1 /*
   2  * Copyright 1999-2009 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(ZERO)
 180 static char cpu_arch[] = ZERO_LIBARCH;
 181 #elif defined(IA64)
 182 static char cpu_arch[] = "ia64";
 183 #elif defined(IA32)
 184 static char cpu_arch[] = "i386";
 185 #elif defined(AMD64)
 186 static char cpu_arch[] = "amd64";
 187 #elif defined(SPARC)
 188 #  ifdef _LP64
 189 static char cpu_arch[] = "sparcv9";
 190 #  else
 191 static char cpu_arch[] = "sparc";
 192 #  endif
 193 #else
 194 #error Add appropriate cpu_arch setting
 195 #endif
 196 
 197 
 198 // pid_t gettid()
 199 //
 200 // Returns the kernel thread id of the currently running thread. Kernel
 201 // thread id is used to access /proc.
 202 //
 203 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 204 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 205 //
 206 pid_t os::Linux::gettid() {
 207   int rslt = syscall(SYS_gettid);
 208   if (rslt == -1) {
 209      // old kernel, no NPTL support
 210      return getpid();
 211   } else {
 212      return (pid_t)rslt;
 213   }
 214 }
 215 
 216 // Most versions of linux have a bug where the number of processors are
 217 // determined by looking at the /proc file system.  In a chroot environment,
 218 // the system call returns 1.  This causes the VM to act as if it is
 219 // a single processor and elide locking (see is_MP() call).
 220 static bool unsafe_chroot_detected = false;
 221 static const char *unstable_chroot_error = "/proc file system not found.\n"
 222                      "Java may be unstable running multithreaded in a chroot "
 223                      "environment on Linux when /proc filesystem is not mounted.";
 224 
 225 void os::Linux::initialize_system_info() {
 226   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 227   if (processor_count() == 1) {
 228     pid_t pid = os::Linux::gettid();
 229     char fname[32];
 230     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 231     FILE *fp = fopen(fname, "r");
 232     if (fp == NULL) {
 233       unsafe_chroot_detected = true;
 234     } else {
 235       fclose(fp);
 236     }
 237   }
 238   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 239   assert(processor_count() > 0, "linux error");
 240 }
 241 
 242 void os::init_system_properties_values() {
 243 //  char arch[12];
 244 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 245 
 246   // The next steps are taken in the product version:
 247   //
 248   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 249   // This library should be located at:
 250   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 251   //
 252   // If "/jre/lib/" appears at the right place in the path, then we
 253   // assume libjvm[_g].so is installed in a JDK and we use this path.
 254   //
 255   // Otherwise exit with message: "Could not create the Java virtual machine."
 256   //
 257   // The following extra steps are taken in the debugging version:
 258   //
 259   // If "/jre/lib/" does NOT appear at the right place in the path
 260   // instead of exit check for $JAVA_HOME environment variable.
 261   //
 262   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 263   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 264   // it looks like libjvm[_g].so is installed there
 265   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 266   //
 267   // Otherwise exit.
 268   //
 269   // Important note: if the location of libjvm.so changes this
 270   // code needs to be changed accordingly.
 271 
 272   // The next few definitions allow the code to be verbatim:
 273 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 274 #define getenv(n) ::getenv(n)
 275 
 276 /*
 277  * See ld(1):
 278  *      The linker uses the following search paths to locate required
 279  *      shared libraries:
 280  *        1: ...
 281  *        ...
 282  *        7: The default directories, normally /lib and /usr/lib.
 283  */
 284 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 285 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 286 #else
 287 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 288 #endif
 289 
 290 #define EXTENSIONS_DIR  "/lib/ext"
 291 #define ENDORSED_DIR    "/lib/endorsed"
 292 #define REG_DIR         "/usr/java/packages"
 293 
 294   {
 295     /* sysclasspath, java_home, dll_dir */
 296     {
 297         char *home_path;
 298         char *dll_path;
 299         char *pslash;
 300         char buf[MAXPATHLEN];
 301         os::jvm_path(buf, sizeof(buf));
 302 
 303         // Found the full path to libjvm.so.
 304         // Now cut the path to <java_home>/jre if we can.
 305         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 306         pslash = strrchr(buf, '/');
 307         if (pslash != NULL)
 308             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 309         dll_path = malloc(strlen(buf) + 1);
 310         if (dll_path == NULL)
 311             return;
 312         strcpy(dll_path, buf);
 313         Arguments::set_dll_dir(dll_path);
 314 
 315         if (pslash != NULL) {
 316             pslash = strrchr(buf, '/');
 317             if (pslash != NULL) {
 318                 *pslash = '\0';       /* get rid of /<arch> */
 319                 pslash = strrchr(buf, '/');
 320                 if (pslash != NULL)
 321                     *pslash = '\0';   /* get rid of /lib */
 322             }
 323         }
 324 
 325         home_path = malloc(strlen(buf) + 1);
 326         if (home_path == NULL)
 327             return;
 328         strcpy(home_path, buf);
 329         Arguments::set_java_home(home_path);
 330 
 331         if (!set_boot_path('/', ':'))
 332             return;
 333     }
 334 
 335     /*
 336      * Where to look for native libraries
 337      *
 338      * Note: Due to a legacy implementation, most of the library path
 339      * is set in the launcher.  This was to accomodate linking restrictions
 340      * on legacy Linux implementations (which are no longer supported).
 341      * Eventually, all the library path setting will be done here.
 342      *
 343      * However, to prevent the proliferation of improperly built native
 344      * libraries, the new path component /usr/java/packages is added here.
 345      * Eventually, all the library path setting will be done here.
 346      */
 347     {
 348         char *ld_library_path;
 349 
 350         /*
 351          * Construct the invariant part of ld_library_path. Note that the
 352          * space for the colon and the trailing null are provided by the
 353          * nulls included by the sizeof operator (so actually we allocate
 354          * a byte more than necessary).
 355          */
 356         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 357             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 358         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 359 
 360         /*
 361          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 362          * should always exist (until the legacy problem cited above is
 363          * addressed).
 364          */
 365         char *v = getenv("LD_LIBRARY_PATH");
 366         if (v != NULL) {
 367             char *t = ld_library_path;
 368             /* That's +1 for the colon and +1 for the trailing '\0' */
 369             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 370             sprintf(ld_library_path, "%s:%s", v, t);
 371         }
 372         Arguments::set_library_path(ld_library_path);
 373     }
 374 
 375     /*
 376      * Extensions directories.
 377      *
 378      * Note that the space for the colon and the trailing null are provided
 379      * by the nulls included by the sizeof operator (so actually one byte more
 380      * than necessary is allocated).
 381      */
 382     {
 383         char *buf = malloc(strlen(Arguments::get_java_home()) +
 384             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 385         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 386             Arguments::get_java_home());
 387         Arguments::set_ext_dirs(buf);
 388     }
 389 
 390     /* Endorsed standards default directory. */
 391     {
 392         char * buf;
 393         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 394         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 395         Arguments::set_endorsed_dirs(buf);
 396     }
 397   }
 398 
 399 #undef malloc
 400 #undef getenv
 401 #undef EXTENSIONS_DIR
 402 #undef ENDORSED_DIR
 403 
 404   // Done
 405   return;
 406 }
 407 
 408 ////////////////////////////////////////////////////////////////////////////////
 409 // breakpoint support
 410 
 411 void os::breakpoint() {
 412   BREAKPOINT;
 413 }
 414 
 415 extern "C" void breakpoint() {
 416   // use debugger to set breakpoint here
 417 }
 418 
 419 ////////////////////////////////////////////////////////////////////////////////
 420 // signal support
 421 
 422 debug_only(static bool signal_sets_initialized = false);
 423 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 424 
 425 bool os::Linux::is_sig_ignored(int sig) {
 426       struct sigaction oact;
 427       sigaction(sig, (struct sigaction*)NULL, &oact);
 428       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 429                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 430       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 431            return true;
 432       else
 433            return false;
 434 }
 435 
 436 void os::Linux::signal_sets_init() {
 437   // Should also have an assertion stating we are still single-threaded.
 438   assert(!signal_sets_initialized, "Already initialized");
 439   // Fill in signals that are necessarily unblocked for all threads in
 440   // the VM. Currently, we unblock the following signals:
 441   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 442   //                         by -Xrs (=ReduceSignalUsage));
 443   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 444   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 445   // the dispositions or masks wrt these signals.
 446   // Programs embedding the VM that want to use the above signals for their
 447   // own purposes must, at this time, use the "-Xrs" option to prevent
 448   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 449   // (See bug 4345157, and other related bugs).
 450   // In reality, though, unblocking these signals is really a nop, since
 451   // these signals are not blocked by default.
 452   sigemptyset(&unblocked_sigs);
 453   sigemptyset(&allowdebug_blocked_sigs);
 454   sigaddset(&unblocked_sigs, SIGILL);
 455   sigaddset(&unblocked_sigs, SIGSEGV);
 456   sigaddset(&unblocked_sigs, SIGBUS);
 457   sigaddset(&unblocked_sigs, SIGFPE);
 458   sigaddset(&unblocked_sigs, SR_signum);
 459 
 460   if (!ReduceSignalUsage) {
 461    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 462       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 463       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 464    }
 465    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 466       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 467       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 468    }
 469    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 470       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 471       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 472    }
 473   }
 474   // Fill in signals that are blocked by all but the VM thread.
 475   sigemptyset(&vm_sigs);
 476   if (!ReduceSignalUsage)
 477     sigaddset(&vm_sigs, BREAK_SIGNAL);
 478   debug_only(signal_sets_initialized = true);
 479 
 480 }
 481 
 482 // These are signals that are unblocked while a thread is running Java.
 483 // (For some reason, they get blocked by default.)
 484 sigset_t* os::Linux::unblocked_signals() {
 485   assert(signal_sets_initialized, "Not initialized");
 486   return &unblocked_sigs;
 487 }
 488 
 489 // These are the signals that are blocked while a (non-VM) thread is
 490 // running Java. Only the VM thread handles these signals.
 491 sigset_t* os::Linux::vm_signals() {
 492   assert(signal_sets_initialized, "Not initialized");
 493   return &vm_sigs;
 494 }
 495 
 496 // These are signals that are blocked during cond_wait to allow debugger in
 497 sigset_t* os::Linux::allowdebug_blocked_signals() {
 498   assert(signal_sets_initialized, "Not initialized");
 499   return &allowdebug_blocked_sigs;
 500 }
 501 
 502 void os::Linux::hotspot_sigmask(Thread* thread) {
 503 
 504   //Save caller's signal mask before setting VM signal mask
 505   sigset_t caller_sigmask;
 506   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 507 
 508   OSThread* osthread = thread->osthread();
 509   osthread->set_caller_sigmask(caller_sigmask);
 510 
 511   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 512 
 513   if (!ReduceSignalUsage) {
 514     if (thread->is_VM_thread()) {
 515       // Only the VM thread handles BREAK_SIGNAL ...
 516       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 517     } else {
 518       // ... all other threads block BREAK_SIGNAL
 519       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 520     }
 521   }
 522 }
 523 
 524 //////////////////////////////////////////////////////////////////////////////
 525 // detecting pthread library
 526 
 527 void os::Linux::libpthread_init() {
 528   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 529   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 530   // generic name for earlier versions.
 531   // Define macros here so we can build HotSpot on old systems.
 532 # ifndef _CS_GNU_LIBC_VERSION
 533 # define _CS_GNU_LIBC_VERSION 2
 534 # endif
 535 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 536 # define _CS_GNU_LIBPTHREAD_VERSION 3
 537 # endif
 538 
 539   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 540   if (n > 0) {
 541      char *str = (char *)malloc(n);
 542      confstr(_CS_GNU_LIBC_VERSION, str, n);
 543      os::Linux::set_glibc_version(str);
 544   } else {
 545      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 546      static char _gnu_libc_version[32];
 547      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 548               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 549      os::Linux::set_glibc_version(_gnu_libc_version);
 550   }
 551 
 552   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 553   if (n > 0) {
 554      char *str = (char *)malloc(n);
 555      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 556      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 557      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 558      // is the case. LinuxThreads has a hard limit on max number of threads.
 559      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 560      // On the other hand, NPTL does not have such a limit, sysconf()
 561      // will return -1 and errno is not changed. Check if it is really NPTL.
 562      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 563          strstr(str, "NPTL") &&
 564          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 565        free(str);
 566        os::Linux::set_libpthread_version("linuxthreads");
 567      } else {
 568        os::Linux::set_libpthread_version(str);
 569      }
 570   } else {
 571     // glibc before 2.3.2 only has LinuxThreads.
 572     os::Linux::set_libpthread_version("linuxthreads");
 573   }
 574 
 575   if (strstr(libpthread_version(), "NPTL")) {
 576      os::Linux::set_is_NPTL();
 577   } else {
 578      os::Linux::set_is_LinuxThreads();
 579   }
 580 
 581   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 582   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 583   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 584      os::Linux::set_is_floating_stack();
 585   }
 586 }
 587 
 588 /////////////////////////////////////////////////////////////////////////////
 589 // thread stack
 590 
 591 // Force Linux kernel to expand current thread stack. If "bottom" is close
 592 // to the stack guard, caller should block all signals.
 593 //
 594 // MAP_GROWSDOWN:
 595 //   A special mmap() flag that is used to implement thread stacks. It tells
 596 //   kernel that the memory region should extend downwards when needed. This
 597 //   allows early versions of LinuxThreads to only mmap the first few pages
 598 //   when creating a new thread. Linux kernel will automatically expand thread
 599 //   stack as needed (on page faults).
 600 //
 601 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 602 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 603 //   region, it's hard to tell if the fault is due to a legitimate stack
 604 //   access or because of reading/writing non-exist memory (e.g. buffer
 605 //   overrun). As a rule, if the fault happens below current stack pointer,
 606 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 607 //   application (see Linux kernel fault.c).
 608 //
 609 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 610 //   stack overflow detection.
 611 //
 612 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 613 //   not use this flag. However, the stack of initial thread is not created
 614 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 615 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 616 //   and then attach the thread to JVM.
 617 //
 618 // To get around the problem and allow stack banging on Linux, we need to
 619 // manually expand thread stack after receiving the SIGSEGV.
 620 //
 621 // There are two ways to expand thread stack to address "bottom", we used
 622 // both of them in JVM before 1.5:
 623 //   1. adjust stack pointer first so that it is below "bottom", and then
 624 //      touch "bottom"
 625 //   2. mmap() the page in question
 626 //
 627 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 628 // if current sp is already near the lower end of page 101, and we need to
 629 // call mmap() to map page 100, it is possible that part of the mmap() frame
 630 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 631 // That will destroy the mmap() frame and cause VM to crash.
 632 //
 633 // The following code works by adjusting sp first, then accessing the "bottom"
 634 // page to force a page fault. Linux kernel will then automatically expand the
 635 // stack mapping.
 636 //
 637 // _expand_stack_to() assumes its frame size is less than page size, which
 638 // should always be true if the function is not inlined.
 639 
 640 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 641 #define NOINLINE
 642 #else
 643 #define NOINLINE __attribute__ ((noinline))
 644 #endif
 645 
 646 static void _expand_stack_to(address bottom) NOINLINE;
 647 
 648 static void _expand_stack_to(address bottom) {
 649   address sp;
 650   size_t size;
 651   volatile char *p;
 652 
 653   // Adjust bottom to point to the largest address within the same page, it
 654   // gives us a one-page buffer if alloca() allocates slightly more memory.
 655   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 656   bottom += os::Linux::page_size() - 1;
 657 
 658   // sp might be slightly above current stack pointer; if that's the case, we
 659   // will alloca() a little more space than necessary, which is OK. Don't use
 660   // os::current_stack_pointer(), as its result can be slightly below current
 661   // stack pointer, causing us to not alloca enough to reach "bottom".
 662   sp = (address)&sp;
 663 
 664   if (sp > bottom) {
 665     size = sp - bottom;
 666     p = (volatile char *)alloca(size);
 667     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 668     p[0] = '\0';
 669   }
 670 }
 671 
 672 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 673   assert(t!=NULL, "just checking");
 674   assert(t->osthread()->expanding_stack(), "expand should be set");
 675   assert(t->stack_base() != NULL, "stack_base was not initialized");
 676 
 677   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 678     sigset_t mask_all, old_sigset;
 679     sigfillset(&mask_all);
 680     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 681     _expand_stack_to(addr);
 682     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 683     return true;
 684   }
 685   return false;
 686 }
 687 
 688 //////////////////////////////////////////////////////////////////////////////
 689 // create new thread
 690 
 691 static address highest_vm_reserved_address();
 692 
 693 // check if it's safe to start a new thread
 694 static bool _thread_safety_check(Thread* thread) {
 695   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 696     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 697     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 698     //   allocated (MAP_FIXED) from high address space. Every thread stack
 699     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 700     //   it to other values if they rebuild LinuxThreads).
 701     //
 702     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 703     // the memory region has already been mmap'ed. That means if we have too
 704     // many threads and/or very large heap, eventually thread stack will
 705     // collide with heap.
 706     //
 707     // Here we try to prevent heap/stack collision by comparing current
 708     // stack bottom with the highest address that has been mmap'ed by JVM
 709     // plus a safety margin for memory maps created by native code.
 710     //
 711     // This feature can be disabled by setting ThreadSafetyMargin to 0
 712     //
 713     if (ThreadSafetyMargin > 0) {
 714       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 715 
 716       // not safe if our stack extends below the safety margin
 717       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 718     } else {
 719       return true;
 720     }
 721   } else {
 722     // Floating stack LinuxThreads or NPTL:
 723     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 724     //   there's not enough space left, pthread_create() will fail. If we come
 725     //   here, that means enough space has been reserved for stack.
 726     return true;
 727   }
 728 }
 729 
 730 // Thread start routine for all newly created threads
 731 static void *java_start(Thread *thread) {
 732   // Try to randomize the cache line index of hot stack frames.
 733   // This helps when threads of the same stack traces evict each other's
 734   // cache lines. The threads can be either from the same JVM instance, or
 735   // from different JVM instances. The benefit is especially true for
 736   // processors with hyperthreading technology.
