1 /*
   2  * Copyright 1999-2008 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  20  * CA 95054 USA or visit www.sun.com if you need additional information or
  21  * have any questions.
  22  *
  23  */
  24 
  25 // do not include  precompiled  header file
  26 # include "incls/_os_linux.cpp.incl"
  27 
  28 // put OS-includes here
  29 # include <sys/types.h>
  30 # include <sys/mman.h>
  31 # include <pthread.h>
  32 # include <signal.h>
  33 # include <errno.h>
  34 # include <dlfcn.h>
  35 # include <stdio.h>
  36 # include <unistd.h>
  37 # include <sys/resource.h>
  38 # include <pthread.h>
  39 # include <sys/stat.h>
  40 # include <sys/time.h>
  41 # include <sys/times.h>
  42 # include <sys/utsname.h>
  43 # include <sys/socket.h>
  44 # include <sys/wait.h>
  45 # include <pwd.h>
  46 # include <poll.h>
  47 # include <semaphore.h>
  48 # include <fcntl.h>
  49 # include <string.h>
  50 # include <syscall.h>
  51 # include <sys/sysinfo.h>
  52 # include <gnu/libc-version.h>
  53 # include <sys/ipc.h>
  54 # include <sys/shm.h>
  55 # include <link.h>
  56 
  57 #define MAX_PATH    (2 * K)
  58 
  59 // for timer info max values which include all bits
  60 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
  61 #define SEC_IN_NANOSECS  1000000000LL
  62 
  63 ////////////////////////////////////////////////////////////////////////////////
  64 // global variables
  65 julong os::Linux::_physical_memory = 0;
  66 
  67 address   os::Linux::_initial_thread_stack_bottom = NULL;
  68 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
  69 
  70 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
  71 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
  72 Mutex* os::Linux::_createThread_lock = NULL;
  73 pthread_t os::Linux::_main_thread;
  74 int os::Linux::_page_size = -1;
  75 bool os::Linux::_is_floating_stack = false;
  76 bool os::Linux::_is_NPTL = false;
  77 bool os::Linux::_supports_fast_thread_cpu_time = false;
  78 const char * os::Linux::_glibc_version = NULL;
  79 const char * os::Linux::_libpthread_version = NULL;
  80 
  81 static jlong initial_time_count=0;
  82 
  83 static int clock_tics_per_sec = 100;
  84 
  85 // For diagnostics to print a message once. see run_periodic_checks
  86 static sigset_t check_signal_done;
  87 static bool check_signals = true;;
  88 
  89 static pid_t _initial_pid = 0;
  90 
  91 /* Signal number used to suspend/resume a thread */
  92 
  93 /* do not use any signal number less than SIGSEGV, see 4355769 */
  94 static int SR_signum = SIGUSR2;
  95 sigset_t SR_sigset;
  96 
  97 /* Used to protect dlsym() calls */
  98 static pthread_mutex_t dl_mutex;
  99 
 100 ////////////////////////////////////////////////////////////////////////////////
 101 // utility functions
 102 
 103 static int SR_initialize();
 104 static int SR_finalize();
 105 
 106 julong os::available_memory() {
 107   return Linux::available_memory();
 108 }
 109 
 110 julong os::Linux::available_memory() {
 111   // values in struct sysinfo are "unsigned long"
 112   struct sysinfo si;
 113   sysinfo(&si);
 114 
 115   return (julong)si.freeram * si.mem_unit;
 116 }
 117 
 118 julong os::physical_memory() {
 119   return Linux::physical_memory();
 120 }
 121 
 122 julong os::allocatable_physical_memory(julong size) {
 123 #ifdef _LP64
 124   return size;
 125 #else
 126   julong result = MIN2(size, (julong)3800*M);
 127    if (!is_allocatable(result)) {
 128      // See comments under solaris for alignment considerations
 129      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 130      result =  MIN2(size, reasonable_size);
 131    }
 132    return result;
 133 #endif // _LP64
 134 }
 135 
 136 ////////////////////////////////////////////////////////////////////////////////
 137 // environment support
 138 
 139 bool os::getenv(const char* name, char* buf, int len) {
 140   const char* val = ::getenv(name);
 141   if (val != NULL && strlen(val) < (size_t)len) {
 142     strcpy(buf, val);
 143     return true;
 144   }
 145   if (len > 0) buf[0] = 0;  // return a null string
 146   return false;
 147 }
 148 
 149 
 150 // Return true if user is running as root.
 151 
 152 bool os::have_special_privileges() {
 153   static bool init = false;
 154   static bool privileges = false;
 155   if (!init) {
 156     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 157     init = true;
 158   }
 159   return privileges;
 160 }
 161 
 162 
 163 #ifndef SYS_gettid
 164 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 165 #ifdef __ia64__
 166 #define SYS_gettid 1105
 167 #elif __i386__
 168 #define SYS_gettid 224
 169 #elif __amd64__
 170 #define SYS_gettid 186
 171 #elif __sparc__
 172 #define SYS_gettid 143
 173 #else
 174 #error define gettid for the arch
 175 #endif
 176 #endif
 177 
 178 // Cpu architecture string
 179 #if   defined(IA64)
 180 static char cpu_arch[] = "ia64";
 181 #elif defined(IA32)
 182 static char cpu_arch[] = "i386";
 183 #elif defined(AMD64)
 184 static char cpu_arch[] = "amd64";
 185 #elif defined(SPARC)
 186 #  ifdef _LP64
 187 static char cpu_arch[] = "sparcv9";
 188 #  else
 189 static char cpu_arch[] = "sparc";
 190 #  endif
 191 #else
 192 #error Add appropriate cpu_arch setting
 193 #endif
 194 
 195 
 196 // pid_t gettid()
 197 //
 198 // Returns the kernel thread id of the currently running thread. Kernel
 199 // thread id is used to access /proc.
 200 //
 201 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 202 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 203 //
 204 pid_t os::Linux::gettid() {
 205   int rslt = syscall(SYS_gettid);
 206   if (rslt == -1) {
 207      // old kernel, no NPTL support
 208      return getpid();
 209   } else {
 210      return (pid_t)rslt;
 211   }
 212 }
 213 
 214 // Most versions of linux have a bug where the number of processors are
 215 // determined by looking at the /proc file system.  In a chroot environment,
 216 // the system call returns 1.  This causes the VM to act as if it is
 217 // a single processor and elide locking (see is_MP() call).
 218 static bool unsafe_chroot_detected = false;
 219 static const char *unstable_chroot_error = "/proc file system not found.\n"
 220                      "Java may be unstable running multithreaded in a chroot "
 221                      "environment on Linux when /proc filesystem is not mounted.";
 222 
 223 void os::Linux::initialize_system_info() {
 224   _processor_count = sysconf(_SC_NPROCESSORS_CONF);
 225   if (_processor_count == 1) {
 226     pid_t pid = os::Linux::gettid();
 227     char fname[32];
 228     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 229     FILE *fp = fopen(fname, "r");
 230     if (fp == NULL) {
 231       unsafe_chroot_detected = true;
 232     } else {
 233       fclose(fp);
 234     }
 235   }
 236   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 237   assert(_processor_count > 0, "linux error");
 238 }
 239 
 240 void os::init_system_properties_values() {
 241 //  char arch[12];
 242 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 243 
 244   // The next steps are taken in the product version:
 245   //
 246   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 247   // This library should be located at:
 248   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 249   //
 250   // If "/jre/lib/" appears at the right place in the path, then we
 251   // assume libjvm[_g].so is installed in a JDK and we use this path.
 252   //
 253   // Otherwise exit with message: "Could not create the Java virtual machine."
 254   //
 255   // The following extra steps are taken in the debugging version:
 256   //
 257   // If "/jre/lib/" does NOT appear at the right place in the path
 258   // instead of exit check for $JAVA_HOME environment variable.
 259   //
 260   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 261   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 262   // it looks like libjvm[_g].so is installed there
 263   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 264   //
 265   // Otherwise exit.
 266   //
 267   // Important note: if the location of libjvm.so changes this
 268   // code needs to be changed accordingly.
 269 
 270   // The next few definitions allow the code to be verbatim:
 271 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 272 #define getenv(n) ::getenv(n)
 273 
 274 /*
 275  * See ld(1):
 276  *      The linker uses the following search paths to locate required
 277  *      shared libraries:
 278  *        1: ...
 279  *        ...
 280  *        7: The default directories, normally /lib and /usr/lib.
 281  */
 282 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 283 
 284 #define EXTENSIONS_DIR  "/lib/ext"
 285 #define ENDORSED_DIR    "/lib/endorsed"
 286 #define REG_DIR         "/usr/java/packages"
 287 
 288   {
 289     /* sysclasspath, java_home, dll_dir */
 290     {
 291         char *home_path;
 292         char *dll_path;
 293         char *pslash;
 294         char buf[MAXPATHLEN];
 295         os::jvm_path(buf, sizeof(buf));
 296 
 297         // Found the full path to libjvm.so.
 298         // Now cut the path to <java_home>/jre if we can.
 299         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 300         pslash = strrchr(buf, '/');
 301         if (pslash != NULL)
 302             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 303         dll_path = malloc(strlen(buf) + 1);
 304         if (dll_path == NULL)
 305             return;
 306         strcpy(dll_path, buf);
 307         Arguments::set_dll_dir(dll_path);
 308 
 309         if (pslash != NULL) {
 310             pslash = strrchr(buf, '/');
 311             if (pslash != NULL) {
 312                 *pslash = '\0';       /* get rid of /<arch> */
 313                 pslash = strrchr(buf, '/');
 314                 if (pslash != NULL)
 315                     *pslash = '\0';   /* get rid of /lib */
 316             }
 317         }
 318 
 319         home_path = malloc(strlen(buf) + 1);
 320         if (home_path == NULL)
 321             return;
 322         strcpy(home_path, buf);
 323         Arguments::set_java_home(home_path);
 324 
 325         if (!set_boot_path('/', ':'))
 326             return;
 327     }
 328 
 329     /*
 330      * Where to look for native libraries
 331      *
 332      * Note: Due to a legacy implementation, most of the library path
 333      * is set in the launcher.  This was to accomodate linking restrictions
 334      * on legacy Linux implementations (which are no longer supported).
 335      * Eventually, all the library path setting will be done here.
 336      *
 337      * However, to prevent the proliferation of improperly built native
 338      * libraries, the new path component /usr/java/packages is added here.
 339      * Eventually, all the library path setting will be done here.
 340      */
 341     {
 342         char *ld_library_path;
 343 
 344         /*
 345          * Construct the invariant part of ld_library_path. Note that the
 346          * space for the colon and the trailing null are provided by the
 347          * nulls included by the sizeof operator (so actually we allocate
 348          * a byte more than necessary).
 349          */
 350         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 351             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 352         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 353 
 354         /*
 355          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 356          * should always exist (until the legacy problem cited above is
 357          * addressed).
 358          */
 359         char *v = getenv("LD_LIBRARY_PATH");
 360         if (v != NULL) {
 361             char *t = ld_library_path;
 362             /* That's +1 for the colon and +1 for the trailing '\0' */
 363             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 364             sprintf(ld_library_path, "%s:%s", v, t);
 365         }
 366         Arguments::set_library_path(ld_library_path);
 367     }
 368 
 369     /*
 370      * Extensions directories.
 371      *
 372      * Note that the space for the colon and the trailing null are provided
 373      * by the nulls included by the sizeof operator (so actually one byte more
 374      * than necessary is allocated).
 375      */
 376     {
 377         char *buf = malloc(strlen(Arguments::get_java_home()) +
 378             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 379         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 380             Arguments::get_java_home());
 381         Arguments::set_ext_dirs(buf);
 382     }
 383 
 384     /* Endorsed standards default directory. */
 385     {
 386         char * buf;
 387         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 388         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 389         Arguments::set_endorsed_dirs(buf);
 390     }
 391   }
 392 
 393 #undef malloc
 394 #undef getenv
 395 #undef EXTENSIONS_DIR
 396 #undef ENDORSED_DIR
 397 
 398   // Done
 399   return;
 400 }
 401 
 402 ////////////////////////////////////////////////////////////////////////////////
 403 // breakpoint support
 404 
 405 void os::breakpoint() {
 406   BREAKPOINT;
 407 }
 408 
 409 extern "C" void breakpoint() {
 410   // use debugger to set breakpoint here
 411 }
 412 
 413 ////////////////////////////////////////////////////////////////////////////////
 414 // signal support
 415 
 416 debug_only(static bool signal_sets_initialized = false);
 417 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 418 
 419 bool os::Linux::is_sig_ignored(int sig) {
 420       struct sigaction oact;
 421       sigaction(sig, (struct sigaction*)NULL, &oact);
 422       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 423                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 424       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 425            return true;
 426       else
 427            return false;
 428 }
 429 
 430 void os::Linux::signal_sets_init() {
 431   // Should also have an assertion stating we are still single-threaded.
 432   assert(!signal_sets_initialized, "Already initialized");
 433   // Fill in signals that are necessarily unblocked for all threads in
 434   // the VM. Currently, we unblock the following signals:
 435   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 436   //                         by -Xrs (=ReduceSignalUsage));
 437   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 438   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 439   // the dispositions or masks wrt these signals.
 440   // Programs embedding the VM that want to use the above signals for their
 441   // own purposes must, at this time, use the "-Xrs" option to prevent
 442   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 443   // (See bug 4345157, and other related bugs).
 444   // In reality, though, unblocking these signals is really a nop, since
 445   // these signals are not blocked by default.
 446   sigemptyset(&unblocked_sigs);
 447   sigemptyset(&allowdebug_blocked_sigs);
 448   sigaddset(&unblocked_sigs, SIGILL);
 449   sigaddset(&unblocked_sigs, SIGSEGV);
 450   sigaddset(&unblocked_sigs, SIGBUS);
 451   sigaddset(&unblocked_sigs, SIGFPE);
 452   sigaddset(&unblocked_sigs, SR_signum);
 453 
 454   if (!ReduceSignalUsage) {
 455    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 456       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 457       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 458    }
 459    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 460       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 461       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 462    }
 463    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 464       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 465       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 466    }
 467   }
 468   // Fill in signals that are blocked by all but the VM thread.
 469   sigemptyset(&vm_sigs);
 470   if (!ReduceSignalUsage)
 471     sigaddset(&vm_sigs, BREAK_SIGNAL);
 472   debug_only(signal_sets_initialized = true);
 473 
 474 }
 475 
 476 // These are signals that are unblocked while a thread is running Java.
 477 // (For some reason, they get blocked by default.)
 478 sigset_t* os::Linux::unblocked_signals() {
 479   assert(signal_sets_initialized, "Not initialized");
 480   return &unblocked_sigs;
 481 }
 482 
 483 // These are the signals that are blocked while a (non-VM) thread is
 484 // running Java. Only the VM thread handles these signals.
 485 sigset_t* os::Linux::vm_signals() {
 486   assert(signal_sets_initialized, "Not initialized");
 487   return &vm_sigs;
 488 }
 489 
 490 // These are signals that are blocked during cond_wait to allow debugger in
 491 sigset_t* os::Linux::allowdebug_blocked_signals() {
 492   assert(signal_sets_initialized, "Not initialized");
 493   return &allowdebug_blocked_sigs;
 494 }
 495 
 496 void os::Linux::hotspot_sigmask(Thread* thread) {
 497 
 498   //Save caller's signal mask before setting VM signal mask
 499   sigset_t caller_sigmask;
 500   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 501 
 502   OSThread* osthread = thread->osthread();
 503   osthread->set_caller_sigmask(caller_sigmask);
 504 
 505   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 506 
 507   if (!ReduceSignalUsage) {
 508     if (thread->is_VM_thread()) {
 509       // Only the VM thread handles BREAK_SIGNAL ...
 510       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 511     } else {
 512       // ... all other threads block BREAK_SIGNAL
 513       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 514     }
 515   }
 516 }
 517 
 518 //////////////////////////////////////////////////////////////////////////////
 519 // detecting pthread library
 520 
 521 void os::Linux::libpthread_init() {
 522   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 523   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 524   // generic name for earlier versions.
 525   // Define macros here so we can build HotSpot on old systems.
 526 # ifndef _CS_GNU_LIBC_VERSION
 527 # define _CS_GNU_LIBC_VERSION 2
 528 # endif
 529 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 530 # define _CS_GNU_LIBPTHREAD_VERSION 3
 531 # endif
 532 
 533   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 534   if (n > 0) {
 535      char *str = (char *)malloc(n);
 536      confstr(_CS_GNU_LIBC_VERSION, str, n);
 537      os::Linux::set_glibc_version(str);
 538   } else {
 539      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 540      static char _gnu_libc_version[32];
 541      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 542               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 543      os::Linux::set_glibc_version(_gnu_libc_version);
 544   }
 545 
 546   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 547   if (n > 0) {
 548      char *str = (char *)malloc(n);
 549      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 550      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 551      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 552      // is the case. LinuxThreads has a hard limit on max number of threads.
 553      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 554      // On the other hand, NPTL does not have such a limit, sysconf()
 555      // will return -1 and errno is not changed. Check if it is really NPTL.
 556      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 557          strstr(str, "NPTL") &&
 558          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 559        free(str);
 560        os::Linux::set_libpthread_version("linuxthreads");
 561      } else {
 562        os::Linux::set_libpthread_version(str);
 563      }
 564   } else {
 565     // glibc before 2.3.2 only has LinuxThreads.
 566     os::Linux::set_libpthread_version("linuxthreads");
 567   }
 568 
 569   if (strstr(libpthread_version(), "NPTL")) {
 570      os::Linux::set_is_NPTL();
 571   } else {
 572      os::Linux::set_is_LinuxThreads();
 573   }
 574 
 575   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 576   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 577   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 578      os::Linux::set_is_floating_stack();
 579   }
 580 }
 581 
 582 /////////////////////////////////////////////////////////////////////////////
 583 // thread stack
 584 
 585 // Force Linux kernel to expand current thread stack. If "bottom" is close
 586 // to the stack guard, caller should block all signals.
 587 //
 588 // MAP_GROWSDOWN:
 589 //   A special mmap() flag that is used to implement thread stacks. It tells
 590 //   kernel that the memory region should extend downwards when needed. This
 591 //   allows early versions of LinuxThreads to only mmap the first few pages
 592 //   when creating a new thread. Linux kernel will automatically expand thread
 593 //   stack as needed (on page faults).
 594 //
 595 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 596 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 597 //   region, it's hard to tell if the fault is due to a legitimate stack
 598 //   access or because of reading/writing non-exist memory (e.g. buffer
 599 //   overrun). As a rule, if the fault happens below current stack pointer,
 600 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 601 //   application (see Linux kernel fault.c).
 602 //
 603 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 604 //   stack overflow detection.
 605 //
 606 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 607 //   not use this flag. However, the stack of initial thread is not created
 608 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 609 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 610 //   and then attach the thread to JVM.
