()) {
1471     info_ptr->max_value = ALL_64_BITS;
1472 
1473     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1474     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1475     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1476   } else {
1477     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1478     info_ptr->max_value = ALL_64_BITS;
1479 
1480     // gettimeofday is a real time clock so it skips
1481     info_ptr->may_skip_backward = true;
1482     info_ptr->may_skip_forward = true;
1483   }
1484 
1485   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1486 }
1487 
1488 // Return the real, user, and system times in seconds from an
1489 // arbitrary fixed point in the past.
1490 bool os::getTimesSecs(double* process_real_time,
1491                       double* process_user_time,
1492                       double* process_system_time) {
1493   struct tms ticks;
1494   clock_t real_ticks = times(&ticks);
1495 
1496   if (real_ticks == (clock_t) (-1)) {
1497     return false;
1498   } else {
1499     double ticks_per_second = (double) clock_tics_per_sec;
1500     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1501     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1502     *process_real_time = ((double) real_ticks) / ticks_per_second;
1503 
1504     return true;
1505   }
1506 }
1507 
1508 
1509 char * os::local_time_string(char *buf, size_t buflen) {
1510   struct tm t;
1511   time_t long_time;
1512   time(&long_time);
1513   localtime_r(&long_time, &t);
1514   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1515                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1516                t.tm_hour, t.tm_min, t.tm_sec);
1517   return buf;
1518 }
1519 
1520 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1521   return localtime_r(clock, res);
1522 }
1523 
1524 ////////////////////////////////////////////////////////////////////////////////
1525 // runtime exit support
1526 
1527 // Note: os::shutdown() might be called very early during initialization, or
1528 // called from signal handler. Before adding something to os::shutdown(), make
1529 // sure it is async-safe and can handle partially initialized VM.
1530 void os::shutdown() {
1531 
1532   // allow PerfMemory to attempt cleanup of any persistent resources
1533   perfMemory_exit();
1534 
1535   // needs to remove object in file system
1536   AttachListener::abort();
1537 
1538   // flush buffered output, finish log files
1539   ostream_abort();
1540 
1541   // Check for abort hook
1542   abort_hook_t abort_hook = Arguments::abort_hook();
1543   if (abort_hook != NULL) {
1544     abort_hook();
1545   }
1546 
1547 }
1548 
1549 // Note: os::abort() might be called very early during initialization, or
1550 // called from signal handler. Before adding something to os::abort(), make
1551 // sure it is async-safe and can handle partially initialized VM.
1552 void os::abort(bool dump_core) {
1553   os::shutdown();
1554   if (dump_core) {
1555 #ifndef PRODUCT
1556     fdStream out(defaultStream::output_fd());
1557     out.print_raw("Current thread is ");
1558     char buf[16];
1559     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1560     out.print_raw_cr(buf);
1561     out.print_raw_cr("Dumping core ...");
1562 #endif
1563     ::abort(); // dump core
1564   }
1565 
1566   ::exit(1);
1567 }
1568 
1569 // Die immediately, no exit hook, no abort hook, no cleanup.
1570 void os::die() {
1571   // _exit() on LinuxThreads only kills current thread
1572   ::abort();
1573 }
1574 
1575 // unused on linux for now.
1576 void os::set_error_file(const char *logfile) {}
1577 
1578 
1579 // This method is a copy of JDK's sysGetLastErrorString
1580 // from src/solaris/hpi/src/system_md.c
1581 
1582 size_t os::lasterror(char *buf, size_t len) {
1583 
1584   if (errno == 0)  return 0;
1585 
1586   const char *s = ::strerror(errno);
1587   size_t n = ::strlen(s);
1588   if (n >= len) {
1589     n = len - 1;
1590   }
1591   ::strncpy(buf, s, n);
1592   buf[n] = '\0';
1593   return n;
1594 }
1595 
1596 intx os::current_thread_id() { return (intx)pthread_self(); }
1597 int os::current_process_id() {
1598 
1599   // Under the old linux thread library, linux gives each thread
1600   // its own process id. Because of this each thread will return
1601   // a different pid if this method were to return the result
1602   // of getpid(2). Linux provides no api that returns the pid
1603   // of the launcher thread for the vm. This implementation
1604   // returns a unique pid, the pid of the launcher thread
1605   // that starts the vm 'process'.
1606 
1607   // Under the NPTL, getpid() returns the same pid as the
1608   // launcher thread rather than a unique pid per thread.
1609   // Use gettid() if you want the old pre NPTL behaviour.
1610 
1611   // if you are looking for the result of a call to getpid() that
1612   // returns a unique pid for the calling thread, then look at the
1613   // OSThread::thread_id() method in osThread_linux.hpp file
1614 
1615   return (int)(_initial_pid ? _initial_pid : getpid());
1616 }
1617 
1618 // DLL functions
1619 
1620 const char* os::dll_file_extension() { return ".so"; }
1621 
1622 // This must be hard coded because it's the system's temporary
1623 // directory not the java application's temp directory, ala java.io.tmpdir.
1624 const char* os::get_temp_directory() { return "/tmp"; }
1625 
1626 static bool file_exists(const char* filename) {
1627   struct stat statbuf;
1628   if (filename == NULL || strlen(filename) == 0) {
1629     return false;
1630   }
1631   return os::stat(filename, &statbuf) == 0;
1632 }
1633 
1634 bool os::dll_build_name(char* buffer, size_t buflen,
1635                         const char* pname, const char* fname) {
1636   bool retval = false;
1637   // Copied from libhpi
1638   const size_t pnamelen = pname ? strlen(pname) : 0;
1639 
1640   // Return error on buffer overflow.
1641   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1642     return retval;
1643   }
1644 
1645   if (pnamelen == 0) {
1646     snprintf(buffer, buflen, "lib%s.so", fname);
1647     retval = true;
1648   } else if (strchr(pname, *os::path_separator()) != NULL) {
1649     int n;
1650     char** pelements = split_path(pname, &n);
1651     if (pelements == NULL) {
1652       return false;
1653     }
1654     for (int i = 0 ; i < n ; i++) {
1655       // Really shouldn't be NULL, but check can't hurt
1656       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1657         continue; // skip the empty path values
1658       }
1659       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1660       if (file_exists(buffer)) {
1661         retval = true;
1662         break;
1663       }
1664     }
1665     // release the storage
1666     for (int i = 0 ; i < n ; i++) {
1667       if (pelements[i] != NULL) {
1668         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1669       }
1670     }
1671     if (pelements != NULL) {
1672       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1673     }
1674   } else {
1675     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1676     retval = true;
1677   }
1678   return retval;
1679 }
1680 
1681 // check if addr is inside libjvm.so
1682 bool os::address_is_in_vm(address addr) {
1683   static address libjvm_base_addr;
1684   Dl_info dlinfo;
1685 
1686   if (libjvm_base_addr == NULL) {
1687     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1688       libjvm_base_addr = (address)dlinfo.dli_fbase;
1689     }
1690     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1691   }
1692 
1693   if (dladdr((void *)addr, &dlinfo) != 0) {
1694     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1695   }
1696 
1697   return false;
1698 }
1699 
1700 bool os::dll_address_to_function_name(address addr, char *buf,
1701                                       int buflen, int *offset) {
1702   // buf is not optional, but offset is optional
1703   assert(buf != NULL, "sanity check");
1704 
1705   Dl_info dlinfo;
1706 
1707   if (dladdr((void*)addr, &dlinfo) != 0) {
1708     // see if we have a matching symbol
1709     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1710       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1711         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1712       }
1713       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1714       return true;
1715     }
1716     // no matching symbol so try for just file info
1717     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1718       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1719                           buf, buflen, offset, dlinfo.dli_fname)) {
1720         return true;
1721       }
1722     }
1723   }
1724 
1725   buf[0] = '\0';
1726   if (offset != NULL) *offset = -1;
1727   return false;
1728 }
1729 
1730 struct _address_to_library_name {
1731   address addr;          // input : memory address
1732   size_t  buflen;        //         size of fname
1733   char*   fname;         // output: library name
1734   address base;          //         library base addr
1735 };
1736 
1737 static int address_to_library_name_callback(struct dl_phdr_info *info,
1738                                             size_t size, void *data) {
1739   int i;
1740   bool found = false;
1741   address libbase = NULL;
1742   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1743 
1744   // iterate through all loadable segments
1745   for (i = 0; i < info->dlpi_phnum; i++) {
1746     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1747     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1748       // base address of a library is the lowest address of its loaded
1749       // segments.
1750       if (libbase == NULL || libbase > segbase) {
1751         libbase = segbase;
1752       }
1753       // see if 'addr' is within current segment
1754       if (segbase <= d->addr &&
1755           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1756         found = true;
1757       }
1758     }
1759   }
1760 
1761   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1762   // so dll_address_to_library_name() can fall through to use dladdr() which
1763   // can figure out executable name from argv[0].
1764   if (found && info->dlpi_name && info->dlpi_name[0]) {
1765     d->base = libbase;
1766     if (d->fname) {
1767       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1768     }
1769     return 1;
1770   }
1771   return 0;
1772 }
1773 
1774 bool os::dll_address_to_library_name(address addr, char* buf,
1775                                      int buflen, int* offset) {
1776   // buf is not optional, but offset is optional
1777   assert(buf != NULL, "sanity check");
1778 
1779   Dl_info dlinfo;
1780   struct _address_to_library_name data;
1781 
1782   // There is a bug in old glibc dladdr() implementation that it could resolve
1783   // to wrong library name if the .so file has a base address != NULL. Here
1784   // we iterate through the program headers of all loaded libraries to find
1785   // out which library 'addr' really belongs to. This workaround can be
1786   // removed once the minimum requirement for glibc is moved to 2.3.x.
1787   data.addr = addr;
1788   data.fname = buf;
1789   data.buflen = buflen;
1790   data.base = NULL;
1791   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1792 
1793   if (rslt) {
1794      // buf already contains library name
1795      if (offset) *offset = addr - data.base;
1796      return true;
1797   }
1798   if (dladdr((void*)addr, &dlinfo) != 0) {
1799     if (dlinfo.dli_fname != NULL) {
1800       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1801     }
1802     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1803       *offset = addr - (address)dlinfo.dli_fbase;
1804     }
1805     return true;
1806   }
1807 
1808   buf[0] = '\0';
1809   if (offset) *offset = -1;
1810   return false;
1811 }
1812 
1813   // Loads .dll/.so and
1814   // in case of error it checks if .dll/.so was built for the
1815   // same architecture as Hotspot is running on
1816 
1817 
1818 // Remember the stack's state. The Linux dynamic linker will change
1819 // the stack to 'executable' at most once, so we must safepoint only once.
1820 bool os::Linux::_stack_is_executable = false;
1821 
1822 // VM operation that loads a library.  This is necessary if stack protection
1823 // of the Java stacks can be lost during loading the library.  If we
1824 // do not stop the Java threads, they can stack overflow before the stacks
1825 // are protected again.
1826 class VM_LinuxDllLoad: public VM_Operation {
1827  private:
1828   const char *_filename;
1829   char *_ebuf;
1830   int _ebuflen;
1831   void *_lib;
1832  public:
1833   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1834     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1835   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1836   void doit() {
1837     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1838     os::Linux::_stack_is_executable = true;
1839   }
1840   void* loaded_library() { return _lib; }
1841 };
1842 
1843 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1844 {
1845   void * result = NULL;
1846   bool load_attempted = false;
1847 
1848   // Check whether the library to load might change execution rights
1849   // of the stack. If they are changed, the protection of the stack
1850   // guard pages will be lost. We need a safepoint to fix this.
1851   //
1852   // See Linux man page execstack(8) for more info.
1853   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1854     ElfFile ef(filename);
1855     if (!ef.specifies_noexecstack()) {
1856       if (!is_init_completed()) {
1857         os::Linux::_stack_is_executable = true;
1858         // This is OK - No Java threads have been created yet, and hence no
1859         // stack guard pages to fix.
1860         //
1861         // This should happen only when you are building JDK7 using a very
1862         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1863         //
1864         // Dynamic loader will make all stacks executable after
1865         // this function returns, and will not do that again.
1866         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1867       } else {
1868         warning("You have loaded library %s which might have disabled stack guard. "
1869                 "The VM will try to fix the stack guard now.\n"
1870                 "It's highly recommended that you fix the library with "
1871                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1872                 filename);
1873 
1874         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1875         JavaThread *jt = JavaThread::current();
1876         if (jt->thread_state() != _thread_in_native) {
1877           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1878           // that requires ExecStack. Cannot enter safe point. Let's give up.
1879           warning("Unable to fix stack guard. Giving up.");
1880         } else {
1881           if (!LoadExecStackDllInVMThread) {
1882             // This is for the case where the DLL has an static
1883             // constructor function that executes JNI code. We cannot
1884             // load such DLLs in the VMThread.
1885             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1886           }
1887 
1888           ThreadInVMfromNative tiv(jt);
1889           debug_only(VMNativeEntryWrapper vew;)
1890 
1891           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1892           VMThread::execute(&op);
1893           if (LoadExecStackDllInVMThread) {
1894             result = op.loaded_library();
1895           }
1896           load_attempted = true;
1897         }
1898       }
1899     }
1900   }
1901 
1902   if (!load_attempted) {
1903     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1904   }
1905 
1906   if (result != NULL) {
1907     // Successful loading
1908     return result;
1909   }
1910 
1911   Elf32_Ehdr elf_head;
1912   int diag_msg_max_length=ebuflen-strlen(ebuf);
1913   char* diag_msg_buf=ebuf+strlen(ebuf);
1914 
1915   if (diag_msg_max_length==0) {
1916     // No more space in ebuf for additional diagnostics message
1917     return NULL;
1918   }
1919 
1920 
1921   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1922 
1923   if (file_descriptor < 0) {
1924     // Can't open library, report dlerror() message
1925     return NULL;
1926   }
1927 
1928   bool failed_to_read_elf_head=
1929     (sizeof(elf_head)!=
1930         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1931 
1932   ::close(file_descriptor);
1933   if (failed_to_read_elf_head) {
1934     // file i/o error - report dlerror() msg
1935     return NULL;
1936   }
1937 
1938   typedef struct {
1939     Elf32_Half  code;         // Actual value as defined in elf.h
1940     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1941     char        elf_class;    // 32 or 64 bit
1942     char        endianess;    // MSB or LSB
1943     char*       name;         // String representation
1944   } arch_t;
1945 
1946   #ifndef EM_486
1947   #define EM_486          6               /* Intel 80486 */
1948   #endif
1949 
1950   static const arch_t arch_array[]={
1951     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1952     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1953     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1954     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1955     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1956     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1957     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1958     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1959     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1960     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1961     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1962     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1963     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1964     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1965     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1966     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1967   };
1968 
1969   #if  (defined IA32)
1970     static  Elf32_Half running_arch_code=EM_386;
1971   #elif   (defined AMD64)
1972     static  Elf32_Half running_arch_code=EM_X86_64;
1973   #elif  (defined IA64)
1974     static  Elf32_Half running_arch_code=EM_IA_64;
1975   #elif  (defined __sparc) && (defined _LP64)
1976     static  Elf32_Half running_arch_code=EM_SPARCV9;
1977   #elif  (defined __sparc) && (!defined _LP64)
1978     static  Elf32_Half running_arch_code=EM_SPARC;
1979   #elif  (defined __powerpc64__)
1980     static  Elf32_Half running_arch_code=EM_PPC64;
1981   #elif  (defined __powerpc__)
1982     static  Elf32_Half running_arch_code=EM_PPC;
1983   #elif  (defined ARM)
1984     static  Elf32_Half running_arch_code=EM_ARM;
1985   #elif  (defined S390)
1986     static  Elf32_Half running_arch_code=EM_S390;
1987   #elif  (defined ALPHA)
1988     static  Elf32_Half running_arch_code=EM_ALPHA;
1989   #elif  (defined MIPSEL)
1990     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1991   #elif  (defined PARISC)
1992     static  Elf32_Half running_arch_code=EM_PARISC;
1993   #elif  (defined MIPS)
1994     static  Elf32_Half running_arch_code=EM_MIPS;
1995   #elif  (defined M68K)
1996     static  Elf32_Half running_arch_code=EM_68K;
1997   #else
1998     #error Method os::dll_load requires that one of following is defined:\
1999          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2000   #endif
2001 
2002   // Identify compatability class for VM's architecture and library's architecture
2003   // Obtain string descriptions for architectures
2004 
2005   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2006   int running_arch_index=-1;
2007 
2008   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2009     if (running_arch_code == arch_array[i].code) {
2010       running_arch_index    = i;
2011     }
2012     if (lib_arch.code == arch_array[i].code) {
2013       lib_arch.compat_class = arch_array[i].compat_class;
2014       lib_arch.name         = arch_array[i].name;
2015     }
2016   }
2017 
2018   assert(running_arch_index != -1,
2019     "Didn't find running architecture code (running_arch_code) in arch_array");
2020   if (running_arch_index == -1) {
2021     // Even though running architecture detection failed
2022     // we may still continue with reporting dlerror() message
2023     return NULL;
2024   }
2025 
2026   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2027     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2028     return NULL;
2029   }
2030 
2031 #ifndef S390
2032   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2033     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2034     return NULL;
2035   }
2036 #endif // !S390
2037 
2038   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2039     if ( lib_arch.name!=NULL ) {
2040       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2041         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2042         lib_arch.name, arch_array[running_arch_index].name);
2043     } else {
2044       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2045       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2046         lib_arch.code,
2047         arch_array[running_arch_index].name);
2048     }
2049   }
2050 
2051   return NULL;
2052 }
2053 
2054 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2055   void * result = ::dlopen(filename, RTLD_LAZY);
2056   if (result == NULL) {
2057     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2058     ebuf[ebuflen-1] = '\0';
2059   }
2060   return result;
2061 }
2062 
2063 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2064   void * result = NULL;
2065   if (LoadExecStackDllInVMThread) {
2066     result = dlopen_helper(filename, ebuf, ebuflen);
2067   }
2068 
2069   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2070   // library that requires an executable stack, or which does not have this
2071   // stack attribute set, dlopen changes the stack attribute to executable. The
2072   // read protection of the guard pages gets lost.
2073   //
2074   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2075   // may have been queued at the same time.