 737   static int counter = 0;
 738   int pid = os::current_process_id();
 739   alloca(((pid ^ counter++) & 7) * 128);
 740 
 741   ThreadLocalStorage::set_thread(thread);
 742 
 743   OSThread* osthread = thread->osthread();
 744   Monitor* sync = osthread->startThread_lock();
 745 
 746   // non floating stack LinuxThreads needs extra check, see above
 747   if (!_thread_safety_check(thread)) {
 748     // notify parent thread
 749     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 750     osthread->set_state(ZOMBIE);
 751     sync->notify_all();
 752     return NULL;
 753   }
 754 
 755   // thread_id is kernel thread id (similar to Solaris LWP id)
 756   osthread->set_thread_id(os::Linux::gettid());
 757 
 758   if (UseNUMA) {
 759     int lgrp_id = os::numa_get_group_id();
 760     if (lgrp_id != -1) {
 761       thread->set_lgrp_id(lgrp_id);
 762     }
 763   }
 764   // initialize signal mask for this thread
 765   os::Linux::hotspot_sigmask(thread);
 766 
 767   // initialize floating point control register
 768   os::Linux::init_thread_fpu_state();
 769 
 770   // handshaking with parent thread
 771   {
 772     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 773 
 774     // notify parent thread
 775     osthread->set_state(INITIALIZED);
 776     sync->notify_all();
 777 
 778     // wait until os::start_thread()
 779     while (osthread->get_state() == INITIALIZED) {
 780       sync->wait(Mutex::_no_safepoint_check_flag);
 781     }
 782   }
 783 
 784   // call one more level start routine
 785   thread->run();
 786 
 787   return 0;
 788 }
 789 
 790 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 791   assert(thread->osthread() == NULL, "caller responsible");
 792 
 793   // Allocate the OSThread object
 794   OSThread* osthread = new OSThread(NULL, NULL);
 795   if (osthread == NULL) {
 796     return false;
 797   }
 798 
 799   // set the correct thread state
 800   osthread->set_thread_type(thr_type);
 801 
 802   // Initial state is ALLOCATED but not INITIALIZED
 803   osthread->set_state(ALLOCATED);
 804 
 805   thread->set_osthread(osthread);
 806 
 807   // init thread attributes
 808   pthread_attr_t attr;
 809   pthread_attr_init(&attr);
 810   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 811 
 812   // stack size
 813   if (os::Linux::supports_variable_stack_size()) {
 814     // calculate stack size if it's not specified by caller
 815     if (stack_size == 0) {
 816       stack_size = os::Linux::default_stack_size(thr_type);
 817 
 818       switch (thr_type) {
 819       case os::java_thread:
 820         // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
 821         if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
 822         break;
 823       case os::compiler_thread:
 824         if (CompilerThreadStackSize > 0) {
 825           stack_size = (size_t)(CompilerThreadStackSize * K);
 826           break;
 827         } // else fall through:
 828           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 829       case os::vm_thread:
 830       case os::pgc_thread:
 831       case os::cgc_thread:
 832       case os::watcher_thread:
 833         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 834         break;
 835       }
 836     }
 837 
 838     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 839     pthread_attr_setstacksize(&attr, stack_size);
 840   } else {
 841     // let pthread_create() pick the default value.
 842   }
 843 
 844   // glibc guard page
 845   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 846 
 847   ThreadState state;
 848 
 849   {
 850     // Serialize thread creation if we are running with fixed stack LinuxThreads
 851     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 852     if (lock) {
 853       os::Linux::createThread_lock()->lock_without_safepoint_check();
 854     }
 855 
 856     pthread_t tid;
 857     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 858 
 859     pthread_attr_destroy(&attr);
 860 
 861     if (ret != 0) {
 862       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 863         perror("pthread_create()");
 864       }
 865       // Need to clean up stuff we've allocated so far
 866       thread->set_osthread(NULL);
 867       delete osthread;
 868       if (lock) os::Linux::createThread_lock()->unlock();
 869       return false;
 870     }
 871 
 872     // Store pthread info into the OSThread
 873     osthread->set_pthread_id(tid);
 874 
 875     // Wait until child thread is either initialized or aborted
 876     {
 877       Monitor* sync_with_child = osthread->startThread_lock();
 878       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 879       while ((state = osthread->get_state()) == ALLOCATED) {
 880         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 881       }
 882     }
 883 
 884     if (lock) {
 885       os::Linux::createThread_lock()->unlock();
 886     }
 887   }
 888 
 889   // Aborted due to thread limit being reached
 890   if (state == ZOMBIE) {
 891       thread->set_osthread(NULL);
 892       delete osthread;
 893       return false;
 894   }
 895 
 896   // The thread is returned suspended (in state INITIALIZED),
 897   // and is started higher up in the call chain
 898   assert(state == INITIALIZED, "race condition");
 899   return true;
 900 }
 901 
 902 /////////////////////////////////////////////////////////////////////////////
 903 // attach existing thread
 904 
 905 // bootstrap the main thread
 906 bool os::create_main_thread(JavaThread* thread) {
 907   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 908   return create_attached_thread(thread);
 909 }
 910 
 911 bool os::create_attached_thread(JavaThread* thread) {
 912 #ifdef ASSERT
 913     thread->verify_not_published();
 914 #endif
 915 
 916   // Allocate the OSThread object
 917   OSThread* osthread = new OSThread(NULL, NULL);
 918 
 919   if (osthread == NULL) {
 920     return false;
 921   }
 922 
 923   // Store pthread info into the OSThread
 924   osthread->set_thread_id(os::Linux::gettid());
 925   osthread->set_pthread_id(::pthread_self());
 926 
 927   // initialize floating point control register
 928   os::Linux::init_thread_fpu_state();
 929 
 930   // Initial thread state is RUNNABLE
 931   osthread->set_state(RUNNABLE);
 932 
 933   thread->set_osthread(osthread);
 934 
 935   if (UseNUMA) {
 936     int lgrp_id = os::numa_get_group_id();
 937     if (lgrp_id != -1) {
 938       thread->set_lgrp_id(lgrp_id);
 939     }
 940   }
 941 
 942   if (os::Linux::is_initial_thread()) {
 943     // If current thread is initial thread, its stack is mapped on demand,
 944     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 945     // the entire stack region to avoid SEGV in stack banging.
 946     // It is also useful to get around the heap-stack-gap problem on SuSE
 947     // kernel (see 4821821 for details). We first expand stack to the top
 948     // of yellow zone, then enable stack yellow zone (order is significant,
 949     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 950     // is no gap between the last two virtual memory regions.
 951 
 952     JavaThread *jt = (JavaThread *)thread;
 953     address addr = jt->stack_yellow_zone_base();
 954     assert(addr != NULL, "initialization problem?");
 955     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 956 
 957     osthread->set_expanding_stack();
 958     os::Linux::manually_expand_stack(jt, addr);
 959     osthread->clear_expanding_stack();
 960   }
 961 
 962   // initialize signal mask for this thread
 963   // and save the caller's signal mask
 964   os::Linux::hotspot_sigmask(thread);
 965 
 966   return true;
 967 }
 968 
 969 void os::pd_start_thread(Thread* thread) {
 970   OSThread * osthread = thread->osthread();
 971   assert(osthread->get_state() != INITIALIZED, "just checking");
 972   Monitor* sync_with_child = osthread->startThread_lock();
 973   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 974   sync_with_child->notify();
 975 }
 976 
 977 // Free Linux resources related to the OSThread
 978 void os::free_thread(OSThread* osthread) {
 979   assert(osthread != NULL, "osthread not set");
 980 
 981   if (Thread::current()->osthread() == osthread) {
 982     // Restore caller's signal mask
 983     sigset_t sigmask = osthread->caller_sigmask();
 984     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 985    }
 986 
 987   delete osthread;
 988 }
 989 
 990 //////////////////////////////////////////////////////////////////////////////
 991 // thread local storage
 992 
 993 int os::allocate_thread_local_storage() {
 994   pthread_key_t key;
 995   int rslt = pthread_key_create(&key, NULL);
 996   assert(rslt == 0, "cannot allocate thread local storage");
 997   return (int)key;
 998 }
 999 
1000 // Note: This is currently not used by VM, as we don't destroy TLS key
1001 // on VM exit.
1002 void os::free_thread_local_storage(int index) {
1003   int rslt = pthread_key_delete((pthread_key_t)index);
1004   assert(rslt == 0, "invalid index");
1005 }
1006 
1007 void os::thread_local_storage_at_put(int index, void* value) {
1008   int rslt = pthread_setspecific((pthread_key_t)index, value);
1009   assert(rslt == 0, "pthread_setspecific failed");
1010 }
1011 
1012 extern "C" Thread* get_thread() {
1013   return ThreadLocalStorage::thread();
1014 }
1015 
1016 //////////////////////////////////////////////////////////////////////////////
1017 // initial thread
1018 
1019 // Check if current thread is the initial thread, similar to Solaris thr_main.
1020 bool os::Linux::is_initial_thread(void) {
1021   char dummy;
1022   // If called before init complete, thread stack bottom will be null.
1023   // Can be called if fatal error occurs before initialization.
1024   if (initial_thread_stack_bottom() == NULL) return false;
1025   assert(initial_thread_stack_bottom() != NULL &&
1026          initial_thread_stack_size()   != 0,
1027          "os::init did not locate initial thread's stack region");
1028   if ((address)&dummy >= initial_thread_stack_bottom() &&
1029       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1030        return true;
1031   else return false;
1032 }
1033 
1034 // Find the virtual memory area that contains addr
1035 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1036   FILE *fp = fopen("/proc/self/maps", "r");
1037   if (fp) {
1038     address low, high;
1039     while (!feof(fp)) {
1040       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1041         if (low <= addr && addr < high) {
1042            if (vma_low)  *vma_low  = low;
1043            if (vma_high) *vma_high = high;
1044            fclose (fp);
1045            return true;
1046         }
1047       }
1048       for (;;) {
1049         int ch = fgetc(fp);
1050         if (ch == EOF || ch == (int)'\n') break;
1051       }
1052     }
1053     fclose(fp);
1054   }
1055   return false;
1056 }
1057 
1058 // Locate initial thread stack. This special handling of initial thread stack
1059 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1060 // bogus value for initial thread.
1061 void os::Linux::capture_initial_stack(size_t max_size) {
1062   // stack size is the easy part, get it from RLIMIT_STACK
1063   size_t stack_size;
1064   struct rlimit rlim;
1065   getrlimit(RLIMIT_STACK, &rlim);
1066   stack_size = rlim.rlim_cur;
1067 
1068   // 6308388: a bug in ld.so will relocate its own .data section to the
1069   //   lower end of primordial stack; reduce ulimit -s value a little bit
1070   //   so we won't install guard page on ld.so's data section.
1071   stack_size -= 2 * page_size();
1072 
1073   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1074   //   7.1, in both cases we will get 2G in return value.
1075   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1076   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1077   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1078   //   in case other parts in glibc still assumes 2M max stack size.
1079   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1080 #ifndef IA64
1081   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1082 #else
1083   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1084   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1085 #endif
1086 
1087   // Try to figure out where the stack base (top) is. This is harder.
1088   //
1089   // When an application is started, glibc saves the initial stack pointer in
1090   // a global variable "__libc_stack_end", which is then used by system
1091   // libraries. __libc_stack_end should be pretty close to stack top. The
1092   // variable is available since the very early days. However, because it is
1093   // a private interface, it could disappear in the future.
1094   //
1095   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1096   // to __libc_stack_end, it is very close to stack top, but isn't the real
1097   // stack top. Note that /proc may not exist if VM is running as a chroot
1098   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1099   // /proc/<pid>/stat could change in the future (though unlikely).
1100   //
1101   // We try __libc_stack_end first. If that doesn't work, look for
1102   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1103   // as a hint, which should work well in most cases.
1104 
1105   uintptr_t stack_start;
1106 
1107   // try __libc_stack_end first
1108   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1109   if (p && *p) {
1110     stack_start = *p;
1111   } else {
1112     // see if we can get the start_stack field from /proc/self/stat
1113     FILE *fp;
1114     int pid;
1115     char state;
1116     int ppid;
1117     int pgrp;
1118     int session;
1119     int nr;
1120     int tpgrp;
1121     unsigned long flags;
1122     unsigned long minflt;
1123     unsigned long cminflt;
1124     unsigned long majflt;
1125     unsigned long cmajflt;
1126     unsigned long utime;
1127     unsigned long stime;
1128     long cutime;
1129     long cstime;
1130     long prio;
1131     long nice;
1132     long junk;
1133     long it_real;
1134     uintptr_t start;
1135     uintptr_t vsize;
1136     uintptr_t rss;
1137     unsigned long rsslim;
1138     uintptr_t scodes;
1139     uintptr_t ecode;
1140     int i;
1141 
1142     // Figure what the primordial thread stack base is. Code is inspired
1143     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1144     // followed by command name surrounded by parentheses, state, etc.
1145     char stat[2048];
1146     int statlen;
1147 
1148     fp = fopen("/proc/self/stat", "r");
1149     if (fp) {
1150       statlen = fread(stat, 1, 2047, fp);
1151       stat[statlen] = '\0';
1152       fclose(fp);
1153 
1154       // Skip pid and the command string. Note that we could be dealing with
1155       // weird command names, e.g. user could decide to rename java launcher
1156       // to "java 1.4.2 :)", then the stat file would look like
1157       //                1234 (java 1.4.2 :)) R ... ...
1158       // We don't really need to know the command string, just find the last
1159       // occurrence of ")" and then start parsing from there. See bug 4726580.
1160       char * s = strrchr(stat, ')');
1161 
1162       i = 0;
1163       if (s) {
1164         // Skip blank chars
1165         do s++; while (isspace(*s));
1166 
1167         /*                                     1   1   1   1   1   1   1   1   1   1   2   2   2   2   2   2   2   2   2 */
1168         /*              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 */
1169         i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld "
1170                    UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT
1171                    " %lu "
1172                    UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT,
1173              &state,          /* 3  %c  */
1174              &ppid,           /* 4  %d  */
1175              &pgrp,           /* 5  %d  */
1176              &session,        /* 6  %d  */
1177              &nr,             /* 7  %d  */
1178              &tpgrp,          /* 8  %d  */
1179              &flags,          /* 9  %lu  */
1180              &minflt,         /* 10 %lu  */
1181              &cminflt,        /* 11 %lu  */
1182              &majflt,         /* 12 %lu  */
1183              &cmajflt,        /* 13 %lu  */
1184              &utime,          /* 14 %lu  */
1185              &stime,          /* 15 %lu  */
1186              &cutime,         /* 16 %ld  */
1187              &cstime,         /* 17 %ld  */
1188              &prio,           /* 18 %ld  */
1189              &nice,           /* 19 %ld  */
1190              &junk,           /* 20 %ld  */
1191              &it_real,        /* 21 %ld  */
1192              &start,          /* 22 UINTX_FORMAT  */
1193              &vsize,          /* 23 UINTX_FORMAT  */
1194              &rss,            /* 24 UINTX_FORMAT  */
1195              &rsslim,         /* 25 %lu  */
1196              &scodes,         /* 26 UINTX_FORMAT  */
1197              &ecode,          /* 27 UINTX_FORMAT  */
1198              &stack_start);   /* 28 UINTX_FORMAT  */
1199       }
1200 
1201       if (i != 28 - 2) {
1202          assert(false, "Bad conversion from /proc/self/stat");
1203          // product mode - assume we are the initial thread, good luck in the
1204          // embedded case.
1205          warning("Can't detect initial thread stack location - bad conversion");
1206          stack_start = (uintptr_t) &rlim;
1207       }
1208     } else {
1209       // For some reason we can't open /proc/self/stat (for example, running on
1210       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1211       // most cases, so don't abort:
1212       warning("Can't detect initial thread stack location - no /proc/self/stat");
1213       stack_start = (uintptr_t) &rlim;
1214     }
1215   }
1216 
1217   // Now we have a pointer (stack_start) very close to the stack top, the
1218   // next thing to do is to figure out the exact location of stack top. We
1219   // can find out the virtual memory area that contains stack_start by
1220   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1221   // and its upper limit is the real stack top. (again, this would fail if
1222   // running inside chroot, because /proc may not exist.)
1223 
1224   uintptr_t stack_top;
1225   address low, high;
1226   if (find_vma((address)stack_start, &low, &high)) {
1227     // success, "high" is the true stack top. (ignore "low", because initial
1228     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1229     stack_top = (uintptr_t)high;
1230   } else {
1231     // failed, likely because /proc/self/maps does not exist
1232     warning("Can't detect initial thread stack location - find_vma failed");
1233     // best effort: stack_start is normally within a few pages below the real
1234     // stack top, use it as stack top, and reduce stack size so we won't put
1235     // guard page outside stack.
1236     stack_top = stack_start;
1237     stack_size -= 16 * page_size();
1238   }
1239 
1240   // stack_top could be partially down the page so align it
1241   stack_top = align_size_up(stack_top, page_size());
1242 
1243   if (max_size && stack_size > max_size) {
1244      _initial_thread_stack_size = max_size;
1245   } else {
1246      _initial_thread_stack_size = stack_size;
1247   }
1248 
1249   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1250   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1251 }
1252 
1253 ////////////////////////////////////////////////////////////////////////////////
1254 // time support
1255 
1256 // Time since start-up in seconds to a fine granularity.
1257 // Used by VMSelfDestructTimer and the MemProfiler.
1258 double os::elapsedTime() {
1259 
1260   return (double)(os::elapsed_counter()) * 0.000001;
1261 }
1262 
1263 jlong os::elapsed_counter() {
1264   timeval time;
1265   int status = gettimeofday(&time, NULL);
1266   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1267 }
1268 
1269 jlong os::elapsed_frequency() {
1270   return (1000 * 1000);
1271 }
1272 
1273 // For now, we say that linux does not support vtime.  I have no idea
1274 // whether it can actually be made to (DLD, 9/13/05).
1275 
1276 bool os::supports_vtime() { return false; }
1277 bool os::enable_vtime()   { return false; }
1278 bool os::vtime_enabled()  { return false; }
1279 double os::elapsedVTime() {
1280   // better than nothing, but not much
1281   return elapsedTime();
1282 }
1283 
1284 jlong os::javaTimeMillis() {
1285   timeval time;
1286   int status = gettimeofday(&time, NULL);
1287   assert(status != -1, "linux error");
1288   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1289 }
1290 
1291 #ifndef CLOCK_MONOTONIC
1292 #define CLOCK_MONOTONIC (1)
1293 #endif
1294 
1295 void os::Linux::clock_init() {
1296   // we do dlopen's in this particular order due to bug in linux
1297   // dynamical loader (see 6348968) leading to crash on exit
1298   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1299   if (handle == NULL) {
1300     handle = dlopen("librt.so", RTLD_LAZY);
1301   }
1302 
1303   if (handle) {
1304     int (*clock_getres_func)(clockid_t, struct timespec*) =
1305            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1306     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1307            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1308     if (clock_getres_func && clock_gettime_func) {
1309       // See if monotonic clock is supported by the kernel. Note that some
1310       // early implementations simply return kernel jiffies (updated every
1311       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1312       // for nano time (though the monotonic property is still nice to have).
