1 /*
   2  * Copyright (c) 1999, 2011, 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 "interpreter/interpreter.hpp"
  33 #include "jvm_linux.h"
  34 #include "memory/allocation.inline.hpp"
  35 #include "memory/filemap.hpp"
  36 #include "mutex_linux.inline.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "os_share_linux.hpp"
  39 #include "prims/jniFastGetField.hpp"
  40 #include "prims/jvm.h"
  41 #include "prims/jvm_misc.hpp"
  42 #include "runtime/arguments.hpp"
  43 #include "runtime/extendedPC.hpp"
  44 #include "runtime/globals.hpp"
  45 #include "runtime/interfaceSupport.hpp"
  46 #include "runtime/java.hpp"
  47 #include "runtime/javaCalls.hpp"
  48 #include "runtime/mutexLocker.hpp"
  49 #include "runtime/objectMonitor.hpp"
  50 #include "runtime/osThread.hpp"
  51 #include "runtime/perfMemory.hpp"
  52 #include "runtime/sharedRuntime.hpp"
  53 #include "runtime/statSampler.hpp"
  54 #include "runtime/stubRoutines.hpp"
  55 #include "runtime/threadCritical.hpp"
  56 #include "runtime/timer.hpp"
  57 #include "services/attachListener.hpp"
  58 #include "services/runtimeService.hpp"
  59 #include "thread_linux.inline.hpp"
  60 #include "utilities/decoder.hpp"
  61 #include "utilities/defaultStream.hpp"
  62 #include "utilities/events.hpp"
  63 #include "utilities/growableArray.hpp"
  64 #include "utilities/vmError.hpp"
  65 #ifdef TARGET_ARCH_x86
  66 # include "assembler_x86.inline.hpp"
  67 # include "nativeInst_x86.hpp"
  68 #endif
  69 #ifdef TARGET_ARCH_sparc
  70 # include "assembler_sparc.inline.hpp"
  71 # include "nativeInst_sparc.hpp"
  72 #endif
  73 #ifdef TARGET_ARCH_zero
  74 # include "assembler_zero.inline.hpp"
  75 # include "nativeInst_zero.hpp"
  76 #endif
  77 #ifdef TARGET_ARCH_arm
  78 # include "assembler_arm.inline.hpp"
  79 # include "nativeInst_arm.hpp"
  80 #endif
  81 #ifdef TARGET_ARCH_ppc
  82 # include "assembler_ppc.inline.hpp"
  83 # include "nativeInst_ppc.hpp"
  84 #endif
  85 #ifdef COMPILER1
  86 #include "c1/c1_Runtime1.hpp"
  87 #endif
  88 #ifdef COMPILER2
  89 #include "opto/runtime.hpp"
  90 #endif
  91 
  92 // put OS-includes here
  93 # include <sys/types.h>
  94 # include <sys/mman.h>
  95 # include <sys/stat.h>
  96 # include <sys/select.h>
  97 # include <pthread.h>
  98 # include <signal.h>
  99 # include <errno.h>
 100 # include <dlfcn.h>
 101 # include <stdio.h>
 102 # include <unistd.h>
 103 # include <sys/resource.h>
 104 # include <pthread.h>
 105 # include <sys/stat.h>
 106 # include <sys/time.h>
 107 # include <sys/times.h>
 108 # include <sys/utsname.h>
 109 # include <sys/socket.h>
 110 # include <sys/wait.h>
 111 # include <pwd.h>
 112 # include <poll.h>
 113 # include <semaphore.h>
 114 # include <fcntl.h>
 115 # include <string.h>
 116 # include <syscall.h>
 117 # include <sys/sysinfo.h>
 118 # include <gnu/libc-version.h>
 119 # include <sys/ipc.h>
 120 # include <sys/shm.h>
 121 # include <link.h>
 122 # include <stdint.h>
 123 # include <inttypes.h>
 124 # include <sys/ioctl.h>
 125 
 126 #define MAX_PATH    (2 * K)
 127 
 128 // for timer info max values which include all bits
 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 130 
 131 #define LARGEPAGES_BIT (1 << 6)
 132 ////////////////////////////////////////////////////////////////////////////////
 133 // global variables
 134 julong os::Linux::_physical_memory = 0;
 135 
 136 address   os::Linux::_initial_thread_stack_bottom = NULL;
 137 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 138 
 139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 141 Mutex* os::Linux::_createThread_lock = NULL;
 142 pthread_t os::Linux::_main_thread;
 143 int os::Linux::_page_size = -1;
 144 bool os::Linux::_is_floating_stack = false;
 145 bool os::Linux::_is_NPTL = false;
 146 bool os::Linux::_supports_fast_thread_cpu_time = false;
 147 const char * os::Linux::_glibc_version = NULL;
 148 const char * os::Linux::_libpthread_version = NULL;
 149 
 150 static jlong initial_time_count=0;
 151 
 152 static int clock_tics_per_sec = 100;
 153 
 154 // For diagnostics to print a message once. see run_periodic_checks
 155 static sigset_t check_signal_done;
 156 static bool check_signals = true;;
 157 
 158 static pid_t _initial_pid = 0;
 159 
 160 /* Signal number used to suspend/resume a thread */
 161 
 162 /* do not use any signal number less than SIGSEGV, see 4355769 */
 163 static int SR_signum = SIGUSR2;
 164 sigset_t SR_sigset;
 165 
 166 /* Used to protect dlsym() calls */
 167 static pthread_mutex_t dl_mutex;
 168 
 169 #ifdef JAVASE_EMBEDDED
 170 class MemNotifyThread: public Thread {
 171   friend class VMStructs;
 172  public:
 173   virtual void run();
 174 
 175  private:
 176   static MemNotifyThread* _memnotify_thread;
 177   int _fd;
 178 
 179  public:
 180 
 181   // Constructor
 182   MemNotifyThread(int fd);
 183 
 184   // Tester
 185   bool is_memnotify_thread() const { return true; }
 186 
 187   // Printing
 188   char* name() const { return (char*)"Linux MemNotify Thread"; }
 189 
 190   // Returns the single instance of the MemNotifyThread
 191   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 192 
 193   // Create and start the single instance of MemNotifyThread
 194   static void start();
 195 };
 196 #endif // JAVASE_EMBEDDED
 197 
 198 // utility functions
 199 
 200 static int SR_initialize();
 201 static int SR_finalize();
 202 
 203 julong os::available_memory() {
 204   return Linux::available_memory();
 205 }
 206 
 207 julong os::Linux::available_memory() {
 208   // values in struct sysinfo are "unsigned long"
 209   struct sysinfo si;
 210   sysinfo(&si);
 211 
 212   return (julong)si.freeram * si.mem_unit;
 213 }
 214 
 215 julong os::physical_memory() {
 216   return Linux::physical_memory();
 217 }
 218 
 219 julong os::allocatable_physical_memory(julong size) {
 220 #ifdef _LP64
 221   return size;
 222 #else
 223   julong result = MIN2(size, (julong)3800*M);
 224    if (!is_allocatable(result)) {
 225      // See comments under solaris for alignment considerations
 226      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 227      result =  MIN2(size, reasonable_size);
 228    }
 229    return result;
 230 #endif // _LP64
 231 }
 232 
 233 ////////////////////////////////////////////////////////////////////////////////
 234 // environment support
 235 
 236 bool os::getenv(const char* name, char* buf, int len) {
 237   const char* val = ::getenv(name);
 238   if (val != NULL && strlen(val) < (size_t)len) {
 239     strcpy(buf, val);
 240     return true;
 241   }
 242   if (len > 0) buf[0] = 0;  // return a null string
 243   return false;
 244 }
 245 
 246 
 247 // Return true if user is running as root.
 248 
 249 bool os::have_special_privileges() {
 250   static bool init = false;
 251   static bool privileges = false;
 252   if (!init) {
 253     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 254     init = true;
 255   }
 256   return privileges;
 257 }
 258 
 259 
 260 #ifndef SYS_gettid
 261 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 262 #ifdef __ia64__
 263 #define SYS_gettid 1105
 264 #elif __i386__
 265 #define SYS_gettid 224
 266 #elif __amd64__
 267 #define SYS_gettid 186
 268 #elif __sparc__
 269 #define SYS_gettid 143
 270 #else
 271 #error define gettid for the arch
 272 #endif
 273 #endif
 274 
 275 // Cpu architecture string
 276 #if   defined(ZERO)
 277 static char cpu_arch[] = ZERO_LIBARCH;
 278 #elif defined(IA64)
 279 static char cpu_arch[] = "ia64";
 280 #elif defined(IA32)
 281 static char cpu_arch[] = "i386";
 282 #elif defined(AMD64)
 283 static char cpu_arch[] = "amd64";
 284 #elif defined(ARM)
 285 static char cpu_arch[] = "arm";
 286 #elif defined(PPC)
 287 static char cpu_arch[] = "ppc";
 288 #elif defined(SPARC)
 289 #  ifdef _LP64
 290 static char cpu_arch[] = "sparcv9";
 291 #  else
 292 static char cpu_arch[] = "sparc";
 293 #  endif
 294 #else
 295 #error Add appropriate cpu_arch setting
 296 #endif
 297 
 298 
 299 // pid_t gettid()
 300 //
 301 // Returns the kernel thread id of the currently running thread. Kernel
 302 // thread id is used to access /proc.
 303 //
 304 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 305 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 306 //
 307 pid_t os::Linux::gettid() {
 308   int rslt = syscall(SYS_gettid);
 309   if (rslt == -1) {
 310      // old kernel, no NPTL support
 311      return getpid();
 312   } else {
 313      return (pid_t)rslt;
 314   }
 315 }
 316 
 317 // Most versions of linux have a bug where the number of processors are
 318 // determined by looking at the /proc file system.  In a chroot environment,
 319 // the system call returns 1.  This causes the VM to act as if it is
 320 // a single processor and elide locking (see is_MP() call).
 321 static bool unsafe_chroot_detected = false;
 322 static const char *unstable_chroot_error = "/proc file system not found.\n"
 323                      "Java may be unstable running multithreaded in a chroot "
 324                      "environment on Linux when /proc filesystem is not mounted.";
 325 
 326 void os::Linux::initialize_system_info() {
 327   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 328   if (processor_count() == 1) {
 329     pid_t pid = os::Linux::gettid();
 330     char fname[32];
 331     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 332     FILE *fp = fopen(fname, "r");
 333     if (fp == NULL) {
 334       unsafe_chroot_detected = true;
 335     } else {
 336       fclose(fp);
 337     }
 338   }
 339   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 340   assert(processor_count() > 0, "linux error");
 341 }
 342 
 343 void os::init_system_properties_values() {
 344 //  char arch[12];
 345 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 346 
 347   // The next steps are taken in the product version:
 348   //
 349   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 350   // This library should be located at:
 351   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 352   //
 353   // If "/jre/lib/" appears at the right place in the path, then we
 354   // assume libjvm[_g].so is installed in a JDK and we use this path.
 355   //
 356   // Otherwise exit with message: "Could not create the Java virtual machine."
 357   //
 358   // The following extra steps are taken in the debugging version:
 359   //
 360   // If "/jre/lib/" does NOT appear at the right place in the path
 361   // instead of exit check for $JAVA_HOME environment variable.
 362   //
 363   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 364   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 365   // it looks like libjvm[_g].so is installed there
 366   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 367   //
 368   // Otherwise exit.
 369   //
 370   // Important note: if the location of libjvm.so changes this
 371   // code needs to be changed accordingly.
 372 
 373   // The next few definitions allow the code to be verbatim:
 374 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 375 #define getenv(n) ::getenv(n)
 376 
 377 /*
 378  * See ld(1):
 379  *      The linker uses the following search paths to locate required
 380  *      shared libraries:
 381  *        1: ...
 382  *        ...
 383  *        7: The default directories, normally /lib and /usr/lib.
 384  */
 385 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 386 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 387 #else
 388 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 389 #endif
 390 
 391 #define EXTENSIONS_DIR  "/lib/ext"
 392 #define ENDORSED_DIR    "/lib/endorsed"
 393 #define REG_DIR         "/usr/java/packages"
 394 
 395   {
 396     /* sysclasspath, java_home, dll_dir */
 397     {
 398         char *home_path;
 399         char *dll_path;
 400         char *pslash;
 401         char buf[MAXPATHLEN];
 402         os::jvm_path(buf, sizeof(buf));
 403 
 404         // Found the full path to libjvm.so.
 405         // Now cut the path to <java_home>/jre if we can.
 406         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 407         pslash = strrchr(buf, '/');
 408         if (pslash != NULL)
 409             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 410         dll_path = malloc(strlen(buf) + 1);
 411         if (dll_path == NULL)
 412             return;
 413         strcpy(dll_path, buf);
 414         Arguments::set_dll_dir(dll_path);
 415 
 416         if (pslash != NULL) {
 417             pslash = strrchr(buf, '/');
 418             if (pslash != NULL) {
 419                 *pslash = '\0';       /* get rid of /<arch> */
 420                 pslash = strrchr(buf, '/');
 421                 if (pslash != NULL)
 422                     *pslash = '\0';   /* get rid of /lib */
 423             }
 424         }
 425 
 426         home_path = malloc(strlen(buf) + 1);
 427         if (home_path == NULL)
 428             return;
 429         strcpy(home_path, buf);
 430         Arguments::set_java_home(home_path);
 431 
 432         if (!set_boot_path('/', ':'))
 433             return;
 434     }
 435 
 436     /*
 437      * Where to look for native libraries
 438      *
 439      * Note: Due to a legacy implementation, most of the library path
 440      * is set in the launcher.  This was to accomodate linking restrictions
 441      * on legacy Linux implementations (which are no longer supported).
 442      * Eventually, all the library path setting will be done here.
 443      *
 444      * However, to prevent the proliferation of improperly built native
 445      * libraries, the new path component /usr/java/packages is added here.
 446      * Eventually, all the library path setting will be done here.
 447      */
 448     {
 449         char *ld_library_path;
 450 
 451         /*
 452          * Construct the invariant part of ld_library_path. Note that the
 453          * space for the colon and the trailing null are provided by the
 454          * nulls included by the sizeof operator (so actually we allocate
 455          * a byte more than necessary).
 456          */
 457         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 458             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 459         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 460 
 461         /*
 462          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 463          * should always exist (until the legacy problem cited above is
 464          * addressed).
 465          */
 466         char *v = getenv("LD_LIBRARY_PATH");
 467         if (v != NULL) {
 468             char *t = ld_library_path;
 469             /* That's +1 for the colon and +1 for the trailing '\0' */
 470             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 471             sprintf(ld_library_path, "%s:%s", v, t);
 472         }
 473         Arguments::set_library_path(ld_library_path);
 474     }
 475 
 476     /*
 477      * Extensions directories.
 478      *
 479      * Note that the space for the colon and the trailing null are provided
 480      * by the nulls included by the sizeof operator (so actually one byte more
 481      * than necessary is allocated).
 482      */
 483     {
 484         char *buf = malloc(strlen(Arguments::get_java_home()) +
 485             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 486         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 487             Arguments::get_java_home());
 488         Arguments::set_ext_dirs(buf);
 489     }
 490 
 491     /* Endorsed standards default directory. */
 492     {
 493         char * buf;
 494         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 495         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 496         Arguments::set_endorsed_dirs(buf);
 497     }
 498   }
 499 
 500 #undef malloc
 501 #undef getenv
 502 #undef EXTENSIONS_DIR
 503 #undef ENDORSED_DIR
 504 
 505   // Done
 506   return;
 507 }
 508 
 509 ////////////////////////////////////////////////////////////////////////////////
 510 // breakpoint support
 511 
 512 void os::breakpoint() {
 513   BREAKPOINT;
 514 }
 515 
 516 extern "C" void breakpoint() {
 517   // use debugger to set breakpoint here
 518 }
 519 
 520 ////////////////////////////////////////////////////////////////////////////////
 521 // signal support
 522 
 523 debug_only(static bool signal_sets_initialized = false);
 524 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 525 
 526 bool os::Linux::is_sig_ignored(int sig) {
 527       struct sigaction oact;
 528       sigaction(sig, (struct sigaction*)NULL, &oact);
 529       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 530                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 531       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 532            return true;
 533       else
 534            return false;
 535 }
 536 
 537 void os::Linux::signal_sets_init() {
 538   // Should also have an assertion stating we are still single-threaded.
 539   assert(!signal_sets_initialized, "Already initialized");
 540   // Fill in signals that are necessarily unblocked for all threads in
 541   // the VM. Currently, we unblock the following signals:
 542   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 543   //                         by -Xrs (=ReduceSignalUsage));
 544   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 545   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 546   // the dispositions or masks wrt these signals.
 547   // Programs embedding the VM that want to use the above signals for their
 548   // own purposes must, at this time, use the "-Xrs" option to prevent
 549   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 550   // (See bug 4345157, and other related bugs).
 551   // In reality, though, unblocking these signals is really a nop, since
 552   // these signals are not blocked by default.
 553   sigemptyset(&unblocked_sigs);
 554   sigemptyset(&allowdebug_blocked_sigs);
 555   sigaddset(&unblocked_sigs, SIGILL);
 556   sigaddset(&unblocked_sigs, SIGSEGV);
 557   sigaddset(&unblocked_sigs, SIGBUS);
 558   sigaddset(&unblocked_sigs, SIGFPE);
 559   sigaddset(&unblocked_sigs, SR_signum);
 560 
 561   if (!ReduceSignalUsage) {
 562    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 563       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 564       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 565    }
 566    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 567       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 568       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 569    }
 570    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 571       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 572       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 573    }
 574   }
 575   // Fill in signals that are blocked by all but the VM thread.
 576   sigemptyset(&vm_sigs);
 577   if (!ReduceSignalUsage)
 578     sigaddset(&vm_sigs, BREAK_SIGNAL);
 579   debug_only(signal_sets_initialized = true);
 580 
 581 }
 582 
 583 // These are signals that are unblocked while a thread is running Java.
 584 // (For some reason, they get blocked by default.)
 585 sigset_t* os::Linux::unblocked_signals() {
 586   assert(signal_sets_initialized, "Not initialized");
 587   return &unblocked_sigs;
 588 }
 589 
 590 // These are the signals that are blocked while a (non-VM) thread is
 591 // running Java. Only the VM thread handles these signals.
 592 sigset_t* os::Linux::vm_signals() {
 593   assert(signal_sets_initialized, "Not initialized");
 594   return &vm_sigs;
 595 }
 596 
 597 // These are signals that are blocked during cond_wait to allow debugger in
 598 sigset_t* os::Linux::allowdebug_blocked_signals() {
 599   assert(signal_sets_initialized, "Not initialized");
 600   return &allowdebug_blocked_sigs;
 601 }
 602 
 603 void os::Linux::hotspot_sigmask(Thread* thread) {
 604 
 605   //Save caller's signal mask before setting VM signal mask
 606   sigset_t caller_sigmask;
 607   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 608 
 609   OSThread* osthread = thread->osthread();
 610   osthread->set_caller_sigmask(caller_sigmask);
 611 
 612   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 613 
 614   if (!ReduceSignalUsage) {
 615     if (thread->is_VM_thread()) {
 616       // Only the VM thread handles BREAK_SIGNAL ...
 617       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 618     } else {
 619       // ... all other threads block BREAK_SIGNAL
 620       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 621     }
 622   }
 623 }
 624 
 625 //////////////////////////////////////////////////////////////////////////////
 626 // detecting pthread library
 627 
 628 void os::Linux::libpthread_init() {
 629   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 630   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 631   // generic name for earlier versions.
 632   // Define macros here so we can build HotSpot on old systems.
 633 # ifndef _CS_GNU_LIBC_VERSION
 634 # define _CS_GNU_LIBC_VERSION 2
 635 # endif
 636 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 637 # define _CS_GNU_LIBPTHREAD_VERSION 3
 638 # endif
 639 
 640   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 641   if (n > 0) {
 642      char *str = (char *)malloc(n);
 643      confstr(_CS_GNU_LIBC_VERSION, str, n);
 644      os::Linux::set_glibc_version(str);
 645   } else {
 646      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 647      static char _gnu_libc_version[32];
 648      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 649               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 650      os::Linux::set_glibc_version(_gnu_libc_version);
 651   }
 652 
 653   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 654   if (n > 0) {
 655      char *str = (char *)malloc(n);
 656      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 657      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 658      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 659      // is the case. LinuxThreads has a hard limit on max number of threads.
 660      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 661      // On the other hand, NPTL does not have such a limit, sysconf()
 662      // will return -1 and errno is not changed. Check if it is really NPTL.
 663      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 664          strstr(str, "NPTL") &&
 665          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 666        free(str);
 667        os::Linux::set_libpthread_version("linuxthreads");
 668      } else {
 669        os::Linux::set_libpthread_version(str);
 670      }
 671   } else {
 672     // glibc before 2.3.2 only has LinuxThreads.
 673     os::Linux::set_libpthread_version("linuxthreads");
 674   }
 675 
 676   if (strstr(libpthread_version(), "NPTL")) {
 677      os::Linux::set_is_NPTL();
 678   } else {
 679      os::Linux::set_is_LinuxThreads();
 680   }
 681 
 682   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 683   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 684   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 685      os::Linux::set_is_floating_stack();
 686   }
 687 }
 688 
 689 /////////////////////////////////////////////////////////////////////////////
 690 // thread stack
 691 
 692 // Force Linux kernel to expand current thread stack. If "bottom" is close
 693 // to the stack guard, caller should block all signals.
