1 /*
   2  * Copyright (c) 1999, 2011, 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 "interpreter/interpreter.hpp"
  33 #include "jvm_linux.h"
  34 #include "memory/allocation.inline.hpp"
  35 #include "memory/filemap.hpp"
  36 #include "mutex_linux.inline.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "os_share_linux.hpp"
  39 #include "prims/jniFastGetField.hpp"
  40 #include "prims/jvm.h"
  41 #include "prims/jvm_misc.hpp"
  42 #include "runtime/arguments.hpp"
  43 #include "runtime/extendedPC.hpp"
  44 #include "runtime/globals.hpp"
  45 #include "runtime/interfaceSupport.hpp"
  46 #include "runtime/java.hpp"
  47 #include "runtime/javaCalls.hpp"
  48 #include "runtime/mutexLocker.hpp"
  49 #include "runtime/objectMonitor.hpp"
  50 #include "runtime/osThread.hpp"
  51 #include "runtime/perfMemory.hpp"
  52 #include "runtime/sharedRuntime.hpp"
  53 #include "runtime/statSampler.hpp"
  54 #include "runtime/stubRoutines.hpp"
  55 #include "runtime/threadCritical.hpp"
  56 #include "runtime/timer.hpp"
  57 #include "services/attachListener.hpp"
  58 #include "services/runtimeService.hpp"
  59 #include "thread_linux.inline.hpp"
  60 #include "utilities/decoder.hpp"
  61 #include "utilities/defaultStream.hpp"
  62 #include "utilities/events.hpp"
  63 #include "utilities/growableArray.hpp"
  64 #include "utilities/vmError.hpp"
  65 #ifdef TARGET_ARCH_x86
  66 # include "assembler_x86.inline.hpp"
  67 # include "nativeInst_x86.hpp"
  68 #endif
  69 #ifdef TARGET_ARCH_sparc
  70 # include "assembler_sparc.inline.hpp"
  71 # include "nativeInst_sparc.hpp"
  72 #endif
  73 #ifdef TARGET_ARCH_zero
  74 # include "assembler_zero.inline.hpp"
  75 # include "nativeInst_zero.hpp"
  76 #endif
  77 #ifdef TARGET_ARCH_arm
  78 # include "assembler_arm.inline.hpp"
  79 # include "nativeInst_arm.hpp"
  80 #endif
  81 #ifdef TARGET_ARCH_ppc
  82 # include "assembler_ppc.inline.hpp"
  83 # include "nativeInst_ppc.hpp"
  84 #endif
  85 #ifdef COMPILER1
  86 #include "c1/c1_Runtime1.hpp"
  87 #endif
  88 #ifdef COMPILER2
  89 #include "opto/runtime.hpp"
  90 #endif
  91 
  92 // put OS-includes here
  93 # include <sys/types.h>
  94 # include <sys/mman.h>
  95 # include <sys/stat.h>
  96 # include <sys/select.h>
  97 # include <pthread.h>
  98 # include <signal.h>
  99 # include <errno.h>
 100 # include <dlfcn.h>
 101 # include <stdio.h>
 102 # include <unistd.h>
 103 # include <sys/resource.h>
 104 # include <pthread.h>
 105 # include <sys/stat.h>
 106 # include <sys/time.h>
 107 # include <sys/times.h>
 108 # include <sys/utsname.h>
 109 # include <sys/socket.h>
 110 # include <sys/wait.h>
 111 # include <pwd.h>
 112 # include <poll.h>
 113 # include <semaphore.h>
 114 # include <fcntl.h>
 115 # include <string.h>
 116 # include <syscall.h>
 117 # include <sys/sysinfo.h>
 118 # include <gnu/libc-version.h>
 119 # include <sys/ipc.h>
 120 # include <sys/shm.h>
 121 # include <link.h>
 122 # include <stdint.h>
 123 # include <inttypes.h>
 124 # include <sys/ioctl.h>
 125 
 126 #define MAX_PATH    (2 * K)
 127 
 128 // for timer info max values which include all bits
 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 130 #define SEC_IN_NANOSECS  1000000000LL
 131 
 132 #define LARGEPAGES_BIT (1 << 6)
 133 ////////////////////////////////////////////////////////////////////////////////
 134 // global variables
 135 julong os::Linux::_physical_memory = 0;
 136 
 137 address   os::Linux::_initial_thread_stack_bottom = NULL;
 138 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 139 
 140 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 141 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 142 Mutex* os::Linux::_createThread_lock = NULL;
 143 pthread_t os::Linux::_main_thread;
 144 int os::Linux::_page_size = -1;
 145 bool os::Linux::_is_floating_stack = false;
 146 bool os::Linux::_is_NPTL = false;
 147 bool os::Linux::_supports_fast_thread_cpu_time = false;
 148 const char * os::Linux::_glibc_version = NULL;
 149 const char * os::Linux::_libpthread_version = NULL;
 150 
 151 static jlong initial_time_count=0;
 152 
 153 static int clock_tics_per_sec = 100;
 154 
 155 // For diagnostics to print a message once. see run_periodic_checks
 156 static sigset_t check_signal_done;
 157 static bool check_signals = true;;
 158 
 159 static pid_t _initial_pid = 0;
 160 
 161 /* Signal number used to suspend/resume a thread */
 162 
 163 /* do not use any signal number less than SIGSEGV, see 4355769 */
 164 static int SR_signum = SIGUSR2;
 165 sigset_t SR_sigset;
 166 
 167 /* Used to protect dlsym() calls */
 168 static pthread_mutex_t dl_mutex;
 169 
 170 #ifdef JAVASE_EMBEDDED
 171 class MemNotifyThread: public Thread {
 172   friend class VMStructs;
 173  public:
 174   virtual void run();
 175 
 176  private:
 177   static MemNotifyThread* _memnotify_thread;
 178   int _fd;
 179 
 180  public:
 181 
 182   // Constructor
 183   MemNotifyThread(int fd);
 184 
 185   // Tester
 186   bool is_memnotify_thread() const { return true; }
 187 
 188   // Printing
 189   char* name() const { return (char*)"Linux MemNotify Thread"; }
 190 
 191   // Returns the single instance of the MemNotifyThread
 192   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 193 
 194   // Create and start the single instance of MemNotifyThread
 195   static void start();
 196 };
 197 #endif // JAVASE_EMBEDDED
 198 
 199 // utility functions
 200 
 201 static int SR_initialize();
 202 static int SR_finalize();
 203 
 204 julong os::available_memory() {
 205   return Linux::available_memory();
 206 }
 207 
 208 julong os::Linux::available_memory() {
 209   // values in struct sysinfo are "unsigned long"
 210   struct sysinfo si;
 211   sysinfo(&si);
 212 
 213   return (julong)si.freeram * si.mem_unit;
 214 }
 215 
 216 julong os::physical_memory() {
 217   return Linux::physical_memory();
 218 }
 219 
 220 julong os::allocatable_physical_memory(julong size) {
 221 #ifdef _LP64
 222   return size;
 223 #else
 224   julong result = MIN2(size, (julong)3800*M);
 225    if (!is_allocatable(result)) {
 226      // See comments under solaris for alignment considerations
 227      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 228      result =  MIN2(size, reasonable_size);
 229    }
 230    return result;
 231 #endif // _LP64
 232 }
 233 
 234 ////////////////////////////////////////////////////////////////////////////////
 235 // environment support
 236 
 237 bool os::getenv(const char* name, char* buf, int len) {
 238   const char* val = ::getenv(name);
 239   if (val != NULL && strlen(val) < (size_t)len) {
 240     strcpy(buf, val);
 241     return true;
 242   }
 243   if (len > 0) buf[0] = 0;  // return a null string
 244   return false;
 245 }
 246 
 247 
 248 // Return true if user is running as root.
 249 
 250 bool os::have_special_privileges() {
 251   static bool init = false;
 252   static bool privileges = false;
 253   if (!init) {
 254     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 255     init = true;
 256   }
 257   return privileges;
 258 }
 259 
 260 
 261 #ifndef SYS_gettid
 262 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 263 #ifdef __ia64__
 264 #define SYS_gettid 1105
 265 #elif __i386__
 266 #define SYS_gettid 224
 267 #elif __amd64__
 268 #define SYS_gettid 186
 269 #elif __sparc__
 270 #define SYS_gettid 143
 271 #else
 272 #error define gettid for the arch
 273 #endif
 274 #endif
 275 
 276 // Cpu architecture string
 277 #if   defined(ZERO)
 278 static char cpu_arch[] = ZERO_LIBARCH;
 279 #elif defined(IA64)
 280 static char cpu_arch[] = "ia64";
 281 #elif defined(IA32)
 282 static char cpu_arch[] = "i386";
 283 #elif defined(AMD64)
 284 static char cpu_arch[] = "amd64";
 285 #elif defined(ARM)
 286 static char cpu_arch[] = "arm";
 287 #elif defined(PPC)
 288 static char cpu_arch[] = "ppc";
 289 #elif defined(SPARC)
 290 #  ifdef _LP64
 291 static char cpu_arch[] = "sparcv9";
 292 #  else
 293 static char cpu_arch[] = "sparc";
 294 #  endif
 295 #else
 296 #error Add appropriate cpu_arch setting
 297 #endif
 298 
 299 
 300 // pid_t gettid()
 301 //
 302 // Returns the kernel thread id of the currently running thread. Kernel
 303 // thread id is used to access /proc.
 304 //
 305 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 306 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 307 //
 308 pid_t os::Linux::gettid() {
 309   int rslt = syscall(SYS_gettid);
 310   if (rslt == -1) {
 311      // old kernel, no NPTL support
 312      return getpid();
 313   } else {
 314      return (pid_t)rslt;
 315   }
 316 }
 317 
 318 // Most versions of linux have a bug where the number of processors are
 319 // determined by looking at the /proc file system.  In a chroot environment,
 320 // the system call returns 1.  This causes the VM to act as if it is
 321 // a single processor and elide locking (see is_MP() call).
 322 static bool unsafe_chroot_detected = false;
 323 static const char *unstable_chroot_error = "/proc file system not found.\n"
 324                      "Java may be unstable running multithreaded in a chroot "
 325                      "environment on Linux when /proc filesystem is not mounted.";
 326 
 327 void os::Linux::initialize_system_info() {
 328   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 329   if (processor_count() == 1) {
 330     pid_t pid = os::Linux::gettid();
 331     char fname[32];
 332     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 333     FILE *fp = fopen(fname, "r");
 334     if (fp == NULL) {
 335       unsafe_chroot_detected = true;
 336     } else {
 337       fclose(fp);
 338     }
 339   }
 340   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 341   assert(processor_count() > 0, "linux error");
 342 }
 343 
 344 void os::init_system_properties_values() {
 345 //  char arch[12];
 346 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 347 
 348   // The next steps are taken in the product version:
 349   //
 350   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 351   // This library should be located at:
 352   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 353   //
 354   // If "/jre/lib/" appears at the right place in the path, then we
 355   // assume libjvm[_g].so is installed in a JDK and we use this path.
 356   //
 357   // Otherwise exit with message: "Could not create the Java virtual machine."
 358   //
 359   // The following extra steps are taken in the debugging version:
 360   //
 361   // If "/jre/lib/" does NOT appear at the right place in the path
 362   // instead of exit check for $JAVA_HOME environment variable.
 363   //
 364   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 365   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 366   // it looks like libjvm[_g].so is installed there
 367   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 368   //
 369   // Otherwise exit.
 370   //
 371   // Important note: if the location of libjvm.so changes this
 372   // code needs to be changed accordingly.
 373 
 374   // The next few definitions allow the code to be verbatim:
 375 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 376 #define getenv(n) ::getenv(n)
 377 
 378 /*
 379  * See ld(1):
 380  *      The linker uses the following search paths to locate required
 381  *      shared libraries:
 382  *        1: ...
 383  *        ...
 384  *        7: The default directories, normally /lib and /usr/lib.
 385  */
 386 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 387 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 388 #else
 389 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 390 #endif
 391 
 392 #define EXTENSIONS_DIR  "/lib/ext"
 393 #define ENDORSED_DIR    "/lib/endorsed"
 394 #define REG_DIR         "/usr/java/packages"
 395 
 396   {
 397     /* sysclasspath, java_home, dll_dir */
 398     {
 399         char *home_path;
 400         char *dll_path;
 401         char *pslash;
 402         char buf[MAXPATHLEN];
 403         os::jvm_path(buf, sizeof(buf));
 404 
 405         // Found the full path to libjvm.so.
 406         // Now cut the path to <java_home>/jre if we can.
 407         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 408         pslash = strrchr(buf, '/');
 409         if (pslash != NULL)
 410             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 411         dll_path = malloc(strlen(buf) + 1);
 412         if (dll_path == NULL)
 413             return;
 414         strcpy(dll_path, buf);
 415         Arguments::set_dll_dir(dll_path);
 416 
 417         if (pslash != NULL) {
 418             pslash = strrchr(buf, '/');
 419             if (pslash != NULL) {
 420                 *pslash = '\0';       /* get rid of /<arch> */
 421                 pslash = strrchr(buf, '/');
 422                 if (pslash != NULL)
 423                     *pslash = '\0';   /* get rid of /lib */
 424             }
 425         }
 426 
 427         home_path = malloc(strlen(buf) + 1);
 428         if (home_path == NULL)
 429             return;
 430         strcpy(home_path, buf);
 431         Arguments::set_java_home(home_path);
 432 
 433         if (!set_boot_path('/', ':'))
 434             return;
 435     }
 436 
 437     /*
 438      * Where to look for native libraries
 439      *
 440      * Note: Due to a legacy implementation, most of the library path
 441      * is set in the launcher.  This was to accomodate linking restrictions
 442      * on legacy Linux implementations (which are no longer supported).
 443      * Eventually, all the library path setting will be done here.
 444      *
 445      * However, to prevent the proliferation of improperly built native
 446      * libraries, the new path component /usr/java/packages is added here.
 447      * Eventually, all the library path setting will be done here.
 448      */
 449     {
 450         char *ld_library_path;
 451 
 452         /*
 453          * Construct the invariant part of ld_library_path. Note that the
 454          * space for the colon and the trailing null are provided by the
 455          * nulls included by the sizeof operator (so actually we allocate
 456          * a byte more than necessary).
 457          */
 458         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 459             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 460         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 461 
 462         /*
 463          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 464          * should always exist (until the legacy problem cited above is
 465          * addressed).
 466          */
 467         char *v = getenv("LD_LIBRARY_PATH");
 468         if (v != NULL) {
 469             char *t = ld_library_path;
 470             /* That's +1 for the colon and +1 for the trailing '\0' */
 471             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 472             sprintf(ld_library_path, "%s:%s", v, t);
 473         }
 474         Arguments::set_library_path(ld_library_path);
 475     }
 476 
 477     /*
 478      * Extensions directories.
 479      *
 480      * Note that the space for the colon and the trailing null are provided
 481      * by the nulls included by the sizeof operator (so actually one byte more
 482      * than necessary is allocated).
 483      */
 484     {
 485         char *buf = malloc(strlen(Arguments::get_java_home()) +
 486             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 487         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 488             Arguments::get_java_home());
 489         Arguments::set_ext_dirs(buf);
 490     }
 491 
 492     /* Endorsed standards default directory. */
 493     {
 494         char * buf;
 495         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 496         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 497         Arguments::set_endorsed_dirs(buf);
 498     }
 499   }
 500 
 501 #undef malloc
 502 #undef getenv
 503 #undef EXTENSIONS_DIR
 504 #undef ENDORSED_DIR
 505 
 506   // Done
 507   return;
 508 }
 509 
 510 ////////////////////////////////////////////////////////////////////////////////
 511 // breakpoint support
 512 
 513 void os::breakpoint() {
 514   BREAKPOINT;
 515 }
 516 
 517 extern "C" void breakpoint() {
 518   // use debugger to set breakpoint here
 519 }
 520 
 521 ////////////////////////////////////////////////////////////////////////////////
 522 // signal support
 523 
 524 debug_only(static bool signal_sets_initialized = false);
 525 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 526 
 527 bool os::Linux::is_sig_ignored(int sig) {
 528       struct sigaction oact;
 529       sigaction(sig, (struct sigaction*)NULL, &oact);
 530       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 531                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 532       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 533            return true;
 534       else
 535            return false;
 536 }
 537 
 538 void os::Linux::signal_sets_init() {
 539   // Should also have an assertion stating we are still single-threaded.
 540   assert(!signal_sets_initialized, "Already initialized");
 541   // Fill in signals that are necessarily unblocked for all threads in
 542   // the VM. Currently, we unblock the following signals:
 543   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 544   //                         by -Xrs (=ReduceSignalUsage));
 545   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 546   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 547   // the dispositions or masks wrt these signals.
 548   // Programs embedding the VM that want to use the above signals for their
 549   // own purposes must, at this time, use the "-Xrs" option to prevent
 550   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 551   // (See bug 4345157, and other related bugs).
 552   // In reality, though, unblocking these signals is really a nop, since
 553   // these signals are not blocked by default.
 554   sigemptyset(&unblocked_sigs);
 555   sigemptyset(&allowdebug_blocked_sigs);
 556   sigaddset(&unblocked_sigs, SIGILL);
 557   sigaddset(&unblocked_sigs, SIGSEGV);
 558   sigaddset(&unblocked_sigs, SIGBUS);
 559   sigaddset(&unblocked_sigs, SIGFPE);
 560   sigaddset(&unblocked_sigs, SR_signum);
 561 
 562   if (!ReduceSignalUsage) {
 563    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 564       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 565       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 566    }
 567    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 568       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 569       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 570    }
 571    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 572       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 573       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 574    }
 575   }
 576   // Fill in signals that are blocked by all but the VM thread.
 577   sigemptyset(&vm_sigs);
 578   if (!ReduceSignalUsage)
 579     sigaddset(&vm_sigs, BREAK_SIGNAL);
 580   debug_only(signal_sets_initialized = true);
 581 
 582 }
 583 
 584 // These are signals that are unblocked while a thread is running Java.
 585 // (For some reason, they get blocked by default.)
 586 sigset_t* os::Linux::unblocked_signals() {
 587   assert(signal_sets_initialized, "Not initialized");
 588   return &unblocked_sigs;
 589 }
 590 
 591 // These are the signals that are blocked while a (non-VM) thread is
 592 // running Java. Only the VM thread handles these signals.
 593 sigset_t* os::Linux::vm_signals() {
 594   assert(signal_sets_initialized, "Not initialized");
 595   return &vm_sigs;
 596 }
 597 
 598 // These are signals that are blocked during cond_wait to allow debugger in
 599 sigset_t* os::Linux::allowdebug_blocked_signals() {
 600   assert(signal_sets_initialized, "Not initialized");
 601   return &allowdebug_blocked_sigs;
 602 }
 603 
 604 void os::Linux::hotspot_sigmask(Thread* thread) {
 605 
 606   //Save caller's signal mask before setting VM signal mask
 607   sigset_t caller_sigmask;
 608   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 609 
 610   OSThread* osthread = thread->osthread();
 611   osthread->set_caller_sigmask(caller_sigmask);
 612 
 613   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 614 
 615   if (!ReduceSignalUsage) {
 616     if (thread->is_VM_thread()) {
 617       // Only the VM thread handles BREAK_SIGNAL ...
 618       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 619     } else {
 620       // ... all other threads block BREAK_SIGNAL
 621       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 622     }
 623   }
 624 }
 625 
 626 //////////////////////////////////////////////////////////////////////////////
 627 // detecting pthread library
 628 
 629 void os::Linux::libpthread_init() {
 630   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 631   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 632   // generic name for earlier versions.
 633   // Define macros here so we can build HotSpot on old systems.
 634 # ifndef _CS_GNU_LIBC_VERSION
 635 # define _CS_GNU_LIBC_VERSION 2
 636 # endif
 637 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 638 # define _CS_GNU_LIBPTHREAD_VERSION 3
 639 # endif
 640 
 641   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 642   if (n > 0) {
 643      char *str = (char *)malloc(n);
 644      confstr(_CS_GNU_LIBC_VERSION, str, n);
 645      os::Linux::set_glibc_version(str);
 646   } else {
 647      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 648      static char _gnu_libc_version[32];
 649      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 650               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 651      os::Linux::set_glibc_version(_gnu_libc_version);
 652   }
 653 
 654   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 655   if (n > 0) {
 656      char *str = (char *)malloc(n);
 657      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 658      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 659      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 660      // is the case. LinuxThreads has a hard limit on max number of threads.
 661      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 662      // On the other hand, NPTL does not have such a limit, sysconf()
 663      // will return -1 and errno is not changed. Check if it is really NPTL.
 664      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 665          strstr(str, "NPTL") &&
 666          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 667        free(str);
 668        os::Linux::set_libpthread_version("linuxthreads");
 669      } else {
 670        os::Linux::set_libpthread_version(str);
 671      }
 672   } else {
 673     // glibc before 2.3.2 only has LinuxThreads.
 674     os::Linux::set_libpthread_version("linuxthreads");
 675   }
 676 
 677   if (strstr(libpthread_version(), "NPTL")) {
 678      os::Linux::set_is_NPTL();
 679   } else {
 680      os::Linux::set_is_LinuxThreads();
 681   }
 682 
 683   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 684   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 685   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 686      os::Linux::set_is_floating_stack();
 687   }
 688 }
 689 
 690 /////////////////////////////////////////////////////////////////////////////
 691 // thread stack
 692 
 693 // Force Linux kernel to expand current thread stack. If "bottom" is close
 694 // to the stack guard, caller should block all signals.