1313       // It's fixed in newer kernels, however clock_getres() still returns
1314       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1315       // resolution for now. Hopefully as people move to new kernels, this
1316       // won't be a problem.
1317       struct timespec res;
1318       struct timespec tp;
1319       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1320           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1321         // yes, monotonic clock is supported
1322         _clock_gettime = clock_gettime_func;
1323       } else {
1324         // close librt if there is no monotonic clock
1325         dlclose(handle);
1326       }
1327     }
1328   }
1329 }
1330 
1331 #ifndef SYS_clock_getres
1332 
1333 #if defined(IA32) || defined(AMD64)
1334 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1335 #else
1336 #error Value of SYS_clock_getres not known on this platform
1337 #endif
1338 
1339 #endif
1340 
1341 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1342 
1343 void os::Linux::fast_thread_clock_init() {
1344   if (!UseLinuxPosixThreadCPUClocks) {
1345     return;
1346   }
1347   clockid_t clockid;
1348   struct timespec tp;
1349   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1350       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1351 
1352   // Switch to using fast clocks for thread cpu time if
1353   // the sys_clock_getres() returns 0 error code.
1354   // Note, that some kernels may support the current thread
1355   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1356   // returned by the pthread_getcpuclockid().
1357   // If the fast Posix clocks are supported then the sys_clock_getres()
1358   // must return at least tp.tv_sec == 0 which means a resolution
1359   // better than 1 sec. This is extra check for reliability.
1360 
1361   if(pthread_getcpuclockid_func &&
1362      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1363      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1364 
1365     _supports_fast_thread_cpu_time = true;
1366     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1367   }
1368 }
1369 
1370 jlong os::javaTimeNanos() {
1371   if (Linux::supports_monotonic_clock()) {
1372     struct timespec tp;
1373     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1374     assert(status == 0, "gettime error");
1375     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1376     return result;
1377   } else {
1378     timeval time;
1379     int status = gettimeofday(&time, NULL);
1380     assert(status != -1, "linux error");
1381     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1382     return 1000 * usecs;
1383   }
1384 }
1385 
1386 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1387   if (Linux::supports_monotonic_clock()) {
1388     info_ptr->max_value = ALL_64_BITS;
1389 
1390     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1391     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1392     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1393   } else {
1394     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1395     info_ptr->max_value = ALL_64_BITS;
1396 
1397     // gettimeofday is a real time clock so it skips
1398     info_ptr->may_skip_backward = true;
1399     info_ptr->may_skip_forward = true;
1400   }
1401 
1402   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1403 }
1404 
1405 // Return the real, user, and system times in seconds from an
1406 // arbitrary fixed point in the past.
1407 bool os::getTimesSecs(double* process_real_time,
1408                       double* process_user_time,
1409                       double* process_system_time) {
1410   struct tms ticks;
1411   clock_t real_ticks = times(&ticks);
1412 
1413   if (real_ticks == (clock_t) (-1)) {
1414     return false;
1415   } else {
1416     double ticks_per_second = (double) clock_tics_per_sec;
1417     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1418     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1419     *process_real_time = ((double) real_ticks) / ticks_per_second;
1420 
1421     return true;
1422   }
1423 }
1424 
1425 
1426 char * os::local_time_string(char *buf, size_t buflen) {
1427   struct tm t;
1428   time_t long_time;
1429   time(&long_time);
1430   localtime_r(&long_time, &t);
1431   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1432                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1433                t.tm_hour, t.tm_min, t.tm_sec);
1434   return buf;
1435 }
1436 
1437 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1438   return localtime_r(clock, res);
1439 }
1440 
1441 ////////////////////////////////////////////////////////////////////////////////
1442 // runtime exit support
1443 
1444 // Note: os::shutdown() might be called very early during initialization, or
1445 // called from signal handler. Before adding something to os::shutdown(), make
1446 // sure it is async-safe and can handle partially initialized VM.
1447 void os::shutdown() {
1448 
1449   // allow PerfMemory to attempt cleanup of any persistent resources
1450   perfMemory_exit();
1451 
1452   // needs to remove object in file system
1453   AttachListener::abort();
1454 
1455   // flush buffered output, finish log files
1456   ostream_abort();
1457 
1458   // Check for abort hook
1459   abort_hook_t abort_hook = Arguments::abort_hook();
1460   if (abort_hook != NULL) {
1461     abort_hook();
1462   }
1463 
1464 }
1465 
1466 // Note: os::abort() might be called very early during initialization, or
1467 // called from signal handler. Before adding something to os::abort(), make
1468 // sure it is async-safe and can handle partially initialized VM.
1469 void os::abort(bool dump_core) {
1470   os::shutdown();
1471   if (dump_core) {
1472 #ifndef PRODUCT
1473     fdStream out(defaultStream::output_fd());
1474     out.print_raw("Current thread is ");
1475     char buf[16];
1476     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1477     out.print_raw_cr(buf);
1478     out.print_raw_cr("Dumping core ...");
1479 #endif
1480     ::abort(); // dump core
1481   }
1482 
1483   ::exit(1);
1484 }
1485 
1486 // Die immediately, no exit hook, no abort hook, no cleanup.
1487 void os::die() {
1488   // _exit() on LinuxThreads only kills current thread
1489   ::abort();
1490 }
1491 
1492 // unused on linux for now.
1493 void os::set_error_file(const char *logfile) {}
1494 
1495 intx os::current_thread_id() { return (intx)pthread_self(); }
1496 int os::current_process_id() {
1497 
1498   // Under the old linux thread library, linux gives each thread
1499   // its own process id. Because of this each thread will return
1500   // a different pid if this method were to return the result
1501   // of getpid(2). Linux provides no api that returns the pid
1502   // of the launcher thread for the vm. This implementation
1503   // returns a unique pid, the pid of the launcher thread
1504   // that starts the vm 'process'.
1505 
1506   // Under the NPTL, getpid() returns the same pid as the
1507   // launcher thread rather than a unique pid per thread.
1508   // Use gettid() if you want the old pre NPTL behaviour.
1509 
1510   // if you are looking for the result of a call to getpid() that
1511   // returns a unique pid for the calling thread, then look at the
1512   // OSThread::thread_id() method in osThread_linux.hpp file
1513 
1514   return (int)(_initial_pid ? _initial_pid : getpid());
1515 }
1516 
1517 // DLL functions
1518 
1519 const char* os::dll_file_extension() { return ".so"; }
1520 
1521 const char* os::get_temp_directory() { return "/tmp/"; }
1522 
1523 static bool file_exists(const char* filename) {
1524   struct stat statbuf;
1525   if (filename == NULL || strlen(filename) == 0) {
1526     return false;
1527   }
1528   return os::stat(filename, &statbuf) == 0;
1529 }
1530 
1531 void os::dll_build_name(char* buffer, size_t buflen,
1532                         const char* pname, const char* fname) {
1533   // Copied from libhpi
1534   const size_t pnamelen = pname ? strlen(pname) : 0;
1535 
1536   // Quietly truncate on buffer overflow.  Should be an error.
1537   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1538       *buffer = '\0';
1539       return;
1540   }
1541 
1542   if (pnamelen == 0) {
1543     snprintf(buffer, buflen, "lib%s.so", fname);
1544   } else if (strchr(pname, *os::path_separator()) != NULL) {
1545     int n;
1546     char** pelements = split_path(pname, &n);
1547     for (int i = 0 ; i < n ; i++) {
1548       // Really shouldn't be NULL, but check can't hurt
1549       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1550         continue; // skip the empty path values
1551       }
1552       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1553       if (file_exists(buffer)) {
1554         break;
1555       }
1556     }
1557     // release the storage
1558     for (int i = 0 ; i < n ; i++) {
1559       if (pelements[i] != NULL) {
1560         FREE_C_HEAP_ARRAY(char, pelements[i]);
1561       }
1562     }
1563     if (pelements != NULL) {
1564       FREE_C_HEAP_ARRAY(char*, pelements);
1565     }
1566   } else {
1567     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1568   }
1569 }
1570 
1571 const char* os::get_current_directory(char *buf, int buflen) {
1572   return getcwd(buf, buflen);
1573 }
1574 
1575 // check if addr is inside libjvm[_g].so
1576 bool os::address_is_in_vm(address addr) {
1577   static address libjvm_base_addr;
1578   Dl_info dlinfo;
1579 
1580   if (libjvm_base_addr == NULL) {
1581     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1582     libjvm_base_addr = (address)dlinfo.dli_fbase;
1583     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1584   }
1585 
1586   if (dladdr((void *)addr, &dlinfo)) {
1587     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1588   }
1589 
1590   return false;
1591 }
1592 
1593 bool os::dll_address_to_function_name(address addr, char *buf,
1594                                       int buflen, int *offset) {
1595   Dl_info dlinfo;
1596 
1597   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1598     if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1599     if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1600     return true;
1601   } else {
1602     if (buf) buf[0] = '\0';
1603     if (offset) *offset = -1;
1604     return false;
1605   }
1606 }
1607 
1608 struct _address_to_library_name {
1609   address addr;          // input : memory address
1610   size_t  buflen;        //         size of fname
1611   char*   fname;         // output: library name
1612   address base;          //         library base addr
1613 };
1614 
1615 static int address_to_library_name_callback(struct dl_phdr_info *info,
1616                                             size_t size, void *data) {
1617   int i;
1618   bool found = false;
1619   address libbase = NULL;
1620   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1621 
1622   // iterate through all loadable segments
1623   for (i = 0; i < info->dlpi_phnum; i++) {
1624     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1625     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1626       // base address of a library is the lowest address of its loaded
1627       // segments.
1628       if (libbase == NULL || libbase > segbase) {
1629         libbase = segbase;
1630       }
1631       // see if 'addr' is within current segment
1632       if (segbase <= d->addr &&
1633           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1634         found = true;
1635       }
1636     }
1637   }
1638 
1639   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1640   // so dll_address_to_library_name() can fall through to use dladdr() which
1641   // can figure out executable name from argv[0].
1642   if (found && info->dlpi_name && info->dlpi_name[0]) {
1643     d->base = libbase;
1644     if (d->fname) {
1645       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1646     }
1647     return 1;
1648   }
1649   return 0;
1650 }
1651 
1652 bool os::dll_address_to_library_name(address addr, char* buf,
1653                                      int buflen, int* offset) {
1654   Dl_info dlinfo;
1655   struct _address_to_library_name data;
1656 
1657   // There is a bug in old glibc dladdr() implementation that it could resolve
1658   // to wrong library name if the .so file has a base address != NULL. Here
1659   // we iterate through the program headers of all loaded libraries to find
1660   // out which library 'addr' really belongs to. This workaround can be
1661   // removed once the minimum requirement for glibc is moved to 2.3.x.
1662   data.addr = addr;
1663   data.fname = buf;
1664   data.buflen = buflen;
1665   data.base = NULL;
1666   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1667 
1668   if (rslt) {
1669      // buf already contains library name
1670      if (offset) *offset = addr - data.base;
1671      return true;
1672   } else if (dladdr((void*)addr, &dlinfo)){
1673      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1674      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1675      return true;
1676   } else {
1677      if (buf) buf[0] = '\0';
1678      if (offset) *offset = -1;
1679      return false;
1680   }
1681 }
1682 
1683   // Loads .dll/.so and
1684   // in case of error it checks if .dll/.so was built for the
1685   // same architecture as Hotspot is running on
1686 
1687 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1688 {
1689   void * result= ::dlopen(filename, RTLD_LAZY);
1690   if (result != NULL) {
1691     // Successful loading
1692     return result;
1693   }
1694 
1695   Elf32_Ehdr elf_head;
1696 
1697   // Read system error message into ebuf
1698   // It may or may not be overwritten below
1699   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1700   ebuf[ebuflen-1]='\0';
1701   int diag_msg_max_length=ebuflen-strlen(ebuf);
1702   char* diag_msg_buf=ebuf+strlen(ebuf);
1703 
1704   if (diag_msg_max_length==0) {
1705     // No more space in ebuf for additional diagnostics message
1706     return NULL;
1707   }
1708 
1709 
1710   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1711 
1712   if (file_descriptor < 0) {
1713     // Can't open library, report dlerror() message
1714     return NULL;
1715   }
1716 
1717   bool failed_to_read_elf_head=
1718     (sizeof(elf_head)!=
1719         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1720 
1721   ::close(file_descriptor);
1722   if (failed_to_read_elf_head) {
1723     // file i/o error - report dlerror() msg
1724     return NULL;
1725   }
1726 
1727   typedef struct {
1728     Elf32_Half  code;         // Actual value as defined in elf.h
1729     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1730     char        elf_class;    // 32 or 64 bit
1731     char        endianess;    // MSB or LSB
1732     char*       name;         // String representation
1733   } arch_t;
1734 
1735   #ifndef EM_486
1736   #define EM_486          6               /* Intel 80486 */
1737   #endif
1738 
1739   static const arch_t arch_array[]={
1740     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1741     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1742     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1743     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1744     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1745     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1746     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1747     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1748     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1749     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1750     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1751     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1752     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1753     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1754     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1755     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1756   };
1757 
1758   #if  (defined IA32)
1759     static  Elf32_Half running_arch_code=EM_386;
1760   #elif   (defined AMD64)
1761     static  Elf32_Half running_arch_code=EM_X86_64;
1762   #elif  (defined IA64)
1763     static  Elf32_Half running_arch_code=EM_IA_64;
1764   #elif  (defined __sparc) && (defined _LP64)
1765     static  Elf32_Half running_arch_code=EM_SPARCV9;
1766   #elif  (defined __sparc) && (!defined _LP64)
1767     static  Elf32_Half running_arch_code=EM_SPARC;
1768   #elif  (defined __powerpc64__)
1769     static  Elf32_Half running_arch_code=EM_PPC64;
1770   #elif  (defined __powerpc__)
1771     static  Elf32_Half running_arch_code=EM_PPC;
1772   #elif  (defined ARM)
1773     static  Elf32_Half running_arch_code=EM_ARM;
1774   #elif  (defined S390)
1775     static  Elf32_Half running_arch_code=EM_S390;
1776   #elif  (defined ALPHA)
1777     static  Elf32_Half running_arch_code=EM_ALPHA;
1778   #elif  (defined MIPSEL)
1779     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1780   #elif  (defined PARISC)
1781     static  Elf32_Half running_arch_code=EM_PARISC;
1782   #elif  (defined MIPS)
1783     static  Elf32_Half running_arch_code=EM_MIPS;
1784   #elif  (defined M68K)
1785     static  Elf32_Half running_arch_code=EM_68K;
1786   #else
1787     #error Method os::dll_load requires that one of following is defined:\
1788          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1789   #endif
1790 
1791   // Identify compatability class for VM's architecture and library's architecture
1792   // Obtain string descriptions for architectures
1793 
1794   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1795   int running_arch_index=-1;
1796 
1797   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1798     if (running_arch_code == arch_array[i].code) {
1799       running_arch_index    = i;
1800     }
1801     if (lib_arch.code == arch_array[i].code) {
1802       lib_arch.compat_class = arch_array[i].compat_class;
1803       lib_arch.name         = arch_array[i].name;
1804     }
1805   }
1806 
1807   assert(running_arch_index != -1,
1808     "Didn't find running architecture code (running_arch_code) in arch_array");
1809   if (running_arch_index == -1) {
1810     // Even though running architecture detection failed
1811     // we may still continue with reporting dlerror() message
1812     return NULL;
1813   }
1814 
1815   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1816     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1817     return NULL;
1818   }
1819 
1820 #ifndef S390
1821   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1822     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1823     return NULL;
1824   }
1825 #endif // !S390
1826 
1827   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1828     if ( lib_arch.name!=NULL ) {
1829       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1830         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1831         lib_arch.name, arch_array[running_arch_index].name);
1832     } else {
1833       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1834       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1835         lib_arch.code,
1836         arch_array[running_arch_index].name);
1837     }
1838   }
1839 
1840   return NULL;
1841 }
1842 
1843 /*
1844  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
1845  * chances are you might want to run the generated bits against glibc-2.0
1846  * libdl.so, so always use locking for any version of glibc.
1847  */
1848 void* os::dll_lookup(void* handle, const char* name) {
1849   pthread_mutex_lock(&dl_mutex);
1850   void* res = dlsym(handle, name);
1851   pthread_mutex_unlock(&dl_mutex);
1852   return res;
1853 }
1854 
1855 
1856 bool _print_ascii_file(const char* filename, outputStream* st) {
1857   int fd = open(filename, O_RDONLY);
1858   if (fd == -1) {
1859      return false;
1860   }
1861 
1862   char buf[32];
1863   int bytes;
1864   while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1865     st->print_raw(buf, bytes);
1866   }
1867 
1868   close(fd);
1869 
1870   return true;
1871 }
1872 
1873 void os::print_dll_info(outputStream *st) {
1874    st->print_cr("Dynamic libraries:");
1875 
1876    char fname[32];
1877    pid_t pid = os::Linux::gettid();
1878 
1879    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1880 
1881    if (!_print_ascii_file(fname, st)) {
1882      st->print("Can not get library information for pid = %d\n", pid);
1883    }
1884 }
1885 
1886 
1887 void os::print_os_info(outputStream* st) {
1888   st->print("OS:");
1889 
1890   // Try to identify popular distros.
1891   // Most Linux distributions have /etc/XXX-release file, which contains
1892   // the OS version string. Some have more than one /etc/XXX-release file
1893   // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1894   // so the order is important.
