1 /*
   2  * Copyright (c) 1999, 2009, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 # define __STDC_FORMAT_MACROS
  26 
  27 // do not include  precompiled  header file
  28 # include "incls/_os_linux.cpp.incl"
  29 
  30 // put OS-includes here
  31 # include <sys/types.h>
  32 # include <sys/mman.h>
  33 # include <pthread.h>
  34 # include <signal.h>
  35 # include <errno.h>
  36 # include <dlfcn.h>
  37 # include <stdio.h>
  38 # include <unistd.h>
  39 # include <sys/resource.h>
  40 # include <pthread.h>
  41 # include <sys/stat.h>
  42 # include <sys/time.h>
  43 # include <sys/times.h>
  44 # include <sys/utsname.h>
  45 # include <sys/socket.h>
  46 # include <sys/wait.h>
  47 # include <pwd.h>
  48 # include <poll.h>
  49 # include <semaphore.h>
  50 # include <fcntl.h>
  51 # include <string.h>
  52 # include <syscall.h>
  53 # include <sys/sysinfo.h>
  54 # include <gnu/libc-version.h>
  55 # include <sys/ipc.h>
  56 # include <sys/shm.h>
  57 # include <link.h>
  58 # include <stdint.h>
  59 # include <inttypes.h>
  60 
  61 #define MAX_PATH    (2 * K)
  62 
  63 // for timer info max values which include all bits
  64 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
  65 #define SEC_IN_NANOSECS  1000000000LL
  66 
  67 ////////////////////////////////////////////////////////////////////////////////
  68 // global variables
  69 julong os::Linux::_physical_memory = 0;
  70 
  71 address   os::Linux::_initial_thread_stack_bottom = NULL;
  72 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
  73 
  74 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
  75 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
  76 Mutex* os::Linux::_createThread_lock = NULL;
  77 pthread_t os::Linux::_main_thread;
  78 int os::Linux::_page_size = -1;
  79 bool os::Linux::_is_floating_stack = false;
  80 bool os::Linux::_is_NPTL = false;
  81 bool os::Linux::_supports_fast_thread_cpu_time = false;
  82 const char * os::Linux::_glibc_version = NULL;
  83 const char * os::Linux::_libpthread_version = NULL;
  84 
  85 static jlong initial_time_count=0;
  86 
  87 static int clock_tics_per_sec = 100;
  88 
  89 // For diagnostics to print a message once. see run_periodic_checks
  90 static sigset_t check_signal_done;
  91 static bool check_signals = true;;
  92 
  93 static pid_t _initial_pid = 0;
  94 
  95 /* Signal number used to suspend/resume a thread */
  96 
  97 /* do not use any signal number less than SIGSEGV, see 4355769 */
  98 static int SR_signum = SIGUSR2;
  99 sigset_t SR_sigset;
 100 
 101 /* Used to protect dlsym() calls */
 102 static pthread_mutex_t dl_mutex;
 103 
 104 ////////////////////////////////////////////////////////////////////////////////
 105 // utility functions
 106 
 107 static int SR_initialize();
 108 static int SR_finalize();
 109 
 110 julong os::available_memory() {
 111   return Linux::available_memory();
 112 }
 113 
 114 julong os::Linux::available_memory() {
 115   // values in struct sysinfo are "unsigned long"
 116   struct sysinfo si;
 117   sysinfo(&si);
 118 
 119   return (julong)si.freeram * si.mem_unit;
 120 }
 121 
 122 julong os::physical_memory() {
 123   return Linux::physical_memory();
 124 }
 125 
 126 julong os::allocatable_physical_memory(julong size) {
 127 #ifdef _LP64
 128   return size;
 129 #else
 130   julong result = MIN2(size, (julong)3800*M);
 131    if (!is_allocatable(result)) {
 132      // See comments under solaris for alignment considerations
 133      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 134      result =  MIN2(size, reasonable_size);
 135    }
 136    return result;
 137 #endif // _LP64
 138 }
 139 
 140 ////////////////////////////////////////////////////////////////////////////////
 141 // environment support
 142 
 143 bool os::getenv(const char* name, char* buf, int len) {
 144   const char* val = ::getenv(name);
 145   if (val != NULL && strlen(val) < (size_t)len) {
 146     strcpy(buf, val);
 147     return true;
 148   }
 149   if (len > 0) buf[0] = 0;  // return a null string
 150   return false;
 151 }
 152 
 153 
 154 // Return true if user is running as root.
 155 
 156 bool os::have_special_privileges() {
 157   static bool init = false;
 158   static bool privileges = false;
 159   if (!init) {
 160     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 161     init = true;
 162   }
 163   return privileges;
 164 }
 165 
 166 
 167 #ifndef SYS_gettid
 168 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 169 #ifdef __ia64__
 170 #define SYS_gettid 1105
 171 #elif __i386__
 172 #define SYS_gettid 224
 173 #elif __amd64__
 174 #define SYS_gettid 186
 175 #elif __sparc__
 176 #define SYS_gettid 143
 177 #else
 178 #error define gettid for the arch
 179 #endif
 180 #endif
 181 
 182 // Cpu architecture string
 183 #if   defined(ZERO)
 184 static char cpu_arch[] = ZERO_LIBARCH;
 185 #elif defined(IA64)
 186 static char cpu_arch[] = "ia64";
 187 #elif defined(IA32)
 188 static char cpu_arch[] = "i386";
 189 #elif defined(AMD64)
 190 static char cpu_arch[] = "amd64";
 191 #elif defined(SPARC)
 192 #  ifdef _LP64
 193 static char cpu_arch[] = "sparcv9";
 194 #  else
 195 static char cpu_arch[] = "sparc";
 196 #  endif
 197 #else
 198 #error Add appropriate cpu_arch setting
 199 #endif
 200 
 201 
 202 // pid_t gettid()
 203 //
 204 // Returns the kernel thread id of the currently running thread. Kernel
 205 // thread id is used to access /proc.
 206 //
 207 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 208 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 209 //
 210 pid_t os::Linux::gettid() {
 211   int rslt = syscall(SYS_gettid);
 212   if (rslt == -1) {
 213      // old kernel, no NPTL support
 214      return getpid();
 215   } else {
 216      return (pid_t)rslt;
 217   }
 218 }
 219 
 220 // Most versions of linux have a bug where the number of processors are
 221 // determined by looking at the /proc file system.  In a chroot environment,
 222 // the system call returns 1.  This causes the VM to act as if it is
 223 // a single processor and elide locking (see is_MP() call).
 224 static bool unsafe_chroot_detected = false;
 225 static const char *unstable_chroot_error = "/proc file system not found.\n"
 226                      "Java may be unstable running multithreaded in a chroot "
 227                      "environment on Linux when /proc filesystem is not mounted.";
 228 
 229 void os::Linux::initialize_system_info() {
 230   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 231   if (processor_count() == 1) {
 232     pid_t pid = os::Linux::gettid();
 233     char fname[32];
 234     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 235     FILE *fp = fopen(fname, "r");
 236     if (fp == NULL) {
 237       unsafe_chroot_detected = true;
 238     } else {
 239       fclose(fp);
 240     }
 241   }
 242   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 243   assert(processor_count() > 0, "linux error");
 244 }
 245 
 246 void os::init_system_properties_values() {
 247 //  char arch[12];
 248 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 249 
 250   // The next steps are taken in the product version:
 251   //
 252   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 253   // This library should be located at:
 254   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 255   //
 256   // If "/jre/lib/" appears at the right place in the path, then we
 257   // assume libjvm[_g].so is installed in a JDK and we use this path.
 258   //
 259   // Otherwise exit with message: "Could not create the Java virtual machine."
 260   //
 261   // The following extra steps are taken in the debugging version:
 262   //
 263   // If "/jre/lib/" does NOT appear at the right place in the path
 264   // instead of exit check for $JAVA_HOME environment variable.
 265   //
 266   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 267   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 268   // it looks like libjvm[_g].so is installed there
 269   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 270   //
 271   // Otherwise exit.
 272   //
 273   // Important note: if the location of libjvm.so changes this
 274   // code needs to be changed accordingly.
 275 
 276   // The next few definitions allow the code to be verbatim:
 277 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 278 #define getenv(n) ::getenv(n)
 279 
 280 /*
 281  * See ld(1):
 282  *      The linker uses the following search paths to locate required
 283  *      shared libraries:
 284  *        1: ...
 285  *        ...
 286  *        7: The default directories, normally /lib and /usr/lib.
 287  */
 288 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 289 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 290 #else
 291 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 292 #endif
 293 
 294 #define EXTENSIONS_DIR  "/lib/ext"
 295 #define ENDORSED_DIR    "/lib/endorsed"
 296 #define REG_DIR         "/usr/java/packages"
 297 
 298   {
 299     /* sysclasspath, java_home, dll_dir */
 300     {
 301         char *home_path;
 302         char *dll_path;
 303         char *pslash;
 304         char buf[MAXPATHLEN];
 305         os::jvm_path(buf, sizeof(buf));
 306 
 307         // Found the full path to libjvm.so.
 308         // Now cut the path to <java_home>/jre if we can.
 309         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 310         pslash = strrchr(buf, '/');
 311         if (pslash != NULL)
 312             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 313         dll_path = malloc(strlen(buf) + 1);
 314         if (dll_path == NULL)
 315             return;
 316         strcpy(dll_path, buf);
 317         Arguments::set_dll_dir(dll_path);
 318 
 319         if (pslash != NULL) {
 320             pslash = strrchr(buf, '/');
 321             if (pslash != NULL) {
 322                 *pslash = '\0';       /* get rid of /<arch> */
 323                 pslash = strrchr(buf, '/');
 324                 if (pslash != NULL)
 325                     *pslash = '\0';   /* get rid of /lib */
 326             }
 327         }
 328 
 329         home_path = malloc(strlen(buf) + 1);
 330         if (home_path == NULL)
 331             return;
 332         strcpy(home_path, buf);
 333         Arguments::set_java_home(home_path);
 334 
 335         if (!set_boot_path('/', ':'))
 336             return;
 337     }
 338 
 339     /*
 340      * Where to look for native libraries
 341      *
 342      * Note: Due to a legacy implementation, most of the library path
 343      * is set in the launcher.  This was to accomodate linking restrictions
 344      * on legacy Linux implementations (which are no longer supported).
 345      * Eventually, all the library path setting will be done here.
 346      *
 347      * However, to prevent the proliferation of improperly built native
 348      * libraries, the new path component /usr/java/packages is added here.
 349      * Eventually, all the library path setting will be done here.
 350      */
 351     {
 352         char *ld_library_path;
 353 
 354         /*
 355          * Construct the invariant part of ld_library_path. Note that the
 356          * space for the colon and the trailing null are provided by the
 357          * nulls included by the sizeof operator (so actually we allocate
 358          * a byte more than necessary).
 359          */
 360         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 361             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 362         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 363 
 364         /*
 365          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 366          * should always exist (until the legacy problem cited above is
 367          * addressed).
 368          */
 369         char *v = getenv("LD_LIBRARY_PATH");
 370         if (v != NULL) {
 371             char *t = ld_library_path;
 372             /* That's +1 for the colon and +1 for the trailing '\0' */
 373             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 374             sprintf(ld_library_path, "%s:%s", v, t);
 375         }
 376         Arguments::set_library_path(ld_library_path);
 377     }
 378 
 379     /*
 380      * Extensions directories.
 381      *
 382      * Note that the space for the colon and the trailing null are provided
 383      * by the nulls included by the sizeof operator (so actually one byte more
 384      * than necessary is allocated).
 385      */
 386     {
 387         char *buf = malloc(strlen(Arguments::get_java_home()) +
 388             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 389         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 390             Arguments::get_java_home());
 391         Arguments::set_ext_dirs(buf);
 392     }
 393 
 394     /* Endorsed standards default directory. */
 395     {
 396         char * buf;
 397         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 398         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 399         Arguments::set_endorsed_dirs(buf);
 400     }
 401   }
 402 
 403 #undef malloc
 404 #undef getenv
 405 #undef EXTENSIONS_DIR
 406 #undef ENDORSED_DIR
 407 
 408   // Done
 409   return;
 410 }
 411 
 412 ////////////////////////////////////////////////////////////////////////////////
 413 // breakpoint support
 414 
 415 void os::breakpoint() {
 416   BREAKPOINT;
 417 }
 418 
 419 extern "C" void breakpoint() {
 420   // use debugger to set breakpoint here
 421 }
 422 
 423 ////////////////////////////////////////////////////////////////////////////////
 424 // signal support
 425 
 426 debug_only(static bool signal_sets_initialized = false);
 427 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 428 
 429 bool os::Linux::is_sig_ignored(int sig) {
 430       struct sigaction oact;
 431       sigaction(sig, (struct sigaction*)NULL, &oact);
 432       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 433                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 434       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 435            return true;
 436       else
 437            return false;
 438 }
 439 
 440 void os::Linux::signal_sets_init() {
 441   // Should also have an assertion stating we are still single-threaded.
 442   assert(!signal_sets_initialized, "Already initialized");
 443   // Fill in signals that are necessarily unblocked for all threads in
 444   // the VM. Currently, we unblock the following signals:
 445   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 446   //                         by -Xrs (=ReduceSignalUsage));
 447   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 448   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 449   // the dispositions or masks wrt these signals.
 450   // Programs embedding the VM that want to use the above signals for their
 451   // own purposes must, at this time, use the "-Xrs" option to prevent
 452   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 453   // (See bug 4345157, and other related bugs).
 454   // In reality, though, unblocking these signals is really a nop, since
 455   // these signals are not blocked by default.
 456   sigemptyset(&unblocked_sigs);
 457   sigemptyset(&allowdebug_blocked_sigs);
 458   sigaddset(&unblocked_sigs, SIGILL);
 459   sigaddset(&unblocked_sigs, SIGSEGV);
 460   sigaddset(&unblocked_sigs, SIGBUS);
 461   sigaddset(&unblocked_sigs, SIGFPE);
 462   sigaddset(&unblocked_sigs, SR_signum);
 463 
 464   if (!ReduceSignalUsage) {
 465    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 466       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 467       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 468    }
 469    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 470       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 471       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 472    }
 473    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 474       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 475       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 476    }
 477   }
 478   // Fill in signals that are blocked by all but the VM thread.
 479   sigemptyset(&vm_sigs);
 480   if (!ReduceSignalUsage)
 481     sigaddset(&vm_sigs, BREAK_SIGNAL);
 482   debug_only(signal_sets_initialized = true);
 483 
 484 }
 485 
 486 // These are signals that are unblocked while a thread is running Java.
 487 // (For some reason, they get blocked by default.)
 488 sigset_t* os::Linux::unblocked_signals() {
 489   assert(signal_sets_initialized, "Not initialized");
 490   return &unblocked_sigs;
 491 }
 492 
 493 // These are the signals that are blocked while a (non-VM) thread is
 494 // running Java. Only the VM thread handles these signals.
 495 sigset_t* os::Linux::vm_signals() {
 496   assert(signal_sets_initialized, "Not initialized");
 497   return &vm_sigs;
 498 }
 499 
 500 // These are signals that are blocked during cond_wait to allow debugger in
 501 sigset_t* os::Linux::allowdebug_blocked_signals() {
 502   assert(signal_sets_initialized, "Not initialized");
 503   return &allowdebug_blocked_sigs;
 504 }
 505 
 506 void os::Linux::hotspot_sigmask(Thread* thread) {
 507 
 508   //Save caller's signal mask before setting VM signal mask
 509   sigset_t caller_sigmask;
 510   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 511 
 512   OSThread* osthread = thread->osthread();
 513   osthread->set_caller_sigmask(caller_sigmask);
 514 
 515   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 516 
 517   if (!ReduceSignalUsage) {
 518     if (thread->is_VM_thread()) {
 519       // Only the VM thread handles BREAK_SIGNAL ...
 520       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 521     } else {
 522       // ... all other threads block BREAK_SIGNAL
 523       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 524     }
 525   }
 526 }
 527 
 528 //////////////////////////////////////////////////////////////////////////////
 529 // detecting pthread library
 530 
 531 void os::Linux::libpthread_init() {
 532   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 533   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 534   // generic name for earlier versions.
 535   // Define macros here so we can build HotSpot on old systems.
 536 # ifndef _CS_GNU_LIBC_VERSION
 537 # define _CS_GNU_LIBC_VERSION 2
 538 # endif
 539 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 540 # define _CS_GNU_LIBPTHREAD_VERSION 3
 541 # endif
 542 
 543   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 544   if (n > 0) {
 545      char *str = (char *)malloc(n);
 546      confstr(_CS_GNU_LIBC_VERSION, str, n);
 547      os::Linux::set_glibc_version(str);
 548   } else {
 549      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 550      static char _gnu_libc_version[32];
 551      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 552               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 553      os::Linux::set_glibc_version(_gnu_libc_version);
 554   }
 555 
 556   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 557   if (n > 0) {
 558      char *str = (char *)malloc(n);
 559      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 560      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 561      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 562      // is the case. LinuxThreads has a hard limit on max number of threads.
 563      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 564      // On the other hand, NPTL does not have such a limit, sysconf()
 565      // will return -1 and errno is not changed. Check if it is really NPTL.
 566      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 567          strstr(str, "NPTL") &&
 568          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 569        free(str);
 570        os::Linux::set_libpthread_version("linuxthreads");
 571      } else {
 572        os::Linux::set_libpthread_version(str);
 573      }
 574   } else {
 575     // glibc before 2.3.2 only has LinuxThreads.
 576     os::Linux::set_libpthread_version("linuxthreads");
 577   }
 578 
 579   if (strstr(libpthread_version(), "NPTL")) {
 580      os::Linux::set_is_NPTL();
 581   } else {
 582      os::Linux::set_is_LinuxThreads();
 583   }
 584 
 585   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 586   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 587   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 588      os::Linux::set_is_floating_stack();
 589   }
 590 }
 591 
 592 /////////////////////////////////////////////////////////////////////////////
 593 // thread stack
 594 
 595 // Force Linux kernel to expand current thread stack. If "bottom" is close
 596 // to the stack guard, caller should block all signals.
 597 //
 598 // MAP_GROWSDOWN:
 599 //   A special mmap() flag that is used to implement thread stacks. It tells
 600 //   kernel that the memory region should extend downwards when needed. This
 601 //   allows early versions of LinuxThreads to only mmap the first few pages
 602 //   when creating a new thread. Linux kernel will automatically expand thread
 603 //   stack as needed (on page faults).
 604 //
 605 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 606 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 607 //   region, it's hard to tell if the fault is due to a legitimate stack
 608 //   access or because of reading/writing non-exist memory (e.g. buffer
 609 //   overrun). As a rule, if the fault happens below current stack pointer,
 610 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 611 //   application (see Linux kernel fault.c).
 612 //
 613 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 614 //   stack overflow detection.
 615 //
 616 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 617 //   not use this flag. However, the stack of initial thread is not created
 618 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 619 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 620 //   and then attach the thread to JVM.
 621 //
 622 // To get around the problem and allow stack banging on Linux, we need to
 623 // manually expand thread stack after receiving the SIGSEGV.
 624 //
 625 // There are two ways to expand thread stack to address "bottom", we used
 626 // both of them in JVM before 1.5:
 627 //   1. adjust stack pointer first so that it is below "bottom", and then
 628 //      touch "bottom"
 629 //   2. mmap() the page in question
 630 //
 631 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 632 // if current sp is already near the lower end of page 101, and we need to
 633 // call mmap() to map page 100, it is possible that part of the mmap() frame
 634 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 635 // That will destroy the mmap() frame and cause VM to crash.
