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_share_linux.hpp"
  40 #include "prims/jniFastGetField.hpp"
  41 #include "prims/jvm.h"
  42 #include "prims/jvm_misc.hpp"
  43 #include "runtime/arguments.hpp"
  44 #include "runtime/extendedPC.hpp"
  45 #include "runtime/globals.hpp"
  46 #include "runtime/interfaceSupport.hpp"
  47 #include "runtime/init.hpp"
  48 #include "runtime/java.hpp"
  49 #include "runtime/javaCalls.hpp"
  50 #include "runtime/mutexLocker.hpp"
  51 #include "runtime/objectMonitor.hpp"
  52 #include "runtime/osThread.hpp"
  53 #include "runtime/perfMemory.hpp"
  54 #include "runtime/sharedRuntime.hpp"
  55 #include "runtime/statSampler.hpp"
  56 #include "runtime/stubRoutines.hpp"
  57 #include "runtime/thread.inline.hpp"
  58 #include "runtime/threadCritical.hpp"
  59 #include "runtime/timer.hpp"
  60 #include "services/attachListener.hpp"
  61 #include "services/memTracker.hpp"
  62 #include "services/runtimeService.hpp"
  63 #include "utilities/decoder.hpp"
  64 #include "utilities/defaultStream.hpp"
  65 #include "utilities/events.hpp"
  66 #include "utilities/elfFile.hpp"
  67 #include "utilities/growableArray.hpp"
  68 #include "utilities/vmError.hpp"
  69 
  70 // put OS-includes here
  71 # include <sys/types.h>
  72 # include <sys/mman.h>
  73 # include <sys/stat.h>
  74 # include <sys/select.h>
  75 # include <pthread.h>
  76 # include <signal.h>
  77 # include <errno.h>
  78 # include <dlfcn.h>
  79 # include <stdio.h>
  80 # include <unistd.h>
  81 # include <sys/resource.h>
  82 # include <pthread.h>
  83 # include <sys/stat.h>
  84 # include <sys/time.h>
  85 # include <sys/times.h>
  86 # include <sys/utsname.h>
  87 # include <sys/socket.h>
  88 # include <sys/wait.h>
  89 # include <pwd.h>
  90 # include <poll.h>
  91 # include <semaphore.h>
  92 # include <fcntl.h>
  93 # include <string.h>
  94 # include <syscall.h>
  95 # include <sys/sysinfo.h>
  96 # include <gnu/libc-version.h>
  97 # include <sys/ipc.h>
  98 # include <sys/shm.h>
  99 # include <link.h>
 100 # include <stdint.h>
 101 # include <inttypes.h>
 102 # include <sys/ioctl.h>
 103 
 104 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 105 // getrusage() is prepared to handle the associated failure.
 106 #ifndef RUSAGE_THREAD
 107 #define RUSAGE_THREAD   (1)               /* only the calling thread */
 108 #endif
 109 
 110 #define MAX_PATH    (2 * K)
 111 
 112 #define MAX_SECS 100000000
 113 
 114 // for timer info max values which include all bits
 115 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 116 
 117 #define LARGEPAGES_BIT (1 << 6)
 118 ////////////////////////////////////////////////////////////////////////////////
 119 // global variables
 120 julong os::Linux::_physical_memory = 0;
 121 
 122 address   os::Linux::_initial_thread_stack_bottom = NULL;
 123 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 124 
 125 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 126 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 127 Mutex* os::Linux::_createThread_lock = NULL;
 128 pthread_t os::Linux::_main_thread;
 129 int os::Linux::_page_size = -1;
 130 const int os::Linux::_vm_default_page_size = (8 * K);
 131 bool os::Linux::_is_floating_stack = false;
 132 bool os::Linux::_is_NPTL = false;
 133 bool os::Linux::_supports_fast_thread_cpu_time = false;
 134 const char * os::Linux::_glibc_version = NULL;
 135 const char * os::Linux::_libpthread_version = NULL;
 136 pthread_condattr_t os::Linux::_condattr[1];
 137 
 138 static jlong initial_time_count=0;
 139 
 140 static int clock_tics_per_sec = 100;
 141 
 142 // For diagnostics to print a message once. see run_periodic_checks
 143 static sigset_t check_signal_done;
 144 static bool check_signals = true;
 145 
 146 static pid_t _initial_pid = 0;
 147 
 148 /* Signal number used to suspend/resume a thread */
 149 
 150 /* do not use any signal number less than SIGSEGV, see 4355769 */
 151 static int SR_signum = SIGUSR2;
 152 sigset_t SR_sigset;
 153 
 154 /* Used to protect dlsym() calls */
 155 static pthread_mutex_t dl_mutex;
 156 
 157 // Declarations
 158 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 159 
 160 #ifdef JAVASE_EMBEDDED
 161 class MemNotifyThread: public Thread {
 162   friend class VMStructs;
 163  public:
 164   virtual void run();
 165 
 166  private:
 167   static MemNotifyThread* _memnotify_thread;
 168   int _fd;
 169 
 170  public:
 171 
 172   // Constructor
 173   MemNotifyThread(int fd);
 174 
 175   // Tester
 176   bool is_memnotify_thread() const { return true; }
 177 
 178   // Printing
 179   char* name() const { return (char*)"Linux MemNotify Thread"; }
 180 
 181   // Returns the single instance of the MemNotifyThread
 182   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 183 
 184   // Create and start the single instance of MemNotifyThread
 185   static void start();
 186 };
 187 #endif // JAVASE_EMBEDDED
 188 
 189 // utility functions
 190 
 191 static int SR_initialize();
 192 
 193 julong os::available_memory() {
 194   return Linux::available_memory();
 195 }
 196 
 197 julong os::Linux::available_memory() {
 198   // values in struct sysinfo are "unsigned long"
 199   struct sysinfo si;
 200   sysinfo(&si);
 201 
 202   return (julong)si.freeram * si.mem_unit;
 203 }
 204 
 205 julong os::physical_memory() {
 206   return Linux::physical_memory();
 207 }
 208 
 209 ////////////////////////////////////////////////////////////////////////////////
 210 // environment support
 211 
 212 bool os::getenv(const char* name, char* buf, int len) {
 213   const char* val = ::getenv(name);
 214   if (val != NULL && strlen(val) < (size_t)len) {
 215     strcpy(buf, val);
 216     return true;
 217   }
 218   if (len > 0) buf[0] = 0;  // return a null string
 219   return false;
 220 }
 221 
 222 
 223 // Return true if user is running as root.
 224 
 225 bool os::have_special_privileges() {
 226   static bool init = false;
 227   static bool privileges = false;
 228   if (!init) {
 229     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 230     init = true;
 231   }
 232   return privileges;
 233 }
 234 
 235 
 236 #ifndef SYS_gettid
 237 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 238 #ifdef __ia64__
 239 #define SYS_gettid 1105
 240 #elif __i386__
 241 #define SYS_gettid 224
 242 #elif __amd64__
 243 #define SYS_gettid 186
 244 #elif __sparc__
 245 #define SYS_gettid 143
 246 #else
 247 #error define gettid for the arch
 248 #endif
 249 #endif
 250 
 251 // Cpu architecture string
 252 #if   defined(ZERO)
 253 static char cpu_arch[] = ZERO_LIBARCH;
 254 #elif defined(IA64)
 255 static char cpu_arch[] = "ia64";
 256 #elif defined(IA32)
 257 static char cpu_arch[] = "i386";
 258 #elif defined(AMD64)
 259 static char cpu_arch[] = "amd64";
 260 #elif defined(ARM)
 261 static char cpu_arch[] = "arm";
 262 #elif defined(PPC32)
 263 static char cpu_arch[] = "ppc";
 264 #elif defined(PPC64)
 265 static char cpu_arch[] = "ppc64";
 266 #elif defined(SPARC)
 267 #  ifdef _LP64
 268 static char cpu_arch[] = "sparcv9";
 269 #  else
 270 static char cpu_arch[] = "sparc";
 271 #  endif
 272 #else
 273 #error Add appropriate cpu_arch setting
 274 #endif
 275 
 276 
 277 // pid_t gettid()
 278 //
 279 // Returns the kernel thread id of the currently running thread. Kernel
 280 // thread id is used to access /proc.
 281 //
 282 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 283 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 284 //
 285 pid_t os::Linux::gettid() {
 286   int rslt = syscall(SYS_gettid);
 287   if (rslt == -1) {
 288      // old kernel, no NPTL support
 289      return getpid();
 290   } else {
 291      return (pid_t)rslt;
 292   }
 293 }
 294 
 295 // Most versions of linux have a bug where the number of processors are
 296 // determined by looking at the /proc file system.  In a chroot environment,
 297 // the system call returns 1.  This causes the VM to act as if it is
 298 // a single processor and elide locking (see is_MP() call).
 299 static bool unsafe_chroot_detected = false;
 300 static const char *unstable_chroot_error = "/proc file system not found.\n"
 301                      "Java may be unstable running multithreaded in a chroot "
 302                      "environment on Linux when /proc filesystem is not mounted.";
 303 
 304 void os::Linux::initialize_system_info() {
 305   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 306   if (processor_count() == 1) {
 307     pid_t pid = os::Linux::gettid();
 308     char fname[32];
 309     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 310     FILE *fp = fopen(fname, "r");
 311     if (fp == NULL) {
 312       unsafe_chroot_detected = true;
 313     } else {
 314       fclose(fp);
 315     }
 316   }
 317   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 318   assert(processor_count() > 0, "linux error");
 319 }
 320 
 321 void os::init_system_properties_values() {
 322 //  char arch[12];
 323 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 324 
 325   // The next steps are taken in the product version:
 326   //
 327   // Obtain the JAVA_HOME value from the location of libjvm.so.
 328   // This library should be located at:
 329   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 330   //
 331   // If "/jre/lib/" appears at the right place in the path, then we
 332   // assume libjvm.so is installed in a JDK and we use this path.
 333   //
 334   // Otherwise exit with message: "Could not create the Java virtual machine."
 335   //
 336   // The following extra steps are taken in the debugging version:
 337   //
 338   // If "/jre/lib/" does NOT appear at the right place in the path
 339   // instead of exit check for $JAVA_HOME environment variable.
 340   //
 341   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 342   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 343   // it looks like libjvm.so is installed there
 344   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 345   //
 346   // Otherwise exit.
 347   //
 348   // Important note: if the location of libjvm.so changes this
 349   // code needs to be changed accordingly.
 350 
 351   // The next few definitions allow the code to be verbatim:
 352 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
 353 #define getenv(n) ::getenv(n)
 354 
 355 /*
 356  * See ld(1):
 357  *      The linker uses the following search paths to locate required
 358  *      shared libraries:
 359  *        1: ...
 360  *        ...
 361  *        7: The default directories, normally /lib and /usr/lib.
 362  */
 363 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 364 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 365 #else
 366 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 367 #endif
 368 
 369 #define EXTENSIONS_DIR  "/lib/ext"
 370 #define ENDORSED_DIR    "/lib/endorsed"
 371 #define REG_DIR         "/usr/java/packages"
 372 
 373   {
 374     /* sysclasspath, java_home, dll_dir */
 375     {
 376         char *home_path;
 377         char *dll_path;
 378         char *pslash;
 379         char buf[MAXPATHLEN];
 380         os::jvm_path(buf, sizeof(buf));
 381 
 382         // Found the full path to libjvm.so.
 383         // Now cut the path to <java_home>/jre if we can.
 384         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 385         pslash = strrchr(buf, '/');
 386         if (pslash != NULL)
 387             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 388         dll_path = malloc(strlen(buf) + 1);
 389         if (dll_path == NULL)
 390             return;
 391         strcpy(dll_path, buf);
 392         Arguments::set_dll_dir(dll_path);
 393 
 394         if (pslash != NULL) {
 395             pslash = strrchr(buf, '/');
 396             if (pslash != NULL) {
 397                 *pslash = '\0';       /* get rid of /<arch> */
 398                 pslash = strrchr(buf, '/');
 399                 if (pslash != NULL)
 400                     *pslash = '\0';   /* get rid of /lib */
 401             }
 402         }
 403 
 404         home_path = malloc(strlen(buf) + 1);
 405         if (home_path == NULL)
 406             return;
 407         strcpy(home_path, buf);
 408         Arguments::set_java_home(home_path);
 409 
 410         if (!set_boot_path('/', ':'))
 411             return;
 412     }
 413 
 414     /*
 415      * Where to look for native libraries
 416      *
 417      * Note: Due to a legacy implementation, most of the library path
 418      * is set in the launcher.  This was to accomodate linking restrictions
 419      * on legacy Linux implementations (which are no longer supported).
 420      * Eventually, all the library path setting will be done here.
 421      *
 422      * However, to prevent the proliferation of improperly built native
 423      * libraries, the new path component /usr/java/packages is added here.
 424      * Eventually, all the library path setting will be done here.
 425      */
 426     {
 427         char *ld_library_path;
 428 
 429         /*
 430          * Construct the invariant part of ld_library_path. Note that the
 431          * space for the colon and the trailing null are provided by the
 432          * nulls included by the sizeof operator (so actually we allocate
 433          * a byte more than necessary).
 434          */
 435         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 436             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 437         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 438 
 439         /*
 440          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 441          * should always exist (until the legacy problem cited above is
 442          * addressed).
 443          */
 444         char *v = getenv("LD_LIBRARY_PATH");
 445         if (v != NULL) {
 446             char *t = ld_library_path;
 447             /* That's +1 for the colon and +1 for the trailing '\0' */
 448             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 449             sprintf(ld_library_path, "%s:%s", v, t);
 450         }
 451         Arguments::set_library_path(ld_library_path);
 452     }
 453 
 454     /*
 455      * Extensions directories.
 456      *
 457      * Note that the space for the colon and the trailing null are provided
 458      * by the nulls included by the sizeof operator (so actually one byte more
 459      * than necessary is allocated).
 460      */
 461     {
 462         char *buf = malloc(strlen(Arguments::get_java_home()) +
 463             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 464         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 465             Arguments::get_java_home());
 466         Arguments::set_ext_dirs(buf);
 467     }
 468 
 469     /* Endorsed standards default directory. */
 470     {
 471         char * buf;
 472         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 473         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 474         Arguments::set_endorsed_dirs(buf);
 475     }
 476   }
 477 
 478 #undef malloc
 479 #undef getenv
 480 #undef EXTENSIONS_DIR
 481 #undef ENDORSED_DIR
 482 
 483   // Done
 484   return;
 485 }
 486 
 487 ////////////////////////////////////////////////////////////////////////////////
 488 // breakpoint support
 489 
 490 void os::breakpoint() {
 491   BREAKPOINT;
 492 }
 493 
 494 extern "C" void breakpoint() {
 495   // use debugger to set breakpoint here
 496 }
 497 
 498 ////////////////////////////////////////////////////////////////////////////////
 499 // signal support
 500 
 501 debug_only(static bool signal_sets_initialized = false);
 502 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 503 
 504 bool os::Linux::is_sig_ignored(int sig) {
 505       struct sigaction oact;
 506       sigaction(sig, (struct sigaction*)NULL, &oact);
 507       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 508                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 509       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 510            return true;
 511       else
 512            return false;
 513 }
 514 
 515 void os::Linux::signal_sets_init() {
 516   // Should also have an assertion stating we are still single-threaded.
 517   assert(!signal_sets_initialized, "Already initialized");
 518   // Fill in signals that are necessarily unblocked for all threads in
 519   // the VM. Currently, we unblock the following signals:
 520   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 521   //                         by -Xrs (=ReduceSignalUsage));
 522   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 523   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 524   // the dispositions or masks wrt these signals.
 525   // Programs embedding the VM that want to use the above signals for their
 526   // own purposes must, at this time, use the "-Xrs" option to prevent
 527   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 528   // (See bug 4345157, and other related bugs).
 529   // In reality, though, unblocking these signals is really a nop, since
 530   // these signals are not blocked by default.
 531   sigemptyset(&unblocked_sigs);
 532   sigemptyset(&allowdebug_blocked_sigs);
 533   sigaddset(&unblocked_sigs, SIGILL);
 534   sigaddset(&unblocked_sigs, SIGSEGV);
 535   sigaddset(&unblocked_sigs, SIGBUS);
 536   sigaddset(&unblocked_sigs, SIGFPE);
 537 #if defined(PPC64)
 538   sigaddset(&unblocked_sigs, SIGTRAP);
 539 #endif
 540   sigaddset(&unblocked_sigs, SR_signum);
 541 
 542   if (!ReduceSignalUsage) {
 543    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 544       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 545       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 546    }
 547    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 548       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 549       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 550    }
 551    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 552       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 553       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 554    }
 555   }
 556   // Fill in signals that are blocked by all but the VM thread.
 557   sigemptyset(&vm_sigs);
 558   if (!ReduceSignalUsage)
 559     sigaddset(&vm_sigs, BREAK_SIGNAL);
 560   debug_only(signal_sets_initialized = true);
 561 
 562 }
 563 
 564 // These are signals that are unblocked while a thread is running Java.
 565 // (For some reason, they get blocked by default.)
 566 sigset_t* os::Linux::unblocked_signals() {
 567   assert(signal_sets_initialized, "Not initialized");
 568   return &unblocked_sigs;
 569 }
 570 
 571 // These are the signals that are blocked while a (non-VM) thread is
 572 // running Java. Only the VM thread handles these signals.
 573 sigset_t* os::Linux::vm_signals() {
 574   assert(signal_sets_initialized, "Not initialized");
 575   return &vm_sigs;
 576 }
 577 
 578 // These are signals that are blocked during cond_wait to allow debugger in
 579 sigset_t* os::Linux::allowdebug_blocked_signals() {
 580   assert(signal_sets_initialized, "Not initialized");
 581   return &allowdebug_blocked_sigs;
 582 }
 583 
 584 void os::Linux::hotspot_sigmask(Thread* thread) {
 585 
 586   //Save caller's signal mask before setting VM signal mask
 587   sigset_t caller_sigmask;
 588   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 589 
 590   OSThread* osthread = thread->osthread();
 591   osthread->set_caller_sigmask(caller_sigmask);
 592 
 593   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 594 
 595   if (!ReduceSignalUsage) {
 596     if (thread->is_VM_thread()) {
 597       // Only the VM thread handles BREAK_SIGNAL ...
 598       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 599     } else {
 600       // ... all other threads block BREAK_SIGNAL
 601       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 602     }
 603   }
 604 }
 605 
 606 //////////////////////////////////////////////////////////////////////////////
 607 // detecting pthread library
 608 
 609 void os::Linux::libpthread_init() {
 610   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 611   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 612   // generic name for earlier versions.
 613   // Define macros here so we can build HotSpot on old systems.
 614 # ifndef _CS_GNU_LIBC_VERSION
 615 # define _CS_GNU_LIBC_VERSION 2
 616 # endif
 617 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 618 # define _CS_GNU_LIBPTHREAD_VERSION 3
 619 # endif
 620 
 621   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 622   if (n > 0) {
 623      char *str = (char *)malloc(n, mtInternal);
 624      confstr(_CS_GNU_LIBC_VERSION, str, n);
 625      os::Linux::set_glibc_version(str);
 626   } else {
 627      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 628      static char _gnu_libc_version[32];
 629      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 630               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 631      os::Linux::set_glibc_version(_gnu_libc_version);
 632   }
 633 
 634   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 635   if (n > 0) {
 636      char *str = (char *)malloc(n, mtInternal);
 637      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 638      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 639      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 640      // is the case. LinuxThreads has a hard limit on max number of threads.
 641      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 642      // On the other hand, NPTL does not have such a limit, sysconf()
 643      // will return -1 and errno is not changed. Check if it is really NPTL.
 644      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 645          strstr(str, "NPTL") &&
 646          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 647        free(str);
 648        os::Linux::set_libpthread_version("linuxthreads");
 649      } else {
 650        os::Linux::set_libpthread_version(str);
 651      }
 652   } else {
 653     // glibc before 2.3.2 only has LinuxThreads.
 654     os::Linux::set_libpthread_version("linuxthreads");
 655   }
 656 
 657   if (strstr(libpthread_version(), "NPTL")) {
 658      os::Linux::set_is_NPTL();
 659   } else {
 660      os::Linux::set_is_LinuxThreads();
 661   }
 662 
 663   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 664   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 665   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 666      os::Linux::set_is_floating_stack();
 667   }
 668 }
 669 
 670 /////////////////////////////////////////////////////////////////////////////
 671 // thread stack
 672 
 673 // Force Linux kernel to expand current thread stack. If "bottom" is close
 674 // to the stack guard, caller should block all signals.
 675 //
 676 // MAP_GROWSDOWN:
 677 //   A special mmap() flag that is used to implement thread stacks. It tells
 678 //   kernel that the memory region should extend downwards when needed. This
 679 //   allows early versions of LinuxThreads to only mmap the first few pages
 680 //   when creating a new thread. Linux kernel will automatically expand thread
 681 //   stack as needed (on page faults).
 682 //
 683 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 684 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 685 //   region, it's hard to tell if the fault is due to a legitimate stack
 686 //   access or because of reading/writing non-exist memory (e.g. buffer
 687 //   overrun). As a rule, if the fault happens below current stack pointer,
 688 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 689 //   application (see Linux kernel fault.c).
 690 //
 691 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 692 //   stack overflow detection.
 693 //
 694 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 695 //   not use this flag. However, the stack of initial thread is not created
 696 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 697 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 698 //   and then attach the thread to JVM.
 699 //
 700 // To get around the problem and allow stack banging on Linux, we need to
 701 // manually expand thread stack after receiving the SIGSEGV.
 702 //
 703 // There are two ways to expand thread stack to address "bottom", we used
 704 // both of them in JVM before 1.5:
 705 //   1. adjust stack pointer first so that it is below "bottom", and then
 706 //      touch "bottom"
 707 //   2. mmap() the page in question
 708 //
 709 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 710 // if current sp is already near the lower end of page 101, and we need to
 711 // call mmap() to map page 100, it is possible that part of the mmap() frame
 712 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 713 // That will destroy the mmap() frame and cause VM to crash.
 714 //
 715 // The following code works by adjusting sp first, then accessing the "bottom"
 716 // page to force a page fault. Linux kernel will then automatically expand the
 717 // stack mapping.
 718 //
 719 // _expand_stack_to() assumes its frame size is less than page size, which
 720 // should always be true if the function is not inlined.
 721 
 722 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 723 #define NOINLINE
 724 #else
 725 #define NOINLINE __attribute__ ((noinline))
 726 #endif
 727 
 728 static void _expand_stack_to(address bottom) NOINLINE;
 729 
 730 static void _expand_stack_to(address bottom) {
 731   address sp;
 732   size_t size;
 733   volatile char *p;
 734 
 735   // Adjust bottom to point to the largest address within the same page, it
 736   // gives us a one-page buffer if alloca() allocates slightly more memory.
 737   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 738   bottom += os::Linux::page_size() - 1;
 739 
 740   // sp might be slightly above current stack pointer; if that's the case, we
 741   // will alloca() a little more space than necessary, which is OK. Don't use
 742   // os::current_stack_pointer(), as its result can be slightly below current
 743   // stack pointer, causing us to not alloca enough to reach "bottom".
 744   sp = (address)&sp;
 745 
 746   if (sp > bottom) {
 747     size = sp - bottom;
 748     p = (volatile char *)alloca(size);
 749     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 750     p[0] = '\0';
 751   }
 752 }
 753 
 754 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 755   assert(t!=NULL, "just checking");
 756   assert(t->osthread()->expanding_stack(), "expand should be set");
 757   assert(t->stack_base() != NULL, "stack_base was not initialized");
 758 
 759   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 760     sigset_t mask_all, old_sigset;
 761     sigfillset(&mask_all);
 762     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 763     _expand_stack_to(addr);
 764     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 765     return true;
 766   }
 767   return false;
 768 }
 769 
 770 //////////////////////////////////////////////////////////////////////////////
 771 // create new thread
 772 
 773 static address highest_vm_reserved_address();
 774 
 775 // check if it's safe to start a new thread
 776 static bool _thread_safety_check(Thread* thread) {
 777   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 778     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 779     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 780     //   allocated (MAP_FIXED) from high address space. Every thread stack
 781     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 782     //   it to other values if they rebuild LinuxThreads).
