1 /*
   2  * Copyright (c) 1999, 2012, 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 
  86 // put OS-includes here
  87 # include <sys/types.h>
  88 # include <sys/mman.h>
  89 # include <sys/stat.h>
  90 # include <sys/select.h>
  91 # include <pthread.h>
  92 # include <signal.h>
  93 # include <errno.h>
  94 # include <dlfcn.h>
  95 # include <stdio.h>
  96 # include <unistd.h>
  97 # include <sys/resource.h>
  98 # include <pthread.h>
  99 # include <sys/stat.h>
 100 # include <sys/time.h>
 101 # include <sys/times.h>
 102 # include <sys/utsname.h>
 103 # include <sys/socket.h>
 104 # include <sys/wait.h>
 105 # include <pwd.h>
 106 # include <poll.h>
 107 # include <semaphore.h>
 108 # include <fcntl.h>
 109 # include <string.h>
 110 # include <syscall.h>
 111 # include <sys/sysinfo.h>
 112 # include <gnu/libc-version.h>
 113 # include <sys/ipc.h>
 114 # include <sys/shm.h>
 115 # include <link.h>
 116 # include <stdint.h>
 117 # include <inttypes.h>
 118 # include <sys/ioctl.h>
 119 
 120 #define MAX_PATH    (2 * K)
 121 
 122 // for timer info max values which include all bits
 123 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 124 
 125 #define LARGEPAGES_BIT (1 << 6)
 126 ////////////////////////////////////////////////////////////////////////////////
 127 // global variables
 128 julong os::Linux::_physical_memory = 0;
 129 
 130 address   os::Linux::_initial_thread_stack_bottom = NULL;
 131 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 132 
 133 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 134 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 135 Mutex* os::Linux::_createThread_lock = NULL;
 136 pthread_t os::Linux::_main_thread;
 137 int os::Linux::_page_size = -1;
 138 bool os::Linux::_is_floating_stack = false;
 139 bool os::Linux::_is_NPTL = false;
 140 bool os::Linux::_supports_fast_thread_cpu_time = false;
 141 const char * os::Linux::_glibc_version = NULL;
 142 const char * os::Linux::_libpthread_version = NULL;
 143 
 144 static jlong initial_time_count=0;
 145 
 146 static int clock_tics_per_sec = 100;
 147 
 148 // For diagnostics to print a message once. see run_periodic_checks
 149 static sigset_t check_signal_done;
 150 static bool check_signals = true;;
 151 
 152 static pid_t _initial_pid = 0;
 153 
 154 /* Signal number used to suspend/resume a thread */
 155 
 156 /* do not use any signal number less than SIGSEGV, see 4355769 */
 157 static int SR_signum = SIGUSR2;
 158 sigset_t SR_sigset;
 159 
 160 /* Used to protect dlsym() calls */
 161 static pthread_mutex_t dl_mutex;
 162 
 163 #ifdef JAVASE_EMBEDDED
 164 class MemNotifyThread: public Thread {
 165   friend class VMStructs;
 166  public:
 167   virtual void run();
 168 
 169  private:
 170   static MemNotifyThread* _memnotify_thread;
 171   int _fd;
 172 
 173  public:
 174 
 175   // Constructor
 176   MemNotifyThread(int fd);
 177 
 178   // Tester
 179   bool is_memnotify_thread() const { return true; }
 180 
 181   // Printing
 182   char* name() const { return (char*)"Linux MemNotify Thread"; }
 183 
 184   // Returns the single instance of the MemNotifyThread
 185   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 186 
 187   // Create and start the single instance of MemNotifyThread
 188   static void start();
 189 };
 190 #endif // JAVASE_EMBEDDED
 191 
 192 // utility functions
 193 
 194 static int SR_initialize();
 195 static int SR_finalize();
 196 
 197 julong os::available_memory() {
 198   return Linux::available_memory();
 199 }
 200 
 201 julong os::Linux::available_memory() {
 202   // values in struct sysinfo are "unsigned long"
 203   struct sysinfo si;
 204   sysinfo(&si);
 205 
 206   return (julong)si.freeram * si.mem_unit;
 207 }
 208 
 209 julong os::physical_memory() {
 210   return Linux::physical_memory();
 211 }
 212 
 213 julong os::allocatable_physical_memory(julong size) {
 214 #ifdef _LP64
 215   return size;
 216 #else
 217   julong result = MIN2(size, (julong)3800*M);
 218    if (!is_allocatable(result)) {
 219      // See comments under solaris for alignment considerations
 220      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 221      result =  MIN2(size, reasonable_size);
 222    }
 223    return result;
 224 #endif // _LP64
 225 }
 226 
 227 ////////////////////////////////////////////////////////////////////////////////
 228 // environment support
 229 
 230 bool os::getenv(const char* name, char* buf, int len) {
 231   const char* val = ::getenv(name);
 232   if (val != NULL && strlen(val) < (size_t)len) {
 233     strcpy(buf, val);
 234     return true;
 235   }
 236   if (len > 0) buf[0] = 0;  // return a null string
 237   return false;
 238 }
 239 
 240 
 241 // Return true if user is running as root.
 242 
 243 bool os::have_special_privileges() {
 244   static bool init = false;
 245   static bool privileges = false;
 246   if (!init) {
 247     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 248     init = true;
 249   }
 250   return privileges;
 251 }
 252 
 253 
 254 #ifndef SYS_gettid
 255 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 256 #ifdef __ia64__
 257 #define SYS_gettid 1105
 258 #elif __i386__
 259 #define SYS_gettid 224
 260 #elif __amd64__
 261 #define SYS_gettid 186
 262 #elif __sparc__
 263 #define SYS_gettid 143
 264 #else
 265 #error define gettid for the arch
 266 #endif
 267 #endif
 268 
 269 // Cpu architecture string
 270 #if   defined(ZERO)
 271 static char cpu_arch[] = ZERO_LIBARCH;
 272 #elif defined(IA64)
 273 static char cpu_arch[] = "ia64";
 274 #elif defined(IA32)
 275 static char cpu_arch[] = "i386";
 276 #elif defined(AMD64)
 277 static char cpu_arch[] = "amd64";
 278 #elif defined(ARM)
 279 static char cpu_arch[] = "arm";
 280 #elif defined(PPC)
 281 static char cpu_arch[] = "ppc";
 282 #elif defined(SPARC)
 283 #  ifdef _LP64
 284 static char cpu_arch[] = "sparcv9";
 285 #  else
 286 static char cpu_arch[] = "sparc";
 287 #  endif
 288 #else
 289 #error Add appropriate cpu_arch setting
 290 #endif
 291 
 292 
 293 // pid_t gettid()
 294 //
 295 // Returns the kernel thread id of the currently running thread. Kernel
 296 // thread id is used to access /proc.
 297 //
 298 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 299 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 300 //
 301 pid_t os::Linux::gettid() {
 302   int rslt = syscall(SYS_gettid);
 303   if (rslt == -1) {
 304      // old kernel, no NPTL support
 305      return getpid();
 306   } else {
 307      return (pid_t)rslt;
 308   }
 309 }
 310 
 311 // Most versions of linux have a bug where the number of processors are
 312 // determined by looking at the /proc file system.  In a chroot environment,
 313 // the system call returns 1.  This causes the VM to act as if it is
 314 // a single processor and elide locking (see is_MP() call).
 315 static bool unsafe_chroot_detected = false;
 316 static const char *unstable_chroot_error = "/proc file system not found.\n"
 317                      "Java may be unstable running multithreaded in a chroot "
 318                      "environment on Linux when /proc filesystem is not mounted.";
 319 
 320 void os::Linux::initialize_system_info() {
 321   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 322   if (processor_count() == 1) {
 323     pid_t pid = os::Linux::gettid();
 324     char fname[32];
 325     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 326     FILE *fp = fopen(fname, "r");
 327     if (fp == NULL) {
 328       unsafe_chroot_detected = true;
 329     } else {
 330       fclose(fp);
 331     }
 332   }
 333   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 334   assert(processor_count() > 0, "linux error");
 335 }
 336 
 337 void os::init_system_properties_values() {
 338 //  char arch[12];
 339 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 340 
 341   // The next steps are taken in the product version:
 342   //
 343   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 344   // This library should be located at:
 345   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 346   //
 347   // If "/jre/lib/" appears at the right place in the path, then we
 348   // assume libjvm[_g].so is installed in a JDK and we use this path.
 349   //
 350   // Otherwise exit with message: "Could not create the Java virtual machine."
 351   //
 352   // The following extra steps are taken in the debugging version:
 353   //
 354   // If "/jre/lib/" does NOT appear at the right place in the path
 355   // instead of exit check for $JAVA_HOME environment variable.
 356   //
 357   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 358   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 359   // it looks like libjvm[_g].so is installed there
 360   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 361   //
 362   // Otherwise exit.
 363   //
 364   // Important note: if the location of libjvm.so changes this
 365   // code needs to be changed accordingly.
 366 
 367   // The next few definitions allow the code to be verbatim:
 368 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
 369 #define getenv(n) ::getenv(n)
 370 
 371 /*
 372  * See ld(1):
 373  *      The linker uses the following search paths to locate required
 374  *      shared libraries:
 375  *        1: ...
 376  *        ...
 377  *        7: The default directories, normally /lib and /usr/lib.
 378  */
 379 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 380 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 381 #else
 382 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 383 #endif
 384 
 385 #define EXTENSIONS_DIR  "/lib/ext"
 386 #define ENDORSED_DIR    "/lib/endorsed"
 387 #define REG_DIR         "/usr/java/packages"
 388 
 389   {
 390     /* sysclasspath, java_home, dll_dir */
 391     {
 392         char *home_path;
 393         char *dll_path;
 394         char *pslash;
 395         char buf[MAXPATHLEN];
 396         os::jvm_path(buf, sizeof(buf));
 397 
 398         // Found the full path to libjvm.so.
 399         // Now cut the path to <java_home>/jre if we can.
 400         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 401         pslash = strrchr(buf, '/');
 402         if (pslash != NULL)
 403             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 404         dll_path = malloc(strlen(buf) + 1);
 405         if (dll_path == NULL)
 406             return;
 407         strcpy(dll_path, buf);
 408         Arguments::set_dll_dir(dll_path);
 409 
 410         if (pslash != NULL) {
 411             pslash = strrchr(buf, '/');
 412             if (pslash != NULL) {
 413                 *pslash = '\0';       /* get rid of /<arch> */
 414                 pslash = strrchr(buf, '/');
 415                 if (pslash != NULL)
 416                     *pslash = '\0';   /* get rid of /lib */
 417             }
 418         }
 419 
 420         home_path = malloc(strlen(buf) + 1);
 421         if (home_path == NULL)
 422             return;
 423         strcpy(home_path, buf);
 424         Arguments::set_java_home(home_path);
 425 
 426         if (!set_boot_path('/', ':'))
 427             return;
 428     }
 429 
 430     /*
 431      * Where to look for native libraries
 432      *
 433      * Note: Due to a legacy implementation, most of the library path
 434      * is set in the launcher.  This was to accomodate linking restrictions
 435      * on legacy Linux implementations (which are no longer supported).
 436      * Eventually, all the library path setting will be done here.
 437      *
 438      * However, to prevent the proliferation of improperly built native
 439      * libraries, the new path component /usr/java/packages is added here.
 440      * Eventually, all the library path setting will be done here.
 441      */
 442     {
 443         char *ld_library_path;
 444 
 445         /*
 446          * Construct the invariant part of ld_library_path. Note that the
 447          * space for the colon and the trailing null are provided by the
 448          * nulls included by the sizeof operator (so actually we allocate
 449          * a byte more than necessary).
 450          */
 451         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 452             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 453         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 454 
 455         /*
 456          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 457          * should always exist (until the legacy problem cited above is
 458          * addressed).
 459          */
 460         char *v = getenv("LD_LIBRARY_PATH");
 461         if (v != NULL) {
 462             char *t = ld_library_path;
 463             /* That's +1 for the colon and +1 for the trailing '\0' */
 464             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 465             sprintf(ld_library_path, "%s:%s", v, t);
 466         }
 467         Arguments::set_library_path(ld_library_path);
 468     }
 469 
 470     /*
 471      * Extensions directories.
 472      *
 473      * Note that the space for the colon and the trailing null are provided
 474      * by the nulls included by the sizeof operator (so actually one byte more
 475      * than necessary is allocated).
 476      */
 477     {
 478         char *buf = malloc(strlen(Arguments::get_java_home()) +
 479             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 480         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 481             Arguments::get_java_home());
 482         Arguments::set_ext_dirs(buf);
 483     }
 484 
 485     /* Endorsed standards default directory. */
 486     {
 487         char * buf;
 488         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 489         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 490         Arguments::set_endorsed_dirs(buf);
 491     }
 492   }
 493 
 494 #undef malloc
 495 #undef getenv
 496 #undef EXTENSIONS_DIR
 497 #undef ENDORSED_DIR
 498 
 499   // Done
 500   return;
 501 }
 502 
 503 ////////////////////////////////////////////////////////////////////////////////
 504 // breakpoint support
 505 
 506 void os::breakpoint() {
 507   BREAKPOINT;
 508 }
 509 
 510 extern "C" void breakpoint() {
 511   // use debugger to set breakpoint here
 512 }
 513 
 514 ////////////////////////////////////////////////////////////////////////////////
 515 // signal support
 516 
 517 debug_only(static bool signal_sets_initialized = false);
 518 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 519 
 520 bool os::Linux::is_sig_ignored(int sig) {
 521       struct sigaction oact;
 522       sigaction(sig, (struct sigaction*)NULL, &oact);
 523       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 524                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 525       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 526            return true;
 527       else
 528            return false;
 529 }
 530 
 531 void os::Linux::signal_sets_init() {
 532   // Should also have an assertion stating we are still single-threaded.
 533   assert(!signal_sets_initialized, "Already initialized");
 534   // Fill in signals that are necessarily unblocked for all threads in
 535   // the VM. Currently, we unblock the following signals:
 536   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 537   //                         by -Xrs (=ReduceSignalUsage));
 538   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 539   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 540   // the dispositions or masks wrt these signals.
 541   // Programs embedding the VM that want to use the above signals for their
 542   // own purposes must, at this time, use the "-Xrs" option to prevent
 543   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 544   // (See bug 4345157, and other related bugs).
 545   // In reality, though, unblocking these signals is really a nop, since
 546   // these signals are not blocked by default.
 547   sigemptyset(&unblocked_sigs);
 548   sigemptyset(&allowdebug_blocked_sigs);
 549   sigaddset(&unblocked_sigs, SIGILL);
 550   sigaddset(&unblocked_sigs, SIGSEGV);
 551   sigaddset(&unblocked_sigs, SIGBUS);
 552   sigaddset(&unblocked_sigs, SIGFPE);
 553   sigaddset(&unblocked_sigs, SR_signum);
 554 
 555   if (!ReduceSignalUsage) {
 556    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 557       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 558       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 559    }
 560    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 561       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 562       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 563    }
 564    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 565       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 566       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 567    }
 568   }
 569   // Fill in signals that are blocked by all but the VM thread.
 570   sigemptyset(&vm_sigs);
 571   if (!ReduceSignalUsage)
 572     sigaddset(&vm_sigs, BREAK_SIGNAL);
 573   debug_only(signal_sets_initialized = true);
 574 
 575 }
 576 
 577 // These are signals that are unblocked while a thread is running Java.
 578 // (For some reason, they get blocked by default.)
 579 sigset_t* os::Linux::unblocked_signals() {
 580   assert(signal_sets_initialized, "Not initialized");
 581   return &unblocked_sigs;
 582 }
 583 
 584 // These are the signals that are blocked while a (non-VM) thread is
 585 // running Java. Only the VM thread handles these signals.
 586 sigset_t* os::Linux::vm_signals() {
 587   assert(signal_sets_initialized, "Not initialized");
 588   return &vm_sigs;
 589 }
 590 
 591 // These are signals that are blocked during cond_wait to allow debugger in
 592 sigset_t* os::Linux::allowdebug_blocked_signals() {
 593   assert(signal_sets_initialized, "Not initialized");
 594   return &allowdebug_blocked_sigs;
 595 }
 596 
 597 void os::Linux::hotspot_sigmask(Thread* thread) {
 598 
 599   //Save caller's signal mask before setting VM signal mask
 600   sigset_t caller_sigmask;
 601   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 602 
 603   OSThread* osthread = thread->osthread();
 604   osthread->set_caller_sigmask(caller_sigmask);
 605 
 606   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 607 
 608   if (!ReduceSignalUsage) {
 609     if (thread->is_VM_thread()) {
 610       // Only the VM thread handles BREAK_SIGNAL ...
 611       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 612     } else {
 613       // ... all other threads block BREAK_SIGNAL
 614       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 615     }
 616   }
 617 }
 618 
 619 //////////////////////////////////////////////////////////////////////////////
 620 // detecting pthread library
 621 
 622 void os::Linux::libpthread_init() {
 623   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 624   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 625   // generic name for earlier versions.
 626   // Define macros here so we can build HotSpot on old systems.
 627 # ifndef _CS_GNU_LIBC_VERSION
 628 # define _CS_GNU_LIBC_VERSION 2
 629 # endif
 630 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 631 # define _CS_GNU_LIBPTHREAD_VERSION 3
 632 # endif
 633 
 634   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 635   if (n > 0) {
 636      char *str = (char *)malloc(n, mtInternal);
 637      confstr(_CS_GNU_LIBC_VERSION, str, n);
 638      os::Linux::set_glibc_version(str);
 639   } else {
 640      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 641      static char _gnu_libc_version[32];
 642      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 643               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 644      os::Linux::set_glibc_version(_gnu_libc_version);
 645   }
 646 
 647   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 648   if (n > 0) {
 649      char *str = (char *)malloc(n, mtInternal);
 650      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 651      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 652      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 653      // is the case. LinuxThreads has a hard limit on max number of threads.
 654      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 655      // On the other hand, NPTL does not have such a limit, sysconf()
 656      // will return -1 and errno is not changed. Check if it is really NPTL.
 657      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 658          strstr(str, "NPTL") &&
 659          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 660        free(str);
 661        os::Linux::set_libpthread_version("linuxthreads");
 662      } else {
 663        os::Linux::set_libpthread_version(str);
 664      }
 665   } else {
 666     // glibc before 2.3.2 only has LinuxThreads.
 667     os::Linux::set_libpthread_version("linuxthreads");
 668   }
 669 
 670   if (strstr(libpthread_version(), "NPTL")) {
 671      os::Linux::set_is_NPTL();
 672   } else {
 673      os::Linux::set_is_LinuxThreads();
 674   }
 675 
 676   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 677   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 678   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 679      os::Linux::set_is_floating_stack();
 680   }
 681 }
 682 
 683 /////////////////////////////////////////////////////////////////////////////
 684 // thread stack
 685 
 686 // Force Linux kernel to expand current thread stack. If "bottom" is close
 687 // to the stack guard, caller should block all signals.