 695 //
 696 // MAP_GROWSDOWN:
 697 //   A special mmap() flag that is used to implement thread stacks. It tells
 698 //   kernel that the memory region should extend downwards when needed. This
 699 //   allows early versions of LinuxThreads to only mmap the first few pages
 700 //   when creating a new thread. Linux kernel will automatically expand thread
 701 //   stack as needed (on page faults).
 702 //
 703 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 704 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 705 //   region, it's hard to tell if the fault is due to a legitimate stack
 706 //   access or because of reading/writing non-exist memory (e.g. buffer
 707 //   overrun). As a rule, if the fault happens below current stack pointer,
 708 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 709 //   application (see Linux kernel fault.c).
 710 //
 711 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 712 //   stack overflow detection.
 713 //
 714 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 715 //   not use this flag. However, the stack of initial thread is not created
 716 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 717 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 718 //   and then attach the thread to JVM.
 719 //
 720 // To get around the problem and allow stack banging on Linux, we need to
 721 // manually expand thread stack after receiving the SIGSEGV.
 722 //
 723 // There are two ways to expand thread stack to address "bottom", we used
 724 // both of them in JVM before 1.5:
 725 //   1. adjust stack pointer first so that it is below "bottom", and then
 726 //      touch "bottom"
 727 //   2. mmap() the page in question
 728 //
 729 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 730 // if current sp is already near the lower end of page 101, and we need to
 731 // call mmap() to map page 100, it is possible that part of the mmap() frame
 732 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 733 // That will destroy the mmap() frame and cause VM to crash.
 734 //
 735 // The following code works by adjusting sp first, then accessing the "bottom"
 736 // page to force a page fault. Linux kernel will then automatically expand the
 737 // stack mapping.
 738 //
 739 // _expand_stack_to() assumes its frame size is less than page size, which
 740 // should always be true if the function is not inlined.
 741 
 742 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 743 #define NOINLINE
 744 #else
 745 #define NOINLINE __attribute__ ((noinline))
 746 #endif
 747 
 748 static void _expand_stack_to(address bottom) NOINLINE;
 749 
 750 static void _expand_stack_to(address bottom) {
 751   address sp;
 752   size_t size;
 753   volatile char *p;
 754 
 755   // Adjust bottom to point to the largest address within the same page, it
 756   // gives us a one-page buffer if alloca() allocates slightly more memory.
 757   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 758   bottom += os::Linux::page_size() - 1;
 759 
 760   // sp might be slightly above current stack pointer; if that's the case, we
 761   // will alloca() a little more space than necessary, which is OK. Don't use
 762   // os::current_stack_pointer(), as its result can be slightly below current
 763   // stack pointer, causing us to not alloca enough to reach "bottom".
 764   sp = (address)&sp;
 765 
 766   if (sp > bottom) {
 767     size = sp - bottom;
 768     p = (volatile char *)alloca(size);
 769     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 770     p[0] = '\0';
 771   }
 772 }
 773 
 774 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 775   assert(t!=NULL, "just checking");
 776   assert(t->osthread()->expanding_stack(), "expand should be set");
 777   assert(t->stack_base() != NULL, "stack_base was not initialized");
 778 
 779   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 780     sigset_t mask_all, old_sigset;
 781     sigfillset(&mask_all);
 782     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 783     _expand_stack_to(addr);
 784     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 785     return true;
 786   }
 787   return false;
 788 }
 789 
 790 //////////////////////////////////////////////////////////////////////////////
 791 // create new thread
 792 
 793 static address highest_vm_reserved_address();
 794 
 795 // check if it's safe to start a new thread
 796 static bool _thread_safety_check(Thread* thread) {
 797   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 798     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 799     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 800     //   allocated (MAP_FIXED) from high address space. Every thread stack
 801     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 802     //   it to other values if they rebuild LinuxThreads).
 803     //
 804     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 805     // the memory region has already been mmap'ed. That means if we have too
 806     // many threads and/or very large heap, eventually thread stack will
 807     // collide with heap.
 808     //
 809     // Here we try to prevent heap/stack collision by comparing current
 810     // stack bottom with the highest address that has been mmap'ed by JVM
 811     // plus a safety margin for memory maps created by native code.
 812     //
 813     // This feature can be disabled by setting ThreadSafetyMargin to 0
 814     //
 815     if (ThreadSafetyMargin > 0) {
 816       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 817 
 818       // not safe if our stack extends below the safety margin
 819       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 820     } else {
 821       return true;
 822     }
 823   } else {
 824     // Floating stack LinuxThreads or NPTL:
 825     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 826     //   there's not enough space left, pthread_create() will fail. If we come
 827     //   here, that means enough space has been reserved for stack.
 828     return true;
 829   }
 830 }
 831 
 832 // Thread start routine for all newly created threads
 833 static void *java_start(Thread *thread) {
 834   // Try to randomize the cache line index of hot stack frames.
 835   // This helps when threads of the same stack traces evict each other's
 836   // cache lines. The threads can be either from the same JVM instance, or
 837   // from different JVM instances. The benefit is especially true for
 838   // processors with hyperthreading technology.
 839   static int counter = 0;
 840   int pid = os::current_process_id();
 841   alloca(((pid ^ counter++) & 7) * 128);
 842 
 843   ThreadLocalStorage::set_thread(thread);
 844 
 845   OSThread* osthread = thread->osthread();
 846   Monitor* sync = osthread->startThread_lock();
 847 
 848   // non floating stack LinuxThreads needs extra check, see above
 849   if (!_thread_safety_check(thread)) {
 850     // notify parent thread
 851     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 852     osthread->set_state(ZOMBIE);
 853     sync->notify_all();
 854     return NULL;
 855   }
 856 
 857   // thread_id is kernel thread id (similar to Solaris LWP id)
 858   osthread->set_thread_id(os::Linux::gettid());
 859 
 860   if (UseNUMA) {
 861     int lgrp_id = os::numa_get_group_id();
 862     if (lgrp_id != -1) {
 863       thread->set_lgrp_id(lgrp_id);
 864     }
 865   }
 866   // initialize signal mask for this thread
 867   os::Linux::hotspot_sigmask(thread);
 868 
 869   // initialize floating point control register
 870   os::Linux::init_thread_fpu_state();
 871 
 872   // handshaking with parent thread
 873   {
 874     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 875 
 876     // notify parent thread
 877     osthread->set_state(INITIALIZED);
 878     sync->notify_all();
 879 
 880     // wait until os::start_thread()
 881     while (osthread->get_state() == INITIALIZED) {
 882       sync->wait(Mutex::_no_safepoint_check_flag);
 883     }
 884   }
 885 
 886   // call one more level start routine
 887   thread->run();
 888 
 889   return 0;
 890 }
 891 
 892 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 893   assert(thread->osthread() == NULL, "caller responsible");
 894 
 895   // Allocate the OSThread object
 896   OSThread* osthread = new OSThread(NULL, NULL);
 897   if (osthread == NULL) {
 898     return false;
 899   }
 900 
 901   // set the correct thread state
 902   osthread->set_thread_type(thr_type);
 903 
 904   // Initial state is ALLOCATED but not INITIALIZED
 905   osthread->set_state(ALLOCATED);
 906 
 907   thread->set_osthread(osthread);
 908 
 909   // init thread attributes
 910   pthread_attr_t attr;
 911   pthread_attr_init(&attr);
 912   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 913 
 914   // stack size
 915   if (os::Linux::supports_variable_stack_size()) {
 916     // calculate stack size if it's not specified by caller
 917     if (stack_size == 0) {
 918       stack_size = os::Linux::default_stack_size(thr_type);
 919 
 920       switch (thr_type) {
 921       case os::java_thread:
 922         // Java threads use ThreadStackSize which default value can be
 923         // changed with the flag -Xss
 924         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 925         stack_size = JavaThread::stack_size_at_create();
 926         break;
 927       case os::compiler_thread:
 928         if (CompilerThreadStackSize > 0) {
 929           stack_size = (size_t)(CompilerThreadStackSize * K);
 930           break;
 931         } // else fall through:
 932           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 933       case os::vm_thread:
 934       case os::pgc_thread:
 935       case os::cgc_thread:
 936       case os::watcher_thread:
 937         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 938         break;
 939       }
 940     }
 941 
 942     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 943     pthread_attr_setstacksize(&attr, stack_size);
 944   } else {
 945     // let pthread_create() pick the default value.
 946   }
 947 
 948   // glibc guard page
 949   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 950 
 951   ThreadState state;
 952 
 953   {
 954     // Serialize thread creation if we are running with fixed stack LinuxThreads
 955     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 956     if (lock) {
 957       os::Linux::createThread_lock()->lock_without_safepoint_check();
 958     }
 959 
 960     pthread_t tid;
 961     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 962 
 963     pthread_attr_destroy(&attr);
 964 
 965     if (ret != 0) {
 966       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 967         perror("pthread_create()");
 968       }
 969       // Need to clean up stuff we've allocated so far
 970       thread->set_osthread(NULL);
 971       delete osthread;
 972       if (lock) os::Linux::createThread_lock()->unlock();
 973       return false;
 974     }
 975 
 976     // Store pthread info into the OSThread
 977     osthread->set_pthread_id(tid);
 978 
 979     // Wait until child thread is either initialized or aborted
 980     {
 981       Monitor* sync_with_child = osthread->startThread_lock();
 982       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 983       while ((state = osthread->get_state()) == ALLOCATED) {
 984         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 985       }
 986     }
 987 
 988     if (lock) {
 989       os::Linux::createThread_lock()->unlock();
 990     }
 991   }
 992 
 993   // Aborted due to thread limit being reached
 994   if (state == ZOMBIE) {
 995       thread->set_osthread(NULL);
 996       delete osthread;
 997       return false;
 998   }
 999 
1000   // The thread is returned suspended (in state INITIALIZED),
1001   // and is started higher up in the call chain
1002   assert(state == INITIALIZED, "race condition");
1003   return true;
1004 }
1005 
1006 /////////////////////////////////////////////////////////////////////////////
1007 // attach existing thread
1008 
1009 // bootstrap the main thread
1010 bool os::create_main_thread(JavaThread* thread) {
1011   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
1012   return create_attached_thread(thread);
1013 }
1014 
1015 bool os::create_attached_thread(JavaThread* thread) {
1016 #ifdef ASSERT
1017     thread->verify_not_published();
1018 #endif
1019 
1020   // Allocate the OSThread object
1021   OSThread* osthread = new OSThread(NULL, NULL);
1022 
1023   if (osthread == NULL) {
1024     return false;
1025   }
1026 
1027   // Store pthread info into the OSThread
1028   osthread->set_thread_id(os::Linux::gettid());
1029   osthread->set_pthread_id(::pthread_self());
1030 
1031   // initialize floating point control register
1032   os::Linux::init_thread_fpu_state();
1033 
1034   // Initial thread state is RUNNABLE
1035   osthread->set_state(RUNNABLE);
1036 
1037   thread->set_osthread(osthread);
1038 
1039   if (UseNUMA) {
1040     int lgrp_id = os::numa_get_group_id();
1041     if (lgrp_id != -1) {
1042       thread->set_lgrp_id(lgrp_id);
1043     }
1044   }
1045 
1046   if (os::Linux::is_initial_thread()) {
1047     // If current thread is initial thread, its stack is mapped on demand,
1048     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1049     // the entire stack region to avoid SEGV in stack banging.
1050     // It is also useful to get around the heap-stack-gap problem on SuSE
1051     // kernel (see 4821821 for details). We first expand stack to the top
1052     // of yellow zone, then enable stack yellow zone (order is significant,
1053     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1054     // is no gap between the last two virtual memory regions.
1055 
1056     JavaThread *jt = (JavaThread *)thread;
1057     address addr = jt->stack_yellow_zone_base();
1058     assert(addr != NULL, "initialization problem?");
1059     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1060 
1061     osthread->set_expanding_stack();
1062     os::Linux::manually_expand_stack(jt, addr);
1063     osthread->clear_expanding_stack();
1064   }
1065 
1066   // initialize signal mask for this thread
1067   // and save the caller's signal mask
1068   os::Linux::hotspot_sigmask(thread);
1069 
1070   return true;
1071 }
1072 
1073 void os::pd_start_thread(Thread* thread) {
1074   OSThread * osthread = thread->osthread();
1075   assert(osthread->get_state() != INITIALIZED, "just checking");
1076   Monitor* sync_with_child = osthread->startThread_lock();
1077   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1078   sync_with_child->notify();
1079 }
1080 
1081 // Free Linux resources related to the OSThread
1082 void os::free_thread(OSThread* osthread) {
1083   assert(osthread != NULL, "osthread not set");
1084 
1085   if (Thread::current()->osthread() == osthread) {
1086     // Restore caller's signal mask
1087     sigset_t sigmask = osthread->caller_sigmask();
1088     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1089    }
1090 
1091   delete osthread;
1092 }
1093 
1094 //////////////////////////////////////////////////////////////////////////////
1095 // thread local storage
1096 
1097 int os::allocate_thread_local_storage() {
1098   pthread_key_t key;
1099   int rslt = pthread_key_create(&key, NULL);
1100   assert(rslt == 0, "cannot allocate thread local storage");
1101   return (int)key;
1102 }
1103 
1104 // Note: This is currently not used by VM, as we don't destroy TLS key
1105 // on VM exit.
1106 void os::free_thread_local_storage(int index) {
1107   int rslt = pthread_key_delete((pthread_key_t)index);
1108   assert(rslt == 0, "invalid index");
1109 }
1110 
1111 void os::thread_local_storage_at_put(int index, void* value) {
1112   int rslt = pthread_setspecific((pthread_key_t)index, value);
1113   assert(rslt == 0, "pthread_setspecific failed");
1114 }
1115 
1116 extern "C" Thread* get_thread() {
1117   return ThreadLocalStorage::thread();
1118 }
1119 
1120 //////////////////////////////////////////////////////////////////////////////
1121 // initial thread
1122 
1123 // Check if current thread is the initial thread, similar to Solaris thr_main.
1124 bool os::Linux::is_initial_thread(void) {
1125   char dummy;
1126   // If called before init complete, thread stack bottom will be null.
1127   // Can be called if fatal error occurs before initialization.
1128   if (initial_thread_stack_bottom() == NULL) return false;
1129   assert(initial_thread_stack_bottom() != NULL &&
1130          initial_thread_stack_size()   != 0,
1131          "os::init did not locate initial thread's stack region");
1132   if ((address)&dummy >= initial_thread_stack_bottom() &&
1133       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1134        return true;
1135   else return false;
1136 }
1137 
1138 // Find the virtual memory area that contains addr
1139 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1140   FILE *fp = fopen("/proc/self/maps", "r");
1141   if (fp) {
1142     address low, high;
1143     while (!feof(fp)) {
1144       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1145         if (low <= addr && addr < high) {
1146            if (vma_low)  *vma_low  = low;
1147            if (vma_high) *vma_high = high;
1148            fclose (fp);
1149            return true;
1150         }
1151       }
1152       for (;;) {
1153         int ch = fgetc(fp);
1154         if (ch == EOF || ch == (int)'\n') break;
1155       }
1156     }
1157     fclose(fp);
1158   }
1159   return false;
1160 }
1161 
1162 // Locate initial thread stack. This special handling of initial thread stack
1163 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1164 // bogus value for initial thread.
1165 void os::Linux::capture_initial_stack(size_t max_size) {
1166   // stack size is the easy part, get it from RLIMIT_STACK
1167   size_t stack_size;
1168   struct rlimit rlim;
1169   getrlimit(RLIMIT_STACK, &rlim);
1170   stack_size = rlim.rlim_cur;
1171 
1172   // 6308388: a bug in ld.so will relocate its own .data section to the
1173   //   lower end of primordial stack; reduce ulimit -s value a little bit
1174   //   so we won't install guard page on ld.so's data section.
1175   stack_size -= 2 * page_size();
1176 
1177   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1178   //   7.1, in both cases we will get 2G in return value.
1179   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1180   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1181   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1182   //   in case other parts in glibc still assumes 2M max stack size.
1183   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1184 #ifndef IA64
1185   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1186 #else
1187   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1188   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1189 #endif
1190 
1191   // Try to figure out where the stack base (top) is. This is harder.
1192   //
1193   // When an application is started, glibc saves the initial stack pointer in
1194   // a global variable "__libc_stack_end", which is then used by system
1195   // libraries. __libc_stack_end should be pretty close to stack top. The
1196   // variable is available since the very early days. However, because it is
1197   // a private interface, it could disappear in the future.
1198   //
1199   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1200   // to __libc_stack_end, it is very close to stack top, but isn't the real
1201   // stack top. Note that /proc may not exist if VM is running as a chroot
1202   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1203   // /proc/<pid>/stat could change in the future (though unlikely).
1204   //
1205   // We try __libc_stack_end first. If that doesn't work, look for
1206   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1207   // as a hint, which should work well in most cases.
1208 
1209   uintptr_t stack_start;
1210 
1211   // try __libc_stack_end first
1212   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1213   if (p && *p) {
1214     stack_start = *p;
1215   } else {
1216     // see if we can get the start_stack field from /proc/self/stat
1217     FILE *fp;
1218     int pid;
1219     char state;
1220     int ppid;
1221     int pgrp;
1222     int session;
1223     int nr;
1224     int tpgrp;
1225     unsigned long flags;
1226     unsigned long minflt;
1227     unsigned long cminflt;
1228     unsigned long majflt;
1229     unsigned long cmajflt;
1230     unsigned long utime;
1231     unsigned long stime;
1232     long cutime;
1233     long cstime;
1234     long prio;
1235     long nice;
1236     long junk;
1237     long it_real;
1238     uintptr_t start;
1239     uintptr_t vsize;
1240     intptr_t rss;
1241     uintptr_t rsslim;
1242     uintptr_t scodes;
1243     uintptr_t ecode;
1244     int i;
1245 
1246     // Figure what the primordial thread stack base is. Code is inspired
1247     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1248     // followed by command name surrounded by parentheses, state, etc.
1249     char stat[2048];
1250     int statlen;
1251 
1252     fp = fopen("/proc/self/stat", "r");
1253     if (fp) {
1254       statlen = fread(stat, 1, 2047, fp);
1255       stat[statlen] = '\0';
1256       fclose(fp);
1257 
1258       // Skip pid and the command string. Note that we could be dealing with
1259       // weird command names, e.g. user could decide to rename java launcher
1260       // to "java 1.4.2 :)", then the stat file would look like
1261       //                1234 (java 1.4.2 :)) R ... ...
1262       // We don't really need to know the command string, just find the last
1263       // occurrence of ")" and then start parsing from there. See bug 4726580.
1264       char * s = strrchr(stat, ')');
1265 
1266       i = 0;
1267       if (s) {
1268         // Skip blank chars
1269         do s++; while (isspace(*s));
1270 
1271 #define _UFM UINTX_FORMAT
1272 #define _DFM INTX_FORMAT
1273 
1274         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1275         /*              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 */
1276         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,
1277              &state,          /* 3  %c  */
1278              &ppid,           /* 4  %d  */
1279              &pgrp,           /* 5  %d  */
1280              &session,        /* 6  %d  */
1281              &nr,             /* 7  %d  */
1282              &tpgrp,          /* 8  %d  */
1283              &flags,          /* 9  %lu  */
1284              &minflt,         /* 10 %lu  */
1285              &cminflt,        /* 11 %lu  */
1286              &majflt,         /* 12 %lu  */
1287              &cmajflt,        /* 13 %lu  */
1288              &utime,          /* 14 %lu  */
1289              &stime,          /* 15 %lu  */
1290              &cutime,         /* 16 %ld  */
1291              &cstime,         /* 17 %ld  */
1292              &prio,           /* 18 %ld  */
1293              &nice,           /* 19 %ld  */
1294              &junk,           /* 20 %ld  */
1295              &it_real,        /* 21 %ld  */
1296              &start,          /* 22 UINTX_FORMAT */
1297              &vsize,          /* 23 UINTX_FORMAT */
1298              &rss,            /* 24 INTX_FORMAT  */
1299              &rsslim,         /* 25 UINTX_FORMAT */
1300              &scodes,         /* 26 UINTX_FORMAT */
1301              &ecode,          /* 27 UINTX_FORMAT */
1302              &stack_start);   /* 28 UINTX_FORMAT */
1303       }
1304 
1305 #undef _UFM
1306 #undef _DFM
1307 
1308       if (i != 28 - 2) {
1309          assert(false, "Bad conversion from /proc/self/stat");
1310          // product mode - assume we are the initial thread, good luck in the
1311          // embedded case.
1312          warning("Can't detect initial thread stack location - bad conversion");
1313          stack_start = (uintptr_t) &rlim;
1314       }
1315     } else {
1316       // For some reason we can't open /proc/self/stat (for example, running on
1317       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1318       // most cases, so don't abort:
1319       warning("Can't detect initial thread stack location - no /proc/self/stat");
1320       stack_start = (uintptr_t) &rlim;
1321     }
1322   }
1323 
1324   // Now we have a pointer (stack_start) very close to the stack top, the
1325   // next thing to do is to figure out the exact location of stack top. We
1326   // can find out the virtual memory area that contains stack_start by
1327   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1328   // and its upper limit is the real stack top. (again, this would fail if
1329   // running inside chroot, because /proc may not exist.)
