1 /*
   2  * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "memory/allocation.inline.hpp"
  36 #include "memory/filemap.hpp"
  37 #include "mutex_linux.inline.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_linux.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "prims/jniFastGetField.hpp"
  42 #include "prims/jvm.h"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.inline.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.hpp"
  49 #include "runtime/init.hpp"
  50 #include "runtime/java.hpp"
  51 #include "runtime/javaCalls.hpp"
  52 #include "runtime/mutexLocker.hpp"
  53 #include "runtime/objectMonitor.hpp"
  54 #include "runtime/orderAccess.inline.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/perfMemory.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/statSampler.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "runtime/thread.inline.hpp"
  61 #include "runtime/threadCritical.hpp"
  62 #include "runtime/timer.hpp"
  63 #include "services/attachListener.hpp"
  64 #include "services/memTracker.hpp"
  65 #include "services/runtimeService.hpp"
  66 #include "utilities/decoder.hpp"
  67 #include "utilities/defaultStream.hpp"
  68 #include "utilities/events.hpp"
  69 #include "utilities/elfFile.hpp"
  70 #include "utilities/growableArray.hpp"
  71 #include "utilities/macros.hpp"
  72 #include "utilities/vmError.hpp"
  73 
  74 // put OS-includes here
  75 # include <sys/types.h>
  76 # include <sys/mman.h>
  77 # include <sys/stat.h>
  78 # include <sys/select.h>
  79 # include <pthread.h>
  80 # include <signal.h>
  81 # include <errno.h>
  82 # include <dlfcn.h>
  83 # include <stdio.h>
  84 # include <unistd.h>
  85 # include <sys/resource.h>
  86 # include <pthread.h>
  87 # include <sys/stat.h>
  88 # include <sys/time.h>
  89 # include <sys/times.h>
  90 # include <sys/utsname.h>
  91 # include <sys/socket.h>
  92 # include <sys/wait.h>
  93 # include <pwd.h>
  94 # include <poll.h>
  95 # include <semaphore.h>
  96 # include <fcntl.h>
  97 # include <string.h>
  98 # include <syscall.h>
  99 # include <sys/sysinfo.h>
 100 # include <gnu/libc-version.h>
 101 # include <sys/ipc.h>
 102 # include <sys/shm.h>
 103 # include <link.h>
 104 # include <stdint.h>
 105 # include <inttypes.h>
 106 # include <sys/ioctl.h>
 107 
 108 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
 109 
 110 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 111 // getrusage() is prepared to handle the associated failure.
 112 #ifndef RUSAGE_THREAD
 113   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 114 #endif
 115 
 116 #define MAX_PATH    (2 * K)
 117 
 118 #define MAX_SECS 100000000
 119 
 120 // for timer info max values which include all bits
 121 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 122 
 123 #define LARGEPAGES_BIT (1 << 6)
 124 ////////////////////////////////////////////////////////////////////////////////
 125 // global variables
 126 julong os::Linux::_physical_memory = 0;
 127 
 128 address   os::Linux::_initial_thread_stack_bottom = NULL;
 129 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 130 
 131 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 132 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 133 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 134 Mutex* os::Linux::_createThread_lock = NULL;
 135 pthread_t os::Linux::_main_thread;
 136 int os::Linux::_page_size = -1;
 137 const int os::Linux::_vm_default_page_size = (8 * K);
 138 bool os::Linux::_is_floating_stack = false;
 139 bool os::Linux::_is_NPTL = false;
 140 bool os::Linux::_supports_fast_thread_cpu_time = false;
 141 const char * os::Linux::_glibc_version = NULL;
 142 const char * os::Linux::_libpthread_version = NULL;
 143 pthread_condattr_t os::Linux::_condattr[1];
 144 
 145 static jlong initial_time_count=0;
 146 
 147 static int clock_tics_per_sec = 100;
 148 
 149 // For diagnostics to print a message once. see run_periodic_checks
 150 static sigset_t check_signal_done;
 151 static bool check_signals = true;
 152 
 153 static pid_t _initial_pid = 0;
 154 
 155 // Signal number used to suspend/resume a thread
 156 
 157 // do not use any signal number less than SIGSEGV, see 4355769
 158 static int SR_signum = SIGUSR2;
 159 sigset_t SR_sigset;
 160 
 161 // Used to protect dlsym() calls
 162 static pthread_mutex_t dl_mutex;
 163 
 164 // Declarations
 165 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 166 
 167 // utility functions
 168 
 169 static int SR_initialize();
 170 
 171 julong os::available_memory() {
 172   return Linux::available_memory();
 173 }
 174 
 175 julong os::Linux::available_memory() {
 176   // values in struct sysinfo are "unsigned long"
 177   struct sysinfo si;
 178   sysinfo(&si);
 179 
 180   return (julong)si.freeram * si.mem_unit;
 181 }
 182 
 183 julong os::physical_memory() {
 184   return Linux::physical_memory();
 185 }
 186 
 187 ////////////////////////////////////////////////////////////////////////////////
 188 // environment support
 189 
 190 bool os::getenv(const char* name, char* buf, int len) {
 191   const char* val = ::getenv(name);
 192   if (val != NULL && strlen(val) < (size_t)len) {
 193     strcpy(buf, val);
 194     return true;
 195   }
 196   if (len > 0) buf[0] = 0;  // return a null string
 197   return false;
 198 }
 199 
 200 
 201 // Return true if user is running as root.
 202 
 203 bool os::have_special_privileges() {
 204   static bool init = false;
 205   static bool privileges = false;
 206   if (!init) {
 207     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 208     init = true;
 209   }
 210   return privileges;
 211 }
 212 
 213 
 214 #ifndef SYS_gettid
 215 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 216   #ifdef __ia64__
 217     #define SYS_gettid 1105
 218   #elif __i386__
 219     #define SYS_gettid 224
 220   #elif __amd64__
 221     #define SYS_gettid 186
 222   #elif __sparc__
 223     #define SYS_gettid 143
 224   #else
 225     #error define gettid for the arch
 226   #endif
 227 #endif
 228 
 229 // Cpu architecture string
 230 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 231 
 232 
 233 // pid_t gettid()
 234 //
 235 // Returns the kernel thread id of the currently running thread. Kernel
 236 // thread id is used to access /proc.
 237 //
 238 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 239 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 240 //
 241 pid_t os::Linux::gettid() {
 242   int rslt = syscall(SYS_gettid);
 243   if (rslt == -1) {
 244     // old kernel, no NPTL support
 245     return getpid();
 246   } else {
 247     return (pid_t)rslt;
 248   }
 249 }
 250 
 251 // Most versions of linux have a bug where the number of processors are
 252 // determined by looking at the /proc file system.  In a chroot environment,
 253 // the system call returns 1.  This causes the VM to act as if it is
 254 // a single processor and elide locking (see is_MP() call).
 255 static bool unsafe_chroot_detected = false;
 256 static const char *unstable_chroot_error = "/proc file system not found.\n"
 257                      "Java may be unstable running multithreaded in a chroot "
 258                      "environment on Linux when /proc filesystem is not mounted.";
 259 
 260 void os::Linux::initialize_system_info() {
 261   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 262   if (processor_count() == 1) {
 263     pid_t pid = os::Linux::gettid();
 264     char fname[32];
 265     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 266     FILE *fp = fopen(fname, "r");
 267     if (fp == NULL) {
 268       unsafe_chroot_detected = true;
 269     } else {
 270       fclose(fp);
 271     }
 272   }
 273   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 274   assert(processor_count() > 0, "linux error");
 275 }
 276 
 277 void os::init_system_properties_values() {
 278   // The next steps are taken in the product version:
 279   //
 280   // Obtain the JAVA_HOME value from the location of libjvm.so.
 281   // This library should be located at:
 282   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 283   //
 284   // If "/jre/lib/" appears at the right place in the path, then we
 285   // assume libjvm.so is installed in a JDK and we use this path.
 286   //
 287   // Otherwise exit with message: "Could not create the Java virtual machine."
 288   //
 289   // The following extra steps are taken in the debugging version:
 290   //
 291   // If "/jre/lib/" does NOT appear at the right place in the path
 292   // instead of exit check for $JAVA_HOME environment variable.
 293   //
 294   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 295   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 296   // it looks like libjvm.so is installed there
 297   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 298   //
 299   // Otherwise exit.
 300   //
 301   // Important note: if the location of libjvm.so changes this
 302   // code needs to be changed accordingly.
 303 
 304   // See ld(1):
 305   //      The linker uses the following search paths to locate required
 306   //      shared libraries:
 307   //        1: ...
 308   //        ...
 309   //        7: The default directories, normally /lib and /usr/lib.
 310 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 311   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 312 #else
 313   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 314 #endif
 315 
 316 // Base path of extensions installed on the system.
 317 #define SYS_EXT_DIR     "/usr/java/packages"
 318 #define EXTENSIONS_DIR  "/lib/ext"
 319 
 320   // Buffer that fits several sprintfs.
 321   // Note that the space for the colon and the trailing null are provided
 322   // by the nulls included by the sizeof operator.
 323   const size_t bufsize =
 324     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 325          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 326   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 327 
 328   // sysclasspath, java_home, dll_dir
 329   {
 330     char *pslash;
 331     os::jvm_path(buf, bufsize);
 332 
 333     // Found the full path to libjvm.so.
 334     // Now cut the path to <java_home>/jre if we can.
 335     pslash = strrchr(buf, '/');
 336     if (pslash != NULL) {
 337       *pslash = '\0';            // Get rid of /libjvm.so.
 338     }
 339     pslash = strrchr(buf, '/');
 340     if (pslash != NULL) {
 341       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 342     }
 343     Arguments::set_dll_dir(buf);
 344 
 345     if (pslash != NULL) {
 346       pslash = strrchr(buf, '/');
 347       if (pslash != NULL) {
 348         *pslash = '\0';          // Get rid of /<arch>.
 349         pslash = strrchr(buf, '/');
 350         if (pslash != NULL) {
 351           *pslash = '\0';        // Get rid of /lib.
 352         }
 353       }
 354     }
 355     Arguments::set_java_home(buf);
 356     set_boot_path('/', ':');
 357   }
 358 
 359   // Where to look for native libraries.
 360   //
 361   // Note: Due to a legacy implementation, most of the library path
 362   // is set in the launcher. This was to accomodate linking restrictions
 363   // on legacy Linux implementations (which are no longer supported).
 364   // Eventually, all the library path setting will be done here.
 365   //
 366   // However, to prevent the proliferation of improperly built native
 367   // libraries, the new path component /usr/java/packages is added here.
 368   // Eventually, all the library path setting will be done here.
 369   {
 370     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 371     // should always exist (until the legacy problem cited above is
 372     // addressed).
 373     const char *v = ::getenv("LD_LIBRARY_PATH");
 374     const char *v_colon = ":";
 375     if (v == NULL) { v = ""; v_colon = ""; }
 376     // That's +1 for the colon and +1 for the trailing '\0'.
 377     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 378                                                      strlen(v) + 1 +
 379                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 380                                                      mtInternal);
 381     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 382     Arguments::set_library_path(ld_library_path);
 383     FREE_C_HEAP_ARRAY(char, ld_library_path);
 384   }
 385 
 386   // Extensions directories.
 387   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 388   Arguments::set_ext_dirs(buf);
 389 
 390   FREE_C_HEAP_ARRAY(char, buf);
 391 
 392 #undef DEFAULT_LIBPATH
 393 #undef SYS_EXT_DIR
 394 #undef EXTENSIONS_DIR
 395 }
 396 
 397 ////////////////////////////////////////////////////////////////////////////////
 398 // breakpoint support
 399 
 400 void os::breakpoint() {
 401   BREAKPOINT;
 402 }
 403 
 404 extern "C" void breakpoint() {
 405   // use debugger to set breakpoint here
 406 }
 407 
 408 ////////////////////////////////////////////////////////////////////////////////
 409 // signal support
 410 
 411 debug_only(static bool signal_sets_initialized = false);
 412 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 413 
 414 bool os::Linux::is_sig_ignored(int sig) {
 415   struct sigaction oact;
 416   sigaction(sig, (struct sigaction*)NULL, &oact);
 417   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 418                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 419   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 420     return true;
 421   } else {
 422     return false;
 423   }
 424 }
 425 
 426 void os::Linux::signal_sets_init() {
 427   // Should also have an assertion stating we are still single-threaded.
 428   assert(!signal_sets_initialized, "Already initialized");
 429   // Fill in signals that are necessarily unblocked for all threads in
 430   // the VM. Currently, we unblock the following signals:
 431   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 432   //                         by -Xrs (=ReduceSignalUsage));
 433   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 434   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 435   // the dispositions or masks wrt these signals.
 436   // Programs embedding the VM that want to use the above signals for their
 437   // own purposes must, at this time, use the "-Xrs" option to prevent
 438   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 439   // (See bug 4345157, and other related bugs).
 440   // In reality, though, unblocking these signals is really a nop, since
 441   // these signals are not blocked by default.
 442   sigemptyset(&unblocked_sigs);
 443   sigemptyset(&allowdebug_blocked_sigs);
 444   sigaddset(&unblocked_sigs, SIGILL);
 445   sigaddset(&unblocked_sigs, SIGSEGV);
 446   sigaddset(&unblocked_sigs, SIGBUS);
 447   sigaddset(&unblocked_sigs, SIGFPE);
 448 #if defined(PPC64)
 449   sigaddset(&unblocked_sigs, SIGTRAP);
 450 #endif
 451   sigaddset(&unblocked_sigs, SR_signum);
 452 
 453   if (!ReduceSignalUsage) {
 454     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 455       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 456       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 457     }
 458     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 459       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 460       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 461     }
 462     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 463       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 464       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 465     }
 466   }
 467   // Fill in signals that are blocked by all but the VM thread.
 468   sigemptyset(&vm_sigs);
 469   if (!ReduceSignalUsage) {
 470     sigaddset(&vm_sigs, BREAK_SIGNAL);
 471   }
 472   debug_only(signal_sets_initialized = true);
 473 
 474 }
 475 
 476 // These are signals that are unblocked while a thread is running Java.
 477 // (For some reason, they get blocked by default.)
 478 sigset_t* os::Linux::unblocked_signals() {
 479   assert(signal_sets_initialized, "Not initialized");
 480   return &unblocked_sigs;
 481 }
 482 
 483 // These are the signals that are blocked while a (non-VM) thread is
 484 // running Java. Only the VM thread handles these signals.
 485 sigset_t* os::Linux::vm_signals() {
 486   assert(signal_sets_initialized, "Not initialized");
 487   return &vm_sigs;
 488 }
 489 
 490 // These are signals that are blocked during cond_wait to allow debugger in
 491 sigset_t* os::Linux::allowdebug_blocked_signals() {
 492   assert(signal_sets_initialized, "Not initialized");
 493   return &allowdebug_blocked_sigs;
 494 }
 495 
 496 void os::Linux::hotspot_sigmask(Thread* thread) {
 497 
 498   //Save caller's signal mask before setting VM signal mask
 499   sigset_t caller_sigmask;
 500   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 501 
 502   OSThread* osthread = thread->osthread();
 503   osthread->set_caller_sigmask(caller_sigmask);
 504 
 505   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 506 
 507   if (!ReduceSignalUsage) {
 508     if (thread->is_VM_thread()) {
 509       // Only the VM thread handles BREAK_SIGNAL ...
 510       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 511     } else {
 512       // ... all other threads block BREAK_SIGNAL
 513       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 514     }
 515   }
 516 }
 517 
 518 //////////////////////////////////////////////////////////////////////////////
 519 // detecting pthread library
 520 
 521 void os::Linux::libpthread_init() {
 522   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 523   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 524   // generic name for earlier versions.
 525   // Define macros here so we can build HotSpot on old systems.
 526 #ifndef _CS_GNU_LIBC_VERSION
 527   #define _CS_GNU_LIBC_VERSION 2
 528 #endif
 529 #ifndef _CS_GNU_LIBPTHREAD_VERSION
 530   #define _CS_GNU_LIBPTHREAD_VERSION 3
 531 #endif
 532 
 533   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 534   if (n > 0) {
 535     char *str = (char *)malloc(n, mtInternal);
 536     confstr(_CS_GNU_LIBC_VERSION, str, n);
 537     os::Linux::set_glibc_version(str);
 538   } else {
 539     // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 540     static char _gnu_libc_version[32];
 541     jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 542                  "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 543     os::Linux::set_glibc_version(_gnu_libc_version);
 544   }
 545 
 546   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 547   if (n > 0) {
 548     char *str = (char *)malloc(n, mtInternal);
 549     confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 550     // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 551     // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 552     // is the case. LinuxThreads has a hard limit on max number of threads.
 553     // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 554     // On the other hand, NPTL does not have such a limit, sysconf()
 555     // will return -1 and errno is not changed. Check if it is really NPTL.
 556     if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 557         strstr(str, "NPTL") &&
 558         sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 559       free(str);
 560       os::Linux::set_libpthread_version("linuxthreads");
 561     } else {
 562       os::Linux::set_libpthread_version(str);
 563     }
 564   } else {
 565     // glibc before 2.3.2 only has LinuxThreads.
 566     os::Linux::set_libpthread_version("linuxthreads");
 567   }
 568 
 569   if (strstr(libpthread_version(), "NPTL")) {
 570     os::Linux::set_is_NPTL();
 571   } else {
 572     os::Linux::set_is_LinuxThreads();
 573   }
 574 
 575   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 576   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 577   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 578     os::Linux::set_is_floating_stack();
 579   }
 580 }
 581 
 582 /////////////////////////////////////////////////////////////////////////////
 583 // thread stack
 584 
 585 // Force Linux kernel to expand current thread stack. If "bottom" is close
 586 // to the stack guard, caller should block all signals.
 587 //
 588 // MAP_GROWSDOWN:
 589 //   A special mmap() flag that is used to implement thread stacks. It tells
 590 //   kernel that the memory region should extend downwards when needed. This
 591 //   allows early versions of LinuxThreads to only mmap the first few pages
 592 //   when creating a new thread. Linux kernel will automatically expand thread
 593 //   stack as needed (on page faults).
 594 //
 595 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 596 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 597 //   region, it's hard to tell if the fault is due to a legitimate stack
 598 //   access or because of reading/writing non-exist memory (e.g. buffer
 599 //   overrun). As a rule, if the fault happens below current stack pointer,
 600 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 601 //   application (see Linux kernel fault.c).
 602 //
 603 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 604 //   stack overflow detection.
 605 //
 606 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 607 //   not use this flag. However, the stack of initial thread is not created
 608 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 609 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 610 //   and then attach the thread to JVM.
 611 //
 612 // To get around the problem and allow stack banging on Linux, we need to
 613 // manually expand thread stack after receiving the SIGSEGV.
 614 //
 615 // There are two ways to expand thread stack to address "bottom", we used
 616 // both of them in JVM before 1.5:
 617 //   1. adjust stack pointer first so that it is below "bottom", and then
 618 //      touch "bottom"
 619 //   2. mmap() the page in question
 620 //
 621 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 622 // if current sp is already near the lower end of page 101, and we need to
 623 // call mmap() to map page 100, it is possible that part of the mmap() frame
 624 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 625 // That will destroy the mmap() frame and cause VM to crash.
 626 //
 627 // The following code works by adjusting sp first, then accessing the "bottom"
 628 // page to force a page fault. Linux kernel will then automatically expand the
 629 // stack mapping.
 630 //
 631 // _expand_stack_to() assumes its frame size is less than page size, which
 632 // should always be true if the function is not inlined.
 633 
 634 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 635   #define NOINLINE
 636 #else
 637   #define NOINLINE __attribute__ ((noinline))
 638 #endif
 639 
 640 static void _expand_stack_to(address bottom) NOINLINE;
 641 
 642 static void _expand_stack_to(address bottom) {
 643   address sp;
 644   size_t size;
 645   volatile char *p;
 646 
 647   // Adjust bottom to point to the largest address within the same page, it
 648   // gives us a one-page buffer if alloca() allocates slightly more memory.
 649   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 650   bottom += os::Linux::page_size() - 1;
 651 
 652   // sp might be slightly above current stack pointer; if that's the case, we
 653   // will alloca() a little more space than necessary, which is OK. Don't use
 654   // os::current_stack_pointer(), as its result can be slightly below current
 655   // stack pointer, causing us to not alloca enough to reach "bottom".
 656   sp = (address)&sp;
 657 
 658   if (sp > bottom) {
 659     size = sp - bottom;
 660     p = (volatile char *)alloca(size);
 661     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 662     p[0] = '\0';
 663   }
 664 }
 665 
 666 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 667   assert(t!=NULL, "just checking");
 668   assert(t->osthread()->expanding_stack(), "expand should be set");
 669   assert(t->stack_base() != NULL, "stack_base was not initialized");
 670 
 671   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 672     sigset_t mask_all, old_sigset;
 673     sigfillset(&mask_all);
 674     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 675     _expand_stack_to(addr);
 676     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 677     return true;
 678   }
 679   return false;
 680 }
 681 
 682 //////////////////////////////////////////////////////////////////////////////
 683 // create new thread
 684 
 685 static address highest_vm_reserved_address();
 686 
 687 // check if it's safe to start a new thread
 688 static bool _thread_safety_check(Thread* thread) {
 689   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 690     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 691     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 692     //   allocated (MAP_FIXED) from high address space. Every thread stack
 693     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 694     //   it to other values if they rebuild LinuxThreads).
 695     //
 696     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 697     // the memory region has already been mmap'ed. That means if we have too
 698     // many threads and/or very large heap, eventually thread stack will
 699     // collide with heap.
 700     //
 701     // Here we try to prevent heap/stack collision by comparing current
 702     // stack bottom with the highest address that has been mmap'ed by JVM
 703     // plus a safety margin for memory maps created by native code.
 704     //
 705     // This feature can be disabled by setting ThreadSafetyMargin to 0
 706     //
 707     if (ThreadSafetyMargin > 0) {
 708       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 709 
 710       // not safe if our stack extends below the safety margin
 711       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 712     } else {
 713       return true;
 714     }
 715   } else {
 716     // Floating stack LinuxThreads or NPTL:
 717     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 718     //   there's not enough space left, pthread_create() will fail. If we come
 719     //   here, that means enough space has been reserved for stack.
 720     return true;
 721   }
 722 }
 723 
 724 // Thread start routine for all newly created threads
 725 static void *java_start(Thread *thread) {
 726   // Try to randomize the cache line index of hot stack frames.
 727   // This helps when threads of the same stack traces evict each other's
 728   // cache lines. The threads can be either from the same JVM instance, or
 729   // from different JVM instances. The benefit is especially true for
 730   // processors with hyperthreading technology.
