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 #define ENDORSED_DIR    "/lib/endorsed"
 341 
 342   // Buffer that fits several sprintfs.
 343   // Note that the space for the colon and the trailing null are provided
 344   // by the nulls included by the sizeof operator.
 345   const size_t bufsize =
 346     MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
 347          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
 348          (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
 349   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 350 
 351   // sysclasspath, java_home, dll_dir
 352   {
 353     char *pslash;
 354     os::jvm_path(buf, bufsize);
 355 
 356     // Found the full path to libjvm.so.
 357     // Now cut the path to <java_home>/jre if we can.
 358     pslash = strrchr(buf, '/');
 359     if (pslash != NULL) {
 360       *pslash = '\0';            // Get rid of /libjvm.so.
 361     }
 362     pslash = strrchr(buf, '/');
 363     if (pslash != NULL) {
 364       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 365     }
 366     Arguments::set_dll_dir(buf);
 367 
 368     if (pslash != NULL) {
 369       pslash = strrchr(buf, '/');
 370       if (pslash != NULL) {
 371         *pslash = '\0';          // Get rid of /<arch>.
 372         pslash = strrchr(buf, '/');
 373         if (pslash != NULL) {
 374           *pslash = '\0';        // Get rid of /lib.
 375         }
 376       }
 377     }
 378     Arguments::set_java_home(buf);
 379     set_boot_path('/', ':');
 380   }
 381 
 382   // Where to look for native libraries.
 383   //
 384   // Note: Due to a legacy implementation, most of the library path
 385   // is set in the launcher. This was to accomodate linking restrictions
 386   // on legacy Linux implementations (which are no longer supported).
 387   // Eventually, all the library path setting will be done here.
 388   //
 389   // However, to prevent the proliferation of improperly built native
 390   // libraries, the new path component /usr/java/packages is added here.
 391   // Eventually, all the library path setting will be done here.
 392   {
 393     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 394     // should always exist (until the legacy problem cited above is
 395     // addressed).
 396     const char *v = ::getenv("LD_LIBRARY_PATH");
 397     const char *v_colon = ":";
 398     if (v == NULL) { v = ""; v_colon = ""; }
 399     // That's +1 for the colon and +1 for the trailing '\0'.
 400     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 401                                                      strlen(v) + 1 +
 402                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 403                                                      mtInternal);
 404     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 405     Arguments::set_library_path(ld_library_path);
 406     FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
 407   }
 408 
 409   // Extensions directories.
 410   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 411   Arguments::set_ext_dirs(buf);
 412 
 413   // Endorsed standards default directory.
 414   sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 415   Arguments::set_endorsed_dirs(buf);
 416 
 417   FREE_C_HEAP_ARRAY(char, buf, mtInternal);
 418 
 419 #undef DEFAULT_LIBPATH
 420 #undef SYS_EXT_DIR
 421 #undef EXTENSIONS_DIR
 422 #undef ENDORSED_DIR
 423 }
 424 
 425 ////////////////////////////////////////////////////////////////////////////////
 426 // breakpoint support
 427 
 428 void os::breakpoint() {
 429   BREAKPOINT;
 430 }
 431 
 432 extern "C" void breakpoint() {
 433   // use debugger to set breakpoint here
 434 }
 435 
 436 ////////////////////////////////////////////////////////////////////////////////
 437 // signal support
 438 
 439 debug_only(static bool signal_sets_initialized = false);
 440 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 441 
 442 bool os::Linux::is_sig_ignored(int sig) {
 443   struct sigaction oact;
 444   sigaction(sig, (struct sigaction*)NULL, &oact);
 445   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 446                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 447   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 448     return true;
 449   } else {
 450     return false;
 451   }
 452 }
 453 
 454 void os::Linux::signal_sets_init() {
 455   // Should also have an assertion stating we are still single-threaded.
 456   assert(!signal_sets_initialized, "Already initialized");
 457   // Fill in signals that are necessarily unblocked for all threads in
 458   // the VM. Currently, we unblock the following signals:
 459   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 460   //                         by -Xrs (=ReduceSignalUsage));
 461   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 462   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 463   // the dispositions or masks wrt these signals.
 464   // Programs embedding the VM that want to use the above signals for their
 465   // own purposes must, at this time, use the "-Xrs" option to prevent
 466   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 467   // (See bug 4345157, and other related bugs).
 468   // In reality, though, unblocking these signals is really a nop, since
 469   // these signals are not blocked by default.
 470   sigemptyset(&unblocked_sigs);
 471   sigemptyset(&allowdebug_blocked_sigs);
 472   sigaddset(&unblocked_sigs, SIGILL);
 473   sigaddset(&unblocked_sigs, SIGSEGV);
 474   sigaddset(&unblocked_sigs, SIGBUS);
 475   sigaddset(&unblocked_sigs, SIGFPE);
 476 #if defined(PPC64)
 477   sigaddset(&unblocked_sigs, SIGTRAP);
 478 #endif
 479   sigaddset(&unblocked_sigs, SR_signum);
 480 
 481   if (!ReduceSignalUsage) {
 482     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 483       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 484       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 485     }
 486     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 487       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 488       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 489     }
 490     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 491       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 492       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 493     }
 494   }
 495   // Fill in signals that are blocked by all but the VM thread.
 496   sigemptyset(&vm_sigs);
 497   if (!ReduceSignalUsage) {
 498     sigaddset(&vm_sigs, BREAK_SIGNAL);
 499   }
 500   debug_only(signal_sets_initialized = true);
 501 
 502 }
 503 
 504 // These are signals that are unblocked while a thread is running Java.
 505 // (For some reason, they get blocked by default.)
 506 sigset_t* os::Linux::unblocked_signals() {
 507   assert(signal_sets_initialized, "Not initialized");
 508   return &unblocked_sigs;
 509 }
 510 
 511 // These are the signals that are blocked while a (non-VM) thread is
 512 // running Java. Only the VM thread handles these signals.
 513 sigset_t* os::Linux::vm_signals() {
 514   assert(signal_sets_initialized, "Not initialized");
 515   return &vm_sigs;
 516 }
 517 
 518 // These are signals that are blocked during cond_wait to allow debugger in
 519 sigset_t* os::Linux::allowdebug_blocked_signals() {
 520   assert(signal_sets_initialized, "Not initialized");
 521   return &allowdebug_blocked_sigs;
 522 }
 523 
 524 void os::Linux::hotspot_sigmask(Thread* thread) {
 525 
 526   //Save caller's signal mask before setting VM signal mask
 527   sigset_t caller_sigmask;
 528   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 529 
 530   OSThread* osthread = thread->osthread();
 531   osthread->set_caller_sigmask(caller_sigmask);
 532 
 533   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 534 
 535   if (!ReduceSignalUsage) {
 536     if (thread->is_VM_thread()) {
 537       // Only the VM thread handles BREAK_SIGNAL ...
 538       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 539     } else {
 540       // ... all other threads block BREAK_SIGNAL
 541       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 542     }
 543   }
 544 }
 545 
 546 //////////////////////////////////////////////////////////////////////////////
 547 // detecting pthread library
 548 
 549 void os::Linux::libpthread_init() {
 550   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 551   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 552   // generic name for earlier versions.
 553   // Define macros here so we can build HotSpot on old systems.
 554 #ifndef _CS_GNU_LIBC_VERSION
 555   #define _CS_GNU_LIBC_VERSION 2
 556 #endif
 557 #ifndef _CS_GNU_LIBPTHREAD_VERSION
 558   #define _CS_GNU_LIBPTHREAD_VERSION 3
 559 #endif
 560 
 561   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 562   if (n > 0) {
 563     char *str = (char *)malloc(n, mtInternal);
 564     confstr(_CS_GNU_LIBC_VERSION, str, n);
 565     os::Linux::set_glibc_version(str);
 566   } else {
 567     // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 568     static char _gnu_libc_version[32];
 569     jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 570                  "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 571     os::Linux::set_glibc_version(_gnu_libc_version);
 572   }
 573 
 574   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 575   if (n > 0) {
 576     char *str = (char *)malloc(n, mtInternal);
 577     confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 578     // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 579     // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 580     // is the case. LinuxThreads has a hard limit on max number of threads.
 581     // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 582     // On the other hand, NPTL does not have such a limit, sysconf()
 583     // will return -1 and errno is not changed. Check if it is really NPTL.
 584     if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 585         strstr(str, "NPTL") &&
 586         sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 587       free(str);
 588       os::Linux::set_libpthread_version("linuxthreads");
 589     } else {
 590       os::Linux::set_libpthread_version(str);
 591     }
 592   } else {
 593     // glibc before 2.3.2 only has LinuxThreads.
 594     os::Linux::set_libpthread_version("linuxthreads");
 595   }
 596 
 597   if (strstr(libpthread_version(), "NPTL")) {
 598     os::Linux::set_is_NPTL();
 599   } else {
 600     os::Linux::set_is_LinuxThreads();
 601   }
 602 
 603   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 604   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 605   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 606     os::Linux::set_is_floating_stack();
 607   }
 608 }
 609 
 610 /////////////////////////////////////////////////////////////////////////////
 611 // thread stack
 612 
 613 // Force Linux kernel to expand current thread stack. If "bottom" is close
 614 // to the stack guard, caller should block all signals.
 615 //
 616 // MAP_GROWSDOWN:
 617 //   A special mmap() flag that is used to implement thread stacks. It tells
 618 //   kernel that the memory region should extend downwards when needed. This
 619 //   allows early versions of LinuxThreads to only mmap the first few pages
 620 //   when creating a new thread. Linux kernel will automatically expand thread
 621 //   stack as needed (on page faults).
 622 //
 623 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 624 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 625 //   region, it's hard to tell if the fault is due to a legitimate stack
 626 //   access or because of reading/writing non-exist memory (e.g. buffer
 627 //   overrun). As a rule, if the fault happens below current stack pointer,
 628 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 629 //   application (see Linux kernel fault.c).
 630 //
 631 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 632 //   stack overflow detection.
 633 //
 634 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 635 //   not use this flag. However, the stack of initial thread is not created
 636 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 637 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 638 //   and then attach the thread to JVM.
 639 //
 640 // To get around the problem and allow stack banging on Linux, we need to
 641 // manually expand thread stack after receiving the SIGSEGV.
 642 //
 643 // There are two ways to expand thread stack to address "bottom", we used
 644 // both of them in JVM before 1.5:
 645 //   1. adjust stack pointer first so that it is below "bottom", and then
 646 //      touch "bottom"
 647 //   2. mmap() the page in question
 648 //
 649 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 650 // if current sp is already near the lower end of page 101, and we need to
 651 // call mmap() to map page 100, it is possible that part of the mmap() frame
 652 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 653 // That will destroy the mmap() frame and cause VM to crash.
 654 //
 655 // The following code works by adjusting sp first, then accessing the "bottom"
 656 // page to force a page fault. Linux kernel will then automatically expand the
 657 // stack mapping.
 658 //
 659 // _expand_stack_to() assumes its frame size is less than page size, which
 660 // should always be true if the function is not inlined.
 661 
 662 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 663   #define NOINLINE
 664 #else
 665   #define NOINLINE __attribute__ ((noinline))
 666 #endif
 667 
 668 static void _expand_stack_to(address bottom) NOINLINE;
 669 
 670 static void _expand_stack_to(address bottom) {
 671   address sp;
 672   size_t size;
 673   volatile char *p;
 674 
 675   // Adjust bottom to point to the largest address within the same page, it
 676   // gives us a one-page buffer if alloca() allocates slightly more memory.
 677   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 678   bottom += os::Linux::page_size() - 1;
 679 
 680   // sp might be slightly above current stack pointer; if that's the case, we
 681   // will alloca() a little more space than necessary, which is OK. Don't use
 682   // os::current_stack_pointer(), as its result can be slightly below current
 683   // stack pointer, causing us to not alloca enough to reach "bottom".
 684   sp = (address)&sp;
 685 
 686   if (sp > bottom) {
 687     size = sp - bottom;
 688     p = (volatile char *)alloca(size);
 689     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 690     p[0] = '\0';
 691   }
 692 }
 693 
 694 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 695   assert(t!=NULL, "just checking");
 696   assert(t->osthread()->expanding_stack(), "expand should be set");
 697   assert(t->stack_base() != NULL, "stack_base was not initialized");
 698 
 699   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 700     sigset_t mask_all, old_sigset;
 701     sigfillset(&mask_all);
 702     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 703     _expand_stack_to(addr);
 704     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 705     return true;
 706   }
 707   return false;
 708 }
 709 
 710 //////////////////////////////////////////////////////////////////////////////
 711 // create new thread
 712 
 713 static address highest_vm_reserved_address();
 714 
 715 // check if it's safe to start a new thread
 716 static bool _thread_safety_check(Thread* thread) {
 717   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 718     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 719     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 720     //   allocated (MAP_FIXED) from high address space. Every thread stack
 721     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 722     //   it to other values if they rebuild LinuxThreads).
 723     //
 724     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 725     // the memory region has already been mmap'ed. That means if we have too
 726     // many threads and/or very large heap, eventually thread stack will
 727     // collide with heap.
 728     //
 729     // Here we try to prevent heap/stack collision by comparing current
 730     // stack bottom with the highest address that has been mmap'ed by JVM
 731     // plus a safety margin for memory maps created by native code.
 732     //
 733     // This feature can be disabled by setting ThreadSafetyMargin to 0
 734     //
 735     if (ThreadSafetyMargin > 0) {
 736       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 737 
 738       // not safe if our stack extends below the safety margin
 739       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 740     } else {
 741       return true;
 742     }
 743   } else {
 744     // Floating stack LinuxThreads or NPTL:
 745     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 746     //   there's not enough space left, pthread_create() will fail. If we come
 747     //   here, that means enough space has been reserved for stack.
 748     return true;
 749   }
 750 }
 751 
 752 // Thread start routine for all newly created threads
 753 static void *java_start(Thread *thread) {
 754   // Try to randomize the cache line index of hot stack frames.
 755   // This helps when threads of the same stack traces evict each other's
 756   // cache lines. The threads can be either from the same JVM instance, or
 757   // from different JVM instances. The benefit is especially true for
 758   // processors with hyperthreading technology.
 759   static int counter = 0;
 760   int pid = os::current_process_id();
 761   alloca(((pid ^ counter++) & 7) * 128);
 762 
 763   ThreadLocalStorage::set_thread(thread);
 764 
 765   OSThread* osthread = thread->osthread();
 766   Monitor* sync = osthread->startThread_lock();
 767 
 768   // non floating stack LinuxThreads needs extra check, see above
 769   if (!_thread_safety_check(thread)) {
 770     // notify parent thread
 771     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 772     osthread->set_state(ZOMBIE);
 773     sync->notify_all();
 774     return NULL;
 775   }
 776 
 777   // thread_id is kernel thread id (similar to Solaris LWP id)
 778   osthread->set_thread_id(os::Linux::gettid());
 779 
 780   if (UseNUMA) {
 781     int lgrp_id = os::numa_get_group_id();
 782     if (lgrp_id != -1) {
 783       thread->set_lgrp_id(lgrp_id);
 784     }
 785   }
 786   // initialize signal mask for this thread
 787   os::Linux::hotspot_sigmask(thread);
 788 
 789   // initialize floating point control register
 790   os::Linux::init_thread_fpu_state();
 791 
 792   // handshaking with parent thread
 793   {
 794     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 795 
 796     // notify parent thread
 797     osthread->set_state(INITIALIZED);
 798     sync->notify_all();
 799 
 800     // wait until os::start_thread()
 801     while (osthread->get_state() == INITIALIZED) {
 802       sync->wait(Mutex::_no_safepoint_check_flag);
 803     }
 804   }
 805 
 806   // call one more level start routine
 807   thread->run();
 808 
 809   return 0;
 810 }
 811 
 812 bool os::create_thread(Thread* thread, ThreadType thr_type,
 813                        size_t stack_size) {
 814   assert(thread->osthread() == NULL, "caller responsible");
 815 
 816   // Allocate the OSThread object
 817   OSThread* osthread = new OSThread(NULL, NULL);
 818   if (osthread == NULL) {
 819     return false;
 820   }
 821 
 822   // set the correct thread state
 823   osthread->set_thread_type(thr_type);
 824 
 825   // Initial state is ALLOCATED but not INITIALIZED
 826   osthread->set_state(ALLOCATED);
 827 
 828   thread->set_osthread(osthread);
 829 
 830   // init thread attributes
 831   pthread_attr_t attr;
 832   pthread_attr_init(&attr);
 833   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 834 
 835   // stack size
 836   if (os::Linux::supports_variable_stack_size()) {
 837     // calculate stack size if it's not specified by caller
 838     if (stack_size == 0) {
 839       stack_size = os::Linux::default_stack_size(thr_type);
 840 
 841       switch (thr_type) {
 842       case os::java_thread:
 843         // Java threads use ThreadStackSize which default value can be
 844         // changed with the flag -Xss
 845         assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 846         stack_size = JavaThread::stack_size_at_create();
 847         break;
 848       case os::compiler_thread:
 849         if (CompilerThreadStackSize > 0) {
 850           stack_size = (size_t)(CompilerThreadStackSize * K);
 851           break;
 852         } // else fall through:
 853           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 854       case os::vm_thread:
 855       case os::pgc_thread:
 856       case os::cgc_thread:
 857       case os::watcher_thread:
 858         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 859         break;
 860       }
 861     }
 862 
 863     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 864     pthread_attr_setstacksize(&attr, stack_size);
 865   } else {
 866     // let pthread_create() pick the default value.
 867   }
 868 
 869   // glibc guard page
 870   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 871 
 872   ThreadState state;
 873 
 874   {
 875     // Serialize thread creation if we are running with fixed stack LinuxThreads
 876     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 877     if (lock) {
 878       os::Linux::createThread_lock()->lock_without_safepoint_check();
 879     }
 880 
 881     pthread_t tid;
 882     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 883 
 884     pthread_attr_destroy(&attr);
 885 
 886     if (ret != 0) {
 887       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 888         perror("pthread_create()");
 889       }
 890       // Need to clean up stuff we've allocated so far
 891       thread->set_osthread(NULL);
 892       delete osthread;
 893       if (lock) os::Linux::createThread_lock()->unlock();
 894       return false;
 895     }
 896 
 897     // Store pthread info into the OSThread
 898     osthread->set_pthread_id(tid);
 899 
 900     // Wait until child thread is either initialized or aborted
 901     {
 902       Monitor* sync_with_child = osthread->startThread_lock();
 903       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 904       while ((state = osthread->get_state()) == ALLOCATED) {
 905         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 906       }
 907     }
 908 
 909     if (lock) {
 910       os::Linux::createThread_lock()->unlock();
 911     }
 912   }
 913 
 914   // Aborted due to thread limit being reached
 915   if (state == ZOMBIE) {
 916     thread->set_osthread(NULL);
 917     delete osthread;
 918     return false;
 919   }
 920 
 921   // The thread is returned suspended (in state INITIALIZED),
 922   // and is started higher up in the call chain
 923   assert(state == INITIALIZED, "race condition");
 924   return true;
 925 }
 926 
 927 /////////////////////////////////////////////////////////////////////////////
 928 // attach existing thread
 929 
 930 // bootstrap the main thread
 931 bool os::create_main_thread(JavaThread* thread) {
 932   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 933   return create_attached_thread(thread);
 934 }
 935 
 936 bool os::create_attached_thread(JavaThread* thread) {
 937 #ifdef ASSERT
 938   thread->verify_not_published();
 939 #endif
 940 
 941   // Allocate the OSThread object
 942   OSThread* osthread = new OSThread(NULL, NULL);
 943 
 944   if (osthread == NULL) {
 945     return false;
 946   }
 947 
 948   // Store pthread info into the OSThread
 949   osthread->set_thread_id(os::Linux::gettid());
 950   osthread->set_pthread_id(::pthread_self());
 951 
 952   // initialize floating point control register
 953   os::Linux::init_thread_fpu_state();
 954 
 955   // Initial thread state is RUNNABLE
 956   osthread->set_state(RUNNABLE);
 957 
 958   thread->set_osthread(osthread);
 959 
 960   if (UseNUMA) {
 961     int lgrp_id = os::numa_get_group_id();
 962     if (lgrp_id != -1) {
 963       thread->set_lgrp_id(lgrp_id);
 964     }
 965   }
 966 
 967   if (os::Linux::is_initial_thread()) {
 968     // If current thread is initial thread, its stack is mapped on demand,
 969     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 970     // the entire stack region to avoid SEGV in stack banging.
 971     // It is also useful to get around the heap-stack-gap problem on SuSE
 972     // kernel (see 4821821 for details). We first expand stack to the top
 973     // of yellow zone, then enable stack yellow zone (order is significant,
 974     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 975     // is no gap between the last two virtual memory regions.
 976 
 977     JavaThread *jt = (JavaThread *)thread;
 978     address addr = jt->stack_yellow_zone_base();
 979     assert(addr != NULL, "initialization problem?");
 980     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 981 
 982     osthread->set_expanding_stack();
 983     os::Linux::manually_expand_stack(jt, addr);
 984     osthread->clear_expanding_stack();
 985   }
 986 
 987   // initialize signal mask for this thread
 988   // and save the caller's signal mask
 989   os::Linux::hotspot_sigmask(thread);
 990 
 991   return true;
 992 }
 993 
 994 void os::pd_start_thread(Thread* thread) {
 995   OSThread * osthread = thread->osthread();
 996   assert(osthread->get_state() != INITIALIZED, "just checking");
 997   Monitor* sync_with_child = osthread->startThread_lock();
 998   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 999   sync_with_child->notify();
1000 }
1001 
1002 // Free Linux resources related to the OSThread
1003 void os::free_thread(OSThread* osthread) {
1004   assert(osthread != NULL, "osthread not set");
1005 
1006   if (Thread::current()->osthread() == osthread) {
1007     // Restore caller's signal mask
1008     sigset_t sigmask = osthread->caller_sigmask();
1009     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1010   }
1011 
1012   delete osthread;
1013 }
1014 
1015 //////////////////////////////////////////////////////////////////////////////
1016 // thread local storage
1017 
1018 // Restore the thread pointer if the destructor is called. This is in case
1019 // someone from JNI code sets up a destructor with pthread_key_create to run
1020 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1021 // will hang or crash. When detachCurrentThread is called the key will be set
1022 // to null and we will not be called again. If detachCurrentThread is never
1023 // called we could loop forever depending on the pthread implementation.