1330 
1331   uintptr_t stack_top;
1332   address low, high;
1333   if (find_vma((address)stack_start, &low, &high)) {
1334     // success, "high" is the true stack top. (ignore "low", because initial
1335     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1336     stack_top = (uintptr_t)high;
1337   } else {
1338     // failed, likely because /proc/self/maps does not exist
1339     warning("Can't detect initial thread stack location - find_vma failed");
1340     // best effort: stack_start is normally within a few pages below the real
1341     // stack top, use it as stack top, and reduce stack size so we won't put
1342     // guard page outside stack.
1343     stack_top = stack_start;
1344     stack_size -= 16 * page_size();
1345   }
1346 
1347   // stack_top could be partially down the page so align it
1348   stack_top = align_size_up(stack_top, page_size());
1349 
1350   if (max_size && stack_size > max_size) {
1351      _initial_thread_stack_size = max_size;
1352   } else {
1353      _initial_thread_stack_size = stack_size;
1354   }
1355 
1356   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1357   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1358 }
1359 
1360 ////////////////////////////////////////////////////////////////////////////////
1361 // time support
1362 
1363 // Time since start-up in seconds to a fine granularity.
1364 // Used by VMSelfDestructTimer and the MemProfiler.
1365 double os::elapsedTime() {
1366 
1367   return (double)(os::elapsed_counter()) * 0.000001;
1368 }
1369 
1370 jlong os::elapsed_counter() {
1371   timeval time;
1372   int status = gettimeofday(&time, NULL);
1373   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1374 }
1375 
1376 jlong os::elapsed_frequency() {
1377   return (1000 * 1000);
1378 }
1379 
1380 // For now, we say that linux does not support vtime.  I have no idea
1381 // whether it can actually be made to (DLD, 9/13/05).
1382 
1383 bool os::supports_vtime() { return false; }
1384 bool os::enable_vtime()   { return false; }
1385 bool os::vtime_enabled()  { return false; }
1386 double os::elapsedVTime() {
1387   // better than nothing, but not much
1388   return elapsedTime();
1389 }
1390 
1391 jlong os::javaTimeMillis() {
1392   timeval time;
1393   int status = gettimeofday(&time, NULL);
1394   assert(status != -1, "linux error");
1395   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1396 }
1397 
1398 #ifndef CLOCK_MONOTONIC
1399 #define CLOCK_MONOTONIC (1)
1400 #endif
1401 
1402 void os::Linux::clock_init() {
1403   // we do dlopen's in this particular order due to bug in linux
1404   // dynamical loader (see 6348968) leading to crash on exit
1405   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1406   if (handle == NULL) {
1407     handle = dlopen("librt.so", RTLD_LAZY);
1408   }
1409 
1410   if (handle) {
1411     int (*clock_getres_func)(clockid_t, struct timespec*) =
1412            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1413     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1414            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1415     if (clock_getres_func && clock_gettime_func) {
1416       // See if monotonic clock is supported by the kernel. Note that some
1417       // early implementations simply return kernel jiffies (updated every
1418       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1419       // for nano time (though the monotonic property is still nice to have).
1420       // It's fixed in newer kernels, however clock_getres() still returns
1421       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1422       // resolution for now. Hopefully as people move to new kernels, this
1423       // won't be a problem.
1424       struct timespec res;
1425       struct timespec tp;
1426       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1427           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1428         // yes, monotonic clock is supported
1429         _clock_gettime = clock_gettime_func;
1430       } else {
1431         // close librt if there is no monotonic clock
1432         dlclose(handle);
1433       }
1434     }
1435   }
1436 }
1437 
1438 #ifndef SYS_clock_getres
1439 
1440 #if defined(IA32) || defined(AMD64)
1441 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1442 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1443 #else
1444 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1445 #define sys_clock_getres(x,y)  -1
1446 #endif
1447 
1448 #else
1449 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1450 #endif
1451 
1452 void os::Linux::fast_thread_clock_init() {
1453   if (!UseLinuxPosixThreadCPUClocks) {
1454     return;
1455   }
1456   clockid_t clockid;
1457   struct timespec tp;
1458   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1459       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1460 
1461   // Switch to using fast clocks for thread cpu time if
1462   // the sys_clock_getres() returns 0 error code.
1463   // Note, that some kernels may support the current thread
1464   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1465   // returned by the pthread_getcpuclockid().
1466   // If the fast Posix clocks are supported then the sys_clock_getres()
1467   // must return at least tp.tv_sec == 0 which means a resolution
1468   // better than 1 sec. This is extra check for reliability.
1469 
1470   if(pthread_getcpuclockid_func &&
1471      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1472      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1473 
1474     _supports_fast_thread_cpu_time = true;
1475     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1476   }
1477 }
1478 
1479 jlong os::javaTimeNanos() {
1480   if (Linux::supports_monotonic_clock()) {
1481     struct timespec tp;
1482     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1483     assert(status == 0, "gettime error");
1484     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1485     return result;
1486   } else {
1487     timeval time;
1488     int status = gettimeofday(&time, NULL);
1489     assert(status != -1, "linux error");
1490     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1491     return 1000 * usecs;
1492   }
1493 }
1494 
1495 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1496   if (Linux::supports_monotonic_clock()) {
1497     info_ptr->max_value = ALL_64_BITS;
1498 
1499     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1500     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1501     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1502   } else {
1503     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1504     info_ptr->max_value = ALL_64_BITS;
1505 
1506     // gettimeofday is a real time clock so it skips
1507     info_ptr->may_skip_backward = true;
1508     info_ptr->may_skip_forward = true;
1509   }
1510 
1511   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1512 }
1513 
1514 // Return the real, user, and system times in seconds from an
1515 // arbitrary fixed point in the past.
1516 bool os::getTimesSecs(double* process_real_time,
1517                       double* process_user_time,
1518                       double* process_system_time) {
1519   struct tms ticks;
1520   clock_t real_ticks = times(&ticks);
1521 
1522   if (real_ticks == (clock_t) (-1)) {
1523     return false;
1524   } else {
1525     double ticks_per_second = (double) clock_tics_per_sec;
1526     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1527     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1528     *process_real_time = ((double) real_ticks) / ticks_per_second;
1529 
1530     return true;
1531   }
1532 }
1533 
1534 
1535 char * os::local_time_string(char *buf, size_t buflen) {
1536   struct tm t;
1537   time_t long_time;
1538   time(&long_time);
1539   localtime_r(&long_time, &t);
1540   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1541                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1542                t.tm_hour, t.tm_min, t.tm_sec);
1543   return buf;
1544 }
1545 
1546 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1547   return localtime_r(clock, res);
1548 }
1549 
1550 ////////////////////////////////////////////////////////////////////////////////
1551 // runtime exit support
1552 
1553 // Note: os::shutdown() might be called very early during initialization, or
1554 // called from signal handler. Before adding something to os::shutdown(), make
1555 // sure it is async-safe and can handle partially initialized VM.
1556 void os::shutdown() {
1557 
1558   // allow PerfMemory to attempt cleanup of any persistent resources
1559   perfMemory_exit();
1560 
1561   // needs to remove object in file system
1562   AttachListener::abort();
1563 
1564   // flush buffered output, finish log files
1565   ostream_abort();
1566 
1567   // Check for abort hook
1568   abort_hook_t abort_hook = Arguments::abort_hook();
1569   if (abort_hook != NULL) {
1570     abort_hook();
1571   }
1572 
1573 }
1574 
1575 // Note: os::abort() might be called very early during initialization, or
1576 // called from signal handler. Before adding something to os::abort(), make
1577 // sure it is async-safe and can handle partially initialized VM.
1578 void os::abort(bool dump_core) {
1579   os::shutdown();
1580   if (dump_core) {
1581 #ifndef PRODUCT
1582     fdStream out(defaultStream::output_fd());
1583     out.print_raw("Current thread is ");
1584     char buf[16];
1585     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1586     out.print_raw_cr(buf);
1587     out.print_raw_cr("Dumping core ...");
1588 #endif
1589     ::abort(); // dump core
1590   }
1591 
1592   ::exit(1);
1593 }
1594 
1595 // Die immediately, no exit hook, no abort hook, no cleanup.
1596 void os::die() {
1597   // _exit() on LinuxThreads only kills current thread
1598   ::abort();
1599 }
1600 
1601 // unused on linux for now.
1602 void os::set_error_file(const char *logfile) {}
1603 
1604 
1605 // This method is a copy of JDK's sysGetLastErrorString
1606 // from src/solaris/hpi/src/system_md.c
1607 
1608 size_t os::lasterror(char *buf, size_t len) {
1609 
1610   if (errno == 0)  return 0;
1611 
1612   const char *s = ::strerror(errno);
1613   size_t n = ::strlen(s);
1614   if (n >= len) {
1615     n = len - 1;
1616   }
1617   ::strncpy(buf, s, n);
1618   buf[n] = '\0';
1619   return n;
1620 }
1621 
1622 intx os::current_thread_id() { return (intx)pthread_self(); }
1623 int os::current_process_id() {
1624 
1625   // Under the old linux thread library, linux gives each thread
1626   // its own process id. Because of this each thread will return
1627   // a different pid if this method were to return the result
1628   // of getpid(2). Linux provides no api that returns the pid
1629   // of the launcher thread for the vm. This implementation
1630   // returns a unique pid, the pid of the launcher thread
1631   // that starts the vm 'process'.
1632 
1633   // Under the NPTL, getpid() returns the same pid as the
1634   // launcher thread rather than a unique pid per thread.
1635   // Use gettid() if you want the old pre NPTL behaviour.
1636 
1637   // if you are looking for the result of a call to getpid() that
1638   // returns a unique pid for the calling thread, then look at the
1639   // OSThread::thread_id() method in osThread_linux.hpp file
1640 
1641   return (int)(_initial_pid ? _initial_pid : getpid());
1642 }
1643 
1644 // DLL functions
1645 
1646 const char* os::dll_file_extension() { return ".so"; }
1647 
1648 // This must be hard coded because it's the system's temporary
1649 // directory not the java application's temp directory, ala java.io.tmpdir.
1650 const char* os::get_temp_directory() { return "/tmp"; }
1651 
1652 static bool file_exists(const char* filename) {
1653   struct stat statbuf;
1654   if (filename == NULL || strlen(filename) == 0) {
1655     return false;
1656   }
1657   return os::stat(filename, &statbuf) == 0;
1658 }
1659 
1660 void os::dll_build_name(char* buffer, size_t buflen,
1661                         const char* pname, const char* fname) {
1662   // Copied from libhpi
1663   const size_t pnamelen = pname ? strlen(pname) : 0;
1664 
1665   // Quietly truncate on buffer overflow.  Should be an error.
1666   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1667       *buffer = '\0';
1668       return;
1669   }
1670 
1671   if (pnamelen == 0) {
1672     snprintf(buffer, buflen, "lib%s.so", fname);
1673   } else if (strchr(pname, *os::path_separator()) != NULL) {
1674     int n;
1675     char** pelements = split_path(pname, &n);
1676     for (int i = 0 ; i < n ; i++) {
1677       // Really shouldn't be NULL, but check can't hurt
1678       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1679         continue; // skip the empty path values
1680       }
1681       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1682       if (file_exists(buffer)) {
1683         break;
1684       }
1685     }
1686     // release the storage
1687     for (int i = 0 ; i < n ; i++) {
1688       if (pelements[i] != NULL) {
1689         FREE_C_HEAP_ARRAY(char, pelements[i]);
1690       }
1691     }
1692     if (pelements != NULL) {
1693       FREE_C_HEAP_ARRAY(char*, pelements);
1694     }
1695   } else {
1696     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1697   }
1698 }
1699 
1700 const char* os::get_current_directory(char *buf, int buflen) {
1701   return getcwd(buf, buflen);
1702 }
1703 
1704 // check if addr is inside libjvm[_g].so
1705 bool os::address_is_in_vm(address addr) {
1706   static address libjvm_base_addr;
1707   Dl_info dlinfo;
1708 
1709   if (libjvm_base_addr == NULL) {
1710     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1711     libjvm_base_addr = (address)dlinfo.dli_fbase;
1712     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1713   }
1714 
1715   if (dladdr((void *)addr, &dlinfo)) {
1716     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1717   }
1718 
1719   return false;
1720 }
1721 
1722 bool os::dll_address_to_function_name(address addr, char *buf,
1723                                       int buflen, int *offset) {
1724   Dl_info dlinfo;
1725 
1726   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1727     if (buf != NULL) {
1728       if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1729         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1730       }
1731     }
1732     if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1733     return true;
1734   } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1735     if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1736        dlinfo.dli_fname, buf, buflen, offset) == Decoder::no_error) {
1737        return true;
1738     }
1739   }
1740 
1741   if (buf != NULL) buf[0] = '\0';
1742   if (offset != NULL) *offset = -1;
1743   return false;
1744 }
1745 
1746 struct _address_to_library_name {
1747   address addr;          // input : memory address
1748   size_t  buflen;        //         size of fname
1749   char*   fname;         // output: library name
1750   address base;          //         library base addr
1751 };
1752 
1753 static int address_to_library_name_callback(struct dl_phdr_info *info,
1754                                             size_t size, void *data) {
1755   int i;
1756   bool found = false;
1757   address libbase = NULL;
1758   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1759 
1760   // iterate through all loadable segments
1761   for (i = 0; i < info->dlpi_phnum; i++) {
1762     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1763     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1764       // base address of a library is the lowest address of its loaded
1765       // segments.
1766       if (libbase == NULL || libbase > segbase) {
1767         libbase = segbase;
1768       }
1769       // see if 'addr' is within current segment
1770       if (segbase <= d->addr &&
1771           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1772         found = true;
1773       }
1774     }
1775   }
1776 
1777   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1778   // so dll_address_to_library_name() can fall through to use dladdr() which
1779   // can figure out executable name from argv[0].
1780   if (found && info->dlpi_name && info->dlpi_name[0]) {
1781     d->base = libbase;
1782     if (d->fname) {
1783       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1784     }
1785     return 1;
1786   }
1787   return 0;
1788 }
1789 
1790 bool os::dll_address_to_library_name(address addr, char* buf,
1791                                      int buflen, int* offset) {
1792   Dl_info dlinfo;
1793   struct _address_to_library_name data;
1794 
1795   // There is a bug in old glibc dladdr() implementation that it could resolve
1796   // to wrong library name if the .so file has a base address != NULL. Here
1797   // we iterate through the program headers of all loaded libraries to find
1798   // out which library 'addr' really belongs to. This workaround can be
1799   // removed once the minimum requirement for glibc is moved to 2.3.x.
1800   data.addr = addr;
1801   data.fname = buf;
1802   data.buflen = buflen;
1803   data.base = NULL;
1804   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1805 
1806   if (rslt) {
1807      // buf already contains library name
1808      if (offset) *offset = addr - data.base;
1809      return true;
1810   } else if (dladdr((void*)addr, &dlinfo)){
1811      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1812      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1813      return true;
1814   } else {
1815      if (buf) buf[0] = '\0';
1816      if (offset) *offset = -1;
1817      return false;
1818   }
1819 }
1820 
1821   // Loads .dll/.so and
1822   // in case of error it checks if .dll/.so was built for the
1823   // same architecture as Hotspot is running on
1824 
1825 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1826 {
1827   void * result= ::dlopen(filename, RTLD_LAZY);
1828   if (result != NULL) {
1829     // Successful loading
1830     return result;
1831   }
1832 
1833   Elf32_Ehdr elf_head;
1834 
1835   // Read system error message into ebuf
1836   // It may or may not be overwritten below
1837   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1838   ebuf[ebuflen-1]='\0';
1839   int diag_msg_max_length=ebuflen-strlen(ebuf);
1840   char* diag_msg_buf=ebuf+strlen(ebuf);
1841 
1842   if (diag_msg_max_length==0) {
1843     // No more space in ebuf for additional diagnostics message
1844     return NULL;
1845   }
1846 
1847 
1848   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1849 
1850   if (file_descriptor < 0) {
1851     // Can't open library, report dlerror() message
1852     return NULL;
1853   }
1854 
1855   bool failed_to_read_elf_head=
1856     (sizeof(elf_head)!=
1857         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1858 
1859   ::close(file_descriptor);
1860   if (failed_to_read_elf_head) {
1861     // file i/o error - report dlerror() msg
1862     return NULL;
1863   }
1864 
1865   typedef struct {
1866     Elf32_Half  code;         // Actual value as defined in elf.h
1867     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1868     char        elf_class;    // 32 or 64 bit
1869     char        endianess;    // MSB or LSB
1870     char*       name;         // String representation
1871   } arch_t;
1872 
1873   #ifndef EM_486
1874   #define EM_486          6               /* Intel 80486 */
1875   #endif
1876 
1877   static const arch_t arch_array[]={
1878     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1879     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1880     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1881     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1882     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1883     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1884     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1885     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1886     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1887     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1888     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1889     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1890     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1891     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1892     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1893     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1894   };
1895 
1896   #if  (defined IA32)
1897     static  Elf32_Half running_arch_code=EM_386;
1898   #elif   (defined AMD64)
1899     static  Elf32_Half running_arch_code=EM_X86_64;
1900   #elif  (defined IA64)
1901     static  Elf32_Half running_arch_code=EM_IA_64;
1902   #elif  (defined __sparc) && (defined _LP64)
1903     static  Elf32_Half running_arch_code=EM_SPARCV9;
1904   #elif  (defined __sparc) && (!defined _LP64)
1905     static  Elf32_Half running_arch_code=EM_SPARC;
1906   #elif  (defined __powerpc64__)
1907     static  Elf32_Half running_arch_code=EM_PPC64;
1908   #elif  (defined __powerpc__)
1909     static  Elf32_Half running_arch_code=EM_PPC;
1910   #elif  (defined ARM)
1911     static  Elf32_Half running_arch_code=EM_ARM;
1912   #elif  (defined S390)
1913     static  Elf32_Half running_arch_code=EM_S390;
1914   #elif  (defined ALPHA)
1915     static  Elf32_Half running_arch_code=EM_ALPHA;
1916   #elif  (defined MIPSEL)
1917     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1918   #elif  (defined PARISC)
1919     static  Elf32_Half running_arch_code=EM_PARISC;
1920   #elif  (defined MIPS)
1921     static  Elf32_Half running_arch_code=EM_MIPS;
1922   #elif  (defined M68K)
1923     static  Elf32_Half running_arch_code=EM_68K;
1924   #else
1925     #error Method os::dll_load requires that one of following is defined:\
1926          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1927   #endif
1928 
1929   // Identify compatability class for VM's architecture and library's architecture
1930   // Obtain string descriptions for architectures
1931 
1932   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1933   int running_arch_index=-1;
1934 
1935   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1936     if (running_arch_code == arch_array[i].code) {
1937       running_arch_index    = i;
1938     }
1939     if (lib_arch.code == arch_array[i].code) {
1940       lib_arch.compat_class = arch_array[i].compat_class;
1941       lib_arch.name         = arch_array[i].name;
1942     }
1943   }
1944 
1945   assert(running_arch_index != -1,
1946     "Didn't find running architecture code (running_arch_code) in arch_array");
1947   if (running_arch_index == -1) {
1948     // Even though running architecture detection failed
1949     // we may still continue with reporting dlerror() message
1950     return NULL;
1951   }
1952 
1953   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1954     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1955     return NULL;
1956   }
1957 
1958 #ifndef S390
1959   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1960     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1961     return NULL;
1962   }
1963 #endif // !S390
1964 
1965   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1966     if ( lib_arch.name!=NULL ) {
1967       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1968         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1969         lib_arch.name, arch_array[running_arch_index].name);
1970     } else {
1971       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1972       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1973         lib_arch.code,
1974         arch_array[running_arch_index].name);
1975     }
1976   }
1977 
1978   return NULL;
1979 }
1980 
1981 /*
1982  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
1983  * chances are you might want to run the generated bits against glibc-2.0
1984  * libdl.so, so always use locking for any version of glibc.
1985  */
1986 void* os::dll_lookup(void* handle, const char* name) {
1987   pthread_mutex_lock(&dl_mutex);
1988   void* res = dlsym(handle, name);
1989   pthread_mutex_unlock(&dl_mutex);
1990   return res;
1991 }
1992 
1993 
1994 static bool _print_ascii_file(const char* filename, outputStream* st) {
1995   int fd = ::open(filename, O_RDONLY);
1996   if (fd == -1) {
1997      return false;
1998   }
1999 
2000   char buf[32];
2001   int bytes;
2002   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2003     st->print_raw(buf, bytes);
2004   }
2005 
2006   ::close(fd);
2007 
2008   return true;
2009 }
2010 
2011 void os::print_dll_info(outputStream *st) {
2012    st->print_cr("Dynamic libraries:");
2013 
2014    char fname[32];
2015    pid_t pid = os::Linux::gettid();
2016 
2017    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2018 
2019    if (!_print_ascii_file(fname, st)) {
2020      st->print("Can not get library information for pid = %d\n", pid);
2021    }
2022 }
2023 
2024 
2025 void os::print_os_info(outputStream* st) {
2026   st->print("OS:");
2027 
2028   // Try to identify popular distros.
2029   // Most Linux distributions have /etc/XXX-release file, which contains
2030   // the OS version string. Some have more than one /etc/XXX-release file
2031   // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2032   // so the order is important.