 694 //
 695 // MAP_GROWSDOWN:
 696 //   A special mmap() flag that is used to implement thread stacks. It tells
 697 //   kernel that the memory region should extend downwards when needed. This
 698 //   allows early versions of LinuxThreads to only mmap the first few pages
 699 //   when creating a new thread. Linux kernel will automatically expand thread
 700 //   stack as needed (on page faults).
 701 //
 702 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 703 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 704 //   region, it's hard to tell if the fault is due to a legitimate stack
 705 //   access or because of reading/writing non-exist memory (e.g. buffer
 706 //   overrun). As a rule, if the fault happens below current stack pointer,
 707 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 708 //   application (see Linux kernel fault.c).
 709 //
 710 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 711 //   stack overflow detection.
 712 //
 713 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 714 //   not use this flag. However, the stack of initial thread is not created
 715 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 716 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 717 //   and then attach the thread to JVM.
 718 //
 719 // To get around the problem and allow stack banging on Linux, we need to
 720 // manually expand thread stack after receiving the SIGSEGV.
 721 //
 722 // There are two ways to expand thread stack to address "bottom", we used
 723 // both of them in JVM before 1.5:
 724 //   1. adjust stack pointer first so that it is below "bottom", and then
 725 //      touch "bottom"
 726 //   2. mmap() the page in question
 727 //
 728 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 729 // if current sp is already near the lower end of page 101, and we need to
 730 // call mmap() to map page 100, it is possible that part of the mmap() frame
 731 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 732 // That will destroy the mmap() frame and cause VM to crash.
 733 //
 734 // The following code works by adjusting sp first, then accessing the "bottom"
 735 // page to force a page fault. Linux kernel will then automatically expand the
 736 // stack mapping.
 737 //
 738 // _expand_stack_to() assumes its frame size is less than page size, which
 739 // should always be true if the function is not inlined.
 740 
 741 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 742 #define NOINLINE
 743 #else
 744 #define NOINLINE __attribute__ ((noinline))
 745 #endif
 746 
 747 static void _expand_stack_to(address bottom) NOINLINE;
 748 
 749 static void _expand_stack_to(address bottom) {
 750   address sp;
 751   size_t size;
 752   volatile char *p;
 753 
 754   // Adjust bottom to point to the largest address within the same page, it
 755   // gives us a one-page buffer if alloca() allocates slightly more memory.
 756   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 757   bottom += os::Linux::page_size() - 1;
 758 
 759   // sp might be slightly above current stack pointer; if that's the case, we
 760   // will alloca() a little more space than necessary, which is OK. Don't use
 761   // os::current_stack_pointer(), as its result can be slightly below current
 762   // stack pointer, causing us to not alloca enough to reach "bottom".
 763   sp = (address)&sp;
 764 
 765   if (sp > bottom) {
 766     size = sp - bottom;
 767     p = (volatile char *)alloca(size);
 768     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 769     p[0] = '\0';
 770   }
 771 }
 772 
 773 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 774   assert(t!=NULL, "just checking");
 775   assert(t->osthread()->expanding_stack(), "expand should be set");
 776   assert(t->stack_base() != NULL, "stack_base was not initialized");
 777 
 778   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 779     sigset_t mask_all, old_sigset;
 780     sigfillset(&mask_all);
 781     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 782     _expand_stack_to(addr);
 783     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 784     return true;
 785   }
 786   return false;
 787 }
 788 
 789 //////////////////////////////////////////////////////////////////////////////
 790 // create new thread
 791 
 792 static address highest_vm_reserved_address();
 793 
 794 // check if it's safe to start a new thread
 795 static bool _thread_safety_check(Thread* thread) {
 796   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 797     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 798     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 799     //   allocated (MAP_FIXED) from high address space. Every thread stack
 800     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 801     //   it to other values if they rebuild LinuxThreads).
 802     //
 803     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 804     // the memory region has already been mmap'ed. That means if we have too
 805     // many threads and/or very large heap, eventually thread stack will
 806     // collide with heap.
 807     //
 808     // Here we try to prevent heap/stack collision by comparing current
 809     // stack bottom with the highest address that has been mmap'ed by JVM
 810     // plus a safety margin for memory maps created by native code.
 811     //
 812     // This feature can be disabled by setting ThreadSafetyMargin to 0
 813     //
 814     if (ThreadSafetyMargin > 0) {
 815       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 816 
 817       // not safe if our stack extends below the safety margin
 818       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 819     } else {
 820       return true;
 821     }
 822   } else {
 823     // Floating stack LinuxThreads or NPTL:
 824     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 825     //   there's not enough space left, pthread_create() will fail. If we come
 826     //   here, that means enough space has been reserved for stack.
 827     return true;
 828   }
 829 }
 830 
 831 // Thread start routine for all newly created threads
 832 static void *java_start(Thread *thread) {
 833   // Try to randomize the cache line index of hot stack frames.
 834   // This helps when threads of the same stack traces evict each other's
 835   // cache lines. The threads can be either from the same JVM instance, or
 836   // from different JVM instances. The benefit is especially true for
 837   // processors with hyperthreading technology.
 838   static int counter = 0;
 839   int pid = os::current_process_id();
 840   alloca(((pid ^ counter++) & 7) * 128);
 841 
 842   ThreadLocalStorage::set_thread(thread);
 843 
 844   OSThread* osthread = thread->osthread();
 845   Monitor* sync = osthread->startThread_lock();
 846 
 847   // non floating stack LinuxThreads needs extra check, see above
 848   if (!_thread_safety_check(thread)) {
 849     // notify parent thread
 850     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 851     osthread->set_state(ZOMBIE);
 852     sync->notify_all();
 853     return NULL;
 854   }
 855 
 856   // thread_id is kernel thread id (similar to Solaris LWP id)
 857   osthread->set_thread_id(os::Linux::gettid());
 858 
 859   if (UseNUMA) {
 860     int lgrp_id = os::numa_get_group_id();
 861     if (lgrp_id != -1) {
 862       thread->set_lgrp_id(lgrp_id);
 863     }
 864   }
 865   // initialize signal mask for this thread
 866   os::Linux::hotspot_sigmask(thread);
 867 
 868   // initialize floating point control register
 869   os::Linux::init_thread_fpu_state();
 870 
 871   // handshaking with parent thread
 872   {
 873     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 874 
 875     // notify parent thread
 876     osthread->set_state(INITIALIZED);
 877     sync->notify_all();
 878 
 879     // wait until os::start_thread()
 880     while (osthread->get_state() == INITIALIZED) {
 881       sync->wait(Mutex::_no_safepoint_check_flag);
 882     }
 883   }
 884 
 885   // call one more level start routine
 886   thread->run();
 887 
 888   return 0;
 889 }
 890 
 891 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 892   assert(thread->osthread() == NULL, "caller responsible");
 893 
 894   // Allocate the OSThread object
 895   OSThread* osthread = new OSThread(NULL, NULL);
 896   if (osthread == NULL) {
 897     return false;
 898   }
 899 
 900   // set the correct thread state
 901   osthread->set_thread_type(thr_type);
 902 
 903   // Initial state is ALLOCATED but not INITIALIZED
 904   osthread->set_state(ALLOCATED);
 905 
 906   thread->set_osthread(osthread);
 907 
 908   // init thread attributes
 909   pthread_attr_t attr;
 910   pthread_attr_init(&attr);
 911   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 912 
 913   // stack size
 914   if (os::Linux::supports_variable_stack_size()) {
 915     // calculate stack size if it's not specified by caller
 916     if (stack_size == 0) {
 917       stack_size = os::Linux::default_stack_size(thr_type);
 918 
 919       switch (thr_type) {
 920       case os::java_thread:
 921         // Java threads use ThreadStackSize which default value can be
 922         // changed with the flag -Xss
 923         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 924         stack_size = JavaThread::stack_size_at_create();
 925         break;
 926       case os::compiler_thread:
 927         if (CompilerThreadStackSize > 0) {
 928           stack_size = (size_t)(CompilerThreadStackSize * K);
 929           break;
 930         } // else fall through:
 931           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 932       case os::vm_thread:
 933       case os::pgc_thread:
 934       case os::cgc_thread:
 935       case os::watcher_thread:
 936         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 937         break;
 938       }
 939     }
 940 
 941     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 942     pthread_attr_setstacksize(&attr, stack_size);
 943   } else {
 944     // let pthread_create() pick the default value.
 945   }
 946 
 947   // glibc guard page
 948   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 949 
 950   ThreadState state;
 951 
 952   {
 953     // Serialize thread creation if we are running with fixed stack LinuxThreads
 954     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 955     if (lock) {
 956       os::Linux::createThread_lock()->lock_without_safepoint_check();
 957     }
 958 
 959     pthread_t tid;
 960     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 961 
 962     pthread_attr_destroy(&attr);
 963 
 964     if (ret != 0) {
 965       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 966         perror("pthread_create()");
 967       }
 968       // Need to clean up stuff we've allocated so far
 969       thread->set_osthread(NULL);
 970       delete osthread;
 971       if (lock) os::Linux::createThread_lock()->unlock();
 972       return false;
 973     }
 974 
 975     // Store pthread info into the OSThread
 976     osthread->set_pthread_id(tid);
 977 
 978     // Wait until child thread is either initialized or aborted
 979     {
 980       Monitor* sync_with_child = osthread->startThread_lock();
 981       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 982       while ((state = osthread->get_state()) == ALLOCATED) {
 983         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 984       }
 985     }
 986 
 987     if (lock) {
 988       os::Linux::createThread_lock()->unlock();
 989     }
 990   }
 991 
 992   // Aborted due to thread limit being reached
 993   if (state == ZOMBIE) {
 994       thread->set_osthread(NULL);
 995       delete osthread;
 996       return false;
 997   }
 998 
 999   // The thread is returned suspended (in state INITIALIZED),
1000   // and is started higher up in the call chain
1001   assert(state == INITIALIZED, "race condition");
1002   return true;
1003 }
1004 
1005 /////////////////////////////////////////////////////////////////////////////
1006 // attach existing thread
1007 
1008 // bootstrap the main thread
1009 bool os::create_main_thread(JavaThread* thread) {
1010   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
1011   return create_attached_thread(thread);
1012 }
1013 
1014 bool os::create_attached_thread(JavaThread* thread) {
1015 #ifdef ASSERT
1016     thread->verify_not_published();
1017 #endif
1018 
1019   // Allocate the OSThread object
1020   OSThread* osthread = new OSThread(NULL, NULL);
1021 
1022   if (osthread == NULL) {
1023     return false;
1024   }
1025 
1026   // Store pthread info into the OSThread
1027   osthread->set_thread_id(os::Linux::gettid());
1028   osthread->set_pthread_id(::pthread_self());
1029 
1030   // initialize floating point control register
1031   os::Linux::init_thread_fpu_state();
1032 
1033   // Initial thread state is RUNNABLE
1034   osthread->set_state(RUNNABLE);
1035 
1036   thread->set_osthread(osthread);
1037 
1038   if (UseNUMA) {
1039     int lgrp_id = os::numa_get_group_id();
1040     if (lgrp_id != -1) {
1041       thread->set_lgrp_id(lgrp_id);
1042     }
1043   }
1044 
1045   if (os::Linux::is_initial_thread()) {
1046     // If current thread is initial thread, its stack is mapped on demand,
1047     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1048     // the entire stack region to avoid SEGV in stack banging.
1049     // It is also useful to get around the heap-stack-gap problem on SuSE
1050     // kernel (see 4821821 for details). We first expand stack to the top
1051     // of yellow zone, then enable stack yellow zone (order is significant,
1052     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1053     // is no gap between the last two virtual memory regions.
1054 
1055     JavaThread *jt = (JavaThread *)thread;
1056     address addr = jt->stack_yellow_zone_base();
1057     assert(addr != NULL, "initialization problem?");
1058     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1059 
1060     osthread->set_expanding_stack();
1061     os::Linux::manually_expand_stack(jt, addr);
1062     osthread->clear_expanding_stack();
1063   }
1064 
1065   // initialize signal mask for this thread
1066   // and save the caller's signal mask
1067   os::Linux::hotspot_sigmask(thread);
1068 
1069   return true;
1070 }
1071 
1072 void os::pd_start_thread(Thread* thread) {
1073   OSThread * osthread = thread->osthread();
1074   assert(osthread->get_state() != INITIALIZED, "just checking");
1075   Monitor* sync_with_child = osthread->startThread_lock();
1076   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1077   sync_with_child->notify();
1078 }
1079 
1080 // Free Linux resources related to the OSThread
1081 void os::free_thread(OSThread* osthread) {
1082   assert(osthread != NULL, "osthread not set");
1083 
1084   if (Thread::current()->osthread() == osthread) {
1085     // Restore caller's signal mask
1086     sigset_t sigmask = osthread->caller_sigmask();
1087     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1088    }
1089 
1090   delete osthread;
1091 }
1092 
1093 //////////////////////////////////////////////////////////////////////////////
1094 // thread local storage
1095 
1096 int os::allocate_thread_local_storage() {
1097   pthread_key_t key;
1098   int rslt = pthread_key_create(&key, NULL);
1099   assert(rslt == 0, "cannot allocate thread local storage");
1100   return (int)key;
1101 }
1102 
1103 // Note: This is currently not used by VM, as we don't destroy TLS key
1104 // on VM exit.
1105 void os::free_thread_local_storage(int index) {
1106   int rslt = pthread_key_delete((pthread_key_t)index);
1107   assert(rslt == 0, "invalid index");
1108 }
1109 
1110 void os::thread_local_storage_at_put(int index, void* value) {
1111   int rslt = pthread_setspecific((pthread_key_t)index, value);
1112   assert(rslt == 0, "pthread_setspecific failed");
1113 }
1114 
1115 extern "C" Thread* get_thread() {
1116   return ThreadLocalStorage::thread();
1117 }
1118 
1119 //////////////////////////////////////////////////////////////////////////////
1120 // initial thread
1121 
1122 // Check if current thread is the initial thread, similar to Solaris thr_main.
1123 bool os::Linux::is_initial_thread(void) {
1124   char dummy;
1125   // If called before init complete, thread stack bottom will be null.
1126   // Can be called if fatal error occurs before initialization.
1127   if (initial_thread_stack_bottom() == NULL) return false;
1128   assert(initial_thread_stack_bottom() != NULL &&
1129          initial_thread_stack_size()   != 0,
1130          "os::init did not locate initial thread's stack region");
1131   if ((address)&dummy >= initial_thread_stack_bottom() &&
1132       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1133        return true;
1134   else return false;
1135 }
1136 
1137 // Find the virtual memory area that contains addr
1138 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1139   FILE *fp = fopen("/proc/self/maps", "r");
1140   if (fp) {
1141     address low, high;
1142     while (!feof(fp)) {
1143       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1144         if (low <= addr && addr < high) {
1145            if (vma_low)  *vma_low  = low;
1146            if (vma_high) *vma_high = high;
1147            fclose (fp);
1148            return true;
1149         }
1150       }
1151       for (;;) {
1152         int ch = fgetc(fp);
1153         if (ch == EOF || ch == (int)'\n') break;
1154       }
1155     }
1156     fclose(fp);
1157   }
1158   return false;
1159 }
1160 
1161 // Locate initial thread stack. This special handling of initial thread stack
1162 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1163 // bogus value for initial thread.
1164 void os::Linux::capture_initial_stack(size_t max_size) {
1165   // stack size is the easy part, get it from RLIMIT_STACK
1166   size_t stack_size;
1167   struct rlimit rlim;
1168   getrlimit(RLIMIT_STACK, &rlim);
1169   stack_size = rlim.rlim_cur;
1170 
1171   // 6308388: a bug in ld.so will relocate its own .data section to the
1172   //   lower end of primordial stack; reduce ulimit -s value a little bit
1173   //   so we won't install guard page on ld.so's data section.
1174   stack_size -= 2 * page_size();
1175 
1176   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1177   //   7.1, in both cases we will get 2G in return value.
1178   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1179   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1180   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1181   //   in case other parts in glibc still assumes 2M max stack size.
1182   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1183 #ifndef IA64
1184   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1185 #else
1186   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1187   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1188 #endif
1189 
1190   // Try to figure out where the stack base (top) is. This is harder.
1191   //
1192   // When an application is started, glibc saves the initial stack pointer in
1193   // a global variable "__libc_stack_end", which is then used by system
1194   // libraries. __libc_stack_end should be pretty close to stack top. The
1195   // variable is available since the very early days. However, because it is
1196   // a private interface, it could disappear in the future.
1197   //
1198   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1199   // to __libc_stack_end, it is very close to stack top, but isn't the real
1200   // stack top. Note that /proc may not exist if VM is running as a chroot
1201   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1202   // /proc/<pid>/stat could change in the future (though unlikely).
1203   //
1204   // We try __libc_stack_end first. If that doesn't work, look for
1205   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1206   // as a hint, which should work well in most cases.
1207 
1208   uintptr_t stack_start;
1209 
1210   // try __libc_stack_end first
1211   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1212   if (p && *p) {
1213     stack_start = *p;
1214   } else {
1215     // see if we can get the start_stack field from /proc/self/stat
1216     FILE *fp;
1217     int pid;
1218     char state;
1219     int ppid;
1220     int pgrp;
1221     int session;
1222     int nr;
1223     int tpgrp;
1224     unsigned long flags;
1225     unsigned long minflt;
1226     unsigned long cminflt;
1227     unsigned long majflt;
1228     unsigned long cmajflt;
1229     unsigned long utime;
1230     unsigned long stime;
1231     long cutime;
1232     long cstime;
1233     long prio;
1234     long nice;
1235     long junk;
1236     long it_real;
1237     uintptr_t start;
1238     uintptr_t vsize;
1239     intptr_t rss;
1240     uintptr_t rsslim;
1241     uintptr_t scodes;
1242     uintptr_t ecode;
1243     int i;
1244 
1245     // Figure what the primordial thread stack base is. Code is inspired
1246     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1247     // followed by command name surrounded by parentheses, state, etc.
1248     char stat[2048];
1249     int statlen;
1250 
1251     fp = fopen("/proc/self/stat", "r");
1252     if (fp) {
1253       statlen = fread(stat, 1, 2047, fp);
1254       stat[statlen] = '\0';
1255       fclose(fp);
1256 
1257       // Skip pid and the command string. Note that we could be dealing with
1258       // weird command names, e.g. user could decide to rename java launcher
1259       // to "java 1.4.2 :)", then the stat file would look like
1260       //                1234 (java 1.4.2 :)) R ... ...
1261       // We don't really need to know the command string, just find the last
1262       // occurrence of ")" and then start parsing from there. See bug 4726580.
1263       char * s = strrchr(stat, ')');
1264 
1265       i = 0;
1266       if (s) {
1267         // Skip blank chars
1268         do s++; while (isspace(*s));
1269 
1270 #define _UFM UINTX_FORMAT
1271 #define _DFM INTX_FORMAT
1272 
1273         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1274         /*              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 */
1275         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,
1276              &state,          /* 3  %c  */
1277              &ppid,           /* 4  %d  */
1278              &pgrp,           /* 5  %d  */
1279              &session,        /* 6  %d  */
1280              &nr,             /* 7  %d  */
1281              &tpgrp,          /* 8  %d  */
1282              &flags,          /* 9  %lu  */
1283              &minflt,         /* 10 %lu  */
1284              &cminflt,        /* 11 %lu  */
1285              &majflt,         /* 12 %lu  */
1286              &cmajflt,        /* 13 %lu  */
1287              &utime,          /* 14 %lu  */
1288              &stime,          /* 15 %lu  */
1289              &cutime,         /* 16 %ld  */
1290              &cstime,         /* 17 %ld  */
1291              &prio,           /* 18 %ld  */
1292              &nice,           /* 19 %ld  */
1293              &junk,           /* 20 %ld  */
1294              &it_real,        /* 21 %ld  */
1295              &start,          /* 22 UINTX_FORMAT */
1296              &vsize,          /* 23 UINTX_FORMAT */
1297              &rss,            /* 24 INTX_FORMAT  */
1298              &rsslim,         /* 25 UINTX_FORMAT */
1299              &scodes,         /* 26 UINTX_FORMAT */
1300              &ecode,          /* 27 UINTX_FORMAT */
1301              &stack_start);   /* 28 UINTX_FORMAT */
1302       }
1303 
1304 #undef _UFM
1305 #undef _DFM
1306 
1307       if (i != 28 - 2) {
1308          assert(false, "Bad conversion from /proc/self/stat");
1309          // product mode - assume we are the initial thread, good luck in the
1310          // embedded case.
1311          warning("Can't detect initial thread stack location - bad conversion");
1312          stack_start = (uintptr_t) &rlim;
1313       }
1314     } else {
1315       // For some reason we can't open /proc/self/stat (for example, running on
1316       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1317       // most cases, so don't abort:
1318       warning("Can't detect initial thread stack location - no /proc/self/stat");
1319       stack_start = (uintptr_t) &rlim;
1320     }
1321   }
1322 
1323   // Now we have a pointer (stack_start) very close to the stack top, the
1324   // next thing to do is to figure out the exact location of stack top. We
1325   // can find out the virtual memory area that contains stack_start by
1326   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1327   // and its upper limit is the real stack top. (again, this would fail if
1328   // running inside chroot, because /proc may not exist.)