 611 //
 612 // To get around the problem and allow stack banging on Linux, we need to
 613 // manually expand thread stack after receiving the SIGSEGV.
 614 //
 615 // There are two ways to expand thread stack to address "bottom", we used
 616 // both of them in JVM before 1.5:
 617 //   1. adjust stack pointer first so that it is below "bottom", and then
 618 //      touch "bottom"
 619 //   2. mmap() the page in question
 620 //
 621 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 622 // if current sp is already near the lower end of page 101, and we need to
 623 // call mmap() to map page 100, it is possible that part of the mmap() frame
 624 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 625 // That will destroy the mmap() frame and cause VM to crash.
 626 //
 627 // The following code works by adjusting sp first, then accessing the "bottom"
 628 // page to force a page fault. Linux kernel will then automatically expand the
 629 // stack mapping.
 630 //
 631 // _expand_stack_to() assumes its frame size is less than page size, which
 632 // should always be true if the function is not inlined.
 633 
 634 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 635 #define NOINLINE
 636 #else
 637 #define NOINLINE __attribute__ ((noinline))
 638 #endif
 639 
 640 static void _expand_stack_to(address bottom) NOINLINE;
 641 
 642 static void _expand_stack_to(address bottom) {
 643   address sp;
 644   size_t size;
 645   volatile char *p;
 646 
 647   // Adjust bottom to point to the largest address within the same page, it
 648   // gives us a one-page buffer if alloca() allocates slightly more memory.
 649   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 650   bottom += os::Linux::page_size() - 1;
 651 
 652   // sp might be slightly above current stack pointer; if that's the case, we
 653   // will alloca() a little more space than necessary, which is OK. Don't use
 654   // os::current_stack_pointer(), as its result can be slightly below current
 655   // stack pointer, causing us to not alloca enough to reach "bottom".
 656   sp = (address)&sp;
 657 
 658   if (sp > bottom) {
 659     size = sp - bottom;
 660     p = (volatile char *)alloca(size);
 661     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 662     p[0] = '\0';
 663   }
 664 }
 665 
 666 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 667   assert(t!=NULL, "just checking");
 668   assert(t->osthread()->expanding_stack(), "expand should be set");
 669   assert(t->stack_base() != NULL, "stack_base was not initialized");
 670 
 671   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 672     sigset_t mask_all, old_sigset;
 673     sigfillset(&mask_all);
 674     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 675     _expand_stack_to(addr);
 676     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 677     return true;
 678   }
 679   return false;
 680 }
 681 
 682 //////////////////////////////////////////////////////////////////////////////
 683 // create new thread
 684 
 685 static address highest_vm_reserved_address();
 686 
 687 // check if it's safe to start a new thread
 688 static bool _thread_safety_check(Thread* thread) {
 689   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 690     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 691     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 692     //   allocated (MAP_FIXED) from high address space. Every thread stack
 693     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 694     //   it to other values if they rebuild LinuxThreads).
 695     //
 696     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 697     // the memory region has already been mmap'ed. That means if we have too
 698     // many threads and/or very large heap, eventually thread stack will
 699     // collide with heap.
 700     //
 701     // Here we try to prevent heap/stack collision by comparing current
 702     // stack bottom with the highest address that has been mmap'ed by JVM
 703     // plus a safety margin for memory maps created by native code.
 704     //
 705     // This feature can be disabled by setting ThreadSafetyMargin to 0
 706     //
 707     if (ThreadSafetyMargin > 0) {
 708       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 709 
 710       // not safe if our stack extends below the safety margin
 711       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 712     } else {
 713       return true;
 714     }
 715   } else {
 716     // Floating stack LinuxThreads or NPTL:
 717     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 718     //   there's not enough space left, pthread_create() will fail. If we come
 719     //   here, that means enough space has been reserved for stack.
 720     return true;
 721   }
 722 }
 723 
 724 // Thread start routine for all newly created threads
 725 static void *java_start(Thread *thread) {
 726   // Try to randomize the cache line index of hot stack frames.
 727   // This helps when threads of the same stack traces evict each other's
 728   // cache lines. The threads can be either from the same JVM instance, or
 729   // from different JVM instances. The benefit is especially true for
 730   // processors with hyperthreading technology.
 731   static int counter = 0;
 732   int pid = os::current_process_id();
 733   alloca(((pid ^ counter++) & 7) * 128);
 734 
 735   ThreadLocalStorage::set_thread(thread);
 736 
 737   OSThread* osthread = thread->osthread();
 738   Monitor* sync = osthread->startThread_lock();
 739 
 740   // non floating stack LinuxThreads needs extra check, see above
 741   if (!_thread_safety_check(thread)) {
 742     // notify parent thread
 743     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 744     osthread->set_state(ZOMBIE);
 745     sync->notify_all();
 746     return NULL;
 747   }
 748 
 749   // thread_id is kernel thread id (similar to Solaris LWP id)
 750   osthread->set_thread_id(os::Linux::gettid());
 751 
 752   if (UseNUMA) {
 753     int lgrp_id = os::numa_get_group_id();
 754     if (lgrp_id != -1) {
 755       thread->set_lgrp_id(lgrp_id);
 756     }
 757   }
 758   // initialize signal mask for this thread
 759   os::Linux::hotspot_sigmask(thread);
 760 
 761   // initialize floating point control register
 762   os::Linux::init_thread_fpu_state();
 763 
 764   // handshaking with parent thread
 765   {
 766     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 767 
 768     // notify parent thread
 769     osthread->set_state(INITIALIZED);
 770     sync->notify_all();
 771 
 772     // wait until os::start_thread()
 773     while (osthread->get_state() == INITIALIZED) {
 774       sync->wait(Mutex::_no_safepoint_check_flag);
 775     }
 776   }
 777 
 778   // call one more level start routine
 779   thread->run();
 780 
 781   return 0;
 782 }
 783 
 784 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 785   assert(thread->osthread() == NULL, "caller responsible");
 786 
 787   // Allocate the OSThread object
 788   OSThread* osthread = new OSThread(NULL, NULL);
 789   if (osthread == NULL) {
 790     return false;
 791   }
 792 
 793   // set the correct thread state
 794   osthread->set_thread_type(thr_type);
 795 
 796   // Initial state is ALLOCATED but not INITIALIZED
 797   osthread->set_state(ALLOCATED);
 798 
 799   thread->set_osthread(osthread);
 800 
 801   // init thread attributes
 802   pthread_attr_t attr;
 803   pthread_attr_init(&attr);
 804   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 805 
 806   // stack size
 807   if (os::Linux::supports_variable_stack_size()) {
 808     // calculate stack size if it's not specified by caller
 809     if (stack_size == 0) {
 810       stack_size = os::Linux::default_stack_size(thr_type);
 811 
 812       switch (thr_type) {
 813       case os::java_thread:
 814         // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
 815         if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
 816         break;
 817       case os::compiler_thread:
 818         if (CompilerThreadStackSize > 0) {
 819           stack_size = (size_t)(CompilerThreadStackSize * K);
 820           break;
 821         } // else fall through:
 822           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 823       case os::vm_thread:
 824       case os::pgc_thread:
 825       case os::cgc_thread:
 826       case os::watcher_thread:
 827         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 828         break;
 829       }
 830     }
 831 
 832     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 833     pthread_attr_setstacksize(&attr, stack_size);
 834   } else {
 835     // let pthread_create() pick the default value.
 836   }
 837 
 838   // glibc guard page
 839   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 840 
 841   ThreadState state;
 842 
 843   {
 844     // Serialize thread creation if we are running with fixed stack LinuxThreads
 845     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 846     if (lock) {
 847       os::Linux::createThread_lock()->lock_without_safepoint_check();
 848     }
 849 
 850     pthread_t tid;
 851     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 852 
 853     pthread_attr_destroy(&attr);
 854 
 855     if (ret != 0) {
 856       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 857         perror("pthread_create()");
 858       }
 859       // Need to clean up stuff we've allocated so far
 860       thread->set_osthread(NULL);
 861       delete osthread;
 862       if (lock) os::Linux::createThread_lock()->unlock();
 863       return false;
 864     }
 865 
 866     // Store pthread info into the OSThread
 867     osthread->set_pthread_id(tid);
 868 
 869     // Wait until child thread is either initialized or aborted
 870     {
 871       Monitor* sync_with_child = osthread->startThread_lock();
 872       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 873       while ((state = osthread->get_state()) == ALLOCATED) {
 874         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 875       }
 876     }
 877 
 878     if (lock) {
 879       os::Linux::createThread_lock()->unlock();
 880     }
 881   }
 882 
 883   // Aborted due to thread limit being reached
 884   if (state == ZOMBIE) {
 885       thread->set_osthread(NULL);
 886       delete osthread;
 887       return false;
 888   }
 889 
 890   // The thread is returned suspended (in state INITIALIZED),
 891   // and is started higher up in the call chain
 892   assert(state == INITIALIZED, "race condition");
 893   return true;
 894 }
 895 
 896 /////////////////////////////////////////////////////////////////////////////
 897 // attach existing thread
 898 
 899 // bootstrap the main thread
 900 bool os::create_main_thread(JavaThread* thread) {
 901   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 902   return create_attached_thread(thread);
 903 }
 904 
 905 bool os::create_attached_thread(JavaThread* thread) {
 906 #ifdef ASSERT
 907     thread->verify_not_published();
 908 #endif
 909 
 910   // Allocate the OSThread object
 911   OSThread* osthread = new OSThread(NULL, NULL);
 912 
 913   if (osthread == NULL) {
 914     return false;
 915   }
 916 
 917   // Store pthread info into the OSThread
 918   osthread->set_thread_id(os::Linux::gettid());
 919   osthread->set_pthread_id(::pthread_self());
 920 
 921   // initialize floating point control register
 922   os::Linux::init_thread_fpu_state();
 923 
 924   // Initial thread state is RUNNABLE
 925   osthread->set_state(RUNNABLE);
 926 
 927   thread->set_osthread(osthread);
 928 
 929   if (UseNUMA) {
 930     int lgrp_id = os::numa_get_group_id();
 931     if (lgrp_id != -1) {
 932       thread->set_lgrp_id(lgrp_id);
 933     }
 934   }
 935 
 936   if (os::Linux::is_initial_thread()) {
 937     // If current thread is initial thread, its stack is mapped on demand,
 938     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 939     // the entire stack region to avoid SEGV in stack banging.
 940     // It is also useful to get around the heap-stack-gap problem on SuSE
 941     // kernel (see 4821821 for details). We first expand stack to the top
 942     // of yellow zone, then enable stack yellow zone (order is significant,
 943     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 944     // is no gap between the last two virtual memory regions.
 945 
 946     JavaThread *jt = (JavaThread *)thread;
 947     address addr = jt->stack_yellow_zone_base();
 948     assert(addr != NULL, "initialization problem?");
 949     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 950 
 951     osthread->set_expanding_stack();
 952     os::Linux::manually_expand_stack(jt, addr);
 953     osthread->clear_expanding_stack();
 954   }
 955 
 956   // initialize signal mask for this thread
 957   // and save the caller's signal mask
 958   os::Linux::hotspot_sigmask(thread);
 959 
 960   return true;
 961 }
 962 
 963 void os::pd_start_thread(Thread* thread) {
 964   OSThread * osthread = thread->osthread();
 965   assert(osthread->get_state() != INITIALIZED, "just checking");
 966   Monitor* sync_with_child = osthread->startThread_lock();
 967   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 968   sync_with_child->notify();
 969 }
 970 
 971 // Free Linux resources related to the OSThread
 972 void os::free_thread(OSThread* osthread) {
 973   assert(osthread != NULL, "osthread not set");
 974 
 975   if (Thread::current()->osthread() == osthread) {
 976     // Restore caller's signal mask
 977     sigset_t sigmask = osthread->caller_sigmask();
 978     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 979    }
 980 
 981   delete osthread;
 982 }
 983 
 984 //////////////////////////////////////////////////////////////////////////////
 985 // thread local storage
 986 
 987 int os::allocate_thread_local_storage() {
 988   pthread_key_t key;
 989   int rslt = pthread_key_create(&key, NULL);
 990   assert(rslt == 0, "cannot allocate thread local storage");
 991   return (int)key;
 992 }
 993 
 994 // Note: This is currently not used by VM, as we don't destroy TLS key
 995 // on VM exit.
 996 void os::free_thread_local_storage(int index) {
 997   int rslt = pthread_key_delete((pthread_key_t)index);
 998   assert(rslt == 0, "invalid index");
 999 }
1000 
1001 void os::thread_local_storage_at_put(int index, void* value) {
1002   int rslt = pthread_setspecific((pthread_key_t)index, value);
1003   assert(rslt == 0, "pthread_setspecific failed");
1004 }
1005 
1006 extern "C" Thread* get_thread() {
1007   return ThreadLocalStorage::thread();
1008 }
1009 
1010 //////////////////////////////////////////////////////////////////////////////
1011 // initial thread
1012 
1013 // Check if current thread is the initial thread, similar to Solaris thr_main.
1014 bool os::Linux::is_initial_thread(void) {
1015   char dummy;
1016   // If called before init complete, thread stack bottom will be null.
1017   // Can be called if fatal error occurs before initialization.
1018   if (initial_thread_stack_bottom() == NULL) return false;
1019   assert(initial_thread_stack_bottom() != NULL &&
1020          initial_thread_stack_size()   != 0,
1021          "os::init did not locate initial thread's stack region");
1022   if ((address)&dummy >= initial_thread_stack_bottom() &&
1023       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1024        return true;
1025   else return false;
1026 }
1027 
1028 // Find the virtual memory area that contains addr
1029 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1030   FILE *fp = fopen("/proc/self/maps", "r");
1031   if (fp) {
1032     address low, high;
1033     while (!feof(fp)) {
1034       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1035         if (low <= addr && addr < high) {
1036            if (vma_low)  *vma_low  = low;
1037            if (vma_high) *vma_high = high;
1038            fclose (fp);
1039            return true;
1040         }
1041       }
1042       for (;;) {
1043         int ch = fgetc(fp);
1044         if (ch == EOF || ch == (int)'\n') break;
1045       }
1046     }
1047     fclose(fp);
1048   }
1049   return false;
1050 }
1051 
1052 // Locate initial thread stack. This special handling of initial thread stack
1053 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1054 // bogus value for initial thread.
1055 void os::Linux::capture_initial_stack(size_t max_size) {
1056   // stack size is the easy part, get it from RLIMIT_STACK
1057   size_t stack_size;
1058   struct rlimit rlim;
1059   getrlimit(RLIMIT_STACK, &rlim);
1060   stack_size = rlim.rlim_cur;
1061 
1062   // 6308388: a bug in ld.so will relocate its own .data section to the
1063   //   lower end of primordial stack; reduce ulimit -s value a little bit
1064   //   so we won't install guard page on ld.so's data section.
1065   stack_size -= 2 * page_size();
1066 
1067   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1068   //   7.1, in both cases we will get 2G in return value.
1069   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1070   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1071   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1072   //   in case other parts in glibc still assumes 2M max stack size.
1073   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1074 #ifndef IA64
1075   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1076 #else
1077   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1078   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1079 #endif
1080 
1081   // Try to figure out where the stack base (top) is. This is harder.
1082   //
1083   // When an application is started, glibc saves the initial stack pointer in
1084   // a global variable "__libc_stack_end", which is then used by system
1085   // libraries. __libc_stack_end should be pretty close to stack top. The
1086   // variable is available since the very early days. However, because it is
1087   // a private interface, it could disappear in the future.
1088   //
1089   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1090   // to __libc_stack_end, it is very close to stack top, but isn't the real
1091   // stack top. Note that /proc may not exist if VM is running as a chroot
1092   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1093   // /proc/<pid>/stat could change in the future (though unlikely).
1094   //
1095   // We try __libc_stack_end first. If that doesn't work, look for
1096   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1097   // as a hint, which should work well in most cases.
1098 
1099   uintptr_t stack_start;
1100 
1101   // try __libc_stack_end first
1102   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1103   if (p && *p) {
1104     stack_start = *p;
1105   } else {
1106     // see if we can get the start_stack field from /proc/self/stat
1107     FILE *fp;
1108     int pid;
1109     char state;
1110     int ppid;
1111     int pgrp;
1112     int session;
1113     int nr;
1114     int tpgrp;
1115     unsigned long flags;
1116     unsigned long minflt;
1117     unsigned long cminflt;
1118     unsigned long majflt;
1119     unsigned long cmajflt;
1120     unsigned long utime;
1121     unsigned long stime;
1122     long cutime;
1123     long cstime;
1124     long prio;
1125     long nice;
1126     long junk;
1127     long it_real;
1128     uintptr_t start;
1129     uintptr_t vsize;
1130     uintptr_t rss;
1131     unsigned long rsslim;
1132     uintptr_t scodes;
1133     uintptr_t ecode;
1134     int i;
1135 
1136     // Figure what the primordial thread stack base is. Code is inspired
1137     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1138     // followed by command name surrounded by parentheses, state, etc.
1139     char stat[2048];
1140     int statlen;
1141 
1142     fp = fopen("/proc/self/stat", "r");
1143     if (fp) {
1144       statlen = fread(stat, 1, 2047, fp);
1145       stat[statlen] = '\0';
1146       fclose(fp);
1147 
1148       // Skip pid and the command string. Note that we could be dealing with
1149       // weird command names, e.g. user could decide to rename java launcher
1150       // to "java 1.4.2 :)", then the stat file would look like
1151       //                1234 (java 1.4.2 :)) R ... ...
1152       // We don't really need to know the command string, just find the last
1153       // occurrence of ")" and then start parsing from there. See bug 4726580.
1154       char * s = strrchr(stat, ')');
1155 
1156       i = 0;
1157       if (s) {
1158         // Skip blank chars
1159         do s++; while (isspace(*s));
1160 
1161         /*                                     1   1   1   1   1   1   1   1   1   1   2   2   2   2   2   2   2   2   2 */
1162         /*              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 */
1163         i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld "
1164                    UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT
1165                    " %lu " 
1166                    UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT,
1167              &state,          /* 3  %c  */
1168              &ppid,           /* 4  %d  */
1169              &pgrp,           /* 5  %d  */
1170              &session,        /* 6  %d  */
1171              &nr,             /* 7  %d  */
1172              &tpgrp,          /* 8  %d  */
1173              &flags,          /* 9  %lu  */
1174              &minflt,         /* 10 %lu  */
1175              &cminflt,        /* 11 %lu  */
1176              &majflt,         /* 12 %lu  */
1177              &cmajflt,        /* 13 %lu  */
1178              &utime,          /* 14 %lu  */
1179              &stime,          /* 15 %lu  */
1180              &cutime,         /* 16 %ld  */
1181              &cstime,         /* 17 %ld  */
1182              &prio,           /* 18 %ld  */
1183              &nice,           /* 19 %ld  */
1184              &junk,           /* 20 %ld  */
1185              &it_real,        /* 21 %ld  */
1186              &start,          /* 22 UINTX_FORMAT  */
1187              &vsize,          /* 23 UINTX_FORMAT  */
1188              &rss,            /* 24 UINTX_FORMAT  */
1189              &rsslim,         /* 25 %lu  */
1190              &scodes,         /* 26 UINTX_FORMAT  */
1191              &ecode,          /* 27 UINTX_FORMAT  */
1192              &stack_start);   /* 28 UINTX_FORMAT  */
1193       }
1194 
1195       if (i != 28 - 2) {
1196          assert(false, "Bad conversion from /proc/self/stat");
1197          // product mode - assume we are the initial thread, good luck in the
1198          // embedded case.