2076 
2077   if (!_stack_is_executable) {
2078     JavaThread *jt = Threads::first();
2079 
2080     while (jt) {
2081       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2082           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2083         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2084                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2085           warning("Attempt to reguard stack yellow zone failed.");
2086         }
2087       }
2088       jt = jt->next();
2089     }
2090   }
2091 
2092   return result;
2093 }
2094 
2095 /*
2096  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2097  * chances are you might want to run the generated bits against glibc-2.0
2098  * libdl.so, so always use locking for any version of glibc.
2099  */
2100 void* os::dll_lookup(void* handle, const char* name) {
2101   pthread_mutex_lock(&dl_mutex);
2102   void* res = dlsym(handle, name);
2103   pthread_mutex_unlock(&dl_mutex);
2104   return res;
2105 }
2106 
2107 void* os::get_default_process_handle() {
2108   return (void*)::dlopen(NULL, RTLD_LAZY);
2109 }
2110 
2111 static bool _print_ascii_file(const char* filename, outputStream* st) {
2112   int fd = ::open(filename, O_RDONLY);
2113   if (fd == -1) {
2114      return false;
2115   }
2116 
2117   char buf[32];
2118   int bytes;
2119   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2120     st->print_raw(buf, bytes);
2121   }
2122 
2123   ::close(fd);
2124 
2125   return true;
2126 }
2127 
2128 void os::print_dll_info(outputStream *st) {
2129    st->print_cr("Dynamic libraries:");
2130 
2131    char fname[32];
2132    pid_t pid = os::Linux::gettid();
2133 
2134    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2135 
2136    if (!_print_ascii_file(fname, st)) {
2137      st->print("Can not get library information for pid = %d\n", pid);
2138    }
2139 }
2140 
2141 void os::print_os_info_brief(outputStream* st) {
2142   os::Linux::print_distro_info(st);
2143 
2144   os::Posix::print_uname_info(st);
2145 
2146   os::Linux::print_libversion_info(st);
2147 
2148 }
2149 
2150 void os::print_os_info(outputStream* st) {
2151   st->print("OS:");
2152 
2153   os::Linux::print_distro_info(st);
2154 
2155   os::Posix::print_uname_info(st);
2156 
2157   // Print warning if unsafe chroot environment detected
2158   if (unsafe_chroot_detected) {
2159     st->print("WARNING!! ");
2160     st->print_cr(unstable_chroot_error);
2161   }
2162 
2163   os::Linux::print_libversion_info(st);
2164 
2165   os::Posix::print_rlimit_info(st);
2166 
2167   os::Posix::print_load_average(st);
2168 
2169   os::Linux::print_full_memory_info(st);
2170 }
2171 
2172 // Try to identify popular distros.
2173 // Most Linux distributions have a /etc/XXX-release file, which contains
2174 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2175 // file that also contains the OS version string. Some have more than one
2176 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2177 // /etc/redhat-release.), so the order is important.
2178 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2179 // their own specific XXX-release file as well as a redhat-release file.
2180 // Because of this the XXX-release file needs to be searched for before the
2181 // redhat-release file.
2182 // Since Red Hat has a lsb-release file that is not very descriptive the
2183 // search for redhat-release needs to be before lsb-release.
2184 // Since the lsb-release file is the new standard it needs to be searched
2185 // before the older style release files.
2186 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2187 // next to last resort.  The os-release file is a new standard that contains
2188 // distribution information and the system-release file seems to be an old
2189 // standard that has been replaced by the lsb-release and os-release files.
2190 // Searching for the debian_version file is the last resort.  It contains
2191 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2192 // "Debian " is printed before the contents of the debian_version file.
2193 void os::Linux::print_distro_info(outputStream* st) {
2194    if (!_print_ascii_file("/etc/oracle-release", st) &&
2195        !_print_ascii_file("/etc/mandriva-release", st) &&
2196        !_print_ascii_file("/etc/mandrake-release", st) &&
2197        !_print_ascii_file("/etc/sun-release", st) &&
2198        !_print_ascii_file("/etc/redhat-release", st) &&
2199        !_print_ascii_file("/etc/lsb-release", st) &&
2200        !_print_ascii_file("/etc/SuSE-release", st) &&
2201        !_print_ascii_file("/etc/turbolinux-release", st) &&
2202        !_print_ascii_file("/etc/gentoo-release", st) &&
2203        !_print_ascii_file("/etc/ltib-release", st) &&
2204        !_print_ascii_file("/etc/angstrom-version", st) &&
2205        !_print_ascii_file("/etc/system-release", st) &&
2206        !_print_ascii_file("/etc/os-release", st)) {
2207 
2208        if (file_exists("/etc/debian_version")) {
2209          st->print("Debian ");
2210          _print_ascii_file("/etc/debian_version", st);
2211        } else {
2212          st->print("Linux");
2213        }
2214    }
2215    st->cr();
2216 }
2217 
2218 void os::Linux::print_libversion_info(outputStream* st) {
2219   // libc, pthread
2220   st->print("libc:");
2221   st->print(os::Linux::glibc_version()); st->print(" ");
2222   st->print(os::Linux::libpthread_version()); st->print(" ");
2223   if (os::Linux::is_LinuxThreads()) {
2224      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2225   }
2226   st->cr();
2227 }
2228 
2229 void os::Linux::print_full_memory_info(outputStream* st) {
2230    st->print("\n/proc/meminfo:\n");
2231    _print_ascii_file("/proc/meminfo", st);
2232    st->cr();
2233 }
2234 
2235 void os::print_memory_info(outputStream* st) {
2236 
2237   st->print("Memory:");
2238   st->print(" %dk page", os::vm_page_size()>>10);
2239 
2240   // values in struct sysinfo are "unsigned long"
2241   struct sysinfo si;
2242   sysinfo(&si);
2243 
2244   st->print(", physical " UINT64_FORMAT "k",
2245             os::physical_memory() >> 10);
2246   st->print("(" UINT64_FORMAT "k free)",
2247             os::available_memory() >> 10);
2248   st->print(", swap " UINT64_FORMAT "k",
2249             ((jlong)si.totalswap * si.mem_unit) >> 10);
2250   st->print("(" UINT64_FORMAT "k free)",
2251             ((jlong)si.freeswap * si.mem_unit) >> 10);
2252   st->cr();
2253 }
2254 
2255 void os::pd_print_cpu_info(outputStream* st) {
2256   st->print("\n/proc/cpuinfo:\n");
2257   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2258     st->print("  <Not Available>");
2259   }
2260   st->cr();
2261 }
2262 
2263 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2264 // but they're the same for all the linux arch that we support
2265 // and they're the same for solaris but there's no common place to put this.
2266 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2267                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2268                           "ILL_COPROC", "ILL_BADSTK" };
2269 
2270 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2271                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2272                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2273 
2274 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2275 
2276 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2277 
2278 void os::print_siginfo(outputStream* st, void* siginfo) {
2279   st->print("siginfo:");
2280 
2281   const int buflen = 100;
2282   char buf[buflen];
2283   siginfo_t *si = (siginfo_t*)siginfo;
2284   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2285   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2286     st->print("si_errno=%s", buf);
2287   } else {
2288     st->print("si_errno=%d", si->si_errno);
2289   }
2290   const int c = si->si_code;
2291   assert(c > 0, "unexpected si_code");
2292   switch (si->si_signo) {
2293   case SIGILL:
2294     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2295     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2296     break;
2297   case SIGFPE:
2298     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2299     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2300     break;
2301   case SIGSEGV:
2302     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2303     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2304     break;
2305   case SIGBUS:
2306     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2307     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2308     break;
2309   default:
2310     st->print(", si_code=%d", si->si_code);
2311     // no si_addr
2312   }
2313 
2314   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2315       UseSharedSpaces) {
2316     FileMapInfo* mapinfo = FileMapInfo::current_info();
2317     if (mapinfo->is_in_shared_space(si->si_addr)) {
2318       st->print("\n\nError accessing class data sharing archive."   \
2319                 " Mapped file inaccessible during execution, "      \
2320                 " possible disk/network problem.");
2321     }
2322   }
2323   st->cr();
2324 }
2325 
2326 
2327 static void print_signal_handler(outputStream* st, int sig,
2328                                  char* buf, size_t buflen);
2329 
2330 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2331   st->print_cr("Signal Handlers:");
2332   print_signal_handler(st, SIGSEGV, buf, buflen);
2333   print_signal_handler(st, SIGBUS , buf, buflen);
2334   print_signal_handler(st, SIGFPE , buf, buflen);
2335   print_signal_handler(st, SIGPIPE, buf, buflen);
2336   print_signal_handler(st, SIGXFSZ, buf, buflen);
2337   print_signal_handler(st, SIGILL , buf, buflen);
2338   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2339   print_signal_handler(st, SR_signum, buf, buflen);
2340   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2341   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2342   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2343   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2344 }
2345 
2346 static char saved_jvm_path[MAXPATHLEN] = {0};
2347 
2348 // Find the full path to the current module, libjvm.so
2349 void os::jvm_path(char *buf, jint buflen) {
2350   // Error checking.
2351   if (buflen < MAXPATHLEN) {
2352     assert(false, "must use a large-enough buffer");
2353     buf[0] = '\0';
2354     return;
2355   }
2356   // Lazy resolve the path to current module.
2357   if (saved_jvm_path[0] != 0) {
2358     strcpy(buf, saved_jvm_path);
2359     return;
2360   }
2361 
2362   char dli_fname[MAXPATHLEN];
2363   bool ret = dll_address_to_library_name(
2364                 CAST_FROM_FN_PTR(address, os::jvm_path),
2365                 dli_fname, sizeof(dli_fname), NULL);
2366   assert(ret, "cannot locate libjvm");
2367   char *rp = NULL;
2368   if (ret && dli_fname[0] != '\0') {
2369     rp = realpath(dli_fname, buf);
2370   }
2371   if (rp == NULL)
2372     return;
2373 
2374   if (Arguments::sun_java_launcher_is_altjvm()) {
2375     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2376     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2377     // If "/jre/lib/" appears at the right place in the string, then
2378     // assume we are installed in a JDK and we're done. Otherwise, check
2379     // for a JAVA_HOME environment variable and fix up the path so it
2380     // looks like libjvm.so is installed there (append a fake suffix
2381     // hotspot/libjvm.so).
2382     const char *p = buf + strlen(buf) - 1;
2383     for (int count = 0; p > buf && count < 5; ++count) {
2384       for (--p; p > buf && *p != '/'; --p)
2385         /* empty */ ;
2386     }
2387 
2388     if (strncmp(p, "/jre/lib/", 9) != 0) {
2389       // Look for JAVA_HOME in the environment.
2390       char* java_home_var = ::getenv("JAVA_HOME");
2391       if (java_home_var != NULL && java_home_var[0] != 0) {
2392         char* jrelib_p;
2393         int len;
2394 
2395         // Check the current module name "libjvm.so".
2396         p = strrchr(buf, '/');
2397         assert(strstr(p, "/libjvm") == p, "invalid library name");
2398 
2399         rp = realpath(java_home_var, buf);
2400         if (rp == NULL)
2401           return;
2402 
2403         // determine if this is a legacy image or modules image
2404         // modules image doesn't have "jre" subdirectory
2405         len = strlen(buf);
2406         jrelib_p = buf + len;
2407         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2408         if (0 != access(buf, F_OK)) {
2409           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2410         }
2411 
2412         if (0 == access(buf, F_OK)) {
2413           // Use current module name "libjvm.so"
2414           len = strlen(buf);
2415           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2416         } else {
2417           // Go back to path of .so
2418           rp = realpath(dli_fname, buf);
2419           if (rp == NULL)
2420             return;
2421         }
2422       }
2423     }
2424   }
2425 
2426   strcpy(saved_jvm_path, buf);
2427 }
2428 
2429 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2430   // no prefix required, not even "_"
2431 }
2432 
2433 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2434   // no suffix required
2435 }
2436 
2437 ////////////////////////////////////////////////////////////////////////////////
2438 // sun.misc.Signal support
2439 
2440 static volatile jint sigint_count = 0;
2441 
2442 static void
2443 UserHandler(int sig, void *siginfo, void *context) {
2444   // 4511530 - sem_post is serialized and handled by the manager thread. When
2445   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2446   // don't want to flood the manager thread with sem_post requests.
2447   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2448       return;
2449 
2450   // Ctrl-C is pressed during error reporting, likely because the error
2451   // handler fails to abort. Let VM die immediately.
2452   if (sig == SIGINT && is_error_reported()) {
2453      os::die();
2454   }
2455 
2456   os::signal_notify(sig);
2457 }
2458 
2459 void* os::user_handler() {
2460   return CAST_FROM_FN_PTR(void*, UserHandler);
2461 }
2462 
2463 class Semaphore : public StackObj {
2464   public:
2465     Semaphore();
2466     ~Semaphore();
2467     void signal();
2468     void wait();
2469     bool trywait();
2470     bool timedwait(unsigned int sec, int nsec);
2471   private:
2472     sem_t _semaphore;
2473 };
2474 
2475 
2476 Semaphore::Semaphore() {
2477   sem_init(&_semaphore, 0, 0);
2478 }
2479 
2480 Semaphore::~Semaphore() {
2481   sem_destroy(&_semaphore);
2482 }
2483 
2484 void Semaphore::signal() {
2485   sem_post(&_semaphore);
2486 }
2487 
2488 void Semaphore::wait() {
2489   sem_wait(&_semaphore);
2490 }
2491 
2492 bool Semaphore::trywait() {
2493   return sem_trywait(&_semaphore) == 0;
2494 }
2495 
2496 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2497   struct timespec ts;
2498   unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2499 
2500   while (1) {
2501     int result = sem_timedwait(&_semaphore, &ts);
2502     if (result == 0) {
2503       return true;
2504     } else if (errno == EINTR) {
2505       continue;
2506     } else if (errno == ETIMEDOUT) {
2507       return false;
2508     } else {
2509       return false;
2510     }
2511   }
2512 }
2513 
2514 extern "C" {
2515   typedef void (*sa_handler_t)(int);
2516   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2517 }
2518 
2519 void* os::signal(int signal_number, void* handler) {
2520   struct sigaction sigAct, oldSigAct;
2521 
2522   sigfillset(&(sigAct.sa_mask));
2523   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2524   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2525 
2526   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2527     // -1 means registration failed
2528     return (void *)-1;
2529   }
2530 
2531   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2532 }
2533 
2534 void os::signal_raise(int signal_number) {
2535   ::raise(signal_number);
2536 }
2537 
2538 /*
2539  * The following code is moved from os.cpp for making this
2540  * code platform specific, which it is by its very nature.
2541  */
2542 
2543 // Will be modified when max signal is changed to be dynamic
2544 int os::sigexitnum_pd() {
2545   return NSIG;
2546 }
2547 
2548 // a counter for each possible signal value
2549 static volatile jint pending_signals[NSIG+1] = { 0 };
2550 
2551 // Linux(POSIX) specific hand shaking semaphore.
2552 static sem_t sig_sem;
2553 static Semaphore sr_semaphore;
2554 
2555 void os::signal_init_pd() {
2556   // Initialize signal structures
2557   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2558 
2559   // Initialize signal semaphore
2560   ::sem_init(&sig_sem, 0, 0);
2561 }
2562 
2563 void os::signal_notify(int sig) {
2564   Atomic::inc(&pending_signals[sig]);
2565   ::sem_post(&sig_sem);
2566 }
2567 
2568 static int check_pending_signals(bool wait) {
2569   Atomic::store(0, &sigint_count);
2570   for (;;) {
2571     for (int i = 0; i < NSIG + 1; i++) {
2572       jint n = pending_signals[i];
2573       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2574         return i;
2575       }
2576     }
2577     if (!wait) {
2578       return -1;
2579     }
2580     JavaThread *thread = JavaThread::current();
2581     ThreadBlockInVM tbivm(thread);
2582 
2583     bool threadIsSuspended;
2584     do {
2585       thread->set_suspend_equivalent();
2586       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2587       ::sem_wait(&sig_sem);
2588 
2589       // were we externally suspended while we were waiting?
2590       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2591       if (threadIsSuspended) {
2592         //
2593         // The semaphore has been incremented, but while we were waiting
2594         // another thread suspended us. We don't want to continue running
2595         // while suspended because that would surprise the thread that
2596         // suspended us.
2597         //
2598         ::sem_post(&sig_sem);
2599 
2600         thread->java_suspend_self();
2601       }
2602     } while (threadIsSuspended);
2603   }
2604 }
2605 
2606 int os::signal_lookup() {
2607   return check_pending_signals(false);
2608 }
2609 
2610 int os::signal_wait() {
2611   return check_pending_signals(true);
2612 }
2613 
2614 ////////////////////////////////////////////////////////////////////////////////
2615 // Virtual Memory
2616 
2617 int os::vm_page_size() {
2618   // Seems redundant as all get out
2619   assert(os::Linux::page_size() != -1, "must call os::init");
2620   return os::Linux::page_size();
2621 }
2622 
2623 // Solaris allocates memory by pages.
2624 int os::vm_allocation_granularity() {
2625   assert(os::Linux::page_size() != -1, "must call os::init");
2626   return os::Linux::page_size();
2627 }
2628 
2629 // Rationale behind this function:
2630 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2631 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2632 //  samples for JITted code. Here we create private executable mapping over the code cache
2633 //  and then we can use standard (well, almost, as mapping can change) way to provide
2634 //  info for the reporting script by storing timestamp and location of symbol
2635 void linux_wrap_code(char* base, size_t size) {
2636   static volatile jint cnt = 0;
2637 
2638   if (!UseOprofile) {
2639     return;
2640   }
2641 
2642   char buf[PATH_MAX+1];
2643   int num = Atomic::add(1, &cnt);
2644 
2645   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2646            os::get_temp_directory(), os::current_process_id(), num);
2647   unlink(buf);
2648 
2649   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2650 
2651   if (fd != -1) {
2652     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2653     if (rv != (off_t)-1) {
2654       if (::write(fd, "", 1) == 1) {
2655         mmap(base, size,
2656              PROT_READ|PROT_WRITE|PROT_EXEC,
2657              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2658       }
2659     }
2660     ::close(fd);
2661     unlink(buf);
2662   }
2663 }
2664 
2665 static bool recoverable_mmap_error(int err) {
2666   // See if the error is one we can let the caller handle. This
2667   // list of errno values comes from JBS-6843484. I can't find a
2668   // Linux man page that documents this specific set of errno
2669   // values so while this list currently matches Solaris, it may
2670   // change as we gain experience with this failure mode.