 731   static int counter = 0;
 732   int pid = os::current_process_id();
 733   alloca(((pid ^ counter++) & 7) * 128);
 734 
 735   ThreadLocalStorage::set_thread(thread);
 736 
 737   OSThread* osthread = thread->osthread();
 738   Monitor* sync = osthread->startThread_lock();
 739 
 740   // non floating stack LinuxThreads needs extra check, see above
 741   if (!_thread_safety_check(thread)) {
 742     // notify parent thread
 743     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 744     osthread->set_state(ZOMBIE);
 745     sync->notify_all();
 746     return NULL;
 747   }
 748 
 749   // thread_id is kernel thread id (similar to Solaris LWP id)
 750   osthread->set_thread_id(os::Linux::gettid());
 751 
 752   if (UseNUMA) {
 753     int lgrp_id = os::numa_get_group_id();
 754     if (lgrp_id != -1) {
 755       thread->set_lgrp_id(lgrp_id);
 756     }
 757   }
 758   // initialize signal mask for this thread
 759   os::Linux::hotspot_sigmask(thread);
 760 
 761   // initialize floating point control register
 762   os::Linux::init_thread_fpu_state();
 763 
 764   // handshaking with parent thread
 765   {
 766     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 767 
 768     // notify parent thread
 769     osthread->set_state(INITIALIZED);
 770     sync->notify_all();
 771 
 772     // wait until os::start_thread()
 773     while (osthread->get_state() == INITIALIZED) {
 774       sync->wait(Mutex::_no_safepoint_check_flag);
 775     }
 776   }
 777 
 778   // call one more level start routine
 779   thread->run();
 780 
 781   return 0;
 782 }
 783 
 784 bool os::create_thread(Thread* thread, ThreadType thr_type,
 785                        size_t stack_size) {
 786   assert(thread->osthread() == NULL, "caller responsible");
 787 
 788   // Allocate the OSThread object
 789   OSThread* osthread = new OSThread(NULL, NULL);
 790   if (osthread == NULL) {
 791     return false;
 792   }
 793 
 794   // set the correct thread state
 795   osthread->set_thread_type(thr_type);
 796 
 797   // Initial state is ALLOCATED but not INITIALIZED
 798   osthread->set_state(ALLOCATED);
 799 
 800   thread->set_osthread(osthread);
 801 
 802   // init thread attributes
 803   pthread_attr_t attr;
 804   pthread_attr_init(&attr);
 805   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 806 
 807   // stack size
 808   if (os::Linux::supports_variable_stack_size()) {
 809     // calculate stack size if it's not specified by caller
 810     if (stack_size == 0) {
 811       stack_size = os::Linux::default_stack_size(thr_type);
 812 
 813       switch (thr_type) {
 814       case os::java_thread:
 815         // Java threads use ThreadStackSize which default value can be
 816         // changed with the flag -Xss
 817         assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 818         stack_size = JavaThread::stack_size_at_create();
 819         break;
 820       case os::compiler_thread:
 821         if (CompilerThreadStackSize > 0) {
 822           stack_size = (size_t)(CompilerThreadStackSize * K);
 823           break;
 824         } // else fall through:
 825           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 826       case os::vm_thread:
 827       case os::pgc_thread:
 828       case os::cgc_thread:
 829       case os::watcher_thread:
 830         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 831         break;
 832       }
 833     }
 834 
 835     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 836     pthread_attr_setstacksize(&attr, stack_size);
 837   } else {
 838     // let pthread_create() pick the default value.
 839   }
 840 
 841   // glibc guard page
 842   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 843 
 844   ThreadState state;
 845 
 846   {
 847     // Serialize thread creation if we are running with fixed stack LinuxThreads
 848     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 849     if (lock) {
 850       os::Linux::createThread_lock()->lock_without_safepoint_check();
 851     }
 852 
 853     pthread_t tid;
 854     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 855 
 856     pthread_attr_destroy(&attr);
 857 
 858     if (ret != 0) {
 859       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 860         perror("pthread_create()");
 861       }
 862       // Need to clean up stuff we've allocated so far
 863       thread->set_osthread(NULL);
 864       delete osthread;
 865       if (lock) os::Linux::createThread_lock()->unlock();
 866       return false;
 867     }
 868 
 869     // Store pthread info into the OSThread
 870     osthread->set_pthread_id(tid);
 871 
 872     // Wait until child thread is either initialized or aborted
 873     {
 874       Monitor* sync_with_child = osthread->startThread_lock();
 875       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 876       while ((state = osthread->get_state()) == ALLOCATED) {
 877         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 878       }
 879     }
 880 
 881     if (lock) {
 882       os::Linux::createThread_lock()->unlock();
 883     }
 884   }
 885 
 886   // Aborted due to thread limit being reached
 887   if (state == ZOMBIE) {
 888     thread->set_osthread(NULL);
 889     delete osthread;
 890     return false;
 891   }
 892 
 893   // The thread is returned suspended (in state INITIALIZED),
 894   // and is started higher up in the call chain
 895   assert(state == INITIALIZED, "race condition");
 896   return true;
 897 }
 898 
 899 /////////////////////////////////////////////////////////////////////////////
 900 // attach existing thread
 901 
 902 // bootstrap the main thread
 903 bool os::create_main_thread(JavaThread* thread) {
 904   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 905   return create_attached_thread(thread);
 906 }
 907 
 908 bool os::create_attached_thread(JavaThread* thread) {
 909 #ifdef ASSERT
 910   thread->verify_not_published();
 911 #endif
 912 
 913   // Allocate the OSThread object
 914   OSThread* osthread = new OSThread(NULL, NULL);
 915 
 916   if (osthread == NULL) {
 917     return false;
 918   }
 919 
 920   // Store pthread info into the OSThread
 921   osthread->set_thread_id(os::Linux::gettid());
 922   osthread->set_pthread_id(::pthread_self());
 923 
 924   // initialize floating point control register
 925   os::Linux::init_thread_fpu_state();
 926 
 927   // Initial thread state is RUNNABLE
 928   osthread->set_state(RUNNABLE);
 929 
 930   thread->set_osthread(osthread);
 931 
 932   if (UseNUMA) {
 933     int lgrp_id = os::numa_get_group_id();
 934     if (lgrp_id != -1) {
 935       thread->set_lgrp_id(lgrp_id);
 936     }
 937   }
 938 
 939   if (os::Linux::is_initial_thread()) {
 940     // If current thread is initial thread, its stack is mapped on demand,
 941     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 942     // the entire stack region to avoid SEGV in stack banging.
 943     // It is also useful to get around the heap-stack-gap problem on SuSE
 944     // kernel (see 4821821 for details). We first expand stack to the top
 945     // of yellow zone, then enable stack yellow zone (order is significant,
 946     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 947     // is no gap between the last two virtual memory regions.
 948 
 949     JavaThread *jt = (JavaThread *)thread;
 950     address addr = jt->stack_yellow_zone_base();
 951     assert(addr != NULL, "initialization problem?");
 952     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 953 
 954     osthread->set_expanding_stack();
 955     os::Linux::manually_expand_stack(jt, addr);
 956     osthread->clear_expanding_stack();
 957   }
 958 
 959   // initialize signal mask for this thread
 960   // and save the caller's signal mask
 961   os::Linux::hotspot_sigmask(thread);
 962 
 963   return true;
 964 }
 965 
 966 void os::pd_start_thread(Thread* thread) {
 967   OSThread * osthread = thread->osthread();
 968   assert(osthread->get_state() != INITIALIZED, "just checking");
 969   Monitor* sync_with_child = osthread->startThread_lock();
 970   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 971   sync_with_child->notify();
 972 }
 973 
 974 // Free Linux resources related to the OSThread
 975 void os::free_thread(OSThread* osthread) {
 976   assert(osthread != NULL, "osthread not set");
 977 
 978   if (Thread::current()->osthread() == osthread) {
 979     // Restore caller's signal mask
 980     sigset_t sigmask = osthread->caller_sigmask();
 981     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 982   }
 983 
 984   delete osthread;
 985 }
 986 
 987 //////////////////////////////////////////////////////////////////////////////
 988 // thread local storage
 989 
 990 // Restore the thread pointer if the destructor is called. This is in case
 991 // someone from JNI code sets up a destructor with pthread_key_create to run
 992 // detachCurrentThread on thread death. Unless we restore the thread pointer we
 993 // will hang or crash. When detachCurrentThread is called the key will be set
 994 // to null and we will not be called again. If detachCurrentThread is never
 995 // called we could loop forever depending on the pthread implementation.
 996 static void restore_thread_pointer(void* p) {
 997   Thread* thread = (Thread*) p;
 998   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
 999 }
1000 
1001 int os::allocate_thread_local_storage() {
1002   pthread_key_t key;
1003   int rslt = pthread_key_create(&key, restore_thread_pointer);
1004   assert(rslt == 0, "cannot allocate thread local storage");
1005   return (int)key;
1006 }
1007 
1008 // Note: This is currently not used by VM, as we don't destroy TLS key
1009 // on VM exit.
1010 void os::free_thread_local_storage(int index) {
1011   int rslt = pthread_key_delete((pthread_key_t)index);
1012   assert(rslt == 0, "invalid index");
1013 }
1014 
1015 void os::thread_local_storage_at_put(int index, void* value) {
1016   int rslt = pthread_setspecific((pthread_key_t)index, value);
1017   assert(rslt == 0, "pthread_setspecific failed");
1018 }
1019 
1020 extern "C" Thread* get_thread() {
1021   return ThreadLocalStorage::thread();
1022 }
1023 
1024 //////////////////////////////////////////////////////////////////////////////
1025 // initial thread
1026 
1027 // Check if current thread is the initial thread, similar to Solaris thr_main.
1028 bool os::Linux::is_initial_thread(void) {
1029   char dummy;
1030   // If called before init complete, thread stack bottom will be null.
1031   // Can be called if fatal error occurs before initialization.
1032   if (initial_thread_stack_bottom() == NULL) return false;
1033   assert(initial_thread_stack_bottom() != NULL &&
1034          initial_thread_stack_size()   != 0,
1035          "os::init did not locate initial thread's stack region");
1036   if ((address)&dummy >= initial_thread_stack_bottom() &&
1037       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1038     return true;
1039   } else {
1040     return false;
1041   }
1042 }
1043 
1044 // Find the virtual memory area that contains addr
1045 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1046   FILE *fp = fopen("/proc/self/maps", "r");
1047   if (fp) {
1048     address low, high;
1049     while (!feof(fp)) {
1050       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1051         if (low <= addr && addr < high) {
1052           if (vma_low)  *vma_low  = low;
1053           if (vma_high) *vma_high = high;
1054           fclose(fp);
1055           return true;
1056         }
1057       }
1058       for (;;) {
1059         int ch = fgetc(fp);
1060         if (ch == EOF || ch == (int)'\n') break;
1061       }
1062     }
1063     fclose(fp);
1064   }
1065   return false;
1066 }
1067 
1068 // Locate initial thread stack. This special handling of initial thread stack
1069 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1070 // bogus value for initial thread.
1071 void os::Linux::capture_initial_stack(size_t max_size) {
1072   // stack size is the easy part, get it from RLIMIT_STACK
1073   size_t stack_size;
1074   struct rlimit rlim;
1075   getrlimit(RLIMIT_STACK, &rlim);
1076   stack_size = rlim.rlim_cur;
1077 
1078   // 6308388: a bug in ld.so will relocate its own .data section to the
1079   //   lower end of primordial stack; reduce ulimit -s value a little bit
1080   //   so we won't install guard page on ld.so's data section.
1081   stack_size -= 2 * page_size();
1082 
1083   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1084   //   7.1, in both cases we will get 2G in return value.
1085   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1086   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1087   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1088   //   in case other parts in glibc still assumes 2M max stack size.
1089   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1090   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1091   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1092     stack_size = 2 * K * K IA64_ONLY(*2);
1093   }
1094   // Try to figure out where the stack base (top) is. This is harder.
1095   //
1096   // When an application is started, glibc saves the initial stack pointer in
1097   // a global variable "__libc_stack_end", which is then used by system
1098   // libraries. __libc_stack_end should be pretty close to stack top. The
1099   // variable is available since the very early days. However, because it is
1100   // a private interface, it could disappear in the future.
1101   //
1102   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1103   // to __libc_stack_end, it is very close to stack top, but isn't the real
1104   // stack top. Note that /proc may not exist if VM is running as a chroot
1105   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1106   // /proc/<pid>/stat could change in the future (though unlikely).
1107   //
1108   // We try __libc_stack_end first. If that doesn't work, look for
1109   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1110   // as a hint, which should work well in most cases.
1111 
1112   uintptr_t stack_start;
1113 
1114   // try __libc_stack_end first
1115   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1116   if (p && *p) {
1117     stack_start = *p;
1118   } else {
1119     // see if we can get the start_stack field from /proc/self/stat
1120     FILE *fp;
1121     int pid;
1122     char state;
1123     int ppid;
1124     int pgrp;
1125     int session;
1126     int nr;
1127     int tpgrp;
1128     unsigned long flags;
1129     unsigned long minflt;
1130     unsigned long cminflt;
1131     unsigned long majflt;
1132     unsigned long cmajflt;
1133     unsigned long utime;
1134     unsigned long stime;
1135     long cutime;
1136     long cstime;
1137     long prio;
1138     long nice;
1139     long junk;
1140     long it_real;
1141     uintptr_t start;
1142     uintptr_t vsize;
1143     intptr_t rss;
1144     uintptr_t rsslim;
1145     uintptr_t scodes;
1146     uintptr_t ecode;
1147     int i;
1148 
1149     // Figure what the primordial thread stack base is. Code is inspired
1150     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1151     // followed by command name surrounded by parentheses, state, etc.
1152     char stat[2048];
1153     int statlen;
1154 
1155     fp = fopen("/proc/self/stat", "r");
1156     if (fp) {
1157       statlen = fread(stat, 1, 2047, fp);
1158       stat[statlen] = '\0';
1159       fclose(fp);
1160 
1161       // Skip pid and the command string. Note that we could be dealing with
1162       // weird command names, e.g. user could decide to rename java launcher
1163       // to "java 1.4.2 :)", then the stat file would look like
1164       //                1234 (java 1.4.2 :)) R ... ...
1165       // We don't really need to know the command string, just find the last
1166       // occurrence of ")" and then start parsing from there. See bug 4726580.
1167       char * s = strrchr(stat, ')');
1168 
1169       i = 0;
1170       if (s) {
1171         // Skip blank chars
1172         do { s++; } while (s && isspace(*s));
1173 
1174 #define _UFM UINTX_FORMAT
1175 #define _DFM INTX_FORMAT
1176 
1177         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1178         //              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
1179         i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1180                    &state,          // 3  %c
1181                    &ppid,           // 4  %d
1182                    &pgrp,           // 5  %d
1183                    &session,        // 6  %d
1184                    &nr,             // 7  %d
1185                    &tpgrp,          // 8  %d
1186                    &flags,          // 9  %lu
1187                    &minflt,         // 10 %lu
1188                    &cminflt,        // 11 %lu
1189                    &majflt,         // 12 %lu
1190                    &cmajflt,        // 13 %lu
1191                    &utime,          // 14 %lu
1192                    &stime,          // 15 %lu
1193                    &cutime,         // 16 %ld
1194                    &cstime,         // 17 %ld
1195                    &prio,           // 18 %ld
1196                    &nice,           // 19 %ld
1197                    &junk,           // 20 %ld
1198                    &it_real,        // 21 %ld
1199                    &start,          // 22 UINTX_FORMAT
1200                    &vsize,          // 23 UINTX_FORMAT
1201                    &rss,            // 24 INTX_FORMAT
1202                    &rsslim,         // 25 UINTX_FORMAT
1203                    &scodes,         // 26 UINTX_FORMAT
1204                    &ecode,          // 27 UINTX_FORMAT
1205                    &stack_start);   // 28 UINTX_FORMAT
1206       }
1207 
1208 #undef _UFM
1209 #undef _DFM
1210 
1211       if (i != 28 - 2) {
1212         assert(false, "Bad conversion from /proc/self/stat");
1213         // product mode - assume we are the initial thread, good luck in the
1214         // embedded case.
1215         warning("Can't detect initial thread stack location - bad conversion");
1216         stack_start = (uintptr_t) &rlim;
1217       }
1218     } else {
1219       // For some reason we can't open /proc/self/stat (for example, running on
1220       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1221       // most cases, so don't abort:
1222       warning("Can't detect initial thread stack location - no /proc/self/stat");
1223       stack_start = (uintptr_t) &rlim;
1224     }
1225   }
1226 
1227   // Now we have a pointer (stack_start) very close to the stack top, the
1228   // next thing to do is to figure out the exact location of stack top. We
1229   // can find out the virtual memory area that contains stack_start by
1230   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1231   // and its upper limit is the real stack top. (again, this would fail if
1232   // running inside chroot, because /proc may not exist.)
1233 
1234   uintptr_t stack_top;
1235   address low, high;
1236   if (find_vma((address)stack_start, &low, &high)) {
1237     // success, "high" is the true stack top. (ignore "low", because initial
1238     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1239     stack_top = (uintptr_t)high;
1240   } else {
1241     // failed, likely because /proc/self/maps does not exist
1242     warning("Can't detect initial thread stack location - find_vma failed");
1243     // best effort: stack_start is normally within a few pages below the real
1244     // stack top, use it as stack top, and reduce stack size so we won't put
1245     // guard page outside stack.
1246     stack_top = stack_start;
1247     stack_size -= 16 * page_size();
1248   }
1249 
1250   // stack_top could be partially down the page so align it
1251   stack_top = align_size_up(stack_top, page_size());
1252 
1253   if (max_size && stack_size > max_size) {
1254     _initial_thread_stack_size = max_size;
1255   } else {
1256     _initial_thread_stack_size = stack_size;
1257   }
1258 
1259   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1260   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1261 }
1262 
1263 ////////////////////////////////////////////////////////////////////////////////
1264 // time support
1265 
1266 // Time since start-up in seconds to a fine granularity.
1267 // Used by VMSelfDestructTimer and the MemProfiler.
1268 double os::elapsedTime() {
1269 
1270   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1271 }
1272 
1273 jlong os::elapsed_counter() {
1274   return javaTimeNanos() - initial_time_count;
1275 }
1276 
1277 jlong os::elapsed_frequency() {
1278   return NANOSECS_PER_SEC; // nanosecond resolution
1279 }
1280 
1281 bool os::supports_vtime() { return true; }
1282 bool os::enable_vtime()   { return false; }
1283 bool os::vtime_enabled()  { return false; }
1284 
1285 double os::elapsedVTime() {
1286   struct rusage usage;
1287   int retval = getrusage(RUSAGE_THREAD, &usage);
1288   if (retval == 0) {
1289     return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1290   } else {
1291     // better than nothing, but not much
1292     return elapsedTime();
1293   }
1294 }
1295 
1296 jlong os::javaTimeMillis() {
1297   timeval time;
1298   int status = gettimeofday(&time, NULL);
1299   assert(status != -1, "linux error");
1300   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1301 }
1302 
1303 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1304   timeval time;
1305   int status = gettimeofday(&time, NULL);
1306   assert(status != -1, "linux error");
1307   seconds = jlong(time.tv_sec);
1308   nanos = jlong(time.tv_usec) * 1000;
1309 }
1310 
1311 
1312 #ifndef CLOCK_MONOTONIC
1313   #define CLOCK_MONOTONIC (1)
1314 #endif
1315 
1316 void os::Linux::clock_init() {
1317   // we do dlopen's in this particular order due to bug in linux
1318   // dynamical loader (see 6348968) leading to crash on exit
1319   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1320   if (handle == NULL) {
1321     handle = dlopen("librt.so", RTLD_LAZY);
1322   }
1323 
1324   if (handle) {
1325     int (*clock_getres_func)(clockid_t, struct timespec*) =
1326            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1327     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1328            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1329     if (clock_getres_func && clock_gettime_func) {
1330       // See if monotonic clock is supported by the kernel. Note that some
1331       // early implementations simply return kernel jiffies (updated every
1332       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1333       // for nano time (though the monotonic property is still nice to have).
1334       // It's fixed in newer kernels, however clock_getres() still returns
1335       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1336       // resolution for now. Hopefully as people move to new kernels, this
1337       // won't be a problem.
1338       struct timespec res;
1339       struct timespec tp;
1340       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1341           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1342         // yes, monotonic clock is supported
1343         _clock_gettime = clock_gettime_func;
1344         return;
1345       } else {
1346         // close librt if there is no monotonic clock
1347         dlclose(handle);
1348       }
1349     }
1350   }
1351   warning("No monotonic clock was available - timed services may " \
1352           "be adversely affected if the time-of-day clock changes");
1353 }
1354 
1355 #ifndef SYS_clock_getres
1356   #if defined(IA32) || defined(AMD64)
1357     #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1358     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1359   #else
1360     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1361     #define sys_clock_getres(x,y)  -1
1362   #endif
1363 #else
1364   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1365 #endif
1366 
1367 void os::Linux::fast_thread_clock_init() {
1368   if (!UseLinuxPosixThreadCPUClocks) {
1369     return;
1370   }
1371   clockid_t clockid;
1372   struct timespec tp;
1373   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1374       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1375 
1376   // Switch to using fast clocks for thread cpu time if
1377   // the sys_clock_getres() returns 0 error code.
1378   // Note, that some kernels may support the current thread
1379   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1380   // returned by the pthread_getcpuclockid().
1381   // If the fast Posix clocks are supported then the sys_clock_getres()
1382   // must return at least tp.tv_sec == 0 which means a resolution
1383   // better than 1 sec. This is extra check for reliability.
1384 
1385   if (pthread_getcpuclockid_func &&
1386       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1387       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1388     _supports_fast_thread_cpu_time = true;
1389     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1390   }
1391 }
1392 
1393 jlong os::javaTimeNanos() {
1394   if (os::supports_monotonic_clock()) {
1395     struct timespec tp;
1396     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1397     assert(status == 0, "gettime error");
1398     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1399     return result;
1400   } else {
1401     timeval time;
1402     int status = gettimeofday(&time, NULL);
1403     assert(status != -1, "linux error");
1404     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1405     return 1000 * usecs;
1406   }
1407 }
1408 
1409 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1410   if (os::supports_monotonic_clock()) {
1411     info_ptr->max_value = ALL_64_BITS;
1412 
1413     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1414     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1415     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1416   } else {
1417     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1418     info_ptr->max_value = ALL_64_BITS;
1419 
1420     // gettimeofday is a real time clock so it skips
1421     info_ptr->may_skip_backward = true;
1422     info_ptr->may_skip_forward = true;
1423   }
1424 
1425   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1426 }
1427 
1428 // Return the real, user, and system times in seconds from an
1429 // arbitrary fixed point in the past.
1430 bool os::getTimesSecs(double* process_real_time,
1431                       double* process_user_time,
1432                       double* process_system_time) {
1433   struct tms ticks;
1434   clock_t real_ticks = times(&ticks);
1435 
1436   if (real_ticks == (clock_t) (-1)) {
1437     return false;
1438   } else {
1439     double ticks_per_second = (double) clock_tics_per_sec;
1440     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1441     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1442     *process_real_time = ((double) real_ticks) / ticks_per_second;
1443 
1444     return true;
1445   }
1446 }
1447 
1448 
1449 char * os::local_time_string(char *buf, size_t buflen) {
1450   struct tm t;
1451   time_t long_time;
1452   time(&long_time);
1453   localtime_r(&long_time, &t);
1454   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1455                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1456                t.tm_hour, t.tm_min, t.tm_sec);
1457   return buf;
1458 }
1459 
1460 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1461   return localtime_r(clock, res);
1462 }
1463 
1464 ////////////////////////////////////////////////////////////////////////////////
1465 // runtime exit support
1466 
1467 // Note: os::shutdown() might be called very early during initialization, or
1468 // called from signal handler. Before adding something to os::shutdown(), make
1469 // sure it is async-safe and can handle partially initialized VM.
1470 void os::shutdown() {
1471 
1472   // allow PerfMemory to attempt cleanup of any persistent resources
1473   perfMemory_exit();
1474 
1475   // needs to remove object in file system
1476   AttachListener::abort();
1477 
1478   // flush buffered output, finish log files
1479   ostream_abort();
1480 
1481   // Check for abort hook
1482   abort_hook_t abort_hook = Arguments::abort_hook();
1483   if (abort_hook != NULL) {
1484     abort_hook();
1485   }
1486 
1487 }
1488 
1489 // Note: os::abort() might be called very early during initialization, or
1490 // called from signal handler. Before adding something to os::abort(), make
1491 // sure it is async-safe and can handle partially initialized VM.
