1 /*
   2  * Copyright (c) 1999, 2014, 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/vmError.hpp"
  72 
  73 // put OS-includes here
  74 # include <sys/types.h>
  75 # include <sys/mman.h>
  76 # include <sys/stat.h>
  77 # include <sys/select.h>
  78 # include <pthread.h>
  79 # include <signal.h>
  80 # include <errno.h>
  81 # include <dlfcn.h>
  82 # include <stdio.h>
  83 # include <unistd.h>
  84 # include <sys/resource.h>
  85 # include <pthread.h>
  86 # include <sys/stat.h>
  87 # include <sys/time.h>
  88 # include <sys/times.h>
  89 # include <sys/utsname.h>
  90 # include <sys/socket.h>
  91 # include <sys/wait.h>
  92 # include <pwd.h>
  93 # include <poll.h>
  94 # include <semaphore.h>
  95 # include <fcntl.h>
  96 # include <string.h>
  97 # include <syscall.h>
  98 # include <sys/sysinfo.h>
  99 # include <gnu/libc-version.h>
 100 # include <sys/ipc.h>
 101 # include <sys/shm.h>
 102 # include <link.h>
 103 # include <stdint.h>
 104 # include <inttypes.h>
 105 # include <sys/ioctl.h>
 106 
 107 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
 108 
 109 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 110 // getrusage() is prepared to handle the associated failure.
 111 #ifndef RUSAGE_THREAD
 112   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 113 #endif
 114 
 115 #define MAX_PATH    (2 * K)
 116 
 117 #define MAX_SECS 100000000
 118 
 119 // for timer info max values which include all bits
 120 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 121 
 122 #define LARGEPAGES_BIT (1 << 6)
 123 ////////////////////////////////////////////////////////////////////////////////
 124 // global variables
 125 julong os::Linux::_physical_memory = 0;
 126 
 127 address   os::Linux::_initial_thread_stack_bottom = NULL;
 128 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 129 
 130 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 131 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 132 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 133 Mutex* os::Linux::_createThread_lock = NULL;
 134 pthread_t os::Linux::_main_thread;
 135 int os::Linux::_page_size = -1;
 136 const int os::Linux::_vm_default_page_size = (8 * K);
 137 bool os::Linux::_is_floating_stack = false;
 138 bool os::Linux::_is_NPTL = false;
 139 bool os::Linux::_supports_fast_thread_cpu_time = false;
 140 const char * os::Linux::_glibc_version = NULL;
 141 const char * os::Linux::_libpthread_version = NULL;
 142 pthread_condattr_t os::Linux::_condattr[1];
 143 
 144 static jlong initial_time_count=0;
 145 
 146 static int clock_tics_per_sec = 100;
 147 
 148 // For diagnostics to print a message once. see run_periodic_checks
 149 static sigset_t check_signal_done;
 150 static bool check_signals = true;
 151 
 152 static pid_t _initial_pid = 0;
 153 
 154 // Signal number used to suspend/resume a thread
 155 
 156 // do not use any signal number less than SIGSEGV, see 4355769
 157 static int SR_signum = SIGUSR2;
 158 sigset_t SR_sigset;
 159 
 160 // Used to protect dlsym() calls
 161 static pthread_mutex_t dl_mutex;
 162 
 163 // Declarations
 164 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 165 
 166 // utility functions
 167 
 168 static int SR_initialize();
 169 
 170 julong os::available_memory() {
 171   return Linux::available_memory();
 172 }
 173 
 174 julong os::Linux::available_memory() {
 175   // values in struct sysinfo are "unsigned long"
 176   struct sysinfo si;
 177   sysinfo(&si);
 178 
 179   return (julong)si.freeram * si.mem_unit;
 180 }
 181 
 182 julong os::physical_memory() {
 183   return Linux::physical_memory();
 184 }
 185 
 186 ////////////////////////////////////////////////////////////////////////////////
 187 // environment support
 188 
 189 bool os::getenv(const char* name, char* buf, int len) {
 190   const char* val = ::getenv(name);
 191   if (val != NULL && strlen(val) < (size_t)len) {
 192     strcpy(buf, val);
 193     return true;
 194   }
 195   if (len > 0) buf[0] = 0;  // return a null string
 196   return false;
 197 }
 198 
 199 
 200 // Return true if user is running as root.
 201 
 202 bool os::have_special_privileges() {
 203   static bool init = false;
 204   static bool privileges = false;
 205   if (!init) {
 206     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 207     init = true;
 208   }
 209   return privileges;
 210 }
 211 
 212 
 213 #ifndef SYS_gettid
 214 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 215   #ifdef __ia64__
 216     #define SYS_gettid 1105
 217   #elif __i386__
 218     #define SYS_gettid 224
 219   #elif __amd64__
 220     #define SYS_gettid 186
 221   #elif __sparc__
 222     #define SYS_gettid 143
 223   #else
 224     #error define gettid for the arch
 225   #endif
 226 #endif
 227 
 228 // Cpu architecture string
 229 #if   defined(ZERO)
 230 static char cpu_arch[] = ZERO_LIBARCH;
 231 #elif defined(IA64)
 232 static char cpu_arch[] = "ia64";
 233 #elif defined(IA32)
 234 static char cpu_arch[] = "i386";
 235 #elif defined(AMD64)
 236 static char cpu_arch[] = "amd64";
 237 #elif defined(ARM)
 238 static char cpu_arch[] = "arm";
 239 #elif defined(PPC32)
 240 static char cpu_arch[] = "ppc";
 241 #elif defined(PPC64)
 242 static char cpu_arch[] = "ppc64";
 243 #elif defined(SPARC)
 244   #ifdef _LP64
 245 static char cpu_arch[] = "sparcv9";
 246   #else
 247 static char cpu_arch[] = "sparc";
 248   #endif
 249 #else
 250   #error Add appropriate cpu_arch setting
 251 #endif
 252 
 253 
 254 // pid_t gettid()
 255 //
 256 // Returns the kernel thread id of the currently running thread. Kernel
 257 // thread id is used to access /proc.
 258 //
 259 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 260 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 261 //
 262 pid_t os::Linux::gettid() {
 263   int rslt = syscall(SYS_gettid);
 264   if (rslt == -1) {
 265     // old kernel, no NPTL support
 266     return getpid();
 267   } else {
 268     return (pid_t)rslt;
 269   }
 270 }
 271 
 272 // Most versions of linux have a bug where the number of processors are
 273 // determined by looking at the /proc file system.  In a chroot environment,
 274 // the system call returns 1.  This causes the VM to act as if it is
 275 // a single processor and elide locking (see is_MP() call).
 276 static bool unsafe_chroot_detected = false;
 277 static const char *unstable_chroot_error = "/proc file system not found.\n"
 278                      "Java may be unstable running multithreaded in a chroot "
 279                      "environment on Linux when /proc filesystem is not mounted.";
 280 
 281 void os::Linux::initialize_system_info() {
 282   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 283   if (processor_count() == 1) {
 284     pid_t pid = os::Linux::gettid();
 285     char fname[32];
 286     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 287     FILE *fp = fopen(fname, "r");
 288     if (fp == NULL) {
 289       unsafe_chroot_detected = true;
 290     } else {
 291       fclose(fp);
 292     }
 293   }
 294   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 295   assert(processor_count() > 0, "linux error");
 296 }
 297 
 298 void os::init_system_properties_values() {
 299   // The next steps are taken in the product version:
 300   //
 301   // Obtain the JAVA_HOME value from the location of libjvm.so.
 302   // This library should be located at:
 303   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 304   //
 305   // If "/jre/lib/" appears at the right place in the path, then we
 306   // assume libjvm.so is installed in a JDK and we use this path.
 307   //
 308   // Otherwise exit with message: "Could not create the Java virtual machine."
 309   //
 310   // The following extra steps are taken in the debugging version:
 311   //
 312   // If "/jre/lib/" does NOT appear at the right place in the path
 313   // instead of exit check for $JAVA_HOME environment variable.
 314   //
 315   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 316   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 317   // it looks like libjvm.so is installed there
 318   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 319   //
 320   // Otherwise exit.
 321   //
 322   // Important note: if the location of libjvm.so changes this
 323   // code needs to be changed accordingly.
 324 
 325   // See ld(1):
 326   //      The linker uses the following search paths to locate required
 327   //      shared libraries:
 328   //        1: ...
 329   //        ...
 330   //        7: The default directories, normally /lib and /usr/lib.
 331 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 332   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 333 #else
 334   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 335 #endif
 336 
 337 // Base path of extensions installed on the system.
 338 #define SYS_EXT_DIR     "/usr/java/packages"
 339 #define EXTENSIONS_DIR  "/lib/ext"