2671   switch (err) {
2672   case EBADF:
2673   case EINVAL:
2674   case ENOTSUP:
2675     // let the caller deal with these errors
2676     return true;
2677 
2678   default:
2679     // Any remaining errors on this OS can cause our reserved mapping
2680     // to be lost. That can cause confusion where different data
2681     // structures think they have the same memory mapped. The worst
2682     // scenario is if both the VM and a library think they have the
2683     // same memory mapped.
2684     return false;
2685   }
2686 }
2687 
2688 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2689                                     int err) {
2690   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2691           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2692           strerror(err), err);
2693 }
2694 
2695 static void warn_fail_commit_memory(char* addr, size_t size,
2696                                     size_t alignment_hint, bool exec,
2697                                     int err) {
2698   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2699           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2700           alignment_hint, exec, strerror(err), err);
2701 }
2702 
2703 // NOTE: Linux kernel does not really reserve the pages for us.
2704 //       All it does is to check if there are enough free pages
2705 //       left at the time of mmap(). This could be a potential
2706 //       problem.
2707 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2708   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2709   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2710                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2711   if (res != (uintptr_t) MAP_FAILED) {
2712     if (UseNUMAInterleaving) {
2713       numa_make_global(addr, size);
2714     }
2715     return 0;
2716   }
2717 
2718   int err = errno;  // save errno from mmap() call above
2719 
2720   if (!recoverable_mmap_error(err)) {
2721     warn_fail_commit_memory(addr, size, exec, err);
2722     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2723   }
2724 
2725   return err;
2726 }
2727 
2728 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2729   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2730 }
2731 
2732 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2733                                   const char* mesg) {
2734   assert(mesg != NULL, "mesg must be specified");
2735   int err = os::Linux::commit_memory_impl(addr, size, exec);
2736   if (err != 0) {
2737     // the caller wants all commit errors to exit with the specified mesg:
2738     warn_fail_commit_memory(addr, size, exec, err);
2739     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2740   }
2741 }
2742 
2743 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2744 #ifndef MAP_HUGETLB
2745 #define MAP_HUGETLB 0x40000
2746 #endif
2747 
2748 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2749 #ifndef MADV_HUGEPAGE
2750 #define MADV_HUGEPAGE 14
2751 #endif
2752 
2753 int os::Linux::commit_memory_impl(char* addr, size_t size,
2754                                   size_t alignment_hint, bool exec) {
2755   int err = os::Linux::commit_memory_impl(addr, size, exec);
2756   if (err == 0) {
2757     realign_memory(addr, size, alignment_hint);
2758   }
2759   return err;
2760 }
2761 
2762 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2763                           bool exec) {
2764   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2765 }
2766 
2767 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2768                                   size_t alignment_hint, bool exec,
2769                                   const char* mesg) {
2770   assert(mesg != NULL, "mesg must be specified");
2771   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2772   if (err != 0) {
2773     // the caller wants all commit errors to exit with the specified mesg:
2774     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2775     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2776   }
2777 }
2778 
2779 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2780   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2781     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2782     // be supported or the memory may already be backed by huge pages.
2783     ::madvise(addr, bytes, MADV_HUGEPAGE);
2784   }
2785 }
2786 
2787 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2788   // This method works by doing an mmap over an existing mmaping and effectively discarding
2789   // the existing pages. However it won't work for SHM-based large pages that cannot be
2790   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2791   // small pages on top of the SHM segment. This method always works for small pages, so we
2792   // allow that in any case.
2793   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2794     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2795   }
2796 }
2797 
2798 void os::numa_make_global(char *addr, size_t bytes) {
2799   Linux::numa_interleave_memory(addr, bytes);
2800 }
2801 
2802 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2803 // bind policy to MPOL_PREFERRED for the current thread.
2804 #define USE_MPOL_PREFERRED 0
2805 
2806 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2807   // To make NUMA and large pages more robust when both enabled, we need to ease
2808   // the requirements on where the memory should be allocated. MPOL_BIND is the
2809   // default policy and it will force memory to be allocated on the specified
2810   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2811   // the specified node, but will not force it. Using this policy will prevent
2812   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2813   // free large pages.
2814   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2815   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2816 }
2817 
2818 bool os::numa_topology_changed()   { return false; }
2819 
2820 size_t os::numa_get_groups_num() {
2821   int max_node = Linux::numa_max_node();
2822   return max_node > 0 ? max_node + 1 : 1;
2823 }
2824 
2825 int os::numa_get_group_id() {
2826   int cpu_id = Linux::sched_getcpu();
2827   if (cpu_id != -1) {
2828     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2829     if (lgrp_id != -1) {
2830       return lgrp_id;
2831     }
2832   }
2833   return 0;
2834 }
2835 
2836 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2837   for (size_t i = 0; i < size; i++) {
2838     ids[i] = i;
2839   }
2840   return size;
2841 }
2842 
2843 bool os::get_page_info(char *start, page_info* info) {
2844   return false;
2845 }
2846 
2847 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2848   return end;
2849 }
2850 
2851 
2852 int os::Linux::sched_getcpu_syscall(void) {
2853   unsigned int cpu;
2854   int retval = -1;
2855 
2856 #if defined(IA32)
2857 # ifndef SYS_getcpu
2858 # define SYS_getcpu 318
2859 # endif
2860   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2861 #elif defined(AMD64)
2862 // Unfortunately we have to bring all these macros here from vsyscall.h
2863 // to be able to compile on old linuxes.
2864 # define __NR_vgetcpu 2
2865 # define VSYSCALL_START (-10UL << 20)
2866 # define VSYSCALL_SIZE 1024
2867 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2868   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2869   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2870   retval = vgetcpu(&cpu, NULL, NULL);
2871 #endif
2872 
2873   return (retval == -1) ? retval : cpu;
2874 }
2875 
2876 // Something to do with the numa-aware allocator needs these symbols
2877 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2878 extern "C" JNIEXPORT void numa_error(char *where) { }
2879 extern "C" JNIEXPORT int fork1() { return fork(); }
2880 
2881 
2882 // If we are running with libnuma version > 2, then we should
2883 // be trying to use symbols with versions 1.1
2884 // If we are running with earlier version, which did not have symbol versions,
2885 // we should use the base version.
2886 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2887   void *f = dlvsym(handle, name, "libnuma_1.1");
2888   if (f == NULL) {
2889     f = dlsym(handle, name);
2890   }
2891   return f;
2892 }
2893 
2894 bool os::Linux::libnuma_init() {
2895   // sched_getcpu() should be in libc.
2896   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2897                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2898 
2899   // If it's not, try a direct syscall.
2900   if (sched_getcpu() == -1)
2901     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2902 
2903   if (sched_getcpu() != -1) { // Does it work?
2904     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2905     if (handle != NULL) {
2906       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2907                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2908       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2909                                        libnuma_dlsym(handle, "numa_max_node")));
2910       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2911                                         libnuma_dlsym(handle, "numa_available")));
2912       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2913                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2914       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2915                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2916       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2917                                             libnuma_dlsym(handle, "numa_set_bind_policy")));
2918 
2919 
2920       if (numa_available() != -1) {
2921         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2922         // Create a cpu -> node mapping
2923         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2924         rebuild_cpu_to_node_map();
2925         return true;
2926       }
2927     }
2928   }
2929   return false;
2930 }
2931 
2932 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2933 // The table is later used in get_node_by_cpu().
2934 void os::Linux::rebuild_cpu_to_node_map() {
2935   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2936                               // in libnuma (possible values are starting from 16,
2937                               // and continuing up with every other power of 2, but less
2938                               // than the maximum number of CPUs supported by kernel), and
2939                               // is a subject to change (in libnuma version 2 the requirements
2940                               // are more reasonable) we'll just hardcode the number they use
2941                               // in the library.
2942   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2943 
2944   size_t cpu_num = os::active_processor_count();
2945   size_t cpu_map_size = NCPUS / BitsPerCLong;
2946   size_t cpu_map_valid_size =
2947     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2948 
2949   cpu_to_node()->clear();
2950   cpu_to_node()->at_grow(cpu_num - 1);
2951   size_t node_num = numa_get_groups_num();
2952 
2953   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2954   for (size_t i = 0; i < node_num; i++) {
2955     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2956       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2957         if (cpu_map[j] != 0) {
2958           for (size_t k = 0; k < BitsPerCLong; k++) {
2959             if (cpu_map[j] & (1UL << k)) {
2960               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2961             }
2962           }
2963         }
2964       }
2965     }
2966   }
2967   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2968 }
2969 
2970 int os::Linux::get_node_by_cpu(int cpu_id) {
2971   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2972     return cpu_to_node()->at(cpu_id);
2973   }
2974   return -1;
2975 }
2976 
2977 GrowableArray<int>* os::Linux::_cpu_to_node;
2978 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2979 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2980 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2981 os::Linux::numa_available_func_t os::Linux::_numa_available;
2982 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2983 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2984 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2985 unsigned long* os::Linux::_numa_all_nodes;
2986 
2987 bool os::pd_uncommit_memory(char* addr, size_t size) {
2988   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2989                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2990   return res  != (uintptr_t) MAP_FAILED;
2991 }
2992 
2993 static
2994 address get_stack_commited_bottom(address bottom, size_t size) {
2995   address nbot = bottom;
2996   address ntop = bottom + size;
2997 
2998   size_t page_sz = os::vm_page_size();
2999   unsigned pages = size / page_sz;
3000 
3001   unsigned char vec[1];
3002   unsigned imin = 1, imax = pages + 1, imid;
3003   int mincore_return_value = 0;
3004 
3005   assert(imin <= imax, "Unexpected page size");
3006 
3007   while (imin < imax) {
3008     imid = (imax + imin) / 2;
3009     nbot = ntop - (imid * page_sz);
3010 
3011     // Use a trick with mincore to check whether the page is mapped or not.
3012     // mincore sets vec to 1 if page resides in memory and to 0 if page
3013     // is swapped output but if page we are asking for is unmapped
3014     // it returns -1,ENOMEM
3015     mincore_return_value = mincore(nbot, page_sz, vec);
3016 
3017     if (mincore_return_value == -1) {
3018       // Page is not mapped go up
3019       // to find first mapped page
3020       if (errno != EAGAIN) {
3021         assert(errno == ENOMEM, "Unexpected mincore errno");
3022         imax = imid;
3023       }
3024     } else {
3025       // Page is mapped go down
3026       // to find first not mapped page
3027       imin = imid + 1;
3028     }
3029   }
3030 
3031   nbot = nbot + page_sz;
3032 
3033   // Adjust stack bottom one page up if last checked page is not mapped
3034   if (mincore_return_value == -1) {
3035     nbot = nbot + page_sz;
3036   }
3037 
3038   return nbot;
3039 }
3040 
3041 
3042 // Linux uses a growable mapping for the stack, and if the mapping for
3043 // the stack guard pages is not removed when we detach a thread the
3044 // stack cannot grow beyond the pages where the stack guard was
3045 // mapped.  If at some point later in the process the stack expands to
3046 // that point, the Linux kernel cannot expand the stack any further
3047 // because the guard pages are in the way, and a segfault occurs.
3048 //
3049 // However, it's essential not to split the stack region by unmapping
3050 // a region (leaving a hole) that's already part of the stack mapping,
3051 // so if the stack mapping has already grown beyond the guard pages at
3052 // the time we create them, we have to truncate the stack mapping.
3053 // So, we need to know the extent of the stack mapping when
3054 // create_stack_guard_pages() is called.
3055 
3056 // We only need this for stacks that are growable: at the time of
3057 // writing thread stacks don't use growable mappings (i.e. those
3058 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3059 // only applies to the main thread.
3060 
3061 // If the (growable) stack mapping already extends beyond the point
3062 // where we're going to put our guard pages, truncate the mapping at
3063 // that point by munmap()ping it.  This ensures that when we later
3064 // munmap() the guard pages we don't leave a hole in the stack
3065 // mapping. This only affects the main/initial thread
3066 
3067 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3068 
3069   if (os::Linux::is_initial_thread()) {
3070     // As we manually grow stack up to bottom inside create_attached_thread(),
3071     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3072     // we don't need to do anything special.
3073     // Check it first, before calling heavy function.
3074     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3075     unsigned char vec[1];
3076 
3077     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3078       // Fallback to slow path on all errors, including EAGAIN
3079       stack_extent = (uintptr_t) get_stack_commited_bottom(
3080                                     os::Linux::initial_thread_stack_bottom(),
3081                                     (size_t)addr - stack_extent);
3082     }
3083 
3084     if (stack_extent < (uintptr_t)addr) {
3085       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3086     }
3087   }
3088 
3089   return os::commit_memory(addr, size, !ExecMem);
3090 }
3091 
3092 // If this is a growable mapping, remove the guard pages entirely by
3093 // munmap()ping them.  If not, just call uncommit_memory(). This only
3094 // affects the main/initial thread, but guard against future OS changes
3095 // It's safe to always unmap guard pages for initial thread because we
3096 // always place it right after end of the mapped region
3097 
3098 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3099   uintptr_t stack_extent, stack_base;
3100 
3101   if (os::Linux::is_initial_thread()) {
3102     return ::munmap(addr, size) == 0;
3103   }
3104 
3105   return os::uncommit_memory(addr, size);
3106 }
3107 
3108 static address _highest_vm_reserved_address = NULL;
3109 
3110 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3111 // at 'requested_addr'. If there are existing memory mappings at the same
3112 // location, however, they will be overwritten. If 'fixed' is false,
3113 // 'requested_addr' is only treated as a hint, the return value may or
3114 // may not start from the requested address. Unlike Linux mmap(), this
3115 // function returns NULL to indicate failure.
3116 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3117   char * addr;
3118   int flags;
3119 
3120   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3121   if (fixed) {
3122     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3123     flags |= MAP_FIXED;
3124   }
3125 
3126   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3127   // touch an uncommitted page. Otherwise, the read/write might
3128   // succeed if we have enough swap space to back the physical page.
3129   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3130                        flags, -1, 0);
3131 
3132   if (addr != MAP_FAILED) {
3133     // anon_mmap() should only get called during VM initialization,
3134     // don't need lock (actually we can skip locking even it can be called
3135     // from multiple threads, because _highest_vm_reserved_address is just a
3136     // hint about the upper limit of non-stack memory regions.)
3137     if ((address)addr + bytes > _highest_vm_reserved_address) {
3138       _highest_vm_reserved_address = (address)addr + bytes;
3139     }
3140   }
3141 
3142   return addr == MAP_FAILED ? NULL : addr;
3143 }
3144 
3145 // Don't update _highest_vm_reserved_address, because there might be memory
3146 // regions above addr + size. If so, releasing a memory region only creates
3147 // a hole in the address space, it doesn't help prevent heap-stack collision.
3148 //
3149 static int anon_munmap(char * addr, size_t size) {
3150   return ::munmap(addr, size) == 0;
3151 }
3152 
3153 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3154                          size_t alignment_hint) {
3155   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3156 }
3157 
3158 bool os::pd_release_memory(char* addr, size_t size) {
3159   return anon_munmap(addr, size);
3160 }
3161 
3162 static address highest_vm_reserved_address() {
3163   return _highest_vm_reserved_address;
3164 }
3165 
3166 static bool linux_mprotect(char* addr, size_t size, int prot) {
3167   // Linux wants the mprotect address argument to be page aligned.
3168   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3169 
3170   // According to SUSv3, mprotect() should only be used with mappings
3171   // established by mmap(), and mmap() always maps whole pages. Unaligned
3172   // 'addr' likely indicates problem in the VM (e.g. trying to change
3173   // protection of malloc'ed or statically allocated memory). Check the
3174   // caller if you hit this assert.
3175   assert(addr == bottom, "sanity check");
3176 
3177   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3178   return ::mprotect(bottom, size, prot) == 0;
3179 }
3180 
3181 // Set protections specified
3182 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3183                         bool is_committed) {
3184   unsigned int p = 0;
3185   switch (prot) {
3186   case MEM_PROT_NONE: p = PROT_NONE; break;
3187   case MEM_PROT_READ: p = PROT_READ; break;
3188   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3189   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3190   default:
3191     ShouldNotReachHere();
3192   }
3193   // is_committed is unused.
3194   return linux_mprotect(addr, bytes, p);
3195 }
3196 
3197 bool os::guard_memory(char* addr, size_t size) {
3198   return linux_mprotect(addr, size, PROT_NONE);
3199 }
3200 
3201 bool os::unguard_memory(char* addr, size_t size) {
3202   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3203 }
3204 
3205 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3206   bool result = false;
3207   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3208                  MAP_ANONYMOUS|MAP_PRIVATE,
3209                  -1, 0);
3210   if (p != MAP_FAILED) {
3211     void *aligned_p = align_ptr_up(p, page_size);
3212 
3213     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3214 
3215     munmap(p, page_size * 2);
3216   }
3217 
3218   if (warn && !result) {
3219     warning("TransparentHugePages is not supported by the operating system.");
3220   }
3221 
3222   return result;
3223 }
3224 
3225 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3226   bool result = false;
3227   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3228                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3229                  -1, 0);
3230 
3231   if (p != MAP_FAILED) {
3232     // We don't know if this really is a huge page or not.
3233     FILE *fp = fopen("/proc/self/maps", "r");
3234     if (fp) {
3235       while (!feof(fp)) {
3236         char chars[257];
3237         long x = 0;
3238         if (fgets(chars, sizeof(chars), fp)) {
3239           if (sscanf(chars, "%lx-%*x", &x) == 1
3240               && x == (long)p) {
3241             if (strstr (chars, "hugepage")) {
3242               result = true;
3243               break;
3244             }
3245           }
3246         }
3247       }
3248       fclose(fp);
3249     }
3250     munmap(p, page_size);
3251   }
3252 
3253   if (warn && !result) {
3254     warning("HugeTLBFS is not supported by the operating system.");
3255   }
3256 
3257   return result;
3258 }
3259 
3260 /*
3261 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3262 *
3263 * From the coredump_filter documentation:
3264 *
3265 * - (bit 0) anonymous private memory
3266 * - (bit 1) anonymous shared memory
3267 * - (bit 2) file-backed private memory
3268 * - (bit 3) file-backed shared memory
3269 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3270 *           effective only if the bit 2 is cleared)
3271 * - (bit 5) hugetlb private memory
3272 * - (bit 6) hugetlb shared memory
3273 */
3274 static void set_coredump_filter(void) {
3275   FILE *f;
3276   long cdm;
3277 
3278   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3279     return;
3280   }
3281 
3282   if (fscanf(f, "%lx", &cdm) != 1) {
3283     fclose(f);
3284     return;
3285   }
3286 
3287   rewind(f);
3288 
3289   if ((cdm & LARGEPAGES_BIT) == 0) {
3290     cdm |= LARGEPAGES_BIT;
3291     fprintf(f, "%#lx", cdm);
3292   }
3293 
3294   fclose(f);
3295 }
3296 
3297 // Large page support
3298 
3299 static size_t _large_page_size = 0;
3300 
3301 size_t os::Linux::find_large_page_size() {
3302   size_t large_page_size = 0;
3303 
3304   // large_page_size on Linux is used to round up heap size. x86 uses either
3305   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3306   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3307   // page as large as 256M.
