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 #define MAX_PATH    (2 * K)
 105 
 106 // for timer info max values which include all bits
 107 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 108 
 109 #define LARGEPAGES_BIT (1 << 6)
 110 ////////////////////////////////////////////////////////////////////////////////
 111 // global variables
 112 julong os::Linux::_physical_memory = 0;
 113 
 114 address   os::Linux::_initial_thread_stack_bottom = NULL;
 115 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 116 
 117 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 118 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 119 Mutex* os::Linux::_createThread_lock = NULL;
 120 pthread_t os::Linux::_main_thread;
 121 int os::Linux::_page_size = -1;
 122 const int os::Linux::_vm_default_page_size = (8 * K);
 123 bool os::Linux::_is_floating_stack = false;
 124 bool os::Linux::_is_NPTL = false;
 125 bool os::Linux::_supports_fast_thread_cpu_time = false;
 126 const char * os::Linux::_glibc_version = NULL;
 127 const char * os::Linux::_libpthread_version = NULL;
 128 
 129 static jlong initial_time_count=0;
 130 
 131 static int clock_tics_per_sec = 100;
 132 
 133 // For diagnostics to print a message once. see run_periodic_checks
 134 static sigset_t check_signal_done;
 135 static bool check_signals = true;;
 136 
 137 static pid_t _initial_pid = 0;
 138 
 139 /* Signal number used to suspend/resume a thread */
 140 
 141 /* do not use any signal number less than SIGSEGV, see 4355769 */
 142 static int SR_signum = SIGUSR2;
 143 sigset_t SR_sigset;
 144 
 145 /* Used to protect dlsym() calls */
 146 static pthread_mutex_t dl_mutex;
 147 
 148 // Declarations
 149 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 150 
 151 #ifdef JAVASE_EMBEDDED
 152 class MemNotifyThread: public Thread {
 153   friend class VMStructs;
 154  public:
 155   virtual void run();
 156 
 157  private:
 158   static MemNotifyThread* _memnotify_thread;
 159   int _fd;
 160 
 161  public:
 162 
 163   // Constructor
 164   MemNotifyThread(int fd);
 165 
 166   // Tester
 167   bool is_memnotify_thread() const { return true; }
 168 
 169   // Printing
 170   char* name() const { return (char*)"Linux MemNotify Thread"; }
 171 
 172   // Returns the single instance of the MemNotifyThread
 173   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 174 
 175   // Create and start the single instance of MemNotifyThread
 176   static void start();
 177 };
 178 #endif // JAVASE_EMBEDDED
 179 
 180 // utility functions
 181 
 182 static int SR_initialize();
 183 
 184 julong os::available_memory() {
 185   return Linux::available_memory();
 186 }
 187 
 188 julong os::Linux::available_memory() {
 189   // values in struct sysinfo are "unsigned long"
 190   struct sysinfo si;
 191   sysinfo(&si);
 192 
 193   return (julong)si.freeram * si.mem_unit;
 194 }
 195 
 196 julong os::physical_memory() {
 197   return Linux::physical_memory();
 198 }
 199 
 200 ////////////////////////////////////////////////////////////////////////////////
 201 // environment support
 202 
 203 bool os::getenv(const char* name, char* buf, int len) {
 204   const char* val = ::getenv(name);
 205   if (val != NULL && strlen(val) < (size_t)len) {
 206     strcpy(buf, val);
 207     return true;
 208   }
 209   if (len > 0) buf[0] = 0;  // return a null string
 210   return false;
 211 }
 212 
 213 
 214 // Return true if user is running as root.
 215 
 216 bool os::have_special_privileges() {
 217   static bool init = false;
 218   static bool privileges = false;
 219   if (!init) {
 220     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 221     init = true;
 222   }
 223   return privileges;
 224 }
 225 
 226 
 227 #ifndef SYS_gettid
 228 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 229 #ifdef __ia64__
 230 #define SYS_gettid 1105
 231 #elif __i386__
 232 #define SYS_gettid 224
 233 #elif __amd64__
 234 #define SYS_gettid 186
 235 #elif __sparc__
 236 #define SYS_gettid 143
 237 #else
 238 #error define gettid for the arch
 239 #endif
 240 #endif
 241 
 242 // Cpu architecture string
 243 #if   defined(ZERO)
 244 static char cpu_arch[] = ZERO_LIBARCH;
 245 #elif defined(IA64)
 246 static char cpu_arch[] = "ia64";
 247 #elif defined(IA32)
 248 static char cpu_arch[] = "i386";
 249 #elif defined(AMD64)
 250 static char cpu_arch[] = "amd64";
 251 #elif defined(ARM)
 252 static char cpu_arch[] = "arm";
 253 #elif defined(PPC)
 254 static char cpu_arch[] = "ppc";
 255 #elif defined(SPARC)
 256 #  ifdef _LP64
 257 static char cpu_arch[] = "sparcv9";
 258 #  else
 259 static char cpu_arch[] = "sparc";
 260 #  endif
 261 #else
 262 #error Add appropriate cpu_arch setting
 263 #endif
 264 
 265 
 266 // pid_t gettid()
 267 //
 268 // Returns the kernel thread id of the currently running thread. Kernel
 269 // thread id is used to access /proc.
 270 //
 271 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 272 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 273 //
 274 pid_t os::Linux::gettid() {
 275   int rslt = syscall(SYS_gettid);
 276   if (rslt == -1) {
 277      // old kernel, no NPTL support
 278      return getpid();
 279   } else {
 280      return (pid_t)rslt;
 281   }
 282 }
 283 
 284 // Most versions of linux have a bug where the number of processors are
 285 // determined by looking at the /proc file system.  In a chroot environment,
 286 // the system call returns 1.  This causes the VM to act as if it is
 287 // a single processor and elide locking (see is_MP() call).
 288 static bool unsafe_chroot_detected = false;
 289 static const char *unstable_chroot_error = "/proc file system not found.\n"
 290                      "Java may be unstable running multithreaded in a chroot "
 291                      "environment on Linux when /proc filesystem is not mounted.";
 292 
 293 void os::Linux::initialize_system_info() {
 294   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 295   if (processor_count() == 1) {
 296     pid_t pid = os::Linux::gettid();
 297     char fname[32];
 298     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 299     FILE *fp = fopen(fname, "r");
 300     if (fp == NULL) {
 301       unsafe_chroot_detected = true;
 302     } else {
 303       fclose(fp);
 304     }
 305   }
 306   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 307   assert(processor_count() > 0, "linux error");
 308 }
 309 
 310 void os::init_system_properties_values() {
 311 //  char arch[12];
 312 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 313 
 314   // The next steps are taken in the product version:
 315   //
 316   // Obtain the JAVA_HOME value from the location of libjvm.so.
 317   // This library should be located at:
 318   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 319   //
 320   // If "/jre/lib/" appears at the right place in the path, then we
 321   // assume libjvm.so is installed in a JDK and we use this path.
 322   //
 323   // Otherwise exit with message: "Could not create the Java virtual machine."
 324   //
 325   // The following extra steps are taken in the debugging version:
 326   //
 327   // If "/jre/lib/" does NOT appear at the right place in the path
 328   // instead of exit check for $JAVA_HOME environment variable.
 329   //
 330   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 331   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 332   // it looks like libjvm.so is installed there
 333   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 334   //
 335   // Otherwise exit.
 336   //
 337   // Important note: if the location of libjvm.so changes this
 338   // code needs to be changed accordingly.
 339 
 340   // The next few definitions allow the code to be verbatim:
 341 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
 342 #define getenv(n) ::getenv(n)
 343 
 344 /*
 345  * See ld(1):
 346  *      The linker uses the following search paths to locate required
 347  *      shared libraries:
 348  *        1: ...
 349  *        ...
 350  *        7: The default directories, normally /lib and /usr/lib.
 351  */
 352 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 353 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 354 #else
 355 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 356 #endif
 357 
 358 #define EXTENSIONS_DIR  "/lib/ext"
 359 #define ENDORSED_DIR    "/lib/endorsed"
 360 #define REG_DIR         "/usr/java/packages"
 361 
 362   {
 363     /* sysclasspath, java_home, dll_dir */
 364     {
 365         char *home_path;
 366         char *dll_path;
 367         char *pslash;
 368         char buf[MAXPATHLEN];
 369         os::jvm_path(buf, sizeof(buf));
 370 
 371         // Found the full path to libjvm.so.
 372         // Now cut the path to <java_home>/jre if we can.
 373         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 374         pslash = strrchr(buf, '/');
 375         if (pslash != NULL)
 376             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 377         dll_path = malloc(strlen(buf) + 1);
 378         if (dll_path == NULL)
 379             return;
 380         strcpy(dll_path, buf);
 381         Arguments::set_dll_dir(dll_path);
 382 
 383         if (pslash != NULL) {
 384             pslash = strrchr(buf, '/');
 385             if (pslash != NULL) {
 386                 *pslash = '\0';       /* get rid of /<arch> */
 387                 pslash = strrchr(buf, '/');
 388                 if (pslash != NULL)
 389                     *pslash = '\0';   /* get rid of /lib */
 390             }
 391         }
 392 
 393         home_path = malloc(strlen(buf) + 1);
 394         if (home_path == NULL)
 395             return;
 396         strcpy(home_path, buf);
 397         Arguments::set_java_home(home_path);
 398 
 399         if (!set_boot_path('/', ':'))
 400             return;
 401     }
 402 
 403     /*
 404      * Where to look for native libraries
 405      *
 406      * Note: Due to a legacy implementation, most of the library path
 407      * is set in the launcher.  This was to accomodate linking restrictions
 408      * on legacy Linux implementations (which are no longer supported).
 409      * Eventually, all the library path setting will be done here.
 410      *
 411      * However, to prevent the proliferation of improperly built native
 412      * libraries, the new path component /usr/java/packages is added here.
 413      * Eventually, all the library path setting will be done here.
 414      */
 415     {
 416         char *ld_library_path;
 417 
 418         /*
 419          * Construct the invariant part of ld_library_path. Note that the
 420          * space for the colon and the trailing null are provided by the
 421          * nulls included by the sizeof operator (so actually we allocate
 422          * a byte more than necessary).
 423          */
 424         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 425             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 426         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 427 
 428         /*
 429          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 430          * should always exist (until the legacy problem cited above is
 431          * addressed).
 432          */
 433         char *v = getenv("LD_LIBRARY_PATH");
 434         if (v != NULL) {
 435             char *t = ld_library_path;
 436             /* That's +1 for the colon and +1 for the trailing '\0' */
 437             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 438             sprintf(ld_library_path, "%s:%s", v, t);
 439         }
 440         Arguments::set_library_path(ld_library_path);
 441     }
 442 
 443     /*
 444      * Extensions directories.
 445      *
 446      * Note that the space for the colon and the trailing null are provided
 447      * by the nulls included by the sizeof operator (so actually one byte more
 448      * than necessary is allocated).
 449      */
 450     {
 451         char *buf = malloc(strlen(Arguments::get_java_home()) +
 452             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 453         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 454             Arguments::get_java_home());
 455         Arguments::set_ext_dirs(buf);
 456     }
 457 
 458     /* Endorsed standards default directory. */
 459     {
 460         char * buf;
 461         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 462         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 463         Arguments::set_endorsed_dirs(buf);
 464     }
 465   }
 466 
 467 #undef malloc
 468 #undef getenv
 469 #undef EXTENSIONS_DIR
 470 #undef ENDORSED_DIR
 471 
 472   // Done
 473   return;
 474 }
 475 
 476 ////////////////////////////////////////////////////////////////////////////////
 477 // breakpoint support
 478 
 479 void os::breakpoint() {
 480   BREAKPOINT;
 481 }
 482 
 483 extern "C" void breakpoint() {
 484   // use debugger to set breakpoint here
 485 }
 486 
 487 ////////////////////////////////////////////////////////////////////////////////
 488 // signal support
 489 
 490 debug_only(static bool signal_sets_initialized = false);
 491 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 492 
 493 bool os::Linux::is_sig_ignored(int sig) {
 494       struct sigaction oact;
 495       sigaction(sig, (struct sigaction*)NULL, &oact);
 496       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 497                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 498       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 499            return true;
 500       else
 501            return false;
 502 }
 503 
 504 void os::Linux::signal_sets_init() {
 505   // Should also have an assertion stating we are still single-threaded.
 506   assert(!signal_sets_initialized, "Already initialized");
 507   // Fill in signals that are necessarily unblocked for all threads in
 508   // the VM. Currently, we unblock the following signals:
 509   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 510   //                         by -Xrs (=ReduceSignalUsage));
 511   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 512   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 513   // the dispositions or masks wrt these signals.
 514   // Programs embedding the VM that want to use the above signals for their
 515   // own purposes must, at this time, use the "-Xrs" option to prevent
 516   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 517   // (See bug 4345157, and other related bugs).
 518   // In reality, though, unblocking these signals is really a nop, since
 519   // these signals are not blocked by default.
 520   sigemptyset(&unblocked_sigs);
 521   sigemptyset(&allowdebug_blocked_sigs);
 522   sigaddset(&unblocked_sigs, SIGILL);
 523   sigaddset(&unblocked_sigs, SIGSEGV);
 524   sigaddset(&unblocked_sigs, SIGBUS);
 525   sigaddset(&unblocked_sigs, SIGFPE);
 526   sigaddset(&unblocked_sigs, SR_signum);
 527 
 528   if (!ReduceSignalUsage) {
 529    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 530       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 531       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 532    }
 533    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 534       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 535       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 536    }
 537    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 538       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 539       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 540    }
 541   }
 542   // Fill in signals that are blocked by all but the VM thread.
 543   sigemptyset(&vm_sigs);
 544   if (!ReduceSignalUsage)
 545     sigaddset(&vm_sigs, BREAK_SIGNAL);
 546   debug_only(signal_sets_initialized = true);
 547 
 548 }
 549 
 550 // These are signals that are unblocked while a thread is running Java.
 551 // (For some reason, they get blocked by default.)
 552 sigset_t* os::Linux::unblocked_signals() {
 553   assert(signal_sets_initialized, "Not initialized");
 554   return &unblocked_sigs;
 555 }
 556 
 557 // These are the signals that are blocked while a (non-VM) thread is
 558 // running Java. Only the VM thread handles these signals.
 559 sigset_t* os::Linux::vm_signals() {
 560   assert(signal_sets_initialized, "Not initialized");
 561   return &vm_sigs;
 562 }
 563 
 564 // These are signals that are blocked during cond_wait to allow debugger in
 565 sigset_t* os::Linux::allowdebug_blocked_signals() {
 566   assert(signal_sets_initialized, "Not initialized");
 567   return &allowdebug_blocked_sigs;
 568 }
 569 
 570 void os::Linux::hotspot_sigmask(Thread* thread) {
 571 
 572   //Save caller's signal mask before setting VM signal mask
 573   sigset_t caller_sigmask;
 574   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 575 
 576   OSThread* osthread = thread->osthread();
 577   osthread->set_caller_sigmask(caller_sigmask);
 578 
 579   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 580 
 581   if (!ReduceSignalUsage) {
 582     if (thread->is_VM_thread()) {
 583       // Only the VM thread handles BREAK_SIGNAL ...
 584       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 585     } else {
 586       // ... all other threads block BREAK_SIGNAL
 587       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 588     }
 589   }
 590 }
 591 
 592 //////////////////////////////////////////////////////////////////////////////
 593 // detecting pthread library
 594 
 595 void os::Linux::libpthread_init() {
 596   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 597   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 598   // generic name for earlier versions.
 599   // Define macros here so we can build HotSpot on old systems.
 600 # ifndef _CS_GNU_LIBC_VERSION
 601 # define _CS_GNU_LIBC_VERSION 2
 602 # endif
 603 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 604 # define _CS_GNU_LIBPTHREAD_VERSION 3
 605 # endif
 606 
 607   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 608   if (n > 0) {
 609      char *str = (char *)malloc(n, mtInternal);
 610      confstr(_CS_GNU_LIBC_VERSION, str, n);
 611      os::Linux::set_glibc_version(str);
 612   } else {
 613      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 614      static char _gnu_libc_version[32];
 615      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 616               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 617      os::Linux::set_glibc_version(_gnu_libc_version);
 618   }
 619 
 620   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 621   if (n > 0) {
 622      char *str = (char *)malloc(n, mtInternal);
 623      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 624      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 625      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 626      // is the case. LinuxThreads has a hard limit on max number of threads.
 627      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 628      // On the other hand, NPTL does not have such a limit, sysconf()
 629      // will return -1 and errno is not changed. Check if it is really NPTL.
 630      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 631          strstr(str, "NPTL") &&
 632          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 633        free(str);
 634        os::Linux::set_libpthread_version("linuxthreads");
 635      } else {
 636        os::Linux::set_libpthread_version(str);
 637      }
 638   } else {
 639     // glibc before 2.3.2 only has LinuxThreads.
 640     os::Linux::set_libpthread_version("linuxthreads");
 641   }
 642 
 643   if (strstr(libpthread_version(), "NPTL")) {
 644      os::Linux::set_is_NPTL();
 645   } else {
 646      os::Linux::set_is_LinuxThreads();
 647   }
 648 
 649   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 650   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 651   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 652      os::Linux::set_is_floating_stack();
 653   }
 654 }
 655 
 656 /////////////////////////////////////////////////////////////////////////////
 657 // thread stack
 658 
 659 // Force Linux kernel to expand current thread stack. If "bottom" is close
 660 // to the stack guard, caller should block all signals.
 661 //
 662 // MAP_GROWSDOWN:
 663 //   A special mmap() flag that is used to implement thread stacks. It tells
 664 //   kernel that the memory region should extend downwards when needed. This
 665 //   allows early versions of LinuxThreads to only mmap the first few pages
 666 //   when creating a new thread. Linux kernel will automatically expand thread
 667 //   stack as needed (on page faults).
 668 //
 669 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 670 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 671 //   region, it's hard to tell if the fault is due to a legitimate stack
 672 //   access or because of reading/writing non-exist memory (e.g. buffer
 673 //   overrun). As a rule, if the fault happens below current stack pointer,
 674 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 675 //   application (see Linux kernel fault.c).
 676 //
 677 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 678 //   stack overflow detection.
 679 //
 680 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 681 //   not use this flag. However, the stack of initial thread is not created
 682 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 683 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 684 //   and then attach the thread to JVM.
 685 //
 686 // To get around the problem and allow stack banging on Linux, we need to
 687 // manually expand thread stack after receiving the SIGSEGV.
 688 //
 689 // There are two ways to expand thread stack to address "bottom", we used
 690 // both of them in JVM before 1.5:
 691 //   1. adjust stack pointer first so that it is below "bottom", and then
 692 //      touch "bottom"
 693 //   2. mmap() the page in question
 694 //
 695 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 696 // if current sp is already near the lower end of page 101, and we need to
 697 // call mmap() to map page 100, it is possible that part of the mmap() frame
 698 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 699 // That will destroy the mmap() frame and cause VM to crash.
 700 //
 701 // The following code works by adjusting sp first, then accessing the "bottom"
 702 // page to force a page fault. Linux kernel will then automatically expand the
 703 // stack mapping.
 704 //
 705 // _expand_stack_to() assumes its frame size is less than page size, which
 706 // should always be true if the function is not inlined.
 707 
 708 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 709 #define NOINLINE
 710 #else
 711 #define NOINLINE __attribute__ ((noinline))
 712 #endif
 713 
 714 static void _expand_stack_to(address bottom) NOINLINE;
 715 
 716 static void _expand_stack_to(address bottom) {
 717   address sp;
 718   size_t size;
 719   volatile char *p;
 720 
 721   // Adjust bottom to point to the largest address within the same page, it
 722   // gives us a one-page buffer if alloca() allocates slightly more memory.
 723   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 724   bottom += os::Linux::page_size() - 1;
 725 
 726   // sp might be slightly above current stack pointer; if that's the case, we
 727   // will alloca() a little more space than necessary, which is OK. Don't use
 728   // os::current_stack_pointer(), as its result can be slightly below current
 729   // stack pointer, causing us to not alloca enough to reach "bottom".
 730   sp = (address)&sp;
 731 
 732   if (sp > bottom) {
 733     size = sp - bottom;
 734     p = (volatile char *)alloca(size);
 735     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 736     p[0] = '\0';
 737   }
 738 }
 739 
 740 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 741   assert(t!=NULL, "just checking");
 742   assert(t->osthread()->expanding_stack(), "expand should be set");
 743   assert(t->stack_base() != NULL, "stack_base was not initialized");
 744 
 745   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 746     sigset_t mask_all, old_sigset;
 747     sigfillset(&mask_all);
 748     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 749     _expand_stack_to(addr);
 750     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 751     return true;
 752   }
 753   return false;
 754 }
 755 
 756 //////////////////////////////////////////////////////////////////////////////
 757 // create new thread
 758 
 759 static address highest_vm_reserved_address();
 760 
 761 // check if it's safe to start a new thread
 762 static bool _thread_safety_check(Thread* thread) {
 763   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 764     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 765     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 766     //   allocated (MAP_FIXED) from high address space. Every thread stack
 767     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 768     //   it to other values if they rebuild LinuxThreads).
 769     //
 770     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 771     // the memory region has already been mmap'ed. That means if we have too
 772     // many threads and/or very large heap, eventually thread stack will
 773     // collide with heap.
 774     //
 775     // Here we try to prevent heap/stack collision by comparing current
 776     // stack bottom with the highest address that has been mmap'ed by JVM
 777     // plus a safety margin for memory maps created by native code.