 340 
 341   // Buffer that fits several sprintfs.
 342   // Note that the space for the colon and the trailing null are provided
 343   // by the nulls included by the sizeof operator.
 344   const size_t bufsize =
 345     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 346          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir

 347   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 348 
 349   // sysclasspath, java_home, dll_dir
 350   {
 351     char *pslash;
 352     os::jvm_path(buf, bufsize);
 353 
 354     // Found the full path to libjvm.so.
 355     // Now cut the path to <java_home>/jre if we can.
 356     pslash = strrchr(buf, '/');
 357     if (pslash != NULL) {
 358       *pslash = '\0';            // Get rid of /libjvm.so.
 359     }
 360     pslash = strrchr(buf, '/');
 361     if (pslash != NULL) {
 362       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 363     }
 364     Arguments::set_dll_dir(buf);
 365 
 366     if (pslash != NULL) {
 367       pslash = strrchr(buf, '/');
 368       if (pslash != NULL) {
 369         *pslash = '\0';          // Get rid of /<arch>.
 370         pslash = strrchr(buf, '/');
 371         if (pslash != NULL) {
 372           *pslash = '\0';        // Get rid of /lib.
 373         }
 374       }
 375     }
 376     Arguments::set_java_home(buf);
 377     set_boot_path('/', ':');
 378   }
 379 
 380   // Where to look for native libraries.
 381   //
 382   // Note: Due to a legacy implementation, most of the library path
 383   // is set in the launcher. This was to accomodate linking restrictions
 384   // on legacy Linux implementations (which are no longer supported).
 385   // Eventually, all the library path setting will be done here.
 386   //
 387   // However, to prevent the proliferation of improperly built native
 388   // libraries, the new path component /usr/java/packages is added here.
 389   // Eventually, all the library path setting will be done here.
 390   {
 391     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 392     // should always exist (until the legacy problem cited above is
 393     // addressed).
 394     const char *v = ::getenv("LD_LIBRARY_PATH");
 395     const char *v_colon = ":";
 396     if (v == NULL) { v = ""; v_colon = ""; }
 397     // That's +1 for the colon and +1 for the trailing '\0'.
 398     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 399                                                      strlen(v) + 1 +
 400                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 401                                                      mtInternal);
 402     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 403     Arguments::set_library_path(ld_library_path);
 404     FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
 405   }
 406 
 407   // Extensions directories.
 408   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 409   Arguments::set_ext_dirs(buf);
 410 




 411   FREE_C_HEAP_ARRAY(char, buf, mtInternal);
 412 
 413 #undef DEFAULT_LIBPATH
 414 #undef SYS_EXT_DIR
 415 #undef EXTENSIONS_DIR

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