1199          warning("Can't detect initial thread stack location - bad conversion");
1200          stack_start = (uintptr_t) &rlim;
1201       }
1202     } else {
1203       // For some reason we can't open /proc/self/stat (for example, running on
1204       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1205       // most cases, so don't abort:
1206       warning("Can't detect initial thread stack location - no /proc/self/stat");
1207       stack_start = (uintptr_t) &rlim;
1208     }
1209   }
1210 
1211   // Now we have a pointer (stack_start) very close to the stack top, the
1212   // next thing to do is to figure out the exact location of stack top. We
1213   // can find out the virtual memory area that contains stack_start by
1214   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1215   // and its upper limit is the real stack top. (again, this would fail if
1216   // running inside chroot, because /proc may not exist.)
1217 
1218   uintptr_t stack_top;
1219   address low, high;
1220   if (find_vma((address)stack_start, &low, &high)) {
1221     // success, "high" is the true stack top. (ignore "low", because initial
1222     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1223     stack_top = (uintptr_t)high;
1224   } else {
1225     // failed, likely because /proc/self/maps does not exist
1226     warning("Can't detect initial thread stack location - find_vma failed");
1227     // best effort: stack_start is normally within a few pages below the real
1228     // stack top, use it as stack top, and reduce stack size so we won't put
1229     // guard page outside stack.
1230     stack_top = stack_start;
1231     stack_size -= 16 * page_size();
1232   }
1233 
1234   // stack_top could be partially down the page so align it
1235   stack_top = align_size_up(stack_top, page_size());
1236 
1237   if (max_size && stack_size > max_size) {
1238      _initial_thread_stack_size = max_size;
1239   } else {
1240      _initial_thread_stack_size = stack_size;
1241   }
1242 
1243   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1244   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1245 }
1246 
1247 ////////////////////////////////////////////////////////////////////////////////
1248 // time support
1249 
1250 // Time since start-up in seconds to a fine granularity.
1251 // Used by VMSelfDestructTimer and the MemProfiler.
1252 double os::elapsedTime() {
1253 
1254   return (double)(os::elapsed_counter()) * 0.000001;
1255 }
1256 
1257 jlong os::elapsed_counter() {
1258   timeval time;
1259   int status = gettimeofday(&time, NULL);
1260   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1261 }
1262 
1263 jlong os::elapsed_frequency() {
1264   return (1000 * 1000);
1265 }
1266 
1267 // For now, we say that linux does not support vtime.  I have no idea
1268 // whether it can actually be made to (DLD, 9/13/05).
1269 
1270 bool os::supports_vtime() { return false; }
1271 bool os::enable_vtime()   { return false; }
1272 bool os::vtime_enabled()  { return false; }
1273 double os::elapsedVTime() {
1274   // better than nothing, but not much
1275   return elapsedTime();
1276 }
1277 
1278 jlong os::javaTimeMillis() {
1279   timeval time;
1280   int status = gettimeofday(&time, NULL);
1281   assert(status != -1, "linux error");
1282   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1283 }
1284 
1285 #ifndef CLOCK_MONOTONIC
1286 #define CLOCK_MONOTONIC (1)
1287 #endif
1288 
1289 void os::Linux::clock_init() {
1290   // we do dlopen's in this particular order due to bug in linux
1291   // dynamical loader (see 6348968) leading to crash on exit
1292   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1293   if (handle == NULL) {
1294     handle = dlopen("librt.so", RTLD_LAZY);
1295   }
1296 
1297   if (handle) {
1298     int (*clock_getres_func)(clockid_t, struct timespec*) =
1299            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1300     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1301            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1302     if (clock_getres_func && clock_gettime_func) {
1303       // See if monotonic clock is supported by the kernel. Note that some
1304       // early implementations simply return kernel jiffies (updated every
1305       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1306       // for nano time (though the monotonic property is still nice to have).
1307       // It's fixed in newer kernels, however clock_getres() still returns
1308       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1309       // resolution for now. Hopefully as people move to new kernels, this
1310       // won't be a problem.
1311       struct timespec res;
1312       struct timespec tp;
1313       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1314           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1315         // yes, monotonic clock is supported
1316         _clock_gettime = clock_gettime_func;
1317       } else {
1318         // close librt if there is no monotonic clock
1319         dlclose(handle);
1320       }
1321     }
1322   }
1323 }
1324 
1325 #ifndef SYS_clock_getres
1326 
1327 #if defined(IA32) || defined(AMD64)
1328 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1329 #else
1330 #error Value of SYS_clock_getres not known on this platform
1331 #endif
1332 
1333 #endif
1334 
1335 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1336 
1337 void os::Linux::fast_thread_clock_init() {
1338   if (!UseLinuxPosixThreadCPUClocks) {
1339     return;
1340   }
1341   clockid_t clockid;
1342   struct timespec tp;
1343   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1344       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1345 
1346   // Switch to using fast clocks for thread cpu time if
1347   // the sys_clock_getres() returns 0 error code.
1348   // Note, that some kernels may support the current thread
1349   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1350   // returned by the pthread_getcpuclockid().
1351   // If the fast Posix clocks are supported then the sys_clock_getres()
1352   // must return at least tp.tv_sec == 0 which means a resolution
1353   // better than 1 sec. This is extra check for reliability.
1354 
1355   if(pthread_getcpuclockid_func &&
1356      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1357      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1358 
1359     _supports_fast_thread_cpu_time = true;
1360     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1361   }
1362 }
1363 
1364 jlong os::javaTimeNanos() {
1365   if (Linux::supports_monotonic_clock()) {
1366     struct timespec tp;
1367     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1368     assert(status == 0, "gettime error");
1369     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1370     return result;
1371   } else {
1372     timeval time;
1373     int status = gettimeofday(&time, NULL);
1374     assert(status != -1, "linux error");
1375     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1376     return 1000 * usecs;
1377   }
1378 }
1379 
1380 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1381   if (Linux::supports_monotonic_clock()) {
1382     info_ptr->max_value = ALL_64_BITS;
1383 
1384     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1385     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1386     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1387   } else {
1388     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1389     info_ptr->max_value = ALL_64_BITS;
1390 
1391     // gettimeofday is a real time clock so it skips
1392     info_ptr->may_skip_backward = true;
1393     info_ptr->may_skip_forward = true;
1394   }
1395 
1396   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1397 }
1398 
1399 // Return the real, user, and system times in seconds from an
1400 // arbitrary fixed point in the past.
1401 bool os::getTimesSecs(double* process_real_time,
1402                       double* process_user_time,
1403                       double* process_system_time) {
1404   struct tms ticks;
1405   clock_t real_ticks = times(&ticks);
1406 
1407   if (real_ticks == (clock_t) (-1)) {
1408     return false;
1409   } else {
1410     double ticks_per_second = (double) clock_tics_per_sec;
1411     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1412     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1413     *process_real_time = ((double) real_ticks) / ticks_per_second;
1414 
1415     return true;
1416   }
1417 }
1418 
1419 
1420 char * os::local_time_string(char *buf, size_t buflen) {
1421   struct tm t;
1422   time_t long_time;
1423   time(&long_time);
1424   localtime_r(&long_time, &t);
1425   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1426                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1427                t.tm_hour, t.tm_min, t.tm_sec);
1428   return buf;
1429 }
1430 
1431 ////////////////////////////////////////////////////////////////////////////////
1432 // runtime exit support
1433 
1434 // Note: os::shutdown() might be called very early during initialization, or
1435 // called from signal handler. Before adding something to os::shutdown(), make
1436 // sure it is async-safe and can handle partially initialized VM.
1437 void os::shutdown() {
1438 
1439   // allow PerfMemory to attempt cleanup of any persistent resources
1440   perfMemory_exit();
1441 
1442   // needs to remove object in file system
1443   AttachListener::abort();
1444 
1445   // flush buffered output, finish log files
1446   ostream_abort();
1447 
1448   // Check for abort hook
1449   abort_hook_t abort_hook = Arguments::abort_hook();
1450   if (abort_hook != NULL) {
1451     abort_hook();
1452   }
1453 
1454 }
1455 
1456 // Note: os::abort() might be called very early during initialization, or
1457 // called from signal handler. Before adding something to os::abort(), make
1458 // sure it is async-safe and can handle partially initialized VM.
1459 void os::abort(bool dump_core) {
1460   os::shutdown();
1461   if (dump_core) {
1462 #ifndef PRODUCT
1463     fdStream out(defaultStream::output_fd());
1464     out.print_raw("Current thread is ");
1465     char buf[16];
1466     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1467     out.print_raw_cr(buf);
1468     out.print_raw_cr("Dumping core ...");
1469 #endif
1470     ::abort(); // dump core
1471   }
1472 
1473   ::exit(1);
1474 }
1475 
1476 // Die immediately, no exit hook, no abort hook, no cleanup.
1477 void os::die() {
1478   // _exit() on LinuxThreads only kills current thread
1479   ::abort();
1480 }
1481 
1482 // unused on linux for now.
1483 void os::set_error_file(const char *logfile) {}
1484 
1485 intx os::current_thread_id() { return (intx)pthread_self(); }
1486 int os::current_process_id() {
1487 
1488   // Under the old linux thread library, linux gives each thread
1489   // its own process id. Because of this each thread will return
1490   // a different pid if this method were to return the result
1491   // of getpid(2). Linux provides no api that returns the pid
1492   // of the launcher thread for the vm. This implementation
1493   // returns a unique pid, the pid of the launcher thread
1494   // that starts the vm 'process'.
1495 
1496   // Under the NPTL, getpid() returns the same pid as the
1497   // launcher thread rather than a unique pid per thread.
1498   // Use gettid() if you want the old pre NPTL behaviour.
1499 
1500   // if you are looking for the result of a call to getpid() that
1501   // returns a unique pid for the calling thread, then look at the
1502   // OSThread::thread_id() method in osThread_linux.hpp file
1503 
1504   return (int)(_initial_pid ? _initial_pid : getpid());
1505 }
1506 
1507 // DLL functions
1508 
1509 const char* os::dll_file_extension() { return ".so"; }
1510 
1511 const char* os::get_temp_directory() { return "/tmp/"; }
1512 
1513 void os::dll_build_name(
1514     char* buffer, size_t buflen, const char* pname, const char* fname) {
1515   // copied from libhpi
1516   const size_t pnamelen = pname ? strlen(pname) : 0;
1517 
1518   /* Quietly truncate on buffer overflow.  Should be an error. */
1519   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1520       *buffer = '\0';
1521       return;
1522   }
1523 
1524   if (pnamelen == 0) {
1525       sprintf(buffer, "lib%s.so", fname);
1526   } else {
1527       sprintf(buffer, "%s/lib%s.so", pname, fname);
1528   }
1529 }
1530 
1531 const char* os::get_current_directory(char *buf, int buflen) {
1532   return getcwd(buf, buflen);
1533 }
1534 
1535 // check if addr is inside libjvm[_g].so
1536 bool os::address_is_in_vm(address addr) {
1537   static address libjvm_base_addr;
1538   Dl_info dlinfo;
1539 
1540   if (libjvm_base_addr == NULL) {
1541     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1542     libjvm_base_addr = (address)dlinfo.dli_fbase;
1543     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1544   }
1545 
1546   if (dladdr((void *)addr, &dlinfo)) {
1547     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1548   }
1549 
1550   return false;
1551 }
1552 
1553 bool os::dll_address_to_function_name(address addr, char *buf,
1554                                       int buflen, int *offset) {
1555   Dl_info dlinfo;
1556 
1557   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1558     if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1559     if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1560     return true;
1561   } else {
1562     if (buf) buf[0] = '\0';
1563     if (offset) *offset = -1;
1564     return false;
1565   }
1566 }
1567 
1568 struct _address_to_library_name {
1569   address addr;          // input : memory address
1570   size_t  buflen;        //         size of fname
1571   char*   fname;         // output: library name
1572   address base;          //         library base addr
1573 };
1574 
1575 static int address_to_library_name_callback(struct dl_phdr_info *info,
1576                                             size_t size, void *data) {
1577   int i;
1578   bool found = false;
1579   address libbase = NULL;
1580   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1581 
1582   // iterate through all loadable segments
1583   for (i = 0; i < info->dlpi_phnum; i++) {
1584     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1585     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1586       // base address of a library is the lowest address of its loaded
1587       // segments.
1588       if (libbase == NULL || libbase > segbase) {
1589         libbase = segbase;
1590       }
1591       // see if 'addr' is within current segment
1592       if (segbase <= d->addr &&
1593           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1594         found = true;
1595       }
1596     }
1597   }
1598 
1599   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1600   // so dll_address_to_library_name() can fall through to use dladdr() which
1601   // can figure out executable name from argv[0].
1602   if (found && info->dlpi_name && info->dlpi_name[0]) {
1603     d->base = libbase;
1604     if (d->fname) {
1605       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1606     }
1607     return 1;
1608   }
1609   return 0;
1610 }
1611 
1612 bool os::dll_address_to_library_name(address addr, char* buf,
1613                                      int buflen, int* offset) {
1614   Dl_info dlinfo;
1615   struct _address_to_library_name data;
1616 
1617   // There is a bug in old glibc dladdr() implementation that it could resolve
1618   // to wrong library name if the .so file has a base address != NULL. Here
1619   // we iterate through the program headers of all loaded libraries to find
1620   // out which library 'addr' really belongs to. This workaround can be
1621   // removed once the minimum requirement for glibc is moved to 2.3.x.
1622   data.addr = addr;
1623   data.fname = buf;
1624   data.buflen = buflen;
1625   data.base = NULL;
1626   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1627 
1628   if (rslt) {
1629      // buf already contains library name
1630      if (offset) *offset = addr - data.base;
1631      return true;
1632   } else if (dladdr((void*)addr, &dlinfo)){
1633      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1634      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1635      return true;
1636   } else {
1637      if (buf) buf[0] = '\0';
1638      if (offset) *offset = -1;
1639      return false;
1640   }
1641 }
1642 
1643   // Loads .dll/.so and
1644   // in case of error it checks if .dll/.so was built for the
1645   // same architecture as Hotspot is running on
1646 
1647 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1648 {
1649   void * result= ::dlopen(filename, RTLD_LAZY);
1650   if (result != NULL) {
1651     // Successful loading
1652     return result;
1653   }
1654 
1655   Elf32_Ehdr elf_head;
1656 
1657   // Read system error message into ebuf
1658   // It may or may not be overwritten below
1659   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1660   ebuf[ebuflen-1]='\0';
1661   int diag_msg_max_length=ebuflen-strlen(ebuf);
1662   char* diag_msg_buf=ebuf+strlen(ebuf);
1663 
1664   if (diag_msg_max_length==0) {
1665     // No more space in ebuf for additional diagnostics message
1666     return NULL;
1667   }
1668 
1669 
1670   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1671 
1672   if (file_descriptor < 0) {
1673     // Can't open library, report dlerror() message
1674     return NULL;
1675   }
1676 
1677   bool failed_to_read_elf_head=
1678     (sizeof(elf_head)!=
1679         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1680 
1681   ::close(file_descriptor);
1682   if (failed_to_read_elf_head) {
1683     // file i/o error - report dlerror() msg
1684     return NULL;
1685   }
1686 
1687   typedef struct {
1688     Elf32_Half  code;         // Actual value as defined in elf.h
1689     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1690     char        elf_class;    // 32 or 64 bit
1691     char        endianess;    // MSB or LSB
1692     char*       name;         // String representation
1693   } arch_t;
1694 
1695   #ifndef EM_486
1696   #define EM_486          6               /* Intel 80486 */
1697   #endif
1698 
1699   static const arch_t arch_array[]={
1700     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1701     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1702     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1703     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1704     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1705     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1706     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1707     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1708     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}
1709   };
1710 
1711   #if  (defined IA32)
1712     static  Elf32_Half running_arch_code=EM_386;
1713   #elif   (defined AMD64)
1714     static  Elf32_Half running_arch_code=EM_X86_64;
1715   #elif  (defined IA64)
1716     static  Elf32_Half running_arch_code=EM_IA_64;
1717   #elif  (defined __sparc) && (defined _LP64)
1718     static  Elf32_Half running_arch_code=EM_SPARCV9;
1719   #elif  (defined __sparc) && (!defined _LP64)
1720     static  Elf32_Half running_arch_code=EM_SPARC;
1721   #elif  (defined __powerpc64__)
1722     static  Elf32_Half running_arch_code=EM_PPC64;
1723   #elif  (defined __powerpc__)
1724     static  Elf32_Half running_arch_code=EM_PPC;
1725   #else
1726     #error Method os::dll_load requires that one of following is defined:\
1727          IA32, AMD64, IA64, __sparc, __powerpc__
1728   #endif
1729 
1730   // Identify compatability class for VM's architecture and library's architecture
1731   // Obtain string descriptions for architectures
1732 
1733   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1734   int running_arch_index=-1;
1735 
1736   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1737     if (running_arch_code == arch_array[i].code) {
1738       running_arch_index    = i;
1739     }
1740     if (lib_arch.code == arch_array[i].code) {
1741       lib_arch.compat_class = arch_array[i].compat_class;
1742       lib_arch.name         = arch_array[i].name;
1743     }
1744   }
1745 
1746   assert(running_arch_index != -1,
1747     "Didn't find running architecture code (running_arch_code) in arch_array");
1748   if (running_arch_index == -1) {
1749     // Even though running architecture detection failed
1750     // we may still continue with reporting dlerror() message
1751     return NULL;
1752   }
1753 
1754   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1755     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1756     return NULL;
1757   }
1758 
1759   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1760     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1761     return NULL;
1762   }
1763 
1764   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1765     if ( lib_arch.name!=NULL ) {
1766       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1767         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1768         lib_arch.name, arch_array[running_arch_index].name);
1769     } else {
1770       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1771       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1772         lib_arch.code,
1773         arch_array[running_arch_index].name);
1774     }
1775   }
1776 
1777   return NULL;
1778 }
1779 
1780 /*
1781  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
1782  * chances are you might want to run the generated bits against glibc-2.0
1783  * libdl.so, so always use locking for any version of glibc.