1024 static void restore_thread_pointer(void* p) {
1025   Thread* thread = (Thread*) p;
1026   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1027 }
1028 
1029 int os::allocate_thread_local_storage() {
1030   pthread_key_t key;
1031   int rslt = pthread_key_create(&key, restore_thread_pointer);
1032   assert(rslt == 0, "cannot allocate thread local storage");
1033   return (int)key;
1034 }
1035 
1036 // Note: This is currently not used by VM, as we don't destroy TLS key
1037 // on VM exit.
1038 void os::free_thread_local_storage(int index) {
1039   int rslt = pthread_key_delete((pthread_key_t)index);
1040   assert(rslt == 0, "invalid index");
1041 }
1042 
1043 void os::thread_local_storage_at_put(int index, void* value) {
1044   int rslt = pthread_setspecific((pthread_key_t)index, value);
1045   assert(rslt == 0, "pthread_setspecific failed");
1046 }
1047 
1048 extern "C" Thread* get_thread() {
1049   return ThreadLocalStorage::thread();
1050 }
1051 
1052 //////////////////////////////////////////////////////////////////////////////
1053 // initial thread
1054 
1055 // Check if current thread is the initial thread, similar to Solaris thr_main.
1056 bool os::Linux::is_initial_thread(void) {
1057   char dummy;
1058   // If called before init complete, thread stack bottom will be null.
1059   // Can be called if fatal error occurs before initialization.
1060   if (initial_thread_stack_bottom() == NULL) return false;
1061   assert(initial_thread_stack_bottom() != NULL &&
1062          initial_thread_stack_size()   != 0,
1063          "os::init did not locate initial thread's stack region");
1064   if ((address)&dummy >= initial_thread_stack_bottom() &&
1065       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1066     return true;
1067   } else {
1068     return false;
1069   }
1070 }
1071 
1072 // Find the virtual memory area that contains addr
1073 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1074   FILE *fp = fopen("/proc/self/maps", "r");
1075   if (fp) {
1076     address low, high;
1077     while (!feof(fp)) {
1078       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1079         if (low <= addr && addr < high) {
1080           if (vma_low)  *vma_low  = low;
1081           if (vma_high) *vma_high = high;
1082           fclose(fp);
1083           return true;
1084         }
1085       }
1086       for (;;) {
1087         int ch = fgetc(fp);
1088         if (ch == EOF || ch == (int)'\n') break;
1089       }
1090     }
1091     fclose(fp);
1092   }
1093   return false;
1094 }
1095 
1096 // Locate initial thread stack. This special handling of initial thread stack
1097 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1098 // bogus value for initial thread.
1099 void os::Linux::capture_initial_stack(size_t max_size) {
1100   // stack size is the easy part, get it from RLIMIT_STACK
1101   size_t stack_size;
1102   struct rlimit rlim;
1103   getrlimit(RLIMIT_STACK, &rlim);
1104   stack_size = rlim.rlim_cur;
1105 
1106   // 6308388: a bug in ld.so will relocate its own .data section to the
1107   //   lower end of primordial stack; reduce ulimit -s value a little bit
1108   //   so we won't install guard page on ld.so's data section.
1109   stack_size -= 2 * page_size();
1110 
1111   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1112   //   7.1, in both cases we will get 2G in return value.
1113   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1114   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1115   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1116   //   in case other parts in glibc still assumes 2M max stack size.
1117   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1118   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1119   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1120     stack_size = 2 * K * K IA64_ONLY(*2);
1121   }
1122   // Try to figure out where the stack base (top) is. This is harder.
1123   //
1124   // When an application is started, glibc saves the initial stack pointer in
1125   // a global variable "__libc_stack_end", which is then used by system
1126   // libraries. __libc_stack_end should be pretty close to stack top. The
1127   // variable is available since the very early days. However, because it is
1128   // a private interface, it could disappear in the future.
1129   //
1130   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1131   // to __libc_stack_end, it is very close to stack top, but isn't the real
1132   // stack top. Note that /proc may not exist if VM is running as a chroot
1133   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1134   // /proc/<pid>/stat could change in the future (though unlikely).
1135   //
1136   // We try __libc_stack_end first. If that doesn't work, look for
1137   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1138   // as a hint, which should work well in most cases.
1139 
1140   uintptr_t stack_start;
1141 
1142   // try __libc_stack_end first
1143   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1144   if (p && *p) {
1145     stack_start = *p;
1146   } else {
1147     // see if we can get the start_stack field from /proc/self/stat
1148     FILE *fp;
1149     int pid;
1150     char state;
1151     int ppid;
1152     int pgrp;
1153     int session;
1154     int nr;
1155     int tpgrp;
1156     unsigned long flags;
1157     unsigned long minflt;
1158     unsigned long cminflt;
1159     unsigned long majflt;
1160     unsigned long cmajflt;
1161     unsigned long utime;
1162     unsigned long stime;
1163     long cutime;
1164     long cstime;
1165     long prio;
1166     long nice;
1167     long junk;
1168     long it_real;
1169     uintptr_t start;
1170     uintptr_t vsize;
1171     intptr_t rss;
1172     uintptr_t rsslim;
1173     uintptr_t scodes;
1174     uintptr_t ecode;
1175     int i;
1176 
1177     // Figure what the primordial thread stack base is. Code is inspired
1178     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1179     // followed by command name surrounded by parentheses, state, etc.
1180     char stat[2048];
1181     int statlen;
1182 
1183     fp = fopen("/proc/self/stat", "r");
1184     if (fp) {
1185       statlen = fread(stat, 1, 2047, fp);
1186       stat[statlen] = '\0';
1187       fclose(fp);
1188 
1189       // Skip pid and the command string. Note that we could be dealing with
1190       // weird command names, e.g. user could decide to rename java launcher
1191       // to "java 1.4.2 :)", then the stat file would look like
1192       //                1234 (java 1.4.2 :)) R ... ...
1193       // We don't really need to know the command string, just find the last
1194       // occurrence of ")" and then start parsing from there. See bug 4726580.
1195       char * s = strrchr(stat, ')');
1196 
1197       i = 0;
1198       if (s) {
1199         // Skip blank chars
1200         do { s++; } while (s && isspace(*s));
1201 
1202 #define _UFM UINTX_FORMAT
1203 #define _DFM INTX_FORMAT
1204 
1205         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1206         //              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
1207         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,
1208                    &state,          // 3  %c
1209                    &ppid,           // 4  %d
1210                    &pgrp,           // 5  %d
1211                    &session,        // 6  %d
1212                    &nr,             // 7  %d
1213                    &tpgrp,          // 8  %d
1214                    &flags,          // 9  %lu
1215                    &minflt,         // 10 %lu
1216                    &cminflt,        // 11 %lu
1217                    &majflt,         // 12 %lu
1218                    &cmajflt,        // 13 %lu
1219                    &utime,          // 14 %lu
1220                    &stime,          // 15 %lu
1221                    &cutime,         // 16 %ld
1222                    &cstime,         // 17 %ld
1223                    &prio,           // 18 %ld
1224                    &nice,           // 19 %ld
1225                    &junk,           // 20 %ld
1226                    &it_real,        // 21 %ld
1227                    &start,          // 22 UINTX_FORMAT
1228                    &vsize,          // 23 UINTX_FORMAT
1229                    &rss,            // 24 INTX_FORMAT
1230                    &rsslim,         // 25 UINTX_FORMAT
1231                    &scodes,         // 26 UINTX_FORMAT
1232                    &ecode,          // 27 UINTX_FORMAT
1233                    &stack_start);   // 28 UINTX_FORMAT
1234       }
1235 
1236 #undef _UFM
1237 #undef _DFM
1238 
1239       if (i != 28 - 2) {
1240         assert(false, "Bad conversion from /proc/self/stat");
1241         // product mode - assume we are the initial thread, good luck in the
1242         // embedded case.
1243         warning("Can't detect initial thread stack location - bad conversion");
1244         stack_start = (uintptr_t) &rlim;
1245       }
1246     } else {
1247       // For some reason we can't open /proc/self/stat (for example, running on
1248       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1249       // most cases, so don't abort:
1250       warning("Can't detect initial thread stack location - no /proc/self/stat");
1251       stack_start = (uintptr_t) &rlim;
1252     }
1253   }
1254 
1255   // Now we have a pointer (stack_start) very close to the stack top, the
1256   // next thing to do is to figure out the exact location of stack top. We
1257   // can find out the virtual memory area that contains stack_start by
1258   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1259   // and its upper limit is the real stack top. (again, this would fail if
1260   // running inside chroot, because /proc may not exist.)
1261 
1262   uintptr_t stack_top;
1263   address low, high;
1264   if (find_vma((address)stack_start, &low, &high)) {
1265     // success, "high" is the true stack top. (ignore "low", because initial
1266     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1267     stack_top = (uintptr_t)high;
1268   } else {
1269     // failed, likely because /proc/self/maps does not exist
1270     warning("Can't detect initial thread stack location - find_vma failed");
1271     // best effort: stack_start is normally within a few pages below the real
1272     // stack top, use it as stack top, and reduce stack size so we won't put
1273     // guard page outside stack.
1274     stack_top = stack_start;
1275     stack_size -= 16 * page_size();
1276   }
1277 
1278   // stack_top could be partially down the page so align it
1279   stack_top = align_size_up(stack_top, page_size());
1280 
1281   if (max_size && stack_size > max_size) {
1282     _initial_thread_stack_size = max_size;
1283   } else {
1284     _initial_thread_stack_size = stack_size;
1285   }
1286 
1287   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1288   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1289 }
1290 
1291 ////////////////////////////////////////////////////////////////////////////////
1292 // time support
1293 
1294 // Time since start-up in seconds to a fine granularity.
1295 // Used by VMSelfDestructTimer and the MemProfiler.
1296 double os::elapsedTime() {
1297 
1298   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1299 }
1300 
1301 jlong os::elapsed_counter() {
1302   return javaTimeNanos() - initial_time_count;
1303 }
1304 
1305 jlong os::elapsed_frequency() {
1306   return NANOSECS_PER_SEC; // nanosecond resolution
1307 }
1308 
1309 bool os::supports_vtime() { return true; }
1310 bool os::enable_vtime()   { return false; }
1311 bool os::vtime_enabled()  { return false; }
1312 
1313 double os::elapsedVTime() {
1314   struct rusage usage;
1315   int retval = getrusage(RUSAGE_THREAD, &usage);
1316   if (retval == 0) {
1317     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);
1318   } else {
1319     // better than nothing, but not much
1320     return elapsedTime();
1321   }
1322 }
1323 
1324 jlong os::javaTimeMillis() {
1325   timeval time;
1326   int status = gettimeofday(&time, NULL);
1327   assert(status != -1, "linux error");
1328   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1329 }
1330 
1331 #ifndef CLOCK_MONOTONIC
1332   #define CLOCK_MONOTONIC (1)
1333 #endif
1334 
1335 void os::Linux::clock_init() {
1336   // we do dlopen's in this particular order due to bug in linux
1337   // dynamical loader (see 6348968) leading to crash on exit
1338   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1339   if (handle == NULL) {
1340     handle = dlopen("librt.so", RTLD_LAZY);
1341   }
1342 
1343   if (handle) {
1344     int (*clock_getres_func)(clockid_t, struct timespec*) =
1345            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1346     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1347            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1348     if (clock_getres_func && clock_gettime_func) {
1349       // See if monotonic clock is supported by the kernel. Note that some
1350       // early implementations simply return kernel jiffies (updated every
1351       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1352       // for nano time (though the monotonic property is still nice to have).
1353       // It's fixed in newer kernels, however clock_getres() still returns
1354       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1355       // resolution for now. Hopefully as people move to new kernels, this
1356       // won't be a problem.
1357       struct timespec res;
1358       struct timespec tp;
1359       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1360           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1361         // yes, monotonic clock is supported
1362         _clock_gettime = clock_gettime_func;
1363         return;
1364       } else {
1365         // close librt if there is no monotonic clock
1366         dlclose(handle);
1367       }
1368     }
1369   }
1370   warning("No monotonic clock was available - timed services may " \
1371           "be adversely affected if the time-of-day clock changes");
1372 }
1373 
1374 #ifndef SYS_clock_getres
1375   #if defined(IA32) || defined(AMD64)
1376     #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1377     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1378   #else
1379     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1380     #define sys_clock_getres(x,y)  -1
1381   #endif
1382 #else
1383   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1384 #endif
1385 
1386 void os::Linux::fast_thread_clock_init() {
1387   if (!UseLinuxPosixThreadCPUClocks) {
1388     return;
1389   }
1390   clockid_t clockid;
1391   struct timespec tp;
1392   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1393       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1394 
1395   // Switch to using fast clocks for thread cpu time if
1396   // the sys_clock_getres() returns 0 error code.
1397   // Note, that some kernels may support the current thread
1398   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1399   // returned by the pthread_getcpuclockid().
1400   // If the fast Posix clocks are supported then the sys_clock_getres()
1401   // must return at least tp.tv_sec == 0 which means a resolution
1402   // better than 1 sec. This is extra check for reliability.
1403 
1404   if (pthread_getcpuclockid_func &&
1405       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1406       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1407     _supports_fast_thread_cpu_time = true;
1408     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1409   }
1410 }
1411 
1412 jlong os::javaTimeNanos() {
1413   if (os::supports_monotonic_clock()) {
1414     struct timespec tp;
1415     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1416     assert(status == 0, "gettime error");
1417     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1418     return result;
1419   } else {
1420     timeval time;
1421     int status = gettimeofday(&time, NULL);
1422     assert(status != -1, "linux error");
1423     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1424     return 1000 * usecs;
1425   }
1426 }
1427 
1428 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1429   if (os::supports_monotonic_clock()) {
1430     info_ptr->max_value = ALL_64_BITS;
1431 
1432     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1433     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1434     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1435   } else {
1436     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1437     info_ptr->max_value = ALL_64_BITS;
1438 
1439     // gettimeofday is a real time clock so it skips
1440     info_ptr->may_skip_backward = true;
1441     info_ptr->may_skip_forward = true;
1442   }
1443 
1444   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1445 }
1446 
1447 // Return the real, user, and system times in seconds from an
1448 // arbitrary fixed point in the past.
1449 bool os::getTimesSecs(double* process_real_time,
1450                       double* process_user_time,
1451                       double* process_system_time) {
1452   struct tms ticks;
1453   clock_t real_ticks = times(&ticks);
1454 
1455   if (real_ticks == (clock_t) (-1)) {
1456     return false;
1457   } else {
1458     double ticks_per_second = (double) clock_tics_per_sec;
1459     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1460     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1461     *process_real_time = ((double) real_ticks) / ticks_per_second;
1462 
1463     return true;
1464   }
1465 }
1466 
1467 
1468 char * os::local_time_string(char *buf, size_t buflen) {
1469   struct tm t;
1470   time_t long_time;
1471   time(&long_time);
1472   localtime_r(&long_time, &t);
1473   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1474                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1475                t.tm_hour, t.tm_min, t.tm_sec);
1476   return buf;
1477 }
1478 
1479 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1480   return localtime_r(clock, res);
1481 }
1482 
1483 ////////////////////////////////////////////////////////////////////////////////
1484 // runtime exit support
1485 
1486 // Note: os::shutdown() might be called very early during initialization, or
1487 // called from signal handler. Before adding something to os::shutdown(), make
1488 // sure it is async-safe and can handle partially initialized VM.
1489 void os::shutdown() {
1490 
1491   // allow PerfMemory to attempt cleanup of any persistent resources
1492   perfMemory_exit();
1493 
1494   // needs to remove object in file system
1495   AttachListener::abort();
1496 
1497   // flush buffered output, finish log files
1498   ostream_abort();
1499 
1500   // Check for abort hook
1501   abort_hook_t abort_hook = Arguments::abort_hook();
1502   if (abort_hook != NULL) {
1503     abort_hook();
1504   }
1505 
1506 }
1507 
1508 // Note: os::abort() might be called very early during initialization, or
1509 // called from signal handler. Before adding something to os::abort(), make
1510 // sure it is async-safe and can handle partially initialized VM.
1511 void os::abort(bool dump_core) {
1512   os::shutdown();
1513   if (dump_core) {
1514 #ifndef PRODUCT
1515     fdStream out(defaultStream::output_fd());
1516     out.print_raw("Current thread is ");
1517     char buf[16];
1518     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1519     out.print_raw_cr(buf);
1520     out.print_raw_cr("Dumping core ...");
1521 #endif
1522     ::abort(); // dump core
1523   }
1524 
1525   ::exit(1);
1526 }
1527 
1528 // Die immediately, no exit hook, no abort hook, no cleanup.
1529 void os::die() {
1530   // _exit() on LinuxThreads only kills current thread
1531   ::abort();
1532 }
1533 
1534 
1535 // This method is a copy of JDK's sysGetLastErrorString
1536 // from src/solaris/hpi/src/system_md.c
1537 
1538 size_t os::lasterror(char *buf, size_t len) {
1539   if (errno == 0)  return 0;
1540 
1541   const char *s = ::strerror(errno);
1542   size_t n = ::strlen(s);
1543   if (n >= len) {
1544     n = len - 1;
1545   }
1546   ::strncpy(buf, s, n);
1547   buf[n] = '\0';
1548   return n;
1549 }
1550 
1551 intx os::current_thread_id() { return (intx)pthread_self(); }
1552 int os::current_process_id() {
1553 
1554   // Under the old linux thread library, linux gives each thread
1555   // its own process id. Because of this each thread will return
1556   // a different pid if this method were to return the result
1557   // of getpid(2). Linux provides no api that returns the pid
1558   // of the launcher thread for the vm. This implementation
1559   // returns a unique pid, the pid of the launcher thread
1560   // that starts the vm 'process'.
1561 
1562   // Under the NPTL, getpid() returns the same pid as the
1563   // launcher thread rather than a unique pid per thread.
1564   // Use gettid() if you want the old pre NPTL behaviour.
1565 
1566   // if you are looking for the result of a call to getpid() that
1567   // returns a unique pid for the calling thread, then look at the
1568   // OSThread::thread_id() method in osThread_linux.hpp file
1569 
1570   return (int)(_initial_pid ? _initial_pid : getpid());
1571 }
1572 
1573 // DLL functions
1574 
1575 const char* os::dll_file_extension() { return ".so"; }
1576 
1577 // This must be hard coded because it's the system's temporary
1578 // directory not the java application's temp directory, ala java.io.tmpdir.
1579 const char* os::get_temp_directory() { return "/tmp"; }
1580 
1581 static bool file_exists(const char* filename) {
1582   struct stat statbuf;
1583   if (filename == NULL || strlen(filename) == 0) {
1584     return false;
1585   }
1586   return os::stat(filename, &statbuf) == 0;
1587 }
1588 
1589 bool os::dll_build_name(char* buffer, size_t buflen,
1590                         const char* pname, const char* fname) {
1591   bool retval = false;
1592   // Copied from libhpi
1593   const size_t pnamelen = pname ? strlen(pname) : 0;
1594 
1595   // Return error on buffer overflow.
1596   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1597     return retval;
1598   }
1599 
1600   if (pnamelen == 0) {
1601     snprintf(buffer, buflen, "lib%s.so", fname);
1602     retval = true;
1603   } else if (strchr(pname, *os::path_separator()) != NULL) {
1604     int n;
1605     char** pelements = split_path(pname, &n);
1606     if (pelements == NULL) {
1607       return false;
1608     }
1609     for (int i = 0; i < n; i++) {
1610       // Really shouldn't be NULL, but check can't hurt
1611       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1612         continue; // skip the empty path values
1613       }
1614       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1615       if (file_exists(buffer)) {
1616         retval = true;
1617         break;
1618       }
1619     }
1620     // release the storage
1621     for (int i = 0; i < n; i++) {
1622       if (pelements[i] != NULL) {
1623         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1624       }
1625     }
1626     if (pelements != NULL) {
1627       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1628     }
1629   } else {
1630     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1631     retval = true;
1632   }
1633   return retval;
1634 }
1635 
1636 // check if addr is inside libjvm.so
1637 bool os::address_is_in_vm(address addr) {
1638   static address libjvm_base_addr;
1639   Dl_info dlinfo;
1640 
1641   if (libjvm_base_addr == NULL) {
1642     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1643       libjvm_base_addr = (address)dlinfo.dli_fbase;
1644     }
1645     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1646   }
1647 
1648   if (dladdr((void *)addr, &dlinfo) != 0) {
1649     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1650   }
1651 
1652   return false;
1653 }
1654 
1655 bool os::dll_address_to_function_name(address addr, char *buf,
1656                                       int buflen, int *offset) {
1657   // buf is not optional, but offset is optional
1658   assert(buf != NULL, "sanity check");
1659 
1660   Dl_info dlinfo;
1661 
1662   if (dladdr((void*)addr, &dlinfo) != 0) {
1663     // see if we have a matching symbol
1664     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1665       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1666         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1667       }
1668       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1669       return true;
1670     }
1671     // no matching symbol so try for just file info
1672     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1673       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1674                           buf, buflen, offset, dlinfo.dli_fname)) {
1675         return true;
1676       }
1677     }
1678   }
1679 
1680   buf[0] = '\0';
1681   if (offset != NULL) *offset = -1;
1682   return false;
1683 }
1684 
1685 struct _address_to_library_name {
1686   address addr;          // input : memory address
1687   size_t  buflen;        //         size of fname
1688   char*   fname;         // output: library name
1689   address base;          //         library base addr
1690 };
1691 
1692 static int address_to_library_name_callback(struct dl_phdr_info *info,
1693                                             size_t size, void *data) {
1694   int i;
1695   bool found = false;
1696   address libbase = NULL;
1697   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1698 
1699   // iterate through all loadable segments
1700   for (i = 0; i < info->dlpi_phnum; i++) {
1701     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1702     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1703       // base address of a library is the lowest address of its loaded
1704       // segments.
1705       if (libbase == NULL || libbase > segbase) {
1706         libbase = segbase;
1707       }
1708       // see if 'addr' is within current segment
1709       if (segbase <= d->addr &&
1710           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1711         found = true;
1712       }
1713     }
1714   }
1715 
1716   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1717   // so dll_address_to_library_name() can fall through to use dladdr() which
1718   // can figure out executable name from argv[0].