 636 //
 637 // The following code works by adjusting sp first, then accessing the "bottom"
 638 // page to force a page fault. Linux kernel will then automatically expand the
 639 // stack mapping.
 640 //
 641 // _expand_stack_to() assumes its frame size is less than page size, which
 642 // should always be true if the function is not inlined.
 643 
 644 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 645 #define NOINLINE
 646 #else
 647 #define NOINLINE __attribute__ ((noinline))
 648 #endif
 649 
 650 static void _expand_stack_to(address bottom) NOINLINE;
 651 
 652 static void _expand_stack_to(address bottom) {
 653   address sp;
 654   size_t size;
 655   volatile char *p;
 656 
 657   // Adjust bottom to point to the largest address within the same page, it
 658   // gives us a one-page buffer if alloca() allocates slightly more memory.
 659   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 660   bottom += os::Linux::page_size() - 1;
 661 
 662   // sp might be slightly above current stack pointer; if that's the case, we
 663   // will alloca() a little more space than necessary, which is OK. Don't use
 664   // os::current_stack_pointer(), as its result can be slightly below current
 665   // stack pointer, causing us to not alloca enough to reach "bottom".
 666   sp = (address)&sp;
 667 
 668   if (sp > bottom) {
 669     size = sp - bottom;
 670     p = (volatile char *)alloca(size);
 671     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 672     p[0] = '\0';
 673   }
 674 }
 675 
 676 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 677   assert(t!=NULL, "just checking");
 678   assert(t->osthread()->expanding_stack(), "expand should be set");
 679   assert(t->stack_base() != NULL, "stack_base was not initialized");
 680 
 681   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 682     sigset_t mask_all, old_sigset;
 683     sigfillset(&mask_all);
 684     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 685     _expand_stack_to(addr);
 686     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 687     return true;
 688   }
 689   return false;
 690 }
 691 
 692 //////////////////////////////////////////////////////////////////////////////
 693 // create new thread
 694 
 695 static address highest_vm_reserved_address();
 696 
 697 // check if it's safe to start a new thread
 698 static bool _thread_safety_check(Thread* thread) {
 699   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 700     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 701     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 702     //   allocated (MAP_FIXED) from high address space. Every thread stack
 703     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 704     //   it to other values if they rebuild LinuxThreads).
 705     //
 706     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 707     // the memory region has already been mmap'ed. That means if we have too
 708     // many threads and/or very large heap, eventually thread stack will
 709     // collide with heap.
 710     //
 711     // Here we try to prevent heap/stack collision by comparing current
 712     // stack bottom with the highest address that has been mmap'ed by JVM
 713     // plus a safety margin for memory maps created by native code.
 714     //
 715     // This feature can be disabled by setting ThreadSafetyMargin to 0
 716     //
 717     if (ThreadSafetyMargin > 0) {
 718       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 719 
 720       // not safe if our stack extends below the safety margin
 721       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 722     } else {
 723       return true;
 724     }
 725   } else {
 726     // Floating stack LinuxThreads or NPTL:
 727     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 728     //   there's not enough space left, pthread_create() will fail. If we come
 729     //   here, that means enough space has been reserved for stack.
 730     return true;
 731   }
 732 }
 733 
 734 // Thread start routine for all newly created threads
 735 static void *java_start(Thread *thread) {
 736   // Try to randomize the cache line index of hot stack frames.
 737   // This helps when threads of the same stack traces evict each other's
 738   // cache lines. The threads can be either from the same JVM instance, or
 739   // from different JVM instances. The benefit is especially true for
 740   // processors with hyperthreading technology.
 741   static int counter = 0;
 742   int pid = os::current_process_id();
 743   alloca(((pid ^ counter++) & 7) * 128);
 744 
 745   ThreadLocalStorage::set_thread(thread);
 746 
 747   OSThread* osthread = thread->osthread();
 748   Monitor* sync = osthread->startThread_lock();
 749 
 750   // non floating stack LinuxThreads needs extra check, see above
 751   if (!_thread_safety_check(thread)) {
 752     // notify parent thread
 753     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 754     osthread->set_state(ZOMBIE);
 755     sync->notify_all();
 756     return NULL;
 757   }
 758 
 759   // thread_id is kernel thread id (similar to Solaris LWP id)
 760   osthread->set_thread_id(os::Linux::gettid());
 761 
 762   if (UseNUMA) {
 763     int lgrp_id = os::numa_get_group_id();
 764     if (lgrp_id != -1) {
 765       thread->set_lgrp_id(lgrp_id);
 766     }
 767   }
 768   // initialize signal mask for this thread
 769   os::Linux::hotspot_sigmask(thread);
 770 
 771   // initialize floating point control register
 772   os::Linux::init_thread_fpu_state();
 773 
 774   // handshaking with parent thread
 775   {
 776     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 777 
 778     // notify parent thread
 779     osthread->set_state(INITIALIZED);
 780     sync->notify_all();
 781 
 782     // wait until os::start_thread()
 783     while (osthread->get_state() == INITIALIZED) {
 784       sync->wait(Mutex::_no_safepoint_check_flag);
 785     }
 786   }
 787 
 788   // call one more level start routine
 789   thread->run();
 790 
 791   return 0;
 792 }
 793 
 794 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 795   assert(thread->osthread() == NULL, "caller responsible");
 796 
 797   // Allocate the OSThread object
 798   OSThread* osthread = new OSThread(NULL, NULL);
 799   if (osthread == NULL) {
 800     return false;
 801   }
 802 
 803   // set the correct thread state
 804   osthread->set_thread_type(thr_type);
 805 
 806   // Initial state is ALLOCATED but not INITIALIZED
 807   osthread->set_state(ALLOCATED);
 808 
 809   thread->set_osthread(osthread);
 810 
 811   // init thread attributes
 812   pthread_attr_t attr;
 813   pthread_attr_init(&attr);
 814   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 815 
 816   // stack size
 817   if (os::Linux::supports_variable_stack_size()) {
 818     // calculate stack size if it's not specified by caller
 819     if (stack_size == 0) {
 820       stack_size = os::Linux::default_stack_size(thr_type);
 821 
 822       switch (thr_type) {
 823       case os::java_thread:
 824         // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
 825         if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
 826         break;
 827       case os::compiler_thread:
 828         if (CompilerThreadStackSize > 0) {
 829           stack_size = (size_t)(CompilerThreadStackSize * K);
 830           break;
 831         } // else fall through:
 832           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 833       case os::vm_thread:
 834       case os::pgc_thread:
 835       case os::cgc_thread:
 836       case os::watcher_thread:
 837         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 838         break;
 839       }
 840     }
 841 
 842     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 843     pthread_attr_setstacksize(&attr, stack_size);
 844   } else {
 845     // let pthread_create() pick the default value.
 846   }
 847 
 848   // glibc guard page
 849   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 850 
 851   ThreadState state;
 852 
 853   {
 854     // Serialize thread creation if we are running with fixed stack LinuxThreads
 855     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 856     if (lock) {
 857       os::Linux::createThread_lock()->lock_without_safepoint_check();
 858     }
 859 
 860     pthread_t tid;
 861     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 862 
 863     pthread_attr_destroy(&attr);
 864 
 865     if (ret != 0) {
 866       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 867         perror("pthread_create()");
 868       }
 869       // Need to clean up stuff we've allocated so far
 870       thread->set_osthread(NULL);
 871       delete osthread;
 872       if (lock) os::Linux::createThread_lock()->unlock();
 873       return false;
 874     }
 875 
 876     // Store pthread info into the OSThread
 877     osthread->set_pthread_id(tid);
 878 
 879     // Wait until child thread is either initialized or aborted
 880     {
 881       Monitor* sync_with_child = osthread->startThread_lock();
 882       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 883       while ((state = osthread->get_state()) == ALLOCATED) {
 884         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 885       }
 886     }
 887 
 888     if (lock) {
 889       os::Linux::createThread_lock()->unlock();
 890     }
 891   }
 892 
 893   // Aborted due to thread limit being reached
 894   if (state == ZOMBIE) {
 895       thread->set_osthread(NULL);
 896       delete osthread;
 897       return false;
 898   }
 899 
 900   // The thread is returned suspended (in state INITIALIZED),
 901   // and is started higher up in the call chain
 902   assert(state == INITIALIZED, "race condition");
 903   return true;
 904 }
 905 
 906 /////////////////////////////////////////////////////////////////////////////
 907 // attach existing thread
 908 
 909 // bootstrap the main thread
 910 bool os::create_main_thread(JavaThread* thread) {
 911   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 912   return create_attached_thread(thread);
 913 }
 914 
 915 bool os::create_attached_thread(JavaThread* thread) {
 916 #ifdef ASSERT
 917     thread->verify_not_published();
 918 #endif
 919 
 920   // Allocate the OSThread object
 921   OSThread* osthread = new OSThread(NULL, NULL);
 922 
 923   if (osthread == NULL) {
 924     return false;
 925   }
 926 
 927   // Store pthread info into the OSThread
 928   osthread->set_thread_id(os::Linux::gettid());
 929   osthread->set_pthread_id(::pthread_self());
 930 
 931   // initialize floating point control register
 932   os::Linux::init_thread_fpu_state();
 933 
 934   // Initial thread state is RUNNABLE
 935   osthread->set_state(RUNNABLE);
 936 
 937   thread->set_osthread(osthread);
 938 
 939   if (UseNUMA) {
 940     int lgrp_id = os::numa_get_group_id();
 941     if (lgrp_id != -1) {
 942       thread->set_lgrp_id(lgrp_id);
 943     }
 944   }
 945 
 946   if (os::Linux::is_initial_thread()) {
 947     // If current thread is initial thread, its stack is mapped on demand,
 948     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 949     // the entire stack region to avoid SEGV in stack banging.
 950     // It is also useful to get around the heap-stack-gap problem on SuSE
 951     // kernel (see 4821821 for details). We first expand stack to the top
 952     // of yellow zone, then enable stack yellow zone (order is significant,
 953     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 954     // is no gap between the last two virtual memory regions.
 955 
 956     JavaThread *jt = (JavaThread *)thread;
 957     address addr = jt->stack_yellow_zone_base();
 958     assert(addr != NULL, "initialization problem?");
 959     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 960 
 961     osthread->set_expanding_stack();
 962     os::Linux::manually_expand_stack(jt, addr);
 963     osthread->clear_expanding_stack();
 964   }
 965 
 966   // initialize signal mask for this thread
 967   // and save the caller's signal mask
 968   os::Linux::hotspot_sigmask(thread);
 969 
 970   return true;
 971 }
 972 
 973 void os::pd_start_thread(Thread* thread) {
 974   OSThread * osthread = thread->osthread();
 975   assert(osthread->get_state() != INITIALIZED, "just checking");
 976   Monitor* sync_with_child = osthread->startThread_lock();
 977   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 978   sync_with_child->notify();
 979 }
 980 
 981 // Free Linux resources related to the OSThread
 982 void os::free_thread(OSThread* osthread) {
 983   assert(osthread != NULL, "osthread not set");
 984 
 985   if (Thread::current()->osthread() == osthread) {
 986     // Restore caller's signal mask
 987     sigset_t sigmask = osthread->caller_sigmask();
 988     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 989    }
 990 
 991   delete osthread;
 992 }
 993 
 994 //////////////////////////////////////////////////////////////////////////////
 995 // thread local storage
 996 
 997 int os::allocate_thread_local_storage() {
 998   pthread_key_t key;
 999   int rslt = pthread_key_create(&key, NULL);
1000   assert(rslt == 0, "cannot allocate thread local storage");
1001   return (int)key;
1002 }
1003 
1004 // Note: This is currently not used by VM, as we don't destroy TLS key
1005 // on VM exit.
1006 void os::free_thread_local_storage(int index) {
1007   int rslt = pthread_key_delete((pthread_key_t)index);
1008   assert(rslt == 0, "invalid index");
1009 }
1010 
1011 void os::thread_local_storage_at_put(int index, void* value) {
1012   int rslt = pthread_setspecific((pthread_key_t)index, value);
1013   assert(rslt == 0, "pthread_setspecific failed");
1014 }
1015 
1016 extern "C" Thread* get_thread() {
1017   return ThreadLocalStorage::thread();
1018 }
1019 
1020 //////////////////////////////////////////////////////////////////////////////
1021 // initial thread
1022 
1023 // Check if current thread is the initial thread, similar to Solaris thr_main.
1024 bool os::Linux::is_initial_thread(void) {
1025   char dummy;
1026   // If called before init complete, thread stack bottom will be null.
1027   // Can be called if fatal error occurs before initialization.
1028   if (initial_thread_stack_bottom() == NULL) return false;
1029   assert(initial_thread_stack_bottom() != NULL &&
1030          initial_thread_stack_size()   != 0,
1031          "os::init did not locate initial thread's stack region");
1032   if ((address)&dummy >= initial_thread_stack_bottom() &&
1033       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1034        return true;
1035   else return false;
1036 }
1037 
1038 // Find the virtual memory area that contains addr
1039 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1040   FILE *fp = fopen("/proc/self/maps", "r");
1041   if (fp) {
1042     address low, high;
1043     while (!feof(fp)) {
1044       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1045         if (low <= addr && addr < high) {
1046            if (vma_low)  *vma_low  = low;
1047            if (vma_high) *vma_high = high;
1048            fclose (fp);
1049            return true;
1050         }
1051       }
1052       for (;;) {
1053         int ch = fgetc(fp);
1054         if (ch == EOF || ch == (int)'\n') break;
1055       }
1056     }
1057     fclose(fp);
1058   }
1059   return false;
1060 }
1061 
1062 // Locate initial thread stack. This special handling of initial thread stack
1063 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1064 // bogus value for initial thread.
1065 void os::Linux::capture_initial_stack(size_t max_size) {
1066   // stack size is the easy part, get it from RLIMIT_STACK
1067   size_t stack_size;
1068   struct rlimit rlim;
1069   getrlimit(RLIMIT_STACK, &rlim);
1070   stack_size = rlim.rlim_cur;
1071 
1072   // 6308388: a bug in ld.so will relocate its own .data section to the
1073   //   lower end of primordial stack; reduce ulimit -s value a little bit
1074   //   so we won't install guard page on ld.so's data section.
1075   stack_size -= 2 * page_size();
1076 
1077   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1078   //   7.1, in both cases we will get 2G in return value.
1079   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1080   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1081   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1082   //   in case other parts in glibc still assumes 2M max stack size.
1083   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1084 #ifndef IA64
1085   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1086 #else
1087   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1088   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1089 #endif
1090 
1091   // Try to figure out where the stack base (top) is. This is harder.
1092   //
1093   // When an application is started, glibc saves the initial stack pointer in
1094   // a global variable "__libc_stack_end", which is then used by system
1095   // libraries. __libc_stack_end should be pretty close to stack top. The
1096   // variable is available since the very early days. However, because it is
1097   // a private interface, it could disappear in the future.
1098   //
1099   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1100   // to __libc_stack_end, it is very close to stack top, but isn't the real
1101   // stack top. Note that /proc may not exist if VM is running as a chroot
1102   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1103   // /proc/<pid>/stat could change in the future (though unlikely).
1104   //
1105   // We try __libc_stack_end first. If that doesn't work, look for
1106   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1107   // as a hint, which should work well in most cases.
1108 
1109   uintptr_t stack_start;
1110 
1111   // try __libc_stack_end first
1112   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1113   if (p && *p) {
1114     stack_start = *p;
1115   } else {
1116     // see if we can get the start_stack field from /proc/self/stat
1117     FILE *fp;
1118     int pid;
1119     char state;
1120     int ppid;
1121     int pgrp;
1122     int session;
1123     int nr;
1124     int tpgrp;
1125     unsigned long flags;
1126     unsigned long minflt;
1127     unsigned long cminflt;
1128     unsigned long majflt;
1129     unsigned long cmajflt;
1130     unsigned long utime;
1131     unsigned long stime;
1132     long cutime;
1133     long cstime;
1134     long prio;
1135     long nice;
1136     long junk;
1137     long it_real;
1138     uintptr_t start;
1139     uintptr_t vsize;
1140     uintptr_t rss;
1141     unsigned long rsslim;
1142     uintptr_t scodes;
1143     uintptr_t ecode;
1144     int i;
1145 
1146     // Figure what the primordial thread stack base is. Code is inspired
1147     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1148     // followed by command name surrounded by parentheses, state, etc.
1149     char stat[2048];
1150     int statlen;
1151 
1152     fp = fopen("/proc/self/stat", "r");
1153     if (fp) {
1154       statlen = fread(stat, 1, 2047, fp);
1155       stat[statlen] = '\0';
1156       fclose(fp);
1157 
1158       // Skip pid and the command string. Note that we could be dealing with
1159       // weird command names, e.g. user could decide to rename java launcher
1160       // to "java 1.4.2 :)", then the stat file would look like
1161       //                1234 (java 1.4.2 :)) R ... ...
1162       // We don't really need to know the command string, just find the last
1163       // occurrence of ")" and then start parsing from there. See bug 4726580.
1164       char * s = strrchr(stat, ')');
1165 
1166       i = 0;
1167       if (s) {
1168         // Skip blank chars
1169         do s++; while (isspace(*s));
1170 
1171         /*                                     1   1   1   1   1   1   1   1   1   1   2   2   2   2   2   2   2   2   2 */
1172         /*              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 */
1173         i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld "
1174                    UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT
1175                    " %lu "
1176                    UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT,
1177              &state,          /* 3  %c  */
1178              &ppid,           /* 4  %d  */
1179              &pgrp,           /* 5  %d  */
1180              &session,        /* 6  %d  */
1181              &nr,             /* 7  %d  */
1182              &tpgrp,          /* 8  %d  */
1183              &flags,          /* 9  %lu  */
1184              &minflt,         /* 10 %lu  */
1185              &cminflt,        /* 11 %lu  */
1186              &majflt,         /* 12 %lu  */
1187              &cmajflt,        /* 13 %lu  */
1188              &utime,          /* 14 %lu  */
1189              &stime,          /* 15 %lu  */
1190              &cutime,         /* 16 %ld  */
1191              &cstime,         /* 17 %ld  */
1192              &prio,           /* 18 %ld  */
1193              &nice,           /* 19 %ld  */
1194              &junk,           /* 20 %ld  */
1195              &it_real,        /* 21 %ld  */
1196              &start,          /* 22 UINTX_FORMAT  */
1197              &vsize,          /* 23 UINTX_FORMAT  */
1198              &rss,            /* 24 UINTX_FORMAT  */
1199              &rsslim,         /* 25 %lu  */
1200              &scodes,         /* 26 UINTX_FORMAT  */
1201              &ecode,          /* 27 UINTX_FORMAT  */
1202              &stack_start);   /* 28 UINTX_FORMAT  */
1203       }
1204 
1205       if (i != 28 - 2) {
1206          assert(false, "Bad conversion from /proc/self/stat");
1207          // product mode - assume we are the initial thread, good luck in the
1208          // embedded case.
1209          warning("Can't detect initial thread stack location - bad conversion");
1210          stack_start = (uintptr_t) &rlim;
1211       }
1212     } else {
1213       // For some reason we can't open /proc/self/stat (for example, running on
1214       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1215       // most cases, so don't abort:
1216       warning("Can't detect initial thread stack location - no /proc/self/stat");
1217       stack_start = (uintptr_t) &rlim;
1218     }
1219   }
1220 
1221   // Now we have a pointer (stack_start) very close to the stack top, the
1222   // next thing to do is to figure out the exact location of stack top. We
1223   // can find out the virtual memory area that contains stack_start by
1224   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1225   // and its upper limit is the real stack top. (again, this would fail if
1226   // running inside chroot, because /proc may not exist.)