1784  */
1785 void* os::dll_lookup(void* handle, const char* name) {
1786   pthread_mutex_lock(&dl_mutex);
1787   void* res = dlsym(handle, name);
1788   pthread_mutex_unlock(&dl_mutex);
1789   return res;
1790 }
1791 
1792 
1793 bool _print_ascii_file(const char* filename, outputStream* st) {
1794   int fd = open(filename, O_RDONLY);
1795   if (fd == -1) {
1796      return false;
1797   }
1798 
1799   char buf[32];
1800   int bytes;
1801   while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1802     st->print_raw(buf, bytes);
1803   }
1804 
1805   close(fd);
1806 
1807   return true;
1808 }
1809 
1810 void os::print_dll_info(outputStream *st) {
1811    st->print_cr("Dynamic libraries:");
1812 
1813    char fname[32];
1814    pid_t pid = os::Linux::gettid();
1815 
1816    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1817 
1818    if (!_print_ascii_file(fname, st)) {
1819      st->print("Can not get library information for pid = %d\n", pid);
1820    }
1821 }
1822 
1823 
1824 void os::print_os_info(outputStream* st) {
1825   st->print("OS:");
1826 
1827   // Try to identify popular distros.
1828   // Most Linux distributions have /etc/XXX-release file, which contains
1829   // the OS version string. Some have more than one /etc/XXX-release file
1830   // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1831   // so the order is important.
1832   if (!_print_ascii_file("/etc/mandrake-release", st) &&
1833       !_print_ascii_file("/etc/sun-release", st) &&
1834       !_print_ascii_file("/etc/redhat-release", st) &&
1835       !_print_ascii_file("/etc/SuSE-release", st) &&
1836       !_print_ascii_file("/etc/turbolinux-release", st) &&
1837       !_print_ascii_file("/etc/gentoo-release", st) &&
1838       !_print_ascii_file("/etc/debian_version", st)) {
1839       st->print("Linux");
1840   }
1841   st->cr();
1842 
1843   // kernel
1844   st->print("uname:");
1845   struct utsname name;
1846   uname(&name);
1847   st->print(name.sysname); st->print(" ");
1848   st->print(name.release); st->print(" ");
1849   st->print(name.version); st->print(" ");
1850   st->print(name.machine);
1851   st->cr();
1852 
1853   // Print warning if unsafe chroot environment detected
1854   if (unsafe_chroot_detected) {
1855     st->print("WARNING!! ");
1856     st->print_cr(unstable_chroot_error);
1857   }
1858 
1859   // libc, pthread
1860   st->print("libc:");
1861   st->print(os::Linux::glibc_version()); st->print(" ");
1862   st->print(os::Linux::libpthread_version()); st->print(" ");
1863   if (os::Linux::is_LinuxThreads()) {
1864      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
1865   }
1866   st->cr();
1867 
1868   // rlimit
1869   st->print("rlimit:");
1870   struct rlimit rlim;
1871 
1872   st->print(" STACK ");
1873   getrlimit(RLIMIT_STACK, &rlim);
1874   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1875   else st->print("%uk", rlim.rlim_cur >> 10);
1876 
1877   st->print(", CORE ");
1878   getrlimit(RLIMIT_CORE, &rlim);
1879   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1880   else st->print("%uk", rlim.rlim_cur >> 10);
1881 
1882   st->print(", NPROC ");
1883   getrlimit(RLIMIT_NPROC, &rlim);
1884   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1885   else st->print("%d", rlim.rlim_cur);
1886 
1887   st->print(", NOFILE ");
1888   getrlimit(RLIMIT_NOFILE, &rlim);
1889   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1890   else st->print("%d", rlim.rlim_cur);
1891 
1892   st->print(", AS ");
1893   getrlimit(RLIMIT_AS, &rlim);
1894   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1895   else st->print("%uk", rlim.rlim_cur >> 10);
1896   st->cr();
1897 
1898   // load average
1899   st->print("load average:");
1900   double loadavg[3];
1901   os::loadavg(loadavg, 3);
1902   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1903   st->cr();
1904 }
1905 
1906 void os::print_memory_info(outputStream* st) {
1907 
1908   st->print("Memory:");
1909   st->print(" %dk page", os::vm_page_size()>>10);
1910 
1911   // values in struct sysinfo are "unsigned long"
1912   struct sysinfo si;
1913   sysinfo(&si);
1914 
1915   st->print(", physical " UINT64_FORMAT "k",
1916             os::physical_memory() >> 10);
1917   st->print("(" UINT64_FORMAT "k free)",
1918             os::available_memory() >> 10);
1919   st->print(", swap " UINT64_FORMAT "k",
1920             ((jlong)si.totalswap * si.mem_unit) >> 10);
1921   st->print("(" UINT64_FORMAT "k free)",
1922             ((jlong)si.freeswap * si.mem_unit) >> 10);
1923   st->cr();
1924 }
1925 
1926 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
1927 // but they're the same for all the linux arch that we support
1928 // and they're the same for solaris but there's no common place to put this.
1929 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
1930                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
1931                           "ILL_COPROC", "ILL_BADSTK" };
1932 
1933 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
1934                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
1935                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
1936 
1937 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
1938 
1939 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
1940 
1941 void os::print_siginfo(outputStream* st, void* siginfo) {
1942   st->print("siginfo:");
1943 
1944   const int buflen = 100;
1945   char buf[buflen];
1946   siginfo_t *si = (siginfo_t*)siginfo;
1947   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
1948   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
1949     st->print("si_errno=%s", buf);
1950   } else {
1951     st->print("si_errno=%d", si->si_errno);
1952   }
1953   const int c = si->si_code;
1954   assert(c > 0, "unexpected si_code");
1955   switch (si->si_signo) {
1956   case SIGILL:
1957     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
1958     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1959     break;
1960   case SIGFPE:
1961     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
1962     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1963     break;
1964   case SIGSEGV:
1965     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
1966     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1967     break;
1968   case SIGBUS:
1969     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
1970     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1971     break;
1972   default:
1973     st->print(", si_code=%d", si->si_code);
1974     // no si_addr
1975   }
1976 
1977   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
1978       UseSharedSpaces) {
1979     FileMapInfo* mapinfo = FileMapInfo::current_info();
1980     if (mapinfo->is_in_shared_space(si->si_addr)) {
1981       st->print("\n\nError accessing class data sharing archive."   \
1982                 " Mapped file inaccessible during execution, "      \
1983                 " possible disk/network problem.");
1984     }
1985   }
1986   st->cr();
1987 }
1988 
1989 
1990 static void print_signal_handler(outputStream* st, int sig,
1991                                  char* buf, size_t buflen);
1992 
1993 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1994   st->print_cr("Signal Handlers:");
1995   print_signal_handler(st, SIGSEGV, buf, buflen);
1996   print_signal_handler(st, SIGBUS , buf, buflen);
1997   print_signal_handler(st, SIGFPE , buf, buflen);
1998   print_signal_handler(st, SIGPIPE, buf, buflen);
1999   print_signal_handler(st, SIGXFSZ, buf, buflen);
2000   print_signal_handler(st, SIGILL , buf, buflen);
2001   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2002   print_signal_handler(st, SR_signum, buf, buflen);
2003   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2004   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2005   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2006   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2007 }
2008 
2009 static char saved_jvm_path[MAXPATHLEN] = {0};
2010 
2011 // Find the full path to the current module, libjvm.so or libjvm_g.so
2012 void os::jvm_path(char *buf, jint len) {
2013   // Error checking.
2014   if (len < MAXPATHLEN) {
2015     assert(false, "must use a large-enough buffer");
2016     buf[0] = '\0';
2017     return;
2018   }
2019   // Lazy resolve the path to current module.
2020   if (saved_jvm_path[0] != 0) {
2021     strcpy(buf, saved_jvm_path);
2022     return;
2023   }
2024 
2025   char dli_fname[MAXPATHLEN];
2026   bool ret = dll_address_to_library_name(
2027                 CAST_FROM_FN_PTR(address, os::jvm_path),
2028                 dli_fname, sizeof(dli_fname), NULL);
2029   assert(ret != 0, "cannot locate libjvm");
2030   if (realpath(dli_fname, buf) == NULL)
2031     return;
2032 
2033   if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2034     // Support for the gamma launcher.  Typical value for buf is
2035     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2036     // the right place in the string, then assume we are installed in a JDK and
2037     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2038     // up the path so it looks like libjvm.so is installed there (append a
2039     // fake suffix hotspot/libjvm.so).
2040     const char *p = buf + strlen(buf) - 1;
2041     for (int count = 0; p > buf && count < 5; ++count) {
2042       for (--p; p > buf && *p != '/'; --p)
2043         /* empty */ ;
2044     }
2045 
2046     if (strncmp(p, "/jre/lib/", 9) != 0) {
2047       // Look for JAVA_HOME in the environment.
2048       char* java_home_var = ::getenv("JAVA_HOME");
2049       if (java_home_var != NULL && java_home_var[0] != 0) {
2050         // Check the current module name "libjvm.so" or "libjvm_g.so".
2051         p = strrchr(buf, '/');
2052         assert(strstr(p, "/libjvm") == p, "invalid library name");
2053         p = strstr(p, "_g") ? "_g" : "";
2054 
2055         if (realpath(java_home_var, buf) == NULL)
2056           return;
2057         sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch);
2058         if (0 == access(buf, F_OK)) {
2059           // Use current module name "libjvm[_g].so" instead of
2060           // "libjvm"debug_only("_g")".so" since for fastdebug version
2061           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2062           // It is used when we are choosing the HPI library's name
2063           // "libhpi[_g].so" in hpi::initialize_get_interface().
2064           sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p);
2065         } else {
2066           // Go back to path of .so
2067           if (realpath(dli_fname, buf) == NULL)
2068             return;
2069         }
2070       }
2071     }
2072   }
2073 
2074   strcpy(saved_jvm_path, buf);
2075 }
2076 
2077 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2078   // no prefix required, not even "_"
2079 }
2080 
2081 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2082   // no suffix required
2083 }
2084 
2085 ////////////////////////////////////////////////////////////////////////////////
2086 // sun.misc.Signal support
2087 
2088 static volatile jint sigint_count = 0;
2089 
2090 static void
2091 UserHandler(int sig, void *siginfo, void *context) {
2092   // 4511530 - sem_post is serialized and handled by the manager thread. When
2093   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2094   // don't want to flood the manager thread with sem_post requests.
2095   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2096       return;
2097 
2098   // Ctrl-C is pressed during error reporting, likely because the error
2099   // handler fails to abort. Let VM die immediately.
2100   if (sig == SIGINT && is_error_reported()) {
2101      os::die();
2102   }
2103 
2104   os::signal_notify(sig);
2105 }
2106 
2107 void* os::user_handler() {
2108   return CAST_FROM_FN_PTR(void*, UserHandler);
2109 }
2110 
2111 extern "C" {
2112   typedef void (*sa_handler_t)(int);
2113   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2114 }
2115 
2116 void* os::signal(int signal_number, void* handler) {
2117   struct sigaction sigAct, oldSigAct;
2118 
2119   sigfillset(&(sigAct.sa_mask));
2120   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2121   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2122 
2123   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2124     // -1 means registration failed
2125     return (void *)-1;
2126   }
2127 
2128   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2129 }
2130 
2131 void os::signal_raise(int signal_number) {
2132   ::raise(signal_number);
2133 }
2134 
2135 /*
2136  * The following code is moved from os.cpp for making this
2137  * code platform specific, which it is by its very nature.
2138  */
2139 
2140 // Will be modified when max signal is changed to be dynamic
2141 int os::sigexitnum_pd() {
2142   return NSIG;
2143 }
2144 
2145 // a counter for each possible signal value
2146 static volatile jint pending_signals[NSIG+1] = { 0 };
2147 
2148 // Linux(POSIX) specific hand shaking semaphore.
2149 static sem_t sig_sem;
2150 
2151 void os::signal_init_pd() {
2152   // Initialize signal structures
2153   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2154 
2155   // Initialize signal semaphore
2156   ::sem_init(&sig_sem, 0, 0);
2157 }
2158 
2159 void os::signal_notify(int sig) {
2160   Atomic::inc(&pending_signals[sig]);
2161   ::sem_post(&sig_sem);
2162 }
2163 
2164 static int check_pending_signals(bool wait) {
2165   Atomic::store(0, &sigint_count);
2166   for (;;) {
2167     for (int i = 0; i < NSIG + 1; i++) {
2168       jint n = pending_signals[i];
2169       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2170         return i;
2171       }
2172     }
2173     if (!wait) {
2174       return -1;
2175     }
2176     JavaThread *thread = JavaThread::current();
2177     ThreadBlockInVM tbivm(thread);
2178 
2179     bool threadIsSuspended;
2180     do {
2181       thread->set_suspend_equivalent();
2182       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2183       ::sem_wait(&sig_sem);
2184 
2185       // were we externally suspended while we were waiting?
2186       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2187       if (threadIsSuspended) {
2188         //
2189         // The semaphore has been incremented, but while we were waiting
2190         // another thread suspended us. We don't want to continue running
2191         // while suspended because that would surprise the thread that
2192         // suspended us.
2193         //
2194         ::sem_post(&sig_sem);
2195 
2196         thread->java_suspend_self();
2197       }
2198     } while (threadIsSuspended);
2199   }
2200 }
2201 
2202 int os::signal_lookup() {
2203   return check_pending_signals(false);
2204 }
2205 
2206 int os::signal_wait() {
2207   return check_pending_signals(true);
2208 }
2209 
2210 ////////////////////////////////////////////////////////////////////////////////
2211 // Virtual Memory
2212 
2213 int os::vm_page_size() {
2214   // Seems redundant as all get out
2215   assert(os::Linux::page_size() != -1, "must call os::init");
2216   return os::Linux::page_size();
2217 }
2218 
2219 // Solaris allocates memory by pages.
2220 int os::vm_allocation_granularity() {
2221   assert(os::Linux::page_size() != -1, "must call os::init");
2222   return os::Linux::page_size();
2223 }
2224 
2225 // Rationale behind this function:
2226 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2227 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2228 //  samples for JITted code. Here we create private executable mapping over the code cache
2229 //  and then we can use standard (well, almost, as mapping can change) way to provide
2230 //  info for the reporting script by storing timestamp and location of symbol
2231 void linux_wrap_code(char* base, size_t size) {
2232   static volatile jint cnt = 0;
2233 
2234   if (!UseOprofile) {
2235     return;
2236   }
2237 
2238   char buf[40];
2239   int num = Atomic::add(1, &cnt);
2240 
2241   sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num);
2242   unlink(buf);
2243 
2244   int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2245 
2246   if (fd != -1) {
2247     off_t rv = lseek(fd, size-2, SEEK_SET);
2248     if (rv != (off_t)-1) {
2249       if (write(fd, "", 1) == 1) {
2250         mmap(base, size,
2251              PROT_READ|PROT_WRITE|PROT_EXEC,
2252              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2253       }
2254     }
2255     close(fd);
2256     unlink(buf);
2257   }
2258 }
2259 
2260 // NOTE: Linux kernel does not really reserve the pages for us.
2261 //       All it does is to check if there are enough free pages
2262 //       left at the time of mmap(). This could be a potential
2263 //       problem.
2264 bool os::commit_memory(char* addr, size_t size) {
2265   uintptr_t res = (uintptr_t) ::mmap(addr, size,
2266                                    PROT_READ|PROT_WRITE|PROT_EXEC,
2267                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2268   return res != (uintptr_t) MAP_FAILED;
2269 }
2270 
2271 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint) {
2272   return commit_memory(addr, size);
2273 }
2274 
2275 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2276 
2277 void os::free_memory(char *addr, size_t bytes) {
2278   uncommit_memory(addr, bytes);
2279 }
2280 
2281 void os::numa_make_global(char *addr, size_t bytes) {
2282   Linux::numa_interleave_memory(addr, bytes);
2283 }
2284 
2285 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2286   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2287 }
2288 
2289 bool os::numa_topology_changed()   { return false; }
2290 
2291 size_t os::numa_get_groups_num() {
2292   int max_node = Linux::numa_max_node();
2293   return max_node > 0 ? max_node + 1 : 1;
2294 }
2295 
2296 int os::numa_get_group_id() {
2297   int cpu_id = Linux::sched_getcpu();
2298   if (cpu_id != -1) {
2299     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2300     if (lgrp_id != -1) {
2301       return lgrp_id;
2302     }
2303   }
2304   return 0;
2305 }
2306 
2307 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2308   for (size_t i = 0; i < size; i++) {
2309     ids[i] = i;
2310   }
2311   return size;
2312 }
2313 
2314 bool os::get_page_info(char *start, page_info* info) {
2315   return false;
2316 }
2317 
2318 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2319   return end;
2320 }
2321 
2322 extern "C" void numa_warn(int number, char *where, ...) { }
2323 extern "C" void numa_error(char *where) { }
2324 
2325 bool os::Linux::libnuma_init() {
2326   // sched_getcpu() should be in libc.
2327   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2328                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2329 
2330   if (sched_getcpu() != -1) { // Does it work?
2331     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2332     if (handle != NULL) {
2333       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2334                                            dlsym(handle, "numa_node_to_cpus")));
2335       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2336                                        dlsym(handle, "numa_max_node")));
2337       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2338                                         dlsym(handle, "numa_available")));
2339       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2340                                             dlsym(handle, "numa_tonode_memory")));
2341       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2342                                             dlsym(handle, "numa_interleave_memory")));
2343 
2344 
2345       if (numa_available() != -1) {
2346         set_numa_all_nodes((unsigned long*)dlsym(handle, "numa_all_nodes"));
2347         // Create a cpu -> node mapping
2348         _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2349         rebuild_cpu_to_node_map();
2350         return true;
2351       }
2352     }
2353   }
2354   return false;
2355 }
2356 
2357 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2358 // The table is later used in get_node_by_cpu().
2359 void os::Linux::rebuild_cpu_to_node_map() {
2360   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2361                               // in libnuma (possible values are starting from 16,
2362                               // and continuing up with every other power of 2, but less
2363                               // than the maximum number of CPUs supported by kernel), and
2364                               // is a subject to change (in libnuma version 2 the requirements
2365                               // are more reasonable) we'll just hardcode the number they use
2366                               // in the library.