 778     //
 779     // This feature can be disabled by setting ThreadSafetyMargin to 0
 780     //
 781     if (ThreadSafetyMargin > 0) {
 782       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 783 
 784       // not safe if our stack extends below the safety margin
 785       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 786     } else {
 787       return true;
 788     }
 789   } else {
 790     // Floating stack LinuxThreads or NPTL:
 791     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 792     //   there's not enough space left, pthread_create() will fail. If we come
 793     //   here, that means enough space has been reserved for stack.
 794     return true;
 795   }
 796 }
 797 
 798 // Thread start routine for all newly created threads
 799 static void *java_start(Thread *thread) {
 800   // Try to randomize the cache line index of hot stack frames.
 801   // This helps when threads of the same stack traces evict each other's
 802   // cache lines. The threads can be either from the same JVM instance, or
 803   // from different JVM instances. The benefit is especially true for
 804   // processors with hyperthreading technology.
 805   static int counter = 0;
 806   int pid = os::current_process_id();
 807   alloca(((pid ^ counter++) & 7) * 128);
 808 
 809   ThreadLocalStorage::set_thread(thread);
 810 
 811   OSThread* osthread = thread->osthread();
 812   Monitor* sync = osthread->startThread_lock();
 813 
 814   // non floating stack LinuxThreads needs extra check, see above
 815   if (!_thread_safety_check(thread)) {
 816     // notify parent thread
 817     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 818     osthread->set_state(ZOMBIE);
 819     sync->notify_all();
 820     return NULL;
 821   }
 822 
 823   // thread_id is kernel thread id (similar to Solaris LWP id)
 824   osthread->set_thread_id(os::Linux::gettid());
 825 
 826   if (UseNUMA) {
 827     int lgrp_id = os::numa_get_group_id();
 828     if (lgrp_id != -1) {
 829       thread->set_lgrp_id(lgrp_id);
 830     }
 831   }
 832   // initialize signal mask for this thread
 833   os::Linux::hotspot_sigmask(thread);
 834 
 835   // initialize floating point control register
 836   os::Linux::init_thread_fpu_state();
 837 
 838   // handshaking with parent thread
 839   {
 840     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 841 
 842     // notify parent thread
 843     osthread->set_state(INITIALIZED);
 844     sync->notify_all();
 845 
 846     // wait until os::start_thread()
 847     while (osthread->get_state() == INITIALIZED) {
 848       sync->wait(Mutex::_no_safepoint_check_flag);
 849     }
 850   }
 851 
 852   // call one more level start routine
 853   thread->run();
 854 
 855   return 0;
 856 }
 857 
 858 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 859   assert(thread->osthread() == NULL, "caller responsible");
 860 
 861   // Allocate the OSThread object
 862   OSThread* osthread = new OSThread(NULL, NULL);
 863   if (osthread == NULL) {
 864     return false;
 865   }
 866 
 867   // set the correct thread state
 868   osthread->set_thread_type(thr_type);
 869 
 870   // Initial state is ALLOCATED but not INITIALIZED
 871   osthread->set_state(ALLOCATED);
 872 
 873   thread->set_osthread(osthread);
 874 
 875   // init thread attributes
 876   pthread_attr_t attr;
 877   pthread_attr_init(&attr);
 878   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 879 
 880   // stack size
 881   if (os::Linux::supports_variable_stack_size()) {
 882     // calculate stack size if it's not specified by caller
 883     if (stack_size == 0) {
 884       stack_size = os::Linux::default_stack_size(thr_type);
 885 
 886       switch (thr_type) {
 887       case os::java_thread:
 888         // Java threads use ThreadStackSize which default value can be
 889         // changed with the flag -Xss
 890         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 891         stack_size = JavaThread::stack_size_at_create();
 892         break;
 893       case os::compiler_thread:
 894         if (CompilerThreadStackSize > 0) {
 895           stack_size = (size_t)(CompilerThreadStackSize * K);
 896           break;
 897         } // else fall through:
 898           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 899       case os::vm_thread:
 900       case os::pgc_thread:
 901       case os::cgc_thread:
 902       case os::watcher_thread:
 903         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 904         break;
 905       }
 906     }
 907 
 908     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 909     pthread_attr_setstacksize(&attr, stack_size);
 910   } else {
 911     // let pthread_create() pick the default value.
 912   }
 913 
 914   // glibc guard page
 915   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 916 
 917   ThreadState state;
 918 
 919   {
 920     // Serialize thread creation if we are running with fixed stack LinuxThreads
 921     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 922     if (lock) {
 923       os::Linux::createThread_lock()->lock_without_safepoint_check();
 924     }
 925 
 926     pthread_t tid;
 927     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 928 
 929     pthread_attr_destroy(&attr);
 930 
 931     if (ret != 0) {
 932       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 933         perror("pthread_create()");
 934       }
 935       // Need to clean up stuff we've allocated so far
 936       thread->set_osthread(NULL);
 937       delete osthread;
 938       if (lock) os::Linux::createThread_lock()->unlock();
 939       return false;
 940     }
 941 
 942     // Store pthread info into the OSThread
 943     osthread->set_pthread_id(tid);
 944 
 945     // Wait until child thread is either initialized or aborted
 946     {
 947       Monitor* sync_with_child = osthread->startThread_lock();
 948       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 949       while ((state = osthread->get_state()) == ALLOCATED) {
 950         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 951       }
 952     }
 953 
 954     if (lock) {
 955       os::Linux::createThread_lock()->unlock();
 956     }
 957   }
 958 
 959   // Aborted due to thread limit being reached
 960   if (state == ZOMBIE) {
 961       thread->set_osthread(NULL);
 962       delete osthread;
 963       return false;
 964   }
 965 
 966   // The thread is returned suspended (in state INITIALIZED),
 967   // and is started higher up in the call chain
 968   assert(state == INITIALIZED, "race condition");
 969   return true;
 970 }
 971 
 972 /////////////////////////////////////////////////////////////////////////////
 973 // attach existing thread
 974 
 975 // bootstrap the main thread
 976 bool os::create_main_thread(JavaThread* thread) {
 977   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 978   return create_attached_thread(thread);
 979 }
 980 
 981 bool os::create_attached_thread(JavaThread* thread) {
 982 #ifdef ASSERT
 983     thread->verify_not_published();
 984 #endif
 985 
 986   // Allocate the OSThread object
 987   OSThread* osthread = new OSThread(NULL, NULL);
 988 
 989   if (osthread == NULL) {
 990     return false;
 991   }
 992 
 993   // Store pthread info into the OSThread
 994   osthread->set_thread_id(os::Linux::gettid());
 995   osthread->set_pthread_id(::pthread_self());
 996 
 997   // initialize floating point control register
 998   os::Linux::init_thread_fpu_state();
 999 
1000   // Initial thread state is RUNNABLE
1001   osthread->set_state(RUNNABLE);
1002 
1003   thread->set_osthread(osthread);
1004 
1005   if (UseNUMA) {
1006     int lgrp_id = os::numa_get_group_id();
1007     if (lgrp_id != -1) {
1008       thread->set_lgrp_id(lgrp_id);
1009     }
1010   }
1011 
1012   if (os::Linux::is_initial_thread()) {
1013     // If current thread is initial thread, its stack is mapped on demand,
1014     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1015     // the entire stack region to avoid SEGV in stack banging.
1016     // It is also useful to get around the heap-stack-gap problem on SuSE
1017     // kernel (see 4821821 for details). We first expand stack to the top
1018     // of yellow zone, then enable stack yellow zone (order is significant,
1019     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1020     // is no gap between the last two virtual memory regions.
1021 
1022     JavaThread *jt = (JavaThread *)thread;
1023     address addr = jt->stack_yellow_zone_base();
1024     assert(addr != NULL, "initialization problem?");
1025     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1026 
1027     osthread->set_expanding_stack();
1028     os::Linux::manually_expand_stack(jt, addr);
1029     osthread->clear_expanding_stack();
1030   }
1031 
1032   // initialize signal mask for this thread
1033   // and save the caller's signal mask
1034   os::Linux::hotspot_sigmask(thread);
1035 
1036   return true;
1037 }
1038 
1039 void os::pd_start_thread(Thread* thread) {
1040   OSThread * osthread = thread->osthread();
1041   assert(osthread->get_state() != INITIALIZED, "just checking");
1042   Monitor* sync_with_child = osthread->startThread_lock();
1043   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1044   sync_with_child->notify();
1045 }
1046 
1047 // Free Linux resources related to the OSThread
1048 void os::free_thread(OSThread* osthread) {
1049   assert(osthread != NULL, "osthread not set");
1050 
1051   if (Thread::current()->osthread() == osthread) {
1052     // Restore caller's signal mask
1053     sigset_t sigmask = osthread->caller_sigmask();
1054     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1055    }
1056 
1057   delete osthread;
1058 }
1059 
1060 //////////////////////////////////////////////////////////////////////////////
1061 // thread local storage
1062 
1063 int os::allocate_thread_local_storage() {
1064   pthread_key_t key;
1065   int rslt = pthread_key_create(&key, NULL);
1066   assert(rslt == 0, "cannot allocate thread local storage");
1067   return (int)key;
1068 }
1069 
1070 // Note: This is currently not used by VM, as we don't destroy TLS key
1071 // on VM exit.
1072 void os::free_thread_local_storage(int index) {
1073   int rslt = pthread_key_delete((pthread_key_t)index);
1074   assert(rslt == 0, "invalid index");
1075 }
1076 
1077 void os::thread_local_storage_at_put(int index, void* value) {
1078   int rslt = pthread_setspecific((pthread_key_t)index, value);
1079   assert(rslt == 0, "pthread_setspecific failed");
1080 }
1081 
1082 extern "C" Thread* get_thread() {
1083   return ThreadLocalStorage::thread();
1084 }
1085 
1086 //////////////////////////////////////////////////////////////////////////////
1087 // initial thread
1088 
1089 // Check if current thread is the initial thread, similar to Solaris thr_main.
1090 bool os::Linux::is_initial_thread(void) {
1091   char dummy;
1092   // If called before init complete, thread stack bottom will be null.
1093   // Can be called if fatal error occurs before initialization.
1094   if (initial_thread_stack_bottom() == NULL) return false;
1095   assert(initial_thread_stack_bottom() != NULL &&
1096          initial_thread_stack_size()   != 0,
1097          "os::init did not locate initial thread's stack region");
1098   if ((address)&dummy >= initial_thread_stack_bottom() &&
1099       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1100        return true;
1101   else return false;
1102 }
1103 
1104 // Find the virtual memory area that contains addr
1105 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1106   FILE *fp = fopen("/proc/self/maps", "r");
1107   if (fp) {
1108     address low, high;
1109     while (!feof(fp)) {
1110       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1111         if (low <= addr && addr < high) {
1112            if (vma_low)  *vma_low  = low;
1113            if (vma_high) *vma_high = high;
1114            fclose (fp);
1115            return true;
1116         }
1117       }
1118       for (;;) {
1119         int ch = fgetc(fp);
1120         if (ch == EOF || ch == (int)'\n') break;
1121       }
1122     }
1123     fclose(fp);
1124   }
1125   return false;
1126 }
1127 
1128 // Locate initial thread stack. This special handling of initial thread stack
1129 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1130 // bogus value for initial thread.
1131 void os::Linux::capture_initial_stack(size_t max_size) {
1132   // stack size is the easy part, get it from RLIMIT_STACK
1133   size_t stack_size;
1134   struct rlimit rlim;
1135   getrlimit(RLIMIT_STACK, &rlim);
1136   stack_size = rlim.rlim_cur;
1137 
1138   // 6308388: a bug in ld.so will relocate its own .data section to the
1139   //   lower end of primordial stack; reduce ulimit -s value a little bit
1140   //   so we won't install guard page on ld.so's data section.
1141   stack_size -= 2 * page_size();
1142 
1143   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1144   //   7.1, in both cases we will get 2G in return value.
1145   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1146   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1147   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1148   //   in case other parts in glibc still assumes 2M max stack size.
1149   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1150   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1151   if (stack_size > 2 * K * K IA64_ONLY(*2))
1152       stack_size = 2 * K * K IA64_ONLY(*2);
1153   // Try to figure out where the stack base (top) is. This is harder.
1154   //
1155   // When an application is started, glibc saves the initial stack pointer in
1156   // a global variable "__libc_stack_end", which is then used by system
1157   // libraries. __libc_stack_end should be pretty close to stack top. The
1158   // variable is available since the very early days. However, because it is
1159   // a private interface, it could disappear in the future.
1160   //
1161   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1162   // to __libc_stack_end, it is very close to stack top, but isn't the real
1163   // stack top. Note that /proc may not exist if VM is running as a chroot
1164   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1165   // /proc/<pid>/stat could change in the future (though unlikely).
1166   //
1167   // We try __libc_stack_end first. If that doesn't work, look for
1168   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1169   // as a hint, which should work well in most cases.
1170 
1171   uintptr_t stack_start;
1172 
1173   // try __libc_stack_end first
1174   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1175   if (p && *p) {
1176     stack_start = *p;
1177   } else {
1178     // see if we can get the start_stack field from /proc/self/stat
1179     FILE *fp;
1180     int pid;
1181     char state;
1182     int ppid;
1183     int pgrp;
1184     int session;
1185     int nr;
1186     int tpgrp;
1187     unsigned long flags;
1188     unsigned long minflt;
1189     unsigned long cminflt;
1190     unsigned long majflt;
1191     unsigned long cmajflt;
1192     unsigned long utime;
1193     unsigned long stime;
1194     long cutime;
1195     long cstime;
1196     long prio;
1197     long nice;
1198     long junk;
1199     long it_real;
1200     uintptr_t start;
1201     uintptr_t vsize;
1202     intptr_t rss;
1203     uintptr_t rsslim;
1204     uintptr_t scodes;
1205     uintptr_t ecode;
1206     int i;
1207 
1208     // Figure what the primordial thread stack base is. Code is inspired
1209     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1210     // followed by command name surrounded by parentheses, state, etc.
1211     char stat[2048];
1212     int statlen;
1213 
1214     fp = fopen("/proc/self/stat", "r");
1215     if (fp) {
1216       statlen = fread(stat, 1, 2047, fp);
1217       stat[statlen] = '\0';
1218       fclose(fp);
1219 
1220       // Skip pid and the command string. Note that we could be dealing with
1221       // weird command names, e.g. user could decide to rename java launcher
1222       // to "java 1.4.2 :)", then the stat file would look like
1223       //                1234 (java 1.4.2 :)) R ... ...
1224       // We don't really need to know the command string, just find the last
1225       // occurrence of ")" and then start parsing from there. See bug 4726580.
1226       char * s = strrchr(stat, ')');
1227 
1228       i = 0;
1229       if (s) {
1230         // Skip blank chars
1231         do s++; while (isspace(*s));
1232 
1233 #define _UFM UINTX_FORMAT
1234 #define _DFM INTX_FORMAT
1235 
1236         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1237         /*              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 */
1238         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,
1239              &state,          /* 3  %c  */
1240              &ppid,           /* 4  %d  */
1241              &pgrp,           /* 5  %d  */
1242              &session,        /* 6  %d  */
1243              &nr,             /* 7  %d  */
1244              &tpgrp,          /* 8  %d  */
1245              &flags,          /* 9  %lu  */
1246              &minflt,         /* 10 %lu  */
1247              &cminflt,        /* 11 %lu  */
1248              &majflt,         /* 12 %lu  */
1249              &cmajflt,        /* 13 %lu  */
1250              &utime,          /* 14 %lu  */
1251              &stime,          /* 15 %lu  */
1252              &cutime,         /* 16 %ld  */
1253              &cstime,         /* 17 %ld  */
1254              &prio,           /* 18 %ld  */
1255              &nice,           /* 19 %ld  */
1256              &junk,           /* 20 %ld  */
1257              &it_real,        /* 21 %ld  */
1258              &start,          /* 22 UINTX_FORMAT */
1259              &vsize,          /* 23 UINTX_FORMAT */
1260              &rss,            /* 24 INTX_FORMAT  */
1261              &rsslim,         /* 25 UINTX_FORMAT */
1262              &scodes,         /* 26 UINTX_FORMAT */
1263              &ecode,          /* 27 UINTX_FORMAT */
1264              &stack_start);   /* 28 UINTX_FORMAT */
1265       }
1266 
1267 #undef _UFM
1268 #undef _DFM
1269 
1270       if (i != 28 - 2) {
1271          assert(false, "Bad conversion from /proc/self/stat");
1272          // product mode - assume we are the initial thread, good luck in the
1273          // embedded case.
1274          warning("Can't detect initial thread stack location - bad conversion");
1275          stack_start = (uintptr_t) &rlim;
1276       }
1277     } else {
1278       // For some reason we can't open /proc/self/stat (for example, running on
1279       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1280       // most cases, so don't abort:
1281       warning("Can't detect initial thread stack location - no /proc/self/stat");
1282       stack_start = (uintptr_t) &rlim;
1283     }
1284   }
1285 
1286   // Now we have a pointer (stack_start) very close to the stack top, the
1287   // next thing to do is to figure out the exact location of stack top. We
1288   // can find out the virtual memory area that contains stack_start by
1289   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1290   // and its upper limit is the real stack top. (again, this would fail if
1291   // running inside chroot, because /proc may not exist.)
1292 
1293   uintptr_t stack_top;
1294   address low, high;
1295   if (find_vma((address)stack_start, &low, &high)) {
1296     // success, "high" is the true stack top. (ignore "low", because initial
1297     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1298     stack_top = (uintptr_t)high;
1299   } else {
1300     // failed, likely because /proc/self/maps does not exist
1301     warning("Can't detect initial thread stack location - find_vma failed");
1302     // best effort: stack_start is normally within a few pages below the real
1303     // stack top, use it as stack top, and reduce stack size so we won't put
1304     // guard page outside stack.
1305     stack_top = stack_start;
1306     stack_size -= 16 * page_size();
1307   }
1308 
1309   // stack_top could be partially down the page so align it
1310   stack_top = align_size_up(stack_top, page_size());
1311 
1312   if (max_size && stack_size > max_size) {
1313      _initial_thread_stack_size = max_size;
1314   } else {
1315      _initial_thread_stack_size = stack_size;
1316   }
1317 
1318   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1319   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1320 }
1321 
1322 ////////////////////////////////////////////////////////////////////////////////
1323 // time support
1324 
1325 // Time since start-up in seconds to a fine granularity.
1326 // Used by VMSelfDestructTimer and the MemProfiler.
1327 double os::elapsedTime() {
1328 
1329   return (double)(os::elapsed_counter()) * 0.000001;
1330 }
1331 
1332 jlong os::elapsed_counter() {
1333   timeval time;
1334   int status = gettimeofday(&time, NULL);
1335   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1336 }
1337 
1338 jlong os::elapsed_frequency() {
1339   return (1000 * 1000);
1340 }
1341 
1342 // For now, we say that linux does not support vtime.  I have no idea
1343 // whether it can actually be made to (DLD, 9/13/05).
1344 
1345 bool os::supports_vtime() { return false; }
1346 bool os::enable_vtime()   { return false; }
1347 bool os::vtime_enabled()  { return false; }
1348 double os::elapsedVTime() {
1349   // better than nothing, but not much
1350   return elapsedTime();
1351 }
1352 
1353 jlong os::javaTimeMillis() {
1354   timeval time;
1355   int status = gettimeofday(&time, NULL);
1356   assert(status != -1, "linux error");
1357   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1358 }
1359 
1360 #ifndef CLOCK_MONOTONIC
1361 #define CLOCK_MONOTONIC (1)
1362 #endif
1363 
1364 void os::Linux::clock_init() {
1365   // we do dlopen's in this particular order due to bug in linux
1366   // dynamical loader (see 6348968) leading to crash on exit
1367   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1368   if (handle == NULL) {
1369     handle = dlopen("librt.so", RTLD_LAZY);
1370   }
1371 
1372   if (handle) {
1373     int (*clock_getres_func)(clockid_t, struct timespec*) =
1374            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1375     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1376            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1377     if (clock_getres_func && clock_gettime_func) {
1378       // See if monotonic clock is supported by the kernel. Note that some
1379       // early implementations simply return kernel jiffies (updated every
1380       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1381       // for nano time (though the monotonic property is still nice to have).
1382       // It's fixed in newer kernels, however clock_getres() still returns
1383       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1384       // resolution for now. Hopefully as people move to new kernels, this
1385       // won't be a problem.
1386       struct timespec res;
1387       struct timespec tp;
1388       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1389           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1390         // yes, monotonic clock is supported
1391         _clock_gettime = clock_gettime_func;
1392       } else {
1393         // close librt if there is no monotonic clock
1394         dlclose(handle);
1395       }
1396     }
1397   }
1398 }
1399 
1400 #ifndef SYS_clock_getres
1401 
1402 #if defined(IA32) || defined(AMD64)
1403 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1404 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1405 #else
1406 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1407 #define sys_clock_getres(x,y)  -1
1408 #endif
1409 
1410 #else
1411 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1412 #endif
1413 
1414 void os::Linux::fast_thread_clock_init() {
1415   if (!UseLinuxPosixThreadCPUClocks) {
1416     return;
1417   }
1418   clockid_t clockid;
1419   struct timespec tp;
1420   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1421       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1422 
1423   // Switch to using fast clocks for thread cpu time if
1424   // the sys_clock_getres() returns 0 error code.
1425   // Note, that some kernels may support the current thread
1426   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1427   // returned by the pthread_getcpuclockid().
1428   // If the fast Posix clocks are supported then the sys_clock_getres()
1429   // must return at least tp.tv_sec == 0 which means a resolution
1430   // better than 1 sec. This is extra check for reliability.