 688 //
 689 // MAP_GROWSDOWN:
 690 //   A special mmap() flag that is used to implement thread stacks. It tells
 691 //   kernel that the memory region should extend downwards when needed. This
 692 //   allows early versions of LinuxThreads to only mmap the first few pages
 693 //   when creating a new thread. Linux kernel will automatically expand thread
 694 //   stack as needed (on page faults).
 695 //
 696 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 697 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 698 //   region, it's hard to tell if the fault is due to a legitimate stack
 699 //   access or because of reading/writing non-exist memory (e.g. buffer
 700 //   overrun). As a rule, if the fault happens below current stack pointer,
 701 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 702 //   application (see Linux kernel fault.c).
 703 //
 704 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 705 //   stack overflow detection.
 706 //
 707 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 708 //   not use this flag. However, the stack of initial thread is not created
 709 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 710 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 711 //   and then attach the thread to JVM.
 712 //
 713 // To get around the problem and allow stack banging on Linux, we need to
 714 // manually expand thread stack after receiving the SIGSEGV.
 715 //
 716 // There are two ways to expand thread stack to address "bottom", we used
 717 // both of them in JVM before 1.5:
 718 //   1. adjust stack pointer first so that it is below "bottom", and then
 719 //      touch "bottom"
 720 //   2. mmap() the page in question
 721 //
 722 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 723 // if current sp is already near the lower end of page 101, and we need to
 724 // call mmap() to map page 100, it is possible that part of the mmap() frame
 725 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 726 // That will destroy the mmap() frame and cause VM to crash.
 727 //
 728 // The following code works by adjusting sp first, then accessing the "bottom"
 729 // page to force a page fault. Linux kernel will then automatically expand the
 730 // stack mapping.
 731 //
 732 // _expand_stack_to() assumes its frame size is less than page size, which
 733 // should always be true if the function is not inlined.
 734 
 735 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 736 #define NOINLINE
 737 #else
 738 #define NOINLINE __attribute__ ((noinline))
 739 #endif
 740 
 741 static void _expand_stack_to(address bottom) NOINLINE;
 742 
 743 static void _expand_stack_to(address bottom) {
 744   address sp;
 745   size_t size;
 746   volatile char *p;
 747 
 748   // Adjust bottom to point to the largest address within the same page, it
 749   // gives us a one-page buffer if alloca() allocates slightly more memory.
 750   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 751   bottom += os::Linux::page_size() - 1;
 752 
 753   // sp might be slightly above current stack pointer; if that's the case, we
 754   // will alloca() a little more space than necessary, which is OK. Don't use
 755   // os::current_stack_pointer(), as its result can be slightly below current
 756   // stack pointer, causing us to not alloca enough to reach "bottom".
 757   sp = (address)&sp;
 758 
 759   if (sp > bottom) {
 760     size = sp - bottom;
 761     p = (volatile char *)alloca(size);
 762     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 763     p[0] = '\0';
 764   }
 765 }
 766 
 767 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 768   assert(t!=NULL, "just checking");
 769   assert(t->osthread()->expanding_stack(), "expand should be set");
 770   assert(t->stack_base() != NULL, "stack_base was not initialized");
 771 
 772   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 773     sigset_t mask_all, old_sigset;
 774     sigfillset(&mask_all);
 775     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 776     _expand_stack_to(addr);
 777     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 778     return true;
 779   }
 780   return false;
 781 }
 782 
 783 //////////////////////////////////////////////////////////////////////////////
 784 // create new thread
 785 
 786 static address highest_vm_reserved_address();
 787 
 788 // check if it's safe to start a new thread
 789 static bool _thread_safety_check(Thread* thread) {
 790   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 791     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 792     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 793     //   allocated (MAP_FIXED) from high address space. Every thread stack
 794     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 795     //   it to other values if they rebuild LinuxThreads).
 796     //
 797     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 798     // the memory region has already been mmap'ed. That means if we have too
 799     // many threads and/or very large heap, eventually thread stack will
 800     // collide with heap.
 801     //
 802     // Here we try to prevent heap/stack collision by comparing current
 803     // stack bottom with the highest address that has been mmap'ed by JVM
 804     // plus a safety margin for memory maps created by native code.
 805     //
 806     // This feature can be disabled by setting ThreadSafetyMargin to 0
 807     //
 808     if (ThreadSafetyMargin > 0) {
 809       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 810 
 811       // not safe if our stack extends below the safety margin
 812       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 813     } else {
 814       return true;
 815     }
 816   } else {
 817     // Floating stack LinuxThreads or NPTL:
 818     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 819     //   there's not enough space left, pthread_create() will fail. If we come
 820     //   here, that means enough space has been reserved for stack.
 821     return true;
 822   }
 823 }
 824 
 825 // Thread start routine for all newly created threads
 826 static void *java_start(Thread *thread) {
 827   // Try to randomize the cache line index of hot stack frames.
 828   // This helps when threads of the same stack traces evict each other's
 829   // cache lines. The threads can be either from the same JVM instance, or
 830   // from different JVM instances. The benefit is especially true for
 831   // processors with hyperthreading technology.
 832   static int counter = 0;
 833   int pid = os::current_process_id();
 834   alloca(((pid ^ counter++) & 7) * 128);
 835 
 836   ThreadLocalStorage::set_thread(thread);
 837 
 838   OSThread* osthread = thread->osthread();
 839   Monitor* sync = osthread->startThread_lock();
 840 
 841   // non floating stack LinuxThreads needs extra check, see above
 842   if (!_thread_safety_check(thread)) {
 843     // notify parent thread
 844     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 845     osthread->set_state(ZOMBIE);
 846     sync->notify_all();
 847     return NULL;
 848   }
 849 
 850   // thread_id is kernel thread id (similar to Solaris LWP id)
 851   osthread->set_thread_id(os::Linux::gettid());
 852 
 853   if (UseNUMA) {
 854     int lgrp_id = os::numa_get_group_id();
 855     if (lgrp_id != -1) {
 856       thread->set_lgrp_id(lgrp_id);
 857     }
 858   }
 859   // initialize signal mask for this thread
 860   os::Linux::hotspot_sigmask(thread);
 861 
 862   // initialize floating point control register
 863   os::Linux::init_thread_fpu_state();
 864 
 865   // handshaking with parent thread
 866   {
 867     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 868 
 869     // notify parent thread
 870     osthread->set_state(INITIALIZED);
 871     sync->notify_all();
 872 
 873     // wait until os::start_thread()
 874     while (osthread->get_state() == INITIALIZED) {
 875       sync->wait(Mutex::_no_safepoint_check_flag);
 876     }
 877   }
 878 
 879   // call one more level start routine
 880   thread->run();
 881 
 882   return 0;
 883 }
 884 
 885 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 886   assert(thread->osthread() == NULL, "caller responsible");
 887 
 888   // Allocate the OSThread object
 889   OSThread* osthread = new OSThread(NULL, NULL);
 890   if (osthread == NULL) {
 891     return false;
 892   }
 893 
 894   // set the correct thread state
 895   osthread->set_thread_type(thr_type);
 896 
 897   // Initial state is ALLOCATED but not INITIALIZED
 898   osthread->set_state(ALLOCATED);
 899 
 900   thread->set_osthread(osthread);
 901 
 902   // init thread attributes
 903   pthread_attr_t attr;
 904   pthread_attr_init(&attr);
 905   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 906 
 907   // stack size
 908   if (os::Linux::supports_variable_stack_size()) {
 909     // calculate stack size if it's not specified by caller
 910     if (stack_size == 0) {
 911       stack_size = os::Linux::default_stack_size(thr_type);
 912 
 913       switch (thr_type) {
 914       case os::java_thread:
 915         // Java threads use ThreadStackSize which default value can be
 916         // changed with the flag -Xss
 917         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 918         stack_size = JavaThread::stack_size_at_create();
 919         break;
 920       case os::compiler_thread:
 921         if (CompilerThreadStackSize > 0) {
 922           stack_size = (size_t)(CompilerThreadStackSize * K);
 923           break;
 924         } // else fall through:
 925           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 926       case os::vm_thread:
 927       case os::pgc_thread:
 928       case os::cgc_thread:
 929       case os::watcher_thread:
 930         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 931         break;
 932       }
 933     }
 934 
 935     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 936     pthread_attr_setstacksize(&attr, stack_size);
 937   } else {
 938     // let pthread_create() pick the default value.
 939   }
 940 
 941   // glibc guard page
 942   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 943 
 944   ThreadState state;
 945 
 946   {
 947     // Serialize thread creation if we are running with fixed stack LinuxThreads
 948     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 949     if (lock) {
 950       os::Linux::createThread_lock()->lock_without_safepoint_check();
 951     }
 952 
 953     pthread_t tid;
 954     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 955 
 956     pthread_attr_destroy(&attr);
 957 
 958     if (ret != 0) {
 959       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 960         perror("pthread_create()");
 961       }
 962       // Need to clean up stuff we've allocated so far
 963       thread->set_osthread(NULL);
 964       delete osthread;
 965       if (lock) os::Linux::createThread_lock()->unlock();
 966       return false;
 967     }
 968 
 969     // Store pthread info into the OSThread
 970     osthread->set_pthread_id(tid);
 971 
 972     // Wait until child thread is either initialized or aborted
 973     {
 974       Monitor* sync_with_child = osthread->startThread_lock();
 975       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 976       while ((state = osthread->get_state()) == ALLOCATED) {
 977         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 978       }
 979     }
 980 
 981     if (lock) {
 982       os::Linux::createThread_lock()->unlock();
 983     }
 984   }
 985 
 986   // Aborted due to thread limit being reached
 987   if (state == ZOMBIE) {
 988       thread->set_osthread(NULL);
 989       delete osthread;
 990       return false;
 991   }
 992 
 993   // The thread is returned suspended (in state INITIALIZED),
 994   // and is started higher up in the call chain
 995   assert(state == INITIALIZED, "race condition");
 996   return true;
 997 }
 998 
 999 /////////////////////////////////////////////////////////////////////////////
1000 // attach existing thread
1001 
1002 // bootstrap the main thread
1003 bool os::create_main_thread(JavaThread* thread) {
1004   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
1005   return create_attached_thread(thread);
1006 }
1007 
1008 bool os::create_attached_thread(JavaThread* thread) {
1009 #ifdef ASSERT
1010     thread->verify_not_published();
1011 #endif
1012 
1013   // Allocate the OSThread object
1014   OSThread* osthread = new OSThread(NULL, NULL);
1015 
1016   if (osthread == NULL) {
1017     return false;
1018   }
1019 
1020   // Store pthread info into the OSThread
1021   osthread->set_thread_id(os::Linux::gettid());
1022   osthread->set_pthread_id(::pthread_self());
1023 
1024   // initialize floating point control register
1025   os::Linux::init_thread_fpu_state();
1026 
1027   // Initial thread state is RUNNABLE
1028   osthread->set_state(RUNNABLE);
1029 
1030   thread->set_osthread(osthread);
1031 
1032   if (UseNUMA) {
1033     int lgrp_id = os::numa_get_group_id();
1034     if (lgrp_id != -1) {
1035       thread->set_lgrp_id(lgrp_id);
1036     }
1037   }
1038 
1039   if (os::Linux::is_initial_thread()) {
1040     // If current thread is initial thread, its stack is mapped on demand,
1041     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1042     // the entire stack region to avoid SEGV in stack banging.
1043     // It is also useful to get around the heap-stack-gap problem on SuSE
1044     // kernel (see 4821821 for details). We first expand stack to the top
1045     // of yellow zone, then enable stack yellow zone (order is significant,
1046     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1047     // is no gap between the last two virtual memory regions.
1048 
1049     JavaThread *jt = (JavaThread *)thread;
1050     address addr = jt->stack_yellow_zone_base();
1051     assert(addr != NULL, "initialization problem?");
1052     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1053 
1054     osthread->set_expanding_stack();
1055     os::Linux::manually_expand_stack(jt, addr);
1056     osthread->clear_expanding_stack();
1057   }
1058 
1059   // initialize signal mask for this thread
1060   // and save the caller's signal mask
1061   os::Linux::hotspot_sigmask(thread);
1062 
1063   return true;
1064 }
1065 
1066 void os::pd_start_thread(Thread* thread) {
1067   OSThread * osthread = thread->osthread();
1068   assert(osthread->get_state() != INITIALIZED, "just checking");
1069   Monitor* sync_with_child = osthread->startThread_lock();
1070   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1071   sync_with_child->notify();
1072 }
1073 
1074 // Free Linux resources related to the OSThread
1075 void os::free_thread(OSThread* osthread) {
1076   assert(osthread != NULL, "osthread not set");
1077 
1078   if (Thread::current()->osthread() == osthread) {
1079     // Restore caller's signal mask
1080     sigset_t sigmask = osthread->caller_sigmask();
1081     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1082    }
1083 
1084   delete osthread;
1085 }
1086 
1087 //////////////////////////////////////////////////////////////////////////////
1088 // thread local storage
1089 
1090 int os::allocate_thread_local_storage() {
1091   pthread_key_t key;
1092   int rslt = pthread_key_create(&key, NULL);
1093   assert(rslt == 0, "cannot allocate thread local storage");
1094   return (int)key;
1095 }
1096 
1097 // Note: This is currently not used by VM, as we don't destroy TLS key
1098 // on VM exit.
1099 void os::free_thread_local_storage(int index) {
1100   int rslt = pthread_key_delete((pthread_key_t)index);
1101   assert(rslt == 0, "invalid index");
1102 }
1103 
1104 void os::thread_local_storage_at_put(int index, void* value) {
1105   int rslt = pthread_setspecific((pthread_key_t)index, value);
1106   assert(rslt == 0, "pthread_setspecific failed");
1107 }
1108 
1109 extern "C" Thread* get_thread() {
1110   return ThreadLocalStorage::thread();
1111 }
1112 
1113 //////////////////////////////////////////////////////////////////////////////
1114 // initial thread
1115 
1116 // Check if current thread is the initial thread, similar to Solaris thr_main.
1117 bool os::Linux::is_initial_thread(void) {
1118   char dummy;
1119   // If called before init complete, thread stack bottom will be null.
1120   // Can be called if fatal error occurs before initialization.
1121   if (initial_thread_stack_bottom() == NULL) return false;
1122   assert(initial_thread_stack_bottom() != NULL &&
1123          initial_thread_stack_size()   != 0,
1124          "os::init did not locate initial thread's stack region");
1125   if ((address)&dummy >= initial_thread_stack_bottom() &&
1126       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1127        return true;
1128   else return false;
1129 }
1130 
1131 // Find the virtual memory area that contains addr
1132 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1133   FILE *fp = fopen("/proc/self/maps", "r");
1134   if (fp) {
1135     address low, high;
1136     while (!feof(fp)) {
1137       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1138         if (low <= addr && addr < high) {
1139            if (vma_low)  *vma_low  = low;
1140            if (vma_high) *vma_high = high;
1141            fclose (fp);
1142            return true;
1143         }
1144       }
1145       for (;;) {
1146         int ch = fgetc(fp);
1147         if (ch == EOF || ch == (int)'\n') break;
1148       }
1149     }
1150     fclose(fp);
1151   }
1152   return false;
1153 }
1154 
1155 // Locate initial thread stack. This special handling of initial thread stack
1156 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1157 // bogus value for initial thread.
1158 void os::Linux::capture_initial_stack(size_t max_size) {
1159   // stack size is the easy part, get it from RLIMIT_STACK
1160   size_t stack_size;
1161   struct rlimit rlim;
1162   getrlimit(RLIMIT_STACK, &rlim);
1163   stack_size = rlim.rlim_cur;
1164 
1165   // 6308388: a bug in ld.so will relocate its own .data section to the
1166   //   lower end of primordial stack; reduce ulimit -s value a little bit
1167   //   so we won't install guard page on ld.so's data section.
1168   stack_size -= 2 * page_size();
1169 
1170   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1171   //   7.1, in both cases we will get 2G in return value.
1172   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1173   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1174   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1175   //   in case other parts in glibc still assumes 2M max stack size.
1176   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1177 #ifndef IA64
1178   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1179 #else
1180   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1181   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1182 #endif
1183 
1184   // Try to figure out where the stack base (top) is. This is harder.
1185   //
1186   // When an application is started, glibc saves the initial stack pointer in
1187   // a global variable "__libc_stack_end", which is then used by system
1188   // libraries. __libc_stack_end should be pretty close to stack top. The
1189   // variable is available since the very early days. However, because it is
1190   // a private interface, it could disappear in the future.
1191   //
1192   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1193   // to __libc_stack_end, it is very close to stack top, but isn't the real
1194   // stack top. Note that /proc may not exist if VM is running as a chroot
1195   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1196   // /proc/<pid>/stat could change in the future (though unlikely).
1197   //
1198   // We try __libc_stack_end first. If that doesn't work, look for
1199   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1200   // as a hint, which should work well in most cases.
1201 
1202   uintptr_t stack_start;
1203 
1204   // try __libc_stack_end first
1205   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1206   if (p && *p) {
1207     stack_start = *p;
1208   } else {
1209     // see if we can get the start_stack field from /proc/self/stat
1210     FILE *fp;
1211     int pid;
1212     char state;
1213     int ppid;
1214     int pgrp;
1215     int session;
1216     int nr;
1217     int tpgrp;
1218     unsigned long flags;
1219     unsigned long minflt;
1220     unsigned long cminflt;
1221     unsigned long majflt;
1222     unsigned long cmajflt;
1223     unsigned long utime;
1224     unsigned long stime;
1225     long cutime;
1226     long cstime;
1227     long prio;
1228     long nice;
1229     long junk;
1230     long it_real;
1231     uintptr_t start;
1232     uintptr_t vsize;
1233     intptr_t rss;
1234     uintptr_t rsslim;
1235     uintptr_t scodes;
1236     uintptr_t ecode;
1237     int i;
1238 
1239     // Figure what the primordial thread stack base is. Code is inspired
1240     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1241     // followed by command name surrounded by parentheses, state, etc.
1242     char stat[2048];
1243     int statlen;
1244 
1245     fp = fopen("/proc/self/stat", "r");
1246     if (fp) {
1247       statlen = fread(stat, 1, 2047, fp);
1248       stat[statlen] = '\0';
1249       fclose(fp);
1250 
1251       // Skip pid and the command string. Note that we could be dealing with
1252       // weird command names, e.g. user could decide to rename java launcher
1253       // to "java 1.4.2 :)", then the stat file would look like
1254       //                1234 (java 1.4.2 :)) R ... ...
1255       // We don't really need to know the command string, just find the last
1256       // occurrence of ")" and then start parsing from there. See bug 4726580.