1492 void os::abort(bool dump_core) {
1493   os::shutdown();
1494   if (dump_core) {
1495 #ifndef PRODUCT
1496     fdStream out(defaultStream::output_fd());
1497     out.print_raw("Current thread is ");
1498     char buf[16];
1499     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1500     out.print_raw_cr(buf);
1501     out.print_raw_cr("Dumping core ...");
1502 #endif
1503     ::abort(); // dump core
1504   }
1505 
1506   ::exit(1);
1507 }
1508 
1509 // Die immediately, no exit hook, no abort hook, no cleanup.
1510 void os::die() {
1511   // _exit() on LinuxThreads only kills current thread
1512   ::abort();
1513 }
1514 
1515 
1516 // This method is a copy of JDK's sysGetLastErrorString
1517 // from src/solaris/hpi/src/system_md.c
1518 
1519 size_t os::lasterror(char *buf, size_t len) {
1520   if (errno == 0)  return 0;
1521 
1522   const char *s = ::strerror(errno);
1523   size_t n = ::strlen(s);
1524   if (n >= len) {
1525     n = len - 1;
1526   }
1527   ::strncpy(buf, s, n);
1528   buf[n] = '\0';
1529   return n;
1530 }
1531 
1532 intx os::current_thread_id() { return (intx)pthread_self(); }
1533 int os::current_process_id() {
1534 
1535   // Under the old linux thread library, linux gives each thread
1536   // its own process id. Because of this each thread will return
1537   // a different pid if this method were to return the result
1538   // of getpid(2). Linux provides no api that returns the pid
1539   // of the launcher thread for the vm. This implementation
1540   // returns a unique pid, the pid of the launcher thread
1541   // that starts the vm 'process'.
1542 
1543   // Under the NPTL, getpid() returns the same pid as the
1544   // launcher thread rather than a unique pid per thread.
1545   // Use gettid() if you want the old pre NPTL behaviour.
1546 
1547   // if you are looking for the result of a call to getpid() that
1548   // returns a unique pid for the calling thread, then look at the
1549   // OSThread::thread_id() method in osThread_linux.hpp file
1550 
1551   return (int)(_initial_pid ? _initial_pid : getpid());
1552 }
1553 
1554 // DLL functions
1555 
1556 const char* os::dll_file_extension() { return ".so"; }
1557 
1558 // This must be hard coded because it's the system's temporary
1559 // directory not the java application's temp directory, ala java.io.tmpdir.
1560 const char* os::get_temp_directory() { return "/tmp"; }
1561 
1562 static bool file_exists(const char* filename) {
1563   struct stat statbuf;
1564   if (filename == NULL || strlen(filename) == 0) {
1565     return false;
1566   }
1567   return os::stat(filename, &statbuf) == 0;
1568 }
1569 
1570 bool os::dll_build_name(char* buffer, size_t buflen,
1571                         const char* pname, const char* fname) {
1572   bool retval = false;
1573   // Copied from libhpi
1574   const size_t pnamelen = pname ? strlen(pname) : 0;
1575 
1576   // Return error on buffer overflow.
1577   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1578     return retval;
1579   }
1580 
1581   if (pnamelen == 0) {
1582     snprintf(buffer, buflen, "lib%s.so", fname);
1583     retval = true;
1584   } else if (strchr(pname, *os::path_separator()) != NULL) {
1585     int n;
1586     char** pelements = split_path(pname, &n);
1587     if (pelements == NULL) {
1588       return false;
1589     }
1590     for (int i = 0; i < n; i++) {
1591       // Really shouldn't be NULL, but check can't hurt
1592       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1593         continue; // skip the empty path values
1594       }
1595       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1596       if (file_exists(buffer)) {
1597         retval = true;
1598         break;
1599       }
1600     }
1601     // release the storage
1602     for (int i = 0; i < n; i++) {
1603       if (pelements[i] != NULL) {
1604         FREE_C_HEAP_ARRAY(char, pelements[i]);
1605       }
1606     }
1607     if (pelements != NULL) {
1608       FREE_C_HEAP_ARRAY(char*, pelements);
1609     }
1610   } else {
1611     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1612     retval = true;
1613   }
1614   return retval;
1615 }
1616 
1617 // check if addr is inside libjvm.so
1618 bool os::address_is_in_vm(address addr) {
1619   static address libjvm_base_addr;
1620   Dl_info dlinfo;
1621 
1622   if (libjvm_base_addr == NULL) {
1623     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1624       libjvm_base_addr = (address)dlinfo.dli_fbase;
1625     }
1626     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1627   }
1628 
1629   if (dladdr((void *)addr, &dlinfo) != 0) {
1630     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1631   }
1632 
1633   return false;
1634 }
1635 
1636 bool os::dll_address_to_function_name(address addr, char *buf,
1637                                       int buflen, int *offset) {
1638   // buf is not optional, but offset is optional
1639   assert(buf != NULL, "sanity check");
1640 
1641   Dl_info dlinfo;
1642 
1643   if (dladdr((void*)addr, &dlinfo) != 0) {
1644     // see if we have a matching symbol
1645     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1646       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1647         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1648       }
1649       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1650       return true;
1651     }
1652     // no matching symbol so try for just file info
1653     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1654       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1655                           buf, buflen, offset, dlinfo.dli_fname)) {
1656         return true;
1657       }
1658     }
1659   }
1660 
1661   buf[0] = '\0';
1662   if (offset != NULL) *offset = -1;
1663   return false;
1664 }
1665 
1666 struct _address_to_library_name {
1667   address addr;          // input : memory address
1668   size_t  buflen;        //         size of fname
1669   char*   fname;         // output: library name
1670   address base;          //         library base addr
1671 };
1672 
1673 static int address_to_library_name_callback(struct dl_phdr_info *info,
1674                                             size_t size, void *data) {
1675   int i;
1676   bool found = false;
1677   address libbase = NULL;
1678   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1679 
1680   // iterate through all loadable segments
1681   for (i = 0; i < info->dlpi_phnum; i++) {
1682     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1683     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1684       // base address of a library is the lowest address of its loaded
1685       // segments.
1686       if (libbase == NULL || libbase > segbase) {
1687         libbase = segbase;
1688       }
1689       // see if 'addr' is within current segment
1690       if (segbase <= d->addr &&
1691           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1692         found = true;
1693       }
1694     }
1695   }
1696 
1697   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1698   // so dll_address_to_library_name() can fall through to use dladdr() which
1699   // can figure out executable name from argv[0].
1700   if (found && info->dlpi_name && info->dlpi_name[0]) {
1701     d->base = libbase;
1702     if (d->fname) {
1703       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1704     }
1705     return 1;
1706   }
1707   return 0;
1708 }
1709 
1710 bool os::dll_address_to_library_name(address addr, char* buf,
1711                                      int buflen, int* offset) {
1712   // buf is not optional, but offset is optional
1713   assert(buf != NULL, "sanity check");
1714 
1715   Dl_info dlinfo;
1716   struct _address_to_library_name data;
1717 
1718   // There is a bug in old glibc dladdr() implementation that it could resolve
1719   // to wrong library name if the .so file has a base address != NULL. Here
1720   // we iterate through the program headers of all loaded libraries to find
1721   // out which library 'addr' really belongs to. This workaround can be
1722   // removed once the minimum requirement for glibc is moved to 2.3.x.
1723   data.addr = addr;
1724   data.fname = buf;
1725   data.buflen = buflen;
1726   data.base = NULL;
1727   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1728 
1729   if (rslt) {
1730     // buf already contains library name
1731     if (offset) *offset = addr - data.base;
1732     return true;
1733   }
1734   if (dladdr((void*)addr, &dlinfo) != 0) {
1735     if (dlinfo.dli_fname != NULL) {
1736       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1737     }
1738     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1739       *offset = addr - (address)dlinfo.dli_fbase;
1740     }
1741     return true;
1742   }
1743 
1744   buf[0] = '\0';
1745   if (offset) *offset = -1;
1746   return false;
1747 }
1748 
1749 // Loads .dll/.so and
1750 // in case of error it checks if .dll/.so was built for the
1751 // same architecture as Hotspot is running on
1752 
1753 
1754 // Remember the stack's state. The Linux dynamic linker will change
1755 // the stack to 'executable' at most once, so we must safepoint only once.
1756 bool os::Linux::_stack_is_executable = false;
1757 
1758 // VM operation that loads a library.  This is necessary if stack protection
1759 // of the Java stacks can be lost during loading the library.  If we
1760 // do not stop the Java threads, they can stack overflow before the stacks
1761 // are protected again.
1762 class VM_LinuxDllLoad: public VM_Operation {
1763  private:
1764   const char *_filename;
1765   char *_ebuf;
1766   int _ebuflen;
1767   void *_lib;
1768  public:
1769   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1770     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1771   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1772   void doit() {
1773     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1774     os::Linux::_stack_is_executable = true;
1775   }
1776   void* loaded_library() { return _lib; }
1777 };
1778 
1779 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1780   void * result = NULL;
1781   bool load_attempted = false;
1782 
1783   // Check whether the library to load might change execution rights
1784   // of the stack. If they are changed, the protection of the stack
1785   // guard pages will be lost. We need a safepoint to fix this.
1786   //
1787   // See Linux man page execstack(8) for more info.
1788   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1789     ElfFile ef(filename);
1790     if (!ef.specifies_noexecstack()) {
1791       if (!is_init_completed()) {
1792         os::Linux::_stack_is_executable = true;
1793         // This is OK - No Java threads have been created yet, and hence no
1794         // stack guard pages to fix.
1795         //
1796         // This should happen only when you are building JDK7 using a very
1797         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1798         //
1799         // Dynamic loader will make all stacks executable after
1800         // this function returns, and will not do that again.
1801         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1802       } else {
1803         warning("You have loaded library %s which might have disabled stack guard. "
1804                 "The VM will try to fix the stack guard now.\n"
1805                 "It's highly recommended that you fix the library with "
1806                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1807                 filename);
1808 
1809         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1810         JavaThread *jt = JavaThread::current();
1811         if (jt->thread_state() != _thread_in_native) {
1812           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1813           // that requires ExecStack. Cannot enter safe point. Let's give up.
1814           warning("Unable to fix stack guard. Giving up.");
1815         } else {
1816           if (!LoadExecStackDllInVMThread) {
1817             // This is for the case where the DLL has an static
1818             // constructor function that executes JNI code. We cannot
1819             // load such DLLs in the VMThread.
1820             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1821           }
1822 
1823           ThreadInVMfromNative tiv(jt);
1824           debug_only(VMNativeEntryWrapper vew;)
1825 
1826           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1827           VMThread::execute(&op);
1828           if (LoadExecStackDllInVMThread) {
1829             result = op.loaded_library();
1830           }
1831           load_attempted = true;
1832         }
1833       }
1834     }
1835   }
1836 
1837   if (!load_attempted) {
1838     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1839   }
1840 
1841   if (result != NULL) {
1842     // Successful loading
1843     return result;
1844   }
1845 
1846   Elf32_Ehdr elf_head;
1847   int diag_msg_max_length=ebuflen-strlen(ebuf);
1848   char* diag_msg_buf=ebuf+strlen(ebuf);
1849 
1850   if (diag_msg_max_length==0) {
1851     // No more space in ebuf for additional diagnostics message
1852     return NULL;
1853   }
1854 
1855 
1856   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1857 
1858   if (file_descriptor < 0) {
1859     // Can't open library, report dlerror() message
1860     return NULL;
1861   }
1862 
1863   bool failed_to_read_elf_head=
1864     (sizeof(elf_head)!=
1865      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1866 
1867   ::close(file_descriptor);
1868   if (failed_to_read_elf_head) {
1869     // file i/o error - report dlerror() msg
1870     return NULL;
1871   }
1872 
1873   typedef struct {
1874     Elf32_Half  code;         // Actual value as defined in elf.h
1875     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1876     char        elf_class;    // 32 or 64 bit
1877     char        endianess;    // MSB or LSB
1878     char*       name;         // String representation
1879   } arch_t;
1880 
1881 #ifndef EM_486
1882   #define EM_486          6               /* Intel 80486 */
1883 #endif
1884 #ifndef EM_AARCH64
1885   #define EM_AARCH64    183               /* ARM AARCH64 */
1886 #endif
1887 
1888   static const arch_t arch_array[]={
1889     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1890     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1891     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1892     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1893     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1894     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1895     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1896     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1897 #if defined(VM_LITTLE_ENDIAN)
1898     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1899 #else
1900     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1901 #endif
1902     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1903     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1904     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1905     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1906     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1907     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1908     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1909     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1910   };
1911 
1912 #if  (defined IA32)
1913   static  Elf32_Half running_arch_code=EM_386;
1914 #elif   (defined AMD64)
1915   static  Elf32_Half running_arch_code=EM_X86_64;
1916 #elif  (defined IA64)
1917   static  Elf32_Half running_arch_code=EM_IA_64;
1918 #elif  (defined __sparc) && (defined _LP64)
1919   static  Elf32_Half running_arch_code=EM_SPARCV9;
1920 #elif  (defined __sparc) && (!defined _LP64)
1921   static  Elf32_Half running_arch_code=EM_SPARC;
1922 #elif  (defined __powerpc64__)
1923   static  Elf32_Half running_arch_code=EM_PPC64;
1924 #elif  (defined __powerpc__)
1925   static  Elf32_Half running_arch_code=EM_PPC;
1926 #elif  (defined ARM)
1927   static  Elf32_Half running_arch_code=EM_ARM;
1928 #elif  (defined S390)
1929   static  Elf32_Half running_arch_code=EM_S390;
1930 #elif  (defined ALPHA)
1931   static  Elf32_Half running_arch_code=EM_ALPHA;
1932 #elif  (defined MIPSEL)
1933   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1934 #elif  (defined PARISC)
1935   static  Elf32_Half running_arch_code=EM_PARISC;
1936 #elif  (defined MIPS)
1937   static  Elf32_Half running_arch_code=EM_MIPS;
1938 #elif  (defined M68K)
1939   static  Elf32_Half running_arch_code=EM_68K;
1940 #elif  (defined AARCH64)
1941   static  Elf32_Half running_arch_code=EM_AARCH64;
1942 #else
1943     #error Method os::dll_load requires that one of following is defined:\
1944          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1945 #endif
1946 
1947   // Identify compatability class for VM's architecture and library's architecture
1948   // Obtain string descriptions for architectures
1949 
1950   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1951   int running_arch_index=-1;
1952 
1953   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1954     if (running_arch_code == arch_array[i].code) {
1955       running_arch_index    = i;
1956     }
1957     if (lib_arch.code == arch_array[i].code) {
1958       lib_arch.compat_class = arch_array[i].compat_class;
1959       lib_arch.name         = arch_array[i].name;
1960     }
1961   }
1962 
1963   assert(running_arch_index != -1,
1964          "Didn't find running architecture code (running_arch_code) in arch_array");
1965   if (running_arch_index == -1) {
1966     // Even though running architecture detection failed
1967     // we may still continue with reporting dlerror() message
1968     return NULL;
1969   }
1970 
1971   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1972     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1973     return NULL;
1974   }
1975 
1976 #ifndef S390
1977   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1978     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1979     return NULL;
1980   }
1981 #endif // !S390
1982 
1983   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1984     if (lib_arch.name!=NULL) {
1985       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1986                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1987                  lib_arch.name, arch_array[running_arch_index].name);
1988     } else {
1989       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1990                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1991                  lib_arch.code,
1992                  arch_array[running_arch_index].name);
1993     }
1994   }
1995 
1996   return NULL;
1997 }
1998 
1999 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
2000                                 int ebuflen) {
2001   void * result = ::dlopen(filename, RTLD_LAZY);
2002   if (result == NULL) {
2003     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2004     ebuf[ebuflen-1] = '\0';
2005   }
2006   return result;
2007 }
2008 
2009 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2010                                        int ebuflen) {
2011   void * result = NULL;
2012   if (LoadExecStackDllInVMThread) {
2013     result = dlopen_helper(filename, ebuf, ebuflen);
2014   }
2015 
2016   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2017   // library that requires an executable stack, or which does not have this
2018   // stack attribute set, dlopen changes the stack attribute to executable. The
2019   // read protection of the guard pages gets lost.
2020   //
2021   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2022   // may have been queued at the same time.
2023 
2024   if (!_stack_is_executable) {
2025     JavaThread *jt = Threads::first();
2026 
2027     while (jt) {
2028       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2029           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2030         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2031                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2032           warning("Attempt to reguard stack yellow zone failed.");
2033         }
2034       }
2035       jt = jt->next();
2036     }
2037   }
2038 
2039   return result;
2040 }
2041 
2042 // glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2043 // chances are you might want to run the generated bits against glibc-2.0
2044 // libdl.so, so always use locking for any version of glibc.
2045 //
2046 void* os::dll_lookup(void* handle, const char* name) {
2047   pthread_mutex_lock(&dl_mutex);
2048   void* res = dlsym(handle, name);
2049   pthread_mutex_unlock(&dl_mutex);
2050   return res;
2051 }
2052 
2053 void* os::get_default_process_handle() {
2054   return (void*)::dlopen(NULL, RTLD_LAZY);
2055 }
2056 
2057 static bool _print_ascii_file(const char* filename, outputStream* st) {
2058   int fd = ::open(filename, O_RDONLY);
2059   if (fd == -1) {
2060     return false;
2061   }
2062 
2063   char buf[32];
2064   int bytes;
2065   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2066     st->print_raw(buf, bytes);
2067   }
2068 
2069   ::close(fd);
2070 
2071   return true;
2072 }
2073 
2074 void os::print_dll_info(outputStream *st) {
2075   st->print_cr("Dynamic libraries:");
2076 
2077   char fname[32];
2078   pid_t pid = os::Linux::gettid();
2079 
2080   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2081 
2082   if (!_print_ascii_file(fname, st)) {
2083     st->print("Can not get library information for pid = %d\n", pid);
2084   }
2085 }
2086 
2087 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2088   FILE *procmapsFile = NULL;
2089 
2090   // Open the procfs maps file for the current process
2091   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2092     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2093     char line[PATH_MAX + 100];
2094 
2095     // Read line by line from 'file'
2096     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2097       u8 base, top, offset, inode;
2098       char permissions[5];
2099       char device[6];
2100       char name[PATH_MAX + 1];
2101 
2102       // Parse fields from line
2103       sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2104 
2105       // Filter by device id '00:00' so that we only get file system mapped files.
2106       if (strcmp(device, "00:00") != 0) {
2107 
2108         // Call callback with the fields of interest
2109         if(callback(name, (address)base, (address)top, param)) {
2110           // Oops abort, callback aborted
2111           fclose(procmapsFile);
2112           return 1;
2113         }
2114       }
2115     }
2116     fclose(procmapsFile);
2117   }
2118   return 0;
2119 }
2120 
2121 void os::print_os_info_brief(outputStream* st) {
2122   os::Linux::print_distro_info(st);
2123 
2124   os::Posix::print_uname_info(st);
2125 
2126   os::Linux::print_libversion_info(st);
2127 
2128 }
2129 
2130 void os::print_os_info(outputStream* st) {
2131   st->print("OS:");
2132 
2133   os::Linux::print_distro_info(st);
2134 
2135   os::Posix::print_uname_info(st);
2136 
2137   // Print warning if unsafe chroot environment detected
2138   if (unsafe_chroot_detected) {
2139     st->print("WARNING!! ");
2140     st->print_cr("%s", unstable_chroot_error);
2141   }
2142 
2143   os::Linux::print_libversion_info(st);
2144 
2145   os::Posix::print_rlimit_info(st);
2146 
2147   os::Posix::print_load_average(st);
2148 
2149   os::Linux::print_full_memory_info(st);
2150 }
2151 
2152 // Try to identify popular distros.
2153 // Most Linux distributions have a /etc/XXX-release file, which contains
2154 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2155 // file that also contains the OS version string. Some have more than one
2156 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2157 // /etc/redhat-release.), so the order is important.
2158 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2159 // their own specific XXX-release file as well as a redhat-release file.
2160 // Because of this the XXX-release file needs to be searched for before the
2161 // redhat-release file.
2162 // Since Red Hat has a lsb-release file that is not very descriptive the
2163 // search for redhat-release needs to be before lsb-release.
2164 // Since the lsb-release file is the new standard it needs to be searched
2165 // before the older style release files.
2166 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2167 // next to last resort.  The os-release file is a new standard that contains
2168 // distribution information and the system-release file seems to be an old
2169 // standard that has been replaced by the lsb-release and os-release files.
2170 // Searching for the debian_version file is the last resort.  It contains
2171 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2172 // "Debian " is printed before the contents of the debian_version file.
2173 void os::Linux::print_distro_info(outputStream* st) {
2174   if (!_print_ascii_file("/etc/oracle-release", st) &&
2175       !_print_ascii_file("/etc/mandriva-release", st) &&
2176       !_print_ascii_file("/etc/mandrake-release", st) &&
2177       !_print_ascii_file("/etc/sun-release", st) &&
2178       !_print_ascii_file("/etc/redhat-release", st) &&
2179       !_print_ascii_file("/etc/lsb-release", st) &&
2180       !_print_ascii_file("/etc/SuSE-release", st) &&
2181       !_print_ascii_file("/etc/turbolinux-release", st) &&
2182       !_print_ascii_file("/etc/gentoo-release", st) &&
2183       !_print_ascii_file("/etc/ltib-release", st) &&
2184       !_print_ascii_file("/etc/angstrom-version", st) &&
2185       !_print_ascii_file("/etc/system-release", st) &&
2186       !_print_ascii_file("/etc/os-release", st)) {
2187 
2188     if (file_exists("/etc/debian_version")) {
2189       st->print("Debian ");
2190       _print_ascii_file("/etc/debian_version", st);
2191     } else {
2192       st->print("Linux");
2193     }
2194   }
2195   st->cr();
2196 }
2197 
2198 void os::Linux::print_libversion_info(outputStream* st) {
2199   // libc, pthread
2200   st->print("libc:");
2201   st->print("%s ", os::Linux::glibc_version());
2202   st->print("%s ", os::Linux::libpthread_version());
2203   if (os::Linux::is_LinuxThreads()) {
2204     st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2205   }
2206   st->cr();
2207 }
2208 
2209 void os::Linux::print_full_memory_info(outputStream* st) {
2210   st->print("\n/proc/meminfo:\n");
2211   _print_ascii_file("/proc/meminfo", st);
2212   st->cr();
2213 }
2214 
2215 void os::print_memory_info(outputStream* st) {
2216 
2217   st->print("Memory:");
2218   st->print(" %dk page", os::vm_page_size()>>10);
2219 
2220   // values in struct sysinfo are "unsigned long"
2221   struct sysinfo si;
2222   sysinfo(&si);
2223 
2224   st->print(", physical " UINT64_FORMAT "k",
2225             os::physical_memory() >> 10);
2226   st->print("(" UINT64_FORMAT "k free)",
2227             os::available_memory() >> 10);
2228   st->print(", swap " UINT64_FORMAT "k",
2229             ((jlong)si.totalswap * si.mem_unit) >> 10);
2230   st->print("(" UINT64_FORMAT "k free)",
2231             ((jlong)si.freeswap * si.mem_unit) >> 10);
2232   st->cr();
2233 }
2234 
2235 void os::pd_print_cpu_info(outputStream* st) {
2236   st->print("\n/proc/cpuinfo:\n");
2237   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2238     st->print("  <Not Available>");
2239   }
2240   st->cr();
2241 }
2242 
2243 void os::print_siginfo(outputStream* st, void* siginfo) {
2244   const siginfo_t* si = (const siginfo_t*)siginfo;
2245 
2246   os::Posix::print_siginfo_brief(st, si);
2247 #if INCLUDE_CDS
2248   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2249       UseSharedSpaces) {
2250     FileMapInfo* mapinfo = FileMapInfo::current_info();
2251     if (mapinfo->is_in_shared_space(si->si_addr)) {
2252       st->print("\n\nError accessing class data sharing archive."   \
2253                 " Mapped file inaccessible during execution, "      \
2254                 " possible disk/network problem.");
2255     }
2256   }
2257 #endif
2258   st->cr();
2259 }
2260 
2261 
2262 static void print_signal_handler(outputStream* st, int sig,
2263                                  char* buf, size_t buflen);
2264 
2265 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2266   st->print_cr("Signal Handlers:");
2267   print_signal_handler(st, SIGSEGV, buf, buflen);
2268   print_signal_handler(st, SIGBUS , buf, buflen);
2269   print_signal_handler(st, SIGFPE , buf, buflen);
2270   print_signal_handler(st, SIGPIPE, buf, buflen);
2271   print_signal_handler(st, SIGXFSZ, buf, buflen);
2272   print_signal_handler(st, SIGILL , buf, buflen);
2273   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2274   print_signal_handler(st, SR_signum, buf, buflen);
2275   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2276   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2277   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2278   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2279 #if defined(PPC64)
2280   print_signal_handler(st, SIGTRAP, buf, buflen);
2281 #endif
2282 }
2283 
2284 static char saved_jvm_path[MAXPATHLEN] = {0};
2285 
2286 // Find the full path to the current module, libjvm.so
2287 void os::jvm_path(char *buf, jint buflen) {
2288   // Error checking.