2033   if (!_print_ascii_file("/etc/mandrake-release", st) &&
2034       !_print_ascii_file("/etc/sun-release", st) &&
2035       !_print_ascii_file("/etc/redhat-release", st) &&
2036       !_print_ascii_file("/etc/SuSE-release", st) &&
2037       !_print_ascii_file("/etc/turbolinux-release", st) &&
2038       !_print_ascii_file("/etc/gentoo-release", st) &&
2039       !_print_ascii_file("/etc/debian_version", st) &&
2040       !_print_ascii_file("/etc/ltib-release", st) &&
2041       !_print_ascii_file("/etc/angstrom-version", st)) {
2042       st->print("Linux");
2043   }
2044   st->cr();
2045 
2046   // kernel
2047   st->print("uname:");
2048   struct utsname name;
2049   uname(&name);
2050   st->print(name.sysname); st->print(" ");
2051   st->print(name.release); st->print(" ");
2052   st->print(name.version); st->print(" ");
2053   st->print(name.machine);
2054   st->cr();
2055 
2056   // Print warning if unsafe chroot environment detected
2057   if (unsafe_chroot_detected) {
2058     st->print("WARNING!! ");
2059     st->print_cr(unstable_chroot_error);
2060   }
2061 
2062   // libc, pthread
2063   st->print("libc:");
2064   st->print(os::Linux::glibc_version()); st->print(" ");
2065   st->print(os::Linux::libpthread_version()); st->print(" ");
2066   if (os::Linux::is_LinuxThreads()) {
2067      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2068   }
2069   st->cr();
2070 
2071   // rlimit
2072   st->print("rlimit:");
2073   struct rlimit rlim;
2074 
2075   st->print(" STACK ");
2076   getrlimit(RLIMIT_STACK, &rlim);
2077   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2078   else st->print("%uk", rlim.rlim_cur >> 10);
2079 
2080   st->print(", CORE ");
2081   getrlimit(RLIMIT_CORE, &rlim);
2082   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2083   else st->print("%uk", rlim.rlim_cur >> 10);
2084 
2085   st->print(", NPROC ");
2086   getrlimit(RLIMIT_NPROC, &rlim);
2087   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2088   else st->print("%d", rlim.rlim_cur);
2089 
2090   st->print(", NOFILE ");
2091   getrlimit(RLIMIT_NOFILE, &rlim);
2092   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2093   else st->print("%d", rlim.rlim_cur);
2094 
2095   st->print(", AS ");
2096   getrlimit(RLIMIT_AS, &rlim);
2097   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2098   else st->print("%uk", rlim.rlim_cur >> 10);
2099   st->cr();
2100 
2101   // load average
2102   st->print("load average:");
2103   double loadavg[3];
2104   os::loadavg(loadavg, 3);
2105   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2106   st->cr();
2107 
2108   // meminfo
2109   st->print("\n/proc/meminfo:\n");
2110   _print_ascii_file("/proc/meminfo", st);
2111   st->cr();
2112 }
2113 
2114 void os::pd_print_cpu_info(outputStream* st) {
2115   st->print("\n/proc/cpuinfo:\n");
2116   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2117     st->print("  <Not Available>");
2118   }
2119   st->cr();
2120 }
2121 
2122 void os::print_memory_info(outputStream* st) {
2123 
2124   st->print("Memory:");
2125   st->print(" %dk page", os::vm_page_size()>>10);
2126 
2127   // values in struct sysinfo are "unsigned long"
2128   struct sysinfo si;
2129   sysinfo(&si);
2130 
2131   st->print(", physical " UINT64_FORMAT "k",
2132             os::physical_memory() >> 10);
2133   st->print("(" UINT64_FORMAT "k free)",
2134             os::available_memory() >> 10);
2135   st->print(", swap " UINT64_FORMAT "k",
2136             ((jlong)si.totalswap * si.mem_unit) >> 10);
2137   st->print("(" UINT64_FORMAT "k free)",
2138             ((jlong)si.freeswap * si.mem_unit) >> 10);
2139   st->cr();
2140 }
2141 
2142 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2143 // but they're the same for all the linux arch that we support
2144 // and they're the same for solaris but there's no common place to put this.
2145 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2146                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2147                           "ILL_COPROC", "ILL_BADSTK" };
2148 
2149 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2150                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2151                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2152 
2153 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2154 
2155 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2156 
2157 void os::print_siginfo(outputStream* st, void* siginfo) {
2158   st->print("siginfo:");
2159 
2160   const int buflen = 100;
2161   char buf[buflen];
2162   siginfo_t *si = (siginfo_t*)siginfo;
2163   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2164   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2165     st->print("si_errno=%s", buf);
2166   } else {
2167     st->print("si_errno=%d", si->si_errno);
2168   }
2169   const int c = si->si_code;
2170   assert(c > 0, "unexpected si_code");
2171   switch (si->si_signo) {
2172   case SIGILL:
2173     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2174     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2175     break;
2176   case SIGFPE:
2177     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2178     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2179     break;
2180   case SIGSEGV:
2181     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2182     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2183     break;
2184   case SIGBUS:
2185     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2186     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2187     break;
2188   default:
2189     st->print(", si_code=%d", si->si_code);
2190     // no si_addr
2191   }
2192 
2193   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2194       UseSharedSpaces) {
2195     FileMapInfo* mapinfo = FileMapInfo::current_info();
2196     if (mapinfo->is_in_shared_space(si->si_addr)) {
2197       st->print("\n\nError accessing class data sharing archive."   \
2198                 " Mapped file inaccessible during execution, "      \
2199                 " possible disk/network problem.");
2200     }
2201   }
2202   st->cr();
2203 }
2204 
2205 
2206 static void print_signal_handler(outputStream* st, int sig,
2207                                  char* buf, size_t buflen);
2208 
2209 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2210   st->print_cr("Signal Handlers:");
2211   print_signal_handler(st, SIGSEGV, buf, buflen);
2212   print_signal_handler(st, SIGBUS , buf, buflen);
2213   print_signal_handler(st, SIGFPE , buf, buflen);
2214   print_signal_handler(st, SIGPIPE, buf, buflen);
2215   print_signal_handler(st, SIGXFSZ, buf, buflen);
2216   print_signal_handler(st, SIGILL , buf, buflen);
2217   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2218   print_signal_handler(st, SR_signum, buf, buflen);
2219   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2220   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2221   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2222   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2223 }
2224 
2225 static char saved_jvm_path[MAXPATHLEN] = {0};
2226 
2227 // Find the full path to the current module, libjvm.so or libjvm_g.so
2228 void os::jvm_path(char *buf, jint buflen) {
2229   // Error checking.
2230   if (buflen < MAXPATHLEN) {
2231     assert(false, "must use a large-enough buffer");
2232     buf[0] = '\0';
2233     return;
2234   }
2235   // Lazy resolve the path to current module.
2236   if (saved_jvm_path[0] != 0) {
2237     strcpy(buf, saved_jvm_path);
2238     return;
2239   }
2240 
2241   char dli_fname[MAXPATHLEN];
2242   bool ret = dll_address_to_library_name(
2243                 CAST_FROM_FN_PTR(address, os::jvm_path),
2244                 dli_fname, sizeof(dli_fname), NULL);
2245   assert(ret != 0, "cannot locate libjvm");
2246   char *rp = realpath(dli_fname, buf);
2247   if (rp == NULL)
2248     return;
2249 
2250   if (Arguments::created_by_gamma_launcher()) {
2251     // Support for the gamma launcher.  Typical value for buf is
2252     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2253     // the right place in the string, then assume we are installed in a JDK and
2254     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2255     // up the path so it looks like libjvm.so is installed there (append a
2256     // fake suffix hotspot/libjvm.so).
2257     const char *p = buf + strlen(buf) - 1;
2258     for (int count = 0; p > buf && count < 5; ++count) {
2259       for (--p; p > buf && *p != '/'; --p)
2260         /* empty */ ;
2261     }
2262 
2263     if (strncmp(p, "/jre/lib/", 9) != 0) {
2264       // Look for JAVA_HOME in the environment.
2265       char* java_home_var = ::getenv("JAVA_HOME");
2266       if (java_home_var != NULL && java_home_var[0] != 0) {
2267         char* jrelib_p;
2268         int len;
2269 
2270         // Check the current module name "libjvm.so" or "libjvm_g.so".
2271         p = strrchr(buf, '/');
2272         assert(strstr(p, "/libjvm") == p, "invalid library name");
2273         p = strstr(p, "_g") ? "_g" : "";
2274 
2275         rp = realpath(java_home_var, buf);
2276         if (rp == NULL)
2277           return;
2278 
2279         // determine if this is a legacy image or modules image
2280         // modules image doesn't have "jre" subdirectory
2281         len = strlen(buf);
2282         jrelib_p = buf + len;
2283         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2284         if (0 != access(buf, F_OK)) {
2285           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2286         }
2287 
2288         if (0 == access(buf, F_OK)) {
2289           // Use current module name "libjvm[_g].so" instead of
2290           // "libjvm"debug_only("_g")".so" since for fastdebug version
2291           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2292           len = strlen(buf);
2293           snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2294         } else {
2295           // Go back to path of .so
2296           rp = realpath(dli_fname, buf);
2297           if (rp == NULL)
2298             return;
2299         }
2300       }
2301     }
2302   }
2303 
2304   strcpy(saved_jvm_path, buf);
2305 }
2306 
2307 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2308   // no prefix required, not even "_"
2309 }
2310 
2311 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2312   // no suffix required
2313 }
2314 
2315 ////////////////////////////////////////////////////////////////////////////////
2316 // sun.misc.Signal support
2317 
2318 static volatile jint sigint_count = 0;
2319 
2320 static void
2321 UserHandler(int sig, void *siginfo, void *context) {
2322   // 4511530 - sem_post is serialized and handled by the manager thread. When
2323   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2324   // don't want to flood the manager thread with sem_post requests.
2325   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2326       return;
2327 
2328   // Ctrl-C is pressed during error reporting, likely because the error
2329   // handler fails to abort. Let VM die immediately.
2330   if (sig == SIGINT && is_error_reported()) {
2331      os::die();
2332   }
2333 
2334   os::signal_notify(sig);
2335 }
2336 
2337 void* os::user_handler() {
2338   return CAST_FROM_FN_PTR(void*, UserHandler);
2339 }
2340 
2341 extern "C" {
2342   typedef void (*sa_handler_t)(int);
2343   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2344 }
2345 
2346 void* os::signal(int signal_number, void* handler) {
2347   struct sigaction sigAct, oldSigAct;
2348 
2349   sigfillset(&(sigAct.sa_mask));
2350   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2351   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2352 
2353   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2354     // -1 means registration failed
2355     return (void *)-1;
2356   }
2357 
2358   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2359 }
2360 
2361 void os::signal_raise(int signal_number) {
2362   ::raise(signal_number);
2363 }
2364 
2365 /*
2366  * The following code is moved from os.cpp for making this
2367  * code platform specific, which it is by its very nature.
2368  */
2369 
2370 // Will be modified when max signal is changed to be dynamic
2371 int os::sigexitnum_pd() {
2372   return NSIG;
2373 }
2374 
2375 // a counter for each possible signal value
2376 static volatile jint pending_signals[NSIG+1] = { 0 };
2377 
2378 // Linux(POSIX) specific hand shaking semaphore.
2379 static sem_t sig_sem;
2380 
2381 void os::signal_init_pd() {
2382   // Initialize signal structures
2383   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2384 
2385   // Initialize signal semaphore
2386   ::sem_init(&sig_sem, 0, 0);
2387 }
2388 
2389 void os::signal_notify(int sig) {
2390   Atomic::inc(&pending_signals[sig]);
2391   ::sem_post(&sig_sem);
2392 }
2393 
2394 static int check_pending_signals(bool wait) {
2395   Atomic::store(0, &sigint_count);
2396   for (;;) {
2397     for (int i = 0; i < NSIG + 1; i++) {
2398       jint n = pending_signals[i];
2399       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2400         return i;
2401       }
2402     }
2403     if (!wait) {
2404       return -1;
2405     }
2406     JavaThread *thread = JavaThread::current();
2407     ThreadBlockInVM tbivm(thread);
2408 
2409     bool threadIsSuspended;
2410     do {
2411       thread->set_suspend_equivalent();
2412       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2413       ::sem_wait(&sig_sem);
2414 
2415       // were we externally suspended while we were waiting?
2416       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2417       if (threadIsSuspended) {
2418         //
2419         // The semaphore has been incremented, but while we were waiting
2420         // another thread suspended us. We don't want to continue running
2421         // while suspended because that would surprise the thread that
2422         // suspended us.
2423         //
2424         ::sem_post(&sig_sem);
2425 
2426         thread->java_suspend_self();
2427       }
2428     } while (threadIsSuspended);
2429   }
2430 }
2431 
2432 int os::signal_lookup() {
2433   return check_pending_signals(false);
2434 }
2435 
2436 int os::signal_wait() {
2437   return check_pending_signals(true);
2438 }
2439 
2440 ////////////////////////////////////////////////////////////////////////////////
2441 // Virtual Memory
2442 
2443 int os::vm_page_size() {
2444   // Seems redundant as all get out
2445   assert(os::Linux::page_size() != -1, "must call os::init");
2446   return os::Linux::page_size();
2447 }
2448 
2449 // Solaris allocates memory by pages.
2450 int os::vm_allocation_granularity() {
2451   assert(os::Linux::page_size() != -1, "must call os::init");
2452   return os::Linux::page_size();
2453 }
2454 
2455 // Rationale behind this function:
2456 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2457 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2458 //  samples for JITted code. Here we create private executable mapping over the code cache
2459 //  and then we can use standard (well, almost, as mapping can change) way to provide
2460 //  info for the reporting script by storing timestamp and location of symbol
2461 void linux_wrap_code(char* base, size_t size) {
2462   static volatile jint cnt = 0;
2463 
2464   if (!UseOprofile) {
2465     return;
2466   }
2467 
2468   char buf[PATH_MAX+1];
2469   int num = Atomic::add(1, &cnt);
2470 
2471   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2472            os::get_temp_directory(), os::current_process_id(), num);
2473   unlink(buf);
2474 
2475   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2476 
2477   if (fd != -1) {
2478     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2479     if (rv != (off_t)-1) {
2480       if (::write(fd, "", 1) == 1) {
2481         mmap(base, size,
2482              PROT_READ|PROT_WRITE|PROT_EXEC,
2483              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2484       }
2485     }
2486     ::close(fd);
2487     unlink(buf);
2488   }
2489 }
2490 
2491 // NOTE: Linux kernel does not really reserve the pages for us.
2492 //       All it does is to check if there are enough free pages
2493 //       left at the time of mmap(). This could be a potential
2494 //       problem.
2495 bool os::commit_memory(char* addr, size_t size, bool exec) {
2496   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2497   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2498                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2499   if (res != (uintptr_t) MAP_FAILED) {
2500     if (UseNUMAInterleaving) {
2501       numa_make_global(addr, size);
2502     }
2503     return true;
2504   }
2505   return false;
2506 }
2507 
2508 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2509 #ifndef MAP_HUGETLB
2510 #define MAP_HUGETLB 0x40000
2511 #endif
2512 
2513 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2514 #ifndef MADV_HUGEPAGE
2515 #define MADV_HUGEPAGE 14
2516 #endif
2517 
2518 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2519                        bool exec) {
2520   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2521     int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2522     uintptr_t res =
2523       (uintptr_t) ::mmap(addr, size, prot,
2524                          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2525                          -1, 0);
2526     if (res != (uintptr_t) MAP_FAILED) {
2527       if (UseNUMAInterleaving) {
2528         numa_make_global(addr, size);
2529       }
2530       return true;
2531     }
2532     // Fall through and try to use small pages
2533   }
2534 
2535   if (commit_memory(addr, size, exec)) {
2536     realign_memory(addr, size, alignment_hint);
2537     return true;
2538   }
2539   return false;
2540 }
2541 
2542 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2543   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2544     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2545     // be supported or the memory may already be backed by huge pages.
2546     ::madvise(addr, bytes, MADV_HUGEPAGE);
2547   }
2548 }
2549 
2550 void os::free_memory(char *addr, size_t bytes) {
2551   commit_memory(addr, bytes, false);
2552 }
2553 
2554 void os::numa_make_global(char *addr, size_t bytes) {
2555   Linux::numa_interleave_memory(addr, bytes);
2556 }
2557 
2558 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2559   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2560 }
2561 
2562 bool os::numa_topology_changed()   { return false; }
2563 
2564 size_t os::numa_get_groups_num() {
2565   int max_node = Linux::numa_max_node();
2566   return max_node > 0 ? max_node + 1 : 1;
2567 }
2568 
2569 int os::numa_get_group_id() {
2570   int cpu_id = Linux::sched_getcpu();
2571   if (cpu_id != -1) {
2572     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2573     if (lgrp_id != -1) {
2574       return lgrp_id;
2575     }
2576   }
2577   return 0;
2578 }
2579 
2580 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2581   for (size_t i = 0; i < size; i++) {
2582     ids[i] = i;
2583   }
2584   return size;
2585 }
2586 
2587 bool os::get_page_info(char *start, page_info* info) {
2588   return false;
2589 }
2590 
2591 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2592   return end;
2593 }
2594 
2595 
2596 int os::Linux::sched_getcpu_syscall(void) {
2597   unsigned int cpu;
2598   int retval = -1;
2599 
2600 #if defined(IA32)
2601 # ifndef SYS_getcpu
2602 # define SYS_getcpu 318
2603 # endif
2604   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2605 #elif defined(AMD64)
2606 // Unfortunately we have to bring all these macros here from vsyscall.h
2607 // to be able to compile on old linuxes.
2608 # define __NR_vgetcpu 2
2609 # define VSYSCALL_START (-10UL << 20)
2610 # define VSYSCALL_SIZE 1024
2611 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2612   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2613   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2614   retval = vgetcpu(&cpu, NULL, NULL);
2615 #endif
2616 
2617   return (retval == -1) ? retval : cpu;
2618 }
2619 
2620 // Something to do with the numa-aware allocator needs these symbols
2621 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2622 extern "C" JNIEXPORT void numa_error(char *where) { }
2623 extern "C" JNIEXPORT int fork1() { return fork(); }
2624 
2625 
2626 // If we are running with libnuma version > 2, then we should
2627 // be trying to use symbols with versions 1.1
2628 // If we are running with earlier version, which did not have symbol versions,
2629 // we should use the base version.
2630 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2631   void *f = dlvsym(handle, name, "libnuma_1.1");
2632   if (f == NULL) {
2633     f = dlsym(handle, name);
2634   }
2635   return f;
2636 }
2637 
2638 bool os::Linux::libnuma_init() {
2639   // sched_getcpu() should be in libc.
2640   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2641                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2642 
2643   // If it's not, try a direct syscall.
2644   if (sched_getcpu() == -1)
2645     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2646 
2647   if (sched_getcpu() != -1) { // Does it work?
2648     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2649     if (handle != NULL) {
2650       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2651                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2652       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2653                                        libnuma_dlsym(handle, "numa_max_node")));
2654       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2655                                         libnuma_dlsym(handle, "numa_available")));
2656       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2657                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2658       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2659                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2660 
2661 
2662       if (numa_available() != -1) {
2663         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2664         // Create a cpu -> node mapping
2665         _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2666         rebuild_cpu_to_node_map();
2667         return true;
2668       }
2669     }
2670   }
2671   return false;
2672 }
2673 
2674 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2675 // The table is later used in get_node_by_cpu().
2676 void os::Linux::rebuild_cpu_to_node_map() {
2677   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2678                               // in libnuma (possible values are starting from 16,
2679                               // and continuing up with every other power of 2, but less
2680                               // than the maximum number of CPUs supported by kernel), and
2681                               // is a subject to change (in libnuma version 2 the requirements
2682                               // are more reasonable) we'll just hardcode the number they use
2683                               // in the library.
2684   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2685 
2686   size_t cpu_num = os::active_processor_count();
2687   size_t cpu_map_size = NCPUS / BitsPerCLong;
2688   size_t cpu_map_valid_size =
2689     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2690 
2691   cpu_to_node()->clear();
2692   cpu_to_node()->at_grow(cpu_num - 1);
2693   size_t node_num = numa_get_groups_num();
2694 
2695   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2696   for (size_t i = 0; i < node_num; i++) {
2697     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2698       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2699         if (cpu_map[j] != 0) {
2700           for (size_t k = 0; k < BitsPerCLong; k++) {
2701             if (cpu_map[j] & (1UL << k)) {
2702               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2703             }
2704           }
2705         }
2706       }
2707     }
2708   }
2709   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2710 }
2711 
2712 int os::Linux::get_node_by_cpu(int cpu_id) {
2713   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2714     return cpu_to_node()->at(cpu_id);
2715   }
2716   return -1;
2717 }
2718 
2719 GrowableArray<int>* os::Linux::_cpu_to_node;
2720 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2721 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2722 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2723 os::Linux::numa_available_func_t os::Linux::_numa_available;
2724 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2725 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2726 unsigned long* os::Linux::_numa_all_nodes;
2727 
2728 bool os::uncommit_memory(char* addr, size_t size) {
2729   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2730                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2731   return res  != (uintptr_t) MAP_FAILED;
2732 }
2733 
2734 // Linux uses a growable mapping for the stack, and if the mapping for
2735 // the stack guard pages is not removed when we detach a thread the
2736 // stack cannot grow beyond the pages where the stack guard was
2737 // mapped.  If at some point later in the process the stack expands to
2738 // that point, the Linux kernel cannot expand the stack any further
2739 // because the guard pages are in the way, and a segfault occurs.
