1 /*
   2  * Copyright (c) 1999, 2013, 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 
 135 static jlong initial_time_count=0;
 136 
 137 static int clock_tics_per_sec = 100;
 138 
 139 // For diagnostics to print a message once. see run_periodic_checks
 140 static sigset_t check_signal_done;
 141 static bool check_signals = true;;
 142 
 143 static pid_t _initial_pid = 0;
 144 
 145 /* Signal number used to suspend/resume a thread */
 146 
 147 /* do not use any signal number less than SIGSEGV, see 4355769 */
 148 static int SR_signum = SIGUSR2;
 149 sigset_t SR_sigset;
 150 
 151 /* Used to protect dlsym() calls */
 152 static pthread_mutex_t dl_mutex;
 153 
 154 // Declarations
 155 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 156 
 157 #ifdef JAVASE_EMBEDDED
 158 class MemNotifyThread: public Thread {
 159   friend class VMStructs;
 160  public:
 161   virtual void run();
 162 
 163  private:
 164   static MemNotifyThread* _memnotify_thread;
 165   int _fd;
 166 
 167  public:
 168 
 169   // Constructor
 170   MemNotifyThread(int fd);
 171 
 172   // Tester
 173   bool is_memnotify_thread() const { return true; }
 174 
 175   // Printing
 176   char* name() const { return (char*)"Linux MemNotify Thread"; }
 177 
 178   // Returns the single instance of the MemNotifyThread
 179   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 180 
 181   // Create and start the single instance of MemNotifyThread
 182   static void start();
 183 };
 184 #endif // JAVASE_EMBEDDED
 185 
 186 // utility functions
 187 
 188 static int SR_initialize();
 189 
 190 julong os::available_memory() {
 191   return Linux::available_memory();
 192 }
 193 
 194 julong os::Linux::available_memory() {
 195   // values in struct sysinfo are "unsigned long"
 196   struct sysinfo si;
 197   sysinfo(&si);
 198 
 199   return (julong)si.freeram * si.mem_unit;
 200 }
 201 
 202 julong os::physical_memory() {
 203   return Linux::physical_memory();
 204 }
 205 
 206 ////////////////////////////////////////////////////////////////////////////////
 207 // environment support
 208 
 209 bool os::getenv(const char* name, char* buf, int len) {
 210   const char* val = ::getenv(name);
 211   if (val != NULL && strlen(val) < (size_t)len) {
 212     strcpy(buf, val);
 213     return true;
 214   }
 215   if (len > 0) buf[0] = 0;  // return a null string
 216   return false;
 217 }
 218 
 219 
 220 // Return true if user is running as root.
 221 
 222 bool os::have_special_privileges() {
 223   static bool init = false;
 224   static bool privileges = false;
 225   if (!init) {
 226     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 227     init = true;
 228   }
 229   return privileges;
 230 }
 231 
 232 
 233 #ifndef SYS_gettid
 234 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 235 #ifdef __ia64__
 236 #define SYS_gettid 1105
 237 #elif __i386__
 238 #define SYS_gettid 224
 239 #elif __amd64__
 240 #define SYS_gettid 186
 241 #elif __sparc__
 242 #define SYS_gettid 143
 243 #else
 244 #error define gettid for the arch
 245 #endif
 246 #endif
 247 
 248 // Cpu architecture string
 249 #if   defined(ZERO)
 250 static char cpu_arch[] = ZERO_LIBARCH;
 251 #elif defined(IA64)
 252 static char cpu_arch[] = "ia64";
 253 #elif defined(IA32)
 254 static char cpu_arch[] = "i386";
 255 #elif defined(AMD64)
 256 static char cpu_arch[] = "amd64";
 257 #elif defined(ARM)
 258 static char cpu_arch[] = "arm";
 259 #elif defined(PPC)
 260 static char cpu_arch[] = "ppc";
 261 #elif defined(SPARC)
 262 #  ifdef _LP64
 263 static char cpu_arch[] = "sparcv9";
 264 #  else
 265 static char cpu_arch[] = "sparc";
 266 #  endif
 267 #else
 268 #error Add appropriate cpu_arch setting
 269 #endif
 270 
 271 
 272 // pid_t gettid()
 273 //
 274 // Returns the kernel thread id of the currently running thread. Kernel
 275 // thread id is used to access /proc.
 276 //
 277 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 278 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 279 //
 280 pid_t os::Linux::gettid() {
 281   int rslt = syscall(SYS_gettid);
 282   if (rslt == -1) {
 283      // old kernel, no NPTL support
 284      return getpid();
 285   } else {
 286      return (pid_t)rslt;
 287   }
 288 }
 289 
 290 // Most versions of linux have a bug where the number of processors are
 291 // determined by looking at the /proc file system.  In a chroot environment,
 292 // the system call returns 1.  This causes the VM to act as if it is
 293 // a single processor and elide locking (see is_MP() call).
 294 static bool unsafe_chroot_detected = false;
 295 static const char *unstable_chroot_error = "/proc file system not found.\n"
 296                      "Java may be unstable running multithreaded in a chroot "
 297                      "environment on Linux when /proc filesystem is not mounted.";
 298 
 299 void os::Linux::initialize_system_info() {
 300   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 301   if (processor_count() == 1) {
 302     pid_t pid = os::Linux::gettid();
 303     char fname[32];
 304     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 305     FILE *fp = fopen(fname, "r");
 306     if (fp == NULL) {
 307       unsafe_chroot_detected = true;
 308     } else {
 309       fclose(fp);
 310     }
 311   }
 312   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 313   assert(processor_count() > 0, "linux error");
 314 }
 315 
 316 void os::init_system_properties_values() {
 317 //  char arch[12];
 318 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 319 
 320   // The next steps are taken in the product version:
 321   //
 322   // Obtain the JAVA_HOME value from the location of libjvm.so.
 323   // This library should be located at:
 324   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 325   //
 326   // If "/jre/lib/" appears at the right place in the path, then we
 327   // assume libjvm.so is installed in a JDK and we use this path.
 328   //
 329   // Otherwise exit with message: "Could not create the Java virtual machine."
 330   //
 331   // The following extra steps are taken in the debugging version:
 332   //
 333   // If "/jre/lib/" does NOT appear at the right place in the path
 334   // instead of exit check for $JAVA_HOME environment variable.
 335   //
 336   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 337   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 338   // it looks like libjvm.so is installed there
 339   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 340   //
 341   // Otherwise exit.
 342   //
 343   // Important note: if the location of libjvm.so changes this
 344   // code needs to be changed accordingly.
 345 
 346   // The next few definitions allow the code to be verbatim:
 347 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
 348 #define getenv(n) ::getenv(n)
 349 
 350 /*
 351  * See ld(1):
 352  *      The linker uses the following search paths to locate required
 353  *      shared libraries:
 354  *        1: ...
 355  *        ...
 356  *        7: The default directories, normally /lib and /usr/lib.
 357  */
 358 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 359 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 360 #else
 361 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 362 #endif
 363 
 364 #define EXTENSIONS_DIR  "/lib/ext"
 365 #define ENDORSED_DIR    "/lib/endorsed"
 366 #define REG_DIR         "/usr/java/packages"
 367 
 368   {
 369     /* sysclasspath, java_home, dll_dir */
 370     {
 371         char *home_path;
 372         char *dll_path;
 373         char *pslash;
 374         char buf[MAXPATHLEN];
 375         os::jvm_path(buf, sizeof(buf));
 376 
 377         // Found the full path to libjvm.so.
 378         // Now cut the path to <java_home>/jre if we can.
 379         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 380         pslash = strrchr(buf, '/');
 381         if (pslash != NULL)
 382             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 383         dll_path = malloc(strlen(buf) + 1);
 384         if (dll_path == NULL)
 385             return;
 386         strcpy(dll_path, buf);
 387         Arguments::set_dll_dir(dll_path);
 388 
 389         if (pslash != NULL) {
 390             pslash = strrchr(buf, '/');
 391             if (pslash != NULL) {
 392                 *pslash = '\0';       /* get rid of /<arch> */
 393                 pslash = strrchr(buf, '/');
 394                 if (pslash != NULL)
 395                     *pslash = '\0';   /* get rid of /lib */
 396             }
 397         }
 398 
 399         home_path = malloc(strlen(buf) + 1);
 400         if (home_path == NULL)
 401             return;
 402         strcpy(home_path, buf);
 403         Arguments::set_java_home(home_path);
 404 
 405         if (!set_boot_path('/', ':'))
 406             return;
 407     }
 408 
 409     /*
 410      * Where to look for native libraries
 411      *
 412      * Note: Due to a legacy implementation, most of the library path
 413      * is set in the launcher.  This was to accomodate linking restrictions
 414      * on legacy Linux implementations (which are no longer supported).
 415      * Eventually, all the library path setting will be done here.
 416      *
 417      * However, to prevent the proliferation of improperly built native
 418      * libraries, the new path component /usr/java/packages is added here.
 419      * Eventually, all the library path setting will be done here.
 420      */
 421     {
 422         char *ld_library_path;
 423 
 424         /*
 425          * Construct the invariant part of ld_library_path. Note that the
 426          * space for the colon and the trailing null are provided by the
 427          * nulls included by the sizeof operator (so actually we allocate
 428          * a byte more than necessary).
 429          */
 430         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 431             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 432         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 433 
 434         /*
 435          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 436          * should always exist (until the legacy problem cited above is
 437          * addressed).
 438          */
 439         char *v = getenv("LD_LIBRARY_PATH");
 440         if (v != NULL) {
 441             char *t = ld_library_path;
 442             /* That's +1 for the colon and +1 for the trailing '\0' */
 443             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 444             sprintf(ld_library_path, "%s:%s", v, t);
 445         }
 446         Arguments::set_library_path(ld_library_path);
 447     }
 448 
 449     /*
 450      * Extensions directories.
 451      *
 452      * Note that the space for the colon and the trailing null are provided
 453      * by the nulls included by the sizeof operator (so actually one byte more
 454      * than necessary is allocated).
 455      */
 456     {
 457         char *buf = malloc(strlen(Arguments::get_java_home()) +
 458             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 459         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 460             Arguments::get_java_home());
 461         Arguments::set_ext_dirs(buf);
 462     }
 463 
 464     /* Endorsed standards default directory. */
 465     {
 466         char * buf;
 467         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 468         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 469         Arguments::set_endorsed_dirs(buf);
 470     }
 471   }
 472 
 473 #undef malloc
 474 #undef getenv
 475 #undef EXTENSIONS_DIR
 476 #undef ENDORSED_DIR
 477 
 478   // Done
 479   return;
 480 }
 481 
 482 ////////////////////////////////////////////////////////////////////////////////
 483 // breakpoint support
 484 
 485 void os::breakpoint() {
 486   BREAKPOINT;
 487 }
 488 
 489 extern "C" void breakpoint() {
 490   // use debugger to set breakpoint here
 491 }
 492 
 493 ////////////////////////////////////////////////////////////////////////////////
 494 // signal support
 495 
 496 debug_only(static bool signal_sets_initialized = false);
 497 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 498 
 499 bool os::Linux::is_sig_ignored(int sig) {
 500       struct sigaction oact;
 501       sigaction(sig, (struct sigaction*)NULL, &oact);
 502       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 503                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 504       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 505            return true;
 506       else
 507            return false;
 508 }
 509 
 510 void os::Linux::signal_sets_init() {
 511   // Should also have an assertion stating we are still single-threaded.
 512   assert(!signal_sets_initialized, "Already initialized");
 513   // Fill in signals that are necessarily unblocked for all threads in
 514   // the VM. Currently, we unblock the following signals:
 515   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 516   //                         by -Xrs (=ReduceSignalUsage));
 517   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 518   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 519   // the dispositions or masks wrt these signals.
 520   // Programs embedding the VM that want to use the above signals for their
 521   // own purposes must, at this time, use the "-Xrs" option to prevent
 522   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 523   // (See bug 4345157, and other related bugs).
 524   // In reality, though, unblocking these signals is really a nop, since
 525   // these signals are not blocked by default.
 526   sigemptyset(&unblocked_sigs);
 527   sigemptyset(&allowdebug_blocked_sigs);
 528   sigaddset(&unblocked_sigs, SIGILL);
 529   sigaddset(&unblocked_sigs, SIGSEGV);
 530   sigaddset(&unblocked_sigs, SIGBUS);
 531   sigaddset(&unblocked_sigs, SIGFPE);
 532   sigaddset(&unblocked_sigs, SR_signum);
 533 
 534   if (!ReduceSignalUsage) {
 535    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 536       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 537       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 538    }
 539    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 540       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 541       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 542    }
 543    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 544       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 545       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 546    }
 547   }
 548   // Fill in signals that are blocked by all but the VM thread.
 549   sigemptyset(&vm_sigs);
 550   if (!ReduceSignalUsage)
 551     sigaddset(&vm_sigs, BREAK_SIGNAL);
 552   debug_only(signal_sets_initialized = true);
 553 
 554 }
 555 
 556 // These are signals that are unblocked while a thread is running Java.
 557 // (For some reason, they get blocked by default.)
 558 sigset_t* os::Linux::unblocked_signals() {
 559   assert(signal_sets_initialized, "Not initialized");
 560   return &unblocked_sigs;
 561 }
 562 
 563 // These are the signals that are blocked while a (non-VM) thread is
 564 // running Java. Only the VM thread handles these signals.
 565 sigset_t* os::Linux::vm_signals() {
 566   assert(signal_sets_initialized, "Not initialized");
 567   return &vm_sigs;
 568 }
 569 
 570 // These are signals that are blocked during cond_wait to allow debugger in
 571 sigset_t* os::Linux::allowdebug_blocked_signals() {
 572   assert(signal_sets_initialized, "Not initialized");
 573   return &allowdebug_blocked_sigs;
 574 }
 575 
 576 void os::Linux::hotspot_sigmask(Thread* thread) {
 577 
 578   //Save caller's signal mask before setting VM signal mask
 579   sigset_t caller_sigmask;
 580   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 581 
 582   OSThread* osthread = thread->osthread();
 583   osthread->set_caller_sigmask(caller_sigmask);
 584 
 585   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 586 
 587   if (!ReduceSignalUsage) {
 588     if (thread->is_VM_thread()) {
 589       // Only the VM thread handles BREAK_SIGNAL ...
 590       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 591     } else {
 592       // ... all other threads block BREAK_SIGNAL
 593       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 594     }
 595   }
 596 }
 597 
 598 //////////////////////////////////////////////////////////////////////////////
 599 // detecting pthread library
 600 
 601 void os::Linux::libpthread_init() {
 602   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 603   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 604   // generic name for earlier versions.
 605   // Define macros here so we can build HotSpot on old systems.
 606 # ifndef _CS_GNU_LIBC_VERSION
 607 # define _CS_GNU_LIBC_VERSION 2
 608 # endif
 609 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 610 # define _CS_GNU_LIBPTHREAD_VERSION 3
 611 # endif
 612 
 613   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 614   if (n > 0) {
 615      char *str = (char *)malloc(n, mtInternal);
 616      confstr(_CS_GNU_LIBC_VERSION, str, n);
 617      os::Linux::set_glibc_version(str);
 618   } else {
 619      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 620      static char _gnu_libc_version[32];
 621      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 622               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 623      os::Linux::set_glibc_version(_gnu_libc_version);
 624   }
 625 
 626   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 627   if (n > 0) {
 628      char *str = (char *)malloc(n, mtInternal);
 629      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 630      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 631      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 632      // is the case. LinuxThreads has a hard limit on max number of threads.
 633      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 634      // On the other hand, NPTL does not have such a limit, sysconf()
 635      // will return -1 and errno is not changed. Check if it is really NPTL.
 636      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 637          strstr(str, "NPTL") &&
 638          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 639        free(str);
 640        os::Linux::set_libpthread_version("linuxthreads");
 641      } else {
 642        os::Linux::set_libpthread_version(str);
 643      }
 644   } else {
 645     // glibc before 2.3.2 only has LinuxThreads.
 646     os::Linux::set_libpthread_version("linuxthreads");
 647   }
 648 
 649   if (strstr(libpthread_version(), "NPTL")) {
 650      os::Linux::set_is_NPTL();
 651   } else {
 652      os::Linux::set_is_LinuxThreads();
 653   }
 654 
 655   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 656   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 657   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 658      os::Linux::set_is_floating_stack();
 659   }
 660 }
 661 
 662 /////////////////////////////////////////////////////////////////////////////
 663 // thread stack
 664 
 665 // Force Linux kernel to expand current thread stack. If "bottom" is close
 666 // to the stack guard, caller should block all signals.
 667 //
 668 // MAP_GROWSDOWN:
 669 //   A special mmap() flag that is used to implement thread stacks. It tells
 670 //   kernel that the memory region should extend downwards when needed. This
 671 //   allows early versions of LinuxThreads to only mmap the first few pages
 672 //   when creating a new thread. Linux kernel will automatically expand thread
 673 //   stack as needed (on page faults).
 674 //
 675 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 676 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 677 //   region, it's hard to tell if the fault is due to a legitimate stack
 678 //   access or because of reading/writing non-exist memory (e.g. buffer
 679 //   overrun). As a rule, if the fault happens below current stack pointer,
 680 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 681 //   application (see Linux kernel fault.c).
 682 //
 683 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 684 //   stack overflow detection.
 685 //
 686 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 687 //   not use this flag. However, the stack of initial thread is not created
 688 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 689 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 690 //   and then attach the thread to JVM.
 691 //
 692 // To get around the problem and allow stack banging on Linux, we need to
 693 // manually expand thread stack after receiving the SIGSEGV.
 694 //
 695 // There are two ways to expand thread stack to address "bottom", we used
 696 // both of them in JVM before 1.5:
 697 //   1. adjust stack pointer first so that it is below "bottom", and then
 698 //      touch "bottom"
 699 //   2. mmap() the page in question
 700 //
 701 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 702 // if current sp is already near the lower end of page 101, and we need to
 703 // call mmap() to map page 100, it is possible that part of the mmap() frame
 704 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 705 // That will destroy the mmap() frame and cause VM to crash.
 706 //
 707 // The following code works by adjusting sp first, then accessing the "bottom"
 708 // page to force a page fault. Linux kernel will then automatically expand the
 709 // stack mapping.
 710 //
 711 // _expand_stack_to() assumes its frame size is less than page size, which
 712 // should always be true if the function is not inlined.
 713 
 714 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 715 #define NOINLINE
 716 #else
 717 #define NOINLINE __attribute__ ((noinline))
 718 #endif
 719 
 720 static void _expand_stack_to(address bottom) NOINLINE;
 721 
 722 static void _expand_stack_to(address bottom) {
 723   address sp;
 724   size_t size;
 725   volatile char *p;
 726 
 727   // Adjust bottom to point to the largest address within the same page, it
 728   // gives us a one-page buffer if alloca() allocates slightly more memory.
 729   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 730   bottom += os::Linux::page_size() - 1;
 731 
 732   // sp might be slightly above current stack pointer; if that's the case, we
 733   // will alloca() a little more space than necessary, which is OK. Don't use
 734   // os::current_stack_pointer(), as its result can be slightly below current
 735   // stack pointer, causing us to not alloca enough to reach "bottom".
 736   sp = (address)&sp;
 737 
 738   if (sp > bottom) {
 739     size = sp - bottom;
 740     p = (volatile char *)alloca(size);
 741     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 742     p[0] = '\0';
 743   }
 744 }
 745 
 746 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 747   assert(t!=NULL, "just checking");
 748   assert(t->osthread()->expanding_stack(), "expand should be set");
 749   assert(t->stack_base() != NULL, "stack_base was not initialized");
 750 
 751   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 752     sigset_t mask_all, old_sigset;
 753     sigfillset(&mask_all);
 754     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 755     _expand_stack_to(addr);
 756     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 757     return true;
 758   }
 759   return false;
 760 }
 761 
 762 //////////////////////////////////////////////////////////////////////////////
 763 // create new thread
 764 
 765 static address highest_vm_reserved_address();
 766 
 767 // check if it's safe to start a new thread
 768 static bool _thread_safety_check(Thread* thread) {
 769   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 770     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 771     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 772     //   allocated (MAP_FIXED) from high address space. Every thread stack
 773     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 774     //   it to other values if they rebuild LinuxThreads).
 775     //
 776     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 777     // the memory region has already been mmap'ed. That means if we have too
 778     // many threads and/or very large heap, eventually thread stack will
 779     // collide with heap.
 780     //
 781     // Here we try to prevent heap/stack collision by comparing current
 782     // stack bottom with the highest address that has been mmap'ed by JVM
 783     // plus a safety margin for memory maps created by native code.
 784     //
 785     // This feature can be disabled by setting ThreadSafetyMargin to 0
 786     //
 787     if (ThreadSafetyMargin > 0) {
 788       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 789 
 790       // not safe if our stack extends below the safety margin
 791       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 792     } else {
 793       return true;
 794     }
 795   } else {
 796     // Floating stack LinuxThreads or NPTL:
 797     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 798     //   there's not enough space left, pthread_create() will fail. If we come
 799     //   here, that means enough space has been reserved for stack.
 800     return true;
 801   }
 802 }
 803 
 804 // Thread start routine for all newly created threads
 805 static void *java_start(Thread *thread) {
 806   // Try to randomize the cache line index of hot stack frames.
 807   // This helps when threads of the same stack traces evict each other's
 808   // cache lines. The threads can be either from the same JVM instance, or
 809   // from different JVM instances. The benefit is especially true for
 810   // processors with hyperthreading technology.