3308   //
3309   // Here we try to figure out page size by parsing /proc/meminfo and looking
3310   // for a line with the following format:
3311   //    Hugepagesize:     2048 kB
3312   //
3313   // If we can't determine the value (e.g. /proc is not mounted, or the text
3314   // format has been changed), we'll use the largest page size supported by
3315   // the processor.
3316 
3317 #ifndef ZERO
3318   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3319                      ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3320 #endif // ZERO
3321 
3322   FILE *fp = fopen("/proc/meminfo", "r");
3323   if (fp) {
3324     while (!feof(fp)) {
3325       int x = 0;
3326       char buf[16];
3327       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3328         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3329           large_page_size = x * K;
3330           break;
3331         }
3332       } else {
3333         // skip to next line
3334         for (;;) {
3335           int ch = fgetc(fp);
3336           if (ch == EOF || ch == (int)'\n') break;
3337         }
3338       }
3339     }
3340     fclose(fp);
3341   }
3342 
3343   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3344     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3345         SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3346         proper_unit_for_byte_size(large_page_size));
3347   }
3348 
3349   return large_page_size;
3350 }
3351 
3352 size_t os::Linux::setup_large_page_size() {
3353   _large_page_size = Linux::find_large_page_size();
3354   const size_t default_page_size = (size_t)Linux::page_size();
3355   if (_large_page_size > default_page_size) {
3356     _page_sizes[0] = _large_page_size;
3357     _page_sizes[1] = default_page_size;
3358     _page_sizes[2] = 0;
3359   }
3360 
3361   return _large_page_size;
3362 }
3363 
3364 bool os::Linux::setup_large_page_type(size_t page_size) {
3365   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3366       FLAG_IS_DEFAULT(UseSHM) &&
3367       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3368 
3369     // The type of large pages has not been specified by the user.
3370 
3371     // Try UseHugeTLBFS and then UseSHM.
3372     UseHugeTLBFS = UseSHM = true;
3373 
3374     // Don't try UseTransparentHugePages since there are known
3375     // performance issues with it turned on. This might change in the future.
3376     UseTransparentHugePages = false;
3377   }
3378 
3379   if (UseTransparentHugePages) {
3380     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3381     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3382       UseHugeTLBFS = false;
3383       UseSHM = false;
3384       return true;
3385     }
3386     UseTransparentHugePages = false;
3387   }
3388 
3389   if (UseHugeTLBFS) {
3390     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3391     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3392       UseSHM = false;
3393       return true;
3394     }
3395     UseHugeTLBFS = false;
3396   }
3397 
3398   return UseSHM;
3399 }
3400 
3401 void os::large_page_init() {
3402   if (!UseLargePages &&
3403       !UseTransparentHugePages &&
3404       !UseHugeTLBFS &&
3405       !UseSHM) {
3406     // Not using large pages.
3407     return;
3408   }
3409 
3410   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3411     // The user explicitly turned off large pages.
3412     // Ignore the rest of the large pages flags.
3413     UseTransparentHugePages = false;
3414     UseHugeTLBFS = false;
3415     UseSHM = false;
3416     return;
3417   }
3418 
3419   size_t large_page_size = Linux::setup_large_page_size();
3420   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3421 
3422   set_coredump_filter();
3423 }
3424 
3425 #ifndef SHM_HUGETLB
3426 #define SHM_HUGETLB 04000
3427 #endif
3428 
3429 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3430   // "exec" is passed in but not used.  Creating the shared image for
3431   // the code cache doesn't have an SHM_X executable permission to check.
3432   assert(UseLargePages && UseSHM, "only for SHM large pages");
3433   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3434 
3435   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3436     return NULL; // Fallback to small pages.
3437   }
3438 
3439   key_t key = IPC_PRIVATE;
3440   char *addr;
3441 
3442   bool warn_on_failure = UseLargePages &&
3443                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3444                          !FLAG_IS_DEFAULT(UseSHM) ||
3445                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3446                         );
3447   char msg[128];
3448 
3449   // Create a large shared memory region to attach to based on size.
3450   // Currently, size is the total size of the heap
3451   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3452   if (shmid == -1) {
3453      // Possible reasons for shmget failure:
3454      // 1. shmmax is too small for Java heap.
3455      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3456      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3457      // 2. not enough large page memory.
3458      //    > check available large pages: cat /proc/meminfo
3459      //    > increase amount of large pages:
3460      //          echo new_value > /proc/sys/vm/nr_hugepages
3461      //      Note 1: different Linux may use different name for this property,
3462      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3463      //      Note 2: it's possible there's enough physical memory available but
3464      //            they are so fragmented after a long run that they can't
3465      //            coalesce into large pages. Try to reserve large pages when
3466      //            the system is still "fresh".
3467      if (warn_on_failure) {
3468        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3469        warning(msg);
3470      }
3471      return NULL;
3472   }
3473 
3474   // attach to the region
3475   addr = (char*)shmat(shmid, req_addr, 0);
3476   int err = errno;
3477 
3478   // Remove shmid. If shmat() is successful, the actual shared memory segment
3479   // will be deleted when it's detached by shmdt() or when the process
3480   // terminates. If shmat() is not successful this will remove the shared
3481   // segment immediately.
3482   shmctl(shmid, IPC_RMID, NULL);
3483 
3484   if ((intptr_t)addr == -1) {
3485      if (warn_on_failure) {
3486        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3487        warning(msg);
3488      }
3489      return NULL;
3490   }
3491 
3492   return addr;
3493 }
3494 
3495 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3496   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3497 
3498   bool warn_on_failure = UseLargePages &&
3499       (!FLAG_IS_DEFAULT(UseLargePages) ||
3500        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3501        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3502 
3503   if (warn_on_failure) {
3504     char msg[128];
3505     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3506         PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3507     warning(msg);
3508   }
3509 }
3510 
3511 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3512   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3513   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3514   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3515 
3516   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3517   char* addr = (char*)::mmap(req_addr, bytes, prot,
3518                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3519                              -1, 0);
3520 
3521   if (addr == MAP_FAILED) {
3522     warn_on_large_pages_failure(req_addr, bytes, errno);
3523     return NULL;
3524   }
3525 
3526   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3527 
3528   return addr;
3529 }
3530 
3531 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3532   size_t large_page_size = os::large_page_size();
3533 
3534   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3535 
3536   // Allocate small pages.
3537 
3538   char* start;
3539   if (req_addr != NULL) {
3540     assert(is_ptr_aligned(req_addr, alignment), "Must be");
3541     assert(is_size_aligned(bytes, alignment), "Must be");
3542     start = os::reserve_memory(bytes, req_addr);
3543     assert(start == NULL || start == req_addr, "Must be");
3544   } else {
3545     start = os::reserve_memory_aligned(bytes, alignment);
3546   }
3547 
3548   if (start == NULL) {
3549     return NULL;
3550   }
3551 
3552   assert(is_ptr_aligned(start, alignment), "Must be");
3553 
3554   // os::reserve_memory_special will record this memory area.
3555   // Need to release it here to prevent overlapping reservations.
3556   MemTracker::record_virtual_memory_release((address)start, bytes);
3557 
3558   char* end = start + bytes;
3559 
3560   // Find the regions of the allocated chunk that can be promoted to large pages.
3561   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3562   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3563 
3564   size_t lp_bytes = lp_end - lp_start;
3565 
3566   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3567 
3568   if (lp_bytes == 0) {
3569     // The mapped region doesn't even span the start and the end of a large page.
3570     // Fall back to allocate a non-special area.
3571     ::munmap(start, end - start);
3572     return NULL;
3573   }
3574 
3575   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3576 
3577 
3578   void* result;
3579 
3580   if (start != lp_start) {
3581     result = ::mmap(start, lp_start - start, prot,
3582                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3583                     -1, 0);
3584     if (result == MAP_FAILED) {
3585       ::munmap(lp_start, end - lp_start);
3586       return NULL;
3587     }
3588   }
3589 
3590   result = ::mmap(lp_start, lp_bytes, prot,
3591                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3592                   -1, 0);
3593   if (result == MAP_FAILED) {
3594     warn_on_large_pages_failure(req_addr, bytes, errno);
3595     // If the mmap above fails, the large pages region will be unmapped and we
3596     // have regions before and after with small pages. Release these regions.
3597     //
3598     // |  mapped  |  unmapped  |  mapped  |
3599     // ^          ^            ^          ^
3600     // start      lp_start     lp_end     end
3601     //
3602     ::munmap(start, lp_start - start);
3603     ::munmap(lp_end, end - lp_end);
3604     return NULL;
3605   }
3606 
3607   if (lp_end != end) {
3608       result = ::mmap(lp_end, end - lp_end, prot,
3609                       MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3610                       -1, 0);
3611     if (result == MAP_FAILED) {
3612       ::munmap(start, lp_end - start);
3613       return NULL;
3614     }
3615   }
3616 
3617   return start;
3618 }
3619 
3620 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3621   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3622   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3623   assert(is_power_of_2(alignment), "Must be");
3624   assert(is_power_of_2(os::large_page_size()), "Must be");
3625   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3626 
3627   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3628     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3629   } else {
3630     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3631   }
3632 }
3633 
3634 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3635   assert(UseLargePages, "only for large pages");
3636 
3637   char* addr;
3638   if (UseSHM) {
3639     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3640   } else {
3641     assert(UseHugeTLBFS, "must be");
3642     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3643   }
3644 
3645   if (addr != NULL) {
3646     if (UseNUMAInterleaving) {
3647       numa_make_global(addr, bytes);
3648     }
3649 
3650     // The memory is committed
3651     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3652   }
3653 
3654   return addr;
3655 }
3656 
3657 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3658   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3659   return shmdt(base) == 0;
3660 }
3661 
3662 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3663   return pd_release_memory(base, bytes);
3664 }
3665 
3666 bool os::release_memory_special(char* base, size_t bytes) {
3667   assert(UseLargePages, "only for large pages");
3668 
3669   MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3670 
3671   bool res;
3672   if (UseSHM) {
3673     res = os::Linux::release_memory_special_shm(base, bytes);
3674   } else {
3675     assert(UseHugeTLBFS, "must be");
3676     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3677   }
3678 
3679   if (res) {
3680     tkr.record((address)base, bytes);
3681   } else {
3682     tkr.discard();
3683   }
3684 
3685   return res;
3686 }
3687 
3688 size_t os::large_page_size() {
3689   return _large_page_size;
3690 }
3691 
3692 // With SysV SHM the entire memory region must be allocated as shared
3693 // memory.
3694 // HugeTLBFS allows application to commit large page memory on demand.
3695 // However, when committing memory with HugeTLBFS fails, the region
3696 // that was supposed to be committed will lose the old reservation
3697 // and allow other threads to steal that memory region. Because of this
3698 // behavior we can't commit HugeTLBFS memory.
3699 bool os::can_commit_large_page_memory() {
3700   return UseTransparentHugePages;
3701 }
3702 
3703 bool os::can_execute_large_page_memory() {
3704   return UseTransparentHugePages || UseHugeTLBFS;
3705 }
3706 
3707 // Reserve memory at an arbitrary address, only if that area is
3708 // available (and not reserved for something else).
3709 
3710 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3711   const int max_tries = 10;
3712   char* base[max_tries];
3713   size_t size[max_tries];
3714   const size_t gap = 0x000000;
3715 
3716   // Assert only that the size is a multiple of the page size, since
3717   // that's all that mmap requires, and since that's all we really know
3718   // about at this low abstraction level.  If we need higher alignment,
3719   // we can either pass an alignment to this method or verify alignment
3720   // in one of the methods further up the call chain.  See bug 5044738.
3721   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3722 
3723   // Repeatedly allocate blocks until the block is allocated at the
3724   // right spot. Give up after max_tries. Note that reserve_memory() will
3725   // automatically update _highest_vm_reserved_address if the call is
3726   // successful. The variable tracks the highest memory address every reserved
3727   // by JVM. It is used to detect heap-stack collision if running with
3728   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3729   // space than needed, it could confuse the collision detecting code. To
3730   // solve the problem, save current _highest_vm_reserved_address and
3731   // calculate the correct value before return.
3732   address old_highest = _highest_vm_reserved_address;
3733 
3734   // Linux mmap allows caller to pass an address as hint; give it a try first,
3735   // if kernel honors the hint then we can return immediately.
3736   char * addr = anon_mmap(requested_addr, bytes, false);
3737   if (addr == requested_addr) {
3738      return requested_addr;
3739   }
3740 
3741   if (addr != NULL) {
3742      // mmap() is successful but it fails to reserve at the requested address
3743      anon_munmap(addr, bytes);
3744   }
3745 
3746   int i;
3747   for (i = 0; i < max_tries; ++i) {
3748     base[i] = reserve_memory(bytes);
3749 
3750     if (base[i] != NULL) {
3751       // Is this the block we wanted?
3752       if (base[i] == requested_addr) {
3753         size[i] = bytes;
3754         break;
3755       }
3756 
3757       // Does this overlap the block we wanted? Give back the overlapped
3758       // parts and try again.
3759 
3760       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3761       if (top_overlap >= 0 && top_overlap < bytes) {
3762         unmap_memory(base[i], top_overlap);
3763         base[i] += top_overlap;
3764         size[i] = bytes - top_overlap;
3765       } else {
3766         size_t bottom_overlap = base[i] + bytes - requested_addr;
3767         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3768           unmap_memory(requested_addr, bottom_overlap);
3769           size[i] = bytes - bottom_overlap;
3770         } else {
3771           size[i] = bytes;
3772         }
3773       }
3774     }
3775   }
3776 
3777   // Give back the unused reserved pieces.
3778 
3779   for (int j = 0; j < i; ++j) {
3780     if (base[j] != NULL) {
3781       unmap_memory(base[j], size[j]);
3782     }
3783   }
3784 
3785   if (i < max_tries) {
3786     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3787     return requested_addr;
3788   } else {
3789     _highest_vm_reserved_address = old_highest;
3790     return NULL;
3791   }
3792 }
3793 
3794 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3795   return ::read(fd, buf, nBytes);
3796 }
3797 
3798 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3799 // Solaris uses poll(), linux uses park().
3800 // Poll() is likely a better choice, assuming that Thread.interrupt()
3801 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3802 // SIGSEGV, see 4355769.
3803 
3804 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3805   assert(thread == Thread::current(),  "thread consistency check");
3806 
3807   ParkEvent * const slp = thread->_SleepEvent ;
3808   slp->reset() ;
3809   OrderAccess::fence() ;
3810 
3811   if (interruptible) {
3812     jlong prevtime = javaTimeNanos();
3813 
3814     for (;;) {
3815       if (os::is_interrupted(thread, true)) {
3816         return OS_INTRPT;
3817       }
3818 
3819       jlong newtime = javaTimeNanos();
3820 
3821       if (newtime - prevtime < 0) {
3822         // time moving backwards, should only happen if no monotonic clock
3823         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3824         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3825       } else {
3826         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3827       }
3828 
3829       if(millis <= 0) {
3830         return OS_OK;
3831       }
3832 
3833       prevtime = newtime;
3834 
3835       {
3836         assert(thread->is_Java_thread(), "sanity check");
3837         JavaThread *jt = (JavaThread *) thread;
3838         ThreadBlockInVM tbivm(jt);
3839         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3840 
3841         jt->set_suspend_equivalent();
3842         // cleared by handle_special_suspend_equivalent_condition() or
3843         // java_suspend_self() via check_and_wait_while_suspended()
3844 
3845         slp->park(millis);
3846 
3847         // were we externally suspended while we were waiting?
3848         jt->check_and_wait_while_suspended();
3849       }
3850     }
3851   } else {
3852     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3853     jlong prevtime = javaTimeNanos();
3854 
3855     for (;;) {
3856       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3857       // the 1st iteration ...
3858       jlong newtime = javaTimeNanos();
3859 
3860       if (newtime - prevtime < 0) {
3861         // time moving backwards, should only happen if no monotonic clock
3862         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3863         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3864       } else {
3865         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3866       }
3867 
3868       if(millis <= 0) break ;
3869 
3870       prevtime = newtime;
3871       slp->park(millis);
3872     }
3873     return OS_OK ;
3874   }
3875 }
3876 
3877 //
3878 // Short sleep, direct OS call.
3879 //
3880 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3881 // sched_yield(2) will actually give up the CPU:
3882 //
3883 //   * Alone on this pariticular CPU, keeps running.
3884 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3885 //     (pre 2.6.39).
3886 //
3887 // So calling this with 0 is an alternative.
3888 //
3889 void os::naked_short_sleep(jlong ms) {
3890   struct timespec req;
3891 
3892   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3893   req.tv_sec = 0;
3894   if (ms > 0) {
3895     req.tv_nsec = (ms % 1000) * 1000000;
3896   }
3897   else {
3898     req.tv_nsec = 1;
3899   }
3900 
3901   nanosleep(&req, NULL);
3902 
3903   return;
3904 }
3905 
3906 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3907 void os::infinite_sleep() {
3908   while (true) {    // sleep forever ...
3909     ::sleep(100);   // ... 100 seconds at a time
3910   }
3911 }
3912 
3913 // Used to convert frequent JVM_Yield() to nops
3914 bool os::dont_yield() {
3915   return DontYieldALot;
3916 }
3917 
3918 void os::yield() {
3919   sched_yield();
3920 }
3921 
3922 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3923 
3924 void os::yield_all(int attempts) {
3925   // Yields to all threads, including threads with lower priorities
3926   // Threads on Linux are all with same priority. The Solaris style
3927   // os::yield_all() with nanosleep(1ms) is not necessary.
3928   sched_yield();
3929 }
3930 
3931 // Called from the tight loops to possibly influence time-sharing heuristics
3932 void os::loop_breaker(int attempts) {
3933   os::yield_all(attempts);
3934 }
3935 
3936 ////////////////////////////////////////////////////////////////////////////////
3937 // thread priority support
3938 
3939 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3940 // only supports dynamic priority, static priority must be zero. For real-time
3941 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3942 // However, for large multi-threaded applications, SCHED_RR is not only slower
3943 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3944 // of 5 runs - Sep 2005).