1257       char * s = strrchr(stat, ')');
1258 
1259       i = 0;
1260       if (s) {
1261         // Skip blank chars
1262         do s++; while (isspace(*s));
1263 
1264 #define _UFM UINTX_FORMAT
1265 #define _DFM INTX_FORMAT
1266 
1267         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1268         /*              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 */
1269         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,
1270              &state,          /* 3  %c  */
1271              &ppid,           /* 4  %d  */
1272              &pgrp,           /* 5  %d  */
1273              &session,        /* 6  %d  */
1274              &nr,             /* 7  %d  */
1275              &tpgrp,          /* 8  %d  */
1276              &flags,          /* 9  %lu  */
1277              &minflt,         /* 10 %lu  */
1278              &cminflt,        /* 11 %lu  */
1279              &majflt,         /* 12 %lu  */
1280              &cmajflt,        /* 13 %lu  */
1281              &utime,          /* 14 %lu  */
1282              &stime,          /* 15 %lu  */
1283              &cutime,         /* 16 %ld  */
1284              &cstime,         /* 17 %ld  */
1285              &prio,           /* 18 %ld  */
1286              &nice,           /* 19 %ld  */
1287              &junk,           /* 20 %ld  */
1288              &it_real,        /* 21 %ld  */
1289              &start,          /* 22 UINTX_FORMAT */
1290              &vsize,          /* 23 UINTX_FORMAT */
1291              &rss,            /* 24 INTX_FORMAT  */
1292              &rsslim,         /* 25 UINTX_FORMAT */
1293              &scodes,         /* 26 UINTX_FORMAT */
1294              &ecode,          /* 27 UINTX_FORMAT */
1295              &stack_start);   /* 28 UINTX_FORMAT */
1296       }
1297 
1298 #undef _UFM
1299 #undef _DFM
1300 
1301       if (i != 28 - 2) {
1302          assert(false, "Bad conversion from /proc/self/stat");
1303          // product mode - assume we are the initial thread, good luck in the
1304          // embedded case.
1305          warning("Can't detect initial thread stack location - bad conversion");
1306          stack_start = (uintptr_t) &rlim;
1307       }
1308     } else {
1309       // For some reason we can't open /proc/self/stat (for example, running on
1310       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1311       // most cases, so don't abort:
1312       warning("Can't detect initial thread stack location - no /proc/self/stat");
1313       stack_start = (uintptr_t) &rlim;
1314     }
1315   }
1316 
1317   // Now we have a pointer (stack_start) very close to the stack top, the
1318   // next thing to do is to figure out the exact location of stack top. We
1319   // can find out the virtual memory area that contains stack_start by
1320   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1321   // and its upper limit is the real stack top. (again, this would fail if
1322   // running inside chroot, because /proc may not exist.)
1323 
1324   uintptr_t stack_top;
1325   address low, high;
1326   if (find_vma((address)stack_start, &low, &high)) {
1327     // success, "high" is the true stack top. (ignore "low", because initial
1328     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1329     stack_top = (uintptr_t)high;
1330   } else {
1331     // failed, likely because /proc/self/maps does not exist
1332     warning("Can't detect initial thread stack location - find_vma failed");
1333     // best effort: stack_start is normally within a few pages below the real
1334     // stack top, use it as stack top, and reduce stack size so we won't put
1335     // guard page outside stack.
1336     stack_top = stack_start;
1337     stack_size -= 16 * page_size();
1338   }
1339 
1340   // stack_top could be partially down the page so align it
1341   stack_top = align_size_up(stack_top, page_size());
1342 
1343   if (max_size && stack_size > max_size) {
1344      _initial_thread_stack_size = max_size;
1345   } else {
1346      _initial_thread_stack_size = stack_size;
1347   }
1348 
1349   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1350   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1351 }
1352 
1353 ////////////////////////////////////////////////////////////////////////////////
1354 // time support
1355 
1356 // Time since start-up in seconds to a fine granularity.
1357 // Used by VMSelfDestructTimer and the MemProfiler.
1358 double os::elapsedTime() {
1359 
1360   return (double)(os::elapsed_counter()) * 0.000001;
1361 }
1362 
1363 jlong os::elapsed_counter() {
1364   timeval time;
1365   int status = gettimeofday(&time, NULL);
1366   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1367 }
1368 
1369 jlong os::elapsed_frequency() {
1370   return (1000 * 1000);
1371 }
1372 
1373 // For now, we say that linux does not support vtime.  I have no idea
1374 // whether it can actually be made to (DLD, 9/13/05).
1375 
1376 bool os::supports_vtime() { return false; }
1377 bool os::enable_vtime()   { return false; }
1378 bool os::vtime_enabled()  { return false; }
1379 double os::elapsedVTime() {
1380   // better than nothing, but not much
1381   return elapsedTime();
1382 }
1383 
1384 jlong os::javaTimeMillis() {
1385   timeval time;
1386   int status = gettimeofday(&time, NULL);
1387   assert(status != -1, "linux error");
1388   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1389 }
1390 
1391 #ifndef CLOCK_MONOTONIC
1392 #define CLOCK_MONOTONIC (1)
1393 #endif
1394 
1395 void os::Linux::clock_init() {
1396   // we do dlopen's in this particular order due to bug in linux
1397   // dynamical loader (see 6348968) leading to crash on exit
1398   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1399   if (handle == NULL) {
1400     handle = dlopen("librt.so", RTLD_LAZY);
1401   }
1402 
1403   if (handle) {
1404     int (*clock_getres_func)(clockid_t, struct timespec*) =
1405            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1406     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1407            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1408     if (clock_getres_func && clock_gettime_func) {
1409       // See if monotonic clock is supported by the kernel. Note that some
1410       // early implementations simply return kernel jiffies (updated every
1411       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1412       // for nano time (though the monotonic property is still nice to have).
1413       // It's fixed in newer kernels, however clock_getres() still returns
1414       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1415       // resolution for now. Hopefully as people move to new kernels, this
1416       // won't be a problem.
1417       struct timespec res;
1418       struct timespec tp;
1419       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1420           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1421         // yes, monotonic clock is supported
1422         _clock_gettime = clock_gettime_func;
1423       } else {
1424         // close librt if there is no monotonic clock
1425         dlclose(handle);
1426       }
1427     }
1428   }
1429 }
1430 
1431 #ifndef SYS_clock_getres
1432 
1433 #if defined(IA32) || defined(AMD64)
1434 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1435 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1436 #else
1437 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1438 #define sys_clock_getres(x,y)  -1
1439 #endif
1440 
1441 #else
1442 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1443 #endif
1444 
1445 void os::Linux::fast_thread_clock_init() {
1446   if (!UseLinuxPosixThreadCPUClocks) {
1447     return;
1448   }
1449   clockid_t clockid;
1450   struct timespec tp;
1451   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1452       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1453 
1454   // Switch to using fast clocks for thread cpu time if
1455   // the sys_clock_getres() returns 0 error code.
1456   // Note, that some kernels may support the current thread
1457   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1458   // returned by the pthread_getcpuclockid().
1459   // If the fast Posix clocks are supported then the sys_clock_getres()
1460   // must return at least tp.tv_sec == 0 which means a resolution
1461   // better than 1 sec. This is extra check for reliability.
1462 
1463   if(pthread_getcpuclockid_func &&
1464      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1465      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1466 
1467     _supports_fast_thread_cpu_time = true;
1468     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1469   }
1470 }
1471 
1472 jlong os::javaTimeNanos() {
1473   if (Linux::supports_monotonic_clock()) {
1474     struct timespec tp;
1475     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1476     assert(status == 0, "gettime error");
1477     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1478     return result;
1479   } else {
1480     timeval time;
1481     int status = gettimeofday(&time, NULL);
1482     assert(status != -1, "linux error");
1483     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1484     return 1000 * usecs;
1485   }
1486 }
1487 
1488 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1489   if (Linux::supports_monotonic_clock()) {
1490     info_ptr->max_value = ALL_64_BITS;
1491 
1492     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1493     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1494     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1495   } else {
1496     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1497     info_ptr->max_value = ALL_64_BITS;
1498 
1499     // gettimeofday is a real time clock so it skips
1500     info_ptr->may_skip_backward = true;
1501     info_ptr->may_skip_forward = true;
1502   }
1503 
1504   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1505 }
1506 
1507 // Return the real, user, and system times in seconds from an
1508 // arbitrary fixed point in the past.
1509 bool os::getTimesSecs(double* process_real_time,
1510                       double* process_user_time,
1511                       double* process_system_time) {
1512   struct tms ticks;
1513   clock_t real_ticks = times(&ticks);
1514 
1515   if (real_ticks == (clock_t) (-1)) {
1516     return false;
1517   } else {
1518     double ticks_per_second = (double) clock_tics_per_sec;
1519     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1520     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1521     *process_real_time = ((double) real_ticks) / ticks_per_second;
1522 
1523     return true;
1524   }
1525 }
1526 
1527 
1528 char * os::local_time_string(char *buf, size_t buflen) {
1529   struct tm t;
1530   time_t long_time;
1531   time(&long_time);
1532   localtime_r(&long_time, &t);
1533   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1534                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1535                t.tm_hour, t.tm_min, t.tm_sec);
1536   return buf;
1537 }
1538 
1539 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1540   return localtime_r(clock, res);
1541 }
1542 
1543 ////////////////////////////////////////////////////////////////////////////////
1544 // runtime exit support
1545 
1546 // Note: os::shutdown() might be called very early during initialization, or
1547 // called from signal handler. Before adding something to os::shutdown(), make
1548 // sure it is async-safe and can handle partially initialized VM.
1549 void os::shutdown() {
1550 
1551   // allow PerfMemory to attempt cleanup of any persistent resources
1552   perfMemory_exit();
1553 
1554   // needs to remove object in file system
1555   AttachListener::abort();
1556 
1557   // flush buffered output, finish log files
1558   ostream_abort();
1559 
1560   // Check for abort hook
1561   abort_hook_t abort_hook = Arguments::abort_hook();
1562   if (abort_hook != NULL) {
1563     abort_hook();
1564   }
1565 
1566 }
1567 
1568 // Note: os::abort() might be called very early during initialization, or
1569 // called from signal handler. Before adding something to os::abort(), make
1570 // sure it is async-safe and can handle partially initialized VM.
1571 void os::abort(bool dump_core) {
1572   os::shutdown();
1573   if (dump_core) {
1574 #ifndef PRODUCT
1575     fdStream out(defaultStream::output_fd());
1576     out.print_raw("Current thread is ");
1577     char buf[16];
1578     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1579     out.print_raw_cr(buf);
1580     out.print_raw_cr("Dumping core ...");
1581 #endif
1582     ::abort(); // dump core
1583   }
1584 
1585   ::exit(1);
1586 }
1587 
1588 // Die immediately, no exit hook, no abort hook, no cleanup.
1589 void os::die() {
1590   // _exit() on LinuxThreads only kills current thread
1591   ::abort();
1592 }
1593 
1594 // unused on linux for now.
1595 void os::set_error_file(const char *logfile) {}
1596 
1597 
1598 // This method is a copy of JDK's sysGetLastErrorString
1599 // from src/solaris/hpi/src/system_md.c
1600 
1601 size_t os::lasterror(char *buf, size_t len) {
1602 
1603   if (errno == 0)  return 0;
1604 
1605   const char *s = ::strerror(errno);
1606   size_t n = ::strlen(s);
1607   if (n >= len) {
1608     n = len - 1;
1609   }
1610   ::strncpy(buf, s, n);
1611   buf[n] = '\0';
1612   return n;
1613 }
1614 
1615 intx os::current_thread_id() { return (intx)pthread_self(); }
1616 int os::current_process_id() {
1617 
1618   // Under the old linux thread library, linux gives each thread
1619   // its own process id. Because of this each thread will return
1620   // a different pid if this method were to return the result
1621   // of getpid(2). Linux provides no api that returns the pid
1622   // of the launcher thread for the vm. This implementation
1623   // returns a unique pid, the pid of the launcher thread
1624   // that starts the vm 'process'.
1625 
1626   // Under the NPTL, getpid() returns the same pid as the
1627   // launcher thread rather than a unique pid per thread.
1628   // Use gettid() if you want the old pre NPTL behaviour.
1629 
1630   // if you are looking for the result of a call to getpid() that
1631   // returns a unique pid for the calling thread, then look at the
1632   // OSThread::thread_id() method in osThread_linux.hpp file
1633 
1634   return (int)(_initial_pid ? _initial_pid : getpid());
1635 }
1636 
1637 // DLL functions
1638 
1639 const char* os::dll_file_extension() { return ".so"; }
1640 
1641 // This must be hard coded because it's the system's temporary
1642 // directory not the java application's temp directory, ala java.io.tmpdir.
1643 const char* os::get_temp_directory() { return "/tmp"; }
1644 
1645 static bool file_exists(const char* filename) {
1646   struct stat statbuf;
1647   if (filename == NULL || strlen(filename) == 0) {
1648     return false;
1649   }
1650   return os::stat(filename, &statbuf) == 0;
1651 }
1652 
1653 void os::dll_build_name(char* buffer, size_t buflen,
1654                         const char* pname, const char* fname) {
1655   // Copied from libhpi
1656   const size_t pnamelen = pname ? strlen(pname) : 0;
1657 
1658   // Quietly truncate on buffer overflow.  Should be an error.
1659   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1660       *buffer = '\0';
1661       return;
1662   }
1663 
1664   if (pnamelen == 0) {
1665     snprintf(buffer, buflen, "lib%s.so", fname);
1666   } else if (strchr(pname, *os::path_separator()) != NULL) {
1667     int n;
1668     char** pelements = split_path(pname, &n);
1669     for (int i = 0 ; i < n ; i++) {
1670       // Really shouldn't be NULL, but check can't hurt
1671       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1672         continue; // skip the empty path values
1673       }
1674       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1675       if (file_exists(buffer)) {
1676         break;
1677       }
1678     }
1679     // release the storage
1680     for (int i = 0 ; i < n ; i++) {
1681       if (pelements[i] != NULL) {
1682         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1683       }
1684     }
1685     if (pelements != NULL) {
1686       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1687     }
1688   } else {
1689     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1690   }
1691 }
1692 
1693 const char* os::get_current_directory(char *buf, int buflen) {
1694   return getcwd(buf, buflen);
1695 }
1696 
1697 // check if addr is inside libjvm[_g].so
1698 bool os::address_is_in_vm(address addr) {
1699   static address libjvm_base_addr;
1700   Dl_info dlinfo;
1701 
1702   if (libjvm_base_addr == NULL) {
1703     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1704     libjvm_base_addr = (address)dlinfo.dli_fbase;
1705     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1706   }
1707 
1708   if (dladdr((void *)addr, &dlinfo)) {
1709     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1710   }
1711 
1712   return false;
1713 }
1714 
1715 bool os::dll_address_to_function_name(address addr, char *buf,
1716                                       int buflen, int *offset) {
1717   Dl_info dlinfo;
1718 
1719   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1720     if (buf != NULL) {
1721       if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1722         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1723       }
1724     }
1725     if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1726     return true;
1727   } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1728     if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1729         buf, buflen, offset, dlinfo.dli_fname)) {
1730        return true;
1731     }
1732   }
1733 
1734   if (buf != NULL) buf[0] = '\0';
1735   if (offset != NULL) *offset = -1;
1736   return false;
1737 }
1738 
1739 struct _address_to_library_name {
1740   address addr;          // input : memory address
1741   size_t  buflen;        //         size of fname
1742   char*   fname;         // output: library name
1743   address base;          //         library base addr
1744 };
1745 
1746 static int address_to_library_name_callback(struct dl_phdr_info *info,
1747                                             size_t size, void *data) {
1748   int i;
1749   bool found = false;
1750   address libbase = NULL;
1751   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1752 
1753   // iterate through all loadable segments
1754   for (i = 0; i < info->dlpi_phnum; i++) {
1755     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1756     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1757       // base address of a library is the lowest address of its loaded
1758       // segments.
1759       if (libbase == NULL || libbase > segbase) {
1760         libbase = segbase;
1761       }
1762       // see if 'addr' is within current segment
1763       if (segbase <= d->addr &&
1764           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1765         found = true;
1766       }
1767     }
1768   }
1769 
1770   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1771   // so dll_address_to_library_name() can fall through to use dladdr() which
1772   // can figure out executable name from argv[0].
1773   if (found && info->dlpi_name && info->dlpi_name[0]) {
1774     d->base = libbase;
1775     if (d->fname) {
1776       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1777     }
1778     return 1;
1779   }
1780   return 0;
1781 }
1782 
1783 bool os::dll_address_to_library_name(address addr, char* buf,
1784                                      int buflen, int* offset) {
1785   Dl_info dlinfo;
1786   struct _address_to_library_name data;
1787 
1788   // There is a bug in old glibc dladdr() implementation that it could resolve
1789   // to wrong library name if the .so file has a base address != NULL. Here
1790   // we iterate through the program headers of all loaded libraries to find
1791   // out which library 'addr' really belongs to. This workaround can be
1792   // removed once the minimum requirement for glibc is moved to 2.3.x.