1227 
1228   uintptr_t stack_top;
1229   address low, high;
1230   if (find_vma((address)stack_start, &low, &high)) {
1231     // success, "high" is the true stack top. (ignore "low", because initial
1232     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1233     stack_top = (uintptr_t)high;
1234   } else {
1235     // failed, likely because /proc/self/maps does not exist
1236     warning("Can't detect initial thread stack location - find_vma failed");
1237     // best effort: stack_start is normally within a few pages below the real
1238     // stack top, use it as stack top, and reduce stack size so we won't put
1239     // guard page outside stack.
1240     stack_top = stack_start;
1241     stack_size -= 16 * page_size();
1242   }
1243 
1244   // stack_top could be partially down the page so align it
1245   stack_top = align_size_up(stack_top, page_size());
1246 
1247   if (max_size && stack_size > max_size) {
1248      _initial_thread_stack_size = max_size;
1249   } else {
1250      _initial_thread_stack_size = stack_size;
1251   }
1252 
1253   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1254   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1255 }
1256 
1257 ////////////////////////////////////////////////////////////////////////////////
1258 // time support
1259 
1260 // Time since start-up in seconds to a fine granularity.
1261 // Used by VMSelfDestructTimer and the MemProfiler.
1262 double os::elapsedTime() {
1263 
1264   return (double)(os::elapsed_counter()) * 0.000001;
1265 }
1266 
1267 jlong os::elapsed_counter() {
1268   timeval time;
1269   int status = gettimeofday(&time, NULL);
1270   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1271 }
1272 
1273 jlong os::elapsed_frequency() {
1274   return (1000 * 1000);
1275 }
1276 
1277 // For now, we say that linux does not support vtime.  I have no idea
1278 // whether it can actually be made to (DLD, 9/13/05).
1279 
1280 bool os::supports_vtime() { return false; }
1281 bool os::enable_vtime()   { return false; }
1282 bool os::vtime_enabled()  { return false; }
1283 double os::elapsedVTime() {
1284   // better than nothing, but not much
1285   return elapsedTime();
1286 }
1287 
1288 jlong os::javaTimeMillis() {
1289   timeval time;
1290   int status = gettimeofday(&time, NULL);
1291   assert(status != -1, "linux error");
1292   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1293 }
1294 
1295 #ifndef CLOCK_MONOTONIC
1296 #define CLOCK_MONOTONIC (1)
1297 #endif
1298 
1299 void os::Linux::clock_init() {
1300   // we do dlopen's in this particular order due to bug in linux
1301   // dynamical loader (see 6348968) leading to crash on exit
1302   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1303   if (handle == NULL) {
1304     handle = dlopen("librt.so", RTLD_LAZY);
1305   }
1306 
1307   if (handle) {
1308     int (*clock_getres_func)(clockid_t, struct timespec*) =
1309            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1310     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1311            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1312     if (clock_getres_func && clock_gettime_func) {
1313       // See if monotonic clock is supported by the kernel. Note that some
1314       // early implementations simply return kernel jiffies (updated every
1315       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1316       // for nano time (though the monotonic property is still nice to have).
1317       // It's fixed in newer kernels, however clock_getres() still returns
1318       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1319       // resolution for now. Hopefully as people move to new kernels, this
1320       // won't be a problem.
1321       struct timespec res;
1322       struct timespec tp;
1323       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1324           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1325         // yes, monotonic clock is supported
1326         _clock_gettime = clock_gettime_func;
1327       } else {
1328         // close librt if there is no monotonic clock
1329         dlclose(handle);
1330       }
1331     }
1332   }
1333 }
1334 
1335 #ifndef SYS_clock_getres
1336 
1337 #if defined(IA32) || defined(AMD64)
1338 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1339 #else
1340 #error Value of SYS_clock_getres not known on this platform
1341 #endif
1342 
1343 #endif
1344 
1345 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1346 
1347 void os::Linux::fast_thread_clock_init() {
1348   if (!UseLinuxPosixThreadCPUClocks) {
1349     return;
1350   }
1351   clockid_t clockid;
1352   struct timespec tp;
1353   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1354       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1355 
1356   // Switch to using fast clocks for thread cpu time if
1357   // the sys_clock_getres() returns 0 error code.
1358   // Note, that some kernels may support the current thread
1359   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1360   // returned by the pthread_getcpuclockid().
1361   // If the fast Posix clocks are supported then the sys_clock_getres()
1362   // must return at least tp.tv_sec == 0 which means a resolution
1363   // better than 1 sec. This is extra check for reliability.
1364 
1365   if(pthread_getcpuclockid_func &&
1366      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1367      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1368 
1369     _supports_fast_thread_cpu_time = true;
1370     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1371   }
1372 }
1373 
1374 jlong os::javaTimeNanos() {
1375   if (Linux::supports_monotonic_clock()) {
1376     struct timespec tp;
1377     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1378     assert(status == 0, "gettime error");
1379     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1380     return result;
1381   } else {
1382     timeval time;
1383     int status = gettimeofday(&time, NULL);
1384     assert(status != -1, "linux error");
1385     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1386     return 1000 * usecs;
1387   }
1388 }
1389 
1390 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1391   if (Linux::supports_monotonic_clock()) {
1392     info_ptr->max_value = ALL_64_BITS;
1393 
1394     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1395     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1396     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1397   } else {
1398     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1399     info_ptr->max_value = ALL_64_BITS;
1400 
1401     // gettimeofday is a real time clock so it skips
1402     info_ptr->may_skip_backward = true;
1403     info_ptr->may_skip_forward = true;
1404   }
1405 
1406   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1407 }
1408 
1409 // Return the real, user, and system times in seconds from an
1410 // arbitrary fixed point in the past.
1411 bool os::getTimesSecs(double* process_real_time,
1412                       double* process_user_time,
1413                       double* process_system_time) {
1414   struct tms ticks;
1415   clock_t real_ticks = times(&ticks);
1416 
1417   if (real_ticks == (clock_t) (-1)) {
1418     return false;
1419   } else {
1420     double ticks_per_second = (double) clock_tics_per_sec;
1421     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1422     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1423     *process_real_time = ((double) real_ticks) / ticks_per_second;
1424 
1425     return true;
1426   }
1427 }
1428 
1429 
1430 char * os::local_time_string(char *buf, size_t buflen) {
1431   struct tm t;
1432   time_t long_time;
1433   time(&long_time);
1434   localtime_r(&long_time, &t);
1435   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1436                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1437                t.tm_hour, t.tm_min, t.tm_sec);
1438   return buf;
1439 }
1440 
1441 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1442   return localtime_r(clock, res);
1443 }
1444 
1445 ////////////////////////////////////////////////////////////////////////////////
1446 // runtime exit support
1447 
1448 // Note: os::shutdown() might be called very early during initialization, or
1449 // called from signal handler. Before adding something to os::shutdown(), make
1450 // sure it is async-safe and can handle partially initialized VM.
1451 void os::shutdown() {
1452 
1453   // allow PerfMemory to attempt cleanup of any persistent resources
1454   perfMemory_exit();
1455 
1456   // needs to remove object in file system
1457   AttachListener::abort();
1458 
1459   // flush buffered output, finish log files
1460   ostream_abort();
1461 
1462   // Check for abort hook
1463   abort_hook_t abort_hook = Arguments::abort_hook();
1464   if (abort_hook != NULL) {
1465     abort_hook();
1466   }
1467 
1468 }
1469 
1470 // Note: os::abort() might be called very early during initialization, or
1471 // called from signal handler. Before adding something to os::abort(), make
1472 // sure it is async-safe and can handle partially initialized VM.
1473 void os::abort(bool dump_core) {
1474   os::shutdown();
1475   if (dump_core) {
1476 #ifndef PRODUCT
1477     fdStream out(defaultStream::output_fd());
1478     out.print_raw("Current thread is ");
1479     char buf[16];
1480     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1481     out.print_raw_cr(buf);
1482     out.print_raw_cr("Dumping core ...");
1483 #endif
1484     ::abort(); // dump core
1485   }
1486 
1487   ::exit(1);
1488 }
1489 
1490 // Die immediately, no exit hook, no abort hook, no cleanup.
1491 void os::die() {
1492   // _exit() on LinuxThreads only kills current thread
1493   ::abort();
1494 }
1495 
1496 // unused on linux for now.
1497 void os::set_error_file(const char *logfile) {}
1498 
1499 intx os::current_thread_id() { return (intx)pthread_self(); }
1500 int os::current_process_id() {
1501 
1502   // Under the old linux thread library, linux gives each thread
1503   // its own process id. Because of this each thread will return
1504   // a different pid if this method were to return the result
1505   // of getpid(2). Linux provides no api that returns the pid
1506   // of the launcher thread for the vm. This implementation
1507   // returns a unique pid, the pid of the launcher thread
1508   // that starts the vm 'process'.
1509 
1510   // Under the NPTL, getpid() returns the same pid as the
1511   // launcher thread rather than a unique pid per thread.
1512   // Use gettid() if you want the old pre NPTL behaviour.
1513 
1514   // if you are looking for the result of a call to getpid() that
1515   // returns a unique pid for the calling thread, then look at the
1516   // OSThread::thread_id() method in osThread_linux.hpp file
1517 
1518   return (int)(_initial_pid ? _initial_pid : getpid());
1519 }
1520 
1521 // DLL functions
1522 
1523 const char* os::dll_file_extension() { return ".so"; }
1524 
1525 const char* os::get_temp_directory() {
1526   const char *prop = Arguments::get_property("java.io.tmpdir");
1527   return prop == NULL ? "/tmp" : prop;
1528 }
1529 
1530 static bool file_exists(const char* filename) {
1531   struct stat statbuf;
1532   if (filename == NULL || strlen(filename) == 0) {
1533     return false;
1534   }
1535   return os::stat(filename, &statbuf) == 0;
1536 }
1537 
1538 void os::dll_build_name(char* buffer, size_t buflen,
1539                         const char* pname, const char* fname) {
1540   // Copied from libhpi
1541   const size_t pnamelen = pname ? strlen(pname) : 0;
1542 
1543   // Quietly truncate on buffer overflow.  Should be an error.
1544   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1545       *buffer = '\0';
1546       return;
1547   }
1548 
1549   if (pnamelen == 0) {
1550     snprintf(buffer, buflen, "lib%s.so", fname);
1551   } else if (strchr(pname, *os::path_separator()) != NULL) {
1552     int n;
1553     char** pelements = split_path(pname, &n);
1554     for (int i = 0 ; i < n ; i++) {
1555       // Really shouldn't be NULL, but check can't hurt
1556       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1557         continue; // skip the empty path values
1558       }
1559       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1560       if (file_exists(buffer)) {
1561         break;
1562       }
1563     }
1564     // release the storage
1565     for (int i = 0 ; i < n ; i++) {
1566       if (pelements[i] != NULL) {
1567         FREE_C_HEAP_ARRAY(char, pelements[i]);
1568       }
1569     }
1570     if (pelements != NULL) {
1571       FREE_C_HEAP_ARRAY(char*, pelements);
1572     }
1573   } else {
1574     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1575   }
1576 }
1577 
1578 const char* os::get_current_directory(char *buf, int buflen) {
1579   return getcwd(buf, buflen);
1580 }
1581 
1582 // check if addr is inside libjvm[_g].so
1583 bool os::address_is_in_vm(address addr) {
1584   static address libjvm_base_addr;
1585   Dl_info dlinfo;
1586 
1587   if (libjvm_base_addr == NULL) {
1588     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1589     libjvm_base_addr = (address)dlinfo.dli_fbase;
1590     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1591   }
1592 
1593   if (dladdr((void *)addr, &dlinfo)) {
1594     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1595   }
1596 
1597   return false;
1598 }
1599 
1600 bool os::dll_address_to_function_name(address addr, char *buf,
1601                                       int buflen, int *offset) {
1602   Dl_info dlinfo;
1603 
1604   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1605     if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1606     if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1607     return true;
1608   } else {
1609     if (buf) buf[0] = '\0';
1610     if (offset) *offset = -1;
1611     return false;
1612   }
1613 }
1614 
1615 struct _address_to_library_name {
1616   address addr;          // input : memory address
1617   size_t  buflen;        //         size of fname
1618   char*   fname;         // output: library name
1619   address base;          //         library base addr
1620 };
1621 
1622 static int address_to_library_name_callback(struct dl_phdr_info *info,
1623                                             size_t size, void *data) {
1624   int i;
1625   bool found = false;
1626   address libbase = NULL;
1627   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1628 
1629   // iterate through all loadable segments
1630   for (i = 0; i < info->dlpi_phnum; i++) {
1631     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1632     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1633       // base address of a library is the lowest address of its loaded
1634       // segments.
1635       if (libbase == NULL || libbase > segbase) {
1636         libbase = segbase;
1637       }
1638       // see if 'addr' is within current segment
1639       if (segbase <= d->addr &&
1640           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1641         found = true;
1642       }
1643     }
1644   }
1645 
1646   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1647   // so dll_address_to_library_name() can fall through to use dladdr() which
1648   // can figure out executable name from argv[0].
1649   if (found && info->dlpi_name && info->dlpi_name[0]) {
1650     d->base = libbase;
1651     if (d->fname) {
1652       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1653     }
1654     return 1;
1655   }
1656   return 0;
1657 }
1658 
1659 bool os::dll_address_to_library_name(address addr, char* buf,
1660                                      int buflen, int* offset) {
1661   Dl_info dlinfo;
1662   struct _address_to_library_name data;
1663 
1664   // There is a bug in old glibc dladdr() implementation that it could resolve
1665   // to wrong library name if the .so file has a base address != NULL. Here
1666   // we iterate through the program headers of all loaded libraries to find
1667   // out which library 'addr' really belongs to. This workaround can be
1668   // removed once the minimum requirement for glibc is moved to 2.3.x.
1669   data.addr = addr;
1670   data.fname = buf;
1671   data.buflen = buflen;
1672   data.base = NULL;
1673   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1674 
1675   if (rslt) {
1676      // buf already contains library name
1677      if (offset) *offset = addr - data.base;
1678      return true;
1679   } else if (dladdr((void*)addr, &dlinfo)){
1680      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1681      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1682      return true;
1683   } else {
1684      if (buf) buf[0] = '\0';
1685      if (offset) *offset = -1;
1686      return false;
1687   }
1688 }
1689 
1690   // Loads .dll/.so and
1691   // in case of error it checks if .dll/.so was built for the
1692   // same architecture as Hotspot is running on
1693 
1694 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1695 {
1696   void * result= ::dlopen(filename, RTLD_LAZY);
1697   if (result != NULL) {
1698     // Successful loading
1699     return result;
1700   }
1701 
1702   Elf32_Ehdr elf_head;
1703 
1704   // Read system error message into ebuf
1705   // It may or may not be overwritten below
1706   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1707   ebuf[ebuflen-1]='\0';
1708   int diag_msg_max_length=ebuflen-strlen(ebuf);
1709   char* diag_msg_buf=ebuf+strlen(ebuf);
1710 
1711   if (diag_msg_max_length==0) {
1712     // No more space in ebuf for additional diagnostics message
1713     return NULL;
1714   }
1715 
1716 
1717   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1718 
1719   if (file_descriptor < 0) {
1720     // Can't open library, report dlerror() message
1721     return NULL;
1722   }
1723 
1724   bool failed_to_read_elf_head=
1725     (sizeof(elf_head)!=
1726         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1727 
1728   ::close(file_descriptor);
1729   if (failed_to_read_elf_head) {
1730     // file i/o error - report dlerror() msg
1731     return NULL;
1732   }
1733 
1734   typedef struct {
1735     Elf32_Half  code;         // Actual value as defined in elf.h
1736     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1737     char        elf_class;    // 32 or 64 bit
1738     char        endianess;    // MSB or LSB
1739     char*       name;         // String representation
1740   } arch_t;
1741 
1742   #ifndef EM_486
1743   #define EM_486          6               /* Intel 80486 */
1744   #endif
1745 
1746   static const arch_t arch_array[]={
1747     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1748     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1749     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1750     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1751     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1752     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1753     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1754     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1755     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1756     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1757     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1758     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1759     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1760     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1761     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1762     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1763   };
1764 
1765   #if  (defined IA32)
1766     static  Elf32_Half running_arch_code=EM_386;
1767   #elif   (defined AMD64)
1768     static  Elf32_Half running_arch_code=EM_X86_64;
1769   #elif  (defined IA64)
1770     static  Elf32_Half running_arch_code=EM_IA_64;
1771   #elif  (defined __sparc) && (defined _LP64)
1772     static  Elf32_Half running_arch_code=EM_SPARCV9;
1773   #elif  (defined __sparc) && (!defined _LP64)
1774     static  Elf32_Half running_arch_code=EM_SPARC;
1775   #elif  (defined __powerpc64__)
1776     static  Elf32_Half running_arch_code=EM_PPC64;
1777   #elif  (defined __powerpc__)
1778     static  Elf32_Half running_arch_code=EM_PPC;
1779   #elif  (defined ARM)
1780     static  Elf32_Half running_arch_code=EM_ARM;
1781   #elif  (defined S390)
1782     static  Elf32_Half running_arch_code=EM_S390;
1783   #elif  (defined ALPHA)
1784     static  Elf32_Half running_arch_code=EM_ALPHA;
1785   #elif  (defined MIPSEL)
1786     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1787   #elif  (defined PARISC)
1788     static  Elf32_Half running_arch_code=EM_PARISC;
1789   #elif  (defined MIPS)
1790     static  Elf32_Half running_arch_code=EM_MIPS;
1791   #elif  (defined M68K)
1792     static  Elf32_Half running_arch_code=EM_68K;
1793   #else
1794     #error Method os::dll_load requires that one of following is defined:\
1795          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1796   #endif
1797 
1798   // Identify compatability class for VM's architecture and library's architecture
1799   // Obtain string descriptions for architectures
1800 
1801   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1802   int running_arch_index=-1;
1803 
1804   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1805     if (running_arch_code == arch_array[i].code) {
1806       running_arch_index    = i;
1807     }
1808     if (lib_arch.code == arch_array[i].code) {
1809       lib_arch.compat_class = arch_array[i].compat_class;
1810       lib_arch.name         = arch_array[i].name;
1811     }
1812   }
1813 
1814   assert(running_arch_index != -1,
1815     "Didn't find running architecture code (running_arch_code) in arch_array");
1816   if (running_arch_index == -1) {
1817     // Even though running architecture detection failed
1818     // we may still continue with reporting dlerror() message
1819     return NULL;
1820   }
1821 
1822   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1823     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1824     return NULL;
1825   }
1826 
1827 #ifndef S390
1828   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1829     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1830     return NULL;
1831   }
1832 #endif // !S390
1833 
1834   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1835     if ( lib_arch.name!=NULL ) {
1836       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1837         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1838         lib_arch.name, arch_array[running_arch_index].name);
1839     } else {
1840       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1841       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1842         lib_arch.code,
1843         arch_array[running_arch_index].name);
1844     }
1845   }
1846 
1847   return NULL;
1848 }
1849 
1850 /*
1851  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
1852  * chances are you might want to run the generated bits against glibc-2.0
1853  * libdl.so, so always use locking for any version of glibc.