3945 //
3946 // The following code actually changes the niceness of kernel-thread/LWP. It
3947 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3948 // not the entire user process, and user level threads are 1:1 mapped to kernel
3949 // threads. It has always been the case, but could change in the future. For
3950 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3951 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3952 
3953 int os::java_to_os_priority[CriticalPriority + 1] = {
3954   19,              // 0 Entry should never be used
3955 
3956    4,              // 1 MinPriority
3957    3,              // 2
3958    2,              // 3
3959 
3960    1,              // 4
3961    0,              // 5 NormPriority
3962   -1,              // 6
3963 
3964   -2,              // 7
3965   -3,              // 8
3966   -4,              // 9 NearMaxPriority
3967 
3968   -5,              // 10 MaxPriority
3969 
3970   -5               // 11 CriticalPriority
3971 };
3972 
3973 static int prio_init() {
3974   if (ThreadPriorityPolicy == 1) {
3975     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3976     // if effective uid is not root. Perhaps, a more elegant way of doing
3977     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3978     if (geteuid() != 0) {
3979       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3980         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3981       }
3982       ThreadPriorityPolicy = 0;
3983     }
3984   }
3985   if (UseCriticalJavaThreadPriority) {
3986     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3987   }
3988   return 0;
3989 }
3990 
3991 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3992   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3993 
3994   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3995   return (ret == 0) ? OS_OK : OS_ERR;
3996 }
3997 
3998 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3999   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4000     *priority_ptr = java_to_os_priority[NormPriority];
4001     return OS_OK;
4002   }
4003 
4004   errno = 0;
4005   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4006   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4007 }
4008 
4009 // Hint to the underlying OS that a task switch would not be good.
4010 // Void return because it's a hint and can fail.
4011 void os::hint_no_preempt() {}
4012 
4013 ////////////////////////////////////////////////////////////////////////////////
4014 // suspend/resume support
4015 
4016 //  the low-level signal-based suspend/resume support is a remnant from the
4017 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4018 //  within hotspot. Now there is a single use-case for this:
4019 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
4020 //      that runs in the watcher thread.
4021 //  The remaining code is greatly simplified from the more general suspension
4022 //  code that used to be used.
4023 //
4024 //  The protocol is quite simple:
4025 //  - suspend:
4026 //      - sends a signal to the target thread
4027 //      - polls the suspend state of the osthread using a yield loop
4028 //      - target thread signal handler (SR_handler) sets suspend state
4029 //        and blocks in sigsuspend until continued
4030 //  - resume:
4031 //      - sets target osthread state to continue
4032 //      - sends signal to end the sigsuspend loop in the SR_handler
4033 //
4034 //  Note that the SR_lock plays no role in this suspend/resume protocol.
4035 //
4036 
4037 static void resume_clear_context(OSThread *osthread) {
4038   osthread->set_ucontext(NULL);
4039   osthread->set_siginfo(NULL);
4040 }
4041 
4042 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4043   osthread->set_ucontext(context);
4044   osthread->set_siginfo(siginfo);
4045 }
4046 
4047 //
4048 // Handler function invoked when a thread's execution is suspended or
4049 // resumed. We have to be careful that only async-safe functions are
4050 // called here (Note: most pthread functions are not async safe and
4051 // should be avoided.)
4052 //
4053 // Note: sigwait() is a more natural fit than sigsuspend() from an
4054 // interface point of view, but sigwait() prevents the signal hander
4055 // from being run. libpthread would get very confused by not having
4056 // its signal handlers run and prevents sigwait()'s use with the
4057 // mutex granting granting signal.
4058 //
4059 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4060 //
4061 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4062   // Save and restore errno to avoid confusing native code with EINTR
4063   // after sigsuspend.
4064   int old_errno = errno;
4065 
4066   Thread* thread = Thread::current();
4067   OSThread* osthread = thread->osthread();
4068   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4069 
4070   os::SuspendResume::State current = osthread->sr.state();
4071   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4072     suspend_save_context(osthread, siginfo, context);
4073 
4074     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4075     os::SuspendResume::State state = osthread->sr.suspended();
4076     if (state == os::SuspendResume::SR_SUSPENDED) {
4077       sigset_t suspend_set;  // signals for sigsuspend()
4078 
4079       // get current set of blocked signals and unblock resume signal
4080       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4081       sigdelset(&suspend_set, SR_signum);
4082 
4083       sr_semaphore.signal();
4084       // wait here until we are resumed
4085       while (1) {
4086         sigsuspend(&suspend_set);
4087 
4088         os::SuspendResume::State result = osthread->sr.running();
4089         if (result == os::SuspendResume::SR_RUNNING) {
4090           sr_semaphore.signal();
4091           break;
4092         }
4093       }
4094 
4095     } else if (state == os::SuspendResume::SR_RUNNING) {
4096       // request was cancelled, continue
4097     } else {
4098       ShouldNotReachHere();
4099     }
4100 
4101     resume_clear_context(osthread);
4102   } else if (current == os::SuspendResume::SR_RUNNING) {
4103     // request was cancelled, continue
4104   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4105     // ignore
4106   } else {
4107     // ignore
4108   }
4109 
4110   errno = old_errno;
4111 }
4112 
4113 
4114 static int SR_initialize() {
4115   struct sigaction act;
4116   char *s;
4117   /* Get signal number to use for suspend/resume */
4118   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4119     int sig = ::strtol(s, 0, 10);
4120     if (sig > 0 || sig < _NSIG) {
4121         SR_signum = sig;
4122     }
4123   }
4124 
4125   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4126         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4127 
4128   sigemptyset(&SR_sigset);
4129   sigaddset(&SR_sigset, SR_signum);
4130 
4131   /* Set up signal handler for suspend/resume */
4132   act.sa_flags = SA_RESTART|SA_SIGINFO;
4133   act.sa_handler = (void (*)(int)) SR_handler;
4134 
4135   // SR_signum is blocked by default.
4136   // 4528190 - We also need to block pthread restart signal (32 on all
4137   // supported Linux platforms). Note that LinuxThreads need to block
4138   // this signal for all threads to work properly. So we don't have
4139   // to use hard-coded signal number when setting up the mask.
4140   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4141 
4142   if (sigaction(SR_signum, &act, 0) == -1) {
4143     return -1;
4144   }
4145 
4146   // Save signal flag
4147   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4148   return 0;
4149 }
4150 
4151 static int sr_notify(OSThread* osthread) {
4152   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4153   assert_status(status == 0, status, "pthread_kill");
4154   return status;
4155 }
4156 
4157 // "Randomly" selected value for how long we want to spin
4158 // before bailing out on suspending a thread, also how often
4159 // we send a signal to a thread we want to resume
4160 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4161 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4162 
4163 // returns true on success and false on error - really an error is fatal
4164 // but this seems the normal response to library errors
4165 static bool do_suspend(OSThread* osthread) {
4166   assert(osthread->sr.is_running(), "thread should be running");
4167   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4168 
4169   // mark as suspended and send signal
4170   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4171     // failed to switch, state wasn't running?
4172     ShouldNotReachHere();
4173     return false;
4174   }
4175 
4176   if (sr_notify(osthread) != 0) {
4177     ShouldNotReachHere();
4178   }
4179 
4180   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4181   while (true) {
4182     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4183       break;
4184     } else {
4185       // timeout
4186       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4187       if (cancelled == os::SuspendResume::SR_RUNNING) {
4188         return false;
4189       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4190         // make sure that we consume the signal on the semaphore as well
4191         sr_semaphore.wait();
4192         break;
4193       } else {
4194         ShouldNotReachHere();
4195         return false;
4196       }
4197     }
4198   }
4199 
4200   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4201   return true;
4202 }
4203 
4204 static void do_resume(OSThread* osthread) {
4205   assert(osthread->sr.is_suspended(), "thread should be suspended");
4206   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4207 
4208   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4209     // failed to switch to WAKEUP_REQUEST
4210     ShouldNotReachHere();
4211     return;
4212   }
4213 
4214   while (true) {
4215     if (sr_notify(osthread) == 0) {
4216       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4217         if (osthread->sr.is_running()) {
4218           return;
4219         }
4220       }
4221     } else {
4222       ShouldNotReachHere();
4223     }
4224   }
4225 
4226   guarantee(osthread->sr.is_running(), "Must be running!");
4227 }
4228 
4229 ////////////////////////////////////////////////////////////////////////////////
4230 // interrupt support
4231 
4232 void os::interrupt(Thread* thread) {
4233   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4234     "possibility of dangling Thread pointer");
4235 
4236   OSThread* osthread = thread->osthread();
4237 
4238   if (!osthread->interrupted()) {
4239     osthread->set_interrupted(true);
4240     // More than one thread can get here with the same value of osthread,
4241     // resulting in multiple notifications.  We do, however, want the store
4242     // to interrupted() to be visible to other threads before we execute unpark().
4243     OrderAccess::fence();
4244     ParkEvent * const slp = thread->_SleepEvent ;
4245     if (slp != NULL) slp->unpark() ;
4246   }
4247 
4248   // For JSR166. Unpark even if interrupt status already was set
4249   if (thread->is_Java_thread())
4250     ((JavaThread*)thread)->parker()->unpark();
4251 
4252   ParkEvent * ev = thread->_ParkEvent ;
4253   if (ev != NULL) ev->unpark() ;
4254 
4255 }
4256 
4257 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4258   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4259     "possibility of dangling Thread pointer");
4260 
4261   OSThread* osthread = thread->osthread();
4262 
4263   bool interrupted = osthread->interrupted();
4264 
4265   if (interrupted && clear_interrupted) {
4266     osthread->set_interrupted(false);
4267     // consider thread->_SleepEvent->reset() ... optional optimization
4268   }
4269 
4270   return interrupted;
4271 }
4272 
4273 ///////////////////////////////////////////////////////////////////////////////////
4274 // signal handling (except suspend/resume)
4275 
4276 // This routine may be used by user applications as a "hook" to catch signals.
4277 // The user-defined signal handler must pass unrecognized signals to this
4278 // routine, and if it returns true (non-zero), then the signal handler must
4279 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4280 // routine will never retun false (zero), but instead will execute a VM panic
4281 // routine kill the process.
4282 //
4283 // If this routine returns false, it is OK to call it again.  This allows
4284 // the user-defined signal handler to perform checks either before or after
4285 // the VM performs its own checks.  Naturally, the user code would be making
4286 // a serious error if it tried to handle an exception (such as a null check
4287 // or breakpoint) that the VM was generating for its own correct operation.
4288 //
4289 // This routine may recognize any of the following kinds of signals:
4290 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4291 // It should be consulted by handlers for any of those signals.
4292 //
4293 // The caller of this routine must pass in the three arguments supplied
4294 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4295 // field of the structure passed to sigaction().  This routine assumes that
4296 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4297 //
4298 // Note that the VM will print warnings if it detects conflicting signal
4299 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4300 //
4301 extern "C" JNIEXPORT int
4302 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4303                         void* ucontext, int abort_if_unrecognized);
4304 
4305 void signalHandler(int sig, siginfo_t* info, void* uc) {
4306   assert(info != NULL && uc != NULL, "it must be old kernel");
4307   int orig_errno = errno;  // Preserve errno value over signal handler.
4308   JVM_handle_linux_signal(sig, info, uc, true);
4309   errno = orig_errno;
4310 }
4311 
4312 
4313 // This boolean allows users to forward their own non-matching signals
4314 // to JVM_handle_linux_signal, harmlessly.
4315 bool os::Linux::signal_handlers_are_installed = false;
4316 
4317 // For signal-chaining
4318 struct sigaction os::Linux::sigact[MAXSIGNUM];
4319 unsigned int os::Linux::sigs = 0;
4320 bool os::Linux::libjsig_is_loaded = false;
4321 typedef struct sigaction *(*get_signal_t)(int);
4322 get_signal_t os::Linux::get_signal_action = NULL;
4323 
4324 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4325   struct sigaction *actp = NULL;
4326 
4327   if (libjsig_is_loaded) {
4328     // Retrieve the old signal handler from libjsig
4329     actp = (*get_signal_action)(sig);
4330   }
4331   if (actp == NULL) {
4332     // Retrieve the preinstalled signal handler from jvm
4333     actp = get_preinstalled_handler(sig);
4334   }
4335 
4336   return actp;
4337 }
4338 
4339 static bool call_chained_handler(struct sigaction *actp, int sig,
4340                                  siginfo_t *siginfo, void *context) {
4341   // Call the old signal handler
4342   if (actp->sa_handler == SIG_DFL) {
4343     // It's more reasonable to let jvm treat it as an unexpected exception
4344     // instead of taking the default action.
4345     return false;
4346   } else if (actp->sa_handler != SIG_IGN) {
4347     if ((actp->sa_flags & SA_NODEFER) == 0) {
4348       // automaticlly block the signal
4349       sigaddset(&(actp->sa_mask), sig);
4350     }
4351 
4352     sa_handler_t hand;
4353     sa_sigaction_t sa;
4354     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4355     // retrieve the chained handler
4356     if (siginfo_flag_set) {
4357       sa = actp->sa_sigaction;
4358     } else {
4359       hand = actp->sa_handler;
4360     }
4361 
4362     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4363       actp->sa_handler = SIG_DFL;
4364     }
4365 
4366     // try to honor the signal mask
4367     sigset_t oset;
4368     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4369 
4370     // call into the chained handler
4371     if (siginfo_flag_set) {
4372       (*sa)(sig, siginfo, context);
4373     } else {
4374       (*hand)(sig);
4375     }
4376 
4377     // restore the signal mask
4378     pthread_sigmask(SIG_SETMASK, &oset, 0);
4379   }
4380   // Tell jvm's signal handler the signal is taken care of.
4381   return true;
4382 }
4383 
4384 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4385   bool chained = false;
4386   // signal-chaining
4387   if (UseSignalChaining) {
4388     struct sigaction *actp = get_chained_signal_action(sig);
4389     if (actp != NULL) {
4390       chained = call_chained_handler(actp, sig, siginfo, context);
4391     }
4392   }
4393   return chained;
4394 }
4395 
4396 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4397   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4398     return &sigact[sig];
4399   }
4400   return NULL;
4401 }
4402 
4403 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4404   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4405   sigact[sig] = oldAct;
4406   sigs |= (unsigned int)1 << sig;
4407 }
4408 
4409 // for diagnostic
4410 int os::Linux::sigflags[MAXSIGNUM];
4411 
4412 int os::Linux::get_our_sigflags(int sig) {
4413   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4414   return sigflags[sig];
4415 }
4416 
4417 void os::Linux::set_our_sigflags(int sig, int flags) {
4418   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4419   sigflags[sig] = flags;
4420 }
4421 
4422 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4423   // Check for overwrite.
4424   struct sigaction oldAct;
4425   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4426 
4427   void* oldhand = oldAct.sa_sigaction
4428                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4429                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4430   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4431       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4432       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4433     if (AllowUserSignalHandlers || !set_installed) {
4434       // Do not overwrite; user takes responsibility to forward to us.
4435       return;
4436     } else if (UseSignalChaining) {
4437       // save the old handler in jvm
4438       save_preinstalled_handler(sig, oldAct);
4439       // libjsig also interposes the sigaction() call below and saves the
4440       // old sigaction on it own.
4441     } else {
4442       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4443                     "%#lx for signal %d.", (long)oldhand, sig));
4444     }
4445   }
4446 
4447   struct sigaction sigAct;
4448   sigfillset(&(sigAct.sa_mask));
4449   sigAct.sa_handler = SIG_DFL;
4450   if (!set_installed) {
4451     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4452   } else {
4453     sigAct.sa_sigaction = signalHandler;
4454     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4455   }
4456   // Save flags, which are set by ours
4457   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4458   sigflags[sig] = sigAct.sa_flags;
4459 
4460   int ret = sigaction(sig, &sigAct, &oldAct);
4461   assert(ret == 0, "check");
4462 
4463   void* oldhand2  = oldAct.sa_sigaction
4464                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4465                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4466   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4467 }
4468 
4469 // install signal handlers for signals that HotSpot needs to
4470 // handle in order to support Java-level exception handling.
4471 
4472 void os::Linux::install_signal_handlers() {
4473   if (!signal_handlers_are_installed) {
4474     signal_handlers_are_installed = true;
4475 
4476     // signal-chaining
4477     typedef void (*signal_setting_t)();
4478     signal_setting_t begin_signal_setting = NULL;
4479     signal_setting_t end_signal_setting = NULL;
4480     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4481                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4482     if (begin_signal_setting != NULL) {
4483       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4484                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4485       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4486                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4487       libjsig_is_loaded = true;
4488       assert(UseSignalChaining, "should enable signal-chaining");
4489     }
4490     if (libjsig_is_loaded) {
4491       // Tell libjsig jvm is setting signal handlers
4492       (*begin_signal_setting)();
4493     }
4494 
4495     set_signal_handler(SIGSEGV, true);
4496     set_signal_handler(SIGPIPE, true);
4497     set_signal_handler(SIGBUS, true);
4498     set_signal_handler(SIGILL, true);
4499     set_signal_handler(SIGFPE, true);
4500     set_signal_handler(SIGXFSZ, true);
4501 
4502     if (libjsig_is_loaded) {
4503       // Tell libjsig jvm finishes setting signal handlers
4504       (*end_signal_setting)();
4505     }
4506 
4507     // We don't activate signal checker if libjsig is in place, we trust ourselves
4508     // and if UserSignalHandler is installed all bets are off.
4509     // Log that signal checking is off only if -verbose:jni is specified.
4510     if (CheckJNICalls) {
4511       if (libjsig_is_loaded) {
4512         if (PrintJNIResolving) {
4513           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4514         }
4515         check_signals = false;
4516       }
4517       if (AllowUserSignalHandlers) {
4518         if (PrintJNIResolving) {
4519           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4520         }
4521         check_signals = false;
4522       }
4523     }
4524   }
4525 }
4526 
4527 // This is the fastest way to get thread cpu time on Linux.
4528 // Returns cpu time (user+sys) for any thread, not only for current.
4529 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4530 // It might work on 2.6.10+ with a special kernel/glibc patch.