1431 
1432   if(pthread_getcpuclockid_func &&
1433      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1434      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1435 
1436     _supports_fast_thread_cpu_time = true;
1437     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1438   }
1439 }
1440 
1441 jlong os::javaTimeNanos() {
1442   if (Linux::supports_monotonic_clock()) {
1443     struct timespec tp;
1444     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1445     assert(status == 0, "gettime error");
1446     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1447     return result;
1448   } else {
1449     timeval time;
1450     int status = gettimeofday(&time, NULL);
1451     assert(status != -1, "linux error");
1452     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1453     return 1000 * usecs;
1454   }
1455 }
1456 
1457 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1458   if (Linux::supports_monotonic_clock()) {
1459     info_ptr->max_value = ALL_64_BITS;
1460 
1461     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1462     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1463     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1464   } else {
1465     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1466     info_ptr->max_value = ALL_64_BITS;
1467 
1468     // gettimeofday is a real time clock so it skips
1469     info_ptr->may_skip_backward = true;
1470     info_ptr->may_skip_forward = true;
1471   }
1472 
1473   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1474 }
1475 
1476 // Return the real, user, and system times in seconds from an
1477 // arbitrary fixed point in the past.
1478 bool os::getTimesSecs(double* process_real_time,
1479                       double* process_user_time,
1480                       double* process_system_time) {
1481   struct tms ticks;
1482   clock_t real_ticks = times(&ticks);
1483 
1484   if (real_ticks == (clock_t) (-1)) {
1485     return false;
1486   } else {
1487     double ticks_per_second = (double) clock_tics_per_sec;
1488     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1489     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1490     *process_real_time = ((double) real_ticks) / ticks_per_second;
1491 
1492     return true;
1493   }
1494 }
1495 
1496 
1497 char * os::local_time_string(char *buf, size_t buflen) {
1498   struct tm t;
1499   time_t long_time;
1500   time(&long_time);
1501   localtime_r(&long_time, &t);
1502   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1503                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1504                t.tm_hour, t.tm_min, t.tm_sec);
1505   return buf;
1506 }
1507 
1508 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1509   return localtime_r(clock, res);
1510 }
1511 
1512 ////////////////////////////////////////////////////////////////////////////////
1513 // runtime exit support
1514 
1515 // Note: os::shutdown() might be called very early during initialization, or
1516 // called from signal handler. Before adding something to os::shutdown(), make
1517 // sure it is async-safe and can handle partially initialized VM.
1518 void os::shutdown() {
1519 
1520   // allow PerfMemory to attempt cleanup of any persistent resources
1521   perfMemory_exit();
1522 
1523   // needs to remove object in file system
1524   AttachListener::abort();
1525 
1526   // flush buffered output, finish log files
1527   ostream_abort();
1528 
1529   // Check for abort hook
1530   abort_hook_t abort_hook = Arguments::abort_hook();
1531   if (abort_hook != NULL) {
1532     abort_hook();
1533   }
1534 
1535 }
1536 
1537 // Note: os::abort() might be called very early during initialization, or
1538 // called from signal handler. Before adding something to os::abort(), make
1539 // sure it is async-safe and can handle partially initialized VM.
1540 void os::abort(bool dump_core) {
1541   os::shutdown();
1542   if (dump_core) {
1543 #ifndef PRODUCT
1544     fdStream out(defaultStream::output_fd());
1545     out.print_raw("Current thread is ");
1546     char buf[16];
1547     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1548     out.print_raw_cr(buf);
1549     out.print_raw_cr("Dumping core ...");
1550 #endif
1551     ::abort(); // dump core
1552   }
1553 
1554   ::exit(1);
1555 }
1556 
1557 // Die immediately, no exit hook, no abort hook, no cleanup.
1558 void os::die() {
1559   // _exit() on LinuxThreads only kills current thread
1560   ::abort();
1561 }
1562 
1563 // unused on linux for now.
1564 void os::set_error_file(const char *logfile) {}
1565 
1566 
1567 // This method is a copy of JDK's sysGetLastErrorString
1568 // from src/solaris/hpi/src/system_md.c
1569 
1570 size_t os::lasterror(char *buf, size_t len) {
1571 
1572   if (errno == 0)  return 0;
1573 
1574   const char *s = ::strerror(errno);
1575   size_t n = ::strlen(s);
1576   if (n >= len) {
1577     n = len - 1;
1578   }
1579   ::strncpy(buf, s, n);
1580   buf[n] = '\0';
1581   return n;
1582 }
1583 
1584 intx os::current_thread_id() { return (intx)pthread_self(); }
1585 int os::current_process_id() {
1586 
1587   // Under the old linux thread library, linux gives each thread
1588   // its own process id. Because of this each thread will return
1589   // a different pid if this method were to return the result
1590   // of getpid(2). Linux provides no api that returns the pid
1591   // of the launcher thread for the vm. This implementation
1592   // returns a unique pid, the pid of the launcher thread
1593   // that starts the vm 'process'.
1594 
1595   // Under the NPTL, getpid() returns the same pid as the
1596   // launcher thread rather than a unique pid per thread.
1597   // Use gettid() if you want the old pre NPTL behaviour.
1598 
1599   // if you are looking for the result of a call to getpid() that
1600   // returns a unique pid for the calling thread, then look at the
1601   // OSThread::thread_id() method in osThread_linux.hpp file
1602 
1603   return (int)(_initial_pid ? _initial_pid : getpid());
1604 }
1605 
1606 // DLL functions
1607 
1608 const char* os::dll_file_extension() { return ".so"; }
1609 
1610 // This must be hard coded because it's the system's temporary
1611 // directory not the java application's temp directory, ala java.io.tmpdir.
1612 const char* os::get_temp_directory() { return "/tmp"; }
1613 
1614 static bool file_exists(const char* filename) {
1615   struct stat statbuf;
1616   if (filename == NULL || strlen(filename) == 0) {
1617     return false;
1618   }
1619   return os::stat(filename, &statbuf) == 0;
1620 }
1621 
1622 bool os::dll_build_name(char* buffer, size_t buflen,
1623                         const char* pname, const char* fname) {
1624   bool retval = false;
1625   // Copied from libhpi
1626   const size_t pnamelen = pname ? strlen(pname) : 0;
1627 
1628   // Return error on buffer overflow.
1629   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1630     return retval;
1631   }
1632 
1633   if (pnamelen == 0) {
1634     snprintf(buffer, buflen, "lib%s.so", fname);
1635     retval = true;
1636   } else if (strchr(pname, *os::path_separator()) != NULL) {
1637     int n;
1638     char** pelements = split_path(pname, &n);
1639     if (pelements == NULL) {
1640       return false;
1641     }
1642     for (int i = 0 ; i < n ; i++) {
1643       // Really shouldn't be NULL, but check can't hurt
1644       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1645         continue; // skip the empty path values
1646       }
1647       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1648       if (file_exists(buffer)) {
1649         retval = true;
1650         break;
1651       }
1652     }
1653     // release the storage
1654     for (int i = 0 ; i < n ; i++) {
1655       if (pelements[i] != NULL) {
1656         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1657       }
1658     }
1659     if (pelements != NULL) {
1660       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1661     }
1662   } else {
1663     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1664     retval = true;
1665   }
1666   return retval;
1667 }
1668 
1669 // check if addr is inside libjvm.so
1670 bool os::address_is_in_vm(address addr) {
1671   static address libjvm_base_addr;
1672   Dl_info dlinfo;
1673 
1674   if (libjvm_base_addr == NULL) {
1675     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1676     libjvm_base_addr = (address)dlinfo.dli_fbase;
1677     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1678   }
1679 
1680   if (dladdr((void *)addr, &dlinfo)) {
1681     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1682   }
1683 
1684   return false;
1685 }
1686 
1687 bool os::dll_address_to_function_name(address addr, char *buf,
1688                                       int buflen, int *offset) {
1689   Dl_info dlinfo;
1690 
1691   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1692     if (buf != NULL) {
1693       if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1694         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1695       }
1696     }
1697     if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1698     return true;
1699   } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1700     if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1701         buf, buflen, offset, dlinfo.dli_fname)) {
1702        return true;
1703     }
1704   }
1705 
1706   if (buf != NULL) buf[0] = '\0';
1707   if (offset != NULL) *offset = -1;
1708   return false;
1709 }
1710 
1711 struct _address_to_library_name {
1712   address addr;          // input : memory address
1713   size_t  buflen;        //         size of fname
1714   char*   fname;         // output: library name
1715   address base;          //         library base addr
1716 };
1717 
1718 static int address_to_library_name_callback(struct dl_phdr_info *info,
1719                                             size_t size, void *data) {
1720   int i;
1721   bool found = false;
1722   address libbase = NULL;
1723   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1724 
1725   // iterate through all loadable segments
1726   for (i = 0; i < info->dlpi_phnum; i++) {
1727     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1728     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1729       // base address of a library is the lowest address of its loaded
1730       // segments.
1731       if (libbase == NULL || libbase > segbase) {
1732         libbase = segbase;
1733       }
1734       // see if 'addr' is within current segment
1735       if (segbase <= d->addr &&
1736           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1737         found = true;
1738       }
1739     }
1740   }
1741 
1742   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1743   // so dll_address_to_library_name() can fall through to use dladdr() which
1744   // can figure out executable name from argv[0].
1745   if (found && info->dlpi_name && info->dlpi_name[0]) {
1746     d->base = libbase;
1747     if (d->fname) {
1748       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1749     }
1750     return 1;
1751   }
1752   return 0;
1753 }
1754 
1755 bool os::dll_address_to_library_name(address addr, char* buf,
1756                                      int buflen, int* offset) {
1757   Dl_info dlinfo;
1758   struct _address_to_library_name data;
1759 
1760   // There is a bug in old glibc dladdr() implementation that it could resolve
1761   // to wrong library name if the .so file has a base address != NULL. Here
1762   // we iterate through the program headers of all loaded libraries to find
1763   // out which library 'addr' really belongs to. This workaround can be
1764   // removed once the minimum requirement for glibc is moved to 2.3.x.
1765   data.addr = addr;
1766   data.fname = buf;
1767   data.buflen = buflen;
1768   data.base = NULL;
1769   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1770 
1771   if (rslt) {
1772      // buf already contains library name
1773      if (offset) *offset = addr - data.base;
1774      return true;
1775   } else if (dladdr((void*)addr, &dlinfo)){
1776      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1777      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1778      return true;
1779   } else {
1780      if (buf) buf[0] = '\0';
1781      if (offset) *offset = -1;
1782      return false;
1783   }
1784 }
1785 
1786   // Loads .dll/.so and
1787   // in case of error it checks if .dll/.so was built for the
1788   // same architecture as Hotspot is running on
1789 
1790 
1791 // Remember the stack's state. The Linux dynamic linker will change
1792 // the stack to 'executable' at most once, so we must safepoint only once.
1793 bool os::Linux::_stack_is_executable = false;
1794 
1795 // VM operation that loads a library.  This is necessary if stack protection
1796 // of the Java stacks can be lost during loading the library.  If we
1797 // do not stop the Java threads, they can stack overflow before the stacks
1798 // are protected again.
1799 class VM_LinuxDllLoad: public VM_Operation {
1800  private:
1801   const char *_filename;
1802   char *_ebuf;
1803   int _ebuflen;
1804   void *_lib;
1805  public:
1806   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1807     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1808   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1809   void doit() {
1810     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1811     os::Linux::_stack_is_executable = true;
1812   }
1813   void* loaded_library() { return _lib; }
1814 };
1815 
1816 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1817 {
1818   void * result = NULL;
1819   bool load_attempted = false;
1820 
1821   // Check whether the library to load might change execution rights
1822   // of the stack. If they are changed, the protection of the stack
1823   // guard pages will be lost. We need a safepoint to fix this.
1824   //
1825   // See Linux man page execstack(8) for more info.
1826   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1827     ElfFile ef(filename);
1828     if (!ef.specifies_noexecstack()) {
1829       if (!is_init_completed()) {
1830         os::Linux::_stack_is_executable = true;
1831         // This is OK - No Java threads have been created yet, and hence no
1832         // stack guard pages to fix.
1833         //
1834         // This should happen only when you are building JDK7 using a very
1835         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1836         //
1837         // Dynamic loader will make all stacks executable after
1838         // this function returns, and will not do that again.
1839         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1840       } else {
1841         warning("You have loaded library %s which might have disabled stack guard. "
1842                 "The VM will try to fix the stack guard now.\n"
1843                 "It's highly recommended that you fix the library with "
1844                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1845                 filename);
1846 
1847         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1848         JavaThread *jt = JavaThread::current();
1849         if (jt->thread_state() != _thread_in_native) {
1850           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1851           // that requires ExecStack. Cannot enter safe point. Let's give up.
1852           warning("Unable to fix stack guard. Giving up.");
1853         } else {
1854           if (!LoadExecStackDllInVMThread) {
1855             // This is for the case where the DLL has an static
1856             // constructor function that executes JNI code. We cannot
1857             // load such DLLs in the VMThread.
1858             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1859           }
1860 
1861           ThreadInVMfromNative tiv(jt);
1862           debug_only(VMNativeEntryWrapper vew;)
1863 
1864           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1865           VMThread::execute(&op);
1866           if (LoadExecStackDllInVMThread) {
1867             result = op.loaded_library();
1868           }
1869           load_attempted = true;
1870         }
1871       }
1872     }
1873   }
1874 
1875   if (!load_attempted) {
1876     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1877   }
1878 
1879   if (result != NULL) {
1880     // Successful loading
1881     return result;
1882   }
1883 
1884   Elf32_Ehdr elf_head;
1885   int diag_msg_max_length=ebuflen-strlen(ebuf);
1886   char* diag_msg_buf=ebuf+strlen(ebuf);
1887 
1888   if (diag_msg_max_length==0) {
1889     // No more space in ebuf for additional diagnostics message
1890     return NULL;
1891   }
1892 
1893 
1894   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1895 
1896   if (file_descriptor < 0) {
1897     // Can't open library, report dlerror() message
1898     return NULL;
1899   }
1900 
1901   bool failed_to_read_elf_head=
1902     (sizeof(elf_head)!=
1903         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1904 
1905   ::close(file_descriptor);
1906   if (failed_to_read_elf_head) {
1907     // file i/o error - report dlerror() msg
1908     return NULL;
1909   }
1910 
1911   typedef struct {
1912     Elf32_Half  code;         // Actual value as defined in elf.h
1913     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1914     char        elf_class;    // 32 or 64 bit
1915     char        endianess;    // MSB or LSB
1916     char*       name;         // String representation
1917   } arch_t;
1918 
1919   #ifndef EM_486
1920   #define EM_486          6               /* Intel 80486 */
1921   #endif
1922 
1923   static const arch_t arch_array[]={
1924     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1925     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1926     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1927     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1928     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1929     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1930     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1931     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1932     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1933     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1934     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1935     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1936     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1937     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1938     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1939     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1940   };
1941 
1942   #if  (defined IA32)
1943     static  Elf32_Half running_arch_code=EM_386;
1944   #elif   (defined AMD64)
1945     static  Elf32_Half running_arch_code=EM_X86_64;
1946   #elif  (defined IA64)
1947     static  Elf32_Half running_arch_code=EM_IA_64;
1948   #elif  (defined __sparc) && (defined _LP64)
1949     static  Elf32_Half running_arch_code=EM_SPARCV9;
1950   #elif  (defined __sparc) && (!defined _LP64)
1951     static  Elf32_Half running_arch_code=EM_SPARC;
1952   #elif  (defined __powerpc64__)
1953     static  Elf32_Half running_arch_code=EM_PPC64;
1954   #elif  (defined __powerpc__)
1955     static  Elf32_Half running_arch_code=EM_PPC;
1956   #elif  (defined ARM)
1957     static  Elf32_Half running_arch_code=EM_ARM;
1958   #elif  (defined S390)
1959     static  Elf32_Half running_arch_code=EM_S390;
1960   #elif  (defined ALPHA)
1961     static  Elf32_Half running_arch_code=EM_ALPHA;
1962   #elif  (defined MIPSEL)
1963     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1964   #elif  (defined PARISC)
1965     static  Elf32_Half running_arch_code=EM_PARISC;
1966   #elif  (defined MIPS)
1967     static  Elf32_Half running_arch_code=EM_MIPS;
1968   #elif  (defined M68K)
1969     static  Elf32_Half running_arch_code=EM_68K;
1970   #else
1971     #error Method os::dll_load requires that one of following is defined:\
1972          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1973   #endif
1974 
1975   // Identify compatability class for VM's architecture and library's architecture
1976   // Obtain string descriptions for architectures
1977 
1978   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1979   int running_arch_index=-1;
1980 
1981   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1982     if (running_arch_code == arch_array[i].code) {
1983       running_arch_index    = i;
1984     }
1985     if (lib_arch.code == arch_array[i].code) {
1986       lib_arch.compat_class = arch_array[i].compat_class;
1987       lib_arch.name         = arch_array[i].name;
1988     }
1989   }
1990 
1991   assert(running_arch_index != -1,
1992     "Didn't find running architecture code (running_arch_code) in arch_array");
1993   if (running_arch_index == -1) {
1994     // Even though running architecture detection failed
1995     // we may still continue with reporting dlerror() message
1996     return NULL;
1997   }
1998 
1999   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2000     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2001     return NULL;
2002   }
2003 
2004 #ifndef S390
2005   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2006     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2007     return NULL;
2008   }
2009 #endif // !S390
2010 
2011   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2012     if ( lib_arch.name!=NULL ) {
2013       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2014         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2015         lib_arch.name, arch_array[running_arch_index].name);
2016     } else {
2017       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2018       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2019         lib_arch.code,
2020         arch_array[running_arch_index].name);
2021     }
2022   }
2023 
2024   return NULL;
2025 }
2026 
2027 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2028   void * result = ::dlopen(filename, RTLD_LAZY);
2029   if (result == NULL) {
2030     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2031     ebuf[ebuflen-1] = '\0';
2032   }
2033   return result;
2034 }
2035 
2036 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2037   void * result = NULL;
2038   if (LoadExecStackDllInVMThread) {
2039     result = dlopen_helper(filename, ebuf, ebuflen);
2040   }
2041 
2042   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2043   // library that requires an executable stack, or which does not have this
2044   // stack attribute set, dlopen changes the stack attribute to executable. The
2045   // read protection of the guard pages gets lost.
2046   //
2047   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2048   // may have been queued at the same time.
2049 
2050   if (!_stack_is_executable) {
2051     JavaThread *jt = Threads::first();
2052 
2053     while (jt) {
2054       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2055           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2056         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2057                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2058           warning("Attempt to reguard stack yellow zone failed.");
2059         }
2060       }
2061       jt = jt->next();
2062     }
2063   }
2064 
2065   return result;
2066 }
2067 
2068 /*
2069  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2070  * chances are you might want to run the generated bits against glibc-2.0
2071  * libdl.so, so always use locking for any version of glibc.
2072  */
2073 void* os::dll_lookup(void* handle, const char* name) {
2074   pthread_mutex_lock(&dl_mutex);
2075   void* res = dlsym(handle, name);
2076   pthread_mutex_unlock(&dl_mutex);
2077   return res;
2078 }
2079 
2080 
2081 static bool _print_ascii_file(const char* filename, outputStream* st) {
2082   int fd = ::open(filename, O_RDONLY);
2083   if (fd == -1) {
2084      return false;
2085   }
2086 
2087   char buf[32];
2088   int bytes;
2089   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2090     st->print_raw(buf, bytes);
2091   }
2092 
2093   ::close(fd);
2094 
2095   return true;
2096 }
2097 
2098 void os::print_dll_info(outputStream *st) {
2099    st->print_cr("Dynamic libraries:");
2100 
2101    char fname[32];
2102    pid_t pid = os::Linux::gettid();
2103 
2104    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2105 
2106    if (!_print_ascii_file(fname, st)) {
2107      st->print("Can not get library information for pid = %d\n", pid);
2108    }
2109 }
2110 
2111 void os::print_os_info_brief(outputStream* st) {
2112   os::Linux::print_distro_info(st);
2113 
2114   os::Posix::print_uname_info(st);
2115 
2116   os::Linux::print_libversion_info(st);
2117 
2118 }
2119 
2120 void os::print_os_info(outputStream* st) {
2121   st->print("OS:");
2122 
2123   os::Linux::print_distro_info(st);
2124 
2125   os::Posix::print_uname_info(st);
2126 
2127   // Print warning if unsafe chroot environment detected
2128   if (unsafe_chroot_detected) {
2129     st->print("WARNING!! ");
2130     st->print_cr(unstable_chroot_error);
2131   }
2132 
2133   os::Linux::print_libversion_info(st);
2134 
2135   os::Posix::print_rlimit_info(st);
2136 
2137   os::Posix::print_load_average(st);
2138 
2139   os::Linux::print_full_memory_info(st);
2140 }
2141 
2142 // Try to identify popular distros.
2143 // Most Linux distributions have /etc/XXX-release file, which contains
2144 // the OS version string. Some have more than one /etc/XXX-release file
2145 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2146 // so the order is important.