 811   static int counter = 0;
 812   int pid = os::current_process_id();
 813   alloca(((pid ^ counter++) & 7) * 128);
 814 
 815   ThreadLocalStorage::set_thread(thread);
 816 
 817   OSThread* osthread = thread->osthread();
 818   Monitor* sync = osthread->startThread_lock();
 819 
 820   // non floating stack LinuxThreads needs extra check, see above
 821   if (!_thread_safety_check(thread)) {
 822     // notify parent thread
 823     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 824     osthread->set_state(ZOMBIE);
 825     sync->notify_all();
 826     return NULL;
 827   }
 828 
 829   // thread_id is kernel thread id (similar to Solaris LWP id)
 830   osthread->set_thread_id(os::Linux::gettid());
 831 
 832   if (UseNUMA) {
 833     int lgrp_id = os::numa_get_group_id();
 834     if (lgrp_id != -1) {
 835       thread->set_lgrp_id(lgrp_id);
 836     }
 837   }
 838   // initialize signal mask for this thread
 839   os::Linux::hotspot_sigmask(thread);
 840 
 841   // initialize floating point control register
 842   os::Linux::init_thread_fpu_state();
 843 
 844   // handshaking with parent thread
 845   {
 846     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 847 
 848     // notify parent thread
 849     osthread->set_state(INITIALIZED);
 850     sync->notify_all();
 851 
 852     // wait until os::start_thread()
 853     while (osthread->get_state() == INITIALIZED) {
 854       sync->wait(Mutex::_no_safepoint_check_flag);
 855     }
 856   }
 857 
 858   // call one more level start routine
 859   thread->run();
 860 
 861   return 0;
 862 }
 863 
 864 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 865   assert(thread->osthread() == NULL, "caller responsible");
 866 
 867   // Allocate the OSThread object
 868   OSThread* osthread = new OSThread(NULL, NULL);
 869   if (osthread == NULL) {
 870     return false;
 871   }
 872 
 873   // set the correct thread state
 874   osthread->set_thread_type(thr_type);
 875 
 876   // Initial state is ALLOCATED but not INITIALIZED
 877   osthread->set_state(ALLOCATED);
 878 
 879   thread->set_osthread(osthread);
 880 
 881   // init thread attributes
 882   pthread_attr_t attr;
 883   pthread_attr_init(&attr);
 884   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 885 
 886   // stack size
 887   if (os::Linux::supports_variable_stack_size()) {
 888     // calculate stack size if it's not specified by caller
 889     if (stack_size == 0) {
 890       stack_size = os::Linux::default_stack_size(thr_type);
 891 
 892       switch (thr_type) {
 893       case os::java_thread:
 894         // Java threads use ThreadStackSize which default value can be
 895         // changed with the flag -Xss
 896         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 897         stack_size = JavaThread::stack_size_at_create();
 898         break;
 899       case os::compiler_thread:
 900         if (CompilerThreadStackSize > 0) {
 901           stack_size = (size_t)(CompilerThreadStackSize * K);
 902           break;
 903         } // else fall through:
 904           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 905       case os::vm_thread:
 906       case os::pgc_thread:
 907       case os::cgc_thread:
 908       case os::watcher_thread:
 909         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 910         break;
 911       }
 912     }
 913 
 914     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 915     pthread_attr_setstacksize(&attr, stack_size);
 916   } else {
 917     // let pthread_create() pick the default value.
 918   }
 919 
 920   // glibc guard page
 921   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 922 
 923   ThreadState state;
 924 
 925   {
 926     // Serialize thread creation if we are running with fixed stack LinuxThreads
 927     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 928     if (lock) {
 929       os::Linux::createThread_lock()->lock_without_safepoint_check();
 930     }
 931 
 932     pthread_t tid;
 933     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 934 
 935     pthread_attr_destroy(&attr);
 936 
 937     if (ret != 0) {
 938       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 939         perror("pthread_create()");
 940       }
 941       // Need to clean up stuff we've allocated so far
 942       thread->set_osthread(NULL);
 943       delete osthread;
 944       if (lock) os::Linux::createThread_lock()->unlock();
 945       return false;
 946     }
 947 
 948     // Store pthread info into the OSThread
 949     osthread->set_pthread_id(tid);
 950 
 951     // Wait until child thread is either initialized or aborted
 952     {
 953       Monitor* sync_with_child = osthread->startThread_lock();
 954       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 955       while ((state = osthread->get_state()) == ALLOCATED) {
 956         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 957       }
 958     }
 959 
 960     if (lock) {
 961       os::Linux::createThread_lock()->unlock();
 962     }
 963   }
 964 
 965   // Aborted due to thread limit being reached
 966   if (state == ZOMBIE) {
 967       thread->set_osthread(NULL);
 968       delete osthread;
 969       return false;
 970   }
 971 
 972   // The thread is returned suspended (in state INITIALIZED),
 973   // and is started higher up in the call chain
 974   assert(state == INITIALIZED, "race condition");
 975   return true;
 976 }
 977 
 978 /////////////////////////////////////////////////////////////////////////////
 979 // attach existing thread
 980 
 981 // bootstrap the main thread
 982 bool os::create_main_thread(JavaThread* thread) {
 983   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 984   return create_attached_thread(thread);
 985 }
 986 
 987 bool os::create_attached_thread(JavaThread* thread) {
 988 #ifdef ASSERT
 989     thread->verify_not_published();
 990 #endif
 991 
 992   // Allocate the OSThread object
 993   OSThread* osthread = new OSThread(NULL, NULL);
 994 
 995   if (osthread == NULL) {
 996     return false;
 997   }
 998 
 999   // Store pthread info into the OSThread
1000   osthread->set_thread_id(os::Linux::gettid());
1001   osthread->set_pthread_id(::pthread_self());
1002 
1003   // initialize floating point control register
1004   os::Linux::init_thread_fpu_state();
1005 
1006   // Initial thread state is RUNNABLE
1007   osthread->set_state(RUNNABLE);
1008 
1009   thread->set_osthread(osthread);
1010 
1011   if (UseNUMA) {
1012     int lgrp_id = os::numa_get_group_id();
1013     if (lgrp_id != -1) {
1014       thread->set_lgrp_id(lgrp_id);
1015     }
1016   }
1017 
1018   if (os::Linux::is_initial_thread()) {
1019     // If current thread is initial thread, its stack is mapped on demand,
1020     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1021     // the entire stack region to avoid SEGV in stack banging.
1022     // It is also useful to get around the heap-stack-gap problem on SuSE
1023     // kernel (see 4821821 for details). We first expand stack to the top
1024     // of yellow zone, then enable stack yellow zone (order is significant,
1025     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1026     // is no gap between the last two virtual memory regions.
1027 
1028     JavaThread *jt = (JavaThread *)thread;
1029     address addr = jt->stack_yellow_zone_base();
1030     assert(addr != NULL, "initialization problem?");
1031     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1032 
1033     osthread->set_expanding_stack();
1034     os::Linux::manually_expand_stack(jt, addr);
1035     osthread->clear_expanding_stack();
1036   }
1037 
1038   // initialize signal mask for this thread
1039   // and save the caller's signal mask
1040   os::Linux::hotspot_sigmask(thread);
1041 
1042   return true;
1043 }
1044 
1045 void os::pd_start_thread(Thread* thread) {
1046   OSThread * osthread = thread->osthread();
1047   assert(osthread->get_state() != INITIALIZED, "just checking");
1048   Monitor* sync_with_child = osthread->startThread_lock();
1049   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1050   sync_with_child->notify();
1051 }
1052 
1053 // Free Linux resources related to the OSThread
1054 void os::free_thread(OSThread* osthread) {
1055   assert(osthread != NULL, "osthread not set");
1056 
1057   if (Thread::current()->osthread() == osthread) {
1058     // Restore caller's signal mask
1059     sigset_t sigmask = osthread->caller_sigmask();
1060     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1061    }
1062 
1063   delete osthread;
1064 }
1065 
1066 //////////////////////////////////////////////////////////////////////////////
1067 // thread local storage
1068 
1069 int os::allocate_thread_local_storage() {
1070   pthread_key_t key;
1071   int rslt = pthread_key_create(&key, NULL);
1072   assert(rslt == 0, "cannot allocate thread local storage");
1073   return (int)key;
1074 }
1075 
1076 // Note: This is currently not used by VM, as we don't destroy TLS key
1077 // on VM exit.
1078 void os::free_thread_local_storage(int index) {
1079   int rslt = pthread_key_delete((pthread_key_t)index);
1080   assert(rslt == 0, "invalid index");
1081 }
1082 
1083 void os::thread_local_storage_at_put(int index, void* value) {
1084   int rslt = pthread_setspecific((pthread_key_t)index, value);
1085   assert(rslt == 0, "pthread_setspecific failed");
1086 }
1087 
1088 extern "C" Thread* get_thread() {
1089   return ThreadLocalStorage::thread();
1090 }
1091 
1092 //////////////////////////////////////////////////////////////////////////////
1093 // initial thread
1094 
1095 // Check if current thread is the initial thread, similar to Solaris thr_main.
1096 bool os::Linux::is_initial_thread(void) {
1097   char dummy;
1098   // If called before init complete, thread stack bottom will be null.
1099   // Can be called if fatal error occurs before initialization.
1100   if (initial_thread_stack_bottom() == NULL) return false;
1101   assert(initial_thread_stack_bottom() != NULL &&
1102          initial_thread_stack_size()   != 0,
1103          "os::init did not locate initial thread's stack region");
1104   if ((address)&dummy >= initial_thread_stack_bottom() &&
1105       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1106        return true;
1107   else return false;
1108 }
1109 
1110 // Find the virtual memory area that contains addr
1111 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1112   FILE *fp = fopen("/proc/self/maps", "r");
1113   if (fp) {
1114     address low, high;
1115     while (!feof(fp)) {
1116       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1117         if (low <= addr && addr < high) {
1118            if (vma_low)  *vma_low  = low;
1119            if (vma_high) *vma_high = high;
1120            fclose (fp);
1121            return true;
1122         }
1123       }
1124       for (;;) {
1125         int ch = fgetc(fp);
1126         if (ch == EOF || ch == (int)'\n') break;
1127       }
1128     }
1129     fclose(fp);
1130   }
1131   return false;
1132 }
1133 
1134 // Locate initial thread stack. This special handling of initial thread stack
1135 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1136 // bogus value for initial thread.
1137 void os::Linux::capture_initial_stack(size_t max_size) {
1138   // stack size is the easy part, get it from RLIMIT_STACK
1139   size_t stack_size;
1140   struct rlimit rlim;
1141   getrlimit(RLIMIT_STACK, &rlim);
1142   stack_size = rlim.rlim_cur;
1143 
1144   // 6308388: a bug in ld.so will relocate its own .data section to the
1145   //   lower end of primordial stack; reduce ulimit -s value a little bit
1146   //   so we won't install guard page on ld.so's data section.
1147   stack_size -= 2 * page_size();
1148 
1149   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1150   //   7.1, in both cases we will get 2G in return value.
1151   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1152   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1153   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1154   //   in case other parts in glibc still assumes 2M max stack size.
1155   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1156   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1157   if (stack_size > 2 * K * K IA64_ONLY(*2))
1158       stack_size = 2 * K * K IA64_ONLY(*2);
1159   // Try to figure out where the stack base (top) is. This is harder.
1160   //
1161   // When an application is started, glibc saves the initial stack pointer in
1162   // a global variable "__libc_stack_end", which is then used by system
1163   // libraries. __libc_stack_end should be pretty close to stack top. The
1164   // variable is available since the very early days. However, because it is
1165   // a private interface, it could disappear in the future.
1166   //
1167   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1168   // to __libc_stack_end, it is very close to stack top, but isn't the real
1169   // stack top. Note that /proc may not exist if VM is running as a chroot
1170   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1171   // /proc/<pid>/stat could change in the future (though unlikely).
1172   //
1173   // We try __libc_stack_end first. If that doesn't work, look for
1174   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1175   // as a hint, which should work well in most cases.
1176 
1177   uintptr_t stack_start;
1178 
1179   // try __libc_stack_end first
1180   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1181   if (p && *p) {
1182     stack_start = *p;
1183   } else {
1184     // see if we can get the start_stack field from /proc/self/stat
1185     FILE *fp;
1186     int pid;
1187     char state;
1188     int ppid;
1189     int pgrp;
1190     int session;
1191     int nr;
1192     int tpgrp;
1193     unsigned long flags;
1194     unsigned long minflt;
1195     unsigned long cminflt;
1196     unsigned long majflt;
1197     unsigned long cmajflt;
1198     unsigned long utime;
1199     unsigned long stime;
1200     long cutime;
1201     long cstime;
1202     long prio;
1203     long nice;
1204     long junk;
1205     long it_real;
1206     uintptr_t start;
1207     uintptr_t vsize;
1208     intptr_t rss;
1209     uintptr_t rsslim;
1210     uintptr_t scodes;
1211     uintptr_t ecode;
1212     int i;
1213 
1214     // Figure what the primordial thread stack base is. Code is inspired
1215     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1216     // followed by command name surrounded by parentheses, state, etc.
1217     char stat[2048];
1218     int statlen;
1219 
1220     fp = fopen("/proc/self/stat", "r");
1221     if (fp) {
1222       statlen = fread(stat, 1, 2047, fp);
1223       stat[statlen] = '\0';
1224       fclose(fp);
1225 
1226       // Skip pid and the command string. Note that we could be dealing with
1227       // weird command names, e.g. user could decide to rename java launcher
1228       // to "java 1.4.2 :)", then the stat file would look like
1229       //                1234 (java 1.4.2 :)) R ... ...
1230       // We don't really need to know the command string, just find the last
1231       // occurrence of ")" and then start parsing from there. See bug 4726580.
1232       char * s = strrchr(stat, ')');
1233 
1234       i = 0;
1235       if (s) {
1236         // Skip blank chars
1237         do s++; while (isspace(*s));
1238 
1239 #define _UFM UINTX_FORMAT
1240 #define _DFM INTX_FORMAT
1241 
1242         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1243         /*              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 */
1244         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,
1245              &state,          /* 3  %c  */
1246              &ppid,           /* 4  %d  */
1247              &pgrp,           /* 5  %d  */
1248              &session,        /* 6  %d  */
1249              &nr,             /* 7  %d  */
1250              &tpgrp,          /* 8  %d  */
1251              &flags,          /* 9  %lu  */
1252              &minflt,         /* 10 %lu  */
1253              &cminflt,        /* 11 %lu  */
1254              &majflt,         /* 12 %lu  */
1255              &cmajflt,        /* 13 %lu  */
1256              &utime,          /* 14 %lu  */
1257              &stime,          /* 15 %lu  */
1258              &cutime,         /* 16 %ld  */
1259              &cstime,         /* 17 %ld  */
1260              &prio,           /* 18 %ld  */
1261              &nice,           /* 19 %ld  */
1262              &junk,           /* 20 %ld  */
1263              &it_real,        /* 21 %ld  */
1264              &start,          /* 22 UINTX_FORMAT */
1265              &vsize,          /* 23 UINTX_FORMAT */
1266              &rss,            /* 24 INTX_FORMAT  */
1267              &rsslim,         /* 25 UINTX_FORMAT */
1268              &scodes,         /* 26 UINTX_FORMAT */
1269              &ecode,          /* 27 UINTX_FORMAT */
1270              &stack_start);   /* 28 UINTX_FORMAT */
1271       }
1272 
1273 #undef _UFM
1274 #undef _DFM
1275 
1276       if (i != 28 - 2) {
1277          assert(false, "Bad conversion from /proc/self/stat");
1278          // product mode - assume we are the initial thread, good luck in the
1279          // embedded case.
1280          warning("Can't detect initial thread stack location - bad conversion");
1281          stack_start = (uintptr_t) &rlim;
1282       }
1283     } else {
1284       // For some reason we can't open /proc/self/stat (for example, running on
1285       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1286       // most cases, so don't abort:
1287       warning("Can't detect initial thread stack location - no /proc/self/stat");
1288       stack_start = (uintptr_t) &rlim;
1289     }
1290   }
1291 
1292   // Now we have a pointer (stack_start) very close to the stack top, the
1293   // next thing to do is to figure out the exact location of stack top. We
1294   // can find out the virtual memory area that contains stack_start by
1295   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1296   // and its upper limit is the real stack top. (again, this would fail if
1297   // running inside chroot, because /proc may not exist.)
1298 
1299   uintptr_t stack_top;
1300   address low, high;
1301   if (find_vma((address)stack_start, &low, &high)) {
1302     // success, "high" is the true stack top. (ignore "low", because initial
1303     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1304     stack_top = (uintptr_t)high;
1305   } else {
1306     // failed, likely because /proc/self/maps does not exist
1307     warning("Can't detect initial thread stack location - find_vma failed");
1308     // best effort: stack_start is normally within a few pages below the real
1309     // stack top, use it as stack top, and reduce stack size so we won't put
1310     // guard page outside stack.
1311     stack_top = stack_start;
1312     stack_size -= 16 * page_size();
1313   }
1314 
1315   // stack_top could be partially down the page so align it
1316   stack_top = align_size_up(stack_top, page_size());
1317 
1318   if (max_size && stack_size > max_size) {
1319      _initial_thread_stack_size = max_size;
1320   } else {
1321      _initial_thread_stack_size = stack_size;
1322   }
1323 
1324   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1325   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1326 }
1327 
1328 ////////////////////////////////////////////////////////////////////////////////
1329 // time support
1330 
1331 // Time since start-up in seconds to a fine granularity.
1332 // Used by VMSelfDestructTimer and the MemProfiler.
1333 double os::elapsedTime() {
1334 
1335   return (double)(os::elapsed_counter()) * 0.000001;
1336 }
1337 
1338 jlong os::elapsed_counter() {
1339   timeval time;
1340   int status = gettimeofday(&time, NULL);
1341   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1342 }
1343 
1344 jlong os::elapsed_frequency() {
1345   return (1000 * 1000);
1346 }
1347 
1348 bool os::supports_vtime() { return true; }
1349 bool os::enable_vtime()   { return false; }
1350 bool os::vtime_enabled()  { return false; }
1351 
1352 double os::elapsedVTime() {
1353   struct rusage usage;
1354   int retval = getrusage(RUSAGE_THREAD, &usage);
1355   if (retval == 0) {
1356     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);
1357   } else {
1358     // better than nothing, but not much
1359     return elapsedTime();
1360   }
1361 }
1362 
1363 jlong os::javaTimeMillis() {
1364   timeval time;
1365   int status = gettimeofday(&time, NULL);
1366   assert(status != -1, "linux error");
1367   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1368 }
1369 
1370 #ifndef CLOCK_MONOTONIC
1371 #define CLOCK_MONOTONIC (1)
1372 #endif
1373 
1374 void os::Linux::clock_init() {
1375   // we do dlopen's in this particular order due to bug in linux
1376   // dynamical loader (see 6348968) leading to crash on exit
1377   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1378   if (handle == NULL) {
1379     handle = dlopen("librt.so", RTLD_LAZY);
1380   }
1381 
1382   if (handle) {
1383     int (*clock_getres_func)(clockid_t, struct timespec*) =
1384            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1385     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1386            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1387     if (clock_getres_func && clock_gettime_func) {
1388       // See if monotonic clock is supported by the kernel. Note that some
1389       // early implementations simply return kernel jiffies (updated every
1390       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1391       // for nano time (though the monotonic property is still nice to have).
1392       // It's fixed in newer kernels, however clock_getres() still returns
1393       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1394       // resolution for now. Hopefully as people move to new kernels, this
1395       // won't be a problem.
1396       struct timespec res;
1397       struct timespec tp;
1398       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1399           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1400         // yes, monotonic clock is supported
1401         _clock_gettime = clock_gettime_func;
1402       } else {
1403         // close librt if there is no monotonic clock
1404         dlclose(handle);
1405       }
1406     }
1407   }
1408 }
1409 
1410 #ifndef SYS_clock_getres
1411 
1412 #if defined(IA32) || defined(AMD64)
1413 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1414 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1415 #else
1416 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1417 #define sys_clock_getres(x,y)  -1
1418 #endif
1419 
1420 #else
1421 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1422 #endif
1423 
1424 void os::Linux::fast_thread_clock_init() {
1425   if (!UseLinuxPosixThreadCPUClocks) {
1426     return;
1427   }
1428   clockid_t clockid;
1429   struct timespec tp;
1430   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1431       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1432 
1433   // Switch to using fast clocks for thread cpu time if
1434   // the sys_clock_getres() returns 0 error code.
1435   // Note, that some kernels may support the current thread
1436   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1437   // returned by the pthread_getcpuclockid().
1438   // If the fast Posix clocks are supported then the sys_clock_getres()
1439   // must return at least tp.tv_sec == 0 which means a resolution
1440   // better than 1 sec. This is extra check for reliability.
1441 
1442   if(pthread_getcpuclockid_func &&
1443      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1444      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1445 
1446     _supports_fast_thread_cpu_time = true;
1447     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1448   }
1449 }
1450 
1451 jlong os::javaTimeNanos() {
1452   if (Linux::supports_monotonic_clock()) {
1453     struct timespec tp;
1454     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1455     assert(status == 0, "gettime error");
1456     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1457     return result;
1458   } else {
1459     timeval time;
1460     int status = gettimeofday(&time, NULL);
1461     assert(status != -1, "linux error");
1462     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1463     return 1000 * usecs;
1464   }
1465 }
1466 
1467 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1468   if (Linux::supports_monotonic_clock()) {
1469     info_ptr->max_value = ALL_64_BITS;
1470 
1471     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1472     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1473     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1474   } else {
1475     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1476     info_ptr->max_value = ALL_64_BITS;
1477 
1478     // gettimeofday is a real time clock so it skips
1479     info_ptr->may_skip_backward = true;
1480     info_ptr->may_skip_forward = true;
1481   }
1482 
1483   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1484 }
1485 
1486 // Return the real, user, and system times in seconds from an
1487 // arbitrary fixed point in the past.