1854  */
1855 void* os::dll_lookup(void* handle, const char* name) {
1856   pthread_mutex_lock(&dl_mutex);
1857   void* res = dlsym(handle, name);
1858   pthread_mutex_unlock(&dl_mutex);
1859   return res;
1860 }
1861 
1862 
1863 bool _print_ascii_file(const char* filename, outputStream* st) {
1864   int fd = open(filename, O_RDONLY);
1865   if (fd == -1) {
1866      return false;
1867   }
1868 
1869   char buf[32];
1870   int bytes;
1871   while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1872     st->print_raw(buf, bytes);
1873   }
1874 
1875   close(fd);
1876 
1877   return true;
1878 }
1879 
1880 void os::print_dll_info(outputStream *st) {
1881    st->print_cr("Dynamic libraries:");
1882 
1883    char fname[32];
1884    pid_t pid = os::Linux::gettid();
1885 
1886    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1887 
1888    if (!_print_ascii_file(fname, st)) {
1889      st->print("Can not get library information for pid = %d\n", pid);
1890    }
1891 }
1892 
1893 
1894 void os::print_os_info(outputStream* st) {
1895   st->print("OS:");
1896 
1897   // Try to identify popular distros.
1898   // Most Linux distributions have /etc/XXX-release file, which contains
1899   // the OS version string. Some have more than one /etc/XXX-release file
1900   // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1901   // so the order is important.
1902   if (!_print_ascii_file("/etc/mandrake-release", st) &&
1903       !_print_ascii_file("/etc/sun-release", st) &&
1904       !_print_ascii_file("/etc/redhat-release", st) &&
1905       !_print_ascii_file("/etc/SuSE-release", st) &&
1906       !_print_ascii_file("/etc/turbolinux-release", st) &&
1907       !_print_ascii_file("/etc/gentoo-release", st) &&
1908       !_print_ascii_file("/etc/debian_version", st)) {
1909       st->print("Linux");
1910   }
1911   st->cr();
1912 
1913   // kernel
1914   st->print("uname:");
1915   struct utsname name;
1916   uname(&name);
1917   st->print(name.sysname); st->print(" ");
1918   st->print(name.release); st->print(" ");
1919   st->print(name.version); st->print(" ");
1920   st->print(name.machine);
1921   st->cr();
1922 
1923   // Print warning if unsafe chroot environment detected
1924   if (unsafe_chroot_detected) {
1925     st->print("WARNING!! ");
1926     st->print_cr(unstable_chroot_error);
1927   }
1928 
1929   // libc, pthread
1930   st->print("libc:");
1931   st->print(os::Linux::glibc_version()); st->print(" ");
1932   st->print(os::Linux::libpthread_version()); st->print(" ");
1933   if (os::Linux::is_LinuxThreads()) {
1934      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
1935   }
1936   st->cr();
1937 
1938   // rlimit
1939   st->print("rlimit:");
1940   struct rlimit rlim;
1941 
1942   st->print(" STACK ");
1943   getrlimit(RLIMIT_STACK, &rlim);
1944   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1945   else st->print("%uk", rlim.rlim_cur >> 10);
1946 
1947   st->print(", CORE ");
1948   getrlimit(RLIMIT_CORE, &rlim);
1949   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1950   else st->print("%uk", rlim.rlim_cur >> 10);
1951 
1952   st->print(", NPROC ");
1953   getrlimit(RLIMIT_NPROC, &rlim);
1954   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1955   else st->print("%d", rlim.rlim_cur);
1956 
1957   st->print(", NOFILE ");
1958   getrlimit(RLIMIT_NOFILE, &rlim);
1959   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1960   else st->print("%d", rlim.rlim_cur);
1961 
1962   st->print(", AS ");
1963   getrlimit(RLIMIT_AS, &rlim);
1964   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1965   else st->print("%uk", rlim.rlim_cur >> 10);
1966   st->cr();
1967 
1968   // load average
1969   st->print("load average:");
1970   double loadavg[3];
1971   os::loadavg(loadavg, 3);
1972   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1973   st->cr();
1974 }
1975 
1976 void os::print_memory_info(outputStream* st) {
1977 
1978   st->print("Memory:");
1979   st->print(" %dk page", os::vm_page_size()>>10);
1980 
1981   // values in struct sysinfo are "unsigned long"
1982   struct sysinfo si;
1983   sysinfo(&si);
1984 
1985   st->print(", physical " UINT64_FORMAT "k",
1986             os::physical_memory() >> 10);
1987   st->print("(" UINT64_FORMAT "k free)",
1988             os::available_memory() >> 10);
1989   st->print(", swap " UINT64_FORMAT "k",
1990             ((jlong)si.totalswap * si.mem_unit) >> 10);
1991   st->print("(" UINT64_FORMAT "k free)",
1992             ((jlong)si.freeswap * si.mem_unit) >> 10);
1993   st->cr();
1994 }
1995 
1996 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
1997 // but they're the same for all the linux arch that we support
1998 // and they're the same for solaris but there's no common place to put this.
1999 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2000                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2001                           "ILL_COPROC", "ILL_BADSTK" };
2002 
2003 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2004                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2005                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2006 
2007 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2008 
2009 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2010 
2011 void os::print_siginfo(outputStream* st, void* siginfo) {
2012   st->print("siginfo:");
2013 
2014   const int buflen = 100;
2015   char buf[buflen];
2016   siginfo_t *si = (siginfo_t*)siginfo;
2017   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2018   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2019     st->print("si_errno=%s", buf);
2020   } else {
2021     st->print("si_errno=%d", si->si_errno);
2022   }
2023   const int c = si->si_code;
2024   assert(c > 0, "unexpected si_code");
2025   switch (si->si_signo) {
2026   case SIGILL:
2027     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2028     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2029     break;
2030   case SIGFPE:
2031     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2032     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2033     break;
2034   case SIGSEGV:
2035     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2036     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2037     break;
2038   case SIGBUS:
2039     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2040     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2041     break;
2042   default:
2043     st->print(", si_code=%d", si->si_code);
2044     // no si_addr
2045   }
2046 
2047   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2048       UseSharedSpaces) {
2049     FileMapInfo* mapinfo = FileMapInfo::current_info();
2050     if (mapinfo->is_in_shared_space(si->si_addr)) {
2051       st->print("\n\nError accessing class data sharing archive."   \
2052                 " Mapped file inaccessible during execution, "      \
2053                 " possible disk/network problem.");
2054     }
2055   }
2056   st->cr();
2057 }
2058 
2059 
2060 static void print_signal_handler(outputStream* st, int sig,
2061                                  char* buf, size_t buflen);
2062 
2063 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2064   st->print_cr("Signal Handlers:");
2065   print_signal_handler(st, SIGSEGV, buf, buflen);
2066   print_signal_handler(st, SIGBUS , buf, buflen);
2067   print_signal_handler(st, SIGFPE , buf, buflen);
2068   print_signal_handler(st, SIGPIPE, buf, buflen);
2069   print_signal_handler(st, SIGXFSZ, buf, buflen);
2070   print_signal_handler(st, SIGILL , buf, buflen);
2071   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2072   print_signal_handler(st, SR_signum, buf, buflen);
2073   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2074   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2075   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2076   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2077 }
2078 
2079 static char saved_jvm_path[MAXPATHLEN] = {0};
2080 
2081 // Find the full path to the current module, libjvm.so or libjvm_g.so
2082 void os::jvm_path(char *buf, jint len) {
2083   // Error checking.
2084   if (len < MAXPATHLEN) {
2085     assert(false, "must use a large-enough buffer");
2086     buf[0] = '\0';
2087     return;
2088   }
2089   // Lazy resolve the path to current module.
2090   if (saved_jvm_path[0] != 0) {
2091     strcpy(buf, saved_jvm_path);
2092     return;
2093   }
2094 
2095   char dli_fname[MAXPATHLEN];
2096   bool ret = dll_address_to_library_name(
2097                 CAST_FROM_FN_PTR(address, os::jvm_path),
2098                 dli_fname, sizeof(dli_fname), NULL);
2099   assert(ret != 0, "cannot locate libjvm");
2100   if (realpath(dli_fname, buf) == NULL)
2101     return;
2102 
2103   if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2104     // Support for the gamma launcher.  Typical value for buf is
2105     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2106     // the right place in the string, then assume we are installed in a JDK and
2107     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2108     // up the path so it looks like libjvm.so is installed there (append a
2109     // fake suffix hotspot/libjvm.so).
2110     const char *p = buf + strlen(buf) - 1;
2111     for (int count = 0; p > buf && count < 5; ++count) {
2112       for (--p; p > buf && *p != '/'; --p)
2113         /* empty */ ;
2114     }
2115 
2116     if (strncmp(p, "/jre/lib/", 9) != 0) {
2117       // Look for JAVA_HOME in the environment.
2118       char* java_home_var = ::getenv("JAVA_HOME");
2119       if (java_home_var != NULL && java_home_var[0] != 0) {
2120         // Check the current module name "libjvm.so" or "libjvm_g.so".
2121         p = strrchr(buf, '/');
2122         assert(strstr(p, "/libjvm") == p, "invalid library name");
2123         p = strstr(p, "_g") ? "_g" : "";
2124 
2125         if (realpath(java_home_var, buf) == NULL)
2126           return;
2127         sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch);
2128         if (0 == access(buf, F_OK)) {
2129           // Use current module name "libjvm[_g].so" instead of
2130           // "libjvm"debug_only("_g")".so" since for fastdebug version
2131           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2132           // It is used when we are choosing the HPI library's name
2133           // "libhpi[_g].so" in hpi::initialize_get_interface().
2134           sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p);
2135         } else {
2136           // Go back to path of .so
2137           if (realpath(dli_fname, buf) == NULL)
2138             return;
2139         }
2140       }
2141     }
2142   }
2143 
2144   strcpy(saved_jvm_path, buf);
2145 }
2146 
2147 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2148   // no prefix required, not even "_"
2149 }
2150 
2151 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2152   // no suffix required
2153 }
2154 
2155 ////////////////////////////////////////////////////////////////////////////////
2156 // sun.misc.Signal support
2157 
2158 static volatile jint sigint_count = 0;
2159 
2160 static void
2161 UserHandler(int sig, void *siginfo, void *context) {
2162   // 4511530 - sem_post is serialized and handled by the manager thread. When
2163   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2164   // don't want to flood the manager thread with sem_post requests.
2165   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2166       return;
2167 
2168   // Ctrl-C is pressed during error reporting, likely because the error
2169   // handler fails to abort. Let VM die immediately.
2170   if (sig == SIGINT && is_error_reported()) {
2171      os::die();
2172   }
2173 
2174   os::signal_notify(sig);
2175 }
2176 
2177 void* os::user_handler() {
2178   return CAST_FROM_FN_PTR(void*, UserHandler);
2179 }
2180 
2181 extern "C" {
2182   typedef void (*sa_handler_t)(int);
2183   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2184 }
2185 
2186 void* os::signal(int signal_number, void* handler) {
2187   struct sigaction sigAct, oldSigAct;
2188 
2189   sigfillset(&(sigAct.sa_mask));
2190   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2191   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2192 
2193   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2194     // -1 means registration failed
2195     return (void *)-1;
2196   }
2197 
2198   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2199 }
2200 
2201 void os::signal_raise(int signal_number) {
2202   ::raise(signal_number);
2203 }
2204 
2205 /*
2206  * The following code is moved from os.cpp for making this
2207  * code platform specific, which it is by its very nature.
2208  */
2209 
2210 // Will be modified when max signal is changed to be dynamic
2211 int os::sigexitnum_pd() {
2212   return NSIG;
2213 }
2214 
2215 // a counter for each possible signal value
2216 static volatile jint pending_signals[NSIG+1] = { 0 };
2217 
2218 // Linux(POSIX) specific hand shaking semaphore.
2219 static sem_t sig_sem;
2220 
2221 void os::signal_init_pd() {
2222   // Initialize signal structures
2223   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2224 
2225   // Initialize signal semaphore
2226   ::sem_init(&sig_sem, 0, 0);
2227 }
2228 
2229 void os::signal_notify(int sig) {
2230   Atomic::inc(&pending_signals[sig]);
2231   ::sem_post(&sig_sem);
2232 }
2233 
2234 static int check_pending_signals(bool wait) {
2235   Atomic::store(0, &sigint_count);
2236   for (;;) {
2237     for (int i = 0; i < NSIG + 1; i++) {
2238       jint n = pending_signals[i];
2239       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2240         return i;
2241       }
2242     }
2243     if (!wait) {
2244       return -1;
2245     }
2246     JavaThread *thread = JavaThread::current();
2247     ThreadBlockInVM tbivm(thread);
2248 
2249     bool threadIsSuspended;
2250     do {
2251       thread->set_suspend_equivalent();
2252       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2253       ::sem_wait(&sig_sem);
2254 
2255       // were we externally suspended while we were waiting?
2256       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2257       if (threadIsSuspended) {
2258         //
2259         // The semaphore has been incremented, but while we were waiting
2260         // another thread suspended us. We don't want to continue running
2261         // while suspended because that would surprise the thread that
2262         // suspended us.
2263         //
2264         ::sem_post(&sig_sem);
2265 
2266         thread->java_suspend_self();
2267       }
2268     } while (threadIsSuspended);
2269   }
2270 }
2271 
2272 int os::signal_lookup() {
2273   return check_pending_signals(false);
2274 }
2275 
2276 int os::signal_wait() {
2277   return check_pending_signals(true);
2278 }
2279 
2280 ////////////////////////////////////////////////////////////////////////////////
2281 // Virtual Memory
2282 
2283 int os::vm_page_size() {
2284   // Seems redundant as all get out
2285   assert(os::Linux::page_size() != -1, "must call os::init");
2286   return os::Linux::page_size();
2287 }
2288 
2289 // Solaris allocates memory by pages.
2290 int os::vm_allocation_granularity() {
2291   assert(os::Linux::page_size() != -1, "must call os::init");
2292   return os::Linux::page_size();
2293 }
2294 
2295 // Rationale behind this function:
2296 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2297 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2298 //  samples for JITted code. Here we create private executable mapping over the code cache
2299 //  and then we can use standard (well, almost, as mapping can change) way to provide
2300 //  info for the reporting script by storing timestamp and location of symbol
2301 void linux_wrap_code(char* base, size_t size) {
2302   static volatile jint cnt = 0;
2303 
2304   if (!UseOprofile) {
2305     return;
2306   }
2307 
2308   char buf[PATH_MAX+1];
2309   int num = Atomic::add(1, &cnt);
2310 
2311   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2312            os::get_temp_directory(), os::current_process_id(), num);
2313   unlink(buf);
2314 
2315   int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2316 
2317   if (fd != -1) {
2318     off_t rv = lseek(fd, size-2, SEEK_SET);
2319     if (rv != (off_t)-1) {
2320       if (write(fd, "", 1) == 1) {
2321         mmap(base, size,
2322              PROT_READ|PROT_WRITE|PROT_EXEC,
2323              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2324       }
2325     }
2326     close(fd);
2327     unlink(buf);
2328   }
2329 }
2330 
2331 // NOTE: Linux kernel does not really reserve the pages for us.
2332 //       All it does is to check if there are enough free pages
2333 //       left at the time of mmap(). This could be a potential
2334 //       problem.
2335 bool os::commit_memory(char* addr, size_t size, bool exec) {
2336   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2337   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2338                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2339   return res != (uintptr_t) MAP_FAILED;
2340 }
2341 
2342 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2343                        bool exec) {
2344   return commit_memory(addr, size, exec);
2345 }
2346 
2347 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2348 
2349 void os::free_memory(char *addr, size_t bytes) {
2350   ::mmap(addr, bytes, PROT_READ | PROT_WRITE,
2351          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2352 }
2353 
2354 void os::numa_make_global(char *addr, size_t bytes) {
2355   Linux::numa_interleave_memory(addr, bytes);
2356 }
2357 
2358 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2359   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2360 }
2361 
2362 bool os::numa_topology_changed()   { return false; }
2363 
2364 size_t os::numa_get_groups_num() {
2365   int max_node = Linux::numa_max_node();
2366   return max_node > 0 ? max_node + 1 : 1;
2367 }
2368 
2369 int os::numa_get_group_id() {
2370   int cpu_id = Linux::sched_getcpu();
2371   if (cpu_id != -1) {
2372     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2373     if (lgrp_id != -1) {
2374       return lgrp_id;
2375     }
2376   }
2377   return 0;
2378 }
2379 
2380 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2381   for (size_t i = 0; i < size; i++) {
2382     ids[i] = i;
2383   }
2384   return size;
2385 }
2386 
2387 bool os::get_page_info(char *start, page_info* info) {
2388   return false;
2389 }
2390 
2391 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2392   return end;
2393 }
2394 
2395 extern "C" void numa_warn(int number, char *where, ...) { }
2396 extern "C" void numa_error(char *where) { }
2397 
2398 
2399 // If we are running with libnuma version > 2, then we should
2400 // be trying to use symbols with versions 1.1
2401 // If we are running with earlier version, which did not have symbol versions,
2402 // we should use the base version.
2403 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2404   void *f = dlvsym(handle, name, "libnuma_1.1");
2405   if (f == NULL) {
2406     f = dlsym(handle, name);
2407   }
2408   return f;
2409 }
2410 
2411 bool os::Linux::libnuma_init() {
2412   // sched_getcpu() should be in libc.
2413   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2414                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2415 
2416   if (sched_getcpu() != -1) { // Does it work?
2417     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2418     if (handle != NULL) {
2419       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2420                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2421       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2422                                        libnuma_dlsym(handle, "numa_max_node")));
2423       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2424                                         libnuma_dlsym(handle, "numa_available")));
2425       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2426                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2427       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2428                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2429 
2430 
2431       if (numa_available() != -1) {
2432         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2433         // Create a cpu -> node mapping
2434         _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2435         rebuild_cpu_to_node_map();
2436         return true;
2437       }
2438     }
2439   }
2440   return false;
2441 }
2442 
2443 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2444 // The table is later used in get_node_by_cpu().
2445 void os::Linux::rebuild_cpu_to_node_map() {
2446   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2447                               // in libnuma (possible values are starting from 16,
2448                               // and continuing up with every other power of 2, but less
2449                               // than the maximum number of CPUs supported by kernel), and
2450                               // is a subject to change (in libnuma version 2 the requirements
2451                               // are more reasonable) we'll just hardcode the number they use
2452                               // in the library.
2453   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2454 
2455   size_t cpu_num = os::active_processor_count();
2456   size_t cpu_map_size = NCPUS / BitsPerCLong;
2457   size_t cpu_map_valid_size =
2458     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2459 
2460   cpu_to_node()->clear();
2461   cpu_to_node()->at_grow(cpu_num - 1);
2462   size_t node_num = numa_get_groups_num();
2463 
2464   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2465   for (size_t i = 0; i < node_num; i++) {
2466     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2467       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2468         if (cpu_map[j] != 0) {
2469           for (size_t k = 0; k < BitsPerCLong; k++) {
2470             if (cpu_map[j] & (1UL << k)) {
2471               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2472             }
2473           }
2474         }
2475       }
2476     }
2477   }
2478   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2479 }
2480 
2481 int os::Linux::get_node_by_cpu(int cpu_id) {
2482   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2483     return cpu_to_node()->at(cpu_id);
2484   }
2485   return -1;
2486 }
2487 
2488 GrowableArray<int>* os::Linux::_cpu_to_node;
2489 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2490 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2491 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2492 os::Linux::numa_available_func_t os::Linux::_numa_available;
2493 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2494 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2495 unsigned long* os::Linux::_numa_all_nodes;
2496 
2497 bool os::uncommit_memory(char* addr, size_t size) {
2498   return ::mmap(addr, size, PROT_NONE,
2499                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0)
2500     != MAP_FAILED;
2501 }
2502 
2503 // Linux uses a growable mapping for the stack, and if the mapping for
2504 // the stack guard pages is not removed when we detach a thread the
2505 // stack cannot grow beyond the pages where the stack guard was
2506 // mapped.  If at some point later in the process the stack expands to
2507 // that point, the Linux kernel cannot expand the stack any further
2508 // because the guard pages are in the way, and a segfault occurs.