2740 //
2741 // However, it's essential not to split the stack region by unmapping
2742 // a region (leaving a hole) that's already part of the stack mapping,
2743 // so if the stack mapping has already grown beyond the guard pages at
2744 // the time we create them, we have to truncate the stack mapping.
2745 // So, we need to know the extent of the stack mapping when
2746 // create_stack_guard_pages() is called.
2747 
2748 // Find the bounds of the stack mapping.  Return true for success.
2749 //
2750 // We only need this for stacks that are growable: at the time of
2751 // writing thread stacks don't use growable mappings (i.e. those
2752 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2753 // only applies to the main thread.
2754 
2755 static
2756 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2757 
2758   char buf[128];
2759   int fd, sz;
2760 
2761   if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2762     return false;
2763   }
2764 
2765   const char kw[] = "[stack]";
2766   const int kwlen = sizeof(kw)-1;
2767 
2768   // Address part of /proc/self/maps couldn't be more than 128 bytes
2769   while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2770      if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2771         // Extract addresses
2772         if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2773            uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2774            if (sp >= *bottom && sp <= *top) {
2775               ::close(fd);
2776               return true;
2777            }
2778         }
2779      }
2780   }
2781 
2782  ::close(fd);
2783   return false;
2784 }
2785 
2786 
2787 // If the (growable) stack mapping already extends beyond the point
2788 // where we're going to put our guard pages, truncate the mapping at
2789 // that point by munmap()ping it.  This ensures that when we later
2790 // munmap() the guard pages we don't leave a hole in the stack
2791 // mapping. This only affects the main/initial thread, but guard
2792 // against future OS changes
2793 bool os::create_stack_guard_pages(char* addr, size_t size) {
2794   uintptr_t stack_extent, stack_base;
2795   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2796   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2797       assert(os::Linux::is_initial_thread(),
2798            "growable stack in non-initial thread");
2799     if (stack_extent < (uintptr_t)addr)
2800       ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2801   }
2802 
2803   return os::commit_memory(addr, size);
2804 }
2805 
2806 // If this is a growable mapping, remove the guard pages entirely by
2807 // munmap()ping them.  If not, just call uncommit_memory(). This only
2808 // affects the main/initial thread, but guard against future OS changes
2809 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2810   uintptr_t stack_extent, stack_base;
2811   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2812   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2813       assert(os::Linux::is_initial_thread(),
2814            "growable stack in non-initial thread");
2815 
2816     return ::munmap(addr, size) == 0;
2817   }
2818 
2819   return os::uncommit_memory(addr, size);
2820 }
2821 
2822 static address _highest_vm_reserved_address = NULL;
2823 
2824 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2825 // at 'requested_addr'. If there are existing memory mappings at the same
2826 // location, however, they will be overwritten. If 'fixed' is false,
2827 // 'requested_addr' is only treated as a hint, the return value may or
2828 // may not start from the requested address. Unlike Linux mmap(), this
2829 // function returns NULL to indicate failure.
2830 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2831   char * addr;
2832   int flags;
2833 
2834   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2835   if (fixed) {
2836     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2837     flags |= MAP_FIXED;
2838   }
2839 
2840   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2841   // to PROT_EXEC if executable when we commit the page.
2842   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2843                        flags, -1, 0);
2844 
2845   if (addr != MAP_FAILED) {
2846     // anon_mmap() should only get called during VM initialization,
2847     // don't need lock (actually we can skip locking even it can be called
2848     // from multiple threads, because _highest_vm_reserved_address is just a
2849     // hint about the upper limit of non-stack memory regions.)
2850     if ((address)addr + bytes > _highest_vm_reserved_address) {
2851       _highest_vm_reserved_address = (address)addr + bytes;
2852     }
2853   }
2854 
2855   return addr == MAP_FAILED ? NULL : addr;
2856 }
2857 
2858 // Don't update _highest_vm_reserved_address, because there might be memory
2859 // regions above addr + size. If so, releasing a memory region only creates
2860 // a hole in the address space, it doesn't help prevent heap-stack collision.
2861 //
2862 static int anon_munmap(char * addr, size_t size) {
2863   return ::munmap(addr, size) == 0;
2864 }
2865 
2866 char* os::reserve_memory(size_t bytes, char* requested_addr,
2867                          size_t alignment_hint) {
2868   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2869 }
2870 
2871 bool os::release_memory(char* addr, size_t size) {
2872   return anon_munmap(addr, size);
2873 }
2874 
2875 static address highest_vm_reserved_address() {
2876   return _highest_vm_reserved_address;
2877 }
2878 
2879 static bool linux_mprotect(char* addr, size_t size, int prot) {
2880   // Linux wants the mprotect address argument to be page aligned.
2881   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2882 
2883   // According to SUSv3, mprotect() should only be used with mappings
2884   // established by mmap(), and mmap() always maps whole pages. Unaligned
2885   // 'addr' likely indicates problem in the VM (e.g. trying to change
2886   // protection of malloc'ed or statically allocated memory). Check the
2887   // caller if you hit this assert.
2888   assert(addr == bottom, "sanity check");
2889 
2890   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2891   return ::mprotect(bottom, size, prot) == 0;
2892 }
2893 
2894 // Set protections specified
2895 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2896                         bool is_committed) {
2897   unsigned int p = 0;
2898   switch (prot) {
2899   case MEM_PROT_NONE: p = PROT_NONE; break;
2900   case MEM_PROT_READ: p = PROT_READ; break;
2901   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2902   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2903   default:
2904     ShouldNotReachHere();
2905   }
2906   // is_committed is unused.
2907   return linux_mprotect(addr, bytes, p);
2908 }
2909 
2910 bool os::guard_memory(char* addr, size_t size) {
2911   return linux_mprotect(addr, size, PROT_NONE);
2912 }
2913 
2914 bool os::unguard_memory(char* addr, size_t size) {
2915   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2916 }
2917 
2918 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2919   bool result = false;
2920   void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2921                   MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2922                   -1, 0);
2923 
2924   if (p != (void *) -1) {
2925     // We don't know if this really is a huge page or not.
2926     FILE *fp = fopen("/proc/self/maps", "r");
2927     if (fp) {
2928       while (!feof(fp)) {
2929         char chars[257];
2930         long x = 0;
2931         if (fgets(chars, sizeof(chars), fp)) {
2932           if (sscanf(chars, "%lx-%*x", &x) == 1
2933               && x == (long)p) {
2934             if (strstr (chars, "hugepage")) {
2935               result = true;
2936               break;
2937             }
2938           }
2939         }
2940       }
2941       fclose(fp);
2942     }
2943     munmap (p, page_size);
2944     if (result)
2945       return true;
2946   }
2947 
2948   if (warn) {
2949     warning("HugeTLBFS is not supported by the operating system.");
2950   }
2951 
2952   return result;
2953 }
2954 
2955 /*
2956 * Set the coredump_filter bits to include largepages in core dump (bit 6)
2957 *
2958 * From the coredump_filter documentation:
2959 *
2960 * - (bit 0) anonymous private memory
2961 * - (bit 1) anonymous shared memory
2962 * - (bit 2) file-backed private memory
2963 * - (bit 3) file-backed shared memory
2964 * - (bit 4) ELF header pages in file-backed private memory areas (it is
2965 *           effective only if the bit 2 is cleared)
2966 * - (bit 5) hugetlb private memory
2967 * - (bit 6) hugetlb shared memory
2968 */
2969 static void set_coredump_filter(void) {
2970   FILE *f;
2971   long cdm;
2972 
2973   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
2974     return;
2975   }
2976 
2977   if (fscanf(f, "%lx", &cdm) != 1) {
2978     fclose(f);
2979     return;
2980   }
2981 
2982   rewind(f);
2983 
2984   if ((cdm & LARGEPAGES_BIT) == 0) {
2985     cdm |= LARGEPAGES_BIT;
2986     fprintf(f, "%#lx", cdm);
2987   }
2988 
2989   fclose(f);
2990 }
2991 
2992 // Large page support
2993 
2994 static size_t _large_page_size = 0;
2995 
2996 void os::large_page_init() {
2997   if (!UseLargePages) {
2998     UseHugeTLBFS = false;
2999     UseSHM = false;
3000     return;
3001   }
3002 
3003   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3004     // If UseLargePages is specified on the command line try both methods,
3005     // if it's default, then try only HugeTLBFS.
3006     if (FLAG_IS_DEFAULT(UseLargePages)) {
3007       UseHugeTLBFS = true;
3008     } else {
3009       UseHugeTLBFS = UseSHM = true;
3010     }
3011   }
3012 
3013   if (LargePageSizeInBytes) {
3014     _large_page_size = LargePageSizeInBytes;
3015   } else {
3016     // large_page_size on Linux is used to round up heap size. x86 uses either
3017     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3018     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3019     // page as large as 256M.
3020     //
3021     // Here we try to figure out page size by parsing /proc/meminfo and looking
3022     // for a line with the following format:
3023     //    Hugepagesize:     2048 kB
3024     //
3025     // If we can't determine the value (e.g. /proc is not mounted, or the text
3026     // format has been changed), we'll use the largest page size supported by
3027     // the processor.
3028 
3029 #ifndef ZERO
3030     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3031                        ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3032 #endif // ZERO
3033 
3034     FILE *fp = fopen("/proc/meminfo", "r");
3035     if (fp) {
3036       while (!feof(fp)) {
3037         int x = 0;
3038         char buf[16];
3039         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3040           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3041             _large_page_size = x * K;
3042             break;
3043           }
3044         } else {
3045           // skip to next line
3046           for (;;) {
3047             int ch = fgetc(fp);
3048             if (ch == EOF || ch == (int)'\n') break;
3049           }
3050         }
3051       }
3052       fclose(fp);
3053     }
3054   }
3055 
3056   // print a warning if any large page related flag is specified on command line
3057   bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3058 
3059   const size_t default_page_size = (size_t)Linux::page_size();
3060   if (_large_page_size > default_page_size) {
3061     _page_sizes[0] = _large_page_size;
3062     _page_sizes[1] = default_page_size;
3063     _page_sizes[2] = 0;
3064   }
3065   UseHugeTLBFS = UseHugeTLBFS &&
3066                  Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3067 
3068   if (UseHugeTLBFS)
3069     UseSHM = false;
3070 
3071   UseLargePages = UseHugeTLBFS || UseSHM;
3072 
3073   set_coredump_filter();
3074 }
3075 
3076 #ifndef SHM_HUGETLB
3077 #define SHM_HUGETLB 04000
3078 #endif
3079 
3080 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3081   // "exec" is passed in but not used.  Creating the shared image for
3082   // the code cache doesn't have an SHM_X executable permission to check.
3083   assert(UseLargePages && UseSHM, "only for SHM large pages");
3084 
3085   key_t key = IPC_PRIVATE;
3086   char *addr;
3087 
3088   bool warn_on_failure = UseLargePages &&
3089                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3090                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3091                         );
3092   char msg[128];
3093 
3094   // Create a large shared memory region to attach to based on size.
3095   // Currently, size is the total size of the heap
3096   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3097   if (shmid == -1) {
3098      // Possible reasons for shmget failure:
3099      // 1. shmmax is too small for Java heap.
3100      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3101      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3102      // 2. not enough large page memory.
3103      //    > check available large pages: cat /proc/meminfo
3104      //    > increase amount of large pages:
3105      //          echo new_value > /proc/sys/vm/nr_hugepages
3106      //      Note 1: different Linux may use different name for this property,
3107      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3108      //      Note 2: it's possible there's enough physical memory available but
3109      //            they are so fragmented after a long run that they can't
3110      //            coalesce into large pages. Try to reserve large pages when
3111      //            the system is still "fresh".
3112      if (warn_on_failure) {
3113        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3114        warning(msg);
3115      }
3116      return NULL;
3117   }
3118 
3119   // attach to the region
3120   addr = (char*)shmat(shmid, req_addr, 0);
3121   int err = errno;
3122 
3123   // Remove shmid. If shmat() is successful, the actual shared memory segment
3124   // will be deleted when it's detached by shmdt() or when the process
3125   // terminates. If shmat() is not successful this will remove the shared
3126   // segment immediately.
3127   shmctl(shmid, IPC_RMID, NULL);
3128 
3129   if ((intptr_t)addr == -1) {
3130      if (warn_on_failure) {
3131        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3132        warning(msg);
3133      }
3134      return NULL;
3135   }
3136 
3137   if ((addr != NULL) && UseNUMAInterleaving) {
3138     numa_make_global(addr, bytes);
3139   }
3140 
3141   return addr;
3142 }
3143 
3144 bool os::release_memory_special(char* base, size_t bytes) {
3145   // detaching the SHM segment will also delete it, see reserve_memory_special()
3146   int rslt = shmdt(base);
3147   return rslt == 0;
3148 }
3149 
3150 size_t os::large_page_size() {
3151   return _large_page_size;
3152 }
3153 
3154 // HugeTLBFS allows application to commit large page memory on demand;
3155 // with SysV SHM the entire memory region must be allocated as shared
3156 // memory.
3157 bool os::can_commit_large_page_memory() {
3158   return UseHugeTLBFS;
3159 }
3160 
3161 bool os::can_execute_large_page_memory() {
3162   return UseHugeTLBFS;
3163 }
3164 
3165 // Reserve memory at an arbitrary address, only if that area is
3166 // available (and not reserved for something else).
3167 
3168 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3169   const int max_tries = 10;
3170   char* base[max_tries];
3171   size_t size[max_tries];
3172   const size_t gap = 0x000000;
3173 
3174   // Assert only that the size is a multiple of the page size, since
3175   // that's all that mmap requires, and since that's all we really know
3176   // about at this low abstraction level.  If we need higher alignment,
3177   // we can either pass an alignment to this method or verify alignment
3178   // in one of the methods further up the call chain.  See bug 5044738.
3179   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3180 
3181   // Repeatedly allocate blocks until the block is allocated at the
3182   // right spot. Give up after max_tries. Note that reserve_memory() will
3183   // automatically update _highest_vm_reserved_address if the call is
3184   // successful. The variable tracks the highest memory address every reserved
3185   // by JVM. It is used to detect heap-stack collision if running with
3186   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3187   // space than needed, it could confuse the collision detecting code. To
3188   // solve the problem, save current _highest_vm_reserved_address and
3189   // calculate the correct value before return.
3190   address old_highest = _highest_vm_reserved_address;
3191 
3192   // Linux mmap allows caller to pass an address as hint; give it a try first,
3193   // if kernel honors the hint then we can return immediately.
3194   char * addr = anon_mmap(requested_addr, bytes, false);
3195   if (addr == requested_addr) {
3196      return requested_addr;
3197   }
3198 
3199   if (addr != NULL) {
3200      // mmap() is successful but it fails to reserve at the requested address
3201      anon_munmap(addr, bytes);
3202   }
3203 
3204   int i;
3205   for (i = 0; i < max_tries; ++i) {
3206     base[i] = reserve_memory(bytes);
3207 
3208     if (base[i] != NULL) {
3209       // Is this the block we wanted?
3210       if (base[i] == requested_addr) {
3211         size[i] = bytes;
3212         break;
3213       }
3214 
3215       // Does this overlap the block we wanted? Give back the overlapped
3216       // parts and try again.
3217 
3218       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3219       if (top_overlap >= 0 && top_overlap < bytes) {
3220         unmap_memory(base[i], top_overlap);
3221         base[i] += top_overlap;
3222         size[i] = bytes - top_overlap;
3223       } else {
3224         size_t bottom_overlap = base[i] + bytes - requested_addr;
3225         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3226           unmap_memory(requested_addr, bottom_overlap);
3227           size[i] = bytes - bottom_overlap;
3228         } else {
3229           size[i] = bytes;
3230         }
3231       }
3232     }
3233   }
3234 
3235   // Give back the unused reserved pieces.
3236 
3237   for (int j = 0; j < i; ++j) {
3238     if (base[j] != NULL) {
3239       unmap_memory(base[j], size[j]);
3240     }
3241   }
3242 
3243   if (i < max_tries) {
3244     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3245     return requested_addr;
3246   } else {
3247     _highest_vm_reserved_address = old_highest;
3248     return NULL;
3249   }
3250 }
3251 
3252 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3253   return ::read(fd, buf, nBytes);
3254 }
3255 
3256 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3257 // Solaris uses poll(), linux uses park().
3258 // Poll() is likely a better choice, assuming that Thread.interrupt()
3259 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3260 // SIGSEGV, see 4355769.
3261 
3262 const int NANOSECS_PER_MILLISECS = 1000000;
3263 
3264 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3265   assert(thread == Thread::current(),  "thread consistency check");
3266 
3267   ParkEvent * const slp = thread->_SleepEvent ;
3268   slp->reset() ;
3269   OrderAccess::fence() ;
3270 
3271   if (interruptible) {
3272     jlong prevtime = javaTimeNanos();
3273 
3274     for (;;) {
3275       if (os::is_interrupted(thread, true)) {
3276         return OS_INTRPT;
3277       }
3278 
3279       jlong newtime = javaTimeNanos();
3280 
3281       if (newtime - prevtime < 0) {
3282         // time moving backwards, should only happen if no monotonic clock
3283         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3284         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3285       } else {
3286         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3287       }
3288 
3289       if(millis <= 0) {
3290         return OS_OK;
3291       }
3292 
3293       prevtime = newtime;
3294 
3295       {
3296         assert(thread->is_Java_thread(), "sanity check");
3297         JavaThread *jt = (JavaThread *) thread;
3298         ThreadBlockInVM tbivm(jt);
3299         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3300 
3301         jt->set_suspend_equivalent();
3302         // cleared by handle_special_suspend_equivalent_condition() or
3303         // java_suspend_self() via check_and_wait_while_suspended()
3304 
3305         slp->park(millis);
3306 
3307         // were we externally suspended while we were waiting?
3308         jt->check_and_wait_while_suspended();
3309       }
3310     }
3311   } else {
3312     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3313     jlong prevtime = javaTimeNanos();
3314 
3315     for (;;) {
3316       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3317       // the 1st iteration ...
3318       jlong newtime = javaTimeNanos();
3319 
3320       if (newtime - prevtime < 0) {
3321         // time moving backwards, should only happen if no monotonic clock
3322         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3323         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3324       } else {
3325         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3326       }
3327 
3328       if(millis <= 0) break ;
3329 
3330       prevtime = newtime;
3331       slp->park(millis);
3332     }
3333     return OS_OK ;
3334   }
3335 }
3336 
3337 int os::naked_sleep() {
3338   // %% make the sleep time an integer flag. for now use 1 millisec.
3339   return os::sleep(Thread::current(), 1, false);
3340 }
3341 
3342 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3343 void os::infinite_sleep() {
3344   while (true) {    // sleep forever ...
3345     ::sleep(100);   // ... 100 seconds at a time
3346   }
3347 }
3348 
3349 // Used to convert frequent JVM_Yield() to nops
3350 bool os::dont_yield() {
3351   return DontYieldALot;
3352 }
3353 
3354 void os::yield() {
3355   sched_yield();
3356 }
3357 
3358 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3359 
3360 void os::yield_all(int attempts) {
3361   // Yields to all threads, including threads with lower priorities
3362   // Threads on Linux are all with same priority. The Solaris style
3363   // os::yield_all() with nanosleep(1ms) is not necessary.
3364   sched_yield();
3365 }
3366 
3367 // Called from the tight loops to possibly influence time-sharing heuristics
3368 void os::loop_breaker(int attempts) {
3369   os::yield_all(attempts);
3370 }
3371 
3372 ////////////////////////////////////////////////////////////////////////////////
3373 // thread priority support
3374 
3375 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3376 // only supports dynamic priority, static priority must be zero. For real-time
3377 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3378 // However, for large multi-threaded applications, SCHED_RR is not only slower
3379 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3380 // of 5 runs - Sep 2005).
3381 //
3382 // The following code actually changes the niceness of kernel-thread/LWP. It
3383 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3384 // not the entire user process, and user level threads are 1:1 mapped to kernel
3385 // threads. It has always been the case, but could change in the future. For
3386 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3387 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3388 
3389 int os::java_to_os_priority[MaxPriority + 1] = {
3390   19,              // 0 Entry should never be used
3391 
3392    4,              // 1 MinPriority
3393    3,              // 2
3394    2,              // 3
3395 
3396    1,              // 4
3397    0,              // 5 NormPriority
3398   -1,              // 6
3399 
3400   -2,              // 7
3401   -3,              // 8
3402   -4,              // 9 NearMaxPriority
3403 
3404   -5               // 10 MaxPriority
3405 };
3406 
3407 static int prio_init() {
3408   if (ThreadPriorityPolicy == 1) {
3409     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3410     // if effective uid is not root. Perhaps, a more elegant way of doing
3411     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3412     if (geteuid() != 0) {
3413       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3414         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3415       }
3416       ThreadPriorityPolicy = 0;
3417     }
3418   }
3419   return 0;
3420 }
3421 
3422 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3423   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3424 
3425   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3426   return (ret == 0) ? OS_OK : OS_ERR;
3427 }
3428 
3429 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3430   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3431     *priority_ptr = java_to_os_priority[NormPriority];
3432     return OS_OK;
3433   }
3434 
3435   errno = 0;
3436   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3437   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3438 }
3439 
3440 // Hint to the underlying OS that a task switch would not be good.
3441 // Void return because it's a hint and can fail.
3442 void os::hint_no_preempt() {}
3443 
3444 ////////////////////////////////////////////////////////////////////////////////
3445 // suspend/resume support
3446 
3447 //  the low-level signal-based suspend/resume support is a remnant from the
3448 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3449 //  within hotspot. Now there is a single use-case for this:
3450 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3451 //      that runs in the watcher thread.