1488 bool os::getTimesSecs(double* process_real_time,
1489                       double* process_user_time,
1490                       double* process_system_time) {
1491   struct tms ticks;
1492   clock_t real_ticks = times(&ticks);
1493 
1494   if (real_ticks == (clock_t) (-1)) {
1495     return false;
1496   } else {
1497     double ticks_per_second = (double) clock_tics_per_sec;
1498     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1499     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1500     *process_real_time = ((double) real_ticks) / ticks_per_second;
1501 
1502     return true;
1503   }
1504 }
1505 
1506 
1507 char * os::local_time_string(char *buf, size_t buflen) {
1508   struct tm t;
1509   time_t long_time;
1510   time(&long_time);
1511   localtime_r(&long_time, &t);
1512   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1513                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1514                t.tm_hour, t.tm_min, t.tm_sec);
1515   return buf;
1516 }
1517 
1518 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1519   return localtime_r(clock, res);
1520 }
1521 
1522 ////////////////////////////////////////////////////////////////////////////////
1523 // runtime exit support
1524 
1525 // Note: os::shutdown() might be called very early during initialization, or
1526 // called from signal handler. Before adding something to os::shutdown(), make
1527 // sure it is async-safe and can handle partially initialized VM.
1528 void os::shutdown() {
1529 
1530   // allow PerfMemory to attempt cleanup of any persistent resources
1531   perfMemory_exit();
1532 
1533   // needs to remove object in file system
1534   AttachListener::abort();
1535 
1536   // flush buffered output, finish log files
1537   ostream_abort();
1538 
1539   // Check for abort hook
1540   abort_hook_t abort_hook = Arguments::abort_hook();
1541   if (abort_hook != NULL) {
1542     abort_hook();
1543   }
1544 
1545 }
1546 
1547 // Note: os::abort() might be called very early during initialization, or
1548 // called from signal handler. Before adding something to os::abort(), make
1549 // sure it is async-safe and can handle partially initialized VM.
1550 void os::abort(bool dump_core) {
1551   os::shutdown();
1552   if (dump_core) {
1553 #ifndef PRODUCT
1554     fdStream out(defaultStream::output_fd());
1555     out.print_raw("Current thread is ");
1556     char buf[16];
1557     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1558     out.print_raw_cr(buf);
1559     out.print_raw_cr("Dumping core ...");
1560 #endif
1561     ::abort(); // dump core
1562   }
1563 
1564   ::exit(1);
1565 }
1566 
1567 // Die immediately, no exit hook, no abort hook, no cleanup.
1568 void os::die() {
1569   // _exit() on LinuxThreads only kills current thread
1570   ::abort();
1571 }
1572 
1573 // unused on linux for now.
1574 void os::set_error_file(const char *logfile) {}
1575 
1576 
1577 // This method is a copy of JDK's sysGetLastErrorString
1578 // from src/solaris/hpi/src/system_md.c
1579 
1580 size_t os::lasterror(char *buf, size_t len) {
1581 
1582   if (errno == 0)  return 0;
1583 
1584   const char *s = ::strerror(errno);
1585   size_t n = ::strlen(s);
1586   if (n >= len) {
1587     n = len - 1;
1588   }
1589   ::strncpy(buf, s, n);
1590   buf[n] = '\0';
1591   return n;
1592 }
1593 
1594 intx os::current_thread_id() { return (intx)pthread_self(); }
1595 int os::current_process_id() {
1596 
1597   // Under the old linux thread library, linux gives each thread
1598   // its own process id. Because of this each thread will return
1599   // a different pid if this method were to return the result
1600   // of getpid(2). Linux provides no api that returns the pid
1601   // of the launcher thread for the vm. This implementation
1602   // returns a unique pid, the pid of the launcher thread
1603   // that starts the vm 'process'.
1604 
1605   // Under the NPTL, getpid() returns the same pid as the
1606   // launcher thread rather than a unique pid per thread.
1607   // Use gettid() if you want the old pre NPTL behaviour.
1608 
1609   // if you are looking for the result of a call to getpid() that
1610   // returns a unique pid for the calling thread, then look at the
1611   // OSThread::thread_id() method in osThread_linux.hpp file
1612 
1613   return (int)(_initial_pid ? _initial_pid : getpid());
1614 }
1615 
1616 // DLL functions
1617 
1618 const char* os::dll_file_extension() { return ".so"; }
1619 
1620 // This must be hard coded because it's the system's temporary
1621 // directory not the java application's temp directory, ala java.io.tmpdir.
1622 const char* os::get_temp_directory() { return "/tmp"; }
1623 
1624 static bool file_exists(const char* filename) {
1625   struct stat statbuf;
1626   if (filename == NULL || strlen(filename) == 0) {
1627     return false;
1628   }
1629   return os::stat(filename, &statbuf) == 0;
1630 }
1631 
1632 bool os::dll_build_name(char* buffer, size_t buflen,
1633                         const char* pname, const char* fname) {
1634   bool retval = false;
1635   // Copied from libhpi
1636   const size_t pnamelen = pname ? strlen(pname) : 0;
1637 
1638   // Return error on buffer overflow.
1639   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1640     return retval;
1641   }
1642 
1643   if (pnamelen == 0) {
1644     snprintf(buffer, buflen, "lib%s.so", fname);
1645     retval = true;
1646   } else if (strchr(pname, *os::path_separator()) != NULL) {
1647     int n;
1648     char** pelements = split_path(pname, &n);
1649     if (pelements == NULL) {
1650       return false;
1651     }
1652     for (int i = 0 ; i < n ; i++) {
1653       // Really shouldn't be NULL, but check can't hurt
1654       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1655         continue; // skip the empty path values
1656       }
1657       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1658       if (file_exists(buffer)) {
1659         retval = true;
1660         break;
1661       }
1662     }
1663     // release the storage
1664     for (int i = 0 ; i < n ; i++) {
1665       if (pelements[i] != NULL) {
1666         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1667       }
1668     }
1669     if (pelements != NULL) {
1670       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1671     }
1672   } else {
1673     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1674     retval = true;
1675   }
1676   return retval;
1677 }
1678 
1679 // check if addr is inside libjvm.so
1680 bool os::address_is_in_vm(address addr) {
1681   static address libjvm_base_addr;
1682   Dl_info dlinfo;
1683 
1684   if (libjvm_base_addr == NULL) {
1685     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1686       libjvm_base_addr = (address)dlinfo.dli_fbase;
1687     }
1688     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1689   }
1690 
1691   if (dladdr((void *)addr, &dlinfo) != 0) {
1692     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1693   }
1694 
1695   return false;
1696 }
1697 
1698 bool os::dll_address_to_function_name(address addr, char *buf,
1699                                       int buflen, int *offset) {
1700   // buf is not optional, but offset is optional
1701   assert(buf != NULL, "sanity check");
1702 
1703   Dl_info dlinfo;
1704 
1705   if (dladdr((void*)addr, &dlinfo) != 0) {
1706     // see if we have a matching symbol
1707     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1708       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1709         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1710       }
1711       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1712       return true;
1713     }
1714     // no matching symbol so try for just file info
1715     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1716       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1717                           buf, buflen, offset, dlinfo.dli_fname)) {
1718         return true;
1719       }
1720     }
1721   }
1722 
1723   buf[0] = '\0';
1724   if (offset != NULL) *offset = -1;
1725   return false;
1726 }
1727 
1728 struct _address_to_library_name {
1729   address addr;          // input : memory address
1730   size_t  buflen;        //         size of fname
1731   char*   fname;         // output: library name
1732   address base;          //         library base addr
1733 };
1734 
1735 static int address_to_library_name_callback(struct dl_phdr_info *info,
1736                                             size_t size, void *data) {
1737   int i;
1738   bool found = false;
1739   address libbase = NULL;
1740   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1741 
1742   // iterate through all loadable segments
1743   for (i = 0; i < info->dlpi_phnum; i++) {
1744     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1745     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1746       // base address of a library is the lowest address of its loaded
1747       // segments.
1748       if (libbase == NULL || libbase > segbase) {
1749         libbase = segbase;
1750       }
1751       // see if 'addr' is within current segment
1752       if (segbase <= d->addr &&
1753           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1754         found = true;
1755       }
1756     }
1757   }
1758 
1759   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1760   // so dll_address_to_library_name() can fall through to use dladdr() which
1761   // can figure out executable name from argv[0].
1762   if (found && info->dlpi_name && info->dlpi_name[0]) {
1763     d->base = libbase;
1764     if (d->fname) {
1765       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1766     }
1767     return 1;
1768   }
1769   return 0;
1770 }
1771 
1772 bool os::dll_address_to_library_name(address addr, char* buf,
1773                                      int buflen, int* offset) {
1774   // buf is not optional, but offset is optional
1775   assert(buf != NULL, "sanity check");
1776 
1777   Dl_info dlinfo;
1778   struct _address_to_library_name data;
1779 
1780   // There is a bug in old glibc dladdr() implementation that it could resolve
1781   // to wrong library name if the .so file has a base address != NULL. Here
1782   // we iterate through the program headers of all loaded libraries to find
1783   // out which library 'addr' really belongs to. This workaround can be
1784   // removed once the minimum requirement for glibc is moved to 2.3.x.
1785   data.addr = addr;
1786   data.fname = buf;
1787   data.buflen = buflen;
1788   data.base = NULL;
1789   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1790 
1791   if (rslt) {
1792      // buf already contains library name
1793      if (offset) *offset = addr - data.base;
1794      return true;
1795   }
1796   if (dladdr((void*)addr, &dlinfo) != 0) {
1797     if (dlinfo.dli_fname != NULL) {
1798       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1799     }
1800     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1801       *offset = addr - (address)dlinfo.dli_fbase;
1802     }
1803     return true;
1804   }
1805 
1806   buf[0] = '\0';
1807   if (offset) *offset = -1;
1808   return false;
1809 }
1810 
1811   // Loads .dll/.so and
1812   // in case of error it checks if .dll/.so was built for the
1813   // same architecture as Hotspot is running on
1814 
1815 
1816 // Remember the stack's state. The Linux dynamic linker will change
1817 // the stack to 'executable' at most once, so we must safepoint only once.
1818 bool os::Linux::_stack_is_executable = false;
1819 
1820 // VM operation that loads a library.  This is necessary if stack protection
1821 // of the Java stacks can be lost during loading the library.  If we
1822 // do not stop the Java threads, they can stack overflow before the stacks
1823 // are protected again.
1824 class VM_LinuxDllLoad: public VM_Operation {
1825  private:
1826   const char *_filename;
1827   char *_ebuf;
1828   int _ebuflen;
1829   void *_lib;
1830  public:
1831   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1832     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1833   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1834   void doit() {
1835     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1836     os::Linux::_stack_is_executable = true;
1837   }
1838   void* loaded_library() { return _lib; }
1839 };
1840 
1841 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1842 {
1843   void * result = NULL;
1844   bool load_attempted = false;
1845 
1846   // Check whether the library to load might change execution rights
1847   // of the stack. If they are changed, the protection of the stack
1848   // guard pages will be lost. We need a safepoint to fix this.
1849   //
1850   // See Linux man page execstack(8) for more info.
1851   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1852     ElfFile ef(filename);
1853     if (!ef.specifies_noexecstack()) {
1854       if (!is_init_completed()) {
1855         os::Linux::_stack_is_executable = true;
1856         // This is OK - No Java threads have been created yet, and hence no
1857         // stack guard pages to fix.
1858         //
1859         // This should happen only when you are building JDK7 using a very
1860         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1861         //
1862         // Dynamic loader will make all stacks executable after
1863         // this function returns, and will not do that again.
1864         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1865       } else {
1866         warning("You have loaded library %s which might have disabled stack guard. "
1867                 "The VM will try to fix the stack guard now.\n"
1868                 "It's highly recommended that you fix the library with "
1869                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1870                 filename);
1871 
1872         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1873         JavaThread *jt = JavaThread::current();
1874         if (jt->thread_state() != _thread_in_native) {
1875           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1876           // that requires ExecStack. Cannot enter safe point. Let's give up.
1877           warning("Unable to fix stack guard. Giving up.");
1878         } else {
1879           if (!LoadExecStackDllInVMThread) {
1880             // This is for the case where the DLL has an static
1881             // constructor function that executes JNI code. We cannot
1882             // load such DLLs in the VMThread.
1883             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1884           }
1885 
1886           ThreadInVMfromNative tiv(jt);
1887           debug_only(VMNativeEntryWrapper vew;)
1888 
1889           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1890           VMThread::execute(&op);
1891           if (LoadExecStackDllInVMThread) {
1892             result = op.loaded_library();
1893           }
1894           load_attempted = true;
1895         }
1896       }
1897     }
1898   }
1899 
1900   if (!load_attempted) {
1901     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1902   }
1903 
1904   if (result != NULL) {
1905     // Successful loading
1906     return result;
1907   }
1908 
1909   Elf32_Ehdr elf_head;
1910   int diag_msg_max_length=ebuflen-strlen(ebuf);
1911   char* diag_msg_buf=ebuf+strlen(ebuf);
1912 
1913   if (diag_msg_max_length==0) {
1914     // No more space in ebuf for additional diagnostics message
1915     return NULL;
1916   }
1917 
1918 
1919   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1920 
1921   if (file_descriptor < 0) {
1922     // Can't open library, report dlerror() message
1923     return NULL;
1924   }
1925 
1926   bool failed_to_read_elf_head=
1927     (sizeof(elf_head)!=
1928         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1929 
1930   ::close(file_descriptor);
1931   if (failed_to_read_elf_head) {
1932     // file i/o error - report dlerror() msg
1933     return NULL;
1934   }
1935 
1936   typedef struct {
1937     Elf32_Half  code;         // Actual value as defined in elf.h
1938     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1939     char        elf_class;    // 32 or 64 bit
1940     char        endianess;    // MSB or LSB
1941     char*       name;         // String representation
1942   } arch_t;
1943 
1944   #ifndef EM_486
1945   #define EM_486          6               /* Intel 80486 */
1946   #endif
1947 
1948   static const arch_t arch_array[]={
1949     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1950     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1951     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1952     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1953     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1954     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1955     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1956     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1957     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1958     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1959     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1960     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1961     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1962     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1963     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1964     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1965   };
1966 
1967   #if  (defined IA32)
1968     static  Elf32_Half running_arch_code=EM_386;
1969   #elif   (defined AMD64)
1970     static  Elf32_Half running_arch_code=EM_X86_64;
1971   #elif  (defined IA64)
1972     static  Elf32_Half running_arch_code=EM_IA_64;
1973   #elif  (defined __sparc) && (defined _LP64)
1974     static  Elf32_Half running_arch_code=EM_SPARCV9;
1975   #elif  (defined __sparc) && (!defined _LP64)
1976     static  Elf32_Half running_arch_code=EM_SPARC;
1977   #elif  (defined __powerpc64__)
1978     static  Elf32_Half running_arch_code=EM_PPC64;
1979   #elif  (defined __powerpc__)
1980     static  Elf32_Half running_arch_code=EM_PPC;
1981   #elif  (defined ARM)
1982     static  Elf32_Half running_arch_code=EM_ARM;
1983   #elif  (defined S390)
1984     static  Elf32_Half running_arch_code=EM_S390;
1985   #elif  (defined ALPHA)
1986     static  Elf32_Half running_arch_code=EM_ALPHA;
1987   #elif  (defined MIPSEL)
1988     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1989   #elif  (defined PARISC)
1990     static  Elf32_Half running_arch_code=EM_PARISC;
1991   #elif  (defined MIPS)
1992     static  Elf32_Half running_arch_code=EM_MIPS;
1993   #elif  (defined M68K)
1994     static  Elf32_Half running_arch_code=EM_68K;
1995   #else
1996     #error Method os::dll_load requires that one of following is defined:\
1997          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1998   #endif
1999 
2000   // Identify compatability class for VM's architecture and library's architecture
2001   // Obtain string descriptions for architectures
2002 
2003   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2004   int running_arch_index=-1;
2005 
2006   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2007     if (running_arch_code == arch_array[i].code) {
2008       running_arch_index    = i;
2009     }
2010     if (lib_arch.code == arch_array[i].code) {
2011       lib_arch.compat_class = arch_array[i].compat_class;
2012       lib_arch.name         = arch_array[i].name;
2013     }
2014   }
2015 
2016   assert(running_arch_index != -1,
2017     "Didn't find running architecture code (running_arch_code) in arch_array");
2018   if (running_arch_index == -1) {
2019     // Even though running architecture detection failed
2020     // we may still continue with reporting dlerror() message
2021     return NULL;
2022   }
2023 
2024   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2025     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2026     return NULL;
2027   }
2028 
2029 #ifndef S390
2030   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2031     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2032     return NULL;
2033   }
2034 #endif // !S390
2035 
2036   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2037     if ( lib_arch.name!=NULL ) {
2038       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2039         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2040         lib_arch.name, arch_array[running_arch_index].name);
2041     } else {
2042       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2043       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2044         lib_arch.code,
2045         arch_array[running_arch_index].name);
2046     }
2047   }
2048 
2049   return NULL;
2050 }
2051 
2052 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2053   void * result = ::dlopen(filename, RTLD_LAZY);
2054   if (result == NULL) {
2055     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2056     ebuf[ebuflen-1] = '\0';
2057   }
2058   return result;
2059 }
2060 
2061 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2062   void * result = NULL;
2063   if (LoadExecStackDllInVMThread) {
2064     result = dlopen_helper(filename, ebuf, ebuflen);
2065   }
2066 
2067   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2068   // library that requires an executable stack, or which does not have this
2069   // stack attribute set, dlopen changes the stack attribute to executable. The
2070   // read protection of the guard pages gets lost.
2071   //
2072   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2073   // may have been queued at the same time.
2074 
2075   if (!_stack_is_executable) {
2076     JavaThread *jt = Threads::first();
2077 
2078     while (jt) {
2079       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2080           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2081         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2082                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2083           warning("Attempt to reguard stack yellow zone failed.");
2084         }
2085       }
2086       jt = jt->next();
2087     }
2088   }
2089 
2090   return result;
2091 }
2092 
2093 /*
2094  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2095  * chances are you might want to run the generated bits against glibc-2.0
2096  * libdl.so, so always use locking for any version of glibc.
2097  */
2098 void* os::dll_lookup(void* handle, const char* name) {
2099   pthread_mutex_lock(&dl_mutex);
2100   void* res = dlsym(handle, name);
2101   pthread_mutex_unlock(&dl_mutex);
2102   return res;
2103 }
2104 
2105 
2106 static bool _print_ascii_file(const char* filename, outputStream* st) {
2107   int fd = ::open(filename, O_RDONLY);
2108   if (fd == -1) {
2109      return false;
2110   }
2111 
2112   char buf[32];
2113   int bytes;
2114   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2115     st->print_raw(buf, bytes);
2116   }
2117 
2118   ::close(fd);
2119 
2120   return true;
2121 }
2122 
2123 void os::print_dll_info(outputStream *st) {
2124    st->print_cr("Dynamic libraries:");
2125 
2126    char fname[32];
2127    pid_t pid = os::Linux::gettid();
2128 
2129    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2130 
2131    if (!_print_ascii_file(fname, st)) {
2132      st->print("Can not get library information for pid = %d\n", pid);
2133    }
2134 }
2135 
2136 void os::print_os_info_brief(outputStream* st) {
2137   os::Linux::print_distro_info(st);
2138 
2139   os::Posix::print_uname_info(st);
2140 
2141   os::Linux::print_libversion_info(st);
2142 
2143 }
2144 
2145 void os::print_os_info(outputStream* st) {
2146   st->print("OS:");
2147 
2148   os::Linux::print_distro_info(st);
2149 
2150   os::Posix::print_uname_info(st);
2151 
2152   // Print warning if unsafe chroot environment detected
2153   if (unsafe_chroot_detected) {
2154     st->print("WARNING!! ");
2155     st->print_cr(unstable_chroot_error);
2156   }
2157 
2158   os::Linux::print_libversion_info(st);
2159 
2160   os::Posix::print_rlimit_info(st);
2161 
2162   os::Posix::print_load_average(st);
2163 
2164   os::Linux::print_full_memory_info(st);
2165 }
2166 
2167 // Try to identify popular distros.
2168 // Most Linux distributions have /etc/XXX-release file, which contains
2169 // the OS version string. Some have more than one /etc/XXX-release file
2170 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2171 // so the order is important.
2172 void os::Linux::print_distro_info(outputStream* st) {
2173   if (!_print_ascii_file("/etc/mandrake-release", st) &&
2174       !_print_ascii_file("/etc/sun-release", st) &&
2175       !_print_ascii_file("/etc/redhat-release", st) &&
2176       !_print_ascii_file("/etc/SuSE-release", st) &&
2177       !_print_ascii_file("/etc/turbolinux-release", st) &&
2178       !_print_ascii_file("/etc/gentoo-release", st) &&
2179       !_print_ascii_file("/etc/debian_version", st) &&
2180       !_print_ascii_file("/etc/ltib-release", st) &&
2181       !_print_ascii_file("/etc/angstrom-version", st)) {
2182       st->print("Linux");
2183   }
2184   st->cr();
2185 }
2186 
2187 void os::Linux::print_libversion_info(outputStream* st) {
2188   // libc, pthread
2189   st->print("libc:");
2190   st->print(os::Linux::glibc_version()); st->print(" ");
2191   st->print(os::Linux::libpthread_version()); st->print(" ");
2192   if (os::Linux::is_LinuxThreads()) {
2193      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2194   }
2195   st->cr();
2196 }
2197 
2198 void os::Linux::print_full_memory_info(outputStream* st) {
2199    st->print("\n/proc/meminfo:\n");
2200    _print_ascii_file("/proc/meminfo", st);
2201    st->cr();
2202 }
2203 
2204 void os::print_memory_info(outputStream* st) {
2205 
2206   st->print("Memory:");
2207   st->print(" %dk page", os::vm_page_size()>>10);
2208 
2209   // values in struct sysinfo are "unsigned long"
2210   struct sysinfo si;
2211   sysinfo(&si);
2212 
2213   st->print(", physical " UINT64_FORMAT "k",
2214             os::physical_memory() >> 10);
2215   st->print("(" UINT64_FORMAT "k free)",
2216             os::available_memory() >> 10);
2217   st->print(", swap " UINT64_FORMAT "k",
2218             ((jlong)si.totalswap * si.mem_unit) >> 10);
2219   st->print("(" UINT64_FORMAT "k free)",
2220             ((jlong)si.freeswap * si.mem_unit) >> 10);
2221   st->cr();
2222 }
2223 
2224 void os::pd_print_cpu_info(outputStream* st) {
2225   st->print("\n/proc/cpuinfo:\n");
2226   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2227     st->print("  <Not Available>");
2228   }
2229   st->cr();
2230 }
2231 
2232 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2233 // but they're the same for all the linux arch that we support
2234 // and they're the same for solaris but there's no common place to put this.