2147 void os::Linux::print_distro_info(outputStream* st) {
2148   if (!_print_ascii_file("/etc/mandrake-release", st) &&
2149       !_print_ascii_file("/etc/sun-release", st) &&
2150       !_print_ascii_file("/etc/redhat-release", st) &&
2151       !_print_ascii_file("/etc/SuSE-release", st) &&
2152       !_print_ascii_file("/etc/turbolinux-release", st) &&
2153       !_print_ascii_file("/etc/gentoo-release", st) &&
2154       !_print_ascii_file("/etc/debian_version", st) &&
2155       !_print_ascii_file("/etc/ltib-release", st) &&
2156       !_print_ascii_file("/etc/angstrom-version", st)) {
2157       st->print("Linux");
2158   }
2159   st->cr();
2160 }
2161 
2162 void os::Linux::print_libversion_info(outputStream* st) {
2163   // libc, pthread
2164   st->print("libc:");
2165   st->print(os::Linux::glibc_version()); st->print(" ");
2166   st->print(os::Linux::libpthread_version()); st->print(" ");
2167   if (os::Linux::is_LinuxThreads()) {
2168      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2169   }
2170   st->cr();
2171 }
2172 
2173 void os::Linux::print_full_memory_info(outputStream* st) {
2174    st->print("\n/proc/meminfo:\n");
2175    _print_ascii_file("/proc/meminfo", st);
2176    st->cr();
2177 }
2178 
2179 void os::print_memory_info(outputStream* st) {
2180 
2181   st->print("Memory:");
2182   st->print(" %dk page", os::vm_page_size()>>10);
2183 
2184   // values in struct sysinfo are "unsigned long"
2185   struct sysinfo si;
2186   sysinfo(&si);
2187 
2188   st->print(", physical " UINT64_FORMAT "k",
2189             os::physical_memory() >> 10);
2190   st->print("(" UINT64_FORMAT "k free)",
2191             os::available_memory() >> 10);
2192   st->print(", swap " UINT64_FORMAT "k",
2193             ((jlong)si.totalswap * si.mem_unit) >> 10);
2194   st->print("(" UINT64_FORMAT "k free)",
2195             ((jlong)si.freeswap * si.mem_unit) >> 10);
2196   st->cr();
2197 }
2198 
2199 void os::pd_print_cpu_info(outputStream* st) {
2200   st->print("\n/proc/cpuinfo:\n");
2201   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2202     st->print("  <Not Available>");
2203   }
2204   st->cr();
2205 }
2206 
2207 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2208 // but they're the same for all the linux arch that we support
2209 // and they're the same for solaris but there's no common place to put this.
2210 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2211                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2212                           "ILL_COPROC", "ILL_BADSTK" };
2213 
2214 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2215                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2216                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2217 
2218 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2219 
2220 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2221 
2222 void os::print_siginfo(outputStream* st, void* siginfo) {
2223   st->print("siginfo:");
2224 
2225   const int buflen = 100;
2226   char buf[buflen];
2227   siginfo_t *si = (siginfo_t*)siginfo;
2228   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2229   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2230     st->print("si_errno=%s", buf);
2231   } else {
2232     st->print("si_errno=%d", si->si_errno);
2233   }
2234   const int c = si->si_code;
2235   assert(c > 0, "unexpected si_code");
2236   switch (si->si_signo) {
2237   case SIGILL:
2238     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2239     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2240     break;
2241   case SIGFPE:
2242     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2243     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2244     break;
2245   case SIGSEGV:
2246     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2247     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2248     break;
2249   case SIGBUS:
2250     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2251     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2252     break;
2253   default:
2254     st->print(", si_code=%d", si->si_code);
2255     // no si_addr
2256   }
2257 
2258   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2259       UseSharedSpaces) {
2260     FileMapInfo* mapinfo = FileMapInfo::current_info();
2261     if (mapinfo->is_in_shared_space(si->si_addr)) {
2262       st->print("\n\nError accessing class data sharing archive."   \
2263                 " Mapped file inaccessible during execution, "      \
2264                 " possible disk/network problem.");
2265     }
2266   }
2267   st->cr();
2268 }
2269 
2270 
2271 static void print_signal_handler(outputStream* st, int sig,
2272                                  char* buf, size_t buflen);
2273 
2274 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2275   st->print_cr("Signal Handlers:");
2276   print_signal_handler(st, SIGSEGV, buf, buflen);
2277   print_signal_handler(st, SIGBUS , buf, buflen);
2278   print_signal_handler(st, SIGFPE , buf, buflen);
2279   print_signal_handler(st, SIGPIPE, buf, buflen);
2280   print_signal_handler(st, SIGXFSZ, buf, buflen);
2281   print_signal_handler(st, SIGILL , buf, buflen);
2282   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2283   print_signal_handler(st, SR_signum, buf, buflen);
2284   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2285   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2286   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2287   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2288 }
2289 
2290 static char saved_jvm_path[MAXPATHLEN] = {0};
2291 
2292 // Find the full path to the current module, libjvm.so
2293 void os::jvm_path(char *buf, jint buflen) {
2294   // Error checking.
2295   if (buflen < MAXPATHLEN) {
2296     assert(false, "must use a large-enough buffer");
2297     buf[0] = '\0';
2298     return;
2299   }
2300   // Lazy resolve the path to current module.
2301   if (saved_jvm_path[0] != 0) {
2302     strcpy(buf, saved_jvm_path);
2303     return;
2304   }
2305 
2306   char dli_fname[MAXPATHLEN];
2307   bool ret = dll_address_to_library_name(
2308                 CAST_FROM_FN_PTR(address, os::jvm_path),
2309                 dli_fname, sizeof(dli_fname), NULL);
2310   assert(ret != 0, "cannot locate libjvm");
2311   char *rp = realpath(dli_fname, buf);
2312   if (rp == NULL)
2313     return;
2314 
2315   if (Arguments::created_by_gamma_launcher()) {
2316     // Support for the gamma launcher.  Typical value for buf is
2317     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2318     // the right place in the string, then assume we are installed in a JDK and
2319     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2320     // up the path so it looks like libjvm.so is installed there (append a
2321     // fake suffix hotspot/libjvm.so).
2322     const char *p = buf + strlen(buf) - 1;
2323     for (int count = 0; p > buf && count < 5; ++count) {
2324       for (--p; p > buf && *p != '/'; --p)
2325         /* empty */ ;
2326     }
2327 
2328     if (strncmp(p, "/jre/lib/", 9) != 0) {
2329       // Look for JAVA_HOME in the environment.
2330       char* java_home_var = ::getenv("JAVA_HOME");
2331       if (java_home_var != NULL && java_home_var[0] != 0) {
2332         char* jrelib_p;
2333         int len;
2334 
2335         // Check the current module name "libjvm.so".
2336         p = strrchr(buf, '/');
2337         assert(strstr(p, "/libjvm") == p, "invalid library name");
2338 
2339         rp = realpath(java_home_var, buf);
2340         if (rp == NULL)
2341           return;
2342 
2343         // determine if this is a legacy image or modules image
2344         // modules image doesn't have "jre" subdirectory
2345         len = strlen(buf);
2346         jrelib_p = buf + len;
2347         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2348         if (0 != access(buf, F_OK)) {
2349           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2350         }
2351 
2352         if (0 == access(buf, F_OK)) {
2353           // Use current module name "libjvm.so"
2354           len = strlen(buf);
2355           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2356         } else {
2357           // Go back to path of .so
2358           rp = realpath(dli_fname, buf);
2359           if (rp == NULL)
2360             return;
2361         }
2362       }
2363     }
2364   }
2365 
2366   strcpy(saved_jvm_path, buf);
2367 }
2368 
2369 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2370   // no prefix required, not even "_"
2371 }
2372 
2373 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2374   // no suffix required
2375 }
2376 
2377 ////////////////////////////////////////////////////////////////////////////////
2378 // sun.misc.Signal support
2379 
2380 static volatile jint sigint_count = 0;
2381 
2382 static void
2383 UserHandler(int sig, void *siginfo, void *context) {
2384   // 4511530 - sem_post is serialized and handled by the manager thread. When
2385   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2386   // don't want to flood the manager thread with sem_post requests.
2387   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2388       return;
2389 
2390   // Ctrl-C is pressed during error reporting, likely because the error
2391   // handler fails to abort. Let VM die immediately.
2392   if (sig == SIGINT && is_error_reported()) {
2393      os::die();
2394   }
2395 
2396   os::signal_notify(sig);
2397 }
2398 
2399 void* os::user_handler() {
2400   return CAST_FROM_FN_PTR(void*, UserHandler);
2401 }
2402 
2403 class Semaphore : public StackObj {
2404   public:
2405     Semaphore();
2406     ~Semaphore();
2407     void signal();
2408     void wait();
2409     bool trywait();
2410     bool timedwait(unsigned int sec, int nsec);
2411   private:
2412     sem_t _semaphore;
2413 };
2414 
2415 
2416 Semaphore::Semaphore() {
2417   sem_init(&_semaphore, 0, 0);
2418 }
2419 
2420 Semaphore::~Semaphore() {
2421   sem_destroy(&_semaphore);
2422 }
2423 
2424 void Semaphore::signal() {
2425   sem_post(&_semaphore);
2426 }
2427 
2428 void Semaphore::wait() {
2429   sem_wait(&_semaphore);
2430 }
2431 
2432 bool Semaphore::trywait() {
2433   return sem_trywait(&_semaphore) == 0;
2434 }
2435 
2436 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2437   struct timespec ts;
2438   unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2439 
2440   while (1) {
2441     int result = sem_timedwait(&_semaphore, &ts);
2442     if (result == 0) {
2443       return true;
2444     } else if (errno == EINTR) {
2445       continue;
2446     } else if (errno == ETIMEDOUT) {
2447       return false;
2448     } else {
2449       return false;
2450     }
2451   }
2452 }
2453 
2454 extern "C" {
2455   typedef void (*sa_handler_t)(int);
2456   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2457 }
2458 
2459 void* os::signal(int signal_number, void* handler) {
2460   struct sigaction sigAct, oldSigAct;
2461 
2462   sigfillset(&(sigAct.sa_mask));
2463   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2464   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2465 
2466   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2467     // -1 means registration failed
2468     return (void *)-1;
2469   }
2470 
2471   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2472 }
2473 
2474 void os::signal_raise(int signal_number) {
2475   ::raise(signal_number);
2476 }
2477 
2478 /*
2479  * The following code is moved from os.cpp for making this
2480  * code platform specific, which it is by its very nature.
2481  */
2482 
2483 // Will be modified when max signal is changed to be dynamic
2484 int os::sigexitnum_pd() {
2485   return NSIG;
2486 }
2487 
2488 // a counter for each possible signal value
2489 static volatile jint pending_signals[NSIG+1] = { 0 };
2490 
2491 // Linux(POSIX) specific hand shaking semaphore.
2492 static sem_t sig_sem;
2493 static Semaphore sr_semaphore;
2494 
2495 void os::signal_init_pd() {
2496   // Initialize signal structures
2497   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2498 
2499   // Initialize signal semaphore
2500   ::sem_init(&sig_sem, 0, 0);
2501 }
2502 
2503 void os::signal_notify(int sig) {
2504   Atomic::inc(&pending_signals[sig]);
2505   ::sem_post(&sig_sem);
2506 }
2507 
2508 static int check_pending_signals(bool wait) {
2509   Atomic::store(0, &sigint_count);
2510   for (;;) {
2511     for (int i = 0; i < NSIG + 1; i++) {
2512       jint n = pending_signals[i];
2513       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2514         return i;
2515       }
2516     }
2517     if (!wait) {
2518       return -1;
2519     }
2520     JavaThread *thread = JavaThread::current();
2521     ThreadBlockInVM tbivm(thread);
2522 
2523     bool threadIsSuspended;
2524     do {
2525       thread->set_suspend_equivalent();
2526       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2527       ::sem_wait(&sig_sem);
2528 
2529       // were we externally suspended while we were waiting?
2530       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2531       if (threadIsSuspended) {
2532         //
2533         // The semaphore has been incremented, but while we were waiting
2534         // another thread suspended us. We don't want to continue running
2535         // while suspended because that would surprise the thread that
2536         // suspended us.
2537         //
2538         ::sem_post(&sig_sem);
2539 
2540         thread->java_suspend_self();
2541       }
2542     } while (threadIsSuspended);
2543   }
2544 }
2545 
2546 int os::signal_lookup() {
2547   return check_pending_signals(false);
2548 }
2549 
2550 int os::signal_wait() {
2551   return check_pending_signals(true);
2552 }
2553 
2554 ////////////////////////////////////////////////////////////////////////////////
2555 // Virtual Memory
2556 
2557 int os::vm_page_size() {
2558   // Seems redundant as all get out
2559   assert(os::Linux::page_size() != -1, "must call os::init");
2560   return os::Linux::page_size();
2561 }
2562 
2563 // Solaris allocates memory by pages.
2564 int os::vm_allocation_granularity() {
2565   assert(os::Linux::page_size() != -1, "must call os::init");
2566   return os::Linux::page_size();
2567 }
2568 
2569 // Rationale behind this function:
2570 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2571 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2572 //  samples for JITted code. Here we create private executable mapping over the code cache
2573 //  and then we can use standard (well, almost, as mapping can change) way to provide
2574 //  info for the reporting script by storing timestamp and location of symbol
2575 void linux_wrap_code(char* base, size_t size) {
2576   static volatile jint cnt = 0;
2577 
2578   if (!UseOprofile) {
2579     return;
2580   }
2581 
2582   char buf[PATH_MAX+1];
2583   int num = Atomic::add(1, &cnt);
2584 
2585   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2586            os::get_temp_directory(), os::current_process_id(), num);
2587   unlink(buf);
2588 
2589   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2590 
2591   if (fd != -1) {
2592     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2593     if (rv != (off_t)-1) {
2594       if (::write(fd, "", 1) == 1) {
2595         mmap(base, size,
2596              PROT_READ|PROT_WRITE|PROT_EXEC,
2597              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2598       }
2599     }
2600     ::close(fd);
2601     unlink(buf);
2602   }
2603 }
2604 
2605 // NOTE: Linux kernel does not really reserve the pages for us.
2606 //       All it does is to check if there are enough free pages
2607 //       left at the time of mmap(). This could be a potential
2608 //       problem.
2609 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2610   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2611   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2612                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2613   if (res != (uintptr_t) MAP_FAILED) {
2614     if (UseNUMAInterleaving) {
2615       numa_make_global(addr, size);
2616     }
2617     return true;
2618   }
2619   return false;
2620 }
2621 
2622 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2623 #ifndef MAP_HUGETLB
2624 #define MAP_HUGETLB 0x40000
2625 #endif
2626 
2627 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2628 #ifndef MADV_HUGEPAGE
2629 #define MADV_HUGEPAGE 14
2630 #endif
2631 
2632 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2633                        bool exec) {
2634   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2635     int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2636     uintptr_t res =
2637       (uintptr_t) ::mmap(addr, size, prot,
2638                          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2639                          -1, 0);
2640     if (res != (uintptr_t) MAP_FAILED) {
2641       if (UseNUMAInterleaving) {
2642         numa_make_global(addr, size);
2643       }
2644       return true;
2645     }
2646     // Fall through and try to use small pages
2647   }
2648 
2649   if (commit_memory(addr, size, exec)) {
2650     realign_memory(addr, size, alignment_hint);
2651     return true;
2652   }
2653   return false;
2654 }
2655 
2656 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2657   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2658     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2659     // be supported or the memory may already be backed by huge pages.
2660     ::madvise(addr, bytes, MADV_HUGEPAGE);
2661   }
2662 }
2663 
2664 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2665   // This method works by doing an mmap over an existing mmaping and effectively discarding
2666   // the existing pages. However it won't work for SHM-based large pages that cannot be
2667   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2668   // small pages on top of the SHM segment. This method always works for small pages, so we
2669   // allow that in any case.
2670   if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2671     commit_memory(addr, bytes, alignment_hint, false);
2672   }
2673 }
2674 
2675 void os::numa_make_global(char *addr, size_t bytes) {
2676   Linux::numa_interleave_memory(addr, bytes);
2677 }
2678 
2679 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2680   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2681 }
2682 
2683 bool os::numa_topology_changed()   { return false; }
2684 
2685 size_t os::numa_get_groups_num() {
2686   int max_node = Linux::numa_max_node();
2687   return max_node > 0 ? max_node + 1 : 1;
2688 }
2689 
2690 int os::numa_get_group_id() {
2691   int cpu_id = Linux::sched_getcpu();
2692   if (cpu_id != -1) {
2693     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2694     if (lgrp_id != -1) {
2695       return lgrp_id;
2696     }
2697   }
2698   return 0;
2699 }
2700 
2701 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2702   for (size_t i = 0; i < size; i++) {
2703     ids[i] = i;
2704   }
2705   return size;
2706 }
2707 
2708 bool os::get_page_info(char *start, page_info* info) {
2709   return false;
2710 }
2711 
2712 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2713   return end;
2714 }
2715 
2716 
2717 int os::Linux::sched_getcpu_syscall(void) {
2718   unsigned int cpu;
2719   int retval = -1;
2720 
2721 #if defined(IA32)
2722 # ifndef SYS_getcpu
2723 # define SYS_getcpu 318
2724 # endif
2725   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2726 #elif defined(AMD64)
2727 // Unfortunately we have to bring all these macros here from vsyscall.h
2728 // to be able to compile on old linuxes.
2729 # define __NR_vgetcpu 2
2730 # define VSYSCALL_START (-10UL << 20)
2731 # define VSYSCALL_SIZE 1024
2732 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2733   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2734   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2735   retval = vgetcpu(&cpu, NULL, NULL);
2736 #endif
2737 
2738   return (retval == -1) ? retval : cpu;
2739 }
2740 
2741 // Something to do with the numa-aware allocator needs these symbols
2742 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2743 extern "C" JNIEXPORT void numa_error(char *where) { }
2744 extern "C" JNIEXPORT int fork1() { return fork(); }
2745 
2746 
2747 // If we are running with libnuma version > 2, then we should
2748 // be trying to use symbols with versions 1.1
2749 // If we are running with earlier version, which did not have symbol versions,
2750 // we should use the base version.
2751 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2752   void *f = dlvsym(handle, name, "libnuma_1.1");
2753   if (f == NULL) {
2754     f = dlsym(handle, name);
2755   }
2756   return f;
2757 }
2758 
2759 bool os::Linux::libnuma_init() {
2760   // sched_getcpu() should be in libc.
2761   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2762                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2763 
2764   // If it's not, try a direct syscall.
2765   if (sched_getcpu() == -1)
2766     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2767 
2768   if (sched_getcpu() != -1) { // Does it work?
2769     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2770     if (handle != NULL) {
2771       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2772                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2773       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2774                                        libnuma_dlsym(handle, "numa_max_node")));
2775       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2776                                         libnuma_dlsym(handle, "numa_available")));
2777       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2778                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2779       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2780                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2781 
2782 
2783       if (numa_available() != -1) {
2784         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2785         // Create a cpu -> node mapping
2786         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2787         rebuild_cpu_to_node_map();
2788         return true;
2789       }
2790     }
2791   }
2792   return false;
2793 }
2794 
2795 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2796 // The table is later used in get_node_by_cpu().
2797 void os::Linux::rebuild_cpu_to_node_map() {
2798   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2799                               // in libnuma (possible values are starting from 16,
2800                               // and continuing up with every other power of 2, but less
2801                               // than the maximum number of CPUs supported by kernel), and
2802                               // is a subject to change (in libnuma version 2 the requirements
2803                               // are more reasonable) we'll just hardcode the number they use
2804                               // in the library.
2805   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2806 
2807   size_t cpu_num = os::active_processor_count();
2808   size_t cpu_map_size = NCPUS / BitsPerCLong;
2809   size_t cpu_map_valid_size =
2810     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2811 
2812   cpu_to_node()->clear();
2813   cpu_to_node()->at_grow(cpu_num - 1);
2814   size_t node_num = numa_get_groups_num();
2815 
2816   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2817   for (size_t i = 0; i < node_num; i++) {
2818     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2819       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2820         if (cpu_map[j] != 0) {
2821           for (size_t k = 0; k < BitsPerCLong; k++) {
2822             if (cpu_map[j] & (1UL << k)) {
2823               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2824             }
2825           }
2826         }
2827       }
2828     }
2829   }
2830   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2831 }
2832 
2833 int os::Linux::get_node_by_cpu(int cpu_id) {
2834   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2835     return cpu_to_node()->at(cpu_id);
2836   }
2837   return -1;
2838 }
2839 
2840 GrowableArray<int>* os::Linux::_cpu_to_node;
2841 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2842 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2843 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2844 os::Linux::numa_available_func_t os::Linux::_numa_available;
2845 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2846 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2847 unsigned long* os::Linux::_numa_all_nodes;
2848 
2849 bool os::pd_uncommit_memory(char* addr, size_t size) {
2850   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2851                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2852   return res  != (uintptr_t) MAP_FAILED;
2853 }
2854 
2855 // Linux uses a growable mapping for the stack, and if the mapping for
2856 // the stack guard pages is not removed when we detach a thread the
2857 // stack cannot grow beyond the pages where the stack guard was
2858 // mapped.  If at some point later in the process the stack expands to
2859 // that point, the Linux kernel cannot expand the stack any further
2860 // because the guard pages are in the way, and a segfault occurs.
2861 //
2862 // However, it's essential not to split the stack region by unmapping
2863 // a region (leaving a hole) that's already part of the stack mapping,
2864 // so if the stack mapping has already grown beyond the guard pages at
2865 // the time we create them, we have to truncate the stack mapping.
2866 // So, we need to know the extent of the stack mapping when
2867 // create_stack_guard_pages() is called.
2868 
2869 // Find the bounds of the stack mapping.  Return true for success.
2870 //
2871 // We only need this for stacks that are growable: at the time of
2872 // writing thread stacks don't use growable mappings (i.e. those
2873 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2874 // only applies to the main thread.