1793   data.addr = addr;
1794   data.fname = buf;
1795   data.buflen = buflen;
1796   data.base = NULL;
1797   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1798 
1799   if (rslt) {
1800      // buf already contains library name
1801      if (offset) *offset = addr - data.base;
1802      return true;
1803   } else if (dladdr((void*)addr, &dlinfo)){
1804      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1805      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1806      return true;
1807   } else {
1808      if (buf) buf[0] = '\0';
1809      if (offset) *offset = -1;
1810      return false;
1811   }
1812 }
1813 
1814   // Loads .dll/.so and
1815   // in case of error it checks if .dll/.so was built for the
1816   // same architecture as Hotspot is running on
1817 
1818 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1819 {
1820   void * result= ::dlopen(filename, RTLD_LAZY);
1821   if (result != NULL) {
1822     // Successful loading
1823     return result;
1824   }
1825 
1826   Elf32_Ehdr elf_head;
1827 
1828   // Read system error message into ebuf
1829   // It may or may not be overwritten below
1830   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1831   ebuf[ebuflen-1]='\0';
1832   int diag_msg_max_length=ebuflen-strlen(ebuf);
1833   char* diag_msg_buf=ebuf+strlen(ebuf);
1834 
1835   if (diag_msg_max_length==0) {
1836     // No more space in ebuf for additional diagnostics message
1837     return NULL;
1838   }
1839 
1840 
1841   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1842 
1843   if (file_descriptor < 0) {
1844     // Can't open library, report dlerror() message
1845     return NULL;
1846   }
1847 
1848   bool failed_to_read_elf_head=
1849     (sizeof(elf_head)!=
1850         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1851 
1852   ::close(file_descriptor);
1853   if (failed_to_read_elf_head) {
1854     // file i/o error - report dlerror() msg
1855     return NULL;
1856   }
1857 
1858   typedef struct {
1859     Elf32_Half  code;         // Actual value as defined in elf.h
1860     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1861     char        elf_class;    // 32 or 64 bit
1862     char        endianess;    // MSB or LSB
1863     char*       name;         // String representation
1864   } arch_t;
1865 
1866   #ifndef EM_486
1867   #define EM_486          6               /* Intel 80486 */
1868   #endif
1869 
1870   static const arch_t arch_array[]={
1871     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1872     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1873     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1874     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1875     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1876     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1877     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1878     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1879     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1880     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1881     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1882     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1883     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1884     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1885     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1886     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1887   };
1888 
1889   #if  (defined IA32)
1890     static  Elf32_Half running_arch_code=EM_386;
1891   #elif   (defined AMD64)
1892     static  Elf32_Half running_arch_code=EM_X86_64;
1893   #elif  (defined IA64)
1894     static  Elf32_Half running_arch_code=EM_IA_64;
1895   #elif  (defined __sparc) && (defined _LP64)
1896     static  Elf32_Half running_arch_code=EM_SPARCV9;
1897   #elif  (defined __sparc) && (!defined _LP64)
1898     static  Elf32_Half running_arch_code=EM_SPARC;
1899   #elif  (defined __powerpc64__)
1900     static  Elf32_Half running_arch_code=EM_PPC64;
1901   #elif  (defined __powerpc__)
1902     static  Elf32_Half running_arch_code=EM_PPC;
1903   #elif  (defined ARM)
1904     static  Elf32_Half running_arch_code=EM_ARM;
1905   #elif  (defined S390)
1906     static  Elf32_Half running_arch_code=EM_S390;
1907   #elif  (defined ALPHA)
1908     static  Elf32_Half running_arch_code=EM_ALPHA;
1909   #elif  (defined MIPSEL)
1910     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1911   #elif  (defined PARISC)
1912     static  Elf32_Half running_arch_code=EM_PARISC;
1913   #elif  (defined MIPS)
1914     static  Elf32_Half running_arch_code=EM_MIPS;
1915   #elif  (defined M68K)
1916     static  Elf32_Half running_arch_code=EM_68K;
1917   #else
1918     #error Method os::dll_load requires that one of following is defined:\
1919          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1920   #endif
1921 
1922   // Identify compatability class for VM's architecture and library's architecture
1923   // Obtain string descriptions for architectures
1924 
1925   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1926   int running_arch_index=-1;
1927 
1928   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1929     if (running_arch_code == arch_array[i].code) {
1930       running_arch_index    = i;
1931     }
1932     if (lib_arch.code == arch_array[i].code) {
1933       lib_arch.compat_class = arch_array[i].compat_class;
1934       lib_arch.name         = arch_array[i].name;
1935     }
1936   }
1937 
1938   assert(running_arch_index != -1,
1939     "Didn't find running architecture code (running_arch_code) in arch_array");
1940   if (running_arch_index == -1) {
1941     // Even though running architecture detection failed
1942     // we may still continue with reporting dlerror() message
1943     return NULL;
1944   }
1945 
1946   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1947     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1948     return NULL;
1949   }
1950 
1951 #ifndef S390
1952   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1953     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1954     return NULL;
1955   }
1956 #endif // !S390
1957 
1958   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1959     if ( lib_arch.name!=NULL ) {
1960       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1961         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1962         lib_arch.name, arch_array[running_arch_index].name);
1963     } else {
1964       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1965       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1966         lib_arch.code,
1967         arch_array[running_arch_index].name);
1968     }
1969   }
1970 
1971   return NULL;
1972 }
1973 
1974 /*
1975  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
1976  * chances are you might want to run the generated bits against glibc-2.0
1977  * libdl.so, so always use locking for any version of glibc.
1978  */
1979 void* os::dll_lookup(void* handle, const char* name) {
1980   pthread_mutex_lock(&dl_mutex);
1981   void* res = dlsym(handle, name);
1982   pthread_mutex_unlock(&dl_mutex);
1983   return res;
1984 }
1985 
1986 
1987 static bool _print_ascii_file(const char* filename, outputStream* st) {
1988   int fd = ::open(filename, O_RDONLY);
1989   if (fd == -1) {
1990      return false;
1991   }
1992 
1993   char buf[32];
1994   int bytes;
1995   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1996     st->print_raw(buf, bytes);
1997   }
1998 
1999   ::close(fd);
2000 
2001   return true;
2002 }
2003 
2004 void os::print_dll_info(outputStream *st) {
2005    st->print_cr("Dynamic libraries:");
2006 
2007    char fname[32];
2008    pid_t pid = os::Linux::gettid();
2009 
2010    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2011 
2012    if (!_print_ascii_file(fname, st)) {
2013      st->print("Can not get library information for pid = %d\n", pid);
2014    }
2015 }
2016 
2017 void os::print_os_info_brief(outputStream* st) {
2018   os::Linux::print_distro_info(st);
2019 
2020   os::Posix::print_uname_info(st);
2021 
2022   os::Linux::print_libversion_info(st);
2023 
2024 }
2025 
2026 void os::print_os_info(outputStream* st) {
2027   st->print("OS:");
2028 
2029   os::Linux::print_distro_info(st);
2030 
2031   os::Posix::print_uname_info(st);
2032 
2033   // Print warning if unsafe chroot environment detected
2034   if (unsafe_chroot_detected) {
2035     st->print("WARNING!! ");
2036     st->print_cr(unstable_chroot_error);
2037   }
2038 
2039   os::Linux::print_libversion_info(st);
2040 
2041   os::Posix::print_rlimit_info(st);
2042 
2043   os::Posix::print_load_average(st);
2044 
2045   os::Linux::print_full_memory_info(st);
2046 }
2047 
2048 // Try to identify popular distros.
2049 // Most Linux distributions have /etc/XXX-release file, which contains
2050 // the OS version string. Some have more than one /etc/XXX-release file
2051 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2052 // so the order is important.
2053 void os::Linux::print_distro_info(outputStream* st) {
2054   if (!_print_ascii_file("/etc/mandrake-release", st) &&
2055       !_print_ascii_file("/etc/sun-release", st) &&
2056       !_print_ascii_file("/etc/redhat-release", st) &&
2057       !_print_ascii_file("/etc/SuSE-release", st) &&
2058       !_print_ascii_file("/etc/turbolinux-release", st) &&
2059       !_print_ascii_file("/etc/gentoo-release", st) &&
2060       !_print_ascii_file("/etc/debian_version", st) &&
2061       !_print_ascii_file("/etc/ltib-release", st) &&
2062       !_print_ascii_file("/etc/angstrom-version", st)) {
2063       st->print("Linux");
2064   }
2065   st->cr();
2066 }
2067 
2068 void os::Linux::print_libversion_info(outputStream* st) {
2069   // libc, pthread
2070   st->print("libc:");
2071   st->print(os::Linux::glibc_version()); st->print(" ");
2072   st->print(os::Linux::libpthread_version()); st->print(" ");
2073   if (os::Linux::is_LinuxThreads()) {
2074      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2075   }
2076   st->cr();
2077 }
2078 
2079 void os::Linux::print_full_memory_info(outputStream* st) {
2080    st->print("\n/proc/meminfo:\n");
2081    _print_ascii_file("/proc/meminfo", st);
2082    st->cr();
2083 }
2084 
2085 void os::print_memory_info(outputStream* st) {
2086 
2087   st->print("Memory:");
2088   st->print(" %dk page", os::vm_page_size()>>10);
2089 
2090   // values in struct sysinfo are "unsigned long"
2091   struct sysinfo si;
2092   sysinfo(&si);
2093 
2094   st->print(", physical " UINT64_FORMAT "k",
2095             os::physical_memory() >> 10);
2096   st->print("(" UINT64_FORMAT "k free)",
2097             os::available_memory() >> 10);
2098   st->print(", swap " UINT64_FORMAT "k",
2099             ((jlong)si.totalswap * si.mem_unit) >> 10);
2100   st->print("(" UINT64_FORMAT "k free)",
2101             ((jlong)si.freeswap * si.mem_unit) >> 10);
2102   st->cr();
2103 }
2104 
2105 void os::pd_print_cpu_info(outputStream* st) {
2106   st->print("\n/proc/cpuinfo:\n");
2107   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2108     st->print("  <Not Available>");
2109   }
2110   st->cr();
2111 }
2112 
2113 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2114 // but they're the same for all the linux arch that we support
2115 // and they're the same for solaris but there's no common place to put this.
2116 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2117                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2118                           "ILL_COPROC", "ILL_BADSTK" };
2119 
2120 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2121                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2122                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2123 
2124 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2125 
2126 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2127 
2128 void os::print_siginfo(outputStream* st, void* siginfo) {
2129   st->print("siginfo:");
2130 
2131   const int buflen = 100;
2132   char buf[buflen];
2133   siginfo_t *si = (siginfo_t*)siginfo;
2134   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2135   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2136     st->print("si_errno=%s", buf);
2137   } else {
2138     st->print("si_errno=%d", si->si_errno);
2139   }
2140   const int c = si->si_code;
2141   assert(c > 0, "unexpected si_code");
2142   switch (si->si_signo) {
2143   case SIGILL:
2144     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2145     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2146     break;
2147   case SIGFPE:
2148     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2149     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2150     break;
2151   case SIGSEGV:
2152     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2153     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2154     break;
2155   case SIGBUS:
2156     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2157     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2158     break;
2159   default:
2160     st->print(", si_code=%d", si->si_code);
2161     // no si_addr
2162   }
2163 
2164   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2165       UseSharedSpaces) {
2166     FileMapInfo* mapinfo = FileMapInfo::current_info();
2167     if (mapinfo->is_in_shared_space(si->si_addr)) {
2168       st->print("\n\nError accessing class data sharing archive."   \
2169                 " Mapped file inaccessible during execution, "      \
2170                 " possible disk/network problem.");
2171     }
2172   }
2173   st->cr();
2174 }
2175 
2176 
2177 static void print_signal_handler(outputStream* st, int sig,
2178                                  char* buf, size_t buflen);
2179 
2180 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2181   st->print_cr("Signal Handlers:");
2182   print_signal_handler(st, SIGSEGV, buf, buflen);
2183   print_signal_handler(st, SIGBUS , buf, buflen);
2184   print_signal_handler(st, SIGFPE , buf, buflen);
2185   print_signal_handler(st, SIGPIPE, buf, buflen);
2186   print_signal_handler(st, SIGXFSZ, buf, buflen);
2187   print_signal_handler(st, SIGILL , buf, buflen);
2188   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2189   print_signal_handler(st, SR_signum, buf, buflen);
2190   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2191   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2192   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2193   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2194 }
2195 
2196 static char saved_jvm_path[MAXPATHLEN] = {0};
2197 
2198 // Find the full path to the current module, libjvm.so or libjvm_g.so
2199 void os::jvm_path(char *buf, jint buflen) {
2200   // Error checking.
2201   if (buflen < MAXPATHLEN) {
2202     assert(false, "must use a large-enough buffer");
2203     buf[0] = '\0';
2204     return;
2205   }
2206   // Lazy resolve the path to current module.
2207   if (saved_jvm_path[0] != 0) {
2208     strcpy(buf, saved_jvm_path);
2209     return;
2210   }
2211 
2212   char dli_fname[MAXPATHLEN];
2213   bool ret = dll_address_to_library_name(
2214                 CAST_FROM_FN_PTR(address, os::jvm_path),
2215                 dli_fname, sizeof(dli_fname), NULL);
2216   assert(ret != 0, "cannot locate libjvm");
2217   char *rp = realpath(dli_fname, buf);
2218   if (rp == NULL)
2219     return;
2220 
2221   if (Arguments::created_by_gamma_launcher()) {
2222     // Support for the gamma launcher.  Typical value for buf is
2223     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2224     // the right place in the string, then assume we are installed in a JDK and
2225     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2226     // up the path so it looks like libjvm.so is installed there (append a
2227     // fake suffix hotspot/libjvm.so).
2228     const char *p = buf + strlen(buf) - 1;
2229     for (int count = 0; p > buf && count < 5; ++count) {
2230       for (--p; p > buf && *p != '/'; --p)
2231         /* empty */ ;
2232     }
2233 
2234     if (strncmp(p, "/jre/lib/", 9) != 0) {
2235       // Look for JAVA_HOME in the environment.
2236       char* java_home_var = ::getenv("JAVA_HOME");
2237       if (java_home_var != NULL && java_home_var[0] != 0) {
2238         char* jrelib_p;
2239         int len;
2240 
2241         // Check the current module name "libjvm.so" or "libjvm_g.so".
2242         p = strrchr(buf, '/');
2243         assert(strstr(p, "/libjvm") == p, "invalid library name");
2244         p = strstr(p, "_g") ? "_g" : "";
2245 
2246         rp = realpath(java_home_var, buf);
2247         if (rp == NULL)
2248           return;
2249 
2250         // determine if this is a legacy image or modules image
2251         // modules image doesn't have "jre" subdirectory
2252         len = strlen(buf);
2253         jrelib_p = buf + len;
2254         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2255         if (0 != access(buf, F_OK)) {
2256           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2257         }
2258 
2259         if (0 == access(buf, F_OK)) {
2260           // Use current module name "libjvm[_g].so" instead of
2261           // "libjvm"debug_only("_g")".so" since for fastdebug version
2262           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2263           len = strlen(buf);
2264           snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2265         } else {
2266           // Go back to path of .so
2267           rp = realpath(dli_fname, buf);
2268           if (rp == NULL)
2269             return;
2270         }
2271       }
2272     }
2273   }
2274 
2275   strcpy(saved_jvm_path, buf);
2276 }
2277 
2278 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2279   // no prefix required, not even "_"
2280 }
2281 
2282 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2283   // no suffix required
2284 }
2285 
2286 ////////////////////////////////////////////////////////////////////////////////
2287 // sun.misc.Signal support
2288 
2289 static volatile jint sigint_count = 0;
2290 
2291 static void
2292 UserHandler(int sig, void *siginfo, void *context) {
2293   // 4511530 - sem_post is serialized and handled by the manager thread. When
2294   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2295   // don't want to flood the manager thread with sem_post requests.
2296   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2297       return;
2298 
2299   // Ctrl-C is pressed during error reporting, likely because the error
2300   // handler fails to abort. Let VM die immediately.
2301   if (sig == SIGINT && is_error_reported()) {
2302      os::die();
2303   }
2304 
2305   os::signal_notify(sig);
2306 }
2307 
2308 void* os::user_handler() {
2309   return CAST_FROM_FN_PTR(void*, UserHandler);
2310 }
2311 
2312 extern "C" {
2313   typedef void (*sa_handler_t)(int);
2314   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2315 }
2316 
2317 void* os::signal(int signal_number, void* handler) {
2318   struct sigaction sigAct, oldSigAct;
2319 
2320   sigfillset(&(sigAct.sa_mask));
2321   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2322   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2323 
2324   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2325     // -1 means registration failed
2326     return (void *)-1;
2327   }
2328 
2329   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2330 }
2331 
2332 void os::signal_raise(int signal_number) {
2333   ::raise(signal_number);
2334 }
2335 
2336 /*
2337  * The following code is moved from os.cpp for making this
2338  * code platform specific, which it is by its very nature.
2339  */
2340 
2341 // Will be modified when max signal is changed to be dynamic
2342 int os::sigexitnum_pd() {
2343   return NSIG;
2344 }
2345 
2346 // a counter for each possible signal value
2347 static volatile jint pending_signals[NSIG+1] = { 0 };
2348 
2349 // Linux(POSIX) specific hand shaking semaphore.
2350 static sem_t sig_sem;
2351 
2352 void os::signal_init_pd() {
2353   // Initialize signal structures
2354   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2355 
2356   // Initialize signal semaphore
2357   ::sem_init(&sig_sem, 0, 0);
2358 }
2359 
2360 void os::signal_notify(int sig) {
2361   Atomic::inc(&pending_signals[sig]);
2362   ::sem_post(&sig_sem);
2363 }
2364 
2365 static int check_pending_signals(bool wait) {
2366   Atomic::store(0, &sigint_count);
2367   for (;;) {
2368     for (int i = 0; i < NSIG + 1; i++) {
2369       jint n = pending_signals[i];
2370       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2371         return i;
2372       }
2373     }
2374     if (!wait) {
2375       return -1;
2376     }
2377     JavaThread *thread = JavaThread::current();
2378     ThreadBlockInVM tbivm(thread);
2379 
2380     bool threadIsSuspended;
2381     do {
2382       thread->set_suspend_equivalent();
2383       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2384       ::sem_wait(&sig_sem);
2385 
2386       // were we externally suspended while we were waiting?
2387       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2388       if (threadIsSuspended) {
2389         //
2390         // The semaphore has been incremented, but while we were waiting
2391         // another thread suspended us. We don't want to continue running
2392         // while suspended because that would surprise the thread that
2393         // suspended us.
2394         //
2395         ::sem_post(&sig_sem);
2396 
2397         thread->java_suspend_self();
2398       }
2399     } while (threadIsSuspended);
2400   }
2401 }
2402 
2403 int os::signal_lookup() {
2404   return check_pending_signals(false);
2405 }
2406 
2407 int os::signal_wait() {
2408   return check_pending_signals(true);
2409 }
2410 
2411 ////////////////////////////////////////////////////////////////////////////////
2412 // Virtual Memory
2413 
2414 int os::vm_page_size() {
2415   // Seems redundant as all get out
2416   assert(os::Linux::page_size() != -1, "must call os::init");
2417   return os::Linux::page_size();
2418 }
2419 
2420 // Solaris allocates memory by pages.
2421 int os::vm_allocation_granularity() {
2422   assert(os::Linux::page_size() != -1, "must call os::init");
2423   return os::Linux::page_size();
2424 }
2425 
2426 // Rationale behind this function:
2427 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2428 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2429 //  samples for JITted code. Here we create private executable mapping over the code cache
2430 //  and then we can use standard (well, almost, as mapping can change) way to provide
2431 //  info for the reporting script by storing timestamp and location of symbol
2432 void linux_wrap_code(char* base, size_t size) {
2433   static volatile jint cnt = 0;
2434 
2435   if (!UseOprofile) {
2436     return;
2437   }
2438 
2439   char buf[PATH_MAX+1];
2440   int num = Atomic::add(1, &cnt);
2441 
2442   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2443            os::get_temp_directory(), os::current_process_id(), num);
2444   unlink(buf);
2445 
2446   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2447 
2448   if (fd != -1) {
2449     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2450     if (rv != (off_t)-1) {
2451       if (::write(fd, "", 1) == 1) {
2452         mmap(base, size,
2453              PROT_READ|PROT_WRITE|PROT_EXEC,
2454              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2455       }
2456     }
2457     ::close(fd);
2458     unlink(buf);
2459   }
2460 }
2461 
2462 // NOTE: Linux kernel does not really reserve the pages for us.