1895   if (!_print_ascii_file("/etc/mandrake-release", st) &&
1896       !_print_ascii_file("/etc/sun-release", st) &&
1897       !_print_ascii_file("/etc/redhat-release", st) &&
1898       !_print_ascii_file("/etc/SuSE-release", st) &&
1899       !_print_ascii_file("/etc/turbolinux-release", st) &&
1900       !_print_ascii_file("/etc/gentoo-release", st) &&
1901       !_print_ascii_file("/etc/debian_version", st)) {
1902       st->print("Linux");
1903   }
1904   st->cr();
1905 
1906   // kernel
1907   st->print("uname:");
1908   struct utsname name;
1909   uname(&name);
1910   st->print(name.sysname); st->print(" ");
1911   st->print(name.release); st->print(" ");
1912   st->print(name.version); st->print(" ");
1913   st->print(name.machine);
1914   st->cr();
1915 
1916   // Print warning if unsafe chroot environment detected
1917   if (unsafe_chroot_detected) {
1918     st->print("WARNING!! ");
1919     st->print_cr(unstable_chroot_error);
1920   }
1921 
1922   // libc, pthread
1923   st->print("libc:");
1924   st->print(os::Linux::glibc_version()); st->print(" ");
1925   st->print(os::Linux::libpthread_version()); st->print(" ");
1926   if (os::Linux::is_LinuxThreads()) {
1927      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
1928   }
1929   st->cr();
1930 
1931   // rlimit
1932   st->print("rlimit:");
1933   struct rlimit rlim;
1934 
1935   st->print(" STACK ");
1936   getrlimit(RLIMIT_STACK, &rlim);
1937   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1938   else st->print("%uk", rlim.rlim_cur >> 10);
1939 
1940   st->print(", CORE ");
1941   getrlimit(RLIMIT_CORE, &rlim);
1942   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1943   else st->print("%uk", rlim.rlim_cur >> 10);
1944 
1945   st->print(", NPROC ");
1946   getrlimit(RLIMIT_NPROC, &rlim);
1947   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1948   else st->print("%d", rlim.rlim_cur);
1949 
1950   st->print(", NOFILE ");
1951   getrlimit(RLIMIT_NOFILE, &rlim);
1952   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1953   else st->print("%d", rlim.rlim_cur);
1954 
1955   st->print(", AS ");
1956   getrlimit(RLIMIT_AS, &rlim);
1957   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1958   else st->print("%uk", rlim.rlim_cur >> 10);
1959   st->cr();
1960 
1961   // load average
1962   st->print("load average:");
1963   double loadavg[3];
1964   os::loadavg(loadavg, 3);
1965   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1966   st->cr();
1967 }
1968 
1969 void os::print_memory_info(outputStream* st) {
1970 
1971   st->print("Memory:");
1972   st->print(" %dk page", os::vm_page_size()>>10);
1973 
1974   // values in struct sysinfo are "unsigned long"
1975   struct sysinfo si;
1976   sysinfo(&si);
1977 
1978   st->print(", physical " UINT64_FORMAT "k",
1979             os::physical_memory() >> 10);
1980   st->print("(" UINT64_FORMAT "k free)",
1981             os::available_memory() >> 10);
1982   st->print(", swap " UINT64_FORMAT "k",
1983             ((jlong)si.totalswap * si.mem_unit) >> 10);
1984   st->print("(" UINT64_FORMAT "k free)",
1985             ((jlong)si.freeswap * si.mem_unit) >> 10);
1986   st->cr();
1987 }
1988 
1989 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
1990 // but they're the same for all the linux arch that we support
1991 // and they're the same for solaris but there's no common place to put this.
1992 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
1993                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
1994                           "ILL_COPROC", "ILL_BADSTK" };
1995 
1996 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
1997                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
1998                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
1999 
2000 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2001 
2002 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2003 
2004 void os::print_siginfo(outputStream* st, void* siginfo) {
2005   st->print("siginfo:");
2006 
2007   const int buflen = 100;
2008   char buf[buflen];
2009   siginfo_t *si = (siginfo_t*)siginfo;
2010   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2011   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2012     st->print("si_errno=%s", buf);
2013   } else {
2014     st->print("si_errno=%d", si->si_errno);
2015   }
2016   const int c = si->si_code;
2017   assert(c > 0, "unexpected si_code");
2018   switch (si->si_signo) {
2019   case SIGILL:
2020     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2021     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2022     break;
2023   case SIGFPE:
2024     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2025     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2026     break;
2027   case SIGSEGV:
2028     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2029     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2030     break;
2031   case SIGBUS:
2032     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2033     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2034     break;
2035   default:
2036     st->print(", si_code=%d", si->si_code);
2037     // no si_addr
2038   }
2039 
2040   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2041       UseSharedSpaces) {
2042     FileMapInfo* mapinfo = FileMapInfo::current_info();
2043     if (mapinfo->is_in_shared_space(si->si_addr)) {
2044       st->print("\n\nError accessing class data sharing archive."   \
2045                 " Mapped file inaccessible during execution, "      \
2046                 " possible disk/network problem.");
2047     }
2048   }
2049   st->cr();
2050 }
2051 
2052 
2053 static void print_signal_handler(outputStream* st, int sig,
2054                                  char* buf, size_t buflen);
2055 
2056 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2057   st->print_cr("Signal Handlers:");
2058   print_signal_handler(st, SIGSEGV, buf, buflen);
2059   print_signal_handler(st, SIGBUS , buf, buflen);
2060   print_signal_handler(st, SIGFPE , buf, buflen);
2061   print_signal_handler(st, SIGPIPE, buf, buflen);
2062   print_signal_handler(st, SIGXFSZ, buf, buflen);
2063   print_signal_handler(st, SIGILL , buf, buflen);
2064   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2065   print_signal_handler(st, SR_signum, buf, buflen);
2066   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2067   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2068   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2069   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2070 }
2071 
2072 static char saved_jvm_path[MAXPATHLEN] = {0};
2073 
2074 // Find the full path to the current module, libjvm.so or libjvm_g.so
2075 void os::jvm_path(char *buf, jint len) {
2076   // Error checking.
2077   if (len < MAXPATHLEN) {
2078     assert(false, "must use a large-enough buffer");
2079     buf[0] = '\0';
2080     return;
2081   }
2082   // Lazy resolve the path to current module.
2083   if (saved_jvm_path[0] != 0) {
2084     strcpy(buf, saved_jvm_path);
2085     return;
2086   }
2087 
2088   char dli_fname[MAXPATHLEN];
2089   bool ret = dll_address_to_library_name(
2090                 CAST_FROM_FN_PTR(address, os::jvm_path),
2091                 dli_fname, sizeof(dli_fname), NULL);
2092   assert(ret != 0, "cannot locate libjvm");
2093   if (realpath(dli_fname, buf) == NULL)
2094     return;
2095 
2096   if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2097     // Support for the gamma launcher.  Typical value for buf is
2098     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2099     // the right place in the string, then assume we are installed in a JDK and
2100     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2101     // up the path so it looks like libjvm.so is installed there (append a
2102     // fake suffix hotspot/libjvm.so).
2103     const char *p = buf + strlen(buf) - 1;
2104     for (int count = 0; p > buf && count < 5; ++count) {
2105       for (--p; p > buf && *p != '/'; --p)
2106         /* empty */ ;
2107     }
2108 
2109     if (strncmp(p, "/jre/lib/", 9) != 0) {
2110       // Look for JAVA_HOME in the environment.
2111       char* java_home_var = ::getenv("JAVA_HOME");
2112       if (java_home_var != NULL && java_home_var[0] != 0) {
2113         // Check the current module name "libjvm.so" or "libjvm_g.so".
2114         p = strrchr(buf, '/');
2115         assert(strstr(p, "/libjvm") == p, "invalid library name");
2116         p = strstr(p, "_g") ? "_g" : "";
2117 
2118         if (realpath(java_home_var, buf) == NULL)
2119           return;
2120         sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch);
2121         if (0 == access(buf, F_OK)) {
2122           // Use current module name "libjvm[_g].so" instead of
2123           // "libjvm"debug_only("_g")".so" since for fastdebug version
2124           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2125           // It is used when we are choosing the HPI library's name
2126           // "libhpi[_g].so" in hpi::initialize_get_interface().
2127           sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p);
2128         } else {
2129           // Go back to path of .so
2130           if (realpath(dli_fname, buf) == NULL)
2131             return;
2132         }
2133       }
2134     }
2135   }
2136 
2137   strcpy(saved_jvm_path, buf);
2138 }
2139 
2140 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2141   // no prefix required, not even "_"
2142 }
2143 
2144 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2145   // no suffix required
2146 }
2147 
2148 ////////////////////////////////////////////////////////////////////////////////
2149 // sun.misc.Signal support
2150 
2151 static volatile jint sigint_count = 0;
2152 
2153 static void
2154 UserHandler(int sig, void *siginfo, void *context) {
2155   // 4511530 - sem_post is serialized and handled by the manager thread. When
2156   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2157   // don't want to flood the manager thread with sem_post requests.
2158   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2159       return;
2160 
2161   // Ctrl-C is pressed during error reporting, likely because the error
2162   // handler fails to abort. Let VM die immediately.
2163   if (sig == SIGINT && is_error_reported()) {
2164      os::die();
2165   }
2166 
2167   os::signal_notify(sig);
2168 }
2169 
2170 void* os::user_handler() {
2171   return CAST_FROM_FN_PTR(void*, UserHandler);
2172 }
2173 
2174 extern "C" {
2175   typedef void (*sa_handler_t)(int);
2176   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2177 }
2178 
2179 void* os::signal(int signal_number, void* handler) {
2180   struct sigaction sigAct, oldSigAct;
2181 
2182   sigfillset(&(sigAct.sa_mask));
2183   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2184   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2185 
2186   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2187     // -1 means registration failed
2188     return (void *)-1;
2189   }
2190 
2191   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2192 }
2193 
2194 void os::signal_raise(int signal_number) {
2195   ::raise(signal_number);
2196 }
2197 
2198 /*
2199  * The following code is moved from os.cpp for making this
2200  * code platform specific, which it is by its very nature.
2201  */
2202 
2203 // Will be modified when max signal is changed to be dynamic
2204 int os::sigexitnum_pd() {
2205   return NSIG;
2206 }
2207 
2208 // a counter for each possible signal value
2209 static volatile jint pending_signals[NSIG+1] = { 0 };
2210 
2211 // Linux(POSIX) specific hand shaking semaphore.
2212 static sem_t sig_sem;
2213 
2214 void os::signal_init_pd() {
2215   // Initialize signal structures
2216   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2217 
2218   // Initialize signal semaphore
2219   ::sem_init(&sig_sem, 0, 0);
2220 }
2221 
2222 void os::signal_notify(int sig) {
2223   Atomic::inc(&pending_signals[sig]);
2224   ::sem_post(&sig_sem);
2225 }
2226 
2227 static int check_pending_signals(bool wait) {
2228   Atomic::store(0, &sigint_count);
2229   for (;;) {
2230     for (int i = 0; i < NSIG + 1; i++) {
2231       jint n = pending_signals[i];
2232       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2233         return i;
2234       }
2235     }
2236     if (!wait) {
2237       return -1;
2238     }
2239     JavaThread *thread = JavaThread::current();
2240     ThreadBlockInVM tbivm(thread);
2241 
2242     bool threadIsSuspended;
2243     do {
2244       thread->set_suspend_equivalent();
2245       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2246       ::sem_wait(&sig_sem);
2247 
2248       // were we externally suspended while we were waiting?
2249       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2250       if (threadIsSuspended) {
2251         //
2252         // The semaphore has been incremented, but while we were waiting
2253         // another thread suspended us. We don't want to continue running
2254         // while suspended because that would surprise the thread that
2255         // suspended us.
2256         //
2257         ::sem_post(&sig_sem);
2258 
2259         thread->java_suspend_self();
2260       }
2261     } while (threadIsSuspended);
2262   }
2263 }
2264 
2265 int os::signal_lookup() {
2266   return check_pending_signals(false);
2267 }
2268 
2269 int os::signal_wait() {
2270   return check_pending_signals(true);
2271 }
2272 
2273 ////////////////////////////////////////////////////////////////////////////////
2274 // Virtual Memory
2275 
2276 int os::vm_page_size() {
2277   // Seems redundant as all get out
2278   assert(os::Linux::page_size() != -1, "must call os::init");
2279   return os::Linux::page_size();
2280 }
2281 
2282 // Solaris allocates memory by pages.
2283 int os::vm_allocation_granularity() {
2284   assert(os::Linux::page_size() != -1, "must call os::init");
2285   return os::Linux::page_size();
2286 }
2287 
2288 // Rationale behind this function:
2289 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2290 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2291 //  samples for JITted code. Here we create private executable mapping over the code cache
2292 //  and then we can use standard (well, almost, as mapping can change) way to provide
2293 //  info for the reporting script by storing timestamp and location of symbol
2294 void linux_wrap_code(char* base, size_t size) {
2295   static volatile jint cnt = 0;
2296 
2297   if (!UseOprofile) {
2298     return;
2299   }
2300 
2301   char buf[40];
2302   int num = Atomic::add(1, &cnt);
2303 
2304   sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num);
2305   unlink(buf);
2306 
2307   int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2308 
2309   if (fd != -1) {
2310     off_t rv = lseek(fd, size-2, SEEK_SET);
2311     if (rv != (off_t)-1) {
2312       if (write(fd, "", 1) == 1) {
2313         mmap(base, size,
2314              PROT_READ|PROT_WRITE|PROT_EXEC,
2315              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2316       }
2317     }
2318     close(fd);
2319     unlink(buf);
2320   }
2321 }
2322 
2323 // NOTE: Linux kernel does not really reserve the pages for us.
2324 //       All it does is to check if there are enough free pages
2325 //       left at the time of mmap(). This could be a potential
2326 //       problem.
2327 bool os::commit_memory(char* addr, size_t size, bool exec) {
2328   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2329   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2330                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2331   return res != (uintptr_t) MAP_FAILED;
2332 }
2333 
2334 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2335                        bool exec) {
2336   return commit_memory(addr, size, exec);
2337 }
2338 
2339 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2340 
2341 void os::free_memory(char *addr, size_t bytes) {
2342   ::mmap(addr, bytes, PROT_READ | PROT_WRITE,
2343          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2344 }
2345 
2346 void os::numa_make_global(char *addr, size_t bytes) {
2347   Linux::numa_interleave_memory(addr, bytes);
2348 }
2349 
2350 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2351   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2352 }
2353 
2354 bool os::numa_topology_changed()   { return false; }
2355 
2356 size_t os::numa_get_groups_num() {
2357   int max_node = Linux::numa_max_node();
2358   return max_node > 0 ? max_node + 1 : 1;
2359 }
2360 
2361 int os::numa_get_group_id() {
2362   int cpu_id = Linux::sched_getcpu();
2363   if (cpu_id != -1) {
2364     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2365     if (lgrp_id != -1) {
2366       return lgrp_id;
2367     }
2368   }
2369   return 0;
2370 }
2371 
2372 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2373   for (size_t i = 0; i < size; i++) {
2374     ids[i] = i;
2375   }
2376   return size;
2377 }
2378 
2379 bool os::get_page_info(char *start, page_info* info) {
2380   return false;
2381 }
2382 
2383 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2384   return end;
2385 }
2386 
2387 extern "C" void numa_warn(int number, char *where, ...) { }
2388 extern "C" void numa_error(char *where) { }
2389 
2390 
2391 // If we are running with libnuma version > 2, then we should
2392 // be trying to use symbols with versions 1.1
2393 // If we are running with earlier version, which did not have symbol versions,
2394 // we should use the base version.
2395 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2396   void *f = dlvsym(handle, name, "libnuma_1.1");
2397   if (f == NULL) {
2398     f = dlsym(handle, name);
2399   }
2400   return f;
2401 }
2402 
2403 bool os::Linux::libnuma_init() {
2404   // sched_getcpu() should be in libc.
2405   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2406                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2407 
2408   if (sched_getcpu() != -1) { // Does it work?
2409     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2410     if (handle != NULL) {
2411       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2412                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2413       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2414                                        libnuma_dlsym(handle, "numa_max_node")));
2415       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2416                                         libnuma_dlsym(handle, "numa_available")));
2417       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2418                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2419       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2420                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2421 
2422 
2423       if (numa_available() != -1) {
2424         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2425         // Create a cpu -> node mapping
2426         _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2427         rebuild_cpu_to_node_map();
2428         return true;
2429       }
2430     }
2431   }
2432   return false;
2433 }
2434 
2435 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2436 // The table is later used in get_node_by_cpu().
2437 void os::Linux::rebuild_cpu_to_node_map() {
2438   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2439                               // in libnuma (possible values are starting from 16,
2440                               // and continuing up with every other power of 2, but less
2441                               // than the maximum number of CPUs supported by kernel), and
2442                               // is a subject to change (in libnuma version 2 the requirements
2443                               // are more reasonable) we'll just hardcode the number they use
2444                               // in the library.
2445   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2446 
2447   size_t cpu_num = os::active_processor_count();
2448   size_t cpu_map_size = NCPUS / BitsPerCLong;
2449   size_t cpu_map_valid_size =
2450     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2451 
2452   cpu_to_node()->clear();
2453   cpu_to_node()->at_grow(cpu_num - 1);
2454   size_t node_num = numa_get_groups_num();
2455 
2456   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2457   for (size_t i = 0; i < node_num; i++) {
2458     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2459       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2460         if (cpu_map[j] != 0) {
2461           for (size_t k = 0; k < BitsPerCLong; k++) {
2462             if (cpu_map[j] & (1UL << k)) {
2463               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2464             }
2465           }
2466         }
2467       }
2468     }
2469   }
2470   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2471 }
2472 
2473 int os::Linux::get_node_by_cpu(int cpu_id) {
2474   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2475     return cpu_to_node()->at(cpu_id);
2476   }
2477   return -1;
2478 }
2479 
2480 GrowableArray<int>* os::Linux::_cpu_to_node;
2481 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2482 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2483 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2484 os::Linux::numa_available_func_t os::Linux::_numa_available;
2485 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2486 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2487 unsigned long* os::Linux::_numa_all_nodes;
2488 
2489 bool os::uncommit_memory(char* addr, size_t size) {
2490   return ::mmap(addr, size, PROT_NONE,
2491                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0)
2492     != MAP_FAILED;
2493 }
2494 
























































































2495 static address _highest_vm_reserved_address = NULL;
2496 
2497 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2498 // at 'requested_addr'. If there are existing memory mappings at the same
2499 // location, however, they will be overwritten. If 'fixed' is false,
2500 // 'requested_addr' is only treated as a hint, the return value may or
2501 // may not start from the requested address. Unlike Linux mmap(), this
2502 // function returns NULL to indicate failure.
2503 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2504   char * addr;
2505   int flags;
2506 
2507   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2508   if (fixed) {
2509     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2510     flags |= MAP_FIXED;
2511   }
2512 
2513   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2514   // to PROT_EXEC if executable when we commit the page.
2515   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2516                        flags, -1, 0);
2517 
2518   if (addr != MAP_FAILED) {
2519     // anon_mmap() should only get called during VM initialization,
2520     // don't need lock (actually we can skip locking even it can be called
2521     // from multiple threads, because _highest_vm_reserved_address is just a
2522     // hint about the upper limit of non-stack memory regions.)