2875 
2876 static
2877 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2878 
2879   char buf[128];
2880   int fd, sz;
2881 
2882   if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2883     return false;
2884   }
2885 
2886   const char kw[] = "[stack]";
2887   const int kwlen = sizeof(kw)-1;
2888 
2889   // Address part of /proc/self/maps couldn't be more than 128 bytes
2890   while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2891      if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2892         // Extract addresses
2893         if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2894            uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2895            if (sp >= *bottom && sp <= *top) {
2896               ::close(fd);
2897               return true;
2898            }
2899         }
2900      }
2901   }
2902 
2903  ::close(fd);
2904   return false;
2905 }
2906 
2907 
2908 // If the (growable) stack mapping already extends beyond the point
2909 // where we're going to put our guard pages, truncate the mapping at
2910 // that point by munmap()ping it.  This ensures that when we later
2911 // munmap() the guard pages we don't leave a hole in the stack
2912 // mapping. This only affects the main/initial thread, but guard
2913 // against future OS changes
2914 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2915   uintptr_t stack_extent, stack_base;
2916   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2917   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2918       assert(os::Linux::is_initial_thread(),
2919            "growable stack in non-initial thread");
2920     if (stack_extent < (uintptr_t)addr)
2921       ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2922   }
2923 
2924   return os::commit_memory(addr, size);
2925 }
2926 
2927 // If this is a growable mapping, remove the guard pages entirely by
2928 // munmap()ping them.  If not, just call uncommit_memory(). This only
2929 // affects the main/initial thread, but guard against future OS changes
2930 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2931   uintptr_t stack_extent, stack_base;
2932   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2933   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2934       assert(os::Linux::is_initial_thread(),
2935            "growable stack in non-initial thread");
2936 
2937     return ::munmap(addr, size) == 0;
2938   }
2939 
2940   return os::uncommit_memory(addr, size);
2941 }
2942 
2943 static address _highest_vm_reserved_address = NULL;
2944 
2945 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2946 // at 'requested_addr'. If there are existing memory mappings at the same
2947 // location, however, they will be overwritten. If 'fixed' is false,
2948 // 'requested_addr' is only treated as a hint, the return value may or
2949 // may not start from the requested address. Unlike Linux mmap(), this
2950 // function returns NULL to indicate failure.
2951 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2952   char * addr;
2953   int flags;
2954 
2955   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2956   if (fixed) {
2957     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2958     flags |= MAP_FIXED;
2959   }
2960 
2961   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
2962   // touch an uncommitted page. Otherwise, the read/write might
2963   // succeed if we have enough swap space to back the physical page.
2964   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
2965                        flags, -1, 0);
2966 
2967   if (addr != MAP_FAILED) {
2968     // anon_mmap() should only get called during VM initialization,
2969     // don't need lock (actually we can skip locking even it can be called
2970     // from multiple threads, because _highest_vm_reserved_address is just a
2971     // hint about the upper limit of non-stack memory regions.)
2972     if ((address)addr + bytes > _highest_vm_reserved_address) {
2973       _highest_vm_reserved_address = (address)addr + bytes;
2974     }
2975   }
2976 
2977   return addr == MAP_FAILED ? NULL : addr;
2978 }
2979 
2980 // Don't update _highest_vm_reserved_address, because there might be memory
2981 // regions above addr + size. If so, releasing a memory region only creates
2982 // a hole in the address space, it doesn't help prevent heap-stack collision.
2983 //
2984 static int anon_munmap(char * addr, size_t size) {
2985   return ::munmap(addr, size) == 0;
2986 }
2987 
2988 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2989                          size_t alignment_hint) {
2990   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2991 }
2992 
2993 bool os::pd_release_memory(char* addr, size_t size) {
2994   return anon_munmap(addr, size);
2995 }
2996 
2997 static address highest_vm_reserved_address() {
2998   return _highest_vm_reserved_address;
2999 }
3000 
3001 static bool linux_mprotect(char* addr, size_t size, int prot) {
3002   // Linux wants the mprotect address argument to be page aligned.
3003   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3004 
3005   // According to SUSv3, mprotect() should only be used with mappings
3006   // established by mmap(), and mmap() always maps whole pages. Unaligned
3007   // 'addr' likely indicates problem in the VM (e.g. trying to change
3008   // protection of malloc'ed or statically allocated memory). Check the
3009   // caller if you hit this assert.
3010   assert(addr == bottom, "sanity check");
3011 
3012   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3013   return ::mprotect(bottom, size, prot) == 0;
3014 }
3015 
3016 // Set protections specified
3017 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3018                         bool is_committed) {
3019   unsigned int p = 0;
3020   switch (prot) {
3021   case MEM_PROT_NONE: p = PROT_NONE; break;
3022   case MEM_PROT_READ: p = PROT_READ; break;
3023   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3024   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3025   default:
3026     ShouldNotReachHere();
3027   }
3028   // is_committed is unused.
3029   return linux_mprotect(addr, bytes, p);
3030 }
3031 
3032 bool os::guard_memory(char* addr, size_t size) {
3033   return linux_mprotect(addr, size, PROT_NONE);
3034 }
3035 
3036 bool os::unguard_memory(char* addr, size_t size) {
3037   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3038 }
3039 
3040 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3041   bool result = false;
3042   void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
3043                   MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3044                   -1, 0);
3045 
3046   if (p != (void *) -1) {
3047     // We don't know if this really is a huge page or not.
3048     FILE *fp = fopen("/proc/self/maps", "r");
3049     if (fp) {
3050       while (!feof(fp)) {
3051         char chars[257];
3052         long x = 0;
3053         if (fgets(chars, sizeof(chars), fp)) {
3054           if (sscanf(chars, "%lx-%*x", &x) == 1
3055               && x == (long)p) {
3056             if (strstr (chars, "hugepage")) {
3057               result = true;
3058               break;
3059             }
3060           }
3061         }
3062       }
3063       fclose(fp);
3064     }
3065     munmap (p, page_size);
3066     if (result)
3067       return true;
3068   }
3069 
3070   if (warn) {
3071     warning("HugeTLBFS is not supported by the operating system.");
3072   }
3073 
3074   return result;
3075 }
3076 
3077 /*
3078 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3079 *
3080 * From the coredump_filter documentation:
3081 *
3082 * - (bit 0) anonymous private memory
3083 * - (bit 1) anonymous shared memory
3084 * - (bit 2) file-backed private memory
3085 * - (bit 3) file-backed shared memory
3086 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3087 *           effective only if the bit 2 is cleared)
3088 * - (bit 5) hugetlb private memory
3089 * - (bit 6) hugetlb shared memory
3090 */
3091 static void set_coredump_filter(void) {
3092   FILE *f;
3093   long cdm;
3094 
3095   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3096     return;
3097   }
3098 
3099   if (fscanf(f, "%lx", &cdm) != 1) {
3100     fclose(f);
3101     return;
3102   }
3103 
3104   rewind(f);
3105 
3106   if ((cdm & LARGEPAGES_BIT) == 0) {
3107     cdm |= LARGEPAGES_BIT;
3108     fprintf(f, "%#lx", cdm);
3109   }
3110 
3111   fclose(f);
3112 }
3113 
3114 // Large page support
3115 
3116 static size_t _large_page_size = 0;
3117 
3118 void os::large_page_init() {
3119   if (!UseLargePages) {
3120     UseHugeTLBFS = false;
3121     UseSHM = false;
3122     return;
3123   }
3124 
3125   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3126     // If UseLargePages is specified on the command line try both methods,
3127     // if it's default, then try only HugeTLBFS.
3128     if (FLAG_IS_DEFAULT(UseLargePages)) {
3129       UseHugeTLBFS = true;
3130     } else {
3131       UseHugeTLBFS = UseSHM = true;
3132     }
3133   }
3134 
3135   if (LargePageSizeInBytes) {
3136     _large_page_size = LargePageSizeInBytes;
3137   } else {
3138     // large_page_size on Linux is used to round up heap size. x86 uses either
3139     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3140     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3141     // page as large as 256M.
3142     //
3143     // Here we try to figure out page size by parsing /proc/meminfo and looking
3144     // for a line with the following format:
3145     //    Hugepagesize:     2048 kB
3146     //
3147     // If we can't determine the value (e.g. /proc is not mounted, or the text
3148     // format has been changed), we'll use the largest page size supported by
3149     // the processor.
3150 
3151 #ifndef ZERO
3152     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3153                        ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3154 #endif // ZERO
3155 
3156     FILE *fp = fopen("/proc/meminfo", "r");
3157     if (fp) {
3158       while (!feof(fp)) {
3159         int x = 0;
3160         char buf[16];
3161         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3162           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3163             _large_page_size = x * K;
3164             break;
3165           }
3166         } else {
3167           // skip to next line
3168           for (;;) {
3169             int ch = fgetc(fp);
3170             if (ch == EOF || ch == (int)'\n') break;
3171           }
3172         }
3173       }
3174       fclose(fp);
3175     }
3176   }
3177 
3178   // print a warning if any large page related flag is specified on command line
3179   bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3180 
3181   const size_t default_page_size = (size_t)Linux::page_size();
3182   if (_large_page_size > default_page_size) {
3183     _page_sizes[0] = _large_page_size;
3184     _page_sizes[1] = default_page_size;
3185     _page_sizes[2] = 0;
3186   }
3187   UseHugeTLBFS = UseHugeTLBFS &&
3188                  Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3189 
3190   if (UseHugeTLBFS)
3191     UseSHM = false;
3192 
3193   UseLargePages = UseHugeTLBFS || UseSHM;
3194 
3195   set_coredump_filter();
3196 }
3197 
3198 #ifndef SHM_HUGETLB
3199 #define SHM_HUGETLB 04000
3200 #endif
3201 
3202 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3203   // "exec" is passed in but not used.  Creating the shared image for
3204   // the code cache doesn't have an SHM_X executable permission to check.
3205   assert(UseLargePages && UseSHM, "only for SHM large pages");
3206 
3207   key_t key = IPC_PRIVATE;
3208   char *addr;
3209 
3210   bool warn_on_failure = UseLargePages &&
3211                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3212                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3213                         );
3214   char msg[128];
3215 
3216   // Create a large shared memory region to attach to based on size.
3217   // Currently, size is the total size of the heap
3218   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3219   if (shmid == -1) {
3220      // Possible reasons for shmget failure:
3221      // 1. shmmax is too small for Java heap.
3222      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3223      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3224      // 2. not enough large page memory.
3225      //    > check available large pages: cat /proc/meminfo
3226      //    > increase amount of large pages:
3227      //          echo new_value > /proc/sys/vm/nr_hugepages
3228      //      Note 1: different Linux may use different name for this property,
3229      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3230      //      Note 2: it's possible there's enough physical memory available but
3231      //            they are so fragmented after a long run that they can't
3232      //            coalesce into large pages. Try to reserve large pages when
3233      //            the system is still "fresh".
3234      if (warn_on_failure) {
3235        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3236        warning(msg);
3237      }
3238      return NULL;
3239   }
3240 
3241   // attach to the region
3242   addr = (char*)shmat(shmid, req_addr, 0);
3243   int err = errno;
3244 
3245   // Remove shmid. If shmat() is successful, the actual shared memory segment
3246   // will be deleted when it's detached by shmdt() or when the process
3247   // terminates. If shmat() is not successful this will remove the shared
3248   // segment immediately.
3249   shmctl(shmid, IPC_RMID, NULL);
3250 
3251   if ((intptr_t)addr == -1) {
3252      if (warn_on_failure) {
3253        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3254        warning(msg);
3255      }
3256      return NULL;
3257   }
3258 
3259   if ((addr != NULL) && UseNUMAInterleaving) {
3260     numa_make_global(addr, bytes);
3261   }
3262 
3263   // The memory is committed
3264   address pc = CALLER_PC;
3265   MemTracker::record_virtual_memory_reserve((address)addr, bytes, pc);
3266   MemTracker::record_virtual_memory_commit((address)addr, bytes, pc);
3267 
3268   return addr;
3269 }
3270 
3271 bool os::release_memory_special(char* base, size_t bytes) {
3272   // detaching the SHM segment will also delete it, see reserve_memory_special()
3273   int rslt = shmdt(base);
3274   if (rslt == 0) {
3275     MemTracker::record_virtual_memory_uncommit((address)base, bytes);
3276     MemTracker::record_virtual_memory_release((address)base, bytes);
3277     return true;
3278   } else {
3279    return false;
3280   }
3281 }
3282 
3283 size_t os::large_page_size() {
3284   return _large_page_size;
3285 }
3286 
3287 // HugeTLBFS allows application to commit large page memory on demand;
3288 // with SysV SHM the entire memory region must be allocated as shared
3289 // memory.
3290 bool os::can_commit_large_page_memory() {
3291   return UseHugeTLBFS;
3292 }
3293 
3294 bool os::can_execute_large_page_memory() {
3295   return UseHugeTLBFS;
3296 }
3297 
3298 // Reserve memory at an arbitrary address, only if that area is
3299 // available (and not reserved for something else).
3300 
3301 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3302   const int max_tries = 10;
3303   char* base[max_tries];
3304   size_t size[max_tries];
3305   const size_t gap = 0x000000;
3306 
3307   // Assert only that the size is a multiple of the page size, since
3308   // that's all that mmap requires, and since that's all we really know
3309   // about at this low abstraction level.  If we need higher alignment,
3310   // we can either pass an alignment to this method or verify alignment
3311   // in one of the methods further up the call chain.  See bug 5044738.
3312   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3313 
3314   // Repeatedly allocate blocks until the block is allocated at the
3315   // right spot. Give up after max_tries. Note that reserve_memory() will
3316   // automatically update _highest_vm_reserved_address if the call is
3317   // successful. The variable tracks the highest memory address every reserved
3318   // by JVM. It is used to detect heap-stack collision if running with
3319   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3320   // space than needed, it could confuse the collision detecting code. To
3321   // solve the problem, save current _highest_vm_reserved_address and
3322   // calculate the correct value before return.
3323   address old_highest = _highest_vm_reserved_address;
3324 
3325   // Linux mmap allows caller to pass an address as hint; give it a try first,
3326   // if kernel honors the hint then we can return immediately.
3327   char * addr = anon_mmap(requested_addr, bytes, false);
3328   if (addr == requested_addr) {
3329      return requested_addr;
3330   }
3331 
3332   if (addr != NULL) {
3333      // mmap() is successful but it fails to reserve at the requested address
3334      anon_munmap(addr, bytes);
3335   }
3336 
3337   int i;
3338   for (i = 0; i < max_tries; ++i) {
3339     base[i] = reserve_memory(bytes);
3340 
3341     if (base[i] != NULL) {
3342       // Is this the block we wanted?
3343       if (base[i] == requested_addr) {
3344         size[i] = bytes;
3345         break;
3346       }
3347 
3348       // Does this overlap the block we wanted? Give back the overlapped
3349       // parts and try again.
3350 
3351       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3352       if (top_overlap >= 0 && top_overlap < bytes) {
3353         unmap_memory(base[i], top_overlap);
3354         base[i] += top_overlap;
3355         size[i] = bytes - top_overlap;
3356       } else {
3357         size_t bottom_overlap = base[i] + bytes - requested_addr;
3358         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3359           unmap_memory(requested_addr, bottom_overlap);
3360           size[i] = bytes - bottom_overlap;
3361         } else {
3362           size[i] = bytes;
3363         }
3364       }
3365     }
3366   }
3367 
3368   // Give back the unused reserved pieces.
3369 
3370   for (int j = 0; j < i; ++j) {
3371     if (base[j] != NULL) {
3372       unmap_memory(base[j], size[j]);
3373     }
3374   }
3375 
3376   if (i < max_tries) {
3377     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3378     return requested_addr;
3379   } else {
3380     _highest_vm_reserved_address = old_highest;
3381     return NULL;
3382   }
3383 }
3384 
3385 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3386   return ::read(fd, buf, nBytes);
3387 }
3388 
3389 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3390 // Solaris uses poll(), linux uses park().
3391 // Poll() is likely a better choice, assuming that Thread.interrupt()
3392 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3393 // SIGSEGV, see 4355769.
3394 
3395 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3396   assert(thread == Thread::current(),  "thread consistency check");
3397 
3398   ParkEvent * const slp = thread->_SleepEvent ;
3399   slp->reset() ;
3400   OrderAccess::fence() ;
3401 
3402   if (interruptible) {
3403     jlong prevtime = javaTimeNanos();
3404 
3405     for (;;) {
3406       if (os::is_interrupted(thread, true)) {
3407         return OS_INTRPT;
3408       }
3409 
3410       jlong newtime = javaTimeNanos();
3411 
3412       if (newtime - prevtime < 0) {
3413         // time moving backwards, should only happen if no monotonic clock
3414         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3415         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3416       } else {
3417         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3418       }
3419 
3420       if(millis <= 0) {
3421         return OS_OK;
3422       }
3423 
3424       prevtime = newtime;
3425 
3426       {
3427         assert(thread->is_Java_thread(), "sanity check");
3428         JavaThread *jt = (JavaThread *) thread;
3429         ThreadBlockInVM tbivm(jt);
3430         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3431 
3432         jt->set_suspend_equivalent();
3433         // cleared by handle_special_suspend_equivalent_condition() or
3434         // java_suspend_self() via check_and_wait_while_suspended()
3435 
3436         slp->park(millis);
3437 
3438         // were we externally suspended while we were waiting?
3439         jt->check_and_wait_while_suspended();
3440       }
3441     }
3442   } else {
3443     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3444     jlong prevtime = javaTimeNanos();
3445 
3446     for (;;) {
3447       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3448       // the 1st iteration ...
3449       jlong newtime = javaTimeNanos();
3450 
3451       if (newtime - prevtime < 0) {
3452         // time moving backwards, should only happen if no monotonic clock
3453         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3454         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3455       } else {
3456         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3457       }
3458 
3459       if(millis <= 0) break ;
3460 
3461       prevtime = newtime;
3462       slp->park(millis);
3463     }
3464     return OS_OK ;
3465   }
3466 }
3467 
3468 int os::naked_sleep() {
3469   // %% make the sleep time an integer flag. for now use 1 millisec.
3470   return os::sleep(Thread::current(), 1, false);
3471 }
3472 
3473 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3474 void os::infinite_sleep() {
3475   while (true) {    // sleep forever ...
3476     ::sleep(100);   // ... 100 seconds at a time
3477   }
3478 }
3479 
3480 // Used to convert frequent JVM_Yield() to nops
3481 bool os::dont_yield() {
3482   return DontYieldALot;
3483 }
3484 
3485 void os::yield() {
3486   sched_yield();
3487 }
3488 
3489 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3490 
3491 void os::yield_all(int attempts) {
3492   // Yields to all threads, including threads with lower priorities
3493   // Threads on Linux are all with same priority. The Solaris style
3494   // os::yield_all() with nanosleep(1ms) is not necessary.
3495   sched_yield();
3496 }
3497 
3498 // Called from the tight loops to possibly influence time-sharing heuristics
3499 void os::loop_breaker(int attempts) {
3500   os::yield_all(attempts);
3501 }
3502 
3503 ////////////////////////////////////////////////////////////////////////////////
3504 // thread priority support
3505 
3506 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3507 // only supports dynamic priority, static priority must be zero. For real-time
3508 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3509 // However, for large multi-threaded applications, SCHED_RR is not only slower
3510 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3511 // of 5 runs - Sep 2005).
3512 //
3513 // The following code actually changes the niceness of kernel-thread/LWP. It
3514 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3515 // not the entire user process, and user level threads are 1:1 mapped to kernel
3516 // threads. It has always been the case, but could change in the future. For
3517 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3518 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3519 
3520 int os::java_to_os_priority[CriticalPriority + 1] = {
3521   19,              // 0 Entry should never be used
3522 
3523    4,              // 1 MinPriority
3524    3,              // 2
3525    2,              // 3
3526 
3527    1,              // 4
3528    0,              // 5 NormPriority
3529   -1,              // 6
3530 
3531   -2,              // 7
3532   -3,              // 8
3533   -4,              // 9 NearMaxPriority
3534 
3535   -5,              // 10 MaxPriority
3536 
3537   -5               // 11 CriticalPriority
3538 };
3539 
3540 static int prio_init() {
3541   if (ThreadPriorityPolicy == 1) {
3542     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3543     // if effective uid is not root. Perhaps, a more elegant way of doing
3544     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3545     if (geteuid() != 0) {
3546       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3547         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3548       }
3549       ThreadPriorityPolicy = 0;
3550     }
3551   }
3552   if (UseCriticalJavaThreadPriority) {
3553     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3554   }
3555   return 0;
3556 }
3557 
3558 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3559   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3560 
3561   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3562   return (ret == 0) ? OS_OK : OS_ERR;
3563 }
3564 
3565 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3566   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3567     *priority_ptr = java_to_os_priority[NormPriority];
3568     return OS_OK;
3569   }
3570 
3571   errno = 0;
3572   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3573   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3574 }
3575 
3576 // Hint to the underlying OS that a task switch would not be good.
3577 // Void return because it's a hint and can fail.
3578 void os::hint_no_preempt() {}
3579 
3580 ////////////////////////////////////////////////////////////////////////////////
3581 // suspend/resume support
3582 
3583 //  the low-level signal-based suspend/resume support is a remnant from the
3584 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3585 //  within hotspot. Now there is a single use-case for this:
3586 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3587 //      that runs in the watcher thread.
3588 //  The remaining code is greatly simplified from the more general suspension
3589 //  code that used to be used.