1329 
1330   uintptr_t stack_top;
1331   address low, high;
1332   if (find_vma((address)stack_start, &low, &high)) {
1333     // success, "high" is the true stack top. (ignore "low", because initial
1334     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1335     stack_top = (uintptr_t)high;
1336   } else {
1337     // failed, likely because /proc/self/maps does not exist
1338     warning("Can't detect initial thread stack location - find_vma failed");
1339     // best effort: stack_start is normally within a few pages below the real
1340     // stack top, use it as stack top, and reduce stack size so we won't put
1341     // guard page outside stack.
1342     stack_top = stack_start;
1343     stack_size -= 16 * page_size();
1344   }
1345 
1346   // stack_top could be partially down the page so align it
1347   stack_top = align_size_up(stack_top, page_size());
1348 
1349   if (max_size && stack_size > max_size) {
1350      _initial_thread_stack_size = max_size;
1351   } else {
1352      _initial_thread_stack_size = stack_size;
1353   }
1354 
1355   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1356   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1357 }
1358 
1359 ////////////////////////////////////////////////////////////////////////////////
1360 // time support
1361 
1362 // Time since start-up in seconds to a fine granularity.
1363 // Used by VMSelfDestructTimer and the MemProfiler.
1364 double os::elapsedTime() {
1365 
1366   return (double)(os::elapsed_counter()) * 0.000001;
1367 }
1368 
1369 jlong os::elapsed_counter() {
1370   timeval time;
1371   int status = gettimeofday(&time, NULL);
1372   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1373 }
1374 
1375 jlong os::elapsed_frequency() {
1376   return (1000 * 1000);
1377 }
1378 
1379 // For now, we say that linux does not support vtime.  I have no idea
1380 // whether it can actually be made to (DLD, 9/13/05).
1381 
1382 bool os::supports_vtime() { return false; }
1383 bool os::enable_vtime()   { return false; }
1384 bool os::vtime_enabled()  { return false; }
1385 double os::elapsedVTime() {
1386   // better than nothing, but not much
1387   return elapsedTime();
1388 }
1389 
1390 jlong os::javaTimeMillis() {
1391   timeval time;
1392   int status = gettimeofday(&time, NULL);
1393   assert(status != -1, "linux error");
1394   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1395 }
1396 
1397 #ifndef CLOCK_MONOTONIC
1398 #define CLOCK_MONOTONIC (1)
1399 #endif
1400 
1401 void os::Linux::clock_init() {
1402   // we do dlopen's in this particular order due to bug in linux
1403   // dynamical loader (see 6348968) leading to crash on exit
1404   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1405   if (handle == NULL) {
1406     handle = dlopen("librt.so", RTLD_LAZY);
1407   }
1408 
1409   if (handle) {
1410     int (*clock_getres_func)(clockid_t, struct timespec*) =
1411            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1412     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1413            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1414     if (clock_getres_func && clock_gettime_func) {
1415       // See if monotonic clock is supported by the kernel. Note that some
1416       // early implementations simply return kernel jiffies (updated every
1417       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1418       // for nano time (though the monotonic property is still nice to have).
1419       // It's fixed in newer kernels, however clock_getres() still returns
1420       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1421       // resolution for now. Hopefully as people move to new kernels, this
1422       // won't be a problem.
1423       struct timespec res;
1424       struct timespec tp;
1425       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1426           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1427         // yes, monotonic clock is supported
1428         _clock_gettime = clock_gettime_func;
1429       } else {
1430         // close librt if there is no monotonic clock
1431         dlclose(handle);
1432       }
1433     }
1434   }
1435 }
1436 
1437 #ifndef SYS_clock_getres
1438 
1439 #if defined(IA32) || defined(AMD64)
1440 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1441 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1442 #else
1443 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1444 #define sys_clock_getres(x,y)  -1
1445 #endif
1446 
1447 #else
1448 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1449 #endif
1450 
1451 void os::Linux::fast_thread_clock_init() {
1452   if (!UseLinuxPosixThreadCPUClocks) {
1453     return;
1454   }
1455   clockid_t clockid;
1456   struct timespec tp;
1457   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1458       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1459 
1460   // Switch to using fast clocks for thread cpu time if
1461   // the sys_clock_getres() returns 0 error code.
1462   // Note, that some kernels may support the current thread
1463   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1464   // returned by the pthread_getcpuclockid().
1465   // If the fast Posix clocks are supported then the sys_clock_getres()
1466   // must return at least tp.tv_sec == 0 which means a resolution
1467   // better than 1 sec. This is extra check for reliability.
1468 
1469   if(pthread_getcpuclockid_func &&
1470      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1471      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1472 
1473     _supports_fast_thread_cpu_time = true;
1474     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1475   }
1476 }
1477 
1478 jlong os::javaTimeNanos() {
1479   if (Linux::supports_monotonic_clock()) {
1480     struct timespec tp;
1481     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1482     assert(status == 0, "gettime error");
1483     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1484     return result;
1485   } else {
1486     timeval time;
1487     int status = gettimeofday(&time, NULL);
1488     assert(status != -1, "linux error");
1489     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1490     return 1000 * usecs;
1491   }
1492 }
1493 
1494 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1495   if (Linux::supports_monotonic_clock()) {
1496     info_ptr->max_value = ALL_64_BITS;
1497 
1498     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1499     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1500     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1501   } else {
1502     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1503     info_ptr->max_value = ALL_64_BITS;
1504 
1505     // gettimeofday is a real time clock so it skips
1506     info_ptr->may_skip_backward = true;
1507     info_ptr->may_skip_forward = true;
1508   }
1509 
1510   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1511 }
1512 
1513 // Return the real, user, and system times in seconds from an
1514 // arbitrary fixed point in the past.
1515 bool os::getTimesSecs(double* process_real_time,
1516                       double* process_user_time,
1517                       double* process_system_time) {
1518   struct tms ticks;
1519   clock_t real_ticks = times(&ticks);
1520 
1521   if (real_ticks == (clock_t) (-1)) {
1522     return false;
1523   } else {
1524     double ticks_per_second = (double) clock_tics_per_sec;
1525     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1526     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1527     *process_real_time = ((double) real_ticks) / ticks_per_second;
1528 
1529     return true;
1530   }
1531 }
1532 
1533 
1534 char * os::local_time_string(char *buf, size_t buflen) {
1535   struct tm t;
1536   time_t long_time;
1537   time(&long_time);
1538   localtime_r(&long_time, &t);
1539   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1540                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1541                t.tm_hour, t.tm_min, t.tm_sec);
1542   return buf;
1543 }
1544 
1545 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1546   return localtime_r(clock, res);
1547 }
1548 
1549 ////////////////////////////////////////////////////////////////////////////////
1550 // runtime exit support
1551 
1552 // Note: os::shutdown() might be called very early during initialization, or
1553 // called from signal handler. Before adding something to os::shutdown(), make
1554 // sure it is async-safe and can handle partially initialized VM.
1555 void os::shutdown() {
1556 
1557   // allow PerfMemory to attempt cleanup of any persistent resources
1558   perfMemory_exit();
1559 
1560   // needs to remove object in file system
1561   AttachListener::abort();
1562 
1563   // flush buffered output, finish log files
1564   ostream_abort();
1565 
1566   // Check for abort hook
1567   abort_hook_t abort_hook = Arguments::abort_hook();
1568   if (abort_hook != NULL) {
1569     abort_hook();
1570   }
1571 
1572 }
1573 
1574 // Note: os::abort() might be called very early during initialization, or
1575 // called from signal handler. Before adding something to os::abort(), make
1576 // sure it is async-safe and can handle partially initialized VM.
1577 void os::abort(bool dump_core) {
1578   os::shutdown();
1579   if (dump_core) {
1580 #ifndef PRODUCT
1581     fdStream out(defaultStream::output_fd());
1582     out.print_raw("Current thread is ");
1583     char buf[16];
1584     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1585     out.print_raw_cr(buf);
1586     out.print_raw_cr("Dumping core ...");
1587 #endif
1588     ::abort(); // dump core
1589   }
1590 
1591   ::exit(1);
1592 }
1593 
1594 // Die immediately, no exit hook, no abort hook, no cleanup.
1595 void os::die() {
1596   // _exit() on LinuxThreads only kills current thread
1597   ::abort();
1598 }
1599 
1600 // unused on linux for now.
1601 void os::set_error_file(const char *logfile) {}
1602 
1603 
1604 // This method is a copy of JDK's sysGetLastErrorString
1605 // from src/solaris/hpi/src/system_md.c
1606 
1607 size_t os::lasterror(char *buf, size_t len) {
1608 
1609   if (errno == 0)  return 0;
1610 
1611   const char *s = ::strerror(errno);
1612   size_t n = ::strlen(s);
1613   if (n >= len) {
1614     n = len - 1;
1615   }
1616   ::strncpy(buf, s, n);
1617   buf[n] = '\0';
1618   return n;
1619 }
1620 
1621 intx os::current_thread_id() { return (intx)pthread_self(); }
1622 int os::current_process_id() {
1623 
1624   // Under the old linux thread library, linux gives each thread
1625   // its own process id. Because of this each thread will return
1626   // a different pid if this method were to return the result
1627   // of getpid(2). Linux provides no api that returns the pid
1628   // of the launcher thread for the vm. This implementation
1629   // returns a unique pid, the pid of the launcher thread
1630   // that starts the vm 'process'.
1631 
1632   // Under the NPTL, getpid() returns the same pid as the
1633   // launcher thread rather than a unique pid per thread.
1634   // Use gettid() if you want the old pre NPTL behaviour.
1635 
1636   // if you are looking for the result of a call to getpid() that
1637   // returns a unique pid for the calling thread, then look at the
1638   // OSThread::thread_id() method in osThread_linux.hpp file
1639 
1640   return (int)(_initial_pid ? _initial_pid : getpid());
1641 }
1642 
1643 // DLL functions
1644 
1645 const char* os::dll_file_extension() { return ".so"; }
1646 
1647 // This must be hard coded because it's the system's temporary
1648 // directory not the java application's temp directory, ala java.io.tmpdir.
1649 const char* os::get_temp_directory() { return "/tmp"; }
1650 
1651 static bool file_exists(const char* filename) {
1652   struct stat statbuf;
1653   if (filename == NULL || strlen(filename) == 0) {
1654     return false;
1655   }
1656   return os::stat(filename, &statbuf) == 0;
1657 }
1658 
1659 void os::dll_build_name(char* buffer, size_t buflen,
1660                         const char* pname, const char* fname) {
1661   // Copied from libhpi
1662   const size_t pnamelen = pname ? strlen(pname) : 0;
1663 
1664   // Quietly truncate on buffer overflow.  Should be an error.
1665   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1666       *buffer = '\0';
1667       return;
1668   }
1669 
1670   if (pnamelen == 0) {
1671     snprintf(buffer, buflen, "lib%s.so", fname);
1672   } else if (strchr(pname, *os::path_separator()) != NULL) {
1673     int n;
1674     char** pelements = split_path(pname, &n);
1675     for (int i = 0 ; i < n ; i++) {
1676       // Really shouldn't be NULL, but check can't hurt
1677       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1678         continue; // skip the empty path values
1679       }
1680       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1681       if (file_exists(buffer)) {
1682         break;
1683       }
1684     }
1685     // release the storage
1686     for (int i = 0 ; i < n ; i++) {
1687       if (pelements[i] != NULL) {
1688         FREE_C_HEAP_ARRAY(char, pelements[i]);
1689       }
1690     }
1691     if (pelements != NULL) {
1692       FREE_C_HEAP_ARRAY(char*, pelements);
1693     }
1694   } else {
1695     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1696   }
1697 }
1698 
1699 const char* os::get_current_directory(char *buf, int buflen) {
1700   return getcwd(buf, buflen);
1701 }
1702 
1703 // check if addr is inside libjvm[_g].so
1704 bool os::address_is_in_vm(address addr) {
1705   static address libjvm_base_addr;
1706   Dl_info dlinfo;
1707 
1708   if (libjvm_base_addr == NULL) {
1709     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1710     libjvm_base_addr = (address)dlinfo.dli_fbase;
1711     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1712   }
1713 
1714   if (dladdr((void *)addr, &dlinfo)) {
1715     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1716   }
1717 
1718   return false;
1719 }
1720 
1721 bool os::dll_address_to_function_name(address addr, char *buf,
1722                                       int buflen, int *offset) {
1723   Dl_info dlinfo;
1724 
1725   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1726     if (buf != NULL) {
1727       if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1728         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1729       }
1730     }
1731     if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1732     return true;
1733   } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1734     if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1735         buf, buflen, offset, dlinfo.dli_fname)) {
1736        return true;
1737     }
1738   }
1739 
1740   if (buf != NULL) buf[0] = '\0';
1741   if (offset != NULL) *offset = -1;
1742   return false;
1743 }
1744 
1745 struct _address_to_library_name {
1746   address addr;          // input : memory address
1747   size_t  buflen;        //         size of fname
1748   char*   fname;         // output: library name
1749   address base;          //         library base addr
1750 };
1751 
1752 static int address_to_library_name_callback(struct dl_phdr_info *info,
1753                                             size_t size, void *data) {
1754   int i;
1755   bool found = false;
1756   address libbase = NULL;
1757   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1758 
1759   // iterate through all loadable segments
1760   for (i = 0; i < info->dlpi_phnum; i++) {
1761     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1762     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1763       // base address of a library is the lowest address of its loaded
1764       // segments.
1765       if (libbase == NULL || libbase > segbase) {
1766         libbase = segbase;
1767       }
1768       // see if 'addr' is within current segment
1769       if (segbase <= d->addr &&
1770           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1771         found = true;
1772       }
1773     }
1774   }
1775 
1776   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1777   // so dll_address_to_library_name() can fall through to use dladdr() which
1778   // can figure out executable name from argv[0].
1779   if (found && info->dlpi_name && info->dlpi_name[0]) {
1780     d->base = libbase;
1781     if (d->fname) {
1782       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1783     }
1784     return 1;
1785   }
1786   return 0;
1787 }
1788 
1789 bool os::dll_address_to_library_name(address addr, char* buf,
1790                                      int buflen, int* offset) {
1791   Dl_info dlinfo;
1792   struct _address_to_library_name data;
1793 
1794   // There is a bug in old glibc dladdr() implementation that it could resolve
1795   // to wrong library name if the .so file has a base address != NULL. Here
1796   // we iterate through the program headers of all loaded libraries to find
1797   // out which library 'addr' really belongs to. This workaround can be
1798   // removed once the minimum requirement for glibc is moved to 2.3.x.
1799   data.addr = addr;
1800   data.fname = buf;
1801   data.buflen = buflen;
1802   data.base = NULL;
1803   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1804 
1805   if (rslt) {
1806      // buf already contains library name
1807      if (offset) *offset = addr - data.base;
1808      return true;
1809   } else if (dladdr((void*)addr, &dlinfo)){
1810      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1811      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1812      return true;
1813   } else {
1814      if (buf) buf[0] = '\0';
1815      if (offset) *offset = -1;
1816      return false;
1817   }
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 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1825 {
1826   void * result= ::dlopen(filename, RTLD_LAZY);
1827   if (result != NULL) {
1828     // Successful loading
1829     return result;
1830   }
1831 
1832   Elf32_Ehdr elf_head;
1833 
1834   // Read system error message into ebuf
1835   // It may or may not be overwritten below
1836   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1837   ebuf[ebuflen-1]='\0';
1838   int diag_msg_max_length=ebuflen-strlen(ebuf);
1839   char* diag_msg_buf=ebuf+strlen(ebuf);
1840 
1841   if (diag_msg_max_length==0) {
1842     // No more space in ebuf for additional diagnostics message
1843     return NULL;
1844   }
1845 
1846 
1847   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1848 
1849   if (file_descriptor < 0) {
1850     // Can't open library, report dlerror() message
1851     return NULL;
1852   }
1853 
1854   bool failed_to_read_elf_head=
1855     (sizeof(elf_head)!=
1856         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1857 
1858   ::close(file_descriptor);
1859   if (failed_to_read_elf_head) {
1860     // file i/o error - report dlerror() msg
1861     return NULL;
1862   }
1863 
1864   typedef struct {
1865     Elf32_Half  code;         // Actual value as defined in elf.h
1866     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1867     char        elf_class;    // 32 or 64 bit
1868     char        endianess;    // MSB or LSB
1869     char*       name;         // String representation
1870   } arch_t;
1871 
1872   #ifndef EM_486
1873   #define EM_486          6               /* Intel 80486 */
1874   #endif
1875 
1876   static const arch_t arch_array[]={
1877     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1878     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1879     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1880     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1881     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1882     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1883     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1884     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1885     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1886     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1887     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1888     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1889     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1890     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1891     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1892     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1893   };
1894 
1895   #if  (defined IA32)
1896     static  Elf32_Half running_arch_code=EM_386;
1897   #elif   (defined AMD64)
1898     static  Elf32_Half running_arch_code=EM_X86_64;
1899   #elif  (defined IA64)
1900     static  Elf32_Half running_arch_code=EM_IA_64;
1901   #elif  (defined __sparc) && (defined _LP64)
1902     static  Elf32_Half running_arch_code=EM_SPARCV9;
1903   #elif  (defined __sparc) && (!defined _LP64)
1904     static  Elf32_Half running_arch_code=EM_SPARC;
1905   #elif  (defined __powerpc64__)
1906     static  Elf32_Half running_arch_code=EM_PPC64;
1907   #elif  (defined __powerpc__)
1908     static  Elf32_Half running_arch_code=EM_PPC;
1909   #elif  (defined ARM)
1910     static  Elf32_Half running_arch_code=EM_ARM;
1911   #elif  (defined S390)
1912     static  Elf32_Half running_arch_code=EM_S390;
1913   #elif  (defined ALPHA)
1914     static  Elf32_Half running_arch_code=EM_ALPHA;
1915   #elif  (defined MIPSEL)
1916     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1917   #elif  (defined PARISC)
1918     static  Elf32_Half running_arch_code=EM_PARISC;
1919   #elif  (defined MIPS)
1920     static  Elf32_Half running_arch_code=EM_MIPS;
1921   #elif  (defined M68K)
1922     static  Elf32_Half running_arch_code=EM_68K;
1923   #else
1924     #error Method os::dll_load requires that one of following is defined:\
1925          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1926   #endif
1927 
1928   // Identify compatability class for VM's architecture and library's architecture
1929   // Obtain string descriptions for architectures
1930 
1931   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1932   int running_arch_index=-1;
1933 
1934   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1935     if (running_arch_code == arch_array[i].code) {
1936       running_arch_index    = i;
1937     }
1938     if (lib_arch.code == arch_array[i].code) {
1939       lib_arch.compat_class = arch_array[i].compat_class;
1940       lib_arch.name         = arch_array[i].name;
1941     }
1942   }
1943 
1944   assert(running_arch_index != -1,
1945     "Didn't find running architecture code (running_arch_code) in arch_array");
1946   if (running_arch_index == -1) {
1947     // Even though running architecture detection failed
1948     // we may still continue with reporting dlerror() message
1949     return NULL;
1950   }
1951 
1952   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1953     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1954     return NULL;
1955   }
1956 
1957 #ifndef S390
1958   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1959     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1960     return NULL;
1961   }
1962 #endif // !S390
1963 
1964   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1965     if ( lib_arch.name!=NULL ) {
1966       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1967         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1968         lib_arch.name, arch_array[running_arch_index].name);
1969     } else {
1970       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1971       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1972         lib_arch.code,
1973         arch_array[running_arch_index].name);
1974     }
1975   }
1976 
1977   return NULL;
1978 }
1979 
1980 /*
1981  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
1982  * chances are you might want to run the generated bits against glibc-2.0
1983  * libdl.so, so always use locking for any version of glibc.
1984  */
1985 void* os::dll_lookup(void* handle, const char* name) {
1986   pthread_mutex_lock(&dl_mutex);
1987   void* res = dlsym(handle, name);
1988   pthread_mutex_unlock(&dl_mutex);
1989   return res;
1990 }
1991 
1992 
1993 static bool _print_ascii_file(const char* filename, outputStream* st) {
1994   int fd = ::open(filename, O_RDONLY);
1995   if (fd == -1) {
1996      return false;
1997   }
1998 
1999   char buf[32];
2000   int bytes;
2001   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2002     st->print_raw(buf, bytes);
2003   }
2004 
2005   ::close(fd);
2006 
2007   return true;
2008 }
2009 
2010 void os::print_dll_info(outputStream *st) {
2011    st->print_cr("Dynamic libraries:");
2012 
2013    char fname[32];
2014    pid_t pid = os::Linux::gettid();
2015 
2016    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2017 
2018    if (!_print_ascii_file(fname, st)) {
2019      st->print("Can not get library information for pid = %d\n", pid);
2020    }
2021 }
2022 
2023 
2024 void os::print_os_info(outputStream* st) {
2025   st->print("OS:");
2026 
2027   // Try to identify popular distros.
2028   // Most Linux distributions have /etc/XXX-release file, which contains
2029   // the OS version string. Some have more than one /etc/XXX-release file
2030   // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2031   // so the order is important.