2523     if ((address)addr + bytes > _highest_vm_reserved_address) {
2524       _highest_vm_reserved_address = (address)addr + bytes;
2525     }
2526   }
2527 
2528   return addr == MAP_FAILED ? NULL : addr;
2529 }
2530 
2531 // Don't update _highest_vm_reserved_address, because there might be memory
2532 // regions above addr + size. If so, releasing a memory region only creates
2533 // a hole in the address space, it doesn't help prevent heap-stack collision.
2534 //
2535 static int anon_munmap(char * addr, size_t size) {
2536   return ::munmap(addr, size) == 0;
2537 }
2538 
2539 char* os::reserve_memory(size_t bytes, char* requested_addr,
2540                          size_t alignment_hint) {
2541   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2542 }
2543 
2544 bool os::release_memory(char* addr, size_t size) {
2545   return anon_munmap(addr, size);
2546 }
2547 
2548 static address highest_vm_reserved_address() {
2549   return _highest_vm_reserved_address;
2550 }
2551 
2552 static bool linux_mprotect(char* addr, size_t size, int prot) {
2553   // Linux wants the mprotect address argument to be page aligned.
2554   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2555 
2556   // According to SUSv3, mprotect() should only be used with mappings
2557   // established by mmap(), and mmap() always maps whole pages. Unaligned
2558   // 'addr' likely indicates problem in the VM (e.g. trying to change
2559   // protection of malloc'ed or statically allocated memory). Check the
2560   // caller if you hit this assert.
2561   assert(addr == bottom, "sanity check");
2562 
2563   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2564   return ::mprotect(bottom, size, prot) == 0;
2565 }
2566 
2567 // Set protections specified
2568 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2569                         bool is_committed) {
2570   unsigned int p = 0;
2571   switch (prot) {
2572   case MEM_PROT_NONE: p = PROT_NONE; break;
2573   case MEM_PROT_READ: p = PROT_READ; break;
2574   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2575   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2576   default:
2577     ShouldNotReachHere();
2578   }
2579   // is_committed is unused.
2580   return linux_mprotect(addr, bytes, p);
2581 }
2582 
2583 bool os::guard_memory(char* addr, size_t size) {
2584   return linux_mprotect(addr, size, PROT_NONE);
2585 }
2586 
2587 bool os::unguard_memory(char* addr, size_t size) {
2588   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2589 }
2590 
2591 // Large page support
2592 
2593 static size_t _large_page_size = 0;
2594 
2595 bool os::large_page_init() {
2596   if (!UseLargePages) return false;
2597 
2598   if (LargePageSizeInBytes) {
2599     _large_page_size = LargePageSizeInBytes;
2600   } else {
2601     // large_page_size on Linux is used to round up heap size. x86 uses either
2602     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2603     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2604     // page as large as 256M.
2605     //
2606     // Here we try to figure out page size by parsing /proc/meminfo and looking
2607     // for a line with the following format:
2608     //    Hugepagesize:     2048 kB
2609     //
2610     // If we can't determine the value (e.g. /proc is not mounted, or the text
2611     // format has been changed), we'll use the largest page size supported by
2612     // the processor.
2613 
2614 #ifndef ZERO
2615     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M);
2616 #endif // ZERO
2617 
2618     FILE *fp = fopen("/proc/meminfo", "r");
2619     if (fp) {
2620       while (!feof(fp)) {
2621         int x = 0;
2622         char buf[16];
2623         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2624           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2625             _large_page_size = x * K;
2626             break;
2627           }
2628         } else {
2629           // skip to next line
2630           for (;;) {
2631             int ch = fgetc(fp);
2632             if (ch == EOF || ch == (int)'\n') break;
2633           }
2634         }
2635       }
2636       fclose(fp);
2637     }
2638   }
2639 
2640   const size_t default_page_size = (size_t)Linux::page_size();
2641   if (_large_page_size > default_page_size) {
2642     _page_sizes[0] = _large_page_size;
2643     _page_sizes[1] = default_page_size;
2644     _page_sizes[2] = 0;
2645   }
2646 
2647   // Large page support is available on 2.6 or newer kernel, some vendors
2648   // (e.g. Redhat) have backported it to their 2.4 based distributions.
2649   // We optimistically assume the support is available. If later it turns out
2650   // not true, VM will automatically switch to use regular page size.
2651   return true;
2652 }
2653 
2654 #ifndef SHM_HUGETLB
2655 #define SHM_HUGETLB 04000
2656 #endif
2657 
2658 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
2659   // "exec" is passed in but not used.  Creating the shared image for
2660   // the code cache doesn't have an SHM_X executable permission to check.
2661   assert(UseLargePages, "only for large pages");
2662 
2663   key_t key = IPC_PRIVATE;
2664   char *addr;
2665 
2666   bool warn_on_failure = UseLargePages &&
2667                         (!FLAG_IS_DEFAULT(UseLargePages) ||
2668                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2669                         );
2670   char msg[128];
2671 
2672   // Create a large shared memory region to attach to based on size.
2673   // Currently, size is the total size of the heap
2674   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2675   if (shmid == -1) {
2676      // Possible reasons for shmget failure:
2677      // 1. shmmax is too small for Java heap.
2678      //    > check shmmax value: cat /proc/sys/kernel/shmmax
2679      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2680      // 2. not enough large page memory.
2681      //    > check available large pages: cat /proc/meminfo
2682      //    > increase amount of large pages:
2683      //          echo new_value > /proc/sys/vm/nr_hugepages
2684      //      Note 1: different Linux may use different name for this property,
2685      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
2686      //      Note 2: it's possible there's enough physical memory available but
2687      //            they are so fragmented after a long run that they can't
2688      //            coalesce into large pages. Try to reserve large pages when
2689      //            the system is still "fresh".
2690      if (warn_on_failure) {
2691        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2692        warning(msg);
2693      }
2694      return NULL;
2695   }
2696 
2697   // attach to the region
2698   addr = (char*)shmat(shmid, NULL, 0);
2699   int err = errno;
2700 
2701   // Remove shmid. If shmat() is successful, the actual shared memory segment
2702   // will be deleted when it's detached by shmdt() or when the process
2703   // terminates. If shmat() is not successful this will remove the shared
2704   // segment immediately.
2705   shmctl(shmid, IPC_RMID, NULL);
2706 
2707   if ((intptr_t)addr == -1) {
2708      if (warn_on_failure) {
2709        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2710        warning(msg);
2711      }
2712      return NULL;
2713   }
2714 
2715   return addr;
2716 }
2717 
2718 bool os::release_memory_special(char* base, size_t bytes) {
2719   // detaching the SHM segment will also delete it, see reserve_memory_special()
2720   int rslt = shmdt(base);
2721   return rslt == 0;
2722 }
2723 
2724 size_t os::large_page_size() {
2725   return _large_page_size;
2726 }
2727 
2728 // Linux does not support anonymous mmap with large page memory. The only way
2729 // to reserve large page memory without file backing is through SysV shared
2730 // memory API. The entire memory region is committed and pinned upfront.
2731 // Hopefully this will change in the future...
2732 bool os::can_commit_large_page_memory() {
2733   return false;
2734 }
2735 
2736 bool os::can_execute_large_page_memory() {
2737   return false;
2738 }
2739 
2740 // Reserve memory at an arbitrary address, only if that area is
2741 // available (and not reserved for something else).
2742 
2743 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2744   const int max_tries = 10;
2745   char* base[max_tries];
2746   size_t size[max_tries];
2747   const size_t gap = 0x000000;
2748 
2749   // Assert only that the size is a multiple of the page size, since
2750   // that's all that mmap requires, and since that's all we really know
2751   // about at this low abstraction level.  If we need higher alignment,
2752   // we can either pass an alignment to this method or verify alignment
2753   // in one of the methods further up the call chain.  See bug 5044738.
2754   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2755 
2756   // Repeatedly allocate blocks until the block is allocated at the
2757   // right spot. Give up after max_tries. Note that reserve_memory() will
2758   // automatically update _highest_vm_reserved_address if the call is
2759   // successful. The variable tracks the highest memory address every reserved
2760   // by JVM. It is used to detect heap-stack collision if running with
2761   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2762   // space than needed, it could confuse the collision detecting code. To
2763   // solve the problem, save current _highest_vm_reserved_address and
2764   // calculate the correct value before return.
2765   address old_highest = _highest_vm_reserved_address;
2766 
2767   // Linux mmap allows caller to pass an address as hint; give it a try first,
2768   // if kernel honors the hint then we can return immediately.
2769   char * addr = anon_mmap(requested_addr, bytes, false);
2770   if (addr == requested_addr) {
2771      return requested_addr;
2772   }
2773 
2774   if (addr != NULL) {
2775      // mmap() is successful but it fails to reserve at the requested address
2776      anon_munmap(addr, bytes);
2777   }
2778 
2779   int i;
2780   for (i = 0; i < max_tries; ++i) {
2781     base[i] = reserve_memory(bytes);
2782 
2783     if (base[i] != NULL) {
2784       // Is this the block we wanted?
2785       if (base[i] == requested_addr) {
2786         size[i] = bytes;
2787         break;
2788       }
2789 
2790       // Does this overlap the block we wanted? Give back the overlapped
2791       // parts and try again.
2792 
2793       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2794       if (top_overlap >= 0 && top_overlap < bytes) {
2795         unmap_memory(base[i], top_overlap);
2796         base[i] += top_overlap;
2797         size[i] = bytes - top_overlap;
2798       } else {
2799         size_t bottom_overlap = base[i] + bytes - requested_addr;
2800         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2801           unmap_memory(requested_addr, bottom_overlap);
2802           size[i] = bytes - bottom_overlap;
2803         } else {
2804           size[i] = bytes;
2805         }
2806       }
2807     }
2808   }
2809 
2810   // Give back the unused reserved pieces.
2811 
2812   for (int j = 0; j < i; ++j) {
2813     if (base[j] != NULL) {
2814       unmap_memory(base[j], size[j]);
2815     }
2816   }
2817 
2818   if (i < max_tries) {
2819     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2820     return requested_addr;
2821   } else {
2822     _highest_vm_reserved_address = old_highest;
2823     return NULL;
2824   }
2825 }
2826 
2827 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2828   return ::read(fd, buf, nBytes);
2829 }
2830 
2831 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
2832 // Solaris uses poll(), linux uses park().
2833 // Poll() is likely a better choice, assuming that Thread.interrupt()
2834 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2835 // SIGSEGV, see 4355769.
2836 
2837 const int NANOSECS_PER_MILLISECS = 1000000;
2838 
2839 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2840   assert(thread == Thread::current(),  "thread consistency check");
2841 
2842   ParkEvent * const slp = thread->_SleepEvent ;
2843   slp->reset() ;
2844   OrderAccess::fence() ;
2845 
2846   if (interruptible) {
2847     jlong prevtime = javaTimeNanos();
2848 
2849     for (;;) {
2850       if (os::is_interrupted(thread, true)) {
2851         return OS_INTRPT;
2852       }
2853 
2854       jlong newtime = javaTimeNanos();
2855 
2856       if (newtime - prevtime < 0) {
2857         // time moving backwards, should only happen if no monotonic clock
2858         // not a guarantee() because JVM should not abort on kernel/glibc bugs
2859         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2860       } else {
2861         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2862       }
2863 
2864       if(millis <= 0) {
2865         return OS_OK;
2866       }
2867 
2868       prevtime = newtime;
2869 
2870       {
2871         assert(thread->is_Java_thread(), "sanity check");
2872         JavaThread *jt = (JavaThread *) thread;
2873         ThreadBlockInVM tbivm(jt);
2874         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
2875 
2876         jt->set_suspend_equivalent();
2877         // cleared by handle_special_suspend_equivalent_condition() or
2878         // java_suspend_self() via check_and_wait_while_suspended()
2879 
2880         slp->park(millis);
2881 
2882         // were we externally suspended while we were waiting?
2883         jt->check_and_wait_while_suspended();
2884       }
2885     }
2886   } else {
2887     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2888     jlong prevtime = javaTimeNanos();
2889 
2890     for (;;) {
2891       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
2892       // the 1st iteration ...
2893       jlong newtime = javaTimeNanos();
2894 
2895       if (newtime - prevtime < 0) {
2896         // time moving backwards, should only happen if no monotonic clock
2897         // not a guarantee() because JVM should not abort on kernel/glibc bugs
2898         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2899       } else {
2900         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2901       }
2902 
2903       if(millis <= 0) break ;
2904 
2905       prevtime = newtime;
2906       slp->park(millis);
2907     }
2908     return OS_OK ;
2909   }
2910 }
2911 
2912 int os::naked_sleep() {
2913   // %% make the sleep time an integer flag. for now use 1 millisec.
2914   return os::sleep(Thread::current(), 1, false);
2915 }
2916 
2917 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2918 void os::infinite_sleep() {
2919   while (true) {    // sleep forever ...
2920     ::sleep(100);   // ... 100 seconds at a time
2921   }
2922 }
2923 
2924 // Used to convert frequent JVM_Yield() to nops
2925 bool os::dont_yield() {
2926   return DontYieldALot;
2927 }
2928 
2929 void os::yield() {
2930   sched_yield();
2931 }
2932 
2933 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
2934 
2935 void os::yield_all(int attempts) {
2936   // Yields to all threads, including threads with lower priorities
2937   // Threads on Linux are all with same priority. The Solaris style
2938   // os::yield_all() with nanosleep(1ms) is not necessary.
2939   sched_yield();
2940 }
2941 
2942 // Called from the tight loops to possibly influence time-sharing heuristics
2943 void os::loop_breaker(int attempts) {
2944   os::yield_all(attempts);
2945 }
2946 
2947 ////////////////////////////////////////////////////////////////////////////////
2948 // thread priority support
2949 
2950 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
2951 // only supports dynamic priority, static priority must be zero. For real-time
2952 // applications, Linux supports SCHED_RR which allows static priority (1-99).
2953 // However, for large multi-threaded applications, SCHED_RR is not only slower
2954 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
2955 // of 5 runs - Sep 2005).
2956 //
2957 // The following code actually changes the niceness of kernel-thread/LWP. It
2958 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
2959 // not the entire user process, and user level threads are 1:1 mapped to kernel
2960 // threads. It has always been the case, but could change in the future. For
2961 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
2962 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
2963 
2964 int os::java_to_os_priority[MaxPriority + 1] = {
2965   19,              // 0 Entry should never be used
2966 
2967    4,              // 1 MinPriority
2968    3,              // 2
2969    2,              // 3
2970 
2971    1,              // 4
2972    0,              // 5 NormPriority
2973   -1,              // 6
2974 
2975   -2,              // 7
2976   -3,              // 8
2977   -4,              // 9 NearMaxPriority
2978 
2979   -5               // 10 MaxPriority
2980 };
2981 
2982 static int prio_init() {
2983   if (ThreadPriorityPolicy == 1) {
2984     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
2985     // if effective uid is not root. Perhaps, a more elegant way of doing
2986     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
2987     if (geteuid() != 0) {
2988       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
2989         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
2990       }
2991       ThreadPriorityPolicy = 0;
2992     }
2993   }
2994   return 0;
2995 }
2996 
2997 OSReturn os::set_native_priority(Thread* thread, int newpri) {
2998   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
2999 
3000   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3001   return (ret == 0) ? OS_OK : OS_ERR;
3002 }
3003 
3004 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3005   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3006     *priority_ptr = java_to_os_priority[NormPriority];
3007     return OS_OK;
3008   }
3009 
3010   errno = 0;
3011   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3012   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3013 }
3014 
3015 // Hint to the underlying OS that a task switch would not be good.
3016 // Void return because it's a hint and can fail.
3017 void os::hint_no_preempt() {}
3018 
3019 ////////////////////////////////////////////////////////////////////////////////
3020 // suspend/resume support
3021 
3022 //  the low-level signal-based suspend/resume support is a remnant from the
3023 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3024 //  within hotspot. Now there is a single use-case for this:
3025 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3026 //      that runs in the watcher thread.
3027 //  The remaining code is greatly simplified from the more general suspension
3028 //  code that used to be used.
3029 //
3030 //  The protocol is quite simple:
3031 //  - suspend:
3032 //      - sends a signal to the target thread
3033 //      - polls the suspend state of the osthread using a yield loop
3034 //      - target thread signal handler (SR_handler) sets suspend state
3035 //        and blocks in sigsuspend until continued
3036 //  - resume:
3037 //      - sets target osthread state to continue
3038 //      - sends signal to end the sigsuspend loop in the SR_handler
3039 //
3040 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3041 //
3042 
3043 static void resume_clear_context(OSThread *osthread) {
3044   osthread->set_ucontext(NULL);
3045   osthread->set_siginfo(NULL);
3046 
3047   // notify the suspend action is completed, we have now resumed
3048   osthread->sr.clear_suspended();
3049 }
3050 
3051 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3052   osthread->set_ucontext(context);
3053   osthread->set_siginfo(siginfo);
3054 }
3055 
3056 //
3057 // Handler function invoked when a thread's execution is suspended or
3058 // resumed. We have to be careful that only async-safe functions are
3059 // called here (Note: most pthread functions are not async safe and
3060 // should be avoided.)
3061 //
3062 // Note: sigwait() is a more natural fit than sigsuspend() from an
3063 // interface point of view, but sigwait() prevents the signal hander
3064 // from being run. libpthread would get very confused by not having
3065 // its signal handlers run and prevents sigwait()'s use with the
3066 // mutex granting granting signal.
3067 //
3068 // Currently only ever called on the VMThread
3069 //
3070 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3071   // Save and restore errno to avoid confusing native code with EINTR
3072   // after sigsuspend.