2235 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2236                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2237                           "ILL_COPROC", "ILL_BADSTK" };
2238 
2239 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2240                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2241                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2242 
2243 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2244 
2245 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2246 
2247 void os::print_siginfo(outputStream* st, void* siginfo) {
2248   st->print("siginfo:");
2249 
2250   const int buflen = 100;
2251   char buf[buflen];
2252   siginfo_t *si = (siginfo_t*)siginfo;
2253   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2254   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2255     st->print("si_errno=%s", buf);
2256   } else {
2257     st->print("si_errno=%d", si->si_errno);
2258   }
2259   const int c = si->si_code;
2260   assert(c > 0, "unexpected si_code");
2261   switch (si->si_signo) {
2262   case SIGILL:
2263     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2264     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2265     break;
2266   case SIGFPE:
2267     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2268     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2269     break;
2270   case SIGSEGV:
2271     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2272     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2273     break;
2274   case SIGBUS:
2275     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2276     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2277     break;
2278   default:
2279     st->print(", si_code=%d", si->si_code);
2280     // no si_addr
2281   }
2282 
2283   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2284       UseSharedSpaces) {
2285     FileMapInfo* mapinfo = FileMapInfo::current_info();
2286     if (mapinfo->is_in_shared_space(si->si_addr)) {
2287       st->print("\n\nError accessing class data sharing archive."   \
2288                 " Mapped file inaccessible during execution, "      \
2289                 " possible disk/network problem.");
2290     }
2291   }
2292   st->cr();
2293 }
2294 
2295 
2296 static void print_signal_handler(outputStream* st, int sig,
2297                                  char* buf, size_t buflen);
2298 
2299 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2300   st->print_cr("Signal Handlers:");
2301   print_signal_handler(st, SIGSEGV, buf, buflen);
2302   print_signal_handler(st, SIGBUS , buf, buflen);
2303   print_signal_handler(st, SIGFPE , buf, buflen);
2304   print_signal_handler(st, SIGPIPE, buf, buflen);
2305   print_signal_handler(st, SIGXFSZ, buf, buflen);
2306   print_signal_handler(st, SIGILL , buf, buflen);
2307   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2308   print_signal_handler(st, SR_signum, buf, buflen);
2309   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2310   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2311   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2312   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2313 }
2314 
2315 static char saved_jvm_path[MAXPATHLEN] = {0};
2316 
2317 // Find the full path to the current module, libjvm.so
2318 void os::jvm_path(char *buf, jint buflen) {
2319   // Error checking.
2320   if (buflen < MAXPATHLEN) {
2321     assert(false, "must use a large-enough buffer");
2322     buf[0] = '\0';
2323     return;
2324   }
2325   // Lazy resolve the path to current module.
2326   if (saved_jvm_path[0] != 0) {
2327     strcpy(buf, saved_jvm_path);
2328     return;
2329   }
2330 
2331   char dli_fname[MAXPATHLEN];
2332   bool ret = dll_address_to_library_name(
2333                 CAST_FROM_FN_PTR(address, os::jvm_path),
2334                 dli_fname, sizeof(dli_fname), NULL);
2335   assert(ret, "cannot locate libjvm");
2336   char *rp = NULL;
2337   if (ret && dli_fname[0] != '\0') {
2338     rp = realpath(dli_fname, buf);
2339   }
2340   if (rp == NULL)
2341     return;
2342 
2343   if (Arguments::created_by_gamma_launcher()) {
2344     // Support for the gamma launcher.  Typical value for buf is
2345     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2346     // the right place in the string, then assume we are installed in a JDK and
2347     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2348     // up the path so it looks like libjvm.so is installed there (append a
2349     // fake suffix hotspot/libjvm.so).
2350     const char *p = buf + strlen(buf) - 1;
2351     for (int count = 0; p > buf && count < 5; ++count) {
2352       for (--p; p > buf && *p != '/'; --p)
2353         /* empty */ ;
2354     }
2355 
2356     if (strncmp(p, "/jre/lib/", 9) != 0) {
2357       // Look for JAVA_HOME in the environment.
2358       char* java_home_var = ::getenv("JAVA_HOME");
2359       if (java_home_var != NULL && java_home_var[0] != 0) {
2360         char* jrelib_p;
2361         int len;
2362 
2363         // Check the current module name "libjvm.so".
2364         p = strrchr(buf, '/');
2365         assert(strstr(p, "/libjvm") == p, "invalid library name");
2366 
2367         rp = realpath(java_home_var, buf);
2368         if (rp == NULL)
2369           return;
2370 
2371         // determine if this is a legacy image or modules image
2372         // modules image doesn't have "jre" subdirectory
2373         len = strlen(buf);
2374         jrelib_p = buf + len;
2375         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2376         if (0 != access(buf, F_OK)) {
2377           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2378         }
2379 
2380         if (0 == access(buf, F_OK)) {
2381           // Use current module name "libjvm.so"
2382           len = strlen(buf);
2383           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2384         } else {
2385           // Go back to path of .so
2386           rp = realpath(dli_fname, buf);
2387           if (rp == NULL)
2388             return;
2389         }
2390       }
2391     }
2392   }
2393 
2394   strcpy(saved_jvm_path, buf);
2395 }
2396 
2397 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2398   // no prefix required, not even "_"
2399 }
2400 
2401 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2402   // no suffix required
2403 }
2404 
2405 ////////////////////////////////////////////////////////////////////////////////
2406 // sun.misc.Signal support
2407 
2408 static volatile jint sigint_count = 0;
2409 
2410 static void
2411 UserHandler(int sig, void *siginfo, void *context) {
2412   // 4511530 - sem_post is serialized and handled by the manager thread. When
2413   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2414   // don't want to flood the manager thread with sem_post requests.
2415   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2416       return;
2417 
2418   // Ctrl-C is pressed during error reporting, likely because the error
2419   // handler fails to abort. Let VM die immediately.
2420   if (sig == SIGINT && is_error_reported()) {
2421      os::die();
2422   }
2423 
2424   os::signal_notify(sig);
2425 }
2426 
2427 void* os::user_handler() {
2428   return CAST_FROM_FN_PTR(void*, UserHandler);
2429 }
2430 
2431 class Semaphore : public StackObj {
2432   public:
2433     Semaphore();
2434     ~Semaphore();
2435     void signal();
2436     void wait();
2437     bool trywait();
2438     bool timedwait(unsigned int sec, int nsec);
2439   private:
2440     sem_t _semaphore;
2441 };
2442 
2443 
2444 Semaphore::Semaphore() {
2445   sem_init(&_semaphore, 0, 0);
2446 }
2447 
2448 Semaphore::~Semaphore() {
2449   sem_destroy(&_semaphore);
2450 }
2451 
2452 void Semaphore::signal() {
2453   sem_post(&_semaphore);
2454 }
2455 
2456 void Semaphore::wait() {
2457   sem_wait(&_semaphore);
2458 }
2459 
2460 bool Semaphore::trywait() {
2461   return sem_trywait(&_semaphore) == 0;
2462 }
2463 
2464 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2465   struct timespec ts;
2466   unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2467 
2468   while (1) {
2469     int result = sem_timedwait(&_semaphore, &ts);
2470     if (result == 0) {
2471       return true;
2472     } else if (errno == EINTR) {
2473       continue;
2474     } else if (errno == ETIMEDOUT) {
2475       return false;
2476     } else {
2477       return false;
2478     }
2479   }
2480 }
2481 
2482 extern "C" {
2483   typedef void (*sa_handler_t)(int);
2484   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2485 }
2486 
2487 void* os::signal(int signal_number, void* handler) {
2488   struct sigaction sigAct, oldSigAct;
2489 
2490   sigfillset(&(sigAct.sa_mask));
2491   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2492   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2493 
2494   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2495     // -1 means registration failed
2496     return (void *)-1;
2497   }
2498 
2499   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2500 }
2501 
2502 void os::signal_raise(int signal_number) {
2503   ::raise(signal_number);
2504 }
2505 
2506 /*
2507  * The following code is moved from os.cpp for making this
2508  * code platform specific, which it is by its very nature.
2509  */
2510 
2511 // Will be modified when max signal is changed to be dynamic
2512 int os::sigexitnum_pd() {
2513   return NSIG;
2514 }
2515 
2516 // a counter for each possible signal value
2517 static volatile jint pending_signals[NSIG+1] = { 0 };
2518 
2519 // Linux(POSIX) specific hand shaking semaphore.
2520 static sem_t sig_sem;
2521 static Semaphore sr_semaphore;
2522 
2523 void os::signal_init_pd() {
2524   // Initialize signal structures
2525   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2526 
2527   // Initialize signal semaphore
2528   ::sem_init(&sig_sem, 0, 0);
2529 }
2530 
2531 void os::signal_notify(int sig) {
2532   Atomic::inc(&pending_signals[sig]);
2533   ::sem_post(&sig_sem);
2534 }
2535 
2536 static int check_pending_signals(bool wait) {
2537   Atomic::store(0, &sigint_count);
2538   for (;;) {
2539     for (int i = 0; i < NSIG + 1; i++) {
2540       jint n = pending_signals[i];
2541       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2542         return i;
2543       }
2544     }
2545     if (!wait) {
2546       return -1;
2547     }
2548     JavaThread *thread = JavaThread::current();
2549     ThreadBlockInVM tbivm(thread);
2550 
2551     bool threadIsSuspended;
2552     do {
2553       thread->set_suspend_equivalent();
2554       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2555       ::sem_wait(&sig_sem);
2556 
2557       // were we externally suspended while we were waiting?
2558       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2559       if (threadIsSuspended) {
2560         //
2561         // The semaphore has been incremented, but while we were waiting
2562         // another thread suspended us. We don't want to continue running
2563         // while suspended because that would surprise the thread that
2564         // suspended us.
2565         //
2566         ::sem_post(&sig_sem);
2567 
2568         thread->java_suspend_self();
2569       }
2570     } while (threadIsSuspended);
2571   }
2572 }
2573 
2574 int os::signal_lookup() {
2575   return check_pending_signals(false);
2576 }
2577 
2578 int os::signal_wait() {
2579   return check_pending_signals(true);
2580 }
2581 
2582 ////////////////////////////////////////////////////////////////////////////////
2583 // Virtual Memory
2584 
2585 int os::vm_page_size() {
2586   // Seems redundant as all get out
2587   assert(os::Linux::page_size() != -1, "must call os::init");
2588   return os::Linux::page_size();
2589 }
2590 
2591 // Solaris allocates memory by pages.
2592 int os::vm_allocation_granularity() {
2593   assert(os::Linux::page_size() != -1, "must call os::init");
2594   return os::Linux::page_size();
2595 }
2596 
2597 // Rationale behind this function:
2598 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2599 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2600 //  samples for JITted code. Here we create private executable mapping over the code cache
2601 //  and then we can use standard (well, almost, as mapping can change) way to provide
2602 //  info for the reporting script by storing timestamp and location of symbol
2603 void linux_wrap_code(char* base, size_t size) {
2604   static volatile jint cnt = 0;
2605 
2606   if (!UseOprofile) {
2607     return;
2608   }
2609 
2610   char buf[PATH_MAX+1];
2611   int num = Atomic::add(1, &cnt);
2612 
2613   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2614            os::get_temp_directory(), os::current_process_id(), num);
2615   unlink(buf);
2616 
2617   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2618 
2619   if (fd != -1) {
2620     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2621     if (rv != (off_t)-1) {
2622       if (::write(fd, "", 1) == 1) {
2623         mmap(base, size,
2624              PROT_READ|PROT_WRITE|PROT_EXEC,
2625              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2626       }
2627     }
2628     ::close(fd);
2629     unlink(buf);
2630   }
2631 }
2632 
2633 static bool recoverable_mmap_error(int err) {
2634   // See if the error is one we can let the caller handle. This
2635   // list of errno values comes from JBS-6843484. I can't find a
2636   // Linux man page that documents this specific set of errno
2637   // values so while this list currently matches Solaris, it may
2638   // change as we gain experience with this failure mode.
2639   switch (err) {
2640   case EBADF:
2641   case EINVAL:
2642   case ENOTSUP:
2643     // let the caller deal with these errors
2644     return true;
2645 
2646   default:
2647     // Any remaining errors on this OS can cause our reserved mapping
2648     // to be lost. That can cause confusion where different data
2649     // structures think they have the same memory mapped. The worst
2650     // scenario is if both the VM and a library think they have the
2651     // same memory mapped.
2652     return false;
2653   }
2654 }
2655 
2656 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2657                                     int err) {
2658   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2659           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2660           strerror(err), err);
2661 }
2662 
2663 static void warn_fail_commit_memory(char* addr, size_t size,
2664                                     size_t alignment_hint, bool exec,
2665                                     int err) {
2666   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2667           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2668           alignment_hint, exec, strerror(err), err);
2669 }
2670 
2671 // NOTE: Linux kernel does not really reserve the pages for us.
2672 //       All it does is to check if there are enough free pages
2673 //       left at the time of mmap(). This could be a potential
2674 //       problem.
2675 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2676   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2677   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2678                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2679   if (res != (uintptr_t) MAP_FAILED) {
2680     if (UseNUMAInterleaving) {
2681       numa_make_global(addr, size);
2682     }
2683     return 0;
2684   }
2685 
2686   int err = errno;  // save errno from mmap() call above
2687 
2688   if (!recoverable_mmap_error(err)) {
2689     warn_fail_commit_memory(addr, size, exec, err);
2690     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2691   }
2692 
2693   return err;
2694 }
2695 
2696 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2697   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2698 }
2699 
2700 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2701                                   const char* mesg) {
2702   assert(mesg != NULL, "mesg must be specified");
2703   int err = os::Linux::commit_memory_impl(addr, size, exec);
2704   if (err != 0) {
2705     // the caller wants all commit errors to exit with the specified mesg:
2706     warn_fail_commit_memory(addr, size, exec, err);
2707     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2708   }
2709 }
2710 
2711 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2712 #ifndef MAP_HUGETLB
2713 #define MAP_HUGETLB 0x40000
2714 #endif
2715 
2716 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2717 #ifndef MADV_HUGEPAGE
2718 #define MADV_HUGEPAGE 14
2719 #endif
2720 
2721 int os::Linux::commit_memory_impl(char* addr, size_t size,
2722                                   size_t alignment_hint, bool exec) {
2723   int err;
2724   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2725     int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2726     uintptr_t res =
2727       (uintptr_t) ::mmap(addr, size, prot,
2728                          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2729                          -1, 0);
2730     if (res != (uintptr_t) MAP_FAILED) {
2731       if (UseNUMAInterleaving) {
2732         numa_make_global(addr, size);
2733       }
2734       return 0;
2735     }
2736 
2737     err = errno;  // save errno from mmap() call above
2738 
2739     if (!recoverable_mmap_error(err)) {
2740       // However, it is not clear that this loss of our reserved mapping
2741       // happens with large pages on Linux or that we cannot recover
2742       // from the loss. For now, we just issue a warning and we don't
2743       // call vm_exit_out_of_memory(). This issue is being tracked by
2744       // JBS-8007074.
2745       warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2746 //    vm_exit_out_of_memory(size, OOM_MMAP_ERROR,
2747 //                          "committing reserved memory.");
2748     }
2749     // Fall through and try to use small pages
2750   }
2751 
2752   err = os::Linux::commit_memory_impl(addr, size, exec);
2753   if (err == 0) {
2754     realign_memory(addr, size, alignment_hint);
2755   }
2756   return err;
2757 }
2758 
2759 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2760                           bool exec) {
2761   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2762 }
2763 
2764 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2765                                   size_t alignment_hint, bool exec,
2766                                   const char* mesg) {
2767   assert(mesg != NULL, "mesg must be specified");
2768   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2769   if (err != 0) {
2770     // the caller wants all commit errors to exit with the specified mesg:
2771     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2772     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2773   }
2774 }
2775 
2776 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2777   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2778     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2779     // be supported or the memory may already be backed by huge pages.
2780     ::madvise(addr, bytes, MADV_HUGEPAGE);
2781   }
2782 }
2783 
2784 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2785   // This method works by doing an mmap over an existing mmaping and effectively discarding
2786   // the existing pages. However it won't work for SHM-based large pages that cannot be
2787   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2788   // small pages on top of the SHM segment. This method always works for small pages, so we
2789   // allow that in any case.
2790   if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2791     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2792   }
2793 }
2794 
2795 void os::numa_make_global(char *addr, size_t bytes) {
2796   Linux::numa_interleave_memory(addr, bytes);
2797 }
2798 
2799 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2800   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2801 }
2802 
2803 bool os::numa_topology_changed()   { return false; }
2804 
2805 size_t os::numa_get_groups_num() {
2806   int max_node = Linux::numa_max_node();
2807   return max_node > 0 ? max_node + 1 : 1;
2808 }
2809 
2810 int os::numa_get_group_id() {
2811   int cpu_id = Linux::sched_getcpu();
2812   if (cpu_id != -1) {
2813     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2814     if (lgrp_id != -1) {
2815       return lgrp_id;
2816     }
2817   }
2818   return 0;
2819 }
2820 
2821 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2822   for (size_t i = 0; i < size; i++) {
2823     ids[i] = i;
2824   }
2825   return size;
2826 }
2827 
2828 bool os::get_page_info(char *start, page_info* info) {
2829   return false;
2830 }
2831 
2832 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2833   return end;
2834 }
2835 
2836 
2837 int os::Linux::sched_getcpu_syscall(void) {
2838   unsigned int cpu;
2839   int retval = -1;
2840 
2841 #if defined(IA32)
2842 # ifndef SYS_getcpu
2843 # define SYS_getcpu 318
2844 # endif
2845   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2846 #elif defined(AMD64)
2847 // Unfortunately we have to bring all these macros here from vsyscall.h
2848 // to be able to compile on old linuxes.
2849 # define __NR_vgetcpu 2
2850 # define VSYSCALL_START (-10UL << 20)
2851 # define VSYSCALL_SIZE 1024
2852 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2853   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2854   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2855   retval = vgetcpu(&cpu, NULL, NULL);
2856 #endif
2857 
2858   return (retval == -1) ? retval : cpu;
2859 }
2860 
2861 // Something to do with the numa-aware allocator needs these symbols
2862 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2863 extern "C" JNIEXPORT void numa_error(char *where) { }
2864 extern "C" JNIEXPORT int fork1() { return fork(); }
2865 
2866 
2867 // If we are running with libnuma version > 2, then we should
2868 // be trying to use symbols with versions 1.1
2869 // If we are running with earlier version, which did not have symbol versions,
2870 // we should use the base version.
2871 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2872   void *f = dlvsym(handle, name, "libnuma_1.1");
2873   if (f == NULL) {
2874     f = dlsym(handle, name);
2875   }
2876   return f;
2877 }
2878 
2879 bool os::Linux::libnuma_init() {
2880   // sched_getcpu() should be in libc.
2881   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2882                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2883 
2884   // If it's not, try a direct syscall.
2885   if (sched_getcpu() == -1)
2886     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2887 
2888   if (sched_getcpu() != -1) { // Does it work?
2889     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2890     if (handle != NULL) {
2891       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2892                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2893       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2894                                        libnuma_dlsym(handle, "numa_max_node")));
2895       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2896                                         libnuma_dlsym(handle, "numa_available")));
2897       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2898                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2899       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2900                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2901 
2902 
2903       if (numa_available() != -1) {
2904         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2905         // Create a cpu -> node mapping
2906         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2907         rebuild_cpu_to_node_map();
2908         return true;
2909       }
2910     }
2911   }
2912   return false;
2913 }
2914 
2915 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2916 // The table is later used in get_node_by_cpu().
2917 void os::Linux::rebuild_cpu_to_node_map() {
2918   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2919                               // in libnuma (possible values are starting from 16,
2920                               // and continuing up with every other power of 2, but less
2921                               // than the maximum number of CPUs supported by kernel), and
2922                               // is a subject to change (in libnuma version 2 the requirements
2923                               // are more reasonable) we'll just hardcode the number they use
2924                               // in the library.
2925   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2926 
2927   size_t cpu_num = os::active_processor_count();
2928   size_t cpu_map_size = NCPUS / BitsPerCLong;
2929   size_t cpu_map_valid_size =
2930     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2931 
2932   cpu_to_node()->clear();
2933   cpu_to_node()->at_grow(cpu_num - 1);
2934   size_t node_num = numa_get_groups_num();
2935 
2936   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2937   for (size_t i = 0; i < node_num; i++) {
2938     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2939       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2940         if (cpu_map[j] != 0) {
2941           for (size_t k = 0; k < BitsPerCLong; k++) {
2942             if (cpu_map[j] & (1UL << k)) {
2943               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2944             }
2945           }
2946         }
2947       }
2948     }
2949   }
2950   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2951 }
2952 
2953 int os::Linux::get_node_by_cpu(int cpu_id) {
2954   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2955     return cpu_to_node()->at(cpu_id);
2956   }
2957   return -1;
2958 }
2959 
2960 GrowableArray<int>* os::Linux::_cpu_to_node;
2961 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2962 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2963 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2964 os::Linux::numa_available_func_t os::Linux::_numa_available;
2965 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2966 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2967 unsigned long* os::Linux::_numa_all_nodes;
2968 
2969 bool os::pd_uncommit_memory(char* addr, size_t size) {
2970   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2971                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2972   return res  != (uintptr_t) MAP_FAILED;
2973 }
2974 
2975 // Linux uses a growable mapping for the stack, and if the mapping for
2976 // the stack guard pages is not removed when we detach a thread the
2977 // stack cannot grow beyond the pages where the stack guard was
2978 // mapped.  If at some point later in the process the stack expands to
2979 // that point, the Linux kernel cannot expand the stack any further
2980 // because the guard pages are in the way, and a segfault occurs.