 783     //
 784     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 785     // the memory region has already been mmap'ed. That means if we have too
 786     // many threads and/or very large heap, eventually thread stack will
 787     // collide with heap.
 788     //
 789     // Here we try to prevent heap/stack collision by comparing current
 790     // stack bottom with the highest address that has been mmap'ed by JVM
 791     // plus a safety margin for memory maps created by native code.
 792     //
 793     // This feature can be disabled by setting ThreadSafetyMargin to 0
 794     //
 795     if (ThreadSafetyMargin > 0) {
 796       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 797 
 798       // not safe if our stack extends below the safety margin
 799       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 800     } else {
 801       return true;
 802     }
 803   } else {
 804     // Floating stack LinuxThreads or NPTL:
 805     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 806     //   there's not enough space left, pthread_create() will fail. If we come
 807     //   here, that means enough space has been reserved for stack.
 808     return true;
 809   }
 810 }
 811 
 812 // Thread start routine for all newly created threads
 813 static void *java_start(Thread *thread) {
 814   // Try to randomize the cache line index of hot stack frames.
 815   // This helps when threads of the same stack traces evict each other's
 816   // cache lines. The threads can be either from the same JVM instance, or
 817   // from different JVM instances. The benefit is especially true for
 818   // processors with hyperthreading technology.
 819   static int counter = 0;
 820   int pid = os::current_process_id();
 821   alloca(((pid ^ counter++) & 7) * 128);
 822 
 823   ThreadLocalStorage::set_thread(thread);
 824 
 825   OSThread* osthread = thread->osthread();
 826   Monitor* sync = osthread->startThread_lock();
 827 
 828   // non floating stack LinuxThreads needs extra check, see above
 829   if (!_thread_safety_check(thread)) {
 830     // notify parent thread
 831     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 832     osthread->set_state(ZOMBIE);
 833     sync->notify_all();
 834     return NULL;
 835   }
 836 
 837   // thread_id is kernel thread id (similar to Solaris LWP id)
 838   osthread->set_thread_id(os::Linux::gettid());
 839 
 840   if (UseNUMA) {
 841     int lgrp_id = os::numa_get_group_id();
 842     if (lgrp_id != -1) {
 843       thread->set_lgrp_id(lgrp_id);
 844     }
 845   }
 846   // initialize signal mask for this thread
 847   os::Linux::hotspot_sigmask(thread);
 848 
 849   // initialize floating point control register
 850   os::Linux::init_thread_fpu_state();
 851 
 852   // handshaking with parent thread
 853   {
 854     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 855 
 856     // notify parent thread
 857     osthread->set_state(INITIALIZED);
 858     sync->notify_all();
 859 
 860     // wait until os::start_thread()
 861     while (osthread->get_state() == INITIALIZED) {
 862       sync->wait(Mutex::_no_safepoint_check_flag);
 863     }
 864   }
 865 
 866   // call one more level start routine
 867   thread->run();
 868 
 869   return 0;
 870 }
 871 
 872 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 873   assert(thread->osthread() == NULL, "caller responsible");
 874 
 875   // Allocate the OSThread object
 876   OSThread* osthread = new OSThread(NULL, NULL);
 877   if (osthread == NULL) {
 878     return false;
 879   }
 880 
 881   // set the correct thread state
 882   osthread->set_thread_type(thr_type);
 883 
 884   // Initial state is ALLOCATED but not INITIALIZED
 885   osthread->set_state(ALLOCATED);
 886 
 887   thread->set_osthread(osthread);
 888 
 889   // init thread attributes
 890   pthread_attr_t attr;
 891   pthread_attr_init(&attr);
 892   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 893 
 894   // stack size
 895   if (os::Linux::supports_variable_stack_size()) {
 896     // calculate stack size if it's not specified by caller
 897     if (stack_size == 0) {
 898       stack_size = os::Linux::default_stack_size(thr_type);
 899 
 900       switch (thr_type) {
 901       case os::java_thread:
 902         // Java threads use ThreadStackSize which default value can be
 903         // changed with the flag -Xss
 904         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 905         stack_size = JavaThread::stack_size_at_create();
 906         break;
 907       case os::compiler_thread:
 908         if (CompilerThreadStackSize > 0) {
 909           stack_size = (size_t)(CompilerThreadStackSize * K);
 910           break;
 911         } // else fall through:
 912           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 913       case os::vm_thread:
 914       case os::pgc_thread:
 915       case os::cgc_thread:
 916       case os::watcher_thread:
 917         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 918         break;
 919       }
 920     }
 921 
 922     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 923     pthread_attr_setstacksize(&attr, stack_size);
 924   } else {
 925     // let pthread_create() pick the default value.
 926   }
 927 
 928   // glibc guard page
 929   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 930 
 931   ThreadState state;
 932 
 933   {
 934     // Serialize thread creation if we are running with fixed stack LinuxThreads
 935     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 936     if (lock) {
 937       os::Linux::createThread_lock()->lock_without_safepoint_check();
 938     }
 939 
 940     pthread_t tid;
 941     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 942 
 943     pthread_attr_destroy(&attr);
 944 
 945     if (ret != 0) {
 946       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 947         perror("pthread_create()");
 948       }
 949       // Need to clean up stuff we've allocated so far
 950       thread->set_osthread(NULL);
 951       delete osthread;
 952       if (lock) os::Linux::createThread_lock()->unlock();
 953       return false;
 954     }
 955 
 956     // Store pthread info into the OSThread
 957     osthread->set_pthread_id(tid);
 958 
 959     // Wait until child thread is either initialized or aborted
 960     {
 961       Monitor* sync_with_child = osthread->startThread_lock();
 962       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 963       while ((state = osthread->get_state()) == ALLOCATED) {
 964         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 965       }
 966     }
 967 
 968     if (lock) {
 969       os::Linux::createThread_lock()->unlock();
 970     }
 971   }
 972 
 973   // Aborted due to thread limit being reached
 974   if (state == ZOMBIE) {
 975       thread->set_osthread(NULL);
 976       delete osthread;
 977       return false;
 978   }
 979 
 980   // The thread is returned suspended (in state INITIALIZED),
 981   // and is started higher up in the call chain
 982   assert(state == INITIALIZED, "race condition");
 983   return true;
 984 }
 985 
 986 /////////////////////////////////////////////////////////////////////////////
 987 // attach existing thread
 988 
 989 // bootstrap the main thread
 990 bool os::create_main_thread(JavaThread* thread) {
 991   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 992   return create_attached_thread(thread);
 993 }
 994 
 995 bool os::create_attached_thread(JavaThread* thread) {
 996 #ifdef ASSERT
 997     thread->verify_not_published();
 998 #endif
 999 
1000   // Allocate the OSThread object
1001   OSThread* osthread = new OSThread(NULL, NULL);
1002 
1003   if (osthread == NULL) {
1004     return false;
1005   }
1006 
1007   // Store pthread info into the OSThread
1008   osthread->set_thread_id(os::Linux::gettid());
1009   osthread->set_pthread_id(::pthread_self());
1010 
1011   // initialize floating point control register
1012   os::Linux::init_thread_fpu_state();
1013 
1014   // Initial thread state is RUNNABLE
1015   osthread->set_state(RUNNABLE);
1016 
1017   thread->set_osthread(osthread);
1018 
1019   if (UseNUMA) {
1020     int lgrp_id = os::numa_get_group_id();
1021     if (lgrp_id != -1) {
1022       thread->set_lgrp_id(lgrp_id);
1023     }
1024   }
1025 
1026   if (os::Linux::is_initial_thread()) {
1027     // If current thread is initial thread, its stack is mapped on demand,
1028     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1029     // the entire stack region to avoid SEGV in stack banging.
1030     // It is also useful to get around the heap-stack-gap problem on SuSE
1031     // kernel (see 4821821 for details). We first expand stack to the top
1032     // of yellow zone, then enable stack yellow zone (order is significant,
1033     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1034     // is no gap between the last two virtual memory regions.
1035 
1036     JavaThread *jt = (JavaThread *)thread;
1037     address addr = jt->stack_yellow_zone_base();
1038     assert(addr != NULL, "initialization problem?");
1039     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1040 
1041     osthread->set_expanding_stack();
1042     os::Linux::manually_expand_stack(jt, addr);
1043     osthread->clear_expanding_stack();
1044   }
1045 
1046   // initialize signal mask for this thread
1047   // and save the caller's signal mask
1048   os::Linux::hotspot_sigmask(thread);
1049 
1050   return true;
1051 }
1052 
1053 void os::pd_start_thread(Thread* thread) {
1054   OSThread * osthread = thread->osthread();
1055   assert(osthread->get_state() != INITIALIZED, "just checking");
1056   Monitor* sync_with_child = osthread->startThread_lock();
1057   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1058   sync_with_child->notify();
1059 }
1060 
1061 // Free Linux resources related to the OSThread
1062 void os::free_thread(OSThread* osthread) {
1063   assert(osthread != NULL, "osthread not set");
1064 
1065   if (Thread::current()->osthread() == osthread) {
1066     // Restore caller's signal mask
1067     sigset_t sigmask = osthread->caller_sigmask();
1068     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1069    }
1070 
1071   delete osthread;
1072 }
1073 
1074 //////////////////////////////////////////////////////////////////////////////
1075 // thread local storage
1076 
1077 int os::allocate_thread_local_storage() {
1078   pthread_key_t key;
1079   int rslt = pthread_key_create(&key, NULL);
1080   assert(rslt == 0, "cannot allocate thread local storage");
1081   return (int)key;
1082 }
1083 
1084 // Note: This is currently not used by VM, as we don't destroy TLS key
1085 // on VM exit.
1086 void os::free_thread_local_storage(int index) {
1087   int rslt = pthread_key_delete((pthread_key_t)index);
1088   assert(rslt == 0, "invalid index");
1089 }
1090 
1091 void os::thread_local_storage_at_put(int index, void* value) {
1092   int rslt = pthread_setspecific((pthread_key_t)index, value);
1093   assert(rslt == 0, "pthread_setspecific failed");
1094 }
1095 
1096 extern "C" Thread* get_thread() {
1097   return ThreadLocalStorage::thread();
1098 }
1099 
1100 //////////////////////////////////////////////////////////////////////////////
1101 // initial thread
1102 
1103 // Check if current thread is the initial thread, similar to Solaris thr_main.
1104 bool os::Linux::is_initial_thread(void) {
1105   char dummy;
1106   // If called before init complete, thread stack bottom will be null.
1107   // Can be called if fatal error occurs before initialization.
1108   if (initial_thread_stack_bottom() == NULL) return false;
1109   assert(initial_thread_stack_bottom() != NULL &&
1110          initial_thread_stack_size()   != 0,
1111          "os::init did not locate initial thread's stack region");
1112   if ((address)&dummy >= initial_thread_stack_bottom() &&
1113       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1114        return true;
1115   else return false;
1116 }
1117 
1118 // Find the virtual memory area that contains addr
1119 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1120   FILE *fp = fopen("/proc/self/maps", "r");
1121   if (fp) {
1122     address low, high;
1123     while (!feof(fp)) {
1124       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1125         if (low <= addr && addr < high) {
1126            if (vma_low)  *vma_low  = low;
1127            if (vma_high) *vma_high = high;
1128            fclose (fp);
1129            return true;
1130         }
1131       }
1132       for (;;) {
1133         int ch = fgetc(fp);
1134         if (ch == EOF || ch == (int)'\n') break;
1135       }
1136     }
1137     fclose(fp);
1138   }
1139   return false;
1140 }
1141 
1142 // Locate initial thread stack. This special handling of initial thread stack
1143 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1144 // bogus value for initial thread.
1145 void os::Linux::capture_initial_stack(size_t max_size) {
1146   // stack size is the easy part, get it from RLIMIT_STACK
1147   size_t stack_size;
1148   struct rlimit rlim;
1149   getrlimit(RLIMIT_STACK, &rlim);
1150   stack_size = rlim.rlim_cur;
1151 
1152   // 6308388: a bug in ld.so will relocate its own .data section to the
1153   //   lower end of primordial stack; reduce ulimit -s value a little bit
1154   //   so we won't install guard page on ld.so's data section.
1155   stack_size -= 2 * page_size();
1156 
1157   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1158   //   7.1, in both cases we will get 2G in return value.
1159   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1160   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1161   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1162   //   in case other parts in glibc still assumes 2M max stack size.
1163   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1164   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1165   if (stack_size > 2 * K * K IA64_ONLY(*2))
1166       stack_size = 2 * K * K IA64_ONLY(*2);
1167   // Try to figure out where the stack base (top) is. This is harder.
1168   //
1169   // When an application is started, glibc saves the initial stack pointer in
1170   // a global variable "__libc_stack_end", which is then used by system
1171   // libraries. __libc_stack_end should be pretty close to stack top. The
1172   // variable is available since the very early days. However, because it is
1173   // a private interface, it could disappear in the future.
1174   //
1175   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1176   // to __libc_stack_end, it is very close to stack top, but isn't the real
1177   // stack top. Note that /proc may not exist if VM is running as a chroot
1178   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1179   // /proc/<pid>/stat could change in the future (though unlikely).
1180   //
1181   // We try __libc_stack_end first. If that doesn't work, look for
1182   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1183   // as a hint, which should work well in most cases.
1184 
1185   uintptr_t stack_start;
1186 
1187   // try __libc_stack_end first
1188   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1189   if (p && *p) {
1190     stack_start = *p;
1191   } else {
1192     // see if we can get the start_stack field from /proc/self/stat
1193     FILE *fp;
1194     int pid;
1195     char state;
1196     int ppid;
1197     int pgrp;
1198     int session;
1199     int nr;
1200     int tpgrp;
1201     unsigned long flags;
1202     unsigned long minflt;
1203     unsigned long cminflt;
1204     unsigned long majflt;
1205     unsigned long cmajflt;
1206     unsigned long utime;
1207     unsigned long stime;
1208     long cutime;
1209     long cstime;
1210     long prio;
1211     long nice;
1212     long junk;
1213     long it_real;
1214     uintptr_t start;
1215     uintptr_t vsize;
1216     intptr_t rss;
1217     uintptr_t rsslim;
1218     uintptr_t scodes;
1219     uintptr_t ecode;
1220     int i;
1221 
1222     // Figure what the primordial thread stack base is. Code is inspired
1223     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1224     // followed by command name surrounded by parentheses, state, etc.
1225     char stat[2048];
1226     int statlen;
1227 
1228     fp = fopen("/proc/self/stat", "r");
1229     if (fp) {
1230       statlen = fread(stat, 1, 2047, fp);
1231       stat[statlen] = '\0';
1232       fclose(fp);
1233 
1234       // Skip pid and the command string. Note that we could be dealing with
1235       // weird command names, e.g. user could decide to rename java launcher
1236       // to "java 1.4.2 :)", then the stat file would look like
1237       //                1234 (java 1.4.2 :)) R ... ...
1238       // We don't really need to know the command string, just find the last
1239       // occurrence of ")" and then start parsing from there. See bug 4726580.
1240       char * s = strrchr(stat, ')');
1241 
1242       i = 0;
1243       if (s) {
1244         // Skip blank chars
1245         do s++; while (isspace(*s));
1246 
1247 #define _UFM UINTX_FORMAT
1248 #define _DFM INTX_FORMAT
1249 
1250         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1251         /*              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 */
1252         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,
1253              &state,          /* 3  %c  */
1254              &ppid,           /* 4  %d  */
1255              &pgrp,           /* 5  %d  */
1256              &session,        /* 6  %d  */
1257              &nr,             /* 7  %d  */
1258              &tpgrp,          /* 8  %d  */
1259              &flags,          /* 9  %lu  */
1260              &minflt,         /* 10 %lu  */
1261              &cminflt,        /* 11 %lu  */
1262              &majflt,         /* 12 %lu  */
1263              &cmajflt,        /* 13 %lu  */
1264              &utime,          /* 14 %lu  */
1265              &stime,          /* 15 %lu  */
1266              &cutime,         /* 16 %ld  */
1267              &cstime,         /* 17 %ld  */
1268              &prio,           /* 18 %ld  */
1269              &nice,           /* 19 %ld  */
1270              &junk,           /* 20 %ld  */
1271              &it_real,        /* 21 %ld  */
1272              &start,          /* 22 UINTX_FORMAT */
1273              &vsize,          /* 23 UINTX_FORMAT */
1274              &rss,            /* 24 INTX_FORMAT  */
1275              &rsslim,         /* 25 UINTX_FORMAT */
1276              &scodes,         /* 26 UINTX_FORMAT */
1277              &ecode,          /* 27 UINTX_FORMAT */
1278              &stack_start);   /* 28 UINTX_FORMAT */
1279       }
1280 
1281 #undef _UFM
1282 #undef _DFM
1283 
1284       if (i != 28 - 2) {
1285          assert(false, "Bad conversion from /proc/self/stat");
1286          // product mode - assume we are the initial thread, good luck in the
1287          // embedded case.
1288          warning("Can't detect initial thread stack location - bad conversion");
1289          stack_start = (uintptr_t) &rlim;
1290       }
1291     } else {
1292       // For some reason we can't open /proc/self/stat (for example, running on
1293       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1294       // most cases, so don't abort:
1295       warning("Can't detect initial thread stack location - no /proc/self/stat");
1296       stack_start = (uintptr_t) &rlim;
1297     }
1298   }
1299 
1300   // Now we have a pointer (stack_start) very close to the stack top, the
1301   // next thing to do is to figure out the exact location of stack top. We
1302   // can find out the virtual memory area that contains stack_start by
1303   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1304   // and its upper limit is the real stack top. (again, this would fail if
1305   // running inside chroot, because /proc may not exist.)
1306 
1307   uintptr_t stack_top;
1308   address low, high;
1309   if (find_vma((address)stack_start, &low, &high)) {
1310     // success, "high" is the true stack top. (ignore "low", because initial
1311     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1312     stack_top = (uintptr_t)high;
1313   } else {
1314     // failed, likely because /proc/self/maps does not exist
1315     warning("Can't detect initial thread stack location - find_vma failed");
1316     // best effort: stack_start is normally within a few pages below the real
1317     // stack top, use it as stack top, and reduce stack size so we won't put
1318     // guard page outside stack.
1319     stack_top = stack_start;
1320     stack_size -= 16 * page_size();
1321   }
1322 
1323   // stack_top could be partially down the page so align it
1324   stack_top = align_size_up(stack_top, page_size());
1325 
1326   if (max_size && stack_size > max_size) {
1327      _initial_thread_stack_size = max_size;
1328   } else {
1329      _initial_thread_stack_size = stack_size;
1330   }
1331 
1332   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1333   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1334 }
1335 
1336 ////////////////////////////////////////////////////////////////////////////////
1337 // time support
1338 
1339 // Time since start-up in seconds to a fine granularity.
1340 // Used by VMSelfDestructTimer and the MemProfiler.
1341 double os::elapsedTime() {
1342 
1343   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1344 }
1345 
1346 jlong os::elapsed_counter() {
1347   return javaTimeNanos() - initial_time_count;
1348 }
1349 
1350 jlong os::elapsed_frequency() {
1351   return NANOSECS_PER_SEC; // nanosecond resolution
1352 }
1353 
1354 bool os::supports_vtime() { return true; }
1355 bool os::enable_vtime()   { return false; }
1356 bool os::vtime_enabled()  { return false; }
1357 
1358 double os::elapsedVTime() {
1359   struct rusage usage;
1360   int retval = getrusage(RUSAGE_THREAD, &usage);
1361   if (retval == 0) {
1362     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);
1363   } else {
1364     // better than nothing, but not much
1365     return elapsedTime();
1366   }
1367 }
1368 
1369 jlong os::javaTimeMillis() {
1370   timeval time;
1371   int status = gettimeofday(&time, NULL);
1372   assert(status != -1, "linux error");
1373   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1374 }
1375 
1376 #ifndef CLOCK_MONOTONIC
1377 #define CLOCK_MONOTONIC (1)
1378 #endif
1379 
1380 void os::Linux::clock_init() {
1381   // we do dlopen's in this particular order due to bug in linux
1382   // dynamical loader (see 6348968) leading to crash on exit
1383   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1384   if (handle == NULL) {
1385     handle = dlopen("librt.so", RTLD_LAZY);
1386   }
1387 
1388   if (handle) {
1389     int (*clock_getres_func)(clockid_t, struct timespec*) =
1390            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1391     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1392            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1393     if (clock_getres_func && clock_gettime_func) {
1394       // See if monotonic clock is supported by the kernel. Note that some
1395       // early implementations simply return kernel jiffies (updated every
1396       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1397       // for nano time (though the monotonic property is still nice to have).
1398       // It's fixed in newer kernels, however clock_getres() still returns
1399       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1400       // resolution for now. Hopefully as people move to new kernels, this
1401       // won't be a problem.
1402       struct timespec res;
1403       struct timespec tp;
1404       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1405           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1406         // yes, monotonic clock is supported
1407         _clock_gettime = clock_gettime_func;
1408         return;
1409       } else {
1410         // close librt if there is no monotonic clock
1411         dlclose(handle);
1412       }
1413     }
1414   }
1415   warning("No monotonic clock was available - timed services may " \
1416           "be adversely affected if the time-of-day clock changes");
1417 }
1418 
1419 #ifndef SYS_clock_getres
1420 
1421 #if defined(IA32) || defined(AMD64)
1422 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1423 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1424 #else
1425 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1426 #define sys_clock_getres(x,y)  -1
1427 #endif
1428 
1429 #else
1430 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1431 #endif
1432 
1433 void os::Linux::fast_thread_clock_init() {
1434   if (!UseLinuxPosixThreadCPUClocks) {
1435     return;
1436   }
1437   clockid_t clockid;
1438   struct timespec tp;
1439   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1440       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1441 
1442   // Switch to using fast clocks for thread cpu time if
1443   // the sys_clock_getres() returns 0 error code.
1444   // Note, that some kernels may support the current thread
1445   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1446   // returned by the pthread_getcpuclockid().
1447   // If the fast Posix clocks are supported then the sys_clock_getres()
1448   // must return at least tp.tv_sec == 0 which means a resolution
1449   // better than 1 sec. This is extra check for reliability.
1450 
1451   if(pthread_getcpuclockid_func &&
1452      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1453      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1454 
1455     _supports_fast_thread_cpu_time = true;
1456     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1457   }
1458 }
1459 
1460 jlong os::javaTimeNanos() {
1461   if (os::supports_monotonic_clock()) {
1462     struct timespec tp;
1463     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1464     assert(status == 0, "gettime error");
1465     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1466     return result;
1467   } else {
1468     timeval time;
1469     int status = gettimeofday(&time, NULL);
1470     assert(status != -1, "linux error");
1471     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1472     return 1000 * usecs;
1473   }
1474 }
1475 
1476 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1477   if (os::supports_monotonic_clock()) {
1478     info_ptr->max_value = ALL_64_BITS;
1479 
1480     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1481     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1482     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1483   } else {
1484     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1485     info_ptr->max_value = ALL_64_BITS;
1486 
1487     // gettimeofday is a real time clock so it skips
1488     info_ptr->may_skip_backward = true;
1489     info_ptr->may_skip_forward = true;
1490   }
1491 
1492   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1493 }
1494 
1495 // Return the real, user, and system times in seconds from an
1496 // arbitrary fixed point in the past.