3452 //  The remaining code is greatly simplified from the more general suspension
3453 //  code that used to be used.
3454 //
3455 //  The protocol is quite simple:
3456 //  - suspend:
3457 //      - sends a signal to the target thread
3458 //      - polls the suspend state of the osthread using a yield loop
3459 //      - target thread signal handler (SR_handler) sets suspend state
3460 //        and blocks in sigsuspend until continued
3461 //  - resume:
3462 //      - sets target osthread state to continue
3463 //      - sends signal to end the sigsuspend loop in the SR_handler
3464 //
3465 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3466 //
3467 
3468 static void resume_clear_context(OSThread *osthread) {
3469   osthread->set_ucontext(NULL);
3470   osthread->set_siginfo(NULL);
3471 
3472   // notify the suspend action is completed, we have now resumed
3473   osthread->sr.clear_suspended();
3474 }
3475 
3476 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3477   osthread->set_ucontext(context);
3478   osthread->set_siginfo(siginfo);
3479 }
3480 
3481 //
3482 // Handler function invoked when a thread's execution is suspended or
3483 // resumed. We have to be careful that only async-safe functions are
3484 // called here (Note: most pthread functions are not async safe and
3485 // should be avoided.)
3486 //
3487 // Note: sigwait() is a more natural fit than sigsuspend() from an
3488 // interface point of view, but sigwait() prevents the signal hander
3489 // from being run. libpthread would get very confused by not having
3490 // its signal handlers run and prevents sigwait()'s use with the
3491 // mutex granting granting signal.
3492 //
3493 // Currently only ever called on the VMThread
3494 //
3495 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3496   // Save and restore errno to avoid confusing native code with EINTR
3497   // after sigsuspend.
3498   int old_errno = errno;
3499 
3500   Thread* thread = Thread::current();
3501   OSThread* osthread = thread->osthread();
3502   assert(thread->is_VM_thread(), "Must be VMThread");
3503   // read current suspend action
3504   int action = osthread->sr.suspend_action();
3505   if (action == SR_SUSPEND) {
3506     suspend_save_context(osthread, siginfo, context);
3507 
3508     // Notify the suspend action is about to be completed. do_suspend()
3509     // waits until SR_SUSPENDED is set and then returns. We will wait
3510     // here for a resume signal and that completes the suspend-other
3511     // action. do_suspend/do_resume is always called as a pair from
3512     // the same thread - so there are no races
3513 
3514     // notify the caller
3515     osthread->sr.set_suspended();
3516 
3517     sigset_t suspend_set;  // signals for sigsuspend()
3518 
3519     // get current set of blocked signals and unblock resume signal
3520     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3521     sigdelset(&suspend_set, SR_signum);
3522 
3523     // wait here until we are resumed
3524     do {
3525       sigsuspend(&suspend_set);
3526       // ignore all returns until we get a resume signal
3527     } while (osthread->sr.suspend_action() != SR_CONTINUE);
3528 
3529     resume_clear_context(osthread);
3530 
3531   } else {
3532     assert(action == SR_CONTINUE, "unexpected sr action");
3533     // nothing special to do - just leave the handler
3534   }
3535 
3536   errno = old_errno;
3537 }
3538 
3539 
3540 static int SR_initialize() {
3541   struct sigaction act;
3542   char *s;
3543   /* Get signal number to use for suspend/resume */
3544   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3545     int sig = ::strtol(s, 0, 10);
3546     if (sig > 0 || sig < _NSIG) {
3547         SR_signum = sig;
3548     }
3549   }
3550 
3551   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3552         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3553 
3554   sigemptyset(&SR_sigset);
3555   sigaddset(&SR_sigset, SR_signum);
3556 
3557   /* Set up signal handler for suspend/resume */
3558   act.sa_flags = SA_RESTART|SA_SIGINFO;
3559   act.sa_handler = (void (*)(int)) SR_handler;
3560 
3561   // SR_signum is blocked by default.
3562   // 4528190 - We also need to block pthread restart signal (32 on all
3563   // supported Linux platforms). Note that LinuxThreads need to block
3564   // this signal for all threads to work properly. So we don't have
3565   // to use hard-coded signal number when setting up the mask.
3566   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3567 
3568   if (sigaction(SR_signum, &act, 0) == -1) {
3569     return -1;
3570   }
3571 
3572   // Save signal flag
3573   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3574   return 0;
3575 }
3576 
3577 static int SR_finalize() {
3578   return 0;
3579 }
3580 
3581 
3582 // returns true on success and false on error - really an error is fatal
3583 // but this seems the normal response to library errors
3584 static bool do_suspend(OSThread* osthread) {
3585   // mark as suspended and send signal
3586   osthread->sr.set_suspend_action(SR_SUSPEND);
3587   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3588   assert_status(status == 0, status, "pthread_kill");
3589 
3590   // check status and wait until notified of suspension
3591   if (status == 0) {
3592     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3593       os::yield_all(i);
3594     }
3595     osthread->sr.set_suspend_action(SR_NONE);
3596     return true;
3597   }
3598   else {
3599     osthread->sr.set_suspend_action(SR_NONE);
3600     return false;
3601   }
3602 }
3603 
3604 static void do_resume(OSThread* osthread) {
3605   assert(osthread->sr.is_suspended(), "thread should be suspended");
3606   osthread->sr.set_suspend_action(SR_CONTINUE);
3607 
3608   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3609   assert_status(status == 0, status, "pthread_kill");
3610   // check status and wait unit notified of resumption
3611   if (status == 0) {
3612     for (int i = 0; osthread->sr.is_suspended(); i++) {
3613       os::yield_all(i);
3614     }
3615   }
3616   osthread->sr.set_suspend_action(SR_NONE);
3617 }
3618 
3619 ////////////////////////////////////////////////////////////////////////////////
3620 // interrupt support
3621 
3622 void os::interrupt(Thread* thread) {
3623   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3624     "possibility of dangling Thread pointer");
3625 
3626   OSThread* osthread = thread->osthread();
3627 
3628   if (!osthread->interrupted()) {
3629     osthread->set_interrupted(true);
3630     // More than one thread can get here with the same value of osthread,
3631     // resulting in multiple notifications.  We do, however, want the store
3632     // to interrupted() to be visible to other threads before we execute unpark().
3633     OrderAccess::fence();
3634     ParkEvent * const slp = thread->_SleepEvent ;
3635     if (slp != NULL) slp->unpark() ;
3636   }
3637 
3638   // For JSR166. Unpark even if interrupt status already was set
3639   if (thread->is_Java_thread())
3640     ((JavaThread*)thread)->parker()->unpark();
3641 
3642   ParkEvent * ev = thread->_ParkEvent ;
3643   if (ev != NULL) ev->unpark() ;
3644 
3645 }
3646 
3647 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3648   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3649     "possibility of dangling Thread pointer");
3650 
3651   OSThread* osthread = thread->osthread();
3652 
3653   bool interrupted = osthread->interrupted();
3654 
3655   if (interrupted && clear_interrupted) {
3656     osthread->set_interrupted(false);
3657     // consider thread->_SleepEvent->reset() ... optional optimization
3658   }
3659 
3660   return interrupted;
3661 }
3662 
3663 ///////////////////////////////////////////////////////////////////////////////////
3664 // signal handling (except suspend/resume)
3665 
3666 // This routine may be used by user applications as a "hook" to catch signals.
3667 // The user-defined signal handler must pass unrecognized signals to this
3668 // routine, and if it returns true (non-zero), then the signal handler must
3669 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3670 // routine will never retun false (zero), but instead will execute a VM panic
3671 // routine kill the process.
3672 //
3673 // If this routine returns false, it is OK to call it again.  This allows
3674 // the user-defined signal handler to perform checks either before or after
3675 // the VM performs its own checks.  Naturally, the user code would be making
3676 // a serious error if it tried to handle an exception (such as a null check
3677 // or breakpoint) that the VM was generating for its own correct operation.
3678 //
3679 // This routine may recognize any of the following kinds of signals:
3680 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3681 // It should be consulted by handlers for any of those signals.
3682 //
3683 // The caller of this routine must pass in the three arguments supplied
3684 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3685 // field of the structure passed to sigaction().  This routine assumes that
3686 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3687 //
3688 // Note that the VM will print warnings if it detects conflicting signal
3689 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3690 //
3691 extern "C" JNIEXPORT int
3692 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3693                         void* ucontext, int abort_if_unrecognized);
3694 
3695 void signalHandler(int sig, siginfo_t* info, void* uc) {
3696   assert(info != NULL && uc != NULL, "it must be old kernel");
3697   JVM_handle_linux_signal(sig, info, uc, true);
3698 }
3699 
3700 
3701 // This boolean allows users to forward their own non-matching signals
3702 // to JVM_handle_linux_signal, harmlessly.
3703 bool os::Linux::signal_handlers_are_installed = false;
3704 
3705 // For signal-chaining
3706 struct sigaction os::Linux::sigact[MAXSIGNUM];
3707 unsigned int os::Linux::sigs = 0;
3708 bool os::Linux::libjsig_is_loaded = false;
3709 typedef struct sigaction *(*get_signal_t)(int);
3710 get_signal_t os::Linux::get_signal_action = NULL;
3711 
3712 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3713   struct sigaction *actp = NULL;
3714 
3715   if (libjsig_is_loaded) {
3716     // Retrieve the old signal handler from libjsig
3717     actp = (*get_signal_action)(sig);
3718   }
3719   if (actp == NULL) {
3720     // Retrieve the preinstalled signal handler from jvm
3721     actp = get_preinstalled_handler(sig);
3722   }
3723 
3724   return actp;
3725 }
3726 
3727 static bool call_chained_handler(struct sigaction *actp, int sig,
3728                                  siginfo_t *siginfo, void *context) {
3729   // Call the old signal handler
3730   if (actp->sa_handler == SIG_DFL) {
3731     // It's more reasonable to let jvm treat it as an unexpected exception
3732     // instead of taking the default action.
3733     return false;
3734   } else if (actp->sa_handler != SIG_IGN) {
3735     if ((actp->sa_flags & SA_NODEFER) == 0) {
3736       // automaticlly block the signal
3737       sigaddset(&(actp->sa_mask), sig);
3738     }
3739 
3740     sa_handler_t hand;
3741     sa_sigaction_t sa;
3742     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3743     // retrieve the chained handler
3744     if (siginfo_flag_set) {
3745       sa = actp->sa_sigaction;
3746     } else {
3747       hand = actp->sa_handler;
3748     }
3749 
3750     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3751       actp->sa_handler = SIG_DFL;
3752     }
3753 
3754     // try to honor the signal mask
3755     sigset_t oset;
3756     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3757 
3758     // call into the chained handler
3759     if (siginfo_flag_set) {
3760       (*sa)(sig, siginfo, context);
3761     } else {
3762       (*hand)(sig);
3763     }
3764 
3765     // restore the signal mask
3766     pthread_sigmask(SIG_SETMASK, &oset, 0);
3767   }
3768   // Tell jvm's signal handler the signal is taken care of.
3769   return true;
3770 }
3771 
3772 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3773   bool chained = false;
3774   // signal-chaining
3775   if (UseSignalChaining) {
3776     struct sigaction *actp = get_chained_signal_action(sig);
3777     if (actp != NULL) {
3778       chained = call_chained_handler(actp, sig, siginfo, context);
3779     }
3780   }
3781   return chained;
3782 }
3783 
3784 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3785   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3786     return &sigact[sig];
3787   }
3788   return NULL;
3789 }
3790 
3791 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3792   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3793   sigact[sig] = oldAct;
3794   sigs |= (unsigned int)1 << sig;
3795 }
3796 
3797 // for diagnostic
3798 int os::Linux::sigflags[MAXSIGNUM];
3799 
3800 int os::Linux::get_our_sigflags(int sig) {
3801   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3802   return sigflags[sig];
3803 }
3804 
3805 void os::Linux::set_our_sigflags(int sig, int flags) {
3806   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3807   sigflags[sig] = flags;
3808 }
3809 
3810 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3811   // Check for overwrite.
3812   struct sigaction oldAct;
3813   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3814 
3815   void* oldhand = oldAct.sa_sigaction
3816                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3817                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3818   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3819       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3820       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3821     if (AllowUserSignalHandlers || !set_installed) {
3822       // Do not overwrite; user takes responsibility to forward to us.
3823       return;
3824     } else if (UseSignalChaining) {
3825       // save the old handler in jvm
3826       save_preinstalled_handler(sig, oldAct);
3827       // libjsig also interposes the sigaction() call below and saves the
3828       // old sigaction on it own.
3829     } else {
3830       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3831                     "%#lx for signal %d.", (long)oldhand, sig));
3832     }
3833   }
3834 
3835   struct sigaction sigAct;
3836   sigfillset(&(sigAct.sa_mask));
3837   sigAct.sa_handler = SIG_DFL;
3838   if (!set_installed) {
3839     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3840   } else {
3841     sigAct.sa_sigaction = signalHandler;
3842     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3843   }
3844   // Save flags, which are set by ours
3845   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3846   sigflags[sig] = sigAct.sa_flags;
3847 
3848   int ret = sigaction(sig, &sigAct, &oldAct);
3849   assert(ret == 0, "check");
3850 
3851   void* oldhand2  = oldAct.sa_sigaction
3852                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3853                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3854   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3855 }
3856 
3857 // install signal handlers for signals that HotSpot needs to
3858 // handle in order to support Java-level exception handling.
3859 
3860 void os::Linux::install_signal_handlers() {
3861   if (!signal_handlers_are_installed) {
3862     signal_handlers_are_installed = true;
3863 
3864     // signal-chaining
3865     typedef void (*signal_setting_t)();
3866     signal_setting_t begin_signal_setting = NULL;
3867     signal_setting_t end_signal_setting = NULL;
3868     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3869                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3870     if (begin_signal_setting != NULL) {
3871       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3872                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3873       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3874                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3875       libjsig_is_loaded = true;
3876       assert(UseSignalChaining, "should enable signal-chaining");
3877     }
3878     if (libjsig_is_loaded) {
3879       // Tell libjsig jvm is setting signal handlers
3880       (*begin_signal_setting)();
3881     }
3882 
3883     set_signal_handler(SIGSEGV, true);
3884     set_signal_handler(SIGPIPE, true);
3885     set_signal_handler(SIGBUS, true);
3886     set_signal_handler(SIGILL, true);
3887     set_signal_handler(SIGFPE, true);
3888     set_signal_handler(SIGXFSZ, true);
3889 
3890     if (libjsig_is_loaded) {
3891       // Tell libjsig jvm finishes setting signal handlers
3892       (*end_signal_setting)();
3893     }
3894 
3895     // We don't activate signal checker if libjsig is in place, we trust ourselves
3896     // and if UserSignalHandler is installed all bets are off.
3897     // Log that signal checking is off only if -verbose:jni is specified.
3898     if (CheckJNICalls) {
3899       if (libjsig_is_loaded) {
3900         if (PrintJNIResolving) {
3901           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3902         }
3903         check_signals = false;
3904       }
3905       if (AllowUserSignalHandlers) {
3906         if (PrintJNIResolving) {
3907           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3908         }
3909         check_signals = false;
3910       }
3911     }
3912   }
3913 }
3914 
3915 // This is the fastest way to get thread cpu time on Linux.
3916 // Returns cpu time (user+sys) for any thread, not only for current.
3917 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3918 // It might work on 2.6.10+ with a special kernel/glibc patch.
3919 // For reference, please, see IEEE Std 1003.1-2004:
3920 //   http://www.unix.org/single_unix_specification
3921 
3922 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3923   struct timespec tp;
3924   int rc = os::Linux::clock_gettime(clockid, &tp);
3925   assert(rc == 0, "clock_gettime is expected to return 0 code");
3926 
3927   return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3928 }
3929 
3930 /////
3931 // glibc on Linux platform uses non-documented flag
3932 // to indicate, that some special sort of signal
3933 // trampoline is used.
3934 // We will never set this flag, and we should
3935 // ignore this flag in our diagnostic
3936 #ifdef SIGNIFICANT_SIGNAL_MASK
3937 #undef SIGNIFICANT_SIGNAL_MASK
3938 #endif
3939 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3940 
3941 static const char* get_signal_handler_name(address handler,
3942                                            char* buf, int buflen) {
3943   int offset;
3944   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3945   if (found) {
3946     // skip directory names
3947     const char *p1, *p2;
3948     p1 = buf;
3949     size_t len = strlen(os::file_separator());
3950     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3951     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3952   } else {
3953     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3954   }
3955   return buf;
3956 }
3957 
3958 static void print_signal_handler(outputStream* st, int sig,
3959                                  char* buf, size_t buflen) {
3960   struct sigaction sa;
3961 
3962   sigaction(sig, NULL, &sa);
3963 
3964   // See comment for SIGNIFICANT_SIGNAL_MASK define
3965   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3966 
3967   st->print("%s: ", os::exception_name(sig, buf, buflen));
3968 
3969   address handler = (sa.sa_flags & SA_SIGINFO)
3970     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3971     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3972 
3973   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3974     st->print("SIG_DFL");
3975   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3976     st->print("SIG_IGN");
3977   } else {
3978     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3979   }
3980 
3981   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3982 
3983   address rh = VMError::get_resetted_sighandler(sig);
3984   // May be, handler was resetted by VMError?
3985   if(rh != NULL) {
3986     handler = rh;
3987     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3988   }
3989 
3990   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
3991 
3992   // Check: is it our handler?
3993   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3994      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3995     // It is our signal handler
3996     // check for flags, reset system-used one!
3997     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3998       st->print(
3999                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4000                 os::Linux::get_our_sigflags(sig));
4001     }
4002   }
4003   st->cr();
4004 }
4005 
4006 
4007 #define DO_SIGNAL_CHECK(sig) \
4008   if (!sigismember(&check_signal_done, sig)) \
4009     os::Linux::check_signal_handler(sig)
4010 
4011 // This method is a periodic task to check for misbehaving JNI applications
4012 // under CheckJNI, we can add any periodic checks here
4013 
4014 void os::run_periodic_checks() {
4015 
4016   if (check_signals == false) return;
4017 
4018   // SEGV and BUS if overridden could potentially prevent
4019   // generation of hs*.log in the event of a crash, debugging
4020   // such a case can be very challenging, so we absolutely
4021   // check the following for a good measure:
4022   DO_SIGNAL_CHECK(SIGSEGV);
4023   DO_SIGNAL_CHECK(SIGILL);
4024   DO_SIGNAL_CHECK(SIGFPE);
4025   DO_SIGNAL_CHECK(SIGBUS);
4026   DO_SIGNAL_CHECK(SIGPIPE);
4027   DO_SIGNAL_CHECK(SIGXFSZ);
4028 
4029 
4030   // ReduceSignalUsage allows the user to override these handlers
4031   // see comments at the very top and jvm_solaris.h
4032   if (!ReduceSignalUsage) {
4033     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4034     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4035     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4036     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4037   }
4038 
4039   DO_SIGNAL_CHECK(SR_signum);
4040   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4041 }
4042 
4043 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4044 
4045 static os_sigaction_t os_sigaction = NULL;
4046 
4047 void os::Linux::check_signal_handler(int sig) {
4048   char buf[O_BUFLEN];
4049   address jvmHandler = NULL;
4050 
4051 
4052   struct sigaction act;
4053   if (os_sigaction == NULL) {
4054     // only trust the default sigaction, in case it has been interposed
4055     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4056     if (os_sigaction == NULL) return;
4057   }
4058 
4059   os_sigaction(sig, (struct sigaction*)NULL, &act);
4060 
4061 
4062   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4063 
4064   address thisHandler = (act.sa_flags & SA_SIGINFO)
4065     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4066     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4067 
4068 
4069   switch(sig) {
4070   case SIGSEGV:
4071   case SIGBUS:
4072   case SIGFPE:
4073   case SIGPIPE:
4074   case SIGILL:
4075   case SIGXFSZ:
4076     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4077     break;
4078 
4079   case SHUTDOWN1_SIGNAL:
4080   case SHUTDOWN2_SIGNAL:
4081   case SHUTDOWN3_SIGNAL:
4082   case BREAK_SIGNAL:
4083     jvmHandler = (address)user_handler();
4084     break;
4085 
4086   case INTERRUPT_SIGNAL:
4087     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4088     break;
4089 
4090   default:
4091     if (sig == SR_signum) {
4092       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4093     } else {
4094       return;
4095     }
4096     break;
4097   }
4098 
4099   if (thisHandler != jvmHandler) {
4100     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4101     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4102     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4103     // No need to check this sig any longer
4104     sigaddset(&check_signal_done, sig);
4105   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4106     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4107     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4108     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4109     // No need to check this sig any longer
4110     sigaddset(&check_signal_done, sig);
4111   }
4112 
4113   // Dump all the signal
4114   if (sigismember(&check_signal_done, sig)) {
4115     print_signal_handlers(tty, buf, O_BUFLEN);
4116   }
4117 }
4118 
4119 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4120 
4121 extern bool signal_name(int signo, char* buf, size_t len);
4122 
4123 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4124   if (0 < exception_code && exception_code <= SIGRTMAX) {
4125     // signal
4126     if (!signal_name(exception_code, buf, size)) {
4127       jio_snprintf(buf, size, "SIG%d", exception_code);
4128     }
4129     return buf;
4130   } else {
4131     return NULL;
4132   }
4133 }
4134 
4135 // this is called _before_ the most of global arguments have been parsed
4136 void os::init(void) {
4137   char dummy;   /* used to get a guess on initial stack address */
4138 //  first_hrtime = gethrtime();
4139 
4140   // With LinuxThreads the JavaMain thread pid (primordial thread)
4141   // is different than the pid of the java launcher thread.