2289   if (buflen < MAXPATHLEN) {
2290     assert(false, "must use a large-enough buffer");
2291     buf[0] = '\0';
2292     return;
2293   }
2294   // Lazy resolve the path to current module.
2295   if (saved_jvm_path[0] != 0) {
2296     strcpy(buf, saved_jvm_path);
2297     return;
2298   }
2299 
2300   char dli_fname[MAXPATHLEN];
2301   bool ret = dll_address_to_library_name(
2302                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2303                                          dli_fname, sizeof(dli_fname), NULL);
2304   assert(ret, "cannot locate libjvm");
2305   char *rp = NULL;
2306   if (ret && dli_fname[0] != '\0') {
2307     rp = realpath(dli_fname, buf);
2308   }
2309   if (rp == NULL) {
2310     return;
2311   }
2312 
2313   if (Arguments::sun_java_launcher_is_altjvm()) {
2314     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2315     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2316     // If "/jre/lib/" appears at the right place in the string, then
2317     // assume we are installed in a JDK and we're done. Otherwise, check
2318     // for a JAVA_HOME environment variable and fix up the path so it
2319     // looks like libjvm.so is installed there (append a fake suffix
2320     // hotspot/libjvm.so).
2321     const char *p = buf + strlen(buf) - 1;
2322     for (int count = 0; p > buf && count < 5; ++count) {
2323       for (--p; p > buf && *p != '/'; --p)
2324         /* empty */ ;
2325     }
2326 
2327     if (strncmp(p, "/jre/lib/", 9) != 0) {
2328       // Look for JAVA_HOME in the environment.
2329       char* java_home_var = ::getenv("JAVA_HOME");
2330       if (java_home_var != NULL && java_home_var[0] != 0) {
2331         char* jrelib_p;
2332         int len;
2333 
2334         // Check the current module name "libjvm.so".
2335         p = strrchr(buf, '/');
2336         if (p == NULL) {
2337           return;
2338         }
2339         assert(strstr(p, "/libjvm") == p, "invalid library name");
2340 
2341         rp = realpath(java_home_var, buf);
2342         if (rp == NULL) {
2343           return;
2344         }
2345 
2346         // determine if this is a legacy image or modules image
2347         // modules image doesn't have "jre" subdirectory
2348         len = strlen(buf);
2349         assert(len < buflen, "Ran out of buffer room");
2350         jrelib_p = buf + len;
2351         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2352         if (0 != access(buf, F_OK)) {
2353           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2354         }
2355 
2356         if (0 == access(buf, F_OK)) {
2357           // Use current module name "libjvm.so"
2358           len = strlen(buf);
2359           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2360         } else {
2361           // Go back to path of .so
2362           rp = realpath(dli_fname, buf);
2363           if (rp == NULL) {
2364             return;
2365           }
2366         }
2367       }
2368     }
2369   }
2370 
2371   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2372   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2373 }
2374 
2375 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2376   // no prefix required, not even "_"
2377 }
2378 
2379 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2380   // no suffix required
2381 }
2382 
2383 ////////////////////////////////////////////////////////////////////////////////
2384 // sun.misc.Signal support
2385 
2386 static volatile jint sigint_count = 0;
2387 
2388 static void UserHandler(int sig, void *siginfo, void *context) {
2389   // 4511530 - sem_post is serialized and handled by the manager thread. When
2390   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2391   // don't want to flood the manager thread with sem_post requests.
2392   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2393     return;
2394   }
2395 
2396   // Ctrl-C is pressed during error reporting, likely because the error
2397   // handler fails to abort. Let VM die immediately.
2398   if (sig == SIGINT && is_error_reported()) {
2399     os::die();
2400   }
2401 
2402   os::signal_notify(sig);
2403 }
2404 
2405 void* os::user_handler() {
2406   return CAST_FROM_FN_PTR(void*, UserHandler);
2407 }
2408 
2409 class Semaphore : public StackObj {
2410  public:
2411   Semaphore();
2412   ~Semaphore();
2413   void signal();
2414   void wait();
2415   bool trywait();
2416   bool timedwait(unsigned int sec, int nsec);
2417  private:
2418   sem_t _semaphore;
2419 };
2420 
2421 Semaphore::Semaphore() {
2422   sem_init(&_semaphore, 0, 0);
2423 }
2424 
2425 Semaphore::~Semaphore() {
2426   sem_destroy(&_semaphore);
2427 }
2428 
2429 void Semaphore::signal() {
2430   sem_post(&_semaphore);
2431 }
2432 
2433 void Semaphore::wait() {
2434   sem_wait(&_semaphore);
2435 }
2436 
2437 bool Semaphore::trywait() {
2438   return sem_trywait(&_semaphore) == 0;
2439 }
2440 
2441 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2442 
2443   struct timespec ts;
2444   // Semaphore's are always associated with CLOCK_REALTIME
2445   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2446   // see unpackTime for discussion on overflow checking
2447   if (sec >= MAX_SECS) {
2448     ts.tv_sec += MAX_SECS;
2449     ts.tv_nsec = 0;
2450   } else {
2451     ts.tv_sec += sec;
2452     ts.tv_nsec += nsec;
2453     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2454       ts.tv_nsec -= NANOSECS_PER_SEC;
2455       ++ts.tv_sec; // note: this must be <= max_secs
2456     }
2457   }
2458 
2459   while (1) {
2460     int result = sem_timedwait(&_semaphore, &ts);
2461     if (result == 0) {
2462       return true;
2463     } else if (errno == EINTR) {
2464       continue;
2465     } else if (errno == ETIMEDOUT) {
2466       return false;
2467     } else {
2468       return false;
2469     }
2470   }
2471 }
2472 
2473 extern "C" {
2474   typedef void (*sa_handler_t)(int);
2475   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2476 }
2477 
2478 void* os::signal(int signal_number, void* handler) {
2479   struct sigaction sigAct, oldSigAct;
2480 
2481   sigfillset(&(sigAct.sa_mask));
2482   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2483   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2484 
2485   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2486     // -1 means registration failed
2487     return (void *)-1;
2488   }
2489 
2490   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2491 }
2492 
2493 void os::signal_raise(int signal_number) {
2494   ::raise(signal_number);
2495 }
2496 
2497 // The following code is moved from os.cpp for making this
2498 // code platform specific, which it is by its very nature.
2499 
2500 // Will be modified when max signal is changed to be dynamic
2501 int os::sigexitnum_pd() {
2502   return NSIG;
2503 }
2504 
2505 // a counter for each possible signal value
2506 static volatile jint pending_signals[NSIG+1] = { 0 };
2507 
2508 // Linux(POSIX) specific hand shaking semaphore.
2509 static sem_t sig_sem;
2510 static Semaphore sr_semaphore;
2511 
2512 void os::signal_init_pd() {
2513   // Initialize signal structures
2514   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2515 
2516   // Initialize signal semaphore
2517   ::sem_init(&sig_sem, 0, 0);
2518 }
2519 
2520 void os::signal_notify(int sig) {
2521   Atomic::inc(&pending_signals[sig]);
2522   ::sem_post(&sig_sem);
2523 }
2524 
2525 static int check_pending_signals(bool wait) {
2526   Atomic::store(0, &sigint_count);
2527   for (;;) {
2528     for (int i = 0; i < NSIG + 1; i++) {
2529       jint n = pending_signals[i];
2530       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2531         return i;
2532       }
2533     }
2534     if (!wait) {
2535       return -1;
2536     }
2537     JavaThread *thread = JavaThread::current();
2538     ThreadBlockInVM tbivm(thread);
2539 
2540     bool threadIsSuspended;
2541     do {
2542       thread->set_suspend_equivalent();
2543       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2544       ::sem_wait(&sig_sem);
2545 
2546       // were we externally suspended while we were waiting?
2547       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2548       if (threadIsSuspended) {
2549         // The semaphore has been incremented, but while we were waiting
2550         // another thread suspended us. We don't want to continue running
2551         // while suspended because that would surprise the thread that
2552         // suspended us.
2553         ::sem_post(&sig_sem);
2554 
2555         thread->java_suspend_self();
2556       }
2557     } while (threadIsSuspended);
2558   }
2559 }
2560 
2561 int os::signal_lookup() {
2562   return check_pending_signals(false);
2563 }
2564 
2565 int os::signal_wait() {
2566   return check_pending_signals(true);
2567 }
2568 
2569 ////////////////////////////////////////////////////////////////////////////////
2570 // Virtual Memory
2571 
2572 int os::vm_page_size() {
2573   // Seems redundant as all get out
2574   assert(os::Linux::page_size() != -1, "must call os::init");
2575   return os::Linux::page_size();
2576 }
2577 
2578 // Solaris allocates memory by pages.
2579 int os::vm_allocation_granularity() {
2580   assert(os::Linux::page_size() != -1, "must call os::init");
2581   return os::Linux::page_size();
2582 }
2583 
2584 // Rationale behind this function:
2585 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2586 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2587 //  samples for JITted code. Here we create private executable mapping over the code cache
2588 //  and then we can use standard (well, almost, as mapping can change) way to provide
2589 //  info for the reporting script by storing timestamp and location of symbol
2590 void linux_wrap_code(char* base, size_t size) {
2591   static volatile jint cnt = 0;
2592 
2593   if (!UseOprofile) {
2594     return;
2595   }
2596 
2597   char buf[PATH_MAX+1];
2598   int num = Atomic::add(1, &cnt);
2599 
2600   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2601            os::get_temp_directory(), os::current_process_id(), num);
2602   unlink(buf);
2603 
2604   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2605 
2606   if (fd != -1) {
2607     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2608     if (rv != (off_t)-1) {
2609       if (::write(fd, "", 1) == 1) {
2610         mmap(base, size,
2611              PROT_READ|PROT_WRITE|PROT_EXEC,
2612              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2613       }
2614     }
2615     ::close(fd);
2616     unlink(buf);
2617   }
2618 }
2619 
2620 static bool recoverable_mmap_error(int err) {
2621   // See if the error is one we can let the caller handle. This
2622   // list of errno values comes from JBS-6843484. I can't find a
2623   // Linux man page that documents this specific set of errno
2624   // values so while this list currently matches Solaris, it may
2625   // change as we gain experience with this failure mode.
2626   switch (err) {
2627   case EBADF:
2628   case EINVAL:
2629   case ENOTSUP:
2630     // let the caller deal with these errors
2631     return true;
2632 
2633   default:
2634     // Any remaining errors on this OS can cause our reserved mapping
2635     // to be lost. That can cause confusion where different data
2636     // structures think they have the same memory mapped. The worst
2637     // scenario is if both the VM and a library think they have the
2638     // same memory mapped.
2639     return false;
2640   }
2641 }
2642 
2643 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2644                                     int err) {
2645   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2646           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2647           strerror(err), err);
2648 }
2649 
2650 static void warn_fail_commit_memory(char* addr, size_t size,
2651                                     size_t alignment_hint, bool exec,
2652                                     int err) {
2653   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2654           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2655           alignment_hint, exec, strerror(err), err);
2656 }
2657 
2658 // NOTE: Linux kernel does not really reserve the pages for us.
2659 //       All it does is to check if there are enough free pages
2660 //       left at the time of mmap(). This could be a potential
2661 //       problem.
2662 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2663   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2664   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2665                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2666   if (res != (uintptr_t) MAP_FAILED) {
2667     if (UseNUMAInterleaving) {
2668       numa_make_global(addr, size);
2669     }
2670     return 0;
2671   }
2672 
2673   int err = errno;  // save errno from mmap() call above
2674 
2675   if (!recoverable_mmap_error(err)) {
2676     warn_fail_commit_memory(addr, size, exec, err);
2677     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2678   }
2679 
2680   return err;
2681 }
2682 
2683 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2684   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2685 }
2686 
2687 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2688                                   const char* mesg) {
2689   assert(mesg != NULL, "mesg must be specified");
2690   int err = os::Linux::commit_memory_impl(addr, size, exec);
2691   if (err != 0) {
2692     // the caller wants all commit errors to exit with the specified mesg:
2693     warn_fail_commit_memory(addr, size, exec, err);
2694     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2695   }
2696 }
2697 
2698 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2699 #ifndef MAP_HUGETLB
2700   #define MAP_HUGETLB 0x40000
2701 #endif
2702 
2703 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2704 #ifndef MADV_HUGEPAGE
2705   #define MADV_HUGEPAGE 14
2706 #endif
2707 
2708 int os::Linux::commit_memory_impl(char* addr, size_t size,
2709                                   size_t alignment_hint, bool exec) {
2710   int err = os::Linux::commit_memory_impl(addr, size, exec);
2711   if (err == 0) {
2712     realign_memory(addr, size, alignment_hint);
2713   }
2714   return err;
2715 }
2716 
2717 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2718                           bool exec) {
2719   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2720 }
2721 
2722 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2723                                   size_t alignment_hint, bool exec,
2724                                   const char* mesg) {
2725   assert(mesg != NULL, "mesg must be specified");
2726   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2727   if (err != 0) {
2728     // the caller wants all commit errors to exit with the specified mesg:
2729     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2730     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2731   }
2732 }
2733 
2734 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2735   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2736     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2737     // be supported or the memory may already be backed by huge pages.
2738     ::madvise(addr, bytes, MADV_HUGEPAGE);
2739   }
2740 }
2741 
2742 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2743   // This method works by doing an mmap over an existing mmaping and effectively discarding
2744   // the existing pages. However it won't work for SHM-based large pages that cannot be
2745   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2746   // small pages on top of the SHM segment. This method always works for small pages, so we
2747   // allow that in any case.
2748   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2749     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2750   }
2751 }
2752 
2753 void os::numa_make_global(char *addr, size_t bytes) {
2754   Linux::numa_interleave_memory(addr, bytes);
2755 }
2756 
2757 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2758 // bind policy to MPOL_PREFERRED for the current thread.
2759 #define USE_MPOL_PREFERRED 0
2760 
2761 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2762   // To make NUMA and large pages more robust when both enabled, we need to ease
2763   // the requirements on where the memory should be allocated. MPOL_BIND is the
2764   // default policy and it will force memory to be allocated on the specified
2765   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2766   // the specified node, but will not force it. Using this policy will prevent
2767   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2768   // free large pages.
2769   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2770   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2771 }
2772 
2773 bool os::numa_topology_changed() { return false; }
2774 
2775 size_t os::numa_get_groups_num() {
2776   int max_node = Linux::numa_max_node();
2777   return max_node > 0 ? max_node + 1 : 1;
2778 }
2779 
2780 int os::numa_get_group_id() {
2781   int cpu_id = Linux::sched_getcpu();
2782   if (cpu_id != -1) {
2783     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2784     if (lgrp_id != -1) {
2785       return lgrp_id;
2786     }
2787   }
2788   return 0;
2789 }
2790 
2791 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2792   for (size_t i = 0; i < size; i++) {
2793     ids[i] = i;
2794   }
2795   return size;
2796 }
2797 
2798 bool os::get_page_info(char *start, page_info* info) {
2799   return false;
2800 }
2801 
2802 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2803                      page_info* page_found) {
2804   return end;
2805 }
2806 
2807 
2808 int os::Linux::sched_getcpu_syscall(void) {
2809   unsigned int cpu;
2810   int retval = -1;
2811 
2812 #if defined(IA32)
2813   #ifndef SYS_getcpu
2814     #define SYS_getcpu 318
2815   #endif
2816   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2817 #elif defined(AMD64)
2818 // Unfortunately we have to bring all these macros here from vsyscall.h
2819 // to be able to compile on old linuxes.
2820   #define __NR_vgetcpu 2
2821   #define VSYSCALL_START (-10UL << 20)
2822   #define VSYSCALL_SIZE 1024
2823   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2824   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2825   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2826   retval = vgetcpu(&cpu, NULL, NULL);
2827 #endif
2828 
2829   return (retval == -1) ? retval : cpu;
2830 }
2831 
2832 // Something to do with the numa-aware allocator needs these symbols
2833 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2834 extern "C" JNIEXPORT void numa_error(char *where) { }
2835 extern "C" JNIEXPORT int fork1() { return fork(); }
2836 
2837 
2838 // If we are running with libnuma version > 2, then we should
2839 // be trying to use symbols with versions 1.1
2840 // If we are running with earlier version, which did not have symbol versions,
2841 // we should use the base version.
2842 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2843   void *f = dlvsym(handle, name, "libnuma_1.1");
2844   if (f == NULL) {
2845     f = dlsym(handle, name);
2846   }
2847   return f;
2848 }
2849 
2850 bool os::Linux::libnuma_init() {
2851   // sched_getcpu() should be in libc.
2852   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2853                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2854 
2855   // If it's not, try a direct syscall.
2856   if (sched_getcpu() == -1) {
2857     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2858                                     (void*)&sched_getcpu_syscall));
2859   }
2860 
2861   if (sched_getcpu() != -1) { // Does it work?
2862     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2863     if (handle != NULL) {
2864       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2865                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2866       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2867                                        libnuma_dlsym(handle, "numa_max_node")));
2868       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2869                                         libnuma_dlsym(handle, "numa_available")));
2870       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2871                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2872       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2873                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2874       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2875                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2876 
2877 
2878       if (numa_available() != -1) {
2879         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2880         // Create a cpu -> node mapping
2881         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2882         rebuild_cpu_to_node_map();
2883         return true;
2884       }
2885     }
2886   }
2887   return false;
2888 }
2889 
2890 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2891 // The table is later used in get_node_by_cpu().
2892 void os::Linux::rebuild_cpu_to_node_map() {
2893   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2894                               // in libnuma (possible values are starting from 16,
2895                               // and continuing up with every other power of 2, but less
2896                               // than the maximum number of CPUs supported by kernel), and
2897                               // is a subject to change (in libnuma version 2 the requirements
2898                               // are more reasonable) we'll just hardcode the number they use
2899                               // in the library.
2900   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2901 
2902   size_t cpu_num = os::active_processor_count();
2903   size_t cpu_map_size = NCPUS / BitsPerCLong;
2904   size_t cpu_map_valid_size =
2905     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2906 
2907   cpu_to_node()->clear();
2908   cpu_to_node()->at_grow(cpu_num - 1);
2909   size_t node_num = numa_get_groups_num();
2910 
2911   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2912   for (size_t i = 0; i < node_num; i++) {
2913     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2914       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2915         if (cpu_map[j] != 0) {
2916           for (size_t k = 0; k < BitsPerCLong; k++) {
2917             if (cpu_map[j] & (1UL << k)) {
2918               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2919             }
2920           }
2921         }
2922       }
2923     }
2924   }
2925   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2926 }
2927 
2928 int os::Linux::get_node_by_cpu(int cpu_id) {
2929   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2930     return cpu_to_node()->at(cpu_id);
2931   }
2932   return -1;
2933 }
2934 
2935 GrowableArray<int>* os::Linux::_cpu_to_node;
2936 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2937 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2938 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2939 os::Linux::numa_available_func_t os::Linux::_numa_available;
2940 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2941 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2942 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2943 unsigned long* os::Linux::_numa_all_nodes;
2944 
2945 bool os::pd_uncommit_memory(char* addr, size_t size) {
2946   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2947                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2948   return res  != (uintptr_t) MAP_FAILED;
2949 }
2950 
2951 static address get_stack_commited_bottom(address bottom, size_t size) {
2952   address nbot = bottom;
2953   address ntop = bottom + size;
2954 
2955   size_t page_sz = os::vm_page_size();
2956   unsigned pages = size / page_sz;
2957 
2958   unsigned char vec[1];
2959   unsigned imin = 1, imax = pages + 1, imid;
2960   int mincore_return_value = 0;
2961 
2962   assert(imin <= imax, "Unexpected page size");
2963 
2964   while (imin < imax) {
2965     imid = (imax + imin) / 2;
2966     nbot = ntop - (imid * page_sz);
2967 
2968     // Use a trick with mincore to check whether the page is mapped or not.
2969     // mincore sets vec to 1 if page resides in memory and to 0 if page
2970     // is swapped output but if page we are asking for is unmapped
2971     // it returns -1,ENOMEM
2972     mincore_return_value = mincore(nbot, page_sz, vec);
2973 
2974     if (mincore_return_value == -1) {
2975       // Page is not mapped go up
2976       // to find first mapped page
2977       if (errno != EAGAIN) {
2978         assert(errno == ENOMEM, "Unexpected mincore errno");
2979         imax = imid;
2980       }
2981     } else {
2982       // Page is mapped go down
2983       // to find first not mapped page
2984       imin = imid + 1;
2985     }
2986   }
2987 
2988   nbot = nbot + page_sz;
2989 
2990   // Adjust stack bottom one page up if last checked page is not mapped
2991   if (mincore_return_value == -1) {
2992     nbot = nbot + page_sz;
2993   }
2994 
2995   return nbot;
2996 }
2997 
2998 
2999 // Linux uses a growable mapping for the stack, and if the mapping for
3000 // the stack guard pages is not removed when we detach a thread the
3001 // stack cannot grow beyond the pages where the stack guard was
3002 // mapped.  If at some point later in the process the stack expands to
3003 // that point, the Linux kernel cannot expand the stack any further
3004 // because the guard pages are in the way, and a segfault occurs.
3005 //
3006 // However, it's essential not to split the stack region by unmapping
3007 // a region (leaving a hole) that's already part of the stack mapping,
3008 // so if the stack mapping has already grown beyond the guard pages at
3009 // the time we create them, we have to truncate the stack mapping.
3010 // So, we need to know the extent of the stack mapping when
3011 // create_stack_guard_pages() is called.
3012 
3013 // We only need this for stacks that are growable: at the time of
3014 // writing thread stacks don't use growable mappings (i.e. those
3015 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3016 // only applies to the main thread.
3017 
3018 // If the (growable) stack mapping already extends beyond the point
3019 // where we're going to put our guard pages, truncate the mapping at
3020 // that point by munmap()ping it.  This ensures that when we later
3021 // munmap() the guard pages we don't leave a hole in the stack
3022 // mapping. This only affects the main/initial thread
3023 
3024 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3025   if (os::Linux::is_initial_thread()) {
3026     // As we manually grow stack up to bottom inside create_attached_thread(),
3027     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3028     // we don't need to do anything special.
3029     // Check it first, before calling heavy function.