4531 // For reference, please, see IEEE Std 1003.1-2004:
4532 //   http://www.unix.org/single_unix_specification
4533 
4534 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4535   struct timespec tp;
4536   int rc = os::Linux::clock_gettime(clockid, &tp);
4537   assert(rc == 0, "clock_gettime is expected to return 0 code");
4538 
4539   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4540 }
4541 
4542 /////
4543 // glibc on Linux platform uses non-documented flag
4544 // to indicate, that some special sort of signal
4545 // trampoline is used.
4546 // We will never set this flag, and we should
4547 // ignore this flag in our diagnostic
4548 #ifdef SIGNIFICANT_SIGNAL_MASK
4549 #undef SIGNIFICANT_SIGNAL_MASK
4550 #endif
4551 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4552 
4553 static const char* get_signal_handler_name(address handler,
4554                                            char* buf, int buflen) {
4555   int offset;
4556   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4557   if (found) {
4558     // skip directory names
4559     const char *p1, *p2;
4560     p1 = buf;
4561     size_t len = strlen(os::file_separator());
4562     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4563     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4564   } else {
4565     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4566   }
4567   return buf;
4568 }
4569 
4570 static void print_signal_handler(outputStream* st, int sig,
4571                                  char* buf, size_t buflen) {
4572   struct sigaction sa;
4573 
4574   sigaction(sig, NULL, &sa);
4575 
4576   // See comment for SIGNIFICANT_SIGNAL_MASK define
4577   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4578 
4579   st->print("%s: ", os::exception_name(sig, buf, buflen));
4580 
4581   address handler = (sa.sa_flags & SA_SIGINFO)
4582     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4583     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4584 
4585   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4586     st->print("SIG_DFL");
4587   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4588     st->print("SIG_IGN");
4589   } else {
4590     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4591   }
4592 
4593   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4594 
4595   address rh = VMError::get_resetted_sighandler(sig);
4596   // May be, handler was resetted by VMError?
4597   if(rh != NULL) {
4598     handler = rh;
4599     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4600   }
4601 
4602   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
4603 
4604   // Check: is it our handler?
4605   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4606      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4607     // It is our signal handler
4608     // check for flags, reset system-used one!
4609     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4610       st->print(
4611                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4612                 os::Linux::get_our_sigflags(sig));
4613     }
4614   }
4615   st->cr();
4616 }
4617 
4618 
4619 #define DO_SIGNAL_CHECK(sig) \
4620   if (!sigismember(&check_signal_done, sig)) \
4621     os::Linux::check_signal_handler(sig)
4622 
4623 // This method is a periodic task to check for misbehaving JNI applications
4624 // under CheckJNI, we can add any periodic checks here
4625 
4626 void os::run_periodic_checks() {
4627 
4628   if (check_signals == false) return;
4629 
4630   // SEGV and BUS if overridden could potentially prevent
4631   // generation of hs*.log in the event of a crash, debugging
4632   // such a case can be very challenging, so we absolutely
4633   // check the following for a good measure:
4634   DO_SIGNAL_CHECK(SIGSEGV);
4635   DO_SIGNAL_CHECK(SIGILL);
4636   DO_SIGNAL_CHECK(SIGFPE);
4637   DO_SIGNAL_CHECK(SIGBUS);
4638   DO_SIGNAL_CHECK(SIGPIPE);
4639   DO_SIGNAL_CHECK(SIGXFSZ);
4640 
4641 
4642   // ReduceSignalUsage allows the user to override these handlers
4643   // see comments at the very top and jvm_solaris.h
4644   if (!ReduceSignalUsage) {
4645     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4646     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4647     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4648     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4649   }
4650 
4651   DO_SIGNAL_CHECK(SR_signum);
4652   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4653 }
4654 
4655 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4656 
4657 static os_sigaction_t os_sigaction = NULL;
4658 
4659 void os::Linux::check_signal_handler(int sig) {
4660   char buf[O_BUFLEN];
4661   address jvmHandler = NULL;
4662 
4663 
4664   struct sigaction act;
4665   if (os_sigaction == NULL) {
4666     // only trust the default sigaction, in case it has been interposed
4667     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4668     if (os_sigaction == NULL) return;
4669   }
4670 
4671   os_sigaction(sig, (struct sigaction*)NULL, &act);
4672 
4673 
4674   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4675 
4676   address thisHandler = (act.sa_flags & SA_SIGINFO)
4677     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4678     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4679 
4680 
4681   switch(sig) {
4682   case SIGSEGV:
4683   case SIGBUS:
4684   case SIGFPE:
4685   case SIGPIPE:
4686   case SIGILL:
4687   case SIGXFSZ:
4688     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4689     break;
4690 
4691   case SHUTDOWN1_SIGNAL:
4692   case SHUTDOWN2_SIGNAL:
4693   case SHUTDOWN3_SIGNAL:
4694   case BREAK_SIGNAL:
4695     jvmHandler = (address)user_handler();
4696     break;
4697 
4698   case INTERRUPT_SIGNAL:
4699     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4700     break;
4701 
4702   default:
4703     if (sig == SR_signum) {
4704       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4705     } else {
4706       return;
4707     }
4708     break;
4709   }
4710 
4711   if (thisHandler != jvmHandler) {
4712     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4713     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4714     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4715     // No need to check this sig any longer
4716     sigaddset(&check_signal_done, sig);
4717   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4718     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4719     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4720     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4721     // No need to check this sig any longer
4722     sigaddset(&check_signal_done, sig);
4723   }
4724 
4725   // Dump all the signal
4726   if (sigismember(&check_signal_done, sig)) {
4727     print_signal_handlers(tty, buf, O_BUFLEN);
4728   }
4729 }
4730 
4731 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4732 
4733 extern bool signal_name(int signo, char* buf, size_t len);
4734 
4735 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4736   if (0 < exception_code && exception_code <= SIGRTMAX) {
4737     // signal
4738     if (!signal_name(exception_code, buf, size)) {
4739       jio_snprintf(buf, size, "SIG%d", exception_code);
4740     }
4741     return buf;
4742   } else {
4743     return NULL;
4744   }
4745 }
4746 
4747 // this is called _before_ the most of global arguments have been parsed
4748 void os::init(void) {
4749   char dummy;   /* used to get a guess on initial stack address */
4750 //  first_hrtime = gethrtime();
4751 
4752   // With LinuxThreads the JavaMain thread pid (primordial thread)
4753   // is different than the pid of the java launcher thread.
4754   // So, on Linux, the launcher thread pid is passed to the VM
4755   // via the sun.java.launcher.pid property.
4756   // Use this property instead of getpid() if it was correctly passed.
4757   // See bug 6351349.
4758   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4759 
4760   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4761 
4762   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4763 
4764   init_random(1234567);
4765 
4766   ThreadCritical::initialize();
4767 
4768   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4769   if (Linux::page_size() == -1) {
4770     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4771                   strerror(errno)));
4772   }
4773   init_page_sizes((size_t) Linux::page_size());
4774 
4775   Linux::initialize_system_info();
4776 
4777   // main_thread points to the aboriginal thread
4778   Linux::_main_thread = pthread_self();
4779 
4780   Linux::clock_init();
4781   initial_time_count = javaTimeNanos();
4782 
4783   // pthread_condattr initialization for monotonic clock
4784   int status;
4785   pthread_condattr_t* _condattr = os::Linux::condAttr();
4786   if ((status = pthread_condattr_init(_condattr)) != 0) {
4787     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4788   }
4789   // Only set the clock if CLOCK_MONOTONIC is available
4790   if (Linux::supports_monotonic_clock()) {
4791     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4792       if (status == EINVAL) {
4793         warning("Unable to use monotonic clock with relative timed-waits" \
4794                 " - changes to the time-of-day clock may have adverse affects");
4795       } else {
4796         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4797       }
4798     }
4799   }
4800   // else it defaults to CLOCK_REALTIME
4801 
4802   pthread_mutex_init(&dl_mutex, NULL);
4803 
4804   // If the pagesize of the VM is greater than 8K determine the appropriate
4805   // number of initial guard pages.  The user can change this with the
4806   // command line arguments, if needed.
4807   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4808     StackYellowPages = 1;
4809     StackRedPages = 1;
4810     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4811   }
4812 }
4813 
4814 // To install functions for atexit system call
4815 extern "C" {
4816   static void perfMemory_exit_helper() {
4817     perfMemory_exit();
4818   }
4819 }
4820 
4821 // this is called _after_ the global arguments have been parsed
4822 jint os::init_2(void)
4823 {
4824   Linux::fast_thread_clock_init();
4825 
4826   // Allocate a single page and mark it as readable for safepoint polling
4827   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4828   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4829 
4830   os::set_polling_page( polling_page );
4831 
4832 #ifndef PRODUCT
4833   if(Verbose && PrintMiscellaneous)
4834     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4835 #endif
4836 
4837   if (!UseMembar) {
4838     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4839     guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4840     os::set_memory_serialize_page( mem_serialize_page );
4841 
4842 #ifndef PRODUCT
4843     if(Verbose && PrintMiscellaneous)
4844       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4845 #endif
4846   }
4847 
4848   // initialize suspend/resume support - must do this before signal_sets_init()
4849   if (SR_initialize() != 0) {
4850     perror("SR_initialize failed");
4851     return JNI_ERR;
4852   }
4853 
4854   Linux::signal_sets_init();
4855   Linux::install_signal_handlers();
4856 
4857   // Check minimum allowable stack size for thread creation and to initialize
4858   // the java system classes, including StackOverflowError - depends on page
4859   // size.  Add a page for compiler2 recursion in main thread.
4860   // Add in 2*BytesPerWord times page size to account for VM stack during
4861   // class initialization depending on 32 or 64 bit VM.
4862   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4863             (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4864                     (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4865 
4866   size_t threadStackSizeInBytes = ThreadStackSize * K;
4867   if (threadStackSizeInBytes != 0 &&
4868       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4869         tty->print_cr("\nThe stack size specified is too small, "
4870                       "Specify at least %dk",
4871                       os::Linux::min_stack_allowed/ K);
4872         return JNI_ERR;
4873   }
4874 
4875   // Make the stack size a multiple of the page size so that
4876   // the yellow/red zones can be guarded.
4877   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4878         vm_page_size()));
4879 
4880   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4881 
4882 #if defined(IA32)
4883   workaround_expand_exec_shield_cs_limit();
4884 #endif
4885 
4886   Linux::libpthread_init();
4887   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4888      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4889           Linux::glibc_version(), Linux::libpthread_version(),
4890           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4891   }
4892 
4893   if (UseNUMA) {
4894     if (!Linux::libnuma_init()) {
4895       UseNUMA = false;
4896     } else {
4897       if ((Linux::numa_max_node() < 1)) {
4898         // There's only one node(they start from 0), disable NUMA.
4899         UseNUMA = false;
4900       }
4901     }
4902     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4903     // we can make the adaptive lgrp chunk resizing work. If the user specified
4904     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4905     // disable adaptive resizing.
4906     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4907       if (FLAG_IS_DEFAULT(UseNUMA)) {
4908         UseNUMA = false;
4909       } else {
4910         if (FLAG_IS_DEFAULT(UseLargePages) &&
4911             FLAG_IS_DEFAULT(UseSHM) &&
4912             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4913           UseLargePages = false;
4914         } else {
4915           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4916           UseAdaptiveSizePolicy = false;
4917           UseAdaptiveNUMAChunkSizing = false;
4918         }
4919       }
4920     }
4921     if (!UseNUMA && ForceNUMA) {
4922       UseNUMA = true;
4923     }
4924   }
4925 
4926   if (MaxFDLimit) {
4927     // set the number of file descriptors to max. print out error
4928     // if getrlimit/setrlimit fails but continue regardless.
4929     struct rlimit nbr_files;
4930     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4931     if (status != 0) {
4932       if (PrintMiscellaneous && (Verbose || WizardMode))
4933         perror("os::init_2 getrlimit failed");
4934     } else {
4935       nbr_files.rlim_cur = nbr_files.rlim_max;
4936       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4937       if (status != 0) {
4938         if (PrintMiscellaneous && (Verbose || WizardMode))
4939           perror("os::init_2 setrlimit failed");
4940       }
4941     }
4942   }
4943 
4944   // Initialize lock used to serialize thread creation (see os::create_thread)
4945   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4946 
4947   // at-exit methods are called in the reverse order of their registration.
4948   // atexit functions are called on return from main or as a result of a
4949   // call to exit(3C). There can be only 32 of these functions registered
4950   // and atexit() does not set errno.
4951 
4952   if (PerfAllowAtExitRegistration) {
4953     // only register atexit functions if PerfAllowAtExitRegistration is set.
4954     // atexit functions can be delayed until process exit time, which
4955     // can be problematic for embedded VM situations. Embedded VMs should
4956     // call DestroyJavaVM() to assure that VM resources are released.
4957 
4958     // note: perfMemory_exit_helper atexit function may be removed in
4959     // the future if the appropriate cleanup code can be added to the
4960     // VM_Exit VMOperation's doit method.
4961     if (atexit(perfMemory_exit_helper) != 0) {
4962       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4963     }
4964   }
4965 
4966   // initialize thread priority policy
4967   prio_init();
4968 
4969   return JNI_OK;
4970 }
4971 
4972 // this is called at the end of vm_initialization
4973 void os::init_3(void)
4974 {
4975 #ifdef JAVASE_EMBEDDED
4976   // Start the MemNotifyThread
4977   if (LowMemoryProtection) {
4978     MemNotifyThread::start();
4979   }
4980   return;
4981 #endif
4982 }
4983 
4984 // Mark the polling page as unreadable
4985 void os::make_polling_page_unreadable(void) {
4986   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4987     fatal("Could not disable polling page");
4988 };
4989 
4990 // Mark the polling page as readable
4991 void os::make_polling_page_readable(void) {
4992   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4993     fatal("Could not enable polling page");
4994   }
4995 };
4996 
4997 int os::active_processor_count() {
4998   // Linux doesn't yet have a (official) notion of processor sets,
4999   // so just return the number of online processors.
5000   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5001   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
5002   return online_cpus;
5003 }
5004 
5005 void os::set_native_thread_name(const char *name) {
5006   // Not yet implemented.
5007   return;
5008 }
5009 
5010 bool os::distribute_processes(uint length, uint* distribution) {
5011   // Not yet implemented.
5012   return false;
5013 }
5014 
5015 bool os::bind_to_processor(uint processor_id) {
5016   // Not yet implemented.
5017   return false;
5018 }
5019 
5020 ///
5021 
5022 void os::SuspendedThreadTask::internal_do_task() {
5023   if (do_suspend(_thread->osthread())) {
5024     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5025     do_task(context);
5026     do_resume(_thread->osthread());
5027   }
5028 }
5029 
5030 class PcFetcher : public os::SuspendedThreadTask {
5031 public:
5032   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5033   ExtendedPC result();
5034 protected:
5035   void do_task(const os::SuspendedThreadTaskContext& context);
5036 private:
5037   ExtendedPC _epc;
5038 };
5039 
5040 ExtendedPC PcFetcher::result() {
5041   guarantee(is_done(), "task is not done yet.");
5042   return _epc;
5043 }
5044 
5045 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5046   Thread* thread = context.thread();
5047   OSThread* osthread = thread->osthread();
5048   if (osthread->ucontext() != NULL) {
5049     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5050   } else {
5051     // NULL context is unexpected, double-check this is the VMThread
5052     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5053   }
5054 }
5055 
5056 // Suspends the target using the signal mechanism and then grabs the PC before
5057 // resuming the target. Used by the flat-profiler only
5058 ExtendedPC os::get_thread_pc(Thread* thread) {
5059   // Make sure that it is called by the watcher for the VMThread
5060   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5061   assert(thread->is_VM_thread(), "Can only be called for VMThread");
5062 
5063   PcFetcher fetcher(thread);
5064   fetcher.run();
5065   return fetcher.result();
5066 }
5067 
5068 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5069 {
5070    if (is_NPTL()) {
5071       return pthread_cond_timedwait(_cond, _mutex, _abstime);
5072    } else {
5073       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5074       // word back to default 64bit precision if condvar is signaled. Java
5075       // wants 53bit precision.  Save and restore current value.
5076       int fpu = get_fpu_control_word();
5077       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5078       set_fpu_control_word(fpu);
5079       return status;
5080    }
5081 }
5082 
5083 ////////////////////////////////////////////////////////////////////////////////
5084 // debug support
5085 
5086 bool os::find(address addr, outputStream* st) {
5087   Dl_info dlinfo;
5088   memset(&dlinfo, 0, sizeof(dlinfo));
5089   if (dladdr(addr, &dlinfo) != 0) {
5090     st->print(PTR_FORMAT ": ", addr);
5091     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5092       st->print("%s+%#x", dlinfo.dli_sname,
5093                  addr - (intptr_t)dlinfo.dli_saddr);
5094     } else if (dlinfo.dli_fbase != NULL) {
5095       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5096     } else {
5097       st->print("<absolute address>");
5098     }
5099     if (dlinfo.dli_fname != NULL) {
5100       st->print(" in %s", dlinfo.dli_fname);
5101     }
5102     if (dlinfo.dli_fbase != NULL) {
5103       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5104     }
5105     st->cr();
5106 
5107     if (Verbose) {
5108       // decode some bytes around the PC
5109       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5110       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5111       address       lowest = (address) dlinfo.dli_sname;
5112       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5113       if (begin < lowest)  begin = lowest;
5114       Dl_info dlinfo2;
5115       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5116           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5117         end = (address) dlinfo2.dli_saddr;
5118       Disassembler::decode(begin, end, st);
5119     }
5120     return true;
5121   }
5122   return false;
5123 }
5124 
5125 ////////////////////////////////////////////////////////////////////////////////
5126 // misc
5127 
5128 // This does not do anything on Linux. This is basically a hook for being
5129 // able to use structured exception handling (thread-local exception filters)
5130 // on, e.g., Win32.