4142   // So, on Linux, the launcher thread pid is passed to the VM
4143   // via the sun.java.launcher.pid property.
4144   // Use this property instead of getpid() if it was correctly passed.
4145   // See bug 6351349.
4146   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4147 
4148   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4149 
4150   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4151 
4152   init_random(1234567);
4153 
4154   ThreadCritical::initialize();
4155 
4156   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4157   if (Linux::page_size() == -1) {
4158     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4159                   strerror(errno)));
4160   }
4161   init_page_sizes((size_t) Linux::page_size());
4162 
4163   Linux::initialize_system_info();
4164 
4165   // main_thread points to the aboriginal thread
4166   Linux::_main_thread = pthread_self();
4167 
4168   Linux::clock_init();
4169   initial_time_count = os::elapsed_counter();
4170   pthread_mutex_init(&dl_mutex, NULL);
4171 }
4172 
4173 // To install functions for atexit system call
4174 extern "C" {
4175   static void perfMemory_exit_helper() {
4176     perfMemory_exit();
4177   }
4178 }
4179 
4180 // this is called _after_ the global arguments have been parsed
4181 jint os::init_2(void)
4182 {
4183   Linux::fast_thread_clock_init();
4184 
4185   // Allocate a single page and mark it as readable for safepoint polling
4186   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4187   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4188 
4189   os::set_polling_page( polling_page );
4190 
4191 #ifndef PRODUCT
4192   if(Verbose && PrintMiscellaneous)
4193     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4194 #endif
4195 
4196   if (!UseMembar) {
4197     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4198     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4199     os::set_memory_serialize_page( mem_serialize_page );
4200 
4201 #ifndef PRODUCT
4202     if(Verbose && PrintMiscellaneous)
4203       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4204 #endif
4205   }
4206 
4207   os::large_page_init();
4208 
4209   // initialize suspend/resume support - must do this before signal_sets_init()
4210   if (SR_initialize() != 0) {
4211     perror("SR_initialize failed");
4212     return JNI_ERR;
4213   }
4214 
4215   Linux::signal_sets_init();
4216   Linux::install_signal_handlers();
4217 
4218   // Check minimum allowable stack size for thread creation and to initialize
4219   // the java system classes, including StackOverflowError - depends on page
4220   // size.  Add a page for compiler2 recursion in main thread.
4221   // Add in 2*BytesPerWord times page size to account for VM stack during
4222   // class initialization depending on 32 or 64 bit VM.
4223   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4224             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4225                     2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4226 
4227   size_t threadStackSizeInBytes = ThreadStackSize * K;
4228   if (threadStackSizeInBytes != 0 &&
4229       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4230         tty->print_cr("\nThe stack size specified is too small, "
4231                       "Specify at least %dk",
4232                       os::Linux::min_stack_allowed/ K);
4233         return JNI_ERR;
4234   }
4235 
4236   // Make the stack size a multiple of the page size so that
4237   // the yellow/red zones can be guarded.
4238   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4239         vm_page_size()));
4240 
4241   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4242 
4243   Linux::libpthread_init();
4244   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4245      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4246           Linux::glibc_version(), Linux::libpthread_version(),
4247           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4248   }
4249 
4250   if (UseNUMA) {
4251     if (!Linux::libnuma_init()) {
4252       UseNUMA = false;
4253     } else {
4254       if ((Linux::numa_max_node() < 1)) {
4255         // There's only one node(they start from 0), disable NUMA.
4256         UseNUMA = false;
4257       }
4258     }
4259     // With SHM large pages we cannot uncommit a page, so there's not way
4260     // we can make the adaptive lgrp chunk resizing work. If the user specified
4261     // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4262     // disable adaptive resizing.
4263     if (UseNUMA && UseLargePages && UseSHM) {
4264       if (!FLAG_IS_DEFAULT(UseNUMA)) {
4265         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4266           UseLargePages = false;
4267         } else {
4268           warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4269           UseAdaptiveSizePolicy = false;
4270           UseAdaptiveNUMAChunkSizing = false;
4271         }
4272       } else {
4273         UseNUMA = false;
4274       }
4275     }
4276     if (!UseNUMA && ForceNUMA) {
4277       UseNUMA = true;
4278     }
4279   }
4280 
4281   if (MaxFDLimit) {
4282     // set the number of file descriptors to max. print out error
4283     // if getrlimit/setrlimit fails but continue regardless.
4284     struct rlimit nbr_files;
4285     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4286     if (status != 0) {
4287       if (PrintMiscellaneous && (Verbose || WizardMode))
4288         perror("os::init_2 getrlimit failed");
4289     } else {
4290       nbr_files.rlim_cur = nbr_files.rlim_max;
4291       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4292       if (status != 0) {
4293         if (PrintMiscellaneous && (Verbose || WizardMode))
4294           perror("os::init_2 setrlimit failed");
4295       }
4296     }
4297   }
4298 
4299   // Initialize lock used to serialize thread creation (see os::create_thread)
4300   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4301 
4302   // at-exit methods are called in the reverse order of their registration.
4303   // atexit functions are called on return from main or as a result of a
4304   // call to exit(3C). There can be only 32 of these functions registered
4305   // and atexit() does not set errno.
4306 
4307   if (PerfAllowAtExitRegistration) {
4308     // only register atexit functions if PerfAllowAtExitRegistration is set.
4309     // atexit functions can be delayed until process exit time, which
4310     // can be problematic for embedded VM situations. Embedded VMs should
4311     // call DestroyJavaVM() to assure that VM resources are released.
4312 
4313     // note: perfMemory_exit_helper atexit function may be removed in
4314     // the future if the appropriate cleanup code can be added to the
4315     // VM_Exit VMOperation's doit method.
4316     if (atexit(perfMemory_exit_helper) != 0) {
4317       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4318     }
4319   }
4320 
4321   // initialize thread priority policy
4322   prio_init();
4323 
4324   return JNI_OK;
4325 }
4326 
4327 // this is called at the end of vm_initialization
4328 void os::init_3(void)
4329 {
4330 #ifdef JAVASE_EMBEDDED
4331   // Start the MemNotifyThread
4332   if (LowMemoryProtection) {
4333     MemNotifyThread::start();
4334   }
4335   return;
4336 #endif
4337 }
4338 
4339 // Mark the polling page as unreadable
4340 void os::make_polling_page_unreadable(void) {
4341   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4342     fatal("Could not disable polling page");
4343 };
4344 
4345 // Mark the polling page as readable
4346 void os::make_polling_page_readable(void) {
4347   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4348     fatal("Could not enable polling page");
4349   }
4350 };
4351 
4352 int os::active_processor_count() {
4353   // Linux doesn't yet have a (official) notion of processor sets,
4354   // so just return the number of online processors.
4355   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4356   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4357   return online_cpus;
4358 }
4359 
4360 bool os::distribute_processes(uint length, uint* distribution) {
4361   // Not yet implemented.
4362   return false;
4363 }
4364 
4365 bool os::bind_to_processor(uint processor_id) {
4366   // Not yet implemented.
4367   return false;
4368 }
4369 
4370 ///
4371 
4372 // Suspends the target using the signal mechanism and then grabs the PC before
4373 // resuming the target. Used by the flat-profiler only
4374 ExtendedPC os::get_thread_pc(Thread* thread) {
4375   // Make sure that it is called by the watcher for the VMThread
4376   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4377   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4378 
4379   ExtendedPC epc;
4380 
4381   OSThread* osthread = thread->osthread();
4382   if (do_suspend(osthread)) {
4383     if (osthread->ucontext() != NULL) {
4384       epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4385     } else {
4386       // NULL context is unexpected, double-check this is the VMThread
4387       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4388     }
4389     do_resume(osthread);
4390   }
4391   // failure means pthread_kill failed for some reason - arguably this is
4392   // a fatal problem, but such problems are ignored elsewhere
4393 
4394   return epc;
4395 }
4396 
4397 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4398 {
4399    if (is_NPTL()) {
4400       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4401    } else {
4402 #ifndef IA64
4403       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4404       // word back to default 64bit precision if condvar is signaled. Java
4405       // wants 53bit precision.  Save and restore current value.
4406       int fpu = get_fpu_control_word();
4407 #endif // IA64
4408       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4409 #ifndef IA64
4410       set_fpu_control_word(fpu);
4411 #endif // IA64
4412       return status;
4413    }
4414 }
4415 
4416 ////////////////////////////////////////////////////////////////////////////////
4417 // debug support
4418 
4419 static address same_page(address x, address y) {
4420   int page_bits = -os::vm_page_size();
4421   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4422     return x;
4423   else if (x > y)
4424     return (address)(intptr_t(y) | ~page_bits) + 1;
4425   else
4426     return (address)(intptr_t(y) & page_bits);
4427 }
4428 
4429 bool os::find(address addr, outputStream* st) {
4430   Dl_info dlinfo;
4431   memset(&dlinfo, 0, sizeof(dlinfo));
4432   if (dladdr(addr, &dlinfo)) {
4433     st->print(PTR_FORMAT ": ", addr);
4434     if (dlinfo.dli_sname != NULL) {
4435       st->print("%s+%#x", dlinfo.dli_sname,
4436                  addr - (intptr_t)dlinfo.dli_saddr);
4437     } else if (dlinfo.dli_fname) {
4438       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4439     } else {
4440       st->print("<absolute address>");
4441     }
4442     if (dlinfo.dli_fname) {
4443       st->print(" in %s", dlinfo.dli_fname);
4444     }
4445     if (dlinfo.dli_fbase) {
4446       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4447     }
4448     st->cr();
4449 
4450     if (Verbose) {
4451       // decode some bytes around the PC
4452       address begin = same_page(addr-40, addr);
4453       address end   = same_page(addr+40, addr);
4454       address       lowest = (address) dlinfo.dli_sname;
4455       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4456       if (begin < lowest)  begin = lowest;
4457       Dl_info dlinfo2;
4458       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4459           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4460         end = (address) dlinfo2.dli_saddr;
4461       Disassembler::decode(begin, end, st);
4462     }
4463     return true;
4464   }
4465   return false;
4466 }
4467 
4468 ////////////////////////////////////////////////////////////////////////////////
4469 // misc
4470 
4471 // This does not do anything on Linux. This is basically a hook for being
4472 // able to use structured exception handling (thread-local exception filters)
4473 // on, e.g., Win32.
4474 void
4475 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4476                          JavaCallArguments* args, Thread* thread) {
4477   f(value, method, args, thread);
4478 }
4479 
4480 void os::print_statistics() {
4481 }
4482 
4483 int os::message_box(const char* title, const char* message) {
4484   int i;
4485   fdStream err(defaultStream::error_fd());
4486   for (i = 0; i < 78; i++) err.print_raw("=");
4487   err.cr();
4488   err.print_raw_cr(title);
4489   for (i = 0; i < 78; i++) err.print_raw("-");
4490   err.cr();
4491   err.print_raw_cr(message);
4492   for (i = 0; i < 78; i++) err.print_raw("=");
4493   err.cr();
4494 
4495   char buf[16];
4496   // Prevent process from exiting upon "read error" without consuming all CPU
4497   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4498 
4499   return buf[0] == 'y' || buf[0] == 'Y';
4500 }
4501 
4502 int os::stat(const char *path, struct stat *sbuf) {
4503   char pathbuf[MAX_PATH];
4504   if (strlen(path) > MAX_PATH - 1) {
4505     errno = ENAMETOOLONG;
4506     return -1;
4507   }
4508   os::native_path(strcpy(pathbuf, path));
4509   return ::stat(pathbuf, sbuf);
4510 }
4511 
4512 bool os::check_heap(bool force) {
4513   return true;
4514 }
4515 
4516 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4517   return ::vsnprintf(buf, count, format, args);
4518 }
4519 
4520 // Is a (classpath) directory empty?
4521 bool os::dir_is_empty(const char* path) {
4522   DIR *dir = NULL;
4523   struct dirent *ptr;
4524 
4525   dir = opendir(path);
4526   if (dir == NULL) return true;
4527 
4528   /* Scan the directory */
4529   bool result = true;
4530   char buf[sizeof(struct dirent) + MAX_PATH];
4531   while (result && (ptr = ::readdir(dir)) != NULL) {
4532     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4533       result = false;
4534     }
4535   }
4536   closedir(dir);
4537   return result;
4538 }
4539 
4540 // This code originates from JDK's sysOpen and open64_w
4541 // from src/solaris/hpi/src/system_md.c
4542 
4543 #ifndef O_DELETE
4544 #define O_DELETE 0x10000
4545 #endif
4546 
4547 // Open a file. Unlink the file immediately after open returns
4548 // if the specified oflag has the O_DELETE flag set.
4549 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4550 
4551 int os::open(const char *path, int oflag, int mode) {
4552 
4553   if (strlen(path) > MAX_PATH - 1) {
4554     errno = ENAMETOOLONG;
4555     return -1;
4556   }
4557   int fd;
4558   int o_delete = (oflag & O_DELETE);
4559   oflag = oflag & ~O_DELETE;
4560 
4561   fd = ::open64(path, oflag, mode);
4562   if (fd == -1) return -1;
4563 
4564   //If the open succeeded, the file might still be a directory
4565   {
4566     struct stat64 buf64;
4567     int ret = ::fstat64(fd, &buf64);
4568     int st_mode = buf64.st_mode;
4569 
4570     if (ret != -1) {
4571       if ((st_mode & S_IFMT) == S_IFDIR) {
4572         errno = EISDIR;
4573         ::close(fd);
4574         return -1;
4575       }
4576     } else {
4577       ::close(fd);
4578       return -1;
4579     }
4580   }
4581 
4582     /*
4583      * All file descriptors that are opened in the JVM and not
4584      * specifically destined for a subprocess should have the
4585      * close-on-exec flag set.  If we don't set it, then careless 3rd
4586      * party native code might fork and exec without closing all
4587      * appropriate file descriptors (e.g. as we do in closeDescriptors in
4588      * UNIXProcess.c), and this in turn might:
4589      *
4590      * - cause end-of-file to fail to be detected on some file
4591      *   descriptors, resulting in mysterious hangs, or
4592      *
4593      * - might cause an fopen in the subprocess to fail on a system
4594      *   suffering from bug 1085341.
4595      *
4596      * (Yes, the default setting of the close-on-exec flag is a Unix
4597      * design flaw)
4598      *
4599      * See:
4600      * 1085341: 32-bit stdio routines should support file descriptors >255
4601      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4602      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4603      */
4604 #ifdef FD_CLOEXEC
4605     {
4606         int flags = ::fcntl(fd, F_GETFD);
4607         if (flags != -1)
4608             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4609     }
4610 #endif
4611 
4612   if (o_delete != 0) {
4613     ::unlink(path);
4614   }
4615   return fd;
4616 }
4617 
4618 
4619 // create binary file, rewriting existing file if required
4620 int os::create_binary_file(const char* path, bool rewrite_existing) {
4621   int oflags = O_WRONLY | O_CREAT;
4622   if (!rewrite_existing) {
4623     oflags |= O_EXCL;
4624   }
4625   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4626 }
4627 
4628 // return current position of file pointer
4629 jlong os::current_file_offset(int fd) {
4630   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4631 }
4632 
4633 // move file pointer to the specified offset
4634 jlong os::seek_to_file_offset(int fd, jlong offset) {
4635   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4636 }
4637 
4638 // This code originates from JDK's sysAvailable
4639 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4640 
4641 int os::available(int fd, jlong *bytes) {
4642   jlong cur, end;
4643   int mode;
4644   struct stat64 buf64;
4645 
4646   if (::fstat64(fd, &buf64) >= 0) {
4647     mode = buf64.st_mode;
4648     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4649       /*
4650       * XXX: is the following call interruptible? If so, this might
4651       * need to go through the INTERRUPT_IO() wrapper as for other
4652       * blocking, interruptible calls in this file.
4653       */
4654       int n;
4655       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4656         *bytes = n;
4657         return 1;
4658       }
4659     }
4660   }
4661   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4662     return 0;
4663   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4664     return 0;
4665   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4666     return 0;
4667   }
4668   *bytes = end - cur;
4669   return 1;
4670 }
4671 
4672 int os::socket_available(int fd, jint *pbytes) {
4673   // Linux doc says EINTR not returned, unlike Solaris
4674   int ret = ::ioctl(fd, FIONREAD, pbytes);
4675 
4676   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4677   // is expected to return 0 on failure and 1 on success to the jdk.
4678   return (ret < 0) ? 0 : 1;
4679 }
4680 
4681 // Map a block of memory.
4682 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4683                      char *addr, size_t bytes, bool read_only,
4684                      bool allow_exec) {
4685   int prot;
4686   int flags;
4687 
4688   if (read_only) {
4689     prot = PROT_READ;
4690     flags = MAP_SHARED;
4691   } else {
4692     prot = PROT_READ | PROT_WRITE;
4693     flags = MAP_PRIVATE;
4694   }
4695 
4696   if (allow_exec) {
4697     prot |= PROT_EXEC;
4698   }
4699 
4700   if (addr != NULL) {
4701     flags |= MAP_FIXED;
4702   }
4703 
4704   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4705                                      fd, file_offset);
4706   if (mapped_address == MAP_FAILED) {
4707     return NULL;
4708   }
4709   return mapped_address;
4710 }
4711 
4712 
4713 // Remap a block of memory.
4714 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4715                        char *addr, size_t bytes, bool read_only,
4716                        bool allow_exec) {
4717   // same as map_memory() on this OS
4718   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4719                         allow_exec);
4720 }
4721 
4722 
4723 // Unmap a block of memory.
4724 bool os::unmap_memory(char* addr, size_t bytes) {
4725   return munmap(addr, bytes) == 0;
4726 }
4727 
4728 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4729 
4730 static clockid_t thread_cpu_clockid(Thread* thread) {
4731   pthread_t tid = thread->osthread()->pthread_id();
4732   clockid_t clockid;
4733 
4734   // Get thread clockid
4735   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4736   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4737   return clockid;
4738 }
4739 
4740 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4741 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4742 // of a thread.
4743 //
4744 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4745 // the fast estimate available on the platform.
4746 
4747 jlong os::current_thread_cpu_time() {
4748   if (os::Linux::supports_fast_thread_cpu_time()) {
4749     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4750   } else {
4751     // return user + sys since the cost is the same
4752     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4753   }
4754 }
4755 
4756 jlong os::thread_cpu_time(Thread* thread) {
4757   // consistent with what current_thread_cpu_time() returns
4758   if (os::Linux::supports_fast_thread_cpu_time()) {
4759     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4760   } else {
4761     return slow_thread_cpu_time(thread, true /* user + sys */);
4762   }
4763 }
4764 
4765 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4766   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4767     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4768   } else {
4769     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4770   }
4771 }
4772 
4773 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4774   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4775     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4776   } else {
4777     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4778   }
4779 }
4780 
4781 //
4782 //  -1 on error.
4783 //
4784 
4785 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4786   static bool proc_pid_cpu_avail = true;
4787   static bool proc_task_unchecked = true;
4788   static const char *proc_stat_path = "/proc/%d/stat";
4789   pid_t  tid = thread->osthread()->thread_id();
4790   int i;
4791   char *s;
4792   char stat[2048];
4793   int statlen;
4794   char proc_name[64];
4795   int count;
4796   long sys_time, user_time;
4797   char string[64];
4798   char cdummy;
4799   int idummy;
4800   long ldummy;
4801   FILE *fp;
4802 
4803   // We first try accessing /proc/<pid>/cpu since this is faster to
4804   // process.  If this file is not present (linux kernels 2.5 and above)
4805   // then we open /proc/<pid>/stat.
4806   if ( proc_pid_cpu_avail ) {
4807     sprintf(proc_name, "/proc/%d/cpu", tid);
4808     fp =  fopen(proc_name, "r");
4809     if ( fp != NULL ) {
4810       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4811       fclose(fp);
4812       if ( count != 3 ) return -1;
4813 
4814       if (user_sys_cpu_time) {
4815         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4816       } else {
4817         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4818       }
4819     }
4820     else proc_pid_cpu_avail = false;
4821   }
4822 
4823   // The /proc/<tid>/stat aggregates per-process usage on
4824   // new Linux kernels 2.6+ where NPTL is supported.
4825   // The /proc/self/task/<tid>/stat still has the per-thread usage.
4826   // See bug 6328462.
4827   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4828   // and possibly in some other cases, so we check its availability.
4829   if (proc_task_unchecked && os::Linux::is_NPTL()) {
4830     // This is executed only once
4831     proc_task_unchecked = false;
4832     fp = fopen("/proc/self/task", "r");
4833     if (fp != NULL) {
4834       proc_stat_path = "/proc/self/task/%d/stat";
4835       fclose(fp);
4836     }
4837   }
4838 
4839   sprintf(proc_name, proc_stat_path, tid);
4840   fp = fopen(proc_name, "r");
4841   if ( fp == NULL ) return -1;
4842   statlen = fread(stat, 1, 2047, fp);
4843   stat[statlen] = '\0';
4844   fclose(fp);
4845 
4846   // Skip pid and the command string. Note that we could be dealing with
4847   // weird command names, e.g. user could decide to rename java launcher
4848   // to "java 1.4.2 :)", then the stat file would look like
4849   //                1234 (java 1.4.2 :)) R ... ...