2032   if (!_print_ascii_file("/etc/mandrake-release", st) &&
2033       !_print_ascii_file("/etc/sun-release", st) &&
2034       !_print_ascii_file("/etc/redhat-release", st) &&
2035       !_print_ascii_file("/etc/SuSE-release", st) &&
2036       !_print_ascii_file("/etc/turbolinux-release", st) &&
2037       !_print_ascii_file("/etc/gentoo-release", st) &&
2038       !_print_ascii_file("/etc/debian_version", st) &&
2039       !_print_ascii_file("/etc/ltib-release", st) &&
2040       !_print_ascii_file("/etc/angstrom-version", st)) {
2041       st->print("Linux");
2042   }
2043   st->cr();
2044 
2045   // kernel
2046   st->print("uname:");
2047   struct utsname name;
2048   uname(&name);
2049   st->print(name.sysname); st->print(" ");
2050   st->print(name.release); st->print(" ");
2051   st->print(name.version); st->print(" ");
2052   st->print(name.machine);
2053   st->cr();
2054 
2055   // Print warning if unsafe chroot environment detected
2056   if (unsafe_chroot_detected) {
2057     st->print("WARNING!! ");
2058     st->print_cr(unstable_chroot_error);
2059   }
2060 
2061   // libc, pthread
2062   st->print("libc:");
2063   st->print(os::Linux::glibc_version()); st->print(" ");
2064   st->print(os::Linux::libpthread_version()); st->print(" ");
2065   if (os::Linux::is_LinuxThreads()) {
2066      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2067   }
2068   st->cr();
2069 
2070   // rlimit
2071   st->print("rlimit:");
2072   struct rlimit rlim;
2073 
2074   st->print(" STACK ");
2075   getrlimit(RLIMIT_STACK, &rlim);
2076   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2077   else st->print("%uk", rlim.rlim_cur >> 10);
2078 
2079   st->print(", CORE ");
2080   getrlimit(RLIMIT_CORE, &rlim);
2081   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2082   else st->print("%uk", rlim.rlim_cur >> 10);
2083 
2084   st->print(", NPROC ");
2085   getrlimit(RLIMIT_NPROC, &rlim);
2086   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2087   else st->print("%d", rlim.rlim_cur);
2088 
2089   st->print(", NOFILE ");
2090   getrlimit(RLIMIT_NOFILE, &rlim);
2091   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2092   else st->print("%d", rlim.rlim_cur);
2093 
2094   st->print(", AS ");
2095   getrlimit(RLIMIT_AS, &rlim);
2096   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2097   else st->print("%uk", rlim.rlim_cur >> 10);
2098   st->cr();
2099 
2100   // load average
2101   st->print("load average:");
2102   double loadavg[3];
2103   os::loadavg(loadavg, 3);
2104   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2105   st->cr();
2106 
2107   // meminfo
2108   st->print("\n/proc/meminfo:\n");
2109   _print_ascii_file("/proc/meminfo", st);
2110   st->cr();
2111 }
2112 
2113 void os::pd_print_cpu_info(outputStream* st) {
2114   st->print("\n/proc/cpuinfo:\n");
2115   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2116     st->print("  <Not Available>");
2117   }
2118   st->cr();
2119 }
2120 
2121 void os::print_memory_info(outputStream* st) {
2122 
2123   st->print("Memory:");
2124   st->print(" %dk page", os::vm_page_size()>>10);
2125 
2126   // values in struct sysinfo are "unsigned long"
2127   struct sysinfo si;
2128   sysinfo(&si);
2129 
2130   st->print(", physical " UINT64_FORMAT "k",
2131             os::physical_memory() >> 10);
2132   st->print("(" UINT64_FORMAT "k free)",
2133             os::available_memory() >> 10);
2134   st->print(", swap " UINT64_FORMAT "k",
2135             ((jlong)si.totalswap * si.mem_unit) >> 10);
2136   st->print("(" UINT64_FORMAT "k free)",
2137             ((jlong)si.freeswap * si.mem_unit) >> 10);
2138   st->cr();
2139 }
2140 
2141 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2142 // but they're the same for all the linux arch that we support
2143 // and they're the same for solaris but there's no common place to put this.
2144 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2145                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2146                           "ILL_COPROC", "ILL_BADSTK" };
2147 
2148 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2149                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2150                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2151 
2152 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2153 
2154 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2155 
2156 void os::print_siginfo(outputStream* st, void* siginfo) {
2157   st->print("siginfo:");
2158 
2159   const int buflen = 100;
2160   char buf[buflen];
2161   siginfo_t *si = (siginfo_t*)siginfo;
2162   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2163   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2164     st->print("si_errno=%s", buf);
2165   } else {
2166     st->print("si_errno=%d", si->si_errno);
2167   }
2168   const int c = si->si_code;
2169   assert(c > 0, "unexpected si_code");
2170   switch (si->si_signo) {
2171   case SIGILL:
2172     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2173     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2174     break;
2175   case SIGFPE:
2176     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2177     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2178     break;
2179   case SIGSEGV:
2180     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2181     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2182     break;
2183   case SIGBUS:
2184     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2185     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2186     break;
2187   default:
2188     st->print(", si_code=%d", si->si_code);
2189     // no si_addr
2190   }
2191 
2192   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2193       UseSharedSpaces) {
2194     FileMapInfo* mapinfo = FileMapInfo::current_info();
2195     if (mapinfo->is_in_shared_space(si->si_addr)) {
2196       st->print("\n\nError accessing class data sharing archive."   \
2197                 " Mapped file inaccessible during execution, "      \
2198                 " possible disk/network problem.");
2199     }
2200   }
2201   st->cr();
2202 }
2203 
2204 
2205 static void print_signal_handler(outputStream* st, int sig,
2206                                  char* buf, size_t buflen);
2207 
2208 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2209   st->print_cr("Signal Handlers:");
2210   print_signal_handler(st, SIGSEGV, buf, buflen);
2211   print_signal_handler(st, SIGBUS , buf, buflen);
2212   print_signal_handler(st, SIGFPE , buf, buflen);
2213   print_signal_handler(st, SIGPIPE, buf, buflen);
2214   print_signal_handler(st, SIGXFSZ, buf, buflen);
2215   print_signal_handler(st, SIGILL , buf, buflen);
2216   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2217   print_signal_handler(st, SR_signum, buf, buflen);
2218   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2219   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2220   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2221   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2222 }
2223 
2224 static char saved_jvm_path[MAXPATHLEN] = {0};
2225 
2226 // Find the full path to the current module, libjvm.so or libjvm_g.so
2227 void os::jvm_path(char *buf, jint buflen) {
2228   // Error checking.
2229   if (buflen < MAXPATHLEN) {
2230     assert(false, "must use a large-enough buffer");
2231     buf[0] = '\0';
2232     return;
2233   }
2234   // Lazy resolve the path to current module.
2235   if (saved_jvm_path[0] != 0) {
2236     strcpy(buf, saved_jvm_path);
2237     return;
2238   }
2239 
2240   char dli_fname[MAXPATHLEN];
2241   bool ret = dll_address_to_library_name(
2242                 CAST_FROM_FN_PTR(address, os::jvm_path),
2243                 dli_fname, sizeof(dli_fname), NULL);
2244   assert(ret != 0, "cannot locate libjvm");
2245   char *rp = realpath(dli_fname, buf);
2246   if (rp == NULL)
2247     return;
2248 
2249   if (Arguments::created_by_gamma_launcher()) {
2250     // Support for the gamma launcher.  Typical value for buf is
2251     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2252     // the right place in the string, then assume we are installed in a JDK and
2253     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2254     // up the path so it looks like libjvm.so is installed there (append a
2255     // fake suffix hotspot/libjvm.so).
2256     const char *p = buf + strlen(buf) - 1;
2257     for (int count = 0; p > buf && count < 5; ++count) {
2258       for (--p; p > buf && *p != '/'; --p)
2259         /* empty */ ;
2260     }
2261 
2262     if (strncmp(p, "/jre/lib/", 9) != 0) {
2263       // Look for JAVA_HOME in the environment.
2264       char* java_home_var = ::getenv("JAVA_HOME");
2265       if (java_home_var != NULL && java_home_var[0] != 0) {
2266         char* jrelib_p;
2267         int len;
2268 
2269         // Check the current module name "libjvm.so" or "libjvm_g.so".
2270         p = strrchr(buf, '/');
2271         assert(strstr(p, "/libjvm") == p, "invalid library name");
2272         p = strstr(p, "_g") ? "_g" : "";
2273 
2274         rp = realpath(java_home_var, buf);
2275         if (rp == NULL)
2276           return;
2277 
2278         // determine if this is a legacy image or modules image
2279         // modules image doesn't have "jre" subdirectory
2280         len = strlen(buf);
2281         jrelib_p = buf + len;
2282         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2283         if (0 != access(buf, F_OK)) {
2284           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2285         }
2286 
2287         if (0 == access(buf, F_OK)) {
2288           // Use current module name "libjvm[_g].so" instead of
2289           // "libjvm"debug_only("_g")".so" since for fastdebug version
2290           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2291           len = strlen(buf);
2292           snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2293         } else {
2294           // Go back to path of .so
2295           rp = realpath(dli_fname, buf);
2296           if (rp == NULL)
2297             return;
2298         }
2299       }
2300     }
2301   }
2302 
2303   strcpy(saved_jvm_path, buf);
2304 }
2305 
2306 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2307   // no prefix required, not even "_"
2308 }
2309 
2310 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2311   // no suffix required
2312 }
2313 
2314 ////////////////////////////////////////////////////////////////////////////////
2315 // sun.misc.Signal support
2316 
2317 static volatile jint sigint_count = 0;
2318 
2319 static void
2320 UserHandler(int sig, void *siginfo, void *context) {
2321   // 4511530 - sem_post is serialized and handled by the manager thread. When
2322   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2323   // don't want to flood the manager thread with sem_post requests.
2324   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2325       return;
2326 
2327   // Ctrl-C is pressed during error reporting, likely because the error
2328   // handler fails to abort. Let VM die immediately.
2329   if (sig == SIGINT && is_error_reported()) {
2330      os::die();
2331   }
2332 
2333   os::signal_notify(sig);
2334 }
2335 
2336 void* os::user_handler() {
2337   return CAST_FROM_FN_PTR(void*, UserHandler);
2338 }
2339 
2340 extern "C" {
2341   typedef void (*sa_handler_t)(int);
2342   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2343 }
2344 
2345 void* os::signal(int signal_number, void* handler) {
2346   struct sigaction sigAct, oldSigAct;
2347 
2348   sigfillset(&(sigAct.sa_mask));
2349   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2350   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2351 
2352   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2353     // -1 means registration failed
2354     return (void *)-1;
2355   }
2356 
2357   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2358 }
2359 
2360 void os::signal_raise(int signal_number) {
2361   ::raise(signal_number);
2362 }
2363 
2364 /*
2365  * The following code is moved from os.cpp for making this
2366  * code platform specific, which it is by its very nature.
2367  */
2368 
2369 // Will be modified when max signal is changed to be dynamic
2370 int os::sigexitnum_pd() {
2371   return NSIG;
2372 }
2373 
2374 // a counter for each possible signal value
2375 static volatile jint pending_signals[NSIG+1] = { 0 };
2376 
2377 // Linux(POSIX) specific hand shaking semaphore.
2378 static sem_t sig_sem;
2379 
2380 void os::signal_init_pd() {
2381   // Initialize signal structures
2382   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2383 
2384   // Initialize signal semaphore
2385   ::sem_init(&sig_sem, 0, 0);
2386 }
2387 
2388 void os::signal_notify(int sig) {
2389   Atomic::inc(&pending_signals[sig]);
2390   ::sem_post(&sig_sem);
2391 }
2392 
2393 static int check_pending_signals(bool wait) {
2394   Atomic::store(0, &sigint_count);
2395   for (;;) {
2396     for (int i = 0; i < NSIG + 1; i++) {
2397       jint n = pending_signals[i];
2398       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2399         return i;
2400       }
2401     }
2402     if (!wait) {
2403       return -1;
2404     }
2405     JavaThread *thread = JavaThread::current();
2406     ThreadBlockInVM tbivm(thread);
2407 
2408     bool threadIsSuspended;
2409     do {
2410       thread->set_suspend_equivalent();
2411       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2412       ::sem_wait(&sig_sem);
2413 
2414       // were we externally suspended while we were waiting?
2415       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2416       if (threadIsSuspended) {
2417         //
2418         // The semaphore has been incremented, but while we were waiting
2419         // another thread suspended us. We don't want to continue running
2420         // while suspended because that would surprise the thread that
2421         // suspended us.
2422         //
2423         ::sem_post(&sig_sem);
2424 
2425         thread->java_suspend_self();
2426       }
2427     } while (threadIsSuspended);
2428   }
2429 }
2430 
2431 int os::signal_lookup() {
2432   return check_pending_signals(false);
2433 }
2434 
2435 int os::signal_wait() {
2436   return check_pending_signals(true);
2437 }
2438 
2439 ////////////////////////////////////////////////////////////////////////////////
2440 // Virtual Memory
2441 
2442 int os::vm_page_size() {
2443   // Seems redundant as all get out
2444   assert(os::Linux::page_size() != -1, "must call os::init");
2445   return os::Linux::page_size();
2446 }
2447 
2448 // Solaris allocates memory by pages.
2449 int os::vm_allocation_granularity() {
2450   assert(os::Linux::page_size() != -1, "must call os::init");
2451   return os::Linux::page_size();
2452 }
2453 
2454 // Rationale behind this function:
2455 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2456 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2457 //  samples for JITted code. Here we create private executable mapping over the code cache
2458 //  and then we can use standard (well, almost, as mapping can change) way to provide
2459 //  info for the reporting script by storing timestamp and location of symbol
2460 void linux_wrap_code(char* base, size_t size) {
2461   static volatile jint cnt = 0;
2462 
2463   if (!UseOprofile) {
2464     return;
2465   }
2466 
2467   char buf[PATH_MAX+1];
2468   int num = Atomic::add(1, &cnt);
2469 
2470   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2471            os::get_temp_directory(), os::current_process_id(), num);
2472   unlink(buf);
2473 
2474   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2475 
2476   if (fd != -1) {
2477     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2478     if (rv != (off_t)-1) {
2479       if (::write(fd, "", 1) == 1) {
2480         mmap(base, size,
2481              PROT_READ|PROT_WRITE|PROT_EXEC,
2482              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2483       }
2484     }
2485     ::close(fd);
2486     unlink(buf);
2487   }
2488 }
2489 
2490 // NOTE: Linux kernel does not really reserve the pages for us.
2491 //       All it does is to check if there are enough free pages
2492 //       left at the time of mmap(). This could be a potential
2493 //       problem.
2494 bool os::commit_memory(char* addr, size_t size, bool exec) {
2495   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2496   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2497                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2498   if (res != (uintptr_t) MAP_FAILED) {
2499     if (UseNUMAInterleaving) {
2500       numa_make_global(addr, size);
2501     }
2502     return true;
2503   }
2504   return false;
2505 }
2506 
2507 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2508 #ifndef MAP_HUGETLB
2509 #define MAP_HUGETLB 0x40000
2510 #endif
2511 
2512 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2513 #ifndef MADV_HUGEPAGE
2514 #define MADV_HUGEPAGE 14
2515 #endif
2516 
2517 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2518                        bool exec) {
2519   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2520     int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2521     uintptr_t res =
2522       (uintptr_t) ::mmap(addr, size, prot,
2523                          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2524                          -1, 0);
2525     if (res != (uintptr_t) MAP_FAILED) {
2526       if (UseNUMAInterleaving) {
2527         numa_make_global(addr, size);
2528       }
2529       return true;
2530     }
2531     // Fall through and try to use small pages
2532   }
2533 
2534   if (commit_memory(addr, size, exec)) {
2535     realign_memory(addr, size, alignment_hint);
2536     return true;
2537   }
2538   return false;
2539 }
2540 
2541 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2542   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2543     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2544     // be supported or the memory may already be backed by huge pages.
2545     ::madvise(addr, bytes, MADV_HUGEPAGE);
2546   }
2547 }
2548 
2549 void os::free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2550   commit_memory(addr, bytes, alignment_hint, false);
2551 }
2552 
2553 void os::numa_make_global(char *addr, size_t bytes) {
2554   Linux::numa_interleave_memory(addr, bytes);
2555 }
2556 
2557 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2558   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2559 }
2560 
2561 bool os::numa_topology_changed()   { return false; }
2562 
2563 size_t os::numa_get_groups_num() {
2564   int max_node = Linux::numa_max_node();
2565   return max_node > 0 ? max_node + 1 : 1;
2566 }
2567 
2568 int os::numa_get_group_id() {
2569   int cpu_id = Linux::sched_getcpu();
2570   if (cpu_id != -1) {
2571     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2572     if (lgrp_id != -1) {
2573       return lgrp_id;
2574     }
2575   }
2576   return 0;
2577 }
2578 
2579 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2580   for (size_t i = 0; i < size; i++) {
2581     ids[i] = i;
2582   }
2583   return size;
2584 }
2585 
2586 bool os::get_page_info(char *start, page_info* info) {
2587   return false;
2588 }
2589 
2590 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2591   return end;
2592 }
2593 
2594 
2595 int os::Linux::sched_getcpu_syscall(void) {
2596   unsigned int cpu;
2597   int retval = -1;
2598 
2599 #if defined(IA32)
2600 # ifndef SYS_getcpu
2601 # define SYS_getcpu 318
2602 # endif
2603   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2604 #elif defined(AMD64)
2605 // Unfortunately we have to bring all these macros here from vsyscall.h
2606 // to be able to compile on old linuxes.
2607 # define __NR_vgetcpu 2
2608 # define VSYSCALL_START (-10UL << 20)
2609 # define VSYSCALL_SIZE 1024
2610 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2611   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2612   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2613   retval = vgetcpu(&cpu, NULL, NULL);
2614 #endif
2615 
2616   return (retval == -1) ? retval : cpu;
2617 }
2618 
2619 // Something to do with the numa-aware allocator needs these symbols
2620 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2621 extern "C" JNIEXPORT void numa_error(char *where) { }
2622 extern "C" JNIEXPORT int fork1() { return fork(); }
2623 
2624 
2625 // If we are running with libnuma version > 2, then we should
2626 // be trying to use symbols with versions 1.1
2627 // If we are running with earlier version, which did not have symbol versions,
2628 // we should use the base version.
2629 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2630   void *f = dlvsym(handle, name, "libnuma_1.1");
2631   if (f == NULL) {
2632     f = dlsym(handle, name);
2633   }
2634   return f;
2635 }
2636 
2637 bool os::Linux::libnuma_init() {
2638   // sched_getcpu() should be in libc.
2639   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2640                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2641 
2642   // If it's not, try a direct syscall.
2643   if (sched_getcpu() == -1)
2644     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2645 
2646   if (sched_getcpu() != -1) { // Does it work?
2647     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2648     if (handle != NULL) {
2649       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2650                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2651       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2652                                        libnuma_dlsym(handle, "numa_max_node")));
2653       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2654                                         libnuma_dlsym(handle, "numa_available")));
2655       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2656                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2657       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2658                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2659 
2660 
2661       if (numa_available() != -1) {
2662         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2663         // Create a cpu -> node mapping
2664         _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2665         rebuild_cpu_to_node_map();
2666         return true;
2667       }
2668     }
2669   }
2670   return false;
2671 }
2672 
2673 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2674 // The table is later used in get_node_by_cpu().
2675 void os::Linux::rebuild_cpu_to_node_map() {
2676   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2677                               // in libnuma (possible values are starting from 16,
2678                               // and continuing up with every other power of 2, but less
2679                               // than the maximum number of CPUs supported by kernel), and
2680                               // is a subject to change (in libnuma version 2 the requirements
2681                               // are more reasonable) we'll just hardcode the number they use
2682                               // in the library.
2683   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2684 
2685   size_t cpu_num = os::active_processor_count();
2686   size_t cpu_map_size = NCPUS / BitsPerCLong;
2687   size_t cpu_map_valid_size =
2688     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2689 
2690   cpu_to_node()->clear();
2691   cpu_to_node()->at_grow(cpu_num - 1);
2692   size_t node_num = numa_get_groups_num();
2693 
2694   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2695   for (size_t i = 0; i < node_num; i++) {
2696     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2697       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2698         if (cpu_map[j] != 0) {
2699           for (size_t k = 0; k < BitsPerCLong; k++) {
2700             if (cpu_map[j] & (1UL << k)) {
2701               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2702             }
2703           }
2704         }
2705       }
2706     }
2707   }
2708   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2709 }
2710 
2711 int os::Linux::get_node_by_cpu(int cpu_id) {
2712   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2713     return cpu_to_node()->at(cpu_id);
2714   }
2715   return -1;
2716 }
2717 
2718 GrowableArray<int>* os::Linux::_cpu_to_node;
2719 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2720 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2721 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2722 os::Linux::numa_available_func_t os::Linux::_numa_available;
2723 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2724 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2725 unsigned long* os::Linux::_numa_all_nodes;
2726 
2727 bool os::uncommit_memory(char* addr, size_t size) {
2728   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2729                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2730   return res  != (uintptr_t) MAP_FAILED;
2731 }
2732 
2733 // Linux uses a growable mapping for the stack, and if the mapping for
2734 // the stack guard pages is not removed when we detach a thread the
2735 // stack cannot grow beyond the pages where the stack guard was
2736 // mapped.  If at some point later in the process the stack expands to
2737 // that point, the Linux kernel cannot expand the stack any further
2738 // because the guard pages are in the way, and a segfault occurs.
2739 //
2740 // However, it's essential not to split the stack region by unmapping
2741 // a region (leaving a hole) that's already part of the stack mapping,
2742 // so if the stack mapping has already grown beyond the guard pages at
2743 // the time we create them, we have to truncate the stack mapping.