3590 //
3591 //  The protocol is quite simple:
3592 //  - suspend:
3593 //      - sends a signal to the target thread
3594 //      - polls the suspend state of the osthread using a yield loop
3595 //      - target thread signal handler (SR_handler) sets suspend state
3596 //        and blocks in sigsuspend until continued
3597 //  - resume:
3598 //      - sets target osthread state to continue
3599 //      - sends signal to end the sigsuspend loop in the SR_handler
3600 //
3601 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3602 //
3603 
3604 static void resume_clear_context(OSThread *osthread) {
3605   osthread->set_ucontext(NULL);
3606   osthread->set_siginfo(NULL);
3607 }
3608 
3609 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3610   osthread->set_ucontext(context);
3611   osthread->set_siginfo(siginfo);
3612 }
3613 
3614 //
3615 // Handler function invoked when a thread's execution is suspended or
3616 // resumed. We have to be careful that only async-safe functions are
3617 // called here (Note: most pthread functions are not async safe and
3618 // should be avoided.)
3619 //
3620 // Note: sigwait() is a more natural fit than sigsuspend() from an
3621 // interface point of view, but sigwait() prevents the signal hander
3622 // from being run. libpthread would get very confused by not having
3623 // its signal handlers run and prevents sigwait()'s use with the
3624 // mutex granting granting signal.
3625 //
3626 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3627 //
3628 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3629   // Save and restore errno to avoid confusing native code with EINTR
3630   // after sigsuspend.
3631   int old_errno = errno;
3632 
3633   Thread* thread = Thread::current();
3634   OSThread* osthread = thread->osthread();
3635   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3636 
3637   os::SuspendResume::State current = osthread->sr.state();
3638   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3639     suspend_save_context(osthread, siginfo, context);
3640 
3641     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3642     os::SuspendResume::State state = osthread->sr.suspended();
3643     if (state == os::SuspendResume::SR_SUSPENDED) {
3644       sigset_t suspend_set;  // signals for sigsuspend()
3645 
3646       // get current set of blocked signals and unblock resume signal
3647       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3648       sigdelset(&suspend_set, SR_signum);
3649 
3650       sr_semaphore.signal();
3651       // wait here until we are resumed
3652       while (1) {
3653         sigsuspend(&suspend_set);
3654 
3655         os::SuspendResume::State result = osthread->sr.running();
3656         if (result == os::SuspendResume::SR_RUNNING) {
3657           sr_semaphore.signal();
3658           break;
3659         }
3660       }
3661 
3662     } else if (state == os::SuspendResume::SR_RUNNING) {
3663       // request was cancelled, continue
3664     } else {
3665       ShouldNotReachHere();
3666     }
3667 
3668     resume_clear_context(osthread);
3669   } else if (current == os::SuspendResume::SR_RUNNING) {
3670     // request was cancelled, continue
3671   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3672     // ignore
3673   } else {
3674     // ignore
3675   }
3676 
3677   errno = old_errno;
3678 }
3679 
3680 
3681 static int SR_initialize() {
3682   struct sigaction act;
3683   char *s;
3684   /* Get signal number to use for suspend/resume */
3685   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3686     int sig = ::strtol(s, 0, 10);
3687     if (sig > 0 || sig < _NSIG) {
3688         SR_signum = sig;
3689     }
3690   }
3691 
3692   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3693         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3694 
3695   sigemptyset(&SR_sigset);
3696   sigaddset(&SR_sigset, SR_signum);
3697 
3698   /* Set up signal handler for suspend/resume */
3699   act.sa_flags = SA_RESTART|SA_SIGINFO;
3700   act.sa_handler = (void (*)(int)) SR_handler;
3701 
3702   // SR_signum is blocked by default.
3703   // 4528190 - We also need to block pthread restart signal (32 on all
3704   // supported Linux platforms). Note that LinuxThreads need to block
3705   // this signal for all threads to work properly. So we don't have
3706   // to use hard-coded signal number when setting up the mask.
3707   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3708 
3709   if (sigaction(SR_signum, &act, 0) == -1) {
3710     return -1;
3711   }
3712 
3713   // Save signal flag
3714   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3715   return 0;
3716 }
3717 
3718 static int sr_notify(OSThread* osthread) {
3719   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3720   assert_status(status == 0, status, "pthread_kill");
3721   return status;
3722 }
3723 
3724 // "Randomly" selected value for how long we want to spin
3725 // before bailing out on suspending a thread, also how often
3726 // we send a signal to a thread we want to resume
3727 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3728 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3729 
3730 // returns true on success and false on error - really an error is fatal
3731 // but this seems the normal response to library errors
3732 static bool do_suspend(OSThread* osthread) {
3733   assert(osthread->sr.is_running(), "thread should be running");
3734   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3735 
3736   // mark as suspended and send signal
3737   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3738     // failed to switch, state wasn't running?
3739     ShouldNotReachHere();
3740     return false;
3741   }
3742 
3743   if (sr_notify(osthread) != 0) {
3744     ShouldNotReachHere();
3745   }
3746 
3747   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3748   while (true) {
3749     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3750       break;
3751     } else {
3752       // timeout
3753       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3754       if (cancelled == os::SuspendResume::SR_RUNNING) {
3755         return false;
3756       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3757         // make sure that we consume the signal on the semaphore as well
3758         sr_semaphore.wait();
3759         break;
3760       } else {
3761         ShouldNotReachHere();
3762         return false;
3763       }
3764     }
3765   }
3766 
3767   guarantee(osthread->sr.is_suspended(), "Must be suspended");
3768   return true;
3769 }
3770 
3771 static void do_resume(OSThread* osthread) {
3772   assert(osthread->sr.is_suspended(), "thread should be suspended");
3773   assert(!sr_semaphore.trywait(), "invalid semaphore state");
3774 
3775   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3776     // failed to switch to WAKEUP_REQUEST
3777     ShouldNotReachHere();
3778     return;
3779   }
3780 
3781   while (true) {
3782     if (sr_notify(osthread) == 0) {
3783       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3784         if (osthread->sr.is_running()) {
3785           return;
3786         }
3787       }
3788     } else {
3789       ShouldNotReachHere();
3790     }
3791   }
3792 
3793   guarantee(osthread->sr.is_running(), "Must be running!");
3794 }
3795 
3796 ////////////////////////////////////////////////////////////////////////////////
3797 // interrupt support
3798 
3799 void os::interrupt(Thread* thread) {
3800   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3801     "possibility of dangling Thread pointer");
3802 
3803   OSThread* osthread = thread->osthread();
3804 
3805   if (!osthread->interrupted()) {
3806     osthread->set_interrupted(true);
3807     // More than one thread can get here with the same value of osthread,
3808     // resulting in multiple notifications.  We do, however, want the store
3809     // to interrupted() to be visible to other threads before we execute unpark().
3810     OrderAccess::fence();
3811     ParkEvent * const slp = thread->_SleepEvent ;
3812     if (slp != NULL) slp->unpark() ;
3813   }
3814 
3815   // For JSR166. Unpark even if interrupt status already was set
3816   if (thread->is_Java_thread())
3817     ((JavaThread*)thread)->parker()->unpark();
3818 
3819   ParkEvent * ev = thread->_ParkEvent ;
3820   if (ev != NULL) ev->unpark() ;
3821 
3822 }
3823 
3824 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3825   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3826     "possibility of dangling Thread pointer");
3827 
3828   OSThread* osthread = thread->osthread();
3829 
3830   bool interrupted = osthread->interrupted();
3831 
3832   if (interrupted && clear_interrupted) {
3833     osthread->set_interrupted(false);
3834     // consider thread->_SleepEvent->reset() ... optional optimization
3835   }
3836 
3837   return interrupted;
3838 }
3839 
3840 ///////////////////////////////////////////////////////////////////////////////////
3841 // signal handling (except suspend/resume)
3842 
3843 // This routine may be used by user applications as a "hook" to catch signals.
3844 // The user-defined signal handler must pass unrecognized signals to this
3845 // routine, and if it returns true (non-zero), then the signal handler must
3846 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3847 // routine will never retun false (zero), but instead will execute a VM panic
3848 // routine kill the process.
3849 //
3850 // If this routine returns false, it is OK to call it again.  This allows
3851 // the user-defined signal handler to perform checks either before or after
3852 // the VM performs its own checks.  Naturally, the user code would be making
3853 // a serious error if it tried to handle an exception (such as a null check
3854 // or breakpoint) that the VM was generating for its own correct operation.
3855 //
3856 // This routine may recognize any of the following kinds of signals:
3857 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3858 // It should be consulted by handlers for any of those signals.
3859 //
3860 // The caller of this routine must pass in the three arguments supplied
3861 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3862 // field of the structure passed to sigaction().  This routine assumes that
3863 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3864 //
3865 // Note that the VM will print warnings if it detects conflicting signal
3866 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3867 //
3868 extern "C" JNIEXPORT int
3869 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3870                         void* ucontext, int abort_if_unrecognized);
3871 
3872 void signalHandler(int sig, siginfo_t* info, void* uc) {
3873   assert(info != NULL && uc != NULL, "it must be old kernel");
3874   int orig_errno = errno;  // Preserve errno value over signal handler.
3875   JVM_handle_linux_signal(sig, info, uc, true);
3876   errno = orig_errno;
3877 }
3878 
3879 
3880 // This boolean allows users to forward their own non-matching signals
3881 // to JVM_handle_linux_signal, harmlessly.
3882 bool os::Linux::signal_handlers_are_installed = false;
3883 
3884 // For signal-chaining
3885 struct sigaction os::Linux::sigact[MAXSIGNUM];
3886 unsigned int os::Linux::sigs = 0;
3887 bool os::Linux::libjsig_is_loaded = false;
3888 typedef struct sigaction *(*get_signal_t)(int);
3889 get_signal_t os::Linux::get_signal_action = NULL;
3890 
3891 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3892   struct sigaction *actp = NULL;
3893 
3894   if (libjsig_is_loaded) {
3895     // Retrieve the old signal handler from libjsig
3896     actp = (*get_signal_action)(sig);
3897   }
3898   if (actp == NULL) {
3899     // Retrieve the preinstalled signal handler from jvm
3900     actp = get_preinstalled_handler(sig);
3901   }
3902 
3903   return actp;
3904 }
3905 
3906 static bool call_chained_handler(struct sigaction *actp, int sig,
3907                                  siginfo_t *siginfo, void *context) {
3908   // Call the old signal handler
3909   if (actp->sa_handler == SIG_DFL) {
3910     // It's more reasonable to let jvm treat it as an unexpected exception
3911     // instead of taking the default action.
3912     return false;
3913   } else if (actp->sa_handler != SIG_IGN) {
3914     if ((actp->sa_flags & SA_NODEFER) == 0) {
3915       // automaticlly block the signal
3916       sigaddset(&(actp->sa_mask), sig);
3917     }
3918 
3919     sa_handler_t hand;
3920     sa_sigaction_t sa;
3921     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3922     // retrieve the chained handler
3923     if (siginfo_flag_set) {
3924       sa = actp->sa_sigaction;
3925     } else {
3926       hand = actp->sa_handler;
3927     }
3928 
3929     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3930       actp->sa_handler = SIG_DFL;
3931     }
3932 
3933     // try to honor the signal mask
3934     sigset_t oset;
3935     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3936 
3937     // call into the chained handler
3938     if (siginfo_flag_set) {
3939       (*sa)(sig, siginfo, context);
3940     } else {
3941       (*hand)(sig);
3942     }
3943 
3944     // restore the signal mask
3945     pthread_sigmask(SIG_SETMASK, &oset, 0);
3946   }
3947   // Tell jvm's signal handler the signal is taken care of.
3948   return true;
3949 }
3950 
3951 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3952   bool chained = false;
3953   // signal-chaining
3954   if (UseSignalChaining) {
3955     struct sigaction *actp = get_chained_signal_action(sig);
3956     if (actp != NULL) {
3957       chained = call_chained_handler(actp, sig, siginfo, context);
3958     }
3959   }
3960   return chained;
3961 }
3962 
3963 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3964   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3965     return &sigact[sig];
3966   }
3967   return NULL;
3968 }
3969 
3970 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3971   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3972   sigact[sig] = oldAct;
3973   sigs |= (unsigned int)1 << sig;
3974 }
3975 
3976 // for diagnostic
3977 int os::Linux::sigflags[MAXSIGNUM];
3978 
3979 int os::Linux::get_our_sigflags(int sig) {
3980   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3981   return sigflags[sig];
3982 }
3983 
3984 void os::Linux::set_our_sigflags(int sig, int flags) {
3985   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3986   sigflags[sig] = flags;
3987 }
3988 
3989 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3990   // Check for overwrite.
3991   struct sigaction oldAct;
3992   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3993 
3994   void* oldhand = oldAct.sa_sigaction
3995                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3996                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3997   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3998       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3999       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4000     if (AllowUserSignalHandlers || !set_installed) {
4001       // Do not overwrite; user takes responsibility to forward to us.
4002       return;
4003     } else if (UseSignalChaining) {
4004       // save the old handler in jvm
4005       save_preinstalled_handler(sig, oldAct);
4006       // libjsig also interposes the sigaction() call below and saves the
4007       // old sigaction on it own.
4008     } else {
4009       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4010                     "%#lx for signal %d.", (long)oldhand, sig));
4011     }
4012   }
4013 
4014   struct sigaction sigAct;
4015   sigfillset(&(sigAct.sa_mask));
4016   sigAct.sa_handler = SIG_DFL;
4017   if (!set_installed) {
4018     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4019   } else {
4020     sigAct.sa_sigaction = signalHandler;
4021     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4022   }
4023   // Save flags, which are set by ours
4024   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4025   sigflags[sig] = sigAct.sa_flags;
4026 
4027   int ret = sigaction(sig, &sigAct, &oldAct);
4028   assert(ret == 0, "check");
4029 
4030   void* oldhand2  = oldAct.sa_sigaction
4031                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4032                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4033   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4034 }
4035 
4036 // install signal handlers for signals that HotSpot needs to
4037 // handle in order to support Java-level exception handling.
4038 
4039 void os::Linux::install_signal_handlers() {
4040   if (!signal_handlers_are_installed) {
4041     signal_handlers_are_installed = true;
4042 
4043     // signal-chaining
4044     typedef void (*signal_setting_t)();
4045     signal_setting_t begin_signal_setting = NULL;
4046     signal_setting_t end_signal_setting = NULL;
4047     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4048                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4049     if (begin_signal_setting != NULL) {
4050       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4051                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4052       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4053                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4054       libjsig_is_loaded = true;
4055       assert(UseSignalChaining, "should enable signal-chaining");
4056     }
4057     if (libjsig_is_loaded) {
4058       // Tell libjsig jvm is setting signal handlers
4059       (*begin_signal_setting)();
4060     }
4061 
4062     set_signal_handler(SIGSEGV, true);
4063     set_signal_handler(SIGPIPE, true);
4064     set_signal_handler(SIGBUS, true);
4065     set_signal_handler(SIGILL, true);
4066     set_signal_handler(SIGFPE, true);
4067     set_signal_handler(SIGXFSZ, true);
4068 
4069     if (libjsig_is_loaded) {
4070       // Tell libjsig jvm finishes setting signal handlers
4071       (*end_signal_setting)();
4072     }
4073 
4074     // We don't activate signal checker if libjsig is in place, we trust ourselves
4075     // and if UserSignalHandler is installed all bets are off.
4076     // Log that signal checking is off only if -verbose:jni is specified.
4077     if (CheckJNICalls) {
4078       if (libjsig_is_loaded) {
4079         if (PrintJNIResolving) {
4080           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4081         }
4082         check_signals = false;
4083       }
4084       if (AllowUserSignalHandlers) {
4085         if (PrintJNIResolving) {
4086           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4087         }
4088         check_signals = false;
4089       }
4090     }
4091   }
4092 }
4093 
4094 // This is the fastest way to get thread cpu time on Linux.
4095 // Returns cpu time (user+sys) for any thread, not only for current.
4096 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4097 // It might work on 2.6.10+ with a special kernel/glibc patch.
4098 // For reference, please, see IEEE Std 1003.1-2004:
4099 //   http://www.unix.org/single_unix_specification
4100 
4101 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4102   struct timespec tp;
4103   int rc = os::Linux::clock_gettime(clockid, &tp);
4104   assert(rc == 0, "clock_gettime is expected to return 0 code");
4105 
4106   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4107 }
4108 
4109 /////
4110 // glibc on Linux platform uses non-documented flag
4111 // to indicate, that some special sort of signal
4112 // trampoline is used.
4113 // We will never set this flag, and we should
4114 // ignore this flag in our diagnostic
4115 #ifdef SIGNIFICANT_SIGNAL_MASK
4116 #undef SIGNIFICANT_SIGNAL_MASK
4117 #endif
4118 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4119 
4120 static const char* get_signal_handler_name(address handler,
4121                                            char* buf, int buflen) {
4122   int offset;
4123   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4124   if (found) {
4125     // skip directory names
4126     const char *p1, *p2;
4127     p1 = buf;
4128     size_t len = strlen(os::file_separator());
4129     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4130     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4131   } else {
4132     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4133   }
4134   return buf;
4135 }
4136 
4137 static void print_signal_handler(outputStream* st, int sig,
4138                                  char* buf, size_t buflen) {
4139   struct sigaction sa;
4140 
4141   sigaction(sig, NULL, &sa);
4142 
4143   // See comment for SIGNIFICANT_SIGNAL_MASK define
4144   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4145 
4146   st->print("%s: ", os::exception_name(sig, buf, buflen));
4147 
4148   address handler = (sa.sa_flags & SA_SIGINFO)
4149     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4150     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4151 
4152   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4153     st->print("SIG_DFL");
4154   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4155     st->print("SIG_IGN");
4156   } else {
4157     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4158   }
4159 
4160   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4161 
4162   address rh = VMError::get_resetted_sighandler(sig);
4163   // May be, handler was resetted by VMError?
4164   if(rh != NULL) {
4165     handler = rh;
4166     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4167   }
4168 
4169   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
4170 
4171   // Check: is it our handler?
4172   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4173      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4174     // It is our signal handler
4175     // check for flags, reset system-used one!
4176     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4177       st->print(
4178                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4179                 os::Linux::get_our_sigflags(sig));
4180     }
4181   }
4182   st->cr();
4183 }
4184 
4185 
4186 #define DO_SIGNAL_CHECK(sig) \
4187   if (!sigismember(&check_signal_done, sig)) \
4188     os::Linux::check_signal_handler(sig)
4189 
4190 // This method is a periodic task to check for misbehaving JNI applications
4191 // under CheckJNI, we can add any periodic checks here
4192 
4193 void os::run_periodic_checks() {
4194 
4195   if (check_signals == false) return;
4196 
4197   // SEGV and BUS if overridden could potentially prevent
4198   // generation of hs*.log in the event of a crash, debugging
4199   // such a case can be very challenging, so we absolutely
4200   // check the following for a good measure:
4201   DO_SIGNAL_CHECK(SIGSEGV);
4202   DO_SIGNAL_CHECK(SIGILL);
4203   DO_SIGNAL_CHECK(SIGFPE);
4204   DO_SIGNAL_CHECK(SIGBUS);
4205   DO_SIGNAL_CHECK(SIGPIPE);
4206   DO_SIGNAL_CHECK(SIGXFSZ);
4207 
4208 
4209   // ReduceSignalUsage allows the user to override these handlers
4210   // see comments at the very top and jvm_solaris.h
4211   if (!ReduceSignalUsage) {
4212     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4213     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4214     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4215     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4216   }
4217 
4218   DO_SIGNAL_CHECK(SR_signum);
4219   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4220 }
4221 
4222 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4223 
4224 static os_sigaction_t os_sigaction = NULL;
4225 
4226 void os::Linux::check_signal_handler(int sig) {
4227   char buf[O_BUFLEN];
4228   address jvmHandler = NULL;
4229 
4230 
4231   struct sigaction act;
4232   if (os_sigaction == NULL) {
4233     // only trust the default sigaction, in case it has been interposed
4234     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4235     if (os_sigaction == NULL) return;
4236   }
4237 
4238   os_sigaction(sig, (struct sigaction*)NULL, &act);
4239 
4240 
4241   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4242 
4243   address thisHandler = (act.sa_flags & SA_SIGINFO)
4244     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4245     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4246 
4247 
4248   switch(sig) {
4249   case SIGSEGV:
4250   case SIGBUS:
4251   case SIGFPE:
4252   case SIGPIPE:
4253   case SIGILL:
4254   case SIGXFSZ:
4255     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4256     break;
4257 
4258   case SHUTDOWN1_SIGNAL:
4259   case SHUTDOWN2_SIGNAL:
4260   case SHUTDOWN3_SIGNAL:
4261   case BREAK_SIGNAL:
4262     jvmHandler = (address)user_handler();
4263     break;
4264 
4265   case INTERRUPT_SIGNAL:
4266     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4267     break;
4268 
4269   default:
4270     if (sig == SR_signum) {
4271       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4272     } else {
4273       return;
4274     }
4275     break;
4276   }
4277 
4278   if (thisHandler != jvmHandler) {
4279     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4280     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4281     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4282     // No need to check this sig any longer
4283     sigaddset(&check_signal_done, sig);
4284   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4285     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4286     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4287     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4288     // No need to check this sig any longer
4289     sigaddset(&check_signal_done, sig);
4290   }
4291 
4292   // Dump all the signal
4293   if (sigismember(&check_signal_done, sig)) {
4294     print_signal_handlers(tty, buf, O_BUFLEN);
4295   }
4296 }
4297 
4298 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4299 
4300 extern bool signal_name(int signo, char* buf, size_t len);
4301 
4302 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4303   if (0 < exception_code && exception_code <= SIGRTMAX) {
4304     // signal
4305     if (!signal_name(exception_code, buf, size)) {
4306       jio_snprintf(buf, size, "SIG%d", exception_code);
4307     }
4308     return buf;
4309   } else {
4310     return NULL;
4311   }
4312 }
4313 
4314 // this is called _before_ the most of global arguments have been parsed
4315 void os::init(void) {
4316   char dummy;   /* used to get a guess on initial stack address */
4317 //  first_hrtime = gethrtime();
4318 
4319   // With LinuxThreads the JavaMain thread pid (primordial thread)
4320   // is different than the pid of the java launcher thread.