1719   if (found && info->dlpi_name && info->dlpi_name[0]) {
1720     d->base = libbase;
1721     if (d->fname) {
1722       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1723     }
1724     return 1;
1725   }
1726   return 0;
1727 }
1728 
1729 bool os::dll_address_to_library_name(address addr, char* buf,
1730                                      int buflen, int* offset) {
1731   // buf is not optional, but offset is optional
1732   assert(buf != NULL, "sanity check");
1733 
1734   Dl_info dlinfo;
1735   struct _address_to_library_name data;
1736 
1737   // There is a bug in old glibc dladdr() implementation that it could resolve
1738   // to wrong library name if the .so file has a base address != NULL. Here
1739   // we iterate through the program headers of all loaded libraries to find
1740   // out which library 'addr' really belongs to. This workaround can be
1741   // removed once the minimum requirement for glibc is moved to 2.3.x.
1742   data.addr = addr;
1743   data.fname = buf;
1744   data.buflen = buflen;
1745   data.base = NULL;
1746   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1747 
1748   if (rslt) {
1749     // buf already contains library name
1750     if (offset) *offset = addr - data.base;
1751     return true;
1752   }
1753   if (dladdr((void*)addr, &dlinfo) != 0) {
1754     if (dlinfo.dli_fname != NULL) {
1755       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1756     }
1757     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1758       *offset = addr - (address)dlinfo.dli_fbase;
1759     }
1760     return true;
1761   }
1762 
1763   buf[0] = '\0';
1764   if (offset) *offset = -1;
1765   return false;
1766 }
1767 
1768 // Loads .dll/.so and
1769 // in case of error it checks if .dll/.so was built for the
1770 // same architecture as Hotspot is running on
1771 
1772 
1773 // Remember the stack's state. The Linux dynamic linker will change
1774 // the stack to 'executable' at most once, so we must safepoint only once.
1775 bool os::Linux::_stack_is_executable = false;
1776 
1777 // VM operation that loads a library.  This is necessary if stack protection
1778 // of the Java stacks can be lost during loading the library.  If we
1779 // do not stop the Java threads, they can stack overflow before the stacks
1780 // are protected again.
1781 class VM_LinuxDllLoad: public VM_Operation {
1782  private:
1783   const char *_filename;
1784   char *_ebuf;
1785   int _ebuflen;
1786   void *_lib;
1787  public:
1788   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1789     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1790   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1791   void doit() {
1792     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1793     os::Linux::_stack_is_executable = true;
1794   }
1795   void* loaded_library() { return _lib; }
1796 };
1797 
1798 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1799   void * result = NULL;
1800   bool load_attempted = false;
1801 
1802   // Check whether the library to load might change execution rights
1803   // of the stack. If they are changed, the protection of the stack
1804   // guard pages will be lost. We need a safepoint to fix this.
1805   //
1806   // See Linux man page execstack(8) for more info.
1807   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1808     ElfFile ef(filename);
1809     if (!ef.specifies_noexecstack()) {
1810       if (!is_init_completed()) {
1811         os::Linux::_stack_is_executable = true;
1812         // This is OK - No Java threads have been created yet, and hence no
1813         // stack guard pages to fix.
1814         //
1815         // This should happen only when you are building JDK7 using a very
1816         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1817         //
1818         // Dynamic loader will make all stacks executable after
1819         // this function returns, and will not do that again.
1820         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1821       } else {
1822         warning("You have loaded library %s which might have disabled stack guard. "
1823                 "The VM will try to fix the stack guard now.\n"
1824                 "It's highly recommended that you fix the library with "
1825                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1826                 filename);
1827 
1828         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1829         JavaThread *jt = JavaThread::current();
1830         if (jt->thread_state() != _thread_in_native) {
1831           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1832           // that requires ExecStack. Cannot enter safe point. Let's give up.
1833           warning("Unable to fix stack guard. Giving up.");
1834         } else {
1835           if (!LoadExecStackDllInVMThread) {
1836             // This is for the case where the DLL has an static
1837             // constructor function that executes JNI code. We cannot
1838             // load such DLLs in the VMThread.
1839             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1840           }
1841 
1842           ThreadInVMfromNative tiv(jt);
1843           debug_only(VMNativeEntryWrapper vew;)
1844 
1845           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1846           VMThread::execute(&op);
1847           if (LoadExecStackDllInVMThread) {
1848             result = op.loaded_library();
1849           }
1850           load_attempted = true;
1851         }
1852       }
1853     }
1854   }
1855 
1856   if (!load_attempted) {
1857     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1858   }
1859 
1860   if (result != NULL) {
1861     // Successful loading
1862     return result;
1863   }
1864 
1865   Elf32_Ehdr elf_head;
1866   int diag_msg_max_length=ebuflen-strlen(ebuf);
1867   char* diag_msg_buf=ebuf+strlen(ebuf);
1868 
1869   if (diag_msg_max_length==0) {
1870     // No more space in ebuf for additional diagnostics message
1871     return NULL;
1872   }
1873 
1874 
1875   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1876 
1877   if (file_descriptor < 0) {
1878     // Can't open library, report dlerror() message
1879     return NULL;
1880   }
1881 
1882   bool failed_to_read_elf_head=
1883     (sizeof(elf_head)!=
1884      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1885 
1886   ::close(file_descriptor);
1887   if (failed_to_read_elf_head) {
1888     // file i/o error - report dlerror() msg
1889     return NULL;
1890   }
1891 
1892   typedef struct {
1893     Elf32_Half  code;         // Actual value as defined in elf.h
1894     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1895     char        elf_class;    // 32 or 64 bit
1896     char        endianess;    // MSB or LSB
1897     char*       name;         // String representation
1898   } arch_t;
1899 
1900 #ifndef EM_486
1901   #define EM_486          6               /* Intel 80486 */
1902 #endif
1903 
1904   static const arch_t arch_array[]={
1905     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1906     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1907     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1908     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1909     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1910     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1911     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1912     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1913 #if defined(VM_LITTLE_ENDIAN)
1914     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1915 #else
1916     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1917 #endif
1918     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1919     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1920     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1921     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1922     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1923     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1924     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1925   };
1926 
1927 #if  (defined IA32)
1928   static  Elf32_Half running_arch_code=EM_386;
1929 #elif   (defined AMD64)
1930   static  Elf32_Half running_arch_code=EM_X86_64;
1931 #elif  (defined IA64)
1932   static  Elf32_Half running_arch_code=EM_IA_64;
1933 #elif  (defined __sparc) && (defined _LP64)
1934   static  Elf32_Half running_arch_code=EM_SPARCV9;
1935 #elif  (defined __sparc) && (!defined _LP64)
1936   static  Elf32_Half running_arch_code=EM_SPARC;
1937 #elif  (defined __powerpc64__)
1938   static  Elf32_Half running_arch_code=EM_PPC64;
1939 #elif  (defined __powerpc__)
1940   static  Elf32_Half running_arch_code=EM_PPC;
1941 #elif  (defined ARM)
1942   static  Elf32_Half running_arch_code=EM_ARM;
1943 #elif  (defined S390)
1944   static  Elf32_Half running_arch_code=EM_S390;
1945 #elif  (defined ALPHA)
1946   static  Elf32_Half running_arch_code=EM_ALPHA;
1947 #elif  (defined MIPSEL)
1948   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1949 #elif  (defined PARISC)
1950   static  Elf32_Half running_arch_code=EM_PARISC;
1951 #elif  (defined MIPS)
1952   static  Elf32_Half running_arch_code=EM_MIPS;
1953 #elif  (defined M68K)
1954   static  Elf32_Half running_arch_code=EM_68K;
1955 #else
1956     #error Method os::dll_load requires that one of following is defined:\
1957          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1958 #endif
1959 
1960   // Identify compatability class for VM's architecture and library's architecture
1961   // Obtain string descriptions for architectures
1962 
1963   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1964   int running_arch_index=-1;
1965 
1966   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1967     if (running_arch_code == arch_array[i].code) {
1968       running_arch_index    = i;
1969     }
1970     if (lib_arch.code == arch_array[i].code) {
1971       lib_arch.compat_class = arch_array[i].compat_class;
1972       lib_arch.name         = arch_array[i].name;
1973     }
1974   }
1975 
1976   assert(running_arch_index != -1,
1977          "Didn't find running architecture code (running_arch_code) in arch_array");
1978   if (running_arch_index == -1) {
1979     // Even though running architecture detection failed
1980     // we may still continue with reporting dlerror() message
1981     return NULL;
1982   }
1983 
1984   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1985     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1986     return NULL;
1987   }
1988 
1989 #ifndef S390
1990   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1991     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1992     return NULL;
1993   }
1994 #endif // !S390
1995 
1996   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1997     if (lib_arch.name!=NULL) {
1998       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1999                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2000                  lib_arch.name, arch_array[running_arch_index].name);
2001     } else {
2002       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2003                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2004                  lib_arch.code,
2005                  arch_array[running_arch_index].name);
2006     }
2007   }
2008 
2009   return NULL;
2010 }
2011 
2012 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
2013                                 int ebuflen) {
2014   void * result = ::dlopen(filename, RTLD_LAZY);
2015   if (result == NULL) {
2016     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2017     ebuf[ebuflen-1] = '\0';
2018   }
2019   return result;
2020 }
2021 
2022 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2023                                        int ebuflen) {
2024   void * result = NULL;
2025   if (LoadExecStackDllInVMThread) {
2026     result = dlopen_helper(filename, ebuf, ebuflen);
2027   }
2028 
2029   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2030   // library that requires an executable stack, or which does not have this
2031   // stack attribute set, dlopen changes the stack attribute to executable. The
2032   // read protection of the guard pages gets lost.
2033   //
2034   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2035   // may have been queued at the same time.
2036 
2037   if (!_stack_is_executable) {
2038     JavaThread *jt = Threads::first();
2039 
2040     while (jt) {
2041       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2042           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2043         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2044                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2045           warning("Attempt to reguard stack yellow zone failed.");
2046         }
2047       }
2048       jt = jt->next();
2049     }
2050   }
2051 
2052   return result;
2053 }
2054 
2055 // glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2056 // chances are you might want to run the generated bits against glibc-2.0
2057 // libdl.so, so always use locking for any version of glibc.
2058 //
2059 void* os::dll_lookup(void* handle, const char* name) {
2060   pthread_mutex_lock(&dl_mutex);
2061   void* res = dlsym(handle, name);
2062   pthread_mutex_unlock(&dl_mutex);
2063   return res;
2064 }
2065 
2066 void* os::get_default_process_handle() {
2067   return (void*)::dlopen(NULL, RTLD_LAZY);
2068 }
2069 
2070 static bool _print_ascii_file(const char* filename, outputStream* st) {
2071   int fd = ::open(filename, O_RDONLY);
2072   if (fd == -1) {
2073     return false;
2074   }
2075 
2076   char buf[32];
2077   int bytes;
2078   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2079     st->print_raw(buf, bytes);
2080   }
2081 
2082   ::close(fd);
2083 
2084   return true;
2085 }
2086 
2087 void os::print_dll_info(outputStream *st) {
2088   st->print_cr("Dynamic libraries:");
2089 
2090   char fname[32];
2091   pid_t pid = os::Linux::gettid();
2092 
2093   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2094 
2095   if (!_print_ascii_file(fname, st)) {
2096     st->print("Can not get library information for pid = %d\n", pid);
2097   }
2098 }
2099 
2100 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2101   FILE *procmapsFile = NULL;
2102 
2103   // Open the procfs maps file for the current process
2104   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2105     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2106     char line[PATH_MAX + 100];
2107 
2108     // Read line by line from 'file'
2109     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2110       u8 base, top, offset, inode;
2111       char permissions[5];
2112       char device[6];
2113       char name[PATH_MAX + 1];
2114 
2115       // Parse fields from line
2116       sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2117 
2118       // Filter by device id '00:00' so that we only get file system mapped files.
2119       if (strcmp(device, "00:00") != 0) {
2120 
2121         // Call callback with the fields of interest
2122         if(callback(name, (address)base, (address)top, param)) {
2123           // Oops abort, callback aborted
2124           fclose(procmapsFile);
2125           return 1;
2126         }
2127       }
2128     }
2129     fclose(procmapsFile);
2130   }
2131   return 0;
2132 }
2133 
2134 void os::print_os_info_brief(outputStream* st) {
2135   os::Linux::print_distro_info(st);
2136 
2137   os::Posix::print_uname_info(st);
2138 
2139   os::Linux::print_libversion_info(st);
2140 
2141 }
2142 
2143 void os::print_os_info(outputStream* st) {
2144   st->print("OS:");
2145 
2146   os::Linux::print_distro_info(st);
2147 
2148   os::Posix::print_uname_info(st);
2149 
2150   // Print warning if unsafe chroot environment detected
2151   if (unsafe_chroot_detected) {
2152     st->print("WARNING!! ");
2153     st->print_cr("%s", unstable_chroot_error);
2154   }
2155 
2156   os::Linux::print_libversion_info(st);
2157 
2158   os::Posix::print_rlimit_info(st);
2159 
2160   os::Posix::print_load_average(st);
2161 
2162   os::Linux::print_full_memory_info(st);
2163 }
2164 
2165 // Try to identify popular distros.
2166 // Most Linux distributions have a /etc/XXX-release file, which contains
2167 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2168 // file that also contains the OS version string. Some have more than one
2169 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2170 // /etc/redhat-release.), so the order is important.
2171 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2172 // their own specific XXX-release file as well as a redhat-release file.
2173 // Because of this the XXX-release file needs to be searched for before the
2174 // redhat-release file.
2175 // Since Red Hat has a lsb-release file that is not very descriptive the
2176 // search for redhat-release needs to be before lsb-release.
2177 // Since the lsb-release file is the new standard it needs to be searched
2178 // before the older style release files.
2179 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2180 // next to last resort.  The os-release file is a new standard that contains
2181 // distribution information and the system-release file seems to be an old
2182 // standard that has been replaced by the lsb-release and os-release files.
2183 // Searching for the debian_version file is the last resort.  It contains
2184 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2185 // "Debian " is printed before the contents of the debian_version file.
2186 void os::Linux::print_distro_info(outputStream* st) {
2187   if (!_print_ascii_file("/etc/oracle-release", st) &&
2188       !_print_ascii_file("/etc/mandriva-release", st) &&
2189       !_print_ascii_file("/etc/mandrake-release", st) &&
2190       !_print_ascii_file("/etc/sun-release", st) &&
2191       !_print_ascii_file("/etc/redhat-release", st) &&
2192       !_print_ascii_file("/etc/lsb-release", st) &&
2193       !_print_ascii_file("/etc/SuSE-release", st) &&
2194       !_print_ascii_file("/etc/turbolinux-release", st) &&
2195       !_print_ascii_file("/etc/gentoo-release", st) &&
2196       !_print_ascii_file("/etc/ltib-release", st) &&
2197       !_print_ascii_file("/etc/angstrom-version", st) &&
2198       !_print_ascii_file("/etc/system-release", st) &&
2199       !_print_ascii_file("/etc/os-release", st)) {
2200 
2201     if (file_exists("/etc/debian_version")) {
2202       st->print("Debian ");
2203       _print_ascii_file("/etc/debian_version", st);
2204     } else {
2205       st->print("Linux");
2206     }
2207   }
2208   st->cr();
2209 }
2210 
2211 void os::Linux::print_libversion_info(outputStream* st) {
2212   // libc, pthread
2213   st->print("libc:");
2214   st->print("%s ", os::Linux::glibc_version());
2215   st->print("%s ", os::Linux::libpthread_version());
2216   if (os::Linux::is_LinuxThreads()) {
2217     st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2218   }
2219   st->cr();
2220 }
2221 
2222 void os::Linux::print_full_memory_info(outputStream* st) {
2223   st->print("\n/proc/meminfo:\n");
2224   _print_ascii_file("/proc/meminfo", st);
2225   st->cr();
2226 }
2227 
2228 void os::print_memory_info(outputStream* st) {
2229 
2230   st->print("Memory:");
2231   st->print(" %dk page", os::vm_page_size()>>10);
2232 
2233   // values in struct sysinfo are "unsigned long"
2234   struct sysinfo si;
2235   sysinfo(&si);
2236 
2237   st->print(", physical " UINT64_FORMAT "k",
2238             os::physical_memory() >> 10);
2239   st->print("(" UINT64_FORMAT "k free)",
2240             os::available_memory() >> 10);
2241   st->print(", swap " UINT64_FORMAT "k",
2242             ((jlong)si.totalswap * si.mem_unit) >> 10);
2243   st->print("(" UINT64_FORMAT "k free)",
2244             ((jlong)si.freeswap * si.mem_unit) >> 10);
2245   st->cr();
2246 }
2247 
2248 void os::pd_print_cpu_info(outputStream* st) {
2249   st->print("\n/proc/cpuinfo:\n");
2250   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2251     st->print("  <Not Available>");
2252   }
2253   st->cr();
2254 }
2255 
2256 void os::print_siginfo(outputStream* st, void* siginfo) {
2257   const siginfo_t* si = (const siginfo_t*)siginfo;
2258 
2259   os::Posix::print_siginfo_brief(st, si);
2260 #if INCLUDE_CDS
2261   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2262       UseSharedSpaces) {
2263     FileMapInfo* mapinfo = FileMapInfo::current_info();
2264     if (mapinfo->is_in_shared_space(si->si_addr)) {
2265       st->print("\n\nError accessing class data sharing archive."   \
2266                 " Mapped file inaccessible during execution, "      \
2267                 " possible disk/network problem.");
2268     }
2269   }
2270 #endif
2271   st->cr();
2272 }
2273 
2274 
2275 static void print_signal_handler(outputStream* st, int sig,
2276                                  char* buf, size_t buflen);
2277 
2278 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2279   st->print_cr("Signal Handlers:");
2280   print_signal_handler(st, SIGSEGV, buf, buflen);
2281   print_signal_handler(st, SIGBUS , buf, buflen);
2282   print_signal_handler(st, SIGFPE , buf, buflen);
2283   print_signal_handler(st, SIGPIPE, buf, buflen);
2284   print_signal_handler(st, SIGXFSZ, buf, buflen);
2285   print_signal_handler(st, SIGILL , buf, buflen);
2286   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2287   print_signal_handler(st, SR_signum, buf, buflen);
2288   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2289   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2290   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2291   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2292 #if defined(PPC64)
2293   print_signal_handler(st, SIGTRAP, buf, buflen);
2294 #endif
2295 }
2296 
2297 static char saved_jvm_path[MAXPATHLEN] = {0};
2298 
2299 // Find the full path to the current module, libjvm.so
2300 void os::jvm_path(char *buf, jint buflen) {
2301   // Error checking.
2302   if (buflen < MAXPATHLEN) {
2303     assert(false, "must use a large-enough buffer");
2304     buf[0] = '\0';
2305     return;
2306   }
2307   // Lazy resolve the path to current module.
2308   if (saved_jvm_path[0] != 0) {
2309     strcpy(buf, saved_jvm_path);
2310     return;
2311   }
2312 
2313   char dli_fname[MAXPATHLEN];
2314   bool ret = dll_address_to_library_name(
2315                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2316                                          dli_fname, sizeof(dli_fname), NULL);
2317   assert(ret, "cannot locate libjvm");
2318   char *rp = NULL;
2319   if (ret && dli_fname[0] != '\0') {
2320     rp = realpath(dli_fname, buf);
2321   }
2322   if (rp == NULL) {
2323     return;
2324   }
2325 
2326   if (Arguments::sun_java_launcher_is_altjvm()) {
2327     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2328     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2329     // If "/jre/lib/" appears at the right place in the string, then
2330     // assume we are installed in a JDK and we're done. Otherwise, check
2331     // for a JAVA_HOME environment variable and fix up the path so it
2332     // looks like libjvm.so is installed there (append a fake suffix
2333     // hotspot/libjvm.so).
2334     const char *p = buf + strlen(buf) - 1;
2335     for (int count = 0; p > buf && count < 5; ++count) {
2336       for (--p; p > buf && *p != '/'; --p)
2337         /* empty */ ;
2338     }
2339 
2340     if (strncmp(p, "/jre/lib/", 9) != 0) {
2341       // Look for JAVA_HOME in the environment.
2342       char* java_home_var = ::getenv("JAVA_HOME");
2343       if (java_home_var != NULL && java_home_var[0] != 0) {
2344         char* jrelib_p;
2345         int len;
2346 
2347         // Check the current module name "libjvm.so".
2348         p = strrchr(buf, '/');
2349         if (p == NULL) {
2350           return;
2351         }
2352         assert(strstr(p, "/libjvm") == p, "invalid library name");
2353 
2354         rp = realpath(java_home_var, buf);
2355         if (rp == NULL) {
2356           return;
2357         }
2358 
2359         // determine if this is a legacy image or modules image
2360         // modules image doesn't have "jre" subdirectory
2361         len = strlen(buf);
2362         assert(len < buflen, "Ran out of buffer room");
2363         jrelib_p = buf + len;
2364         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2365         if (0 != access(buf, F_OK)) {
2366           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2367         }
2368 
2369         if (0 == access(buf, F_OK)) {
2370           // Use current module name "libjvm.so"
2371           len = strlen(buf);
2372           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2373         } else {
2374           // Go back to path of .so
2375           rp = realpath(dli_fname, buf);
2376           if (rp == NULL) {
2377             return;
2378           }
2379         }
2380       }
2381     }
2382   }
2383 
2384   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2385   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2386 }
2387 
2388 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2389   // no prefix required, not even "_"
2390 }
2391 
2392 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2393   // no suffix required
2394 }
2395 
2396 ////////////////////////////////////////////////////////////////////////////////
2397 // sun.misc.Signal support
2398 
2399 static volatile jint sigint_count = 0;
2400 
2401 static void UserHandler(int sig, void *siginfo, void *context) {
2402   // 4511530 - sem_post is serialized and handled by the manager thread. When
2403   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2404   // don't want to flood the manager thread with sem_post requests.
2405   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2406     return;
2407   }
2408 
2409   // Ctrl-C is pressed during error reporting, likely because the error
2410   // handler fails to abort. Let VM die immediately.