2367   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2368 
2369   size_t cpu_num = os::active_processor_count();
2370   size_t cpu_map_size = NCPUS / BitsPerCLong;
2371   size_t cpu_map_valid_size =
2372     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2373 
2374   cpu_to_node()->clear();
2375   cpu_to_node()->at_grow(cpu_num - 1);
2376   size_t node_num = numa_get_groups_num();
2377 
2378   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2379   for (size_t i = 0; i < node_num; i++) {
2380     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2381       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2382         if (cpu_map[j] != 0) {
2383           for (size_t k = 0; k < BitsPerCLong; k++) {
2384             if (cpu_map[j] & (1UL << k)) {
2385               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2386             }
2387           }
2388         }
2389       }
2390     }
2391   }
2392   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2393 }
2394 
2395 int os::Linux::get_node_by_cpu(int cpu_id) {
2396   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2397     return cpu_to_node()->at(cpu_id);
2398   }
2399   return -1;
2400 }
2401 
2402 GrowableArray<int>* os::Linux::_cpu_to_node;
2403 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2404 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2405 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2406 os::Linux::numa_available_func_t os::Linux::_numa_available;
2407 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2408 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2409 unsigned long* os::Linux::_numa_all_nodes;
2410 
2411 bool os::uncommit_memory(char* addr, size_t size) {
2412   return ::mmap(addr, size,
2413                 PROT_READ|PROT_WRITE|PROT_EXEC,
2414                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0)
2415     != MAP_FAILED;
2416 }
2417 
2418 static address _highest_vm_reserved_address = NULL;
2419 
2420 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2421 // at 'requested_addr'. If there are existing memory mappings at the same
2422 // location, however, they will be overwritten. If 'fixed' is false,
2423 // 'requested_addr' is only treated as a hint, the return value may or
2424 // may not start from the requested address. Unlike Linux mmap(), this
2425 // function returns NULL to indicate failure.
2426 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2427   char * addr;
2428   int flags;
2429 
2430   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2431   if (fixed) {
2432     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2433     flags |= MAP_FIXED;
2434   }
2435 
2436   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE|PROT_EXEC,
2437                        flags, -1, 0);
2438 
2439   if (addr != MAP_FAILED) {
2440     // anon_mmap() should only get called during VM initialization,
2441     // don't need lock (actually we can skip locking even it can be called
2442     // from multiple threads, because _highest_vm_reserved_address is just a
2443     // hint about the upper limit of non-stack memory regions.)
2444     if ((address)addr + bytes > _highest_vm_reserved_address) {
2445       _highest_vm_reserved_address = (address)addr + bytes;
2446     }
2447   }
2448 
2449   return addr == MAP_FAILED ? NULL : addr;
2450 }
2451 
2452 // Don't update _highest_vm_reserved_address, because there might be memory
2453 // regions above addr + size. If so, releasing a memory region only creates
2454 // a hole in the address space, it doesn't help prevent heap-stack collision.
2455 //
2456 static int anon_munmap(char * addr, size_t size) {
2457   return ::munmap(addr, size) == 0;
2458 }
2459 
2460 char* os::reserve_memory(size_t bytes, char* requested_addr,
2461                          size_t alignment_hint) {
2462   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2463 }
2464 
2465 bool os::release_memory(char* addr, size_t size) {
2466   return anon_munmap(addr, size);
2467 }
2468 
2469 static address highest_vm_reserved_address() {
2470   return _highest_vm_reserved_address;
2471 }
2472 
2473 static bool linux_mprotect(char* addr, size_t size, int prot) {
2474   // Linux wants the mprotect address argument to be page aligned.
2475   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2476 
2477   // According to SUSv3, mprotect() should only be used with mappings
2478   // established by mmap(), and mmap() always maps whole pages. Unaligned
2479   // 'addr' likely indicates problem in the VM (e.g. trying to change
2480   // protection of malloc'ed or statically allocated memory). Check the
2481   // caller if you hit this assert.
2482   assert(addr == bottom, "sanity check");
2483 
2484   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2485   return ::mprotect(bottom, size, prot) == 0;
2486 }
2487 
2488 // Set protections specified
2489 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2490                         bool is_committed) {
2491   unsigned int p = 0;
2492   switch (prot) {
2493   case MEM_PROT_NONE: p = PROT_NONE; break;
2494   case MEM_PROT_READ: p = PROT_READ; break;
2495   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2496   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2497   default:
2498     ShouldNotReachHere();
2499   }
2500   // is_committed is unused.
2501   return linux_mprotect(addr, bytes, p);
2502 }
2503 
2504 bool os::guard_memory(char* addr, size_t size) {
2505   return linux_mprotect(addr, size, PROT_NONE);
2506 }
2507 
2508 bool os::unguard_memory(char* addr, size_t size) {
2509   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2510 }
2511 
2512 // Large page support
2513 
2514 static size_t _large_page_size = 0;
2515 
2516 bool os::large_page_init() {
2517   if (!UseLargePages) return false;
2518 
2519   if (LargePageSizeInBytes) {
2520     _large_page_size = LargePageSizeInBytes;
2521   } else {
2522     // large_page_size on Linux is used to round up heap size. x86 uses either
2523     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2524     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2525     // page as large as 256M.
2526     //
2527     // Here we try to figure out page size by parsing /proc/meminfo and looking
2528     // for a line with the following format:
2529     //    Hugepagesize:     2048 kB
2530     //
2531     // If we can't determine the value (e.g. /proc is not mounted, or the text
2532     // format has been changed), we'll use the largest page size supported by
2533     // the processor.
2534 
2535     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M);
2536 
2537     FILE *fp = fopen("/proc/meminfo", "r");
2538     if (fp) {
2539       while (!feof(fp)) {
2540         int x = 0;
2541         char buf[16];
2542         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2543           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2544             _large_page_size = x * K;
2545             break;
2546           }
2547         } else {
2548           // skip to next line
2549           for (;;) {
2550             int ch = fgetc(fp);
2551             if (ch == EOF || ch == (int)'\n') break;
2552           }
2553         }
2554       }
2555       fclose(fp);
2556     }
2557   }
2558 
2559   const size_t default_page_size = (size_t)Linux::page_size();
2560   if (_large_page_size > default_page_size) {
2561     _page_sizes[0] = _large_page_size;
2562     _page_sizes[1] = default_page_size;
2563     _page_sizes[2] = 0;
2564   }
2565 
2566   // Large page support is available on 2.6 or newer kernel, some vendors
2567   // (e.g. Redhat) have backported it to their 2.4 based distributions.
2568   // We optimistically assume the support is available. If later it turns out
2569   // not true, VM will automatically switch to use regular page size.
2570   return true;
2571 }
2572 
2573 #ifndef SHM_HUGETLB
2574 #define SHM_HUGETLB 04000
2575 #endif
2576 
2577 char* os::reserve_memory_special(size_t bytes) {
2578   assert(UseLargePages, "only for large pages");
2579 
2580   key_t key = IPC_PRIVATE;
2581   char *addr;
2582 
2583   bool warn_on_failure = UseLargePages &&
2584                         (!FLAG_IS_DEFAULT(UseLargePages) ||
2585                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2586                         );
2587   char msg[128];
2588 
2589   // Create a large shared memory region to attach to based on size.
2590   // Currently, size is the total size of the heap
2591   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2592   if (shmid == -1) {
2593      // Possible reasons for shmget failure:
2594      // 1. shmmax is too small for Java heap.
2595      //    > check shmmax value: cat /proc/sys/kernel/shmmax
2596      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2597      // 2. not enough large page memory.
2598      //    > check available large pages: cat /proc/meminfo
2599      //    > increase amount of large pages:
2600      //          echo new_value > /proc/sys/vm/nr_hugepages
2601      //      Note 1: different Linux may use different name for this property,
2602      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
2603      //      Note 2: it's possible there's enough physical memory available but
2604      //            they are so fragmented after a long run that they can't
2605      //            coalesce into large pages. Try to reserve large pages when
2606      //            the system is still "fresh".
2607      if (warn_on_failure) {
2608        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2609        warning(msg);
2610      }
2611      return NULL;
2612   }
2613 
2614   // attach to the region
2615   addr = (char*)shmat(shmid, NULL, 0);
2616   int err = errno;
2617 
2618   // Remove shmid. If shmat() is successful, the actual shared memory segment
2619   // will be deleted when it's detached by shmdt() or when the process
2620   // terminates. If shmat() is not successful this will remove the shared
2621   // segment immediately.
2622   shmctl(shmid, IPC_RMID, NULL);
2623 
2624   if ((intptr_t)addr == -1) {
2625      if (warn_on_failure) {
2626        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2627        warning(msg);
2628      }
2629      return NULL;
2630   }
2631 
2632   return addr;
2633 }
2634 
2635 bool os::release_memory_special(char* base, size_t bytes) {
2636   // detaching the SHM segment will also delete it, see reserve_memory_special()
2637   int rslt = shmdt(base);
2638   return rslt == 0;
2639 }
2640 
2641 size_t os::large_page_size() {
2642   return _large_page_size;
2643 }
2644 
2645 // Linux does not support anonymous mmap with large page memory. The only way
2646 // to reserve large page memory without file backing is through SysV shared
2647 // memory API. The entire memory region is committed and pinned upfront.
2648 // Hopefully this will change in the future...
2649 bool os::can_commit_large_page_memory() {
2650   return false;
2651 }
2652 
2653 bool os::can_execute_large_page_memory() {
2654   return false;
2655 }
2656 
2657 // Reserve memory at an arbitrary address, only if that area is
2658 // available (and not reserved for something else).
2659 
2660 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2661   const int max_tries = 10;
2662   char* base[max_tries];
2663   size_t size[max_tries];
2664   const size_t gap = 0x000000;
2665 
2666   // Assert only that the size is a multiple of the page size, since
2667   // that's all that mmap requires, and since that's all we really know
2668   // about at this low abstraction level.  If we need higher alignment,
2669   // we can either pass an alignment to this method or verify alignment
2670   // in one of the methods further up the call chain.  See bug 5044738.
2671   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2672 
2673   // Repeatedly allocate blocks until the block is allocated at the
2674   // right spot. Give up after max_tries. Note that reserve_memory() will
2675   // automatically update _highest_vm_reserved_address if the call is
2676   // successful. The variable tracks the highest memory address every reserved
2677   // by JVM. It is used to detect heap-stack collision if running with
2678   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2679   // space than needed, it could confuse the collision detecting code. To
2680   // solve the problem, save current _highest_vm_reserved_address and
2681   // calculate the correct value before return.
2682   address old_highest = _highest_vm_reserved_address;
2683 
2684   // Linux mmap allows caller to pass an address as hint; give it a try first,
2685   // if kernel honors the hint then we can return immediately.
2686   char * addr = anon_mmap(requested_addr, bytes, false);
2687   if (addr == requested_addr) {
2688      return requested_addr;
2689   }
2690 
2691   if (addr != NULL) {
2692      // mmap() is successful but it fails to reserve at the requested address
2693      anon_munmap(addr, bytes);
2694   }
2695 
2696   int i;
2697   for (i = 0; i < max_tries; ++i) {
2698     base[i] = reserve_memory(bytes);
2699 
2700     if (base[i] != NULL) {
2701       // Is this the block we wanted?
2702       if (base[i] == requested_addr) {
2703         size[i] = bytes;
2704         break;
2705       }
2706 
2707       // Does this overlap the block we wanted? Give back the overlapped
2708       // parts and try again.
2709 
2710       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2711       if (top_overlap >= 0 && top_overlap < bytes) {
2712         unmap_memory(base[i], top_overlap);
2713         base[i] += top_overlap;
2714         size[i] = bytes - top_overlap;
2715       } else {
2716         size_t bottom_overlap = base[i] + bytes - requested_addr;
2717         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2718           unmap_memory(requested_addr, bottom_overlap);
2719           size[i] = bytes - bottom_overlap;
2720         } else {
2721           size[i] = bytes;
2722         }
2723       }
2724     }
2725   }
2726 
2727   // Give back the unused reserved pieces.
2728 
2729   for (int j = 0; j < i; ++j) {
2730     if (base[j] != NULL) {
2731       unmap_memory(base[j], size[j]);
2732     }
2733   }
2734 
2735   if (i < max_tries) {
2736     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2737     return requested_addr;
2738   } else {
2739     _highest_vm_reserved_address = old_highest;
2740     return NULL;
2741   }
2742 }
2743 
2744 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2745   return ::read(fd, buf, nBytes);
2746 }
2747 
2748 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
2749 // Solaris uses poll(), linux uses park().
2750 // Poll() is likely a better choice, assuming that Thread.interrupt()
2751 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2752 // SIGSEGV, see 4355769.
2753 
2754 const int NANOSECS_PER_MILLISECS = 1000000;
2755 
2756 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2757   assert(thread == Thread::current(),  "thread consistency check");
2758 
2759   ParkEvent * const slp = thread->_SleepEvent ;
2760   slp->reset() ;
2761   OrderAccess::fence() ;
2762 
2763   if (interruptible) {
2764     jlong prevtime = javaTimeNanos();
2765 
2766     for (;;) {
2767       if (os::is_interrupted(thread, true)) {
2768         return OS_INTRPT;
2769       }
2770 
2771       jlong newtime = javaTimeNanos();
2772 
2773       if (newtime - prevtime < 0) {
2774         // time moving backwards, should only happen if no monotonic clock
2775         // not a guarantee() because JVM should not abort on kernel/glibc bugs
2776         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2777       } else {
2778         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2779       }
2780 
2781       if(millis <= 0) {
2782         return OS_OK;
2783       }
2784 
2785       prevtime = newtime;
2786 
2787       {
2788         assert(thread->is_Java_thread(), "sanity check");
2789         JavaThread *jt = (JavaThread *) thread;
2790         ThreadBlockInVM tbivm(jt);
2791         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
2792 
2793         jt->set_suspend_equivalent();
2794         // cleared by handle_special_suspend_equivalent_condition() or
2795         // java_suspend_self() via check_and_wait_while_suspended()
2796 
2797         slp->park(millis);
2798 
2799         // were we externally suspended while we were waiting?
2800         jt->check_and_wait_while_suspended();
2801       }
2802     }
2803   } else {
2804     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2805     jlong prevtime = javaTimeNanos();
2806 
2807     for (;;) {
2808       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
2809       // the 1st iteration ...
2810       jlong newtime = javaTimeNanos();
2811 
2812       if (newtime - prevtime < 0) {
2813         // time moving backwards, should only happen if no monotonic clock
2814         // not a guarantee() because JVM should not abort on kernel/glibc bugs
2815         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2816       } else {
2817         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2818       }
2819 
2820       if(millis <= 0) break ;
2821 
2822       prevtime = newtime;
2823       slp->park(millis);
2824     }
2825     return OS_OK ;
2826   }
2827 }
2828 
2829 int os::naked_sleep() {
2830   // %% make the sleep time an integer flag. for now use 1 millisec.
2831   return os::sleep(Thread::current(), 1, false);
2832 }
2833 
2834 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2835 void os::infinite_sleep() {
2836   while (true) {    // sleep forever ...
2837     ::sleep(100);   // ... 100 seconds at a time
2838   }
2839 }
2840 
2841 // Used to convert frequent JVM_Yield() to nops
2842 bool os::dont_yield() {
2843   return DontYieldALot;
2844 }
2845 
2846 void os::yield() {
2847   sched_yield();
2848 }
2849 
2850 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
2851 
2852 void os::yield_all(int attempts) {
2853   // Yields to all threads, including threads with lower priorities
2854   // Threads on Linux are all with same priority. The Solaris style
2855   // os::yield_all() with nanosleep(1ms) is not necessary.
2856   sched_yield();
2857 }
2858 
2859 // Called from the tight loops to possibly influence time-sharing heuristics
2860 void os::loop_breaker(int attempts) {
2861   os::yield_all(attempts);
2862 }
2863 
2864 ////////////////////////////////////////////////////////////////////////////////
2865 // thread priority support
2866 
2867 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
2868 // only supports dynamic priority, static priority must be zero. For real-time
2869 // applications, Linux supports SCHED_RR which allows static priority (1-99).
2870 // However, for large multi-threaded applications, SCHED_RR is not only slower
2871 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
2872 // of 5 runs - Sep 2005).
2873 //
2874 // The following code actually changes the niceness of kernel-thread/LWP. It
2875 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
2876 // not the entire user process, and user level threads are 1:1 mapped to kernel
2877 // threads. It has always been the case, but could change in the future. For
2878 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
2879 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
2880 
2881 int os::java_to_os_priority[MaxPriority + 1] = {
2882   19,              // 0 Entry should never be used
2883 
2884    4,              // 1 MinPriority
2885    3,              // 2
2886    2,              // 3
2887 
2888    1,              // 4
2889    0,              // 5 NormPriority
2890   -1,              // 6
2891 
2892   -2,              // 7
2893   -3,              // 8
2894   -4,              // 9 NearMaxPriority
2895 
2896   -5               // 10 MaxPriority
2897 };
2898 
2899 static int prio_init() {
2900   if (ThreadPriorityPolicy == 1) {
2901     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
2902     // if effective uid is not root. Perhaps, a more elegant way of doing
2903     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
2904     if (geteuid() != 0) {
2905       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
2906         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
2907       }
2908       ThreadPriorityPolicy = 0;
2909     }
2910   }
2911   return 0;
2912 }
2913 
2914 OSReturn os::set_native_priority(Thread* thread, int newpri) {
2915   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
2916 
2917   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
2918   return (ret == 0) ? OS_OK : OS_ERR;
2919 }
2920 
2921 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
2922   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
2923     *priority_ptr = java_to_os_priority[NormPriority];
2924     return OS_OK;
2925   }
2926 
2927   errno = 0;
2928   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
2929   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
2930 }
2931 
2932 // Hint to the underlying OS that a task switch would not be good.
2933 // Void return because it's a hint and can fail.
2934 void os::hint_no_preempt() {}
2935 
2936 ////////////////////////////////////////////////////////////////////////////////
2937 // suspend/resume support
2938 
2939 //  the low-level signal-based suspend/resume support is a remnant from the
2940 //  old VM-suspension that used to be for java-suspension, safepoints etc,
2941 //  within hotspot. Now there is a single use-case for this:
2942 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
2943 //      that runs in the watcher thread.
2944 //  The remaining code is greatly simplified from the more general suspension
2945 //  code that used to be used.
2946 //
2947 //  The protocol is quite simple:
2948 //  - suspend:
2949 //      - sends a signal to the target thread
2950 //      - polls the suspend state of the osthread using a yield loop
2951 //      - target thread signal handler (SR_handler) sets suspend state
2952 //        and blocks in sigsuspend until continued
2953 //  - resume:
2954 //      - sets target osthread state to continue
2955 //      - sends signal to end the sigsuspend loop in the SR_handler
2956 //
2957 //  Note that the SR_lock plays no role in this suspend/resume protocol.
2958 //
2959 
2960 static void resume_clear_context(OSThread *osthread) {
2961   osthread->set_ucontext(NULL);
2962   osthread->set_siginfo(NULL);
2963 
2964   // notify the suspend action is completed, we have now resumed
2965   osthread->sr.clear_suspended();
2966 }
2967 
2968 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
2969   osthread->set_ucontext(context);
2970   osthread->set_siginfo(siginfo);
2971 }
2972 
2973 //
2974 // Handler function invoked when a thread's execution is suspended or
2975 // resumed. We have to be careful that only async-safe functions are
2976 // called here (Note: most pthread functions are not async safe and
2977 // should be avoided.)
2978 //
2979 // Note: sigwait() is a more natural fit than sigsuspend() from an
2980 // interface point of view, but sigwait() prevents the signal hander
2981 // from being run. libpthread would get very confused by not having
2982 // its signal handlers run and prevents sigwait()'s use with the
2983 // mutex granting granting signal.