2981 //
2982 // However, it's essential not to split the stack region by unmapping
2983 // a region (leaving a hole) that's already part of the stack mapping,
2984 // so if the stack mapping has already grown beyond the guard pages at
2985 // the time we create them, we have to truncate the stack mapping.
2986 // So, we need to know the extent of the stack mapping when
2987 // create_stack_guard_pages() is called.
2988 
2989 // Find the bounds of the stack mapping.  Return true for success.
2990 //
2991 // We only need this for stacks that are growable: at the time of
2992 // writing thread stacks don't use growable mappings (i.e. those
2993 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2994 // only applies to the main thread.
2995 
2996 static
2997 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2998 
2999   char buf[128];
3000   int fd, sz;
3001 
3002   if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
3003     return false;
3004   }
3005 
3006   const char kw[] = "[stack]";
3007   const int kwlen = sizeof(kw)-1;
3008 
3009   // Address part of /proc/self/maps couldn't be more than 128 bytes
3010   while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
3011      if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
3012         // Extract addresses
3013         if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
3014            uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
3015            if (sp >= *bottom && sp <= *top) {
3016               ::close(fd);
3017               return true;
3018            }
3019         }
3020      }
3021   }
3022 
3023  ::close(fd);
3024   return false;
3025 }
3026 
3027 
3028 // If the (growable) stack mapping already extends beyond the point
3029 // where we're going to put our guard pages, truncate the mapping at
3030 // that point by munmap()ping it.  This ensures that when we later
3031 // munmap() the guard pages we don't leave a hole in the stack
3032 // mapping. This only affects the main/initial thread, but guard
3033 // against future OS changes
3034 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3035   uintptr_t stack_extent, stack_base;
3036   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
3037   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
3038       assert(os::Linux::is_initial_thread(),
3039            "growable stack in non-initial thread");
3040     if (stack_extent < (uintptr_t)addr)
3041       ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
3042   }
3043 
3044   return os::commit_memory(addr, size, !ExecMem);
3045 }
3046 
3047 // If this is a growable mapping, remove the guard pages entirely by
3048 // munmap()ping them.  If not, just call uncommit_memory(). This only
3049 // affects the main/initial thread, but guard against future OS changes
3050 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3051   uintptr_t stack_extent, stack_base;
3052   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
3053   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
3054       assert(os::Linux::is_initial_thread(),
3055            "growable stack in non-initial thread");
3056 
3057     return ::munmap(addr, size) == 0;
3058   }
3059 
3060   return os::uncommit_memory(addr, size);
3061 }
3062 
3063 static address _highest_vm_reserved_address = NULL;
3064 
3065 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3066 // at 'requested_addr'. If there are existing memory mappings at the same
3067 // location, however, they will be overwritten. If 'fixed' is false,
3068 // 'requested_addr' is only treated as a hint, the return value may or
3069 // may not start from the requested address. Unlike Linux mmap(), this
3070 // function returns NULL to indicate failure.
3071 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3072   char * addr;
3073   int flags;
3074 
3075   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3076   if (fixed) {
3077     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3078     flags |= MAP_FIXED;
3079   }
3080 
3081   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3082   // touch an uncommitted page. Otherwise, the read/write might
3083   // succeed if we have enough swap space to back the physical page.
3084   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3085                        flags, -1, 0);
3086 
3087   if (addr != MAP_FAILED) {
3088     // anon_mmap() should only get called during VM initialization,
3089     // don't need lock (actually we can skip locking even it can be called
3090     // from multiple threads, because _highest_vm_reserved_address is just a
3091     // hint about the upper limit of non-stack memory regions.)
3092     if ((address)addr + bytes > _highest_vm_reserved_address) {
3093       _highest_vm_reserved_address = (address)addr + bytes;
3094     }
3095   }
3096 
3097   return addr == MAP_FAILED ? NULL : addr;
3098 }
3099 
3100 // Don't update _highest_vm_reserved_address, because there might be memory
3101 // regions above addr + size. If so, releasing a memory region only creates
3102 // a hole in the address space, it doesn't help prevent heap-stack collision.
3103 //
3104 static int anon_munmap(char * addr, size_t size) {
3105   return ::munmap(addr, size) == 0;
3106 }
3107 
3108 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3109                          size_t alignment_hint) {
3110   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3111 }
3112 
3113 bool os::pd_release_memory(char* addr, size_t size) {
3114   return anon_munmap(addr, size);
3115 }
3116 
3117 static address highest_vm_reserved_address() {
3118   return _highest_vm_reserved_address;
3119 }
3120 
3121 static bool linux_mprotect(char* addr, size_t size, int prot) {
3122   // Linux wants the mprotect address argument to be page aligned.
3123   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3124 
3125   // According to SUSv3, mprotect() should only be used with mappings
3126   // established by mmap(), and mmap() always maps whole pages. Unaligned
3127   // 'addr' likely indicates problem in the VM (e.g. trying to change
3128   // protection of malloc'ed or statically allocated memory). Check the
3129   // caller if you hit this assert.
3130   assert(addr == bottom, "sanity check");
3131 
3132   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3133   return ::mprotect(bottom, size, prot) == 0;
3134 }
3135 
3136 // Set protections specified
3137 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3138                         bool is_committed) {
3139   unsigned int p = 0;
3140   switch (prot) {
3141   case MEM_PROT_NONE: p = PROT_NONE; break;
3142   case MEM_PROT_READ: p = PROT_READ; break;
3143   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3144   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3145   default:
3146     ShouldNotReachHere();
3147   }
3148   // is_committed is unused.
3149   return linux_mprotect(addr, bytes, p);
3150 }
3151 
3152 bool os::guard_memory(char* addr, size_t size) {
3153   return linux_mprotect(addr, size, PROT_NONE);
3154 }
3155 
3156 bool os::unguard_memory(char* addr, size_t size) {
3157   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3158 }
3159 
3160 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3161   bool result = false;
3162   void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
3163                   MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3164                   -1, 0);
3165 
3166   if (p != MAP_FAILED) {
3167     // We don't know if this really is a huge page or not.
3168     FILE *fp = fopen("/proc/self/maps", "r");
3169     if (fp) {
3170       while (!feof(fp)) {
3171         char chars[257];
3172         long x = 0;
3173         if (fgets(chars, sizeof(chars), fp)) {
3174           if (sscanf(chars, "%lx-%*x", &x) == 1
3175               && x == (long)p) {
3176             if (strstr (chars, "hugepage")) {
3177               result = true;
3178               break;
3179             }
3180           }
3181         }
3182       }
3183       fclose(fp);
3184     }
3185     munmap (p, page_size);
3186     if (result)
3187       return true;
3188   }
3189 
3190   if (warn) {
3191     warning("HugeTLBFS is not supported by the operating system.");
3192   }
3193 
3194   return result;
3195 }
3196 
3197 /*
3198 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3199 *
3200 * From the coredump_filter documentation:
3201 *
3202 * - (bit 0) anonymous private memory
3203 * - (bit 1) anonymous shared memory
3204 * - (bit 2) file-backed private memory
3205 * - (bit 3) file-backed shared memory
3206 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3207 *           effective only if the bit 2 is cleared)
3208 * - (bit 5) hugetlb private memory
3209 * - (bit 6) hugetlb shared memory
3210 */
3211 static void set_coredump_filter(void) {
3212   FILE *f;
3213   long cdm;
3214 
3215   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3216     return;
3217   }
3218 
3219   if (fscanf(f, "%lx", &cdm) != 1) {
3220     fclose(f);
3221     return;
3222   }
3223 
3224   rewind(f);
3225 
3226   if ((cdm & LARGEPAGES_BIT) == 0) {
3227     cdm |= LARGEPAGES_BIT;
3228     fprintf(f, "%#lx", cdm);
3229   }
3230 
3231   fclose(f);
3232 }
3233 
3234 // Large page support
3235 
3236 static size_t _large_page_size = 0;
3237 
3238 void os::large_page_init() {
3239   if (!UseLargePages) {
3240     UseHugeTLBFS = false;
3241     UseSHM = false;
3242     return;
3243   }
3244 
3245   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3246     // If UseLargePages is specified on the command line try both methods,
3247     // if it's default, then try only HugeTLBFS.
3248     if (FLAG_IS_DEFAULT(UseLargePages)) {
3249       UseHugeTLBFS = true;
3250     } else {
3251       UseHugeTLBFS = UseSHM = true;
3252     }
3253   }
3254 
3255   if (LargePageSizeInBytes) {
3256     _large_page_size = LargePageSizeInBytes;
3257   } else {
3258     // large_page_size on Linux is used to round up heap size. x86 uses either
3259     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3260     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3261     // page as large as 256M.
3262     //
3263     // Here we try to figure out page size by parsing /proc/meminfo and looking
3264     // for a line with the following format:
3265     //    Hugepagesize:     2048 kB
3266     //
3267     // If we can't determine the value (e.g. /proc is not mounted, or the text
3268     // format has been changed), we'll use the largest page size supported by
3269     // the processor.
3270 
3271 #ifndef ZERO
3272     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3273                        ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3274 #endif // ZERO
3275 
3276     FILE *fp = fopen("/proc/meminfo", "r");
3277     if (fp) {
3278       while (!feof(fp)) {
3279         int x = 0;
3280         char buf[16];
3281         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3282           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3283             _large_page_size = x * K;
3284             break;
3285           }
3286         } else {
3287           // skip to next line
3288           for (;;) {
3289             int ch = fgetc(fp);
3290             if (ch == EOF || ch == (int)'\n') break;
3291           }
3292         }
3293       }
3294       fclose(fp);
3295     }
3296   }
3297 
3298   // print a warning if any large page related flag is specified on command line
3299   bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3300 
3301   const size_t default_page_size = (size_t)Linux::page_size();
3302   if (_large_page_size > default_page_size) {
3303     _page_sizes[0] = _large_page_size;
3304     _page_sizes[1] = default_page_size;
3305     _page_sizes[2] = 0;
3306   }
3307   UseHugeTLBFS = UseHugeTLBFS &&
3308                  Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3309 
3310   if (UseHugeTLBFS)
3311     UseSHM = false;
3312 
3313   UseLargePages = UseHugeTLBFS || UseSHM;
3314 
3315   set_coredump_filter();
3316 }
3317 
3318 #ifndef SHM_HUGETLB
3319 #define SHM_HUGETLB 04000
3320 #endif
3321 
3322 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3323   // "exec" is passed in but not used.  Creating the shared image for
3324   // the code cache doesn't have an SHM_X executable permission to check.
3325   assert(UseLargePages && UseSHM, "only for SHM large pages");
3326 
3327   key_t key = IPC_PRIVATE;
3328   char *addr;
3329 
3330   bool warn_on_failure = UseLargePages &&
3331                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3332                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3333                         );
3334   char msg[128];
3335 
3336   // Create a large shared memory region to attach to based on size.
3337   // Currently, size is the total size of the heap
3338   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3339   if (shmid == -1) {
3340      // Possible reasons for shmget failure:
3341      // 1. shmmax is too small for Java heap.
3342      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3343      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3344      // 2. not enough large page memory.
3345      //    > check available large pages: cat /proc/meminfo
3346      //    > increase amount of large pages:
3347      //          echo new_value > /proc/sys/vm/nr_hugepages
3348      //      Note 1: different Linux may use different name for this property,
3349      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3350      //      Note 2: it's possible there's enough physical memory available but
3351      //            they are so fragmented after a long run that they can't
3352      //            coalesce into large pages. Try to reserve large pages when
3353      //            the system is still "fresh".
3354      if (warn_on_failure) {
3355        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3356        warning(msg);
3357      }
3358      return NULL;
3359   }
3360 
3361   // attach to the region
3362   addr = (char*)shmat(shmid, req_addr, 0);
3363   int err = errno;
3364 
3365   // Remove shmid. If shmat() is successful, the actual shared memory segment
3366   // will be deleted when it's detached by shmdt() or when the process
3367   // terminates. If shmat() is not successful this will remove the shared
3368   // segment immediately.
3369   shmctl(shmid, IPC_RMID, NULL);
3370 
3371   if ((intptr_t)addr == -1) {
3372      if (warn_on_failure) {
3373        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3374        warning(msg);
3375      }
3376      return NULL;
3377   }
3378 
3379   if ((addr != NULL) && UseNUMAInterleaving) {
3380     numa_make_global(addr, bytes);
3381   }
3382 
3383   // The memory is committed
3384   MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3385 
3386   return addr;
3387 }
3388 
3389 bool os::release_memory_special(char* base, size_t bytes) {
3390   MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3391   // detaching the SHM segment will also delete it, see reserve_memory_special()
3392   int rslt = shmdt(base);
3393   if (rslt == 0) {
3394     tkr.record((address)base, bytes);
3395     return true;
3396   } else {
3397     tkr.discard();
3398     return false;
3399   }
3400 }
3401 
3402 size_t os::large_page_size() {
3403   return _large_page_size;
3404 }
3405 
3406 // HugeTLBFS allows application to commit large page memory on demand;
3407 // with SysV SHM the entire memory region must be allocated as shared
3408 // memory.
3409 bool os::can_commit_large_page_memory() {
3410   return UseHugeTLBFS;
3411 }
3412 
3413 bool os::can_execute_large_page_memory() {
3414   return UseHugeTLBFS;
3415 }
3416 
3417 // Reserve memory at an arbitrary address, only if that area is
3418 // available (and not reserved for something else).
3419 
3420 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3421   const int max_tries = 10;
3422   char* base[max_tries];
3423   size_t size[max_tries];
3424   const size_t gap = 0x000000;
3425 
3426   // Assert only that the size is a multiple of the page size, since
3427   // that's all that mmap requires, and since that's all we really know
3428   // about at this low abstraction level.  If we need higher alignment,
3429   // we can either pass an alignment to this method or verify alignment
3430   // in one of the methods further up the call chain.  See bug 5044738.
3431   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3432 
3433   // Repeatedly allocate blocks until the block is allocated at the
3434   // right spot. Give up after max_tries. Note that reserve_memory() will
3435   // automatically update _highest_vm_reserved_address if the call is
3436   // successful. The variable tracks the highest memory address every reserved
3437   // by JVM. It is used to detect heap-stack collision if running with
3438   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3439   // space than needed, it could confuse the collision detecting code. To
3440   // solve the problem, save current _highest_vm_reserved_address and
3441   // calculate the correct value before return.
3442   address old_highest = _highest_vm_reserved_address;
3443 
3444   // Linux mmap allows caller to pass an address as hint; give it a try first,
3445   // if kernel honors the hint then we can return immediately.
3446   char * addr = anon_mmap(requested_addr, bytes, false);
3447   if (addr == requested_addr) {
3448      return requested_addr;
3449   }
3450 
3451   if (addr != NULL) {
3452      // mmap() is successful but it fails to reserve at the requested address
3453      anon_munmap(addr, bytes);
3454   }
3455 
3456   int i;
3457   for (i = 0; i < max_tries; ++i) {
3458     base[i] = reserve_memory(bytes);
3459 
3460     if (base[i] != NULL) {
3461       // Is this the block we wanted?
3462       if (base[i] == requested_addr) {
3463         size[i] = bytes;
3464         break;
3465       }
3466 
3467       // Does this overlap the block we wanted? Give back the overlapped
3468       // parts and try again.
3469 
3470       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3471       if (top_overlap >= 0 && top_overlap < bytes) {
3472         unmap_memory(base[i], top_overlap);
3473         base[i] += top_overlap;
3474         size[i] = bytes - top_overlap;
3475       } else {
3476         size_t bottom_overlap = base[i] + bytes - requested_addr;
3477         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3478           unmap_memory(requested_addr, bottom_overlap);
3479           size[i] = bytes - bottom_overlap;
3480         } else {
3481           size[i] = bytes;
3482         }
3483       }
3484     }
3485   }
3486 
3487   // Give back the unused reserved pieces.
3488 
3489   for (int j = 0; j < i; ++j) {
3490     if (base[j] != NULL) {
3491       unmap_memory(base[j], size[j]);
3492     }
3493   }
3494 
3495   if (i < max_tries) {
3496     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3497     return requested_addr;
3498   } else {
3499     _highest_vm_reserved_address = old_highest;
3500     return NULL;
3501   }
3502 }
3503 
3504 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3505   return ::read(fd, buf, nBytes);
3506 }
3507 
3508 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3509 // Solaris uses poll(), linux uses park().
3510 // Poll() is likely a better choice, assuming that Thread.interrupt()
3511 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3512 // SIGSEGV, see 4355769.
3513 
3514 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3515   assert(thread == Thread::current(),  "thread consistency check");
3516 
3517   ParkEvent * const slp = thread->_SleepEvent ;
3518   slp->reset() ;
3519   OrderAccess::fence() ;
3520 
3521   if (interruptible) {
3522     jlong prevtime = javaTimeNanos();
3523 
3524     for (;;) {
3525       if (os::is_interrupted(thread, true)) {
3526         return OS_INTRPT;
3527       }
3528 
3529       jlong newtime = javaTimeNanos();
3530 
3531       if (newtime - prevtime < 0) {
3532         // time moving backwards, should only happen if no monotonic clock
3533         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3534         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3535       } else {
3536         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3537       }
3538 
3539       if(millis <= 0) {
3540         return OS_OK;
3541       }
3542 
3543       prevtime = newtime;
3544 
3545       {
3546         assert(thread->is_Java_thread(), "sanity check");
3547         JavaThread *jt = (JavaThread *) thread;
3548         ThreadBlockInVM tbivm(jt);
3549         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3550 
3551         jt->set_suspend_equivalent();
3552         // cleared by handle_special_suspend_equivalent_condition() or
3553         // java_suspend_self() via check_and_wait_while_suspended()
3554 
3555         slp->park(millis);
3556 
3557         // were we externally suspended while we were waiting?
3558         jt->check_and_wait_while_suspended();
3559       }
3560     }
3561   } else {
3562     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3563     jlong prevtime = javaTimeNanos();
3564 
3565     for (;;) {
3566       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3567       // the 1st iteration ...
3568       jlong newtime = javaTimeNanos();
3569 
3570       if (newtime - prevtime < 0) {
3571         // time moving backwards, should only happen if no monotonic clock
3572         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3573         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3574       } else {
3575         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3576       }
3577 
3578       if(millis <= 0) break ;
3579 
3580       prevtime = newtime;
3581       slp->park(millis);
3582     }
3583     return OS_OK ;
3584   }
3585 }
3586 
3587 int os::naked_sleep() {
3588   // %% make the sleep time an integer flag. for now use 1 millisec.
3589   return os::sleep(Thread::current(), 1, false);
3590 }
3591 
3592 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3593 void os::infinite_sleep() {
3594   while (true) {    // sleep forever ...
3595     ::sleep(100);   // ... 100 seconds at a time
3596   }
3597 }
3598 
3599 // Used to convert frequent JVM_Yield() to nops
3600 bool os::dont_yield() {
3601   return DontYieldALot;
3602 }
3603 
3604 void os::yield() {
3605   sched_yield();
3606 }
3607 
3608 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3609 
3610 void os::yield_all(int attempts) {
3611   // Yields to all threads, including threads with lower priorities
3612   // Threads on Linux are all with same priority. The Solaris style
3613   // os::yield_all() with nanosleep(1ms) is not necessary.
3614   sched_yield();
3615 }
3616 
3617 // Called from the tight loops to possibly influence time-sharing heuristics
3618 void os::loop_breaker(int attempts) {
3619   os::yield_all(attempts);
3620 }
3621 
3622 ////////////////////////////////////////////////////////////////////////////////
3623 // thread priority support
3624 
3625 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3626 // only supports dynamic priority, static priority must be zero. For real-time
3627 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3628 // However, for large multi-threaded applications, SCHED_RR is not only slower
3629 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3630 // of 5 runs - Sep 2005).
3631 //
3632 // The following code actually changes the niceness of kernel-thread/LWP. It
3633 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3634 // not the entire user process, and user level threads are 1:1 mapped to kernel
3635 // threads. It has always been the case, but could change in the future. For
3636 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3637 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3638 
3639 int os::java_to_os_priority[CriticalPriority + 1] = {
3640   19,              // 0 Entry should never be used
3641 
3642    4,              // 1 MinPriority
3643    3,              // 2
3644    2,              // 3
3645 
3646    1,              // 4
3647    0,              // 5 NormPriority
3648   -1,              // 6
3649 
3650   -2,              // 7
3651   -3,              // 8
3652   -4,              // 9 NearMaxPriority
3653 
3654   -5,              // 10 MaxPriority
3655 
3656   -5               // 11 CriticalPriority
3657 };
3658 
3659 static int prio_init() {
3660   if (ThreadPriorityPolicy == 1) {
3661     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3662     // if effective uid is not root. Perhaps, a more elegant way of doing
3663     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3664     if (geteuid() != 0) {
3665       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3666         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3667       }
3668       ThreadPriorityPolicy = 0;
3669     }
3670   }
3671   if (UseCriticalJavaThreadPriority) {
3672     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3673   }
3674   return 0;
3675 }
3676 
3677 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3678   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3679 
3680   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3681   return (ret == 0) ? OS_OK : OS_ERR;
3682 }
3683 
3684 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3685   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3686     *priority_ptr = java_to_os_priority[NormPriority];
3687     return OS_OK;
3688   }
3689 
3690   errno = 0;
3691   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3692   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3693 }
3694 
3695 // Hint to the underlying OS that a task switch would not be good.
3696 // Void return because it's a hint and can fail.