2411   if (sig == SIGINT && is_error_reported()) {
2412     os::die();
2413   }
2414 
2415   os::signal_notify(sig);
2416 }
2417 
2418 void* os::user_handler() {
2419   return CAST_FROM_FN_PTR(void*, UserHandler);
2420 }
2421 
2422 class Semaphore : public StackObj {
2423  public:
2424   Semaphore();
2425   ~Semaphore();
2426   void signal();
2427   void wait();
2428   bool trywait();
2429   bool timedwait(unsigned int sec, int nsec);
2430  private:
2431   sem_t _semaphore;
2432 };
2433 
2434 Semaphore::Semaphore() {
2435   sem_init(&_semaphore, 0, 0);
2436 }
2437 
2438 Semaphore::~Semaphore() {
2439   sem_destroy(&_semaphore);
2440 }
2441 
2442 void Semaphore::signal() {
2443   sem_post(&_semaphore);
2444 }
2445 
2446 void Semaphore::wait() {
2447   sem_wait(&_semaphore);
2448 }
2449 
2450 bool Semaphore::trywait() {
2451   return sem_trywait(&_semaphore) == 0;
2452 }
2453 
2454 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2455 
2456   struct timespec ts;
2457   // Semaphore's are always associated with CLOCK_REALTIME
2458   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2459   // see unpackTime for discussion on overflow checking
2460   if (sec >= MAX_SECS) {
2461     ts.tv_sec += MAX_SECS;
2462     ts.tv_nsec = 0;
2463   } else {
2464     ts.tv_sec += sec;
2465     ts.tv_nsec += nsec;
2466     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2467       ts.tv_nsec -= NANOSECS_PER_SEC;
2468       ++ts.tv_sec; // note: this must be <= max_secs
2469     }
2470   }
2471 
2472   while (1) {
2473     int result = sem_timedwait(&_semaphore, &ts);
2474     if (result == 0) {
2475       return true;
2476     } else if (errno == EINTR) {
2477       continue;
2478     } else if (errno == ETIMEDOUT) {
2479       return false;
2480     } else {
2481       return false;
2482     }
2483   }
2484 }
2485 
2486 extern "C" {
2487   typedef void (*sa_handler_t)(int);
2488   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2489 }
2490 
2491 void* os::signal(int signal_number, void* handler) {
2492   struct sigaction sigAct, oldSigAct;
2493 
2494   sigfillset(&(sigAct.sa_mask));
2495   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2496   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2497 
2498   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2499     // -1 means registration failed
2500     return (void *)-1;
2501   }
2502 
2503   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2504 }
2505 
2506 void os::signal_raise(int signal_number) {
2507   ::raise(signal_number);
2508 }
2509 
2510 // The following code is moved from os.cpp for making this
2511 // code platform specific, which it is by its very nature.
2512 
2513 // Will be modified when max signal is changed to be dynamic
2514 int os::sigexitnum_pd() {
2515   return NSIG;
2516 }
2517 
2518 // a counter for each possible signal value
2519 static volatile jint pending_signals[NSIG+1] = { 0 };
2520 
2521 // Linux(POSIX) specific hand shaking semaphore.
2522 static sem_t sig_sem;
2523 static Semaphore sr_semaphore;
2524 
2525 void os::signal_init_pd() {
2526   // Initialize signal structures
2527   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2528 
2529   // Initialize signal semaphore
2530   ::sem_init(&sig_sem, 0, 0);
2531 }
2532 
2533 void os::signal_notify(int sig) {
2534   Atomic::inc(&pending_signals[sig]);
2535   ::sem_post(&sig_sem);
2536 }
2537 
2538 static int check_pending_signals(bool wait) {
2539   Atomic::store(0, &sigint_count);
2540   for (;;) {
2541     for (int i = 0; i < NSIG + 1; i++) {
2542       jint n = pending_signals[i];
2543       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2544         return i;
2545       }
2546     }
2547     if (!wait) {
2548       return -1;
2549     }
2550     JavaThread *thread = JavaThread::current();
2551     ThreadBlockInVM tbivm(thread);
2552 
2553     bool threadIsSuspended;
2554     do {
2555       thread->set_suspend_equivalent();
2556       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2557       ::sem_wait(&sig_sem);
2558 
2559       // were we externally suspended while we were waiting?
2560       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2561       if (threadIsSuspended) {
2562         // The semaphore has been incremented, but while we were waiting
2563         // another thread suspended us. We don't want to continue running
2564         // while suspended because that would surprise the thread that
2565         // suspended us.
2566         ::sem_post(&sig_sem);
2567 
2568         thread->java_suspend_self();
2569       }
2570     } while (threadIsSuspended);
2571   }
2572 }
2573 
2574 int os::signal_lookup() {
2575   return check_pending_signals(false);
2576 }
2577 
2578 int os::signal_wait() {
2579   return check_pending_signals(true);
2580 }
2581 
2582 ////////////////////////////////////////////////////////////////////////////////
2583 // Virtual Memory
2584 
2585 int os::vm_page_size() {
2586   // Seems redundant as all get out
2587   assert(os::Linux::page_size() != -1, "must call os::init");
2588   return os::Linux::page_size();
2589 }
2590 
2591 // Solaris allocates memory by pages.
2592 int os::vm_allocation_granularity() {
2593   assert(os::Linux::page_size() != -1, "must call os::init");
2594   return os::Linux::page_size();
2595 }
2596 
2597 // Rationale behind this function:
2598 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2599 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2600 //  samples for JITted code. Here we create private executable mapping over the code cache
2601 //  and then we can use standard (well, almost, as mapping can change) way to provide
2602 //  info for the reporting script by storing timestamp and location of symbol
2603 void linux_wrap_code(char* base, size_t size) {
2604   static volatile jint cnt = 0;
2605 
2606   if (!UseOprofile) {
2607     return;
2608   }
2609 
2610   char buf[PATH_MAX+1];
2611   int num = Atomic::add(1, &cnt);
2612 
2613   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2614            os::get_temp_directory(), os::current_process_id(), num);
2615   unlink(buf);
2616 
2617   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2618 
2619   if (fd != -1) {
2620     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2621     if (rv != (off_t)-1) {
2622       if (::write(fd, "", 1) == 1) {
2623         mmap(base, size,
2624              PROT_READ|PROT_WRITE|PROT_EXEC,
2625              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2626       }
2627     }
2628     ::close(fd);
2629     unlink(buf);
2630   }
2631 }
2632 
2633 static bool recoverable_mmap_error(int err) {
2634   // See if the error is one we can let the caller handle. This
2635   // list of errno values comes from JBS-6843484. I can't find a
2636   // Linux man page that documents this specific set of errno
2637   // values so while this list currently matches Solaris, it may
2638   // change as we gain experience with this failure mode.
2639   switch (err) {
2640   case EBADF:
2641   case EINVAL:
2642   case ENOTSUP:
2643     // let the caller deal with these errors
2644     return true;
2645 
2646   default:
2647     // Any remaining errors on this OS can cause our reserved mapping
2648     // to be lost. That can cause confusion where different data
2649     // structures think they have the same memory mapped. The worst
2650     // scenario is if both the VM and a library think they have the
2651     // same memory mapped.
2652     return false;
2653   }
2654 }
2655 
2656 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2657                                     int err) {
2658   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2659           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2660           strerror(err), err);
2661 }
2662 
2663 static void warn_fail_commit_memory(char* addr, size_t size,
2664                                     size_t alignment_hint, bool exec,
2665                                     int err) {
2666   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2667           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2668           alignment_hint, exec, strerror(err), err);
2669 }
2670 
2671 // NOTE: Linux kernel does not really reserve the pages for us.
2672 //       All it does is to check if there are enough free pages
2673 //       left at the time of mmap(). This could be a potential
2674 //       problem.
2675 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2676   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2677   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2678                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2679   if (res != (uintptr_t) MAP_FAILED) {
2680     if (UseNUMAInterleaving) {
2681       numa_make_global(addr, size);
2682     }
2683     return 0;
2684   }
2685 
2686   int err = errno;  // save errno from mmap() call above
2687 
2688   if (!recoverable_mmap_error(err)) {
2689     warn_fail_commit_memory(addr, size, exec, err);
2690     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2691   }
2692 
2693   return err;
2694 }
2695 
2696 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2697   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2698 }
2699 
2700 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2701                                   const char* mesg) {
2702   assert(mesg != NULL, "mesg must be specified");
2703   int err = os::Linux::commit_memory_impl(addr, size, exec);
2704   if (err != 0) {
2705     // the caller wants all commit errors to exit with the specified mesg:
2706     warn_fail_commit_memory(addr, size, exec, err);
2707     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2708   }
2709 }
2710 
2711 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2712 #ifndef MAP_HUGETLB
2713   #define MAP_HUGETLB 0x40000
2714 #endif
2715 
2716 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2717 #ifndef MADV_HUGEPAGE
2718   #define MADV_HUGEPAGE 14
2719 #endif
2720 
2721 int os::Linux::commit_memory_impl(char* addr, size_t size,
2722                                   size_t alignment_hint, bool exec) {
2723   int err = os::Linux::commit_memory_impl(addr, size, exec);
2724   if (err == 0) {
2725     realign_memory(addr, size, alignment_hint);
2726   }
2727   return err;
2728 }
2729 
2730 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2731                           bool exec) {
2732   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2733 }
2734 
2735 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2736                                   size_t alignment_hint, bool exec,
2737                                   const char* mesg) {
2738   assert(mesg != NULL, "mesg must be specified");
2739   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2740   if (err != 0) {
2741     // the caller wants all commit errors to exit with the specified mesg:
2742     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2743     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2744   }
2745 }
2746 
2747 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2748   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2749     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2750     // be supported or the memory may already be backed by huge pages.
2751     ::madvise(addr, bytes, MADV_HUGEPAGE);
2752   }
2753 }
2754 
2755 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2756   // This method works by doing an mmap over an existing mmaping and effectively discarding
2757   // the existing pages. However it won't work for SHM-based large pages that cannot be
2758   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2759   // small pages on top of the SHM segment. This method always works for small pages, so we
2760   // allow that in any case.
2761   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2762     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2763   }
2764 }
2765 
2766 void os::numa_make_global(char *addr, size_t bytes) {
2767   Linux::numa_interleave_memory(addr, bytes);
2768 }
2769 
2770 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2771 // bind policy to MPOL_PREFERRED for the current thread.
2772 #define USE_MPOL_PREFERRED 0
2773 
2774 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2775   // To make NUMA and large pages more robust when both enabled, we need to ease
2776   // the requirements on where the memory should be allocated. MPOL_BIND is the
2777   // default policy and it will force memory to be allocated on the specified
2778   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2779   // the specified node, but will not force it. Using this policy will prevent
2780   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2781   // free large pages.
2782   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2783   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2784 }
2785 
2786 bool os::numa_topology_changed() { return false; }
2787 
2788 size_t os::numa_get_groups_num() {
2789   int max_node = Linux::numa_max_node();
2790   return max_node > 0 ? max_node + 1 : 1;
2791 }
2792 
2793 int os::numa_get_group_id() {
2794   int cpu_id = Linux::sched_getcpu();
2795   if (cpu_id != -1) {
2796     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2797     if (lgrp_id != -1) {
2798       return lgrp_id;
2799     }
2800   }
2801   return 0;
2802 }
2803 
2804 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2805   for (size_t i = 0; i < size; i++) {
2806     ids[i] = i;
2807   }
2808   return size;
2809 }
2810 
2811 bool os::get_page_info(char *start, page_info* info) {
2812   return false;
2813 }
2814 
2815 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2816                      page_info* page_found) {
2817   return end;
2818 }
2819 
2820 
2821 int os::Linux::sched_getcpu_syscall(void) {
2822   unsigned int cpu;
2823   int retval = -1;
2824 
2825 #if defined(IA32)
2826   #ifndef SYS_getcpu
2827     #define SYS_getcpu 318
2828   #endif
2829   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2830 #elif defined(AMD64)
2831 // Unfortunately we have to bring all these macros here from vsyscall.h
2832 // to be able to compile on old linuxes.
2833   #define __NR_vgetcpu 2
2834   #define VSYSCALL_START (-10UL << 20)
2835   #define VSYSCALL_SIZE 1024
2836   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2837   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2838   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2839   retval = vgetcpu(&cpu, NULL, NULL);
2840 #endif
2841 
2842   return (retval == -1) ? retval : cpu;
2843 }
2844 
2845 // Something to do with the numa-aware allocator needs these symbols
2846 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2847 extern "C" JNIEXPORT void numa_error(char *where) { }
2848 extern "C" JNIEXPORT int fork1() { return fork(); }
2849 
2850 
2851 // If we are running with libnuma version > 2, then we should
2852 // be trying to use symbols with versions 1.1
2853 // If we are running with earlier version, which did not have symbol versions,
2854 // we should use the base version.
2855 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2856   void *f = dlvsym(handle, name, "libnuma_1.1");
2857   if (f == NULL) {
2858     f = dlsym(handle, name);
2859   }
2860   return f;
2861 }
2862 
2863 bool os::Linux::libnuma_init() {
2864   // sched_getcpu() should be in libc.
2865   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2866                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2867 
2868   // If it's not, try a direct syscall.
2869   if (sched_getcpu() == -1) {
2870     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2871                                     (void*)&sched_getcpu_syscall));
2872   }
2873 
2874   if (sched_getcpu() != -1) { // Does it work?
2875     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2876     if (handle != NULL) {
2877       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2878                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2879       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2880                                        libnuma_dlsym(handle, "numa_max_node")));
2881       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2882                                         libnuma_dlsym(handle, "numa_available")));
2883       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2884                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2885       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2886                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2887       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2888                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2889 
2890 
2891       if (numa_available() != -1) {
2892         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2893         // Create a cpu -> node mapping
2894         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2895         rebuild_cpu_to_node_map();
2896         return true;
2897       }
2898     }
2899   }
2900   return false;
2901 }
2902 
2903 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2904 // The table is later used in get_node_by_cpu().
2905 void os::Linux::rebuild_cpu_to_node_map() {
2906   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2907                               // in libnuma (possible values are starting from 16,
2908                               // and continuing up with every other power of 2, but less
2909                               // than the maximum number of CPUs supported by kernel), and
2910                               // is a subject to change (in libnuma version 2 the requirements
2911                               // are more reasonable) we'll just hardcode the number they use
2912                               // in the library.
2913   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2914 
2915   size_t cpu_num = os::active_processor_count();
2916   size_t cpu_map_size = NCPUS / BitsPerCLong;
2917   size_t cpu_map_valid_size =
2918     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2919 
2920   cpu_to_node()->clear();
2921   cpu_to_node()->at_grow(cpu_num - 1);
2922   size_t node_num = numa_get_groups_num();
2923 
2924   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2925   for (size_t i = 0; i < node_num; i++) {
2926     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2927       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2928         if (cpu_map[j] != 0) {
2929           for (size_t k = 0; k < BitsPerCLong; k++) {
2930             if (cpu_map[j] & (1UL << k)) {
2931               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2932             }
2933           }
2934         }
2935       }
2936     }
2937   }
2938   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2939 }
2940 
2941 int os::Linux::get_node_by_cpu(int cpu_id) {
2942   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2943     return cpu_to_node()->at(cpu_id);
2944   }
2945   return -1;
2946 }
2947 
2948 GrowableArray<int>* os::Linux::_cpu_to_node;
2949 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2950 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2951 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2952 os::Linux::numa_available_func_t os::Linux::_numa_available;
2953 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2954 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2955 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2956 unsigned long* os::Linux::_numa_all_nodes;
2957 
2958 bool os::pd_uncommit_memory(char* addr, size_t size) {
2959   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2960                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2961   return res  != (uintptr_t) MAP_FAILED;
2962 }
2963 
2964 static address get_stack_commited_bottom(address bottom, size_t size) {
2965   address nbot = bottom;
2966   address ntop = bottom + size;
2967 
2968   size_t page_sz = os::vm_page_size();
2969   unsigned pages = size / page_sz;
2970 
2971   unsigned char vec[1];
2972   unsigned imin = 1, imax = pages + 1, imid;
2973   int mincore_return_value = 0;
2974 
2975   assert(imin <= imax, "Unexpected page size");
2976 
2977   while (imin < imax) {
2978     imid = (imax + imin) / 2;
2979     nbot = ntop - (imid * page_sz);
2980 
2981     // Use a trick with mincore to check whether the page is mapped or not.
2982     // mincore sets vec to 1 if page resides in memory and to 0 if page
2983     // is swapped output but if page we are asking for is unmapped
2984     // it returns -1,ENOMEM
2985     mincore_return_value = mincore(nbot, page_sz, vec);
2986 
2987     if (mincore_return_value == -1) {
2988       // Page is not mapped go up
2989       // to find first mapped page
2990       if (errno != EAGAIN) {
2991         assert(errno == ENOMEM, "Unexpected mincore errno");
2992         imax = imid;
2993       }
2994     } else {
2995       // Page is mapped go down
2996       // to find first not mapped page
2997       imin = imid + 1;
2998     }
2999   }
3000 
3001   nbot = nbot + page_sz;
3002 
3003   // Adjust stack bottom one page up if last checked page is not mapped
3004   if (mincore_return_value == -1) {
3005     nbot = nbot + page_sz;
3006   }
3007 
3008   return nbot;
3009 }
3010 
3011 
3012 // Linux uses a growable mapping for the stack, and if the mapping for
3013 // the stack guard pages is not removed when we detach a thread the
3014 // stack cannot grow beyond the pages where the stack guard was
3015 // mapped.  If at some point later in the process the stack expands to
3016 // that point, the Linux kernel cannot expand the stack any further
3017 // because the guard pages are in the way, and a segfault occurs.
3018 //
3019 // However, it's essential not to split the stack region by unmapping
3020 // a region (leaving a hole) that's already part of the stack mapping,
3021 // so if the stack mapping has already grown beyond the guard pages at
3022 // the time we create them, we have to truncate the stack mapping.
3023 // So, we need to know the extent of the stack mapping when
3024 // create_stack_guard_pages() is called.
3025 
3026 // We only need this for stacks that are growable: at the time of
3027 // writing thread stacks don't use growable mappings (i.e. those
3028 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3029 // only applies to the main thread.
3030 
3031 // If the (growable) stack mapping already extends beyond the point
3032 // where we're going to put our guard pages, truncate the mapping at
3033 // that point by munmap()ping it.  This ensures that when we later
3034 // munmap() the guard pages we don't leave a hole in the stack
3035 // mapping. This only affects the main/initial thread
3036 
3037 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3038   if (os::Linux::is_initial_thread()) {
3039     // As we manually grow stack up to bottom inside create_attached_thread(),
3040     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3041     // we don't need to do anything special.
3042     // Check it first, before calling heavy function.
3043     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3044     unsigned char vec[1];
3045 
3046     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3047       // Fallback to slow path on all errors, including EAGAIN
3048       stack_extent = (uintptr_t) get_stack_commited_bottom(
3049                                                            os::Linux::initial_thread_stack_bottom(),
3050                                                            (size_t)addr - stack_extent);
3051     }
3052 
3053     if (stack_extent < (uintptr_t)addr) {
3054       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3055     }
3056   }
3057 
3058   return os::commit_memory(addr, size, !ExecMem);
3059 }
3060 
3061 // If this is a growable mapping, remove the guard pages entirely by
3062 // munmap()ping them.  If not, just call uncommit_memory(). This only
3063 // affects the main/initial thread, but guard against future OS changes
3064 // It's safe to always unmap guard pages for initial thread because we
3065 // always place it right after end of the mapped region
3066 
3067 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3068   uintptr_t stack_extent, stack_base;
3069 
3070   if (os::Linux::is_initial_thread()) {
3071     return ::munmap(addr, size) == 0;
3072   }
3073 
3074   return os::uncommit_memory(addr, size);
3075 }
3076 
3077 static address _highest_vm_reserved_address = NULL;
3078 
3079 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3080 // at 'requested_addr'. If there are existing memory mappings at the same
3081 // location, however, they will be overwritten. If 'fixed' is false,
3082 // 'requested_addr' is only treated as a hint, the return value may or
3083 // may not start from the requested address. Unlike Linux mmap(), this
3084 // function returns NULL to indicate failure.
3085 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3086   char * addr;
3087   int flags;
3088 
3089   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3090   if (fixed) {
3091     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3092     flags |= MAP_FIXED;
3093   }
3094 
3095   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3096   // touch an uncommitted page. Otherwise, the read/write might
3097   // succeed if we have enough swap space to back the physical page.
3098   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3099                        flags, -1, 0);
3100 
3101   if (addr != MAP_FAILED) {
3102     // anon_mmap() should only get called during VM initialization,
3103     // don't need lock (actually we can skip locking even it can be called
3104     // from multiple threads, because _highest_vm_reserved_address is just a
3105     // hint about the upper limit of non-stack memory regions.)
3106     if ((address)addr + bytes > _highest_vm_reserved_address) {
3107       _highest_vm_reserved_address = (address)addr + bytes;
3108     }
3109   }
3110 
3111   return addr == MAP_FAILED ? NULL : addr;
3112 }
3113 
3114 // Don't update _highest_vm_reserved_address, because there might be memory
3115 // regions above addr + size. If so, releasing a memory region only creates
3116 // a hole in the address space, it doesn't help prevent heap-stack collision.
3117 //
3118 static int anon_munmap(char * addr, size_t size) {
3119   return ::munmap(addr, size) == 0;
3120 }
3121 
3122 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3123                             size_t alignment_hint) {
3124   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3125 }
3126 
3127 bool os::pd_release_memory(char* addr, size_t size) {
3128   return anon_munmap(addr, size);
3129 }
3130 
3131 static address highest_vm_reserved_address() {
3132   return _highest_vm_reserved_address;
3133 }
3134 
3135 static bool linux_mprotect(char* addr, size_t size, int prot) {
3136   // Linux wants the mprotect address argument to be page aligned.
3137   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3138 
3139   // According to SUSv3, mprotect() should only be used with mappings
3140   // established by mmap(), and mmap() always maps whole pages. Unaligned
3141   // 'addr' likely indicates problem in the VM (e.g. trying to change
3142   // protection of malloc'ed or statically allocated memory). Check the
3143   // caller if you hit this assert.
3144   assert(addr == bottom, "sanity check");
3145 
3146   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3147   return ::mprotect(bottom, size, prot) == 0;
3148 }
3149 
3150 // Set protections specified
3151 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3152                         bool is_committed) {
3153   unsigned int p = 0;
3154   switch (prot) {
3155   case MEM_PROT_NONE: p = PROT_NONE; break;
3156   case MEM_PROT_READ: p = PROT_READ; break;
3157   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3158   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3159   default:
3160     ShouldNotReachHere();
3161   }
3162   // is_committed is unused.