3073   int old_errno = errno;
3074 
3075   Thread* thread = Thread::current();
3076   OSThread* osthread = thread->osthread();
3077   assert(thread->is_VM_thread(), "Must be VMThread");
3078   // read current suspend action
3079   int action = osthread->sr.suspend_action();
3080   if (action == SR_SUSPEND) {
3081     suspend_save_context(osthread, siginfo, context);
3082 
3083     // Notify the suspend action is about to be completed. do_suspend()
3084     // waits until SR_SUSPENDED is set and then returns. We will wait
3085     // here for a resume signal and that completes the suspend-other
3086     // action. do_suspend/do_resume is always called as a pair from
3087     // the same thread - so there are no races
3088 
3089     // notify the caller
3090     osthread->sr.set_suspended();
3091 
3092     sigset_t suspend_set;  // signals for sigsuspend()
3093 
3094     // get current set of blocked signals and unblock resume signal
3095     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3096     sigdelset(&suspend_set, SR_signum);
3097 
3098     // wait here until we are resumed
3099     do {
3100       sigsuspend(&suspend_set);
3101       // ignore all returns until we get a resume signal
3102     } while (osthread->sr.suspend_action() != SR_CONTINUE);
3103 
3104     resume_clear_context(osthread);
3105 
3106   } else {
3107     assert(action == SR_CONTINUE, "unexpected sr action");
3108     // nothing special to do - just leave the handler
3109   }
3110 
3111   errno = old_errno;
3112 }
3113 
3114 
3115 static int SR_initialize() {
3116   struct sigaction act;
3117   char *s;
3118   /* Get signal number to use for suspend/resume */
3119   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3120     int sig = ::strtol(s, 0, 10);
3121     if (sig > 0 || sig < _NSIG) {
3122         SR_signum = sig;
3123     }
3124   }
3125 
3126   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3127         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3128 
3129   sigemptyset(&SR_sigset);
3130   sigaddset(&SR_sigset, SR_signum);
3131 
3132   /* Set up signal handler for suspend/resume */
3133   act.sa_flags = SA_RESTART|SA_SIGINFO;
3134   act.sa_handler = (void (*)(int)) SR_handler;
3135 
3136   // SR_signum is blocked by default.
3137   // 4528190 - We also need to block pthread restart signal (32 on all
3138   // supported Linux platforms). Note that LinuxThreads need to block
3139   // this signal for all threads to work properly. So we don't have
3140   // to use hard-coded signal number when setting up the mask.
3141   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3142 
3143   if (sigaction(SR_signum, &act, 0) == -1) {
3144     return -1;
3145   }
3146 
3147   // Save signal flag
3148   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3149   return 0;
3150 }
3151 
3152 static int SR_finalize() {
3153   return 0;
3154 }
3155 
3156 
3157 // returns true on success and false on error - really an error is fatal
3158 // but this seems the normal response to library errors
3159 static bool do_suspend(OSThread* osthread) {
3160   // mark as suspended and send signal
3161   osthread->sr.set_suspend_action(SR_SUSPEND);
3162   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3163   assert_status(status == 0, status, "pthread_kill");
3164 
3165   // check status and wait until notified of suspension
3166   if (status == 0) {
3167     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3168       os::yield_all(i);
3169     }
3170     osthread->sr.set_suspend_action(SR_NONE);
3171     return true;
3172   }
3173   else {
3174     osthread->sr.set_suspend_action(SR_NONE);
3175     return false;
3176   }
3177 }
3178 
3179 static void do_resume(OSThread* osthread) {
3180   assert(osthread->sr.is_suspended(), "thread should be suspended");
3181   osthread->sr.set_suspend_action(SR_CONTINUE);
3182 
3183   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3184   assert_status(status == 0, status, "pthread_kill");
3185   // check status and wait unit notified of resumption
3186   if (status == 0) {
3187     for (int i = 0; osthread->sr.is_suspended(); i++) {
3188       os::yield_all(i);
3189     }
3190   }
3191   osthread->sr.set_suspend_action(SR_NONE);
3192 }
3193 
3194 ////////////////////////////////////////////////////////////////////////////////
3195 // interrupt support
3196 
3197 void os::interrupt(Thread* thread) {
3198   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3199     "possibility of dangling Thread pointer");
3200 
3201   OSThread* osthread = thread->osthread();
3202 
3203   if (!osthread->interrupted()) {
3204     osthread->set_interrupted(true);
3205     // More than one thread can get here with the same value of osthread,
3206     // resulting in multiple notifications.  We do, however, want the store
3207     // to interrupted() to be visible to other threads before we execute unpark().
3208     OrderAccess::fence();
3209     ParkEvent * const slp = thread->_SleepEvent ;
3210     if (slp != NULL) slp->unpark() ;
3211   }
3212 
3213   // For JSR166. Unpark even if interrupt status already was set
3214   if (thread->is_Java_thread())
3215     ((JavaThread*)thread)->parker()->unpark();
3216 
3217   ParkEvent * ev = thread->_ParkEvent ;
3218   if (ev != NULL) ev->unpark() ;
3219 
3220 }
3221 
3222 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3223   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3224     "possibility of dangling Thread pointer");
3225 
3226   OSThread* osthread = thread->osthread();
3227 
3228   bool interrupted = osthread->interrupted();
3229 
3230   if (interrupted && clear_interrupted) {
3231     osthread->set_interrupted(false);
3232     // consider thread->_SleepEvent->reset() ... optional optimization
3233   }
3234 
3235   return interrupted;
3236 }
3237 
3238 ///////////////////////////////////////////////////////////////////////////////////
3239 // signal handling (except suspend/resume)
3240 
3241 // This routine may be used by user applications as a "hook" to catch signals.
3242 // The user-defined signal handler must pass unrecognized signals to this
3243 // routine, and if it returns true (non-zero), then the signal handler must
3244 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3245 // routine will never retun false (zero), but instead will execute a VM panic
3246 // routine kill the process.
3247 //
3248 // If this routine returns false, it is OK to call it again.  This allows
3249 // the user-defined signal handler to perform checks either before or after
3250 // the VM performs its own checks.  Naturally, the user code would be making
3251 // a serious error if it tried to handle an exception (such as a null check
3252 // or breakpoint) that the VM was generating for its own correct operation.
3253 //
3254 // This routine may recognize any of the following kinds of signals:
3255 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3256 // It should be consulted by handlers for any of those signals.
3257 //
3258 // The caller of this routine must pass in the three arguments supplied
3259 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3260 // field of the structure passed to sigaction().  This routine assumes that
3261 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3262 //
3263 // Note that the VM will print warnings if it detects conflicting signal
3264 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3265 //
3266 extern "C" int
3267 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3268                         void* ucontext, int abort_if_unrecognized);
3269 
3270 void signalHandler(int sig, siginfo_t* info, void* uc) {
3271   assert(info != NULL && uc != NULL, "it must be old kernel");
3272   JVM_handle_linux_signal(sig, info, uc, true);
3273 }
3274 
3275 
3276 // This boolean allows users to forward their own non-matching signals
3277 // to JVM_handle_linux_signal, harmlessly.
3278 bool os::Linux::signal_handlers_are_installed = false;
3279 
3280 // For signal-chaining
3281 struct sigaction os::Linux::sigact[MAXSIGNUM];
3282 unsigned int os::Linux::sigs = 0;
3283 bool os::Linux::libjsig_is_loaded = false;
3284 typedef struct sigaction *(*get_signal_t)(int);
3285 get_signal_t os::Linux::get_signal_action = NULL;
3286 
3287 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3288   struct sigaction *actp = NULL;
3289 
3290   if (libjsig_is_loaded) {
3291     // Retrieve the old signal handler from libjsig
3292     actp = (*get_signal_action)(sig);
3293   }
3294   if (actp == NULL) {
3295     // Retrieve the preinstalled signal handler from jvm
3296     actp = get_preinstalled_handler(sig);
3297   }
3298 
3299   return actp;
3300 }
3301 
3302 static bool call_chained_handler(struct sigaction *actp, int sig,
3303                                  siginfo_t *siginfo, void *context) {
3304   // Call the old signal handler
3305   if (actp->sa_handler == SIG_DFL) {
3306     // It's more reasonable to let jvm treat it as an unexpected exception
3307     // instead of taking the default action.
3308     return false;
3309   } else if (actp->sa_handler != SIG_IGN) {
3310     if ((actp->sa_flags & SA_NODEFER) == 0) {
3311       // automaticlly block the signal
3312       sigaddset(&(actp->sa_mask), sig);
3313     }
3314 
3315     sa_handler_t hand;
3316     sa_sigaction_t sa;
3317     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3318     // retrieve the chained handler
3319     if (siginfo_flag_set) {
3320       sa = actp->sa_sigaction;
3321     } else {
3322       hand = actp->sa_handler;
3323     }
3324 
3325     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3326       actp->sa_handler = SIG_DFL;
3327     }
3328 
3329     // try to honor the signal mask
3330     sigset_t oset;
3331     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3332 
3333     // call into the chained handler
3334     if (siginfo_flag_set) {
3335       (*sa)(sig, siginfo, context);
3336     } else {
3337       (*hand)(sig);
3338     }
3339 
3340     // restore the signal mask
3341     pthread_sigmask(SIG_SETMASK, &oset, 0);
3342   }
3343   // Tell jvm's signal handler the signal is taken care of.
3344   return true;
3345 }
3346 
3347 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3348   bool chained = false;
3349   // signal-chaining
3350   if (UseSignalChaining) {
3351     struct sigaction *actp = get_chained_signal_action(sig);
3352     if (actp != NULL) {
3353       chained = call_chained_handler(actp, sig, siginfo, context);
3354     }
3355   }
3356   return chained;
3357 }
3358 
3359 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3360   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3361     return &sigact[sig];
3362   }
3363   return NULL;
3364 }
3365 
3366 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3367   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3368   sigact[sig] = oldAct;
3369   sigs |= (unsigned int)1 << sig;
3370 }
3371 
3372 // for diagnostic
3373 int os::Linux::sigflags[MAXSIGNUM];
3374 
3375 int os::Linux::get_our_sigflags(int sig) {
3376   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3377   return sigflags[sig];
3378 }
3379 
3380 void os::Linux::set_our_sigflags(int sig, int flags) {
3381   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3382   sigflags[sig] = flags;
3383 }
3384 
3385 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3386   // Check for overwrite.
3387   struct sigaction oldAct;
3388   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3389 
3390   void* oldhand = oldAct.sa_sigaction
3391                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3392                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3393   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3394       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3395       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3396     if (AllowUserSignalHandlers || !set_installed) {
3397       // Do not overwrite; user takes responsibility to forward to us.
3398       return;
3399     } else if (UseSignalChaining) {
3400       // save the old handler in jvm
3401       save_preinstalled_handler(sig, oldAct);
3402       // libjsig also interposes the sigaction() call below and saves the
3403       // old sigaction on it own.
3404     } else {
3405       fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig);
3406     }
3407   }
3408 
3409   struct sigaction sigAct;
3410   sigfillset(&(sigAct.sa_mask));
3411   sigAct.sa_handler = SIG_DFL;
3412   if (!set_installed) {
3413     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3414   } else {
3415     sigAct.sa_sigaction = signalHandler;
3416     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3417   }
3418   // Save flags, which are set by ours
3419   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3420   sigflags[sig] = sigAct.sa_flags;
3421 
3422   int ret = sigaction(sig, &sigAct, &oldAct);
3423   assert(ret == 0, "check");
3424 
3425   void* oldhand2  = oldAct.sa_sigaction
3426                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3427                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3428   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3429 }
3430 
3431 // install signal handlers for signals that HotSpot needs to
3432 // handle in order to support Java-level exception handling.
3433 
3434 void os::Linux::install_signal_handlers() {
3435   if (!signal_handlers_are_installed) {
3436     signal_handlers_are_installed = true;
3437 
3438     // signal-chaining
3439     typedef void (*signal_setting_t)();
3440     signal_setting_t begin_signal_setting = NULL;
3441     signal_setting_t end_signal_setting = NULL;
3442     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3443                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3444     if (begin_signal_setting != NULL) {
3445       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3446                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3447       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3448                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3449       libjsig_is_loaded = true;
3450       assert(UseSignalChaining, "should enable signal-chaining");
3451     }
3452     if (libjsig_is_loaded) {
3453       // Tell libjsig jvm is setting signal handlers
3454       (*begin_signal_setting)();
3455     }
3456 
3457     set_signal_handler(SIGSEGV, true);
3458     set_signal_handler(SIGPIPE, true);
3459     set_signal_handler(SIGBUS, true);
3460     set_signal_handler(SIGILL, true);
3461     set_signal_handler(SIGFPE, true);
3462     set_signal_handler(SIGXFSZ, true);
3463 
3464     if (libjsig_is_loaded) {
3465       // Tell libjsig jvm finishes setting signal handlers
3466       (*end_signal_setting)();
3467     }
3468 
3469     // We don't activate signal checker if libjsig is in place, we trust ourselves
3470     // and if UserSignalHandler is installed all bets are off
3471     if (CheckJNICalls) {
3472       if (libjsig_is_loaded) {
3473         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3474         check_signals = false;
3475       }
3476       if (AllowUserSignalHandlers) {
3477         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3478         check_signals = false;
3479       }
3480     }
3481   }
3482 }
3483 
3484 // This is the fastest way to get thread cpu time on Linux.
3485 // Returns cpu time (user+sys) for any thread, not only for current.
3486 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3487 // It might work on 2.6.10+ with a special kernel/glibc patch.
3488 // For reference, please, see IEEE Std 1003.1-2004:
3489 //   http://www.unix.org/single_unix_specification
3490 
3491 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3492   struct timespec tp;
3493   int rc = os::Linux::clock_gettime(clockid, &tp);
3494   assert(rc == 0, "clock_gettime is expected to return 0 code");
3495 
3496   return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3497 }
3498 
3499 /////
3500 // glibc on Linux platform uses non-documented flag
3501 // to indicate, that some special sort of signal
3502 // trampoline is used.
3503 // We will never set this flag, and we should
3504 // ignore this flag in our diagnostic
3505 #ifdef SIGNIFICANT_SIGNAL_MASK
3506 #undef SIGNIFICANT_SIGNAL_MASK
3507 #endif
3508 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3509 
3510 static const char* get_signal_handler_name(address handler,
3511                                            char* buf, int buflen) {
3512   int offset;
3513   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3514   if (found) {
3515     // skip directory names
3516     const char *p1, *p2;
3517     p1 = buf;
3518     size_t len = strlen(os::file_separator());
3519     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3520     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3521   } else {
3522     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3523   }
3524   return buf;
3525 }
3526 
3527 static void print_signal_handler(outputStream* st, int sig,
3528                                  char* buf, size_t buflen) {
3529   struct sigaction sa;
3530 
3531   sigaction(sig, NULL, &sa);
3532 
3533   // See comment for SIGNIFICANT_SIGNAL_MASK define
3534   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3535 
3536   st->print("%s: ", os::exception_name(sig, buf, buflen));
3537 
3538   address handler = (sa.sa_flags & SA_SIGINFO)
3539     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3540     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3541 
3542   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3543     st->print("SIG_DFL");
3544   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3545     st->print("SIG_IGN");
3546   } else {
3547     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3548   }
3549 
3550   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3551 
3552   address rh = VMError::get_resetted_sighandler(sig);
3553   // May be, handler was resetted by VMError?
3554   if(rh != NULL) {
3555     handler = rh;
3556     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3557   }
3558 
3559   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
3560 
3561   // Check: is it our handler?
3562   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3563      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3564     // It is our signal handler
3565     // check for flags, reset system-used one!
3566     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3567       st->print(
3568                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3569                 os::Linux::get_our_sigflags(sig));
3570     }
3571   }
3572   st->cr();
3573 }
3574 
3575 
3576 #define DO_SIGNAL_CHECK(sig) \
3577   if (!sigismember(&check_signal_done, sig)) \
3578     os::Linux::check_signal_handler(sig)
3579 
3580 // This method is a periodic task to check for misbehaving JNI applications
3581 // under CheckJNI, we can add any periodic checks here
3582 
3583 void os::run_periodic_checks() {
3584 
3585   if (check_signals == false) return;
3586 
3587   // SEGV and BUS if overridden could potentially prevent
3588   // generation of hs*.log in the event of a crash, debugging
3589   // such a case can be very challenging, so we absolutely
3590   // check the following for a good measure:
3591   DO_SIGNAL_CHECK(SIGSEGV);
3592   DO_SIGNAL_CHECK(SIGILL);
3593   DO_SIGNAL_CHECK(SIGFPE);
3594   DO_SIGNAL_CHECK(SIGBUS);
3595   DO_SIGNAL_CHECK(SIGPIPE);
3596   DO_SIGNAL_CHECK(SIGXFSZ);
3597 
3598 
3599   // ReduceSignalUsage allows the user to override these handlers
3600   // see comments at the very top and jvm_solaris.h
3601   if (!ReduceSignalUsage) {
3602     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3603     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3604     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3605     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3606   }
3607 
3608   DO_SIGNAL_CHECK(SR_signum);
3609   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3610 }
3611 
3612 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3613 
3614 static os_sigaction_t os_sigaction = NULL;
3615 
3616 void os::Linux::check_signal_handler(int sig) {
3617   char buf[O_BUFLEN];
3618   address jvmHandler = NULL;
3619 
3620 
3621   struct sigaction act;
3622   if (os_sigaction == NULL) {
3623     // only trust the default sigaction, in case it has been interposed
3624     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3625     if (os_sigaction == NULL) return;
3626   }
3627 
3628   os_sigaction(sig, (struct sigaction*)NULL, &act);
3629 
3630 
3631   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3632 
3633   address thisHandler = (act.sa_flags & SA_SIGINFO)
3634     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3635     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3636 
3637 
3638   switch(sig) {
3639   case SIGSEGV:
3640   case SIGBUS:
3641   case SIGFPE:
3642   case SIGPIPE:
3643   case SIGILL:
3644   case SIGXFSZ:
3645     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3646     break;
3647 
3648   case SHUTDOWN1_SIGNAL:
3649   case SHUTDOWN2_SIGNAL:
3650   case SHUTDOWN3_SIGNAL:
3651   case BREAK_SIGNAL:
3652     jvmHandler = (address)user_handler();
3653     break;
3654 
3655   case INTERRUPT_SIGNAL:
3656     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3657     break;
3658 
3659   default:
3660     if (sig == SR_signum) {
3661       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3662     } else {
3663       return;
3664     }
3665     break;
3666   }
3667 
3668   if (thisHandler != jvmHandler) {
3669     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3670     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3671     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3672     // No need to check this sig any longer
3673     sigaddset(&check_signal_done, sig);
3674   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3675     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3676     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3677     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
3678     // No need to check this sig any longer
3679     sigaddset(&check_signal_done, sig);
3680   }
3681 
3682   // Dump all the signal
3683   if (sigismember(&check_signal_done, sig)) {
3684     print_signal_handlers(tty, buf, O_BUFLEN);
3685   }
3686 }
3687 
3688 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3689 
3690 extern bool signal_name(int signo, char* buf, size_t len);
3691 
3692 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3693   if (0 < exception_code && exception_code <= SIGRTMAX) {
3694     // signal
3695     if (!signal_name(exception_code, buf, size)) {
3696       jio_snprintf(buf, size, "SIG%d", exception_code);
3697     }
3698     return buf;
3699   } else {
3700     return NULL;
3701   }
3702 }
3703 
3704 // this is called _before_ the most of global arguments have been parsed
3705 void os::init(void) {
3706   char dummy;   /* used to get a guess on initial stack address */
3707 //  first_hrtime = gethrtime();
3708 
3709   // With LinuxThreads the JavaMain thread pid (primordial thread)
3710   // is different than the pid of the java launcher thread.