2463 //       All it does is to check if there are enough free pages
2464 //       left at the time of mmap(). This could be a potential
2465 //       problem.
2466 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2467   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2468   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2469                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2470   if (res != (uintptr_t) MAP_FAILED) {
2471     if (UseNUMAInterleaving) {
2472       numa_make_global(addr, size);
2473     }
2474     return true;
2475   }
2476   return false;
2477 }
2478 
2479 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2480 #ifndef MAP_HUGETLB
2481 #define MAP_HUGETLB 0x40000
2482 #endif
2483 
2484 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2485 #ifndef MADV_HUGEPAGE
2486 #define MADV_HUGEPAGE 14
2487 #endif
2488 
2489 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2490                        bool exec) {
2491   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2492     int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2493     uintptr_t res =
2494       (uintptr_t) ::mmap(addr, size, prot,
2495                          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2496                          -1, 0);
2497     if (res != (uintptr_t) MAP_FAILED) {
2498       if (UseNUMAInterleaving) {
2499         numa_make_global(addr, size);
2500       }
2501       return true;
2502     }
2503     // Fall through and try to use small pages
2504   }
2505 
2506   if (commit_memory(addr, size, exec)) {
2507     realign_memory(addr, size, alignment_hint);
2508     return true;
2509   }
2510   return false;
2511 }
2512 
2513 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2514   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2515     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2516     // be supported or the memory may already be backed by huge pages.
2517     ::madvise(addr, bytes, MADV_HUGEPAGE);
2518   }
2519 }
2520 
2521 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2522   // This method works by doing an mmap over an existing mmaping and effectively discarding
2523   // the existing pages. However it won't work for SHM-based large pages that cannot be
2524   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2525   // small pages on top of the SHM segment. This method always works for small pages, so we
2526   // allow that in any case.
2527   if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2528     commit_memory(addr, bytes, alignment_hint, false);
2529   }
2530 }
2531 
2532 void os::numa_make_global(char *addr, size_t bytes) {
2533   Linux::numa_interleave_memory(addr, bytes);
2534 }
2535 
2536 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2537   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2538 }
2539 
2540 bool os::numa_topology_changed()   { return false; }
2541 
2542 size_t os::numa_get_groups_num() {
2543   int max_node = Linux::numa_max_node();
2544   return max_node > 0 ? max_node + 1 : 1;
2545 }
2546 
2547 int os::numa_get_group_id() {
2548   int cpu_id = Linux::sched_getcpu();
2549   if (cpu_id != -1) {
2550     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2551     if (lgrp_id != -1) {
2552       return lgrp_id;
2553     }
2554   }
2555   return 0;
2556 }
2557 
2558 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2559   for (size_t i = 0; i < size; i++) {
2560     ids[i] = i;
2561   }
2562   return size;
2563 }
2564 
2565 bool os::get_page_info(char *start, page_info* info) {
2566   return false;
2567 }
2568 
2569 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2570   return end;
2571 }
2572 
2573 
2574 int os::Linux::sched_getcpu_syscall(void) {
2575   unsigned int cpu;
2576   int retval = -1;
2577 
2578 #if defined(IA32)
2579 # ifndef SYS_getcpu
2580 # define SYS_getcpu 318
2581 # endif
2582   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2583 #elif defined(AMD64)
2584 // Unfortunately we have to bring all these macros here from vsyscall.h
2585 // to be able to compile on old linuxes.
2586 # define __NR_vgetcpu 2
2587 # define VSYSCALL_START (-10UL << 20)
2588 # define VSYSCALL_SIZE 1024
2589 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2590   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2591   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2592   retval = vgetcpu(&cpu, NULL, NULL);
2593 #endif
2594 
2595   return (retval == -1) ? retval : cpu;
2596 }
2597 
2598 // Something to do with the numa-aware allocator needs these symbols
2599 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2600 extern "C" JNIEXPORT void numa_error(char *where) { }
2601 extern "C" JNIEXPORT int fork1() { return fork(); }
2602 
2603 
2604 // If we are running with libnuma version > 2, then we should
2605 // be trying to use symbols with versions 1.1
2606 // If we are running with earlier version, which did not have symbol versions,
2607 // we should use the base version.
2608 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2609   void *f = dlvsym(handle, name, "libnuma_1.1");
2610   if (f == NULL) {
2611     f = dlsym(handle, name);
2612   }
2613   return f;
2614 }
2615 
2616 bool os::Linux::libnuma_init() {
2617   // sched_getcpu() should be in libc.
2618   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2619                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2620 
2621   // If it's not, try a direct syscall.
2622   if (sched_getcpu() == -1)
2623     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2624 
2625   if (sched_getcpu() != -1) { // Does it work?
2626     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2627     if (handle != NULL) {
2628       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2629                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2630       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2631                                        libnuma_dlsym(handle, "numa_max_node")));
2632       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2633                                         libnuma_dlsym(handle, "numa_available")));
2634       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2635                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2636       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2637                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2638 
2639 
2640       if (numa_available() != -1) {
2641         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2642         // Create a cpu -> node mapping
2643         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2644         rebuild_cpu_to_node_map();
2645         return true;
2646       }
2647     }
2648   }
2649   return false;
2650 }
2651 
2652 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2653 // The table is later used in get_node_by_cpu().
2654 void os::Linux::rebuild_cpu_to_node_map() {
2655   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2656                               // in libnuma (possible values are starting from 16,
2657                               // and continuing up with every other power of 2, but less
2658                               // than the maximum number of CPUs supported by kernel), and
2659                               // is a subject to change (in libnuma version 2 the requirements
2660                               // are more reasonable) we'll just hardcode the number they use
2661                               // in the library.
2662   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2663 
2664   size_t cpu_num = os::active_processor_count();
2665   size_t cpu_map_size = NCPUS / BitsPerCLong;
2666   size_t cpu_map_valid_size =
2667     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2668 
2669   cpu_to_node()->clear();
2670   cpu_to_node()->at_grow(cpu_num - 1);
2671   size_t node_num = numa_get_groups_num();
2672 
2673   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2674   for (size_t i = 0; i < node_num; i++) {
2675     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2676       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2677         if (cpu_map[j] != 0) {
2678           for (size_t k = 0; k < BitsPerCLong; k++) {
2679             if (cpu_map[j] & (1UL << k)) {
2680               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2681             }
2682           }
2683         }
2684       }
2685     }
2686   }
2687   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2688 }
2689 
2690 int os::Linux::get_node_by_cpu(int cpu_id) {
2691   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2692     return cpu_to_node()->at(cpu_id);
2693   }
2694   return -1;
2695 }
2696 
2697 GrowableArray<int>* os::Linux::_cpu_to_node;
2698 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2699 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2700 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2701 os::Linux::numa_available_func_t os::Linux::_numa_available;
2702 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2703 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2704 unsigned long* os::Linux::_numa_all_nodes;
2705 
2706 bool os::pd_uncommit_memory(char* addr, size_t size) {
2707   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2708                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2709   return res  != (uintptr_t) MAP_FAILED;
2710 }
2711 
2712 // Linux uses a growable mapping for the stack, and if the mapping for
2713 // the stack guard pages is not removed when we detach a thread the
2714 // stack cannot grow beyond the pages where the stack guard was
2715 // mapped.  If at some point later in the process the stack expands to
2716 // that point, the Linux kernel cannot expand the stack any further
2717 // because the guard pages are in the way, and a segfault occurs.
2718 //
2719 // However, it's essential not to split the stack region by unmapping
2720 // a region (leaving a hole) that's already part of the stack mapping,
2721 // so if the stack mapping has already grown beyond the guard pages at
2722 // the time we create them, we have to truncate the stack mapping.
2723 // So, we need to know the extent of the stack mapping when
2724 // create_stack_guard_pages() is called.
2725 
2726 // Find the bounds of the stack mapping.  Return true for success.
2727 //
2728 // We only need this for stacks that are growable: at the time of
2729 // writing thread stacks don't use growable mappings (i.e. those
2730 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2731 // only applies to the main thread.
2732 
2733 static
2734 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2735 
2736   char buf[128];
2737   int fd, sz;
2738 
2739   if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2740     return false;
2741   }
2742 
2743   const char kw[] = "[stack]";
2744   const int kwlen = sizeof(kw)-1;
2745 
2746   // Address part of /proc/self/maps couldn't be more than 128 bytes
2747   while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2748      if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2749         // Extract addresses
2750         if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2751            uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2752            if (sp >= *bottom && sp <= *top) {
2753               ::close(fd);
2754               return true;
2755            }
2756         }
2757      }
2758   }
2759 
2760  ::close(fd);
2761   return false;
2762 }
2763 
2764 
2765 // If the (growable) stack mapping already extends beyond the point
2766 // where we're going to put our guard pages, truncate the mapping at
2767 // that point by munmap()ping it.  This ensures that when we later
2768 // munmap() the guard pages we don't leave a hole in the stack
2769 // mapping. This only affects the main/initial thread, but guard
2770 // against future OS changes
2771 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2772   uintptr_t stack_extent, stack_base;
2773   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2774   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2775       assert(os::Linux::is_initial_thread(),
2776            "growable stack in non-initial thread");
2777     if (stack_extent < (uintptr_t)addr)
2778       ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2779   }
2780 
2781   return os::commit_memory(addr, size);
2782 }
2783 
2784 // If this is a growable mapping, remove the guard pages entirely by
2785 // munmap()ping them.  If not, just call uncommit_memory(). This only
2786 // affects the main/initial thread, but guard against future OS changes
2787 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2788   uintptr_t stack_extent, stack_base;
2789   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2790   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2791       assert(os::Linux::is_initial_thread(),
2792            "growable stack in non-initial thread");
2793 
2794     return ::munmap(addr, size) == 0;
2795   }
2796 
2797   return os::uncommit_memory(addr, size);
2798 }
2799 
2800 static address _highest_vm_reserved_address = NULL;
2801 
2802 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2803 // at 'requested_addr'. If there are existing memory mappings at the same
2804 // location, however, they will be overwritten. If 'fixed' is false,
2805 // 'requested_addr' is only treated as a hint, the return value may or
2806 // may not start from the requested address. Unlike Linux mmap(), this
2807 // function returns NULL to indicate failure.
2808 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2809   char * addr;
2810   int flags;
2811 
2812   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2813   if (fixed) {
2814     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2815     flags |= MAP_FIXED;
2816   }
2817 
2818   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2819   // to PROT_EXEC if executable when we commit the page.
2820   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2821                        flags, -1, 0);
2822 
2823   if (addr != MAP_FAILED) {
2824     // anon_mmap() should only get called during VM initialization,
2825     // don't need lock (actually we can skip locking even it can be called
2826     // from multiple threads, because _highest_vm_reserved_address is just a
2827     // hint about the upper limit of non-stack memory regions.)
2828     if ((address)addr + bytes > _highest_vm_reserved_address) {
2829       _highest_vm_reserved_address = (address)addr + bytes;
2830     }
2831   }
2832 
2833   return addr == MAP_FAILED ? NULL : addr;
2834 }
2835 
2836 // Don't update _highest_vm_reserved_address, because there might be memory
2837 // regions above addr + size. If so, releasing a memory region only creates
2838 // a hole in the address space, it doesn't help prevent heap-stack collision.
2839 //
2840 static int anon_munmap(char * addr, size_t size) {
2841   return ::munmap(addr, size) == 0;
2842 }
2843 
2844 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2845                          size_t alignment_hint) {
2846   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2847 }
2848 
2849 bool os::pd_release_memory(char* addr, size_t size) {
2850   return anon_munmap(addr, size);
2851 }
2852 
2853 static address highest_vm_reserved_address() {
2854   return _highest_vm_reserved_address;
2855 }
2856 
2857 static bool linux_mprotect(char* addr, size_t size, int prot) {
2858   // Linux wants the mprotect address argument to be page aligned.
2859   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2860 
2861   // According to SUSv3, mprotect() should only be used with mappings
2862   // established by mmap(), and mmap() always maps whole pages. Unaligned
2863   // 'addr' likely indicates problem in the VM (e.g. trying to change
2864   // protection of malloc'ed or statically allocated memory). Check the
2865   // caller if you hit this assert.
2866   assert(addr == bottom, "sanity check");
2867 
2868   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2869   return ::mprotect(bottom, size, prot) == 0;
2870 }
2871 
2872 // Set protections specified
2873 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2874                         bool is_committed) {
2875   unsigned int p = 0;
2876   switch (prot) {
2877   case MEM_PROT_NONE: p = PROT_NONE; break;
2878   case MEM_PROT_READ: p = PROT_READ; break;
2879   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2880   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2881   default:
2882     ShouldNotReachHere();
2883   }
2884   // is_committed is unused.
2885   return linux_mprotect(addr, bytes, p);
2886 }
2887 
2888 bool os::guard_memory(char* addr, size_t size) {
2889   return linux_mprotect(addr, size, PROT_NONE);
2890 }
2891 
2892 bool os::unguard_memory(char* addr, size_t size) {
2893   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2894 }
2895 
2896 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2897   bool result = false;
2898   void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2899                   MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2900                   -1, 0);
2901 
2902   if (p != (void *) -1) {
2903     // We don't know if this really is a huge page or not.
2904     FILE *fp = fopen("/proc/self/maps", "r");
2905     if (fp) {
2906       while (!feof(fp)) {
2907         char chars[257];
2908         long x = 0;
2909         if (fgets(chars, sizeof(chars), fp)) {
2910           if (sscanf(chars, "%lx-%*x", &x) == 1
2911               && x == (long)p) {
2912             if (strstr (chars, "hugepage")) {
2913               result = true;
2914               break;
2915             }
2916           }
2917         }
2918       }
2919       fclose(fp);
2920     }
2921     munmap (p, page_size);
2922     if (result)
2923       return true;
2924   }
2925 
2926   if (warn) {
2927     warning("HugeTLBFS is not supported by the operating system.");
2928   }
2929 
2930   return result;
2931 }
2932 
2933 /*
2934 * Set the coredump_filter bits to include largepages in core dump (bit 6)
2935 *
2936 * From the coredump_filter documentation:
2937 *
2938 * - (bit 0) anonymous private memory
2939 * - (bit 1) anonymous shared memory
2940 * - (bit 2) file-backed private memory
2941 * - (bit 3) file-backed shared memory
2942 * - (bit 4) ELF header pages in file-backed private memory areas (it is
2943 *           effective only if the bit 2 is cleared)
2944 * - (bit 5) hugetlb private memory
2945 * - (bit 6) hugetlb shared memory
2946 */
2947 static void set_coredump_filter(void) {
2948   FILE *f;
2949   long cdm;
2950 
2951   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
2952     return;
2953   }
2954 
2955   if (fscanf(f, "%lx", &cdm) != 1) {
2956     fclose(f);
2957     return;
2958   }
2959 
2960   rewind(f);
2961 
2962   if ((cdm & LARGEPAGES_BIT) == 0) {
2963     cdm |= LARGEPAGES_BIT;
2964     fprintf(f, "%#lx", cdm);
2965   }
2966 
2967   fclose(f);
2968 }
2969 
2970 // Large page support
2971 
2972 static size_t _large_page_size = 0;
2973 
2974 void os::large_page_init() {
2975   if (!UseLargePages) {
2976     UseHugeTLBFS = false;
2977     UseSHM = false;
2978     return;
2979   }
2980 
2981   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
2982     // If UseLargePages is specified on the command line try both methods,
2983     // if it's default, then try only HugeTLBFS.
2984     if (FLAG_IS_DEFAULT(UseLargePages)) {
2985       UseHugeTLBFS = true;
2986     } else {
2987       UseHugeTLBFS = UseSHM = true;
2988     }
2989   }
2990 
2991   if (LargePageSizeInBytes) {
2992     _large_page_size = LargePageSizeInBytes;
2993   } else {
2994     // large_page_size on Linux is used to round up heap size. x86 uses either
2995     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2996     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2997     // page as large as 256M.
2998     //
2999     // Here we try to figure out page size by parsing /proc/meminfo and looking
3000     // for a line with the following format:
3001     //    Hugepagesize:     2048 kB
3002     //
3003     // If we can't determine the value (e.g. /proc is not mounted, or the text
3004     // format has been changed), we'll use the largest page size supported by
3005     // the processor.
3006 
3007 #ifndef ZERO
3008     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3009                        ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3010 #endif // ZERO
3011 
3012     FILE *fp = fopen("/proc/meminfo", "r");
3013     if (fp) {
3014       while (!feof(fp)) {
3015         int x = 0;
3016         char buf[16];
3017         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3018           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3019             _large_page_size = x * K;
3020             break;
3021           }
3022         } else {
3023           // skip to next line
3024           for (;;) {
3025             int ch = fgetc(fp);
3026             if (ch == EOF || ch == (int)'\n') break;
3027           }
3028         }
3029       }
3030       fclose(fp);
3031     }
3032   }
3033 
3034   // print a warning if any large page related flag is specified on command line
3035   bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3036 
3037   const size_t default_page_size = (size_t)Linux::page_size();
3038   if (_large_page_size > default_page_size) {
3039     _page_sizes[0] = _large_page_size;
3040     _page_sizes[1] = default_page_size;
3041     _page_sizes[2] = 0;
3042   }
3043   UseHugeTLBFS = UseHugeTLBFS &&
3044                  Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3045 
3046   if (UseHugeTLBFS)
3047     UseSHM = false;
3048 
3049   UseLargePages = UseHugeTLBFS || UseSHM;
3050 
3051   set_coredump_filter();
3052 }
3053 
3054 #ifndef SHM_HUGETLB
3055 #define SHM_HUGETLB 04000
3056 #endif
3057 
3058 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3059   // "exec" is passed in but not used.  Creating the shared image for
3060   // the code cache doesn't have an SHM_X executable permission to check.
3061   assert(UseLargePages && UseSHM, "only for SHM large pages");
3062 
3063   key_t key = IPC_PRIVATE;
3064   char *addr;
3065 
3066   bool warn_on_failure = UseLargePages &&
3067                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3068                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3069                         );
3070   char msg[128];
3071 
3072   // Create a large shared memory region to attach to based on size.
3073   // Currently, size is the total size of the heap
3074   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3075   if (shmid == -1) {
3076      // Possible reasons for shmget failure:
3077      // 1. shmmax is too small for Java heap.
3078      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3079      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3080      // 2. not enough large page memory.
3081      //    > check available large pages: cat /proc/meminfo
3082      //    > increase amount of large pages:
3083      //          echo new_value > /proc/sys/vm/nr_hugepages
3084      //      Note 1: different Linux may use different name for this property,
3085      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3086      //      Note 2: it's possible there's enough physical memory available but
3087      //            they are so fragmented after a long run that they can't
3088      //            coalesce into large pages. Try to reserve large pages when
3089      //            the system is still "fresh".