3163   return linux_mprotect(addr, bytes, p);
3164 }
3165 
3166 bool os::guard_memory(char* addr, size_t size) {
3167   return linux_mprotect(addr, size, PROT_NONE);
3168 }
3169 
3170 bool os::unguard_memory(char* addr, size_t size) {
3171   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3172 }
3173 
3174 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3175                                                     size_t page_size) {
3176   bool result = false;
3177   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3178                  MAP_ANONYMOUS|MAP_PRIVATE,
3179                  -1, 0);
3180   if (p != MAP_FAILED) {
3181     void *aligned_p = align_ptr_up(p, page_size);
3182 
3183     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3184 
3185     munmap(p, page_size * 2);
3186   }
3187 
3188   if (warn && !result) {
3189     warning("TransparentHugePages is not supported by the operating system.");
3190   }
3191 
3192   return result;
3193 }
3194 
3195 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3196   bool result = false;
3197   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3198                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3199                  -1, 0);
3200 
3201   if (p != MAP_FAILED) {
3202     // We don't know if this really is a huge page or not.
3203     FILE *fp = fopen("/proc/self/maps", "r");
3204     if (fp) {
3205       while (!feof(fp)) {
3206         char chars[257];
3207         long x = 0;
3208         if (fgets(chars, sizeof(chars), fp)) {
3209           if (sscanf(chars, "%lx-%*x", &x) == 1
3210               && x == (long)p) {
3211             if (strstr (chars, "hugepage")) {
3212               result = true;
3213               break;
3214             }
3215           }
3216         }
3217       }
3218       fclose(fp);
3219     }
3220     munmap(p, page_size);
3221   }
3222 
3223   if (warn && !result) {
3224     warning("HugeTLBFS is not supported by the operating system.");
3225   }
3226 
3227   return result;
3228 }
3229 
3230 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3231 //
3232 // From the coredump_filter documentation:
3233 //
3234 // - (bit 0) anonymous private memory
3235 // - (bit 1) anonymous shared memory
3236 // - (bit 2) file-backed private memory
3237 // - (bit 3) file-backed shared memory
3238 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3239 //           effective only if the bit 2 is cleared)
3240 // - (bit 5) hugetlb private memory
3241 // - (bit 6) hugetlb shared memory
3242 //
3243 static void set_coredump_filter(void) {
3244   FILE *f;
3245   long cdm;
3246 
3247   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3248     return;
3249   }
3250 
3251   if (fscanf(f, "%lx", &cdm) != 1) {
3252     fclose(f);
3253     return;
3254   }
3255 
3256   rewind(f);
3257 
3258   if ((cdm & LARGEPAGES_BIT) == 0) {
3259     cdm |= LARGEPAGES_BIT;
3260     fprintf(f, "%#lx", cdm);
3261   }
3262 
3263   fclose(f);
3264 }
3265 
3266 // Large page support
3267 
3268 static size_t _large_page_size = 0;
3269 
3270 size_t os::Linux::find_large_page_size() {
3271   size_t large_page_size = 0;
3272 
3273   // large_page_size on Linux is used to round up heap size. x86 uses either
3274   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3275   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3276   // page as large as 256M.
3277   //
3278   // Here we try to figure out page size by parsing /proc/meminfo and looking
3279   // for a line with the following format:
3280   //    Hugepagesize:     2048 kB
3281   //
3282   // If we can't determine the value (e.g. /proc is not mounted, or the text
3283   // format has been changed), we'll use the largest page size supported by
3284   // the processor.
3285 
3286 #ifndef ZERO
3287   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3288                      ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3289 #endif // ZERO
3290 
3291   FILE *fp = fopen("/proc/meminfo", "r");
3292   if (fp) {
3293     while (!feof(fp)) {
3294       int x = 0;
3295       char buf[16];
3296       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3297         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3298           large_page_size = x * K;
3299           break;
3300         }
3301       } else {
3302         // skip to next line
3303         for (;;) {
3304           int ch = fgetc(fp);
3305           if (ch == EOF || ch == (int)'\n') break;
3306         }
3307       }
3308     }
3309     fclose(fp);
3310   }
3311 
3312   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3313     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3314             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3315             proper_unit_for_byte_size(large_page_size));
3316   }
3317 
3318   return large_page_size;
3319 }
3320 
3321 size_t os::Linux::setup_large_page_size() {
3322   _large_page_size = Linux::find_large_page_size();
3323   const size_t default_page_size = (size_t)Linux::page_size();
3324   if (_large_page_size > default_page_size) {
3325     _page_sizes[0] = _large_page_size;
3326     _page_sizes[1] = default_page_size;
3327     _page_sizes[2] = 0;
3328   }
3329 
3330   return _large_page_size;
3331 }
3332 
3333 bool os::Linux::setup_large_page_type(size_t page_size) {
3334   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3335       FLAG_IS_DEFAULT(UseSHM) &&
3336       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3337 
3338     // The type of large pages has not been specified by the user.
3339 
3340     // Try UseHugeTLBFS and then UseSHM.
3341     UseHugeTLBFS = UseSHM = true;
3342 
3343     // Don't try UseTransparentHugePages since there are known
3344     // performance issues with it turned on. This might change in the future.
3345     UseTransparentHugePages = false;
3346   }
3347 
3348   if (UseTransparentHugePages) {
3349     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3350     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3351       UseHugeTLBFS = false;
3352       UseSHM = false;
3353       return true;
3354     }
3355     UseTransparentHugePages = false;
3356   }
3357 
3358   if (UseHugeTLBFS) {
3359     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3360     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3361       UseSHM = false;
3362       return true;
3363     }
3364     UseHugeTLBFS = false;
3365   }
3366 
3367   return UseSHM;
3368 }
3369 
3370 void os::large_page_init() {
3371   if (!UseLargePages &&
3372       !UseTransparentHugePages &&
3373       !UseHugeTLBFS &&
3374       !UseSHM) {
3375     // Not using large pages.
3376     return;
3377   }
3378 
3379   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3380     // The user explicitly turned off large pages.
3381     // Ignore the rest of the large pages flags.
3382     UseTransparentHugePages = false;
3383     UseHugeTLBFS = false;
3384     UseSHM = false;
3385     return;
3386   }
3387 
3388   size_t large_page_size = Linux::setup_large_page_size();
3389   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3390 
3391   set_coredump_filter();
3392 }
3393 
3394 #ifndef SHM_HUGETLB
3395   #define SHM_HUGETLB 04000
3396 #endif
3397 
3398 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3399                                             char* req_addr, bool exec) {
3400   // "exec" is passed in but not used.  Creating the shared image for
3401   // the code cache doesn't have an SHM_X executable permission to check.
3402   assert(UseLargePages && UseSHM, "only for SHM large pages");
3403   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3404 
3405   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3406     return NULL; // Fallback to small pages.
3407   }
3408 
3409   key_t key = IPC_PRIVATE;
3410   char *addr;
3411 
3412   bool warn_on_failure = UseLargePages &&
3413                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3414                          !FLAG_IS_DEFAULT(UseSHM) ||
3415                          !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3416   char msg[128];
3417 
3418   // Create a large shared memory region to attach to based on size.
3419   // Currently, size is the total size of the heap
3420   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3421   if (shmid == -1) {
3422     // Possible reasons for shmget failure:
3423     // 1. shmmax is too small for Java heap.
3424     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3425     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3426     // 2. not enough large page memory.
3427     //    > check available large pages: cat /proc/meminfo
3428     //    > increase amount of large pages:
3429     //          echo new_value > /proc/sys/vm/nr_hugepages
3430     //      Note 1: different Linux may use different name for this property,
3431     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3432     //      Note 2: it's possible there's enough physical memory available but
3433     //            they are so fragmented after a long run that they can't
3434     //            coalesce into large pages. Try to reserve large pages when
3435     //            the system is still "fresh".
3436     if (warn_on_failure) {
3437       jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3438       warning("%s", msg);
3439     }
3440     return NULL;
3441   }
3442 
3443   // attach to the region
3444   addr = (char*)shmat(shmid, req_addr, 0);
3445   int err = errno;
3446 
3447   // Remove shmid. If shmat() is successful, the actual shared memory segment
3448   // will be deleted when it's detached by shmdt() or when the process
3449   // terminates. If shmat() is not successful this will remove the shared
3450   // segment immediately.
3451   shmctl(shmid, IPC_RMID, NULL);
3452 
3453   if ((intptr_t)addr == -1) {
3454     if (warn_on_failure) {
3455       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3456       warning("%s", msg);
3457     }
3458     return NULL;
3459   }
3460 
3461   return addr;
3462 }
3463 
3464 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3465                                         int error) {
3466   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3467 
3468   bool warn_on_failure = UseLargePages &&
3469       (!FLAG_IS_DEFAULT(UseLargePages) ||
3470        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3471        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3472 
3473   if (warn_on_failure) {
3474     char msg[128];
3475     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3476                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3477     warning("%s", msg);
3478   }
3479 }
3480 
3481 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3482                                                         char* req_addr,
3483                                                         bool exec) {
3484   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3485   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3486   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3487 
3488   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3489   char* addr = (char*)::mmap(req_addr, bytes, prot,
3490                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3491                              -1, 0);
3492 
3493   if (addr == MAP_FAILED) {
3494     warn_on_large_pages_failure(req_addr, bytes, errno);
3495     return NULL;
3496   }
3497 
3498   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3499 
3500   return addr;
3501 }
3502 
3503 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3504                                                          size_t alignment,
3505                                                          char* req_addr,
3506                                                          bool exec) {
3507   size_t large_page_size = os::large_page_size();
3508 
3509   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3510 
3511   // Allocate small pages.
3512 
3513   char* start;
3514   if (req_addr != NULL) {
3515     assert(is_ptr_aligned(req_addr, alignment), "Must be");
3516     assert(is_size_aligned(bytes, alignment), "Must be");
3517     start = os::reserve_memory(bytes, req_addr);
3518     assert(start == NULL || start == req_addr, "Must be");
3519   } else {
3520     start = os::reserve_memory_aligned(bytes, alignment);
3521   }
3522 
3523   if (start == NULL) {
3524     return NULL;
3525   }
3526 
3527   assert(is_ptr_aligned(start, alignment), "Must be");
3528 
3529   if (MemTracker::tracking_level() > NMT_minimal) {
3530     // os::reserve_memory_special will record this memory area.
3531     // Need to release it here to prevent overlapping reservations.
3532     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3533     tkr.record((address)start, bytes);
3534   }
3535 
3536   char* end = start + bytes;
3537 
3538   // Find the regions of the allocated chunk that can be promoted to large pages.
3539   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3540   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3541 
3542   size_t lp_bytes = lp_end - lp_start;
3543 
3544   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3545 
3546   if (lp_bytes == 0) {
3547     // The mapped region doesn't even span the start and the end of a large page.
3548     // Fall back to allocate a non-special area.
3549     ::munmap(start, end - start);
3550     return NULL;
3551   }
3552 
3553   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3554 
3555 
3556   void* result;
3557 
3558   if (start != lp_start) {
3559     result = ::mmap(start, lp_start - start, prot,
3560                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3561                     -1, 0);
3562     if (result == MAP_FAILED) {
3563       ::munmap(lp_start, end - lp_start);
3564       return NULL;
3565     }
3566   }
3567 
3568   result = ::mmap(lp_start, lp_bytes, prot,
3569                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3570                   -1, 0);
3571   if (result == MAP_FAILED) {
3572     warn_on_large_pages_failure(req_addr, bytes, errno);
3573     // If the mmap above fails, the large pages region will be unmapped and we
3574     // have regions before and after with small pages. Release these regions.
3575     //
3576     // |  mapped  |  unmapped  |  mapped  |
3577     // ^          ^            ^          ^
3578     // start      lp_start     lp_end     end
3579     //
3580     ::munmap(start, lp_start - start);
3581     ::munmap(lp_end, end - lp_end);
3582     return NULL;
3583   }
3584 
3585   if (lp_end != end) {
3586     result = ::mmap(lp_end, end - lp_end, prot,
3587                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3588                     -1, 0);
3589     if (result == MAP_FAILED) {
3590       ::munmap(start, lp_end - start);
3591       return NULL;
3592     }
3593   }
3594 
3595   return start;
3596 }
3597 
3598 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3599                                                    size_t alignment,
3600                                                    char* req_addr,
3601                                                    bool exec) {
3602   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3603   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3604   assert(is_power_of_2(alignment), "Must be");
3605   assert(is_power_of_2(os::large_page_size()), "Must be");
3606   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3607 
3608   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3609     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3610   } else {
3611     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3612   }
3613 }
3614 
3615 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3616                                  char* req_addr, bool exec) {
3617   assert(UseLargePages, "only for large pages");
3618 
3619   char* addr;
3620   if (UseSHM) {
3621     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3622   } else {
3623     assert(UseHugeTLBFS, "must be");
3624     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3625   }
3626 
3627   if (addr != NULL) {
3628     if (UseNUMAInterleaving) {
3629       numa_make_global(addr, bytes);
3630     }
3631 
3632     // The memory is committed
3633     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3634   }
3635 
3636   return addr;
3637 }
3638 
3639 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3640   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3641   return shmdt(base) == 0;
3642 }
3643 
3644 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3645   return pd_release_memory(base, bytes);
3646 }
3647 
3648 bool os::release_memory_special(char* base, size_t bytes) {
3649   bool res;
3650   if (MemTracker::tracking_level() > NMT_minimal) {
3651     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3652     res = os::Linux::release_memory_special_impl(base, bytes);
3653     if (res) {
3654       tkr.record((address)base, bytes);
3655     }
3656 
3657   } else {
3658     res = os::Linux::release_memory_special_impl(base, bytes);
3659   }
3660   return res;
3661 }
3662 
3663 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3664   assert(UseLargePages, "only for large pages");
3665   bool res;
3666 
3667   if (UseSHM) {
3668     res = os::Linux::release_memory_special_shm(base, bytes);
3669   } else {
3670     assert(UseHugeTLBFS, "must be");
3671     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3672   }
3673   return res;
3674 }
3675 
3676 size_t os::large_page_size() {
3677   return _large_page_size;
3678 }
3679 
3680 // With SysV SHM the entire memory region must be allocated as shared
3681 // memory.
3682 // HugeTLBFS allows application to commit large page memory on demand.
3683 // However, when committing memory with HugeTLBFS fails, the region
3684 // that was supposed to be committed will lose the old reservation
3685 // and allow other threads to steal that memory region. Because of this
3686 // behavior we can't commit HugeTLBFS memory.
3687 bool os::can_commit_large_page_memory() {
3688   return UseTransparentHugePages;
3689 }
3690 
3691 bool os::can_execute_large_page_memory() {
3692   return UseTransparentHugePages || UseHugeTLBFS;
3693 }
3694 
3695 // Reserve memory at an arbitrary address, only if that area is
3696 // available (and not reserved for something else).
3697 
3698 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3699   const int max_tries = 10;
3700   char* base[max_tries];
3701   size_t size[max_tries];
3702   const size_t gap = 0x000000;
3703 
3704   // Assert only that the size is a multiple of the page size, since
3705   // that's all that mmap requires, and since that's all we really know
3706   // about at this low abstraction level.  If we need higher alignment,
3707   // we can either pass an alignment to this method or verify alignment
3708   // in one of the methods further up the call chain.  See bug 5044738.
3709   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3710 
3711   // Repeatedly allocate blocks until the block is allocated at the
3712   // right spot. Give up after max_tries. Note that reserve_memory() will
3713   // automatically update _highest_vm_reserved_address if the call is
3714   // successful. The variable tracks the highest memory address every reserved
3715   // by JVM. It is used to detect heap-stack collision if running with
3716   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3717   // space than needed, it could confuse the collision detecting code. To
3718   // solve the problem, save current _highest_vm_reserved_address and
3719   // calculate the correct value before return.
3720   address old_highest = _highest_vm_reserved_address;
3721 
3722   // Linux mmap allows caller to pass an address as hint; give it a try first,
3723   // if kernel honors the hint then we can return immediately.
3724   char * addr = anon_mmap(requested_addr, bytes, false);
3725   if (addr == requested_addr) {
3726     return requested_addr;
3727   }
3728 
3729   if (addr != NULL) {
3730     // mmap() is successful but it fails to reserve at the requested address
3731     anon_munmap(addr, bytes);
3732   }
3733 
3734   int i;
3735   for (i = 0; i < max_tries; ++i) {
3736     base[i] = reserve_memory(bytes);
3737 
3738     if (base[i] != NULL) {
3739       // Is this the block we wanted?
3740       if (base[i] == requested_addr) {
3741         size[i] = bytes;
3742         break;
3743       }
3744 
3745       // Does this overlap the block we wanted? Give back the overlapped
3746       // parts and try again.
3747 
3748       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3749       if (top_overlap >= 0 && top_overlap < bytes) {
3750         unmap_memory(base[i], top_overlap);
3751         base[i] += top_overlap;
3752         size[i] = bytes - top_overlap;
3753       } else {
3754         size_t bottom_overlap = base[i] + bytes - requested_addr;
3755         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3756           unmap_memory(requested_addr, bottom_overlap);
3757           size[i] = bytes - bottom_overlap;
3758         } else {
3759           size[i] = bytes;
3760         }
3761       }
3762     }
3763   }
3764 
3765   // Give back the unused reserved pieces.
3766 
3767   for (int j = 0; j < i; ++j) {
3768     if (base[j] != NULL) {
3769       unmap_memory(base[j], size[j]);
3770     }
3771   }
3772 
3773   if (i < max_tries) {
3774     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3775     return requested_addr;
3776   } else {
3777     _highest_vm_reserved_address = old_highest;
3778     return NULL;
3779   }
3780 }
3781 
3782 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3783   return ::read(fd, buf, nBytes);
3784 }
3785 




3786 // Short sleep, direct OS call.
3787 //
3788 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3789 // sched_yield(2) will actually give up the CPU:
3790 //
3791 //   * Alone on this pariticular CPU, keeps running.
3792 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3793 //     (pre 2.6.39).
3794 //
3795 // So calling this with 0 is an alternative.
3796 //
3797 void os::naked_short_sleep(jlong ms) {
3798   struct timespec req;
3799 
3800   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3801   req.tv_sec = 0;
3802   if (ms > 0) {
3803     req.tv_nsec = (ms % 1000) * 1000000;
3804   } else {
3805     req.tv_nsec = 1;
3806   }
3807 
3808   nanosleep(&req, NULL);
3809 
3810   return;
3811 }
3812 
3813 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3814 void os::infinite_sleep() {
3815   while (true) {    // sleep forever ...
3816     ::sleep(100);   // ... 100 seconds at a time
3817   }
3818 }
3819 
3820 // Used to convert frequent JVM_Yield() to nops
3821 bool os::dont_yield() {
3822   return DontYieldALot;
3823 }
3824 
3825 void os::naked_yield() {
3826   sched_yield();
3827 }
3828 
3829 ////////////////////////////////////////////////////////////////////////////////
3830 // thread priority support
3831 
3832 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3833 // only supports dynamic priority, static priority must be zero. For real-time
3834 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3835 // However, for large multi-threaded applications, SCHED_RR is not only slower
3836 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3837 // of 5 runs - Sep 2005).
3838 //
3839 // The following code actually changes the niceness of kernel-thread/LWP. It
3840 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3841 // not the entire user process, and user level threads are 1:1 mapped to kernel
3842 // threads. It has always been the case, but could change in the future. For
3843 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3844 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3845 
3846 int os::java_to_os_priority[CriticalPriority + 1] = {
3847   19,              // 0 Entry should never be used
3848 
3849    4,              // 1 MinPriority
3850    3,              // 2
3851    2,              // 3
3852 
3853    1,              // 4
3854    0,              // 5 NormPriority
3855   -1,              // 6
3856 
3857   -2,              // 7
3858   -3,              // 8
3859   -4,              // 9 NearMaxPriority
3860 
3861   -5,              // 10 MaxPriority
3862 
3863   -5               // 11 CriticalPriority
3864 };
3865 
3866 static int prio_init() {
3867   if (ThreadPriorityPolicy == 1) {
3868     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3869     // if effective uid is not root. Perhaps, a more elegant way of doing
3870     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3871     if (geteuid() != 0) {
3872       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3873         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3874       }
3875       ThreadPriorityPolicy = 0;
3876     }
3877   }
3878   if (UseCriticalJavaThreadPriority) {
3879     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3880   }
3881   return 0;
3882 }
3883 
3884 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3885   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3886 
3887   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3888   return (ret == 0) ? OS_OK : OS_ERR;
3889 }
3890 
3891 OSReturn os::get_native_priority(const Thread* const thread,
3892                                  int *priority_ptr) {
3893   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3894     *priority_ptr = java_to_os_priority[NormPriority];
3895     return OS_OK;
3896   }
3897 
3898   errno = 0;
3899   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3900   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3901 }
3902 
3903 // Hint to the underlying OS that a task switch would not be good.
3904 // Void return because it's a hint and can fail.
3905 void os::hint_no_preempt() {}
3906 
3907 ////////////////////////////////////////////////////////////////////////////////
3908 // suspend/resume support
3909 
3910 //  the low-level signal-based suspend/resume support is a remnant from the
3911 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3912 //  within hotspot. Now there is a single use-case for this:
3913 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3914 //      that runs in the watcher thread.
3915 //  The remaining code is greatly simplified from the more general suspension
3916 //  code that used to be used.
3917 //
3918 //  The protocol is quite simple:
3919 //  - suspend:
3920 //      - sends a signal to the target thread
3921 //      - polls the suspend state of the osthread using a yield loop
3922 //      - target thread signal handler (SR_handler) sets suspend state
3923 //        and blocks in sigsuspend until continued
3924 //  - resume:
3925 //      - sets target osthread state to continue
3926 //      - sends signal to end the sigsuspend loop in the SR_handler
3927 //
3928 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3929 
3930 static void resume_clear_context(OSThread *osthread) {
3931   osthread->set_ucontext(NULL);
3932   osthread->set_siginfo(NULL);
3933 }
3934 
3935 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3936                                  ucontext_t* context) {
3937   osthread->set_ucontext(context);
3938   osthread->set_siginfo(siginfo);
3939 }
3940 
3941 // Handler function invoked when a thread's execution is suspended or
3942 // resumed. We have to be careful that only async-safe functions are
3943 // called here (Note: most pthread functions are not async safe and
3944 // should be avoided.)