2744 // So, we need to know the extent of the stack mapping when
2745 // create_stack_guard_pages() is called.
2746 
2747 // Find the bounds of the stack mapping.  Return true for success.
2748 //
2749 // We only need this for stacks that are growable: at the time of
2750 // writing thread stacks don't use growable mappings (i.e. those
2751 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2752 // only applies to the main thread.
2753 
2754 static
2755 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2756 
2757   char buf[128];
2758   int fd, sz;
2759 
2760   if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2761     return false;
2762   }
2763 
2764   const char kw[] = "[stack]";
2765   const int kwlen = sizeof(kw)-1;
2766 
2767   // Address part of /proc/self/maps couldn't be more than 128 bytes
2768   while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2769      if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2770         // Extract addresses
2771         if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2772            uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2773            if (sp >= *bottom && sp <= *top) {
2774               ::close(fd);
2775               return true;
2776            }
2777         }
2778      }
2779   }
2780 
2781  ::close(fd);
2782   return false;
2783 }
2784 
2785 
2786 // If the (growable) stack mapping already extends beyond the point
2787 // where we're going to put our guard pages, truncate the mapping at
2788 // that point by munmap()ping it.  This ensures that when we later
2789 // munmap() the guard pages we don't leave a hole in the stack
2790 // mapping. This only affects the main/initial thread, but guard
2791 // against future OS changes
2792 bool os::create_stack_guard_pages(char* addr, size_t size) {
2793   uintptr_t stack_extent, stack_base;
2794   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2795   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2796       assert(os::Linux::is_initial_thread(),
2797            "growable stack in non-initial thread");
2798     if (stack_extent < (uintptr_t)addr)
2799       ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2800   }
2801 
2802   return os::commit_memory(addr, size);
2803 }
2804 
2805 // If this is a growable mapping, remove the guard pages entirely by
2806 // munmap()ping them.  If not, just call uncommit_memory(). This only
2807 // affects the main/initial thread, but guard against future OS changes
2808 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2809   uintptr_t stack_extent, stack_base;
2810   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2811   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2812       assert(os::Linux::is_initial_thread(),
2813            "growable stack in non-initial thread");
2814 
2815     return ::munmap(addr, size) == 0;
2816   }
2817 
2818   return os::uncommit_memory(addr, size);
2819 }
2820 
2821 static address _highest_vm_reserved_address = NULL;
2822 
2823 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2824 // at 'requested_addr'. If there are existing memory mappings at the same
2825 // location, however, they will be overwritten. If 'fixed' is false,
2826 // 'requested_addr' is only treated as a hint, the return value may or
2827 // may not start from the requested address. Unlike Linux mmap(), this
2828 // function returns NULL to indicate failure.
2829 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2830   char * addr;
2831   int flags;
2832 
2833   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2834   if (fixed) {
2835     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2836     flags |= MAP_FIXED;
2837   }
2838 
2839   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2840   // to PROT_EXEC if executable when we commit the page.
2841   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2842                        flags, -1, 0);
2843 
2844   if (addr != MAP_FAILED) {
2845     // anon_mmap() should only get called during VM initialization,
2846     // don't need lock (actually we can skip locking even it can be called
2847     // from multiple threads, because _highest_vm_reserved_address is just a
2848     // hint about the upper limit of non-stack memory regions.)
2849     if ((address)addr + bytes > _highest_vm_reserved_address) {
2850       _highest_vm_reserved_address = (address)addr + bytes;
2851     }
2852   }
2853 
2854   return addr == MAP_FAILED ? NULL : addr;
2855 }
2856 
2857 // Don't update _highest_vm_reserved_address, because there might be memory
2858 // regions above addr + size. If so, releasing a memory region only creates
2859 // a hole in the address space, it doesn't help prevent heap-stack collision.
2860 //
2861 static int anon_munmap(char * addr, size_t size) {
2862   return ::munmap(addr, size) == 0;
2863 }
2864 
2865 char* os::reserve_memory(size_t bytes, char* requested_addr,
2866                          size_t alignment_hint) {
2867   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2868 }
2869 
2870 bool os::release_memory(char* addr, size_t size) {
2871   return anon_munmap(addr, size);
2872 }
2873 
2874 static address highest_vm_reserved_address() {
2875   return _highest_vm_reserved_address;
2876 }
2877 
2878 static bool linux_mprotect(char* addr, size_t size, int prot) {
2879   // Linux wants the mprotect address argument to be page aligned.
2880   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2881 
2882   // According to SUSv3, mprotect() should only be used with mappings
2883   // established by mmap(), and mmap() always maps whole pages. Unaligned
2884   // 'addr' likely indicates problem in the VM (e.g. trying to change
2885   // protection of malloc'ed or statically allocated memory). Check the
2886   // caller if you hit this assert.
2887   assert(addr == bottom, "sanity check");
2888 
2889   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2890   return ::mprotect(bottom, size, prot) == 0;
2891 }
2892 
2893 // Set protections specified
2894 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2895                         bool is_committed) {
2896   unsigned int p = 0;
2897   switch (prot) {
2898   case MEM_PROT_NONE: p = PROT_NONE; break;
2899   case MEM_PROT_READ: p = PROT_READ; break;
2900   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2901   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2902   default:
2903     ShouldNotReachHere();
2904   }
2905   // is_committed is unused.
2906   return linux_mprotect(addr, bytes, p);
2907 }
2908 
2909 bool os::guard_memory(char* addr, size_t size) {
2910   return linux_mprotect(addr, size, PROT_NONE);
2911 }
2912 
2913 bool os::unguard_memory(char* addr, size_t size) {
2914   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2915 }
2916 
2917 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2918   bool result = false;
2919   void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2920                   MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2921                   -1, 0);
2922 
2923   if (p != (void *) -1) {
2924     // We don't know if this really is a huge page or not.
2925     FILE *fp = fopen("/proc/self/maps", "r");
2926     if (fp) {
2927       while (!feof(fp)) {
2928         char chars[257];
2929         long x = 0;
2930         if (fgets(chars, sizeof(chars), fp)) {
2931           if (sscanf(chars, "%lx-%*x", &x) == 1
2932               && x == (long)p) {
2933             if (strstr (chars, "hugepage")) {
2934               result = true;
2935               break;
2936             }
2937           }
2938         }
2939       }
2940       fclose(fp);
2941     }
2942     munmap (p, page_size);
2943     if (result)
2944       return true;
2945   }
2946 
2947   if (warn) {
2948     warning("HugeTLBFS is not supported by the operating system.");
2949   }
2950 
2951   return result;
2952 }
2953 
2954 /*
2955 * Set the coredump_filter bits to include largepages in core dump (bit 6)
2956 *
2957 * From the coredump_filter documentation:
2958 *
2959 * - (bit 0) anonymous private memory
2960 * - (bit 1) anonymous shared memory
2961 * - (bit 2) file-backed private memory
2962 * - (bit 3) file-backed shared memory
2963 * - (bit 4) ELF header pages in file-backed private memory areas (it is
2964 *           effective only if the bit 2 is cleared)
2965 * - (bit 5) hugetlb private memory
2966 * - (bit 6) hugetlb shared memory
2967 */
2968 static void set_coredump_filter(void) {
2969   FILE *f;
2970   long cdm;
2971 
2972   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
2973     return;
2974   }
2975 
2976   if (fscanf(f, "%lx", &cdm) != 1) {
2977     fclose(f);
2978     return;
2979   }
2980 
2981   rewind(f);
2982 
2983   if ((cdm & LARGEPAGES_BIT) == 0) {
2984     cdm |= LARGEPAGES_BIT;
2985     fprintf(f, "%#lx", cdm);
2986   }
2987 
2988   fclose(f);
2989 }
2990 
2991 // Large page support
2992 
2993 static size_t _large_page_size = 0;
2994 
2995 void os::large_page_init() {
2996   if (!UseLargePages) {
2997     UseHugeTLBFS = false;
2998     UseSHM = false;
2999     return;
3000   }
3001 
3002   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3003     // If UseLargePages is specified on the command line try both methods,
3004     // if it's default, then try only HugeTLBFS.
3005     if (FLAG_IS_DEFAULT(UseLargePages)) {
3006       UseHugeTLBFS = true;
3007     } else {
3008       UseHugeTLBFS = UseSHM = true;
3009     }
3010   }
3011 
3012   if (LargePageSizeInBytes) {
3013     _large_page_size = LargePageSizeInBytes;
3014   } else {
3015     // large_page_size on Linux is used to round up heap size. x86 uses either
3016     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3017     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3018     // page as large as 256M.
3019     //
3020     // Here we try to figure out page size by parsing /proc/meminfo and looking
3021     // for a line with the following format:
3022     //    Hugepagesize:     2048 kB
3023     //
3024     // If we can't determine the value (e.g. /proc is not mounted, or the text
3025     // format has been changed), we'll use the largest page size supported by
3026     // the processor.
3027 
3028 #ifndef ZERO
3029     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3030                        ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3031 #endif // ZERO
3032 
3033     FILE *fp = fopen("/proc/meminfo", "r");
3034     if (fp) {
3035       while (!feof(fp)) {
3036         int x = 0;
3037         char buf[16];
3038         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3039           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3040             _large_page_size = x * K;
3041             break;
3042           }
3043         } else {
3044           // skip to next line
3045           for (;;) {
3046             int ch = fgetc(fp);
3047             if (ch == EOF || ch == (int)'\n') break;
3048           }
3049         }
3050       }
3051       fclose(fp);
3052     }
3053   }
3054 
3055   // print a warning if any large page related flag is specified on command line
3056   bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3057 
3058   const size_t default_page_size = (size_t)Linux::page_size();
3059   if (_large_page_size > default_page_size) {
3060     _page_sizes[0] = _large_page_size;
3061     _page_sizes[1] = default_page_size;
3062     _page_sizes[2] = 0;
3063   }
3064   UseHugeTLBFS = UseHugeTLBFS &&
3065                  Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3066 
3067   if (UseHugeTLBFS)
3068     UseSHM = false;
3069 
3070   UseLargePages = UseHugeTLBFS || UseSHM;
3071 
3072   set_coredump_filter();
3073 }
3074 
3075 #ifndef SHM_HUGETLB
3076 #define SHM_HUGETLB 04000
3077 #endif
3078 
3079 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3080   // "exec" is passed in but not used.  Creating the shared image for
3081   // the code cache doesn't have an SHM_X executable permission to check.
3082   assert(UseLargePages && UseSHM, "only for SHM large pages");
3083 
3084   key_t key = IPC_PRIVATE;
3085   char *addr;
3086 
3087   bool warn_on_failure = UseLargePages &&
3088                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3089                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3090                         );
3091   char msg[128];
3092 
3093   // Create a large shared memory region to attach to based on size.
3094   // Currently, size is the total size of the heap
3095   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3096   if (shmid == -1) {
3097      // Possible reasons for shmget failure:
3098      // 1. shmmax is too small for Java heap.
3099      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3100      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3101      // 2. not enough large page memory.
3102      //    > check available large pages: cat /proc/meminfo
3103      //    > increase amount of large pages:
3104      //          echo new_value > /proc/sys/vm/nr_hugepages
3105      //      Note 1: different Linux may use different name for this property,
3106      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3107      //      Note 2: it's possible there's enough physical memory available but
3108      //            they are so fragmented after a long run that they can't
3109      //            coalesce into large pages. Try to reserve large pages when
3110      //            the system is still "fresh".
3111      if (warn_on_failure) {
3112        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3113        warning(msg);
3114      }
3115      return NULL;
3116   }
3117 
3118   // attach to the region
3119   addr = (char*)shmat(shmid, req_addr, 0);
3120   int err = errno;
3121 
3122   // Remove shmid. If shmat() is successful, the actual shared memory segment
3123   // will be deleted when it's detached by shmdt() or when the process
3124   // terminates. If shmat() is not successful this will remove the shared
3125   // segment immediately.
3126   shmctl(shmid, IPC_RMID, NULL);
3127 
3128   if ((intptr_t)addr == -1) {
3129      if (warn_on_failure) {
3130        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3131        warning(msg);
3132      }
3133      return NULL;
3134   }
3135 
3136   if ((addr != NULL) && UseNUMAInterleaving) {
3137     numa_make_global(addr, bytes);
3138   }
3139 
3140   return addr;
3141 }
3142 
3143 bool os::release_memory_special(char* base, size_t bytes) {
3144   // detaching the SHM segment will also delete it, see reserve_memory_special()
3145   int rslt = shmdt(base);
3146   return rslt == 0;
3147 }
3148 
3149 size_t os::large_page_size() {
3150   return _large_page_size;
3151 }
3152 
3153 // HugeTLBFS allows application to commit large page memory on demand;
3154 // with SysV SHM the entire memory region must be allocated as shared
3155 // memory.
3156 bool os::can_commit_large_page_memory() {
3157   return UseHugeTLBFS;
3158 }
3159 
3160 bool os::can_execute_large_page_memory() {
3161   return UseHugeTLBFS;
3162 }
3163 
3164 // Reserve memory at an arbitrary address, only if that area is
3165 // available (and not reserved for something else).
3166 
3167 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3168   const int max_tries = 10;
3169   char* base[max_tries];
3170   size_t size[max_tries];
3171   const size_t gap = 0x000000;
3172 
3173   // Assert only that the size is a multiple of the page size, since
3174   // that's all that mmap requires, and since that's all we really know
3175   // about at this low abstraction level.  If we need higher alignment,
3176   // we can either pass an alignment to this method or verify alignment
3177   // in one of the methods further up the call chain.  See bug 5044738.
3178   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3179 
3180   // Repeatedly allocate blocks until the block is allocated at the
3181   // right spot. Give up after max_tries. Note that reserve_memory() will
3182   // automatically update _highest_vm_reserved_address if the call is
3183   // successful. The variable tracks the highest memory address every reserved
3184   // by JVM. It is used to detect heap-stack collision if running with
3185   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3186   // space than needed, it could confuse the collision detecting code. To
3187   // solve the problem, save current _highest_vm_reserved_address and
3188   // calculate the correct value before return.
3189   address old_highest = _highest_vm_reserved_address;
3190 
3191   // Linux mmap allows caller to pass an address as hint; give it a try first,
3192   // if kernel honors the hint then we can return immediately.
3193   char * addr = anon_mmap(requested_addr, bytes, false);
3194   if (addr == requested_addr) {
3195      return requested_addr;
3196   }
3197 
3198   if (addr != NULL) {
3199      // mmap() is successful but it fails to reserve at the requested address
3200      anon_munmap(addr, bytes);
3201   }
3202 
3203   int i;
3204   for (i = 0; i < max_tries; ++i) {
3205     base[i] = reserve_memory(bytes);
3206 
3207     if (base[i] != NULL) {
3208       // Is this the block we wanted?
3209       if (base[i] == requested_addr) {
3210         size[i] = bytes;
3211         break;
3212       }
3213 
3214       // Does this overlap the block we wanted? Give back the overlapped
3215       // parts and try again.
3216 
3217       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3218       if (top_overlap >= 0 && top_overlap < bytes) {
3219         unmap_memory(base[i], top_overlap);
3220         base[i] += top_overlap;
3221         size[i] = bytes - top_overlap;
3222       } else {
3223         size_t bottom_overlap = base[i] + bytes - requested_addr;
3224         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3225           unmap_memory(requested_addr, bottom_overlap);
3226           size[i] = bytes - bottom_overlap;
3227         } else {
3228           size[i] = bytes;
3229         }
3230       }
3231     }
3232   }
3233 
3234   // Give back the unused reserved pieces.
3235 
3236   for (int j = 0; j < i; ++j) {
3237     if (base[j] != NULL) {
3238       unmap_memory(base[j], size[j]);
3239     }
3240   }
3241 
3242   if (i < max_tries) {
3243     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3244     return requested_addr;
3245   } else {
3246     _highest_vm_reserved_address = old_highest;
3247     return NULL;
3248   }
3249 }
3250 
3251 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3252   return ::read(fd, buf, nBytes);
3253 }
3254 
3255 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3256 // Solaris uses poll(), linux uses park().
3257 // Poll() is likely a better choice, assuming that Thread.interrupt()
3258 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3259 // SIGSEGV, see 4355769.
3260 
3261 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3262   assert(thread == Thread::current(),  "thread consistency check");
3263 
3264   ParkEvent * const slp = thread->_SleepEvent ;
3265   slp->reset() ;
3266   OrderAccess::fence() ;
3267 
3268   if (interruptible) {
3269     jlong prevtime = javaTimeNanos();
3270 
3271     for (;;) {
3272       if (os::is_interrupted(thread, true)) {
3273         return OS_INTRPT;
3274       }
3275 
3276       jlong newtime = javaTimeNanos();
3277 
3278       if (newtime - prevtime < 0) {
3279         // time moving backwards, should only happen if no monotonic clock
3280         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3281         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3282       } else {
3283         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3284       }
3285 
3286       if(millis <= 0) {
3287         return OS_OK;
3288       }
3289 
3290       prevtime = newtime;
3291 
3292       {
3293         assert(thread->is_Java_thread(), "sanity check");
3294         JavaThread *jt = (JavaThread *) thread;
3295         ThreadBlockInVM tbivm(jt);
3296         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3297 
3298         jt->set_suspend_equivalent();
3299         // cleared by handle_special_suspend_equivalent_condition() or
3300         // java_suspend_self() via check_and_wait_while_suspended()
3301 
3302         slp->park(millis);
3303 
3304         // were we externally suspended while we were waiting?
3305         jt->check_and_wait_while_suspended();
3306       }
3307     }
3308   } else {
3309     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3310     jlong prevtime = javaTimeNanos();
3311 
3312     for (;;) {
3313       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3314       // the 1st iteration ...
3315       jlong newtime = javaTimeNanos();
3316 
3317       if (newtime - prevtime < 0) {
3318         // time moving backwards, should only happen if no monotonic clock
3319         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3320         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3321       } else {
3322         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3323       }
3324 
3325       if(millis <= 0) break ;
3326 
3327       prevtime = newtime;
3328       slp->park(millis);
3329     }
3330     return OS_OK ;
3331   }
3332 }
3333 
3334 int os::naked_sleep() {
3335   // %% make the sleep time an integer flag. for now use 1 millisec.
3336   return os::sleep(Thread::current(), 1, false);
3337 }
3338 
3339 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3340 void os::infinite_sleep() {
3341   while (true) {    // sleep forever ...
3342     ::sleep(100);   // ... 100 seconds at a time
3343   }
3344 }
3345 
3346 // Used to convert frequent JVM_Yield() to nops
3347 bool os::dont_yield() {
3348   return DontYieldALot;
3349 }
3350 
3351 void os::yield() {
3352   sched_yield();
3353 }
3354 
3355 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3356 
3357 void os::yield_all(int attempts) {
3358   // Yields to all threads, including threads with lower priorities
3359   // Threads on Linux are all with same priority. The Solaris style
3360   // os::yield_all() with nanosleep(1ms) is not necessary.
3361   sched_yield();
3362 }
3363 
3364 // Called from the tight loops to possibly influence time-sharing heuristics
3365 void os::loop_breaker(int attempts) {
3366   os::yield_all(attempts);
3367 }
3368 
3369 ////////////////////////////////////////////////////////////////////////////////
3370 // thread priority support
3371 
3372 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3373 // only supports dynamic priority, static priority must be zero. For real-time
3374 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3375 // However, for large multi-threaded applications, SCHED_RR is not only slower
3376 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3377 // of 5 runs - Sep 2005).
3378 //
3379 // The following code actually changes the niceness of kernel-thread/LWP. It
3380 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3381 // not the entire user process, and user level threads are 1:1 mapped to kernel
3382 // threads. It has always been the case, but could change in the future. For
3383 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3384 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3385 
3386 int os::java_to_os_priority[MaxPriority + 1] = {
3387   19,              // 0 Entry should never be used
3388 
3389    4,              // 1 MinPriority
3390    3,              // 2
3391    2,              // 3
3392 
3393    1,              // 4
3394    0,              // 5 NormPriority
3395   -1,              // 6
3396 
3397   -2,              // 7
3398   -3,              // 8
3399   -4,              // 9 NearMaxPriority
3400 
3401   -5               // 10 MaxPriority


3402 };
3403 
3404 static int prio_init() {
3405   if (ThreadPriorityPolicy == 1) {
3406     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3407     // if effective uid is not root. Perhaps, a more elegant way of doing
3408     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3409     if (geteuid() != 0) {
3410       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3411         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3412       }
3413       ThreadPriorityPolicy = 0;
3414     }
3415   }
3416   return 0;
3417 }
3418 
3419 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3420   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3421 
3422   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3423   return (ret == 0) ? OS_OK : OS_ERR;
3424 }
3425 
3426 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3427   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3428     *priority_ptr = java_to_os_priority[NormPriority];
3429     return OS_OK;
3430   }
3431 
3432   errno = 0;
3433   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3434   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3435 }
3436 
3437 // Hint to the underlying OS that a task switch would not be good.
3438 // Void return because it's a hint and can fail.
3439 void os::hint_no_preempt() {}
3440 
3441 ////////////////////////////////////////////////////////////////////////////////
3442 // suspend/resume support
3443 
3444 //  the low-level signal-based suspend/resume support is a remnant from the
3445 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3446 //  within hotspot. Now there is a single use-case for this:
3447 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3448 //      that runs in the watcher thread.
3449 //  The remaining code is greatly simplified from the more general suspension
3450 //  code that used to be used.