1497 bool os::getTimesSecs(double* process_real_time,
1498                       double* process_user_time,
1499                       double* process_system_time) {
1500   struct tms ticks;
1501   clock_t real_ticks = times(&ticks);
1502 
1503   if (real_ticks == (clock_t) (-1)) {
1504     return false;
1505   } else {
1506     double ticks_per_second = (double) clock_tics_per_sec;
1507     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1508     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1509     *process_real_time = ((double) real_ticks) / ticks_per_second;
1510 
1511     return true;
1512   }
1513 }
1514 
1515 
1516 char * os::local_time_string(char *buf, size_t buflen) {
1517   struct tm t;
1518   time_t long_time;
1519   time(&long_time);
1520   localtime_r(&long_time, &t);
1521   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1522                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1523                t.tm_hour, t.tm_min, t.tm_sec);
1524   return buf;
1525 }
1526 
1527 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1528   return localtime_r(clock, res);
1529 }
1530 
1531 ////////////////////////////////////////////////////////////////////////////////
1532 // runtime exit support
1533 
1534 // Note: os::shutdown() might be called very early during initialization, or
1535 // called from signal handler. Before adding something to os::shutdown(), make
1536 // sure it is async-safe and can handle partially initialized VM.
1537 void os::shutdown() {
1538 
1539   // allow PerfMemory to attempt cleanup of any persistent resources
1540   perfMemory_exit();
1541 
1542   // needs to remove object in file system
1543   AttachListener::abort();
1544 
1545   // flush buffered output, finish log files
1546   ostream_abort();
1547 
1548   // Check for abort hook
1549   abort_hook_t abort_hook = Arguments::abort_hook();
1550   if (abort_hook != NULL) {
1551     abort_hook();
1552   }
1553 
1554 }
1555 
1556 // Note: os::abort() might be called very early during initialization, or
1557 // called from signal handler. Before adding something to os::abort(), make
1558 // sure it is async-safe and can handle partially initialized VM.
1559 void os::abort(bool dump_core) {
1560   os::shutdown();
1561   if (dump_core) {
1562 #ifndef PRODUCT
1563     fdStream out(defaultStream::output_fd());
1564     out.print_raw("Current thread is ");
1565     char buf[16];
1566     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1567     out.print_raw_cr(buf);
1568     out.print_raw_cr("Dumping core ...");
1569 #endif
1570     ::abort(); // dump core
1571   }
1572 
1573   ::exit(1);
1574 }
1575 
1576 // Die immediately, no exit hook, no abort hook, no cleanup.
1577 void os::die() {
1578   // _exit() on LinuxThreads only kills current thread
1579   ::abort();
1580 }
1581 
1582 // unused on linux for now.
1583 void os::set_error_file(const char *logfile) {}
1584 
1585 
1586 // This method is a copy of JDK's sysGetLastErrorString
1587 // from src/solaris/hpi/src/system_md.c
1588 
1589 size_t os::lasterror(char *buf, size_t len) {
1590 
1591   if (errno == 0)  return 0;
1592 
1593   const char *s = ::strerror(errno);
1594   size_t n = ::strlen(s);
1595   if (n >= len) {
1596     n = len - 1;
1597   }
1598   ::strncpy(buf, s, n);
1599   buf[n] = '\0';
1600   return n;
1601 }
1602 
1603 intx os::current_thread_id() { return (intx)pthread_self(); }
1604 int os::current_process_id() {
1605 
1606   // Under the old linux thread library, linux gives each thread
1607   // its own process id. Because of this each thread will return
1608   // a different pid if this method were to return the result
1609   // of getpid(2). Linux provides no api that returns the pid
1610   // of the launcher thread for the vm. This implementation
1611   // returns a unique pid, the pid of the launcher thread
1612   // that starts the vm 'process'.
1613 
1614   // Under the NPTL, getpid() returns the same pid as the
1615   // launcher thread rather than a unique pid per thread.
1616   // Use gettid() if you want the old pre NPTL behaviour.
1617 
1618   // if you are looking for the result of a call to getpid() that
1619   // returns a unique pid for the calling thread, then look at the
1620   // OSThread::thread_id() method in osThread_linux.hpp file
1621 
1622   return (int)(_initial_pid ? _initial_pid : getpid());
1623 }
1624 
1625 // DLL functions
1626 
1627 const char* os::dll_file_extension() { return ".so"; }
1628 
1629 // This must be hard coded because it's the system's temporary
1630 // directory not the java application's temp directory, ala java.io.tmpdir.
1631 const char* os::get_temp_directory() { return "/tmp"; }
1632 
1633 static bool file_exists(const char* filename) {
1634   struct stat statbuf;
1635   if (filename == NULL || strlen(filename) == 0) {
1636     return false;
1637   }
1638   return os::stat(filename, &statbuf) == 0;
1639 }
1640 
1641 bool os::dll_build_name(char* buffer, size_t buflen,
1642                         const char* pname, const char* fname) {
1643   bool retval = false;
1644   // Copied from libhpi
1645   const size_t pnamelen = pname ? strlen(pname) : 0;
1646 
1647   // Return error on buffer overflow.
1648   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1649     return retval;
1650   }
1651 
1652   if (pnamelen == 0) {
1653     snprintf(buffer, buflen, "lib%s.so", fname);
1654     retval = true;
1655   } else if (strchr(pname, *os::path_separator()) != NULL) {
1656     int n;
1657     char** pelements = split_path(pname, &n);
1658     if (pelements == NULL) {
1659       return false;
1660     }
1661     for (int i = 0 ; i < n ; i++) {
1662       // Really shouldn't be NULL, but check can't hurt
1663       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1664         continue; // skip the empty path values
1665       }
1666       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1667       if (file_exists(buffer)) {
1668         retval = true;
1669         break;
1670       }
1671     }
1672     // release the storage
1673     for (int i = 0 ; i < n ; i++) {
1674       if (pelements[i] != NULL) {
1675         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1676       }
1677     }
1678     if (pelements != NULL) {
1679       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1680     }
1681   } else {
1682     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1683     retval = true;
1684   }
1685   return retval;
1686 }
1687 
1688 // check if addr is inside libjvm.so
1689 bool os::address_is_in_vm(address addr) {
1690   static address libjvm_base_addr;
1691   Dl_info dlinfo;
1692 
1693   if (libjvm_base_addr == NULL) {
1694     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1695       libjvm_base_addr = (address)dlinfo.dli_fbase;
1696     }
1697     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1698   }
1699 
1700   if (dladdr((void *)addr, &dlinfo) != 0) {
1701     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1702   }
1703 
1704   return false;
1705 }
1706 
1707 bool os::dll_address_to_function_name(address addr, char *buf,
1708                                       int buflen, int *offset) {
1709   // buf is not optional, but offset is optional
1710   assert(buf != NULL, "sanity check");
1711 
1712   Dl_info dlinfo;
1713 
1714   if (dladdr((void*)addr, &dlinfo) != 0) {
1715     // see if we have a matching symbol
1716     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1717       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1718         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1719       }
1720       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1721       return true;
1722     }
1723     // no matching symbol so try for just file info
1724     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1725       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1726                           buf, buflen, offset, dlinfo.dli_fname)) {
1727         return true;
1728       }
1729     }
1730   }
1731 
1732   buf[0] = '\0';
1733   if (offset != NULL) *offset = -1;
1734   return false;
1735 }
1736 
1737 struct _address_to_library_name {
1738   address addr;          // input : memory address
1739   size_t  buflen;        //         size of fname
1740   char*   fname;         // output: library name
1741   address base;          //         library base addr
1742 };
1743 
1744 static int address_to_library_name_callback(struct dl_phdr_info *info,
1745                                             size_t size, void *data) {
1746   int i;
1747   bool found = false;
1748   address libbase = NULL;
1749   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1750 
1751   // iterate through all loadable segments
1752   for (i = 0; i < info->dlpi_phnum; i++) {
1753     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1754     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1755       // base address of a library is the lowest address of its loaded
1756       // segments.
1757       if (libbase == NULL || libbase > segbase) {
1758         libbase = segbase;
1759       }
1760       // see if 'addr' is within current segment
1761       if (segbase <= d->addr &&
1762           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1763         found = true;
1764       }
1765     }
1766   }
1767 
1768   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1769   // so dll_address_to_library_name() can fall through to use dladdr() which
1770   // can figure out executable name from argv[0].
1771   if (found && info->dlpi_name && info->dlpi_name[0]) {
1772     d->base = libbase;
1773     if (d->fname) {
1774       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1775     }
1776     return 1;
1777   }
1778   return 0;
1779 }
1780 
1781 bool os::dll_address_to_library_name(address addr, char* buf,
1782                                      int buflen, int* offset) {
1783   // buf is not optional, but offset is optional
1784   assert(buf != NULL, "sanity check");
1785 
1786   Dl_info dlinfo;
1787   struct _address_to_library_name data;
1788 
1789   // There is a bug in old glibc dladdr() implementation that it could resolve
1790   // to wrong library name if the .so file has a base address != NULL. Here
1791   // we iterate through the program headers of all loaded libraries to find
1792   // out which library 'addr' really belongs to. This workaround can be
1793   // removed once the minimum requirement for glibc is moved to 2.3.x.
1794   data.addr = addr;
1795   data.fname = buf;
1796   data.buflen = buflen;
1797   data.base = NULL;
1798   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1799 
1800   if (rslt) {
1801      // buf already contains library name
1802      if (offset) *offset = addr - data.base;
1803      return true;
1804   }
1805   if (dladdr((void*)addr, &dlinfo) != 0) {
1806     if (dlinfo.dli_fname != NULL) {
1807       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1808     }
1809     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1810       *offset = addr - (address)dlinfo.dli_fbase;
1811     }
1812     return true;
1813   }
1814 
1815   buf[0] = '\0';
1816   if (offset) *offset = -1;
1817   return false;
1818 }
1819 
1820   // Loads .dll/.so and
1821   // in case of error it checks if .dll/.so was built for the
1822   // same architecture as Hotspot is running on
1823 
1824 
1825 // Remember the stack's state. The Linux dynamic linker will change
1826 // the stack to 'executable' at most once, so we must safepoint only once.
1827 bool os::Linux::_stack_is_executable = false;
1828 
1829 // VM operation that loads a library.  This is necessary if stack protection
1830 // of the Java stacks can be lost during loading the library.  If we
1831 // do not stop the Java threads, they can stack overflow before the stacks
1832 // are protected again.
1833 class VM_LinuxDllLoad: public VM_Operation {
1834  private:
1835   const char *_filename;
1836   char *_ebuf;
1837   int _ebuflen;
1838   void *_lib;
1839  public:
1840   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1841     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1842   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1843   void doit() {
1844     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1845     os::Linux::_stack_is_executable = true;
1846   }
1847   void* loaded_library() { return _lib; }
1848 };
1849 
1850 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1851 {
1852   void * result = NULL;
1853   bool load_attempted = false;
1854 
1855   // Check whether the library to load might change execution rights
1856   // of the stack. If they are changed, the protection of the stack
1857   // guard pages will be lost. We need a safepoint to fix this.
1858   //
1859   // See Linux man page execstack(8) for more info.
1860   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1861     ElfFile ef(filename);
1862     if (!ef.specifies_noexecstack()) {
1863       if (!is_init_completed()) {
1864         os::Linux::_stack_is_executable = true;
1865         // This is OK - No Java threads have been created yet, and hence no
1866         // stack guard pages to fix.
1867         //
1868         // This should happen only when you are building JDK7 using a very
1869         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1870         //
1871         // Dynamic loader will make all stacks executable after
1872         // this function returns, and will not do that again.
1873         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1874       } else {
1875         warning("You have loaded library %s which might have disabled stack guard. "
1876                 "The VM will try to fix the stack guard now.\n"
1877                 "It's highly recommended that you fix the library with "
1878                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1879                 filename);
1880 
1881         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1882         JavaThread *jt = JavaThread::current();
1883         if (jt->thread_state() != _thread_in_native) {
1884           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1885           // that requires ExecStack. Cannot enter safe point. Let's give up.
1886           warning("Unable to fix stack guard. Giving up.");
1887         } else {
1888           if (!LoadExecStackDllInVMThread) {
1889             // This is for the case where the DLL has an static
1890             // constructor function that executes JNI code. We cannot
1891             // load such DLLs in the VMThread.
1892             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1893           }
1894 
1895           ThreadInVMfromNative tiv(jt);
1896           debug_only(VMNativeEntryWrapper vew;)
1897 
1898           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1899           VMThread::execute(&op);
1900           if (LoadExecStackDllInVMThread) {
1901             result = op.loaded_library();
1902           }
1903           load_attempted = true;
1904         }
1905       }
1906     }
1907   }
1908 
1909   if (!load_attempted) {
1910     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1911   }
1912 
1913   if (result != NULL) {
1914     // Successful loading
1915     return result;
1916   }
1917 
1918   Elf32_Ehdr elf_head;
1919   int diag_msg_max_length=ebuflen-strlen(ebuf);
1920   char* diag_msg_buf=ebuf+strlen(ebuf);
1921 
1922   if (diag_msg_max_length==0) {
1923     // No more space in ebuf for additional diagnostics message
1924     return NULL;
1925   }
1926 
1927 
1928   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1929 
1930   if (file_descriptor < 0) {
1931     // Can't open library, report dlerror() message
1932     return NULL;
1933   }
1934 
1935   bool failed_to_read_elf_head=
1936     (sizeof(elf_head)!=
1937         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1938 
1939   ::close(file_descriptor);
1940   if (failed_to_read_elf_head) {
1941     // file i/o error - report dlerror() msg
1942     return NULL;
1943   }
1944 
1945   typedef struct {
1946     Elf32_Half  code;         // Actual value as defined in elf.h
1947     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1948     char        elf_class;    // 32 or 64 bit
1949     char        endianess;    // MSB or LSB
1950     char*       name;         // String representation
1951   } arch_t;
1952 
1953   #ifndef EM_486
1954   #define EM_486          6               /* Intel 80486 */
1955   #endif
1956 
1957   static const arch_t arch_array[]={
1958     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1959     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1960     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1961     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1962     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1963     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1964     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1965     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1966     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1967     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1968     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1969     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1970     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1971     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1972     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1973     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1974   };
1975 
1976   #if  (defined IA32)
1977     static  Elf32_Half running_arch_code=EM_386;
1978   #elif   (defined AMD64)
1979     static  Elf32_Half running_arch_code=EM_X86_64;
1980   #elif  (defined IA64)
1981     static  Elf32_Half running_arch_code=EM_IA_64;
1982   #elif  (defined __sparc) && (defined _LP64)
1983     static  Elf32_Half running_arch_code=EM_SPARCV9;
1984   #elif  (defined __sparc) && (!defined _LP64)
1985     static  Elf32_Half running_arch_code=EM_SPARC;
1986   #elif  (defined __powerpc64__)
1987     static  Elf32_Half running_arch_code=EM_PPC64;
1988   #elif  (defined __powerpc__)
1989     static  Elf32_Half running_arch_code=EM_PPC;
1990   #elif  (defined ARM)
1991     static  Elf32_Half running_arch_code=EM_ARM;
1992   #elif  (defined S390)
1993     static  Elf32_Half running_arch_code=EM_S390;
1994   #elif  (defined ALPHA)
1995     static  Elf32_Half running_arch_code=EM_ALPHA;
1996   #elif  (defined MIPSEL)
1997     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1998   #elif  (defined PARISC)
1999     static  Elf32_Half running_arch_code=EM_PARISC;
2000   #elif  (defined MIPS)
2001     static  Elf32_Half running_arch_code=EM_MIPS;
2002   #elif  (defined M68K)
2003     static  Elf32_Half running_arch_code=EM_68K;
2004   #else
2005     #error Method os::dll_load requires that one of following is defined:\
2006          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2007   #endif
2008 
2009   // Identify compatability class for VM's architecture and library's architecture
2010   // Obtain string descriptions for architectures
2011 
2012   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2013   int running_arch_index=-1;
2014 
2015   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2016     if (running_arch_code == arch_array[i].code) {
2017       running_arch_index    = i;
2018     }
2019     if (lib_arch.code == arch_array[i].code) {
2020       lib_arch.compat_class = arch_array[i].compat_class;
2021       lib_arch.name         = arch_array[i].name;
2022     }
2023   }
2024 
2025   assert(running_arch_index != -1,
2026     "Didn't find running architecture code (running_arch_code) in arch_array");
2027   if (running_arch_index == -1) {
2028     // Even though running architecture detection failed
2029     // we may still continue with reporting dlerror() message
2030     return NULL;
2031   }
2032 
2033   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2034     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2035     return NULL;
2036   }
2037 
2038 #ifndef S390
2039   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2040     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2041     return NULL;
2042   }
2043 #endif // !S390
2044 
2045   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2046     if ( lib_arch.name!=NULL ) {
2047       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2048         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2049         lib_arch.name, arch_array[running_arch_index].name);
2050     } else {
2051       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2052       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2053         lib_arch.code,
2054         arch_array[running_arch_index].name);
2055     }
2056   }
2057 
2058   return NULL;
2059 }
2060 
2061 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2062   void * result = ::dlopen(filename, RTLD_LAZY);
2063   if (result == NULL) {
2064     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2065     ebuf[ebuflen-1] = '\0';
2066   }
2067   return result;
2068 }
2069 
2070 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2071   void * result = NULL;
2072   if (LoadExecStackDllInVMThread) {
2073     result = dlopen_helper(filename, ebuf, ebuflen);
2074   }
2075 
2076   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2077   // library that requires an executable stack, or which does not have this
2078   // stack attribute set, dlopen changes the stack attribute to executable. The
2079   // read protection of the guard pages gets lost.
2080   //
2081   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2082   // may have been queued at the same time.
2083 
2084   if (!_stack_is_executable) {
2085     JavaThread *jt = Threads::first();
2086 
2087     while (jt) {
2088       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2089           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2090         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2091                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2092           warning("Attempt to reguard stack yellow zone failed.");
2093         }
2094       }
2095       jt = jt->next();
2096     }
2097   }
2098 
2099   return result;
2100 }
2101 
2102 /*
2103  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2104  * chances are you might want to run the generated bits against glibc-2.0
2105  * libdl.so, so always use locking for any version of glibc.
2106  */
2107 void* os::dll_lookup(void* handle, const char* name) {
2108   pthread_mutex_lock(&dl_mutex);
2109   void* res = dlsym(handle, name);
2110   pthread_mutex_unlock(&dl_mutex);
2111   return res;
2112 }
2113 
2114 void* os::get_default_process_handle() {
2115   return (void*)::dlopen(NULL, RTLD_LAZY);
2116 }
2117 
2118 static bool _print_ascii_file(const char* filename, outputStream* st) {
2119   int fd = ::open(filename, O_RDONLY);
2120   if (fd == -1) {
2121      return false;
2122   }
2123 
2124   char buf[32];
2125   int bytes;
2126   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2127     st->print_raw(buf, bytes);
2128   }
2129 
2130   ::close(fd);
2131 
2132   return true;
2133 }
2134 
2135 void os::print_dll_info(outputStream *st) {
2136    st->print_cr("Dynamic libraries:");
2137 
2138    char fname[32];
2139    pid_t pid = os::Linux::gettid();
2140 
2141    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2142 
2143    if (!_print_ascii_file(fname, st)) {
2144      st->print("Can not get library information for pid = %d\n", pid);
2145    }
2146 }
2147 
2148 void os::print_os_info_brief(outputStream* st) {
2149   os::Linux::print_distro_info(st);
2150 
2151   os::Posix::print_uname_info(st);
2152 
2153   os::Linux::print_libversion_info(st);
2154 
2155 }
2156 
2157 void os::print_os_info(outputStream* st) {
2158   st->print("OS:");
2159 
2160   os::Linux::print_distro_info(st);
2161 
2162   os::Posix::print_uname_info(st);
2163 
2164   // Print warning if unsafe chroot environment detected
2165   if (unsafe_chroot_detected) {
2166     st->print("WARNING!! ");
2167     st->print_cr(unstable_chroot_error);
2168   }
2169 
2170   os::Linux::print_libversion_info(st);
2171 
2172   os::Posix::print_rlimit_info(st);
2173 
2174   os::Posix::print_load_average(st);
2175 
2176   os::Linux::print_full_memory_info(st);
2177 }
2178 
2179 // Try to identify popular distros.
2180 // Most Linux distributions have a /etc/XXX-release file, which contains
2181 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2182 // file that also contains the OS version string. Some have more than one
2183 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2184 // /etc/redhat-release.), so the order is important.
2185 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2186 // their own specific XXX-release file as well as a redhat-release file.
2187 // Because of this the XXX-release file needs to be searched for before the
2188 // redhat-release file.
2189 // Since Red Hat has a lsb-release file that is not very descriptive the
2190 // search for redhat-release needs to be before lsb-release.
2191 // Since the lsb-release file is the new standard it needs to be searched
2192 // before the older style release files.
2193 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2194 // next to last resort.  The os-release file is a new standard that contains
2195 // distribution information and the system-release file seems to be an old
2196 // standard that has been replaced by the lsb-release and os-release files.
2197 // Searching for the debian_version file is the last resort.  It contains
2198 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2199 // "Debian " is printed before the contents of the debian_version file.
2200 void os::Linux::print_distro_info(outputStream* st) {
2201    if (!_print_ascii_file("/etc/oracle-release", st) &&
2202        !_print_ascii_file("/etc/mandriva-release", st) &&
2203        !_print_ascii_file("/etc/mandrake-release", st) &&
2204        !_print_ascii_file("/etc/sun-release", st) &&
2205        !_print_ascii_file("/etc/redhat-release", st) &&
2206        !_print_ascii_file("/etc/lsb-release", st) &&
2207        !_print_ascii_file("/etc/SuSE-release", st) &&
2208        !_print_ascii_file("/etc/turbolinux-release", st) &&
2209        !_print_ascii_file("/etc/gentoo-release", st) &&
2210        !_print_ascii_file("/etc/ltib-release", st) &&
2211        !_print_ascii_file("/etc/angstrom-version", st) &&
2212        !_print_ascii_file("/etc/system-release", st) &&
2213        !_print_ascii_file("/etc/os-release", st)) {
2214 
2215        if (file_exists("/etc/debian_version")) {
2216          st->print("Debian ");
2217          _print_ascii_file("/etc/debian_version", st);
2218        } else {
2219          st->print("Linux");
2220        }
2221    }
2222    st->cr();
2223 }
2224 
2225 void os::Linux::print_libversion_info(outputStream* st) {
2226   // libc, pthread
2227   st->print("libc:");
2228   st->print(os::Linux::glibc_version()); st->print(" ");
2229   st->print(os::Linux::libpthread_version()); st->print(" ");
2230   if (os::Linux::is_LinuxThreads()) {
2231      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2232   }
2233   st->cr();
2234 }
2235 
2236 void os::Linux::print_full_memory_info(outputStream* st) {
2237    st->print("\n/proc/meminfo:\n");
2238    _print_ascii_file("/proc/meminfo", st);
2239    st->cr();
2240 }
2241 
2242 void os::print_memory_info(outputStream* st) {
2243 
2244   st->print("Memory:");
2245   st->print(" %dk page", os::vm_page_size()>>10);
2246 
2247   // values in struct sysinfo are "unsigned long"
2248   struct sysinfo si;
2249   sysinfo(&si);
2250 
2251   st->print(", physical " UINT64_FORMAT "k",
2252             os::physical_memory() >> 10);
2253   st->print("(" UINT64_FORMAT "k free)",
2254             os::available_memory() >> 10);
2255   st->print(", swap " UINT64_FORMAT "k",
2256             ((jlong)si.totalswap * si.mem_unit) >> 10);
2257   st->print("(" UINT64_FORMAT "k free)",
2258             ((jlong)si.freeswap * si.mem_unit) >> 10);
2259   st->cr();
2260 }
2261 
2262 void os::pd_print_cpu_info(outputStream* st) {
2263   st->print("\n/proc/cpuinfo:\n");
2264   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2265     st->print("  <Not Available>");
2266   }
2267   st->cr();
2268 }
2269 
2270 void os::print_siginfo(outputStream* st, void* siginfo) {
2271   const siginfo_t* si = (const siginfo_t*)siginfo;
2272 
2273   os::Posix::print_siginfo_brief(st, si);
2274 
2275   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2276       UseSharedSpaces) {
2277     FileMapInfo* mapinfo = FileMapInfo::current_info();
2278     if (mapinfo->is_in_shared_space(si->si_addr)) {
2279       st->print("\n\nError accessing class data sharing archive."   \
2280                 " Mapped file inaccessible during execution, "      \
2281                 " possible disk/network problem.");
2282     }
2283   }
2284   st->cr();
2285 }
2286 
2287 
2288 static void print_signal_handler(outputStream* st, int sig,
2289                                  char* buf, size_t buflen);
2290 
2291 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2292   st->print_cr("Signal Handlers:");
2293   print_signal_handler(st, SIGSEGV, buf, buflen);
2294   print_signal_handler(st, SIGBUS , buf, buflen);
2295   print_signal_handler(st, SIGFPE , buf, buflen);
2296   print_signal_handler(st, SIGPIPE, buf, buflen);
2297   print_signal_handler(st, SIGXFSZ, buf, buflen);
2298   print_signal_handler(st, SIGILL , buf, buflen);
2299   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2300   print_signal_handler(st, SR_signum, buf, buflen);
2301   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2302   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2303   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2304   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2305 #if defined(PPC64)
2306   print_signal_handler(st, SIGTRAP, buf, buflen);
2307 #endif
2308 }
2309 
2310 static char saved_jvm_path[MAXPATHLEN] = {0};
2311 
2312 // Find the full path to the current module, libjvm.so
2313 void os::jvm_path(char *buf, jint buflen) {
2314   // Error checking.