3945 //
3946 // Note: sigwait() is a more natural fit than sigsuspend() from an
3947 // interface point of view, but sigwait() prevents the signal hander
3948 // from being run. libpthread would get very confused by not having
3949 // its signal handlers run and prevents sigwait()'s use with the
3950 // mutex granting granting signal.
3951 //
3952 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3953 //
3954 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3955   // Save and restore errno to avoid confusing native code with EINTR
3956   // after sigsuspend.
3957   int old_errno = errno;
3958 
3959   Thread* thread = Thread::current();
3960   OSThread* osthread = thread->osthread();
3961   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3962 
3963   os::SuspendResume::State current = osthread->sr.state();
3964   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3965     suspend_save_context(osthread, siginfo, context);
3966 
3967     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3968     os::SuspendResume::State state = osthread->sr.suspended();
3969     if (state == os::SuspendResume::SR_SUSPENDED) {
3970       sigset_t suspend_set;  // signals for sigsuspend()
3971 
3972       // get current set of blocked signals and unblock resume signal
3973       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3974       sigdelset(&suspend_set, SR_signum);
3975 
3976       sr_semaphore.signal();
3977       // wait here until we are resumed
3978       while (1) {
3979         sigsuspend(&suspend_set);
3980 
3981         os::SuspendResume::State result = osthread->sr.running();
3982         if (result == os::SuspendResume::SR_RUNNING) {
3983           sr_semaphore.signal();
3984           break;
3985         }
3986       }
3987 
3988     } else if (state == os::SuspendResume::SR_RUNNING) {
3989       // request was cancelled, continue
3990     } else {
3991       ShouldNotReachHere();
3992     }
3993 
3994     resume_clear_context(osthread);
3995   } else if (current == os::SuspendResume::SR_RUNNING) {
3996     // request was cancelled, continue
3997   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3998     // ignore
3999   } else {
4000     // ignore
4001   }
4002 
4003   errno = old_errno;
4004 }
4005 
4006 
4007 static int SR_initialize() {
4008   struct sigaction act;
4009   char *s;
4010   // Get signal number to use for suspend/resume
4011   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4012     int sig = ::strtol(s, 0, 10);
4013     if (sig > 0 || sig < _NSIG) {
4014       SR_signum = sig;
4015     }
4016   }
4017 
4018   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4019          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4020 
4021   sigemptyset(&SR_sigset);
4022   sigaddset(&SR_sigset, SR_signum);
4023 
4024   // Set up signal handler for suspend/resume
4025   act.sa_flags = SA_RESTART|SA_SIGINFO;
4026   act.sa_handler = (void (*)(int)) SR_handler;
4027 
4028   // SR_signum is blocked by default.
4029   // 4528190 - We also need to block pthread restart signal (32 on all
4030   // supported Linux platforms). Note that LinuxThreads need to block
4031   // this signal for all threads to work properly. So we don't have
4032   // to use hard-coded signal number when setting up the mask.
4033   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4034 
4035   if (sigaction(SR_signum, &act, 0) == -1) {
4036     return -1;
4037   }
4038 
4039   // Save signal flag
4040   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4041   return 0;
4042 }
4043 
4044 static int sr_notify(OSThread* osthread) {
4045   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4046   assert_status(status == 0, status, "pthread_kill");
4047   return status;
4048 }
4049 
4050 // "Randomly" selected value for how long we want to spin
4051 // before bailing out on suspending a thread, also how often
4052 // we send a signal to a thread we want to resume
4053 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4054 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4055 
4056 // returns true on success and false on error - really an error is fatal
4057 // but this seems the normal response to library errors
4058 static bool do_suspend(OSThread* osthread) {
4059   assert(osthread->sr.is_running(), "thread should be running");
4060   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4061 
4062   // mark as suspended and send signal
4063   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4064     // failed to switch, state wasn't running?
4065     ShouldNotReachHere();
4066     return false;
4067   }
4068 
4069   if (sr_notify(osthread) != 0) {
4070     ShouldNotReachHere();
4071   }
4072 
4073   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4074   while (true) {
4075     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4076       break;
4077     } else {
4078       // timeout
4079       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4080       if (cancelled == os::SuspendResume::SR_RUNNING) {
4081         return false;
4082       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4083         // make sure that we consume the signal on the semaphore as well
4084         sr_semaphore.wait();
4085         break;
4086       } else {
4087         ShouldNotReachHere();
4088         return false;
4089       }
4090     }
4091   }
4092 
4093   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4094   return true;
4095 }
4096 
4097 static void do_resume(OSThread* osthread) {
4098   assert(osthread->sr.is_suspended(), "thread should be suspended");
4099   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4100 
4101   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4102     // failed to switch to WAKEUP_REQUEST
4103     ShouldNotReachHere();
4104     return;
4105   }
4106 
4107   while (true) {
4108     if (sr_notify(osthread) == 0) {
4109       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4110         if (osthread->sr.is_running()) {
4111           return;
4112         }
4113       }
4114     } else {
4115       ShouldNotReachHere();
4116     }
4117   }
4118 
4119   guarantee(osthread->sr.is_running(), "Must be running!");
4120 }
4121 
4122 ///////////////////////////////////////////////////////////////////////////////////
4123 // signal handling (except suspend/resume)
4124 
4125 // This routine may be used by user applications as a "hook" to catch signals.
4126 // The user-defined signal handler must pass unrecognized signals to this
4127 // routine, and if it returns true (non-zero), then the signal handler must
4128 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4129 // routine will never retun false (zero), but instead will execute a VM panic
4130 // routine kill the process.
4131 //
4132 // If this routine returns false, it is OK to call it again.  This allows
4133 // the user-defined signal handler to perform checks either before or after
4134 // the VM performs its own checks.  Naturally, the user code would be making
4135 // a serious error if it tried to handle an exception (such as a null check
4136 // or breakpoint) that the VM was generating for its own correct operation.
4137 //
4138 // This routine may recognize any of the following kinds of signals:
4139 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4140 // It should be consulted by handlers for any of those signals.
4141 //
4142 // The caller of this routine must pass in the three arguments supplied
4143 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4144 // field of the structure passed to sigaction().  This routine assumes that
4145 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4146 //
4147 // Note that the VM will print warnings if it detects conflicting signal
4148 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4149 //
4150 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4151                                                  siginfo_t* siginfo,
4152                                                  void* ucontext,
4153                                                  int abort_if_unrecognized);
4154 
4155 void signalHandler(int sig, siginfo_t* info, void* uc) {
4156   assert(info != NULL && uc != NULL, "it must be old kernel");
4157   int orig_errno = errno;  // Preserve errno value over signal handler.
4158   JVM_handle_linux_signal(sig, info, uc, true);
4159   errno = orig_errno;
4160 }
4161 
4162 
4163 // This boolean allows users to forward their own non-matching signals
4164 // to JVM_handle_linux_signal, harmlessly.
4165 bool os::Linux::signal_handlers_are_installed = false;
4166 
4167 // For signal-chaining
4168 struct sigaction os::Linux::sigact[MAXSIGNUM];
4169 unsigned int os::Linux::sigs = 0;
4170 bool os::Linux::libjsig_is_loaded = false;
4171 typedef struct sigaction *(*get_signal_t)(int);
4172 get_signal_t os::Linux::get_signal_action = NULL;
4173 
4174 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4175   struct sigaction *actp = NULL;
4176 
4177   if (libjsig_is_loaded) {
4178     // Retrieve the old signal handler from libjsig
4179     actp = (*get_signal_action)(sig);
4180   }
4181   if (actp == NULL) {
4182     // Retrieve the preinstalled signal handler from jvm
4183     actp = get_preinstalled_handler(sig);
4184   }
4185 
4186   return actp;
4187 }
4188 
4189 static bool call_chained_handler(struct sigaction *actp, int sig,
4190                                  siginfo_t *siginfo, void *context) {
4191   // Call the old signal handler
4192   if (actp->sa_handler == SIG_DFL) {
4193     // It's more reasonable to let jvm treat it as an unexpected exception
4194     // instead of taking the default action.
4195     return false;
4196   } else if (actp->sa_handler != SIG_IGN) {
4197     if ((actp->sa_flags & SA_NODEFER) == 0) {
4198       // automaticlly block the signal
4199       sigaddset(&(actp->sa_mask), sig);
4200     }
4201 
4202     sa_handler_t hand;
4203     sa_sigaction_t sa;
4204     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4205     // retrieve the chained handler
4206     if (siginfo_flag_set) {
4207       sa = actp->sa_sigaction;
4208     } else {
4209       hand = actp->sa_handler;
4210     }
4211 
4212     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4213       actp->sa_handler = SIG_DFL;
4214     }
4215 
4216     // try to honor the signal mask
4217     sigset_t oset;
4218     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4219 
4220     // call into the chained handler
4221     if (siginfo_flag_set) {
4222       (*sa)(sig, siginfo, context);
4223     } else {
4224       (*hand)(sig);
4225     }
4226 
4227     // restore the signal mask
4228     pthread_sigmask(SIG_SETMASK, &oset, 0);
4229   }
4230   // Tell jvm's signal handler the signal is taken care of.
4231   return true;
4232 }
4233 
4234 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4235   bool chained = false;
4236   // signal-chaining
4237   if (UseSignalChaining) {
4238     struct sigaction *actp = get_chained_signal_action(sig);
4239     if (actp != NULL) {
4240       chained = call_chained_handler(actp, sig, siginfo, context);
4241     }
4242   }
4243   return chained;
4244 }
4245 
4246 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4247   if ((((unsigned int)1 << sig) & sigs) != 0) {
4248     return &sigact[sig];
4249   }
4250   return NULL;
4251 }
4252 
4253 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4254   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4255   sigact[sig] = oldAct;
4256   sigs |= (unsigned int)1 << sig;
4257 }
4258 
4259 // for diagnostic
4260 int os::Linux::sigflags[MAXSIGNUM];
4261 
4262 int os::Linux::get_our_sigflags(int sig) {
4263   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4264   return sigflags[sig];
4265 }
4266 
4267 void os::Linux::set_our_sigflags(int sig, int flags) {
4268   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4269   sigflags[sig] = flags;
4270 }
4271 
4272 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4273   // Check for overwrite.
4274   struct sigaction oldAct;
4275   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4276 
4277   void* oldhand = oldAct.sa_sigaction
4278                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4279                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4280   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4281       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4282       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4283     if (AllowUserSignalHandlers || !set_installed) {
4284       // Do not overwrite; user takes responsibility to forward to us.
4285       return;
4286     } else if (UseSignalChaining) {
4287       // save the old handler in jvm
4288       save_preinstalled_handler(sig, oldAct);
4289       // libjsig also interposes the sigaction() call below and saves the
4290       // old sigaction on it own.
4291     } else {
4292       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4293                     "%#lx for signal %d.", (long)oldhand, sig));
4294     }
4295   }
4296 
4297   struct sigaction sigAct;
4298   sigfillset(&(sigAct.sa_mask));
4299   sigAct.sa_handler = SIG_DFL;
4300   if (!set_installed) {
4301     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4302   } else {
4303     sigAct.sa_sigaction = signalHandler;
4304     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4305   }
4306   // Save flags, which are set by ours
4307   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4308   sigflags[sig] = sigAct.sa_flags;
4309 
4310   int ret = sigaction(sig, &sigAct, &oldAct);
4311   assert(ret == 0, "check");
4312 
4313   void* oldhand2  = oldAct.sa_sigaction
4314                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4315                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4316   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4317 }
4318 
4319 // install signal handlers for signals that HotSpot needs to
4320 // handle in order to support Java-level exception handling.
4321 
4322 void os::Linux::install_signal_handlers() {
4323   if (!signal_handlers_are_installed) {
4324     signal_handlers_are_installed = true;
4325 
4326     // signal-chaining
4327     typedef void (*signal_setting_t)();
4328     signal_setting_t begin_signal_setting = NULL;
4329     signal_setting_t end_signal_setting = NULL;
4330     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4331                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4332     if (begin_signal_setting != NULL) {
4333       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4334                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4335       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4336                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4337       libjsig_is_loaded = true;
4338       assert(UseSignalChaining, "should enable signal-chaining");
4339     }
4340     if (libjsig_is_loaded) {
4341       // Tell libjsig jvm is setting signal handlers
4342       (*begin_signal_setting)();
4343     }
4344 
4345     set_signal_handler(SIGSEGV, true);
4346     set_signal_handler(SIGPIPE, true);
4347     set_signal_handler(SIGBUS, true);
4348     set_signal_handler(SIGILL, true);
4349     set_signal_handler(SIGFPE, true);
4350 #if defined(PPC64)
4351     set_signal_handler(SIGTRAP, true);
4352 #endif
4353     set_signal_handler(SIGXFSZ, true);
4354 
4355     if (libjsig_is_loaded) {
4356       // Tell libjsig jvm finishes setting signal handlers
4357       (*end_signal_setting)();
4358     }
4359 
4360     // We don't activate signal checker if libjsig is in place, we trust ourselves
4361     // and if UserSignalHandler is installed all bets are off.
4362     // Log that signal checking is off only if -verbose:jni is specified.
4363     if (CheckJNICalls) {
4364       if (libjsig_is_loaded) {
4365         if (PrintJNIResolving) {
4366           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4367         }
4368         check_signals = false;
4369       }
4370       if (AllowUserSignalHandlers) {
4371         if (PrintJNIResolving) {
4372           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4373         }
4374         check_signals = false;
4375       }
4376     }
4377   }
4378 }
4379 
4380 // This is the fastest way to get thread cpu time on Linux.
4381 // Returns cpu time (user+sys) for any thread, not only for current.
4382 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4383 // It might work on 2.6.10+ with a special kernel/glibc patch.
4384 // For reference, please, see IEEE Std 1003.1-2004:
4385 //   http://www.unix.org/single_unix_specification
4386 
4387 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4388   struct timespec tp;
4389   int rc = os::Linux::clock_gettime(clockid, &tp);
4390   assert(rc == 0, "clock_gettime is expected to return 0 code");
4391 
4392   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4393 }
4394 
4395 /////
4396 // glibc on Linux platform uses non-documented flag
4397 // to indicate, that some special sort of signal
4398 // trampoline is used.
4399 // We will never set this flag, and we should
4400 // ignore this flag in our diagnostic
4401 #ifdef SIGNIFICANT_SIGNAL_MASK
4402   #undef SIGNIFICANT_SIGNAL_MASK
4403 #endif
4404 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4405 
4406 static const char* get_signal_handler_name(address handler,
4407                                            char* buf, int buflen) {
4408   int offset;
4409   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4410   if (found) {
4411     // skip directory names
4412     const char *p1, *p2;
4413     p1 = buf;
4414     size_t len = strlen(os::file_separator());
4415     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4416     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4417   } else {
4418     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4419   }
4420   return buf;
4421 }
4422 
4423 static void print_signal_handler(outputStream* st, int sig,
4424                                  char* buf, size_t buflen) {
4425   struct sigaction sa;
4426 
4427   sigaction(sig, NULL, &sa);
4428 
4429   // See comment for SIGNIFICANT_SIGNAL_MASK define
4430   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4431 
4432   st->print("%s: ", os::exception_name(sig, buf, buflen));
4433 
4434   address handler = (sa.sa_flags & SA_SIGINFO)
4435     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4436     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4437 
4438   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4439     st->print("SIG_DFL");
4440   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4441     st->print("SIG_IGN");
4442   } else {
4443     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4444   }
4445 
4446   st->print(", sa_mask[0]=");
4447   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4448 
4449   address rh = VMError::get_resetted_sighandler(sig);
4450   // May be, handler was resetted by VMError?
4451   if (rh != NULL) {
4452     handler = rh;
4453     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4454   }
4455 
4456   st->print(", sa_flags=");
4457   os::Posix::print_sa_flags(st, sa.sa_flags);
4458 
4459   // Check: is it our handler?
4460   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4461       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4462     // It is our signal handler
4463     // check for flags, reset system-used one!
4464     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4465       st->print(
4466                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4467                 os::Linux::get_our_sigflags(sig));
4468     }
4469   }
4470   st->cr();
4471 }
4472 
4473 
4474 #define DO_SIGNAL_CHECK(sig)                      \
4475   do {                                            \
4476     if (!sigismember(&check_signal_done, sig)) {  \
4477       os::Linux::check_signal_handler(sig);       \
4478     }                                             \
4479   } while (0)
4480 
4481 // This method is a periodic task to check for misbehaving JNI applications
4482 // under CheckJNI, we can add any periodic checks here
4483 
4484 void os::run_periodic_checks() {
4485   if (check_signals == false) return;
4486 
4487   // SEGV and BUS if overridden could potentially prevent
4488   // generation of hs*.log in the event of a crash, debugging
4489   // such a case can be very challenging, so we absolutely
4490   // check the following for a good measure:
4491   DO_SIGNAL_CHECK(SIGSEGV);
4492   DO_SIGNAL_CHECK(SIGILL);
4493   DO_SIGNAL_CHECK(SIGFPE);
4494   DO_SIGNAL_CHECK(SIGBUS);
4495   DO_SIGNAL_CHECK(SIGPIPE);
4496   DO_SIGNAL_CHECK(SIGXFSZ);
4497 #if defined(PPC64)
4498   DO_SIGNAL_CHECK(SIGTRAP);
4499 #endif
4500 
4501   // ReduceSignalUsage allows the user to override these handlers
4502   // see comments at the very top and jvm_solaris.h
4503   if (!ReduceSignalUsage) {
4504     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4505     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4506     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4507     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4508   }
4509 
4510   DO_SIGNAL_CHECK(SR_signum);
4511   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4512 }
4513 
4514 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4515 
4516 static os_sigaction_t os_sigaction = NULL;
4517 
4518 void os::Linux::check_signal_handler(int sig) {
4519   char buf[O_BUFLEN];
4520   address jvmHandler = NULL;
4521 
4522 
4523   struct sigaction act;
4524   if (os_sigaction == NULL) {
4525     // only trust the default sigaction, in case it has been interposed
4526     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4527     if (os_sigaction == NULL) return;
4528   }
4529 
4530   os_sigaction(sig, (struct sigaction*)NULL, &act);
4531 
4532 
4533   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4534 
4535   address thisHandler = (act.sa_flags & SA_SIGINFO)
4536     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4537     : CAST_FROM_FN_PTR(address, act.sa_handler);
4538 
4539 
4540   switch (sig) {
4541   case SIGSEGV:
4542   case SIGBUS:
4543   case SIGFPE:
4544   case SIGPIPE:
4545   case SIGILL:
4546   case SIGXFSZ:
4547     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4548     break;
4549 
4550   case SHUTDOWN1_SIGNAL:
4551   case SHUTDOWN2_SIGNAL:
4552   case SHUTDOWN3_SIGNAL:
4553   case BREAK_SIGNAL:
4554     jvmHandler = (address)user_handler();
4555     break;
4556 
4557   case INTERRUPT_SIGNAL:
4558     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4559     break;
4560 
4561   default:
4562     if (sig == SR_signum) {
4563       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4564     } else {
4565       return;
4566     }
4567     break;
4568   }
4569 
4570   if (thisHandler != jvmHandler) {
4571     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4572     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4573     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4574     // No need to check this sig any longer
4575     sigaddset(&check_signal_done, sig);
4576     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4577     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4578       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4579                     exception_name(sig, buf, O_BUFLEN));
4580     }
4581   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4582     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4583     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4584     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4585     // No need to check this sig any longer
4586     sigaddset(&check_signal_done, sig);
4587   }
4588 
4589   // Dump all the signal
4590   if (sigismember(&check_signal_done, sig)) {
4591     print_signal_handlers(tty, buf, O_BUFLEN);
4592   }
4593 }
4594 
4595 extern void report_error(char* file_name, int line_no, char* title,
4596                          char* format, ...);
4597 
4598 extern bool signal_name(int signo, char* buf, size_t len);
4599 
4600 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4601   if (0 < exception_code && exception_code <= SIGRTMAX) {
4602     // signal
4603     if (!signal_name(exception_code, buf, size)) {
4604       jio_snprintf(buf, size, "SIG%d", exception_code);
4605     }
4606     return buf;
4607   } else {
4608     return NULL;
4609   }
4610 }
4611 
4612 // this is called _before_ the most of global arguments have been parsed
4613 void os::init(void) {
4614   char dummy;   // used to get a guess on initial stack address
4615 //  first_hrtime = gethrtime();
4616 
4617   // With LinuxThreads the JavaMain thread pid (primordial thread)
4618   // is different than the pid of the java launcher thread.
4619   // So, on Linux, the launcher thread pid is passed to the VM
4620   // via the sun.java.launcher.pid property.
4621   // Use this property instead of getpid() if it was correctly passed.
4622   // See bug 6351349.
4623   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4624 
4625   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4626 
4627   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4628 
4629   init_random(1234567);
4630 
4631   ThreadCritical::initialize();
4632 
4633   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4634   if (Linux::page_size() == -1) {
4635     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4636                   strerror(errno)));
4637   }
4638   init_page_sizes((size_t) Linux::page_size());
4639 
4640   Linux::initialize_system_info();
4641 
4642   // main_thread points to the aboriginal thread
4643   Linux::_main_thread = pthread_self();
4644 
4645   Linux::clock_init();
4646   initial_time_count = javaTimeNanos();
4647 
4648   // pthread_condattr initialization for monotonic clock
4649   int status;
4650   pthread_condattr_t* _condattr = os::Linux::condAttr();
4651   if ((status = pthread_condattr_init(_condattr)) != 0) {
4652     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4653   }
4654   // Only set the clock if CLOCK_MONOTONIC is available
4655   if (os::supports_monotonic_clock()) {
4656     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4657       if (status == EINVAL) {
4658         warning("Unable to use monotonic clock with relative timed-waits" \
4659                 " - changes to the time-of-day clock may have adverse affects");
4660       } else {
4661         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4662       }
4663     }
4664   }
4665   // else it defaults to CLOCK_REALTIME
4666 
4667   pthread_mutex_init(&dl_mutex, NULL);
4668 
4669   // If the pagesize of the VM is greater than 8K determine the appropriate
4670   // number of initial guard pages.  The user can change this with the
4671   // command line arguments, if needed.