3090      if (warn_on_failure) {
3091        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3092        warning(msg);
3093      }
3094      return NULL;
3095   }
3096 
3097   // attach to the region
3098   addr = (char*)shmat(shmid, req_addr, 0);
3099   int err = errno;
3100 
3101   // Remove shmid. If shmat() is successful, the actual shared memory segment
3102   // will be deleted when it's detached by shmdt() or when the process
3103   // terminates. If shmat() is not successful this will remove the shared
3104   // segment immediately.
3105   shmctl(shmid, IPC_RMID, NULL);
3106 
3107   if ((intptr_t)addr == -1) {
3108      if (warn_on_failure) {
3109        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3110        warning(msg);
3111      }
3112      return NULL;
3113   }
3114 
3115   if ((addr != NULL) && UseNUMAInterleaving) {
3116     numa_make_global(addr, bytes);
3117   }
3118 
3119   return addr;
3120 }
3121 
3122 bool os::release_memory_special(char* base, size_t bytes) {
3123   // detaching the SHM segment will also delete it, see reserve_memory_special()
3124   int rslt = shmdt(base);
3125   return rslt == 0;
3126 }
3127 
3128 size_t os::large_page_size() {
3129   return _large_page_size;
3130 }
3131 
3132 // HugeTLBFS allows application to commit large page memory on demand;
3133 // with SysV SHM the entire memory region must be allocated as shared
3134 // memory.
3135 bool os::can_commit_large_page_memory() {
3136   return UseHugeTLBFS;
3137 }
3138 
3139 bool os::can_execute_large_page_memory() {
3140   return UseHugeTLBFS;
3141 }
3142 
3143 // Reserve memory at an arbitrary address, only if that area is
3144 // available (and not reserved for something else).
3145 
3146 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3147   const int max_tries = 10;
3148   char* base[max_tries];
3149   size_t size[max_tries];
3150   const size_t gap = 0x000000;
3151 
3152   // Assert only that the size is a multiple of the page size, since
3153   // that's all that mmap requires, and since that's all we really know
3154   // about at this low abstraction level.  If we need higher alignment,
3155   // we can either pass an alignment to this method or verify alignment
3156   // in one of the methods further up the call chain.  See bug 5044738.
3157   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3158 
3159   // Repeatedly allocate blocks until the block is allocated at the
3160   // right spot. Give up after max_tries. Note that reserve_memory() will
3161   // automatically update _highest_vm_reserved_address if the call is
3162   // successful. The variable tracks the highest memory address every reserved
3163   // by JVM. It is used to detect heap-stack collision if running with
3164   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3165   // space than needed, it could confuse the collision detecting code. To
3166   // solve the problem, save current _highest_vm_reserved_address and
3167   // calculate the correct value before return.
3168   address old_highest = _highest_vm_reserved_address;
3169 
3170   // Linux mmap allows caller to pass an address as hint; give it a try first,
3171   // if kernel honors the hint then we can return immediately.
3172   char * addr = anon_mmap(requested_addr, bytes, false);
3173   if (addr == requested_addr) {
3174      return requested_addr;
3175   }
3176 
3177   if (addr != NULL) {
3178      // mmap() is successful but it fails to reserve at the requested address
3179      anon_munmap(addr, bytes);
3180   }
3181 
3182   int i;
3183   for (i = 0; i < max_tries; ++i) {
3184     base[i] = reserve_memory(bytes);
3185 
3186     if (base[i] != NULL) {
3187       // Is this the block we wanted?
3188       if (base[i] == requested_addr) {
3189         size[i] = bytes;
3190         break;
3191       }
3192 
3193       // Does this overlap the block we wanted? Give back the overlapped
3194       // parts and try again.
3195 
3196       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3197       if (top_overlap >= 0 && top_overlap < bytes) {
3198         unmap_memory(base[i], top_overlap);
3199         base[i] += top_overlap;
3200         size[i] = bytes - top_overlap;
3201       } else {
3202         size_t bottom_overlap = base[i] + bytes - requested_addr;
3203         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3204           unmap_memory(requested_addr, bottom_overlap);
3205           size[i] = bytes - bottom_overlap;
3206         } else {
3207           size[i] = bytes;
3208         }
3209       }
3210     }
3211   }
3212 
3213   // Give back the unused reserved pieces.
3214 
3215   for (int j = 0; j < i; ++j) {
3216     if (base[j] != NULL) {
3217       unmap_memory(base[j], size[j]);
3218     }
3219   }
3220 
3221   if (i < max_tries) {
3222     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3223     return requested_addr;
3224   } else {
3225     _highest_vm_reserved_address = old_highest;
3226     return NULL;
3227   }
3228 }
3229 
3230 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3231   return ::read(fd, buf, nBytes);
3232 }
3233 
3234 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3235 // Solaris uses poll(), linux uses park().
3236 // Poll() is likely a better choice, assuming that Thread.interrupt()
3237 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3238 // SIGSEGV, see 4355769.
3239 
3240 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3241   assert(thread == Thread::current(),  "thread consistency check");
3242 
3243   ParkEvent * const slp = thread->_SleepEvent ;
3244   slp->reset() ;
3245   OrderAccess::fence() ;
3246 
3247   if (interruptible) {
3248     jlong prevtime = javaTimeNanos();
3249 
3250     for (;;) {
3251       if (os::is_interrupted(thread, true)) {
3252         return OS_INTRPT;
3253       }
3254 
3255       jlong newtime = javaTimeNanos();
3256 
3257       if (newtime - prevtime < 0) {
3258         // time moving backwards, should only happen if no monotonic clock
3259         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3260         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3261       } else {
3262         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3263       }
3264 
3265       if(millis <= 0) {
3266         return OS_OK;
3267       }
3268 
3269       prevtime = newtime;
3270 
3271       {
3272         assert(thread->is_Java_thread(), "sanity check");
3273         JavaThread *jt = (JavaThread *) thread;
3274         ThreadBlockInVM tbivm(jt);
3275         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3276 
3277         jt->set_suspend_equivalent();
3278         // cleared by handle_special_suspend_equivalent_condition() or
3279         // java_suspend_self() via check_and_wait_while_suspended()
3280 
3281         slp->park(millis);
3282 
3283         // were we externally suspended while we were waiting?
3284         jt->check_and_wait_while_suspended();
3285       }
3286     }
3287   } else {
3288     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3289     jlong prevtime = javaTimeNanos();
3290 
3291     for (;;) {
3292       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3293       // the 1st iteration ...
3294       jlong newtime = javaTimeNanos();
3295 
3296       if (newtime - prevtime < 0) {
3297         // time moving backwards, should only happen if no monotonic clock
3298         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3299         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3300       } else {
3301         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3302       }
3303 
3304       if(millis <= 0) break ;
3305 
3306       prevtime = newtime;
3307       slp->park(millis);
3308     }
3309     return OS_OK ;
3310   }
3311 }
3312 
3313 int os::naked_sleep() {
3314   // %% make the sleep time an integer flag. for now use 1 millisec.
3315   return os::sleep(Thread::current(), 1, false);
3316 }
3317 
3318 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3319 void os::infinite_sleep() {
3320   while (true) {    // sleep forever ...
3321     ::sleep(100);   // ... 100 seconds at a time
3322   }
3323 }
3324 
3325 // Used to convert frequent JVM_Yield() to nops
3326 bool os::dont_yield() {
3327   return DontYieldALot;
3328 }
3329 
3330 void os::yield() {
3331   sched_yield();
3332 }
3333 
3334 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3335 
3336 void os::yield_all(int attempts) {
3337   // Yields to all threads, including threads with lower priorities
3338   // Threads on Linux are all with same priority. The Solaris style
3339   // os::yield_all() with nanosleep(1ms) is not necessary.
3340   sched_yield();
3341 }
3342 
3343 // Called from the tight loops to possibly influence time-sharing heuristics
3344 void os::loop_breaker(int attempts) {
3345   os::yield_all(attempts);
3346 }
3347 
3348 ////////////////////////////////////////////////////////////////////////////////
3349 // thread priority support
3350 
3351 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3352 // only supports dynamic priority, static priority must be zero. For real-time
3353 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3354 // However, for large multi-threaded applications, SCHED_RR is not only slower
3355 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3356 // of 5 runs - Sep 2005).
3357 //
3358 // The following code actually changes the niceness of kernel-thread/LWP. It
3359 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3360 // not the entire user process, and user level threads are 1:1 mapped to kernel
3361 // threads. It has always been the case, but could change in the future. For
3362 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3363 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3364 
3365 int os::java_to_os_priority[CriticalPriority + 1] = {
3366   19,              // 0 Entry should never be used
3367 
3368    4,              // 1 MinPriority
3369    3,              // 2
3370    2,              // 3
3371 
3372    1,              // 4
3373    0,              // 5 NormPriority
3374   -1,              // 6
3375 
3376   -2,              // 7
3377   -3,              // 8
3378   -4,              // 9 NearMaxPriority
3379 
3380   -5,              // 10 MaxPriority
3381 
3382   -5               // 11 CriticalPriority
3383 };
3384 
3385 static int prio_init() {
3386   if (ThreadPriorityPolicy == 1) {
3387     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3388     // if effective uid is not root. Perhaps, a more elegant way of doing
3389     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3390     if (geteuid() != 0) {
3391       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3392         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3393       }
3394       ThreadPriorityPolicy = 0;
3395     }
3396   }
3397   if (UseCriticalJavaThreadPriority) {
3398     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3399   }
3400   return 0;
3401 }
3402 
3403 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3404   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3405 
3406   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3407   return (ret == 0) ? OS_OK : OS_ERR;
3408 }
3409 
3410 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3411   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3412     *priority_ptr = java_to_os_priority[NormPriority];
3413     return OS_OK;
3414   }
3415 
3416   errno = 0;
3417   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3418   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3419 }
3420 
3421 // Hint to the underlying OS that a task switch would not be good.
3422 // Void return because it's a hint and can fail.
3423 void os::hint_no_preempt() {}
3424 
3425 ////////////////////////////////////////////////////////////////////////////////
3426 // suspend/resume support
3427 
3428 //  the low-level signal-based suspend/resume support is a remnant from the
3429 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3430 //  within hotspot. Now there is a single use-case for this:
3431 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3432 //      that runs in the watcher thread.
3433 //  The remaining code is greatly simplified from the more general suspension
3434 //  code that used to be used.
3435 //
3436 //  The protocol is quite simple:
3437 //  - suspend:
3438 //      - sends a signal to the target thread
3439 //      - polls the suspend state of the osthread using a yield loop
3440 //      - target thread signal handler (SR_handler) sets suspend state
3441 //        and blocks in sigsuspend until continued
3442 //  - resume:
3443 //      - sets target osthread state to continue
3444 //      - sends signal to end the sigsuspend loop in the SR_handler
3445 //
3446 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3447 //
3448 
3449 static void resume_clear_context(OSThread *osthread) {
3450   osthread->set_ucontext(NULL);
3451   osthread->set_siginfo(NULL);
3452 
3453   // notify the suspend action is completed, we have now resumed
3454   osthread->sr.clear_suspended();
3455 }
3456 
3457 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3458   osthread->set_ucontext(context);
3459   osthread->set_siginfo(siginfo);
3460 }
3461 
3462 //
3463 // Handler function invoked when a thread's execution is suspended or
3464 // resumed. We have to be careful that only async-safe functions are
3465 // called here (Note: most pthread functions are not async safe and
3466 // should be avoided.)
3467 //
3468 // Note: sigwait() is a more natural fit than sigsuspend() from an
3469 // interface point of view, but sigwait() prevents the signal hander
3470 // from being run. libpthread would get very confused by not having
3471 // its signal handlers run and prevents sigwait()'s use with the
3472 // mutex granting granting signal.
3473 //
3474 // Currently only ever called on the VMThread
3475 //
3476 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3477   // Save and restore errno to avoid confusing native code with EINTR
3478   // after sigsuspend.
3479   int old_errno = errno;
3480 
3481   Thread* thread = Thread::current();
3482   OSThread* osthread = thread->osthread();
3483   assert(thread->is_VM_thread(), "Must be VMThread");
3484   // read current suspend action
3485   int action = osthread->sr.suspend_action();
3486   if (action == SR_SUSPEND) {
3487     suspend_save_context(osthread, siginfo, context);
3488 
3489     // Notify the suspend action is about to be completed. do_suspend()
3490     // waits until SR_SUSPENDED is set and then returns. We will wait
3491     // here for a resume signal and that completes the suspend-other
3492     // action. do_suspend/do_resume is always called as a pair from
3493     // the same thread - so there are no races
3494 
3495     // notify the caller
3496     osthread->sr.set_suspended();
3497 
3498     sigset_t suspend_set;  // signals for sigsuspend()
3499 
3500     // get current set of blocked signals and unblock resume signal
3501     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3502     sigdelset(&suspend_set, SR_signum);
3503 
3504     // wait here until we are resumed
3505     do {
3506       sigsuspend(&suspend_set);
3507       // ignore all returns until we get a resume signal
3508     } while (osthread->sr.suspend_action() != SR_CONTINUE);
3509 
3510     resume_clear_context(osthread);
3511 
3512   } else {
3513     assert(action == SR_CONTINUE, "unexpected sr action");
3514     // nothing special to do - just leave the handler
3515   }
3516 
3517   errno = old_errno;
3518 }
3519 
3520 
3521 static int SR_initialize() {
3522   struct sigaction act;
3523   char *s;
3524   /* Get signal number to use for suspend/resume */
3525   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3526     int sig = ::strtol(s, 0, 10);
3527     if (sig > 0 || sig < _NSIG) {
3528         SR_signum = sig;
3529     }
3530   }
3531 
3532   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3533         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3534 
3535   sigemptyset(&SR_sigset);
3536   sigaddset(&SR_sigset, SR_signum);
3537 
3538   /* Set up signal handler for suspend/resume */
3539   act.sa_flags = SA_RESTART|SA_SIGINFO;
3540   act.sa_handler = (void (*)(int)) SR_handler;
3541 
3542   // SR_signum is blocked by default.
3543   // 4528190 - We also need to block pthread restart signal (32 on all
3544   // supported Linux platforms). Note that LinuxThreads need to block
3545   // this signal for all threads to work properly. So we don't have
3546   // to use hard-coded signal number when setting up the mask.
3547   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3548 
3549   if (sigaction(SR_signum, &act, 0) == -1) {
3550     return -1;
3551   }
3552 
3553   // Save signal flag
3554   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3555   return 0;
3556 }
3557 
3558 static int SR_finalize() {
3559   return 0;
3560 }
3561 
3562 
3563 // returns true on success and false on error - really an error is fatal
3564 // but this seems the normal response to library errors
3565 static bool do_suspend(OSThread* osthread) {
3566   // mark as suspended and send signal
3567   osthread->sr.set_suspend_action(SR_SUSPEND);
3568   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3569   assert_status(status == 0, status, "pthread_kill");
3570 
3571   // check status and wait until notified of suspension
3572   if (status == 0) {
3573     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3574       os::yield_all(i);
3575     }
3576     osthread->sr.set_suspend_action(SR_NONE);
3577     return true;
3578   }
3579   else {
3580     osthread->sr.set_suspend_action(SR_NONE);
3581     return false;
3582   }
3583 }
3584 
3585 static void do_resume(OSThread* osthread) {
3586   assert(osthread->sr.is_suspended(), "thread should be suspended");
3587   osthread->sr.set_suspend_action(SR_CONTINUE);
3588 
3589   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3590   assert_status(status == 0, status, "pthread_kill");
3591   // check status and wait unit notified of resumption
3592   if (status == 0) {
3593     for (int i = 0; osthread->sr.is_suspended(); i++) {
3594       os::yield_all(i);
3595     }
3596   }
3597   osthread->sr.set_suspend_action(SR_NONE);
3598 }
3599 
3600 ////////////////////////////////////////////////////////////////////////////////
3601 // interrupt support
3602 
3603 void os::interrupt(Thread* thread) {
3604   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3605     "possibility of dangling Thread pointer");
3606 
3607   OSThread* osthread = thread->osthread();
3608 
3609   if (!osthread->interrupted()) {
3610     osthread->set_interrupted(true);
3611     // More than one thread can get here with the same value of osthread,
3612     // resulting in multiple notifications.  We do, however, want the store
3613     // to interrupted() to be visible to other threads before we execute unpark().
3614     OrderAccess::fence();
3615     ParkEvent * const slp = thread->_SleepEvent ;
3616     if (slp != NULL) slp->unpark() ;
3617   }
3618 
3619   // For JSR166. Unpark even if interrupt status already was set
3620   if (thread->is_Java_thread())
3621     ((JavaThread*)thread)->parker()->unpark();
3622 
3623   ParkEvent * ev = thread->_ParkEvent ;
3624   if (ev != NULL) ev->unpark() ;
3625 
3626 }
3627 
3628 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3629   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3630     "possibility of dangling Thread pointer");
3631 
3632   OSThread* osthread = thread->osthread();
3633 
3634   bool interrupted = osthread->interrupted();
3635 
3636   if (interrupted && clear_interrupted) {
3637     osthread->set_interrupted(false);
3638     // consider thread->_SleepEvent->reset() ... optional optimization
3639   }
3640 
3641   return interrupted;
3642 }
3643 
3644 ///////////////////////////////////////////////////////////////////////////////////
3645 // signal handling (except suspend/resume)
3646 
3647 // This routine may be used by user applications as a "hook" to catch signals.
3648 // The user-defined signal handler must pass unrecognized signals to this
3649 // routine, and if it returns true (non-zero), then the signal handler must
3650 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3651 // routine will never retun false (zero), but instead will execute a VM panic
3652 // routine kill the process.
3653 //
3654 // If this routine returns false, it is OK to call it again.  This allows
3655 // the user-defined signal handler to perform checks either before or after
3656 // the VM performs its own checks.  Naturally, the user code would be making
3657 // a serious error if it tried to handle an exception (such as a null check
3658 // or breakpoint) that the VM was generating for its own correct operation.
3659 //
3660 // This routine may recognize any of the following kinds of signals:
3661 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3662 // It should be consulted by handlers for any of those signals.
3663 //
3664 // The caller of this routine must pass in the three arguments supplied
3665 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3666 // field of the structure passed to sigaction().  This routine assumes that
3667 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3668 //
3669 // Note that the VM will print warnings if it detects conflicting signal
3670 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3671 //
3672 extern "C" JNIEXPORT int
3673 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3674                         void* ucontext, int abort_if_unrecognized);
3675 
3676 void signalHandler(int sig, siginfo_t* info, void* uc) {
3677   assert(info != NULL && uc != NULL, "it must be old kernel");
3678   JVM_handle_linux_signal(sig, info, uc, true);
3679 }
3680 
3681 
3682 // This boolean allows users to forward their own non-matching signals
3683 // to JVM_handle_linux_signal, harmlessly.