3451 //
3452 //  The protocol is quite simple:
3453 //  - suspend:
3454 //      - sends a signal to the target thread
3455 //      - polls the suspend state of the osthread using a yield loop
3456 //      - target thread signal handler (SR_handler) sets suspend state
3457 //        and blocks in sigsuspend until continued
3458 //  - resume:
3459 //      - sets target osthread state to continue
3460 //      - sends signal to end the sigsuspend loop in the SR_handler
3461 //
3462 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3463 //
3464 
3465 static void resume_clear_context(OSThread *osthread) {
3466   osthread->set_ucontext(NULL);
3467   osthread->set_siginfo(NULL);
3468 
3469   // notify the suspend action is completed, we have now resumed
3470   osthread->sr.clear_suspended();
3471 }
3472 
3473 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3474   osthread->set_ucontext(context);
3475   osthread->set_siginfo(siginfo);
3476 }
3477 
3478 //
3479 // Handler function invoked when a thread's execution is suspended or
3480 // resumed. We have to be careful that only async-safe functions are
3481 // called here (Note: most pthread functions are not async safe and
3482 // should be avoided.)
3483 //
3484 // Note: sigwait() is a more natural fit than sigsuspend() from an
3485 // interface point of view, but sigwait() prevents the signal hander
3486 // from being run. libpthread would get very confused by not having
3487 // its signal handlers run and prevents sigwait()'s use with the
3488 // mutex granting granting signal.
3489 //
3490 // Currently only ever called on the VMThread
3491 //
3492 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3493   // Save and restore errno to avoid confusing native code with EINTR
3494   // after sigsuspend.
3495   int old_errno = errno;
3496 
3497   Thread* thread = Thread::current();
3498   OSThread* osthread = thread->osthread();
3499   assert(thread->is_VM_thread(), "Must be VMThread");
3500   // read current suspend action
3501   int action = osthread->sr.suspend_action();
3502   if (action == SR_SUSPEND) {
3503     suspend_save_context(osthread, siginfo, context);
3504 
3505     // Notify the suspend action is about to be completed. do_suspend()
3506     // waits until SR_SUSPENDED is set and then returns. We will wait
3507     // here for a resume signal and that completes the suspend-other
3508     // action. do_suspend/do_resume is always called as a pair from
3509     // the same thread - so there are no races
3510 
3511     // notify the caller
3512     osthread->sr.set_suspended();
3513 
3514     sigset_t suspend_set;  // signals for sigsuspend()
3515 
3516     // get current set of blocked signals and unblock resume signal
3517     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3518     sigdelset(&suspend_set, SR_signum);
3519 
3520     // wait here until we are resumed
3521     do {
3522       sigsuspend(&suspend_set);
3523       // ignore all returns until we get a resume signal
3524     } while (osthread->sr.suspend_action() != SR_CONTINUE);
3525 
3526     resume_clear_context(osthread);
3527 
3528   } else {
3529     assert(action == SR_CONTINUE, "unexpected sr action");
3530     // nothing special to do - just leave the handler
3531   }
3532 
3533   errno = old_errno;
3534 }
3535 
3536 
3537 static int SR_initialize() {
3538   struct sigaction act;
3539   char *s;
3540   /* Get signal number to use for suspend/resume */
3541   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3542     int sig = ::strtol(s, 0, 10);
3543     if (sig > 0 || sig < _NSIG) {
3544         SR_signum = sig;
3545     }
3546   }
3547 
3548   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3549         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3550 
3551   sigemptyset(&SR_sigset);
3552   sigaddset(&SR_sigset, SR_signum);
3553 
3554   /* Set up signal handler for suspend/resume */
3555   act.sa_flags = SA_RESTART|SA_SIGINFO;
3556   act.sa_handler = (void (*)(int)) SR_handler;
3557 
3558   // SR_signum is blocked by default.
3559   // 4528190 - We also need to block pthread restart signal (32 on all
3560   // supported Linux platforms). Note that LinuxThreads need to block
3561   // this signal for all threads to work properly. So we don't have
3562   // to use hard-coded signal number when setting up the mask.
3563   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3564 
3565   if (sigaction(SR_signum, &act, 0) == -1) {
3566     return -1;
3567   }
3568 
3569   // Save signal flag
3570   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3571   return 0;
3572 }
3573 
3574 static int SR_finalize() {
3575   return 0;
3576 }
3577 
3578 
3579 // returns true on success and false on error - really an error is fatal
3580 // but this seems the normal response to library errors
3581 static bool do_suspend(OSThread* osthread) {
3582   // mark as suspended and send signal
3583   osthread->sr.set_suspend_action(SR_SUSPEND);
3584   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3585   assert_status(status == 0, status, "pthread_kill");
3586 
3587   // check status and wait until notified of suspension
3588   if (status == 0) {
3589     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3590       os::yield_all(i);
3591     }
3592     osthread->sr.set_suspend_action(SR_NONE);
3593     return true;
3594   }
3595   else {
3596     osthread->sr.set_suspend_action(SR_NONE);
3597     return false;
3598   }
3599 }
3600 
3601 static void do_resume(OSThread* osthread) {
3602   assert(osthread->sr.is_suspended(), "thread should be suspended");
3603   osthread->sr.set_suspend_action(SR_CONTINUE);
3604 
3605   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3606   assert_status(status == 0, status, "pthread_kill");
3607   // check status and wait unit notified of resumption
3608   if (status == 0) {
3609     for (int i = 0; osthread->sr.is_suspended(); i++) {
3610       os::yield_all(i);
3611     }
3612   }
3613   osthread->sr.set_suspend_action(SR_NONE);
3614 }
3615 
3616 ////////////////////////////////////////////////////////////////////////////////
3617 // interrupt support
3618 
3619 void os::interrupt(Thread* thread) {
3620   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3621     "possibility of dangling Thread pointer");
3622 
3623   OSThread* osthread = thread->osthread();
3624 
3625   if (!osthread->interrupted()) {
3626     osthread->set_interrupted(true);
3627     // More than one thread can get here with the same value of osthread,
3628     // resulting in multiple notifications.  We do, however, want the store
3629     // to interrupted() to be visible to other threads before we execute unpark().
3630     OrderAccess::fence();
3631     ParkEvent * const slp = thread->_SleepEvent ;
3632     if (slp != NULL) slp->unpark() ;
3633   }
3634 
3635   // For JSR166. Unpark even if interrupt status already was set
3636   if (thread->is_Java_thread())
3637     ((JavaThread*)thread)->parker()->unpark();
3638 
3639   ParkEvent * ev = thread->_ParkEvent ;
3640   if (ev != NULL) ev->unpark() ;
3641 
3642 }
3643 
3644 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3645   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3646     "possibility of dangling Thread pointer");
3647 
3648   OSThread* osthread = thread->osthread();
3649 
3650   bool interrupted = osthread->interrupted();
3651 
3652   if (interrupted && clear_interrupted) {
3653     osthread->set_interrupted(false);
3654     // consider thread->_SleepEvent->reset() ... optional optimization
3655   }
3656 
3657   return interrupted;
3658 }
3659 
3660 ///////////////////////////////////////////////////////////////////////////////////
3661 // signal handling (except suspend/resume)
3662 
3663 // This routine may be used by user applications as a "hook" to catch signals.
3664 // The user-defined signal handler must pass unrecognized signals to this
3665 // routine, and if it returns true (non-zero), then the signal handler must
3666 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3667 // routine will never retun false (zero), but instead will execute a VM panic
3668 // routine kill the process.
3669 //
3670 // If this routine returns false, it is OK to call it again.  This allows
3671 // the user-defined signal handler to perform checks either before or after
3672 // the VM performs its own checks.  Naturally, the user code would be making
3673 // a serious error if it tried to handle an exception (such as a null check
3674 // or breakpoint) that the VM was generating for its own correct operation.
3675 //
3676 // This routine may recognize any of the following kinds of signals:
3677 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3678 // It should be consulted by handlers for any of those signals.
3679 //
3680 // The caller of this routine must pass in the three arguments supplied
3681 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3682 // field of the structure passed to sigaction().  This routine assumes that
3683 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3684 //
3685 // Note that the VM will print warnings if it detects conflicting signal
3686 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3687 //
3688 extern "C" JNIEXPORT int
3689 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3690                         void* ucontext, int abort_if_unrecognized);
3691 
3692 void signalHandler(int sig, siginfo_t* info, void* uc) {
3693   assert(info != NULL && uc != NULL, "it must be old kernel");
3694   JVM_handle_linux_signal(sig, info, uc, true);
3695 }
3696 
3697 
3698 // This boolean allows users to forward their own non-matching signals
3699 // to JVM_handle_linux_signal, harmlessly.
3700 bool os::Linux::signal_handlers_are_installed = false;
3701 
3702 // For signal-chaining
3703 struct sigaction os::Linux::sigact[MAXSIGNUM];
3704 unsigned int os::Linux::sigs = 0;
3705 bool os::Linux::libjsig_is_loaded = false;
3706 typedef struct sigaction *(*get_signal_t)(int);
3707 get_signal_t os::Linux::get_signal_action = NULL;
3708 
3709 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3710   struct sigaction *actp = NULL;
3711 
3712   if (libjsig_is_loaded) {
3713     // Retrieve the old signal handler from libjsig
3714     actp = (*get_signal_action)(sig);
3715   }
3716   if (actp == NULL) {
3717     // Retrieve the preinstalled signal handler from jvm
3718     actp = get_preinstalled_handler(sig);
3719   }
3720 
3721   return actp;
3722 }
3723 
3724 static bool call_chained_handler(struct sigaction *actp, int sig,
3725                                  siginfo_t *siginfo, void *context) {
3726   // Call the old signal handler
3727   if (actp->sa_handler == SIG_DFL) {
3728     // It's more reasonable to let jvm treat it as an unexpected exception
3729     // instead of taking the default action.
3730     return false;
3731   } else if (actp->sa_handler != SIG_IGN) {
3732     if ((actp->sa_flags & SA_NODEFER) == 0) {
3733       // automaticlly block the signal
3734       sigaddset(&(actp->sa_mask), sig);
3735     }
3736 
3737     sa_handler_t hand;
3738     sa_sigaction_t sa;
3739     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3740     // retrieve the chained handler
3741     if (siginfo_flag_set) {
3742       sa = actp->sa_sigaction;
3743     } else {
3744       hand = actp->sa_handler;
3745     }
3746 
3747     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3748       actp->sa_handler = SIG_DFL;
3749     }
3750 
3751     // try to honor the signal mask
3752     sigset_t oset;
3753     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3754 
3755     // call into the chained handler
3756     if (siginfo_flag_set) {
3757       (*sa)(sig, siginfo, context);
3758     } else {
3759       (*hand)(sig);
3760     }
3761 
3762     // restore the signal mask
3763     pthread_sigmask(SIG_SETMASK, &oset, 0);
3764   }
3765   // Tell jvm's signal handler the signal is taken care of.
3766   return true;
3767 }
3768 
3769 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3770   bool chained = false;
3771   // signal-chaining
3772   if (UseSignalChaining) {
3773     struct sigaction *actp = get_chained_signal_action(sig);
3774     if (actp != NULL) {
3775       chained = call_chained_handler(actp, sig, siginfo, context);
3776     }
3777   }
3778   return chained;
3779 }
3780 
3781 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3782   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3783     return &sigact[sig];
3784   }
3785   return NULL;
3786 }
3787 
3788 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3789   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3790   sigact[sig] = oldAct;
3791   sigs |= (unsigned int)1 << sig;
3792 }
3793 
3794 // for diagnostic
3795 int os::Linux::sigflags[MAXSIGNUM];
3796 
3797 int os::Linux::get_our_sigflags(int sig) {
3798   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3799   return sigflags[sig];
3800 }
3801 
3802 void os::Linux::set_our_sigflags(int sig, int flags) {
3803   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3804   sigflags[sig] = flags;
3805 }
3806 
3807 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3808   // Check for overwrite.
3809   struct sigaction oldAct;
3810   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3811 
3812   void* oldhand = oldAct.sa_sigaction
3813                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3814                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3815   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3816       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3817       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3818     if (AllowUserSignalHandlers || !set_installed) {
3819       // Do not overwrite; user takes responsibility to forward to us.
3820       return;
3821     } else if (UseSignalChaining) {
3822       // save the old handler in jvm
3823       save_preinstalled_handler(sig, oldAct);
3824       // libjsig also interposes the sigaction() call below and saves the
3825       // old sigaction on it own.
3826     } else {
3827       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3828                     "%#lx for signal %d.", (long)oldhand, sig));
3829     }
3830   }
3831 
3832   struct sigaction sigAct;
3833   sigfillset(&(sigAct.sa_mask));
3834   sigAct.sa_handler = SIG_DFL;
3835   if (!set_installed) {
3836     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3837   } else {
3838     sigAct.sa_sigaction = signalHandler;
3839     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3840   }
3841   // Save flags, which are set by ours
3842   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3843   sigflags[sig] = sigAct.sa_flags;
3844 
3845   int ret = sigaction(sig, &sigAct, &oldAct);
3846   assert(ret == 0, "check");
3847 
3848   void* oldhand2  = oldAct.sa_sigaction
3849                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3850                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3851   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3852 }
3853 
3854 // install signal handlers for signals that HotSpot needs to
3855 // handle in order to support Java-level exception handling.
3856 
3857 void os::Linux::install_signal_handlers() {
3858   if (!signal_handlers_are_installed) {
3859     signal_handlers_are_installed = true;
3860 
3861     // signal-chaining
3862     typedef void (*signal_setting_t)();
3863     signal_setting_t begin_signal_setting = NULL;
3864     signal_setting_t end_signal_setting = NULL;
3865     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3866                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3867     if (begin_signal_setting != NULL) {
3868       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3869                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3870       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3871                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3872       libjsig_is_loaded = true;
3873       assert(UseSignalChaining, "should enable signal-chaining");
3874     }
3875     if (libjsig_is_loaded) {
3876       // Tell libjsig jvm is setting signal handlers
3877       (*begin_signal_setting)();
3878     }
3879 
3880     set_signal_handler(SIGSEGV, true);
3881     set_signal_handler(SIGPIPE, true);
3882     set_signal_handler(SIGBUS, true);
3883     set_signal_handler(SIGILL, true);
3884     set_signal_handler(SIGFPE, true);
3885     set_signal_handler(SIGXFSZ, true);
3886 
3887     if (libjsig_is_loaded) {
3888       // Tell libjsig jvm finishes setting signal handlers
3889       (*end_signal_setting)();
3890     }
3891 
3892     // We don't activate signal checker if libjsig is in place, we trust ourselves
3893     // and if UserSignalHandler is installed all bets are off.
3894     // Log that signal checking is off only if -verbose:jni is specified.
3895     if (CheckJNICalls) {
3896       if (libjsig_is_loaded) {
3897         if (PrintJNIResolving) {
3898           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3899         }
3900         check_signals = false;
3901       }
3902       if (AllowUserSignalHandlers) {
3903         if (PrintJNIResolving) {
3904           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3905         }
3906         check_signals = false;
3907       }
3908     }
3909   }
3910 }
3911 
3912 // This is the fastest way to get thread cpu time on Linux.
3913 // Returns cpu time (user+sys) for any thread, not only for current.
3914 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3915 // It might work on 2.6.10+ with a special kernel/glibc patch.
3916 // For reference, please, see IEEE Std 1003.1-2004:
3917 //   http://www.unix.org/single_unix_specification
3918 
3919 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3920   struct timespec tp;
3921   int rc = os::Linux::clock_gettime(clockid, &tp);
3922   assert(rc == 0, "clock_gettime is expected to return 0 code");
3923 
3924   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
3925 }
3926 
3927 /////
3928 // glibc on Linux platform uses non-documented flag
3929 // to indicate, that some special sort of signal
3930 // trampoline is used.
3931 // We will never set this flag, and we should
3932 // ignore this flag in our diagnostic
3933 #ifdef SIGNIFICANT_SIGNAL_MASK
3934 #undef SIGNIFICANT_SIGNAL_MASK
3935 #endif
3936 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3937 
3938 static const char* get_signal_handler_name(address handler,
3939                                            char* buf, int buflen) {
3940   int offset;
3941   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3942   if (found) {
3943     // skip directory names
3944     const char *p1, *p2;
3945     p1 = buf;
3946     size_t len = strlen(os::file_separator());
3947     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3948     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3949   } else {
3950     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3951   }
3952   return buf;
3953 }
3954 
3955 static void print_signal_handler(outputStream* st, int sig,
3956                                  char* buf, size_t buflen) {
3957   struct sigaction sa;
3958 
3959   sigaction(sig, NULL, &sa);
3960 
3961   // See comment for SIGNIFICANT_SIGNAL_MASK define
3962   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3963 
3964   st->print("%s: ", os::exception_name(sig, buf, buflen));
3965 
3966   address handler = (sa.sa_flags & SA_SIGINFO)
3967     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3968     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3969 
3970   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3971     st->print("SIG_DFL");
3972   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3973     st->print("SIG_IGN");
3974   } else {
3975     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3976   }
3977 
3978   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3979 
3980   address rh = VMError::get_resetted_sighandler(sig);
3981   // May be, handler was resetted by VMError?
3982   if(rh != NULL) {
3983     handler = rh;
3984     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3985   }
3986 
3987   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
3988 
3989   // Check: is it our handler?
3990   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3991      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3992     // It is our signal handler
3993     // check for flags, reset system-used one!
3994     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3995       st->print(
3996                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3997                 os::Linux::get_our_sigflags(sig));
3998     }
3999   }
4000   st->cr();
4001 }
4002 
4003 
4004 #define DO_SIGNAL_CHECK(sig) \
4005   if (!sigismember(&check_signal_done, sig)) \
4006     os::Linux::check_signal_handler(sig)
4007 
4008 // This method is a periodic task to check for misbehaving JNI applications
4009 // under CheckJNI, we can add any periodic checks here
4010 
4011 void os::run_periodic_checks() {
4012 
4013   if (check_signals == false) return;
4014 
4015   // SEGV and BUS if overridden could potentially prevent
4016   // generation of hs*.log in the event of a crash, debugging
4017   // such a case can be very challenging, so we absolutely
4018   // check the following for a good measure:
4019   DO_SIGNAL_CHECK(SIGSEGV);
4020   DO_SIGNAL_CHECK(SIGILL);
4021   DO_SIGNAL_CHECK(SIGFPE);
4022   DO_SIGNAL_CHECK(SIGBUS);
4023   DO_SIGNAL_CHECK(SIGPIPE);
4024   DO_SIGNAL_CHECK(SIGXFSZ);
4025 
4026 
4027   // ReduceSignalUsage allows the user to override these handlers
4028   // see comments at the very top and jvm_solaris.h
4029   if (!ReduceSignalUsage) {
4030     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4031     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4032     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4033     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4034   }
4035 
4036   DO_SIGNAL_CHECK(SR_signum);
4037   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4038 }
4039 
4040 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4041 
4042 static os_sigaction_t os_sigaction = NULL;
4043 
4044 void os::Linux::check_signal_handler(int sig) {
4045   char buf[O_BUFLEN];
4046   address jvmHandler = NULL;
4047 
4048 
4049   struct sigaction act;
4050   if (os_sigaction == NULL) {
4051     // only trust the default sigaction, in case it has been interposed
4052     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4053     if (os_sigaction == NULL) return;
4054   }
4055 
4056   os_sigaction(sig, (struct sigaction*)NULL, &act);
4057 
4058 
4059   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4060 
4061   address thisHandler = (act.sa_flags & SA_SIGINFO)
4062     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4063     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4064 
4065 
4066   switch(sig) {
4067   case SIGSEGV:
4068   case SIGBUS:
4069   case SIGFPE:
4070   case SIGPIPE:
4071   case SIGILL:
4072   case SIGXFSZ:
4073     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4074     break;
4075 
4076   case SHUTDOWN1_SIGNAL:
4077   case SHUTDOWN2_SIGNAL:
4078   case SHUTDOWN3_SIGNAL:
4079   case BREAK_SIGNAL:
4080     jvmHandler = (address)user_handler();
4081     break;
4082 
4083   case INTERRUPT_SIGNAL:
4084     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4085     break;
4086 
4087   default:
4088     if (sig == SR_signum) {
4089       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4090     } else {
4091       return;
4092     }
4093     break;
4094   }
4095 
4096   if (thisHandler != jvmHandler) {
4097     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4098     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4099     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4100     // No need to check this sig any longer
4101     sigaddset(&check_signal_done, sig);
4102   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4103     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4104     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4105     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4106     // No need to check this sig any longer
4107     sigaddset(&check_signal_done, sig);
4108   }
4109 
4110   // Dump all the signal
4111   if (sigismember(&check_signal_done, sig)) {
4112     print_signal_handlers(tty, buf, O_BUFLEN);
4113   }
4114 }
4115 
4116 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4117 
4118 extern bool signal_name(int signo, char* buf, size_t len);
4119 
4120 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4121   if (0 < exception_code && exception_code <= SIGRTMAX) {
4122     // signal
4123     if (!signal_name(exception_code, buf, size)) {
4124       jio_snprintf(buf, size, "SIG%d", exception_code);
4125     }
4126     return buf;
4127   } else {
4128     return NULL;
4129   }
4130 }
4131 
4132 // this is called _before_ the most of global arguments have been parsed
4133 void os::init(void) {
4134   char dummy;   /* used to get a guess on initial stack address */
4135 //  first_hrtime = gethrtime();
4136 
4137   // With LinuxThreads the JavaMain thread pid (primordial thread)
4138   // is different than the pid of the java launcher thread.