4672   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4673     StackYellowPages = 1;
4674     StackRedPages = 1;
4675     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4676   }
4677 
4678   // retrieve entry point for pthread_setname_np
4679   Linux::_pthread_setname_np =
4680     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4681 
4682 }
4683 
4684 // To install functions for atexit system call
4685 extern "C" {
4686   static void perfMemory_exit_helper() {
4687     perfMemory_exit();
4688   }
4689 }
4690 
4691 // this is called _after_ the global arguments have been parsed
4692 jint os::init_2(void) {
4693   Linux::fast_thread_clock_init();
4694 
4695   // Allocate a single page and mark it as readable for safepoint polling
4696   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4697   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4698 
4699   os::set_polling_page(polling_page);
4700 
4701 #ifndef PRODUCT
4702   if (Verbose && PrintMiscellaneous) {
4703     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4704                (intptr_t)polling_page);
4705   }
4706 #endif
4707 
4708   if (!UseMembar) {
4709     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4710     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4711     os::set_memory_serialize_page(mem_serialize_page);
4712 
4713 #ifndef PRODUCT
4714     if (Verbose && PrintMiscellaneous) {
4715       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n",
4716                  (intptr_t)mem_serialize_page);
4717     }
4718 #endif
4719   }
4720 
4721   // initialize suspend/resume support - must do this before signal_sets_init()
4722   if (SR_initialize() != 0) {
4723     perror("SR_initialize failed");
4724     return JNI_ERR;
4725   }
4726 
4727   Linux::signal_sets_init();
4728   Linux::install_signal_handlers();
4729 
4730   // Check minimum allowable stack size for thread creation and to initialize
4731   // the java system classes, including StackOverflowError - depends on page
4732   // size.  Add a page for compiler2 recursion in main thread.
4733   // Add in 2*BytesPerWord times page size to account for VM stack during
4734   // class initialization depending on 32 or 64 bit VM.
4735   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4736                                       (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4737                                       (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4738 
4739   size_t threadStackSizeInBytes = ThreadStackSize * K;
4740   if (threadStackSizeInBytes != 0 &&
4741       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4742     tty->print_cr("\nThe stack size specified is too small, "
4743                   "Specify at least %dk",
4744                   os::Linux::min_stack_allowed/ K);
4745     return JNI_ERR;
4746   }
4747 
4748   // Make the stack size a multiple of the page size so that
4749   // the yellow/red zones can be guarded.
4750   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4751                                                 vm_page_size()));
4752 
4753   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4754 
4755 #if defined(IA32)
4756   workaround_expand_exec_shield_cs_limit();
4757 #endif
4758 
4759   Linux::libpthread_init();
4760   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4761     tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4762                   Linux::glibc_version(), Linux::libpthread_version(),
4763                   Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4764   }
4765 
4766   if (UseNUMA) {
4767     if (!Linux::libnuma_init()) {
4768       UseNUMA = false;
4769     } else {
4770       if ((Linux::numa_max_node() < 1)) {
4771         // There's only one node(they start from 0), disable NUMA.
4772         UseNUMA = false;
4773       }
4774     }
4775     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4776     // we can make the adaptive lgrp chunk resizing work. If the user specified
4777     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4778     // disable adaptive resizing.
4779     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4780       if (FLAG_IS_DEFAULT(UseNUMA)) {
4781         UseNUMA = false;
4782       } else {
4783         if (FLAG_IS_DEFAULT(UseLargePages) &&
4784             FLAG_IS_DEFAULT(UseSHM) &&
4785             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4786           UseLargePages = false;
4787         } else {
4788           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4789           UseAdaptiveSizePolicy = false;
4790           UseAdaptiveNUMAChunkSizing = false;
4791         }
4792       }
4793     }
4794     if (!UseNUMA && ForceNUMA) {
4795       UseNUMA = true;
4796     }
4797   }
4798 
4799   if (MaxFDLimit) {
4800     // set the number of file descriptors to max. print out error
4801     // if getrlimit/setrlimit fails but continue regardless.
4802     struct rlimit nbr_files;
4803     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4804     if (status != 0) {
4805       if (PrintMiscellaneous && (Verbose || WizardMode)) {
4806         perror("os::init_2 getrlimit failed");
4807       }
4808     } else {
4809       nbr_files.rlim_cur = nbr_files.rlim_max;
4810       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4811       if (status != 0) {
4812         if (PrintMiscellaneous && (Verbose || WizardMode)) {
4813           perror("os::init_2 setrlimit failed");
4814         }
4815       }
4816     }
4817   }
4818 
4819   // Initialize lock used to serialize thread creation (see os::create_thread)
4820   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4821 
4822   // at-exit methods are called in the reverse order of their registration.
4823   // atexit functions are called on return from main or as a result of a
4824   // call to exit(3C). There can be only 32 of these functions registered
4825   // and atexit() does not set errno.
4826 
4827   if (PerfAllowAtExitRegistration) {
4828     // only register atexit functions if PerfAllowAtExitRegistration is set.
4829     // atexit functions can be delayed until process exit time, which
4830     // can be problematic for embedded VM situations. Embedded VMs should
4831     // call DestroyJavaVM() to assure that VM resources are released.
4832 
4833     // note: perfMemory_exit_helper atexit function may be removed in
4834     // the future if the appropriate cleanup code can be added to the
4835     // VM_Exit VMOperation's doit method.
4836     if (atexit(perfMemory_exit_helper) != 0) {
4837       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4838     }
4839   }
4840 
4841   // initialize thread priority policy
4842   prio_init();
4843 
4844   return JNI_OK;
4845 }
4846 
4847 // Mark the polling page as unreadable
4848 void os::make_polling_page_unreadable(void) {
4849   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4850     fatal("Could not disable polling page");
4851   }
4852 }
4853 
4854 // Mark the polling page as readable
4855 void os::make_polling_page_readable(void) {
4856   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4857     fatal("Could not enable polling page");
4858   }
4859 }
4860 
4861 int os::active_processor_count() {
4862   // Linux doesn't yet have a (official) notion of processor sets,
4863   // so just return the number of online processors.
4864   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4865   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4866   return online_cpus;
4867 }
4868 
4869 void os::set_native_thread_name(const char *name) {
4870   if (Linux::_pthread_setname_np) {
4871     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4872     snprintf(buf, sizeof(buf), "%s", name);
4873     buf[sizeof(buf) - 1] = '\0';
4874     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4875     // ERANGE should not happen; all other errors should just be ignored.
4876     assert(rc != ERANGE, "pthread_setname_np failed");
4877   }
4878 }
4879 
4880 bool os::distribute_processes(uint length, uint* distribution) {
4881   // Not yet implemented.
4882   return false;
4883 }
4884 
4885 bool os::bind_to_processor(uint processor_id) {
4886   // Not yet implemented.
4887   return false;
4888 }
4889 
4890 ///
4891 
4892 void os::SuspendedThreadTask::internal_do_task() {
4893   if (do_suspend(_thread->osthread())) {
4894     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4895     do_task(context);
4896     do_resume(_thread->osthread());
4897   }
4898 }
4899 
4900 class PcFetcher : public os::SuspendedThreadTask {
4901  public:
4902   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4903   ExtendedPC result();
4904  protected:
4905   void do_task(const os::SuspendedThreadTaskContext& context);
4906  private:
4907   ExtendedPC _epc;
4908 };
4909 
4910 ExtendedPC PcFetcher::result() {
4911   guarantee(is_done(), "task is not done yet.");
4912   return _epc;
4913 }
4914 
4915 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4916   Thread* thread = context.thread();
4917   OSThread* osthread = thread->osthread();
4918   if (osthread->ucontext() != NULL) {
4919     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4920   } else {
4921     // NULL context is unexpected, double-check this is the VMThread
4922     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4923   }
4924 }
4925 
4926 // Suspends the target using the signal mechanism and then grabs the PC before
4927 // resuming the target. Used by the flat-profiler only
4928 ExtendedPC os::get_thread_pc(Thread* thread) {
4929   // Make sure that it is called by the watcher for the VMThread
4930   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4931   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4932 
4933   PcFetcher fetcher(thread);
4934   fetcher.run();
4935   return fetcher.result();
4936 }
4937 
4938 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4939                                    pthread_mutex_t *_mutex,
4940                                    const struct timespec *_abstime) {
4941   if (is_NPTL()) {
4942     return pthread_cond_timedwait(_cond, _mutex, _abstime);
4943   } else {
4944     // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4945     // word back to default 64bit precision if condvar is signaled. Java
4946     // wants 53bit precision.  Save and restore current value.
4947     int fpu = get_fpu_control_word();
4948     int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4949     set_fpu_control_word(fpu);
4950     return status;
4951   }
4952 }
4953 
4954 ////////////////////////////////////////////////////////////////////////////////
4955 // debug support
4956 
4957 bool os::find(address addr, outputStream* st) {
4958   Dl_info dlinfo;
4959   memset(&dlinfo, 0, sizeof(dlinfo));
4960   if (dladdr(addr, &dlinfo) != 0) {
4961     st->print(PTR_FORMAT ": ", addr);
4962     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4963       st->print("%s+%#x", dlinfo.dli_sname,
4964                 addr - (intptr_t)dlinfo.dli_saddr);
4965     } else if (dlinfo.dli_fbase != NULL) {
4966       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4967     } else {
4968       st->print("<absolute address>");
4969     }
4970     if (dlinfo.dli_fname != NULL) {
4971       st->print(" in %s", dlinfo.dli_fname);
4972     }
4973     if (dlinfo.dli_fbase != NULL) {
4974       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4975     }
4976     st->cr();
4977 
4978     if (Verbose) {
4979       // decode some bytes around the PC
4980       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4981       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4982       address       lowest = (address) dlinfo.dli_sname;
4983       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4984       if (begin < lowest)  begin = lowest;
4985       Dl_info dlinfo2;
4986       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4987           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4988         end = (address) dlinfo2.dli_saddr;
4989       }
4990       Disassembler::decode(begin, end, st);
4991     }
4992     return true;
4993   }
4994   return false;
4995 }
4996 
4997 ////////////////////////////////////////////////////////////////////////////////
4998 // misc
4999 
5000 // This does not do anything on Linux. This is basically a hook for being
5001 // able to use structured exception handling (thread-local exception filters)
5002 // on, e.g., Win32.
5003 void
5004 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5005                          JavaCallArguments* args, Thread* thread) {
5006   f(value, method, args, thread);
5007 }
5008 
5009 void os::print_statistics() {
5010 }
5011 
5012 int os::message_box(const char* title, const char* message) {
5013   int i;
5014   fdStream err(defaultStream::error_fd());
5015   for (i = 0; i < 78; i++) err.print_raw("=");
5016   err.cr();
5017   err.print_raw_cr(title);
5018   for (i = 0; i < 78; i++) err.print_raw("-");
5019   err.cr();
5020   err.print_raw_cr(message);
5021   for (i = 0; i < 78; i++) err.print_raw("=");
5022   err.cr();
5023 
5024   char buf[16];
5025   // Prevent process from exiting upon "read error" without consuming all CPU
5026   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5027 
5028   return buf[0] == 'y' || buf[0] == 'Y';
5029 }
5030 
5031 int os::stat(const char *path, struct stat *sbuf) {
5032   char pathbuf[MAX_PATH];
5033   if (strlen(path) > MAX_PATH - 1) {
5034     errno = ENAMETOOLONG;
5035     return -1;
5036   }
5037   os::native_path(strcpy(pathbuf, path));
5038   return ::stat(pathbuf, sbuf);
5039 }
5040 
5041 bool os::check_heap(bool force) {
5042   return true;
5043 }
5044 
5045 // Is a (classpath) directory empty?
5046 bool os::dir_is_empty(const char* path) {
5047   DIR *dir = NULL;
5048   struct dirent *ptr;
5049 
5050   dir = opendir(path);
5051   if (dir == NULL) return true;
5052 
5053   // Scan the directory
5054   bool result = true;
5055   char buf[sizeof(struct dirent) + MAX_PATH];
5056   while (result && (ptr = ::readdir(dir)) != NULL) {
5057     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5058       result = false;
5059     }
5060   }
5061   closedir(dir);
5062   return result;
5063 }
5064 
5065 // This code originates from JDK's sysOpen and open64_w
5066 // from src/solaris/hpi/src/system_md.c
5067 
5068 int os::open(const char *path, int oflag, int mode) {
5069   if (strlen(path) > MAX_PATH - 1) {
5070     errno = ENAMETOOLONG;
5071     return -1;
5072   }
5073 
5074   // All file descriptors that are opened in the Java process and not
5075   // specifically destined for a subprocess should have the close-on-exec
5076   // flag set.  If we don't set it, then careless 3rd party native code
5077   // might fork and exec without closing all appropriate file descriptors
5078   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5079   // turn might:
5080   //
5081   // - cause end-of-file to fail to be detected on some file
5082   //   descriptors, resulting in mysterious hangs, or
5083   //
5084   // - might cause an fopen in the subprocess to fail on a system
5085   //   suffering from bug 1085341.
5086   //
5087   // (Yes, the default setting of the close-on-exec flag is a Unix
5088   // design flaw)
5089   //
5090   // See:
5091   // 1085341: 32-bit stdio routines should support file descriptors >255
5092   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5093   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5094   //
5095   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5096   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5097   // because it saves a system call and removes a small window where the flag
5098   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5099   // and we fall back to using FD_CLOEXEC (see below).
5100 #ifdef O_CLOEXEC
5101   oflag |= O_CLOEXEC;
5102 #endif
5103 
5104   int fd = ::open64(path, oflag, mode);
5105   if (fd == -1) return -1;
5106 
5107   //If the open succeeded, the file might still be a directory
5108   {
5109     struct stat64 buf64;
5110     int ret = ::fstat64(fd, &buf64);
5111     int st_mode = buf64.st_mode;
5112 
5113     if (ret != -1) {
5114       if ((st_mode & S_IFMT) == S_IFDIR) {
5115         errno = EISDIR;
5116         ::close(fd);
5117         return -1;
5118       }
5119     } else {
5120       ::close(fd);
5121       return -1;
5122     }
5123   }
5124 
5125 #ifdef FD_CLOEXEC
5126   // Validate that the use of the O_CLOEXEC flag on open above worked.
5127   // With recent kernels, we will perform this check exactly once.
5128   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5129   if (!O_CLOEXEC_is_known_to_work) {
5130     int flags = ::fcntl(fd, F_GETFD);
5131     if (flags != -1) {
5132       if ((flags & FD_CLOEXEC) != 0)
5133         O_CLOEXEC_is_known_to_work = 1;
5134       else
5135         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5136     }
5137   }
5138 #endif
5139 
5140   return fd;
5141 }
5142 
5143 
5144 // create binary file, rewriting existing file if required
5145 int os::create_binary_file(const char* path, bool rewrite_existing) {
5146   int oflags = O_WRONLY | O_CREAT;
5147   if (!rewrite_existing) {
5148     oflags |= O_EXCL;
5149   }
5150   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5151 }
5152 
5153 // return current position of file pointer
5154 jlong os::current_file_offset(int fd) {
5155   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5156 }
5157 
5158 // move file pointer to the specified offset
5159 jlong os::seek_to_file_offset(int fd, jlong offset) {
5160   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5161 }
5162 
5163 // This code originates from JDK's sysAvailable
5164 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5165 
5166 int os::available(int fd, jlong *bytes) {
5167   jlong cur, end;
5168   int mode;
5169   struct stat64 buf64;
5170 
5171   if (::fstat64(fd, &buf64) >= 0) {
5172     mode = buf64.st_mode;
5173     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5174       // XXX: is the following call interruptible? If so, this might
5175       // need to go through the INTERRUPT_IO() wrapper as for other
5176       // blocking, interruptible calls in this file.
5177       int n;
5178       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5179         *bytes = n;
5180         return 1;
5181       }
5182     }
5183   }
5184   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5185     return 0;
5186   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5187     return 0;
5188   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5189     return 0;
5190   }
5191   *bytes = end - cur;
5192   return 1;
5193 }
5194 
5195 // Map a block of memory.
5196 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5197                         char *addr, size_t bytes, bool read_only,
5198                         bool allow_exec) {
5199   int prot;
5200   int flags = MAP_PRIVATE;
5201 
5202   if (read_only) {
5203     prot = PROT_READ;
5204   } else {
5205     prot = PROT_READ | PROT_WRITE;
5206   }
5207 
5208   if (allow_exec) {
5209     prot |= PROT_EXEC;
5210   }
5211 
5212   if (addr != NULL) {
5213     flags |= MAP_FIXED;
5214   }
5215 
5216   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5217                                      fd, file_offset);
5218   if (mapped_address == MAP_FAILED) {
5219     return NULL;
5220   }
5221   return mapped_address;
5222 }
5223 
5224 
5225 // Remap a block of memory.
5226 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5227                           char *addr, size_t bytes, bool read_only,
5228                           bool allow_exec) {
5229   // same as map_memory() on this OS
5230   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5231                         allow_exec);
5232 }
5233 
5234 
5235 // Unmap a block of memory.
5236 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5237   return munmap(addr, bytes) == 0;
5238 }
5239 
5240 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5241 
5242 static clockid_t thread_cpu_clockid(Thread* thread) {
5243   pthread_t tid = thread->osthread()->pthread_id();
5244   clockid_t clockid;
5245 
5246   // Get thread clockid
5247   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5248   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5249   return clockid;
5250 }
5251 
5252 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5253 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5254 // of a thread.
5255 //
5256 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5257 // the fast estimate available on the platform.
5258 
5259 jlong os::current_thread_cpu_time() {
5260   if (os::Linux::supports_fast_thread_cpu_time()) {
5261     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5262   } else {
5263     // return user + sys since the cost is the same
5264     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5265   }
5266 }
5267 
5268 jlong os::thread_cpu_time(Thread* thread) {
5269   // consistent with what current_thread_cpu_time() returns
5270   if (os::Linux::supports_fast_thread_cpu_time()) {
5271     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5272   } else {
5273     return slow_thread_cpu_time(thread, true /* user + sys */);
5274   }
5275 }
5276 
5277 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5278   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5279     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5280   } else {
5281     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5282   }
5283 }
5284 
5285 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5286   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5287     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5288   } else {
5289     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5290   }
5291 }
5292 
5293 //  -1 on error.
5294 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5295   pid_t  tid = thread->osthread()->thread_id();
5296   char *s;
5297   char stat[2048];
5298   int statlen;
5299   char proc_name[64];
5300   int count;
5301   long sys_time, user_time;
5302   char cdummy;
5303   int idummy;
5304   long ldummy;
5305   FILE *fp;
5306 
5307   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5308   fp = fopen(proc_name, "r");
5309   if (fp == NULL) return -1;
5310   statlen = fread(stat, 1, 2047, fp);
5311   stat[statlen] = '\0';
5312   fclose(fp);
5313 
5314   // Skip pid and the command string. Note that we could be dealing with
5315   // weird command names, e.g. user could decide to rename java launcher
5316   // to "java 1.4.2 :)", then the stat file would look like
5317   //                1234 (java 1.4.2 :)) R ... ...
5318   // We don't really need to know the command string, just find the last
5319   // occurrence of ")" and then start parsing from there. See bug 4726580.
5320   s = strrchr(stat, ')');
5321   if (s == NULL) return -1;
5322 
5323   // Skip blank chars
5324   do { s++; } while (s && isspace(*s));
5325 
5326   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5327                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5328                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5329                  &user_time, &sys_time);
5330   if (count != 13) return -1;
5331   if (user_sys_cpu_time) {
5332     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5333   } else {
5334     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5335   }
5336 }
5337 
5338 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5339   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5340   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5341   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5342   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5343 }
5344 
5345 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5346   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5347   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5348   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5349   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5350 }
5351 
5352 bool os::is_thread_cpu_time_supported() {
5353   return true;
5354 }
5355 
5356 // System loadavg support.  Returns -1 if load average cannot be obtained.
5357 // Linux doesn't yet have a (official) notion of processor sets,
5358 // so just return the system wide load average.
5359 int os::loadavg(double loadavg[], int nelem) {
5360   return ::getloadavg(loadavg, nelem);
5361 }
5362 
5363 void os::pause() {
5364   char filename[MAX_PATH];
5365   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5366     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5367   } else {
5368     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5369   }
5370 
5371   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5372   if (fd != -1) {
5373     struct stat buf;
5374     ::close(fd);
5375     while (::stat(filename, &buf) == 0) {
5376       (void)::poll(NULL, 0, 100);
5377     }
5378   } else {
5379     jio_fprintf(stderr,
5380                 "Could not open pause file '%s', continuing immediately.\n", filename);
5381   }
5382 }
5383 
5384 
5385 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5386 // comment paragraphs are worth repeating here:
5387 //
5388 // Assumption:
5389 //    Only one parker can exist on an event, which is why we allocate
5390 //    them per-thread. Multiple unparkers can coexist.
5391 //
5392 // _Event serves as a restricted-range semaphore.
5393 //   -1 : thread is blocked, i.e. there is a waiter
5394 //    0 : neutral: thread is running or ready,
5395 //        could have been signaled after a wait started
5396 //    1 : signaled - thread is running or ready
5397 //
5398 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5399 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5400 // For specifics regarding the bug see GLIBC BUGID 261237 :
5401 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5402 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5403 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5404 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5405 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5406 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5407 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5408 // of libpthread avoids the problem, but isn't practical.
5409 //
5410 // Possible remedies:
5411 //
5412 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5413 //      This is palliative and probabilistic, however.  If the thread is preempted
5414 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5415 //      than the minimum period may have passed, and the abstime may be stale (in the
5416 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5417 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5418 //
5419 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5420 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5421 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5422 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5423 //      thread.
5424 //
5425 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5426 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5427 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5428 //      This also works well.  In fact it avoids kernel-level scalability impediments
5429 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5430 //      timers in a graceful fashion.
5431 //
5432 // 4.   When the abstime value is in the past it appears that control returns
5433 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5434 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5435 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5436 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5437 //      It may be possible to avoid reinitialization by checking the return
5438 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5439 //      condvar we must establish the invariant that cond_signal() is only called
5440 //      within critical sections protected by the adjunct mutex.  This prevents
5441 //      cond_signal() from "seeing" a condvar that's in the midst of being
5442 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5443 //      desirable signal-after-unlock optimization that avoids futile context switching.
5444 //
5445 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5446 //      structure when a condvar is used or initialized.  cond_destroy()  would
5447 //      release the helper structure.  Our reinitialize-after-timedwait fix
5448 //      put excessive stress on malloc/free and locks protecting the c-heap.