3030     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3031     unsigned char vec[1];
3032 
3033     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3034       // Fallback to slow path on all errors, including EAGAIN
3035       stack_extent = (uintptr_t) get_stack_commited_bottom(
3036                                                            os::Linux::initial_thread_stack_bottom(),
3037                                                            (size_t)addr - stack_extent);
3038     }
3039 
3040     if (stack_extent < (uintptr_t)addr) {
3041       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3042     }
3043   }
3044 
3045   return os::commit_memory(addr, size, !ExecMem);
3046 }
3047 
3048 // If this is a growable mapping, remove the guard pages entirely by
3049 // munmap()ping them.  If not, just call uncommit_memory(). This only
3050 // affects the main/initial thread, but guard against future OS changes
3051 // It's safe to always unmap guard pages for initial thread because we
3052 // always place it right after end of the mapped region
3053 
3054 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3055   uintptr_t stack_extent, stack_base;
3056 
3057   if (os::Linux::is_initial_thread()) {
3058     return ::munmap(addr, size) == 0;
3059   }
3060 
3061   return os::uncommit_memory(addr, size);
3062 }
3063 
3064 static address _highest_vm_reserved_address = NULL;
3065 
3066 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3067 // at 'requested_addr'. If there are existing memory mappings at the same
3068 // location, however, they will be overwritten. If 'fixed' is false,
3069 // 'requested_addr' is only treated as a hint, the return value may or
3070 // may not start from the requested address. Unlike Linux mmap(), this
3071 // function returns NULL to indicate failure.
3072 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3073   char * addr;
3074   int flags;
3075 
3076   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3077   if (fixed) {
3078     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3079     flags |= MAP_FIXED;
3080   }
3081 
3082   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3083   // touch an uncommitted page. Otherwise, the read/write might
3084   // succeed if we have enough swap space to back the physical page.
3085   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3086                        flags, -1, 0);
3087 
3088   if (addr != MAP_FAILED) {
3089     // anon_mmap() should only get called during VM initialization,
3090     // don't need lock (actually we can skip locking even it can be called
3091     // from multiple threads, because _highest_vm_reserved_address is just a
3092     // hint about the upper limit of non-stack memory regions.)
3093     if ((address)addr + bytes > _highest_vm_reserved_address) {
3094       _highest_vm_reserved_address = (address)addr + bytes;
3095     }
3096   }
3097 
3098   return addr == MAP_FAILED ? NULL : addr;
3099 }
3100 
3101 // Don't update _highest_vm_reserved_address, because there might be memory
3102 // regions above addr + size. If so, releasing a memory region only creates
3103 // a hole in the address space, it doesn't help prevent heap-stack collision.
3104 //
3105 static int anon_munmap(char * addr, size_t size) {
3106   return ::munmap(addr, size) == 0;
3107 }
3108 
3109 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3110                             size_t alignment_hint) {
3111   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3112 }
3113 
3114 bool os::pd_release_memory(char* addr, size_t size) {
3115   return anon_munmap(addr, size);
3116 }
3117 
3118 static address highest_vm_reserved_address() {
3119   return _highest_vm_reserved_address;
3120 }
3121 
3122 static bool linux_mprotect(char* addr, size_t size, int prot) {
3123   // Linux wants the mprotect address argument to be page aligned.
3124   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3125 
3126   // According to SUSv3, mprotect() should only be used with mappings
3127   // established by mmap(), and mmap() always maps whole pages. Unaligned
3128   // 'addr' likely indicates problem in the VM (e.g. trying to change
3129   // protection of malloc'ed or statically allocated memory). Check the
3130   // caller if you hit this assert.
3131   assert(addr == bottom, "sanity check");
3132 
3133   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3134   return ::mprotect(bottom, size, prot) == 0;
3135 }
3136 
3137 // Set protections specified
3138 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3139                         bool is_committed) {
3140   unsigned int p = 0;
3141   switch (prot) {
3142   case MEM_PROT_NONE: p = PROT_NONE; break;
3143   case MEM_PROT_READ: p = PROT_READ; break;
3144   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3145   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3146   default:
3147     ShouldNotReachHere();
3148   }
3149   // is_committed is unused.
3150   return linux_mprotect(addr, bytes, p);
3151 }
3152 
3153 bool os::guard_memory(char* addr, size_t size) {
3154   return linux_mprotect(addr, size, PROT_NONE);
3155 }
3156 
3157 bool os::unguard_memory(char* addr, size_t size) {
3158   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3159 }
3160 
3161 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3162                                                     size_t page_size) {
3163   bool result = false;
3164   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3165                  MAP_ANONYMOUS|MAP_PRIVATE,
3166                  -1, 0);
3167   if (p != MAP_FAILED) {
3168     void *aligned_p = align_ptr_up(p, page_size);
3169 
3170     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3171 
3172     munmap(p, page_size * 2);
3173   }
3174 
3175   if (warn && !result) {
3176     warning("TransparentHugePages is not supported by the operating system.");
3177   }
3178 
3179   return result;
3180 }
3181 
3182 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3183   bool result = false;
3184   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3185                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3186                  -1, 0);
3187 
3188   if (p != MAP_FAILED) {
3189     // We don't know if this really is a huge page or not.
3190     FILE *fp = fopen("/proc/self/maps", "r");
3191     if (fp) {
3192       while (!feof(fp)) {
3193         char chars[257];
3194         long x = 0;
3195         if (fgets(chars, sizeof(chars), fp)) {
3196           if (sscanf(chars, "%lx-%*x", &x) == 1
3197               && x == (long)p) {
3198             if (strstr (chars, "hugepage")) {
3199               result = true;
3200               break;
3201             }
3202           }
3203         }
3204       }
3205       fclose(fp);
3206     }
3207     munmap(p, page_size);
3208   }
3209 
3210   if (warn && !result) {
3211     warning("HugeTLBFS is not supported by the operating system.");
3212   }
3213 
3214   return result;
3215 }
3216 
3217 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3218 //
3219 // From the coredump_filter documentation:
3220 //
3221 // - (bit 0) anonymous private memory
3222 // - (bit 1) anonymous shared memory
3223 // - (bit 2) file-backed private memory
3224 // - (bit 3) file-backed shared memory
3225 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3226 //           effective only if the bit 2 is cleared)
3227 // - (bit 5) hugetlb private memory
3228 // - (bit 6) hugetlb shared memory
3229 //
3230 static void set_coredump_filter(void) {
3231   FILE *f;
3232   long cdm;
3233 
3234   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3235     return;
3236   }
3237 
3238   if (fscanf(f, "%lx", &cdm) != 1) {
3239     fclose(f);
3240     return;
3241   }
3242 
3243   rewind(f);
3244 
3245   if ((cdm & LARGEPAGES_BIT) == 0) {
3246     cdm |= LARGEPAGES_BIT;
3247     fprintf(f, "%#lx", cdm);
3248   }
3249 
3250   fclose(f);
3251 }
3252 
3253 // Large page support
3254 
3255 static size_t _large_page_size = 0;
3256 
3257 size_t os::Linux::find_large_page_size() {
3258   size_t large_page_size = 0;
3259 
3260   // large_page_size on Linux is used to round up heap size. x86 uses either
3261   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3262   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3263   // page as large as 256M.
3264   //
3265   // Here we try to figure out page size by parsing /proc/meminfo and looking
3266   // for a line with the following format:
3267   //    Hugepagesize:     2048 kB
3268   //
3269   // If we can't determine the value (e.g. /proc is not mounted, or the text
3270   // format has been changed), we'll use the largest page size supported by
3271   // the processor.
3272 
3273 #ifndef ZERO
3274   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3275                      ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3276 #endif // ZERO
3277 
3278   FILE *fp = fopen("/proc/meminfo", "r");
3279   if (fp) {
3280     while (!feof(fp)) {
3281       int x = 0;
3282       char buf[16];
3283       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3284         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3285           large_page_size = x * K;
3286           break;
3287         }
3288       } else {
3289         // skip to next line
3290         for (;;) {
3291           int ch = fgetc(fp);
3292           if (ch == EOF || ch == (int)'\n') break;
3293         }
3294       }
3295     }
3296     fclose(fp);
3297   }
3298 
3299   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3300     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3301             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3302             proper_unit_for_byte_size(large_page_size));
3303   }
3304 
3305   return large_page_size;
3306 }
3307 
3308 size_t os::Linux::setup_large_page_size() {
3309   _large_page_size = Linux::find_large_page_size();
3310   const size_t default_page_size = (size_t)Linux::page_size();
3311   if (_large_page_size > default_page_size) {
3312     _page_sizes[0] = _large_page_size;
3313     _page_sizes[1] = default_page_size;
3314     _page_sizes[2] = 0;
3315   }
3316 
3317   return _large_page_size;
3318 }
3319 
3320 bool os::Linux::setup_large_page_type(size_t page_size) {
3321   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3322       FLAG_IS_DEFAULT(UseSHM) &&
3323       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3324 
3325     // The type of large pages has not been specified by the user.
3326 
3327     // Try UseHugeTLBFS and then UseSHM.
3328     UseHugeTLBFS = UseSHM = true;
3329 
3330     // Don't try UseTransparentHugePages since there are known
3331     // performance issues with it turned on. This might change in the future.
3332     UseTransparentHugePages = false;
3333   }
3334 
3335   if (UseTransparentHugePages) {
3336     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3337     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3338       UseHugeTLBFS = false;
3339       UseSHM = false;
3340       return true;
3341     }
3342     UseTransparentHugePages = false;
3343   }
3344 
3345   if (UseHugeTLBFS) {
3346     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3347     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3348       UseSHM = false;
3349       return true;
3350     }
3351     UseHugeTLBFS = false;
3352   }
3353 
3354   return UseSHM;
3355 }
3356 
3357 void os::large_page_init() {
3358   if (!UseLargePages &&
3359       !UseTransparentHugePages &&
3360       !UseHugeTLBFS &&
3361       !UseSHM) {
3362     // Not using large pages.
3363     return;
3364   }
3365 
3366   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3367     // The user explicitly turned off large pages.
3368     // Ignore the rest of the large pages flags.
3369     UseTransparentHugePages = false;
3370     UseHugeTLBFS = false;
3371     UseSHM = false;
3372     return;
3373   }
3374 
3375   size_t large_page_size = Linux::setup_large_page_size();
3376   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3377 
3378   set_coredump_filter();
3379 }
3380 
3381 #ifndef SHM_HUGETLB
3382   #define SHM_HUGETLB 04000
3383 #endif
3384 
3385 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3386                                             char* req_addr, bool exec) {
3387   // "exec" is passed in but not used.  Creating the shared image for
3388   // the code cache doesn't have an SHM_X executable permission to check.
3389   assert(UseLargePages && UseSHM, "only for SHM large pages");
3390   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3391 
3392   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3393     return NULL; // Fallback to small pages.
3394   }
3395 
3396   key_t key = IPC_PRIVATE;
3397   char *addr;
3398 
3399   bool warn_on_failure = UseLargePages &&
3400                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3401                          !FLAG_IS_DEFAULT(UseSHM) ||
3402                          !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3403   char msg[128];
3404 
3405   // Create a large shared memory region to attach to based on size.
3406   // Currently, size is the total size of the heap
3407   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3408   if (shmid == -1) {
3409     // Possible reasons for shmget failure:
3410     // 1. shmmax is too small for Java heap.
3411     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3412     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3413     // 2. not enough large page memory.
3414     //    > check available large pages: cat /proc/meminfo
3415     //    > increase amount of large pages:
3416     //          echo new_value > /proc/sys/vm/nr_hugepages
3417     //      Note 1: different Linux may use different name for this property,
3418     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3419     //      Note 2: it's possible there's enough physical memory available but
3420     //            they are so fragmented after a long run that they can't
3421     //            coalesce into large pages. Try to reserve large pages when
3422     //            the system is still "fresh".
3423     if (warn_on_failure) {
3424       jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3425       warning("%s", msg);
3426     }
3427     return NULL;
3428   }
3429 
3430   // attach to the region
3431   addr = (char*)shmat(shmid, req_addr, 0);
3432   int err = errno;
3433 
3434   // Remove shmid. If shmat() is successful, the actual shared memory segment
3435   // will be deleted when it's detached by shmdt() or when the process
3436   // terminates. If shmat() is not successful this will remove the shared
3437   // segment immediately.
3438   shmctl(shmid, IPC_RMID, NULL);
3439 
3440   if ((intptr_t)addr == -1) {
3441     if (warn_on_failure) {
3442       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3443       warning("%s", msg);
3444     }
3445     return NULL;
3446   }
3447 
3448   return addr;
3449 }
3450 
3451 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3452                                         int error) {
3453   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3454 
3455   bool warn_on_failure = UseLargePages &&
3456       (!FLAG_IS_DEFAULT(UseLargePages) ||
3457        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3458        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3459 
3460   if (warn_on_failure) {
3461     char msg[128];
3462     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3463                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3464     warning("%s", msg);
3465   }
3466 }
3467 
3468 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3469                                                         char* req_addr,
3470                                                         bool exec) {
3471   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3472   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3473   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3474 
3475   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3476   char* addr = (char*)::mmap(req_addr, bytes, prot,
3477                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3478                              -1, 0);
3479 
3480   if (addr == MAP_FAILED) {
3481     warn_on_large_pages_failure(req_addr, bytes, errno);
3482     return NULL;
3483   }
3484 
3485   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3486 
3487   return addr;
3488 }
3489 
3490 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3491                                                          size_t alignment,
3492                                                          char* req_addr,
3493                                                          bool exec) {
3494   size_t large_page_size = os::large_page_size();
3495 
3496   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3497 
3498   // Allocate small pages.
3499 
3500   char* start;
3501   if (req_addr != NULL) {
3502     assert(is_ptr_aligned(req_addr, alignment), "Must be");
3503     assert(is_size_aligned(bytes, alignment), "Must be");
3504     start = os::reserve_memory(bytes, req_addr);
3505     assert(start == NULL || start == req_addr, "Must be");
3506   } else {
3507     start = os::reserve_memory_aligned(bytes, alignment);
3508   }
3509 
3510   if (start == NULL) {
3511     return NULL;
3512   }
3513 
3514   assert(is_ptr_aligned(start, alignment), "Must be");
3515 
3516   if (MemTracker::tracking_level() > NMT_minimal) {
3517     // os::reserve_memory_special will record this memory area.
3518     // Need to release it here to prevent overlapping reservations.
3519     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3520     tkr.record((address)start, bytes);
3521   }
3522 
3523   char* end = start + bytes;
3524 
3525   // Find the regions of the allocated chunk that can be promoted to large pages.
3526   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3527   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3528 
3529   size_t lp_bytes = lp_end - lp_start;
3530 
3531   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3532 
3533   if (lp_bytes == 0) {
3534     // The mapped region doesn't even span the start and the end of a large page.
3535     // Fall back to allocate a non-special area.
3536     ::munmap(start, end - start);
3537     return NULL;
3538   }
3539 
3540   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3541 
3542 
3543   void* result;
3544 
3545   if (start != lp_start) {
3546     result = ::mmap(start, lp_start - start, prot,
3547                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3548                     -1, 0);
3549     if (result == MAP_FAILED) {
3550       ::munmap(lp_start, end - lp_start);
3551       return NULL;
3552     }
3553   }
3554 
3555   result = ::mmap(lp_start, lp_bytes, prot,
3556                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3557                   -1, 0);
3558   if (result == MAP_FAILED) {
3559     warn_on_large_pages_failure(req_addr, bytes, errno);
3560     // If the mmap above fails, the large pages region will be unmapped and we
3561     // have regions before and after with small pages. Release these regions.
3562     //
3563     // |  mapped  |  unmapped  |  mapped  |
3564     // ^          ^            ^          ^
3565     // start      lp_start     lp_end     end
3566     //
3567     ::munmap(start, lp_start - start);
3568     ::munmap(lp_end, end - lp_end);
3569     return NULL;
3570   }
3571 
3572   if (lp_end != end) {
3573     result = ::mmap(lp_end, end - lp_end, prot,
3574                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3575                     -1, 0);
3576     if (result == MAP_FAILED) {
3577       ::munmap(start, lp_end - start);
3578       return NULL;
3579     }
3580   }
3581 
3582   return start;
3583 }
3584 
3585 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3586                                                    size_t alignment,
3587                                                    char* req_addr,
3588                                                    bool exec) {
3589   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3590   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3591   assert(is_power_of_2(alignment), "Must be");
3592   assert(is_power_of_2(os::large_page_size()), "Must be");
3593   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3594 
3595   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3596     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3597   } else {
3598     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3599   }
3600 }
3601 
3602 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3603                                  char* req_addr, bool exec) {
3604   assert(UseLargePages, "only for large pages");
3605 
3606   char* addr;
3607   if (UseSHM) {
3608     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3609   } else {
3610     assert(UseHugeTLBFS, "must be");
3611     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3612   }
3613 
3614   if (addr != NULL) {
3615     if (UseNUMAInterleaving) {
3616       numa_make_global(addr, bytes);
3617     }
3618 
3619     // The memory is committed
3620     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3621   }
3622 
3623   return addr;
3624 }
3625 
3626 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3627   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3628   return shmdt(base) == 0;
3629 }
3630 
3631 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3632   return pd_release_memory(base, bytes);
3633 }
3634 
3635 bool os::release_memory_special(char* base, size_t bytes) {
3636   bool res;
3637   if (MemTracker::tracking_level() > NMT_minimal) {
3638     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3639     res = os::Linux::release_memory_special_impl(base, bytes);
3640     if (res) {
3641       tkr.record((address)base, bytes);
3642     }
3643 
3644   } else {
3645     res = os::Linux::release_memory_special_impl(base, bytes);
3646   }
3647   return res;
3648 }
3649 
3650 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3651   assert(UseLargePages, "only for large pages");
3652   bool res;
3653 
3654   if (UseSHM) {
3655     res = os::Linux::release_memory_special_shm(base, bytes);
3656   } else {
3657     assert(UseHugeTLBFS, "must be");
3658     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3659   }
3660   return res;
3661 }
3662 
3663 size_t os::large_page_size() {
3664   return _large_page_size;
3665 }
3666 
3667 // With SysV SHM the entire memory region must be allocated as shared
3668 // memory.
3669 // HugeTLBFS allows application to commit large page memory on demand.
3670 // However, when committing memory with HugeTLBFS fails, the region
3671 // that was supposed to be committed will lose the old reservation
3672 // and allow other threads to steal that memory region. Because of this
3673 // behavior we can't commit HugeTLBFS memory.
3674 bool os::can_commit_large_page_memory() {
3675   return UseTransparentHugePages;
3676 }
3677 
3678 bool os::can_execute_large_page_memory() {
3679   return UseTransparentHugePages || UseHugeTLBFS;
3680 }
3681 
3682 // Reserve memory at an arbitrary address, only if that area is
3683 // available (and not reserved for something else).
3684 
3685 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3686   const int max_tries = 10;
3687   char* base[max_tries];
3688   size_t size[max_tries];
3689   const size_t gap = 0x000000;
3690 
3691   // Assert only that the size is a multiple of the page size, since
3692   // that's all that mmap requires, and since that's all we really know
3693   // about at this low abstraction level.  If we need higher alignment,
3694   // we can either pass an alignment to this method or verify alignment
3695   // in one of the methods further up the call chain.  See bug 5044738.
3696   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3697 
3698   // Repeatedly allocate blocks until the block is allocated at the
3699   // right spot. Give up after max_tries. Note that reserve_memory() will
3700   // automatically update _highest_vm_reserved_address if the call is
3701   // successful. The variable tracks the highest memory address every reserved
3702   // by JVM. It is used to detect heap-stack collision if running with
3703   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3704   // space than needed, it could confuse the collision detecting code. To
3705   // solve the problem, save current _highest_vm_reserved_address and
3706   // calculate the correct value before return.
3707   address old_highest = _highest_vm_reserved_address;
3708 
3709   // Linux mmap allows caller to pass an address as hint; give it a try first,
3710   // if kernel honors the hint then we can return immediately.
3711   char * addr = anon_mmap(requested_addr, bytes, false);
3712   if (addr == requested_addr) {
3713     return requested_addr;
3714   }
3715 
3716   if (addr != NULL) {
3717     // mmap() is successful but it fails to reserve at the requested address
3718     anon_munmap(addr, bytes);
3719   }
3720 
3721   int i;
3722   for (i = 0; i < max_tries; ++i) {
3723     base[i] = reserve_memory(bytes);
3724 
3725     if (base[i] != NULL) {
3726       // Is this the block we wanted?
3727       if (base[i] == requested_addr) {
3728         size[i] = bytes;
3729         break;
3730       }
3731 
3732       // Does this overlap the block we wanted? Give back the overlapped
3733       // parts and try again.
3734 
3735       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3736       if (top_overlap >= 0 && top_overlap < bytes) {
3737         unmap_memory(base[i], top_overlap);
3738         base[i] += top_overlap;
3739         size[i] = bytes - top_overlap;
3740       } else {
3741         size_t bottom_overlap = base[i] + bytes - requested_addr;
3742         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3743           unmap_memory(requested_addr, bottom_overlap);
3744           size[i] = bytes - bottom_overlap;
3745         } else {
3746           size[i] = bytes;
3747         }
3748       }
3749     }
3750   }
3751 
3752   // Give back the unused reserved pieces.
3753 
3754   for (int j = 0; j < i; ++j) {
3755     if (base[j] != NULL) {
3756       unmap_memory(base[j], size[j]);
3757     }
3758   }
3759 
3760   if (i < max_tries) {
3761     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3762     return requested_addr;
3763   } else {
3764     _highest_vm_reserved_address = old_highest;
3765     return NULL;
3766   }
3767 }
3768 
3769 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3770   return ::read(fd, buf, nBytes);
3771 }
3772 
3773 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3774   return ::pread(fd, buf, nBytes, offset);
3775 }
3776 
3777 // Short sleep, direct OS call.
3778 //
3779 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3780 // sched_yield(2) will actually give up the CPU:
3781 //
3782 //   * Alone on this pariticular CPU, keeps running.
3783 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3784 //     (pre 2.6.39).
3785 //
3786 // So calling this with 0 is an alternative.
3787 //
3788 void os::naked_short_sleep(jlong ms) {
3789   struct timespec req;
3790 
3791   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3792   req.tv_sec = 0;
3793   if (ms > 0) {
3794     req.tv_nsec = (ms % 1000) * 1000000;
3795   } else {
3796     req.tv_nsec = 1;
3797   }
3798 
3799   nanosleep(&req, NULL);
3800 
3801   return;
3802 }
3803 
3804 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3805 void os::infinite_sleep() {
3806   while (true) {    // sleep forever ...
3807     ::sleep(100);   // ... 100 seconds at a time
3808   }
3809 }
3810 
3811 // Used to convert frequent JVM_Yield() to nops
3812 bool os::dont_yield() {
3813   return DontYieldALot;
3814 }
3815 
3816 void os::naked_yield() {
3817   sched_yield();
3818 }
3819 
3820 ////////////////////////////////////////////////////////////////////////////////
3821 // thread priority support
3822 
3823 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3824 // only supports dynamic priority, static priority must be zero. For real-time
3825 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3826 // However, for large multi-threaded applications, SCHED_RR is not only slower
3827 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3828 // of 5 runs - Sep 2005).