3697 void os::hint_no_preempt() {}
3698 
3699 ////////////////////////////////////////////////////////////////////////////////
3700 // suspend/resume support
3701 
3702 //  the low-level signal-based suspend/resume support is a remnant from the
3703 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3704 //  within hotspot. Now there is a single use-case for this:
3705 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3706 //      that runs in the watcher thread.
3707 //  The remaining code is greatly simplified from the more general suspension
3708 //  code that used to be used.
3709 //
3710 //  The protocol is quite simple:
3711 //  - suspend:
3712 //      - sends a signal to the target thread
3713 //      - polls the suspend state of the osthread using a yield loop
3714 //      - target thread signal handler (SR_handler) sets suspend state
3715 //        and blocks in sigsuspend until continued
3716 //  - resume:
3717 //      - sets target osthread state to continue
3718 //      - sends signal to end the sigsuspend loop in the SR_handler
3719 //
3720 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3721 //
3722 
3723 static void resume_clear_context(OSThread *osthread) {
3724   osthread->set_ucontext(NULL);
3725   osthread->set_siginfo(NULL);
3726 }
3727 
3728 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3729   osthread->set_ucontext(context);
3730   osthread->set_siginfo(siginfo);
3731 }
3732 
3733 //
3734 // Handler function invoked when a thread's execution is suspended or
3735 // resumed. We have to be careful that only async-safe functions are
3736 // called here (Note: most pthread functions are not async safe and
3737 // should be avoided.)
3738 //
3739 // Note: sigwait() is a more natural fit than sigsuspend() from an
3740 // interface point of view, but sigwait() prevents the signal hander
3741 // from being run. libpthread would get very confused by not having
3742 // its signal handlers run and prevents sigwait()'s use with the
3743 // mutex granting granting signal.
3744 //
3745 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3746 //
3747 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3748   // Save and restore errno to avoid confusing native code with EINTR
3749   // after sigsuspend.
3750   int old_errno = errno;
3751 
3752   Thread* thread = Thread::current();
3753   OSThread* osthread = thread->osthread();
3754   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3755 
3756   os::SuspendResume::State current = osthread->sr.state();
3757   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3758     suspend_save_context(osthread, siginfo, context);
3759 
3760     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3761     os::SuspendResume::State state = osthread->sr.suspended();
3762     if (state == os::SuspendResume::SR_SUSPENDED) {
3763       sigset_t suspend_set;  // signals for sigsuspend()
3764 
3765       // get current set of blocked signals and unblock resume signal
3766       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3767       sigdelset(&suspend_set, SR_signum);
3768 
3769       sr_semaphore.signal();
3770       // wait here until we are resumed
3771       while (1) {
3772         sigsuspend(&suspend_set);
3773 
3774         os::SuspendResume::State result = osthread->sr.running();
3775         if (result == os::SuspendResume::SR_RUNNING) {
3776           sr_semaphore.signal();
3777           break;
3778         }
3779       }
3780 
3781     } else if (state == os::SuspendResume::SR_RUNNING) {
3782       // request was cancelled, continue
3783     } else {
3784       ShouldNotReachHere();
3785     }
3786 
3787     resume_clear_context(osthread);
3788   } else if (current == os::SuspendResume::SR_RUNNING) {
3789     // request was cancelled, continue
3790   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3791     // ignore
3792   } else {
3793     // ignore
3794   }
3795 
3796   errno = old_errno;
3797 }
3798 
3799 
3800 static int SR_initialize() {
3801   struct sigaction act;
3802   char *s;
3803   /* Get signal number to use for suspend/resume */
3804   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3805     int sig = ::strtol(s, 0, 10);
3806     if (sig > 0 || sig < _NSIG) {
3807         SR_signum = sig;
3808     }
3809   }
3810 
3811   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3812         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3813 
3814   sigemptyset(&SR_sigset);
3815   sigaddset(&SR_sigset, SR_signum);
3816 
3817   /* Set up signal handler for suspend/resume */
3818   act.sa_flags = SA_RESTART|SA_SIGINFO;
3819   act.sa_handler = (void (*)(int)) SR_handler;
3820 
3821   // SR_signum is blocked by default.
3822   // 4528190 - We also need to block pthread restart signal (32 on all
3823   // supported Linux platforms). Note that LinuxThreads need to block
3824   // this signal for all threads to work properly. So we don't have
3825   // to use hard-coded signal number when setting up the mask.
3826   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3827 
3828   if (sigaction(SR_signum, &act, 0) == -1) {
3829     return -1;
3830   }
3831 
3832   // Save signal flag
3833   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3834   return 0;
3835 }
3836 
3837 static int sr_notify(OSThread* osthread) {
3838   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3839   assert_status(status == 0, status, "pthread_kill");
3840   return status;
3841 }
3842 
3843 // "Randomly" selected value for how long we want to spin
3844 // before bailing out on suspending a thread, also how often
3845 // we send a signal to a thread we want to resume
3846 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3847 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3848 
3849 // returns true on success and false on error - really an error is fatal
3850 // but this seems the normal response to library errors
3851 static bool do_suspend(OSThread* osthread) {
3852   assert(osthread->sr.is_running(), "thread should be running");
3853   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3854 
3855   // mark as suspended and send signal
3856   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3857     // failed to switch, state wasn't running?
3858     ShouldNotReachHere();
3859     return false;
3860   }
3861 
3862   if (sr_notify(osthread) != 0) {
3863     ShouldNotReachHere();
3864   }
3865 
3866   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3867   while (true) {
3868     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3869       break;
3870     } else {
3871       // timeout
3872       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3873       if (cancelled == os::SuspendResume::SR_RUNNING) {
3874         return false;
3875       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3876         // make sure that we consume the signal on the semaphore as well
3877         sr_semaphore.wait();
3878         break;
3879       } else {
3880         ShouldNotReachHere();
3881         return false;
3882       }
3883     }
3884   }
3885 
3886   guarantee(osthread->sr.is_suspended(), "Must be suspended");
3887   return true;
3888 }
3889 
3890 static void do_resume(OSThread* osthread) {
3891   assert(osthread->sr.is_suspended(), "thread should be suspended");
3892   assert(!sr_semaphore.trywait(), "invalid semaphore state");
3893 
3894   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3895     // failed to switch to WAKEUP_REQUEST
3896     ShouldNotReachHere();
3897     return;
3898   }
3899 
3900   while (true) {
3901     if (sr_notify(osthread) == 0) {
3902       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3903         if (osthread->sr.is_running()) {
3904           return;
3905         }
3906       }
3907     } else {
3908       ShouldNotReachHere();
3909     }
3910   }
3911 
3912   guarantee(osthread->sr.is_running(), "Must be running!");
3913 }
3914 
3915 ////////////////////////////////////////////////////////////////////////////////
3916 // interrupt support
3917 
3918 void os::interrupt(Thread* thread) {
3919   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3920     "possibility of dangling Thread pointer");
3921 
3922   OSThread* osthread = thread->osthread();
3923 
3924   if (!osthread->interrupted()) {
3925     osthread->set_interrupted(true);
3926     // More than one thread can get here with the same value of osthread,
3927     // resulting in multiple notifications.  We do, however, want the store
3928     // to interrupted() to be visible to other threads before we execute unpark().
3929     OrderAccess::fence();
3930     ParkEvent * const slp = thread->_SleepEvent ;
3931     if (slp != NULL) slp->unpark() ;
3932   }
3933 
3934   // For JSR166. Unpark even if interrupt status already was set
3935   if (thread->is_Java_thread())
3936     ((JavaThread*)thread)->parker()->unpark();
3937 
3938   ParkEvent * ev = thread->_ParkEvent ;
3939   if (ev != NULL) ev->unpark() ;
3940 
3941 }
3942 
3943 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3944   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3945     "possibility of dangling Thread pointer");
3946 
3947   OSThread* osthread = thread->osthread();
3948 
3949   bool interrupted = osthread->interrupted();
3950 
3951   if (interrupted && clear_interrupted) {
3952     osthread->set_interrupted(false);
3953     // consider thread->_SleepEvent->reset() ... optional optimization
3954   }
3955 
3956   return interrupted;
3957 }
3958 
3959 ///////////////////////////////////////////////////////////////////////////////////
3960 // signal handling (except suspend/resume)
3961 
3962 // This routine may be used by user applications as a "hook" to catch signals.
3963 // The user-defined signal handler must pass unrecognized signals to this
3964 // routine, and if it returns true (non-zero), then the signal handler must
3965 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3966 // routine will never retun false (zero), but instead will execute a VM panic
3967 // routine kill the process.
3968 //
3969 // If this routine returns false, it is OK to call it again.  This allows
3970 // the user-defined signal handler to perform checks either before or after
3971 // the VM performs its own checks.  Naturally, the user code would be making
3972 // a serious error if it tried to handle an exception (such as a null check
3973 // or breakpoint) that the VM was generating for its own correct operation.
3974 //
3975 // This routine may recognize any of the following kinds of signals:
3976 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3977 // It should be consulted by handlers for any of those signals.
3978 //
3979 // The caller of this routine must pass in the three arguments supplied
3980 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3981 // field of the structure passed to sigaction().  This routine assumes that
3982 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3983 //
3984 // Note that the VM will print warnings if it detects conflicting signal
3985 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3986 //
3987 extern "C" JNIEXPORT int
3988 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3989                         void* ucontext, int abort_if_unrecognized);
3990 
3991 void signalHandler(int sig, siginfo_t* info, void* uc) {
3992   assert(info != NULL && uc != NULL, "it must be old kernel");
3993   int orig_errno = errno;  // Preserve errno value over signal handler.
3994   JVM_handle_linux_signal(sig, info, uc, true);
3995   errno = orig_errno;
3996 }
3997 
3998 
3999 // This boolean allows users to forward their own non-matching signals
4000 // to JVM_handle_linux_signal, harmlessly.
4001 bool os::Linux::signal_handlers_are_installed = false;
4002 
4003 // For signal-chaining
4004 struct sigaction os::Linux::sigact[MAXSIGNUM];
4005 unsigned int os::Linux::sigs = 0;
4006 bool os::Linux::libjsig_is_loaded = false;
4007 typedef struct sigaction *(*get_signal_t)(int);
4008 get_signal_t os::Linux::get_signal_action = NULL;
4009 
4010 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4011   struct sigaction *actp = NULL;
4012 
4013   if (libjsig_is_loaded) {
4014     // Retrieve the old signal handler from libjsig
4015     actp = (*get_signal_action)(sig);
4016   }
4017   if (actp == NULL) {
4018     // Retrieve the preinstalled signal handler from jvm
4019     actp = get_preinstalled_handler(sig);
4020   }
4021 
4022   return actp;
4023 }
4024 
4025 static bool call_chained_handler(struct sigaction *actp, int sig,
4026                                  siginfo_t *siginfo, void *context) {
4027   // Call the old signal handler
4028   if (actp->sa_handler == SIG_DFL) {
4029     // It's more reasonable to let jvm treat it as an unexpected exception
4030     // instead of taking the default action.
4031     return false;
4032   } else if (actp->sa_handler != SIG_IGN) {
4033     if ((actp->sa_flags & SA_NODEFER) == 0) {
4034       // automaticlly block the signal
4035       sigaddset(&(actp->sa_mask), sig);
4036     }
4037 
4038     sa_handler_t hand;
4039     sa_sigaction_t sa;
4040     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4041     // retrieve the chained handler
4042     if (siginfo_flag_set) {
4043       sa = actp->sa_sigaction;
4044     } else {
4045       hand = actp->sa_handler;
4046     }
4047 
4048     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4049       actp->sa_handler = SIG_DFL;
4050     }
4051 
4052     // try to honor the signal mask
4053     sigset_t oset;
4054     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4055 
4056     // call into the chained handler
4057     if (siginfo_flag_set) {
4058       (*sa)(sig, siginfo, context);
4059     } else {
4060       (*hand)(sig);
4061     }
4062 
4063     // restore the signal mask
4064     pthread_sigmask(SIG_SETMASK, &oset, 0);
4065   }
4066   // Tell jvm's signal handler the signal is taken care of.
4067   return true;
4068 }
4069 
4070 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4071   bool chained = false;
4072   // signal-chaining
4073   if (UseSignalChaining) {
4074     struct sigaction *actp = get_chained_signal_action(sig);
4075     if (actp != NULL) {
4076       chained = call_chained_handler(actp, sig, siginfo, context);
4077     }
4078   }
4079   return chained;
4080 }
4081 
4082 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4083   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4084     return &sigact[sig];
4085   }
4086   return NULL;
4087 }
4088 
4089 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4090   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4091   sigact[sig] = oldAct;
4092   sigs |= (unsigned int)1 << sig;
4093 }
4094 
4095 // for diagnostic
4096 int os::Linux::sigflags[MAXSIGNUM];
4097 
4098 int os::Linux::get_our_sigflags(int sig) {
4099   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4100   return sigflags[sig];
4101 }
4102 
4103 void os::Linux::set_our_sigflags(int sig, int flags) {
4104   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4105   sigflags[sig] = flags;
4106 }
4107 
4108 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4109   // Check for overwrite.
4110   struct sigaction oldAct;
4111   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4112 
4113   void* oldhand = oldAct.sa_sigaction
4114                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4115                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4116   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4117       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4118       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4119     if (AllowUserSignalHandlers || !set_installed) {
4120       // Do not overwrite; user takes responsibility to forward to us.
4121       return;
4122     } else if (UseSignalChaining) {
4123       // save the old handler in jvm
4124       save_preinstalled_handler(sig, oldAct);
4125       // libjsig also interposes the sigaction() call below and saves the
4126       // old sigaction on it own.
4127     } else {
4128       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4129                     "%#lx for signal %d.", (long)oldhand, sig));
4130     }
4131   }
4132 
4133   struct sigaction sigAct;
4134   sigfillset(&(sigAct.sa_mask));
4135   sigAct.sa_handler = SIG_DFL;
4136   if (!set_installed) {
4137     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4138   } else {
4139     sigAct.sa_sigaction = signalHandler;
4140     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4141   }
4142   // Save flags, which are set by ours
4143   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4144   sigflags[sig] = sigAct.sa_flags;
4145 
4146   int ret = sigaction(sig, &sigAct, &oldAct);
4147   assert(ret == 0, "check");
4148 
4149   void* oldhand2  = oldAct.sa_sigaction
4150                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4151                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4152   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4153 }
4154 
4155 // install signal handlers for signals that HotSpot needs to
4156 // handle in order to support Java-level exception handling.
4157 
4158 void os::Linux::install_signal_handlers() {
4159   if (!signal_handlers_are_installed) {
4160     signal_handlers_are_installed = true;
4161 
4162     // signal-chaining
4163     typedef void (*signal_setting_t)();
4164     signal_setting_t begin_signal_setting = NULL;
4165     signal_setting_t end_signal_setting = NULL;
4166     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4167                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4168     if (begin_signal_setting != NULL) {
4169       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4170                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4171       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4172                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4173       libjsig_is_loaded = true;
4174       assert(UseSignalChaining, "should enable signal-chaining");
4175     }
4176     if (libjsig_is_loaded) {
4177       // Tell libjsig jvm is setting signal handlers
4178       (*begin_signal_setting)();
4179     }
4180 
4181     set_signal_handler(SIGSEGV, true);
4182     set_signal_handler(SIGPIPE, true);
4183     set_signal_handler(SIGBUS, true);
4184     set_signal_handler(SIGILL, true);
4185     set_signal_handler(SIGFPE, true);
4186     set_signal_handler(SIGXFSZ, true);
4187 
4188     if (libjsig_is_loaded) {
4189       // Tell libjsig jvm finishes setting signal handlers
4190       (*end_signal_setting)();
4191     }
4192 
4193     // We don't activate signal checker if libjsig is in place, we trust ourselves
4194     // and if UserSignalHandler is installed all bets are off.
4195     // Log that signal checking is off only if -verbose:jni is specified.
4196     if (CheckJNICalls) {
4197       if (libjsig_is_loaded) {
4198         if (PrintJNIResolving) {
4199           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4200         }
4201         check_signals = false;
4202       }
4203       if (AllowUserSignalHandlers) {
4204         if (PrintJNIResolving) {
4205           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4206         }
4207         check_signals = false;
4208       }
4209     }
4210   }
4211 }
4212 
4213 // This is the fastest way to get thread cpu time on Linux.
4214 // Returns cpu time (user+sys) for any thread, not only for current.
4215 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4216 // It might work on 2.6.10+ with a special kernel/glibc patch.
4217 // For reference, please, see IEEE Std 1003.1-2004:
4218 //   http://www.unix.org/single_unix_specification
4219 
4220 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4221   struct timespec tp;
4222   int rc = os::Linux::clock_gettime(clockid, &tp);
4223   assert(rc == 0, "clock_gettime is expected to return 0 code");
4224 
4225   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4226 }
4227 
4228 /////
4229 // glibc on Linux platform uses non-documented flag
4230 // to indicate, that some special sort of signal
4231 // trampoline is used.
4232 // We will never set this flag, and we should
4233 // ignore this flag in our diagnostic
4234 #ifdef SIGNIFICANT_SIGNAL_MASK
4235 #undef SIGNIFICANT_SIGNAL_MASK
4236 #endif
4237 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4238 
4239 static const char* get_signal_handler_name(address handler,
4240                                            char* buf, int buflen) {
4241   int offset;
4242   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4243   if (found) {
4244     // skip directory names
4245     const char *p1, *p2;
4246     p1 = buf;
4247     size_t len = strlen(os::file_separator());
4248     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4249     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4250   } else {
4251     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4252   }
4253   return buf;
4254 }
4255 
4256 static void print_signal_handler(outputStream* st, int sig,
4257                                  char* buf, size_t buflen) {
4258   struct sigaction sa;
4259 
4260   sigaction(sig, NULL, &sa);
4261 
4262   // See comment for SIGNIFICANT_SIGNAL_MASK define
4263   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4264 
4265   st->print("%s: ", os::exception_name(sig, buf, buflen));
4266 
4267   address handler = (sa.sa_flags & SA_SIGINFO)
4268     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4269     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4270 
4271   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4272     st->print("SIG_DFL");
4273   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4274     st->print("SIG_IGN");
4275   } else {
4276     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4277   }
4278 
4279   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4280 
4281   address rh = VMError::get_resetted_sighandler(sig);
4282   // May be, handler was resetted by VMError?
4283   if(rh != NULL) {
4284     handler = rh;
4285     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4286   }
4287 
4288   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
4289 
4290   // Check: is it our handler?
4291   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4292      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4293     // It is our signal handler
4294     // check for flags, reset system-used one!
4295     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4296       st->print(
4297                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4298                 os::Linux::get_our_sigflags(sig));
4299     }
4300   }
4301   st->cr();
4302 }
4303 
4304 
4305 #define DO_SIGNAL_CHECK(sig) \
4306   if (!sigismember(&check_signal_done, sig)) \
4307     os::Linux::check_signal_handler(sig)
4308 
4309 // This method is a periodic task to check for misbehaving JNI applications
4310 // under CheckJNI, we can add any periodic checks here
4311 
4312 void os::run_periodic_checks() {
4313 
4314   if (check_signals == false) return;
4315 
4316   // SEGV and BUS if overridden could potentially prevent
4317   // generation of hs*.log in the event of a crash, debugging
4318   // such a case can be very challenging, so we absolutely
4319   // check the following for a good measure:
4320   DO_SIGNAL_CHECK(SIGSEGV);
4321   DO_SIGNAL_CHECK(SIGILL);
4322   DO_SIGNAL_CHECK(SIGFPE);
4323   DO_SIGNAL_CHECK(SIGBUS);
4324   DO_SIGNAL_CHECK(SIGPIPE);
4325   DO_SIGNAL_CHECK(SIGXFSZ);
4326 
4327 
4328   // ReduceSignalUsage allows the user to override these handlers
4329   // see comments at the very top and jvm_solaris.h
4330   if (!ReduceSignalUsage) {
4331     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4332     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4333     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4334     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4335   }
4336 
4337   DO_SIGNAL_CHECK(SR_signum);
4338   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4339 }
4340 
4341 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4342 
4343 static os_sigaction_t os_sigaction = NULL;
4344 
4345 void os::Linux::check_signal_handler(int sig) {
4346   char buf[O_BUFLEN];
4347   address jvmHandler = NULL;
4348 
4349 
4350   struct sigaction act;
4351   if (os_sigaction == NULL) {
4352     // only trust the default sigaction, in case it has been interposed
4353     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4354     if (os_sigaction == NULL) return;
4355   }
4356 
4357   os_sigaction(sig, (struct sigaction*)NULL, &act);
4358 
4359 
4360   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4361 
4362   address thisHandler = (act.sa_flags & SA_SIGINFO)
4363     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4364     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4365 
4366 
4367   switch(sig) {
4368   case SIGSEGV:
4369   case SIGBUS:
4370   case SIGFPE:
4371   case SIGPIPE:
4372   case SIGILL:
4373   case SIGXFSZ:
4374     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4375     break;
4376 
4377   case SHUTDOWN1_SIGNAL:
4378   case SHUTDOWN2_SIGNAL:
4379   case SHUTDOWN3_SIGNAL:
4380   case BREAK_SIGNAL:
4381     jvmHandler = (address)user_handler();
4382     break;
4383 
4384   case INTERRUPT_SIGNAL:
4385     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4386     break;
4387 
4388   default:
4389     if (sig == SR_signum) {
4390       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4391     } else {
4392       return;
4393     }
4394     break;
4395   }
4396 
4397   if (thisHandler != jvmHandler) {
4398     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4399     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4400     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4401     // No need to check this sig any longer
4402     sigaddset(&check_signal_done, sig);
4403   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4404     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4405     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4406     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4407     // No need to check this sig any longer
4408     sigaddset(&check_signal_done, sig);
4409   }
4410 
4411   // Dump all the signal
4412   if (sigismember(&check_signal_done, sig)) {
4413     print_signal_handlers(tty, buf, O_BUFLEN);
4414   }
4415 }
4416 
4417 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4418 
4419 extern bool signal_name(int signo, char* buf, size_t len);
4420 
4421 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4422   if (0 < exception_code && exception_code <= SIGRTMAX) {
4423     // signal
4424     if (!signal_name(exception_code, buf, size)) {
4425       jio_snprintf(buf, size, "SIG%d", exception_code);
4426     }
4427     return buf;
4428   } else {
4429     return NULL;
4430   }
4431 }
4432 
4433 // this is called _before_ the most of global arguments have been parsed
4434 void os::init(void) {
4435   char dummy;   /* used to get a guess on initial stack address */
4436 //  first_hrtime = gethrtime();
4437 
4438   // With LinuxThreads the JavaMain thread pid (primordial thread)
4439   // is different than the pid of the java launcher thread.