2315   if (buflen < MAXPATHLEN) {
2316     assert(false, "must use a large-enough buffer");
2317     buf[0] = '\0';
2318     return;
2319   }
2320   // Lazy resolve the path to current module.
2321   if (saved_jvm_path[0] != 0) {
2322     strcpy(buf, saved_jvm_path);
2323     return;
2324   }
2325 
2326   char dli_fname[MAXPATHLEN];
2327   bool ret = dll_address_to_library_name(
2328                 CAST_FROM_FN_PTR(address, os::jvm_path),
2329                 dli_fname, sizeof(dli_fname), NULL);
2330   assert(ret, "cannot locate libjvm");
2331   char *rp = NULL;
2332   if (ret && dli_fname[0] != '\0') {
2333     rp = realpath(dli_fname, buf);
2334   }
2335   if (rp == NULL)
2336     return;
2337 
2338   if (Arguments::sun_java_launcher_is_altjvm()) {
2339     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2340     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2341     // If "/jre/lib/" appears at the right place in the string, then
2342     // assume we are installed in a JDK and we're done. Otherwise, check
2343     // for a JAVA_HOME environment variable and fix up the path so it
2344     // looks like libjvm.so is installed there (append a fake suffix
2345     // hotspot/libjvm.so).
2346     const char *p = buf + strlen(buf) - 1;
2347     for (int count = 0; p > buf && count < 5; ++count) {
2348       for (--p; p > buf && *p != '/'; --p)
2349         /* empty */ ;
2350     }
2351 
2352     if (strncmp(p, "/jre/lib/", 9) != 0) {
2353       // Look for JAVA_HOME in the environment.
2354       char* java_home_var = ::getenv("JAVA_HOME");
2355       if (java_home_var != NULL && java_home_var[0] != 0) {
2356         char* jrelib_p;
2357         int len;
2358 
2359         // Check the current module name "libjvm.so".
2360         p = strrchr(buf, '/');
2361         assert(strstr(p, "/libjvm") == p, "invalid library name");
2362 
2363         rp = realpath(java_home_var, buf);
2364         if (rp == NULL)
2365           return;
2366 
2367         // determine if this is a legacy image or modules image
2368         // modules image doesn't have "jre" subdirectory
2369         len = strlen(buf);
2370         jrelib_p = buf + len;
2371         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2372         if (0 != access(buf, F_OK)) {
2373           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2374         }
2375 
2376         if (0 == access(buf, F_OK)) {
2377           // Use current module name "libjvm.so"
2378           len = strlen(buf);
2379           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2380         } else {
2381           // Go back to path of .so
2382           rp = realpath(dli_fname, buf);
2383           if (rp == NULL)
2384             return;
2385         }
2386       }
2387     }
2388   }
2389 
2390   strcpy(saved_jvm_path, buf);
2391 }
2392 
2393 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2394   // no prefix required, not even "_"
2395 }
2396 
2397 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2398   // no suffix required
2399 }
2400 
2401 ////////////////////////////////////////////////////////////////////////////////
2402 // sun.misc.Signal support
2403 
2404 static volatile jint sigint_count = 0;
2405 
2406 static void
2407 UserHandler(int sig, void *siginfo, void *context) {
2408   // 4511530 - sem_post is serialized and handled by the manager thread. When
2409   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2410   // don't want to flood the manager thread with sem_post requests.
2411   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2412       return;
2413 
2414   // Ctrl-C is pressed during error reporting, likely because the error
2415   // handler fails to abort. Let VM die immediately.
2416   if (sig == SIGINT && is_error_reported()) {
2417      os::die();
2418   }
2419 
2420   os::signal_notify(sig);
2421 }
2422 
2423 void* os::user_handler() {
2424   return CAST_FROM_FN_PTR(void*, UserHandler);
2425 }
2426 
2427 class Semaphore : public StackObj {
2428   public:
2429     Semaphore();
2430     ~Semaphore();
2431     void signal();
2432     void wait();
2433     bool trywait();
2434     bool timedwait(unsigned int sec, int nsec);
2435   private:
2436     sem_t _semaphore;
2437 };
2438 
2439 Semaphore::Semaphore() {
2440   sem_init(&_semaphore, 0, 0);
2441 }
2442 
2443 Semaphore::~Semaphore() {
2444   sem_destroy(&_semaphore);
2445 }
2446 
2447 void Semaphore::signal() {
2448   sem_post(&_semaphore);
2449 }
2450 
2451 void Semaphore::wait() {
2452   sem_wait(&_semaphore);
2453 }
2454 
2455 bool Semaphore::trywait() {
2456   return sem_trywait(&_semaphore) == 0;
2457 }
2458 
2459 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2460 
2461   struct timespec ts;
2462   // Semaphore's are always associated with CLOCK_REALTIME
2463   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2464   // see unpackTime for discussion on overflow checking
2465   if (sec >= MAX_SECS) {
2466     ts.tv_sec += MAX_SECS;
2467     ts.tv_nsec = 0;
2468   } else {
2469     ts.tv_sec += sec;
2470     ts.tv_nsec += nsec;
2471     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2472       ts.tv_nsec -= NANOSECS_PER_SEC;
2473       ++ts.tv_sec; // note: this must be <= max_secs
2474     }
2475   }
2476 
2477   while (1) {
2478     int result = sem_timedwait(&_semaphore, &ts);
2479     if (result == 0) {
2480       return true;
2481     } else if (errno == EINTR) {
2482       continue;
2483     } else if (errno == ETIMEDOUT) {
2484       return false;
2485     } else {
2486       return false;
2487     }
2488   }
2489 }
2490 
2491 extern "C" {
2492   typedef void (*sa_handler_t)(int);
2493   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2494 }
2495 
2496 void* os::signal(int signal_number, void* handler) {
2497   struct sigaction sigAct, oldSigAct;
2498 
2499   sigfillset(&(sigAct.sa_mask));
2500   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2501   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2502 
2503   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2504     // -1 means registration failed
2505     return (void *)-1;
2506   }
2507 
2508   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2509 }
2510 
2511 void os::signal_raise(int signal_number) {
2512   ::raise(signal_number);
2513 }
2514 
2515 /*
2516  * The following code is moved from os.cpp for making this
2517  * code platform specific, which it is by its very nature.
2518  */
2519 
2520 // Will be modified when max signal is changed to be dynamic
2521 int os::sigexitnum_pd() {
2522   return NSIG;
2523 }
2524 
2525 // a counter for each possible signal value
2526 static volatile jint pending_signals[NSIG+1] = { 0 };
2527 
2528 // Linux(POSIX) specific hand shaking semaphore.
2529 static sem_t sig_sem;
2530 static Semaphore sr_semaphore;
2531 
2532 void os::signal_init_pd() {
2533   // Initialize signal structures
2534   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2535 
2536   // Initialize signal semaphore
2537   ::sem_init(&sig_sem, 0, 0);
2538 }
2539 
2540 void os::signal_notify(int sig) {
2541   Atomic::inc(&pending_signals[sig]);
2542   ::sem_post(&sig_sem);
2543 }
2544 
2545 static int check_pending_signals(bool wait) {
2546   Atomic::store(0, &sigint_count);
2547   for (;;) {
2548     for (int i = 0; i < NSIG + 1; i++) {
2549       jint n = pending_signals[i];
2550       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2551         return i;
2552       }
2553     }
2554     if (!wait) {
2555       return -1;
2556     }
2557     JavaThread *thread = JavaThread::current();
2558     ThreadBlockInVM tbivm(thread);
2559 
2560     bool threadIsSuspended;
2561     do {
2562       thread->set_suspend_equivalent();
2563       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2564       ::sem_wait(&sig_sem);
2565 
2566       // were we externally suspended while we were waiting?
2567       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2568       if (threadIsSuspended) {
2569         //
2570         // The semaphore has been incremented, but while we were waiting
2571         // another thread suspended us. We don't want to continue running
2572         // while suspended because that would surprise the thread that
2573         // suspended us.
2574         //
2575         ::sem_post(&sig_sem);
2576 
2577         thread->java_suspend_self();
2578       }
2579     } while (threadIsSuspended);
2580   }
2581 }
2582 
2583 int os::signal_lookup() {
2584   return check_pending_signals(false);
2585 }
2586 
2587 int os::signal_wait() {
2588   return check_pending_signals(true);
2589 }
2590 
2591 ////////////////////////////////////////////////////////////////////////////////
2592 // Virtual Memory
2593 
2594 int os::vm_page_size() {
2595   // Seems redundant as all get out
2596   assert(os::Linux::page_size() != -1, "must call os::init");
2597   return os::Linux::page_size();
2598 }
2599 
2600 // Solaris allocates memory by pages.
2601 int os::vm_allocation_granularity() {
2602   assert(os::Linux::page_size() != -1, "must call os::init");
2603   return os::Linux::page_size();
2604 }
2605 
2606 // Rationale behind this function:
2607 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2608 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2609 //  samples for JITted code. Here we create private executable mapping over the code cache
2610 //  and then we can use standard (well, almost, as mapping can change) way to provide
2611 //  info for the reporting script by storing timestamp and location of symbol
2612 void linux_wrap_code(char* base, size_t size) {
2613   static volatile jint cnt = 0;
2614 
2615   if (!UseOprofile) {
2616     return;
2617   }
2618 
2619   char buf[PATH_MAX+1];
2620   int num = Atomic::add(1, &cnt);
2621 
2622   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2623            os::get_temp_directory(), os::current_process_id(), num);
2624   unlink(buf);
2625 
2626   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2627 
2628   if (fd != -1) {
2629     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2630     if (rv != (off_t)-1) {
2631       if (::write(fd, "", 1) == 1) {
2632         mmap(base, size,
2633              PROT_READ|PROT_WRITE|PROT_EXEC,
2634              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2635       }
2636     }
2637     ::close(fd);
2638     unlink(buf);
2639   }
2640 }
2641 
2642 static bool recoverable_mmap_error(int err) {
2643   // See if the error is one we can let the caller handle. This
2644   // list of errno values comes from JBS-6843484. I can't find a
2645   // Linux man page that documents this specific set of errno
2646   // values so while this list currently matches Solaris, it may
2647   // change as we gain experience with this failure mode.
2648   switch (err) {
2649   case EBADF:
2650   case EINVAL:
2651   case ENOTSUP:
2652     // let the caller deal with these errors
2653     return true;
2654 
2655   default:
2656     // Any remaining errors on this OS can cause our reserved mapping
2657     // to be lost. That can cause confusion where different data
2658     // structures think they have the same memory mapped. The worst
2659     // scenario is if both the VM and a library think they have the
2660     // same memory mapped.
2661     return false;
2662   }
2663 }
2664 
2665 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2666                                     int err) {
2667   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2668           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2669           strerror(err), err);
2670 }
2671 
2672 static void warn_fail_commit_memory(char* addr, size_t size,
2673                                     size_t alignment_hint, bool exec,
2674                                     int err) {
2675   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2676           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2677           alignment_hint, exec, strerror(err), err);
2678 }
2679 
2680 // NOTE: Linux kernel does not really reserve the pages for us.
2681 //       All it does is to check if there are enough free pages
2682 //       left at the time of mmap(). This could be a potential
2683 //       problem.
2684 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2685   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2686   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2687                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2688   if (res != (uintptr_t) MAP_FAILED) {
2689     if (UseNUMAInterleaving) {
2690       numa_make_global(addr, size);
2691     }
2692     return 0;
2693   }
2694 
2695   int err = errno;  // save errno from mmap() call above
2696 
2697   if (!recoverable_mmap_error(err)) {
2698     warn_fail_commit_memory(addr, size, exec, err);
2699     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2700   }
2701 
2702   return err;
2703 }
2704 
2705 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2706   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2707 }
2708 
2709 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2710                                   const char* mesg) {
2711   assert(mesg != NULL, "mesg must be specified");
2712   int err = os::Linux::commit_memory_impl(addr, size, exec);
2713   if (err != 0) {
2714     // the caller wants all commit errors to exit with the specified mesg:
2715     warn_fail_commit_memory(addr, size, exec, err);
2716     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2717   }
2718 }
2719 
2720 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2721 #ifndef MAP_HUGETLB
2722 #define MAP_HUGETLB 0x40000
2723 #endif
2724 
2725 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2726 #ifndef MADV_HUGEPAGE
2727 #define MADV_HUGEPAGE 14
2728 #endif
2729 
2730 int os::Linux::commit_memory_impl(char* addr, size_t size,
2731                                   size_t alignment_hint, bool exec) {
2732   int err = os::Linux::commit_memory_impl(addr, size, exec);
2733   if (err == 0) {
2734     realign_memory(addr, size, alignment_hint);
2735   }
2736   return err;
2737 }
2738 
2739 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2740                           bool exec) {
2741   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2742 }
2743 
2744 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2745                                   size_t alignment_hint, bool exec,
2746                                   const char* mesg) {
2747   assert(mesg != NULL, "mesg must be specified");
2748   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2749   if (err != 0) {
2750     // the caller wants all commit errors to exit with the specified mesg:
2751     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2752     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2753   }
2754 }
2755 
2756 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2757   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2758     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2759     // be supported or the memory may already be backed by huge pages.
2760     ::madvise(addr, bytes, MADV_HUGEPAGE);
2761   }
2762 }
2763 
2764 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2765   // This method works by doing an mmap over an existing mmaping and effectively discarding
2766   // the existing pages. However it won't work for SHM-based large pages that cannot be
2767   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2768   // small pages on top of the SHM segment. This method always works for small pages, so we
2769   // allow that in any case.
2770   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2771     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2772   }
2773 }
2774 
2775 void os::numa_make_global(char *addr, size_t bytes) {
2776   Linux::numa_interleave_memory(addr, bytes);
2777 }
2778 
2779 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2780 // bind policy to MPOL_PREFERRED for the current thread.
2781 #define USE_MPOL_PREFERRED 0
2782 
2783 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2784   // To make NUMA and large pages more robust when both enabled, we need to ease
2785   // the requirements on where the memory should be allocated. MPOL_BIND is the
2786   // default policy and it will force memory to be allocated on the specified
2787   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2788   // the specified node, but will not force it. Using this policy will prevent
2789   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2790   // free large pages.
2791   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2792   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2793 }
2794 
2795 bool os::numa_topology_changed()   { return false; }
2796 
2797 size_t os::numa_get_groups_num() {
2798   int max_node = Linux::numa_max_node();
2799   return max_node > 0 ? max_node + 1 : 1;
2800 }
2801 
2802 int os::numa_get_group_id() {
2803   int cpu_id = Linux::sched_getcpu();
2804   if (cpu_id != -1) {
2805     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2806     if (lgrp_id != -1) {
2807       return lgrp_id;
2808     }
2809   }
2810   return 0;
2811 }
2812 
2813 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2814   for (size_t i = 0; i < size; i++) {
2815     ids[i] = i;
2816   }
2817   return size;
2818 }
2819 
2820 bool os::get_page_info(char *start, page_info* info) {
2821   return false;
2822 }
2823 
2824 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2825   return end;
2826 }
2827 
2828 
2829 int os::Linux::sched_getcpu_syscall(void) {
2830   unsigned int cpu;
2831   int retval = -1;
2832 
2833 #if defined(IA32)
2834 # ifndef SYS_getcpu
2835 # define SYS_getcpu 318
2836 # endif
2837   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2838 #elif defined(AMD64)
2839 // Unfortunately we have to bring all these macros here from vsyscall.h
2840 // to be able to compile on old linuxes.
2841 # define __NR_vgetcpu 2
2842 # define VSYSCALL_START (-10UL << 20)
2843 # define VSYSCALL_SIZE 1024
2844 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2845   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2846   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2847   retval = vgetcpu(&cpu, NULL, NULL);
2848 #endif
2849 
2850   return (retval == -1) ? retval : cpu;
2851 }
2852 
2853 // Something to do with the numa-aware allocator needs these symbols
2854 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2855 extern "C" JNIEXPORT void numa_error(char *where) { }
2856 extern "C" JNIEXPORT int fork1() { return fork(); }
2857 
2858 
2859 // If we are running with libnuma version > 2, then we should
2860 // be trying to use symbols with versions 1.1
2861 // If we are running with earlier version, which did not have symbol versions,
2862 // we should use the base version.
2863 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2864   void *f = dlvsym(handle, name, "libnuma_1.1");
2865   if (f == NULL) {
2866     f = dlsym(handle, name);
2867   }
2868   return f;
2869 }
2870 
2871 bool os::Linux::libnuma_init() {
2872   // sched_getcpu() should be in libc.
2873   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2874                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2875 
2876   // If it's not, try a direct syscall.
2877   if (sched_getcpu() == -1)
2878     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2879 
2880   if (sched_getcpu() != -1) { // Does it work?
2881     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2882     if (handle != NULL) {
2883       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2884                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2885       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2886                                        libnuma_dlsym(handle, "numa_max_node")));
2887       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2888                                         libnuma_dlsym(handle, "numa_available")));
2889       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2890                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2891       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2892                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2893       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2894                                             libnuma_dlsym(handle, "numa_set_bind_policy")));
2895 
2896 
2897       if (numa_available() != -1) {
2898         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2899         // Create a cpu -> node mapping
2900         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2901         rebuild_cpu_to_node_map();
2902         return true;
2903       }
2904     }
2905   }
2906   return false;
2907 }
2908 
2909 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2910 // The table is later used in get_node_by_cpu().
2911 void os::Linux::rebuild_cpu_to_node_map() {
2912   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2913                               // in libnuma (possible values are starting from 16,
2914                               // and continuing up with every other power of 2, but less
2915                               // than the maximum number of CPUs supported by kernel), and
2916                               // is a subject to change (in libnuma version 2 the requirements
2917                               // are more reasonable) we'll just hardcode the number they use
2918                               // in the library.
2919   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2920 
2921   size_t cpu_num = os::active_processor_count();
2922   size_t cpu_map_size = NCPUS / BitsPerCLong;
2923   size_t cpu_map_valid_size =
2924     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2925 
2926   cpu_to_node()->clear();
2927   cpu_to_node()->at_grow(cpu_num - 1);
2928   size_t node_num = numa_get_groups_num();
2929 
2930   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2931   for (size_t i = 0; i < node_num; i++) {
2932     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2933       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2934         if (cpu_map[j] != 0) {
2935           for (size_t k = 0; k < BitsPerCLong; k++) {
2936             if (cpu_map[j] & (1UL << k)) {
2937               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2938             }
2939           }
2940         }
2941       }
2942     }
2943   }
2944   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2945 }
2946 
2947 int os::Linux::get_node_by_cpu(int cpu_id) {
2948   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2949     return cpu_to_node()->at(cpu_id);
2950   }
2951   return -1;
2952 }
2953 
2954 GrowableArray<int>* os::Linux::_cpu_to_node;
2955 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2956 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2957 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2958 os::Linux::numa_available_func_t os::Linux::_numa_available;
2959 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2960 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2961 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2962 unsigned long* os::Linux::_numa_all_nodes;
2963 
2964 bool os::pd_uncommit_memory(char* addr, size_t size) {
2965   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2966                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2967   return res  != (uintptr_t) MAP_FAILED;
2968 }
2969 
2970 static
2971 address get_stack_commited_bottom(address bottom, size_t size) {
2972   address nbot = bottom;
2973   address ntop = bottom + size;
2974 
2975   size_t page_sz = os::vm_page_size();
2976   unsigned pages = size / page_sz;
2977 
2978   unsigned char vec[1];
2979   unsigned imin = 1, imax = pages + 1, imid;
2980   int mincore_return_value = 0;
2981 
2982   assert(imin <= imax, "Unexpected page size");
2983 
2984   while (imin < imax) {
2985     imid = (imax + imin) / 2;
2986     nbot = ntop - (imid * page_sz);
2987 
2988     // Use a trick with mincore to check whether the page is mapped or not.
2989     // mincore sets vec to 1 if page resides in memory and to 0 if page
2990     // is swapped output but if page we are asking for is unmapped
2991     // it returns -1,ENOMEM
2992     mincore_return_value = mincore(nbot, page_sz, vec);
2993 
2994     if (mincore_return_value == -1) {
2995       // Page is not mapped go up
2996       // to find first mapped page
2997       if (errno != EAGAIN) {
2998         assert(errno == ENOMEM, "Unexpected mincore errno");
2999         imax = imid;
3000       }
3001     } else {
3002       // Page is mapped go down
3003       // to find first not mapped page
3004       imin = imid + 1;
3005     }
3006   }
3007 
3008   nbot = nbot + page_sz;
3009 
3010   // Adjust stack bottom one page up if last checked page is not mapped
3011   if (mincore_return_value == -1) {
3012     nbot = nbot + page_sz;
3013   }
3014 
3015   return nbot;
3016 }
3017 
3018 
3019 // Linux uses a growable mapping for the stack, and if the mapping for
3020 // the stack guard pages is not removed when we detach a thread the
3021 // stack cannot grow beyond the pages where the stack guard was
3022 // mapped.  If at some point later in the process the stack expands to
3023 // that point, the Linux kernel cannot expand the stack any further
3024 // because the guard pages are in the way, and a segfault occurs.
3025 //
3026 // However, it's essential not to split the stack region by unmapping
3027 // a region (leaving a hole) that's already part of the stack mapping,
3028 // so if the stack mapping has already grown beyond the guard pages at
3029 // the time we create them, we have to truncate the stack mapping.
3030 // So, we need to know the extent of the stack mapping when
3031 // create_stack_guard_pages() is called.
3032 
3033 // We only need this for stacks that are growable: at the time of
3034 // writing thread stacks don't use growable mappings (i.e. those
3035 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3036 // only applies to the main thread.