5131 void
5132 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5133                          JavaCallArguments* args, Thread* thread) {
5134   f(value, method, args, thread);
5135 }
5136 
5137 void os::print_statistics() {
5138 }
5139 
5140 int os::message_box(const char* title, const char* message) {
5141   int i;
5142   fdStream err(defaultStream::error_fd());
5143   for (i = 0; i < 78; i++) err.print_raw("=");
5144   err.cr();
5145   err.print_raw_cr(title);
5146   for (i = 0; i < 78; i++) err.print_raw("-");
5147   err.cr();
5148   err.print_raw_cr(message);
5149   for (i = 0; i < 78; i++) err.print_raw("=");
5150   err.cr();
5151 
5152   char buf[16];
5153   // Prevent process from exiting upon "read error" without consuming all CPU
5154   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5155 
5156   return buf[0] == 'y' || buf[0] == 'Y';
5157 }
5158 
5159 int os::stat(const char *path, struct stat *sbuf) {
5160   char pathbuf[MAX_PATH];
5161   if (strlen(path) > MAX_PATH - 1) {
5162     errno = ENAMETOOLONG;
5163     return -1;
5164   }
5165   os::native_path(strcpy(pathbuf, path));
5166   return ::stat(pathbuf, sbuf);
5167 }
5168 
5169 bool os::check_heap(bool force) {
5170   return true;
5171 }
5172 
5173 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5174   return ::vsnprintf(buf, count, format, args);
5175 }
5176 
5177 // Is a (classpath) directory empty?
5178 bool os::dir_is_empty(const char* path) {
5179   DIR *dir = NULL;
5180   struct dirent *ptr;
5181 
5182   dir = opendir(path);
5183   if (dir == NULL) return true;
5184 
5185   /* Scan the directory */
5186   bool result = true;
5187   char buf[sizeof(struct dirent) + MAX_PATH];
5188   while (result && (ptr = ::readdir(dir)) != NULL) {
5189     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5190       result = false;
5191     }
5192   }
5193   closedir(dir);
5194   return result;
5195 }
5196 
5197 // This code originates from JDK's sysOpen and open64_w
5198 // from src/solaris/hpi/src/system_md.c
5199 
5200 #ifndef O_DELETE
5201 #define O_DELETE 0x10000
5202 #endif
5203 
5204 // Open a file. Unlink the file immediately after open returns
5205 // if the specified oflag has the O_DELETE flag set.
5206 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5207 
5208 int os::open(const char *path, int oflag, int mode) {
5209 
5210   if (strlen(path) > MAX_PATH - 1) {
5211     errno = ENAMETOOLONG;
5212     return -1;
5213   }
5214   int fd;
5215   int o_delete = (oflag & O_DELETE);
5216   oflag = oflag & ~O_DELETE;
5217 
5218   fd = ::open64(path, oflag, mode);
5219   if (fd == -1) return -1;
5220 
5221   //If the open succeeded, the file might still be a directory
5222   {
5223     struct stat64 buf64;
5224     int ret = ::fstat64(fd, &buf64);
5225     int st_mode = buf64.st_mode;
5226 
5227     if (ret != -1) {
5228       if ((st_mode & S_IFMT) == S_IFDIR) {
5229         errno = EISDIR;
5230         ::close(fd);
5231         return -1;
5232       }
5233     } else {
5234       ::close(fd);
5235       return -1;
5236     }
5237   }
5238 
5239     /*
5240      * All file descriptors that are opened in the JVM and not
5241      * specifically destined for a subprocess should have the
5242      * close-on-exec flag set.  If we don't set it, then careless 3rd
5243      * party native code might fork and exec without closing all
5244      * appropriate file descriptors (e.g. as we do in closeDescriptors in
5245      * UNIXProcess.c), and this in turn might:
5246      *
5247      * - cause end-of-file to fail to be detected on some file
5248      *   descriptors, resulting in mysterious hangs, or
5249      *
5250      * - might cause an fopen in the subprocess to fail on a system
5251      *   suffering from bug 1085341.
5252      *
5253      * (Yes, the default setting of the close-on-exec flag is a Unix
5254      * design flaw)
5255      *
5256      * See:
5257      * 1085341: 32-bit stdio routines should support file descriptors >255
5258      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5259      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5260      */
5261 #ifdef FD_CLOEXEC
5262     {
5263         int flags = ::fcntl(fd, F_GETFD);
5264         if (flags != -1)
5265             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5266     }
5267 #endif
5268 
5269   if (o_delete != 0) {
5270     ::unlink(path);
5271   }
5272   return fd;
5273 }
5274 
5275 
5276 // create binary file, rewriting existing file if required
5277 int os::create_binary_file(const char* path, bool rewrite_existing) {
5278   int oflags = O_WRONLY | O_CREAT;
5279   if (!rewrite_existing) {
5280     oflags |= O_EXCL;
5281   }
5282   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5283 }
5284 
5285 // return current position of file pointer
5286 jlong os::current_file_offset(int fd) {
5287   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5288 }
5289 
5290 // move file pointer to the specified offset
5291 jlong os::seek_to_file_offset(int fd, jlong offset) {
5292   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5293 }
5294 
5295 // This code originates from JDK's sysAvailable
5296 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5297 
5298 int os::available(int fd, jlong *bytes) {
5299   jlong cur, end;
5300   int mode;
5301   struct stat64 buf64;
5302 
5303   if (::fstat64(fd, &buf64) >= 0) {
5304     mode = buf64.st_mode;
5305     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5306       /*
5307       * XXX: is the following call interruptible? If so, this might
5308       * need to go through the INTERRUPT_IO() wrapper as for other
5309       * blocking, interruptible calls in this file.
5310       */
5311       int n;
5312       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5313         *bytes = n;
5314         return 1;
5315       }
5316     }
5317   }
5318   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5319     return 0;
5320   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5321     return 0;
5322   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5323     return 0;
5324   }
5325   *bytes = end - cur;
5326   return 1;
5327 }
5328 
5329 int os::socket_available(int fd, jint *pbytes) {
5330   // Linux doc says EINTR not returned, unlike Solaris
5331   int ret = ::ioctl(fd, FIONREAD, pbytes);
5332 
5333   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5334   // is expected to return 0 on failure and 1 on success to the jdk.
5335   return (ret < 0) ? 0 : 1;
5336 }
5337 
5338 // Map a block of memory.
5339 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5340                      char *addr, size_t bytes, bool read_only,
5341                      bool allow_exec) {
5342   int prot;
5343   int flags = MAP_PRIVATE;
5344 
5345   if (read_only) {
5346     prot = PROT_READ;
5347   } else {
5348     prot = PROT_READ | PROT_WRITE;
5349   }
5350 
5351   if (allow_exec) {
5352     prot |= PROT_EXEC;
5353   }
5354 
5355   if (addr != NULL) {
5356     flags |= MAP_FIXED;
5357   }
5358 
5359   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5360                                      fd, file_offset);
5361   if (mapped_address == MAP_FAILED) {
5362     return NULL;
5363   }
5364   return mapped_address;
5365 }
5366 
5367 
5368 // Remap a block of memory.
5369 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5370                        char *addr, size_t bytes, bool read_only,
5371                        bool allow_exec) {
5372   // same as map_memory() on this OS
5373   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5374                         allow_exec);
5375 }
5376 
5377 
5378 // Unmap a block of memory.
5379 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5380   return munmap(addr, bytes) == 0;
5381 }
5382 
5383 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5384 
5385 static clockid_t thread_cpu_clockid(Thread* thread) {
5386   pthread_t tid = thread->osthread()->pthread_id();
5387   clockid_t clockid;
5388 
5389   // Get thread clockid
5390   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5391   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5392   return clockid;
5393 }
5394 
5395 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5396 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5397 // of a thread.
5398 //
5399 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5400 // the fast estimate available on the platform.
5401 
5402 jlong os::current_thread_cpu_time() {
5403   if (os::Linux::supports_fast_thread_cpu_time()) {
5404     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5405   } else {
5406     // return user + sys since the cost is the same
5407     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5408   }
5409 }
5410 
5411 jlong os::thread_cpu_time(Thread* thread) {
5412   // consistent with what current_thread_cpu_time() returns
5413   if (os::Linux::supports_fast_thread_cpu_time()) {
5414     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5415   } else {
5416     return slow_thread_cpu_time(thread, true /* user + sys */);
5417   }
5418 }
5419 
5420 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5421   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5422     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5423   } else {
5424     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5425   }
5426 }
5427 
5428 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5429   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5430     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5431   } else {
5432     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5433   }
5434 }
5435 
5436 //
5437 //  -1 on error.
5438 //
5439 
5440 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5441   static bool proc_task_unchecked = true;
5442   static const char *proc_stat_path = "/proc/%d/stat";
5443   pid_t  tid = thread->osthread()->thread_id();
5444   char *s;
5445   char stat[2048];
5446   int statlen;
5447   char proc_name[64];
5448   int count;
5449   long sys_time, user_time;
5450   char cdummy;
5451   int idummy;
5452   long ldummy;
5453   FILE *fp;
5454 
5455   // The /proc/<tid>/stat aggregates per-process usage on
5456   // new Linux kernels 2.6+ where NPTL is supported.
5457   // The /proc/self/task/<tid>/stat still has the per-thread usage.
5458   // See bug 6328462.
5459   // There possibly can be cases where there is no directory
5460   // /proc/self/task, so we check its availability.
5461   if (proc_task_unchecked && os::Linux::is_NPTL()) {
5462     // This is executed only once
5463     proc_task_unchecked = false;
5464     fp = fopen("/proc/self/task", "r");
5465     if (fp != NULL) {
5466       proc_stat_path = "/proc/self/task/%d/stat";
5467       fclose(fp);
5468     }
5469   }
5470 
5471   sprintf(proc_name, proc_stat_path, tid);
5472   fp = fopen(proc_name, "r");
5473   if ( fp == NULL ) return -1;
5474   statlen = fread(stat, 1, 2047, fp);
5475   stat[statlen] = '\0';
5476   fclose(fp);
5477 
5478   // Skip pid and the command string. Note that we could be dealing with
5479   // weird command names, e.g. user could decide to rename java launcher
5480   // to "java 1.4.2 :)", then the stat file would look like
5481   //                1234 (java 1.4.2 :)) R ... ...
5482   // We don't really need to know the command string, just find the last
5483   // occurrence of ")" and then start parsing from there. See bug 4726580.
5484   s = strrchr(stat, ')');
5485   if (s == NULL ) return -1;
5486 
5487   // Skip blank chars
5488   do s++; while (isspace(*s));
5489 
5490   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5491                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5492                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5493                  &user_time, &sys_time);
5494   if ( count != 13 ) return -1;
5495   if (user_sys_cpu_time) {
5496     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5497   } else {
5498     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5499   }
5500 }
5501 
5502 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5503   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5504   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5505   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5506   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5507 }
5508 
5509 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5510   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5511   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5512   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5513   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5514 }
5515 
5516 bool os::is_thread_cpu_time_supported() {
5517   return true;
5518 }
5519 
5520 // System loadavg support.  Returns -1 if load average cannot be obtained.
5521 // Linux doesn't yet have a (official) notion of processor sets,
5522 // so just return the system wide load average.
5523 int os::loadavg(double loadavg[], int nelem) {
5524   return ::getloadavg(loadavg, nelem);
5525 }
5526 
5527 void os::pause() {
5528   char filename[MAX_PATH];
5529   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5530     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5531   } else {
5532     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5533   }
5534 
5535   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5536   if (fd != -1) {
5537     struct stat buf;
5538     ::close(fd);
5539     while (::stat(filename, &buf) == 0) {
5540       (void)::poll(NULL, 0, 100);
5541     }
5542   } else {
5543     jio_fprintf(stderr,
5544       "Could not open pause file '%s', continuing immediately.\n", filename);
5545   }
5546 }
5547 
5548 
5549 // Refer to the comments in os_solaris.cpp park-unpark.
5550 //
5551 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5552 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5553 // For specifics regarding the bug see GLIBC BUGID 261237 :
5554 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5555 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5556 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5557 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5558 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5559 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5560 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5561 // of libpthread avoids the problem, but isn't practical.
5562 //
5563 // Possible remedies:
5564 //
5565 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5566 //      This is palliative and probabilistic, however.  If the thread is preempted
5567 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5568 //      than the minimum period may have passed, and the abstime may be stale (in the
5569 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5570 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5571 //
5572 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5573 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5574 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5575 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5576 //      thread.
5577 //
5578 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5579 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5580 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5581 //      This also works well.  In fact it avoids kernel-level scalability impediments
5582 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5583 //      timers in a graceful fashion.
5584 //
5585 // 4.   When the abstime value is in the past it appears that control returns
5586 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5587 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5588 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5589 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5590 //      It may be possible to avoid reinitialization by checking the return
5591 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5592 //      condvar we must establish the invariant that cond_signal() is only called
5593 //      within critical sections protected by the adjunct mutex.  This prevents
5594 //      cond_signal() from "seeing" a condvar that's in the midst of being
5595 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5596 //      desirable signal-after-unlock optimization that avoids futile context switching.
5597 //
5598 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5599 //      structure when a condvar is used or initialized.  cond_destroy()  would
5600 //      release the helper structure.  Our reinitialize-after-timedwait fix
5601 //      put excessive stress on malloc/free and locks protecting the c-heap.
5602 //
5603 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5604 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5605 // and only enabling the work-around for vulnerable environments.
5606 
5607 // utility to compute the abstime argument to timedwait:
5608 // millis is the relative timeout time
5609 // abstime will be the absolute timeout time
5610 // TODO: replace compute_abstime() with unpackTime()
5611 
5612 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5613   if (millis < 0)  millis = 0;
5614 
5615   jlong seconds = millis / 1000;
5616   millis %= 1000;
5617   if (seconds > 50000000) { // see man cond_timedwait(3T)
5618     seconds = 50000000;
5619   }
5620 
5621   if (os::Linux::supports_monotonic_clock()) {
5622     struct timespec now;
5623     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5624     assert_status(status == 0, status, "clock_gettime");
5625     abstime->tv_sec = now.tv_sec  + seconds;
5626     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5627     if (nanos >= NANOSECS_PER_SEC) {
5628       abstime->tv_sec += 1;
5629       nanos -= NANOSECS_PER_SEC;
5630     }
5631     abstime->tv_nsec = nanos;
5632   } else {
5633     struct timeval now;
5634     int status = gettimeofday(&now, NULL);
5635     assert(status == 0, "gettimeofday");
5636     abstime->tv_sec = now.tv_sec  + seconds;
5637     long usec = now.tv_usec + millis * 1000;
5638     if (usec >= 1000000) {
5639       abstime->tv_sec += 1;
5640       usec -= 1000000;
5641     }
5642     abstime->tv_nsec = usec * 1000;
5643   }
5644   return abstime;
5645 }
5646 
5647 
5648 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5649 // Conceptually TryPark() should be equivalent to park(0).
5650 
5651 int os::PlatformEvent::TryPark() {
5652   for (;;) {
5653     const int v = _Event ;
5654     guarantee ((v == 0) || (v == 1), "invariant") ;
5655     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5656   }
5657 }
5658 
5659 void os::PlatformEvent::park() {       // AKA "down()"
5660   // Invariant: Only the thread associated with the Event/PlatformEvent
5661   // may call park().
5662   // TODO: assert that _Assoc != NULL or _Assoc == Self
5663   int v ;
5664   for (;;) {
5665       v = _Event ;
5666       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5667   }
5668   guarantee (v >= 0, "invariant") ;
5669   if (v == 0) {
5670      // Do this the hard way by blocking ...
5671      int status = pthread_mutex_lock(_mutex);
5672      assert_status(status == 0, status, "mutex_lock");
5673      guarantee (_nParked == 0, "invariant") ;
5674      ++ _nParked ;
5675      while (_Event < 0) {
5676         status = pthread_cond_wait(_cond, _mutex);
5677         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5678         // Treat this the same as if the wait was interrupted
5679         if (status == ETIME) { status = EINTR; }
5680         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5681      }
5682      -- _nParked ;
5683 
5684     _Event = 0 ;
5685      status = pthread_mutex_unlock(_mutex);
5686      assert_status(status == 0, status, "mutex_unlock");
5687     // Paranoia to ensure our locked and lock-free paths interact
5688     // correctly with each other.
5689     OrderAccess::fence();
5690   }
5691   guarantee (_Event >= 0, "invariant") ;
5692 }
5693 
5694 int os::PlatformEvent::park(jlong millis) {
5695   guarantee (_nParked == 0, "invariant") ;
5696 
5697   int v ;
5698   for (;;) {
5699       v = _Event ;
5700       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5701   }
5702   guarantee (v >= 0, "invariant") ;
5703   if (v != 0) return OS_OK ;
5704 
5705   // We do this the hard way, by blocking the thread.
5706   // Consider enforcing a minimum timeout value.
5707   struct timespec abst;
5708   compute_abstime(&abst, millis);
5709 
5710   int ret = OS_TIMEOUT;
5711   int status = pthread_mutex_lock(_mutex);
5712   assert_status(status == 0, status, "mutex_lock");
5713   guarantee (_nParked == 0, "invariant") ;
5714   ++_nParked ;
5715 
5716   // Object.wait(timo) will return because of
5717   // (a) notification
5718   // (b) timeout
5719   // (c) thread.interrupt
5720   //
5721   // Thread.interrupt and object.notify{All} both call Event::set.
5722   // That is, we treat thread.interrupt as a special case of notification.
5723   // The underlying Solaris implementation, cond_timedwait, admits
5724   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5725   // JVM from making those visible to Java code.  As such, we must
5726   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5727   //
5728   // TODO: properly differentiate simultaneous notify+interrupt.
5729   // In that case, we should propagate the notify to another waiter.
5730 
5731   while (_Event < 0) {
5732     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5733     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5734       pthread_cond_destroy (_cond);
5735       pthread_cond_init (_cond, os::Linux::condAttr()) ;
5736     }
5737     assert_status(status == 0 || status == EINTR ||
5738                   status == ETIME || status == ETIMEDOUT,
5739                   status, "cond_timedwait");
5740     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5741     if (status == ETIME || status == ETIMEDOUT) break ;
5742     // We consume and ignore EINTR and spurious wakeups.
5743   }
5744   --_nParked ;
5745   if (_Event >= 0) {
5746      ret = OS_OK;
5747   }
5748   _Event = 0 ;
5749   status = pthread_mutex_unlock(_mutex);
5750   assert_status(status == 0, status, "mutex_unlock");
5751   assert (_nParked == 0, "invariant") ;
5752   // Paranoia to ensure our locked and lock-free paths interact
5753   // correctly with each other.
5754   OrderAccess::fence();
5755   return ret;
5756 }
5757 
5758 void os::PlatformEvent::unpark() {
5759   // Transitions for _Event:
5760   //    0 :=> 1
5761   //    1 :=> 1
5762   //   -1 :=> either 0 or 1; must signal target thread
5763   //          That is, we can safely transition _Event from -1 to either
5764   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
5765   //          unpark() calls.