2509 //
2510 // However, it's essential not to split the stack region by unmapping
2511 // a region (leaving a hole) that's already part of the stack mapping,
2512 // so if the stack mapping has already grown beyond the guard pages at
2513 // the time we create them, we have to truncate the stack mapping.
2514 // So, we need to know the extent of the stack mapping when
2515 // create_stack_guard_pages() is called.
2516 
2517 // Find the bounds of the stack mapping.  Return true for success.
2518 //
2519 // We only need this for stacks that are growable: at the time of
2520 // writing thread stacks don't use growable mappings (i.e. those
2521 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2522 // only applies to the main thread.
2523 static bool
2524 get_stack_bounds(uintptr_t *bottom, uintptr_t *top)
2525 {
2526   FILE *f = fopen("/proc/self/maps", "r");
2527   if (f == NULL)
2528     return false;
2529 
2530   while (!feof(f)) {
2531     size_t dummy;
2532     char *str = NULL;
2533     ssize_t len = getline(&str, &dummy, f);
2534     if (len == -1) {
2535       fclose(f);
2536       return false;
2537     }
2538 
2539     if (len > 0 && str[len-1] == '\n') {
2540       str[len-1] = 0;
2541       len--;
2542     }
2543 
2544     static const char *stack_str = "[stack]";
2545     if (len > (ssize_t)strlen(stack_str)
2546        && (strcmp(str + len - strlen(stack_str), stack_str) == 0)) {
2547       if (sscanf(str, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2548         uintptr_t sp = (uintptr_t)__builtin_frame_address(0);
2549         if (sp >= *bottom && sp <= *top) {
2550           free(str);
2551           fclose(f);
2552           return true;
2553         }
2554       }
2555     }
2556     free(str);
2557   }
2558   fclose(f);
2559   return false;
2560 }
2561 
2562 // If the (growable) stack mapping already extends beyond the point
2563 // where we're going to put our guard pages, truncate the mapping at
2564 // that point by munmap()ping it.  This ensures that when we later
2565 // munmap() the guard pages we don't leave a hole in the stack
2566 // mapping.
2567 bool os::create_stack_guard_pages(char* addr, size_t size) {
2568   uintptr_t stack_extent, stack_base;
2569   if (get_stack_bounds(&stack_extent, &stack_base)) {
2570     if (stack_extent < (uintptr_t)addr)
2571       ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2572   }
2573 
2574   return os::commit_memory(addr, size);
2575 }
2576 
2577 // If this is a growable mapping, remove the guard pages entirely by
2578 // munmap()ping them.  If not, just call uncommit_memory().
2579 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2580   uintptr_t stack_extent, stack_base;
2581   if (get_stack_bounds(&stack_extent, &stack_base)) {
2582     return ::munmap(addr, size) == 0;
2583   }
2584 
2585   return os::uncommit_memory(addr, size);
2586 }
2587 
2588 static address _highest_vm_reserved_address = NULL;
2589 
2590 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2591 // at 'requested_addr'. If there are existing memory mappings at the same
2592 // location, however, they will be overwritten. If 'fixed' is false,
2593 // 'requested_addr' is only treated as a hint, the return value may or
2594 // may not start from the requested address. Unlike Linux mmap(), this
2595 // function returns NULL to indicate failure.
2596 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2597   char * addr;
2598   int flags;
2599 
2600   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2601   if (fixed) {
2602     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2603     flags |= MAP_FIXED;
2604   }
2605 
2606   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2607   // to PROT_EXEC if executable when we commit the page.
2608   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2609                        flags, -1, 0);
2610 
2611   if (addr != MAP_FAILED) {
2612     // anon_mmap() should only get called during VM initialization,
2613     // don't need lock (actually we can skip locking even it can be called
2614     // from multiple threads, because _highest_vm_reserved_address is just a
2615     // hint about the upper limit of non-stack memory regions.)
2616     if ((address)addr + bytes > _highest_vm_reserved_address) {
2617       _highest_vm_reserved_address = (address)addr + bytes;
2618     }
2619   }
2620 
2621   return addr == MAP_FAILED ? NULL : addr;
2622 }
2623 
2624 // Don't update _highest_vm_reserved_address, because there might be memory
2625 // regions above addr + size. If so, releasing a memory region only creates
2626 // a hole in the address space, it doesn't help prevent heap-stack collision.
2627 //
2628 static int anon_munmap(char * addr, size_t size) {
2629   return ::munmap(addr, size) == 0;
2630 }
2631 
2632 char* os::reserve_memory(size_t bytes, char* requested_addr,
2633                          size_t alignment_hint) {
2634   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2635 }
2636 
2637 bool os::release_memory(char* addr, size_t size) {
2638   return anon_munmap(addr, size);
2639 }
2640 
2641 static address highest_vm_reserved_address() {
2642   return _highest_vm_reserved_address;
2643 }
2644 
2645 static bool linux_mprotect(char* addr, size_t size, int prot) {
2646   // Linux wants the mprotect address argument to be page aligned.
2647   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2648 
2649   // According to SUSv3, mprotect() should only be used with mappings
2650   // established by mmap(), and mmap() always maps whole pages. Unaligned
2651   // 'addr' likely indicates problem in the VM (e.g. trying to change
2652   // protection of malloc'ed or statically allocated memory). Check the
2653   // caller if you hit this assert.
2654   assert(addr == bottom, "sanity check");
2655 
2656   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2657   return ::mprotect(bottom, size, prot) == 0;
2658 }
2659 
2660 // Set protections specified
2661 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2662                         bool is_committed) {
2663   unsigned int p = 0;
2664   switch (prot) {
2665   case MEM_PROT_NONE: p = PROT_NONE; break;
2666   case MEM_PROT_READ: p = PROT_READ; break;
2667   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2668   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2669   default:
2670     ShouldNotReachHere();
2671   }
2672   // is_committed is unused.
2673   return linux_mprotect(addr, bytes, p);
2674 }
2675 
2676 bool os::guard_memory(char* addr, size_t size) {
2677   return linux_mprotect(addr, size, PROT_NONE);
2678 }
2679 
2680 bool os::unguard_memory(char* addr, size_t size) {
2681   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2682 }
2683 
2684 // Large page support
2685 
2686 static size_t _large_page_size = 0;
2687 
2688 bool os::large_page_init() {
2689   if (!UseLargePages) return false;
2690 
2691   if (LargePageSizeInBytes) {
2692     _large_page_size = LargePageSizeInBytes;
2693   } else {
2694     // large_page_size on Linux is used to round up heap size. x86 uses either
2695     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2696     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2697     // page as large as 256M.
2698     //
2699     // Here we try to figure out page size by parsing /proc/meminfo and looking
2700     // for a line with the following format:
2701     //    Hugepagesize:     2048 kB
2702     //
2703     // If we can't determine the value (e.g. /proc is not mounted, or the text
2704     // format has been changed), we'll use the largest page size supported by
2705     // the processor.
2706 
2707 #ifndef ZERO
2708     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M);
2709 #endif // ZERO
2710 
2711     FILE *fp = fopen("/proc/meminfo", "r");
2712     if (fp) {
2713       while (!feof(fp)) {
2714         int x = 0;
2715         char buf[16];
2716         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2717           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2718             _large_page_size = x * K;
2719             break;
2720           }
2721         } else {
2722           // skip to next line
2723           for (;;) {
2724             int ch = fgetc(fp);
2725             if (ch == EOF || ch == (int)'\n') break;
2726           }
2727         }
2728       }
2729       fclose(fp);
2730     }
2731   }
2732 
2733   const size_t default_page_size = (size_t)Linux::page_size();
2734   if (_large_page_size > default_page_size) {
2735     _page_sizes[0] = _large_page_size;
2736     _page_sizes[1] = default_page_size;
2737     _page_sizes[2] = 0;
2738   }
2739 
2740   // Large page support is available on 2.6 or newer kernel, some vendors
2741   // (e.g. Redhat) have backported it to their 2.4 based distributions.
2742   // We optimistically assume the support is available. If later it turns out
2743   // not true, VM will automatically switch to use regular page size.
2744   return true;
2745 }
2746 
2747 #ifndef SHM_HUGETLB
2748 #define SHM_HUGETLB 04000
2749 #endif
2750 
2751 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
2752   // "exec" is passed in but not used.  Creating the shared image for
2753   // the code cache doesn't have an SHM_X executable permission to check.
2754   assert(UseLargePages, "only for large pages");
2755 
2756   key_t key = IPC_PRIVATE;
2757   char *addr;
2758 
2759   bool warn_on_failure = UseLargePages &&
2760                         (!FLAG_IS_DEFAULT(UseLargePages) ||
2761                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2762                         );
2763   char msg[128];
2764 
2765   // Create a large shared memory region to attach to based on size.
2766   // Currently, size is the total size of the heap
2767   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2768   if (shmid == -1) {
2769      // Possible reasons for shmget failure:
2770      // 1. shmmax is too small for Java heap.
2771      //    > check shmmax value: cat /proc/sys/kernel/shmmax
2772      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2773      // 2. not enough large page memory.
2774      //    > check available large pages: cat /proc/meminfo
2775      //    > increase amount of large pages:
2776      //          echo new_value > /proc/sys/vm/nr_hugepages
2777      //      Note 1: different Linux may use different name for this property,
2778      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
2779      //      Note 2: it's possible there's enough physical memory available but
2780      //            they are so fragmented after a long run that they can't
2781      //            coalesce into large pages. Try to reserve large pages when
2782      //            the system is still "fresh".
2783      if (warn_on_failure) {
2784        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2785        warning(msg);
2786      }
2787      return NULL;
2788   }
2789 
2790   // attach to the region
2791   addr = (char*)shmat(shmid, req_addr, 0);
2792   int err = errno;
2793 
2794   // Remove shmid. If shmat() is successful, the actual shared memory segment
2795   // will be deleted when it's detached by shmdt() or when the process
2796   // terminates. If shmat() is not successful this will remove the shared
2797   // segment immediately.
2798   shmctl(shmid, IPC_RMID, NULL);
2799 
2800   if ((intptr_t)addr == -1) {
2801      if (warn_on_failure) {
2802        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2803        warning(msg);
2804      }
2805      return NULL;
2806   }
2807 
2808   return addr;
2809 }
2810 
2811 bool os::release_memory_special(char* base, size_t bytes) {
2812   // detaching the SHM segment will also delete it, see reserve_memory_special()
2813   int rslt = shmdt(base);
2814   return rslt == 0;
2815 }
2816 
2817 size_t os::large_page_size() {
2818   return _large_page_size;
2819 }
2820 
2821 // Linux does not support anonymous mmap with large page memory. The only way
2822 // to reserve large page memory without file backing is through SysV shared
2823 // memory API. The entire memory region is committed and pinned upfront.
2824 // Hopefully this will change in the future...
2825 bool os::can_commit_large_page_memory() {
2826   return false;
2827 }
2828 
2829 bool os::can_execute_large_page_memory() {
2830   return false;
2831 }
2832 
2833 // Reserve memory at an arbitrary address, only if that area is
2834 // available (and not reserved for something else).
2835 
2836 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2837   const int max_tries = 10;
2838   char* base[max_tries];
2839   size_t size[max_tries];
2840   const size_t gap = 0x000000;
2841 
2842   // Assert only that the size is a multiple of the page size, since
2843   // that's all that mmap requires, and since that's all we really know
2844   // about at this low abstraction level.  If we need higher alignment,
2845   // we can either pass an alignment to this method or verify alignment
2846   // in one of the methods further up the call chain.  See bug 5044738.
2847   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2848 
2849   // Repeatedly allocate blocks until the block is allocated at the
2850   // right spot. Give up after max_tries. Note that reserve_memory() will
2851   // automatically update _highest_vm_reserved_address if the call is
2852   // successful. The variable tracks the highest memory address every reserved
2853   // by JVM. It is used to detect heap-stack collision if running with
2854   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2855   // space than needed, it could confuse the collision detecting code. To
2856   // solve the problem, save current _highest_vm_reserved_address and
2857   // calculate the correct value before return.
2858   address old_highest = _highest_vm_reserved_address;
2859 
2860   // Linux mmap allows caller to pass an address as hint; give it a try first,
2861   // if kernel honors the hint then we can return immediately.
2862   char * addr = anon_mmap(requested_addr, bytes, false);
2863   if (addr == requested_addr) {
2864      return requested_addr;
2865   }
2866 
2867   if (addr != NULL) {
2868      // mmap() is successful but it fails to reserve at the requested address
2869      anon_munmap(addr, bytes);
2870   }
2871 
2872   int i;
2873   for (i = 0; i < max_tries; ++i) {
2874     base[i] = reserve_memory(bytes);
2875 
2876     if (base[i] != NULL) {
2877       // Is this the block we wanted?
2878       if (base[i] == requested_addr) {
2879         size[i] = bytes;
2880         break;
2881       }
2882 
2883       // Does this overlap the block we wanted? Give back the overlapped
2884       // parts and try again.
2885 
2886       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2887       if (top_overlap >= 0 && top_overlap < bytes) {
2888         unmap_memory(base[i], top_overlap);
2889         base[i] += top_overlap;
2890         size[i] = bytes - top_overlap;
2891       } else {
2892         size_t bottom_overlap = base[i] + bytes - requested_addr;
2893         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2894           unmap_memory(requested_addr, bottom_overlap);
2895           size[i] = bytes - bottom_overlap;
2896         } else {
2897           size[i] = bytes;
2898         }
2899       }
2900     }
2901   }
2902 
2903   // Give back the unused reserved pieces.
2904 
2905   for (int j = 0; j < i; ++j) {
2906     if (base[j] != NULL) {
2907       unmap_memory(base[j], size[j]);
2908     }
2909   }
2910 
2911   if (i < max_tries) {
2912     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2913     return requested_addr;
2914   } else {
2915     _highest_vm_reserved_address = old_highest;
2916     return NULL;
2917   }
2918 }
2919 
2920 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2921   return ::read(fd, buf, nBytes);
2922 }
2923 
2924 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
2925 // Solaris uses poll(), linux uses park().
2926 // Poll() is likely a better choice, assuming that Thread.interrupt()
2927 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2928 // SIGSEGV, see 4355769.
2929 
2930 const int NANOSECS_PER_MILLISECS = 1000000;
2931 
2932 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2933   assert(thread == Thread::current(),  "thread consistency check");
2934 
2935   ParkEvent * const slp = thread->_SleepEvent ;
2936   slp->reset() ;
2937   OrderAccess::fence() ;
2938 
2939   if (interruptible) {
2940     jlong prevtime = javaTimeNanos();
2941 
2942     for (;;) {
2943       if (os::is_interrupted(thread, true)) {
2944         return OS_INTRPT;
2945       }
2946 
2947       jlong newtime = javaTimeNanos();
2948 
2949       if (newtime - prevtime < 0) {
2950         // time moving backwards, should only happen if no monotonic clock
2951         // not a guarantee() because JVM should not abort on kernel/glibc bugs
2952         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2953       } else {
2954         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2955       }
2956 
2957       if(millis <= 0) {
2958         return OS_OK;
2959       }
2960 
2961       prevtime = newtime;
2962 
2963       {
2964         assert(thread->is_Java_thread(), "sanity check");
2965         JavaThread *jt = (JavaThread *) thread;
2966         ThreadBlockInVM tbivm(jt);
2967         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
2968 
2969         jt->set_suspend_equivalent();
2970         // cleared by handle_special_suspend_equivalent_condition() or
2971         // java_suspend_self() via check_and_wait_while_suspended()
2972 
2973         slp->park(millis);
2974 
2975         // were we externally suspended while we were waiting?
2976         jt->check_and_wait_while_suspended();
2977       }
2978     }
2979   } else {
2980     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2981     jlong prevtime = javaTimeNanos();
2982 
2983     for (;;) {
2984       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
2985       // the 1st iteration ...
2986       jlong newtime = javaTimeNanos();
2987 
2988       if (newtime - prevtime < 0) {
2989         // time moving backwards, should only happen if no monotonic clock
2990         // not a guarantee() because JVM should not abort on kernel/glibc bugs
2991         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2992       } else {
2993         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2994       }
2995 
2996       if(millis <= 0) break ;
2997 
2998       prevtime = newtime;
2999       slp->park(millis);
3000     }
3001     return OS_OK ;
3002   }
3003 }
3004 
3005 int os::naked_sleep() {
3006   // %% make the sleep time an integer flag. for now use 1 millisec.
3007   return os::sleep(Thread::current(), 1, false);
3008 }
3009 
3010 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3011 void os::infinite_sleep() {
3012   while (true) {    // sleep forever ...
3013     ::sleep(100);   // ... 100 seconds at a time
3014   }
3015 }
3016 
3017 // Used to convert frequent JVM_Yield() to nops
3018 bool os::dont_yield() {
3019   return DontYieldALot;
3020 }
3021 
3022 void os::yield() {
3023   sched_yield();
3024 }
3025 
3026 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3027 
3028 void os::yield_all(int attempts) {
3029   // Yields to all threads, including threads with lower priorities
3030   // Threads on Linux are all with same priority. The Solaris style
3031   // os::yield_all() with nanosleep(1ms) is not necessary.
3032   sched_yield();
3033 }
3034 
3035 // Called from the tight loops to possibly influence time-sharing heuristics
3036 void os::loop_breaker(int attempts) {
3037   os::yield_all(attempts);
3038 }
3039 
3040 ////////////////////////////////////////////////////////////////////////////////
3041 // thread priority support
3042 
3043 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3044 // only supports dynamic priority, static priority must be zero. For real-time
3045 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3046 // However, for large multi-threaded applications, SCHED_RR is not only slower
3047 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3048 // of 5 runs - Sep 2005).
3049 //
3050 // The following code actually changes the niceness of kernel-thread/LWP. It
3051 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3052 // not the entire user process, and user level threads are 1:1 mapped to kernel
3053 // threads. It has always been the case, but could change in the future. For
3054 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3055 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3056 
3057 int os::java_to_os_priority[MaxPriority + 1] = {
3058   19,              // 0 Entry should never be used
3059 
3060    4,              // 1 MinPriority
3061    3,              // 2
3062    2,              // 3
3063 
3064    1,              // 4
3065    0,              // 5 NormPriority
3066   -1,              // 6
3067 
3068   -2,              // 7
3069   -3,              // 8
3070   -4,              // 9 NearMaxPriority
3071 
3072   -5               // 10 MaxPriority
3073 };
3074 
3075 static int prio_init() {
3076   if (ThreadPriorityPolicy == 1) {
3077     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3078     // if effective uid is not root. Perhaps, a more elegant way of doing
3079     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3080     if (geteuid() != 0) {
3081       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3082         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3083       }
3084       ThreadPriorityPolicy = 0;
3085     }
3086   }
3087   return 0;
3088 }
3089 
3090 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3091   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3092 
3093   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3094   return (ret == 0) ? OS_OK : OS_ERR;
3095 }
3096 
3097 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3098   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3099     *priority_ptr = java_to_os_priority[NormPriority];
3100     return OS_OK;
3101   }
3102 
3103   errno = 0;
3104   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3105   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3106 }
3107 
3108 // Hint to the underlying OS that a task switch would not be good.