3037 
3038 // If the (growable) stack mapping already extends beyond the point
3039 // where we're going to put our guard pages, truncate the mapping at
3040 // that point by munmap()ping it.  This ensures that when we later
3041 // munmap() the guard pages we don't leave a hole in the stack
3042 // mapping. This only affects the main/initial thread
3043 
3044 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3045 
3046   if (os::Linux::is_initial_thread()) {
3047     // As we manually grow stack up to bottom inside create_attached_thread(),
3048     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3049     // we don't need to do anything special.
3050     // Check it first, before calling heavy function.
3051     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3052     unsigned char vec[1];
3053 
3054     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3055       // Fallback to slow path on all errors, including EAGAIN
3056       stack_extent = (uintptr_t) get_stack_commited_bottom(
3057                                     os::Linux::initial_thread_stack_bottom(),
3058                                     (size_t)addr - stack_extent);
3059     }
3060 
3061     if (stack_extent < (uintptr_t)addr) {
3062       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3063     }
3064   }
3065 
3066   return os::commit_memory(addr, size, !ExecMem);
3067 }
3068 
3069 // If this is a growable mapping, remove the guard pages entirely by
3070 // munmap()ping them.  If not, just call uncommit_memory(). This only
3071 // affects the main/initial thread, but guard against future OS changes
3072 // It's safe to always unmap guard pages for initial thread because we
3073 // always place it right after end of the mapped region
3074 
3075 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3076   uintptr_t stack_extent, stack_base;
3077 
3078   if (os::Linux::is_initial_thread()) {
3079     return ::munmap(addr, size) == 0;
3080   }
3081 
3082   return os::uncommit_memory(addr, size);
3083 }
3084 
3085 static address _highest_vm_reserved_address = NULL;
3086 
3087 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3088 // at 'requested_addr'. If there are existing memory mappings at the same
3089 // location, however, they will be overwritten. If 'fixed' is false,
3090 // 'requested_addr' is only treated as a hint, the return value may or
3091 // may not start from the requested address. Unlike Linux mmap(), this
3092 // function returns NULL to indicate failure.
3093 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3094   char * addr;
3095   int flags;
3096 
3097   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3098   if (fixed) {
3099     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3100     flags |= MAP_FIXED;
3101   }
3102 
3103   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3104   // touch an uncommitted page. Otherwise, the read/write might
3105   // succeed if we have enough swap space to back the physical page.
3106   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3107                        flags, -1, 0);
3108 
3109   if (addr != MAP_FAILED) {
3110     // anon_mmap() should only get called during VM initialization,
3111     // don't need lock (actually we can skip locking even it can be called
3112     // from multiple threads, because _highest_vm_reserved_address is just a
3113     // hint about the upper limit of non-stack memory regions.)
3114     if ((address)addr + bytes > _highest_vm_reserved_address) {
3115       _highest_vm_reserved_address = (address)addr + bytes;
3116     }
3117   }
3118 
3119   return addr == MAP_FAILED ? NULL : addr;
3120 }
3121 
3122 // Don't update _highest_vm_reserved_address, because there might be memory
3123 // regions above addr + size. If so, releasing a memory region only creates
3124 // a hole in the address space, it doesn't help prevent heap-stack collision.
3125 //
3126 static int anon_munmap(char * addr, size_t size) {
3127   return ::munmap(addr, size) == 0;
3128 }
3129 
3130 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3131                          size_t alignment_hint) {
3132   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3133 }
3134 
3135 bool os::pd_release_memory(char* addr, size_t size) {
3136   return anon_munmap(addr, size);
3137 }
3138 
3139 static address highest_vm_reserved_address() {
3140   return _highest_vm_reserved_address;
3141 }
3142 
3143 static bool linux_mprotect(char* addr, size_t size, int prot) {
3144   // Linux wants the mprotect address argument to be page aligned.
3145   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3146 
3147   // According to SUSv3, mprotect() should only be used with mappings
3148   // established by mmap(), and mmap() always maps whole pages. Unaligned
3149   // 'addr' likely indicates problem in the VM (e.g. trying to change
3150   // protection of malloc'ed or statically allocated memory). Check the
3151   // caller if you hit this assert.
3152   assert(addr == bottom, "sanity check");
3153 
3154   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3155   return ::mprotect(bottom, size, prot) == 0;
3156 }
3157 
3158 // Set protections specified
3159 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3160                         bool is_committed) {
3161   unsigned int p = 0;
3162   switch (prot) {
3163   case MEM_PROT_NONE: p = PROT_NONE; break;
3164   case MEM_PROT_READ: p = PROT_READ; break;
3165   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3166   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3167   default:
3168     ShouldNotReachHere();
3169   }
3170   // is_committed is unused.
3171   return linux_mprotect(addr, bytes, p);
3172 }
3173 
3174 bool os::guard_memory(char* addr, size_t size) {
3175   return linux_mprotect(addr, size, PROT_NONE);
3176 }
3177 
3178 bool os::unguard_memory(char* addr, size_t size) {
3179   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3180 }
3181 
3182 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3183   bool result = false;
3184   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3185                  MAP_ANONYMOUS|MAP_PRIVATE,
3186                  -1, 0);
3187   if (p != MAP_FAILED) {
3188     void *aligned_p = align_ptr_up(p, page_size);
3189 
3190     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3191 
3192     munmap(p, page_size * 2);
3193   }
3194 
3195   if (warn && !result) {
3196     warning("TransparentHugePages is not supported by the operating system.");
3197   }
3198 
3199   return result;
3200 }
3201 
3202 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3203   bool result = false;
3204   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3205                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3206                  -1, 0);
3207 
3208   if (p != MAP_FAILED) {
3209     // We don't know if this really is a huge page or not.
3210     FILE *fp = fopen("/proc/self/maps", "r");
3211     if (fp) {
3212       while (!feof(fp)) {
3213         char chars[257];
3214         long x = 0;
3215         if (fgets(chars, sizeof(chars), fp)) {
3216           if (sscanf(chars, "%lx-%*x", &x) == 1
3217               && x == (long)p) {
3218             if (strstr (chars, "hugepage")) {
3219               result = true;
3220               break;
3221             }
3222           }
3223         }
3224       }
3225       fclose(fp);
3226     }
3227     munmap(p, page_size);
3228   }
3229 
3230   if (warn && !result) {
3231     warning("HugeTLBFS is not supported by the operating system.");
3232   }
3233 
3234   return result;
3235 }
3236 
3237 /*
3238 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3239 *
3240 * From the coredump_filter documentation:
3241 *
3242 * - (bit 0) anonymous private memory
3243 * - (bit 1) anonymous shared memory
3244 * - (bit 2) file-backed private memory
3245 * - (bit 3) file-backed shared memory
3246 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3247 *           effective only if the bit 2 is cleared)
3248 * - (bit 5) hugetlb private memory
3249 * - (bit 6) hugetlb shared memory
3250 */
3251 static void set_coredump_filter(void) {
3252   FILE *f;
3253   long cdm;
3254 
3255   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3256     return;
3257   }
3258 
3259   if (fscanf(f, "%lx", &cdm) != 1) {
3260     fclose(f);
3261     return;
3262   }
3263 
3264   rewind(f);
3265 
3266   if ((cdm & LARGEPAGES_BIT) == 0) {
3267     cdm |= LARGEPAGES_BIT;
3268     fprintf(f, "%#lx", cdm);
3269   }
3270 
3271   fclose(f);
3272 }
3273 
3274 // Large page support
3275 
3276 static size_t _large_page_size = 0;
3277 
3278 size_t os::Linux::find_large_page_size() {
3279   size_t large_page_size = 0;
3280 
3281   // large_page_size on Linux is used to round up heap size. x86 uses either
3282   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3283   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3284   // page as large as 256M.
3285   //
3286   // Here we try to figure out page size by parsing /proc/meminfo and looking
3287   // for a line with the following format:
3288   //    Hugepagesize:     2048 kB
3289   //
3290   // If we can't determine the value (e.g. /proc is not mounted, or the text
3291   // format has been changed), we'll use the largest page size supported by
3292   // the processor.
3293 
3294 #ifndef ZERO
3295   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3296                      ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3297 #endif // ZERO
3298 
3299   FILE *fp = fopen("/proc/meminfo", "r");
3300   if (fp) {
3301     while (!feof(fp)) {
3302       int x = 0;
3303       char buf[16];
3304       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3305         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3306           large_page_size = x * K;
3307           break;
3308         }
3309       } else {
3310         // skip to next line
3311         for (;;) {
3312           int ch = fgetc(fp);
3313           if (ch == EOF || ch == (int)'\n') break;
3314         }
3315       }
3316     }
3317     fclose(fp);
3318   }
3319 
3320   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3321     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3322         SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3323         proper_unit_for_byte_size(large_page_size));
3324   }
3325 
3326   return large_page_size;
3327 }
3328 
3329 size_t os::Linux::setup_large_page_size() {
3330   _large_page_size = Linux::find_large_page_size();
3331   const size_t default_page_size = (size_t)Linux::page_size();
3332   if (_large_page_size > default_page_size) {
3333     _page_sizes[0] = _large_page_size;
3334     _page_sizes[1] = default_page_size;
3335     _page_sizes[2] = 0;
3336   }
3337 
3338   return _large_page_size;
3339 }
3340 
3341 bool os::Linux::setup_large_page_type(size_t page_size) {
3342   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3343       FLAG_IS_DEFAULT(UseSHM) &&
3344       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3345 
3346     // The type of large pages has not been specified by the user.
3347 
3348     // Try UseHugeTLBFS and then UseSHM.
3349     UseHugeTLBFS = UseSHM = true;
3350 
3351     // Don't try UseTransparentHugePages since there are known
3352     // performance issues with it turned on. This might change in the future.
3353     UseTransparentHugePages = false;
3354   }
3355 
3356   if (UseTransparentHugePages) {
3357     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3358     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3359       UseHugeTLBFS = false;
3360       UseSHM = false;
3361       return true;
3362     }
3363     UseTransparentHugePages = false;
3364   }
3365 
3366   if (UseHugeTLBFS) {
3367     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3368     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3369       UseSHM = false;
3370       return true;
3371     }
3372     UseHugeTLBFS = false;
3373   }
3374 
3375   return UseSHM;
3376 }
3377 
3378 void os::large_page_init() {
3379   if (!UseLargePages &&
3380       !UseTransparentHugePages &&
3381       !UseHugeTLBFS &&
3382       !UseSHM) {
3383     // Not using large pages.
3384     return;
3385   }
3386 
3387   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3388     // The user explicitly turned off large pages.
3389     // Ignore the rest of the large pages flags.
3390     UseTransparentHugePages = false;
3391     UseHugeTLBFS = false;
3392     UseSHM = false;
3393     return;
3394   }
3395 
3396   size_t large_page_size = Linux::setup_large_page_size();
3397   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3398 
3399   set_coredump_filter();
3400 }
3401 
3402 #ifndef SHM_HUGETLB
3403 #define SHM_HUGETLB 04000
3404 #endif
3405 
3406 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3407   // "exec" is passed in but not used.  Creating the shared image for
3408   // the code cache doesn't have an SHM_X executable permission to check.
3409   assert(UseLargePages && UseSHM, "only for SHM large pages");
3410   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3411 
3412   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3413     return NULL; // Fallback to small pages.
3414   }
3415 
3416   key_t key = IPC_PRIVATE;
3417   char *addr;
3418 
3419   bool warn_on_failure = UseLargePages &&
3420                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3421                          !FLAG_IS_DEFAULT(UseSHM) ||
3422                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3423                         );
3424   char msg[128];
3425 
3426   // Create a large shared memory region to attach to based on size.
3427   // Currently, size is the total size of the heap
3428   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3429   if (shmid == -1) {
3430      // Possible reasons for shmget failure:
3431      // 1. shmmax is too small for Java heap.
3432      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3433      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3434      // 2. not enough large page memory.
3435      //    > check available large pages: cat /proc/meminfo
3436      //    > increase amount of large pages:
3437      //          echo new_value > /proc/sys/vm/nr_hugepages
3438      //      Note 1: different Linux may use different name for this property,
3439      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3440      //      Note 2: it's possible there's enough physical memory available but
3441      //            they are so fragmented after a long run that they can't
3442      //            coalesce into large pages. Try to reserve large pages when
3443      //            the system is still "fresh".
3444      if (warn_on_failure) {
3445        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3446        warning(msg);
3447      }
3448      return NULL;
3449   }
3450 
3451   // attach to the region
3452   addr = (char*)shmat(shmid, req_addr, 0);
3453   int err = errno;
3454 
3455   // Remove shmid. If shmat() is successful, the actual shared memory segment
3456   // will be deleted when it's detached by shmdt() or when the process
3457   // terminates. If shmat() is not successful this will remove the shared
3458   // segment immediately.
3459   shmctl(shmid, IPC_RMID, NULL);
3460 
3461   if ((intptr_t)addr == -1) {
3462      if (warn_on_failure) {
3463        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3464        warning(msg);
3465      }
3466      return NULL;
3467   }
3468 
3469   return addr;
3470 }
3471 
3472 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3473   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3474 
3475   bool warn_on_failure = UseLargePages &&
3476       (!FLAG_IS_DEFAULT(UseLargePages) ||
3477        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3478        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3479 
3480   if (warn_on_failure) {
3481     char msg[128];
3482     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3483         PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3484     warning(msg);
3485   }
3486 }
3487 
3488 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3489   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3490   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3491   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3492 
3493   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3494   char* addr = (char*)::mmap(req_addr, bytes, prot,
3495                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3496                              -1, 0);
3497 
3498   if (addr == MAP_FAILED) {
3499     warn_on_large_pages_failure(req_addr, bytes, errno);
3500     return NULL;
3501   }
3502 
3503   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3504 
3505   return addr;
3506 }
3507 
3508 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3509   size_t large_page_size = os::large_page_size();
3510 
3511   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3512 
3513   // Allocate small pages.
3514 
3515   char* start;
3516   if (req_addr != NULL) {
3517     assert(is_ptr_aligned(req_addr, alignment), "Must be");
3518     assert(is_size_aligned(bytes, alignment), "Must be");
3519     start = os::reserve_memory(bytes, req_addr);
3520     assert(start == NULL || start == req_addr, "Must be");
3521   } else {
3522     start = os::reserve_memory_aligned(bytes, alignment);
3523   }
3524 
3525   if (start == NULL) {
3526     return NULL;
3527   }
3528 
3529   assert(is_ptr_aligned(start, alignment), "Must be");
3530 
3531   // os::reserve_memory_special will record this memory area.
3532   // Need to release it here to prevent overlapping reservations.
3533   MemTracker::record_virtual_memory_release((address)start, bytes);
3534 
3535   char* end = start + bytes;
3536 
3537   // Find the regions of the allocated chunk that can be promoted to large pages.
3538   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3539   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3540 
3541   size_t lp_bytes = lp_end - lp_start;
3542 
3543   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3544 
3545   if (lp_bytes == 0) {
3546     // The mapped region doesn't even span the start and the end of a large page.
3547     // Fall back to allocate a non-special area.
3548     ::munmap(start, end - start);
3549     return NULL;
3550   }
3551 
3552   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3553 
3554 
3555   void* result;
3556 
3557   if (start != lp_start) {
3558     result = ::mmap(start, lp_start - start, prot,
3559                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3560                     -1, 0);
3561     if (result == MAP_FAILED) {
3562       ::munmap(lp_start, end - lp_start);
3563       return NULL;
3564     }
3565   }
3566 
3567   result = ::mmap(lp_start, lp_bytes, prot,
3568                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3569                   -1, 0);
3570   if (result == MAP_FAILED) {
3571     warn_on_large_pages_failure(req_addr, bytes, errno);
3572     // If the mmap above fails, the large pages region will be unmapped and we
3573     // have regions before and after with small pages. Release these regions.
3574     //
3575     // |  mapped  |  unmapped  |  mapped  |
3576     // ^          ^            ^          ^
3577     // start      lp_start     lp_end     end
3578     //
3579     ::munmap(start, lp_start - start);
3580     ::munmap(lp_end, end - lp_end);
3581     return NULL;
3582   }
3583 
3584   if (lp_end != end) {
3585       result = ::mmap(lp_end, end - lp_end, prot,
3586                       MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3587                       -1, 0);
3588     if (result == MAP_FAILED) {
3589       ::munmap(start, lp_end - start);
3590       return NULL;
3591     }
3592   }
3593 
3594   return start;
3595 }
3596 
3597 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3598   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3599   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3600   assert(is_power_of_2(alignment), "Must be");
3601   assert(is_power_of_2(os::large_page_size()), "Must be");
3602   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3603 
3604   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3605     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3606   } else {
3607     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3608   }
3609 }
3610 
3611 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3612   assert(UseLargePages, "only for large pages");
3613 
3614   char* addr;
3615   if (UseSHM) {
3616     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3617   } else {
3618     assert(UseHugeTLBFS, "must be");
3619     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3620   }
3621 
3622   if (addr != NULL) {
3623     if (UseNUMAInterleaving) {
3624       numa_make_global(addr, bytes);
3625     }
3626 
3627     // The memory is committed
3628     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3629   }
3630 
3631   return addr;
3632 }
3633 
3634 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3635   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3636   return shmdt(base) == 0;
3637 }
3638 
3639 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3640   return pd_release_memory(base, bytes);
3641 }
3642 
3643 bool os::release_memory_special(char* base, size_t bytes) {
3644   assert(UseLargePages, "only for large pages");
3645 
3646   MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3647 
3648   bool res;
3649   if (UseSHM) {
3650     res = os::Linux::release_memory_special_shm(base, bytes);
3651   } else {
3652     assert(UseHugeTLBFS, "must be");
3653     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3654   }
3655 
3656   if (res) {
3657     tkr.record((address)base, bytes);
3658   } else {
3659     tkr.discard();
3660   }
3661 
3662   return res;
3663 }
3664 
3665 size_t os::large_page_size() {
3666   return _large_page_size;
3667 }
3668 
3669 // With SysV SHM the entire memory region must be allocated as shared
3670 // memory.
3671 // HugeTLBFS allows application to commit large page memory on demand.
3672 // However, when committing memory with HugeTLBFS fails, the region
3673 // that was supposed to be committed will lose the old reservation
3674 // and allow other threads to steal that memory region. Because of this
3675 // behavior we can't commit HugeTLBFS memory.
3676 bool os::can_commit_large_page_memory() {
3677   return UseTransparentHugePages;
3678 }
3679 
3680 bool os::can_execute_large_page_memory() {
3681   return UseTransparentHugePages || UseHugeTLBFS;
3682 }
3683 
3684 // Reserve memory at an arbitrary address, only if that area is
3685 // available (and not reserved for something else).
3686 
3687 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3688   const int max_tries = 10;
3689   char* base[max_tries];
3690   size_t size[max_tries];
3691   const size_t gap = 0x000000;
3692 
3693   // Assert only that the size is a multiple of the page size, since
3694   // that's all that mmap requires, and since that's all we really know
3695   // about at this low abstraction level.  If we need higher alignment,
3696   // we can either pass an alignment to this method or verify alignment
3697   // in one of the methods further up the call chain.  See bug 5044738.
3698   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3699 
3700   // Repeatedly allocate blocks until the block is allocated at the
3701   // right spot. Give up after max_tries. Note that reserve_memory() will
3702   // automatically update _highest_vm_reserved_address if the call is
3703   // successful. The variable tracks the highest memory address every reserved
3704   // by JVM. It is used to detect heap-stack collision if running with
3705   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3706   // space than needed, it could confuse the collision detecting code. To
3707   // solve the problem, save current _highest_vm_reserved_address and
3708   // calculate the correct value before return.
3709   address old_highest = _highest_vm_reserved_address;
3710 
3711   // Linux mmap allows caller to pass an address as hint; give it a try first,
3712   // if kernel honors the hint then we can return immediately.
3713   char * addr = anon_mmap(requested_addr, bytes, false);
3714   if (addr == requested_addr) {
3715      return requested_addr;
3716   }
3717 
3718   if (addr != NULL) {
3719      // mmap() is successful but it fails to reserve at the requested address
3720      anon_munmap(addr, bytes);
3721   }
3722 
3723   int i;
3724   for (i = 0; i < max_tries; ++i) {
3725     base[i] = reserve_memory(bytes);
3726 
3727     if (base[i] != NULL) {
3728       // Is this the block we wanted?
3729       if (base[i] == requested_addr) {
3730         size[i] = bytes;
3731         break;
3732       }
3733 
3734       // Does this overlap the block we wanted? Give back the overlapped
3735       // parts and try again.
3736 
3737       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3738       if (top_overlap >= 0 && top_overlap < bytes) {
3739         unmap_memory(base[i], top_overlap);
3740         base[i] += top_overlap;
3741         size[i] = bytes - top_overlap;
3742       } else {
3743         size_t bottom_overlap = base[i] + bytes - requested_addr;
3744         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3745           unmap_memory(requested_addr, bottom_overlap);
3746           size[i] = bytes - bottom_overlap;
3747         } else {
3748           size[i] = bytes;
3749         }
3750       }
3751     }
3752   }
3753 
3754   // Give back the unused reserved pieces.
3755 
3756   for (int j = 0; j < i; ++j) {
3757     if (base[j] != NULL) {
3758       unmap_memory(base[j], size[j]);
3759     }
3760   }
3761 
3762   if (i < max_tries) {
3763     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3764     return requested_addr;
3765   } else {
3766     _highest_vm_reserved_address = old_highest;
3767     return NULL;
3768   }
3769 }
3770 
3771 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3772   return ::read(fd, buf, nBytes);
3773 }
3774 
3775 //
3776 // Short sleep, direct OS call.
3777 //
3778 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3779 // sched_yield(2) will actually give up the CPU:
3780 //
3781 //   * Alone on this pariticular CPU, keeps running.
3782 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3783 //     (pre 2.6.39).
3784 //
3785 // So calling this with 0 is an alternative.
3786 //
3787 void os::naked_short_sleep(jlong ms) {
3788   struct timespec req;
3789 
3790   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3791   req.tv_sec = 0;
3792   if (ms > 0) {
3793     req.tv_nsec = (ms % 1000) * 1000000;
3794   }
3795   else {
3796     req.tv_nsec = 1;
3797   }
3798 
3799   nanosleep(&req, NULL);
3800 
3801   return;
3802 }
3803 
3804 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3805 void os::infinite_sleep() {
3806   while (true) {    // sleep forever ...
3807     ::sleep(100);   // ... 100 seconds at a time
3808   }
3809 }
3810 
3811 // Used to convert frequent JVM_Yield() to nops
3812 bool os::dont_yield() {
3813   return DontYieldALot;
3814 }
3815 
3816 void os::yield() {
3817   sched_yield();
3818 }
3819 
3820 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3821 
3822 void os::yield_all(int attempts) {
3823   // Yields to all threads, including threads with lower priorities
3824   // Threads on Linux are all with same priority. The Solaris style
3825   // os::yield_all() with nanosleep(1ms) is not necessary.
3826   sched_yield();
3827 }
3828 
3829 // Called from the tight loops to possibly influence time-sharing heuristics
3830 void os::loop_breaker(int attempts) {
3831   os::yield_all(attempts);
3832 }
3833 
3834 ////////////////////////////////////////////////////////////////////////////////
3835 // thread priority support
3836 
3837 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3838 // only supports dynamic priority, static priority must be zero. For real-time
3839 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3840 // However, for large multi-threaded applications, SCHED_RR is not only slower
3841 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3842 // of 5 runs - Sep 2005).