3684 bool os::Linux::signal_handlers_are_installed = false;
3685 
3686 // For signal-chaining
3687 struct sigaction os::Linux::sigact[MAXSIGNUM];
3688 unsigned int os::Linux::sigs = 0;
3689 bool os::Linux::libjsig_is_loaded = false;
3690 typedef struct sigaction *(*get_signal_t)(int);
3691 get_signal_t os::Linux::get_signal_action = NULL;
3692 
3693 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3694   struct sigaction *actp = NULL;
3695 
3696   if (libjsig_is_loaded) {
3697     // Retrieve the old signal handler from libjsig
3698     actp = (*get_signal_action)(sig);
3699   }
3700   if (actp == NULL) {
3701     // Retrieve the preinstalled signal handler from jvm
3702     actp = get_preinstalled_handler(sig);
3703   }
3704 
3705   return actp;
3706 }
3707 
3708 static bool call_chained_handler(struct sigaction *actp, int sig,
3709                                  siginfo_t *siginfo, void *context) {
3710   // Call the old signal handler
3711   if (actp->sa_handler == SIG_DFL) {
3712     // It's more reasonable to let jvm treat it as an unexpected exception
3713     // instead of taking the default action.
3714     return false;
3715   } else if (actp->sa_handler != SIG_IGN) {
3716     if ((actp->sa_flags & SA_NODEFER) == 0) {
3717       // automaticlly block the signal
3718       sigaddset(&(actp->sa_mask), sig);
3719     }
3720 
3721     sa_handler_t hand;
3722     sa_sigaction_t sa;
3723     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3724     // retrieve the chained handler
3725     if (siginfo_flag_set) {
3726       sa = actp->sa_sigaction;
3727     } else {
3728       hand = actp->sa_handler;
3729     }
3730 
3731     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3732       actp->sa_handler = SIG_DFL;
3733     }
3734 
3735     // try to honor the signal mask
3736     sigset_t oset;
3737     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3738 
3739     // call into the chained handler
3740     if (siginfo_flag_set) {
3741       (*sa)(sig, siginfo, context);
3742     } else {
3743       (*hand)(sig);
3744     }
3745 
3746     // restore the signal mask
3747     pthread_sigmask(SIG_SETMASK, &oset, 0);
3748   }
3749   // Tell jvm's signal handler the signal is taken care of.
3750   return true;
3751 }
3752 
3753 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3754   bool chained = false;
3755   // signal-chaining
3756   if (UseSignalChaining) {
3757     struct sigaction *actp = get_chained_signal_action(sig);
3758     if (actp != NULL) {
3759       chained = call_chained_handler(actp, sig, siginfo, context);
3760     }
3761   }
3762   return chained;
3763 }
3764 
3765 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3766   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3767     return &sigact[sig];
3768   }
3769   return NULL;
3770 }
3771 
3772 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3773   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3774   sigact[sig] = oldAct;
3775   sigs |= (unsigned int)1 << sig;
3776 }
3777 
3778 // for diagnostic
3779 int os::Linux::sigflags[MAXSIGNUM];
3780 
3781 int os::Linux::get_our_sigflags(int sig) {
3782   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3783   return sigflags[sig];
3784 }
3785 
3786 void os::Linux::set_our_sigflags(int sig, int flags) {
3787   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3788   sigflags[sig] = flags;
3789 }
3790 
3791 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3792   // Check for overwrite.
3793   struct sigaction oldAct;
3794   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3795 
3796   void* oldhand = oldAct.sa_sigaction
3797                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3798                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3799   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3800       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3801       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3802     if (AllowUserSignalHandlers || !set_installed) {
3803       // Do not overwrite; user takes responsibility to forward to us.
3804       return;
3805     } else if (UseSignalChaining) {
3806       // save the old handler in jvm
3807       save_preinstalled_handler(sig, oldAct);
3808       // libjsig also interposes the sigaction() call below and saves the
3809       // old sigaction on it own.
3810     } else {
3811       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3812                     "%#lx for signal %d.", (long)oldhand, sig));
3813     }
3814   }
3815 
3816   struct sigaction sigAct;
3817   sigfillset(&(sigAct.sa_mask));
3818   sigAct.sa_handler = SIG_DFL;
3819   if (!set_installed) {
3820     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3821   } else {
3822     sigAct.sa_sigaction = signalHandler;
3823     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3824   }
3825   // Save flags, which are set by ours
3826   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3827   sigflags[sig] = sigAct.sa_flags;
3828 
3829   int ret = sigaction(sig, &sigAct, &oldAct);
3830   assert(ret == 0, "check");
3831 
3832   void* oldhand2  = oldAct.sa_sigaction
3833                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3834                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3835   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3836 }
3837 
3838 // install signal handlers for signals that HotSpot needs to
3839 // handle in order to support Java-level exception handling.
3840 
3841 void os::Linux::install_signal_handlers() {
3842   if (!signal_handlers_are_installed) {
3843     signal_handlers_are_installed = true;
3844 
3845     // signal-chaining
3846     typedef void (*signal_setting_t)();
3847     signal_setting_t begin_signal_setting = NULL;
3848     signal_setting_t end_signal_setting = NULL;
3849     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3850                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3851     if (begin_signal_setting != NULL) {
3852       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3853                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3854       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3855                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3856       libjsig_is_loaded = true;
3857       assert(UseSignalChaining, "should enable signal-chaining");
3858     }
3859     if (libjsig_is_loaded) {
3860       // Tell libjsig jvm is setting signal handlers
3861       (*begin_signal_setting)();
3862     }
3863 
3864     set_signal_handler(SIGSEGV, true);
3865     set_signal_handler(SIGPIPE, true);
3866     set_signal_handler(SIGBUS, true);
3867     set_signal_handler(SIGILL, true);
3868     set_signal_handler(SIGFPE, true);
3869     set_signal_handler(SIGXFSZ, true);
3870 
3871     if (libjsig_is_loaded) {
3872       // Tell libjsig jvm finishes setting signal handlers
3873       (*end_signal_setting)();
3874     }
3875 
3876     // We don't activate signal checker if libjsig is in place, we trust ourselves
3877     // and if UserSignalHandler is installed all bets are off.
3878     // Log that signal checking is off only if -verbose:jni is specified.
3879     if (CheckJNICalls) {
3880       if (libjsig_is_loaded) {
3881         if (PrintJNIResolving) {
3882           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3883         }
3884         check_signals = false;
3885       }
3886       if (AllowUserSignalHandlers) {
3887         if (PrintJNIResolving) {
3888           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3889         }
3890         check_signals = false;
3891       }
3892     }
3893   }
3894 }
3895 
3896 // This is the fastest way to get thread cpu time on Linux.
3897 // Returns cpu time (user+sys) for any thread, not only for current.
3898 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3899 // It might work on 2.6.10+ with a special kernel/glibc patch.
3900 // For reference, please, see IEEE Std 1003.1-2004:
3901 //   http://www.unix.org/single_unix_specification
3902 
3903 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3904   struct timespec tp;
3905   int rc = os::Linux::clock_gettime(clockid, &tp);
3906   assert(rc == 0, "clock_gettime is expected to return 0 code");
3907 
3908   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
3909 }
3910 
3911 /////
3912 // glibc on Linux platform uses non-documented flag
3913 // to indicate, that some special sort of signal
3914 // trampoline is used.
3915 // We will never set this flag, and we should
3916 // ignore this flag in our diagnostic
3917 #ifdef SIGNIFICANT_SIGNAL_MASK
3918 #undef SIGNIFICANT_SIGNAL_MASK
3919 #endif
3920 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3921 
3922 static const char* get_signal_handler_name(address handler,
3923                                            char* buf, int buflen) {
3924   int offset;
3925   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3926   if (found) {
3927     // skip directory names
3928     const char *p1, *p2;
3929     p1 = buf;
3930     size_t len = strlen(os::file_separator());
3931     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3932     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3933   } else {
3934     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3935   }
3936   return buf;
3937 }
3938 
3939 static void print_signal_handler(outputStream* st, int sig,
3940                                  char* buf, size_t buflen) {
3941   struct sigaction sa;
3942 
3943   sigaction(sig, NULL, &sa);
3944 
3945   // See comment for SIGNIFICANT_SIGNAL_MASK define
3946   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3947 
3948   st->print("%s: ", os::exception_name(sig, buf, buflen));
3949 
3950   address handler = (sa.sa_flags & SA_SIGINFO)
3951     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3952     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3953 
3954   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3955     st->print("SIG_DFL");
3956   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3957     st->print("SIG_IGN");
3958   } else {
3959     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3960   }
3961 
3962   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3963 
3964   address rh = VMError::get_resetted_sighandler(sig);
3965   // May be, handler was resetted by VMError?
3966   if(rh != NULL) {
3967     handler = rh;
3968     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3969   }
3970 
3971   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
3972 
3973   // Check: is it our handler?
3974   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3975      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3976     // It is our signal handler
3977     // check for flags, reset system-used one!
3978     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3979       st->print(
3980                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3981                 os::Linux::get_our_sigflags(sig));
3982     }
3983   }
3984   st->cr();
3985 }
3986 
3987 
3988 #define DO_SIGNAL_CHECK(sig) \
3989   if (!sigismember(&check_signal_done, sig)) \
3990     os::Linux::check_signal_handler(sig)
3991 
3992 // This method is a periodic task to check for misbehaving JNI applications
3993 // under CheckJNI, we can add any periodic checks here
3994 
3995 void os::run_periodic_checks() {
3996 
3997   if (check_signals == false) return;
3998 
3999   // SEGV and BUS if overridden could potentially prevent
4000   // generation of hs*.log in the event of a crash, debugging
4001   // such a case can be very challenging, so we absolutely
4002   // check the following for a good measure:
4003   DO_SIGNAL_CHECK(SIGSEGV);
4004   DO_SIGNAL_CHECK(SIGILL);
4005   DO_SIGNAL_CHECK(SIGFPE);
4006   DO_SIGNAL_CHECK(SIGBUS);
4007   DO_SIGNAL_CHECK(SIGPIPE);
4008   DO_SIGNAL_CHECK(SIGXFSZ);
4009 
4010 
4011   // ReduceSignalUsage allows the user to override these handlers
4012   // see comments at the very top and jvm_solaris.h
4013   if (!ReduceSignalUsage) {
4014     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4015     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4016     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4017     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4018   }
4019 
4020   DO_SIGNAL_CHECK(SR_signum);
4021   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4022 }
4023 
4024 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4025 
4026 static os_sigaction_t os_sigaction = NULL;
4027 
4028 void os::Linux::check_signal_handler(int sig) {
4029   char buf[O_BUFLEN];
4030   address jvmHandler = NULL;
4031 
4032 
4033   struct sigaction act;
4034   if (os_sigaction == NULL) {
4035     // only trust the default sigaction, in case it has been interposed
4036     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4037     if (os_sigaction == NULL) return;
4038   }
4039 
4040   os_sigaction(sig, (struct sigaction*)NULL, &act);
4041 
4042 
4043   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4044 
4045   address thisHandler = (act.sa_flags & SA_SIGINFO)
4046     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4047     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4048 
4049 
4050   switch(sig) {
4051   case SIGSEGV:
4052   case SIGBUS:
4053   case SIGFPE:
4054   case SIGPIPE:
4055   case SIGILL:
4056   case SIGXFSZ:
4057     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4058     break;
4059 
4060   case SHUTDOWN1_SIGNAL:
4061   case SHUTDOWN2_SIGNAL:
4062   case SHUTDOWN3_SIGNAL:
4063   case BREAK_SIGNAL:
4064     jvmHandler = (address)user_handler();
4065     break;
4066 
4067   case INTERRUPT_SIGNAL:
4068     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4069     break;
4070 
4071   default:
4072     if (sig == SR_signum) {
4073       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4074     } else {
4075       return;
4076     }
4077     break;
4078   }
4079 
4080   if (thisHandler != jvmHandler) {
4081     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4082     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4083     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4084     // No need to check this sig any longer
4085     sigaddset(&check_signal_done, sig);
4086   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4087     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4088     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4089     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4090     // No need to check this sig any longer
4091     sigaddset(&check_signal_done, sig);
4092   }
4093 
4094   // Dump all the signal
4095   if (sigismember(&check_signal_done, sig)) {
4096     print_signal_handlers(tty, buf, O_BUFLEN);
4097   }
4098 }
4099 
4100 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4101 
4102 extern bool signal_name(int signo, char* buf, size_t len);
4103 
4104 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4105   if (0 < exception_code && exception_code <= SIGRTMAX) {
4106     // signal
4107     if (!signal_name(exception_code, buf, size)) {
4108       jio_snprintf(buf, size, "SIG%d", exception_code);
4109     }
4110     return buf;
4111   } else {
4112     return NULL;
4113   }
4114 }
4115 
4116 // this is called _before_ the most of global arguments have been parsed
4117 void os::init(void) {
4118   char dummy;   /* used to get a guess on initial stack address */
4119 //  first_hrtime = gethrtime();
4120 
4121   // With LinuxThreads the JavaMain thread pid (primordial thread)
4122   // is different than the pid of the java launcher thread.
4123   // So, on Linux, the launcher thread pid is passed to the VM
4124   // via the sun.java.launcher.pid property.
4125   // Use this property instead of getpid() if it was correctly passed.
4126   // See bug 6351349.
4127   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4128 
4129   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4130 
4131   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4132 
4133   init_random(1234567);
4134 
4135   ThreadCritical::initialize();
4136 
4137   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4138   if (Linux::page_size() == -1) {
4139     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4140                   strerror(errno)));
4141   }
4142   init_page_sizes((size_t) Linux::page_size());
4143 
4144   Linux::initialize_system_info();
4145 
4146   // main_thread points to the aboriginal thread
4147   Linux::_main_thread = pthread_self();
4148 
4149   Linux::clock_init();
4150   initial_time_count = os::elapsed_counter();
4151   pthread_mutex_init(&dl_mutex, NULL);
4152 }
4153 
4154 // To install functions for atexit system call
4155 extern "C" {
4156   static void perfMemory_exit_helper() {
4157     perfMemory_exit();
4158   }
4159 }
4160 
4161 // this is called _after_ the global arguments have been parsed
4162 jint os::init_2(void)
4163 {
4164   Linux::fast_thread_clock_init();
4165 
4166   // Allocate a single page and mark it as readable for safepoint polling
4167   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4168   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4169 
4170   os::set_polling_page( polling_page );
4171 
4172 #ifndef PRODUCT
4173   if(Verbose && PrintMiscellaneous)
4174     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4175 #endif
4176 
4177   if (!UseMembar) {
4178     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4179     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4180     os::set_memory_serialize_page( mem_serialize_page );
4181 
4182 #ifndef PRODUCT
4183     if(Verbose && PrintMiscellaneous)
4184       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4185 #endif
4186   }
4187 
4188   os::large_page_init();
4189 
4190   // initialize suspend/resume support - must do this before signal_sets_init()
4191   if (SR_initialize() != 0) {
4192     perror("SR_initialize failed");
4193     return JNI_ERR;
4194   }
4195 
4196   Linux::signal_sets_init();
4197   Linux::install_signal_handlers();
4198 
4199   // Check minimum allowable stack size for thread creation and to initialize
4200   // the java system classes, including StackOverflowError - depends on page
4201   // size.  Add a page for compiler2 recursion in main thread.
4202   // Add in 2*BytesPerWord times page size to account for VM stack during
4203   // class initialization depending on 32 or 64 bit VM.
4204   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4205             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4206                     2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4207 
4208   size_t threadStackSizeInBytes = ThreadStackSize * K;
4209   if (threadStackSizeInBytes != 0 &&
4210       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4211         tty->print_cr("\nThe stack size specified is too small, "
4212                       "Specify at least %dk",
4213                       os::Linux::min_stack_allowed/ K);
4214         return JNI_ERR;
4215   }
4216 
4217   // Make the stack size a multiple of the page size so that
4218   // the yellow/red zones can be guarded.
4219   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4220         vm_page_size()));
4221 
4222   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4223 
4224   Linux::libpthread_init();
4225   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4226      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4227           Linux::glibc_version(), Linux::libpthread_version(),
4228           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4229   }
4230 
4231   if (UseNUMA) {
4232     if (!Linux::libnuma_init()) {
4233       UseNUMA = false;
4234     } else {
4235       if ((Linux::numa_max_node() < 1)) {
4236         // There's only one node(they start from 0), disable NUMA.
4237         UseNUMA = false;
4238       }
4239     }
4240     // With SHM large pages we cannot uncommit a page, so there's not way
4241     // we can make the adaptive lgrp chunk resizing work. If the user specified
4242     // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4243     // disable adaptive resizing.
4244     if (UseNUMA && UseLargePages && UseSHM) {
4245       if (!FLAG_IS_DEFAULT(UseNUMA)) {
4246         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4247           UseLargePages = false;
4248         } else {
4249           warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4250           UseAdaptiveSizePolicy = false;
4251           UseAdaptiveNUMAChunkSizing = false;
4252         }
4253       } else {
4254         UseNUMA = false;
4255       }
4256     }
4257     if (!UseNUMA && ForceNUMA) {
4258       UseNUMA = true;
4259     }
4260   }
4261 
4262   if (MaxFDLimit) {
4263     // set the number of file descriptors to max. print out error
4264     // if getrlimit/setrlimit fails but continue regardless.
4265     struct rlimit nbr_files;
4266     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4267     if (status != 0) {
4268       if (PrintMiscellaneous && (Verbose || WizardMode))
4269         perror("os::init_2 getrlimit failed");
4270     } else {
4271       nbr_files.rlim_cur = nbr_files.rlim_max;
4272       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4273       if (status != 0) {
4274         if (PrintMiscellaneous && (Verbose || WizardMode))
4275           perror("os::init_2 setrlimit failed");
4276       }
4277     }
4278   }
4279 
4280   // Initialize lock used to serialize thread creation (see os::create_thread)
4281   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4282 
4283   // at-exit methods are called in the reverse order of their registration.
4284   // atexit functions are called on return from main or as a result of a
4285   // call to exit(3C). There can be only 32 of these functions registered
4286   // and atexit() does not set errno.
4287 
4288   if (PerfAllowAtExitRegistration) {
4289     // only register atexit functions if PerfAllowAtExitRegistration is set.
4290     // atexit functions can be delayed until process exit time, which
4291     // can be problematic for embedded VM situations. Embedded VMs should
4292     // call DestroyJavaVM() to assure that VM resources are released.
4293 
4294     // note: perfMemory_exit_helper atexit function may be removed in
4295     // the future if the appropriate cleanup code can be added to the
4296     // VM_Exit VMOperation's doit method.