5449 //
5450 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5451 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5452 // and only enabling the work-around for vulnerable environments.
5453 
5454 // utility to compute the abstime argument to timedwait:
5455 // millis is the relative timeout time
5456 // abstime will be the absolute timeout time
5457 // TODO: replace compute_abstime() with unpackTime()
5458 
5459 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5460   if (millis < 0)  millis = 0;
5461 
5462   jlong seconds = millis / 1000;
5463   millis %= 1000;
5464   if (seconds > 50000000) { // see man cond_timedwait(3T)
5465     seconds = 50000000;
5466   }
5467 
5468   if (os::supports_monotonic_clock()) {
5469     struct timespec now;
5470     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5471     assert_status(status == 0, status, "clock_gettime");
5472     abstime->tv_sec = now.tv_sec  + seconds;
5473     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5474     if (nanos >= NANOSECS_PER_SEC) {
5475       abstime->tv_sec += 1;
5476       nanos -= NANOSECS_PER_SEC;
5477     }
5478     abstime->tv_nsec = nanos;
5479   } else {
5480     struct timeval now;
5481     int status = gettimeofday(&now, NULL);
5482     assert(status == 0, "gettimeofday");
5483     abstime->tv_sec = now.tv_sec  + seconds;
5484     long usec = now.tv_usec + millis * 1000;
5485     if (usec >= 1000000) {
5486       abstime->tv_sec += 1;
5487       usec -= 1000000;
5488     }
5489     abstime->tv_nsec = usec * 1000;
5490   }
5491   return abstime;
5492 }
5493 
5494 void os::PlatformEvent::park() {       // AKA "down()"
5495   // Transitions for _Event:
5496   //   -1 => -1 : illegal
5497   //    1 =>  0 : pass - return immediately
5498   //    0 => -1 : block; then set _Event to 0 before returning
5499 
5500   // Invariant: Only the thread associated with the Event/PlatformEvent
5501   // may call park().
5502   // TODO: assert that _Assoc != NULL or _Assoc == Self
5503   assert(_nParked == 0, "invariant");
5504 
5505   int v;
5506   for (;;) {
5507     v = _Event;
5508     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5509   }
5510   guarantee(v >= 0, "invariant");
5511   if (v == 0) {
5512     // Do this the hard way by blocking ...
5513     int status = pthread_mutex_lock(_mutex);
5514     assert_status(status == 0, status, "mutex_lock");
5515     guarantee(_nParked == 0, "invariant");
5516     ++_nParked;
5517     while (_Event < 0) {
5518       status = pthread_cond_wait(_cond, _mutex);
5519       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5520       // Treat this the same as if the wait was interrupted
5521       if (status == ETIME) { status = EINTR; }
5522       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5523     }
5524     --_nParked;
5525 
5526     _Event = 0;
5527     status = pthread_mutex_unlock(_mutex);
5528     assert_status(status == 0, status, "mutex_unlock");
5529     // Paranoia to ensure our locked and lock-free paths interact
5530     // correctly with each other.
5531     OrderAccess::fence();
5532   }
5533   guarantee(_Event >= 0, "invariant");
5534 }
5535 
5536 int os::PlatformEvent::park(jlong millis) {
5537   // Transitions for _Event:
5538   //   -1 => -1 : illegal
5539   //    1 =>  0 : pass - return immediately
5540   //    0 => -1 : block; then set _Event to 0 before returning
5541 
5542   guarantee(_nParked == 0, "invariant");
5543 
5544   int v;
5545   for (;;) {
5546     v = _Event;
5547     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5548   }
5549   guarantee(v >= 0, "invariant");
5550   if (v != 0) return OS_OK;
5551 
5552   // We do this the hard way, by blocking the thread.
5553   // Consider enforcing a minimum timeout value.
5554   struct timespec abst;
5555   compute_abstime(&abst, millis);
5556 
5557   int ret = OS_TIMEOUT;
5558   int status = pthread_mutex_lock(_mutex);
5559   assert_status(status == 0, status, "mutex_lock");
5560   guarantee(_nParked == 0, "invariant");
5561   ++_nParked;
5562 
5563   // Object.wait(timo) will return because of
5564   // (a) notification
5565   // (b) timeout
5566   // (c) thread.interrupt
5567   //
5568   // Thread.interrupt and object.notify{All} both call Event::set.
5569   // That is, we treat thread.interrupt as a special case of notification.
5570   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5571   // We assume all ETIME returns are valid.
5572   //
5573   // TODO: properly differentiate simultaneous notify+interrupt.
5574   // In that case, we should propagate the notify to another waiter.
5575 
5576   while (_Event < 0) {
5577     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5578     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5579       pthread_cond_destroy(_cond);
5580       pthread_cond_init(_cond, os::Linux::condAttr());
5581     }
5582     assert_status(status == 0 || status == EINTR ||
5583                   status == ETIME || status == ETIMEDOUT,
5584                   status, "cond_timedwait");
5585     if (!FilterSpuriousWakeups) break;                 // previous semantics
5586     if (status == ETIME || status == ETIMEDOUT) break;
5587     // We consume and ignore EINTR and spurious wakeups.
5588   }
5589   --_nParked;
5590   if (_Event >= 0) {
5591     ret = OS_OK;
5592   }
5593   _Event = 0;
5594   status = pthread_mutex_unlock(_mutex);
5595   assert_status(status == 0, status, "mutex_unlock");
5596   assert(_nParked == 0, "invariant");
5597   // Paranoia to ensure our locked and lock-free paths interact
5598   // correctly with each other.
5599   OrderAccess::fence();
5600   return ret;
5601 }
5602 
5603 void os::PlatformEvent::unpark() {
5604   // Transitions for _Event:
5605   //    0 => 1 : just return
5606   //    1 => 1 : just return
5607   //   -1 => either 0 or 1; must signal target thread
5608   //         That is, we can safely transition _Event from -1 to either
5609   //         0 or 1.
5610   // See also: "Semaphores in Plan 9" by Mullender & Cox
5611   //
5612   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5613   // that it will take two back-to-back park() calls for the owning
5614   // thread to block. This has the benefit of forcing a spurious return
5615   // from the first park() call after an unpark() call which will help
5616   // shake out uses of park() and unpark() without condition variables.
5617 
5618   if (Atomic::xchg(1, &_Event) >= 0) return;
5619 
5620   // Wait for the thread associated with the event to vacate
5621   int status = pthread_mutex_lock(_mutex);
5622   assert_status(status == 0, status, "mutex_lock");
5623   int AnyWaiters = _nParked;
5624   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5625   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5626     AnyWaiters = 0;
5627     pthread_cond_signal(_cond);
5628   }
5629   status = pthread_mutex_unlock(_mutex);
5630   assert_status(status == 0, status, "mutex_unlock");
5631   if (AnyWaiters != 0) {
5632     // Note that we signal() *after* dropping the lock for "immortal" Events.
5633     // This is safe and avoids a common class of  futile wakeups.  In rare
5634     // circumstances this can cause a thread to return prematurely from
5635     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5636     // will simply re-test the condition and re-park itself.
5637     // This provides particular benefit if the underlying platform does not
5638     // provide wait morphing.
5639     status = pthread_cond_signal(_cond);
5640     assert_status(status == 0, status, "cond_signal");
5641   }
5642 }
5643 
5644 
5645 // JSR166
5646 // -------------------------------------------------------
5647 
5648 // The solaris and linux implementations of park/unpark are fairly
5649 // conservative for now, but can be improved. They currently use a
5650 // mutex/condvar pair, plus a a count.
5651 // Park decrements count if > 0, else does a condvar wait.  Unpark
5652 // sets count to 1 and signals condvar.  Only one thread ever waits
5653 // on the condvar. Contention seen when trying to park implies that someone
5654 // is unparking you, so don't wait. And spurious returns are fine, so there
5655 // is no need to track notifications.
5656 
5657 // This code is common to linux and solaris and will be moved to a
5658 // common place in dolphin.
5659 //
5660 // The passed in time value is either a relative time in nanoseconds
5661 // or an absolute time in milliseconds. Either way it has to be unpacked
5662 // into suitable seconds and nanoseconds components and stored in the
5663 // given timespec structure.
5664 // Given time is a 64-bit value and the time_t used in the timespec is only
5665 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5666 // overflow if times way in the future are given. Further on Solaris versions
5667 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5668 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5669 // As it will be 28 years before "now + 100000000" will overflow we can
5670 // ignore overflow and just impose a hard-limit on seconds using the value
5671 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5672 // years from "now".
5673 
5674 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5675   assert(time > 0, "convertTime");
5676   time_t max_secs = 0;
5677 
5678   if (!os::supports_monotonic_clock() || isAbsolute) {
5679     struct timeval now;
5680     int status = gettimeofday(&now, NULL);
5681     assert(status == 0, "gettimeofday");
5682 
5683     max_secs = now.tv_sec + MAX_SECS;
5684 
5685     if (isAbsolute) {
5686       jlong secs = time / 1000;
5687       if (secs > max_secs) {
5688         absTime->tv_sec = max_secs;
5689       } else {
5690         absTime->tv_sec = secs;
5691       }
5692       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5693     } else {
5694       jlong secs = time / NANOSECS_PER_SEC;
5695       if (secs >= MAX_SECS) {
5696         absTime->tv_sec = max_secs;
5697         absTime->tv_nsec = 0;
5698       } else {
5699         absTime->tv_sec = now.tv_sec + secs;
5700         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5701         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5702           absTime->tv_nsec -= NANOSECS_PER_SEC;
5703           ++absTime->tv_sec; // note: this must be <= max_secs
5704         }
5705       }
5706     }
5707   } else {
5708     // must be relative using monotonic clock
5709     struct timespec now;
5710     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5711     assert_status(status == 0, status, "clock_gettime");
5712     max_secs = now.tv_sec + MAX_SECS;
5713     jlong secs = time / NANOSECS_PER_SEC;
5714     if (secs >= MAX_SECS) {
5715       absTime->tv_sec = max_secs;
5716       absTime->tv_nsec = 0;
5717     } else {
5718       absTime->tv_sec = now.tv_sec + secs;
5719       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5720       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5721         absTime->tv_nsec -= NANOSECS_PER_SEC;
5722         ++absTime->tv_sec; // note: this must be <= max_secs
5723       }
5724     }
5725   }
5726   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5727   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5728   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5729   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5730 }
5731 
5732 void Parker::park(bool isAbsolute, jlong time) {
5733   // Ideally we'd do something useful while spinning, such
5734   // as calling unpackTime().
5735 
5736   // Optional fast-path check:
5737   // Return immediately if a permit is available.
5738   // We depend on Atomic::xchg() having full barrier semantics
5739   // since we are doing a lock-free update to _counter.
5740   if (Atomic::xchg(0, &_counter) > 0) return;
5741 
5742   Thread* thread = Thread::current();
5743   assert(thread->is_Java_thread(), "Must be JavaThread");
5744   JavaThread *jt = (JavaThread *)thread;
5745 
5746   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5747   // Check interrupt before trying to wait
5748   if (Thread::is_interrupted(thread, false)) {
5749     return;
5750   }
5751 
5752   // Next, demultiplex/decode time arguments
5753   timespec absTime;
5754   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5755     return;
5756   }
5757   if (time > 0) {
5758     unpackTime(&absTime, isAbsolute, time);
5759   }
5760 
5761 
5762   // Enter safepoint region
5763   // Beware of deadlocks such as 6317397.
5764   // The per-thread Parker:: mutex is a classic leaf-lock.
5765   // In particular a thread must never block on the Threads_lock while
5766   // holding the Parker:: mutex.  If safepoints are pending both the
5767   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5768   ThreadBlockInVM tbivm(jt);
5769 
5770   // Don't wait if cannot get lock since interference arises from
5771   // unblocking.  Also. check interrupt before trying wait
5772   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5773     return;
5774   }
5775 
5776   int status;
5777   if (_counter > 0)  { // no wait needed
5778     _counter = 0;
5779     status = pthread_mutex_unlock(_mutex);
5780     assert(status == 0, "invariant");
5781     // Paranoia to ensure our locked and lock-free paths interact
5782     // correctly with each other and Java-level accesses.
5783     OrderAccess::fence();
5784     return;
5785   }
5786 
5787 #ifdef ASSERT
5788   // Don't catch signals while blocked; let the running threads have the signals.
5789   // (This allows a debugger to break into the running thread.)
5790   sigset_t oldsigs;
5791   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5792   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5793 #endif
5794 
5795   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5796   jt->set_suspend_equivalent();
5797   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5798 
5799   assert(_cur_index == -1, "invariant");
5800   if (time == 0) {
5801     _cur_index = REL_INDEX; // arbitrary choice when not timed
5802     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5803   } else {
5804     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5805     status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5806     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5807       pthread_cond_destroy(&_cond[_cur_index]);
5808       pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5809     }
5810   }
5811   _cur_index = -1;
5812   assert_status(status == 0 || status == EINTR ||
5813                 status == ETIME || status == ETIMEDOUT,
5814                 status, "cond_timedwait");
5815 
5816 #ifdef ASSERT
5817   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5818 #endif
5819 
5820   _counter = 0;
5821   status = pthread_mutex_unlock(_mutex);
5822   assert_status(status == 0, status, "invariant");
5823   // Paranoia to ensure our locked and lock-free paths interact
5824   // correctly with each other and Java-level accesses.
5825   OrderAccess::fence();
5826 
5827   // If externally suspended while waiting, re-suspend
5828   if (jt->handle_special_suspend_equivalent_condition()) {
5829     jt->java_suspend_self();
5830   }
5831 }
5832 
5833 void Parker::unpark() {
5834   int status = pthread_mutex_lock(_mutex);
5835   assert(status == 0, "invariant");
5836   const int s = _counter;
5837   _counter = 1;
5838   if (s < 1) {
5839     // thread might be parked
5840     if (_cur_index != -1) {
5841       // thread is definitely parked
5842       if (WorkAroundNPTLTimedWaitHang) {
5843         status = pthread_cond_signal(&_cond[_cur_index]);
5844         assert(status == 0, "invariant");
5845         status = pthread_mutex_unlock(_mutex);
5846         assert(status == 0, "invariant");
5847       } else {
5848         status = pthread_mutex_unlock(_mutex);
5849         assert(status == 0, "invariant");
5850         status = pthread_cond_signal(&_cond[_cur_index]);
5851         assert(status == 0, "invariant");
5852       }
5853     } else {
5854       pthread_mutex_unlock(_mutex);
5855       assert(status == 0, "invariant");
5856     }
5857   } else {
5858     pthread_mutex_unlock(_mutex);
5859     assert(status == 0, "invariant");
5860   }
5861 }
5862 
5863 
5864 extern char** environ;
5865 
5866 #ifndef __NR_fork
5867   #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5868 #endif
5869 
5870 #ifndef __NR_execve
5871   #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5872 #endif
5873 
5874 // Run the specified command in a separate process. Return its exit value,
5875 // or -1 on failure (e.g. can't fork a new process).
5876 // Unlike system(), this function can be called from signal handler. It
5877 // doesn't block SIGINT et al.
5878 int os::fork_and_exec(char* cmd) {
5879   const char * argv[4] = {"sh", "-c", cmd, NULL};
5880 
5881   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5882   // pthread_atfork handlers and reset pthread library. All we need is a
5883   // separate process to execve. Make a direct syscall to fork process.
5884   // On IA64 there's no fork syscall, we have to use fork() and hope for
5885   // the best...
5886   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5887   IA64_ONLY(fork();)
5888 
5889   if (pid < 0) {
5890     // fork failed
5891     return -1;
5892 
5893   } else if (pid == 0) {
5894     // child process
5895 
5896     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5897     // first to kill every thread on the thread list. Because this list is
5898     // not reset by fork() (see notes above), execve() will instead kill
5899     // every thread in the parent process. We know this is the only thread
5900     // in the new process, so make a system call directly.
5901     // IA64 should use normal execve() from glibc to match the glibc fork()
5902     // above.
5903     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5904     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5905 
5906     // execve failed
5907     _exit(-1);
5908 
5909   } else  {
5910     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5911     // care about the actual exit code, for now.
5912 
5913     int status;
5914 
5915     // Wait for the child process to exit.  This returns immediately if
5916     // the child has already exited. */
5917     while (waitpid(pid, &status, 0) < 0) {
5918       switch (errno) {
5919       case ECHILD: return 0;
5920       case EINTR: break;
5921       default: return -1;
5922       }
5923     }
5924 
5925     if (WIFEXITED(status)) {
5926       // The child exited normally; get its exit code.
5927       return WEXITSTATUS(status);
5928     } else if (WIFSIGNALED(status)) {
5929       // The child exited because of a signal
5930       // The best value to return is 0x80 + signal number,
5931       // because that is what all Unix shells do, and because
5932       // it allows callers to distinguish between process exit and
5933       // process death by signal.
5934       return 0x80 + WTERMSIG(status);
5935     } else {
5936       // Unknown exit code; pass it through
5937       return status;
5938     }
5939   }
5940 }
5941 
5942 // is_headless_jre()
5943 //
5944 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5945 // in order to report if we are running in a headless jre
5946 //
5947 // Since JDK8 xawt/libmawt.so was moved into the same directory
5948 // as libawt.so, and renamed libawt_xawt.so
5949 //
5950 bool os::is_headless_jre() {
5951   struct stat statbuf;
5952   char buf[MAXPATHLEN];
5953   char libmawtpath[MAXPATHLEN];
5954   const char *xawtstr  = "/xawt/libmawt.so";
5955   const char *new_xawtstr = "/libawt_xawt.so";
5956   char *p;
5957 
5958   // Get path to libjvm.so
5959   os::jvm_path(buf, sizeof(buf));
5960 
5961   // Get rid of libjvm.so
5962   p = strrchr(buf, '/');
5963   if (p == NULL) {
5964     return false;
5965   } else {
5966     *p = '\0';
5967   }
5968 
5969   // Get rid of client or server
5970   p = strrchr(buf, '/');
5971   if (p == NULL) {
5972     return false;
5973   } else {
5974     *p = '\0';
5975   }
5976 
5977   // check xawt/libmawt.so
5978   strcpy(libmawtpath, buf);
5979   strcat(libmawtpath, xawtstr);
5980   if (::stat(libmawtpath, &statbuf) == 0) return false;
5981 
5982   // check libawt_xawt.so
5983   strcpy(libmawtpath, buf);
5984   strcat(libmawtpath, new_xawtstr);
5985   if (::stat(libmawtpath, &statbuf) == 0) return false;
5986 
5987   return true;
5988 }
5989 
5990 // Get the default path to the core file
5991 // Returns the length of the string
5992 int os::get_core_path(char* buffer, size_t bufferSize) {
5993   const char* p = get_current_directory(buffer, bufferSize);
5994 
5995   if (p == NULL) {
5996     assert(p != NULL, "failed to get current directory");
5997     return 0;
5998   }
5999 
6000   return strlen(buffer);
6001 }
6002 
6003 /////////////// Unit tests ///////////////
6004 
6005 #ifndef PRODUCT
6006 
6007 #define test_log(...)              \
6008   do {                             \
6009     if (VerboseInternalVMTests) {  \
6010       tty->print_cr(__VA_ARGS__);  \
6011       tty->flush();                \
6012     }                              \
6013   } while (false)
6014 
6015 class TestReserveMemorySpecial : AllStatic {
6016  public:
6017   static void small_page_write(void* addr, size_t size) {
6018     size_t page_size = os::vm_page_size();
6019 
6020     char* end = (char*)addr + size;
6021     for (char* p = (char*)addr; p < end; p += page_size) {
6022       *p = 1;
6023     }
6024   }
6025 
6026   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6027     if (!UseHugeTLBFS) {
6028       return;
6029     }
6030 
6031     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6032 
6033     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6034 
6035     if (addr != NULL) {
6036       small_page_write(addr, size);
6037 
6038       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6039     }
6040   }
6041 
6042   static void test_reserve_memory_special_huge_tlbfs_only() {
6043     if (!UseHugeTLBFS) {
6044       return;
6045     }
6046 
6047     size_t lp = os::large_page_size();
6048 
6049     for (size_t size = lp; size <= lp * 10; size += lp) {
6050       test_reserve_memory_special_huge_tlbfs_only(size);
6051     }
6052   }
6053 
6054   static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6055     if (!UseHugeTLBFS) {
6056       return;
6057     }
6058 
6059     test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6060              size, alignment);
6061 
6062     assert(size >= os::large_page_size(), "Incorrect input to test");
6063 
6064     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6065 
6066     if (addr != NULL) {
6067       small_page_write(addr, size);
6068 
6069       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6070     }
6071   }
6072 
6073   static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6074     size_t lp = os::large_page_size();
6075     size_t ag = os::vm_allocation_granularity();
6076 
6077     for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6078       test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6079     }
6080   }
6081 
6082   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6083     size_t lp = os::large_page_size();
6084     size_t ag = os::vm_allocation_granularity();
6085 
6086     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6087     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6088     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6089     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6090     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6091     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6092     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6093     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6094     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6095   }
6096 
6097   static void test_reserve_memory_special_huge_tlbfs() {
6098     if (!UseHugeTLBFS) {
6099       return;
6100     }
6101 
6102     test_reserve_memory_special_huge_tlbfs_only();
6103     test_reserve_memory_special_huge_tlbfs_mixed();
6104   }
6105 
6106   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6107     if (!UseSHM) {
6108       return;
6109     }
6110 
6111     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6112 
6113     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6114 
6115     if (addr != NULL) {
6116       assert(is_ptr_aligned(addr, alignment), "Check");
6117       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6118 
6119       small_page_write(addr, size);
6120 
6121       os::Linux::release_memory_special_shm(addr, size);
6122     }
6123   }
6124 
6125   static void test_reserve_memory_special_shm() {
6126     size_t lp = os::large_page_size();
6127     size_t ag = os::vm_allocation_granularity();
6128 
6129     for (size_t size = ag; size < lp * 3; size += ag) {
6130       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6131         test_reserve_memory_special_shm(size, alignment);
6132       }
6133     }
6134   }
6135 
6136   static void test() {
6137     test_reserve_memory_special_huge_tlbfs();
6138     test_reserve_memory_special_shm();
6139   }
6140 };
6141 
6142 void TestReserveMemorySpecial_test() {
6143   TestReserveMemorySpecial::test();
6144 }
6145 
6146 #endif
--- EOF ---