3843 //
3844 // The following code actually changes the niceness of kernel-thread/LWP. It
3845 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3846 // not the entire user process, and user level threads are 1:1 mapped to kernel
3847 // threads. It has always been the case, but could change in the future. For
3848 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3849 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3850 
3851 int os::java_to_os_priority[CriticalPriority + 1] = {
3852   19,              // 0 Entry should never be used
3853 
3854    4,              // 1 MinPriority
3855    3,              // 2
3856    2,              // 3
3857 
3858    1,              // 4
3859    0,              // 5 NormPriority
3860   -1,              // 6
3861 
3862   -2,              // 7
3863   -3,              // 8
3864   -4,              // 9 NearMaxPriority
3865 
3866   -5,              // 10 MaxPriority
3867 
3868   -5               // 11 CriticalPriority
3869 };
3870 
3871 static int prio_init() {
3872   if (ThreadPriorityPolicy == 1) {
3873     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3874     // if effective uid is not root. Perhaps, a more elegant way of doing
3875     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3876     if (geteuid() != 0) {
3877       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3878         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3879       }
3880       ThreadPriorityPolicy = 0;
3881     }
3882   }
3883   if (UseCriticalJavaThreadPriority) {
3884     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3885   }
3886   return 0;
3887 }
3888 
3889 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3890   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3891 
3892   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3893   return (ret == 0) ? OS_OK : OS_ERR;
3894 }
3895 
3896 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3897   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3898     *priority_ptr = java_to_os_priority[NormPriority];
3899     return OS_OK;
3900   }
3901 
3902   errno = 0;
3903   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3904   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3905 }
3906 
3907 // Hint to the underlying OS that a task switch would not be good.
3908 // Void return because it's a hint and can fail.
3909 void os::hint_no_preempt() {}
3910 
3911 ////////////////////////////////////////////////////////////////////////////////
3912 // suspend/resume support
3913 
3914 //  the low-level signal-based suspend/resume support is a remnant from the
3915 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3916 //  within hotspot. Now there is a single use-case for this:
3917 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3918 //      that runs in the watcher thread.
3919 //  The remaining code is greatly simplified from the more general suspension
3920 //  code that used to be used.
3921 //
3922 //  The protocol is quite simple:
3923 //  - suspend:
3924 //      - sends a signal to the target thread
3925 //      - polls the suspend state of the osthread using a yield loop
3926 //      - target thread signal handler (SR_handler) sets suspend state
3927 //        and blocks in sigsuspend until continued
3928 //  - resume:
3929 //      - sets target osthread state to continue
3930 //      - sends signal to end the sigsuspend loop in the SR_handler
3931 //
3932 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3933 //
3934 
3935 static void resume_clear_context(OSThread *osthread) {
3936   osthread->set_ucontext(NULL);
3937   osthread->set_siginfo(NULL);
3938 }
3939 
3940 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3941   osthread->set_ucontext(context);
3942   osthread->set_siginfo(siginfo);
3943 }
3944 
3945 //
3946 // Handler function invoked when a thread's execution is suspended or
3947 // resumed. We have to be careful that only async-safe functions are
3948 // called here (Note: most pthread functions are not async safe and
3949 // should be avoided.)
3950 //
3951 // Note: sigwait() is a more natural fit than sigsuspend() from an
3952 // interface point of view, but sigwait() prevents the signal hander
3953 // from being run. libpthread would get very confused by not having
3954 // its signal handlers run and prevents sigwait()'s use with the
3955 // mutex granting granting signal.
3956 //
3957 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3958 //
3959 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3960   // Save and restore errno to avoid confusing native code with EINTR
3961   // after sigsuspend.
3962   int old_errno = errno;
3963 
3964   Thread* thread = Thread::current();
3965   OSThread* osthread = thread->osthread();
3966   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3967 
3968   os::SuspendResume::State current = osthread->sr.state();
3969   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3970     suspend_save_context(osthread, siginfo, context);
3971 
3972     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3973     os::SuspendResume::State state = osthread->sr.suspended();
3974     if (state == os::SuspendResume::SR_SUSPENDED) {
3975       sigset_t suspend_set;  // signals for sigsuspend()
3976 
3977       // get current set of blocked signals and unblock resume signal
3978       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3979       sigdelset(&suspend_set, SR_signum);
3980 
3981       sr_semaphore.signal();
3982       // wait here until we are resumed
3983       while (1) {
3984         sigsuspend(&suspend_set);
3985 
3986         os::SuspendResume::State result = osthread->sr.running();
3987         if (result == os::SuspendResume::SR_RUNNING) {
3988           sr_semaphore.signal();
3989           break;
3990         }
3991       }
3992 
3993     } else if (state == os::SuspendResume::SR_RUNNING) {
3994       // request was cancelled, continue
3995     } else {
3996       ShouldNotReachHere();
3997     }
3998 
3999     resume_clear_context(osthread);
4000   } else if (current == os::SuspendResume::SR_RUNNING) {
4001     // request was cancelled, continue
4002   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4003     // ignore
4004   } else {
4005     // ignore
4006   }
4007 
4008   errno = old_errno;
4009 }
4010 
4011 
4012 static int SR_initialize() {
4013   struct sigaction act;
4014   char *s;
4015   /* Get signal number to use for suspend/resume */
4016   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4017     int sig = ::strtol(s, 0, 10);
4018     if (sig > 0 || sig < _NSIG) {
4019         SR_signum = sig;
4020     }
4021   }
4022 
4023   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4024         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4025 
4026   sigemptyset(&SR_sigset);
4027   sigaddset(&SR_sigset, SR_signum);
4028 
4029   /* Set up signal handler for suspend/resume */
4030   act.sa_flags = SA_RESTART|SA_SIGINFO;
4031   act.sa_handler = (void (*)(int)) SR_handler;
4032 
4033   // SR_signum is blocked by default.
4034   // 4528190 - We also need to block pthread restart signal (32 on all
4035   // supported Linux platforms). Note that LinuxThreads need to block
4036   // this signal for all threads to work properly. So we don't have
4037   // to use hard-coded signal number when setting up the mask.
4038   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4039 
4040   if (sigaction(SR_signum, &act, 0) == -1) {
4041     return -1;
4042   }
4043 
4044   // Save signal flag
4045   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4046   return 0;
4047 }
4048 
4049 static int sr_notify(OSThread* osthread) {
4050   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4051   assert_status(status == 0, status, "pthread_kill");
4052   return status;
4053 }
4054 
4055 // "Randomly" selected value for how long we want to spin
4056 // before bailing out on suspending a thread, also how often
4057 // we send a signal to a thread we want to resume
4058 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4059 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4060 
4061 // returns true on success and false on error - really an error is fatal
4062 // but this seems the normal response to library errors
4063 static bool do_suspend(OSThread* osthread) {
4064   assert(osthread->sr.is_running(), "thread should be running");
4065   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4066 
4067   // mark as suspended and send signal
4068   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4069     // failed to switch, state wasn't running?
4070     ShouldNotReachHere();
4071     return false;
4072   }
4073 
4074   if (sr_notify(osthread) != 0) {
4075     ShouldNotReachHere();
4076   }
4077 
4078   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4079   while (true) {
4080     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4081       break;
4082     } else {
4083       // timeout
4084       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4085       if (cancelled == os::SuspendResume::SR_RUNNING) {
4086         return false;
4087       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4088         // make sure that we consume the signal on the semaphore as well
4089         sr_semaphore.wait();
4090         break;
4091       } else {
4092         ShouldNotReachHere();
4093         return false;
4094       }
4095     }
4096   }
4097 
4098   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4099   return true;
4100 }
4101 
4102 static void do_resume(OSThread* osthread) {
4103   assert(osthread->sr.is_suspended(), "thread should be suspended");
4104   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4105 
4106   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4107     // failed to switch to WAKEUP_REQUEST
4108     ShouldNotReachHere();
4109     return;
4110   }
4111 
4112   while (true) {
4113     if (sr_notify(osthread) == 0) {
4114       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4115         if (osthread->sr.is_running()) {
4116           return;
4117         }
4118       }
4119     } else {
4120       ShouldNotReachHere();
4121     }
4122   }
4123 
4124   guarantee(osthread->sr.is_running(), "Must be running!");
4125 }
4126 
4127 ///////////////////////////////////////////////////////////////////////////////////
4128 // signal handling (except suspend/resume)
4129 
4130 // This routine may be used by user applications as a "hook" to catch signals.
4131 // The user-defined signal handler must pass unrecognized signals to this
4132 // routine, and if it returns true (non-zero), then the signal handler must
4133 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4134 // routine will never retun false (zero), but instead will execute a VM panic
4135 // routine kill the process.
4136 //
4137 // If this routine returns false, it is OK to call it again.  This allows
4138 // the user-defined signal handler to perform checks either before or after
4139 // the VM performs its own checks.  Naturally, the user code would be making
4140 // a serious error if it tried to handle an exception (such as a null check
4141 // or breakpoint) that the VM was generating for its own correct operation.
4142 //
4143 // This routine may recognize any of the following kinds of signals:
4144 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4145 // It should be consulted by handlers for any of those signals.
4146 //
4147 // The caller of this routine must pass in the three arguments supplied
4148 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4149 // field of the structure passed to sigaction().  This routine assumes that
4150 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4151 //
4152 // Note that the VM will print warnings if it detects conflicting signal
4153 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4154 //
4155 extern "C" JNIEXPORT int
4156 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4157                         void* ucontext, int abort_if_unrecognized);
4158 
4159 void signalHandler(int sig, siginfo_t* info, void* uc) {
4160   assert(info != NULL && uc != NULL, "it must be old kernel");
4161   int orig_errno = errno;  // Preserve errno value over signal handler.
4162   JVM_handle_linux_signal(sig, info, uc, true);
4163   errno = orig_errno;
4164 }
4165 
4166 
4167 // This boolean allows users to forward their own non-matching signals
4168 // to JVM_handle_linux_signal, harmlessly.
4169 bool os::Linux::signal_handlers_are_installed = false;
4170 
4171 // For signal-chaining
4172 struct sigaction os::Linux::sigact[MAXSIGNUM];
4173 unsigned int os::Linux::sigs = 0;
4174 bool os::Linux::libjsig_is_loaded = false;
4175 typedef struct sigaction *(*get_signal_t)(int);
4176 get_signal_t os::Linux::get_signal_action = NULL;
4177 
4178 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4179   struct sigaction *actp = NULL;
4180 
4181   if (libjsig_is_loaded) {
4182     // Retrieve the old signal handler from libjsig
4183     actp = (*get_signal_action)(sig);
4184   }
4185   if (actp == NULL) {
4186     // Retrieve the preinstalled signal handler from jvm
4187     actp = get_preinstalled_handler(sig);
4188   }
4189 
4190   return actp;
4191 }
4192 
4193 static bool call_chained_handler(struct sigaction *actp, int sig,
4194                                  siginfo_t *siginfo, void *context) {
4195   // Call the old signal handler
4196   if (actp->sa_handler == SIG_DFL) {
4197     // It's more reasonable to let jvm treat it as an unexpected exception
4198     // instead of taking the default action.
4199     return false;
4200   } else if (actp->sa_handler != SIG_IGN) {
4201     if ((actp->sa_flags & SA_NODEFER) == 0) {
4202       // automaticlly block the signal
4203       sigaddset(&(actp->sa_mask), sig);
4204     }
4205 
4206     sa_handler_t hand;
4207     sa_sigaction_t sa;
4208     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4209     // retrieve the chained handler
4210     if (siginfo_flag_set) {
4211       sa = actp->sa_sigaction;
4212     } else {
4213       hand = actp->sa_handler;
4214     }
4215 
4216     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4217       actp->sa_handler = SIG_DFL;
4218     }
4219 
4220     // try to honor the signal mask
4221     sigset_t oset;
4222     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4223 
4224     // call into the chained handler
4225     if (siginfo_flag_set) {
4226       (*sa)(sig, siginfo, context);
4227     } else {
4228       (*hand)(sig);
4229     }
4230 
4231     // restore the signal mask
4232     pthread_sigmask(SIG_SETMASK, &oset, 0);
4233   }
4234   // Tell jvm's signal handler the signal is taken care of.
4235   return true;
4236 }
4237 
4238 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4239   bool chained = false;
4240   // signal-chaining
4241   if (UseSignalChaining) {
4242     struct sigaction *actp = get_chained_signal_action(sig);
4243     if (actp != NULL) {
4244       chained = call_chained_handler(actp, sig, siginfo, context);
4245     }
4246   }
4247   return chained;
4248 }
4249 
4250 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4251   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4252     return &sigact[sig];
4253   }
4254   return NULL;
4255 }
4256 
4257 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4258   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4259   sigact[sig] = oldAct;
4260   sigs |= (unsigned int)1 << sig;
4261 }
4262 
4263 // for diagnostic
4264 int os::Linux::sigflags[MAXSIGNUM];
4265 
4266 int os::Linux::get_our_sigflags(int sig) {
4267   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4268   return sigflags[sig];
4269 }
4270 
4271 void os::Linux::set_our_sigflags(int sig, int flags) {
4272   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4273   sigflags[sig] = flags;
4274 }
4275 
4276 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4277   // Check for overwrite.
4278   struct sigaction oldAct;
4279   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4280 
4281   void* oldhand = oldAct.sa_sigaction
4282                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4283                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4284   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4285       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4286       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4287     if (AllowUserSignalHandlers || !set_installed) {
4288       // Do not overwrite; user takes responsibility to forward to us.
4289       return;
4290     } else if (UseSignalChaining) {
4291       // save the old handler in jvm
4292       save_preinstalled_handler(sig, oldAct);
4293       // libjsig also interposes the sigaction() call below and saves the
4294       // old sigaction on it own.
4295     } else {
4296       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4297                     "%#lx for signal %d.", (long)oldhand, sig));
4298     }
4299   }
4300 
4301   struct sigaction sigAct;
4302   sigfillset(&(sigAct.sa_mask));
4303   sigAct.sa_handler = SIG_DFL;
4304   if (!set_installed) {
4305     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4306   } else {
4307     sigAct.sa_sigaction = signalHandler;
4308     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4309   }
4310   // Save flags, which are set by ours
4311   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4312   sigflags[sig] = sigAct.sa_flags;
4313 
4314   int ret = sigaction(sig, &sigAct, &oldAct);
4315   assert(ret == 0, "check");
4316 
4317   void* oldhand2  = oldAct.sa_sigaction
4318                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4319                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4320   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4321 }
4322 
4323 // install signal handlers for signals that HotSpot needs to
4324 // handle in order to support Java-level exception handling.
4325 
4326 void os::Linux::install_signal_handlers() {
4327   if (!signal_handlers_are_installed) {
4328     signal_handlers_are_installed = true;
4329 
4330     // signal-chaining
4331     typedef void (*signal_setting_t)();
4332     signal_setting_t begin_signal_setting = NULL;
4333     signal_setting_t end_signal_setting = NULL;
4334     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4335                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4336     if (begin_signal_setting != NULL) {
4337       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4338                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4339       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4340                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4341       libjsig_is_loaded = true;
4342       assert(UseSignalChaining, "should enable signal-chaining");
4343     }
4344     if (libjsig_is_loaded) {
4345       // Tell libjsig jvm is setting signal handlers
4346       (*begin_signal_setting)();
4347     }
4348 
4349     set_signal_handler(SIGSEGV, true);
4350     set_signal_handler(SIGPIPE, true);
4351     set_signal_handler(SIGBUS, true);
4352     set_signal_handler(SIGILL, true);
4353     set_signal_handler(SIGFPE, true);
4354 #if defined(PPC64)
4355     set_signal_handler(SIGTRAP, true);
4356 #endif
4357     set_signal_handler(SIGXFSZ, true);
4358 
4359     if (libjsig_is_loaded) {
4360       // Tell libjsig jvm finishes setting signal handlers
4361       (*end_signal_setting)();
4362     }
4363 
4364     // We don't activate signal checker if libjsig is in place, we trust ourselves
4365     // and if UserSignalHandler is installed all bets are off.
4366     // Log that signal checking is off only if -verbose:jni is specified.
4367     if (CheckJNICalls) {
4368       if (libjsig_is_loaded) {
4369         if (PrintJNIResolving) {
4370           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4371         }
4372         check_signals = false;
4373       }
4374       if (AllowUserSignalHandlers) {
4375         if (PrintJNIResolving) {
4376           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4377         }
4378         check_signals = false;
4379       }
4380     }
4381   }
4382 }
4383 
4384 // This is the fastest way to get thread cpu time on Linux.
4385 // Returns cpu time (user+sys) for any thread, not only for current.
4386 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4387 // It might work on 2.6.10+ with a special kernel/glibc patch.
4388 // For reference, please, see IEEE Std 1003.1-2004:
4389 //   http://www.unix.org/single_unix_specification
4390 
4391 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4392   struct timespec tp;
4393   int rc = os::Linux::clock_gettime(clockid, &tp);
4394   assert(rc == 0, "clock_gettime is expected to return 0 code");
4395 
4396   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4397 }
4398 
4399 /////
4400 // glibc on Linux platform uses non-documented flag
4401 // to indicate, that some special sort of signal
4402 // trampoline is used.
4403 // We will never set this flag, and we should
4404 // ignore this flag in our diagnostic
4405 #ifdef SIGNIFICANT_SIGNAL_MASK
4406 #undef SIGNIFICANT_SIGNAL_MASK
4407 #endif
4408 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4409 
4410 static const char* get_signal_handler_name(address handler,
4411                                            char* buf, int buflen) {
4412   int offset;
4413   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4414   if (found) {
4415     // skip directory names
4416     const char *p1, *p2;
4417     p1 = buf;
4418     size_t len = strlen(os::file_separator());
4419     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4420     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4421   } else {
4422     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4423   }
4424   return buf;
4425 }
4426 
4427 static void print_signal_handler(outputStream* st, int sig,
4428                                  char* buf, size_t buflen) {
4429   struct sigaction sa;
4430 
4431   sigaction(sig, NULL, &sa);
4432 
4433   // See comment for SIGNIFICANT_SIGNAL_MASK define
4434   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4435 
4436   st->print("%s: ", os::exception_name(sig, buf, buflen));
4437 
4438   address handler = (sa.sa_flags & SA_SIGINFO)
4439     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4440     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4441 
4442   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4443     st->print("SIG_DFL");
4444   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4445     st->print("SIG_IGN");
4446   } else {
4447     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4448   }
4449 
4450   st->print(", sa_mask[0]=");
4451   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4452 
4453   address rh = VMError::get_resetted_sighandler(sig);
4454   // May be, handler was resetted by VMError?
4455   if(rh != NULL) {
4456     handler = rh;
4457     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4458   }
4459 
4460   st->print(", sa_flags=");
4461   os::Posix::print_sa_flags(st, sa.sa_flags);
4462 
4463   // Check: is it our handler?
4464   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4465      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4466     // It is our signal handler
4467     // check for flags, reset system-used one!
4468     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4469       st->print(
4470                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4471                 os::Linux::get_our_sigflags(sig));
4472     }
4473   }
4474   st->cr();
4475 }
4476 
4477 
4478 #define DO_SIGNAL_CHECK(sig) \
4479   if (!sigismember(&check_signal_done, sig)) \
4480     os::Linux::check_signal_handler(sig)
4481 
4482 // This method is a periodic task to check for misbehaving JNI applications
4483 // under CheckJNI, we can add any periodic checks here
4484 
4485 void os::run_periodic_checks() {
4486 
4487   if (check_signals == false) return;
4488 
4489   // SEGV and BUS if overridden could potentially prevent
4490   // generation of hs*.log in the event of a crash, debugging
4491   // such a case can be very challenging, so we absolutely
4492   // check the following for a good measure:
4493   DO_SIGNAL_CHECK(SIGSEGV);
4494   DO_SIGNAL_CHECK(SIGILL);
4495   DO_SIGNAL_CHECK(SIGFPE);
4496   DO_SIGNAL_CHECK(SIGBUS);
4497   DO_SIGNAL_CHECK(SIGPIPE);
4498   DO_SIGNAL_CHECK(SIGXFSZ);
4499 #if defined(PPC64)
4500   DO_SIGNAL_CHECK(SIGTRAP);
4501 #endif
4502 
4503   // ReduceSignalUsage allows the user to override these handlers
4504   // see comments at the very top and jvm_solaris.h
4505   if (!ReduceSignalUsage) {
4506     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4507     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4508     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4509     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4510   }
4511 
4512   DO_SIGNAL_CHECK(SR_signum);
4513   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4514 }
4515 
4516 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4517 
4518 static os_sigaction_t os_sigaction = NULL;
4519 
4520 void os::Linux::check_signal_handler(int sig) {
4521   char buf[O_BUFLEN];
4522   address jvmHandler = NULL;
4523 
4524 
4525   struct sigaction act;
4526   if (os_sigaction == NULL) {
4527     // only trust the default sigaction, in case it has been interposed
4528     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4529     if (os_sigaction == NULL) return;
4530   }
4531 
4532   os_sigaction(sig, (struct sigaction*)NULL, &act);
4533 
4534 
4535   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4536 
4537   address thisHandler = (act.sa_flags & SA_SIGINFO)
4538     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4539     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4540 
4541 
4542   switch(sig) {
4543   case SIGSEGV:
4544   case SIGBUS:
4545   case SIGFPE:
4546   case SIGPIPE:
4547   case SIGILL:
4548   case SIGXFSZ:
4549     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4550     break;
4551 
4552   case SHUTDOWN1_SIGNAL:
4553   case SHUTDOWN2_SIGNAL:
4554   case SHUTDOWN3_SIGNAL:
4555   case BREAK_SIGNAL:
4556     jvmHandler = (address)user_handler();
4557     break;
4558 
4559   case INTERRUPT_SIGNAL:
4560     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4561     break;
4562 
4563   default:
4564     if (sig == SR_signum) {
4565       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4566     } else {
4567       return;
4568     }
4569     break;
4570   }
4571 
4572   if (thisHandler != jvmHandler) {
4573     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4574     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4575     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4576     // No need to check this sig any longer
4577     sigaddset(&check_signal_done, sig);
4578   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4579     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4580     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4581     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4582     // No need to check this sig any longer
4583     sigaddset(&check_signal_done, sig);
4584   }
4585 
4586   // Dump all the signal
4587   if (sigismember(&check_signal_done, sig)) {
4588     print_signal_handlers(tty, buf, O_BUFLEN);
4589   }
4590 }
4591 
4592 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4593 
4594 extern bool signal_name(int signo, char* buf, size_t len);
4595 
4596 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4597   if (0 < exception_code && exception_code <= SIGRTMAX) {
4598     // signal
4599     if (!signal_name(exception_code, buf, size)) {
4600       jio_snprintf(buf, size, "SIG%d", exception_code);
4601     }
4602     return buf;
4603   } else {
4604     return NULL;
4605   }
4606 }
4607 
4608 // this is called _before_ the most of global arguments have been parsed
4609 void os::init(void) {
4610   char dummy;   /* used to get a guess on initial stack address */
4611 //  first_hrtime = gethrtime();
4612 
4613   // With LinuxThreads the JavaMain thread pid (primordial thread)
4614   // is different than the pid of the java launcher thread.
4615   // So, on Linux, the launcher thread pid is passed to the VM
4616   // via the sun.java.launcher.pid property.
4617   // Use this property instead of getpid() if it was correctly passed.
4618   // See bug 6351349.