2984 //
2985 // Currently only ever called on the VMThread
2986 //
2987 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
2988   // Save and restore errno to avoid confusing native code with EINTR
2989   // after sigsuspend.
2990   int old_errno = errno;
2991 
2992   Thread* thread = Thread::current();
2993   OSThread* osthread = thread->osthread();
2994   assert(thread->is_VM_thread(), "Must be VMThread");
2995   // read current suspend action
2996   int action = osthread->sr.suspend_action();
2997   if (action == SR_SUSPEND) {
2998     suspend_save_context(osthread, siginfo, context);
2999 
3000     // Notify the suspend action is about to be completed. do_suspend()
3001     // waits until SR_SUSPENDED is set and then returns. We will wait
3002     // here for a resume signal and that completes the suspend-other
3003     // action. do_suspend/do_resume is always called as a pair from
3004     // the same thread - so there are no races
3005 
3006     // notify the caller
3007     osthread->sr.set_suspended();
3008 
3009     sigset_t suspend_set;  // signals for sigsuspend()
3010 
3011     // get current set of blocked signals and unblock resume signal
3012     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3013     sigdelset(&suspend_set, SR_signum);
3014 
3015     // wait here until we are resumed
3016     do {
3017       sigsuspend(&suspend_set);
3018       // ignore all returns until we get a resume signal
3019     } while (osthread->sr.suspend_action() != SR_CONTINUE);
3020 
3021     resume_clear_context(osthread);
3022 
3023   } else {
3024     assert(action == SR_CONTINUE, "unexpected sr action");
3025     // nothing special to do - just leave the handler
3026   }
3027 
3028   errno = old_errno;
3029 }
3030 
3031 
3032 static int SR_initialize() {
3033   struct sigaction act;
3034   char *s;
3035   /* Get signal number to use for suspend/resume */
3036   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3037     int sig = ::strtol(s, 0, 10);
3038     if (sig > 0 || sig < _NSIG) {
3039         SR_signum = sig;
3040     }
3041   }
3042 
3043   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3044         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3045 
3046   sigemptyset(&SR_sigset);
3047   sigaddset(&SR_sigset, SR_signum);
3048 
3049   /* Set up signal handler for suspend/resume */
3050   act.sa_flags = SA_RESTART|SA_SIGINFO;
3051   act.sa_handler = (void (*)(int)) SR_handler;
3052 
3053   // SR_signum is blocked by default.
3054   // 4528190 - We also need to block pthread restart signal (32 on all
3055   // supported Linux platforms). Note that LinuxThreads need to block
3056   // this signal for all threads to work properly. So we don't have
3057   // to use hard-coded signal number when setting up the mask.
3058   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3059 
3060   if (sigaction(SR_signum, &act, 0) == -1) {
3061     return -1;
3062   }
3063 
3064   // Save signal flag
3065   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3066   return 0;
3067 }
3068 
3069 static int SR_finalize() {
3070   return 0;
3071 }
3072 
3073 
3074 // returns true on success and false on error - really an error is fatal
3075 // but this seems the normal response to library errors
3076 static bool do_suspend(OSThread* osthread) {
3077   // mark as suspended and send signal
3078   osthread->sr.set_suspend_action(SR_SUSPEND);
3079   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3080   assert_status(status == 0, status, "pthread_kill");
3081 
3082   // check status and wait until notified of suspension
3083   if (status == 0) {
3084     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3085       os::yield_all(i);
3086     }
3087     osthread->sr.set_suspend_action(SR_NONE);
3088     return true;
3089   }
3090   else {
3091     osthread->sr.set_suspend_action(SR_NONE);
3092     return false;
3093   }
3094 }
3095 
3096 static void do_resume(OSThread* osthread) {
3097   assert(osthread->sr.is_suspended(), "thread should be suspended");
3098   osthread->sr.set_suspend_action(SR_CONTINUE);
3099 
3100   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3101   assert_status(status == 0, status, "pthread_kill");
3102   // check status and wait unit notified of resumption
3103   if (status == 0) {
3104     for (int i = 0; osthread->sr.is_suspended(); i++) {
3105       os::yield_all(i);
3106     }
3107   }
3108   osthread->sr.set_suspend_action(SR_NONE);
3109 }
3110 
3111 ////////////////////////////////////////////////////////////////////////////////
3112 // interrupt support
3113 
3114 void os::interrupt(Thread* thread) {
3115   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3116     "possibility of dangling Thread pointer");
3117 
3118   OSThread* osthread = thread->osthread();
3119 
3120   if (!osthread->interrupted()) {
3121     osthread->set_interrupted(true);
3122     // More than one thread can get here with the same value of osthread,
3123     // resulting in multiple notifications.  We do, however, want the store
3124     // to interrupted() to be visible to other threads before we execute unpark().
3125     OrderAccess::fence();
3126     ParkEvent * const slp = thread->_SleepEvent ;
3127     if (slp != NULL) slp->unpark() ;
3128   }
3129 
3130   // For JSR166. Unpark even if interrupt status already was set
3131   if (thread->is_Java_thread())
3132     ((JavaThread*)thread)->parker()->unpark();
3133 
3134   ParkEvent * ev = thread->_ParkEvent ;
3135   if (ev != NULL) ev->unpark() ;
3136 
3137 }
3138 
3139 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3140   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3141     "possibility of dangling Thread pointer");
3142 
3143   OSThread* osthread = thread->osthread();
3144 
3145   bool interrupted = osthread->interrupted();
3146 
3147   if (interrupted && clear_interrupted) {
3148     osthread->set_interrupted(false);
3149     // consider thread->_SleepEvent->reset() ... optional optimization
3150   }
3151 
3152   return interrupted;
3153 }
3154 
3155 ///////////////////////////////////////////////////////////////////////////////////
3156 // signal handling (except suspend/resume)
3157 
3158 // This routine may be used by user applications as a "hook" to catch signals.
3159 // The user-defined signal handler must pass unrecognized signals to this
3160 // routine, and if it returns true (non-zero), then the signal handler must
3161 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3162 // routine will never retun false (zero), but instead will execute a VM panic
3163 // routine kill the process.
3164 //
3165 // If this routine returns false, it is OK to call it again.  This allows
3166 // the user-defined signal handler to perform checks either before or after
3167 // the VM performs its own checks.  Naturally, the user code would be making
3168 // a serious error if it tried to handle an exception (such as a null check
3169 // or breakpoint) that the VM was generating for its own correct operation.
3170 //
3171 // This routine may recognize any of the following kinds of signals:
3172 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3173 // It should be consulted by handlers for any of those signals.
3174 //
3175 // The caller of this routine must pass in the three arguments supplied
3176 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3177 // field of the structure passed to sigaction().  This routine assumes that
3178 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3179 //
3180 // Note that the VM will print warnings if it detects conflicting signal
3181 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3182 //
3183 extern "C" int
3184 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3185                         void* ucontext, int abort_if_unrecognized);
3186 
3187 void signalHandler(int sig, siginfo_t* info, void* uc) {
3188   assert(info != NULL && uc != NULL, "it must be old kernel");
3189   JVM_handle_linux_signal(sig, info, uc, true);
3190 }
3191 
3192 
3193 // This boolean allows users to forward their own non-matching signals
3194 // to JVM_handle_linux_signal, harmlessly.
3195 bool os::Linux::signal_handlers_are_installed = false;
3196 
3197 // For signal-chaining
3198 struct sigaction os::Linux::sigact[MAXSIGNUM];
3199 unsigned int os::Linux::sigs = 0;
3200 bool os::Linux::libjsig_is_loaded = false;
3201 typedef struct sigaction *(*get_signal_t)(int);
3202 get_signal_t os::Linux::get_signal_action = NULL;
3203 
3204 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3205   struct sigaction *actp = NULL;
3206 
3207   if (libjsig_is_loaded) {
3208     // Retrieve the old signal handler from libjsig
3209     actp = (*get_signal_action)(sig);
3210   }
3211   if (actp == NULL) {
3212     // Retrieve the preinstalled signal handler from jvm
3213     actp = get_preinstalled_handler(sig);
3214   }
3215 
3216   return actp;
3217 }
3218 
3219 static bool call_chained_handler(struct sigaction *actp, int sig,
3220                                  siginfo_t *siginfo, void *context) {
3221   // Call the old signal handler
3222   if (actp->sa_handler == SIG_DFL) {
3223     // It's more reasonable to let jvm treat it as an unexpected exception
3224     // instead of taking the default action.
3225     return false;
3226   } else if (actp->sa_handler != SIG_IGN) {
3227     if ((actp->sa_flags & SA_NODEFER) == 0) {
3228       // automaticlly block the signal
3229       sigaddset(&(actp->sa_mask), sig);
3230     }
3231 
3232     sa_handler_t hand;
3233     sa_sigaction_t sa;
3234     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3235     // retrieve the chained handler
3236     if (siginfo_flag_set) {
3237       sa = actp->sa_sigaction;
3238     } else {
3239       hand = actp->sa_handler;
3240     }
3241 
3242     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3243       actp->sa_handler = SIG_DFL;
3244     }
3245 
3246     // try to honor the signal mask
3247     sigset_t oset;
3248     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3249 
3250     // call into the chained handler
3251     if (siginfo_flag_set) {
3252       (*sa)(sig, siginfo, context);
3253     } else {
3254       (*hand)(sig);
3255     }
3256 
3257     // restore the signal mask
3258     pthread_sigmask(SIG_SETMASK, &oset, 0);
3259   }
3260   // Tell jvm's signal handler the signal is taken care of.
3261   return true;
3262 }
3263 
3264 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3265   bool chained = false;
3266   // signal-chaining
3267   if (UseSignalChaining) {
3268     struct sigaction *actp = get_chained_signal_action(sig);
3269     if (actp != NULL) {
3270       chained = call_chained_handler(actp, sig, siginfo, context);
3271     }
3272   }
3273   return chained;
3274 }
3275 
3276 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3277   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3278     return &sigact[sig];
3279   }
3280   return NULL;
3281 }
3282 
3283 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3284   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3285   sigact[sig] = oldAct;
3286   sigs |= (unsigned int)1 << sig;
3287 }
3288 
3289 // for diagnostic
3290 int os::Linux::sigflags[MAXSIGNUM];
3291 
3292 int os::Linux::get_our_sigflags(int sig) {
3293   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3294   return sigflags[sig];
3295 }
3296 
3297 void os::Linux::set_our_sigflags(int sig, int flags) {
3298   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3299   sigflags[sig] = flags;
3300 }
3301 
3302 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3303   // Check for overwrite.
3304   struct sigaction oldAct;
3305   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3306 
3307   void* oldhand = oldAct.sa_sigaction
3308                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3309                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3310   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3311       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3312       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3313     if (AllowUserSignalHandlers || !set_installed) {
3314       // Do not overwrite; user takes responsibility to forward to us.
3315       return;
3316     } else if (UseSignalChaining) {
3317       // save the old handler in jvm
3318       save_preinstalled_handler(sig, oldAct);
3319       // libjsig also interposes the sigaction() call below and saves the
3320       // old sigaction on it own.
3321     } else {
3322       fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig);
3323     }
3324   }
3325 
3326   struct sigaction sigAct;
3327   sigfillset(&(sigAct.sa_mask));
3328   sigAct.sa_handler = SIG_DFL;
3329   if (!set_installed) {
3330     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3331   } else {
3332     sigAct.sa_sigaction = signalHandler;
3333     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3334   }
3335   // Save flags, which are set by ours
3336   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3337   sigflags[sig] = sigAct.sa_flags;
3338 
3339   int ret = sigaction(sig, &sigAct, &oldAct);
3340   assert(ret == 0, "check");
3341 
3342   void* oldhand2  = oldAct.sa_sigaction
3343                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3344                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3345   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3346 }
3347 
3348 // install signal handlers for signals that HotSpot needs to
3349 // handle in order to support Java-level exception handling.
3350 
3351 void os::Linux::install_signal_handlers() {
3352   if (!signal_handlers_are_installed) {
3353     signal_handlers_are_installed = true;
3354 
3355     // signal-chaining
3356     typedef void (*signal_setting_t)();
3357     signal_setting_t begin_signal_setting = NULL;
3358     signal_setting_t end_signal_setting = NULL;
3359     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3360                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3361     if (begin_signal_setting != NULL) {
3362       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3363                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3364       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3365                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3366       libjsig_is_loaded = true;
3367       assert(UseSignalChaining, "should enable signal-chaining");
3368     }
3369     if (libjsig_is_loaded) {
3370       // Tell libjsig jvm is setting signal handlers
3371       (*begin_signal_setting)();
3372     }
3373 
3374     set_signal_handler(SIGSEGV, true);
3375     set_signal_handler(SIGPIPE, true);
3376     set_signal_handler(SIGBUS, true);
3377     set_signal_handler(SIGILL, true);
3378     set_signal_handler(SIGFPE, true);
3379     set_signal_handler(SIGXFSZ, true);
3380 
3381     if (libjsig_is_loaded) {
3382       // Tell libjsig jvm finishes setting signal handlers
3383       (*end_signal_setting)();
3384     }
3385 
3386     // We don't activate signal checker if libjsig is in place, we trust ourselves
3387     // and if UserSignalHandler is installed all bets are off
3388     if (CheckJNICalls) {
3389       if (libjsig_is_loaded) {
3390         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3391         check_signals = false;
3392       }
3393       if (AllowUserSignalHandlers) {
3394         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3395         check_signals = false;
3396       }
3397     }
3398   }
3399 }
3400 
3401 // This is the fastest way to get thread cpu time on Linux.
3402 // Returns cpu time (user+sys) for any thread, not only for current.
3403 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3404 // It might work on 2.6.10+ with a special kernel/glibc patch.
3405 // For reference, please, see IEEE Std 1003.1-2004:
3406 //   http://www.unix.org/single_unix_specification
3407 
3408 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3409   struct timespec tp;
3410   int rc = os::Linux::clock_gettime(clockid, &tp);
3411   assert(rc == 0, "clock_gettime is expected to return 0 code");
3412 
3413   return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3414 }
3415 
3416 /////
3417 // glibc on Linux platform uses non-documented flag
3418 // to indicate, that some special sort of signal
3419 // trampoline is used.
3420 // We will never set this flag, and we should
3421 // ignore this flag in our diagnostic
3422 #ifdef SIGNIFICANT_SIGNAL_MASK
3423 #undef SIGNIFICANT_SIGNAL_MASK
3424 #endif
3425 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3426 
3427 static const char* get_signal_handler_name(address handler,
3428                                            char* buf, int buflen) {
3429   int offset;
3430   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3431   if (found) {
3432     // skip directory names
3433     const char *p1, *p2;
3434     p1 = buf;
3435     size_t len = strlen(os::file_separator());
3436     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3437     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3438   } else {
3439     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3440   }
3441   return buf;
3442 }
3443 
3444 static void print_signal_handler(outputStream* st, int sig,
3445                                  char* buf, size_t buflen) {
3446   struct sigaction sa;
3447 
3448   sigaction(sig, NULL, &sa);
3449 
3450   // See comment for SIGNIFICANT_SIGNAL_MASK define
3451   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3452 
3453   st->print("%s: ", os::exception_name(sig, buf, buflen));
3454 
3455   address handler = (sa.sa_flags & SA_SIGINFO)
3456     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3457     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3458 
3459   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3460     st->print("SIG_DFL");
3461   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3462     st->print("SIG_IGN");
3463   } else {
3464     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3465   }
3466 
3467   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3468 
3469   address rh = VMError::get_resetted_sighandler(sig);
3470   // May be, handler was resetted by VMError?
3471   if(rh != NULL) {
3472     handler = rh;
3473     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3474   }
3475 
3476   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
3477 
3478   // Check: is it our handler?
3479   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3480      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3481     // It is our signal handler
3482     // check for flags, reset system-used one!
3483     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3484       st->print(
3485                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3486                 os::Linux::get_our_sigflags(sig));
3487     }
3488   }
3489   st->cr();
3490 }
3491 
3492 
3493 #define DO_SIGNAL_CHECK(sig) \
3494   if (!sigismember(&check_signal_done, sig)) \
3495     os::Linux::check_signal_handler(sig)
3496 
3497 // This method is a periodic task to check for misbehaving JNI applications
3498 // under CheckJNI, we can add any periodic checks here
3499 
3500 void os::run_periodic_checks() {
3501 
3502   if (check_signals == false) return;
3503 
3504   // SEGV and BUS if overridden could potentially prevent
3505   // generation of hs*.log in the event of a crash, debugging
3506   // such a case can be very challenging, so we absolutely
3507   // check the following for a good measure:
3508   DO_SIGNAL_CHECK(SIGSEGV);
3509   DO_SIGNAL_CHECK(SIGILL);
3510   DO_SIGNAL_CHECK(SIGFPE);
3511   DO_SIGNAL_CHECK(SIGBUS);
3512   DO_SIGNAL_CHECK(SIGPIPE);
3513   DO_SIGNAL_CHECK(SIGXFSZ);
3514 
3515 
3516   // ReduceSignalUsage allows the user to override these handlers
3517   // see comments at the very top and jvm_solaris.h
3518   if (!ReduceSignalUsage) {
3519     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3520     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3521     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3522     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3523   }
3524 
3525   DO_SIGNAL_CHECK(SR_signum);
3526   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3527 }
3528 
3529 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3530 
3531 static os_sigaction_t os_sigaction = NULL;
3532 
3533 void os::Linux::check_signal_handler(int sig) {
3534   char buf[O_BUFLEN];
3535   address jvmHandler = NULL;
3536 
3537 
3538   struct sigaction act;
3539   if (os_sigaction == NULL) {
3540     // only trust the default sigaction, in case it has been interposed
3541     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3542     if (os_sigaction == NULL) return;
3543   }
3544 
3545   os_sigaction(sig, (struct sigaction*)NULL, &act);
3546 
3547 
3548   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3549 
3550   address thisHandler = (act.sa_flags & SA_SIGINFO)
3551     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3552     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3553 
3554 
3555   switch(sig) {
3556   case SIGSEGV:
3557   case SIGBUS:
3558   case SIGFPE:
3559   case SIGPIPE:
3560   case SIGILL:
3561   case SIGXFSZ:
3562     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3563     break;
3564 
3565   case SHUTDOWN1_SIGNAL:
3566   case SHUTDOWN2_SIGNAL:
3567   case SHUTDOWN3_SIGNAL:
3568   case BREAK_SIGNAL:
3569     jvmHandler = (address)user_handler();
3570     break;
3571 
3572   case INTERRUPT_SIGNAL:
3573     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3574     break;
3575 
3576   default:
3577     if (sig == SR_signum) {
3578       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3579     } else {
3580       return;
3581     }
3582     break;
3583   }
3584 
3585   if (thisHandler != jvmHandler) {
3586     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3587     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3588     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3589     // No need to check this sig any longer
3590     sigaddset(&check_signal_done, sig);
3591   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3592     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3593     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3594     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
3595     // No need to check this sig any longer
3596     sigaddset(&check_signal_done, sig);
3597   }
3598 
3599   // Dump all the signal
3600   if (sigismember(&check_signal_done, sig)) {
3601     print_signal_handlers(tty, buf, O_BUFLEN);
3602   }
3603 }
3604 
3605 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3606 
3607 extern bool signal_name(int signo, char* buf, size_t len);
3608 
3609 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3610   if (0 < exception_code && exception_code <= SIGRTMAX) {
3611     // signal
3612     if (!signal_name(exception_code, buf, size)) {
3613       jio_snprintf(buf, size, "SIG%d", exception_code);
3614     }
3615     return buf;
3616   } else {
3617     return NULL;
3618   }
3619 }
3620 
3621 // this is called _before_ the most of global arguments have been parsed
3622 void os::init(void) {
3623   char dummy;   /* used to get a guess on initial stack address */
3624 //  first_hrtime = gethrtime();
3625 
3626   // With LinuxThreads the JavaMain thread pid (primordial thread)
3627   // is different than the pid of the java launcher thread.