4850   // We don't really need to know the command string, just find the last
4851   // occurrence of ")" and then start parsing from there. See bug 4726580.
4852   s = strrchr(stat, ')');
4853   i = 0;
4854   if (s == NULL ) return -1;
4855 
4856   // Skip blank chars
4857   do s++; while (isspace(*s));
4858 
4859   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4860                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4861                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4862                  &user_time, &sys_time);
4863   if ( count != 13 ) return -1;
4864   if (user_sys_cpu_time) {
4865     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4866   } else {
4867     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4868   }
4869 }
4870 
4871 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4872   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4873   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4874   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4875   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4876 }
4877 
4878 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4879   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4880   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4881   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4882   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4883 }
4884 
4885 bool os::is_thread_cpu_time_supported() {
4886   return true;
4887 }
4888 
4889 // System loadavg support.  Returns -1 if load average cannot be obtained.
4890 // Linux doesn't yet have a (official) notion of processor sets,
4891 // so just return the system wide load average.
4892 int os::loadavg(double loadavg[], int nelem) {
4893   return ::getloadavg(loadavg, nelem);
4894 }
4895 
4896 void os::pause() {
4897   char filename[MAX_PATH];
4898   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4899     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4900   } else {
4901     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4902   }
4903 
4904   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4905   if (fd != -1) {
4906     struct stat buf;
4907     ::close(fd);
4908     while (::stat(filename, &buf) == 0) {
4909       (void)::poll(NULL, 0, 100);
4910     }
4911   } else {
4912     jio_fprintf(stderr,
4913       "Could not open pause file '%s', continuing immediately.\n", filename);
4914   }
4915 }
4916 
4917 
4918 // Refer to the comments in os_solaris.cpp park-unpark.
4919 //
4920 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4921 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4922 // For specifics regarding the bug see GLIBC BUGID 261237 :
4923 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4924 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4925 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4926 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4927 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4928 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4929 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4930 // of libpthread avoids the problem, but isn't practical.
4931 //
4932 // Possible remedies:
4933 //
4934 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4935 //      This is palliative and probabilistic, however.  If the thread is preempted
4936 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4937 //      than the minimum period may have passed, and the abstime may be stale (in the
4938 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
4939 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4940 //
4941 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4942 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
4943 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4944 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
4945 //      thread.
4946 //
4947 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4948 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
4949 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4950 //      This also works well.  In fact it avoids kernel-level scalability impediments
4951 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4952 //      timers in a graceful fashion.
4953 //
4954 // 4.   When the abstime value is in the past it appears that control returns
4955 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4956 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
4957 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
4958 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
4959 //      It may be possible to avoid reinitialization by checking the return
4960 //      value from pthread_cond_timedwait().  In addition to reinitializing the
4961 //      condvar we must establish the invariant that cond_signal() is only called
4962 //      within critical sections protected by the adjunct mutex.  This prevents
4963 //      cond_signal() from "seeing" a condvar that's in the midst of being
4964 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
4965 //      desirable signal-after-unlock optimization that avoids futile context switching.
4966 //
4967 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
4968 //      structure when a condvar is used or initialized.  cond_destroy()  would
4969 //      release the helper structure.  Our reinitialize-after-timedwait fix
4970 //      put excessive stress on malloc/free and locks protecting the c-heap.
4971 //
4972 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
4973 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4974 // and only enabling the work-around for vulnerable environments.
4975 
4976 // utility to compute the abstime argument to timedwait:
4977 // millis is the relative timeout time
4978 // abstime will be the absolute timeout time
4979 // TODO: replace compute_abstime() with unpackTime()
4980 
4981 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4982   if (millis < 0)  millis = 0;
4983   struct timeval now;
4984   int status = gettimeofday(&now, NULL);
4985   assert(status == 0, "gettimeofday");
4986   jlong seconds = millis / 1000;
4987   millis %= 1000;
4988   if (seconds > 50000000) { // see man cond_timedwait(3T)
4989     seconds = 50000000;
4990   }
4991   abstime->tv_sec = now.tv_sec  + seconds;
4992   long       usec = now.tv_usec + millis * 1000;
4993   if (usec >= 1000000) {
4994     abstime->tv_sec += 1;
4995     usec -= 1000000;
4996   }
4997   abstime->tv_nsec = usec * 1000;
4998   return abstime;
4999 }
5000 
5001 
5002 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5003 // Conceptually TryPark() should be equivalent to park(0).
5004 
5005 int os::PlatformEvent::TryPark() {
5006   for (;;) {
5007     const int v = _Event ;
5008     guarantee ((v == 0) || (v == 1), "invariant") ;
5009     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5010   }
5011 }
5012 
5013 void os::PlatformEvent::park() {       // AKA "down()"
5014   // Invariant: Only the thread associated with the Event/PlatformEvent
5015   // may call park().
5016   // TODO: assert that _Assoc != NULL or _Assoc == Self
5017   int v ;
5018   for (;;) {
5019       v = _Event ;
5020       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5021   }
5022   guarantee (v >= 0, "invariant") ;
5023   if (v == 0) {
5024      // Do this the hard way by blocking ...
5025      int status = pthread_mutex_lock(_mutex);
5026      assert_status(status == 0, status, "mutex_lock");
5027      guarantee (_nParked == 0, "invariant") ;
5028      ++ _nParked ;
5029      while (_Event < 0) {
5030         status = pthread_cond_wait(_cond, _mutex);
5031         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5032         // Treat this the same as if the wait was interrupted
5033         if (status == ETIME) { status = EINTR; }
5034         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5035      }
5036      -- _nParked ;
5037 
5038     // In theory we could move the ST of 0 into _Event past the unlock(),
5039     // but then we'd need a MEMBAR after the ST.
5040     _Event = 0 ;
5041      status = pthread_mutex_unlock(_mutex);
5042      assert_status(status == 0, status, "mutex_unlock");
5043   }
5044   guarantee (_Event >= 0, "invariant") ;
5045 }
5046 
5047 int os::PlatformEvent::park(jlong millis) {
5048   guarantee (_nParked == 0, "invariant") ;
5049 
5050   int v ;
5051   for (;;) {
5052       v = _Event ;
5053       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5054   }
5055   guarantee (v >= 0, "invariant") ;
5056   if (v != 0) return OS_OK ;
5057 
5058   // We do this the hard way, by blocking the thread.
5059   // Consider enforcing a minimum timeout value.
5060   struct timespec abst;
5061   compute_abstime(&abst, millis);
5062 
5063   int ret = OS_TIMEOUT;
5064   int status = pthread_mutex_lock(_mutex);
5065   assert_status(status == 0, status, "mutex_lock");
5066   guarantee (_nParked == 0, "invariant") ;
5067   ++_nParked ;
5068 
5069   // Object.wait(timo) will return because of
5070   // (a) notification
5071   // (b) timeout
5072   // (c) thread.interrupt
5073   //
5074   // Thread.interrupt and object.notify{All} both call Event::set.
5075   // That is, we treat thread.interrupt as a special case of notification.
5076   // The underlying Solaris implementation, cond_timedwait, admits
5077   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5078   // JVM from making those visible to Java code.  As such, we must
5079   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5080   //
5081   // TODO: properly differentiate simultaneous notify+interrupt.
5082   // In that case, we should propagate the notify to another waiter.
5083 
5084   while (_Event < 0) {
5085     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5086     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5087       pthread_cond_destroy (_cond);
5088       pthread_cond_init (_cond, NULL) ;
5089     }
5090     assert_status(status == 0 || status == EINTR ||
5091                   status == ETIME || status == ETIMEDOUT,
5092                   status, "cond_timedwait");
5093     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5094     if (status == ETIME || status == ETIMEDOUT) break ;
5095     // We consume and ignore EINTR and spurious wakeups.
5096   }
5097   --_nParked ;
5098   if (_Event >= 0) {
5099      ret = OS_OK;
5100   }
5101   _Event = 0 ;
5102   status = pthread_mutex_unlock(_mutex);
5103   assert_status(status == 0, status, "mutex_unlock");
5104   assert (_nParked == 0, "invariant") ;
5105   return ret;
5106 }
5107 
5108 void os::PlatformEvent::unpark() {
5109   int v, AnyWaiters ;
5110   for (;;) {
5111       v = _Event ;
5112       if (v > 0) {
5113          // The LD of _Event could have reordered or be satisfied
5114          // by a read-aside from this processor's write buffer.
5115          // To avoid problems execute a barrier and then
5116          // ratify the value.
5117          OrderAccess::fence() ;
5118          if (_Event == v) return ;
5119          continue ;
5120       }
5121       if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5122   }
5123   if (v < 0) {
5124      // Wait for the thread associated with the event to vacate
5125      int status = pthread_mutex_lock(_mutex);
5126      assert_status(status == 0, status, "mutex_lock");
5127      AnyWaiters = _nParked ;
5128      assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5129      if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5130         AnyWaiters = 0 ;
5131         pthread_cond_signal (_cond);
5132      }
5133      status = pthread_mutex_unlock(_mutex);
5134      assert_status(status == 0, status, "mutex_unlock");
5135      if (AnyWaiters != 0) {
5136         status = pthread_cond_signal(_cond);
5137         assert_status(status == 0, status, "cond_signal");
5138      }
5139   }
5140 
5141   // Note that we signal() _after dropping the lock for "immortal" Events.
5142   // This is safe and avoids a common class of  futile wakeups.  In rare
5143   // circumstances this can cause a thread to return prematurely from
5144   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5145   // simply re-test the condition and re-park itself.
5146 }
5147 
5148 
5149 // JSR166
5150 // -------------------------------------------------------
5151 
5152 /*
5153  * The solaris and linux implementations of park/unpark are fairly
5154  * conservative for now, but can be improved. They currently use a
5155  * mutex/condvar pair, plus a a count.
5156  * Park decrements count if > 0, else does a condvar wait.  Unpark
5157  * sets count to 1 and signals condvar.  Only one thread ever waits
5158  * on the condvar. Contention seen when trying to park implies that someone
5159  * is unparking you, so don't wait. And spurious returns are fine, so there
5160  * is no need to track notifications.
5161  */
5162 
5163 
5164 #define NANOSECS_PER_SEC 1000000000
5165 #define NANOSECS_PER_MILLISEC 1000000
5166 #define MAX_SECS 100000000
5167 /*
5168  * This code is common to linux and solaris and will be moved to a
5169  * common place in dolphin.
5170  *
5171  * The passed in time value is either a relative time in nanoseconds
5172  * or an absolute time in milliseconds. Either way it has to be unpacked
5173  * into suitable seconds and nanoseconds components and stored in the
5174  * given timespec structure.
5175  * Given time is a 64-bit value and the time_t used in the timespec is only
5176  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5177  * overflow if times way in the future are given. Further on Solaris versions
5178  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5179  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5180  * As it will be 28 years before "now + 100000000" will overflow we can
5181  * ignore overflow and just impose a hard-limit on seconds using the value
5182  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5183  * years from "now".
5184  */
5185 
5186 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5187   assert (time > 0, "convertTime");
5188 
5189   struct timeval now;
5190   int status = gettimeofday(&now, NULL);
5191   assert(status == 0, "gettimeofday");
5192 
5193   time_t max_secs = now.tv_sec + MAX_SECS;
5194 
5195   if (isAbsolute) {
5196     jlong secs = time / 1000;
5197     if (secs > max_secs) {
5198       absTime->tv_sec = max_secs;
5199     }
5200     else {
5201       absTime->tv_sec = secs;
5202     }
5203     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5204   }
5205   else {
5206     jlong secs = time / NANOSECS_PER_SEC;
5207     if (secs >= MAX_SECS) {
5208       absTime->tv_sec = max_secs;
5209       absTime->tv_nsec = 0;
5210     }
5211     else {
5212       absTime->tv_sec = now.tv_sec + secs;
5213       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5214       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5215         absTime->tv_nsec -= NANOSECS_PER_SEC;
5216         ++absTime->tv_sec; // note: this must be <= max_secs
5217       }
5218     }
5219   }
5220   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5221   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5222   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5223   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5224 }
5225 
5226 void Parker::park(bool isAbsolute, jlong time) {
5227   // Optional fast-path check:
5228   // Return immediately if a permit is available.
5229   if (_counter > 0) {
5230       _counter = 0 ;
5231       OrderAccess::fence();
5232       return ;
5233   }
5234 
5235   Thread* thread = Thread::current();
5236   assert(thread->is_Java_thread(), "Must be JavaThread");
5237   JavaThread *jt = (JavaThread *)thread;
5238 
5239   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5240   // Check interrupt before trying to wait
5241   if (Thread::is_interrupted(thread, false)) {
5242     return;
5243   }
5244 
5245   // Next, demultiplex/decode time arguments
5246   timespec absTime;
5247   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5248     return;
5249   }
5250   if (time > 0) {
5251     unpackTime(&absTime, isAbsolute, time);
5252   }
5253 
5254 
5255   // Enter safepoint region
5256   // Beware of deadlocks such as 6317397.
5257   // The per-thread Parker:: mutex is a classic leaf-lock.
5258   // In particular a thread must never block on the Threads_lock while
5259   // holding the Parker:: mutex.  If safepoints are pending both the
5260   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5261   ThreadBlockInVM tbivm(jt);
5262 
5263   // Don't wait if cannot get lock since interference arises from
5264   // unblocking.  Also. check interrupt before trying wait
5265   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5266     return;
5267   }
5268 
5269   int status ;
5270   if (_counter > 0)  { // no wait needed
5271     _counter = 0;
5272     status = pthread_mutex_unlock(_mutex);
5273     assert (status == 0, "invariant") ;
5274     OrderAccess::fence();
5275     return;
5276   }
5277 
5278 #ifdef ASSERT
5279   // Don't catch signals while blocked; let the running threads have the signals.
5280   // (This allows a debugger to break into the running thread.)
5281   sigset_t oldsigs;
5282   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5283   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5284 #endif
5285 
5286   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5287   jt->set_suspend_equivalent();
5288   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5289 
5290   if (time == 0) {
5291     status = pthread_cond_wait (_cond, _mutex) ;
5292   } else {
5293     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5294     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5295       pthread_cond_destroy (_cond) ;
5296       pthread_cond_init    (_cond, NULL);
5297     }
5298   }
5299   assert_status(status == 0 || status == EINTR ||
5300                 status == ETIME || status == ETIMEDOUT,
5301                 status, "cond_timedwait");
5302 
5303 #ifdef ASSERT
5304   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5305 #endif
5306 
5307   _counter = 0 ;
5308   status = pthread_mutex_unlock(_mutex) ;
5309   assert_status(status == 0, status, "invariant") ;
5310   // If externally suspended while waiting, re-suspend
5311   if (jt->handle_special_suspend_equivalent_condition()) {
5312     jt->java_suspend_self();
5313   }
5314 
5315   OrderAccess::fence();
5316 }
5317 
5318 void Parker::unpark() {
5319   int s, status ;
5320   status = pthread_mutex_lock(_mutex);
5321   assert (status == 0, "invariant") ;
5322   s = _counter;
5323   _counter = 1;
5324   if (s < 1) {
5325      if (WorkAroundNPTLTimedWaitHang) {
5326         status = pthread_cond_signal (_cond) ;
5327         assert (status == 0, "invariant") ;
5328         status = pthread_mutex_unlock(_mutex);
5329         assert (status == 0, "invariant") ;
5330      } else {
5331         status = pthread_mutex_unlock(_mutex);
5332         assert (status == 0, "invariant") ;
5333         status = pthread_cond_signal (_cond) ;
5334         assert (status == 0, "invariant") ;
5335      }
5336   } else {
5337     pthread_mutex_unlock(_mutex);
5338     assert (status == 0, "invariant") ;
5339   }
5340 }
5341 
5342 
5343 extern char** environ;
5344 
5345 #ifndef __NR_fork
5346 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5347 #endif
5348 
5349 #ifndef __NR_execve
5350 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5351 #endif
5352 
5353 // Run the specified command in a separate process. Return its exit value,
5354 // or -1 on failure (e.g. can't fork a new process).
5355 // Unlike system(), this function can be called from signal handler. It
5356 // doesn't block SIGINT et al.
5357 int os::fork_and_exec(char* cmd) {
5358   const char * argv[4] = {"sh", "-c", cmd, NULL};
5359 
5360   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5361   // pthread_atfork handlers and reset pthread library. All we need is a
5362   // separate process to execve. Make a direct syscall to fork process.
5363   // On IA64 there's no fork syscall, we have to use fork() and hope for
5364   // the best...
5365   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5366               IA64_ONLY(fork();)
5367 
5368   if (pid < 0) {
5369     // fork failed
5370     return -1;
5371 
5372   } else if (pid == 0) {
5373     // child process
5374 
5375     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5376     // first to kill every thread on the thread list. Because this list is
5377     // not reset by fork() (see notes above), execve() will instead kill
5378     // every thread in the parent process. We know this is the only thread
5379     // in the new process, so make a system call directly.
5380     // IA64 should use normal execve() from glibc to match the glibc fork()
5381     // above.
5382     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5383     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5384 
5385     // execve failed
5386     _exit(-1);
5387 
5388   } else  {
5389     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5390     // care about the actual exit code, for now.
5391 
5392     int status;
5393 
5394     // Wait for the child process to exit.  This returns immediately if
5395     // the child has already exited. */
5396     while (waitpid(pid, &status, 0) < 0) {
5397         switch (errno) {
5398         case ECHILD: return 0;
5399         case EINTR: break;
5400         default: return -1;
5401         }
5402     }
5403 
5404     if (WIFEXITED(status)) {
5405        // The child exited normally; get its exit code.
5406        return WEXITSTATUS(status);
5407     } else if (WIFSIGNALED(status)) {
5408        // The child exited because of a signal
5409        // The best value to return is 0x80 + signal number,
5410        // because that is what all Unix shells do, and because
5411        // it allows callers to distinguish between process exit and
5412        // process death by signal.
5413        return 0x80 + WTERMSIG(status);
5414     } else {
5415        // Unknown exit code; pass it through
5416        return status;
5417     }
5418   }
5419 }
5420 
5421 // is_headless_jre()
5422 //
5423 // Test for the existence of libmawt in motif21 or xawt directories
5424 // in order to report if we are running in a headless jre
5425 //
5426 bool os::is_headless_jre() {
5427     struct stat statbuf;
5428     char buf[MAXPATHLEN];
5429     char libmawtpath[MAXPATHLEN];
5430     const char *xawtstr  = "/xawt/libmawt.so";
5431     const char *motifstr = "/motif21/libmawt.so";
5432     char *p;
5433 
5434     // Get path to libjvm.so
5435     os::jvm_path(buf, sizeof(buf));
5436 
5437     // Get rid of libjvm.so
5438     p = strrchr(buf, '/');
5439     if (p == NULL) return false;
5440     else *p = '\0';
5441 
5442     // Get rid of client or server
5443     p = strrchr(buf, '/');
5444     if (p == NULL) return false;
5445     else *p = '\0';
5446 
5447     // check xawt/libmawt.so
5448     strcpy(libmawtpath, buf);
5449     strcat(libmawtpath, xawtstr);
5450     if (::stat(libmawtpath, &statbuf) == 0) return false;
5451 
5452     // check motif21/libmawt.so
5453     strcpy(libmawtpath, buf);
5454     strcat(libmawtpath, motifstr);
5455     if (::stat(libmawtpath, &statbuf) == 0) return false;
5456 
5457     return true;
5458 }
5459 
5460 
5461 #ifdef JAVASE_EMBEDDED
5462 //
5463 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5464 //
5465 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5466 
5467 // ctor
5468 //
5469 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5470   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5471   _fd = fd;
5472 
5473   if (os::create_thread(this, os::os_thread)) {
5474     _memnotify_thread = this;
5475     os::set_priority(this, NearMaxPriority);
5476     os::start_thread(this);
5477   }
5478 }
5479 
5480 // Where all the work gets done
5481 //
5482 void MemNotifyThread::run() {
5483   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5484 
5485   // Set up the select arguments
5486   fd_set rfds;
5487   if (_fd != -1) {
5488     FD_ZERO(&rfds);
5489     FD_SET(_fd, &rfds);
5490   }
5491 
5492   // Now wait for the mem_notify device to wake up
5493   while (1) {
5494     // Wait for the mem_notify device to signal us..
5495     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5496     if (rc == -1) {
5497       perror("select!\n");
5498       break;
5499     } else if (rc) {
5500       //ssize_t free_before = os::available_memory();
5501       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5502 
5503       // The kernel is telling us there is not much memory left...
5504       // try to do something about that
5505 
5506       // If we are not already in a GC, try one.
5507       if (!Universe::heap()->is_gc_active()) {
5508         Universe::heap()->collect(GCCause::_allocation_failure);
5509 
5510         //ssize_t free_after = os::available_memory();
5511         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5512         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5513       }
5514       // We might want to do something like the following if we find the GC's are not helping...
5515       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5516     }
5517   }
5518 }
5519 
5520 //
5521 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5522 //
5523 void MemNotifyThread::start() {
5524   int    fd;
5525   fd = open ("/dev/mem_notify", O_RDONLY, 0);
5526   if (fd < 0) {
5527       return;
5528   }
5529 
5530   if (memnotify_thread() == NULL) {
5531     new MemNotifyThread(fd);
5532   }
5533 }
5534 #endif // JAVASE_EMBEDDED