4619   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4620 
4621   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4622 
4623   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4624 
4625   init_random(1234567);
4626 
4627   ThreadCritical::initialize();
4628 
4629   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4630   if (Linux::page_size() == -1) {
4631     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4632                   strerror(errno)));
4633   }
4634   init_page_sizes((size_t) Linux::page_size());
4635 
4636   Linux::initialize_system_info();
4637 
4638   // main_thread points to the aboriginal thread
4639   Linux::_main_thread = pthread_self();
4640 
4641   Linux::clock_init();
4642   initial_time_count = javaTimeNanos();
4643 
4644   // pthread_condattr initialization for monotonic clock
4645   int status;
4646   pthread_condattr_t* _condattr = os::Linux::condAttr();
4647   if ((status = pthread_condattr_init(_condattr)) != 0) {
4648     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4649   }
4650   // Only set the clock if CLOCK_MONOTONIC is available
4651   if (os::supports_monotonic_clock()) {
4652     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4653       if (status == EINVAL) {
4654         warning("Unable to use monotonic clock with relative timed-waits" \
4655                 " - changes to the time-of-day clock may have adverse affects");
4656       } else {
4657         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4658       }
4659     }
4660   }
4661   // else it defaults to CLOCK_REALTIME
4662 
4663   pthread_mutex_init(&dl_mutex, NULL);
4664 
4665   // If the pagesize of the VM is greater than 8K determine the appropriate
4666   // number of initial guard pages.  The user can change this with the
4667   // command line arguments, if needed.
4668   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4669     StackYellowPages = 1;
4670     StackRedPages = 1;
4671     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4672   }
4673 }
4674 
4675 // To install functions for atexit system call
4676 extern "C" {
4677   static void perfMemory_exit_helper() {
4678     perfMemory_exit();
4679   }
4680 }
4681 
4682 // this is called _after_ the global arguments have been parsed
4683 jint os::init_2(void)
4684 {
4685   Linux::fast_thread_clock_init();
4686 
4687   // Allocate a single page and mark it as readable for safepoint polling
4688   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4689   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4690 
4691   os::set_polling_page( polling_page );
4692 
4693 #ifndef PRODUCT
4694   if(Verbose && PrintMiscellaneous)
4695     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4696 #endif
4697 
4698   if (!UseMembar) {
4699     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4700     guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4701     os::set_memory_serialize_page( mem_serialize_page );
4702 
4703 #ifndef PRODUCT
4704     if(Verbose && PrintMiscellaneous)
4705       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4706 #endif
4707   }
4708 
4709   // initialize suspend/resume support - must do this before signal_sets_init()
4710   if (SR_initialize() != 0) {
4711     perror("SR_initialize failed");
4712     return JNI_ERR;
4713   }
4714 
4715   Linux::signal_sets_init();
4716   Linux::install_signal_handlers();
4717 
4718   // Check minimum allowable stack size for thread creation and to initialize
4719   // the java system classes, including StackOverflowError - depends on page
4720   // size.  Add a page for compiler2 recursion in main thread.
4721   // Add in 2*BytesPerWord times page size to account for VM stack during
4722   // class initialization depending on 32 or 64 bit VM.
4723   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4724             (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4725                     (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4726 
4727   size_t threadStackSizeInBytes = ThreadStackSize * K;
4728   if (threadStackSizeInBytes != 0 &&
4729       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4730         tty->print_cr("\nThe stack size specified is too small, "
4731                       "Specify at least %dk",
4732                       os::Linux::min_stack_allowed/ K);
4733         return JNI_ERR;
4734   }
4735 
4736   // Make the stack size a multiple of the page size so that
4737   // the yellow/red zones can be guarded.
4738   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4739         vm_page_size()));
4740 
4741   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4742 
4743 #if defined(IA32)
4744   workaround_expand_exec_shield_cs_limit();
4745 #endif
4746 
4747   Linux::libpthread_init();
4748   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4749      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4750           Linux::glibc_version(), Linux::libpthread_version(),
4751           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4752   }
4753 
4754   if (UseNUMA) {
4755     if (!Linux::libnuma_init()) {
4756       UseNUMA = false;
4757     } else {
4758       if ((Linux::numa_max_node() < 1)) {
4759         // There's only one node(they start from 0), disable NUMA.
4760         UseNUMA = false;
4761       }
4762     }
4763     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4764     // we can make the adaptive lgrp chunk resizing work. If the user specified
4765     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4766     // disable adaptive resizing.
4767     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4768       if (FLAG_IS_DEFAULT(UseNUMA)) {
4769         UseNUMA = false;
4770       } else {
4771         if (FLAG_IS_DEFAULT(UseLargePages) &&
4772             FLAG_IS_DEFAULT(UseSHM) &&
4773             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4774           UseLargePages = false;
4775         } else {
4776           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4777           UseAdaptiveSizePolicy = false;
4778           UseAdaptiveNUMAChunkSizing = false;
4779         }
4780       }
4781     }
4782     if (!UseNUMA && ForceNUMA) {
4783       UseNUMA = true;
4784     }
4785   }
4786 
4787   if (MaxFDLimit) {
4788     // set the number of file descriptors to max. print out error
4789     // if getrlimit/setrlimit fails but continue regardless.
4790     struct rlimit nbr_files;
4791     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4792     if (status != 0) {
4793       if (PrintMiscellaneous && (Verbose || WizardMode))
4794         perror("os::init_2 getrlimit failed");
4795     } else {
4796       nbr_files.rlim_cur = nbr_files.rlim_max;
4797       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4798       if (status != 0) {
4799         if (PrintMiscellaneous && (Verbose || WizardMode))
4800           perror("os::init_2 setrlimit failed");
4801       }
4802     }
4803   }
4804 
4805   // Initialize lock used to serialize thread creation (see os::create_thread)
4806   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4807 
4808   // at-exit methods are called in the reverse order of their registration.
4809   // atexit functions are called on return from main or as a result of a
4810   // call to exit(3C). There can be only 32 of these functions registered
4811   // and atexit() does not set errno.
4812 
4813   if (PerfAllowAtExitRegistration) {
4814     // only register atexit functions if PerfAllowAtExitRegistration is set.
4815     // atexit functions can be delayed until process exit time, which
4816     // can be problematic for embedded VM situations. Embedded VMs should
4817     // call DestroyJavaVM() to assure that VM resources are released.
4818 
4819     // note: perfMemory_exit_helper atexit function may be removed in
4820     // the future if the appropriate cleanup code can be added to the
4821     // VM_Exit VMOperation's doit method.
4822     if (atexit(perfMemory_exit_helper) != 0) {
4823       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4824     }
4825   }
4826 
4827   // initialize thread priority policy
4828   prio_init();
4829 
4830   return JNI_OK;
4831 }
4832 
4833 // this is called at the end of vm_initialization
4834 void os::init_3(void) {
4835 #ifdef JAVASE_EMBEDDED
4836   // Start the MemNotifyThread
4837   if (LowMemoryProtection) {
4838     MemNotifyThread::start();
4839   }
4840   return;
4841 #endif
4842 }
4843 
4844 // Mark the polling page as unreadable
4845 void os::make_polling_page_unreadable(void) {
4846   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4847     fatal("Could not disable polling page");
4848 };
4849 
4850 // Mark the polling page as readable
4851 void os::make_polling_page_readable(void) {
4852   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4853     fatal("Could not enable polling page");
4854   }
4855 };
4856 
4857 int os::active_processor_count() {
4858   // Linux doesn't yet have a (official) notion of processor sets,
4859   // so just return the number of online processors.
4860   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4861   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4862   return online_cpus;
4863 }
4864 
4865 void os::set_native_thread_name(const char *name) {
4866   // Not yet implemented.
4867   return;
4868 }
4869 
4870 bool os::distribute_processes(uint length, uint* distribution) {
4871   // Not yet implemented.
4872   return false;
4873 }
4874 
4875 bool os::bind_to_processor(uint processor_id) {
4876   // Not yet implemented.
4877   return false;
4878 }
4879 
4880 ///
4881 
4882 void os::SuspendedThreadTask::internal_do_task() {
4883   if (do_suspend(_thread->osthread())) {
4884     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4885     do_task(context);
4886     do_resume(_thread->osthread());
4887   }
4888 }
4889 
4890 class PcFetcher : public os::SuspendedThreadTask {
4891 public:
4892   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4893   ExtendedPC result();
4894 protected:
4895   void do_task(const os::SuspendedThreadTaskContext& context);
4896 private:
4897   ExtendedPC _epc;
4898 };
4899 
4900 ExtendedPC PcFetcher::result() {
4901   guarantee(is_done(), "task is not done yet.");
4902   return _epc;
4903 }
4904 
4905 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4906   Thread* thread = context.thread();
4907   OSThread* osthread = thread->osthread();
4908   if (osthread->ucontext() != NULL) {
4909     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4910   } else {
4911     // NULL context is unexpected, double-check this is the VMThread
4912     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4913   }
4914 }
4915 
4916 // Suspends the target using the signal mechanism and then grabs the PC before
4917 // resuming the target. Used by the flat-profiler only
4918 ExtendedPC os::get_thread_pc(Thread* thread) {
4919   // Make sure that it is called by the watcher for the VMThread
4920   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4921   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4922 
4923   PcFetcher fetcher(thread);
4924   fetcher.run();
4925   return fetcher.result();
4926 }
4927 
4928 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4929 {
4930    if (is_NPTL()) {
4931       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4932    } else {
4933       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4934       // word back to default 64bit precision if condvar is signaled. Java
4935       // wants 53bit precision.  Save and restore current value.
4936       int fpu = get_fpu_control_word();
4937       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4938       set_fpu_control_word(fpu);
4939       return status;
4940    }
4941 }
4942 
4943 ////////////////////////////////////////////////////////////////////////////////
4944 // debug support
4945 
4946 bool os::find(address addr, outputStream* st) {
4947   Dl_info dlinfo;
4948   memset(&dlinfo, 0, sizeof(dlinfo));
4949   if (dladdr(addr, &dlinfo) != 0) {
4950     st->print(PTR_FORMAT ": ", addr);
4951     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4952       st->print("%s+%#x", dlinfo.dli_sname,
4953                  addr - (intptr_t)dlinfo.dli_saddr);
4954     } else if (dlinfo.dli_fbase != NULL) {
4955       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4956     } else {
4957       st->print("<absolute address>");
4958     }
4959     if (dlinfo.dli_fname != NULL) {
4960       st->print(" in %s", dlinfo.dli_fname);
4961     }
4962     if (dlinfo.dli_fbase != NULL) {
4963       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4964     }
4965     st->cr();
4966 
4967     if (Verbose) {
4968       // decode some bytes around the PC
4969       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4970       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4971       address       lowest = (address) dlinfo.dli_sname;
4972       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4973       if (begin < lowest)  begin = lowest;
4974       Dl_info dlinfo2;
4975       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4976           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4977         end = (address) dlinfo2.dli_saddr;
4978       Disassembler::decode(begin, end, st);
4979     }
4980     return true;
4981   }
4982   return false;
4983 }
4984 
4985 ////////////////////////////////////////////////////////////////////////////////
4986 // misc
4987 
4988 // This does not do anything on Linux. This is basically a hook for being
4989 // able to use structured exception handling (thread-local exception filters)
4990 // on, e.g., Win32.
4991 void
4992 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4993                          JavaCallArguments* args, Thread* thread) {
4994   f(value, method, args, thread);
4995 }
4996 
4997 void os::print_statistics() {
4998 }
4999 
5000 int os::message_box(const char* title, const char* message) {
5001   int i;
5002   fdStream err(defaultStream::error_fd());
5003   for (i = 0; i < 78; i++) err.print_raw("=");
5004   err.cr();
5005   err.print_raw_cr(title);
5006   for (i = 0; i < 78; i++) err.print_raw("-");
5007   err.cr();
5008   err.print_raw_cr(message);
5009   for (i = 0; i < 78; i++) err.print_raw("=");
5010   err.cr();
5011 
5012   char buf[16];
5013   // Prevent process from exiting upon "read error" without consuming all CPU
5014   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5015 
5016   return buf[0] == 'y' || buf[0] == 'Y';
5017 }
5018 
5019 int os::stat(const char *path, struct stat *sbuf) {
5020   char pathbuf[MAX_PATH];
5021   if (strlen(path) > MAX_PATH - 1) {
5022     errno = ENAMETOOLONG;
5023     return -1;
5024   }
5025   os::native_path(strcpy(pathbuf, path));
5026   return ::stat(pathbuf, sbuf);
5027 }
5028 
5029 bool os::check_heap(bool force) {
5030   return true;
5031 }
5032 
5033 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5034   return ::vsnprintf(buf, count, format, args);
5035 }
5036 
5037 // Is a (classpath) directory empty?
5038 bool os::dir_is_empty(const char* path) {
5039   DIR *dir = NULL;
5040   struct dirent *ptr;
5041 
5042   dir = opendir(path);
5043   if (dir == NULL) return true;
5044 
5045   /* Scan the directory */
5046   bool result = true;
5047   char buf[sizeof(struct dirent) + MAX_PATH];
5048   while (result && (ptr = ::readdir(dir)) != NULL) {
5049     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5050       result = false;
5051     }
5052   }
5053   closedir(dir);
5054   return result;
5055 }
5056 
5057 // This code originates from JDK's sysOpen and open64_w
5058 // from src/solaris/hpi/src/system_md.c
5059 
5060 #ifndef O_DELETE
5061 #define O_DELETE 0x10000
5062 #endif
5063 
5064 // Open a file. Unlink the file immediately after open returns
5065 // if the specified oflag has the O_DELETE flag set.
5066 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5067 
5068 int os::open(const char *path, int oflag, int mode) {
5069 
5070   if (strlen(path) > MAX_PATH - 1) {
5071     errno = ENAMETOOLONG;
5072     return -1;
5073   }
5074   int fd;
5075   int o_delete = (oflag & O_DELETE);
5076   oflag = oflag & ~O_DELETE;
5077 
5078   fd = ::open64(path, oflag, mode);
5079   if (fd == -1) return -1;
5080 
5081   //If the open succeeded, the file might still be a directory
5082   {
5083     struct stat64 buf64;
5084     int ret = ::fstat64(fd, &buf64);
5085     int st_mode = buf64.st_mode;
5086 
5087     if (ret != -1) {
5088       if ((st_mode & S_IFMT) == S_IFDIR) {
5089         errno = EISDIR;
5090         ::close(fd);
5091         return -1;
5092       }
5093     } else {
5094       ::close(fd);
5095       return -1;
5096     }
5097   }
5098 
5099     /*
5100      * All file descriptors that are opened in the JVM and not
5101      * specifically destined for a subprocess should have the
5102      * close-on-exec flag set.  If we don't set it, then careless 3rd
5103      * party native code might fork and exec without closing all
5104      * appropriate file descriptors (e.g. as we do in closeDescriptors in
5105      * UNIXProcess.c), and this in turn might:
5106      *
5107      * - cause end-of-file to fail to be detected on some file
5108      *   descriptors, resulting in mysterious hangs, or
5109      *
5110      * - might cause an fopen in the subprocess to fail on a system
5111      *   suffering from bug 1085341.
5112      *
5113      * (Yes, the default setting of the close-on-exec flag is a Unix
5114      * design flaw)
5115      *
5116      * See:
5117      * 1085341: 32-bit stdio routines should support file descriptors >255
5118      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5119      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5120      */
5121 #ifdef FD_CLOEXEC
5122     {
5123         int flags = ::fcntl(fd, F_GETFD);
5124         if (flags != -1)
5125             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5126     }
5127 #endif
5128 
5129   if (o_delete != 0) {
5130     ::unlink(path);
5131   }
5132   return fd;
5133 }
5134 
5135 
5136 // create binary file, rewriting existing file if required
5137 int os::create_binary_file(const char* path, bool rewrite_existing) {
5138   int oflags = O_WRONLY | O_CREAT;
5139   if (!rewrite_existing) {
5140     oflags |= O_EXCL;
5141   }
5142   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5143 }
5144 
5145 // return current position of file pointer
5146 jlong os::current_file_offset(int fd) {
5147   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5148 }
5149 
5150 // move file pointer to the specified offset
5151 jlong os::seek_to_file_offset(int fd, jlong offset) {
5152   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5153 }
5154 
5155 // This code originates from JDK's sysAvailable
5156 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5157 
5158 int os::available(int fd, jlong *bytes) {
5159   jlong cur, end;
5160   int mode;
5161   struct stat64 buf64;
5162 
5163   if (::fstat64(fd, &buf64) >= 0) {
5164     mode = buf64.st_mode;
5165     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5166       /*
5167       * XXX: is the following call interruptible? If so, this might
5168       * need to go through the INTERRUPT_IO() wrapper as for other
5169       * blocking, interruptible calls in this file.
5170       */
5171       int n;
5172       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5173         *bytes = n;
5174         return 1;
5175       }
5176     }
5177   }
5178   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5179     return 0;
5180   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5181     return 0;
5182   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5183     return 0;
5184   }
5185   *bytes = end - cur;
5186   return 1;
5187 }
5188 
5189 int os::socket_available(int fd, jint *pbytes) {
5190   // Linux doc says EINTR not returned, unlike Solaris
5191   int ret = ::ioctl(fd, FIONREAD, pbytes);
5192 
5193   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5194   // is expected to return 0 on failure and 1 on success to the jdk.
5195   return (ret < 0) ? 0 : 1;
5196 }
5197 
5198 // Map a block of memory.
5199 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5200                      char *addr, size_t bytes, bool read_only,
5201                      bool allow_exec) {
5202   int prot;
5203   int flags = MAP_PRIVATE;
5204 
5205   if (read_only) {
5206     prot = PROT_READ;
5207   } else {
5208     prot = PROT_READ | PROT_WRITE;
5209   }
5210 
5211   if (allow_exec) {
5212     prot |= PROT_EXEC;
5213   }
5214 
5215   if (addr != NULL) {
5216     flags |= MAP_FIXED;
5217   }
5218 
5219   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5220                                      fd, file_offset);
5221   if (mapped_address == MAP_FAILED) {
5222     return NULL;
5223   }
5224   return mapped_address;
5225 }
5226 
5227 
5228 // Remap a block of memory.
5229 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5230                        char *addr, size_t bytes, bool read_only,
5231                        bool allow_exec) {
5232   // same as map_memory() on this OS
5233   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5234                         allow_exec);
5235 }
5236 
5237 
5238 // Unmap a block of memory.
5239 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5240   return munmap(addr, bytes) == 0;
5241 }
5242 
5243 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5244 
5245 static clockid_t thread_cpu_clockid(Thread* thread) {
5246   pthread_t tid = thread->osthread()->pthread_id();
5247   clockid_t clockid;
5248 
5249   // Get thread clockid
5250   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5251   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5252   return clockid;
5253 }
5254 
5255 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5256 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5257 // of a thread.
5258 //
5259 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5260 // the fast estimate available on the platform.
5261 
5262 jlong os::current_thread_cpu_time() {
5263   if (os::Linux::supports_fast_thread_cpu_time()) {
5264     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5265   } else {
5266     // return user + sys since the cost is the same
5267     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5268   }
5269 }
5270 
5271 jlong os::thread_cpu_time(Thread* thread) {
5272   // consistent with what current_thread_cpu_time() returns
5273   if (os::Linux::supports_fast_thread_cpu_time()) {
5274     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5275   } else {
5276     return slow_thread_cpu_time(thread, true /* user + sys */);
5277   }
5278 }
5279 
5280 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5281   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5282     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5283   } else {
5284     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5285   }
5286 }
5287 
5288 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5289   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5290     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5291   } else {
5292     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5293   }
5294 }
5295 
5296 //
5297 //  -1 on error.
5298 //
5299 
5300 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5301   static bool proc_task_unchecked = true;
5302   pid_t  tid = thread->osthread()->thread_id();
5303   char *s;
5304   char stat[2048];
5305   int statlen;
5306   char proc_name[64];
5307   int count;
5308   long sys_time, user_time;
5309   char cdummy;
5310   int idummy;
5311   long ldummy;
5312   FILE *fp;
5313 
5314   snprintf(proc_name, 64, "/proc/%d/stat", tid);
5315 
5316   // The /proc/<tid>/stat aggregates per-process usage on
5317   // new Linux kernels 2.6+ where NPTL is supported.
5318   // The /proc/self/task/<tid>/stat still has the per-thread usage.
5319   // See bug 6328462.
5320   // There possibly can be cases where there is no directory
5321   // /proc/self/task, so we check its availability.
5322   if (proc_task_unchecked && os::Linux::is_NPTL()) {
5323     // This is executed only once
5324     proc_task_unchecked = false;
5325     fp = fopen("/proc/self/task", "r");
5326     if (fp != NULL) {
5327       snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5328       fclose(fp);
5329     }
5330   }
5331 
5332   fp = fopen(proc_name, "r");
5333   if ( fp == NULL ) return -1;
5334   statlen = fread(stat, 1, 2047, fp);
5335   stat[statlen] = '\0';
5336   fclose(fp);
5337 
5338   // Skip pid and the command string. Note that we could be dealing with
5339   // weird command names, e.g. user could decide to rename java launcher
5340   // to "java 1.4.2 :)", then the stat file would look like
5341   //                1234 (java 1.4.2 :)) R ... ...
5342   // We don't really need to know the command string, just find the last
5343   // occurrence of ")" and then start parsing from there. See bug 4726580.
5344   s = strrchr(stat, ')');
5345   if (s == NULL ) return -1;
5346 
5347   // Skip blank chars
5348   do s++; while (isspace(*s));
5349 
5350   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5351                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5352                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5353                  &user_time, &sys_time);
5354   if ( count != 13 ) return -1;
5355   if (user_sys_cpu_time) {
5356     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5357   } else {
5358     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5359   }
5360 }
5361 
5362 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5363   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5364   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5365   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5366   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5367 }
5368 
5369 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5370   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5371   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5372   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5373   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5374 }
5375 
5376 bool os::is_thread_cpu_time_supported() {
5377   return true;
5378 }
5379 
5380 // System loadavg support.  Returns -1 if load average cannot be obtained.
5381 // Linux doesn't yet have a (official) notion of processor sets,
5382 // so just return the system wide load average.
5383 int os::loadavg(double loadavg[], int nelem) {
5384   return ::getloadavg(loadavg, nelem);
5385 }
5386 
5387 void os::pause() {
5388   char filename[MAX_PATH];
5389   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5390     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5391   } else {
5392     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5393   }
5394 
5395   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5396   if (fd != -1) {
5397     struct stat buf;
5398     ::close(fd);
5399     while (::stat(filename, &buf) == 0) {
5400       (void)::poll(NULL, 0, 100);
5401     }
5402   } else {
5403     jio_fprintf(stderr,
5404       "Could not open pause file '%s', continuing immediately.\n", filename);
5405   }
5406 }
5407 
5408 
5409 // Refer to the comments in os_solaris.cpp park-unpark.
5410 //
5411 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5412 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5413 // For specifics regarding the bug see GLIBC BUGID 261237 :
5414 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5415 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5416 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5417 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5418 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5419 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5420 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5421 // of libpthread avoids the problem, but isn't practical.
5422 //
5423 // Possible remedies:
5424 //
5425 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5426 //      This is palliative and probabilistic, however.  If the thread is preempted
5427 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5428 //      than the minimum period may have passed, and the abstime may be stale (in the
5429 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5430 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5431 //
5432 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5433 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5434 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5435 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5436 //      thread.
5437 //
5438 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5439 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5440 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5441 //      This also works well.  In fact it avoids kernel-level scalability impediments
5442 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5443 //      timers in a graceful fashion.
5444 //
5445 // 4.   When the abstime value is in the past it appears that control returns
5446 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5447 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5448 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5449 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5450 //      It may be possible to avoid reinitialization by checking the return
5451 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5452 //      condvar we must establish the invariant that cond_signal() is only called
5453 //      within critical sections protected by the adjunct mutex.  This prevents
5454 //      cond_signal() from "seeing" a condvar that's in the midst of being
5455 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5456 //      desirable signal-after-unlock optimization that avoids futile context switching.
5457 //
5458 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5459 //      structure when a condvar is used or initialized.  cond_destroy()  would
5460 //      release the helper structure.  Our reinitialize-after-timedwait fix
5461 //      put excessive stress on malloc/free and locks protecting the c-heap.