4139   // So, on Linux, the launcher thread pid is passed to the VM
4140   // via the sun.java.launcher.pid property.
4141   // Use this property instead of getpid() if it was correctly passed.
4142   // See bug 6351349.
4143   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4144 
4145   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4146 
4147   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4148 
4149   init_random(1234567);
4150 
4151   ThreadCritical::initialize();
4152 
4153   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4154   if (Linux::page_size() == -1) {
4155     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4156                   strerror(errno)));
4157   }
4158   init_page_sizes((size_t) Linux::page_size());
4159 
4160   Linux::initialize_system_info();
4161 
4162   // main_thread points to the aboriginal thread
4163   Linux::_main_thread = pthread_self();
4164 
4165   Linux::clock_init();
4166   initial_time_count = os::elapsed_counter();
4167   pthread_mutex_init(&dl_mutex, NULL);
4168 }
4169 
4170 // To install functions for atexit system call
4171 extern "C" {
4172   static void perfMemory_exit_helper() {
4173     perfMemory_exit();
4174   }
4175 }
4176 
4177 // this is called _after_ the global arguments have been parsed
4178 jint os::init_2(void)
4179 {
4180   Linux::fast_thread_clock_init();
4181 
4182   // Allocate a single page and mark it as readable for safepoint polling
4183   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4184   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4185 
4186   os::set_polling_page( polling_page );
4187 
4188 #ifndef PRODUCT
4189   if(Verbose && PrintMiscellaneous)
4190     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4191 #endif
4192 
4193   if (!UseMembar) {
4194     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4195     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4196     os::set_memory_serialize_page( mem_serialize_page );
4197 
4198 #ifndef PRODUCT
4199     if(Verbose && PrintMiscellaneous)
4200       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4201 #endif
4202   }
4203 
4204   os::large_page_init();
4205 
4206   // initialize suspend/resume support - must do this before signal_sets_init()
4207   if (SR_initialize() != 0) {
4208     perror("SR_initialize failed");
4209     return JNI_ERR;
4210   }
4211 
4212   Linux::signal_sets_init();
4213   Linux::install_signal_handlers();
4214 
4215   // Check minimum allowable stack size for thread creation and to initialize
4216   // the java system classes, including StackOverflowError - depends on page
4217   // size.  Add a page for compiler2 recursion in main thread.
4218   // Add in 2*BytesPerWord times page size to account for VM stack during
4219   // class initialization depending on 32 or 64 bit VM.
4220   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4221             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4222                     2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4223 
4224   size_t threadStackSizeInBytes = ThreadStackSize * K;
4225   if (threadStackSizeInBytes != 0 &&
4226       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4227         tty->print_cr("\nThe stack size specified is too small, "
4228                       "Specify at least %dk",
4229                       os::Linux::min_stack_allowed/ K);
4230         return JNI_ERR;
4231   }
4232 
4233   // Make the stack size a multiple of the page size so that
4234   // the yellow/red zones can be guarded.
4235   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4236         vm_page_size()));
4237 
4238   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4239 
4240   Linux::libpthread_init();
4241   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4242      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4243           Linux::glibc_version(), Linux::libpthread_version(),
4244           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4245   }
4246 
4247   if (UseNUMA) {
4248     if (!Linux::libnuma_init()) {
4249       UseNUMA = false;
4250     } else {
4251       if ((Linux::numa_max_node() < 1)) {
4252         // There's only one node(they start from 0), disable NUMA.
4253         UseNUMA = false;
4254       }
4255     }
4256     // With SHM large pages we cannot uncommit a page, so there's not way
4257     // we can make the adaptive lgrp chunk resizing work. If the user specified
4258     // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4259     // disable adaptive resizing.
4260     if (UseNUMA && UseLargePages && UseSHM) {
4261       if (!FLAG_IS_DEFAULT(UseNUMA)) {
4262         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4263           UseLargePages = false;
4264         } else {
4265           warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4266           UseAdaptiveSizePolicy = false;
4267           UseAdaptiveNUMAChunkSizing = false;
4268         }
4269       } else {
4270         UseNUMA = false;
4271       }
4272     }
4273     if (!UseNUMA && ForceNUMA) {
4274       UseNUMA = true;
4275     }
4276   }
4277 
4278   if (MaxFDLimit) {
4279     // set the number of file descriptors to max. print out error
4280     // if getrlimit/setrlimit fails but continue regardless.
4281     struct rlimit nbr_files;
4282     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4283     if (status != 0) {
4284       if (PrintMiscellaneous && (Verbose || WizardMode))
4285         perror("os::init_2 getrlimit failed");
4286     } else {
4287       nbr_files.rlim_cur = nbr_files.rlim_max;
4288       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4289       if (status != 0) {
4290         if (PrintMiscellaneous && (Verbose || WizardMode))
4291           perror("os::init_2 setrlimit failed");
4292       }
4293     }
4294   }
4295 
4296   // Initialize lock used to serialize thread creation (see os::create_thread)
4297   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4298 
4299   // at-exit methods are called in the reverse order of their registration.
4300   // atexit functions are called on return from main or as a result of a
4301   // call to exit(3C). There can be only 32 of these functions registered
4302   // and atexit() does not set errno.
4303 
4304   if (PerfAllowAtExitRegistration) {
4305     // only register atexit functions if PerfAllowAtExitRegistration is set.
4306     // atexit functions can be delayed until process exit time, which
4307     // can be problematic for embedded VM situations. Embedded VMs should
4308     // call DestroyJavaVM() to assure that VM resources are released.
4309 
4310     // note: perfMemory_exit_helper atexit function may be removed in
4311     // the future if the appropriate cleanup code can be added to the
4312     // VM_Exit VMOperation's doit method.
4313     if (atexit(perfMemory_exit_helper) != 0) {
4314       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4315     }
4316   }
4317 
4318   // initialize thread priority policy
4319   prio_init();
4320 
4321   return JNI_OK;
4322 }
4323 
4324 // this is called at the end of vm_initialization
4325 void os::init_3(void)
4326 {
4327 #ifdef JAVASE_EMBEDDED
4328   // Start the MemNotifyThread
4329   if (LowMemoryProtection) {
4330     MemNotifyThread::start();
4331   }
4332   return;
4333 #endif
4334 }
4335 
4336 // Mark the polling page as unreadable
4337 void os::make_polling_page_unreadable(void) {
4338   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4339     fatal("Could not disable polling page");
4340 };
4341 
4342 // Mark the polling page as readable
4343 void os::make_polling_page_readable(void) {
4344   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4345     fatal("Could not enable polling page");
4346   }
4347 };
4348 
4349 int os::active_processor_count() {
4350   // Linux doesn't yet have a (official) notion of processor sets,
4351   // so just return the number of online processors.
4352   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4353   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4354   return online_cpus;
4355 }
4356 
4357 void os::set_native_thread_name(const char *name) {
4358   // Not yet implemented.
4359   return;
4360 }
4361 
4362 bool os::distribute_processes(uint length, uint* distribution) {
4363   // Not yet implemented.
4364   return false;
4365 }
4366 
4367 bool os::bind_to_processor(uint processor_id) {
4368   // Not yet implemented.
4369   return false;
4370 }
4371 
4372 ///
4373 
4374 // Suspends the target using the signal mechanism and then grabs the PC before
4375 // resuming the target. Used by the flat-profiler only
4376 ExtendedPC os::get_thread_pc(Thread* thread) {
4377   // Make sure that it is called by the watcher for the VMThread
4378   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4379   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4380 
4381   ExtendedPC epc;
4382 
4383   OSThread* osthread = thread->osthread();
4384   if (do_suspend(osthread)) {
4385     if (osthread->ucontext() != NULL) {
4386       epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4387     } else {
4388       // NULL context is unexpected, double-check this is the VMThread
4389       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4390     }
4391     do_resume(osthread);
4392   }
4393   // failure means pthread_kill failed for some reason - arguably this is
4394   // a fatal problem, but such problems are ignored elsewhere
4395 
4396   return epc;
4397 }
4398 
4399 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4400 {
4401    if (is_NPTL()) {
4402       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4403    } else {
4404 #ifndef IA64
4405       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4406       // word back to default 64bit precision if condvar is signaled. Java
4407       // wants 53bit precision.  Save and restore current value.
4408       int fpu = get_fpu_control_word();
4409 #endif // IA64
4410       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4411 #ifndef IA64
4412       set_fpu_control_word(fpu);
4413 #endif // IA64
4414       return status;
4415    }
4416 }
4417 
4418 ////////////////////////////////////////////////////////////////////////////////
4419 // debug support
4420 
4421 static address same_page(address x, address y) {
4422   int page_bits = -os::vm_page_size();
4423   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4424     return x;
4425   else if (x > y)
4426     return (address)(intptr_t(y) | ~page_bits) + 1;
4427   else
4428     return (address)(intptr_t(y) & page_bits);
4429 }
4430 
4431 bool os::find(address addr, outputStream* st) {
4432   Dl_info dlinfo;
4433   memset(&dlinfo, 0, sizeof(dlinfo));
4434   if (dladdr(addr, &dlinfo)) {
4435     st->print(PTR_FORMAT ": ", addr);
4436     if (dlinfo.dli_sname != NULL) {
4437       st->print("%s+%#x", dlinfo.dli_sname,
4438                  addr - (intptr_t)dlinfo.dli_saddr);
4439     } else if (dlinfo.dli_fname) {
4440       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4441     } else {
4442       st->print("<absolute address>");
4443     }
4444     if (dlinfo.dli_fname) {
4445       st->print(" in %s", dlinfo.dli_fname);
4446     }
4447     if (dlinfo.dli_fbase) {
4448       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4449     }
4450     st->cr();
4451 
4452     if (Verbose) {
4453       // decode some bytes around the PC
4454       address begin = same_page(addr-40, addr);
4455       address end   = same_page(addr+40, addr);
4456       address       lowest = (address) dlinfo.dli_sname;
4457       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4458       if (begin < lowest)  begin = lowest;
4459       Dl_info dlinfo2;
4460       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4461           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4462         end = (address) dlinfo2.dli_saddr;
4463       Disassembler::decode(begin, end, st);
4464     }
4465     return true;
4466   }
4467   return false;
4468 }
4469 
4470 ////////////////////////////////////////////////////////////////////////////////
4471 // misc
4472 
4473 // This does not do anything on Linux. This is basically a hook for being
4474 // able to use structured exception handling (thread-local exception filters)
4475 // on, e.g., Win32.
4476 void
4477 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4478                          JavaCallArguments* args, Thread* thread) {
4479   f(value, method, args, thread);
4480 }
4481 
4482 void os::print_statistics() {
4483 }
4484 
4485 int os::message_box(const char* title, const char* message) {
4486   int i;
4487   fdStream err(defaultStream::error_fd());
4488   for (i = 0; i < 78; i++) err.print_raw("=");
4489   err.cr();
4490   err.print_raw_cr(title);
4491   for (i = 0; i < 78; i++) err.print_raw("-");
4492   err.cr();
4493   err.print_raw_cr(message);
4494   for (i = 0; i < 78; i++) err.print_raw("=");
4495   err.cr();
4496 
4497   char buf[16];
4498   // Prevent process from exiting upon "read error" without consuming all CPU
4499   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4500 
4501   return buf[0] == 'y' || buf[0] == 'Y';
4502 }
4503 
4504 int os::stat(const char *path, struct stat *sbuf) {
4505   char pathbuf[MAX_PATH];
4506   if (strlen(path) > MAX_PATH - 1) {
4507     errno = ENAMETOOLONG;
4508     return -1;
4509   }
4510   os::native_path(strcpy(pathbuf, path));
4511   return ::stat(pathbuf, sbuf);
4512 }
4513 
4514 bool os::check_heap(bool force) {
4515   return true;
4516 }
4517 
4518 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4519   return ::vsnprintf(buf, count, format, args);
4520 }
4521 
4522 // Is a (classpath) directory empty?
4523 bool os::dir_is_empty(const char* path) {
4524   DIR *dir = NULL;
4525   struct dirent *ptr;
4526 
4527   dir = opendir(path);
4528   if (dir == NULL) return true;
4529 
4530   /* Scan the directory */
4531   bool result = true;
4532   char buf[sizeof(struct dirent) + MAX_PATH];
4533   while (result && (ptr = ::readdir(dir)) != NULL) {
4534     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4535       result = false;
4536     }
4537   }
4538   closedir(dir);
4539   return result;
4540 }
4541 
4542 // This code originates from JDK's sysOpen and open64_w
4543 // from src/solaris/hpi/src/system_md.c
4544 
4545 #ifndef O_DELETE
4546 #define O_DELETE 0x10000
4547 #endif
4548 
4549 // Open a file. Unlink the file immediately after open returns
4550 // if the specified oflag has the O_DELETE flag set.
4551 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4552 
4553 int os::open(const char *path, int oflag, int mode) {
4554 
4555   if (strlen(path) > MAX_PATH - 1) {
4556     errno = ENAMETOOLONG;
4557     return -1;
4558   }
4559   int fd;
4560   int o_delete = (oflag & O_DELETE);
4561   oflag = oflag & ~O_DELETE;
4562 
4563   fd = ::open64(path, oflag, mode);
4564   if (fd == -1) return -1;
4565 
4566   //If the open succeeded, the file might still be a directory
4567   {
4568     struct stat64 buf64;
4569     int ret = ::fstat64(fd, &buf64);
4570     int st_mode = buf64.st_mode;
4571 
4572     if (ret != -1) {
4573       if ((st_mode & S_IFMT) == S_IFDIR) {
4574         errno = EISDIR;
4575         ::close(fd);
4576         return -1;
4577       }
4578     } else {
4579       ::close(fd);
4580       return -1;
4581     }
4582   }
4583 
4584     /*
4585      * All file descriptors that are opened in the JVM and not
4586      * specifically destined for a subprocess should have the
4587      * close-on-exec flag set.  If we don't set it, then careless 3rd
4588      * party native code might fork and exec without closing all
4589      * appropriate file descriptors (e.g. as we do in closeDescriptors in
4590      * UNIXProcess.c), and this in turn might:
4591      *
4592      * - cause end-of-file to fail to be detected on some file
4593      *   descriptors, resulting in mysterious hangs, or
4594      *
4595      * - might cause an fopen in the subprocess to fail on a system
4596      *   suffering from bug 1085341.
4597      *
4598      * (Yes, the default setting of the close-on-exec flag is a Unix
4599      * design flaw)
4600      *
4601      * See:
4602      * 1085341: 32-bit stdio routines should support file descriptors >255
4603      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4604      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4605      */
4606 #ifdef FD_CLOEXEC
4607     {
4608         int flags = ::fcntl(fd, F_GETFD);
4609         if (flags != -1)
4610             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4611     }
4612 #endif
4613 
4614   if (o_delete != 0) {
4615     ::unlink(path);
4616   }
4617   return fd;
4618 }
4619 
4620 
4621 // create binary file, rewriting existing file if required
4622 int os::create_binary_file(const char* path, bool rewrite_existing) {
4623   int oflags = O_WRONLY | O_CREAT;
4624   if (!rewrite_existing) {
4625     oflags |= O_EXCL;
4626   }
4627   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4628 }
4629 
4630 // return current position of file pointer
4631 jlong os::current_file_offset(int fd) {
4632   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4633 }
4634 
4635 // move file pointer to the specified offset
4636 jlong os::seek_to_file_offset(int fd, jlong offset) {
4637   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4638 }
4639 
4640 // This code originates from JDK's sysAvailable
4641 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4642 
4643 int os::available(int fd, jlong *bytes) {
4644   jlong cur, end;
4645   int mode;
4646   struct stat64 buf64;
4647 
4648   if (::fstat64(fd, &buf64) >= 0) {
4649     mode = buf64.st_mode;
4650     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4651       /*
4652       * XXX: is the following call interruptible? If so, this might
4653       * need to go through the INTERRUPT_IO() wrapper as for other
4654       * blocking, interruptible calls in this file.
4655       */
4656       int n;
4657       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4658         *bytes = n;
4659         return 1;
4660       }
4661     }
4662   }
4663   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4664     return 0;
4665   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4666     return 0;
4667   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4668     return 0;
4669   }
4670   *bytes = end - cur;
4671   return 1;
4672 }
4673 
4674 int os::socket_available(int fd, jint *pbytes) {
4675   // Linux doc says EINTR not returned, unlike Solaris
4676   int ret = ::ioctl(fd, FIONREAD, pbytes);
4677 
4678   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4679   // is expected to return 0 on failure and 1 on success to the jdk.
4680   return (ret < 0) ? 0 : 1;
4681 }
4682 
4683 // Map a block of memory.
4684 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4685                      char *addr, size_t bytes, bool read_only,
4686                      bool allow_exec) {
4687   int prot;
4688   int flags;
4689 
4690   if (read_only) {
4691     prot = PROT_READ;
4692     flags = MAP_SHARED;
4693   } else {
4694     prot = PROT_READ | PROT_WRITE;
4695     flags = MAP_PRIVATE;
4696   }
4697 
4698   if (allow_exec) {
4699     prot |= PROT_EXEC;
4700   }
4701 
4702   if (addr != NULL) {
4703     flags |= MAP_FIXED;
4704   }
4705 
4706   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4707                                      fd, file_offset);
4708   if (mapped_address == MAP_FAILED) {
4709     return NULL;
4710   }
4711   return mapped_address;
4712 }
4713 
4714 
4715 // Remap a block of memory.
4716 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4717                        char *addr, size_t bytes, bool read_only,
4718                        bool allow_exec) {
4719   // same as map_memory() on this OS
4720   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4721                         allow_exec);
4722 }
4723 
4724 
4725 // Unmap a block of memory.
4726 bool os::unmap_memory(char* addr, size_t bytes) {
4727   return munmap(addr, bytes) == 0;
4728 }
4729 
4730 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4731 
4732 static clockid_t thread_cpu_clockid(Thread* thread) {
4733   pthread_t tid = thread->osthread()->pthread_id();
4734   clockid_t clockid;
4735 
4736   // Get thread clockid
4737   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4738   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4739   return clockid;
4740 }
4741 
4742 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4743 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4744 // of a thread.
4745 //
4746 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4747 // the fast estimate available on the platform.
4748 
4749 jlong os::current_thread_cpu_time() {
4750   if (os::Linux::supports_fast_thread_cpu_time()) {
4751     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4752   } else {
4753     // return user + sys since the cost is the same
4754     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4755   }
4756 }
4757 
4758 jlong os::thread_cpu_time(Thread* thread) {
4759   // consistent with what current_thread_cpu_time() returns
4760   if (os::Linux::supports_fast_thread_cpu_time()) {
4761     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4762   } else {
4763     return slow_thread_cpu_time(thread, true /* user + sys */);
4764   }
4765 }
4766 
4767 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4768   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4769     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4770   } else {
4771     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4772   }
4773 }
4774 
4775 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4776   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4777     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4778   } else {
4779     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4780   }
4781 }
4782 
4783 //
4784 //  -1 on error.
4785 //
4786 
4787 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4788   static bool proc_pid_cpu_avail = true;
4789   static bool proc_task_unchecked = true;
4790   static const char *proc_stat_path = "/proc/%d/stat";
4791   pid_t  tid = thread->osthread()->thread_id();
4792   int i;
4793   char *s;
4794   char stat[2048];
4795   int statlen;
4796   char proc_name[64];
4797   int count;
4798   long sys_time, user_time;
4799   char string[64];
4800   char cdummy;
4801   int idummy;
4802   long ldummy;
4803   FILE *fp;
4804 
4805   // We first try accessing /proc/<pid>/cpu since this is faster to
4806   // process.  If this file is not present (linux kernels 2.5 and above)
4807   // then we open /proc/<pid>/stat.
4808   if ( proc_pid_cpu_avail ) {
4809     sprintf(proc_name, "/proc/%d/cpu", tid);
4810     fp =  fopen(proc_name, "r");
4811     if ( fp != NULL ) {
4812       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4813       fclose(fp);
4814       if ( count != 3 ) return -1;
4815 
4816       if (user_sys_cpu_time) {
4817         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4818       } else {
4819         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4820       }
4821     }
4822     else proc_pid_cpu_avail = false;
4823   }
4824 
4825   // The /proc/<tid>/stat aggregates per-process usage on
4826   // new Linux kernels 2.6+ where NPTL is supported.
4827   // The /proc/self/task/<tid>/stat still has the per-thread usage.
4828   // See bug 6328462.
4829   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4830   // and possibly in some other cases, so we check its availability.
4831   if (proc_task_unchecked && os::Linux::is_NPTL()) {
4832     // This is executed only once
4833     proc_task_unchecked = false;
4834     fp = fopen("/proc/self/task", "r");
4835     if (fp != NULL) {
4836       proc_stat_path = "/proc/self/task/%d/stat";
4837       fclose(fp);
4838     }
4839   }
4840 
4841   sprintf(proc_name, proc_stat_path, tid);
4842   fp = fopen(proc_name, "r");
4843   if ( fp == NULL ) return -1;
4844   statlen = fread(stat, 1, 2047, fp);
4845   stat[statlen] = '\0';
4846   fclose(fp);
4847 
4848   // Skip pid and the command string. Note that we could be dealing with
4849   // weird command names, e.g. user could decide to rename java launcher
4850   // to "java 1.4.2 :)", then the stat file would look like
4851   //                1234 (java 1.4.2 :)) R ... ...