3829 //
3830 // The following code actually changes the niceness of kernel-thread/LWP. It
3831 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3832 // not the entire user process, and user level threads are 1:1 mapped to kernel
3833 // threads. It has always been the case, but could change in the future. For
3834 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3835 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3836 
3837 int os::java_to_os_priority[CriticalPriority + 1] = {
3838   19,              // 0 Entry should never be used
3839 
3840    4,              // 1 MinPriority
3841    3,              // 2
3842    2,              // 3
3843 
3844    1,              // 4
3845    0,              // 5 NormPriority
3846   -1,              // 6
3847 
3848   -2,              // 7
3849   -3,              // 8
3850   -4,              // 9 NearMaxPriority
3851 
3852   -5,              // 10 MaxPriority
3853 
3854   -5               // 11 CriticalPriority
3855 };
3856 
3857 static int prio_init() {
3858   if (ThreadPriorityPolicy == 1) {
3859     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3860     // if effective uid is not root. Perhaps, a more elegant way of doing
3861     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3862     if (geteuid() != 0) {
3863       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3864         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3865       }
3866       ThreadPriorityPolicy = 0;
3867     }
3868   }
3869   if (UseCriticalJavaThreadPriority) {
3870     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3871   }
3872   return 0;
3873 }
3874 
3875 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3876   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3877 
3878   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3879   return (ret == 0) ? OS_OK : OS_ERR;
3880 }
3881 
3882 OSReturn os::get_native_priority(const Thread* const thread,
3883                                  int *priority_ptr) {
3884   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3885     *priority_ptr = java_to_os_priority[NormPriority];
3886     return OS_OK;
3887   }
3888 
3889   errno = 0;
3890   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3891   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3892 }
3893 
3894 // Hint to the underlying OS that a task switch would not be good.
3895 // Void return because it's a hint and can fail.
3896 void os::hint_no_preempt() {}
3897 
3898 ////////////////////////////////////////////////////////////////////////////////
3899 // suspend/resume support
3900 
3901 //  the low-level signal-based suspend/resume support is a remnant from the
3902 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3903 //  within hotspot. Now there is a single use-case for this:
3904 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3905 //      that runs in the watcher thread.
3906 //  The remaining code is greatly simplified from the more general suspension
3907 //  code that used to be used.
3908 //
3909 //  The protocol is quite simple:
3910 //  - suspend:
3911 //      - sends a signal to the target thread
3912 //      - polls the suspend state of the osthread using a yield loop
3913 //      - target thread signal handler (SR_handler) sets suspend state
3914 //        and blocks in sigsuspend until continued
3915 //  - resume:
3916 //      - sets target osthread state to continue
3917 //      - sends signal to end the sigsuspend loop in the SR_handler
3918 //
3919 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3920 
3921 static void resume_clear_context(OSThread *osthread) {
3922   osthread->set_ucontext(NULL);
3923   osthread->set_siginfo(NULL);
3924 }
3925 
3926 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3927                                  ucontext_t* context) {
3928   osthread->set_ucontext(context);
3929   osthread->set_siginfo(siginfo);
3930 }
3931 
3932 // Handler function invoked when a thread's execution is suspended or
3933 // resumed. We have to be careful that only async-safe functions are
3934 // called here (Note: most pthread functions are not async safe and
3935 // should be avoided.)
3936 //
3937 // Note: sigwait() is a more natural fit than sigsuspend() from an
3938 // interface point of view, but sigwait() prevents the signal hander
3939 // from being run. libpthread would get very confused by not having
3940 // its signal handlers run and prevents sigwait()'s use with the
3941 // mutex granting granting signal.
3942 //
3943 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3944 //
3945 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3946   // Save and restore errno to avoid confusing native code with EINTR
3947   // after sigsuspend.
3948   int old_errno = errno;
3949 
3950   Thread* thread = Thread::current();
3951   OSThread* osthread = thread->osthread();
3952   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3953 
3954   os::SuspendResume::State current = osthread->sr.state();
3955   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3956     suspend_save_context(osthread, siginfo, context);
3957 
3958     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3959     os::SuspendResume::State state = osthread->sr.suspended();
3960     if (state == os::SuspendResume::SR_SUSPENDED) {
3961       sigset_t suspend_set;  // signals for sigsuspend()
3962 
3963       // get current set of blocked signals and unblock resume signal
3964       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3965       sigdelset(&suspend_set, SR_signum);
3966 
3967       sr_semaphore.signal();
3968       // wait here until we are resumed
3969       while (1) {
3970         sigsuspend(&suspend_set);
3971 
3972         os::SuspendResume::State result = osthread->sr.running();
3973         if (result == os::SuspendResume::SR_RUNNING) {
3974           sr_semaphore.signal();
3975           break;
3976         }
3977       }
3978 
3979     } else if (state == os::SuspendResume::SR_RUNNING) {
3980       // request was cancelled, continue
3981     } else {
3982       ShouldNotReachHere();
3983     }
3984 
3985     resume_clear_context(osthread);
3986   } else if (current == os::SuspendResume::SR_RUNNING) {
3987     // request was cancelled, continue
3988   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3989     // ignore
3990   } else {
3991     // ignore
3992   }
3993 
3994   errno = old_errno;
3995 }
3996 
3997 
3998 static int SR_initialize() {
3999   struct sigaction act;
4000   char *s;
4001   // Get signal number to use for suspend/resume
4002   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4003     int sig = ::strtol(s, 0, 10);
4004     if (sig > 0 || sig < _NSIG) {
4005       SR_signum = sig;
4006     }
4007   }
4008 
4009   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4010          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4011 
4012   sigemptyset(&SR_sigset);
4013   sigaddset(&SR_sigset, SR_signum);
4014 
4015   // Set up signal handler for suspend/resume
4016   act.sa_flags = SA_RESTART|SA_SIGINFO;
4017   act.sa_handler = (void (*)(int)) SR_handler;
4018 
4019   // SR_signum is blocked by default.
4020   // 4528190 - We also need to block pthread restart signal (32 on all
4021   // supported Linux platforms). Note that LinuxThreads need to block
4022   // this signal for all threads to work properly. So we don't have
4023   // to use hard-coded signal number when setting up the mask.
4024   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4025 
4026   if (sigaction(SR_signum, &act, 0) == -1) {
4027     return -1;
4028   }
4029 
4030   // Save signal flag
4031   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4032   return 0;
4033 }
4034 
4035 static int sr_notify(OSThread* osthread) {
4036   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4037   assert_status(status == 0, status, "pthread_kill");
4038   return status;
4039 }
4040 
4041 // "Randomly" selected value for how long we want to spin
4042 // before bailing out on suspending a thread, also how often
4043 // we send a signal to a thread we want to resume
4044 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4045 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4046 
4047 // returns true on success and false on error - really an error is fatal
4048 // but this seems the normal response to library errors
4049 static bool do_suspend(OSThread* osthread) {
4050   assert(osthread->sr.is_running(), "thread should be running");
4051   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4052 
4053   // mark as suspended and send signal
4054   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4055     // failed to switch, state wasn't running?
4056     ShouldNotReachHere();
4057     return false;
4058   }
4059 
4060   if (sr_notify(osthread) != 0) {
4061     ShouldNotReachHere();
4062   }
4063 
4064   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4065   while (true) {
4066     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4067       break;
4068     } else {
4069       // timeout
4070       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4071       if (cancelled == os::SuspendResume::SR_RUNNING) {
4072         return false;
4073       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4074         // make sure that we consume the signal on the semaphore as well
4075         sr_semaphore.wait();
4076         break;
4077       } else {
4078         ShouldNotReachHere();
4079         return false;
4080       }
4081     }
4082   }
4083 
4084   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4085   return true;
4086 }
4087 
4088 static void do_resume(OSThread* osthread) {
4089   assert(osthread->sr.is_suspended(), "thread should be suspended");
4090   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4091 
4092   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4093     // failed to switch to WAKEUP_REQUEST
4094     ShouldNotReachHere();
4095     return;
4096   }
4097 
4098   while (true) {
4099     if (sr_notify(osthread) == 0) {
4100       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4101         if (osthread->sr.is_running()) {
4102           return;
4103         }
4104       }
4105     } else {
4106       ShouldNotReachHere();
4107     }
4108   }
4109 
4110   guarantee(osthread->sr.is_running(), "Must be running!");
4111 }
4112 
4113 ///////////////////////////////////////////////////////////////////////////////////
4114 // signal handling (except suspend/resume)
4115 
4116 // This routine may be used by user applications as a "hook" to catch signals.
4117 // The user-defined signal handler must pass unrecognized signals to this
4118 // routine, and if it returns true (non-zero), then the signal handler must
4119 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4120 // routine will never retun false (zero), but instead will execute a VM panic
4121 // routine kill the process.
4122 //
4123 // If this routine returns false, it is OK to call it again.  This allows
4124 // the user-defined signal handler to perform checks either before or after
4125 // the VM performs its own checks.  Naturally, the user code would be making
4126 // a serious error if it tried to handle an exception (such as a null check
4127 // or breakpoint) that the VM was generating for its own correct operation.
4128 //
4129 // This routine may recognize any of the following kinds of signals:
4130 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4131 // It should be consulted by handlers for any of those signals.
4132 //
4133 // The caller of this routine must pass in the three arguments supplied
4134 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4135 // field of the structure passed to sigaction().  This routine assumes that
4136 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4137 //
4138 // Note that the VM will print warnings if it detects conflicting signal
4139 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4140 //
4141 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4142                                                  siginfo_t* siginfo,
4143                                                  void* ucontext,
4144                                                  int abort_if_unrecognized);
4145 
4146 void signalHandler(int sig, siginfo_t* info, void* uc) {
4147   assert(info != NULL && uc != NULL, "it must be old kernel");
4148   int orig_errno = errno;  // Preserve errno value over signal handler.
4149   JVM_handle_linux_signal(sig, info, uc, true);
4150   errno = orig_errno;
4151 }
4152 
4153 
4154 // This boolean allows users to forward their own non-matching signals
4155 // to JVM_handle_linux_signal, harmlessly.
4156 bool os::Linux::signal_handlers_are_installed = false;
4157 
4158 // For signal-chaining
4159 struct sigaction os::Linux::sigact[MAXSIGNUM];
4160 unsigned int os::Linux::sigs = 0;
4161 bool os::Linux::libjsig_is_loaded = false;
4162 typedef struct sigaction *(*get_signal_t)(int);
4163 get_signal_t os::Linux::get_signal_action = NULL;
4164 
4165 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4166   struct sigaction *actp = NULL;
4167 
4168   if (libjsig_is_loaded) {
4169     // Retrieve the old signal handler from libjsig
4170     actp = (*get_signal_action)(sig);
4171   }
4172   if (actp == NULL) {
4173     // Retrieve the preinstalled signal handler from jvm
4174     actp = get_preinstalled_handler(sig);
4175   }
4176 
4177   return actp;
4178 }
4179 
4180 static bool call_chained_handler(struct sigaction *actp, int sig,
4181                                  siginfo_t *siginfo, void *context) {
4182   // Call the old signal handler
4183   if (actp->sa_handler == SIG_DFL) {
4184     // It's more reasonable to let jvm treat it as an unexpected exception
4185     // instead of taking the default action.
4186     return false;
4187   } else if (actp->sa_handler != SIG_IGN) {
4188     if ((actp->sa_flags & SA_NODEFER) == 0) {
4189       // automaticlly block the signal
4190       sigaddset(&(actp->sa_mask), sig);
4191     }
4192 
4193     sa_handler_t hand;
4194     sa_sigaction_t sa;
4195     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4196     // retrieve the chained handler
4197     if (siginfo_flag_set) {
4198       sa = actp->sa_sigaction;
4199     } else {
4200       hand = actp->sa_handler;
4201     }
4202 
4203     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4204       actp->sa_handler = SIG_DFL;
4205     }
4206 
4207     // try to honor the signal mask
4208     sigset_t oset;
4209     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4210 
4211     // call into the chained handler
4212     if (siginfo_flag_set) {
4213       (*sa)(sig, siginfo, context);
4214     } else {
4215       (*hand)(sig);
4216     }
4217 
4218     // restore the signal mask
4219     pthread_sigmask(SIG_SETMASK, &oset, 0);
4220   }
4221   // Tell jvm's signal handler the signal is taken care of.
4222   return true;
4223 }
4224 
4225 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4226   bool chained = false;
4227   // signal-chaining
4228   if (UseSignalChaining) {
4229     struct sigaction *actp = get_chained_signal_action(sig);
4230     if (actp != NULL) {
4231       chained = call_chained_handler(actp, sig, siginfo, context);
4232     }
4233   }
4234   return chained;
4235 }
4236 
4237 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4238   if ((((unsigned int)1 << sig) & sigs) != 0) {
4239     return &sigact[sig];
4240   }
4241   return NULL;
4242 }
4243 
4244 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4245   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4246   sigact[sig] = oldAct;
4247   sigs |= (unsigned int)1 << sig;
4248 }
4249 
4250 // for diagnostic
4251 int os::Linux::sigflags[MAXSIGNUM];
4252 
4253 int os::Linux::get_our_sigflags(int sig) {
4254   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4255   return sigflags[sig];
4256 }
4257 
4258 void os::Linux::set_our_sigflags(int sig, int flags) {
4259   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4260   sigflags[sig] = flags;
4261 }
4262 
4263 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4264   // Check for overwrite.
4265   struct sigaction oldAct;
4266   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4267 
4268   void* oldhand = oldAct.sa_sigaction
4269                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4270                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4271   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4272       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4273       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4274     if (AllowUserSignalHandlers || !set_installed) {
4275       // Do not overwrite; user takes responsibility to forward to us.
4276       return;
4277     } else if (UseSignalChaining) {
4278       // save the old handler in jvm
4279       save_preinstalled_handler(sig, oldAct);
4280       // libjsig also interposes the sigaction() call below and saves the
4281       // old sigaction on it own.
4282     } else {
4283       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4284                     "%#lx for signal %d.", (long)oldhand, sig));
4285     }
4286   }
4287 
4288   struct sigaction sigAct;
4289   sigfillset(&(sigAct.sa_mask));
4290   sigAct.sa_handler = SIG_DFL;
4291   if (!set_installed) {
4292     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4293   } else {
4294     sigAct.sa_sigaction = signalHandler;
4295     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4296   }
4297   // Save flags, which are set by ours
4298   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4299   sigflags[sig] = sigAct.sa_flags;
4300 
4301   int ret = sigaction(sig, &sigAct, &oldAct);
4302   assert(ret == 0, "check");
4303 
4304   void* oldhand2  = oldAct.sa_sigaction
4305                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4306                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4307   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4308 }
4309 
4310 // install signal handlers for signals that HotSpot needs to
4311 // handle in order to support Java-level exception handling.
4312 
4313 void os::Linux::install_signal_handlers() {
4314   if (!signal_handlers_are_installed) {
4315     signal_handlers_are_installed = true;
4316 
4317     // signal-chaining
4318     typedef void (*signal_setting_t)();
4319     signal_setting_t begin_signal_setting = NULL;
4320     signal_setting_t end_signal_setting = NULL;
4321     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4322                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4323     if (begin_signal_setting != NULL) {
4324       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4325                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4326       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4327                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4328       libjsig_is_loaded = true;
4329       assert(UseSignalChaining, "should enable signal-chaining");
4330     }
4331     if (libjsig_is_loaded) {
4332       // Tell libjsig jvm is setting signal handlers
4333       (*begin_signal_setting)();
4334     }
4335 
4336     set_signal_handler(SIGSEGV, true);
4337     set_signal_handler(SIGPIPE, true);
4338     set_signal_handler(SIGBUS, true);
4339     set_signal_handler(SIGILL, true);
4340     set_signal_handler(SIGFPE, true);
4341 #if defined(PPC64)
4342     set_signal_handler(SIGTRAP, true);
4343 #endif
4344     set_signal_handler(SIGXFSZ, true);
4345 
4346     if (libjsig_is_loaded) {
4347       // Tell libjsig jvm finishes setting signal handlers
4348       (*end_signal_setting)();
4349     }
4350 
4351     // We don't activate signal checker if libjsig is in place, we trust ourselves
4352     // and if UserSignalHandler is installed all bets are off.
4353     // Log that signal checking is off only if -verbose:jni is specified.
4354     if (CheckJNICalls) {
4355       if (libjsig_is_loaded) {
4356         if (PrintJNIResolving) {
4357           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4358         }
4359         check_signals = false;
4360       }
4361       if (AllowUserSignalHandlers) {
4362         if (PrintJNIResolving) {
4363           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4364         }
4365         check_signals = false;
4366       }
4367     }
4368   }
4369 }
4370 
4371 // This is the fastest way to get thread cpu time on Linux.
4372 // Returns cpu time (user+sys) for any thread, not only for current.
4373 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4374 // It might work on 2.6.10+ with a special kernel/glibc patch.
4375 // For reference, please, see IEEE Std 1003.1-2004:
4376 //   http://www.unix.org/single_unix_specification
4377 
4378 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4379   struct timespec tp;
4380   int rc = os::Linux::clock_gettime(clockid, &tp);
4381   assert(rc == 0, "clock_gettime is expected to return 0 code");
4382 
4383   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4384 }
4385 
4386 /////
4387 // glibc on Linux platform uses non-documented flag
4388 // to indicate, that some special sort of signal
4389 // trampoline is used.
4390 // We will never set this flag, and we should
4391 // ignore this flag in our diagnostic
4392 #ifdef SIGNIFICANT_SIGNAL_MASK
4393   #undef SIGNIFICANT_SIGNAL_MASK
4394 #endif
4395 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4396 
4397 static const char* get_signal_handler_name(address handler,
4398                                            char* buf, int buflen) {
4399   int offset;
4400   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4401   if (found) {
4402     // skip directory names
4403     const char *p1, *p2;
4404     p1 = buf;
4405     size_t len = strlen(os::file_separator());
4406     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4407     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4408   } else {
4409     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4410   }
4411   return buf;
4412 }
4413 
4414 static void print_signal_handler(outputStream* st, int sig,
4415                                  char* buf, size_t buflen) {
4416   struct sigaction sa;
4417 
4418   sigaction(sig, NULL, &sa);
4419 
4420   // See comment for SIGNIFICANT_SIGNAL_MASK define
4421   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4422 
4423   st->print("%s: ", os::exception_name(sig, buf, buflen));
4424 
4425   address handler = (sa.sa_flags & SA_SIGINFO)
4426     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4427     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4428 
4429   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4430     st->print("SIG_DFL");
4431   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4432     st->print("SIG_IGN");
4433   } else {
4434     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4435   }
4436 
4437   st->print(", sa_mask[0]=");
4438   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4439 
4440   address rh = VMError::get_resetted_sighandler(sig);
4441   // May be, handler was resetted by VMError?
4442   if (rh != NULL) {
4443     handler = rh;
4444     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4445   }
4446 
4447   st->print(", sa_flags=");
4448   os::Posix::print_sa_flags(st, sa.sa_flags);
4449 
4450   // Check: is it our handler?
4451   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4452       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4453     // It is our signal handler
4454     // check for flags, reset system-used one!
4455     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4456       st->print(
4457                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4458                 os::Linux::get_our_sigflags(sig));
4459     }
4460   }
4461   st->cr();
4462 }
4463 
4464 
4465 #define DO_SIGNAL_CHECK(sig)                      \
4466   do {                                            \
4467     if (!sigismember(&check_signal_done, sig)) {  \
4468       os::Linux::check_signal_handler(sig);       \
4469     }                                             \
4470   } while (0)
4471 
4472 // This method is a periodic task to check for misbehaving JNI applications
4473 // under CheckJNI, we can add any periodic checks here
4474 
4475 void os::run_periodic_checks() {
4476   if (check_signals == false) return;
4477 
4478   // SEGV and BUS if overridden could potentially prevent
4479   // generation of hs*.log in the event of a crash, debugging
4480   // such a case can be very challenging, so we absolutely
4481   // check the following for a good measure:
4482   DO_SIGNAL_CHECK(SIGSEGV);
4483   DO_SIGNAL_CHECK(SIGILL);
4484   DO_SIGNAL_CHECK(SIGFPE);
4485   DO_SIGNAL_CHECK(SIGBUS);
4486   DO_SIGNAL_CHECK(SIGPIPE);
4487   DO_SIGNAL_CHECK(SIGXFSZ);
4488 #if defined(PPC64)
4489   DO_SIGNAL_CHECK(SIGTRAP);
4490 #endif
4491 
4492   // ReduceSignalUsage allows the user to override these handlers
4493   // see comments at the very top and jvm_solaris.h
4494   if (!ReduceSignalUsage) {
4495     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4496     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4497     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4498     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4499   }
4500 
4501   DO_SIGNAL_CHECK(SR_signum);
4502   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4503 }
4504 
4505 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4506 
4507 static os_sigaction_t os_sigaction = NULL;
4508 
4509 void os::Linux::check_signal_handler(int sig) {
4510   char buf[O_BUFLEN];
4511   address jvmHandler = NULL;
4512 
4513 
4514   struct sigaction act;
4515   if (os_sigaction == NULL) {
4516     // only trust the default sigaction, in case it has been interposed
4517     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4518     if (os_sigaction == NULL) return;
4519   }
4520 
4521   os_sigaction(sig, (struct sigaction*)NULL, &act);
4522 
4523 
4524   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4525 
4526   address thisHandler = (act.sa_flags & SA_SIGINFO)
4527     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4528     : CAST_FROM_FN_PTR(address, act.sa_handler);
4529 
4530 
4531   switch (sig) {
4532   case SIGSEGV:
4533   case SIGBUS:
4534   case SIGFPE:
4535   case SIGPIPE:
4536   case SIGILL:
4537   case SIGXFSZ:
4538     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4539     break;
4540 
4541   case SHUTDOWN1_SIGNAL:
4542   case SHUTDOWN2_SIGNAL:
4543   case SHUTDOWN3_SIGNAL:
4544   case BREAK_SIGNAL:
4545     jvmHandler = (address)user_handler();
4546     break;
4547 
4548   case INTERRUPT_SIGNAL:
4549     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4550     break;
4551 
4552   default:
4553     if (sig == SR_signum) {
4554       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4555     } else {
4556       return;
4557     }
4558     break;
4559   }
4560 
4561   if (thisHandler != jvmHandler) {
4562     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4563     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4564     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4565     // No need to check this sig any longer
4566     sigaddset(&check_signal_done, sig);
4567     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4568     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4569       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4570                     exception_name(sig, buf, O_BUFLEN));
4571     }
4572   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4573     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4574     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4575     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4576     // No need to check this sig any longer
4577     sigaddset(&check_signal_done, sig);
4578   }
4579 
4580   // Dump all the signal
4581   if (sigismember(&check_signal_done, sig)) {
4582     print_signal_handlers(tty, buf, O_BUFLEN);
4583   }
4584 }
4585 
4586 extern void report_error(char* file_name, int line_no, char* title,
4587                          char* format, ...);
4588 
4589 extern bool signal_name(int signo, char* buf, size_t len);
4590 
4591 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4592   if (0 < exception_code && exception_code <= SIGRTMAX) {
4593     // signal
4594     if (!signal_name(exception_code, buf, size)) {
4595       jio_snprintf(buf, size, "SIG%d", exception_code);
4596     }
4597     return buf;
4598   } else {
4599     return NULL;
4600   }
4601 }
4602 
4603 // this is called _before_ the most of global arguments have been parsed
4604 void os::init(void) {
4605   char dummy;   // used to get a guess on initial stack address
4606 //  first_hrtime = gethrtime();
4607 
4608   // With LinuxThreads the JavaMain thread pid (primordial thread)
4609   // is different than the pid of the java launcher thread.
4610   // So, on Linux, the launcher thread pid is passed to the VM
4611   // via the sun.java.launcher.pid property.
4612   // Use this property instead of getpid() if it was correctly passed.
4613   // See bug 6351349.