5462 //
5463 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5464 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5465 // and only enabling the work-around for vulnerable environments.
5466 
5467 // utility to compute the abstime argument to timedwait:
5468 // millis is the relative timeout time
5469 // abstime will be the absolute timeout time
5470 // TODO: replace compute_abstime() with unpackTime()
5471 
5472 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5473   if (millis < 0)  millis = 0;
5474 
5475   jlong seconds = millis / 1000;
5476   millis %= 1000;
5477   if (seconds > 50000000) { // see man cond_timedwait(3T)
5478     seconds = 50000000;
5479   }
5480 
5481   if (os::supports_monotonic_clock()) {
5482     struct timespec now;
5483     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5484     assert_status(status == 0, status, "clock_gettime");
5485     abstime->tv_sec = now.tv_sec  + seconds;
5486     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5487     if (nanos >= NANOSECS_PER_SEC) {
5488       abstime->tv_sec += 1;
5489       nanos -= NANOSECS_PER_SEC;
5490     }
5491     abstime->tv_nsec = nanos;
5492   } else {
5493     struct timeval now;
5494     int status = gettimeofday(&now, NULL);
5495     assert(status == 0, "gettimeofday");
5496     abstime->tv_sec = now.tv_sec  + seconds;
5497     long usec = now.tv_usec + millis * 1000;
5498     if (usec >= 1000000) {
5499       abstime->tv_sec += 1;
5500       usec -= 1000000;
5501     }
5502     abstime->tv_nsec = usec * 1000;
5503   }
5504   return abstime;
5505 }
5506 
5507 
5508 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5509 // Conceptually TryPark() should be equivalent to park(0).
5510 
5511 int os::PlatformEvent::TryPark() {
5512   for (;;) {
5513     const int v = _Event ;
5514     guarantee ((v == 0) || (v == 1), "invariant") ;
5515     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5516   }
5517 }
5518 
5519 void os::PlatformEvent::park() {       // AKA "down()"
5520   // Invariant: Only the thread associated with the Event/PlatformEvent
5521   // may call park().
5522   // TODO: assert that _Assoc != NULL or _Assoc == Self
5523   int v ;
5524   for (;;) {
5525       v = _Event ;
5526       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5527   }
5528   guarantee (v >= 0, "invariant") ;
5529   if (v == 0) {
5530      // Do this the hard way by blocking ...
5531      int status = pthread_mutex_lock(_mutex);
5532      assert_status(status == 0, status, "mutex_lock");
5533      guarantee (_nParked == 0, "invariant") ;
5534      ++ _nParked ;
5535      while (_Event < 0) {
5536         status = pthread_cond_wait(_cond, _mutex);
5537         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5538         // Treat this the same as if the wait was interrupted
5539         if (status == ETIME) { status = EINTR; }
5540         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5541      }
5542      -- _nParked ;
5543 
5544     _Event = 0 ;
5545      status = pthread_mutex_unlock(_mutex);
5546      assert_status(status == 0, status, "mutex_unlock");
5547     // Paranoia to ensure our locked and lock-free paths interact
5548     // correctly with each other.
5549     OrderAccess::fence();
5550   }
5551   guarantee (_Event >= 0, "invariant") ;
5552 }
5553 
5554 int os::PlatformEvent::park(jlong millis) {
5555   guarantee (_nParked == 0, "invariant") ;
5556 
5557   int v ;
5558   for (;;) {
5559       v = _Event ;
5560       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5561   }
5562   guarantee (v >= 0, "invariant") ;
5563   if (v != 0) return OS_OK ;
5564 
5565   // We do this the hard way, by blocking the thread.
5566   // Consider enforcing a minimum timeout value.
5567   struct timespec abst;
5568   compute_abstime(&abst, millis);
5569 
5570   int ret = OS_TIMEOUT;
5571   int status = pthread_mutex_lock(_mutex);
5572   assert_status(status == 0, status, "mutex_lock");
5573   guarantee (_nParked == 0, "invariant") ;
5574   ++_nParked ;
5575 
5576   // Object.wait(timo) will return because of
5577   // (a) notification
5578   // (b) timeout
5579   // (c) thread.interrupt
5580   //
5581   // Thread.interrupt and object.notify{All} both call Event::set.
5582   // That is, we treat thread.interrupt as a special case of notification.
5583   // The underlying Solaris implementation, cond_timedwait, admits
5584   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5585   // JVM from making those visible to Java code.  As such, we must
5586   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5587   //
5588   // TODO: properly differentiate simultaneous notify+interrupt.
5589   // In that case, we should propagate the notify to another waiter.
5590 
5591   while (_Event < 0) {
5592     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5593     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5594       pthread_cond_destroy (_cond);
5595       pthread_cond_init (_cond, os::Linux::condAttr()) ;
5596     }
5597     assert_status(status == 0 || status == EINTR ||
5598                   status == ETIME || status == ETIMEDOUT,
5599                   status, "cond_timedwait");
5600     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5601     if (status == ETIME || status == ETIMEDOUT) break ;
5602     // We consume and ignore EINTR and spurious wakeups.
5603   }
5604   --_nParked ;
5605   if (_Event >= 0) {
5606      ret = OS_OK;
5607   }
5608   _Event = 0 ;
5609   status = pthread_mutex_unlock(_mutex);
5610   assert_status(status == 0, status, "mutex_unlock");
5611   assert (_nParked == 0, "invariant") ;
5612   // Paranoia to ensure our locked and lock-free paths interact
5613   // correctly with each other.
5614   OrderAccess::fence();
5615   return ret;
5616 }
5617 
5618 void os::PlatformEvent::unpark() {
5619   // Transitions for _Event:
5620   //    0 :=> 1
5621   //    1 :=> 1
5622   //   -1 :=> either 0 or 1; must signal target thread
5623   //          That is, we can safely transition _Event from -1 to either
5624   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
5625   //          unpark() calls.
5626   // See also: "Semaphores in Plan 9" by Mullender & Cox
5627   //
5628   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5629   // that it will take two back-to-back park() calls for the owning
5630   // thread to block. This has the benefit of forcing a spurious return
5631   // from the first park() call after an unpark() call which will help
5632   // shake out uses of park() and unpark() without condition variables.
5633 
5634   if (Atomic::xchg(1, &_Event) >= 0) return;
5635 
5636   // Wait for the thread associated with the event to vacate
5637   int status = pthread_mutex_lock(_mutex);
5638   assert_status(status == 0, status, "mutex_lock");
5639   int AnyWaiters = _nParked;
5640   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5641   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5642     AnyWaiters = 0;
5643     pthread_cond_signal(_cond);
5644   }
5645   status = pthread_mutex_unlock(_mutex);
5646   assert_status(status == 0, status, "mutex_unlock");
5647   if (AnyWaiters != 0) {
5648     status = pthread_cond_signal(_cond);
5649     assert_status(status == 0, status, "cond_signal");
5650   }
5651 
5652   // Note that we signal() _after dropping the lock for "immortal" Events.
5653   // This is safe and avoids a common class of  futile wakeups.  In rare
5654   // circumstances this can cause a thread to return prematurely from
5655   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5656   // simply re-test the condition and re-park itself.
5657 }
5658 
5659 
5660 // JSR166
5661 // -------------------------------------------------------
5662 
5663 /*
5664  * The solaris and linux implementations of park/unpark are fairly
5665  * conservative for now, but can be improved. They currently use a
5666  * mutex/condvar pair, plus a a count.
5667  * Park decrements count if > 0, else does a condvar wait.  Unpark
5668  * sets count to 1 and signals condvar.  Only one thread ever waits
5669  * on the condvar. Contention seen when trying to park implies that someone
5670  * is unparking you, so don't wait. And spurious returns are fine, so there
5671  * is no need to track notifications.
5672  */
5673 
5674 /*
5675  * This code is common to linux and solaris and will be moved to a
5676  * common place in dolphin.
5677  *
5678  * The passed in time value is either a relative time in nanoseconds
5679  * or an absolute time in milliseconds. Either way it has to be unpacked
5680  * into suitable seconds and nanoseconds components and stored in the
5681  * given timespec structure.
5682  * Given time is a 64-bit value and the time_t used in the timespec is only
5683  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5684  * overflow if times way in the future are given. Further on Solaris versions
5685  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5686  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5687  * As it will be 28 years before "now + 100000000" will overflow we can
5688  * ignore overflow and just impose a hard-limit on seconds using the value
5689  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5690  * years from "now".
5691  */
5692 
5693 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5694   assert (time > 0, "convertTime");
5695   time_t max_secs = 0;
5696 
5697   if (!os::supports_monotonic_clock() || isAbsolute) {
5698     struct timeval now;
5699     int status = gettimeofday(&now, NULL);
5700     assert(status == 0, "gettimeofday");
5701 
5702     max_secs = now.tv_sec + MAX_SECS;
5703 
5704     if (isAbsolute) {
5705       jlong secs = time / 1000;
5706       if (secs > max_secs) {
5707         absTime->tv_sec = max_secs;
5708       } else {
5709         absTime->tv_sec = secs;
5710       }
5711       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5712     } else {
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_usec*1000;
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   } else {
5727     // must be relative using monotonic clock
5728     struct timespec now;
5729     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5730     assert_status(status == 0, status, "clock_gettime");
5731     max_secs = now.tv_sec + MAX_SECS;
5732     jlong secs = time / NANOSECS_PER_SEC;
5733     if (secs >= MAX_SECS) {
5734       absTime->tv_sec = max_secs;
5735       absTime->tv_nsec = 0;
5736     } else {
5737       absTime->tv_sec = now.tv_sec + secs;
5738       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5739       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5740         absTime->tv_nsec -= NANOSECS_PER_SEC;
5741         ++absTime->tv_sec; // note: this must be <= max_secs
5742       }
5743     }
5744   }
5745   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5746   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5747   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5748   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5749 }
5750 
5751 void Parker::park(bool isAbsolute, jlong time) {
5752   // Ideally we'd do something useful while spinning, such
5753   // as calling unpackTime().
5754 
5755   // Optional fast-path check:
5756   // Return immediately if a permit is available.
5757   // We depend on Atomic::xchg() having full barrier semantics
5758   // since we are doing a lock-free update to _counter.
5759   if (Atomic::xchg(0, &_counter) > 0) return;
5760 
5761   Thread* thread = Thread::current();
5762   assert(thread->is_Java_thread(), "Must be JavaThread");
5763   JavaThread *jt = (JavaThread *)thread;
5764 
5765   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5766   // Check interrupt before trying to wait
5767   if (Thread::is_interrupted(thread, false)) {
5768     return;
5769   }
5770 
5771   // Next, demultiplex/decode time arguments
5772   timespec absTime;
5773   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5774     return;
5775   }
5776   if (time > 0) {
5777     unpackTime(&absTime, isAbsolute, time);
5778   }
5779 
5780 
5781   // Enter safepoint region
5782   // Beware of deadlocks such as 6317397.
5783   // The per-thread Parker:: mutex is a classic leaf-lock.
5784   // In particular a thread must never block on the Threads_lock while
5785   // holding the Parker:: mutex.  If safepoints are pending both the
5786   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5787   ThreadBlockInVM tbivm(jt);
5788 
5789   // Don't wait if cannot get lock since interference arises from
5790   // unblocking.  Also. check interrupt before trying wait
5791   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5792     return;
5793   }
5794 
5795   int status ;
5796   if (_counter > 0)  { // no wait needed
5797     _counter = 0;
5798     status = pthread_mutex_unlock(_mutex);
5799     assert (status == 0, "invariant") ;
5800     // Paranoia to ensure our locked and lock-free paths interact
5801     // correctly with each other and Java-level accesses.
5802     OrderAccess::fence();
5803     return;
5804   }
5805 
5806 #ifdef ASSERT
5807   // Don't catch signals while blocked; let the running threads have the signals.
5808   // (This allows a debugger to break into the running thread.)
5809   sigset_t oldsigs;
5810   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5811   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5812 #endif
5813 
5814   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5815   jt->set_suspend_equivalent();
5816   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5817 
5818   assert(_cur_index == -1, "invariant");
5819   if (time == 0) {
5820     _cur_index = REL_INDEX; // arbitrary choice when not timed
5821     status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5822   } else {
5823     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5824     status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5825     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5826       pthread_cond_destroy (&_cond[_cur_index]) ;
5827       pthread_cond_init    (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5828     }
5829   }
5830   _cur_index = -1;
5831   assert_status(status == 0 || status == EINTR ||
5832                 status == ETIME || status == ETIMEDOUT,
5833                 status, "cond_timedwait");
5834 
5835 #ifdef ASSERT
5836   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5837 #endif
5838 
5839   _counter = 0 ;
5840   status = pthread_mutex_unlock(_mutex) ;
5841   assert_status(status == 0, status, "invariant") ;
5842   // Paranoia to ensure our locked and lock-free paths interact
5843   // correctly with each other and Java-level accesses.
5844   OrderAccess::fence();
5845 
5846   // If externally suspended while waiting, re-suspend
5847   if (jt->handle_special_suspend_equivalent_condition()) {
5848     jt->java_suspend_self();
5849   }
5850 }
5851 
5852 void Parker::unpark() {
5853   int s, status ;
5854   status = pthread_mutex_lock(_mutex);
5855   assert (status == 0, "invariant") ;
5856   s = _counter;
5857   _counter = 1;
5858   if (s < 1) {
5859     // thread might be parked
5860     if (_cur_index != -1) {
5861       // thread is definitely parked
5862       if (WorkAroundNPTLTimedWaitHang) {
5863         status = pthread_cond_signal (&_cond[_cur_index]);
5864         assert (status == 0, "invariant");
5865         status = pthread_mutex_unlock(_mutex);
5866         assert (status == 0, "invariant");
5867       } else {
5868         status = pthread_mutex_unlock(_mutex);
5869         assert (status == 0, "invariant");
5870         status = pthread_cond_signal (&_cond[_cur_index]);
5871         assert (status == 0, "invariant");
5872       }
5873     } else {
5874       pthread_mutex_unlock(_mutex);
5875       assert (status == 0, "invariant") ;
5876     }
5877   } else {
5878     pthread_mutex_unlock(_mutex);
5879     assert (status == 0, "invariant") ;
5880   }
5881 }
5882 
5883 
5884 extern char** environ;
5885 
5886 #ifndef __NR_fork
5887 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5888 #endif
5889 
5890 #ifndef __NR_execve
5891 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5892 #endif
5893 
5894 // Run the specified command in a separate process. Return its exit value,
5895 // or -1 on failure (e.g. can't fork a new process).
5896 // Unlike system(), this function can be called from signal handler. It
5897 // doesn't block SIGINT et al.
5898 int os::fork_and_exec(char* cmd) {
5899   const char * argv[4] = {"sh", "-c", cmd, NULL};
5900 
5901   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5902   // pthread_atfork handlers and reset pthread library. All we need is a
5903   // separate process to execve. Make a direct syscall to fork process.
5904   // On IA64 there's no fork syscall, we have to use fork() and hope for
5905   // the best...
5906   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5907               IA64_ONLY(fork();)
5908 
5909   if (pid < 0) {
5910     // fork failed
5911     return -1;
5912 
5913   } else if (pid == 0) {
5914     // child process
5915 
5916     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5917     // first to kill every thread on the thread list. Because this list is
5918     // not reset by fork() (see notes above), execve() will instead kill
5919     // every thread in the parent process. We know this is the only thread
5920     // in the new process, so make a system call directly.
5921     // IA64 should use normal execve() from glibc to match the glibc fork()
5922     // above.
5923     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5924     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5925 
5926     // execve failed
5927     _exit(-1);
5928 
5929   } else  {
5930     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5931     // care about the actual exit code, for now.
5932 
5933     int status;
5934 
5935     // Wait for the child process to exit.  This returns immediately if
5936     // the child has already exited. */
5937     while (waitpid(pid, &status, 0) < 0) {
5938         switch (errno) {
5939         case ECHILD: return 0;
5940         case EINTR: break;
5941         default: return -1;
5942         }
5943     }
5944 
5945     if (WIFEXITED(status)) {
5946        // The child exited normally; get its exit code.
5947        return WEXITSTATUS(status);
5948     } else if (WIFSIGNALED(status)) {
5949        // The child exited because of a signal
5950        // The best value to return is 0x80 + signal number,
5951        // because that is what all Unix shells do, and because
5952        // it allows callers to distinguish between process exit and
5953        // process death by signal.
5954        return 0x80 + WTERMSIG(status);
5955     } else {
5956        // Unknown exit code; pass it through
5957        return status;
5958     }
5959   }
5960 }
5961 
5962 // is_headless_jre()
5963 //
5964 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5965 // in order to report if we are running in a headless jre
5966 //
5967 // Since JDK8 xawt/libmawt.so was moved into the same directory
5968 // as libawt.so, and renamed libawt_xawt.so
5969 //
5970 bool os::is_headless_jre() {
5971     struct stat statbuf;
5972     char buf[MAXPATHLEN];
5973     char libmawtpath[MAXPATHLEN];
5974     const char *xawtstr  = "/xawt/libmawt.so";
5975     const char *new_xawtstr = "/libawt_xawt.so";
5976     char *p;
5977 
5978     // Get path to libjvm.so
5979     os::jvm_path(buf, sizeof(buf));
5980 
5981     // Get rid of libjvm.so
5982     p = strrchr(buf, '/');
5983     if (p == NULL) return false;
5984     else *p = '\0';
5985 
5986     // Get rid of client or server
5987     p = strrchr(buf, '/');
5988     if (p == NULL) return false;
5989     else *p = '\0';
5990 
5991     // check xawt/libmawt.so
5992     strcpy(libmawtpath, buf);
5993     strcat(libmawtpath, xawtstr);
5994     if (::stat(libmawtpath, &statbuf) == 0) return false;
5995 
5996     // check libawt_xawt.so
5997     strcpy(libmawtpath, buf);
5998     strcat(libmawtpath, new_xawtstr);
5999     if (::stat(libmawtpath, &statbuf) == 0) return false;
6000 
6001     return true;
6002 }
6003 
6004 // Get the default path to the core file
6005 // Returns the length of the string
6006 int os::get_core_path(char* buffer, size_t bufferSize) {
6007   const char* p = get_current_directory(buffer, bufferSize);
6008 
6009   if (p == NULL) {
6010     assert(p != NULL, "failed to get current directory");
6011     return 0;
6012   }
6013 
6014   return strlen(buffer);
6015 }
6016 
6017 #ifdef JAVASE_EMBEDDED
6018 //
6019 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6020 //
6021 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6022 
6023 // ctor
6024 //
6025 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6026   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6027   _fd = fd;
6028 
6029   if (os::create_thread(this, os::os_thread)) {
6030     _memnotify_thread = this;
6031     os::set_priority(this, NearMaxPriority);
6032     os::start_thread(this);
6033   }
6034 }
6035 
6036 // Where all the work gets done
6037 //
6038 void MemNotifyThread::run() {
6039   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6040 
6041   // Set up the select arguments
6042   fd_set rfds;
6043   if (_fd != -1) {
6044     FD_ZERO(&rfds);
6045     FD_SET(_fd, &rfds);
6046   }
6047 
6048   // Now wait for the mem_notify device to wake up
6049   while (1) {
6050     // Wait for the mem_notify device to signal us..
6051     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6052     if (rc == -1) {
6053       perror("select!\n");
6054       break;
6055     } else if (rc) {
6056       //ssize_t free_before = os::available_memory();
6057       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6058 
6059       // The kernel is telling us there is not much memory left...
6060       // try to do something about that
6061 
6062       // If we are not already in a GC, try one.
6063       if (!Universe::heap()->is_gc_active()) {
6064         Universe::heap()->collect(GCCause::_allocation_failure);
6065 
6066         //ssize_t free_after = os::available_memory();
6067         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6068         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6069       }
6070       // We might want to do something like the following if we find the GC's are not helping...
6071       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6072     }
6073   }
6074 }
6075 
6076 //
6077 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6078 //
6079 void MemNotifyThread::start() {
6080   int    fd;
6081   fd = open ("/dev/mem_notify", O_RDONLY, 0);
6082   if (fd < 0) {
6083       return;
6084   }
6085 
6086   if (memnotify_thread() == NULL) {
6087     new MemNotifyThread(fd);
6088   }
6089 }
6090 
6091 #endif // JAVASE_EMBEDDED
6092 
6093 
6094 /////////////// Unit tests ///////////////
6095 
6096 #ifndef PRODUCT
6097 
6098 #define test_log(...) \
6099   do {\
6100     if (VerboseInternalVMTests) { \
6101       tty->print_cr(__VA_ARGS__); \
6102       tty->flush(); \
6103     }\
6104   } while (false)
6105 
6106 class TestReserveMemorySpecial : AllStatic {
6107  public:
6108   static void small_page_write(void* addr, size_t size) {
6109     size_t page_size = os::vm_page_size();
6110 
6111     char* end = (char*)addr + size;
6112     for (char* p = (char*)addr; p < end; p += page_size) {
6113       *p = 1;
6114     }
6115   }
6116 
6117   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6118     if (!UseHugeTLBFS) {
6119       return;
6120     }
6121 
6122     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6123 
6124     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6125 
6126     if (addr != NULL) {
6127       small_page_write(addr, size);
6128 
6129       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6130     }
6131   }
6132 
6133   static void test_reserve_memory_special_huge_tlbfs_only() {
6134     if (!UseHugeTLBFS) {
6135       return;
6136     }
6137 
6138     size_t lp = os::large_page_size();
6139 
6140     for (size_t size = lp; size <= lp * 10; size += lp) {
6141       test_reserve_memory_special_huge_tlbfs_only(size);
6142     }
6143   }
6144 
6145   static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6146     if (!UseHugeTLBFS) {
6147         return;
6148     }
6149 
6150     test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6151         size, alignment);
6152 
6153     assert(size >= os::large_page_size(), "Incorrect input to test");
6154 
6155     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6156 
6157     if (addr != NULL) {
6158       small_page_write(addr, size);
6159 
6160       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6161     }
6162   }
6163 
6164   static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6165     size_t lp = os::large_page_size();
6166     size_t ag = os::vm_allocation_granularity();
6167 
6168     for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6169       test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6170     }
6171   }
6172 
6173   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6174     size_t lp = os::large_page_size();
6175     size_t ag = os::vm_allocation_granularity();
6176 
6177     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6178     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6179     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6180     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6181     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6182     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6183     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6184     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6185     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6186   }
6187 
6188   static void test_reserve_memory_special_huge_tlbfs() {
6189     if (!UseHugeTLBFS) {
6190       return;
6191     }
6192 
6193     test_reserve_memory_special_huge_tlbfs_only();
6194     test_reserve_memory_special_huge_tlbfs_mixed();
6195   }
6196 
6197   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6198     if (!UseSHM) {
6199       return;
6200     }
6201 
6202     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6203 
6204     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6205 
6206     if (addr != NULL) {
6207       assert(is_ptr_aligned(addr, alignment), "Check");
6208       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6209 
6210       small_page_write(addr, size);
6211 
6212       os::Linux::release_memory_special_shm(addr, size);
6213     }
6214   }
6215 
6216   static void test_reserve_memory_special_shm() {
6217     size_t lp = os::large_page_size();
6218     size_t ag = os::vm_allocation_granularity();
6219 
6220     for (size_t size = ag; size < lp * 3; size += ag) {
6221       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6222         test_reserve_memory_special_shm(size, alignment);
6223       }
6224     }
6225   }
6226 
6227   static void test() {
6228     test_reserve_memory_special_huge_tlbfs();
6229     test_reserve_memory_special_shm();
6230   }
6231 };
6232 
6233 void TestReserveMemorySpecial_test() {
6234   TestReserveMemorySpecial::test();
6235 }
6236 
6237 #endif