4321   // So, on Linux, the launcher thread pid is passed to the VM
4322   // via the sun.java.launcher.pid property.
4323   // Use this property instead of getpid() if it was correctly passed.
4324   // See bug 6351349.
4325   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4326 
4327   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4328 
4329   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4330 
4331   init_random(1234567);
4332 
4333   ThreadCritical::initialize();
4334 
4335   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4336   if (Linux::page_size() == -1) {
4337     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4338                   strerror(errno)));
4339   }
4340   init_page_sizes((size_t) Linux::page_size());
4341 
4342   Linux::initialize_system_info();
4343 
4344   // main_thread points to the aboriginal thread
4345   Linux::_main_thread = pthread_self();
4346 
4347   Linux::clock_init();
4348   initial_time_count = os::elapsed_counter();
4349   pthread_mutex_init(&dl_mutex, NULL);
4350 
4351   // If the pagesize of the VM is greater than 8K determine the appropriate
4352   // number of initial guard pages.  The user can change this with the
4353   // command line arguments, if needed.
4354   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4355     StackYellowPages = 1;
4356     StackRedPages = 1;
4357     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4358   }
4359 }
4360 
4361 // To install functions for atexit system call
4362 extern "C" {
4363   static void perfMemory_exit_helper() {
4364     perfMemory_exit();
4365   }
4366 }
4367 
4368 // this is called _after_ the global arguments have been parsed
4369 jint os::init_2(void)
4370 {
4371   Linux::fast_thread_clock_init();
4372 
4373   // Allocate a single page and mark it as readable for safepoint polling
4374   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4375   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4376 
4377   os::set_polling_page( polling_page );
4378 
4379 #ifndef PRODUCT
4380   if(Verbose && PrintMiscellaneous)
4381     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4382 #endif
4383 
4384   if (!UseMembar) {
4385     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4386     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4387     os::set_memory_serialize_page( mem_serialize_page );
4388 
4389 #ifndef PRODUCT
4390     if(Verbose && PrintMiscellaneous)
4391       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4392 #endif
4393   }
4394 
4395   os::large_page_init();
4396 
4397   // initialize suspend/resume support - must do this before signal_sets_init()
4398   if (SR_initialize() != 0) {
4399     perror("SR_initialize failed");
4400     return JNI_ERR;
4401   }
4402 
4403   Linux::signal_sets_init();
4404   Linux::install_signal_handlers();
4405 
4406   // Check minimum allowable stack size for thread creation and to initialize
4407   // the java system classes, including StackOverflowError - depends on page
4408   // size.  Add a page for compiler2 recursion in main thread.
4409   // Add in 2*BytesPerWord times page size to account for VM stack during
4410   // class initialization depending on 32 or 64 bit VM.
4411   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4412             (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4413                     (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4414 
4415   size_t threadStackSizeInBytes = ThreadStackSize * K;
4416   if (threadStackSizeInBytes != 0 &&
4417       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4418         tty->print_cr("\nThe stack size specified is too small, "
4419                       "Specify at least %dk",
4420                       os::Linux::min_stack_allowed/ K);
4421         return JNI_ERR;
4422   }
4423 
4424   // Make the stack size a multiple of the page size so that
4425   // the yellow/red zones can be guarded.
4426   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4427         vm_page_size()));
4428 
4429   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4430 
4431   Linux::libpthread_init();
4432   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4433      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4434           Linux::glibc_version(), Linux::libpthread_version(),
4435           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4436   }
4437 
4438   if (UseNUMA) {
4439     if (!Linux::libnuma_init()) {
4440       UseNUMA = false;
4441     } else {
4442       if ((Linux::numa_max_node() < 1)) {
4443         // There's only one node(they start from 0), disable NUMA.
4444         UseNUMA = false;
4445       }
4446     }
4447     // With SHM large pages we cannot uncommit a page, so there's not way
4448     // we can make the adaptive lgrp chunk resizing work. If the user specified
4449     // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4450     // disable adaptive resizing.
4451     if (UseNUMA && UseLargePages && UseSHM) {
4452       if (!FLAG_IS_DEFAULT(UseNUMA)) {
4453         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4454           UseLargePages = false;
4455         } else {
4456           warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4457           UseAdaptiveSizePolicy = false;
4458           UseAdaptiveNUMAChunkSizing = false;
4459         }
4460       } else {
4461         UseNUMA = false;
4462       }
4463     }
4464     if (!UseNUMA && ForceNUMA) {
4465       UseNUMA = true;
4466     }
4467   }
4468 
4469   if (MaxFDLimit) {
4470     // set the number of file descriptors to max. print out error
4471     // if getrlimit/setrlimit fails but continue regardless.
4472     struct rlimit nbr_files;
4473     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4474     if (status != 0) {
4475       if (PrintMiscellaneous && (Verbose || WizardMode))
4476         perror("os::init_2 getrlimit failed");
4477     } else {
4478       nbr_files.rlim_cur = nbr_files.rlim_max;
4479       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4480       if (status != 0) {
4481         if (PrintMiscellaneous && (Verbose || WizardMode))
4482           perror("os::init_2 setrlimit failed");
4483       }
4484     }
4485   }
4486 
4487   // Initialize lock used to serialize thread creation (see os::create_thread)
4488   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4489 
4490   // at-exit methods are called in the reverse order of their registration.
4491   // atexit functions are called on return from main or as a result of a
4492   // call to exit(3C). There can be only 32 of these functions registered
4493   // and atexit() does not set errno.
4494 
4495   if (PerfAllowAtExitRegistration) {
4496     // only register atexit functions if PerfAllowAtExitRegistration is set.
4497     // atexit functions can be delayed until process exit time, which
4498     // can be problematic for embedded VM situations. Embedded VMs should
4499     // call DestroyJavaVM() to assure that VM resources are released.
4500 
4501     // note: perfMemory_exit_helper atexit function may be removed in
4502     // the future if the appropriate cleanup code can be added to the
4503     // VM_Exit VMOperation's doit method.
4504     if (atexit(perfMemory_exit_helper) != 0) {
4505       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4506     }
4507   }
4508 
4509   // initialize thread priority policy
4510   prio_init();
4511 
4512   return JNI_OK;
4513 }
4514 
4515 // this is called at the end of vm_initialization
4516 void os::init_3(void)
4517 {
4518 #ifdef JAVASE_EMBEDDED
4519   // Start the MemNotifyThread
4520   if (LowMemoryProtection) {
4521     MemNotifyThread::start();
4522   }
4523   return;
4524 #endif
4525 }
4526 
4527 // Mark the polling page as unreadable
4528 void os::make_polling_page_unreadable(void) {
4529   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4530     fatal("Could not disable polling page");
4531 };
4532 
4533 // Mark the polling page as readable
4534 void os::make_polling_page_readable(void) {
4535   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4536     fatal("Could not enable polling page");
4537   }
4538 };
4539 
4540 int os::active_processor_count() {
4541   // Linux doesn't yet have a (official) notion of processor sets,
4542   // so just return the number of online processors.
4543   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4544   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4545   return online_cpus;
4546 }
4547 
4548 void os::set_native_thread_name(const char *name) {
4549   // Not yet implemented.
4550   return;
4551 }
4552 
4553 bool os::distribute_processes(uint length, uint* distribution) {
4554   // Not yet implemented.
4555   return false;
4556 }
4557 
4558 bool os::bind_to_processor(uint processor_id) {
4559   // Not yet implemented.
4560   return false;
4561 }
4562 
4563 ///
4564 
4565 void os::SuspendedThreadTask::internal_do_task() {
4566   if (do_suspend(_thread->osthread())) {
4567     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4568     do_task(context);
4569     do_resume(_thread->osthread());
4570   }
4571 }
4572 
4573 class PcFetcher : public os::SuspendedThreadTask {
4574 public:
4575   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4576   ExtendedPC result();
4577 protected:
4578   void do_task(const os::SuspendedThreadTaskContext& context);
4579 private:
4580   ExtendedPC _epc;
4581 };
4582 
4583 ExtendedPC PcFetcher::result() {
4584   guarantee(is_done(), "task is not done yet.");
4585   return _epc;
4586 }
4587 
4588 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4589   Thread* thread = context.thread();
4590   OSThread* osthread = thread->osthread();
4591   if (osthread->ucontext() != NULL) {
4592     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4593   } else {
4594     // NULL context is unexpected, double-check this is the VMThread
4595     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4596   }
4597 }
4598 
4599 // Suspends the target using the signal mechanism and then grabs the PC before
4600 // resuming the target. Used by the flat-profiler only
4601 ExtendedPC os::get_thread_pc(Thread* thread) {
4602   // Make sure that it is called by the watcher for the VMThread
4603   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4604   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4605 
4606   PcFetcher fetcher(thread);
4607   fetcher.run();
4608   return fetcher.result();
4609 }
4610 
4611 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4612 {
4613    if (is_NPTL()) {
4614       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4615    } else {
4616       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4617       // word back to default 64bit precision if condvar is signaled. Java
4618       // wants 53bit precision.  Save and restore current value.
4619       int fpu = get_fpu_control_word();
4620       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4621       set_fpu_control_word(fpu);
4622       return status;
4623    }
4624 }
4625 
4626 ////////////////////////////////////////////////////////////////////////////////
4627 // debug support
4628 
4629 bool os::find(address addr, outputStream* st) {
4630   Dl_info dlinfo;
4631   memset(&dlinfo, 0, sizeof(dlinfo));
4632   if (dladdr(addr, &dlinfo)) {
4633     st->print(PTR_FORMAT ": ", addr);
4634     if (dlinfo.dli_sname != NULL) {
4635       st->print("%s+%#x", dlinfo.dli_sname,
4636                  addr - (intptr_t)dlinfo.dli_saddr);
4637     } else if (dlinfo.dli_fname) {
4638       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4639     } else {
4640       st->print("<absolute address>");
4641     }
4642     if (dlinfo.dli_fname) {
4643       st->print(" in %s", dlinfo.dli_fname);
4644     }
4645     if (dlinfo.dli_fbase) {
4646       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4647     }
4648     st->cr();
4649 
4650     if (Verbose) {
4651       // decode some bytes around the PC
4652       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4653       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4654       address       lowest = (address) dlinfo.dli_sname;
4655       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4656       if (begin < lowest)  begin = lowest;
4657       Dl_info dlinfo2;
4658       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4659           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4660         end = (address) dlinfo2.dli_saddr;
4661       Disassembler::decode(begin, end, st);
4662     }
4663     return true;
4664   }
4665   return false;
4666 }
4667 
4668 ////////////////////////////////////////////////////////////////////////////////
4669 // misc
4670 
4671 // This does not do anything on Linux. This is basically a hook for being
4672 // able to use structured exception handling (thread-local exception filters)
4673 // on, e.g., Win32.
4674 void
4675 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4676                          JavaCallArguments* args, Thread* thread) {
4677   f(value, method, args, thread);
4678 }
4679 
4680 void os::print_statistics() {
4681 }
4682 
4683 int os::message_box(const char* title, const char* message) {
4684   int i;
4685   fdStream err(defaultStream::error_fd());
4686   for (i = 0; i < 78; i++) err.print_raw("=");
4687   err.cr();
4688   err.print_raw_cr(title);
4689   for (i = 0; i < 78; i++) err.print_raw("-");
4690   err.cr();
4691   err.print_raw_cr(message);
4692   for (i = 0; i < 78; i++) err.print_raw("=");
4693   err.cr();
4694 
4695   char buf[16];
4696   // Prevent process from exiting upon "read error" without consuming all CPU
4697   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4698 
4699   return buf[0] == 'y' || buf[0] == 'Y';
4700 }
4701 
4702 int os::stat(const char *path, struct stat *sbuf) {
4703   char pathbuf[MAX_PATH];
4704   if (strlen(path) > MAX_PATH - 1) {
4705     errno = ENAMETOOLONG;
4706     return -1;
4707   }
4708   os::native_path(strcpy(pathbuf, path));
4709   return ::stat(pathbuf, sbuf);
4710 }
4711 
4712 bool os::check_heap(bool force) {
4713   return true;
4714 }
4715 
4716 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4717   return ::vsnprintf(buf, count, format, args);
4718 }
4719 
4720 // Is a (classpath) directory empty?
4721 bool os::dir_is_empty(const char* path) {
4722   DIR *dir = NULL;
4723   struct dirent *ptr;
4724 
4725   dir = opendir(path);
4726   if (dir == NULL) return true;
4727 
4728   /* Scan the directory */
4729   bool result = true;
4730   char buf[sizeof(struct dirent) + MAX_PATH];
4731   while (result && (ptr = ::readdir(dir)) != NULL) {
4732     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4733       result = false;
4734     }
4735   }
4736   closedir(dir);
4737   return result;
4738 }
4739 
4740 // This code originates from JDK's sysOpen and open64_w
4741 // from src/solaris/hpi/src/system_md.c
4742 
4743 #ifndef O_DELETE
4744 #define O_DELETE 0x10000
4745 #endif
4746 
4747 // Open a file. Unlink the file immediately after open returns
4748 // if the specified oflag has the O_DELETE flag set.
4749 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4750 
4751 int os::open(const char *path, int oflag, int mode) {
4752 
4753   if (strlen(path) > MAX_PATH - 1) {
4754     errno = ENAMETOOLONG;
4755     return -1;
4756   }
4757   int fd;
4758   int o_delete = (oflag & O_DELETE);
4759   oflag = oflag & ~O_DELETE;
4760 
4761   fd = ::open64(path, oflag, mode);
4762   if (fd == -1) return -1;
4763 
4764   //If the open succeeded, the file might still be a directory
4765   {
4766     struct stat64 buf64;
4767     int ret = ::fstat64(fd, &buf64);
4768     int st_mode = buf64.st_mode;
4769 
4770     if (ret != -1) {
4771       if ((st_mode & S_IFMT) == S_IFDIR) {
4772         errno = EISDIR;
4773         ::close(fd);
4774         return -1;
4775       }
4776     } else {
4777       ::close(fd);
4778       return -1;
4779     }
4780   }
4781 
4782     /*
4783      * All file descriptors that are opened in the JVM and not
4784      * specifically destined for a subprocess should have the
4785      * close-on-exec flag set.  If we don't set it, then careless 3rd
4786      * party native code might fork and exec without closing all
4787      * appropriate file descriptors (e.g. as we do in closeDescriptors in
4788      * UNIXProcess.c), and this in turn might:
4789      *
4790      * - cause end-of-file to fail to be detected on some file
4791      *   descriptors, resulting in mysterious hangs, or
4792      *
4793      * - might cause an fopen in the subprocess to fail on a system
4794      *   suffering from bug 1085341.
4795      *
4796      * (Yes, the default setting of the close-on-exec flag is a Unix
4797      * design flaw)
4798      *
4799      * See:
4800      * 1085341: 32-bit stdio routines should support file descriptors >255
4801      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4802      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4803      */
4804 #ifdef FD_CLOEXEC
4805     {
4806         int flags = ::fcntl(fd, F_GETFD);
4807         if (flags != -1)
4808             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4809     }
4810 #endif
4811 
4812   if (o_delete != 0) {
4813     ::unlink(path);
4814   }
4815   return fd;
4816 }
4817 
4818 
4819 // create binary file, rewriting existing file if required
4820 int os::create_binary_file(const char* path, bool rewrite_existing) {
4821   int oflags = O_WRONLY | O_CREAT;
4822   if (!rewrite_existing) {
4823     oflags |= O_EXCL;
4824   }
4825   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4826 }
4827 
4828 // return current position of file pointer
4829 jlong os::current_file_offset(int fd) {
4830   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4831 }
4832 
4833 // move file pointer to the specified offset
4834 jlong os::seek_to_file_offset(int fd, jlong offset) {
4835   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4836 }
4837 
4838 // This code originates from JDK's sysAvailable
4839 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4840 
4841 int os::available(int fd, jlong *bytes) {
4842   jlong cur, end;
4843   int mode;
4844   struct stat64 buf64;
4845 
4846   if (::fstat64(fd, &buf64) >= 0) {
4847     mode = buf64.st_mode;
4848     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4849       /*
4850       * XXX: is the following call interruptible? If so, this might
4851       * need to go through the INTERRUPT_IO() wrapper as for other
4852       * blocking, interruptible calls in this file.
4853       */
4854       int n;
4855       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4856         *bytes = n;
4857         return 1;
4858       }
4859     }
4860   }
4861   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4862     return 0;
4863   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4864     return 0;
4865   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4866     return 0;
4867   }
4868   *bytes = end - cur;
4869   return 1;
4870 }
4871 
4872 int os::socket_available(int fd, jint *pbytes) {
4873   // Linux doc says EINTR not returned, unlike Solaris
4874   int ret = ::ioctl(fd, FIONREAD, pbytes);
4875 
4876   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4877   // is expected to return 0 on failure and 1 on success to the jdk.
4878   return (ret < 0) ? 0 : 1;
4879 }
4880 
4881 // Map a block of memory.
4882 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4883                      char *addr, size_t bytes, bool read_only,
4884                      bool allow_exec) {
4885   int prot;
4886   int flags = MAP_PRIVATE;
4887 
4888   if (read_only) {
4889     prot = PROT_READ;
4890   } else {
4891     prot = PROT_READ | PROT_WRITE;
4892   }
4893 
4894   if (allow_exec) {
4895     prot |= PROT_EXEC;
4896   }
4897 
4898   if (addr != NULL) {
4899     flags |= MAP_FIXED;
4900   }
4901 
4902   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4903                                      fd, file_offset);
4904   if (mapped_address == MAP_FAILED) {
4905     return NULL;
4906   }
4907   return mapped_address;
4908 }
4909 
4910 
4911 // Remap a block of memory.
4912 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4913                        char *addr, size_t bytes, bool read_only,
4914                        bool allow_exec) {
4915   // same as map_memory() on this OS
4916   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4917                         allow_exec);
4918 }
4919 
4920 
4921 // Unmap a block of memory.
4922 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4923   return munmap(addr, bytes) == 0;
4924 }
4925 
4926 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4927 
4928 static clockid_t thread_cpu_clockid(Thread* thread) {
4929   pthread_t tid = thread->osthread()->pthread_id();
4930   clockid_t clockid;
4931 
4932   // Get thread clockid
4933   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4934   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4935   return clockid;
4936 }
4937 
4938 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4939 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4940 // of a thread.
4941 //
4942 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4943 // the fast estimate available on the platform.
4944 
4945 jlong os::current_thread_cpu_time() {
4946   if (os::Linux::supports_fast_thread_cpu_time()) {
4947     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4948   } else {
4949     // return user + sys since the cost is the same
4950     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4951   }
4952 }
4953 
4954 jlong os::thread_cpu_time(Thread* thread) {
4955   // consistent with what current_thread_cpu_time() returns
4956   if (os::Linux::supports_fast_thread_cpu_time()) {
4957     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4958   } else {
4959     return slow_thread_cpu_time(thread, true /* user + sys */);
4960   }
4961 }
4962 
4963 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4964   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4965     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4966   } else {
4967     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4968   }
4969 }
4970 
4971 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4972   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4973     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4974   } else {
4975     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4976   }
4977 }
4978 
4979 //
4980 //  -1 on error.
4981 //
4982 
4983 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4984   static bool proc_task_unchecked = true;
4985   static const char *proc_stat_path = "/proc/%d/stat";
4986   pid_t  tid = thread->osthread()->thread_id();
4987   char *s;
4988   char stat[2048];
4989   int statlen;
4990   char proc_name[64];
4991   int count;
4992   long sys_time, user_time;
4993   char cdummy;
4994   int idummy;
4995   long ldummy;
4996   FILE *fp;
4997 
4998   // The /proc/<tid>/stat aggregates per-process usage on
4999   // new Linux kernels 2.6+ where NPTL is supported.
5000   // The /proc/self/task/<tid>/stat still has the per-thread usage.
5001   // See bug 6328462.
5002   // There possibly can be cases where there is no directory
5003   // /proc/self/task, so we check its availability.
5004   if (proc_task_unchecked && os::Linux::is_NPTL()) {
5005     // This is executed only once
5006     proc_task_unchecked = false;
5007     fp = fopen("/proc/self/task", "r");
5008     if (fp != NULL) {
5009       proc_stat_path = "/proc/self/task/%d/stat";
5010       fclose(fp);
5011     }
5012   }
5013 
5014   sprintf(proc_name, proc_stat_path, tid);
5015   fp = fopen(proc_name, "r");
5016   if ( fp == NULL ) return -1;
5017   statlen = fread(stat, 1, 2047, fp);
5018   stat[statlen] = '\0';
5019   fclose(fp);
5020 
5021   // Skip pid and the command string. Note that we could be dealing with
5022   // weird command names, e.g. user could decide to rename java launcher
5023   // to "java 1.4.2 :)", then the stat file would look like
5024   //                1234 (java 1.4.2 :)) R ... ...
5025   // We don't really need to know the command string, just find the last
5026   // occurrence of ")" and then start parsing from there. See bug 4726580.
5027   s = strrchr(stat, ')');
5028   if (s == NULL ) return -1;
5029 
5030   // Skip blank chars
5031   do s++; while (isspace(*s));
5032 
5033   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5034                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5035                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5036                  &user_time, &sys_time);
5037   if ( count != 13 ) return -1;
5038   if (user_sys_cpu_time) {
5039     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5040   } else {
5041     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5042   }
5043 }
5044 
5045 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5046   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5047   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5048   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5049   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5050 }
5051 
5052 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5053   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5054   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5055   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5056   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5057 }
5058 
5059 bool os::is_thread_cpu_time_supported() {
5060   return true;
5061 }
5062 
5063 // System loadavg support.  Returns -1 if load average cannot be obtained.