3109 // Void return because it's a hint and can fail.
3110 void os::hint_no_preempt() {}
3111 
3112 ////////////////////////////////////////////////////////////////////////////////
3113 // suspend/resume support
3114 
3115 //  the low-level signal-based suspend/resume support is a remnant from the
3116 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3117 //  within hotspot. Now there is a single use-case for this:
3118 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3119 //      that runs in the watcher thread.
3120 //  The remaining code is greatly simplified from the more general suspension
3121 //  code that used to be used.
3122 //
3123 //  The protocol is quite simple:
3124 //  - suspend:
3125 //      - sends a signal to the target thread
3126 //      - polls the suspend state of the osthread using a yield loop
3127 //      - target thread signal handler (SR_handler) sets suspend state
3128 //        and blocks in sigsuspend until continued
3129 //  - resume:
3130 //      - sets target osthread state to continue
3131 //      - sends signal to end the sigsuspend loop in the SR_handler
3132 //
3133 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3134 //
3135 
3136 static void resume_clear_context(OSThread *osthread) {
3137   osthread->set_ucontext(NULL);
3138   osthread->set_siginfo(NULL);
3139 
3140   // notify the suspend action is completed, we have now resumed
3141   osthread->sr.clear_suspended();
3142 }
3143 
3144 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3145   osthread->set_ucontext(context);
3146   osthread->set_siginfo(siginfo);
3147 }
3148 
3149 //
3150 // Handler function invoked when a thread's execution is suspended or
3151 // resumed. We have to be careful that only async-safe functions are
3152 // called here (Note: most pthread functions are not async safe and
3153 // should be avoided.)
3154 //
3155 // Note: sigwait() is a more natural fit than sigsuspend() from an
3156 // interface point of view, but sigwait() prevents the signal hander
3157 // from being run. libpthread would get very confused by not having
3158 // its signal handlers run and prevents sigwait()'s use with the
3159 // mutex granting granting signal.
3160 //
3161 // Currently only ever called on the VMThread
3162 //
3163 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3164   // Save and restore errno to avoid confusing native code with EINTR
3165   // after sigsuspend.
3166   int old_errno = errno;
3167 
3168   Thread* thread = Thread::current();
3169   OSThread* osthread = thread->osthread();
3170   assert(thread->is_VM_thread(), "Must be VMThread");
3171   // read current suspend action
3172   int action = osthread->sr.suspend_action();
3173   if (action == SR_SUSPEND) {
3174     suspend_save_context(osthread, siginfo, context);
3175 
3176     // Notify the suspend action is about to be completed. do_suspend()
3177     // waits until SR_SUSPENDED is set and then returns. We will wait
3178     // here for a resume signal and that completes the suspend-other
3179     // action. do_suspend/do_resume is always called as a pair from
3180     // the same thread - so there are no races
3181 
3182     // notify the caller
3183     osthread->sr.set_suspended();
3184 
3185     sigset_t suspend_set;  // signals for sigsuspend()
3186 
3187     // get current set of blocked signals and unblock resume signal
3188     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3189     sigdelset(&suspend_set, SR_signum);
3190 
3191     // wait here until we are resumed
3192     do {
3193       sigsuspend(&suspend_set);
3194       // ignore all returns until we get a resume signal
3195     } while (osthread->sr.suspend_action() != SR_CONTINUE);
3196 
3197     resume_clear_context(osthread);
3198 
3199   } else {
3200     assert(action == SR_CONTINUE, "unexpected sr action");
3201     // nothing special to do - just leave the handler
3202   }
3203 
3204   errno = old_errno;
3205 }
3206 
3207 
3208 static int SR_initialize() {
3209   struct sigaction act;
3210   char *s;
3211   /* Get signal number to use for suspend/resume */
3212   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3213     int sig = ::strtol(s, 0, 10);
3214     if (sig > 0 || sig < _NSIG) {
3215         SR_signum = sig;
3216     }
3217   }
3218 
3219   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3220         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3221 
3222   sigemptyset(&SR_sigset);
3223   sigaddset(&SR_sigset, SR_signum);
3224 
3225   /* Set up signal handler for suspend/resume */
3226   act.sa_flags = SA_RESTART|SA_SIGINFO;
3227   act.sa_handler = (void (*)(int)) SR_handler;
3228 
3229   // SR_signum is blocked by default.
3230   // 4528190 - We also need to block pthread restart signal (32 on all
3231   // supported Linux platforms). Note that LinuxThreads need to block
3232   // this signal for all threads to work properly. So we don't have
3233   // to use hard-coded signal number when setting up the mask.
3234   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3235 
3236   if (sigaction(SR_signum, &act, 0) == -1) {
3237     return -1;
3238   }
3239 
3240   // Save signal flag
3241   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3242   return 0;
3243 }
3244 
3245 static int SR_finalize() {
3246   return 0;
3247 }
3248 
3249 
3250 // returns true on success and false on error - really an error is fatal
3251 // but this seems the normal response to library errors
3252 static bool do_suspend(OSThread* osthread) {
3253   // mark as suspended and send signal
3254   osthread->sr.set_suspend_action(SR_SUSPEND);
3255   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3256   assert_status(status == 0, status, "pthread_kill");
3257 
3258   // check status and wait until notified of suspension
3259   if (status == 0) {
3260     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3261       os::yield_all(i);
3262     }
3263     osthread->sr.set_suspend_action(SR_NONE);
3264     return true;
3265   }
3266   else {
3267     osthread->sr.set_suspend_action(SR_NONE);
3268     return false;
3269   }
3270 }
3271 
3272 static void do_resume(OSThread* osthread) {
3273   assert(osthread->sr.is_suspended(), "thread should be suspended");
3274   osthread->sr.set_suspend_action(SR_CONTINUE);
3275 
3276   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3277   assert_status(status == 0, status, "pthread_kill");
3278   // check status and wait unit notified of resumption
3279   if (status == 0) {
3280     for (int i = 0; osthread->sr.is_suspended(); i++) {
3281       os::yield_all(i);
3282     }
3283   }
3284   osthread->sr.set_suspend_action(SR_NONE);
3285 }
3286 
3287 ////////////////////////////////////////////////////////////////////////////////
3288 // interrupt support
3289 
3290 void os::interrupt(Thread* thread) {
3291   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3292     "possibility of dangling Thread pointer");
3293 
3294   OSThread* osthread = thread->osthread();
3295 
3296   if (!osthread->interrupted()) {
3297     osthread->set_interrupted(true);
3298     // More than one thread can get here with the same value of osthread,
3299     // resulting in multiple notifications.  We do, however, want the store
3300     // to interrupted() to be visible to other threads before we execute unpark().
3301     OrderAccess::fence();
3302     ParkEvent * const slp = thread->_SleepEvent ;
3303     if (slp != NULL) slp->unpark() ;
3304   }
3305 
3306   // For JSR166. Unpark even if interrupt status already was set
3307   if (thread->is_Java_thread())
3308     ((JavaThread*)thread)->parker()->unpark();
3309 
3310   ParkEvent * ev = thread->_ParkEvent ;
3311   if (ev != NULL) ev->unpark() ;
3312 
3313 }
3314 
3315 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3316   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3317     "possibility of dangling Thread pointer");
3318 
3319   OSThread* osthread = thread->osthread();
3320 
3321   bool interrupted = osthread->interrupted();
3322 
3323   if (interrupted && clear_interrupted) {
3324     osthread->set_interrupted(false);
3325     // consider thread->_SleepEvent->reset() ... optional optimization
3326   }
3327 
3328   return interrupted;
3329 }
3330 
3331 ///////////////////////////////////////////////////////////////////////////////////
3332 // signal handling (except suspend/resume)
3333 
3334 // This routine may be used by user applications as a "hook" to catch signals.
3335 // The user-defined signal handler must pass unrecognized signals to this
3336 // routine, and if it returns true (non-zero), then the signal handler must
3337 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3338 // routine will never retun false (zero), but instead will execute a VM panic
3339 // routine kill the process.
3340 //
3341 // If this routine returns false, it is OK to call it again.  This allows
3342 // the user-defined signal handler to perform checks either before or after
3343 // the VM performs its own checks.  Naturally, the user code would be making
3344 // a serious error if it tried to handle an exception (such as a null check
3345 // or breakpoint) that the VM was generating for its own correct operation.
3346 //
3347 // This routine may recognize any of the following kinds of signals:
3348 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3349 // It should be consulted by handlers for any of those signals.
3350 //
3351 // The caller of this routine must pass in the three arguments supplied
3352 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3353 // field of the structure passed to sigaction().  This routine assumes that
3354 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3355 //
3356 // Note that the VM will print warnings if it detects conflicting signal
3357 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3358 //
3359 extern "C" int
3360 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3361                         void* ucontext, int abort_if_unrecognized);
3362 
3363 void signalHandler(int sig, siginfo_t* info, void* uc) {
3364   assert(info != NULL && uc != NULL, "it must be old kernel");
3365   JVM_handle_linux_signal(sig, info, uc, true);
3366 }
3367 
3368 
3369 // This boolean allows users to forward their own non-matching signals
3370 // to JVM_handle_linux_signal, harmlessly.
3371 bool os::Linux::signal_handlers_are_installed = false;
3372 
3373 // For signal-chaining
3374 struct sigaction os::Linux::sigact[MAXSIGNUM];
3375 unsigned int os::Linux::sigs = 0;
3376 bool os::Linux::libjsig_is_loaded = false;
3377 typedef struct sigaction *(*get_signal_t)(int);
3378 get_signal_t os::Linux::get_signal_action = NULL;
3379 
3380 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3381   struct sigaction *actp = NULL;
3382 
3383   if (libjsig_is_loaded) {
3384     // Retrieve the old signal handler from libjsig
3385     actp = (*get_signal_action)(sig);
3386   }
3387   if (actp == NULL) {
3388     // Retrieve the preinstalled signal handler from jvm
3389     actp = get_preinstalled_handler(sig);
3390   }
3391 
3392   return actp;
3393 }
3394 
3395 static bool call_chained_handler(struct sigaction *actp, int sig,
3396                                  siginfo_t *siginfo, void *context) {
3397   // Call the old signal handler
3398   if (actp->sa_handler == SIG_DFL) {
3399     // It's more reasonable to let jvm treat it as an unexpected exception
3400     // instead of taking the default action.
3401     return false;
3402   } else if (actp->sa_handler != SIG_IGN) {
3403     if ((actp->sa_flags & SA_NODEFER) == 0) {
3404       // automaticlly block the signal
3405       sigaddset(&(actp->sa_mask), sig);
3406     }
3407 
3408     sa_handler_t hand;
3409     sa_sigaction_t sa;
3410     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3411     // retrieve the chained handler
3412     if (siginfo_flag_set) {
3413       sa = actp->sa_sigaction;
3414     } else {
3415       hand = actp->sa_handler;
3416     }
3417 
3418     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3419       actp->sa_handler = SIG_DFL;
3420     }
3421 
3422     // try to honor the signal mask
3423     sigset_t oset;
3424     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3425 
3426     // call into the chained handler
3427     if (siginfo_flag_set) {
3428       (*sa)(sig, siginfo, context);
3429     } else {
3430       (*hand)(sig);
3431     }
3432 
3433     // restore the signal mask
3434     pthread_sigmask(SIG_SETMASK, &oset, 0);
3435   }
3436   // Tell jvm's signal handler the signal is taken care of.
3437   return true;
3438 }
3439 
3440 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3441   bool chained = false;
3442   // signal-chaining
3443   if (UseSignalChaining) {
3444     struct sigaction *actp = get_chained_signal_action(sig);
3445     if (actp != NULL) {
3446       chained = call_chained_handler(actp, sig, siginfo, context);
3447     }
3448   }
3449   return chained;
3450 }
3451 
3452 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3453   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3454     return &sigact[sig];
3455   }
3456   return NULL;
3457 }
3458 
3459 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3460   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3461   sigact[sig] = oldAct;
3462   sigs |= (unsigned int)1 << sig;
3463 }
3464 
3465 // for diagnostic
3466 int os::Linux::sigflags[MAXSIGNUM];
3467 
3468 int os::Linux::get_our_sigflags(int sig) {
3469   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3470   return sigflags[sig];
3471 }
3472 
3473 void os::Linux::set_our_sigflags(int sig, int flags) {
3474   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3475   sigflags[sig] = flags;
3476 }
3477 
3478 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3479   // Check for overwrite.
3480   struct sigaction oldAct;
3481   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3482 
3483   void* oldhand = oldAct.sa_sigaction
3484                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3485                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3486   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3487       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3488       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3489     if (AllowUserSignalHandlers || !set_installed) {
3490       // Do not overwrite; user takes responsibility to forward to us.
3491       return;
3492     } else if (UseSignalChaining) {
3493       // save the old handler in jvm
3494       save_preinstalled_handler(sig, oldAct);
3495       // libjsig also interposes the sigaction() call below and saves the
3496       // old sigaction on it own.
3497     } else {
3498       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3499                     "%#lx for signal %d.", (long)oldhand, sig));
3500     }
3501   }
3502 
3503   struct sigaction sigAct;
3504   sigfillset(&(sigAct.sa_mask));
3505   sigAct.sa_handler = SIG_DFL;
3506   if (!set_installed) {
3507     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3508   } else {
3509     sigAct.sa_sigaction = signalHandler;
3510     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3511   }
3512   // Save flags, which are set by ours
3513   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3514   sigflags[sig] = sigAct.sa_flags;
3515 
3516   int ret = sigaction(sig, &sigAct, &oldAct);
3517   assert(ret == 0, "check");
3518 
3519   void* oldhand2  = oldAct.sa_sigaction
3520                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3521                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3522   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3523 }
3524 
3525 // install signal handlers for signals that HotSpot needs to
3526 // handle in order to support Java-level exception handling.
3527 
3528 void os::Linux::install_signal_handlers() {
3529   if (!signal_handlers_are_installed) {
3530     signal_handlers_are_installed = true;
3531 
3532     // signal-chaining
3533     typedef void (*signal_setting_t)();
3534     signal_setting_t begin_signal_setting = NULL;
3535     signal_setting_t end_signal_setting = NULL;
3536     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3537                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3538     if (begin_signal_setting != NULL) {
3539       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3540                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3541       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3542                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3543       libjsig_is_loaded = true;
3544       assert(UseSignalChaining, "should enable signal-chaining");
3545     }
3546     if (libjsig_is_loaded) {
3547       // Tell libjsig jvm is setting signal handlers
3548       (*begin_signal_setting)();
3549     }
3550 
3551     set_signal_handler(SIGSEGV, true);
3552     set_signal_handler(SIGPIPE, true);
3553     set_signal_handler(SIGBUS, true);
3554     set_signal_handler(SIGILL, true);
3555     set_signal_handler(SIGFPE, true);
3556     set_signal_handler(SIGXFSZ, true);
3557 
3558     if (libjsig_is_loaded) {
3559       // Tell libjsig jvm finishes setting signal handlers
3560       (*end_signal_setting)();
3561     }
3562 
3563     // We don't activate signal checker if libjsig is in place, we trust ourselves
3564     // and if UserSignalHandler is installed all bets are off
3565     if (CheckJNICalls) {
3566       if (libjsig_is_loaded) {
3567         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3568         check_signals = false;
3569       }
3570       if (AllowUserSignalHandlers) {
3571         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3572         check_signals = false;
3573       }
3574     }
3575   }
3576 }
3577 
3578 // This is the fastest way to get thread cpu time on Linux.
3579 // Returns cpu time (user+sys) for any thread, not only for current.
3580 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3581 // It might work on 2.6.10+ with a special kernel/glibc patch.
3582 // For reference, please, see IEEE Std 1003.1-2004:
3583 //   http://www.unix.org/single_unix_specification
3584 
3585 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3586   struct timespec tp;
3587   int rc = os::Linux::clock_gettime(clockid, &tp);
3588   assert(rc == 0, "clock_gettime is expected to return 0 code");
3589 
3590   return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3591 }
3592 
3593 /////
3594 // glibc on Linux platform uses non-documented flag
3595 // to indicate, that some special sort of signal
3596 // trampoline is used.
3597 // We will never set this flag, and we should
3598 // ignore this flag in our diagnostic
3599 #ifdef SIGNIFICANT_SIGNAL_MASK
3600 #undef SIGNIFICANT_SIGNAL_MASK
3601 #endif
3602 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3603 
3604 static const char* get_signal_handler_name(address handler,
3605                                            char* buf, int buflen) {
3606   int offset;
3607   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3608   if (found) {
3609     // skip directory names
3610     const char *p1, *p2;
3611     p1 = buf;
3612     size_t len = strlen(os::file_separator());
3613     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3614     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3615   } else {
3616     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3617   }
3618   return buf;
3619 }
3620 
3621 static void print_signal_handler(outputStream* st, int sig,
3622                                  char* buf, size_t buflen) {
3623   struct sigaction sa;
3624 
3625   sigaction(sig, NULL, &sa);
3626 
3627   // See comment for SIGNIFICANT_SIGNAL_MASK define
3628   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3629 
3630   st->print("%s: ", os::exception_name(sig, buf, buflen));
3631 
3632   address handler = (sa.sa_flags & SA_SIGINFO)
3633     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3634     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3635 
3636   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3637     st->print("SIG_DFL");
3638   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3639     st->print("SIG_IGN");
3640   } else {
3641     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3642   }
3643 
3644   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3645 
3646   address rh = VMError::get_resetted_sighandler(sig);
3647   // May be, handler was resetted by VMError?
3648   if(rh != NULL) {
3649     handler = rh;
3650     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3651   }
3652 
3653   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
3654 
3655   // Check: is it our handler?
3656   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3657      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3658     // It is our signal handler
3659     // check for flags, reset system-used one!