3628   // So, on Linux, the launcher thread pid is passed to the VM
3629   // via the sun.java.launcher.pid property.
3630   // Use this property instead of getpid() if it was correctly passed.
3631   // See bug 6351349.
3632   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3633 
3634   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3635 
3636   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3637 
3638   init_random(1234567);
3639 
3640   ThreadCritical::initialize();
3641 
3642   Linux::set_page_size(sysconf(_SC_PAGESIZE));
3643   if (Linux::page_size() == -1) {
3644     fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno));
3645   }
3646   init_page_sizes((size_t) Linux::page_size());
3647 
3648   Linux::initialize_system_info();
3649 
3650   // main_thread points to the aboriginal thread
3651   Linux::_main_thread = pthread_self();
3652 
3653   Linux::clock_init();
3654   initial_time_count = os::elapsed_counter();
3655   pthread_mutex_init(&dl_mutex, NULL);
3656 }
3657 
3658 // To install functions for atexit system call
3659 extern "C" {
3660   static void perfMemory_exit_helper() {
3661     perfMemory_exit();
3662   }
3663 }
3664 
3665 // this is called _after_ the global arguments have been parsed
3666 jint os::init_2(void)
3667 {
3668   Linux::fast_thread_clock_init();
3669 
3670   // Allocate a single page and mark it as readable for safepoint polling
3671   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3672   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3673 
3674   os::set_polling_page( polling_page );
3675 
3676 #ifndef PRODUCT
3677   if(Verbose && PrintMiscellaneous)
3678     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3679 #endif
3680 
3681   if (!UseMembar) {
3682     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3683     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3684     os::set_memory_serialize_page( mem_serialize_page );
3685 
3686 #ifndef PRODUCT
3687     if(Verbose && PrintMiscellaneous)
3688       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3689 #endif
3690   }
3691 
3692   FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3693 
3694   // initialize suspend/resume support - must do this before signal_sets_init()
3695   if (SR_initialize() != 0) {
3696     perror("SR_initialize failed");
3697     return JNI_ERR;
3698   }
3699 
3700   Linux::signal_sets_init();
3701   Linux::install_signal_handlers();
3702 
3703   size_t threadStackSizeInBytes = ThreadStackSize * K;
3704   if (threadStackSizeInBytes != 0 &&
3705       threadStackSizeInBytes < Linux::min_stack_allowed) {
3706         tty->print_cr("\nThe stack size specified is too small, "
3707                       "Specify at least %dk",
3708                       Linux::min_stack_allowed / K);
3709         return JNI_ERR;
3710   }
3711 
3712   // Make the stack size a multiple of the page size so that
3713   // the yellow/red zones can be guarded.
3714   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3715         vm_page_size()));
3716 
3717   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
3718 
3719   Linux::libpthread_init();
3720   if (PrintMiscellaneous && (Verbose || WizardMode)) {
3721      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
3722           Linux::glibc_version(), Linux::libpthread_version(),
3723           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
3724   }
3725 
3726   if (UseNUMA) {
3727     if (!Linux::libnuma_init()) {
3728       UseNUMA = false;
3729     } else {
3730       if ((Linux::numa_max_node() < 1)) {
3731         // There's only one node(they start from 0), disable NUMA.
3732         UseNUMA = false;
3733       }
3734     }
3735     if (!UseNUMA && ForceNUMA) {
3736       UseNUMA = true;
3737     }
3738   }
3739 
3740   if (MaxFDLimit) {
3741     // set the number of file descriptors to max. print out error
3742     // if getrlimit/setrlimit fails but continue regardless.
3743     struct rlimit nbr_files;
3744     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3745     if (status != 0) {
3746       if (PrintMiscellaneous && (Verbose || WizardMode))
3747         perror("os::init_2 getrlimit failed");
3748     } else {
3749       nbr_files.rlim_cur = nbr_files.rlim_max;
3750       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3751       if (status != 0) {
3752         if (PrintMiscellaneous && (Verbose || WizardMode))
3753           perror("os::init_2 setrlimit failed");
3754       }
3755     }
3756   }
3757 
3758   // Initialize lock used to serialize thread creation (see os::create_thread)
3759   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
3760 
3761   // Initialize HPI.
3762   jint hpi_result = hpi::initialize();
3763   if (hpi_result != JNI_OK) {
3764     tty->print_cr("There was an error trying to initialize the HPI library.");
3765     return hpi_result;
3766   }
3767 
3768   // at-exit methods are called in the reverse order of their registration.
3769   // atexit functions are called on return from main or as a result of a
3770   // call to exit(3C). There can be only 32 of these functions registered
3771   // and atexit() does not set errno.
3772 
3773   if (PerfAllowAtExitRegistration) {
3774     // only register atexit functions if PerfAllowAtExitRegistration is set.
3775     // atexit functions can be delayed until process exit time, which
3776     // can be problematic for embedded VM situations. Embedded VMs should
3777     // call DestroyJavaVM() to assure that VM resources are released.
3778 
3779     // note: perfMemory_exit_helper atexit function may be removed in
3780     // the future if the appropriate cleanup code can be added to the
3781     // VM_Exit VMOperation's doit method.
3782     if (atexit(perfMemory_exit_helper) != 0) {
3783       warning("os::init2 atexit(perfMemory_exit_helper) failed");
3784     }
3785   }
3786 
3787   // initialize thread priority policy
3788   prio_init();
3789 
3790   return JNI_OK;
3791 }
3792 
3793 // Mark the polling page as unreadable
3794 void os::make_polling_page_unreadable(void) {
3795   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
3796     fatal("Could not disable polling page");
3797 };
3798 
3799 // Mark the polling page as readable
3800 void os::make_polling_page_readable(void) {
3801   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
3802     fatal("Could not enable polling page");
3803   }
3804 };
3805 
3806 int os::active_processor_count() {
3807   // Linux doesn't yet have a (official) notion of processor sets,
3808   // so just return the number of online processors.
3809   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
3810   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
3811   return online_cpus;
3812 }
3813 
3814 bool os::distribute_processes(uint length, uint* distribution) {
3815   // Not yet implemented.
3816   return false;
3817 }
3818 
3819 bool os::bind_to_processor(uint processor_id) {
3820   // Not yet implemented.
3821   return false;
3822 }
3823 
3824 ///
3825 
3826 // Suspends the target using the signal mechanism and then grabs the PC before
3827 // resuming the target. Used by the flat-profiler only
3828 ExtendedPC os::get_thread_pc(Thread* thread) {
3829   // Make sure that it is called by the watcher for the VMThread
3830   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
3831   assert(thread->is_VM_thread(), "Can only be called for VMThread");
3832 
3833   ExtendedPC epc;
3834 
3835   OSThread* osthread = thread->osthread();
3836   if (do_suspend(osthread)) {
3837     if (osthread->ucontext() != NULL) {
3838       epc = os::Linux::ucontext_get_pc(osthread->ucontext());
3839     } else {
3840       // NULL context is unexpected, double-check this is the VMThread
3841       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3842     }
3843     do_resume(osthread);
3844   }
3845   // failure means pthread_kill failed for some reason - arguably this is
3846   // a fatal problem, but such problems are ignored elsewhere
3847 
3848   return epc;
3849 }
3850 
3851 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
3852 {
3853    if (is_NPTL()) {
3854       return pthread_cond_timedwait(_cond, _mutex, _abstime);
3855    } else {
3856 #ifndef IA64
3857       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
3858       // word back to default 64bit precision if condvar is signaled. Java
3859       // wants 53bit precision.  Save and restore current value.
3860       int fpu = get_fpu_control_word();
3861 #endif // IA64
3862       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
3863 #ifndef IA64
3864       set_fpu_control_word(fpu);
3865 #endif // IA64
3866       return status;
3867    }
3868 }
3869 
3870 ////////////////////////////////////////////////////////////////////////////////
3871 // debug support
3872 
3873 #ifndef PRODUCT
3874 static address same_page(address x, address y) {
3875   int page_bits = -os::vm_page_size();
3876   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
3877     return x;
3878   else if (x > y)
3879     return (address)(intptr_t(y) | ~page_bits) + 1;
3880   else
3881     return (address)(intptr_t(y) & page_bits);
3882 }
3883 
3884 bool os::find(address addr) {
3885   Dl_info dlinfo;
3886   memset(&dlinfo, 0, sizeof(dlinfo));
3887   if (dladdr(addr, &dlinfo)) {
3888     tty->print(PTR_FORMAT ": ", addr);
3889     if (dlinfo.dli_sname != NULL) {
3890       tty->print("%s+%#x", dlinfo.dli_sname,
3891                  addr - (intptr_t)dlinfo.dli_saddr);
3892     } else if (dlinfo.dli_fname) {
3893       tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
3894     } else {
3895       tty->print("<absolute address>");
3896     }
3897     if (dlinfo.dli_fname) {
3898       tty->print(" in %s", dlinfo.dli_fname);
3899     }
3900     if (dlinfo.dli_fbase) {
3901       tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
3902     }
3903     tty->cr();
3904 
3905     if (Verbose) {
3906       // decode some bytes around the PC
3907       address begin = same_page(addr-40, addr);
3908       address end   = same_page(addr+40, addr);
3909       address       lowest = (address) dlinfo.dli_sname;
3910       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
3911       if (begin < lowest)  begin = lowest;
3912       Dl_info dlinfo2;
3913       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
3914           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
3915         end = (address) dlinfo2.dli_saddr;
3916       Disassembler::decode(begin, end);
3917     }
3918     return true;
3919   }
3920   return false;
3921 }
3922 
3923 #endif
3924 
3925 ////////////////////////////////////////////////////////////////////////////////
3926 // misc
3927 
3928 // This does not do anything on Linux. This is basically a hook for being
3929 // able to use structured exception handling (thread-local exception filters)
3930 // on, e.g., Win32.
3931 void
3932 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
3933                          JavaCallArguments* args, Thread* thread) {
3934   f(value, method, args, thread);
3935 }
3936 
3937 void os::print_statistics() {
3938 }
3939 
3940 int os::message_box(const char* title, const char* message) {
3941   int i;
3942   fdStream err(defaultStream::error_fd());
3943   for (i = 0; i < 78; i++) err.print_raw("=");
3944   err.cr();
3945   err.print_raw_cr(title);
3946   for (i = 0; i < 78; i++) err.print_raw("-");
3947   err.cr();
3948   err.print_raw_cr(message);
3949   for (i = 0; i < 78; i++) err.print_raw("=");
3950   err.cr();
3951 
3952   char buf[16];
3953   // Prevent process from exiting upon "read error" without consuming all CPU
3954   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3955 
3956   return buf[0] == 'y' || buf[0] == 'Y';
3957 }
3958 
3959 int os::stat(const char *path, struct stat *sbuf) {
3960   char pathbuf[MAX_PATH];
3961   if (strlen(path) > MAX_PATH - 1) {
3962     errno = ENAMETOOLONG;
3963     return -1;
3964   }
3965   hpi::native_path(strcpy(pathbuf, path));
3966   return ::stat(pathbuf, sbuf);
3967 }
3968 
3969 bool os::check_heap(bool force) {
3970   return true;
3971 }
3972 
3973 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
3974   return ::vsnprintf(buf, count, format, args);
3975 }
3976 
3977 // Is a (classpath) directory empty?
3978 bool os::dir_is_empty(const char* path) {
3979   DIR *dir = NULL;
3980   struct dirent *ptr;
3981 
3982   dir = opendir(path);
3983   if (dir == NULL) return true;
3984 
3985   /* Scan the directory */
3986   bool result = true;
3987   char buf[sizeof(struct dirent) + MAX_PATH];
3988   while (result && (ptr = ::readdir(dir)) != NULL) {
3989     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
3990       result = false;
3991     }
3992   }
3993   closedir(dir);
3994   return result;
3995 }
3996 
3997 // create binary file, rewriting existing file if required
3998 int os::create_binary_file(const char* path, bool rewrite_existing) {
3999   int oflags = O_WRONLY | O_CREAT;
4000   if (!rewrite_existing) {
4001     oflags |= O_EXCL;
4002   }
4003   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4004 }
4005 
4006 // return current position of file pointer
4007 jlong os::current_file_offset(int fd) {
4008   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4009 }
4010 
4011 // move file pointer to the specified offset
4012 jlong os::seek_to_file_offset(int fd, jlong offset) {
4013   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4014 }
4015 
4016 // Map a block of memory.
4017 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4018                      char *addr, size_t bytes, bool read_only,
4019                      bool allow_exec) {
4020   int prot;
4021   int flags;
4022 
4023   if (read_only) {
4024     prot = PROT_READ;
4025     flags = MAP_SHARED;
4026   } else {
4027     prot = PROT_READ | PROT_WRITE;
4028     flags = MAP_PRIVATE;
4029   }
4030 
4031   if (allow_exec) {
4032     prot |= PROT_EXEC;
4033   }
4034 
4035   if (addr != NULL) {
4036     flags |= MAP_FIXED;
4037   }
4038 
4039   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4040                                      fd, file_offset);
4041   if (mapped_address == MAP_FAILED) {
4042     return NULL;
4043   }
4044   return mapped_address;
4045 }
4046 
4047 
4048 // Remap a block of memory.
4049 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4050                        char *addr, size_t bytes, bool read_only,
4051                        bool allow_exec) {
4052   // same as map_memory() on this OS
4053   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4054                         allow_exec);
4055 }
4056 
4057 
4058 // Unmap a block of memory.
4059 bool os::unmap_memory(char* addr, size_t bytes) {
4060   return munmap(addr, bytes) == 0;
4061 }
4062 
4063 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4064 
4065 static clockid_t thread_cpu_clockid(Thread* thread) {
4066   pthread_t tid = thread->osthread()->pthread_id();
4067   clockid_t clockid;
4068 
4069   // Get thread clockid
4070   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4071   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4072   return clockid;
4073 }
4074 
4075 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4076 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4077 // of a thread.
4078 //
4079 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4080 // the fast estimate available on the platform.
4081 
4082 jlong os::current_thread_cpu_time() {
4083   if (os::Linux::supports_fast_thread_cpu_time()) {
4084     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4085   } else {
4086     // return user + sys since the cost is the same
4087     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4088   }
4089 }
4090 
4091 jlong os::thread_cpu_time(Thread* thread) {
4092   // consistent with what current_thread_cpu_time() returns
4093   if (os::Linux::supports_fast_thread_cpu_time()) {
4094     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4095   } else {
4096     return slow_thread_cpu_time(thread, true /* user + sys */);
4097   }
4098 }
4099 
4100 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4101   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4102     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4103   } else {
4104     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4105   }
4106 }
4107 
4108 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4109   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4110     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4111   } else {
4112     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4113   }
4114 }
4115 
4116 //
4117 //  -1 on error.
4118 //
4119 
4120 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4121   static bool proc_pid_cpu_avail = true;
4122   static bool proc_task_unchecked = true;
4123   static const char *proc_stat_path = "/proc/%d/stat";
4124   pid_t  tid = thread->osthread()->thread_id();
4125   int i;
4126   char *s;
4127   char stat[2048];
4128   int statlen;
4129   char proc_name[64];
4130   int count;
4131   long sys_time, user_time;
4132   char string[64];
4133   int idummy;
4134   long ldummy;
4135   FILE *fp;
4136 
4137   // We first try accessing /proc/<pid>/cpu since this is faster to
4138   // process.  If this file is not present (linux kernels 2.5 and above)
4139   // then we open /proc/<pid>/stat.
4140   if ( proc_pid_cpu_avail ) {
4141     sprintf(proc_name, "/proc/%d/cpu", tid);
4142     fp =  fopen(proc_name, "r");
4143     if ( fp != NULL ) {
4144       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4145       fclose(fp);
4146       if ( count != 3 ) return -1;
4147 
4148       if (user_sys_cpu_time) {
4149         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4150       } else {
4151         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4152       }
4153     }
4154     else proc_pid_cpu_avail = false;
4155   }
4156 
4157   // The /proc/<tid>/stat aggregates per-process usage on
4158   // new Linux kernels 2.6+ where NPTL is supported.
4159   // The /proc/self/task/<tid>/stat still has the per-thread usage.
4160   // See bug 6328462.
4161   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4162   // and possibly in some other cases, so we check its availability.
4163   if (proc_task_unchecked && os::Linux::is_NPTL()) {
4164     // This is executed only once
4165     proc_task_unchecked = false;
4166     fp = fopen("/proc/self/task", "r");
4167     if (fp != NULL) {
4168       proc_stat_path = "/proc/self/task/%d/stat";
4169       fclose(fp);
4170     }
4171   }
4172 
4173   sprintf(proc_name, proc_stat_path, tid);
4174   fp = fopen(proc_name, "r");
4175   if ( fp == NULL ) return -1;
4176   statlen = fread(stat, 1, 2047, fp);
4177   stat[statlen] = '\0';
4178   fclose(fp);
4179 
4180   // Skip pid and the command string. Note that we could be dealing with
4181   // weird command names, e.g. user could decide to rename java launcher
4182   // to "java 1.4.2 :)", then the stat file would look like
4183   //                1234 (java 1.4.2 :)) R ... ...
4184   // We don't really need to know the command string, just find the last
4185   // occurrence of ")" and then start parsing from there. See bug 4726580.
4186   s = strrchr(stat, ')');
4187   i = 0;
4188   if (s == NULL ) return -1;
4189 
4190   // Skip blank chars
4191   do s++; while (isspace(*s));
4192 
4193   count = sscanf(s,"%*c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4194                  &idummy, &idummy, &idummy, &idummy, &idummy,
4195                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4196                  &user_time, &sys_time);
4197   if ( count != 12 ) return -1;
4198   if (user_sys_cpu_time) {
4199     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4200   } else {
4201     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4202   }
4203 }
4204 
4205 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4206   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4207   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4208   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4209   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4210 }
4211 
4212 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4213   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4214   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4215   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4216   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4217 }
4218 
4219 bool os::is_thread_cpu_time_supported() {
4220   return true;
4221 }
4222 
4223 // System loadavg support.  Returns -1 if load average cannot be obtained.