4614   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4615 
4616   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4617 
4618   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4619 
4620   init_random(1234567);
4621 
4622   ThreadCritical::initialize();
4623 
4624   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4625   if (Linux::page_size() == -1) {
4626     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4627                   strerror(errno)));
4628   }
4629   init_page_sizes((size_t) Linux::page_size());
4630 
4631   Linux::initialize_system_info();
4632 
4633   // main_thread points to the aboriginal thread
4634   Linux::_main_thread = pthread_self();
4635 
4636   Linux::clock_init();
4637   initial_time_count = javaTimeNanos();
4638 
4639   // pthread_condattr initialization for monotonic clock
4640   int status;
4641   pthread_condattr_t* _condattr = os::Linux::condAttr();
4642   if ((status = pthread_condattr_init(_condattr)) != 0) {
4643     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4644   }
4645   // Only set the clock if CLOCK_MONOTONIC is available
4646   if (os::supports_monotonic_clock()) {
4647     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4648       if (status == EINVAL) {
4649         warning("Unable to use monotonic clock with relative timed-waits" \
4650                 " - changes to the time-of-day clock may have adverse affects");
4651       } else {
4652         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4653       }
4654     }
4655   }
4656   // else it defaults to CLOCK_REALTIME
4657 
4658   pthread_mutex_init(&dl_mutex, NULL);
4659 
4660   // If the pagesize of the VM is greater than 8K determine the appropriate
4661   // number of initial guard pages.  The user can change this with the
4662   // command line arguments, if needed.
4663   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4664     StackYellowPages = 1;
4665     StackRedPages = 1;
4666     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4667   }
4668 
4669   // retrieve entry point for pthread_setname_np
4670   Linux::_pthread_setname_np =
4671     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4672 
4673 }
4674 
4675 // To install functions for atexit system call
4676 extern "C" {
4677   static void perfMemory_exit_helper() {
4678     perfMemory_exit();
4679   }
4680 }
4681 
4682 // this is called _after_ the global arguments have been parsed
4683 jint os::init_2(void) {
4684   Linux::fast_thread_clock_init();
4685 
4686   // Allocate a single page and mark it as readable for safepoint polling
4687   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4688   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4689 
4690   os::set_polling_page(polling_page);
4691 
4692 #ifndef PRODUCT
4693   if (Verbose && PrintMiscellaneous) {
4694     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4695                (intptr_t)polling_page);
4696   }
4697 #endif
4698 
4699   if (!UseMembar) {
4700     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4701     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4702     os::set_memory_serialize_page(mem_serialize_page);
4703 
4704 #ifndef PRODUCT
4705     if (Verbose && PrintMiscellaneous) {
4706       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n",
4707                  (intptr_t)mem_serialize_page);
4708     }
4709 #endif
4710   }
4711 
4712   // initialize suspend/resume support - must do this before signal_sets_init()
4713   if (SR_initialize() != 0) {
4714     perror("SR_initialize failed");
4715     return JNI_ERR;
4716   }
4717 
4718   Linux::signal_sets_init();
4719   Linux::install_signal_handlers();
4720 
4721   // Check minimum allowable stack size for thread creation and to initialize
4722   // the java system classes, including StackOverflowError - depends on page
4723   // size.  Add a page for compiler2 recursion in main thread.
4724   // Add in 2*BytesPerWord times page size to account for VM stack during
4725   // class initialization depending on 32 or 64 bit VM.
4726   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4727                                       (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4728                                       (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4729 
4730   size_t threadStackSizeInBytes = ThreadStackSize * K;
4731   if (threadStackSizeInBytes != 0 &&
4732       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4733     tty->print_cr("\nThe stack size specified is too small, "
4734                   "Specify at least %dk",
4735                   os::Linux::min_stack_allowed/ K);
4736     return JNI_ERR;
4737   }
4738 
4739   // Make the stack size a multiple of the page size so that
4740   // the yellow/red zones can be guarded.
4741   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4742                                                 vm_page_size()));
4743 
4744   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4745 
4746 #if defined(IA32)
4747   workaround_expand_exec_shield_cs_limit();
4748 #endif
4749 
4750   Linux::libpthread_init();
4751   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4752     tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4753                   Linux::glibc_version(), Linux::libpthread_version(),
4754                   Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4755   }
4756 
4757   if (UseNUMA) {
4758     if (!Linux::libnuma_init()) {
4759       UseNUMA = false;
4760     } else {
4761       if ((Linux::numa_max_node() < 1)) {
4762         // There's only one node(they start from 0), disable NUMA.
4763         UseNUMA = false;
4764       }
4765     }
4766     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4767     // we can make the adaptive lgrp chunk resizing work. If the user specified
4768     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4769     // disable adaptive resizing.
4770     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4771       if (FLAG_IS_DEFAULT(UseNUMA)) {
4772         UseNUMA = false;
4773       } else {
4774         if (FLAG_IS_DEFAULT(UseLargePages) &&
4775             FLAG_IS_DEFAULT(UseSHM) &&
4776             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4777           UseLargePages = false;
4778         } else {
4779           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4780           UseAdaptiveSizePolicy = false;
4781           UseAdaptiveNUMAChunkSizing = false;
4782         }
4783       }
4784     }
4785     if (!UseNUMA && ForceNUMA) {
4786       UseNUMA = true;
4787     }
4788   }
4789 
4790   if (MaxFDLimit) {
4791     // set the number of file descriptors to max. print out error
4792     // if getrlimit/setrlimit fails but continue regardless.
4793     struct rlimit nbr_files;
4794     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4795     if (status != 0) {
4796       if (PrintMiscellaneous && (Verbose || WizardMode)) {
4797         perror("os::init_2 getrlimit failed");
4798       }
4799     } else {
4800       nbr_files.rlim_cur = nbr_files.rlim_max;
4801       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4802       if (status != 0) {
4803         if (PrintMiscellaneous && (Verbose || WizardMode)) {
4804           perror("os::init_2 setrlimit failed");
4805         }
4806       }
4807     }
4808   }
4809 
4810   // Initialize lock used to serialize thread creation (see os::create_thread)
4811   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4812 
4813   // at-exit methods are called in the reverse order of their registration.
4814   // atexit functions are called on return from main or as a result of a
4815   // call to exit(3C). There can be only 32 of these functions registered
4816   // and atexit() does not set errno.
4817 
4818   if (PerfAllowAtExitRegistration) {
4819     // only register atexit functions if PerfAllowAtExitRegistration is set.
4820     // atexit functions can be delayed until process exit time, which
4821     // can be problematic for embedded VM situations. Embedded VMs should
4822     // call DestroyJavaVM() to assure that VM resources are released.
4823 
4824     // note: perfMemory_exit_helper atexit function may be removed in
4825     // the future if the appropriate cleanup code can be added to the
4826     // VM_Exit VMOperation's doit method.
4827     if (atexit(perfMemory_exit_helper) != 0) {
4828       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4829     }
4830   }
4831 
4832   // initialize thread priority policy
4833   prio_init();
4834 
4835   return JNI_OK;
4836 }
4837 
4838 // Mark the polling page as unreadable
4839 void os::make_polling_page_unreadable(void) {
4840   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4841     fatal("Could not disable polling page");
4842   }
4843 }
4844 
4845 // Mark the polling page as readable
4846 void os::make_polling_page_readable(void) {
4847   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4848     fatal("Could not enable polling page");
4849   }
4850 }
4851 
4852 int os::active_processor_count() {
4853   // Linux doesn't yet have a (official) notion of processor sets,
4854   // so just return the number of online processors.
4855   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4856   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4857   return online_cpus;
4858 }
4859 
4860 void os::set_native_thread_name(const char *name) {
4861   if (Linux::_pthread_setname_np) {
4862     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4863     snprintf(buf, sizeof(buf), "%s", name);
4864     buf[sizeof(buf) - 1] = '\0';
4865     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4866     // ERANGE should not happen; all other errors should just be ignored.
4867     assert(rc != ERANGE, "pthread_setname_np failed");
4868   }
4869 }
4870 
4871 bool os::distribute_processes(uint length, uint* distribution) {
4872   // Not yet implemented.
4873   return false;
4874 }
4875 
4876 bool os::bind_to_processor(uint processor_id) {
4877   // Not yet implemented.
4878   return false;
4879 }
4880 
4881 ///
4882 
4883 void os::SuspendedThreadTask::internal_do_task() {
4884   if (do_suspend(_thread->osthread())) {
4885     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4886     do_task(context);
4887     do_resume(_thread->osthread());
4888   }
4889 }
4890 
4891 class PcFetcher : public os::SuspendedThreadTask {
4892  public:
4893   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4894   ExtendedPC result();
4895  protected:
4896   void do_task(const os::SuspendedThreadTaskContext& context);
4897  private:
4898   ExtendedPC _epc;
4899 };
4900 
4901 ExtendedPC PcFetcher::result() {
4902   guarantee(is_done(), "task is not done yet.");
4903   return _epc;
4904 }
4905 
4906 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4907   Thread* thread = context.thread();
4908   OSThread* osthread = thread->osthread();
4909   if (osthread->ucontext() != NULL) {
4910     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4911   } else {
4912     // NULL context is unexpected, double-check this is the VMThread
4913     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4914   }
4915 }
4916 
4917 // Suspends the target using the signal mechanism and then grabs the PC before
4918 // resuming the target. Used by the flat-profiler only
4919 ExtendedPC os::get_thread_pc(Thread* thread) {
4920   // Make sure that it is called by the watcher for the VMThread
4921   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4922   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4923 
4924   PcFetcher fetcher(thread);
4925   fetcher.run();
4926   return fetcher.result();
4927 }
4928 
4929 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4930                                    pthread_mutex_t *_mutex,
4931                                    const struct timespec *_abstime) {
4932   if (is_NPTL()) {
4933     return pthread_cond_timedwait(_cond, _mutex, _abstime);
4934   } else {
4935     // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4936     // word back to default 64bit precision if condvar is signaled. Java
4937     // wants 53bit precision.  Save and restore current value.
4938     int fpu = get_fpu_control_word();
4939     int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4940     set_fpu_control_word(fpu);
4941     return status;
4942   }
4943 }
4944 
4945 ////////////////////////////////////////////////////////////////////////////////
4946 // debug support
4947 
4948 bool os::find(address addr, outputStream* st) {
4949   Dl_info dlinfo;
4950   memset(&dlinfo, 0, sizeof(dlinfo));
4951   if (dladdr(addr, &dlinfo) != 0) {
4952     st->print(PTR_FORMAT ": ", addr);
4953     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4954       st->print("%s+%#x", dlinfo.dli_sname,
4955                 addr - (intptr_t)dlinfo.dli_saddr);
4956     } else if (dlinfo.dli_fbase != NULL) {
4957       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4958     } else {
4959       st->print("<absolute address>");
4960     }
4961     if (dlinfo.dli_fname != NULL) {
4962       st->print(" in %s", dlinfo.dli_fname);
4963     }
4964     if (dlinfo.dli_fbase != NULL) {
4965       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4966     }
4967     st->cr();
4968 
4969     if (Verbose) {
4970       // decode some bytes around the PC
4971       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4972       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4973       address       lowest = (address) dlinfo.dli_sname;
4974       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4975       if (begin < lowest)  begin = lowest;
4976       Dl_info dlinfo2;
4977       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4978           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4979         end = (address) dlinfo2.dli_saddr;
4980       }
4981       Disassembler::decode(begin, end, st);
4982     }
4983     return true;
4984   }
4985   return false;
4986 }
4987 
4988 ////////////////////////////////////////////////////////////////////////////////
4989 // misc
4990 
4991 // This does not do anything on Linux. This is basically a hook for being
4992 // able to use structured exception handling (thread-local exception filters)
4993 // on, e.g., Win32.
4994 void
4995 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4996                          JavaCallArguments* args, Thread* thread) {
4997   f(value, method, args, thread);
4998 }
4999 
5000 void os::print_statistics() {
5001 }
5002 
5003 int os::message_box(const char* title, const char* message) {
5004   int i;
5005   fdStream err(defaultStream::error_fd());
5006   for (i = 0; i < 78; i++) err.print_raw("=");
5007   err.cr();
5008   err.print_raw_cr(title);
5009   for (i = 0; i < 78; i++) err.print_raw("-");
5010   err.cr();
5011   err.print_raw_cr(message);
5012   for (i = 0; i < 78; i++) err.print_raw("=");
5013   err.cr();
5014 
5015   char buf[16];
5016   // Prevent process from exiting upon "read error" without consuming all CPU
5017   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5018 
5019   return buf[0] == 'y' || buf[0] == 'Y';
5020 }
5021 
5022 int os::stat(const char *path, struct stat *sbuf) {
5023   char pathbuf[MAX_PATH];
5024   if (strlen(path) > MAX_PATH - 1) {
5025     errno = ENAMETOOLONG;
5026     return -1;
5027   }
5028   os::native_path(strcpy(pathbuf, path));
5029   return ::stat(pathbuf, sbuf);
5030 }
5031 
5032 bool os::check_heap(bool force) {
5033   return true;
5034 }
5035 
5036 // Is a (classpath) directory empty?
5037 bool os::dir_is_empty(const char* path) {
5038   DIR *dir = NULL;
5039   struct dirent *ptr;
5040 
5041   dir = opendir(path);
5042   if (dir == NULL) return true;
5043 
5044   // Scan the directory
5045   bool result = true;
5046   char buf[sizeof(struct dirent) + MAX_PATH];
5047   while (result && (ptr = ::readdir(dir)) != NULL) {
5048     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5049       result = false;
5050     }
5051   }
5052   closedir(dir);
5053   return result;
5054 }
5055 
5056 // This code originates from JDK's sysOpen and open64_w
5057 // from src/solaris/hpi/src/system_md.c
5058 
5059 int os::open(const char *path, int oflag, int mode) {
5060   if (strlen(path) > MAX_PATH - 1) {
5061     errno = ENAMETOOLONG;
5062     return -1;
5063   }
5064 
5065   // All file descriptors that are opened in the Java process and not
5066   // specifically destined for a subprocess should have the close-on-exec
5067   // flag set.  If we don't set it, then careless 3rd party native code
5068   // might fork and exec without closing all appropriate file descriptors
5069   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5070   // turn might:
5071   //
5072   // - cause end-of-file to fail to be detected on some file
5073   //   descriptors, resulting in mysterious hangs, or
5074   //
5075   // - might cause an fopen in the subprocess to fail on a system
5076   //   suffering from bug 1085341.
5077   //
5078   // (Yes, the default setting of the close-on-exec flag is a Unix
5079   // design flaw)
5080   //
5081   // See:
5082   // 1085341: 32-bit stdio routines should support file descriptors >255
5083   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5084   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5085   //
5086   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5087   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5088   // because it saves a system call and removes a small window where the flag
5089   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5090   // and we fall back to using FD_CLOEXEC (see below).
5091 #ifdef O_CLOEXEC
5092   oflag |= O_CLOEXEC;
5093 #endif
5094 
5095   int fd = ::open64(path, oflag, mode);
5096   if (fd == -1) return -1;
5097 
5098   //If the open succeeded, the file might still be a directory
5099   {
5100     struct stat64 buf64;
5101     int ret = ::fstat64(fd, &buf64);
5102     int st_mode = buf64.st_mode;
5103 
5104     if (ret != -1) {
5105       if ((st_mode & S_IFMT) == S_IFDIR) {
5106         errno = EISDIR;
5107         ::close(fd);
5108         return -1;
5109       }
5110     } else {
5111       ::close(fd);
5112       return -1;
5113     }
5114   }
5115 
5116 #ifdef FD_CLOEXEC
5117   // Validate that the use of the O_CLOEXEC flag on open above worked.
5118   // With recent kernels, we will perform this check exactly once.
5119   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5120   if (!O_CLOEXEC_is_known_to_work) {
5121     int flags = ::fcntl(fd, F_GETFD);
5122     if (flags != -1) {
5123       if ((flags & FD_CLOEXEC) != 0)
5124         O_CLOEXEC_is_known_to_work = 1;
5125       else
5126         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5127     }
5128   }
5129 #endif
5130 
5131   return fd;
5132 }
5133 
5134 
5135 // create binary file, rewriting existing file if required
5136 int os::create_binary_file(const char* path, bool rewrite_existing) {
5137   int oflags = O_WRONLY | O_CREAT;
5138   if (!rewrite_existing) {
5139     oflags |= O_EXCL;
5140   }
5141   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5142 }
5143 
5144 // return current position of file pointer
5145 jlong os::current_file_offset(int fd) {
5146   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5147 }
5148 
5149 // move file pointer to the specified offset
5150 jlong os::seek_to_file_offset(int fd, jlong offset) {
5151   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5152 }
5153 
5154 // This code originates from JDK's sysAvailable
5155 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5156 
5157 int os::available(int fd, jlong *bytes) {
5158   jlong cur, end;
5159   int mode;
5160   struct stat64 buf64;
5161 
5162   if (::fstat64(fd, &buf64) >= 0) {
5163     mode = buf64.st_mode;
5164     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5165       // XXX: is the following call interruptible? If so, this might
5166       // need to go through the INTERRUPT_IO() wrapper as for other
5167       // blocking, interruptible calls in this file.
5168       int n;
5169       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5170         *bytes = n;
5171         return 1;
5172       }
5173     }
5174   }
5175   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5176     return 0;
5177   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5178     return 0;
5179   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5180     return 0;
5181   }
5182   *bytes = end - cur;
5183   return 1;
5184 }
5185 
5186 // Map a block of memory.
5187 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5188                         char *addr, size_t bytes, bool read_only,
5189                         bool allow_exec) {
5190   int prot;
5191   int flags = MAP_PRIVATE;
5192 
5193   if (read_only) {
5194     prot = PROT_READ;
5195   } else {
5196     prot = PROT_READ | PROT_WRITE;
5197   }
5198 
5199   if (allow_exec) {
5200     prot |= PROT_EXEC;
5201   }
5202 
5203   if (addr != NULL) {
5204     flags |= MAP_FIXED;
5205   }
5206 
5207   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5208                                      fd, file_offset);
5209   if (mapped_address == MAP_FAILED) {
5210     return NULL;
5211   }
5212   return mapped_address;
5213 }
5214 
5215 
5216 // Remap a block of memory.
5217 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5218                           char *addr, size_t bytes, bool read_only,
5219                           bool allow_exec) {
5220   // same as map_memory() on this OS
5221   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5222                         allow_exec);
5223 }
5224 
5225 
5226 // Unmap a block of memory.
5227 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5228   return munmap(addr, bytes) == 0;
5229 }
5230 
5231 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5232 
5233 static clockid_t thread_cpu_clockid(Thread* thread) {
5234   pthread_t tid = thread->osthread()->pthread_id();
5235   clockid_t clockid;
5236 
5237   // Get thread clockid
5238   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5239   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5240   return clockid;
5241 }
5242 
5243 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5244 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5245 // of a thread.
5246 //
5247 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5248 // the fast estimate available on the platform.
5249 
5250 jlong os::current_thread_cpu_time() {
5251   if (os::Linux::supports_fast_thread_cpu_time()) {
5252     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5253   } else {
5254     // return user + sys since the cost is the same
5255     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5256   }
5257 }
5258 
5259 jlong os::thread_cpu_time(Thread* thread) {
5260   // consistent with what current_thread_cpu_time() returns
5261   if (os::Linux::supports_fast_thread_cpu_time()) {
5262     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5263   } else {
5264     return slow_thread_cpu_time(thread, true /* user + sys */);
5265   }
5266 }
5267 
5268 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5269   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5270     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5271   } else {
5272     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5273   }
5274 }
5275 
5276 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5277   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5278     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5279   } else {
5280     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5281   }
5282 }
5283 
5284 //  -1 on error.
5285 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5286   pid_t  tid = thread->osthread()->thread_id();
5287   char *s;
5288   char stat[2048];
5289   int statlen;
5290   char proc_name[64];
5291   int count;
5292   long sys_time, user_time;
5293   char cdummy;
5294   int idummy;
5295   long ldummy;
5296   FILE *fp;
5297 
5298   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5299   fp = fopen(proc_name, "r");
5300   if (fp == NULL) return -1;
5301   statlen = fread(stat, 1, 2047, fp);
5302   stat[statlen] = '\0';
5303   fclose(fp);
5304 
5305   // Skip pid and the command string. Note that we could be dealing with
5306   // weird command names, e.g. user could decide to rename java launcher
5307   // to "java 1.4.2 :)", then the stat file would look like
5308   //                1234 (java 1.4.2 :)) R ... ...
5309   // We don't really need to know the command string, just find the last
5310   // occurrence of ")" and then start parsing from there. See bug 4726580.
5311   s = strrchr(stat, ')');
5312   if (s == NULL) return -1;
5313 
5314   // Skip blank chars
5315   do { s++; } while (s && isspace(*s));
5316 
5317   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5318                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5319                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5320                  &user_time, &sys_time);
5321   if (count != 13) return -1;
5322   if (user_sys_cpu_time) {
5323     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5324   } else {
5325     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5326   }
5327 }
5328 
5329 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5330   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5331   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5332   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5333   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5334 }
5335 
5336 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5337   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5338   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5339   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5340   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5341 }
5342 
5343 bool os::is_thread_cpu_time_supported() {
5344   return true;
5345 }
5346 
5347 // System loadavg support.  Returns -1 if load average cannot be obtained.
5348 // Linux doesn't yet have a (official) notion of processor sets,
5349 // so just return the system wide load average.
5350 int os::loadavg(double loadavg[], int nelem) {
5351   return ::getloadavg(loadavg, nelem);
5352 }
5353 
5354 void os::pause() {
5355   char filename[MAX_PATH];
5356   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5357     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5358   } else {
5359     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5360   }
5361 
5362   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5363   if (fd != -1) {
5364     struct stat buf;
5365     ::close(fd);
5366     while (::stat(filename, &buf) == 0) {
5367       (void)::poll(NULL, 0, 100);
5368     }
5369   } else {
5370     jio_fprintf(stderr,
5371                 "Could not open pause file '%s', continuing immediately.\n", filename);
5372   }
5373 }
5374 
5375 
5376 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5377 // comment paragraphs are worth repeating here:
5378 //
5379 // Assumption:
5380 //    Only one parker can exist on an event, which is why we allocate
5381 //    them per-thread. Multiple unparkers can coexist.
5382 //
5383 // _Event serves as a restricted-range semaphore.
5384 //   -1 : thread is blocked, i.e. there is a waiter
5385 //    0 : neutral: thread is running or ready,
5386 //        could have been signaled after a wait started
5387 //    1 : signaled - thread is running or ready
5388 //
5389 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5390 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5391 // For specifics regarding the bug see GLIBC BUGID 261237 :
5392 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5393 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5394 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5395 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5396 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5397 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5398 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5399 // of libpthread avoids the problem, but isn't practical.
5400 //
5401 // Possible remedies:
5402 //
5403 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5404 //      This is palliative and probabilistic, however.  If the thread is preempted
5405 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5406 //      than the minimum period may have passed, and the abstime may be stale (in the
5407 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5408 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5409 //
5410 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5411 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5412 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5413 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5414 //      thread.
5415 //
5416 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5417 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5418 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5419 //      This also works well.  In fact it avoids kernel-level scalability impediments
5420 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5421 //      timers in a graceful fashion.
5422 //
5423 // 4.   When the abstime value is in the past it appears that control returns
5424 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5425 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5426 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5427 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5428 //      It may be possible to avoid reinitialization by checking the return
5429 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5430 //      condvar we must establish the invariant that cond_signal() is only called
5431 //      within critical sections protected by the adjunct mutex.  This prevents
5432 //      cond_signal() from "seeing" a condvar that's in the midst of being
5433 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5434 //      desirable signal-after-unlock optimization that avoids futile context switching.
5435 //
5436 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5437 //      structure when a condvar is used or initialized.  cond_destroy()  would
5438 //      release the helper structure.  Our reinitialize-after-timedwait fix
5439 //      put excessive stress on malloc/free and locks protecting the c-heap.
5440 //
5441 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5442 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5443 // and only enabling the work-around for vulnerable environments.