3660     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3661       st->print(
3662                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3663                 os::Linux::get_our_sigflags(sig));
3664     }
3665   }
3666   st->cr();
3667 }
3668 
3669 
3670 #define DO_SIGNAL_CHECK(sig) \
3671   if (!sigismember(&check_signal_done, sig)) \
3672     os::Linux::check_signal_handler(sig)
3673 
3674 // This method is a periodic task to check for misbehaving JNI applications
3675 // under CheckJNI, we can add any periodic checks here
3676 
3677 void os::run_periodic_checks() {
3678 
3679   if (check_signals == false) return;
3680 
3681   // SEGV and BUS if overridden could potentially prevent
3682   // generation of hs*.log in the event of a crash, debugging
3683   // such a case can be very challenging, so we absolutely
3684   // check the following for a good measure:
3685   DO_SIGNAL_CHECK(SIGSEGV);
3686   DO_SIGNAL_CHECK(SIGILL);
3687   DO_SIGNAL_CHECK(SIGFPE);
3688   DO_SIGNAL_CHECK(SIGBUS);
3689   DO_SIGNAL_CHECK(SIGPIPE);
3690   DO_SIGNAL_CHECK(SIGXFSZ);
3691 
3692 
3693   // ReduceSignalUsage allows the user to override these handlers
3694   // see comments at the very top and jvm_solaris.h
3695   if (!ReduceSignalUsage) {
3696     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3697     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3698     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3699     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3700   }
3701 
3702   DO_SIGNAL_CHECK(SR_signum);
3703   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3704 }
3705 
3706 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3707 
3708 static os_sigaction_t os_sigaction = NULL;
3709 
3710 void os::Linux::check_signal_handler(int sig) {
3711   char buf[O_BUFLEN];
3712   address jvmHandler = NULL;
3713 
3714 
3715   struct sigaction act;
3716   if (os_sigaction == NULL) {
3717     // only trust the default sigaction, in case it has been interposed
3718     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3719     if (os_sigaction == NULL) return;
3720   }
3721 
3722   os_sigaction(sig, (struct sigaction*)NULL, &act);
3723 
3724 
3725   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3726 
3727   address thisHandler = (act.sa_flags & SA_SIGINFO)
3728     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3729     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3730 
3731 
3732   switch(sig) {
3733   case SIGSEGV:
3734   case SIGBUS:
3735   case SIGFPE:
3736   case SIGPIPE:
3737   case SIGILL:
3738   case SIGXFSZ:
3739     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3740     break;
3741 
3742   case SHUTDOWN1_SIGNAL:
3743   case SHUTDOWN2_SIGNAL:
3744   case SHUTDOWN3_SIGNAL:
3745   case BREAK_SIGNAL:
3746     jvmHandler = (address)user_handler();
3747     break;
3748 
3749   case INTERRUPT_SIGNAL:
3750     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3751     break;
3752 
3753   default:
3754     if (sig == SR_signum) {
3755       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3756     } else {
3757       return;
3758     }
3759     break;
3760   }
3761 
3762   if (thisHandler != jvmHandler) {
3763     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3764     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3765     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3766     // No need to check this sig any longer
3767     sigaddset(&check_signal_done, sig);
3768   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3769     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3770     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3771     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
3772     // No need to check this sig any longer
3773     sigaddset(&check_signal_done, sig);
3774   }
3775 
3776   // Dump all the signal
3777   if (sigismember(&check_signal_done, sig)) {
3778     print_signal_handlers(tty, buf, O_BUFLEN);
3779   }
3780 }
3781 
3782 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3783 
3784 extern bool signal_name(int signo, char* buf, size_t len);
3785 
3786 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3787   if (0 < exception_code && exception_code <= SIGRTMAX) {
3788     // signal
3789     if (!signal_name(exception_code, buf, size)) {
3790       jio_snprintf(buf, size, "SIG%d", exception_code);
3791     }
3792     return buf;
3793   } else {
3794     return NULL;
3795   }
3796 }
3797 
3798 // this is called _before_ the most of global arguments have been parsed
3799 void os::init(void) {
3800   char dummy;   /* used to get a guess on initial stack address */
3801 //  first_hrtime = gethrtime();
3802 
3803   // With LinuxThreads the JavaMain thread pid (primordial thread)
3804   // is different than the pid of the java launcher thread.
3805   // So, on Linux, the launcher thread pid is passed to the VM
3806   // via the sun.java.launcher.pid property.
3807   // Use this property instead of getpid() if it was correctly passed.
3808   // See bug 6351349.
3809   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3810 
3811   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3812 
3813   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3814 
3815   init_random(1234567);
3816 
3817   ThreadCritical::initialize();
3818 
3819   Linux::set_page_size(sysconf(_SC_PAGESIZE));
3820   if (Linux::page_size() == -1) {
3821     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
3822                   strerror(errno)));
3823   }
3824   init_page_sizes((size_t) Linux::page_size());
3825 
3826   Linux::initialize_system_info();
3827 
3828   // main_thread points to the aboriginal thread
3829   Linux::_main_thread = pthread_self();
3830 
3831   Linux::clock_init();
3832   initial_time_count = os::elapsed_counter();
3833   pthread_mutex_init(&dl_mutex, NULL);
3834 }
3835 
3836 // To install functions for atexit system call
3837 extern "C" {
3838   static void perfMemory_exit_helper() {
3839     perfMemory_exit();
3840   }
3841 }
3842 
3843 // this is called _after_ the global arguments have been parsed
3844 jint os::init_2(void)
3845 {
3846   Linux::fast_thread_clock_init();
3847 
3848   if (StackShadowPages <= 0) {
3849     tty->print_cr("\nThe -XX:StackShadowPages parameter must be set to positive value!");
3850     return JNI_ERR;
3851   }
3852 
3853   // Allocate a single page and mark it as readable for safepoint polling
3854   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3855   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3856 
3857   os::set_polling_page( polling_page );
3858 
3859 #ifndef PRODUCT
3860   if(Verbose && PrintMiscellaneous)
3861     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3862 #endif
3863 
3864   if (!UseMembar) {
3865     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3866     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3867     os::set_memory_serialize_page( mem_serialize_page );
3868 
3869 #ifndef PRODUCT
3870     if(Verbose && PrintMiscellaneous)
3871       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3872 #endif
3873   }
3874 
3875   FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3876 
3877   // initialize suspend/resume support - must do this before signal_sets_init()
3878   if (SR_initialize() != 0) {
3879     perror("SR_initialize failed");
3880     return JNI_ERR;
3881   }
3882 
3883   Linux::signal_sets_init();
3884   Linux::install_signal_handlers();
3885 
3886   size_t threadStackSizeInBytes = ThreadStackSize * K;
3887   if (threadStackSizeInBytes != 0 &&
3888       threadStackSizeInBytes < Linux::min_stack_allowed) {
3889         tty->print_cr("\nThe stack size specified is too small, "
3890                       "Specify at least %dk",
3891                       Linux::min_stack_allowed / K);
3892         return JNI_ERR;
3893   }
3894 
3895   // Make the stack size a multiple of the page size so that
3896   // the yellow/red zones can be guarded.
3897   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3898         vm_page_size()));
3899 
3900   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
3901 
3902   Linux::libpthread_init();
3903   if (PrintMiscellaneous && (Verbose || WizardMode)) {
3904      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
3905           Linux::glibc_version(), Linux::libpthread_version(),
3906           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
3907   }
3908 
3909   if (UseNUMA) {
3910     if (!Linux::libnuma_init()) {
3911       UseNUMA = false;
3912     } else {
3913       if ((Linux::numa_max_node() < 1)) {
3914         // There's only one node(they start from 0), disable NUMA.
3915         UseNUMA = false;
3916       }
3917     }
3918     if (!UseNUMA && ForceNUMA) {
3919       UseNUMA = true;
3920     }
3921   }
3922 
3923   if (MaxFDLimit) {
3924     // set the number of file descriptors to max. print out error
3925     // if getrlimit/setrlimit fails but continue regardless.
3926     struct rlimit nbr_files;
3927     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3928     if (status != 0) {
3929       if (PrintMiscellaneous && (Verbose || WizardMode))
3930         perror("os::init_2 getrlimit failed");
3931     } else {
3932       nbr_files.rlim_cur = nbr_files.rlim_max;
3933       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3934       if (status != 0) {
3935         if (PrintMiscellaneous && (Verbose || WizardMode))
3936           perror("os::init_2 setrlimit failed");
3937       }
3938     }
3939   }
3940 
3941   // Initialize lock used to serialize thread creation (see os::create_thread)
3942   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
3943 
3944   // Initialize HPI.
3945   jint hpi_result = hpi::initialize();
3946   if (hpi_result != JNI_OK) {
3947     tty->print_cr("There was an error trying to initialize the HPI library.");
3948     return hpi_result;
3949   }
3950 
3951   // at-exit methods are called in the reverse order of their registration.
3952   // atexit functions are called on return from main or as a result of a
3953   // call to exit(3C). There can be only 32 of these functions registered
3954   // and atexit() does not set errno.
3955 
3956   if (PerfAllowAtExitRegistration) {
3957     // only register atexit functions if PerfAllowAtExitRegistration is set.
3958     // atexit functions can be delayed until process exit time, which
3959     // can be problematic for embedded VM situations. Embedded VMs should
3960     // call DestroyJavaVM() to assure that VM resources are released.
3961 
3962     // note: perfMemory_exit_helper atexit function may be removed in
3963     // the future if the appropriate cleanup code can be added to the
3964     // VM_Exit VMOperation's doit method.
3965     if (atexit(perfMemory_exit_helper) != 0) {
3966       warning("os::init2 atexit(perfMemory_exit_helper) failed");
3967     }
3968   }
3969 
3970   // initialize thread priority policy
3971   prio_init();
3972 
3973   return JNI_OK;
3974 }
3975 
3976 // Mark the polling page as unreadable
3977 void os::make_polling_page_unreadable(void) {
3978   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
3979     fatal("Could not disable polling page");
3980 };
3981 
3982 // Mark the polling page as readable
3983 void os::make_polling_page_readable(void) {
3984   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
3985     fatal("Could not enable polling page");
3986   }
3987 };
3988 
3989 int os::active_processor_count() {
3990   // Linux doesn't yet have a (official) notion of processor sets,
3991   // so just return the number of online processors.
3992   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
3993   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
3994   return online_cpus;
3995 }
3996 
3997 bool os::distribute_processes(uint length, uint* distribution) {
3998   // Not yet implemented.
3999   return false;
4000 }
4001 
4002 bool os::bind_to_processor(uint processor_id) {
4003   // Not yet implemented.
4004   return false;
4005 }
4006 
4007 ///
4008 
4009 // Suspends the target using the signal mechanism and then grabs the PC before
4010 // resuming the target. Used by the flat-profiler only
4011 ExtendedPC os::get_thread_pc(Thread* thread) {
4012   // Make sure that it is called by the watcher for the VMThread
4013   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4014   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4015 
4016   ExtendedPC epc;
4017 
4018   OSThread* osthread = thread->osthread();
4019   if (do_suspend(osthread)) {
4020     if (osthread->ucontext() != NULL) {
4021       epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4022     } else {
4023       // NULL context is unexpected, double-check this is the VMThread
4024       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4025     }
4026     do_resume(osthread);
4027   }
4028   // failure means pthread_kill failed for some reason - arguably this is
4029   // a fatal problem, but such problems are ignored elsewhere
4030 
4031   return epc;
4032 }
4033 
4034 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4035 {
4036    if (is_NPTL()) {
4037       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4038    } else {
4039 #ifndef IA64
4040       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4041       // word back to default 64bit precision if condvar is signaled. Java
4042       // wants 53bit precision.  Save and restore current value.
4043       int fpu = get_fpu_control_word();
4044 #endif // IA64
4045       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4046 #ifndef IA64
4047       set_fpu_control_word(fpu);
4048 #endif // IA64
4049       return status;
4050    }
4051 }
4052 
4053 ////////////////////////////////////////////////////////////////////////////////
4054 // debug support
4055 
4056 #ifndef PRODUCT
4057 static address same_page(address x, address y) {
4058   int page_bits = -os::vm_page_size();
4059   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4060     return x;
4061   else if (x > y)
4062     return (address)(intptr_t(y) | ~page_bits) + 1;
4063   else
4064     return (address)(intptr_t(y) & page_bits);
4065 }
4066 
4067 bool os::find(address addr) {
4068   Dl_info dlinfo;
4069   memset(&dlinfo, 0, sizeof(dlinfo));
4070   if (dladdr(addr, &dlinfo)) {
4071     tty->print(PTR_FORMAT ": ", addr);
4072     if (dlinfo.dli_sname != NULL) {
4073       tty->print("%s+%#x", dlinfo.dli_sname,
4074                  addr - (intptr_t)dlinfo.dli_saddr);
4075     } else if (dlinfo.dli_fname) {
4076       tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4077     } else {
4078       tty->print("<absolute address>");
4079     }
4080     if (dlinfo.dli_fname) {
4081       tty->print(" in %s", dlinfo.dli_fname);
4082     }
4083     if (dlinfo.dli_fbase) {
4084       tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4085     }
4086     tty->cr();
4087 
4088     if (Verbose) {
4089       // decode some bytes around the PC
4090       address begin = same_page(addr-40, addr);
4091       address end   = same_page(addr+40, addr);
4092       address       lowest = (address) dlinfo.dli_sname;
4093       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4094       if (begin < lowest)  begin = lowest;
4095       Dl_info dlinfo2;
4096       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4097           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4098         end = (address) dlinfo2.dli_saddr;
4099       Disassembler::decode(begin, end);
4100     }
4101     return true;
4102   }
4103   return false;
4104 }
4105 
4106 #endif
4107 
4108 ////////////////////////////////////////////////////////////////////////////////
4109 // misc
4110 
4111 // This does not do anything on Linux. This is basically a hook for being
4112 // able to use structured exception handling (thread-local exception filters)
4113 // on, e.g., Win32.
4114 void
4115 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4116                          JavaCallArguments* args, Thread* thread) {
4117   f(value, method, args, thread);
4118 }
4119 
4120 void os::print_statistics() {
4121 }
4122 
4123 int os::message_box(const char* title, const char* message) {
4124   int i;
4125   fdStream err(defaultStream::error_fd());
4126   for (i = 0; i < 78; i++) err.print_raw("=");
4127   err.cr();
4128   err.print_raw_cr(title);
4129   for (i = 0; i < 78; i++) err.print_raw("-");
4130   err.cr();
4131   err.print_raw_cr(message);
4132   for (i = 0; i < 78; i++) err.print_raw("=");
4133   err.cr();
4134 
4135   char buf[16];
4136   // Prevent process from exiting upon "read error" without consuming all CPU
4137   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4138 
4139   return buf[0] == 'y' || buf[0] == 'Y';
4140 }
4141 
4142 int os::stat(const char *path, struct stat *sbuf) {
4143   char pathbuf[MAX_PATH];
4144   if (strlen(path) > MAX_PATH - 1) {
4145     errno = ENAMETOOLONG;
4146     return -1;
4147   }
4148   hpi::native_path(strcpy(pathbuf, path));
4149   return ::stat(pathbuf, sbuf);
4150 }
4151 
4152 bool os::check_heap(bool force) {
4153   return true;
4154 }
4155 
4156 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4157   return ::vsnprintf(buf, count, format, args);
4158 }
4159 
4160 // Is a (classpath) directory empty?
4161 bool os::dir_is_empty(const char* path) {
4162   DIR *dir = NULL;
4163   struct dirent *ptr;
4164 
4165   dir = opendir(path);
4166   if (dir == NULL) return true;
4167 
4168   /* Scan the directory */
4169   bool result = true;
4170   char buf[sizeof(struct dirent) + MAX_PATH];
4171   while (result && (ptr = ::readdir(dir)) != NULL) {
4172     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4173       result = false;
4174     }
4175   }
4176   closedir(dir);
4177   return result;
4178 }
4179 
4180 // create binary file, rewriting existing file if required
4181 int os::create_binary_file(const char* path, bool rewrite_existing) {
4182   int oflags = O_WRONLY | O_CREAT;
4183   if (!rewrite_existing) {
4184     oflags |= O_EXCL;
4185   }
4186   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4187 }
4188 
4189 // return current position of file pointer
4190 jlong os::current_file_offset(int fd) {
4191   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4192 }
4193 
4194 // move file pointer to the specified offset
4195 jlong os::seek_to_file_offset(int fd, jlong offset) {
4196   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4197 }
4198 
4199 // Map a block of memory.
4200 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4201                      char *addr, size_t bytes, bool read_only,
4202                      bool allow_exec) {
4203   int prot;
4204   int flags;
4205 
4206   if (read_only) {
4207     prot = PROT_READ;
4208     flags = MAP_SHARED;
4209   } else {
4210     prot = PROT_READ | PROT_WRITE;
4211     flags = MAP_PRIVATE;
4212   }
4213 
4214   if (allow_exec) {
4215     prot |= PROT_EXEC;
4216   }
4217 
4218   if (addr != NULL) {
4219     flags |= MAP_FIXED;
4220   }
4221 
4222   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4223                                      fd, file_offset);
4224   if (mapped_address == MAP_FAILED) {
4225     return NULL;
4226   }
4227   return mapped_address;
4228 }
4229 
4230 
4231 // Remap a block of memory.
4232 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4233                        char *addr, size_t bytes, bool read_only,
4234                        bool allow_exec) {
4235   // same as map_memory() on this OS
4236   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4237                         allow_exec);
4238 }
4239 
4240 
4241 // Unmap a block of memory.
4242 bool os::unmap_memory(char* addr, size_t bytes) {
4243   return munmap(addr, bytes) == 0;
4244 }
4245 
4246 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4247 
4248 static clockid_t thread_cpu_clockid(Thread* thread) {
4249   pthread_t tid = thread->osthread()->pthread_id();
4250   clockid_t clockid;
4251 
4252   // Get thread clockid
4253   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4254   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4255   return clockid;
4256 }
4257 
4258 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4259 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4260 // of a thread.
4261 //
4262 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4263 // the fast estimate available on the platform.
4264 
4265 jlong os::current_thread_cpu_time() {
4266   if (os::Linux::supports_fast_thread_cpu_time()) {
4267     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4268   } else {
4269     // return user + sys since the cost is the same
4270     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4271   }
4272 }
4273 
4274 jlong os::thread_cpu_time(Thread* thread) {
4275   // consistent with what current_thread_cpu_time() returns
4276   if (os::Linux::supports_fast_thread_cpu_time()) {
4277     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4278   } else {
4279     return slow_thread_cpu_time(thread, true /* user + sys */);
4280   }
4281 }
4282 
4283 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4284   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4285     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4286   } else {
4287     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4288   }
4289 }
4290 
4291 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4292   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4293     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4294   } else {
4295     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4296   }
4297 }
4298 
4299 //
4300 //  -1 on error.
4301 //
4302 
4303 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4304   static bool proc_pid_cpu_avail = true;
4305   static bool proc_task_unchecked = true;
4306   static const char *proc_stat_path = "/proc/%d/stat";
4307   pid_t  tid = thread->osthread()->thread_id();
4308   int i;
4309   char *s;
4310   char stat[2048];
4311   int statlen;
4312   char proc_name[64];
4313   int count;
4314   long sys_time, user_time;
4315   char string[64];
4316   int idummy;
4317   long ldummy;
4318   FILE *fp;
4319 
4320   // We first try accessing /proc/<pid>/cpu since this is faster to
4321   // process.  If this file is not present (linux kernels 2.5 and above)
4322   // then we open /proc/<pid>/stat.
4323   if ( proc_pid_cpu_avail ) {
4324     sprintf(proc_name, "/proc/%d/cpu", tid);
4325     fp =  fopen(proc_name, "r");
4326     if ( fp != NULL ) {
4327       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4328       fclose(fp);
4329       if ( count != 3 ) return -1;
4330 
4331       if (user_sys_cpu_time) {
4332         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4333       } else {
4334         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4335       }
4336     }
4337     else proc_pid_cpu_avail = false;
4338   }
4339 
4340   // The /proc/<tid>/stat aggregates per-process usage on
4341   // new Linux kernels 2.6+ where NPTL is supported.
4342   // The /proc/self/task/<tid>/stat still has the per-thread usage.
4343   // See bug 6328462.
4344   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4345   // and possibly in some other cases, so we check its availability.
4346   if (proc_task_unchecked && os::Linux::is_NPTL()) {
4347     // This is executed only once
4348     proc_task_unchecked = false;
4349     fp = fopen("/proc/self/task", "r");
4350     if (fp != NULL) {
4351       proc_stat_path = "/proc/self/task/%d/stat";
4352       fclose(fp);
4353     }
4354   }
4355 
4356   sprintf(proc_name, proc_stat_path, tid);
4357   fp = fopen(proc_name, "r");
4358   if ( fp == NULL ) return -1;
4359   statlen = fread(stat, 1, 2047, fp);
4360   stat[statlen] = '\0';
4361   fclose(fp);
4362 
4363   // Skip pid and the command string. Note that we could be dealing with
4364   // weird command names, e.g. user could decide to rename java launcher
4365   // to "java 1.4.2 :)", then the stat file would look like
4366   //                1234 (java 1.4.2 :)) R ... ...
4367   // We don't really need to know the command string, just find the last
4368   // occurrence of ")" and then start parsing from there. See bug 4726580.