5064 // Linux doesn't yet have a (official) notion of processor sets,
5065 // so just return the system wide load average.
5066 int os::loadavg(double loadavg[], int nelem) {
5067   return ::getloadavg(loadavg, nelem);
5068 }
5069 
5070 void os::pause() {
5071   char filename[MAX_PATH];
5072   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5073     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5074   } else {
5075     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5076   }
5077 
5078   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5079   if (fd != -1) {
5080     struct stat buf;
5081     ::close(fd);
5082     while (::stat(filename, &buf) == 0) {
5083       (void)::poll(NULL, 0, 100);
5084     }
5085   } else {
5086     jio_fprintf(stderr,
5087       "Could not open pause file '%s', continuing immediately.\n", filename);
5088   }
5089 }
5090 
5091 
5092 // Refer to the comments in os_solaris.cpp park-unpark.
5093 //
5094 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5095 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5096 // For specifics regarding the bug see GLIBC BUGID 261237 :
5097 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5098 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5099 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5100 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5101 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5102 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5103 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5104 // of libpthread avoids the problem, but isn't practical.
5105 //
5106 // Possible remedies:
5107 //
5108 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5109 //      This is palliative and probabilistic, however.  If the thread is preempted
5110 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5111 //      than the minimum period may have passed, and the abstime may be stale (in the
5112 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5113 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5114 //
5115 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5116 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5117 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5118 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5119 //      thread.
5120 //
5121 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5122 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5123 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5124 //      This also works well.  In fact it avoids kernel-level scalability impediments
5125 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5126 //      timers in a graceful fashion.
5127 //
5128 // 4.   When the abstime value is in the past it appears that control returns
5129 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5130 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5131 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5132 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5133 //      It may be possible to avoid reinitialization by checking the return
5134 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5135 //      condvar we must establish the invariant that cond_signal() is only called
5136 //      within critical sections protected by the adjunct mutex.  This prevents
5137 //      cond_signal() from "seeing" a condvar that's in the midst of being
5138 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5139 //      desirable signal-after-unlock optimization that avoids futile context switching.
5140 //
5141 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5142 //      structure when a condvar is used or initialized.  cond_destroy()  would
5143 //      release the helper structure.  Our reinitialize-after-timedwait fix
5144 //      put excessive stress on malloc/free and locks protecting the c-heap.
5145 //
5146 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5147 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5148 // and only enabling the work-around for vulnerable environments.
5149 
5150 // utility to compute the abstime argument to timedwait:
5151 // millis is the relative timeout time
5152 // abstime will be the absolute timeout time
5153 // TODO: replace compute_abstime() with unpackTime()
5154 
5155 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5156   if (millis < 0)  millis = 0;
5157   struct timeval now;
5158   int status = gettimeofday(&now, NULL);
5159   assert(status == 0, "gettimeofday");
5160   jlong seconds = millis / 1000;
5161   millis %= 1000;
5162   if (seconds > 50000000) { // see man cond_timedwait(3T)
5163     seconds = 50000000;
5164   }
5165   abstime->tv_sec = now.tv_sec  + seconds;
5166   long       usec = now.tv_usec + millis * 1000;
5167   if (usec >= 1000000) {
5168     abstime->tv_sec += 1;
5169     usec -= 1000000;
5170   }
5171   abstime->tv_nsec = usec * 1000;
5172   return abstime;
5173 }
5174 
5175 
5176 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5177 // Conceptually TryPark() should be equivalent to park(0).
5178 
5179 int os::PlatformEvent::TryPark() {
5180   for (;;) {
5181     const int v = _Event ;
5182     guarantee ((v == 0) || (v == 1), "invariant") ;
5183     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5184   }
5185 }
5186 
5187 void os::PlatformEvent::park() {       // AKA "down()"
5188   // Invariant: Only the thread associated with the Event/PlatformEvent
5189   // may call park().
5190   // TODO: assert that _Assoc != NULL or _Assoc == Self
5191   int v ;
5192   for (;;) {
5193       v = _Event ;
5194       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5195   }
5196   guarantee (v >= 0, "invariant") ;
5197   if (v == 0) {
5198      // Do this the hard way by blocking ...
5199      int status = pthread_mutex_lock(_mutex);
5200      assert_status(status == 0, status, "mutex_lock");
5201      guarantee (_nParked == 0, "invariant") ;
5202      ++ _nParked ;
5203      while (_Event < 0) {
5204         status = pthread_cond_wait(_cond, _mutex);
5205         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5206         // Treat this the same as if the wait was interrupted
5207         if (status == ETIME) { status = EINTR; }
5208         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5209      }
5210      -- _nParked ;
5211 
5212     _Event = 0 ;
5213      status = pthread_mutex_unlock(_mutex);
5214      assert_status(status == 0, status, "mutex_unlock");
5215     // Paranoia to ensure our locked and lock-free paths interact
5216     // correctly with each other.
5217     OrderAccess::fence();
5218   }
5219   guarantee (_Event >= 0, "invariant") ;
5220 }
5221 
5222 int os::PlatformEvent::park(jlong millis) {
5223   guarantee (_nParked == 0, "invariant") ;
5224 
5225   int v ;
5226   for (;;) {
5227       v = _Event ;
5228       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5229   }
5230   guarantee (v >= 0, "invariant") ;
5231   if (v != 0) return OS_OK ;
5232 
5233   // We do this the hard way, by blocking the thread.
5234   // Consider enforcing a minimum timeout value.
5235   struct timespec abst;
5236   compute_abstime(&abst, millis);
5237 
5238   int ret = OS_TIMEOUT;
5239   int status = pthread_mutex_lock(_mutex);
5240   assert_status(status == 0, status, "mutex_lock");
5241   guarantee (_nParked == 0, "invariant") ;
5242   ++_nParked ;
5243 
5244   // Object.wait(timo) will return because of
5245   // (a) notification
5246   // (b) timeout
5247   // (c) thread.interrupt
5248   //
5249   // Thread.interrupt and object.notify{All} both call Event::set.
5250   // That is, we treat thread.interrupt as a special case of notification.
5251   // The underlying Solaris implementation, cond_timedwait, admits
5252   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5253   // JVM from making those visible to Java code.  As such, we must
5254   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5255   //
5256   // TODO: properly differentiate simultaneous notify+interrupt.
5257   // In that case, we should propagate the notify to another waiter.
5258 
5259   while (_Event < 0) {
5260     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5261     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5262       pthread_cond_destroy (_cond);
5263       pthread_cond_init (_cond, NULL) ;
5264     }
5265     assert_status(status == 0 || status == EINTR ||
5266                   status == ETIME || status == ETIMEDOUT,
5267                   status, "cond_timedwait");
5268     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5269     if (status == ETIME || status == ETIMEDOUT) break ;
5270     // We consume and ignore EINTR and spurious wakeups.
5271   }
5272   --_nParked ;
5273   if (_Event >= 0) {
5274      ret = OS_OK;
5275   }
5276   _Event = 0 ;
5277   status = pthread_mutex_unlock(_mutex);
5278   assert_status(status == 0, status, "mutex_unlock");
5279   assert (_nParked == 0, "invariant") ;
5280   // Paranoia to ensure our locked and lock-free paths interact
5281   // correctly with each other.
5282   OrderAccess::fence();
5283   return ret;
5284 }
5285 
5286 void os::PlatformEvent::unpark() {
5287   // Transitions for _Event:
5288   //    0 :=> 1
5289   //    1 :=> 1
5290   //   -1 :=> either 0 or 1; must signal target thread
5291   //          That is, we can safely transition _Event from -1 to either
5292   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
5293   //          unpark() calls.
5294   // See also: "Semaphores in Plan 9" by Mullender & Cox
5295   //
5296   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5297   // that it will take two back-to-back park() calls for the owning
5298   // thread to block. This has the benefit of forcing a spurious return
5299   // from the first park() call after an unpark() call which will help
5300   // shake out uses of park() and unpark() without condition variables.
5301 
5302   if (Atomic::xchg(1, &_Event) >= 0) return;
5303 
5304   // Wait for the thread associated with the event to vacate
5305   int status = pthread_mutex_lock(_mutex);
5306   assert_status(status == 0, status, "mutex_lock");
5307   int AnyWaiters = _nParked;
5308   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5309   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5310     AnyWaiters = 0;
5311     pthread_cond_signal(_cond);
5312   }
5313   status = pthread_mutex_unlock(_mutex);
5314   assert_status(status == 0, status, "mutex_unlock");
5315   if (AnyWaiters != 0) {
5316     status = pthread_cond_signal(_cond);
5317     assert_status(status == 0, status, "cond_signal");
5318   }
5319 
5320   // Note that we signal() _after dropping the lock for "immortal" Events.
5321   // This is safe and avoids a common class of  futile wakeups.  In rare
5322   // circumstances this can cause a thread to return prematurely from
5323   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5324   // simply re-test the condition and re-park itself.
5325 }
5326 
5327 
5328 // JSR166
5329 // -------------------------------------------------------
5330 
5331 /*
5332  * The solaris and linux implementations of park/unpark are fairly
5333  * conservative for now, but can be improved. They currently use a
5334  * mutex/condvar pair, plus a a count.
5335  * Park decrements count if > 0, else does a condvar wait.  Unpark
5336  * sets count to 1 and signals condvar.  Only one thread ever waits
5337  * on the condvar. Contention seen when trying to park implies that someone
5338  * is unparking you, so don't wait. And spurious returns are fine, so there
5339  * is no need to track notifications.
5340  */
5341 
5342 #define MAX_SECS 100000000
5343 /*
5344  * This code is common to linux and solaris and will be moved to a
5345  * common place in dolphin.
5346  *
5347  * The passed in time value is either a relative time in nanoseconds
5348  * or an absolute time in milliseconds. Either way it has to be unpacked
5349  * into suitable seconds and nanoseconds components and stored in the
5350  * given timespec structure.
5351  * Given time is a 64-bit value and the time_t used in the timespec is only
5352  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5353  * overflow if times way in the future are given. Further on Solaris versions
5354  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5355  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5356  * As it will be 28 years before "now + 100000000" will overflow we can
5357  * ignore overflow and just impose a hard-limit on seconds using the value
5358  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5359  * years from "now".
5360  */
5361 
5362 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5363   assert (time > 0, "convertTime");
5364 
5365   struct timeval now;
5366   int status = gettimeofday(&now, NULL);
5367   assert(status == 0, "gettimeofday");
5368 
5369   time_t max_secs = now.tv_sec + MAX_SECS;
5370 
5371   if (isAbsolute) {
5372     jlong secs = time / 1000;
5373     if (secs > max_secs) {
5374       absTime->tv_sec = max_secs;
5375     }
5376     else {
5377       absTime->tv_sec = secs;
5378     }
5379     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5380   }
5381   else {
5382     jlong secs = time / NANOSECS_PER_SEC;
5383     if (secs >= MAX_SECS) {
5384       absTime->tv_sec = max_secs;
5385       absTime->tv_nsec = 0;
5386     }
5387     else {
5388       absTime->tv_sec = now.tv_sec + secs;
5389       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5390       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5391         absTime->tv_nsec -= NANOSECS_PER_SEC;
5392         ++absTime->tv_sec; // note: this must be <= max_secs
5393       }
5394     }
5395   }
5396   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5397   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5398   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5399   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5400 }
5401 
5402 void Parker::park(bool isAbsolute, jlong time) {
5403   // Ideally we'd do something useful while spinning, such
5404   // as calling unpackTime().
5405 
5406   // Optional fast-path check:
5407   // Return immediately if a permit is available.
5408   // We depend on Atomic::xchg() having full barrier semantics
5409   // since we are doing a lock-free update to _counter.
5410   if (Atomic::xchg(0, &_counter) > 0) return;
5411 
5412   Thread* thread = Thread::current();
5413   assert(thread->is_Java_thread(), "Must be JavaThread");
5414   JavaThread *jt = (JavaThread *)thread;
5415 
5416   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5417   // Check interrupt before trying to wait
5418   if (Thread::is_interrupted(thread, false)) {
5419     return;
5420   }
5421 
5422   // Next, demultiplex/decode time arguments
5423   timespec absTime;
5424   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5425     return;
5426   }
5427   if (time > 0) {
5428     unpackTime(&absTime, isAbsolute, time);
5429   }
5430 
5431 
5432   // Enter safepoint region
5433   // Beware of deadlocks such as 6317397.
5434   // The per-thread Parker:: mutex is a classic leaf-lock.
5435   // In particular a thread must never block on the Threads_lock while
5436   // holding the Parker:: mutex.  If safepoints are pending both the
5437   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5438   ThreadBlockInVM tbivm(jt);
5439 
5440   // Don't wait if cannot get lock since interference arises from
5441   // unblocking.  Also. check interrupt before trying wait
5442   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5443     return;
5444   }
5445 
5446   int status ;
5447   if (_counter > 0)  { // no wait needed
5448     _counter = 0;
5449     status = pthread_mutex_unlock(_mutex);
5450     assert (status == 0, "invariant") ;
5451     // Paranoia to ensure our locked and lock-free paths interact
5452     // correctly with each other and Java-level accesses.
5453     OrderAccess::fence();
5454     return;
5455   }
5456 
5457 #ifdef ASSERT
5458   // Don't catch signals while blocked; let the running threads have the signals.
5459   // (This allows a debugger to break into the running thread.)
5460   sigset_t oldsigs;
5461   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5462   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5463 #endif
5464 
5465   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5466   jt->set_suspend_equivalent();
5467   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5468 
5469   if (time == 0) {
5470     status = pthread_cond_wait (_cond, _mutex) ;
5471   } else {
5472     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5473     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5474       pthread_cond_destroy (_cond) ;
5475       pthread_cond_init    (_cond, NULL);
5476     }
5477   }
5478   assert_status(status == 0 || status == EINTR ||
5479                 status == ETIME || status == ETIMEDOUT,
5480                 status, "cond_timedwait");
5481 
5482 #ifdef ASSERT
5483   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5484 #endif
5485 
5486   _counter = 0 ;
5487   status = pthread_mutex_unlock(_mutex) ;
5488   assert_status(status == 0, status, "invariant") ;
5489   // Paranoia to ensure our locked and lock-free paths interact
5490   // correctly with each other and Java-level accesses.
5491   OrderAccess::fence();
5492 
5493   // If externally suspended while waiting, re-suspend
5494   if (jt->handle_special_suspend_equivalent_condition()) {
5495     jt->java_suspend_self();
5496   }
5497 }
5498 
5499 void Parker::unpark() {
5500   int s, status ;
5501   status = pthread_mutex_lock(_mutex);
5502   assert (status == 0, "invariant") ;
5503   s = _counter;
5504   _counter = 1;
5505   if (s < 1) {
5506      if (WorkAroundNPTLTimedWaitHang) {
5507         status = pthread_cond_signal (_cond) ;
5508         assert (status == 0, "invariant") ;
5509         status = pthread_mutex_unlock(_mutex);
5510         assert (status == 0, "invariant") ;
5511      } else {
5512         status = pthread_mutex_unlock(_mutex);
5513         assert (status == 0, "invariant") ;
5514         status = pthread_cond_signal (_cond) ;
5515         assert (status == 0, "invariant") ;
5516      }
5517   } else {
5518     pthread_mutex_unlock(_mutex);
5519     assert (status == 0, "invariant") ;
5520   }
5521 }
5522 
5523 
5524 extern char** environ;
5525 
5526 #ifndef __NR_fork
5527 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5528 #endif
5529 
5530 #ifndef __NR_execve
5531 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5532 #endif
5533 
5534 // Run the specified command in a separate process. Return its exit value,
5535 // or -1 on failure (e.g. can't fork a new process).
5536 // Unlike system(), this function can be called from signal handler. It
5537 // doesn't block SIGINT et al.
5538 int os::fork_and_exec(char* cmd) {
5539   const char * argv[4] = {"sh", "-c", cmd, NULL};
5540 
5541   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5542   // pthread_atfork handlers and reset pthread library. All we need is a
5543   // separate process to execve. Make a direct syscall to fork process.
5544   // On IA64 there's no fork syscall, we have to use fork() and hope for
5545   // the best...
5546   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5547               IA64_ONLY(fork();)
5548 
5549   if (pid < 0) {
5550     // fork failed
5551     return -1;
5552 
5553   } else if (pid == 0) {
5554     // child process
5555 
5556     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5557     // first to kill every thread on the thread list. Because this list is
5558     // not reset by fork() (see notes above), execve() will instead kill
5559     // every thread in the parent process. We know this is the only thread
5560     // in the new process, so make a system call directly.
5561     // IA64 should use normal execve() from glibc to match the glibc fork()
5562     // above.
5563     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5564     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5565 
5566     // execve failed
5567     _exit(-1);
5568 
5569   } else  {
5570     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5571     // care about the actual exit code, for now.
5572 
5573     int status;
5574 
5575     // Wait for the child process to exit.  This returns immediately if
5576     // the child has already exited. */
5577     while (waitpid(pid, &status, 0) < 0) {
5578         switch (errno) {
5579         case ECHILD: return 0;
5580         case EINTR: break;
5581         default: return -1;
5582         }
5583     }
5584 
5585     if (WIFEXITED(status)) {
5586        // The child exited normally; get its exit code.
5587        return WEXITSTATUS(status);
5588     } else if (WIFSIGNALED(status)) {
5589        // The child exited because of a signal
5590        // The best value to return is 0x80 + signal number,
5591        // because that is what all Unix shells do, and because
5592        // it allows callers to distinguish between process exit and
5593        // process death by signal.
5594        return 0x80 + WTERMSIG(status);
5595     } else {
5596        // Unknown exit code; pass it through
5597        return status;
5598     }
5599   }
5600 }
5601 
5602 // is_headless_jre()
5603 //
5604 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5605 // in order to report if we are running in a headless jre
5606 //
5607 // Since JDK8 xawt/libmawt.so was moved into the same directory
5608 // as libawt.so, and renamed libawt_xawt.so
5609 //
5610 bool os::is_headless_jre() {
5611     struct stat statbuf;
5612     char buf[MAXPATHLEN];
5613     char libmawtpath[MAXPATHLEN];
5614     const char *xawtstr  = "/xawt/libmawt.so";
5615     const char *new_xawtstr = "/libawt_xawt.so";
5616     char *p;
5617 
5618     // Get path to libjvm.so
5619     os::jvm_path(buf, sizeof(buf));
5620 
5621     // Get rid of libjvm.so
5622     p = strrchr(buf, '/');
5623     if (p == NULL) return false;
5624     else *p = '\0';
5625 
5626     // Get rid of client or server
5627     p = strrchr(buf, '/');
5628     if (p == NULL) return false;
5629     else *p = '\0';
5630 
5631     // check xawt/libmawt.so
5632     strcpy(libmawtpath, buf);
5633     strcat(libmawtpath, xawtstr);
5634     if (::stat(libmawtpath, &statbuf) == 0) return false;
5635 
5636     // check libawt_xawt.so
5637     strcpy(libmawtpath, buf);
5638     strcat(libmawtpath, new_xawtstr);
5639     if (::stat(libmawtpath, &statbuf) == 0) return false;
5640 
5641     return true;
5642 }
5643 
5644 // Get the default path to the core file
5645 // Returns the length of the string
5646 int os::get_core_path(char* buffer, size_t bufferSize) {
5647   const char* p = get_current_directory(buffer, bufferSize);
5648 
5649   if (p == NULL) {
5650     assert(p != NULL, "failed to get current directory");
5651     return 0;
5652   }
5653 
5654   return strlen(buffer);
5655 }
5656 
5657 #ifdef JAVASE_EMBEDDED
5658 //
5659 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5660 //
5661 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5662 
5663 // ctor
5664 //
5665 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5666   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5667   _fd = fd;
5668 
5669   if (os::create_thread(this, os::os_thread)) {
5670     _memnotify_thread = this;
5671     os::set_priority(this, NearMaxPriority);
5672     os::start_thread(this);
5673   }
5674 }
5675 
5676 // Where all the work gets done
5677 //
5678 void MemNotifyThread::run() {
5679   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5680 
5681   // Set up the select arguments
5682   fd_set rfds;
5683   if (_fd != -1) {
5684     FD_ZERO(&rfds);
5685     FD_SET(_fd, &rfds);
5686   }
5687 
5688   // Now wait for the mem_notify device to wake up
5689   while (1) {
5690     // Wait for the mem_notify device to signal us..
5691     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5692     if (rc == -1) {
5693       perror("select!\n");
5694       break;
5695     } else if (rc) {
5696       //ssize_t free_before = os::available_memory();
5697       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5698 
5699       // The kernel is telling us there is not much memory left...
5700       // try to do something about that
5701 
5702       // If we are not already in a GC, try one.
5703       if (!Universe::heap()->is_gc_active()) {
5704         Universe::heap()->collect(GCCause::_allocation_failure);
5705 
5706         //ssize_t free_after = os::available_memory();
5707         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5708         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5709       }
5710       // We might want to do something like the following if we find the GC's are not helping...
5711       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5712     }
5713   }
5714 }
5715 
5716 //
5717 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5718 //
5719 void MemNotifyThread::start() {
5720   int    fd;
5721   fd = open ("/dev/mem_notify", O_RDONLY, 0);
5722   if (fd < 0) {
5723       return;
5724   }
5725 
5726   if (memnotify_thread() == NULL) {
5727     new MemNotifyThread(fd);
5728   }
5729 }
5730 
5731 #endif // JAVASE_EMBEDDED