5444 
5445 // utility to compute the abstime argument to timedwait:
5446 // millis is the relative timeout time
5447 // abstime will be the absolute timeout time
5448 // TODO: replace compute_abstime() with unpackTime()
5449 
5450 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5451   if (millis < 0)  millis = 0;
5452 
5453   jlong seconds = millis / 1000;
5454   millis %= 1000;
5455   if (seconds > 50000000) { // see man cond_timedwait(3T)
5456     seconds = 50000000;
5457   }
5458 
5459   if (os::supports_monotonic_clock()) {
5460     struct timespec now;
5461     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5462     assert_status(status == 0, status, "clock_gettime");
5463     abstime->tv_sec = now.tv_sec  + seconds;
5464     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5465     if (nanos >= NANOSECS_PER_SEC) {
5466       abstime->tv_sec += 1;
5467       nanos -= NANOSECS_PER_SEC;
5468     }
5469     abstime->tv_nsec = nanos;
5470   } else {
5471     struct timeval now;
5472     int status = gettimeofday(&now, NULL);
5473     assert(status == 0, "gettimeofday");
5474     abstime->tv_sec = now.tv_sec  + seconds;
5475     long usec = now.tv_usec + millis * 1000;
5476     if (usec >= 1000000) {
5477       abstime->tv_sec += 1;
5478       usec -= 1000000;
5479     }
5480     abstime->tv_nsec = usec * 1000;
5481   }
5482   return abstime;
5483 }
5484 
5485 void os::PlatformEvent::park() {       // AKA "down()"
5486   // Transitions for _Event:
5487   //   -1 => -1 : illegal
5488   //    1 =>  0 : pass - return immediately
5489   //    0 => -1 : block; then set _Event to 0 before returning
5490 
5491   // Invariant: Only the thread associated with the Event/PlatformEvent
5492   // may call park().
5493   // TODO: assert that _Assoc != NULL or _Assoc == Self
5494   assert(_nParked == 0, "invariant");
5495 
5496   int v;
5497   for (;;) {
5498     v = _Event;
5499     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5500   }
5501   guarantee(v >= 0, "invariant");
5502   if (v == 0) {
5503     // Do this the hard way by blocking ...
5504     int status = pthread_mutex_lock(_mutex);
5505     assert_status(status == 0, status, "mutex_lock");
5506     guarantee(_nParked == 0, "invariant");
5507     ++_nParked;
5508     while (_Event < 0) {
5509       status = pthread_cond_wait(_cond, _mutex);
5510       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5511       // Treat this the same as if the wait was interrupted
5512       if (status == ETIME) { status = EINTR; }
5513       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5514     }
5515     --_nParked;
5516 
5517     _Event = 0;
5518     status = pthread_mutex_unlock(_mutex);
5519     assert_status(status == 0, status, "mutex_unlock");
5520     // Paranoia to ensure our locked and lock-free paths interact
5521     // correctly with each other.
5522     OrderAccess::fence();
5523   }
5524   guarantee(_Event >= 0, "invariant");
5525 }
5526 
5527 int os::PlatformEvent::park(jlong millis) {
5528   // Transitions for _Event:
5529   //   -1 => -1 : illegal
5530   //    1 =>  0 : pass - return immediately
5531   //    0 => -1 : block; then set _Event to 0 before returning
5532 
5533   guarantee(_nParked == 0, "invariant");
5534 
5535   int v;
5536   for (;;) {
5537     v = _Event;
5538     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5539   }
5540   guarantee(v >= 0, "invariant");
5541   if (v != 0) return OS_OK;
5542 
5543   // We do this the hard way, by blocking the thread.
5544   // Consider enforcing a minimum timeout value.
5545   struct timespec abst;
5546   compute_abstime(&abst, millis);
5547 
5548   int ret = OS_TIMEOUT;
5549   int status = pthread_mutex_lock(_mutex);
5550   assert_status(status == 0, status, "mutex_lock");
5551   guarantee(_nParked == 0, "invariant");
5552   ++_nParked;
5553 
5554   // Object.wait(timo) will return because of
5555   // (a) notification
5556   // (b) timeout
5557   // (c) thread.interrupt
5558   //
5559   // Thread.interrupt and object.notify{All} both call Event::set.
5560   // That is, we treat thread.interrupt as a special case of notification.
5561   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5562   // We assume all ETIME returns are valid.
5563   //
5564   // TODO: properly differentiate simultaneous notify+interrupt.
5565   // In that case, we should propagate the notify to another waiter.
5566 
5567   while (_Event < 0) {
5568     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5569     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5570       pthread_cond_destroy(_cond);
5571       pthread_cond_init(_cond, os::Linux::condAttr());
5572     }
5573     assert_status(status == 0 || status == EINTR ||
5574                   status == ETIME || status == ETIMEDOUT,
5575                   status, "cond_timedwait");
5576     if (!FilterSpuriousWakeups) break;                 // previous semantics
5577     if (status == ETIME || status == ETIMEDOUT) break;
5578     // We consume and ignore EINTR and spurious wakeups.
5579   }
5580   --_nParked;
5581   if (_Event >= 0) {
5582     ret = OS_OK;
5583   }
5584   _Event = 0;
5585   status = pthread_mutex_unlock(_mutex);
5586   assert_status(status == 0, status, "mutex_unlock");
5587   assert(_nParked == 0, "invariant");
5588   // Paranoia to ensure our locked and lock-free paths interact
5589   // correctly with each other.
5590   OrderAccess::fence();
5591   return ret;
5592 }
5593 
5594 void os::PlatformEvent::unpark() {
5595   // Transitions for _Event:
5596   //    0 => 1 : just return
5597   //    1 => 1 : just return
5598   //   -1 => either 0 or 1; must signal target thread
5599   //         That is, we can safely transition _Event from -1 to either
5600   //         0 or 1.
5601   // See also: "Semaphores in Plan 9" by Mullender & Cox
5602   //
5603   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5604   // that it will take two back-to-back park() calls for the owning
5605   // thread to block. This has the benefit of forcing a spurious return
5606   // from the first park() call after an unpark() call which will help
5607   // shake out uses of park() and unpark() without condition variables.
5608 
5609   if (Atomic::xchg(1, &_Event) >= 0) return;
5610 
5611   // Wait for the thread associated with the event to vacate
5612   int status = pthread_mutex_lock(_mutex);
5613   assert_status(status == 0, status, "mutex_lock");
5614   int AnyWaiters = _nParked;
5615   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5616   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5617     AnyWaiters = 0;
5618     pthread_cond_signal(_cond);
5619   }
5620   status = pthread_mutex_unlock(_mutex);
5621   assert_status(status == 0, status, "mutex_unlock");
5622   if (AnyWaiters != 0) {
5623     // Note that we signal() *after* dropping the lock for "immortal" Events.
5624     // This is safe and avoids a common class of  futile wakeups.  In rare
5625     // circumstances this can cause a thread to return prematurely from
5626     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5627     // will simply re-test the condition and re-park itself.
5628     // This provides particular benefit if the underlying platform does not
5629     // provide wait morphing.
5630     status = pthread_cond_signal(_cond);
5631     assert_status(status == 0, status, "cond_signal");
5632   }
5633 }
5634 
5635 
5636 // JSR166
5637 // -------------------------------------------------------
5638 
5639 // The solaris and linux implementations of park/unpark are fairly
5640 // conservative for now, but can be improved. They currently use a
5641 // mutex/condvar pair, plus a a count.
5642 // Park decrements count if > 0, else does a condvar wait.  Unpark
5643 // sets count to 1 and signals condvar.  Only one thread ever waits
5644 // on the condvar. Contention seen when trying to park implies that someone
5645 // is unparking you, so don't wait. And spurious returns are fine, so there
5646 // is no need to track notifications.
5647 
5648 // This code is common to linux and solaris and will be moved to a
5649 // common place in dolphin.
5650 //
5651 // The passed in time value is either a relative time in nanoseconds
5652 // or an absolute time in milliseconds. Either way it has to be unpacked
5653 // into suitable seconds and nanoseconds components and stored in the
5654 // given timespec structure.
5655 // Given time is a 64-bit value and the time_t used in the timespec is only
5656 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5657 // overflow if times way in the future are given. Further on Solaris versions
5658 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5659 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5660 // As it will be 28 years before "now + 100000000" will overflow we can
5661 // ignore overflow and just impose a hard-limit on seconds using the value
5662 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5663 // years from "now".
5664 
5665 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5666   assert(time > 0, "convertTime");
5667   time_t max_secs = 0;
5668 
5669   if (!os::supports_monotonic_clock() || isAbsolute) {
5670     struct timeval now;
5671     int status = gettimeofday(&now, NULL);
5672     assert(status == 0, "gettimeofday");
5673 
5674     max_secs = now.tv_sec + MAX_SECS;
5675 
5676     if (isAbsolute) {
5677       jlong secs = time / 1000;
5678       if (secs > max_secs) {
5679         absTime->tv_sec = max_secs;
5680       } else {
5681         absTime->tv_sec = secs;
5682       }
5683       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5684     } else {
5685       jlong secs = time / NANOSECS_PER_SEC;
5686       if (secs >= MAX_SECS) {
5687         absTime->tv_sec = max_secs;
5688         absTime->tv_nsec = 0;
5689       } else {
5690         absTime->tv_sec = now.tv_sec + secs;
5691         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5692         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5693           absTime->tv_nsec -= NANOSECS_PER_SEC;
5694           ++absTime->tv_sec; // note: this must be <= max_secs
5695         }
5696       }
5697     }
5698   } else {
5699     // must be relative using monotonic clock
5700     struct timespec now;
5701     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5702     assert_status(status == 0, status, "clock_gettime");
5703     max_secs = now.tv_sec + MAX_SECS;
5704     jlong secs = time / NANOSECS_PER_SEC;
5705     if (secs >= MAX_SECS) {
5706       absTime->tv_sec = max_secs;
5707       absTime->tv_nsec = 0;
5708     } else {
5709       absTime->tv_sec = now.tv_sec + secs;
5710       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5711       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5712         absTime->tv_nsec -= NANOSECS_PER_SEC;
5713         ++absTime->tv_sec; // note: this must be <= max_secs
5714       }
5715     }
5716   }
5717   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5718   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5719   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5720   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5721 }
5722 
5723 void Parker::park(bool isAbsolute, jlong time) {
5724   // Ideally we'd do something useful while spinning, such
5725   // as calling unpackTime().
5726 
5727   // Optional fast-path check:
5728   // Return immediately if a permit is available.
5729   // We depend on Atomic::xchg() having full barrier semantics
5730   // since we are doing a lock-free update to _counter.
5731   if (Atomic::xchg(0, &_counter) > 0) return;
5732 
5733   Thread* thread = Thread::current();
5734   assert(thread->is_Java_thread(), "Must be JavaThread");
5735   JavaThread *jt = (JavaThread *)thread;
5736 
5737   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5738   // Check interrupt before trying to wait
5739   if (Thread::is_interrupted(thread, false)) {
5740     return;
5741   }
5742 
5743   // Next, demultiplex/decode time arguments
5744   timespec absTime;
5745   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5746     return;
5747   }
5748   if (time > 0) {
5749     unpackTime(&absTime, isAbsolute, time);
5750   }
5751 
5752 
5753   // Enter safepoint region
5754   // Beware of deadlocks such as 6317397.
5755   // The per-thread Parker:: mutex is a classic leaf-lock.
5756   // In particular a thread must never block on the Threads_lock while
5757   // holding the Parker:: mutex.  If safepoints are pending both the
5758   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5759   ThreadBlockInVM tbivm(jt);
5760 
5761   // Don't wait if cannot get lock since interference arises from
5762   // unblocking.  Also. check interrupt before trying wait
5763   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5764     return;
5765   }
5766 
5767   int status;
5768   if (_counter > 0)  { // no wait needed
5769     _counter = 0;
5770     status = pthread_mutex_unlock(_mutex);
5771     assert(status == 0, "invariant");
5772     // Paranoia to ensure our locked and lock-free paths interact
5773     // correctly with each other and Java-level accesses.
5774     OrderAccess::fence();
5775     return;
5776   }
5777 
5778 #ifdef ASSERT
5779   // Don't catch signals while blocked; let the running threads have the signals.
5780   // (This allows a debugger to break into the running thread.)
5781   sigset_t oldsigs;
5782   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5783   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5784 #endif
5785 
5786   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5787   jt->set_suspend_equivalent();
5788   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5789 
5790   assert(_cur_index == -1, "invariant");
5791   if (time == 0) {
5792     _cur_index = REL_INDEX; // arbitrary choice when not timed
5793     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5794   } else {
5795     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5796     status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5797     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5798       pthread_cond_destroy(&_cond[_cur_index]);
5799       pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5800     }
5801   }
5802   _cur_index = -1;
5803   assert_status(status == 0 || status == EINTR ||
5804                 status == ETIME || status == ETIMEDOUT,
5805                 status, "cond_timedwait");
5806 
5807 #ifdef ASSERT
5808   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5809 #endif
5810 
5811   _counter = 0;
5812   status = pthread_mutex_unlock(_mutex);
5813   assert_status(status == 0, status, "invariant");
5814   // Paranoia to ensure our locked and lock-free paths interact
5815   // correctly with each other and Java-level accesses.
5816   OrderAccess::fence();
5817 
5818   // If externally suspended while waiting, re-suspend
5819   if (jt->handle_special_suspend_equivalent_condition()) {
5820     jt->java_suspend_self();
5821   }
5822 }
5823 
5824 void Parker::unpark() {
5825   int status = pthread_mutex_lock(_mutex);
5826   assert(status == 0, "invariant");
5827   const int s = _counter;
5828   _counter = 1;
5829   if (s < 1) {
5830     // thread might be parked
5831     if (_cur_index != -1) {
5832       // thread is definitely parked
5833       if (WorkAroundNPTLTimedWaitHang) {
5834         status = pthread_cond_signal(&_cond[_cur_index]);
5835         assert(status == 0, "invariant");
5836         status = pthread_mutex_unlock(_mutex);
5837         assert(status == 0, "invariant");
5838       } else {
5839         status = pthread_mutex_unlock(_mutex);
5840         assert(status == 0, "invariant");
5841         status = pthread_cond_signal(&_cond[_cur_index]);
5842         assert(status == 0, "invariant");
5843       }
5844     } else {
5845       pthread_mutex_unlock(_mutex);
5846       assert(status == 0, "invariant");
5847     }
5848   } else {
5849     pthread_mutex_unlock(_mutex);
5850     assert(status == 0, "invariant");
5851   }
5852 }
5853 
5854 
5855 extern char** environ;
5856 
5857 #ifndef __NR_fork
5858   #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) AARCH64_ONLY(1079)
5859 #endif
5860 
5861 #ifndef __NR_execve
5862   #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) AARCH64_ONLY(221)
5863 #endif
5864 
5865 // Run the specified command in a separate process. Return its exit value,
5866 // or -1 on failure (e.g. can't fork a new process).
5867 // Unlike system(), this function can be called from signal handler. It
5868 // doesn't block SIGINT et al.
5869 int os::fork_and_exec(char* cmd) {
5870   const char * argv[4] = {"sh", "-c", cmd, NULL};
5871 
5872   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5873   // pthread_atfork handlers and reset pthread library. All we need is a
5874   // separate process to execve. Make a direct syscall to fork process.
5875   // On IA64 there's no fork syscall, we have to use fork() and hope for
5876   // the best...
5877   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5878   IA64_ONLY(fork();)
5879 
5880   if (pid < 0) {
5881     // fork failed
5882     return -1;
5883 
5884   } else if (pid == 0) {
5885     // child process
5886 
5887     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5888     // first to kill every thread on the thread list. Because this list is
5889     // not reset by fork() (see notes above), execve() will instead kill
5890     // every thread in the parent process. We know this is the only thread
5891     // in the new process, so make a system call directly.
5892     // IA64 should use normal execve() from glibc to match the glibc fork()
5893     // above.
5894     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5895     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5896 
5897     // execve failed
5898     _exit(-1);
5899 
5900   } else  {
5901     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5902     // care about the actual exit code, for now.
5903 
5904     int status;
5905 
5906     // Wait for the child process to exit.  This returns immediately if
5907     // the child has already exited. */
5908     while (waitpid(pid, &status, 0) < 0) {
5909       switch (errno) {
5910       case ECHILD: return 0;
5911       case EINTR: break;
5912       default: return -1;
5913       }
5914     }
5915 
5916     if (WIFEXITED(status)) {
5917       // The child exited normally; get its exit code.
5918       return WEXITSTATUS(status);
5919     } else if (WIFSIGNALED(status)) {
5920       // The child exited because of a signal
5921       // The best value to return is 0x80 + signal number,
5922       // because that is what all Unix shells do, and because
5923       // it allows callers to distinguish between process exit and
5924       // process death by signal.
5925       return 0x80 + WTERMSIG(status);
5926     } else {
5927       // Unknown exit code; pass it through
5928       return status;
5929     }
5930   }
5931 }
5932 
5933 // is_headless_jre()
5934 //
5935 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5936 // in order to report if we are running in a headless jre
5937 //
5938 // Since JDK8 xawt/libmawt.so was moved into the same directory
5939 // as libawt.so, and renamed libawt_xawt.so
5940 //
5941 bool os::is_headless_jre() {
5942   struct stat statbuf;
5943   char buf[MAXPATHLEN];
5944   char libmawtpath[MAXPATHLEN];
5945   const char *xawtstr  = "/xawt/libmawt.so";
5946   const char *new_xawtstr = "/libawt_xawt.so";
5947   char *p;
5948 
5949   // Get path to libjvm.so
5950   os::jvm_path(buf, sizeof(buf));
5951 
5952   // Get rid of libjvm.so
5953   p = strrchr(buf, '/');
5954   if (p == NULL) {
5955     return false;
5956   } else {
5957     *p = '\0';
5958   }
5959 
5960   // Get rid of client or server
5961   p = strrchr(buf, '/');
5962   if (p == NULL) {
5963     return false;
5964   } else {
5965     *p = '\0';
5966   }
5967 
5968   // check xawt/libmawt.so
5969   strcpy(libmawtpath, buf);
5970   strcat(libmawtpath, xawtstr);
5971   if (::stat(libmawtpath, &statbuf) == 0) return false;
5972 
5973   // check libawt_xawt.so
5974   strcpy(libmawtpath, buf);
5975   strcat(libmawtpath, new_xawtstr);
5976   if (::stat(libmawtpath, &statbuf) == 0) return false;
5977 
5978   return true;
5979 }
5980 
5981 // Get the default path to the core file
5982 // Returns the length of the string
5983 int os::get_core_path(char* buffer, size_t bufferSize) {
5984   /*
5985    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5986    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5987    */
5988   const int core_pattern_len = 129;
5989   char core_pattern[core_pattern_len] = {0};
5990 
5991   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5992   if (core_pattern_file != -1) {
5993     ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5994     ::close(core_pattern_file);
5995 
5996     if (ret > 0) {
5997       char *last_char = core_pattern + strlen(core_pattern) - 1;
5998 
5999       if (*last_char == '\n') {
6000         *last_char = '\0';
6001       }
6002     }
6003   }
6004 
6005   if (strlen(core_pattern) == 0) {
6006     return 0;
6007   }
6008 
6009   char *pid_pos = strstr(core_pattern, "%p");
6010   size_t written;
6011 
6012   if (core_pattern[0] == '/') {
6013     written = jio_snprintf(buffer, bufferSize, core_pattern);
6014   } else {
6015     char cwd[PATH_MAX];
6016 
6017     const char* p = get_current_directory(cwd, PATH_MAX);
6018     if (p == NULL) {
6019       assert(p != NULL, "failed to get current directory");
6020       return 0;
6021     }
6022 
6023     if (core_pattern[0] == '|') {
6024       written = jio_snprintf(buffer, bufferSize,
6025                         "\"%s\" (or dumping to %s/core.%d)",
6026                                      &core_pattern[1], p, current_process_id());
6027     } else {
6028       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6029     }
6030   }
6031 
6032   if ((written >= 0) && (written < bufferSize)
6033             && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6034     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6035 
6036     if (core_uses_pid_file != -1) {
6037       char core_uses_pid = 0;
6038       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6039       ::close(core_uses_pid_file);
6040 
6041       if (core_uses_pid == '1'){
6042         jio_snprintf(buffer + written, bufferSize - written,
6043                                           ".%d", current_process_id());
6044       }
6045     }
6046   }
6047 
6048   return strlen(buffer);
6049 }
6050 
6051 /////////////// Unit tests ///////////////
6052 
6053 #ifndef PRODUCT
6054 
6055 #define test_log(...)              \
6056   do {                             \
6057     if (VerboseInternalVMTests) {  \
6058       tty->print_cr(__VA_ARGS__);  \
6059       tty->flush();                \
6060     }                              \
6061   } while (false)
6062 
6063 class TestReserveMemorySpecial : AllStatic {
6064  public:
6065   static void small_page_write(void* addr, size_t size) {
6066     size_t page_size = os::vm_page_size();
6067 
6068     char* end = (char*)addr + size;
6069     for (char* p = (char*)addr; p < end; p += page_size) {
6070       *p = 1;
6071     }
6072   }
6073 
6074   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6075     if (!UseHugeTLBFS) {
6076       return;
6077     }
6078 
6079     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6080 
6081     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6082 
6083     if (addr != NULL) {
6084       small_page_write(addr, size);
6085 
6086       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6087     }
6088   }
6089 
6090   static void test_reserve_memory_special_huge_tlbfs_only() {
6091     if (!UseHugeTLBFS) {
6092       return;
6093     }
6094 
6095     size_t lp = os::large_page_size();
6096 
6097     for (size_t size = lp; size <= lp * 10; size += lp) {
6098       test_reserve_memory_special_huge_tlbfs_only(size);
6099     }
6100   }
6101 
6102   static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6103     if (!UseHugeTLBFS) {
6104       return;
6105     }
6106 
6107     test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6108              size, alignment);
6109 
6110     assert(size >= os::large_page_size(), "Incorrect input to test");
6111 
6112     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6113 
6114     if (addr != NULL) {
6115       small_page_write(addr, size);
6116 
6117       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6118     }
6119   }
6120 
6121   static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6122     size_t lp = os::large_page_size();
6123     size_t ag = os::vm_allocation_granularity();
6124 
6125     for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6126       test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6127     }
6128   }
6129 
6130   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6131     size_t lp = os::large_page_size();
6132     size_t ag = os::vm_allocation_granularity();
6133 
6134     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6135     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6136     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6137     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6138     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6139     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6140     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6141     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6142     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6143   }
6144 
6145   static void test_reserve_memory_special_huge_tlbfs() {
6146     if (!UseHugeTLBFS) {
6147       return;
6148     }
6149 
6150     test_reserve_memory_special_huge_tlbfs_only();
6151     test_reserve_memory_special_huge_tlbfs_mixed();
6152   }
6153 
6154   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6155     if (!UseSHM) {
6156       return;
6157     }
6158 
6159     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6160 
6161     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6162 
6163     if (addr != NULL) {
6164       assert(is_ptr_aligned(addr, alignment), "Check");
6165       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6166 
6167       small_page_write(addr, size);
6168 
6169       os::Linux::release_memory_special_shm(addr, size);
6170     }
6171   }
6172 
6173   static void test_reserve_memory_special_shm() {
6174     size_t lp = os::large_page_size();
6175     size_t ag = os::vm_allocation_granularity();
6176 
6177     for (size_t size = ag; size < lp * 3; size += ag) {
6178       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6179         test_reserve_memory_special_shm(size, alignment);
6180       }
6181     }
6182   }
6183 
6184   static void test() {
6185     test_reserve_memory_special_huge_tlbfs();
6186     test_reserve_memory_special_shm();
6187   }
6188 };
6189 
6190 void TestReserveMemorySpecial_test() {
6191   TestReserveMemorySpecial::test();
6192 }
6193 
6194 #endif