3711   // So, on Linux, the launcher thread pid is passed to the VM
3712   // via the sun.java.launcher.pid property.
3713   // Use this property instead of getpid() if it was correctly passed.
3714   // See bug 6351349.
3715   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3716 
3717   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3718 
3719   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3720 
3721   init_random(1234567);
3722 
3723   ThreadCritical::initialize();
3724 
3725   Linux::set_page_size(sysconf(_SC_PAGESIZE));
3726   if (Linux::page_size() == -1) {
3727     fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno));
3728   }
3729   init_page_sizes((size_t) Linux::page_size());
3730 
3731   Linux::initialize_system_info();
3732 
3733   // main_thread points to the aboriginal thread
3734   Linux::_main_thread = pthread_self();
3735 
3736   Linux::clock_init();
3737   initial_time_count = os::elapsed_counter();
3738   pthread_mutex_init(&dl_mutex, NULL);
3739 }
3740 
3741 // To install functions for atexit system call
3742 extern "C" {
3743   static void perfMemory_exit_helper() {
3744     perfMemory_exit();
3745   }
3746 }
3747 
3748 // this is called _after_ the global arguments have been parsed
3749 jint os::init_2(void)
3750 {
3751   Linux::fast_thread_clock_init();
3752 
3753   // Allocate a single page and mark it as readable for safepoint polling
3754   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3755   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3756 
3757   os::set_polling_page( polling_page );
3758 
3759 #ifndef PRODUCT
3760   if(Verbose && PrintMiscellaneous)
3761     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3762 #endif
3763 
3764   if (!UseMembar) {
3765     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3766     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3767     os::set_memory_serialize_page( mem_serialize_page );
3768 
3769 #ifndef PRODUCT
3770     if(Verbose && PrintMiscellaneous)
3771       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3772 #endif
3773   }
3774 
3775   FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3776 
3777   // initialize suspend/resume support - must do this before signal_sets_init()
3778   if (SR_initialize() != 0) {
3779     perror("SR_initialize failed");
3780     return JNI_ERR;
3781   }
3782 
3783   Linux::signal_sets_init();
3784   Linux::install_signal_handlers();
3785 
3786   size_t threadStackSizeInBytes = ThreadStackSize * K;
3787   if (threadStackSizeInBytes != 0 &&
3788       threadStackSizeInBytes < Linux::min_stack_allowed) {
3789         tty->print_cr("\nThe stack size specified is too small, "
3790                       "Specify at least %dk",
3791                       Linux::min_stack_allowed / K);
3792         return JNI_ERR;
3793   }
3794 
3795   // Make the stack size a multiple of the page size so that
3796   // the yellow/red zones can be guarded.
3797   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3798         vm_page_size()));
3799 
3800   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
3801 
3802   Linux::libpthread_init();
3803   if (PrintMiscellaneous && (Verbose || WizardMode)) {
3804      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
3805           Linux::glibc_version(), Linux::libpthread_version(),
3806           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
3807   }
3808 
3809   if (UseNUMA) {
3810     if (!Linux::libnuma_init()) {
3811       UseNUMA = false;
3812     } else {
3813       if ((Linux::numa_max_node() < 1)) {
3814         // There's only one node(they start from 0), disable NUMA.
3815         UseNUMA = false;
3816       }
3817     }
3818     if (!UseNUMA && ForceNUMA) {
3819       UseNUMA = true;
3820     }
3821   }
3822 
3823   if (MaxFDLimit) {
3824     // set the number of file descriptors to max. print out error
3825     // if getrlimit/setrlimit fails but continue regardless.
3826     struct rlimit nbr_files;
3827     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3828     if (status != 0) {
3829       if (PrintMiscellaneous && (Verbose || WizardMode))
3830         perror("os::init_2 getrlimit failed");
3831     } else {
3832       nbr_files.rlim_cur = nbr_files.rlim_max;
3833       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3834       if (status != 0) {
3835         if (PrintMiscellaneous && (Verbose || WizardMode))
3836           perror("os::init_2 setrlimit failed");
3837       }
3838     }
3839   }
3840 
3841   // Initialize lock used to serialize thread creation (see os::create_thread)
3842   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
3843 
3844   // Initialize HPI.
3845   jint hpi_result = hpi::initialize();
3846   if (hpi_result != JNI_OK) {
3847     tty->print_cr("There was an error trying to initialize the HPI library.");
3848     return hpi_result;
3849   }
3850 
3851   // at-exit methods are called in the reverse order of their registration.
3852   // atexit functions are called on return from main or as a result of a
3853   // call to exit(3C). There can be only 32 of these functions registered
3854   // and atexit() does not set errno.
3855 
3856   if (PerfAllowAtExitRegistration) {
3857     // only register atexit functions if PerfAllowAtExitRegistration is set.
3858     // atexit functions can be delayed until process exit time, which
3859     // can be problematic for embedded VM situations. Embedded VMs should
3860     // call DestroyJavaVM() to assure that VM resources are released.
3861 
3862     // note: perfMemory_exit_helper atexit function may be removed in
3863     // the future if the appropriate cleanup code can be added to the
3864     // VM_Exit VMOperation's doit method.
3865     if (atexit(perfMemory_exit_helper) != 0) {
3866       warning("os::init2 atexit(perfMemory_exit_helper) failed");
3867     }
3868   }
3869 
3870   // initialize thread priority policy
3871   prio_init();
3872 
3873   return JNI_OK;
3874 }
3875 
3876 // Mark the polling page as unreadable
3877 void os::make_polling_page_unreadable(void) {
3878   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
3879     fatal("Could not disable polling page");
3880 };
3881 
3882 // Mark the polling page as readable
3883 void os::make_polling_page_readable(void) {
3884   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
3885     fatal("Could not enable polling page");
3886   }
3887 };
3888 
3889 int os::active_processor_count() {
3890   // Linux doesn't yet have a (official) notion of processor sets,
3891   // so just return the number of online processors.
3892   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
3893   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
3894   return online_cpus;
3895 }
3896 
3897 bool os::distribute_processes(uint length, uint* distribution) {
3898   // Not yet implemented.
3899   return false;
3900 }
3901 
3902 bool os::bind_to_processor(uint processor_id) {
3903   // Not yet implemented.
3904   return false;
3905 }
3906 
3907 ///
3908 
3909 // Suspends the target using the signal mechanism and then grabs the PC before
3910 // resuming the target. Used by the flat-profiler only
3911 ExtendedPC os::get_thread_pc(Thread* thread) {
3912   // Make sure that it is called by the watcher for the VMThread
3913   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
3914   assert(thread->is_VM_thread(), "Can only be called for VMThread");
3915 
3916   ExtendedPC epc;
3917 
3918   OSThread* osthread = thread->osthread();
3919   if (do_suspend(osthread)) {
3920     if (osthread->ucontext() != NULL) {
3921       epc = os::Linux::ucontext_get_pc(osthread->ucontext());
3922     } else {
3923       // NULL context is unexpected, double-check this is the VMThread
3924       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3925     }
3926     do_resume(osthread);
3927   }
3928   // failure means pthread_kill failed for some reason - arguably this is
3929   // a fatal problem, but such problems are ignored elsewhere
3930 
3931   return epc;
3932 }
3933 
3934 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
3935 {
3936    if (is_NPTL()) {
3937       return pthread_cond_timedwait(_cond, _mutex, _abstime);
3938    } else {
3939 #ifndef IA64
3940       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
3941       // word back to default 64bit precision if condvar is signaled. Java
3942       // wants 53bit precision.  Save and restore current value.
3943       int fpu = get_fpu_control_word();
3944 #endif // IA64
3945       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
3946 #ifndef IA64
3947       set_fpu_control_word(fpu);
3948 #endif // IA64
3949       return status;
3950    }
3951 }
3952 
3953 ////////////////////////////////////////////////////////////////////////////////
3954 // debug support
3955 
3956 #ifndef PRODUCT
3957 static address same_page(address x, address y) {
3958   int page_bits = -os::vm_page_size();
3959   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
3960     return x;
3961   else if (x > y)
3962     return (address)(intptr_t(y) | ~page_bits) + 1;
3963   else
3964     return (address)(intptr_t(y) & page_bits);
3965 }
3966 
3967 bool os::find(address addr) {
3968   Dl_info dlinfo;
3969   memset(&dlinfo, 0, sizeof(dlinfo));
3970   if (dladdr(addr, &dlinfo)) {
3971     tty->print(PTR_FORMAT ": ", addr);
3972     if (dlinfo.dli_sname != NULL) {
3973       tty->print("%s+%#x", dlinfo.dli_sname,
3974                  addr - (intptr_t)dlinfo.dli_saddr);
3975     } else if (dlinfo.dli_fname) {
3976       tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
3977     } else {
3978       tty->print("<absolute address>");
3979     }
3980     if (dlinfo.dli_fname) {
3981       tty->print(" in %s", dlinfo.dli_fname);
3982     }
3983     if (dlinfo.dli_fbase) {
3984       tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
3985     }
3986     tty->cr();
3987 
3988     if (Verbose) {
3989       // decode some bytes around the PC
3990       address begin = same_page(addr-40, addr);
3991       address end   = same_page(addr+40, addr);
3992       address       lowest = (address) dlinfo.dli_sname;
3993       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
3994       if (begin < lowest)  begin = lowest;
3995       Dl_info dlinfo2;
3996       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
3997           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
3998         end = (address) dlinfo2.dli_saddr;
3999       Disassembler::decode(begin, end);
4000     }
4001     return true;
4002   }
4003   return false;
4004 }
4005 
4006 #endif
4007 
4008 ////////////////////////////////////////////////////////////////////////////////
4009 // misc
4010 
4011 // This does not do anything on Linux. This is basically a hook for being
4012 // able to use structured exception handling (thread-local exception filters)
4013 // on, e.g., Win32.
4014 void
4015 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4016                          JavaCallArguments* args, Thread* thread) {
4017   f(value, method, args, thread);
4018 }
4019 
4020 void os::print_statistics() {
4021 }
4022 
4023 int os::message_box(const char* title, const char* message) {
4024   int i;
4025   fdStream err(defaultStream::error_fd());
4026   for (i = 0; i < 78; i++) err.print_raw("=");
4027   err.cr();
4028   err.print_raw_cr(title);
4029   for (i = 0; i < 78; i++) err.print_raw("-");
4030   err.cr();
4031   err.print_raw_cr(message);
4032   for (i = 0; i < 78; i++) err.print_raw("=");
4033   err.cr();
4034 
4035   char buf[16];
4036   // Prevent process from exiting upon "read error" without consuming all CPU
4037   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4038 
4039   return buf[0] == 'y' || buf[0] == 'Y';
4040 }
4041 
4042 int os::stat(const char *path, struct stat *sbuf) {
4043   char pathbuf[MAX_PATH];
4044   if (strlen(path) > MAX_PATH - 1) {
4045     errno = ENAMETOOLONG;
4046     return -1;
4047   }
4048   hpi::native_path(strcpy(pathbuf, path));
4049   return ::stat(pathbuf, sbuf);
4050 }
4051 
4052 bool os::check_heap(bool force) {
4053   return true;
4054 }
4055 
4056 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4057   return ::vsnprintf(buf, count, format, args);
4058 }
4059 
4060 // Is a (classpath) directory empty?
4061 bool os::dir_is_empty(const char* path) {
4062   DIR *dir = NULL;
4063   struct dirent *ptr;
4064 
4065   dir = opendir(path);
4066   if (dir == NULL) return true;
4067 
4068   /* Scan the directory */
4069   bool result = true;
4070   char buf[sizeof(struct dirent) + MAX_PATH];
4071   while (result && (ptr = ::readdir(dir)) != NULL) {
4072     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4073       result = false;
4074     }
4075   }
4076   closedir(dir);
4077   return result;
4078 }
4079 
4080 // create binary file, rewriting existing file if required
4081 int os::create_binary_file(const char* path, bool rewrite_existing) {
4082   int oflags = O_WRONLY | O_CREAT;
4083   if (!rewrite_existing) {
4084     oflags |= O_EXCL;
4085   }
4086   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4087 }
4088 
4089 // return current position of file pointer
4090 jlong os::current_file_offset(int fd) {
4091   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4092 }
4093 
4094 // move file pointer to the specified offset
4095 jlong os::seek_to_file_offset(int fd, jlong offset) {
4096   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4097 }
4098 
4099 // Map a block of memory.
4100 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4101                      char *addr, size_t bytes, bool read_only,
4102                      bool allow_exec) {
4103   int prot;
4104   int flags;
4105 
4106   if (read_only) {
4107     prot = PROT_READ;
4108     flags = MAP_SHARED;
4109   } else {
4110     prot = PROT_READ | PROT_WRITE;
4111     flags = MAP_PRIVATE;
4112   }
4113 
4114   if (allow_exec) {
4115     prot |= PROT_EXEC;
4116   }
4117 
4118   if (addr != NULL) {
4119     flags |= MAP_FIXED;
4120   }
4121 
4122   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4123                                      fd, file_offset);
4124   if (mapped_address == MAP_FAILED) {
4125     return NULL;
4126   }
4127   return mapped_address;
4128 }
4129 
4130 
4131 // Remap a block of memory.
4132 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4133                        char *addr, size_t bytes, bool read_only,
4134                        bool allow_exec) {
4135   // same as map_memory() on this OS
4136   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4137                         allow_exec);
4138 }
4139 
4140 
4141 // Unmap a block of memory.
4142 bool os::unmap_memory(char* addr, size_t bytes) {
4143   return munmap(addr, bytes) == 0;
4144 }
4145 
4146 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4147 
4148 static clockid_t thread_cpu_clockid(Thread* thread) {
4149   pthread_t tid = thread->osthread()->pthread_id();
4150   clockid_t clockid;
4151 
4152   // Get thread clockid
4153   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4154   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4155   return clockid;
4156 }
4157 
4158 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4159 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4160 // of a thread.
4161 //
4162 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4163 // the fast estimate available on the platform.
4164 
4165 jlong os::current_thread_cpu_time() {
4166   if (os::Linux::supports_fast_thread_cpu_time()) {
4167     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4168   } else {
4169     // return user + sys since the cost is the same
4170     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4171   }
4172 }
4173 
4174 jlong os::thread_cpu_time(Thread* thread) {
4175   // consistent with what current_thread_cpu_time() returns
4176   if (os::Linux::supports_fast_thread_cpu_time()) {
4177     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4178   } else {
4179     return slow_thread_cpu_time(thread, true /* user + sys */);
4180   }
4181 }
4182 
4183 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4184   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4185     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4186   } else {
4187     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4188   }
4189 }
4190 
4191 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4192   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4193     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4194   } else {
4195     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4196   }
4197 }
4198 
4199 //
4200 //  -1 on error.
4201 //
4202 
4203 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4204   static bool proc_pid_cpu_avail = true;
4205   static bool proc_task_unchecked = true;
4206   static const char *proc_stat_path = "/proc/%d/stat";
4207   pid_t  tid = thread->osthread()->thread_id();
4208   int i;
4209   char *s;
4210   char stat[2048];
4211   int statlen;
4212   char proc_name[64];
4213   int count;
4214   long sys_time, user_time;
4215   char string[64];
4216   int idummy;
4217   long ldummy;
4218   FILE *fp;
4219 
4220   // We first try accessing /proc/<pid>/cpu since this is faster to
4221   // process.  If this file is not present (linux kernels 2.5 and above)
4222   // then we open /proc/<pid>/stat.
4223   if ( proc_pid_cpu_avail ) {
4224     sprintf(proc_name, "/proc/%d/cpu", tid);
4225     fp =  fopen(proc_name, "r");
4226     if ( fp != NULL ) {
4227       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4228       fclose(fp);
4229       if ( count != 3 ) return -1;
4230 
4231       if (user_sys_cpu_time) {
4232         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4233       } else {
4234         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4235       }
4236     }
4237     else proc_pid_cpu_avail = false;
4238   }
4239 
4240   // The /proc/<tid>/stat aggregates per-process usage on
4241   // new Linux kernels 2.6+ where NPTL is supported.
4242   // The /proc/self/task/<tid>/stat still has the per-thread usage.
4243   // See bug 6328462.
4244   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4245   // and possibly in some other cases, so we check its availability.
4246   if (proc_task_unchecked && os::Linux::is_NPTL()) {
4247     // This is executed only once
4248     proc_task_unchecked = false;
4249     fp = fopen("/proc/self/task", "r");
4250     if (fp != NULL) {
4251       proc_stat_path = "/proc/self/task/%d/stat";
4252       fclose(fp);
4253     }
4254   }
4255 
4256   sprintf(proc_name, proc_stat_path, tid);
4257   fp = fopen(proc_name, "r");
4258   if ( fp == NULL ) return -1;
4259   statlen = fread(stat, 1, 2047, fp);
4260   stat[statlen] = '\0';
4261   fclose(fp);
4262 
4263   // Skip pid and the command string. Note that we could be dealing with
4264   // weird command names, e.g. user could decide to rename java launcher
4265   // to "java 1.4.2 :)", then the stat file would look like
4266   //                1234 (java 1.4.2 :)) R ... ...
4267   // We don't really need to know the command string, just find the last
4268   // occurrence of ")" and then start parsing from there. See bug 4726580.
4269   s = strrchr(stat, ')');
4270   i = 0;
4271   if (s == NULL ) return -1;
4272 
4273   // Skip blank chars
4274   do s++; while (isspace(*s));
4275 
4276   count = sscanf(s,"%*c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4277                  &idummy, &idummy, &idummy, &idummy, &idummy,
4278                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4279                  &user_time, &sys_time);
4280   if ( count != 12 ) return -1;
4281   if (user_sys_cpu_time) {
4282     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4283   } else {
4284     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4285   }
4286 }
4287 
4288 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4289   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4290   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4291   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4292   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4293 }
4294 
4295 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4296   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4297   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4298   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4299   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4300 }
4301 
4302 bool os::is_thread_cpu_time_supported() {
4303   return true;
4304 }
4305 
4306 // System loadavg support.  Returns -1 if load average cannot be obtained.
4307 // Linux doesn't yet have a (official) notion of processor sets,
4308 // so just return the system wide load average.