4440   // So, on Linux, the launcher thread pid is passed to the VM
4441   // via the sun.java.launcher.pid property.
4442   // Use this property instead of getpid() if it was correctly passed.
4443   // See bug 6351349.
4444   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4445 
4446   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4447 
4448   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4449 
4450   init_random(1234567);
4451 
4452   ThreadCritical::initialize();
4453 
4454   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4455   if (Linux::page_size() == -1) {
4456     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4457                   strerror(errno)));
4458   }
4459   init_page_sizes((size_t) Linux::page_size());
4460 
4461   Linux::initialize_system_info();
4462 
4463   // main_thread points to the aboriginal thread
4464   Linux::_main_thread = pthread_self();
4465 
4466   Linux::clock_init();
4467   initial_time_count = os::elapsed_counter();
4468   pthread_mutex_init(&dl_mutex, NULL);
4469 
4470   // If the pagesize of the VM is greater than 8K determine the appropriate
4471   // number of initial guard pages.  The user can change this with the
4472   // command line arguments, if needed.
4473   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4474     StackYellowPages = 1;
4475     StackRedPages = 1;
4476     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4477   }
4478 }
4479 
4480 // To install functions for atexit system call
4481 extern "C" {
4482   static void perfMemory_exit_helper() {
4483     perfMemory_exit();
4484   }
4485 }
4486 
4487 // this is called _after_ the global arguments have been parsed
4488 jint os::init_2(void)
4489 {
4490   Linux::fast_thread_clock_init();
4491 
4492   // Allocate a single page and mark it as readable for safepoint polling
4493   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4494   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4495 
4496   os::set_polling_page( polling_page );
4497 
4498 #ifndef PRODUCT
4499   if(Verbose && PrintMiscellaneous)
4500     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4501 #endif
4502 
4503   if (!UseMembar) {
4504     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4505     guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4506     os::set_memory_serialize_page( mem_serialize_page );
4507 
4508 #ifndef PRODUCT
4509     if(Verbose && PrintMiscellaneous)
4510       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4511 #endif
4512   }
4513 
4514   os::large_page_init();
4515 
4516   // initialize suspend/resume support - must do this before signal_sets_init()
4517   if (SR_initialize() != 0) {
4518     perror("SR_initialize failed");
4519     return JNI_ERR;
4520   }
4521 
4522   Linux::signal_sets_init();
4523   Linux::install_signal_handlers();
4524 
4525   // Check minimum allowable stack size for thread creation and to initialize
4526   // the java system classes, including StackOverflowError - depends on page
4527   // size.  Add a page for compiler2 recursion in main thread.
4528   // Add in 2*BytesPerWord times page size to account for VM stack during
4529   // class initialization depending on 32 or 64 bit VM.
4530   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4531             (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4532                     (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4533 
4534   size_t threadStackSizeInBytes = ThreadStackSize * K;
4535   if (threadStackSizeInBytes != 0 &&
4536       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4537         tty->print_cr("\nThe stack size specified is too small, "
4538                       "Specify at least %dk",
4539                       os::Linux::min_stack_allowed/ K);
4540         return JNI_ERR;
4541   }
4542 
4543   // Make the stack size a multiple of the page size so that
4544   // the yellow/red zones can be guarded.
4545   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4546         vm_page_size()));
4547 
4548   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4549 
4550   Linux::libpthread_init();
4551   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4552      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4553           Linux::glibc_version(), Linux::libpthread_version(),
4554           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4555   }
4556 
4557   if (UseNUMA) {
4558     if (!Linux::libnuma_init()) {
4559       UseNUMA = false;
4560     } else {
4561       if ((Linux::numa_max_node() < 1)) {
4562         // There's only one node(they start from 0), disable NUMA.
4563         UseNUMA = false;
4564       }
4565     }
4566     // With SHM large pages we cannot uncommit a page, so there's not way
4567     // we can make the adaptive lgrp chunk resizing work. If the user specified
4568     // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4569     // disable adaptive resizing.
4570     if (UseNUMA && UseLargePages && UseSHM) {
4571       if (!FLAG_IS_DEFAULT(UseNUMA)) {
4572         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4573           UseLargePages = false;
4574         } else {
4575           warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4576           UseAdaptiveSizePolicy = false;
4577           UseAdaptiveNUMAChunkSizing = false;
4578         }
4579       } else {
4580         UseNUMA = false;
4581       }
4582     }
4583     if (!UseNUMA && ForceNUMA) {
4584       UseNUMA = true;
4585     }
4586   }
4587 
4588   if (MaxFDLimit) {
4589     // set the number of file descriptors to max. print out error
4590     // if getrlimit/setrlimit fails but continue regardless.
4591     struct rlimit nbr_files;
4592     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4593     if (status != 0) {
4594       if (PrintMiscellaneous && (Verbose || WizardMode))
4595         perror("os::init_2 getrlimit failed");
4596     } else {
4597       nbr_files.rlim_cur = nbr_files.rlim_max;
4598       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4599       if (status != 0) {
4600         if (PrintMiscellaneous && (Verbose || WizardMode))
4601           perror("os::init_2 setrlimit failed");
4602       }
4603     }
4604   }
4605 
4606   // Initialize lock used to serialize thread creation (see os::create_thread)
4607   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4608 
4609   // at-exit methods are called in the reverse order of their registration.
4610   // atexit functions are called on return from main or as a result of a
4611   // call to exit(3C). There can be only 32 of these functions registered
4612   // and atexit() does not set errno.
4613 
4614   if (PerfAllowAtExitRegistration) {
4615     // only register atexit functions if PerfAllowAtExitRegistration is set.
4616     // atexit functions can be delayed until process exit time, which
4617     // can be problematic for embedded VM situations. Embedded VMs should
4618     // call DestroyJavaVM() to assure that VM resources are released.
4619 
4620     // note: perfMemory_exit_helper atexit function may be removed in
4621     // the future if the appropriate cleanup code can be added to the
4622     // VM_Exit VMOperation's doit method.
4623     if (atexit(perfMemory_exit_helper) != 0) {
4624       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4625     }
4626   }
4627 
4628   // initialize thread priority policy
4629   prio_init();
4630 
4631   return JNI_OK;
4632 }
4633 
4634 // this is called at the end of vm_initialization
4635 void os::init_3(void)
4636 {
4637 #ifdef JAVASE_EMBEDDED
4638   // Start the MemNotifyThread
4639   if (LowMemoryProtection) {
4640     MemNotifyThread::start();
4641   }
4642   return;
4643 #endif
4644 }
4645 
4646 // Mark the polling page as unreadable
4647 void os::make_polling_page_unreadable(void) {
4648   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4649     fatal("Could not disable polling page");
4650 };
4651 
4652 // Mark the polling page as readable
4653 void os::make_polling_page_readable(void) {
4654   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4655     fatal("Could not enable polling page");
4656   }
4657 };
4658 
4659 int os::active_processor_count() {
4660   // Linux doesn't yet have a (official) notion of processor sets,
4661   // so just return the number of online processors.
4662   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4663   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4664   return online_cpus;
4665 }
4666 
4667 void os::set_native_thread_name(const char *name) {
4668   // Not yet implemented.
4669   return;
4670 }
4671 
4672 bool os::distribute_processes(uint length, uint* distribution) {
4673   // Not yet implemented.
4674   return false;
4675 }
4676 
4677 bool os::bind_to_processor(uint processor_id) {
4678   // Not yet implemented.
4679   return false;
4680 }
4681 
4682 ///
4683 
4684 void os::SuspendedThreadTask::internal_do_task() {
4685   if (do_suspend(_thread->osthread())) {
4686     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4687     do_task(context);
4688     do_resume(_thread->osthread());
4689   }
4690 }
4691 
4692 class PcFetcher : public os::SuspendedThreadTask {
4693 public:
4694   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4695   ExtendedPC result();
4696 protected:
4697   void do_task(const os::SuspendedThreadTaskContext& context);
4698 private:
4699   ExtendedPC _epc;
4700 };
4701 
4702 ExtendedPC PcFetcher::result() {
4703   guarantee(is_done(), "task is not done yet.");
4704   return _epc;
4705 }
4706 
4707 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4708   Thread* thread = context.thread();
4709   OSThread* osthread = thread->osthread();
4710   if (osthread->ucontext() != NULL) {
4711     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4712   } else {
4713     // NULL context is unexpected, double-check this is the VMThread
4714     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4715   }
4716 }
4717 
4718 // Suspends the target using the signal mechanism and then grabs the PC before
4719 // resuming the target. Used by the flat-profiler only
4720 ExtendedPC os::get_thread_pc(Thread* thread) {
4721   // Make sure that it is called by the watcher for the VMThread
4722   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4723   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4724 
4725   PcFetcher fetcher(thread);
4726   fetcher.run();
4727   return fetcher.result();
4728 }
4729 
4730 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4731 {
4732    if (is_NPTL()) {
4733       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4734    } else {
4735       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4736       // word back to default 64bit precision if condvar is signaled. Java
4737       // wants 53bit precision.  Save and restore current value.
4738       int fpu = get_fpu_control_word();
4739       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4740       set_fpu_control_word(fpu);
4741       return status;
4742    }
4743 }
4744 
4745 ////////////////////////////////////////////////////////////////////////////////
4746 // debug support
4747 
4748 bool os::find(address addr, outputStream* st) {
4749   Dl_info dlinfo;
4750   memset(&dlinfo, 0, sizeof(dlinfo));
4751   if (dladdr(addr, &dlinfo) != 0) {
4752     st->print(PTR_FORMAT ": ", addr);
4753     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4754       st->print("%s+%#x", dlinfo.dli_sname,
4755                  addr - (intptr_t)dlinfo.dli_saddr);
4756     } else if (dlinfo.dli_fbase != NULL) {
4757       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4758     } else {
4759       st->print("<absolute address>");
4760     }
4761     if (dlinfo.dli_fname != NULL) {
4762       st->print(" in %s", dlinfo.dli_fname);
4763     }
4764     if (dlinfo.dli_fbase != NULL) {
4765       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4766     }
4767     st->cr();
4768 
4769     if (Verbose) {
4770       // decode some bytes around the PC
4771       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4772       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4773       address       lowest = (address) dlinfo.dli_sname;
4774       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4775       if (begin < lowest)  begin = lowest;
4776       Dl_info dlinfo2;
4777       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4778           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4779         end = (address) dlinfo2.dli_saddr;
4780       Disassembler::decode(begin, end, st);
4781     }
4782     return true;
4783   }
4784   return false;
4785 }
4786 
4787 ////////////////////////////////////////////////////////////////////////////////
4788 // misc
4789 
4790 // This does not do anything on Linux. This is basically a hook for being
4791 // able to use structured exception handling (thread-local exception filters)
4792 // on, e.g., Win32.
4793 void
4794 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4795                          JavaCallArguments* args, Thread* thread) {
4796   f(value, method, args, thread);
4797 }
4798 
4799 void os::print_statistics() {
4800 }
4801 
4802 int os::message_box(const char* title, const char* message) {
4803   int i;
4804   fdStream err(defaultStream::error_fd());
4805   for (i = 0; i < 78; i++) err.print_raw("=");
4806   err.cr();
4807   err.print_raw_cr(title);
4808   for (i = 0; i < 78; i++) err.print_raw("-");
4809   err.cr();
4810   err.print_raw_cr(message);
4811   for (i = 0; i < 78; i++) err.print_raw("=");
4812   err.cr();
4813 
4814   char buf[16];
4815   // Prevent process from exiting upon "read error" without consuming all CPU
4816   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4817 
4818   return buf[0] == 'y' || buf[0] == 'Y';
4819 }
4820 
4821 int os::stat(const char *path, struct stat *sbuf) {
4822   char pathbuf[MAX_PATH];
4823   if (strlen(path) > MAX_PATH - 1) {
4824     errno = ENAMETOOLONG;
4825     return -1;
4826   }
4827   os::native_path(strcpy(pathbuf, path));
4828   return ::stat(pathbuf, sbuf);
4829 }
4830 
4831 bool os::check_heap(bool force) {
4832   return true;
4833 }
4834 
4835 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4836   return ::vsnprintf(buf, count, format, args);
4837 }
4838 
4839 // Is a (classpath) directory empty?
4840 bool os::dir_is_empty(const char* path) {
4841   DIR *dir = NULL;
4842   struct dirent *ptr;
4843 
4844   dir = opendir(path);
4845   if (dir == NULL) return true;
4846 
4847   /* Scan the directory */
4848   bool result = true;
4849   char buf[sizeof(struct dirent) + MAX_PATH];
4850   while (result && (ptr = ::readdir(dir)) != NULL) {
4851     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4852       result = false;
4853     }
4854   }
4855   closedir(dir);
4856   return result;
4857 }
4858 
4859 // This code originates from JDK's sysOpen and open64_w
4860 // from src/solaris/hpi/src/system_md.c
4861 
4862 #ifndef O_DELETE
4863 #define O_DELETE 0x10000
4864 #endif
4865 
4866 // Open a file. Unlink the file immediately after open returns
4867 // if the specified oflag has the O_DELETE flag set.
4868 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4869 
4870 int os::open(const char *path, int oflag, int mode) {
4871 
4872   if (strlen(path) > MAX_PATH - 1) {
4873     errno = ENAMETOOLONG;
4874     return -1;
4875   }
4876   int fd;
4877   int o_delete = (oflag & O_DELETE);
4878   oflag = oflag & ~O_DELETE;
4879 
4880   fd = ::open64(path, oflag, mode);
4881   if (fd == -1) return -1;
4882 
4883   //If the open succeeded, the file might still be a directory
4884   {
4885     struct stat64 buf64;
4886     int ret = ::fstat64(fd, &buf64);
4887     int st_mode = buf64.st_mode;
4888 
4889     if (ret != -1) {
4890       if ((st_mode & S_IFMT) == S_IFDIR) {
4891         errno = EISDIR;
4892         ::close(fd);
4893         return -1;
4894       }
4895     } else {
4896       ::close(fd);
4897       return -1;
4898     }
4899   }
4900 
4901     /*
4902      * All file descriptors that are opened in the JVM and not
4903      * specifically destined for a subprocess should have the
4904      * close-on-exec flag set.  If we don't set it, then careless 3rd
4905      * party native code might fork and exec without closing all
4906      * appropriate file descriptors (e.g. as we do in closeDescriptors in
4907      * UNIXProcess.c), and this in turn might:
4908      *
4909      * - cause end-of-file to fail to be detected on some file
4910      *   descriptors, resulting in mysterious hangs, or
4911      *
4912      * - might cause an fopen in the subprocess to fail on a system
4913      *   suffering from bug 1085341.
4914      *
4915      * (Yes, the default setting of the close-on-exec flag is a Unix
4916      * design flaw)
4917      *
4918      * See:
4919      * 1085341: 32-bit stdio routines should support file descriptors >255
4920      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4921      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4922      */
4923 #ifdef FD_CLOEXEC
4924     {
4925         int flags = ::fcntl(fd, F_GETFD);
4926         if (flags != -1)
4927             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4928     }
4929 #endif
4930 
4931   if (o_delete != 0) {
4932     ::unlink(path);
4933   }
4934   return fd;
4935 }
4936 
4937 
4938 // create binary file, rewriting existing file if required
4939 int os::create_binary_file(const char* path, bool rewrite_existing) {
4940   int oflags = O_WRONLY | O_CREAT;
4941   if (!rewrite_existing) {
4942     oflags |= O_EXCL;
4943   }
4944   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4945 }
4946 
4947 // return current position of file pointer
4948 jlong os::current_file_offset(int fd) {
4949   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4950 }
4951 
4952 // move file pointer to the specified offset
4953 jlong os::seek_to_file_offset(int fd, jlong offset) {
4954   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4955 }
4956 
4957 // This code originates from JDK's sysAvailable
4958 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4959 
4960 int os::available(int fd, jlong *bytes) {
4961   jlong cur, end;
4962   int mode;
4963   struct stat64 buf64;
4964 
4965   if (::fstat64(fd, &buf64) >= 0) {
4966     mode = buf64.st_mode;
4967     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4968       /*
4969       * XXX: is the following call interruptible? If so, this might
4970       * need to go through the INTERRUPT_IO() wrapper as for other
4971       * blocking, interruptible calls in this file.
4972       */
4973       int n;
4974       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4975         *bytes = n;
4976         return 1;
4977       }
4978     }
4979   }
4980   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4981     return 0;
4982   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4983     return 0;
4984   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4985     return 0;
4986   }
4987   *bytes = end - cur;
4988   return 1;
4989 }
4990 
4991 int os::socket_available(int fd, jint *pbytes) {
4992   // Linux doc says EINTR not returned, unlike Solaris
4993   int ret = ::ioctl(fd, FIONREAD, pbytes);
4994 
4995   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4996   // is expected to return 0 on failure and 1 on success to the jdk.
4997   return (ret < 0) ? 0 : 1;
4998 }
4999 
5000 // Map a block of memory.
5001 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5002                      char *addr, size_t bytes, bool read_only,
5003                      bool allow_exec) {
5004   int prot;
5005   int flags = MAP_PRIVATE;
5006 
5007   if (read_only) {
5008     prot = PROT_READ;
5009   } else {
5010     prot = PROT_READ | PROT_WRITE;
5011   }
5012 
5013   if (allow_exec) {
5014     prot |= PROT_EXEC;
5015   }
5016 
5017   if (addr != NULL) {
5018     flags |= MAP_FIXED;
5019   }
5020 
5021   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5022                                      fd, file_offset);
5023   if (mapped_address == MAP_FAILED) {
5024     return NULL;
5025   }
5026   return mapped_address;
5027 }
5028 
5029 
5030 // Remap a block of memory.
5031 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5032                        char *addr, size_t bytes, bool read_only,
5033                        bool allow_exec) {
5034   // same as map_memory() on this OS
5035   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5036                         allow_exec);
5037 }
5038 
5039 
5040 // Unmap a block of memory.
5041 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5042   return munmap(addr, bytes) == 0;
5043 }
5044 
5045 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5046 
5047 static clockid_t thread_cpu_clockid(Thread* thread) {
5048   pthread_t tid = thread->osthread()->pthread_id();
5049   clockid_t clockid;
5050 
5051   // Get thread clockid
5052   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5053   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5054   return clockid;
5055 }
5056 
5057 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5058 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5059 // of a thread.
5060 //
5061 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5062 // the fast estimate available on the platform.
5063 
5064 jlong os::current_thread_cpu_time() {
5065   if (os::Linux::supports_fast_thread_cpu_time()) {
5066     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5067   } else {
5068     // return user + sys since the cost is the same
5069     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5070   }
5071 }
5072 
5073 jlong os::thread_cpu_time(Thread* thread) {
5074   // consistent with what current_thread_cpu_time() returns
5075   if (os::Linux::supports_fast_thread_cpu_time()) {
5076     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5077   } else {
5078     return slow_thread_cpu_time(thread, true /* user + sys */);
5079   }
5080 }
5081 
5082 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5083   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5084     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5085   } else {
5086     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5087   }
5088 }
5089 
5090 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5091   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5092     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5093   } else {
5094     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5095   }
5096 }
5097 
5098 //
5099 //  -1 on error.
5100 //
5101 
5102 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5103   static bool proc_task_unchecked = true;
5104   static const char *proc_stat_path = "/proc/%d/stat";
5105   pid_t  tid = thread->osthread()->thread_id();
5106   char *s;
5107   char stat[2048];
5108   int statlen;
5109   char proc_name[64];
5110   int count;
5111   long sys_time, user_time;
5112   char cdummy;
5113   int idummy;
5114   long ldummy;
5115   FILE *fp;
5116 
5117   // The /proc/<tid>/stat aggregates per-process usage on
5118   // new Linux kernels 2.6+ where NPTL is supported.
5119   // The /proc/self/task/<tid>/stat still has the per-thread usage.
5120   // See bug 6328462.
5121   // There possibly can be cases where there is no directory
5122   // /proc/self/task, so we check its availability.
5123   if (proc_task_unchecked && os::Linux::is_NPTL()) {
5124     // This is executed only once
5125     proc_task_unchecked = false;
5126     fp = fopen("/proc/self/task", "r");
5127     if (fp != NULL) {
5128       proc_stat_path = "/proc/self/task/%d/stat";
5129       fclose(fp);
5130     }
5131   }
5132 
5133   sprintf(proc_name, proc_stat_path, tid);
5134   fp = fopen(proc_name, "r");
5135   if ( fp == NULL ) return -1;
5136   statlen = fread(stat, 1, 2047, fp);
5137   stat[statlen] = '\0';
5138   fclose(fp);
5139 
5140   // Skip pid and the command string. Note that we could be dealing with
5141   // weird command names, e.g. user could decide to rename java launcher
5142   // to "java 1.4.2 :)", then the stat file would look like
5143   //                1234 (java 1.4.2 :)) R ... ...
5144   // We don't really need to know the command string, just find the last
5145   // occurrence of ")" and then start parsing from there. See bug 4726580.