4224 // Linux doesn't yet have a (official) notion of processor sets,
4225 // so just return the system wide load average.
4226 int os::loadavg(double loadavg[], int nelem) {
4227   return ::getloadavg(loadavg, nelem);
4228 }
4229 
4230 void os::pause() {
4231   char filename[MAX_PATH];
4232   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4233     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4234   } else {
4235     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4236   }
4237 
4238   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4239   if (fd != -1) {
4240     struct stat buf;
4241     close(fd);
4242     while (::stat(filename, &buf) == 0) {
4243       (void)::poll(NULL, 0, 100);
4244     }
4245   } else {
4246     jio_fprintf(stderr,
4247       "Could not open pause file '%s', continuing immediately.\n", filename);
4248   }
4249 }
4250 
4251 extern "C" {
4252 
4253 /**
4254  * NOTE: the following code is to keep the green threads code
4255  * in the libjava.so happy. Once the green threads is removed,
4256  * these code will no longer be needed.
4257  */
4258 int
4259 jdk_waitpid(pid_t pid, int* status, int options) {
4260     return waitpid(pid, status, options);
4261 }
4262 
4263 int
4264 fork1() {
4265     return fork();
4266 }
4267 
4268 int
4269 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4270     return sem_init(sem, pshared, value);
4271 }
4272 
4273 int
4274 jdk_sem_post(sem_t *sem) {
4275     return sem_post(sem);
4276 }
4277 
4278 int
4279 jdk_sem_wait(sem_t *sem) {
4280     return sem_wait(sem);
4281 }
4282 
4283 int
4284 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4285     return pthread_sigmask(how , newmask, oldmask);
4286 }
4287 
4288 }
4289 
4290 // Refer to the comments in os_solaris.cpp park-unpark.
4291 //
4292 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4293 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4294 // For specifics regarding the bug see GLIBC BUGID 261237 :
4295 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4296 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4297 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4298 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4299 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4300 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4301 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4302 // of libpthread avoids the problem, but isn't practical.
4303 //
4304 // Possible remedies:
4305 //
4306 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4307 //      This is palliative and probabilistic, however.  If the thread is preempted
4308 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4309 //      than the minimum period may have passed, and the abstime may be stale (in the
4310 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
4311 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4312 //
4313 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4314 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
4315 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4316 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
4317 //      thread.
4318 //
4319 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4320 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
4321 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4322 //      This also works well.  In fact it avoids kernel-level scalability impediments
4323 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4324 //      timers in a graceful fashion.
4325 //
4326 // 4.   When the abstime value is in the past it appears that control returns
4327 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4328 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
4329 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
4330 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
4331 //      It may be possible to avoid reinitialization by checking the return
4332 //      value from pthread_cond_timedwait().  In addition to reinitializing the
4333 //      condvar we must establish the invariant that cond_signal() is only called
4334 //      within critical sections protected by the adjunct mutex.  This prevents
4335 //      cond_signal() from "seeing" a condvar that's in the midst of being
4336 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
4337 //      desirable signal-after-unlock optimization that avoids futile context switching.
4338 //
4339 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
4340 //      structure when a condvar is used or initialized.  cond_destroy()  would
4341 //      release the helper structure.  Our reinitialize-after-timedwait fix
4342 //      put excessive stress on malloc/free and locks protecting the c-heap.
4343 //
4344 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
4345 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4346 // and only enabling the work-around for vulnerable environments.
4347 
4348 // utility to compute the abstime argument to timedwait:
4349 // millis is the relative timeout time
4350 // abstime will be the absolute timeout time
4351 // TODO: replace compute_abstime() with unpackTime()
4352 
4353 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4354   if (millis < 0)  millis = 0;
4355   struct timeval now;
4356   int status = gettimeofday(&now, NULL);
4357   assert(status == 0, "gettimeofday");
4358   jlong seconds = millis / 1000;
4359   millis %= 1000;
4360   if (seconds > 50000000) { // see man cond_timedwait(3T)
4361     seconds = 50000000;
4362   }
4363   abstime->tv_sec = now.tv_sec  + seconds;
4364   long       usec = now.tv_usec + millis * 1000;
4365   if (usec >= 1000000) {
4366     abstime->tv_sec += 1;
4367     usec -= 1000000;
4368   }
4369   abstime->tv_nsec = usec * 1000;
4370   return abstime;
4371 }
4372 
4373 
4374 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4375 // Conceptually TryPark() should be equivalent to park(0).
4376 
4377 int os::PlatformEvent::TryPark() {
4378   for (;;) {
4379     const int v = _Event ;
4380     guarantee ((v == 0) || (v == 1), "invariant") ;
4381     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
4382   }
4383 }
4384 
4385 void os::PlatformEvent::park() {       // AKA "down()"
4386   // Invariant: Only the thread associated with the Event/PlatformEvent
4387   // may call park().
4388   // TODO: assert that _Assoc != NULL or _Assoc == Self
4389   int v ;
4390   for (;;) {
4391       v = _Event ;
4392       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4393   }
4394   guarantee (v >= 0, "invariant") ;
4395   if (v == 0) {
4396      // Do this the hard way by blocking ...
4397      int status = pthread_mutex_lock(_mutex);
4398      assert_status(status == 0, status, "mutex_lock");
4399      guarantee (_nParked == 0, "invariant") ;
4400      ++ _nParked ;
4401      while (_Event < 0) {
4402         status = pthread_cond_wait(_cond, _mutex);
4403         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4404         // Treat this the same as if the wait was interrupted
4405         if (status == ETIME) { status = EINTR; }
4406         assert_status(status == 0 || status == EINTR, status, "cond_wait");
4407      }
4408      -- _nParked ;
4409 
4410     // In theory we could move the ST of 0 into _Event past the unlock(),
4411     // but then we'd need a MEMBAR after the ST.
4412     _Event = 0 ;
4413      status = pthread_mutex_unlock(_mutex);
4414      assert_status(status == 0, status, "mutex_unlock");
4415   }
4416   guarantee (_Event >= 0, "invariant") ;
4417 }
4418 
4419 int os::PlatformEvent::park(jlong millis) {
4420   guarantee (_nParked == 0, "invariant") ;
4421 
4422   int v ;
4423   for (;;) {
4424       v = _Event ;
4425       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4426   }
4427   guarantee (v >= 0, "invariant") ;
4428   if (v != 0) return OS_OK ;
4429 
4430   // We do this the hard way, by blocking the thread.
4431   // Consider enforcing a minimum timeout value.
4432   struct timespec abst;
4433   compute_abstime(&abst, millis);
4434 
4435   int ret = OS_TIMEOUT;
4436   int status = pthread_mutex_lock(_mutex);
4437   assert_status(status == 0, status, "mutex_lock");
4438   guarantee (_nParked == 0, "invariant") ;
4439   ++_nParked ;
4440 
4441   // Object.wait(timo) will return because of
4442   // (a) notification
4443   // (b) timeout
4444   // (c) thread.interrupt
4445   //
4446   // Thread.interrupt and object.notify{All} both call Event::set.
4447   // That is, we treat thread.interrupt as a special case of notification.
4448   // The underlying Solaris implementation, cond_timedwait, admits
4449   // spurious/premature wakeups, but the JLS/JVM spec prevents the
4450   // JVM from making those visible to Java code.  As such, we must
4451   // filter out spurious wakeups.  We assume all ETIME returns are valid.
4452   //
4453   // TODO: properly differentiate simultaneous notify+interrupt.
4454   // In that case, we should propagate the notify to another waiter.
4455 
4456   while (_Event < 0) {
4457     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4458     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4459       pthread_cond_destroy (_cond);
4460       pthread_cond_init (_cond, NULL) ;
4461     }
4462     assert_status(status == 0 || status == EINTR ||
4463                   status == ETIME || status == ETIMEDOUT,
4464                   status, "cond_timedwait");
4465     if (!FilterSpuriousWakeups) break ;                 // previous semantics
4466     if (status == ETIME || status == ETIMEDOUT) break ;
4467     // We consume and ignore EINTR and spurious wakeups.
4468   }
4469   --_nParked ;
4470   if (_Event >= 0) {
4471      ret = OS_OK;
4472   }
4473   _Event = 0 ;
4474   status = pthread_mutex_unlock(_mutex);
4475   assert_status(status == 0, status, "mutex_unlock");
4476   assert (_nParked == 0, "invariant") ;
4477   return ret;
4478 }
4479 
4480 void os::PlatformEvent::unpark() {
4481   int v, AnyWaiters ;
4482   for (;;) {
4483       v = _Event ;
4484       if (v > 0) {
4485          // The LD of _Event could have reordered or be satisfied
4486          // by a read-aside from this processor's write buffer.
4487          // To avoid problems execute a barrier and then
4488          // ratify the value.
4489          OrderAccess::fence() ;
4490          if (_Event == v) return ;
4491          continue ;
4492       }
4493       if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4494   }
4495   if (v < 0) {
4496      // Wait for the thread associated with the event to vacate
4497      int status = pthread_mutex_lock(_mutex);
4498      assert_status(status == 0, status, "mutex_lock");
4499      AnyWaiters = _nParked ;
4500      assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4501      if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4502         AnyWaiters = 0 ;
4503         pthread_cond_signal (_cond);
4504      }
4505      status = pthread_mutex_unlock(_mutex);
4506      assert_status(status == 0, status, "mutex_unlock");
4507      if (AnyWaiters != 0) {
4508         status = pthread_cond_signal(_cond);
4509         assert_status(status == 0, status, "cond_signal");
4510      }
4511   }
4512 
4513   // Note that we signal() _after dropping the lock for "immortal" Events.
4514   // This is safe and avoids a common class of  futile wakeups.  In rare
4515   // circumstances this can cause a thread to return prematurely from
4516   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4517   // simply re-test the condition and re-park itself.
4518 }
4519 
4520 
4521 // JSR166
4522 // -------------------------------------------------------
4523 
4524 /*
4525  * The solaris and linux implementations of park/unpark are fairly
4526  * conservative for now, but can be improved. They currently use a
4527  * mutex/condvar pair, plus a a count.
4528  * Park decrements count if > 0, else does a condvar wait.  Unpark
4529  * sets count to 1 and signals condvar.  Only one thread ever waits
4530  * on the condvar. Contention seen when trying to park implies that someone
4531  * is unparking you, so don't wait. And spurious returns are fine, so there
4532  * is no need to track notifications.
4533  */
4534 
4535 
4536 #define NANOSECS_PER_SEC 1000000000
4537 #define NANOSECS_PER_MILLISEC 1000000
4538 #define MAX_SECS 100000000
4539 /*
4540  * This code is common to linux and solaris and will be moved to a
4541  * common place in dolphin.
4542  *
4543  * The passed in time value is either a relative time in nanoseconds
4544  * or an absolute time in milliseconds. Either way it has to be unpacked
4545  * into suitable seconds and nanoseconds components and stored in the
4546  * given timespec structure.
4547  * Given time is a 64-bit value and the time_t used in the timespec is only
4548  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4549  * overflow if times way in the future are given. Further on Solaris versions
4550  * prior to 10 there is a restriction (see cond_timedwait) that the specified
4551  * number of seconds, in abstime, is less than current_time  + 100,000,000.
4552  * As it will be 28 years before "now + 100000000" will overflow we can
4553  * ignore overflow and just impose a hard-limit on seconds using the value
4554  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4555  * years from "now".
4556  */
4557 
4558 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4559   assert (time > 0, "convertTime");
4560 
4561   struct timeval now;
4562   int status = gettimeofday(&now, NULL);
4563   assert(status == 0, "gettimeofday");
4564 
4565   time_t max_secs = now.tv_sec + MAX_SECS;
4566 
4567   if (isAbsolute) {
4568     jlong secs = time / 1000;
4569     if (secs > max_secs) {
4570       absTime->tv_sec = max_secs;
4571     }
4572     else {
4573       absTime->tv_sec = secs;
4574     }
4575     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4576   }
4577   else {
4578     jlong secs = time / NANOSECS_PER_SEC;
4579     if (secs >= MAX_SECS) {
4580       absTime->tv_sec = max_secs;
4581       absTime->tv_nsec = 0;
4582     }
4583     else {
4584       absTime->tv_sec = now.tv_sec + secs;
4585       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4586       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4587         absTime->tv_nsec -= NANOSECS_PER_SEC;
4588         ++absTime->tv_sec; // note: this must be <= max_secs
4589       }
4590     }
4591   }
4592   assert(absTime->tv_sec >= 0, "tv_sec < 0");
4593   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4594   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4595   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4596 }
4597 
4598 void Parker::park(bool isAbsolute, jlong time) {
4599   // Optional fast-path check:
4600   // Return immediately if a permit is available.
4601   if (_counter > 0) {
4602       _counter = 0 ;
4603       return ;
4604   }
4605 
4606   Thread* thread = Thread::current();
4607   assert(thread->is_Java_thread(), "Must be JavaThread");
4608   JavaThread *jt = (JavaThread *)thread;
4609 
4610   // Optional optimization -- avoid state transitions if there's an interrupt pending.
4611   // Check interrupt before trying to wait
4612   if (Thread::is_interrupted(thread, false)) {
4613     return;
4614   }
4615 
4616   // Next, demultiplex/decode time arguments
4617   timespec absTime;
4618   if (time < 0) { // don't wait at all
4619     return;
4620   }
4621   if (time > 0) {
4622     unpackTime(&absTime, isAbsolute, time);
4623   }
4624 
4625 
4626   // Enter safepoint region
4627   // Beware of deadlocks such as 6317397.
4628   // The per-thread Parker:: mutex is a classic leaf-lock.
4629   // In particular a thread must never block on the Threads_lock while
4630   // holding the Parker:: mutex.  If safepoints are pending both the
4631   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4632   ThreadBlockInVM tbivm(jt);
4633 
4634   // Don't wait if cannot get lock since interference arises from
4635   // unblocking.  Also. check interrupt before trying wait
4636   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4637     return;
4638   }
4639 
4640   int status ;
4641   if (_counter > 0)  { // no wait needed
4642     _counter = 0;
4643     status = pthread_mutex_unlock(_mutex);
4644     assert (status == 0, "invariant") ;
4645     return;
4646   }
4647 
4648 #ifdef ASSERT
4649   // Don't catch signals while blocked; let the running threads have the signals.
4650   // (This allows a debugger to break into the running thread.)
4651   sigset_t oldsigs;
4652   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4653   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4654 #endif
4655 
4656   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4657   jt->set_suspend_equivalent();
4658   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4659 
4660   if (time == 0) {
4661     status = pthread_cond_wait (_cond, _mutex) ;
4662   } else {
4663     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4664     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4665       pthread_cond_destroy (_cond) ;
4666       pthread_cond_init    (_cond, NULL);
4667     }
4668   }
4669   assert_status(status == 0 || status == EINTR ||
4670                 status == ETIME || status == ETIMEDOUT,
4671                 status, "cond_timedwait");
4672 
4673 #ifdef ASSERT
4674   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4675 #endif
4676 
4677   _counter = 0 ;
4678   status = pthread_mutex_unlock(_mutex) ;
4679   assert_status(status == 0, status, "invariant") ;
4680   // If externally suspended while waiting, re-suspend
4681   if (jt->handle_special_suspend_equivalent_condition()) {
4682     jt->java_suspend_self();
4683   }
4684 
4685 }
4686 
4687 void Parker::unpark() {
4688   int s, status ;
4689   status = pthread_mutex_lock(_mutex);
4690   assert (status == 0, "invariant") ;
4691   s = _counter;
4692   _counter = 1;
4693   if (s < 1) {
4694      if (WorkAroundNPTLTimedWaitHang) {
4695         status = pthread_cond_signal (_cond) ;
4696         assert (status == 0, "invariant") ;
4697         status = pthread_mutex_unlock(_mutex);
4698         assert (status == 0, "invariant") ;
4699      } else {
4700         status = pthread_mutex_unlock(_mutex);
4701         assert (status == 0, "invariant") ;
4702         status = pthread_cond_signal (_cond) ;
4703         assert (status == 0, "invariant") ;
4704      }
4705   } else {
4706     pthread_mutex_unlock(_mutex);
4707     assert (status == 0, "invariant") ;
4708   }
4709 }
4710 
4711 
4712 extern char** environ;
4713 
4714 #ifndef __NR_fork
4715 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
4716 #endif
4717 
4718 #ifndef __NR_execve
4719 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
4720 #endif
4721 
4722 // Run the specified command in a separate process. Return its exit value,
4723 // or -1 on failure (e.g. can't fork a new process).
4724 // Unlike system(), this function can be called from signal handler. It
4725 // doesn't block SIGINT et al.
4726 int os::fork_and_exec(char* cmd) {
4727   const char * argv[4] = {"sh", "-c", cmd, NULL};
4728 
4729   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
4730   // pthread_atfork handlers and reset pthread library. All we need is a
4731   // separate process to execve. Make a direct syscall to fork process.
4732   // On IA64 there's no fork syscall, we have to use fork() and hope for
4733   // the best...
4734   pid_t pid = NOT_IA64(syscall(__NR_fork);)
4735               IA64_ONLY(fork();)
4736 
4737   if (pid < 0) {
4738     // fork failed
4739     return -1;
4740 
4741   } else if (pid == 0) {
4742     // child process
4743 
4744     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
4745     // first to kill every thread on the thread list. Because this list is
4746     // not reset by fork() (see notes above), execve() will instead kill
4747     // every thread in the parent process. We know this is the only thread
4748     // in the new process, so make a system call directly.
4749     // IA64 should use normal execve() from glibc to match the glibc fork()
4750     // above.
4751     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
4752     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
4753 
4754     // execve failed
4755     _exit(-1);
4756 
4757   } else  {
4758     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4759     // care about the actual exit code, for now.
4760 
4761     int status;
4762 
4763     // Wait for the child process to exit.  This returns immediately if
4764     // the child has already exited. */
4765     while (waitpid(pid, &status, 0) < 0) {
4766         switch (errno) {
4767         case ECHILD: return 0;
4768         case EINTR: break;
4769         default: return -1;
4770         }
4771     }
4772 
4773     if (WIFEXITED(status)) {
4774        // The child exited normally; get its exit code.
4775        return WEXITSTATUS(status);
4776     } else if (WIFSIGNALED(status)) {
4777        // The child exited because of a signal
4778        // The best value to return is 0x80 + signal number,
4779        // because that is what all Unix shells do, and because
4780        // it allows callers to distinguish between process exit and
4781        // process death by signal.
4782        return 0x80 + WTERMSIG(status);
4783     } else {
4784        // Unknown exit code; pass it through
4785        return status;
4786     }
4787   }
4788 }