4369   s = strrchr(stat, ')');
4370   i = 0;
4371   if (s == NULL ) return -1;
4372 
4373   // Skip blank chars
4374   do s++; while (isspace(*s));
4375 
4376   count = sscanf(s,"%*c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4377                  &idummy, &idummy, &idummy, &idummy, &idummy,
4378                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4379                  &user_time, &sys_time);
4380   if ( count != 12 ) return -1;
4381   if (user_sys_cpu_time) {
4382     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4383   } else {
4384     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4385   }
4386 }
4387 
4388 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4389   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4390   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4391   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4392   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4393 }
4394 
4395 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4396   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4397   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4398   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4399   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4400 }
4401 
4402 bool os::is_thread_cpu_time_supported() {
4403   return true;
4404 }
4405 
4406 // System loadavg support.  Returns -1 if load average cannot be obtained.
4407 // Linux doesn't yet have a (official) notion of processor sets,
4408 // so just return the system wide load average.
4409 int os::loadavg(double loadavg[], int nelem) {
4410   return ::getloadavg(loadavg, nelem);
4411 }
4412 
4413 void os::pause() {
4414   char filename[MAX_PATH];
4415   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4416     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4417   } else {
4418     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4419   }
4420 
4421   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4422   if (fd != -1) {
4423     struct stat buf;
4424     close(fd);
4425     while (::stat(filename, &buf) == 0) {
4426       (void)::poll(NULL, 0, 100);
4427     }
4428   } else {
4429     jio_fprintf(stderr,
4430       "Could not open pause file '%s', continuing immediately.\n", filename);
4431   }
4432 }
4433 
4434 extern "C" {
4435 
4436 /**
4437  * NOTE: the following code is to keep the green threads code
4438  * in the libjava.so happy. Once the green threads is removed,
4439  * these code will no longer be needed.
4440  */
4441 int
4442 jdk_waitpid(pid_t pid, int* status, int options) {
4443     return waitpid(pid, status, options);
4444 }
4445 
4446 int
4447 fork1() {
4448     return fork();
4449 }
4450 
4451 int
4452 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4453     return sem_init(sem, pshared, value);
4454 }
4455 
4456 int
4457 jdk_sem_post(sem_t *sem) {
4458     return sem_post(sem);
4459 }
4460 
4461 int
4462 jdk_sem_wait(sem_t *sem) {
4463     return sem_wait(sem);
4464 }
4465 
4466 int
4467 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4468     return pthread_sigmask(how , newmask, oldmask);
4469 }
4470 
4471 }
4472 
4473 // Refer to the comments in os_solaris.cpp park-unpark.
4474 //
4475 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4476 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4477 // For specifics regarding the bug see GLIBC BUGID 261237 :
4478 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4479 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4480 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4481 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4482 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4483 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4484 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4485 // of libpthread avoids the problem, but isn't practical.
4486 //
4487 // Possible remedies:
4488 //
4489 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4490 //      This is palliative and probabilistic, however.  If the thread is preempted
4491 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4492 //      than the minimum period may have passed, and the abstime may be stale (in the
4493 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
4494 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4495 //
4496 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4497 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
4498 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4499 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
4500 //      thread.
4501 //
4502 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4503 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
4504 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4505 //      This also works well.  In fact it avoids kernel-level scalability impediments
4506 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4507 //      timers in a graceful fashion.
4508 //
4509 // 4.   When the abstime value is in the past it appears that control returns
4510 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4511 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
4512 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
4513 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
4514 //      It may be possible to avoid reinitialization by checking the return
4515 //      value from pthread_cond_timedwait().  In addition to reinitializing the
4516 //      condvar we must establish the invariant that cond_signal() is only called
4517 //      within critical sections protected by the adjunct mutex.  This prevents
4518 //      cond_signal() from "seeing" a condvar that's in the midst of being
4519 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
4520 //      desirable signal-after-unlock optimization that avoids futile context switching.
4521 //
4522 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
4523 //      structure when a condvar is used or initialized.  cond_destroy()  would
4524 //      release the helper structure.  Our reinitialize-after-timedwait fix
4525 //      put excessive stress on malloc/free and locks protecting the c-heap.
4526 //
4527 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
4528 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4529 // and only enabling the work-around for vulnerable environments.
4530 
4531 // utility to compute the abstime argument to timedwait:
4532 // millis is the relative timeout time
4533 // abstime will be the absolute timeout time
4534 // TODO: replace compute_abstime() with unpackTime()
4535 
4536 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4537   if (millis < 0)  millis = 0;
4538   struct timeval now;
4539   int status = gettimeofday(&now, NULL);
4540   assert(status == 0, "gettimeofday");
4541   jlong seconds = millis / 1000;
4542   millis %= 1000;
4543   if (seconds > 50000000) { // see man cond_timedwait(3T)
4544     seconds = 50000000;
4545   }
4546   abstime->tv_sec = now.tv_sec  + seconds;
4547   long       usec = now.tv_usec + millis * 1000;
4548   if (usec >= 1000000) {
4549     abstime->tv_sec += 1;
4550     usec -= 1000000;
4551   }
4552   abstime->tv_nsec = usec * 1000;
4553   return abstime;
4554 }
4555 
4556 
4557 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4558 // Conceptually TryPark() should be equivalent to park(0).
4559 
4560 int os::PlatformEvent::TryPark() {
4561   for (;;) {
4562     const int v = _Event ;
4563     guarantee ((v == 0) || (v == 1), "invariant") ;
4564     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
4565   }
4566 }
4567 
4568 void os::PlatformEvent::park() {       // AKA "down()"
4569   // Invariant: Only the thread associated with the Event/PlatformEvent
4570   // may call park().
4571   // TODO: assert that _Assoc != NULL or _Assoc == Self
4572   int v ;
4573   for (;;) {
4574       v = _Event ;
4575       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4576   }
4577   guarantee (v >= 0, "invariant") ;
4578   if (v == 0) {
4579      // Do this the hard way by blocking ...
4580      int status = pthread_mutex_lock(_mutex);
4581      assert_status(status == 0, status, "mutex_lock");
4582      guarantee (_nParked == 0, "invariant") ;
4583      ++ _nParked ;
4584      while (_Event < 0) {
4585         status = pthread_cond_wait(_cond, _mutex);
4586         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4587         // Treat this the same as if the wait was interrupted
4588         if (status == ETIME) { status = EINTR; }
4589         assert_status(status == 0 || status == EINTR, status, "cond_wait");
4590      }
4591      -- _nParked ;
4592 
4593     // In theory we could move the ST of 0 into _Event past the unlock(),
4594     // but then we'd need a MEMBAR after the ST.
4595     _Event = 0 ;
4596      status = pthread_mutex_unlock(_mutex);
4597      assert_status(status == 0, status, "mutex_unlock");
4598   }
4599   guarantee (_Event >= 0, "invariant") ;
4600 }
4601 
4602 int os::PlatformEvent::park(jlong millis) {
4603   guarantee (_nParked == 0, "invariant") ;
4604 
4605   int v ;
4606   for (;;) {
4607       v = _Event ;
4608       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4609   }
4610   guarantee (v >= 0, "invariant") ;
4611   if (v != 0) return OS_OK ;
4612 
4613   // We do this the hard way, by blocking the thread.
4614   // Consider enforcing a minimum timeout value.
4615   struct timespec abst;
4616   compute_abstime(&abst, millis);
4617 
4618   int ret = OS_TIMEOUT;
4619   int status = pthread_mutex_lock(_mutex);
4620   assert_status(status == 0, status, "mutex_lock");
4621   guarantee (_nParked == 0, "invariant") ;
4622   ++_nParked ;
4623 
4624   // Object.wait(timo) will return because of
4625   // (a) notification
4626   // (b) timeout
4627   // (c) thread.interrupt
4628   //
4629   // Thread.interrupt and object.notify{All} both call Event::set.
4630   // That is, we treat thread.interrupt as a special case of notification.
4631   // The underlying Solaris implementation, cond_timedwait, admits
4632   // spurious/premature wakeups, but the JLS/JVM spec prevents the
4633   // JVM from making those visible to Java code.  As such, we must
4634   // filter out spurious wakeups.  We assume all ETIME returns are valid.
4635   //
4636   // TODO: properly differentiate simultaneous notify+interrupt.
4637   // In that case, we should propagate the notify to another waiter.
4638 
4639   while (_Event < 0) {
4640     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4641     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4642       pthread_cond_destroy (_cond);
4643       pthread_cond_init (_cond, NULL) ;
4644     }
4645     assert_status(status == 0 || status == EINTR ||
4646                   status == ETIME || status == ETIMEDOUT,
4647                   status, "cond_timedwait");
4648     if (!FilterSpuriousWakeups) break ;                 // previous semantics
4649     if (status == ETIME || status == ETIMEDOUT) break ;
4650     // We consume and ignore EINTR and spurious wakeups.
4651   }
4652   --_nParked ;
4653   if (_Event >= 0) {
4654      ret = OS_OK;
4655   }
4656   _Event = 0 ;
4657   status = pthread_mutex_unlock(_mutex);
4658   assert_status(status == 0, status, "mutex_unlock");
4659   assert (_nParked == 0, "invariant") ;
4660   return ret;
4661 }
4662 
4663 void os::PlatformEvent::unpark() {
4664   int v, AnyWaiters ;
4665   for (;;) {
4666       v = _Event ;
4667       if (v > 0) {
4668          // The LD of _Event could have reordered or be satisfied
4669          // by a read-aside from this processor's write buffer.
4670          // To avoid problems execute a barrier and then
4671          // ratify the value.
4672          OrderAccess::fence() ;
4673          if (_Event == v) return ;
4674          continue ;
4675       }
4676       if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4677   }
4678   if (v < 0) {
4679      // Wait for the thread associated with the event to vacate
4680      int status = pthread_mutex_lock(_mutex);
4681      assert_status(status == 0, status, "mutex_lock");
4682      AnyWaiters = _nParked ;
4683      assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4684      if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4685         AnyWaiters = 0 ;
4686         pthread_cond_signal (_cond);
4687      }
4688      status = pthread_mutex_unlock(_mutex);
4689      assert_status(status == 0, status, "mutex_unlock");
4690      if (AnyWaiters != 0) {
4691         status = pthread_cond_signal(_cond);
4692         assert_status(status == 0, status, "cond_signal");
4693      }
4694   }
4695 
4696   // Note that we signal() _after dropping the lock for "immortal" Events.
4697   // This is safe and avoids a common class of  futile wakeups.  In rare
4698   // circumstances this can cause a thread to return prematurely from
4699   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4700   // simply re-test the condition and re-park itself.
4701 }
4702 
4703 
4704 // JSR166
4705 // -------------------------------------------------------
4706 
4707 /*
4708  * The solaris and linux implementations of park/unpark are fairly
4709  * conservative for now, but can be improved. They currently use a
4710  * mutex/condvar pair, plus a a count.
4711  * Park decrements count if > 0, else does a condvar wait.  Unpark
4712  * sets count to 1 and signals condvar.  Only one thread ever waits
4713  * on the condvar. Contention seen when trying to park implies that someone
4714  * is unparking you, so don't wait. And spurious returns are fine, so there
4715  * is no need to track notifications.
4716  */
4717 
4718 
4719 #define NANOSECS_PER_SEC 1000000000
4720 #define NANOSECS_PER_MILLISEC 1000000
4721 #define MAX_SECS 100000000
4722 /*
4723  * This code is common to linux and solaris and will be moved to a
4724  * common place in dolphin.
4725  *
4726  * The passed in time value is either a relative time in nanoseconds
4727  * or an absolute time in milliseconds. Either way it has to be unpacked
4728  * into suitable seconds and nanoseconds components and stored in the
4729  * given timespec structure.
4730  * Given time is a 64-bit value and the time_t used in the timespec is only
4731  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4732  * overflow if times way in the future are given. Further on Solaris versions
4733  * prior to 10 there is a restriction (see cond_timedwait) that the specified
4734  * number of seconds, in abstime, is less than current_time  + 100,000,000.
4735  * As it will be 28 years before "now + 100000000" will overflow we can
4736  * ignore overflow and just impose a hard-limit on seconds using the value
4737  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4738  * years from "now".
4739  */
4740 
4741 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4742   assert (time > 0, "convertTime");
4743 
4744   struct timeval now;
4745   int status = gettimeofday(&now, NULL);
4746   assert(status == 0, "gettimeofday");
4747 
4748   time_t max_secs = now.tv_sec + MAX_SECS;
4749 
4750   if (isAbsolute) {
4751     jlong secs = time / 1000;
4752     if (secs > max_secs) {
4753       absTime->tv_sec = max_secs;
4754     }
4755     else {
4756       absTime->tv_sec = secs;
4757     }
4758     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4759   }
4760   else {
4761     jlong secs = time / NANOSECS_PER_SEC;
4762     if (secs >= MAX_SECS) {
4763       absTime->tv_sec = max_secs;
4764       absTime->tv_nsec = 0;
4765     }
4766     else {
4767       absTime->tv_sec = now.tv_sec + secs;
4768       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4769       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4770         absTime->tv_nsec -= NANOSECS_PER_SEC;
4771         ++absTime->tv_sec; // note: this must be <= max_secs
4772       }
4773     }
4774   }
4775   assert(absTime->tv_sec >= 0, "tv_sec < 0");
4776   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4777   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4778   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4779 }
4780 
4781 void Parker::park(bool isAbsolute, jlong time) {
4782   // Optional fast-path check:
4783   // Return immediately if a permit is available.
4784   if (_counter > 0) {
4785       _counter = 0 ;
4786       OrderAccess::fence();
4787       return ;
4788   }
4789 
4790   Thread* thread = Thread::current();
4791   assert(thread->is_Java_thread(), "Must be JavaThread");
4792   JavaThread *jt = (JavaThread *)thread;
4793 
4794   // Optional optimization -- avoid state transitions if there's an interrupt pending.
4795   // Check interrupt before trying to wait
4796   if (Thread::is_interrupted(thread, false)) {
4797     return;
4798   }
4799 
4800   // Next, demultiplex/decode time arguments
4801   timespec absTime;
4802   if (time < 0) { // don't wait at all
4803     return;
4804   }
4805   if (time > 0) {
4806     unpackTime(&absTime, isAbsolute, time);
4807   }
4808 
4809 
4810   // Enter safepoint region
4811   // Beware of deadlocks such as 6317397.
4812   // The per-thread Parker:: mutex is a classic leaf-lock.
4813   // In particular a thread must never block on the Threads_lock while
4814   // holding the Parker:: mutex.  If safepoints are pending both the
4815   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4816   ThreadBlockInVM tbivm(jt);
4817 
4818   // Don't wait if cannot get lock since interference arises from
4819   // unblocking.  Also. check interrupt before trying wait
4820   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4821     return;
4822   }
4823 
4824   int status ;
4825   if (_counter > 0)  { // no wait needed
4826     _counter = 0;
4827     status = pthread_mutex_unlock(_mutex);
4828     assert (status == 0, "invariant") ;
4829     OrderAccess::fence();
4830     return;
4831   }
4832 
4833 #ifdef ASSERT
4834   // Don't catch signals while blocked; let the running threads have the signals.
4835   // (This allows a debugger to break into the running thread.)
4836   sigset_t oldsigs;
4837   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4838   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4839 #endif
4840 
4841   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4842   jt->set_suspend_equivalent();
4843   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4844 
4845   if (time == 0) {
4846     status = pthread_cond_wait (_cond, _mutex) ;
4847   } else {
4848     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4849     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4850       pthread_cond_destroy (_cond) ;
4851       pthread_cond_init    (_cond, NULL);
4852     }
4853   }
4854   assert_status(status == 0 || status == EINTR ||
4855                 status == ETIME || status == ETIMEDOUT,
4856                 status, "cond_timedwait");
4857 
4858 #ifdef ASSERT
4859   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4860 #endif
4861 
4862   _counter = 0 ;
4863   status = pthread_mutex_unlock(_mutex) ;
4864   assert_status(status == 0, status, "invariant") ;
4865   // If externally suspended while waiting, re-suspend
4866   if (jt->handle_special_suspend_equivalent_condition()) {
4867     jt->java_suspend_self();
4868   }
4869 
4870   OrderAccess::fence();
4871 }
4872 
4873 void Parker::unpark() {
4874   int s, status ;
4875   status = pthread_mutex_lock(_mutex);
4876   assert (status == 0, "invariant") ;
4877   s = _counter;
4878   _counter = 1;
4879   if (s < 1) {
4880      if (WorkAroundNPTLTimedWaitHang) {
4881         status = pthread_cond_signal (_cond) ;
4882         assert (status == 0, "invariant") ;
4883         status = pthread_mutex_unlock(_mutex);
4884         assert (status == 0, "invariant") ;
4885      } else {
4886         status = pthread_mutex_unlock(_mutex);
4887         assert (status == 0, "invariant") ;
4888         status = pthread_cond_signal (_cond) ;
4889         assert (status == 0, "invariant") ;
4890      }
4891   } else {
4892     pthread_mutex_unlock(_mutex);
4893     assert (status == 0, "invariant") ;
4894   }
4895 }
4896 
4897 
4898 extern char** environ;
4899 
4900 #ifndef __NR_fork
4901 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
4902 #endif
4903 
4904 #ifndef __NR_execve
4905 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
4906 #endif
4907 
4908 // Run the specified command in a separate process. Return its exit value,
4909 // or -1 on failure (e.g. can't fork a new process).
4910 // Unlike system(), this function can be called from signal handler. It
4911 // doesn't block SIGINT et al.
4912 int os::fork_and_exec(char* cmd) {
4913   const char * argv[4] = {"sh", "-c", cmd, NULL};
4914 
4915   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
4916   // pthread_atfork handlers and reset pthread library. All we need is a
4917   // separate process to execve. Make a direct syscall to fork process.
4918   // On IA64 there's no fork syscall, we have to use fork() and hope for
4919   // the best...
4920   pid_t pid = NOT_IA64(syscall(__NR_fork);)
4921               IA64_ONLY(fork();)
4922 
4923   if (pid < 0) {
4924     // fork failed
4925     return -1;
4926 
4927   } else if (pid == 0) {
4928     // child process
4929 
4930     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
4931     // first to kill every thread on the thread list. Because this list is
4932     // not reset by fork() (see notes above), execve() will instead kill
4933     // every thread in the parent process. We know this is the only thread
4934     // in the new process, so make a system call directly.
4935     // IA64 should use normal execve() from glibc to match the glibc fork()
4936     // above.
4937     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
4938     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
4939 
4940     // execve failed
4941     _exit(-1);
4942 
4943   } else  {
4944     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4945     // care about the actual exit code, for now.
4946 
4947     int status;
4948 
4949     // Wait for the child process to exit.  This returns immediately if
4950     // the child has already exited. */
4951     while (waitpid(pid, &status, 0) < 0) {
4952         switch (errno) {
4953         case ECHILD: return 0;
4954         case EINTR: break;
4955         default: return -1;
4956         }
4957     }
4958 
4959     if (WIFEXITED(status)) {
4960        // The child exited normally; get its exit code.
4961        return WEXITSTATUS(status);
4962     } else if (WIFSIGNALED(status)) {
4963        // The child exited because of a signal
4964        // The best value to return is 0x80 + signal number,
4965        // because that is what all Unix shells do, and because
4966        // it allows callers to distinguish between process exit and
4967        // process death by signal.
4968        return 0x80 + WTERMSIG(status);
4969     } else {
4970        // Unknown exit code; pass it through
4971        return status;
4972     }
4973   }
4974 }