5766   // See also: "Semaphores in Plan 9" by Mullender & Cox
5767   //
5768   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5769   // that it will take two back-to-back park() calls for the owning
5770   // thread to block. This has the benefit of forcing a spurious return
5771   // from the first park() call after an unpark() call which will help
5772   // shake out uses of park() and unpark() without condition variables.
5773 
5774   if (Atomic::xchg(1, &_Event) >= 0) return;
5775 
5776   // Wait for the thread associated with the event to vacate
5777   int status = pthread_mutex_lock(_mutex);
5778   assert_status(status == 0, status, "mutex_lock");
5779   int AnyWaiters = _nParked;
5780   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5781   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5782     AnyWaiters = 0;
5783     pthread_cond_signal(_cond);
5784   }
5785   status = pthread_mutex_unlock(_mutex);
5786   assert_status(status == 0, status, "mutex_unlock");
5787   if (AnyWaiters != 0) {
5788     status = pthread_cond_signal(_cond);
5789     assert_status(status == 0, status, "cond_signal");
5790   }
5791 
5792   // Note that we signal() _after dropping the lock for "immortal" Events.
5793   // This is safe and avoids a common class of  futile wakeups.  In rare
5794   // circumstances this can cause a thread to return prematurely from
5795   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5796   // simply re-test the condition and re-park itself.
5797 }
5798 
5799 
5800 // JSR166
5801 // -------------------------------------------------------
5802 
5803 /*
5804  * The solaris and linux implementations of park/unpark are fairly
5805  * conservative for now, but can be improved. They currently use a
5806  * mutex/condvar pair, plus a a count.
5807  * Park decrements count if > 0, else does a condvar wait.  Unpark
5808  * sets count to 1 and signals condvar.  Only one thread ever waits
5809  * on the condvar. Contention seen when trying to park implies that someone
5810  * is unparking you, so don't wait. And spurious returns are fine, so there
5811  * is no need to track notifications.
5812  */
5813 
5814 #define MAX_SECS 100000000
5815 /*
5816  * This code is common to linux and solaris and will be moved to a
5817  * common place in dolphin.
5818  *
5819  * The passed in time value is either a relative time in nanoseconds
5820  * or an absolute time in milliseconds. Either way it has to be unpacked
5821  * into suitable seconds and nanoseconds components and stored in the
5822  * given timespec structure.
5823  * Given time is a 64-bit value and the time_t used in the timespec is only
5824  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5825  * overflow if times way in the future are given. Further on Solaris versions
5826  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5827  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5828  * As it will be 28 years before "now + 100000000" will overflow we can
5829  * ignore overflow and just impose a hard-limit on seconds using the value
5830  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5831  * years from "now".
5832  */
5833 
5834 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5835   assert (time > 0, "convertTime");
5836   time_t max_secs = 0;
5837 
5838   if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5839     struct timeval now;
5840     int status = gettimeofday(&now, NULL);
5841     assert(status == 0, "gettimeofday");
5842 
5843     max_secs = now.tv_sec + MAX_SECS;
5844 
5845     if (isAbsolute) {
5846       jlong secs = time / 1000;
5847       if (secs > max_secs) {
5848         absTime->tv_sec = max_secs;
5849       } else {
5850         absTime->tv_sec = secs;
5851       }
5852       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5853     } else {
5854       jlong secs = time / NANOSECS_PER_SEC;
5855       if (secs >= MAX_SECS) {
5856         absTime->tv_sec = max_secs;
5857         absTime->tv_nsec = 0;
5858       } else {
5859         absTime->tv_sec = now.tv_sec + secs;
5860         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5861         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5862           absTime->tv_nsec -= NANOSECS_PER_SEC;
5863           ++absTime->tv_sec; // note: this must be <= max_secs
5864         }
5865       }
5866     }
5867   } else {
5868     // must be relative using monotonic clock
5869     struct timespec now;
5870     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5871     assert_status(status == 0, status, "clock_gettime");
5872     max_secs = now.tv_sec + MAX_SECS;
5873     jlong secs = time / NANOSECS_PER_SEC;
5874     if (secs >= MAX_SECS) {
5875       absTime->tv_sec = max_secs;
5876       absTime->tv_nsec = 0;
5877     } else {
5878       absTime->tv_sec = now.tv_sec + secs;
5879       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5880       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5881         absTime->tv_nsec -= NANOSECS_PER_SEC;
5882         ++absTime->tv_sec; // note: this must be <= max_secs
5883       }
5884     }
5885   }
5886   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5887   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5888   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5889   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5890 }
5891 
5892 void Parker::park(bool isAbsolute, jlong time) {
5893   // Ideally we'd do something useful while spinning, such
5894   // as calling unpackTime().
5895 
5896   // Optional fast-path check:
5897   // Return immediately if a permit is available.
5898   // We depend on Atomic::xchg() having full barrier semantics
5899   // since we are doing a lock-free update to _counter.
5900   if (Atomic::xchg(0, &_counter) > 0) return;
5901 
5902   Thread* thread = Thread::current();
5903   assert(thread->is_Java_thread(), "Must be JavaThread");
5904   JavaThread *jt = (JavaThread *)thread;
5905 
5906   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5907   // Check interrupt before trying to wait
5908   if (Thread::is_interrupted(thread, false)) {
5909     return;
5910   }
5911 
5912   // Next, demultiplex/decode time arguments
5913   timespec absTime;
5914   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5915     return;
5916   }
5917   if (time > 0) {
5918     unpackTime(&absTime, isAbsolute, time);
5919   }
5920 
5921 
5922   // Enter safepoint region
5923   // Beware of deadlocks such as 6317397.
5924   // The per-thread Parker:: mutex is a classic leaf-lock.
5925   // In particular a thread must never block on the Threads_lock while
5926   // holding the Parker:: mutex.  If safepoints are pending both the
5927   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5928   ThreadBlockInVM tbivm(jt);
5929 
5930   // Don't wait if cannot get lock since interference arises from
5931   // unblocking.  Also. check interrupt before trying wait
5932   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5933     return;
5934   }
5935 
5936   int status ;
5937   if (_counter > 0)  { // no wait needed
5938     _counter = 0;
5939     status = pthread_mutex_unlock(_mutex);
5940     assert (status == 0, "invariant") ;
5941     // Paranoia to ensure our locked and lock-free paths interact
5942     // correctly with each other and Java-level accesses.
5943     OrderAccess::fence();
5944     return;
5945   }
5946 
5947 #ifdef ASSERT
5948   // Don't catch signals while blocked; let the running threads have the signals.
5949   // (This allows a debugger to break into the running thread.)
5950   sigset_t oldsigs;
5951   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5952   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5953 #endif
5954 
5955   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5956   jt->set_suspend_equivalent();
5957   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5958 
5959   assert(_cur_index == -1, "invariant");
5960   if (time == 0) {
5961     _cur_index = REL_INDEX; // arbitrary choice when not timed
5962     status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5963   } else {
5964     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5965     status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5966     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5967       pthread_cond_destroy (&_cond[_cur_index]) ;
5968       pthread_cond_init    (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5969     }
5970   }
5971   _cur_index = -1;
5972   assert_status(status == 0 || status == EINTR ||
5973                 status == ETIME || status == ETIMEDOUT,
5974                 status, "cond_timedwait");
5975 
5976 #ifdef ASSERT
5977   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5978 #endif
5979 
5980   _counter = 0 ;
5981   status = pthread_mutex_unlock(_mutex) ;
5982   assert_status(status == 0, status, "invariant") ;
5983   // Paranoia to ensure our locked and lock-free paths interact
5984   // correctly with each other and Java-level accesses.
5985   OrderAccess::fence();
5986 
5987   // If externally suspended while waiting, re-suspend
5988   if (jt->handle_special_suspend_equivalent_condition()) {
5989     jt->java_suspend_self();
5990   }
5991 }
5992 
5993 void Parker::unpark() {
5994   int s, status ;
5995   status = pthread_mutex_lock(_mutex);
5996   assert (status == 0, "invariant") ;
5997   s = _counter;
5998   _counter = 1;
5999   if (s < 1) {
6000     // thread might be parked
6001     if (_cur_index != -1) {
6002       // thread is definitely parked
6003       if (WorkAroundNPTLTimedWaitHang) {
6004         status = pthread_cond_signal (&_cond[_cur_index]);
6005         assert (status == 0, "invariant");
6006         status = pthread_mutex_unlock(_mutex);
6007         assert (status == 0, "invariant");
6008       } else {
6009         status = pthread_mutex_unlock(_mutex);
6010         assert (status == 0, "invariant");
6011         status = pthread_cond_signal (&_cond[_cur_index]);
6012         assert (status == 0, "invariant");
6013       }
6014     } else {
6015       pthread_mutex_unlock(_mutex);
6016       assert (status == 0, "invariant") ;
6017     }
6018   } else {
6019     pthread_mutex_unlock(_mutex);
6020     assert (status == 0, "invariant") ;
6021   }
6022 }
6023 
6024 
6025 extern char** environ;
6026 
6027 #ifndef __NR_fork
6028 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
6029 #endif
6030 
6031 #ifndef __NR_execve
6032 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
6033 #endif
6034 
6035 // Run the specified command in a separate process. Return its exit value,
6036 // or -1 on failure (e.g. can't fork a new process).
6037 // Unlike system(), this function can be called from signal handler. It
6038 // doesn't block SIGINT et al.
6039 int os::fork_and_exec(char* cmd) {
6040   const char * argv[4] = {"sh", "-c", cmd, NULL};
6041 
6042   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
6043   // pthread_atfork handlers and reset pthread library. All we need is a
6044   // separate process to execve. Make a direct syscall to fork process.
6045   // On IA64 there's no fork syscall, we have to use fork() and hope for
6046   // the best...
6047   pid_t pid = NOT_IA64(syscall(__NR_fork);)
6048               IA64_ONLY(fork();)
6049 
6050   if (pid < 0) {
6051     // fork failed
6052     return -1;
6053 
6054   } else if (pid == 0) {
6055     // child process
6056 
6057     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
6058     // first to kill every thread on the thread list. Because this list is
6059     // not reset by fork() (see notes above), execve() will instead kill
6060     // every thread in the parent process. We know this is the only thread
6061     // in the new process, so make a system call directly.
6062     // IA64 should use normal execve() from glibc to match the glibc fork()
6063     // above.
6064     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
6065     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
6066 
6067     // execve failed
6068     _exit(-1);
6069 
6070   } else  {
6071     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6072     // care about the actual exit code, for now.
6073 
6074     int status;
6075 
6076     // Wait for the child process to exit.  This returns immediately if
6077     // the child has already exited. */
6078     while (waitpid(pid, &status, 0) < 0) {
6079         switch (errno) {
6080         case ECHILD: return 0;
6081         case EINTR: break;
6082         default: return -1;
6083         }
6084     }
6085 
6086     if (WIFEXITED(status)) {
6087        // The child exited normally; get its exit code.
6088        return WEXITSTATUS(status);
6089     } else if (WIFSIGNALED(status)) {
6090        // The child exited because of a signal
6091        // The best value to return is 0x80 + signal number,
6092        // because that is what all Unix shells do, and because
6093        // it allows callers to distinguish between process exit and
6094        // process death by signal.
6095        return 0x80 + WTERMSIG(status);
6096     } else {
6097        // Unknown exit code; pass it through
6098        return status;
6099     }
6100   }
6101 }
6102 
6103 // is_headless_jre()
6104 //
6105 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6106 // in order to report if we are running in a headless jre
6107 //
6108 // Since JDK8 xawt/libmawt.so was moved into the same directory
6109 // as libawt.so, and renamed libawt_xawt.so
6110 //
6111 bool os::is_headless_jre() {
6112     struct stat statbuf;
6113     char buf[MAXPATHLEN];
6114     char libmawtpath[MAXPATHLEN];
6115     const char *xawtstr  = "/xawt/libmawt.so";
6116     const char *new_xawtstr = "/libawt_xawt.so";
6117     char *p;
6118 
6119     // Get path to libjvm.so
6120     os::jvm_path(buf, sizeof(buf));
6121 
6122     // Get rid of libjvm.so
6123     p = strrchr(buf, '/');
6124     if (p == NULL) return false;
6125     else *p = '\0';
6126 
6127     // Get rid of client or server
6128     p = strrchr(buf, '/');
6129     if (p == NULL) return false;
6130     else *p = '\0';
6131 
6132     // check xawt/libmawt.so
6133     strcpy(libmawtpath, buf);
6134     strcat(libmawtpath, xawtstr);
6135     if (::stat(libmawtpath, &statbuf) == 0) return false;
6136 
6137     // check libawt_xawt.so
6138     strcpy(libmawtpath, buf);
6139     strcat(libmawtpath, new_xawtstr);
6140     if (::stat(libmawtpath, &statbuf) == 0) return false;
6141 
6142     return true;
6143 }
6144 
6145 // Get the default path to the core file
6146 // Returns the length of the string
6147 int os::get_core_path(char* buffer, size_t bufferSize) {
6148   const char* p = get_current_directory(buffer, bufferSize);
6149 
6150   if (p == NULL) {
6151     assert(p != NULL, "failed to get current directory");
6152     return 0;
6153   }
6154 
6155   return strlen(buffer);
6156 }
6157 
6158 #ifdef JAVASE_EMBEDDED
6159 //
6160 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6161 //
6162 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6163 
6164 // ctor
6165 //
6166 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6167   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6168   _fd = fd;
6169 
6170   if (os::create_thread(this, os::os_thread)) {
6171     _memnotify_thread = this;
6172     os::set_priority(this, NearMaxPriority);
6173     os::start_thread(this);
6174   }
6175 }
6176 
6177 // Where all the work gets done
6178 //
6179 void MemNotifyThread::run() {
6180   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6181 
6182   // Set up the select arguments
6183   fd_set rfds;
6184   if (_fd != -1) {
6185     FD_ZERO(&rfds);
6186     FD_SET(_fd, &rfds);
6187   }
6188 
6189   // Now wait for the mem_notify device to wake up
6190   while (1) {
6191     // Wait for the mem_notify device to signal us..
6192     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6193     if (rc == -1) {
6194       perror("select!\n");
6195       break;
6196     } else if (rc) {
6197       //ssize_t free_before = os::available_memory();
6198       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6199 
6200       // The kernel is telling us there is not much memory left...
6201       // try to do something about that
6202 
6203       // If we are not already in a GC, try one.
6204       if (!Universe::heap()->is_gc_active()) {
6205         Universe::heap()->collect(GCCause::_allocation_failure);
6206 
6207         //ssize_t free_after = os::available_memory();
6208         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6209         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6210       }
6211       // We might want to do something like the following if we find the GC's are not helping...
6212       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6213     }
6214   }
6215 }
6216 
6217 //
6218 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6219 //
6220 void MemNotifyThread::start() {
6221   int    fd;
6222   fd = open ("/dev/mem_notify", O_RDONLY, 0);
6223   if (fd < 0) {
6224       return;
6225   }
6226 
6227   if (memnotify_thread() == NULL) {
6228     new MemNotifyThread(fd);
6229   }
6230 }
6231 
6232 #endif // JAVASE_EMBEDDED
6233 
6234 
6235 /////////////// Unit tests ///////////////
6236 
6237 #ifndef PRODUCT
6238 
6239 #define test_log(...) \
6240   do {\
6241     if (VerboseInternalVMTests) { \
6242       tty->print_cr(__VA_ARGS__); \
6243       tty->flush(); \
6244     }\
6245   } while (false)
6246 
6247 class TestReserveMemorySpecial : AllStatic {
6248  public:
6249   static void small_page_write(void* addr, size_t size) {
6250     size_t page_size = os::vm_page_size();
6251 
6252     char* end = (char*)addr + size;
6253     for (char* p = (char*)addr; p < end; p += page_size) {
6254       *p = 1;
6255     }
6256   }
6257 
6258   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6259     if (!UseHugeTLBFS) {
6260       return;
6261     }
6262 
6263     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6264 
6265     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6266 
6267     if (addr != NULL) {
6268       small_page_write(addr, size);
6269 
6270       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6271     }
6272   }
6273 
6274   static void test_reserve_memory_special_huge_tlbfs_only() {
6275     if (!UseHugeTLBFS) {
6276       return;
6277     }
6278 
6279     size_t lp = os::large_page_size();
6280 
6281     for (size_t size = lp; size <= lp * 10; size += lp) {
6282       test_reserve_memory_special_huge_tlbfs_only(size);
6283     }
6284   }
6285 
6286   static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6287     if (!UseHugeTLBFS) {
6288         return;
6289     }
6290 
6291     test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6292         size, alignment);
6293 
6294     assert(size >= os::large_page_size(), "Incorrect input to test");
6295 
6296     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6297 
6298     if (addr != NULL) {
6299       small_page_write(addr, size);
6300 
6301       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6302     }
6303   }
6304 
6305   static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6306     size_t lp = os::large_page_size();
6307     size_t ag = os::vm_allocation_granularity();
6308 
6309     for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6310       test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6311     }
6312   }
6313 
6314   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6315     size_t lp = os::large_page_size();
6316     size_t ag = os::vm_allocation_granularity();
6317 
6318     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6319     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6320     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6321     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6322     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6323     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6324     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6325     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6326     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6327   }
6328 
6329   static void test_reserve_memory_special_huge_tlbfs() {
6330     if (!UseHugeTLBFS) {
6331       return;
6332     }
6333 
6334     test_reserve_memory_special_huge_tlbfs_only();
6335     test_reserve_memory_special_huge_tlbfs_mixed();
6336   }
6337 
6338   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6339     if (!UseSHM) {
6340       return;
6341     }
6342 
6343     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6344 
6345     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6346 
6347     if (addr != NULL) {
6348       assert(is_ptr_aligned(addr, alignment), "Check");
6349       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6350 
6351       small_page_write(addr, size);
6352 
6353       os::Linux::release_memory_special_shm(addr, size);
6354     }
6355   }
6356 
6357   static void test_reserve_memory_special_shm() {
6358     size_t lp = os::large_page_size();
6359     size_t ag = os::vm_allocation_granularity();
6360 
6361     for (size_t size = ag; size < lp * 3; size += ag) {
6362       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6363         test_reserve_memory_special_shm(size, alignment);
6364       }
6365     }
6366   }
6367 
6368   static void test() {
6369     test_reserve_memory_special_huge_tlbfs();
6370     test_reserve_memory_special_shm();
6371   }
6372 };
6373 
6374 void TestReserveMemorySpecial_test() {
6375   TestReserveMemorySpecial::test();
6376 }
6377 
6378 #endif