5146   s = strrchr(stat, ')');
5147   if (s == NULL ) return -1;
5148 
5149   // Skip blank chars
5150   do s++; while (isspace(*s));
5151 
5152   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5153                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5154                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5155                  &user_time, &sys_time);
5156   if ( count != 13 ) return -1;
5157   if (user_sys_cpu_time) {
5158     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5159   } else {
5160     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5161   }
5162 }
5163 
5164 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5165   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5166   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5167   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5168   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5169 }
5170 
5171 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5172   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5173   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5174   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5175   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5176 }
5177 
5178 bool os::is_thread_cpu_time_supported() {
5179   return true;
5180 }
5181 
5182 // System loadavg support.  Returns -1 if load average cannot be obtained.
5183 // Linux doesn't yet have a (official) notion of processor sets,
5184 // so just return the system wide load average.
5185 int os::loadavg(double loadavg[], int nelem) {
5186   return ::getloadavg(loadavg, nelem);
5187 }
5188 
5189 void os::pause() {
5190   char filename[MAX_PATH];
5191   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5192     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5193   } else {
5194     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5195   }
5196 
5197   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5198   if (fd != -1) {
5199     struct stat buf;
5200     ::close(fd);
5201     while (::stat(filename, &buf) == 0) {
5202       (void)::poll(NULL, 0, 100);
5203     }
5204   } else {
5205     jio_fprintf(stderr,
5206       "Could not open pause file '%s', continuing immediately.\n", filename);
5207   }
5208 }
5209 
5210 
5211 // Refer to the comments in os_solaris.cpp park-unpark.
5212 //
5213 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5214 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5215 // For specifics regarding the bug see GLIBC BUGID 261237 :
5216 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5217 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5218 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5219 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5220 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5221 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5222 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5223 // of libpthread avoids the problem, but isn't practical.
5224 //
5225 // Possible remedies:
5226 //
5227 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5228 //      This is palliative and probabilistic, however.  If the thread is preempted
5229 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5230 //      than the minimum period may have passed, and the abstime may be stale (in the
5231 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5232 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5233 //
5234 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5235 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5236 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5237 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5238 //      thread.
5239 //
5240 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5241 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5242 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5243 //      This also works well.  In fact it avoids kernel-level scalability impediments
5244 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5245 //      timers in a graceful fashion.
5246 //
5247 // 4.   When the abstime value is in the past it appears that control returns
5248 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5249 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5250 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5251 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5252 //      It may be possible to avoid reinitialization by checking the return
5253 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5254 //      condvar we must establish the invariant that cond_signal() is only called
5255 //      within critical sections protected by the adjunct mutex.  This prevents
5256 //      cond_signal() from "seeing" a condvar that's in the midst of being
5257 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5258 //      desirable signal-after-unlock optimization that avoids futile context switching.
5259 //
5260 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5261 //      structure when a condvar is used or initialized.  cond_destroy()  would
5262 //      release the helper structure.  Our reinitialize-after-timedwait fix
5263 //      put excessive stress on malloc/free and locks protecting the c-heap.
5264 //
5265 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5266 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5267 // and only enabling the work-around for vulnerable environments.
5268 
5269 // utility to compute the abstime argument to timedwait:
5270 // millis is the relative timeout time
5271 // abstime will be the absolute timeout time
5272 // TODO: replace compute_abstime() with unpackTime()
5273 
5274 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5275   if (millis < 0)  millis = 0;
5276   struct timeval now;
5277   int status = gettimeofday(&now, NULL);
5278   assert(status == 0, "gettimeofday");
5279   jlong seconds = millis / 1000;
5280   millis %= 1000;
5281   if (seconds > 50000000) { // see man cond_timedwait(3T)
5282     seconds = 50000000;
5283   }
5284   abstime->tv_sec = now.tv_sec  + seconds;
5285   long       usec = now.tv_usec + millis * 1000;
5286   if (usec >= 1000000) {
5287     abstime->tv_sec += 1;
5288     usec -= 1000000;
5289   }
5290   abstime->tv_nsec = usec * 1000;
5291   return abstime;
5292 }
5293 
5294 
5295 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5296 // Conceptually TryPark() should be equivalent to park(0).
5297 
5298 int os::PlatformEvent::TryPark() {
5299   for (;;) {
5300     const int v = _Event ;
5301     guarantee ((v == 0) || (v == 1), "invariant") ;
5302     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5303   }
5304 }
5305 
5306 void os::PlatformEvent::park() {       // AKA "down()"
5307   // Invariant: Only the thread associated with the Event/PlatformEvent
5308   // may call park().
5309   // TODO: assert that _Assoc != NULL or _Assoc == Self
5310   int v ;
5311   for (;;) {
5312       v = _Event ;
5313       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5314   }
5315   guarantee (v >= 0, "invariant") ;
5316   if (v == 0) {
5317      // Do this the hard way by blocking ...
5318      int status = pthread_mutex_lock(_mutex);
5319      assert_status(status == 0, status, "mutex_lock");
5320      guarantee (_nParked == 0, "invariant") ;
5321      ++ _nParked ;
5322      while (_Event < 0) {
5323         status = pthread_cond_wait(_cond, _mutex);
5324         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5325         // Treat this the same as if the wait was interrupted
5326         if (status == ETIME) { status = EINTR; }
5327         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5328      }
5329      -- _nParked ;
5330 
5331     _Event = 0 ;
5332      status = pthread_mutex_unlock(_mutex);
5333      assert_status(status == 0, status, "mutex_unlock");
5334     // Paranoia to ensure our locked and lock-free paths interact
5335     // correctly with each other.
5336     OrderAccess::fence();
5337   }
5338   guarantee (_Event >= 0, "invariant") ;
5339 }
5340 
5341 int os::PlatformEvent::park(jlong millis) {
5342   guarantee (_nParked == 0, "invariant") ;
5343 
5344   int v ;
5345   for (;;) {
5346       v = _Event ;
5347       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5348   }
5349   guarantee (v >= 0, "invariant") ;
5350   if (v != 0) return OS_OK ;
5351 
5352   // We do this the hard way, by blocking the thread.
5353   // Consider enforcing a minimum timeout value.
5354   struct timespec abst;
5355   compute_abstime(&abst, millis);
5356 
5357   int ret = OS_TIMEOUT;
5358   int status = pthread_mutex_lock(_mutex);
5359   assert_status(status == 0, status, "mutex_lock");
5360   guarantee (_nParked == 0, "invariant") ;
5361   ++_nParked ;
5362 
5363   // Object.wait(timo) will return because of
5364   // (a) notification
5365   // (b) timeout
5366   // (c) thread.interrupt
5367   //
5368   // Thread.interrupt and object.notify{All} both call Event::set.
5369   // That is, we treat thread.interrupt as a special case of notification.
5370   // The underlying Solaris implementation, cond_timedwait, admits
5371   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5372   // JVM from making those visible to Java code.  As such, we must
5373   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5374   //
5375   // TODO: properly differentiate simultaneous notify+interrupt.
5376   // In that case, we should propagate the notify to another waiter.
5377 
5378   while (_Event < 0) {
5379     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5380     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5381       pthread_cond_destroy (_cond);
5382       pthread_cond_init (_cond, NULL) ;
5383     }
5384     assert_status(status == 0 || status == EINTR ||
5385                   status == ETIME || status == ETIMEDOUT,
5386                   status, "cond_timedwait");
5387     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5388     if (status == ETIME || status == ETIMEDOUT) break ;
5389     // We consume and ignore EINTR and spurious wakeups.
5390   }
5391   --_nParked ;
5392   if (_Event >= 0) {
5393      ret = OS_OK;
5394   }
5395   _Event = 0 ;
5396   status = pthread_mutex_unlock(_mutex);
5397   assert_status(status == 0, status, "mutex_unlock");
5398   assert (_nParked == 0, "invariant") ;
5399   // Paranoia to ensure our locked and lock-free paths interact
5400   // correctly with each other.
5401   OrderAccess::fence();
5402   return ret;
5403 }
5404 
5405 void os::PlatformEvent::unpark() {
5406   // Transitions for _Event:
5407   //    0 :=> 1
5408   //    1 :=> 1
5409   //   -1 :=> either 0 or 1; must signal target thread
5410   //          That is, we can safely transition _Event from -1 to either
5411   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
5412   //          unpark() calls.
5413   // See also: "Semaphores in Plan 9" by Mullender & Cox
5414   //
5415   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5416   // that it will take two back-to-back park() calls for the owning
5417   // thread to block. This has the benefit of forcing a spurious return
5418   // from the first park() call after an unpark() call which will help
5419   // shake out uses of park() and unpark() without condition variables.
5420 
5421   if (Atomic::xchg(1, &_Event) >= 0) return;
5422 
5423   // Wait for the thread associated with the event to vacate
5424   int status = pthread_mutex_lock(_mutex);
5425   assert_status(status == 0, status, "mutex_lock");
5426   int AnyWaiters = _nParked;
5427   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5428   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5429     AnyWaiters = 0;
5430     pthread_cond_signal(_cond);
5431   }
5432   status = pthread_mutex_unlock(_mutex);
5433   assert_status(status == 0, status, "mutex_unlock");
5434   if (AnyWaiters != 0) {
5435     status = pthread_cond_signal(_cond);
5436     assert_status(status == 0, status, "cond_signal");
5437   }
5438 
5439   // Note that we signal() _after dropping the lock for "immortal" Events.
5440   // This is safe and avoids a common class of  futile wakeups.  In rare
5441   // circumstances this can cause a thread to return prematurely from
5442   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5443   // simply re-test the condition and re-park itself.
5444 }
5445 
5446 
5447 // JSR166
5448 // -------------------------------------------------------
5449 
5450 /*
5451  * The solaris and linux implementations of park/unpark are fairly
5452  * conservative for now, but can be improved. They currently use a
5453  * mutex/condvar pair, plus a a count.
5454  * Park decrements count if > 0, else does a condvar wait.  Unpark
5455  * sets count to 1 and signals condvar.  Only one thread ever waits
5456  * on the condvar. Contention seen when trying to park implies that someone
5457  * is unparking you, so don't wait. And spurious returns are fine, so there
5458  * is no need to track notifications.
5459  */
5460 
5461 #define MAX_SECS 100000000
5462 /*
5463  * This code is common to linux and solaris and will be moved to a
5464  * common place in dolphin.
5465  *
5466  * The passed in time value is either a relative time in nanoseconds
5467  * or an absolute time in milliseconds. Either way it has to be unpacked
5468  * into suitable seconds and nanoseconds components and stored in the
5469  * given timespec structure.
5470  * Given time is a 64-bit value and the time_t used in the timespec is only
5471  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5472  * overflow if times way in the future are given. Further on Solaris versions
5473  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5474  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5475  * As it will be 28 years before "now + 100000000" will overflow we can
5476  * ignore overflow and just impose a hard-limit on seconds using the value
5477  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5478  * years from "now".
5479  */
5480 
5481 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5482   assert (time > 0, "convertTime");
5483 
5484   struct timeval now;
5485   int status = gettimeofday(&now, NULL);
5486   assert(status == 0, "gettimeofday");
5487 
5488   time_t max_secs = now.tv_sec + MAX_SECS;
5489 
5490   if (isAbsolute) {
5491     jlong secs = time / 1000;
5492     if (secs > max_secs) {
5493       absTime->tv_sec = max_secs;
5494     }
5495     else {
5496       absTime->tv_sec = secs;
5497     }
5498     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5499   }
5500   else {
5501     jlong secs = time / NANOSECS_PER_SEC;
5502     if (secs >= MAX_SECS) {
5503       absTime->tv_sec = max_secs;
5504       absTime->tv_nsec = 0;
5505     }
5506     else {
5507       absTime->tv_sec = now.tv_sec + secs;
5508       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5509       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5510         absTime->tv_nsec -= NANOSECS_PER_SEC;
5511         ++absTime->tv_sec; // note: this must be <= max_secs
5512       }
5513     }
5514   }
5515   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5516   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5517   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5518   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5519 }
5520 
5521 void Parker::park(bool isAbsolute, jlong time) {
5522   // Ideally we'd do something useful while spinning, such
5523   // as calling unpackTime().
5524 
5525   // Optional fast-path check:
5526   // Return immediately if a permit is available.
5527   // We depend on Atomic::xchg() having full barrier semantics
5528   // since we are doing a lock-free update to _counter.
5529   if (Atomic::xchg(0, &_counter) > 0) return;
5530 
5531   Thread* thread = Thread::current();
5532   assert(thread->is_Java_thread(), "Must be JavaThread");
5533   JavaThread *jt = (JavaThread *)thread;
5534 
5535   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5536   // Check interrupt before trying to wait
5537   if (Thread::is_interrupted(thread, false)) {
5538     return;
5539   }
5540 
5541   // Next, demultiplex/decode time arguments
5542   timespec absTime;
5543   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5544     return;
5545   }
5546   if (time > 0) {
5547     unpackTime(&absTime, isAbsolute, time);
5548   }
5549 
5550 
5551   // Enter safepoint region
5552   // Beware of deadlocks such as 6317397.
5553   // The per-thread Parker:: mutex is a classic leaf-lock.
5554   // In particular a thread must never block on the Threads_lock while
5555   // holding the Parker:: mutex.  If safepoints are pending both the
5556   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5557   ThreadBlockInVM tbivm(jt);
5558 
5559   // Don't wait if cannot get lock since interference arises from
5560   // unblocking.  Also. check interrupt before trying wait
5561   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5562     return;
5563   }
5564 
5565   int status ;
5566   if (_counter > 0)  { // no wait needed
5567     _counter = 0;
5568     status = pthread_mutex_unlock(_mutex);
5569     assert (status == 0, "invariant") ;
5570     // Paranoia to ensure our locked and lock-free paths interact
5571     // correctly with each other and Java-level accesses.
5572     OrderAccess::fence();
5573     return;
5574   }
5575 
5576 #ifdef ASSERT
5577   // Don't catch signals while blocked; let the running threads have the signals.
5578   // (This allows a debugger to break into the running thread.)
5579   sigset_t oldsigs;
5580   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5581   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5582 #endif
5583 
5584   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5585   jt->set_suspend_equivalent();
5586   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5587 
5588   if (time == 0) {
5589     status = pthread_cond_wait (_cond, _mutex) ;
5590   } else {
5591     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5592     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5593       pthread_cond_destroy (_cond) ;
5594       pthread_cond_init    (_cond, NULL);
5595     }
5596   }
5597   assert_status(status == 0 || status == EINTR ||
5598                 status == ETIME || status == ETIMEDOUT,
5599                 status, "cond_timedwait");
5600 
5601 #ifdef ASSERT
5602   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5603 #endif
5604 
5605   _counter = 0 ;
5606   status = pthread_mutex_unlock(_mutex) ;
5607   assert_status(status == 0, status, "invariant") ;
5608   // Paranoia to ensure our locked and lock-free paths interact
5609   // correctly with each other and Java-level accesses.
5610   OrderAccess::fence();
5611 
5612   // If externally suspended while waiting, re-suspend
5613   if (jt->handle_special_suspend_equivalent_condition()) {
5614     jt->java_suspend_self();
5615   }
5616 }
5617 
5618 void Parker::unpark() {
5619   int s, status ;
5620   status = pthread_mutex_lock(_mutex);
5621   assert (status == 0, "invariant") ;
5622   s = _counter;
5623   _counter = 1;
5624   if (s < 1) {
5625      if (WorkAroundNPTLTimedWaitHang) {
5626         status = pthread_cond_signal (_cond) ;
5627         assert (status == 0, "invariant") ;
5628         status = pthread_mutex_unlock(_mutex);
5629         assert (status == 0, "invariant") ;
5630      } else {
5631         status = pthread_mutex_unlock(_mutex);
5632         assert (status == 0, "invariant") ;
5633         status = pthread_cond_signal (_cond) ;
5634         assert (status == 0, "invariant") ;
5635      }
5636   } else {
5637     pthread_mutex_unlock(_mutex);
5638     assert (status == 0, "invariant") ;
5639   }
5640 }
5641 
5642 
5643 extern char** environ;
5644 
5645 #ifndef __NR_fork
5646 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5647 #endif
5648 
5649 #ifndef __NR_execve
5650 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5651 #endif
5652 
5653 // Run the specified command in a separate process. Return its exit value,
5654 // or -1 on failure (e.g. can't fork a new process).
5655 // Unlike system(), this function can be called from signal handler. It
5656 // doesn't block SIGINT et al.
5657 int os::fork_and_exec(char* cmd) {
5658   const char * argv[4] = {"sh", "-c", cmd, NULL};
5659 
5660   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5661   // pthread_atfork handlers and reset pthread library. All we need is a
5662   // separate process to execve. Make a direct syscall to fork process.
5663   // On IA64 there's no fork syscall, we have to use fork() and hope for
5664   // the best...
5665   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5666               IA64_ONLY(fork();)
5667 
5668   if (pid < 0) {
5669     // fork failed
5670     return -1;
5671 
5672   } else if (pid == 0) {
5673     // child process
5674 
5675     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5676     // first to kill every thread on the thread list. Because this list is
5677     // not reset by fork() (see notes above), execve() will instead kill
5678     // every thread in the parent process. We know this is the only thread
5679     // in the new process, so make a system call directly.
5680     // IA64 should use normal execve() from glibc to match the glibc fork()
5681     // above.
5682     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5683     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5684 
5685     // execve failed
5686     _exit(-1);
5687 
5688   } else  {
5689     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5690     // care about the actual exit code, for now.
5691 
5692     int status;
5693 
5694     // Wait for the child process to exit.  This returns immediately if
5695     // the child has already exited. */
5696     while (waitpid(pid, &status, 0) < 0) {
5697         switch (errno) {
5698         case ECHILD: return 0;
5699         case EINTR: break;
5700         default: return -1;
5701         }
5702     }
5703 
5704     if (WIFEXITED(status)) {
5705        // The child exited normally; get its exit code.
5706        return WEXITSTATUS(status);
5707     } else if (WIFSIGNALED(status)) {
5708        // The child exited because of a signal
5709        // The best value to return is 0x80 + signal number,
5710        // because that is what all Unix shells do, and because
5711        // it allows callers to distinguish between process exit and
5712        // process death by signal.
5713        return 0x80 + WTERMSIG(status);
5714     } else {
5715        // Unknown exit code; pass it through
5716        return status;
5717     }
5718   }
5719 }
5720 
5721 // is_headless_jre()
5722 //
5723 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5724 // in order to report if we are running in a headless jre
5725 //
5726 // Since JDK8 xawt/libmawt.so was moved into the same directory
5727 // as libawt.so, and renamed libawt_xawt.so
5728 //
5729 bool os::is_headless_jre() {
5730     struct stat statbuf;
5731     char buf[MAXPATHLEN];
5732     char libmawtpath[MAXPATHLEN];
5733     const char *xawtstr  = "/xawt/libmawt.so";
5734     const char *new_xawtstr = "/libawt_xawt.so";
5735     char *p;
5736 
5737     // Get path to libjvm.so
5738     os::jvm_path(buf, sizeof(buf));
5739 
5740     // Get rid of libjvm.so
5741     p = strrchr(buf, '/');
5742     if (p == NULL) return false;
5743     else *p = '\0';
5744 
5745     // Get rid of client or server
5746     p = strrchr(buf, '/');
5747     if (p == NULL) return false;
5748     else *p = '\0';
5749 
5750     // check xawt/libmawt.so
5751     strcpy(libmawtpath, buf);
5752     strcat(libmawtpath, xawtstr);
5753     if (::stat(libmawtpath, &statbuf) == 0) return false;
5754 
5755     // check libawt_xawt.so
5756     strcpy(libmawtpath, buf);
5757     strcat(libmawtpath, new_xawtstr);
5758     if (::stat(libmawtpath, &statbuf) == 0) return false;
5759 
5760     return true;
5761 }
5762 
5763 // Get the default path to the core file
5764 // Returns the length of the string
5765 int os::get_core_path(char* buffer, size_t bufferSize) {
5766   const char* p = get_current_directory(buffer, bufferSize);
5767 
5768   if (p == NULL) {
5769     assert(p != NULL, "failed to get current directory");
5770     return 0;
5771   }
5772 
5773   return strlen(buffer);
5774 }
5775 
5776 #ifdef JAVASE_EMBEDDED
5777 //
5778 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5779 //
5780 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5781 
5782 // ctor
5783 //
5784 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5785   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5786   _fd = fd;
5787 
5788   if (os::create_thread(this, os::os_thread)) {
5789     _memnotify_thread = this;
5790     os::set_priority(this, NearMaxPriority);
5791     os::start_thread(this);
5792   }
5793 }
5794 
5795 // Where all the work gets done
5796 //
5797 void MemNotifyThread::run() {
5798   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5799 
5800   // Set up the select arguments
5801   fd_set rfds;
5802   if (_fd != -1) {
5803     FD_ZERO(&rfds);
5804     FD_SET(_fd, &rfds);
5805   }
5806 
5807   // Now wait for the mem_notify device to wake up
5808   while (1) {
5809     // Wait for the mem_notify device to signal us..
5810     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5811     if (rc == -1) {
5812       perror("select!\n");
5813       break;
5814     } else if (rc) {
5815       //ssize_t free_before = os::available_memory();
5816       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5817 
5818       // The kernel is telling us there is not much memory left...
5819       // try to do something about that
5820 
5821       // If we are not already in a GC, try one.
5822       if (!Universe::heap()->is_gc_active()) {
5823         Universe::heap()->collect(GCCause::_allocation_failure);
5824 
5825         //ssize_t free_after = os::available_memory();
5826         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5827         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5828       }
5829       // We might want to do something like the following if we find the GC's are not helping...
5830       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5831     }
5832   }
5833 }
5834 
5835 //
5836 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5837 //
5838 void MemNotifyThread::start() {
5839   int    fd;
5840   fd = open ("/dev/mem_notify", O_RDONLY, 0);
5841   if (fd < 0) {
5842       return;
5843   }
5844 
5845   if (memnotify_thread() == NULL) {
5846     new MemNotifyThread(fd);
5847   }
5848 }
5849 
5850 #endif // JAVASE_EMBEDDED