4852   // We don't really need to know the command string, just find the last
4853   // occurrence of ")" and then start parsing from there. See bug 4726580.
4854   s = strrchr(stat, ')');
4855   i = 0;
4856   if (s == NULL ) return -1;
4857 
4858   // Skip blank chars
4859   do s++; while (isspace(*s));
4860 
4861   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4862                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4863                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4864                  &user_time, &sys_time);
4865   if ( count != 13 ) return -1;
4866   if (user_sys_cpu_time) {
4867     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4868   } else {
4869     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4870   }
4871 }
4872 
4873 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4874   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4875   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4876   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4877   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4878 }
4879 
4880 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4881   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4882   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4883   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4884   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4885 }
4886 
4887 bool os::is_thread_cpu_time_supported() {
4888   return true;
4889 }
4890 
4891 // System loadavg support.  Returns -1 if load average cannot be obtained.
4892 // Linux doesn't yet have a (official) notion of processor sets,
4893 // so just return the system wide load average.
4894 int os::loadavg(double loadavg[], int nelem) {
4895   return ::getloadavg(loadavg, nelem);
4896 }
4897 
4898 void os::pause() {
4899   char filename[MAX_PATH];
4900   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4901     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4902   } else {
4903     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4904   }
4905 
4906   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4907   if (fd != -1) {
4908     struct stat buf;
4909     ::close(fd);
4910     while (::stat(filename, &buf) == 0) {
4911       (void)::poll(NULL, 0, 100);
4912     }
4913   } else {
4914     jio_fprintf(stderr,
4915       "Could not open pause file '%s', continuing immediately.\n", filename);
4916   }
4917 }
4918 
4919 
4920 // Refer to the comments in os_solaris.cpp park-unpark.
4921 //
4922 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4923 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4924 // For specifics regarding the bug see GLIBC BUGID 261237 :
4925 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4926 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4927 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4928 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4929 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4930 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4931 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4932 // of libpthread avoids the problem, but isn't practical.
4933 //
4934 // Possible remedies:
4935 //
4936 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4937 //      This is palliative and probabilistic, however.  If the thread is preempted
4938 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4939 //      than the minimum period may have passed, and the abstime may be stale (in the
4940 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
4941 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4942 //
4943 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4944 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
4945 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4946 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
4947 //      thread.
4948 //
4949 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4950 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
4951 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4952 //      This also works well.  In fact it avoids kernel-level scalability impediments
4953 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4954 //      timers in a graceful fashion.
4955 //
4956 // 4.   When the abstime value is in the past it appears that control returns
4957 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4958 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
4959 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
4960 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
4961 //      It may be possible to avoid reinitialization by checking the return
4962 //      value from pthread_cond_timedwait().  In addition to reinitializing the
4963 //      condvar we must establish the invariant that cond_signal() is only called
4964 //      within critical sections protected by the adjunct mutex.  This prevents
4965 //      cond_signal() from "seeing" a condvar that's in the midst of being
4966 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
4967 //      desirable signal-after-unlock optimization that avoids futile context switching.
4968 //
4969 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
4970 //      structure when a condvar is used or initialized.  cond_destroy()  would
4971 //      release the helper structure.  Our reinitialize-after-timedwait fix
4972 //      put excessive stress on malloc/free and locks protecting the c-heap.
4973 //
4974 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
4975 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4976 // and only enabling the work-around for vulnerable environments.
4977 
4978 // utility to compute the abstime argument to timedwait:
4979 // millis is the relative timeout time
4980 // abstime will be the absolute timeout time
4981 // TODO: replace compute_abstime() with unpackTime()
4982 
4983 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4984   if (millis < 0)  millis = 0;
4985   struct timeval now;
4986   int status = gettimeofday(&now, NULL);
4987   assert(status == 0, "gettimeofday");
4988   jlong seconds = millis / 1000;
4989   millis %= 1000;
4990   if (seconds > 50000000) { // see man cond_timedwait(3T)
4991     seconds = 50000000;
4992   }
4993   abstime->tv_sec = now.tv_sec  + seconds;
4994   long       usec = now.tv_usec + millis * 1000;
4995   if (usec >= 1000000) {
4996     abstime->tv_sec += 1;
4997     usec -= 1000000;
4998   }
4999   abstime->tv_nsec = usec * 1000;
5000   return abstime;
5001 }
5002 
5003 
5004 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5005 // Conceptually TryPark() should be equivalent to park(0).
5006 
5007 int os::PlatformEvent::TryPark() {
5008   for (;;) {
5009     const int v = _Event ;
5010     guarantee ((v == 0) || (v == 1), "invariant") ;
5011     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5012   }
5013 }
5014 
5015 void os::PlatformEvent::park() {       // AKA "down()"
5016   // Invariant: Only the thread associated with the Event/PlatformEvent
5017   // may call park().
5018   // TODO: assert that _Assoc != NULL or _Assoc == Self
5019   int v ;
5020   for (;;) {
5021       v = _Event ;
5022       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5023   }
5024   guarantee (v >= 0, "invariant") ;
5025   if (v == 0) {
5026      // Do this the hard way by blocking ...
5027      int status = pthread_mutex_lock(_mutex);
5028      assert_status(status == 0, status, "mutex_lock");
5029      guarantee (_nParked == 0, "invariant") ;
5030      ++ _nParked ;
5031      while (_Event < 0) {
5032         status = pthread_cond_wait(_cond, _mutex);
5033         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5034         // Treat this the same as if the wait was interrupted
5035         if (status == ETIME) { status = EINTR; }
5036         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5037      }
5038      -- _nParked ;
5039 
5040     // In theory we could move the ST of 0 into _Event past the unlock(),
5041     // but then we'd need a MEMBAR after the ST.
5042     _Event = 0 ;
5043      status = pthread_mutex_unlock(_mutex);
5044      assert_status(status == 0, status, "mutex_unlock");
5045   }
5046   guarantee (_Event >= 0, "invariant") ;
5047 }
5048 
5049 int os::PlatformEvent::park(jlong millis) {
5050   guarantee (_nParked == 0, "invariant") ;
5051 
5052   int v ;
5053   for (;;) {
5054       v = _Event ;
5055       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5056   }
5057   guarantee (v >= 0, "invariant") ;
5058   if (v != 0) return OS_OK ;
5059 
5060   // We do this the hard way, by blocking the thread.
5061   // Consider enforcing a minimum timeout value.
5062   struct timespec abst;
5063   compute_abstime(&abst, millis);
5064 
5065   int ret = OS_TIMEOUT;
5066   int status = pthread_mutex_lock(_mutex);
5067   assert_status(status == 0, status, "mutex_lock");
5068   guarantee (_nParked == 0, "invariant") ;
5069   ++_nParked ;
5070 
5071   // Object.wait(timo) will return because of
5072   // (a) notification
5073   // (b) timeout
5074   // (c) thread.interrupt
5075   //
5076   // Thread.interrupt and object.notify{All} both call Event::set.
5077   // That is, we treat thread.interrupt as a special case of notification.
5078   // The underlying Solaris implementation, cond_timedwait, admits
5079   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5080   // JVM from making those visible to Java code.  As such, we must
5081   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5082   //
5083   // TODO: properly differentiate simultaneous notify+interrupt.
5084   // In that case, we should propagate the notify to another waiter.
5085 
5086   while (_Event < 0) {
5087     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5088     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5089       pthread_cond_destroy (_cond);
5090       pthread_cond_init (_cond, NULL) ;
5091     }
5092     assert_status(status == 0 || status == EINTR ||
5093                   status == ETIME || status == ETIMEDOUT,
5094                   status, "cond_timedwait");
5095     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5096     if (status == ETIME || status == ETIMEDOUT) break ;
5097     // We consume and ignore EINTR and spurious wakeups.
5098   }
5099   --_nParked ;
5100   if (_Event >= 0) {
5101      ret = OS_OK;
5102   }
5103   _Event = 0 ;
5104   status = pthread_mutex_unlock(_mutex);
5105   assert_status(status == 0, status, "mutex_unlock");
5106   assert (_nParked == 0, "invariant") ;
5107   return ret;
5108 }
5109 
5110 void os::PlatformEvent::unpark() {
5111   int v, AnyWaiters ;
5112   for (;;) {
5113       v = _Event ;
5114       if (v > 0) {
5115          // The LD of _Event could have reordered or be satisfied
5116          // by a read-aside from this processor's write buffer.
5117          // To avoid problems execute a barrier and then
5118          // ratify the value.
5119          OrderAccess::fence() ;
5120          if (_Event == v) return ;
5121          continue ;
5122       }
5123       if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5124   }
5125   if (v < 0) {
5126      // Wait for the thread associated with the event to vacate
5127      int status = pthread_mutex_lock(_mutex);
5128      assert_status(status == 0, status, "mutex_lock");
5129      AnyWaiters = _nParked ;
5130      assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5131      if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5132         AnyWaiters = 0 ;
5133         pthread_cond_signal (_cond);
5134      }
5135      status = pthread_mutex_unlock(_mutex);
5136      assert_status(status == 0, status, "mutex_unlock");
5137      if (AnyWaiters != 0) {
5138         status = pthread_cond_signal(_cond);
5139         assert_status(status == 0, status, "cond_signal");
5140      }
5141   }
5142 
5143   // Note that we signal() _after dropping the lock for "immortal" Events.
5144   // This is safe and avoids a common class of  futile wakeups.  In rare
5145   // circumstances this can cause a thread to return prematurely from
5146   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5147   // simply re-test the condition and re-park itself.
5148 }
5149 
5150 
5151 // JSR166
5152 // -------------------------------------------------------
5153 
5154 /*
5155  * The solaris and linux implementations of park/unpark are fairly
5156  * conservative for now, but can be improved. They currently use a
5157  * mutex/condvar pair, plus a a count.
5158  * Park decrements count if > 0, else does a condvar wait.  Unpark
5159  * sets count to 1 and signals condvar.  Only one thread ever waits
5160  * on the condvar. Contention seen when trying to park implies that someone
5161  * is unparking you, so don't wait. And spurious returns are fine, so there
5162  * is no need to track notifications.
5163  */
5164 
5165 #define MAX_SECS 100000000
5166 /*
5167  * This code is common to linux and solaris and will be moved to a
5168  * common place in dolphin.
5169  *
5170  * The passed in time value is either a relative time in nanoseconds
5171  * or an absolute time in milliseconds. Either way it has to be unpacked
5172  * into suitable seconds and nanoseconds components and stored in the
5173  * given timespec structure.
5174  * Given time is a 64-bit value and the time_t used in the timespec is only
5175  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5176  * overflow if times way in the future are given. Further on Solaris versions
5177  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5178  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5179  * As it will be 28 years before "now + 100000000" will overflow we can
5180  * ignore overflow and just impose a hard-limit on seconds using the value
5181  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5182  * years from "now".
5183  */
5184 
5185 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5186   assert (time > 0, "convertTime");
5187 
5188   struct timeval now;
5189   int status = gettimeofday(&now, NULL);
5190   assert(status == 0, "gettimeofday");
5191 
5192   time_t max_secs = now.tv_sec + MAX_SECS;
5193 
5194   if (isAbsolute) {
5195     jlong secs = time / 1000;
5196     if (secs > max_secs) {
5197       absTime->tv_sec = max_secs;
5198     }
5199     else {
5200       absTime->tv_sec = secs;
5201     }
5202     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5203   }
5204   else {
5205     jlong secs = time / NANOSECS_PER_SEC;
5206     if (secs >= MAX_SECS) {
5207       absTime->tv_sec = max_secs;
5208       absTime->tv_nsec = 0;
5209     }
5210     else {
5211       absTime->tv_sec = now.tv_sec + secs;
5212       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5213       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5214         absTime->tv_nsec -= NANOSECS_PER_SEC;
5215         ++absTime->tv_sec; // note: this must be <= max_secs
5216       }
5217     }
5218   }
5219   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5220   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5221   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5222   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5223 }
5224 
5225 void Parker::park(bool isAbsolute, jlong time) {
5226   // Optional fast-path check:
5227   // Return immediately if a permit is available.
5228   if (_counter > 0) {
5229       _counter = 0 ;
5230       OrderAccess::fence();
5231       return ;
5232   }
5233 
5234   Thread* thread = Thread::current();
5235   assert(thread->is_Java_thread(), "Must be JavaThread");
5236   JavaThread *jt = (JavaThread *)thread;
5237 
5238   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5239   // Check interrupt before trying to wait
5240   if (Thread::is_interrupted(thread, false)) {
5241     return;
5242   }
5243 
5244   // Next, demultiplex/decode time arguments
5245   timespec absTime;
5246   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5247     return;
5248   }
5249   if (time > 0) {
5250     unpackTime(&absTime, isAbsolute, time);
5251   }
5252 
5253 
5254   // Enter safepoint region
5255   // Beware of deadlocks such as 6317397.
5256   // The per-thread Parker:: mutex is a classic leaf-lock.
5257   // In particular a thread must never block on the Threads_lock while
5258   // holding the Parker:: mutex.  If safepoints are pending both the
5259   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5260   ThreadBlockInVM tbivm(jt);
5261 
5262   // Don't wait if cannot get lock since interference arises from
5263   // unblocking.  Also. check interrupt before trying wait
5264   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5265     return;
5266   }
5267 
5268   int status ;
5269   if (_counter > 0)  { // no wait needed
5270     _counter = 0;
5271     status = pthread_mutex_unlock(_mutex);
5272     assert (status == 0, "invariant") ;
5273     OrderAccess::fence();
5274     return;
5275   }
5276 
5277 #ifdef ASSERT
5278   // Don't catch signals while blocked; let the running threads have the signals.
5279   // (This allows a debugger to break into the running thread.)
5280   sigset_t oldsigs;
5281   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5282   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5283 #endif
5284 
5285   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5286   jt->set_suspend_equivalent();
5287   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5288 
5289   if (time == 0) {
5290     status = pthread_cond_wait (_cond, _mutex) ;
5291   } else {
5292     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5293     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5294       pthread_cond_destroy (_cond) ;
5295       pthread_cond_init    (_cond, NULL);
5296     }
5297   }
5298   assert_status(status == 0 || status == EINTR ||
5299                 status == ETIME || status == ETIMEDOUT,
5300                 status, "cond_timedwait");
5301 
5302 #ifdef ASSERT
5303   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5304 #endif
5305 
5306   _counter = 0 ;
5307   status = pthread_mutex_unlock(_mutex) ;
5308   assert_status(status == 0, status, "invariant") ;
5309   // If externally suspended while waiting, re-suspend
5310   if (jt->handle_special_suspend_equivalent_condition()) {
5311     jt->java_suspend_self();
5312   }
5313 
5314   OrderAccess::fence();
5315 }
5316 
5317 void Parker::unpark() {
5318   int s, status ;
5319   status = pthread_mutex_lock(_mutex);
5320   assert (status == 0, "invariant") ;
5321   s = _counter;
5322   _counter = 1;
5323   if (s < 1) {
5324      if (WorkAroundNPTLTimedWaitHang) {
5325         status = pthread_cond_signal (_cond) ;
5326         assert (status == 0, "invariant") ;
5327         status = pthread_mutex_unlock(_mutex);
5328         assert (status == 0, "invariant") ;
5329      } else {
5330         status = pthread_mutex_unlock(_mutex);
5331         assert (status == 0, "invariant") ;
5332         status = pthread_cond_signal (_cond) ;
5333         assert (status == 0, "invariant") ;
5334      }
5335   } else {
5336     pthread_mutex_unlock(_mutex);
5337     assert (status == 0, "invariant") ;
5338   }
5339 }
5340 
5341 
5342 extern char** environ;
5343 
5344 #ifndef __NR_fork
5345 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5346 #endif
5347 
5348 #ifndef __NR_execve
5349 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5350 #endif
5351 
5352 // Run the specified command in a separate process. Return its exit value,
5353 // or -1 on failure (e.g. can't fork a new process).
5354 // Unlike system(), this function can be called from signal handler. It
5355 // doesn't block SIGINT et al.
5356 int os::fork_and_exec(char* cmd) {
5357   const char * argv[4] = {"sh", "-c", cmd, NULL};
5358 
5359   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5360   // pthread_atfork handlers and reset pthread library. All we need is a
5361   // separate process to execve. Make a direct syscall to fork process.
5362   // On IA64 there's no fork syscall, we have to use fork() and hope for
5363   // the best...
5364   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5365               IA64_ONLY(fork();)
5366 
5367   if (pid < 0) {
5368     // fork failed
5369     return -1;
5370 
5371   } else if (pid == 0) {
5372     // child process
5373 
5374     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5375     // first to kill every thread on the thread list. Because this list is
5376     // not reset by fork() (see notes above), execve() will instead kill
5377     // every thread in the parent process. We know this is the only thread
5378     // in the new process, so make a system call directly.
5379     // IA64 should use normal execve() from glibc to match the glibc fork()
5380     // above.
5381     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5382     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5383 
5384     // execve failed
5385     _exit(-1);
5386 
5387   } else  {
5388     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5389     // care about the actual exit code, for now.
5390 
5391     int status;
5392 
5393     // Wait for the child process to exit.  This returns immediately if
5394     // the child has already exited. */
5395     while (waitpid(pid, &status, 0) < 0) {
5396         switch (errno) {
5397         case ECHILD: return 0;
5398         case EINTR: break;
5399         default: return -1;
5400         }
5401     }
5402 
5403     if (WIFEXITED(status)) {
5404        // The child exited normally; get its exit code.
5405        return WEXITSTATUS(status);
5406     } else if (WIFSIGNALED(status)) {
5407        // The child exited because of a signal
5408        // The best value to return is 0x80 + signal number,
5409        // because that is what all Unix shells do, and because
5410        // it allows callers to distinguish between process exit and
5411        // process death by signal.
5412        return 0x80 + WTERMSIG(status);
5413     } else {
5414        // Unknown exit code; pass it through
5415        return status;
5416     }
5417   }
5418 }
5419 
5420 // is_headless_jre()
5421 //
5422 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5423 // in order to report if we are running in a headless jre
5424 //
5425 // Since JDK8 xawt/libmawt.so was moved into the same directory
5426 // as libawt.so, and renamed libawt_xawt.so
5427 //
5428 bool os::is_headless_jre() {
5429     struct stat statbuf;
5430     char buf[MAXPATHLEN];
5431     char libmawtpath[MAXPATHLEN];
5432     const char *xawtstr  = "/xawt/libmawt.so";
5433     const char *new_xawtstr = "/libawt_xawt.so";
5434     char *p;
5435 
5436     // Get path to libjvm.so
5437     os::jvm_path(buf, sizeof(buf));
5438 
5439     // Get rid of libjvm.so
5440     p = strrchr(buf, '/');
5441     if (p == NULL) return false;
5442     else *p = '\0';
5443 
5444     // Get rid of client or server
5445     p = strrchr(buf, '/');
5446     if (p == NULL) return false;
5447     else *p = '\0';
5448 
5449     // check xawt/libmawt.so
5450     strcpy(libmawtpath, buf);
5451     strcat(libmawtpath, xawtstr);
5452     if (::stat(libmawtpath, &statbuf) == 0) return false;
5453 
5454     // check libawt_xawt.so
5455     strcpy(libmawtpath, buf);
5456     strcat(libmawtpath, new_xawtstr);
5457     if (::stat(libmawtpath, &statbuf) == 0) return false;
5458 
5459     return true;
5460 }
5461 
5462 
5463 #ifdef JAVASE_EMBEDDED
5464 //
5465 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5466 //
5467 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5468 
5469 // ctor
5470 //
5471 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5472   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5473   _fd = fd;
5474 
5475   if (os::create_thread(this, os::os_thread)) {
5476     _memnotify_thread = this;
5477     os::set_priority(this, NearMaxPriority);
5478     os::start_thread(this);
5479   }
5480 }
5481 
5482 // Where all the work gets done
5483 //
5484 void MemNotifyThread::run() {
5485   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5486 
5487   // Set up the select arguments
5488   fd_set rfds;
5489   if (_fd != -1) {
5490     FD_ZERO(&rfds);
5491     FD_SET(_fd, &rfds);
5492   }
5493 
5494   // Now wait for the mem_notify device to wake up
5495   while (1) {
5496     // Wait for the mem_notify device to signal us..
5497     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5498     if (rc == -1) {
5499       perror("select!\n");
5500       break;
5501     } else if (rc) {
5502       //ssize_t free_before = os::available_memory();
5503       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5504 
5505       // The kernel is telling us there is not much memory left...
5506       // try to do something about that
5507 
5508       // If we are not already in a GC, try one.
5509       if (!Universe::heap()->is_gc_active()) {
5510         Universe::heap()->collect(GCCause::_allocation_failure);
5511 
5512         //ssize_t free_after = os::available_memory();
5513         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5514         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5515       }
5516       // We might want to do something like the following if we find the GC's are not helping...
5517       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5518     }
5519   }
5520 }
5521 
5522 //
5523 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5524 //
5525 void MemNotifyThread::start() {
5526   int    fd;
5527   fd = open ("/dev/mem_notify", O_RDONLY, 0);
5528   if (fd < 0) {
5529       return;
5530   }
5531 
5532   if (memnotify_thread() == NULL) {
5533     new MemNotifyThread(fd);
5534   }
5535 }
5536 #endif // JAVASE_EMBEDDED
--- EOF ---