4309 int os::loadavg(double loadavg[], int nelem) {
4310   return ::getloadavg(loadavg, nelem);
4311 }
4312 
4313 void os::pause() {
4314   char filename[MAX_PATH];
4315   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4316     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4317   } else {
4318     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4319   }
4320 
4321   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4322   if (fd != -1) {
4323     struct stat buf;
4324     close(fd);
4325     while (::stat(filename, &buf) == 0) {
4326       (void)::poll(NULL, 0, 100);
4327     }
4328   } else {
4329     jio_fprintf(stderr,
4330       "Could not open pause file '%s', continuing immediately.\n", filename);
4331   }
4332 }
4333 
4334 extern "C" {
4335 
4336 /**
4337  * NOTE: the following code is to keep the green threads code
4338  * in the libjava.so happy. Once the green threads is removed,
4339  * these code will no longer be needed.
4340  */
4341 int
4342 jdk_waitpid(pid_t pid, int* status, int options) {
4343     return waitpid(pid, status, options);
4344 }
4345 
4346 int
4347 fork1() {
4348     return fork();
4349 }
4350 
4351 int
4352 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4353     return sem_init(sem, pshared, value);
4354 }
4355 
4356 int
4357 jdk_sem_post(sem_t *sem) {
4358     return sem_post(sem);
4359 }
4360 
4361 int
4362 jdk_sem_wait(sem_t *sem) {
4363     return sem_wait(sem);
4364 }
4365 
4366 int
4367 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4368     return pthread_sigmask(how , newmask, oldmask);
4369 }
4370 
4371 }
4372 
4373 // Refer to the comments in os_solaris.cpp park-unpark.
4374 //
4375 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4376 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4377 // For specifics regarding the bug see GLIBC BUGID 261237 :
4378 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4379 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4380 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4381 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4382 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4383 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4384 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4385 // of libpthread avoids the problem, but isn't practical.
4386 //
4387 // Possible remedies:
4388 //
4389 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4390 //      This is palliative and probabilistic, however.  If the thread is preempted
4391 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4392 //      than the minimum period may have passed, and the abstime may be stale (in the
4393 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
4394 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4395 //
4396 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4397 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
4398 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4399 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
4400 //      thread.
4401 //
4402 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4403 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
4404 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4405 //      This also works well.  In fact it avoids kernel-level scalability impediments
4406 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4407 //      timers in a graceful fashion.
4408 //
4409 // 4.   When the abstime value is in the past it appears that control returns
4410 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4411 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
4412 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
4413 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
4414 //      It may be possible to avoid reinitialization by checking the return
4415 //      value from pthread_cond_timedwait().  In addition to reinitializing the
4416 //      condvar we must establish the invariant that cond_signal() is only called
4417 //      within critical sections protected by the adjunct mutex.  This prevents
4418 //      cond_signal() from "seeing" a condvar that's in the midst of being
4419 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
4420 //      desirable signal-after-unlock optimization that avoids futile context switching.
4421 //
4422 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
4423 //      structure when a condvar is used or initialized.  cond_destroy()  would
4424 //      release the helper structure.  Our reinitialize-after-timedwait fix
4425 //      put excessive stress on malloc/free and locks protecting the c-heap.
4426 //
4427 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
4428 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4429 // and only enabling the work-around for vulnerable environments.
4430 
4431 // utility to compute the abstime argument to timedwait:
4432 // millis is the relative timeout time
4433 // abstime will be the absolute timeout time
4434 // TODO: replace compute_abstime() with unpackTime()
4435 
4436 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4437   if (millis < 0)  millis = 0;
4438   struct timeval now;
4439   int status = gettimeofday(&now, NULL);
4440   assert(status == 0, "gettimeofday");
4441   jlong seconds = millis / 1000;
4442   millis %= 1000;
4443   if (seconds > 50000000) { // see man cond_timedwait(3T)
4444     seconds = 50000000;
4445   }
4446   abstime->tv_sec = now.tv_sec  + seconds;
4447   long       usec = now.tv_usec + millis * 1000;
4448   if (usec >= 1000000) {
4449     abstime->tv_sec += 1;
4450     usec -= 1000000;
4451   }
4452   abstime->tv_nsec = usec * 1000;
4453   return abstime;
4454 }
4455 
4456 
4457 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4458 // Conceptually TryPark() should be equivalent to park(0).
4459 
4460 int os::PlatformEvent::TryPark() {
4461   for (;;) {
4462     const int v = _Event ;
4463     guarantee ((v == 0) || (v == 1), "invariant") ;
4464     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
4465   }
4466 }
4467 
4468 void os::PlatformEvent::park() {       // AKA "down()"
4469   // Invariant: Only the thread associated with the Event/PlatformEvent
4470   // may call park().
4471   // TODO: assert that _Assoc != NULL or _Assoc == Self
4472   int v ;
4473   for (;;) {
4474       v = _Event ;
4475       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4476   }
4477   guarantee (v >= 0, "invariant") ;
4478   if (v == 0) {
4479      // Do this the hard way by blocking ...
4480      int status = pthread_mutex_lock(_mutex);
4481      assert_status(status == 0, status, "mutex_lock");
4482      guarantee (_nParked == 0, "invariant") ;
4483      ++ _nParked ;
4484      while (_Event < 0) {
4485         status = pthread_cond_wait(_cond, _mutex);
4486         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4487         // Treat this the same as if the wait was interrupted
4488         if (status == ETIME) { status = EINTR; }
4489         assert_status(status == 0 || status == EINTR, status, "cond_wait");
4490      }
4491      -- _nParked ;
4492 
4493     // In theory we could move the ST of 0 into _Event past the unlock(),
4494     // but then we'd need a MEMBAR after the ST.
4495     _Event = 0 ;
4496      status = pthread_mutex_unlock(_mutex);
4497      assert_status(status == 0, status, "mutex_unlock");
4498   }
4499   guarantee (_Event >= 0, "invariant") ;
4500 }
4501 
4502 int os::PlatformEvent::park(jlong millis) {
4503   guarantee (_nParked == 0, "invariant") ;
4504 
4505   int v ;
4506   for (;;) {
4507       v = _Event ;
4508       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4509   }
4510   guarantee (v >= 0, "invariant") ;
4511   if (v != 0) return OS_OK ;
4512 
4513   // We do this the hard way, by blocking the thread.
4514   // Consider enforcing a minimum timeout value.
4515   struct timespec abst;
4516   compute_abstime(&abst, millis);
4517 
4518   int ret = OS_TIMEOUT;
4519   int status = pthread_mutex_lock(_mutex);
4520   assert_status(status == 0, status, "mutex_lock");
4521   guarantee (_nParked == 0, "invariant") ;
4522   ++_nParked ;
4523 
4524   // Object.wait(timo) will return because of
4525   // (a) notification
4526   // (b) timeout
4527   // (c) thread.interrupt
4528   //
4529   // Thread.interrupt and object.notify{All} both call Event::set.
4530   // That is, we treat thread.interrupt as a special case of notification.
4531   // The underlying Solaris implementation, cond_timedwait, admits
4532   // spurious/premature wakeups, but the JLS/JVM spec prevents the
4533   // JVM from making those visible to Java code.  As such, we must
4534   // filter out spurious wakeups.  We assume all ETIME returns are valid.
4535   //
4536   // TODO: properly differentiate simultaneous notify+interrupt.
4537   // In that case, we should propagate the notify to another waiter.
4538 
4539   while (_Event < 0) {
4540     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4541     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4542       pthread_cond_destroy (_cond);
4543       pthread_cond_init (_cond, NULL) ;
4544     }
4545     assert_status(status == 0 || status == EINTR ||
4546                   status == ETIME || status == ETIMEDOUT,
4547                   status, "cond_timedwait");
4548     if (!FilterSpuriousWakeups) break ;                 // previous semantics
4549     if (status == ETIME || status == ETIMEDOUT) break ;
4550     // We consume and ignore EINTR and spurious wakeups.
4551   }
4552   --_nParked ;
4553   if (_Event >= 0) {
4554      ret = OS_OK;
4555   }
4556   _Event = 0 ;
4557   status = pthread_mutex_unlock(_mutex);
4558   assert_status(status == 0, status, "mutex_unlock");
4559   assert (_nParked == 0, "invariant") ;
4560   return ret;
4561 }
4562 
4563 void os::PlatformEvent::unpark() {
4564   int v, AnyWaiters ;
4565   for (;;) {
4566       v = _Event ;
4567       if (v > 0) {
4568          // The LD of _Event could have reordered or be satisfied
4569          // by a read-aside from this processor's write buffer.
4570          // To avoid problems execute a barrier and then
4571          // ratify the value.
4572          OrderAccess::fence() ;
4573          if (_Event == v) return ;
4574          continue ;
4575       }
4576       if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4577   }
4578   if (v < 0) {
4579      // Wait for the thread associated with the event to vacate
4580      int status = pthread_mutex_lock(_mutex);
4581      assert_status(status == 0, status, "mutex_lock");
4582      AnyWaiters = _nParked ;
4583      assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4584      if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4585         AnyWaiters = 0 ;
4586         pthread_cond_signal (_cond);
4587      }
4588      status = pthread_mutex_unlock(_mutex);
4589      assert_status(status == 0, status, "mutex_unlock");
4590      if (AnyWaiters != 0) {
4591         status = pthread_cond_signal(_cond);
4592         assert_status(status == 0, status, "cond_signal");
4593      }
4594   }
4595 
4596   // Note that we signal() _after dropping the lock for "immortal" Events.
4597   // This is safe and avoids a common class of  futile wakeups.  In rare
4598   // circumstances this can cause a thread to return prematurely from
4599   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4600   // simply re-test the condition and re-park itself.
4601 }
4602 
4603 
4604 // JSR166
4605 // -------------------------------------------------------
4606 
4607 /*
4608  * The solaris and linux implementations of park/unpark are fairly
4609  * conservative for now, but can be improved. They currently use a
4610  * mutex/condvar pair, plus a a count.
4611  * Park decrements count if > 0, else does a condvar wait.  Unpark
4612  * sets count to 1 and signals condvar.  Only one thread ever waits
4613  * on the condvar. Contention seen when trying to park implies that someone
4614  * is unparking you, so don't wait. And spurious returns are fine, so there
4615  * is no need to track notifications.
4616  */
4617 
4618 
4619 #define NANOSECS_PER_SEC 1000000000
4620 #define NANOSECS_PER_MILLISEC 1000000
4621 #define MAX_SECS 100000000
4622 /*
4623  * This code is common to linux and solaris and will be moved to a
4624  * common place in dolphin.
4625  *
4626  * The passed in time value is either a relative time in nanoseconds
4627  * or an absolute time in milliseconds. Either way it has to be unpacked
4628  * into suitable seconds and nanoseconds components and stored in the
4629  * given timespec structure.
4630  * Given time is a 64-bit value and the time_t used in the timespec is only
4631  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4632  * overflow if times way in the future are given. Further on Solaris versions
4633  * prior to 10 there is a restriction (see cond_timedwait) that the specified
4634  * number of seconds, in abstime, is less than current_time  + 100,000,000.
4635  * As it will be 28 years before "now + 100000000" will overflow we can
4636  * ignore overflow and just impose a hard-limit on seconds using the value
4637  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4638  * years from "now".
4639  */
4640 
4641 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4642   assert (time > 0, "convertTime");
4643 
4644   struct timeval now;
4645   int status = gettimeofday(&now, NULL);
4646   assert(status == 0, "gettimeofday");
4647 
4648   time_t max_secs = now.tv_sec + MAX_SECS;
4649 
4650   if (isAbsolute) {
4651     jlong secs = time / 1000;
4652     if (secs > max_secs) {
4653       absTime->tv_sec = max_secs;
4654     }
4655     else {
4656       absTime->tv_sec = secs;
4657     }
4658     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4659   }
4660   else {
4661     jlong secs = time / NANOSECS_PER_SEC;
4662     if (secs >= MAX_SECS) {
4663       absTime->tv_sec = max_secs;
4664       absTime->tv_nsec = 0;
4665     }
4666     else {
4667       absTime->tv_sec = now.tv_sec + secs;
4668       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4669       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4670         absTime->tv_nsec -= NANOSECS_PER_SEC;
4671         ++absTime->tv_sec; // note: this must be <= max_secs
4672       }
4673     }
4674   }
4675   assert(absTime->tv_sec >= 0, "tv_sec < 0");
4676   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4677   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4678   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4679 }
4680 
4681 void Parker::park(bool isAbsolute, jlong time) {
4682   // Optional fast-path check:
4683   // Return immediately if a permit is available.
4684   if (_counter > 0) {
4685       _counter = 0 ;
4686       OrderAccess::fence();
4687       return ;
4688   }
4689 
4690   Thread* thread = Thread::current();
4691   assert(thread->is_Java_thread(), "Must be JavaThread");
4692   JavaThread *jt = (JavaThread *)thread;
4693 
4694   // Optional optimization -- avoid state transitions if there's an interrupt pending.
4695   // Check interrupt before trying to wait
4696   if (Thread::is_interrupted(thread, false)) {
4697     return;
4698   }
4699 
4700   // Next, demultiplex/decode time arguments
4701   timespec absTime;
4702   if (time < 0) { // don't wait at all
4703     return;
4704   }
4705   if (time > 0) {
4706     unpackTime(&absTime, isAbsolute, time);
4707   }
4708 
4709 
4710   // Enter safepoint region
4711   // Beware of deadlocks such as 6317397.
4712   // The per-thread Parker:: mutex is a classic leaf-lock.
4713   // In particular a thread must never block on the Threads_lock while
4714   // holding the Parker:: mutex.  If safepoints are pending both the
4715   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4716   ThreadBlockInVM tbivm(jt);
4717 
4718   // Don't wait if cannot get lock since interference arises from
4719   // unblocking.  Also. check interrupt before trying wait
4720   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4721     return;
4722   }
4723 
4724   int status ;
4725   if (_counter > 0)  { // no wait needed
4726     _counter = 0;
4727     status = pthread_mutex_unlock(_mutex);
4728     assert (status == 0, "invariant") ;
4729     OrderAccess::fence();
4730     return;
4731   }
4732 
4733 #ifdef ASSERT
4734   // Don't catch signals while blocked; let the running threads have the signals.
4735   // (This allows a debugger to break into the running thread.)
4736   sigset_t oldsigs;
4737   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4738   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4739 #endif
4740 
4741   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4742   jt->set_suspend_equivalent();
4743   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4744 
4745   if (time == 0) {
4746     status = pthread_cond_wait (_cond, _mutex) ;
4747   } else {
4748     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4749     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4750       pthread_cond_destroy (_cond) ;
4751       pthread_cond_init    (_cond, NULL);
4752     }
4753   }
4754   assert_status(status == 0 || status == EINTR ||
4755                 status == ETIME || status == ETIMEDOUT,
4756                 status, "cond_timedwait");
4757 
4758 #ifdef ASSERT
4759   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4760 #endif
4761 
4762   _counter = 0 ;
4763   status = pthread_mutex_unlock(_mutex) ;
4764   assert_status(status == 0, status, "invariant") ;
4765   // If externally suspended while waiting, re-suspend
4766   if (jt->handle_special_suspend_equivalent_condition()) {
4767     jt->java_suspend_self();
4768   }
4769 
4770   OrderAccess::fence();
4771 }
4772 
4773 void Parker::unpark() {
4774   int s, status ;
4775   status = pthread_mutex_lock(_mutex);
4776   assert (status == 0, "invariant") ;
4777   s = _counter;
4778   _counter = 1;
4779   if (s < 1) {
4780      if (WorkAroundNPTLTimedWaitHang) {
4781         status = pthread_cond_signal (_cond) ;
4782         assert (status == 0, "invariant") ;
4783         status = pthread_mutex_unlock(_mutex);
4784         assert (status == 0, "invariant") ;
4785      } else {
4786         status = pthread_mutex_unlock(_mutex);
4787         assert (status == 0, "invariant") ;
4788         status = pthread_cond_signal (_cond) ;
4789         assert (status == 0, "invariant") ;
4790      }
4791   } else {
4792     pthread_mutex_unlock(_mutex);
4793     assert (status == 0, "invariant") ;
4794   }
4795 }
4796 
4797 
4798 extern char** environ;
4799 
4800 #ifndef __NR_fork
4801 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
4802 #endif
4803 
4804 #ifndef __NR_execve
4805 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
4806 #endif
4807 
4808 // Run the specified command in a separate process. Return its exit value,
4809 // or -1 on failure (e.g. can't fork a new process).
4810 // Unlike system(), this function can be called from signal handler. It
4811 // doesn't block SIGINT et al.
4812 int os::fork_and_exec(char* cmd) {
4813   const char * argv[4] = {"sh", "-c", cmd, NULL};
4814 
4815   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
4816   // pthread_atfork handlers and reset pthread library. All we need is a
4817   // separate process to execve. Make a direct syscall to fork process.
4818   // On IA64 there's no fork syscall, we have to use fork() and hope for
4819   // the best...
4820   pid_t pid = NOT_IA64(syscall(__NR_fork);)
4821               IA64_ONLY(fork();)
4822 
4823   if (pid < 0) {
4824     // fork failed
4825     return -1;
4826 
4827   } else if (pid == 0) {
4828     // child process
4829 
4830     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
4831     // first to kill every thread on the thread list. Because this list is
4832     // not reset by fork() (see notes above), execve() will instead kill
4833     // every thread in the parent process. We know this is the only thread
4834     // in the new process, so make a system call directly.
4835     // IA64 should use normal execve() from glibc to match the glibc fork()
4836     // above.
4837     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
4838     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
4839 
4840     // execve failed
4841     _exit(-1);
4842 
4843   } else  {
4844     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4845     // care about the actual exit code, for now.
4846 
4847     int status;
4848 
4849     // Wait for the child process to exit.  This returns immediately if
4850     // the child has already exited. */
4851     while (waitpid(pid, &status, 0) < 0) {
4852         switch (errno) {
4853         case ECHILD: return 0;
4854         case EINTR: break;
4855         default: return -1;
4856         }
4857     }
4858 
4859     if (WIFEXITED(status)) {
4860        // The child exited normally; get its exit code.
4861        return WEXITSTATUS(status);
4862     } else if (WIFSIGNALED(status)) {
4863        // The child exited because of a signal
4864        // The best value to return is 0x80 + signal number,
4865        // because that is what all Unix shells do, and because
4866        // it allows callers to distinguish between process exit and
4867        // process death by signal.
4868        return 0x80 + WTERMSIG(status);
4869     } else {
4870        // Unknown exit code; pass it through
4871        return status;
4872     }
4873   }
4874 }
--- EOF ---