4297     if (atexit(perfMemory_exit_helper) != 0) {
4298       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4299     }
4300   }
4301 
4302   // initialize thread priority policy
4303   prio_init();
4304 
4305   return JNI_OK;
4306 }
4307 
4308 // this is called at the end of vm_initialization
4309 void os::init_3(void)
4310 {
4311 #ifdef JAVASE_EMBEDDED
4312   // Start the MemNotifyThread
4313   if (LowMemoryProtection) {
4314     MemNotifyThread::start();
4315   }
4316   return;
4317 #endif
4318 }
4319 
4320 // Mark the polling page as unreadable
4321 void os::make_polling_page_unreadable(void) {
4322   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4323     fatal("Could not disable polling page");
4324 };
4325 
4326 // Mark the polling page as readable
4327 void os::make_polling_page_readable(void) {
4328   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4329     fatal("Could not enable polling page");
4330   }
4331 };
4332 
4333 int os::active_processor_count() {
4334   // Linux doesn't yet have a (official) notion of processor sets,
4335   // so just return the number of online processors.
4336   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4337   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4338   return online_cpus;
4339 }
4340 
4341 void os::set_native_thread_name(const char *name) {
4342   // Not yet implemented.
4343   return;
4344 }
4345 
4346 bool os::distribute_processes(uint length, uint* distribution) {
4347   // Not yet implemented.
4348   return false;
4349 }
4350 
4351 bool os::bind_to_processor(uint processor_id) {
4352   // Not yet implemented.
4353   return false;
4354 }
4355 
4356 ///
4357 
4358 // Suspends the target using the signal mechanism and then grabs the PC before
4359 // resuming the target. Used by the flat-profiler only
4360 ExtendedPC os::get_thread_pc(Thread* thread) {
4361   // Make sure that it is called by the watcher for the VMThread
4362   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4363   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4364 
4365   ExtendedPC epc;
4366 
4367   OSThread* osthread = thread->osthread();
4368   if (do_suspend(osthread)) {
4369     if (osthread->ucontext() != NULL) {
4370       epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4371     } else {
4372       // NULL context is unexpected, double-check this is the VMThread
4373       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4374     }
4375     do_resume(osthread);
4376   }
4377   // failure means pthread_kill failed for some reason - arguably this is
4378   // a fatal problem, but such problems are ignored elsewhere
4379 
4380   return epc;
4381 }
4382 
4383 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4384 {
4385    if (is_NPTL()) {
4386       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4387    } else {
4388 #ifndef IA64
4389       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4390       // word back to default 64bit precision if condvar is signaled. Java
4391       // wants 53bit precision.  Save and restore current value.
4392       int fpu = get_fpu_control_word();
4393 #endif // IA64
4394       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4395 #ifndef IA64
4396       set_fpu_control_word(fpu);
4397 #endif // IA64
4398       return status;
4399    }
4400 }
4401 
4402 ////////////////////////////////////////////////////////////////////////////////
4403 // debug support
4404 
4405 static address same_page(address x, address y) {
4406   int page_bits = -os::vm_page_size();
4407   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4408     return x;
4409   else if (x > y)
4410     return (address)(intptr_t(y) | ~page_bits) + 1;
4411   else
4412     return (address)(intptr_t(y) & page_bits);
4413 }
4414 
4415 bool os::find(address addr, outputStream* st) {
4416   Dl_info dlinfo;
4417   memset(&dlinfo, 0, sizeof(dlinfo));
4418   if (dladdr(addr, &dlinfo)) {
4419     st->print(PTR_FORMAT ": ", addr);
4420     if (dlinfo.dli_sname != NULL) {
4421       st->print("%s+%#x", dlinfo.dli_sname,
4422                  addr - (intptr_t)dlinfo.dli_saddr);
4423     } else if (dlinfo.dli_fname) {
4424       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4425     } else {
4426       st->print("<absolute address>");
4427     }
4428     if (dlinfo.dli_fname) {
4429       st->print(" in %s", dlinfo.dli_fname);
4430     }
4431     if (dlinfo.dli_fbase) {
4432       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4433     }
4434     st->cr();
4435 
4436     if (Verbose) {
4437       // decode some bytes around the PC
4438       address begin = same_page(addr-40, addr);
4439       address end   = same_page(addr+40, addr);
4440       address       lowest = (address) dlinfo.dli_sname;
4441       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4442       if (begin < lowest)  begin = lowest;
4443       Dl_info dlinfo2;
4444       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4445           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4446         end = (address) dlinfo2.dli_saddr;
4447       Disassembler::decode(begin, end, st);
4448     }
4449     return true;
4450   }
4451   return false;
4452 }
4453 
4454 ////////////////////////////////////////////////////////////////////////////////
4455 // misc
4456 
4457 // This does not do anything on Linux. This is basically a hook for being
4458 // able to use structured exception handling (thread-local exception filters)
4459 // on, e.g., Win32.
4460 void
4461 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4462                          JavaCallArguments* args, Thread* thread) {
4463   f(value, method, args, thread);
4464 }
4465 
4466 void os::print_statistics() {
4467 }
4468 
4469 int os::message_box(const char* title, const char* message) {
4470   int i;
4471   fdStream err(defaultStream::error_fd());
4472   for (i = 0; i < 78; i++) err.print_raw("=");
4473   err.cr();
4474   err.print_raw_cr(title);
4475   for (i = 0; i < 78; i++) err.print_raw("-");
4476   err.cr();
4477   err.print_raw_cr(message);
4478   for (i = 0; i < 78; i++) err.print_raw("=");
4479   err.cr();
4480 
4481   char buf[16];
4482   // Prevent process from exiting upon "read error" without consuming all CPU
4483   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4484 
4485   return buf[0] == 'y' || buf[0] == 'Y';
4486 }
4487 
4488 int os::stat(const char *path, struct stat *sbuf) {
4489   char pathbuf[MAX_PATH];
4490   if (strlen(path) > MAX_PATH - 1) {
4491     errno = ENAMETOOLONG;
4492     return -1;
4493   }
4494   os::native_path(strcpy(pathbuf, path));
4495   return ::stat(pathbuf, sbuf);
4496 }
4497 
4498 bool os::check_heap(bool force) {
4499   return true;
4500 }
4501 
4502 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4503   return ::vsnprintf(buf, count, format, args);
4504 }
4505 
4506 // Is a (classpath) directory empty?
4507 bool os::dir_is_empty(const char* path) {
4508   DIR *dir = NULL;
4509   struct dirent *ptr;
4510 
4511   dir = opendir(path);
4512   if (dir == NULL) return true;
4513 
4514   /* Scan the directory */
4515   bool result = true;
4516   char buf[sizeof(struct dirent) + MAX_PATH];
4517   while (result && (ptr = ::readdir(dir)) != NULL) {
4518     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4519       result = false;
4520     }
4521   }
4522   closedir(dir);
4523   return result;
4524 }
4525 
4526 // This code originates from JDK's sysOpen and open64_w
4527 // from src/solaris/hpi/src/system_md.c
4528 
4529 #ifndef O_DELETE
4530 #define O_DELETE 0x10000
4531 #endif
4532 
4533 // Open a file. Unlink the file immediately after open returns
4534 // if the specified oflag has the O_DELETE flag set.
4535 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4536 
4537 int os::open(const char *path, int oflag, int mode) {
4538 
4539   if (strlen(path) > MAX_PATH - 1) {
4540     errno = ENAMETOOLONG;
4541     return -1;
4542   }
4543   int fd;
4544   int o_delete = (oflag & O_DELETE);
4545   oflag = oflag & ~O_DELETE;
4546 
4547   fd = ::open64(path, oflag, mode);
4548   if (fd == -1) return -1;
4549 
4550   //If the open succeeded, the file might still be a directory
4551   {
4552     struct stat64 buf64;
4553     int ret = ::fstat64(fd, &buf64);
4554     int st_mode = buf64.st_mode;
4555 
4556     if (ret != -1) {
4557       if ((st_mode & S_IFMT) == S_IFDIR) {
4558         errno = EISDIR;
4559         ::close(fd);
4560         return -1;
4561       }
4562     } else {
4563       ::close(fd);
4564       return -1;
4565     }
4566   }
4567 
4568     /*
4569      * All file descriptors that are opened in the JVM and not
4570      * specifically destined for a subprocess should have the
4571      * close-on-exec flag set.  If we don't set it, then careless 3rd
4572      * party native code might fork and exec without closing all
4573      * appropriate file descriptors (e.g. as we do in closeDescriptors in
4574      * UNIXProcess.c), and this in turn might:
4575      *
4576      * - cause end-of-file to fail to be detected on some file
4577      *   descriptors, resulting in mysterious hangs, or
4578      *
4579      * - might cause an fopen in the subprocess to fail on a system
4580      *   suffering from bug 1085341.
4581      *
4582      * (Yes, the default setting of the close-on-exec flag is a Unix
4583      * design flaw)
4584      *
4585      * See:
4586      * 1085341: 32-bit stdio routines should support file descriptors >255
4587      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4588      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4589      */
4590 #ifdef FD_CLOEXEC
4591     {
4592         int flags = ::fcntl(fd, F_GETFD);
4593         if (flags != -1)
4594             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4595     }
4596 #endif
4597 
4598   if (o_delete != 0) {
4599     ::unlink(path);
4600   }
4601   return fd;
4602 }
4603 
4604 
4605 // create binary file, rewriting existing file if required
4606 int os::create_binary_file(const char* path, bool rewrite_existing) {
4607   int oflags = O_WRONLY | O_CREAT;
4608   if (!rewrite_existing) {
4609     oflags |= O_EXCL;
4610   }
4611   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4612 }
4613 
4614 // return current position of file pointer
4615 jlong os::current_file_offset(int fd) {
4616   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4617 }
4618 
4619 // move file pointer to the specified offset
4620 jlong os::seek_to_file_offset(int fd, jlong offset) {
4621   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4622 }
4623 
4624 // This code originates from JDK's sysAvailable
4625 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4626 
4627 int os::available(int fd, jlong *bytes) {
4628   jlong cur, end;
4629   int mode;
4630   struct stat64 buf64;
4631 
4632   if (::fstat64(fd, &buf64) >= 0) {
4633     mode = buf64.st_mode;
4634     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4635       /*
4636       * XXX: is the following call interruptible? If so, this might
4637       * need to go through the INTERRUPT_IO() wrapper as for other
4638       * blocking, interruptible calls in this file.
4639       */
4640       int n;
4641       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4642         *bytes = n;
4643         return 1;
4644       }
4645     }
4646   }
4647   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4648     return 0;
4649   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4650     return 0;
4651   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4652     return 0;
4653   }
4654   *bytes = end - cur;
4655   return 1;
4656 }
4657 
4658 int os::socket_available(int fd, jint *pbytes) {
4659   // Linux doc says EINTR not returned, unlike Solaris
4660   int ret = ::ioctl(fd, FIONREAD, pbytes);
4661 
4662   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4663   // is expected to return 0 on failure and 1 on success to the jdk.
4664   return (ret < 0) ? 0 : 1;
4665 }
4666 
4667 // Map a block of memory.
4668 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4669                      char *addr, size_t bytes, bool read_only,
4670                      bool allow_exec) {
4671   int prot;
4672   int flags = MAP_PRIVATE;
4673 
4674   if (read_only) {
4675     prot = PROT_READ;
4676   } else {
4677     prot = PROT_READ | PROT_WRITE;
4678   }
4679 
4680   if (allow_exec) {
4681     prot |= PROT_EXEC;
4682   }
4683 
4684   if (addr != NULL) {
4685     flags |= MAP_FIXED;
4686   }
4687 
4688   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4689                                      fd, file_offset);
4690   if (mapped_address == MAP_FAILED) {
4691     return NULL;
4692   }
4693   return mapped_address;
4694 }
4695 
4696 
4697 // Remap a block of memory.
4698 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4699                        char *addr, size_t bytes, bool read_only,
4700                        bool allow_exec) {
4701   // same as map_memory() on this OS
4702   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4703                         allow_exec);
4704 }
4705 
4706 
4707 // Unmap a block of memory.
4708 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4709   return munmap(addr, bytes) == 0;
4710 }
4711 
4712 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4713 
4714 static clockid_t thread_cpu_clockid(Thread* thread) {
4715   pthread_t tid = thread->osthread()->pthread_id();
4716   clockid_t clockid;
4717 
4718   // Get thread clockid
4719   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4720   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4721   return clockid;
4722 }
4723 
4724 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4725 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4726 // of a thread.
4727 //
4728 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4729 // the fast estimate available on the platform.
4730 
4731 jlong os::current_thread_cpu_time() {
4732   if (os::Linux::supports_fast_thread_cpu_time()) {
4733     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4734   } else {
4735     // return user + sys since the cost is the same
4736     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4737   }
4738 }
4739 
4740 jlong os::thread_cpu_time(Thread* thread) {
4741   // consistent with what current_thread_cpu_time() returns
4742   if (os::Linux::supports_fast_thread_cpu_time()) {
4743     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4744   } else {
4745     return slow_thread_cpu_time(thread, true /* user + sys */);
4746   }
4747 }
4748 
4749 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4750   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4751     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4752   } else {
4753     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4754   }
4755 }
4756 
4757 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4758   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4759     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4760   } else {
4761     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4762   }
4763 }
4764 
4765 //
4766 //  -1 on error.
4767 //
4768 
4769 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4770   static bool proc_pid_cpu_avail = true;
4771   static bool proc_task_unchecked = true;
4772   static const char *proc_stat_path = "/proc/%d/stat";
4773   pid_t  tid = thread->osthread()->thread_id();
4774   int i;
4775   char *s;
4776   char stat[2048];
4777   int statlen;
4778   char proc_name[64];
4779   int count;
4780   long sys_time, user_time;
4781   char string[64];
4782   char cdummy;
4783   int idummy;
4784   long ldummy;
4785   FILE *fp;
4786 
4787   // We first try accessing /proc/<pid>/cpu since this is faster to
4788   // process.  If this file is not present (linux kernels 2.5 and above)
4789   // then we open /proc/<pid>/stat.
4790   if ( proc_pid_cpu_avail ) {
4791     sprintf(proc_name, "/proc/%d/cpu", tid);
4792     fp =  fopen(proc_name, "r");
4793     if ( fp != NULL ) {
4794       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4795       fclose(fp);
4796       if ( count != 3 ) return -1;
4797 
4798       if (user_sys_cpu_time) {
4799         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4800       } else {
4801         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4802       }
4803     }
4804     else proc_pid_cpu_avail = false;
4805   }
4806 
4807   // The /proc/<tid>/stat aggregates per-process usage on
4808   // new Linux kernels 2.6+ where NPTL is supported.
4809   // The /proc/self/task/<tid>/stat still has the per-thread usage.
4810   // See bug 6328462.
4811   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4812   // and possibly in some other cases, so we check its availability.
4813   if (proc_task_unchecked && os::Linux::is_NPTL()) {
4814     // This is executed only once
4815     proc_task_unchecked = false;
4816     fp = fopen("/proc/self/task", "r");
4817     if (fp != NULL) {
4818       proc_stat_path = "/proc/self/task/%d/stat";
4819       fclose(fp);
4820     }
4821   }
4822 
4823   sprintf(proc_name, proc_stat_path, tid);
4824   fp = fopen(proc_name, "r");
4825   if ( fp == NULL ) return -1;
4826   statlen = fread(stat, 1, 2047, fp);
4827   stat[statlen] = '\0';
4828   fclose(fp);
4829 
4830   // Skip pid and the command string. Note that we could be dealing with
4831   // weird command names, e.g. user could decide to rename java launcher
4832   // to "java 1.4.2 :)", then the stat file would look like
4833   //                1234 (java 1.4.2 :)) R ... ...
4834   // We don't really need to know the command string, just find the last
4835   // occurrence of ")" and then start parsing from there. See bug 4726580.
4836   s = strrchr(stat, ')');
4837   i = 0;
4838   if (s == NULL ) return -1;
4839 
4840   // Skip blank chars
4841   do s++; while (isspace(*s));
4842 
4843   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4844                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4845                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4846                  &user_time, &sys_time);
4847   if ( count != 13 ) return -1;
4848   if (user_sys_cpu_time) {
4849     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4850   } else {
4851     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4852   }
4853 }
4854 
4855 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4856   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4857   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4858   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4859   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4860 }
4861 
4862 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4863   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4864   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4865   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4866   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4867 }
4868 
4869 bool os::is_thread_cpu_time_supported() {
4870   return true;
4871 }
4872 
4873 // System loadavg support.  Returns -1 if load average cannot be obtained.
4874 // Linux doesn't yet have a (official) notion of processor sets,
4875 // so just return the system wide load average.
4876 int os::loadavg(double loadavg[], int nelem) {
4877   return ::getloadavg(loadavg, nelem);
4878 }
4879 
4880 void os::pause() {
4881   char filename[MAX_PATH];
4882   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4883     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4884   } else {
4885     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4886   }
4887 
4888   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4889   if (fd != -1) {
4890     struct stat buf;
4891     ::close(fd);
4892     while (::stat(filename, &buf) == 0) {
4893       (void)::poll(NULL, 0, 100);
4894     }
4895   } else {
4896     jio_fprintf(stderr,
4897       "Could not open pause file '%s', continuing immediately.\n", filename);
4898   }
4899 }
4900 
4901 
4902 // Refer to the comments in os_solaris.cpp park-unpark.
4903 //
4904 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4905 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4906 // For specifics regarding the bug see GLIBC BUGID 261237 :
4907 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4908 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4909 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4910 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4911 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4912 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4913 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4914 // of libpthread avoids the problem, but isn't practical.
4915 //
4916 // Possible remedies:
4917 //
4918 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4919 //      This is palliative and probabilistic, however.  If the thread is preempted
4920 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4921 //      than the minimum period may have passed, and the abstime may be stale (in the
4922 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
4923 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4924 //
4925 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4926 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
4927 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4928 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
4929 //      thread.
4930 //
4931 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4932 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
4933 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4934 //      This also works well.  In fact it avoids kernel-level scalability impediments
4935 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4936 //      timers in a graceful fashion.
4937 //
4938 // 4.   When the abstime value is in the past it appears that control returns
4939 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4940 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
4941 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
4942 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
4943 //      It may be possible to avoid reinitialization by checking the return
4944 //      value from pthread_cond_timedwait().  In addition to reinitializing the
4945 //      condvar we must establish the invariant that cond_signal() is only called
4946 //      within critical sections protected by the adjunct mutex.  This prevents
4947 //      cond_signal() from "seeing" a condvar that's in the midst of being
4948 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
4949 //      desirable signal-after-unlock optimization that avoids futile context switching.
4950 //
4951 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
4952 //      structure when a condvar is used or initialized.  cond_destroy()  would
4953 //      release the helper structure.  Our reinitialize-after-timedwait fix
4954 //      put excessive stress on malloc/free and locks protecting the c-heap.
4955 //
4956 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
4957 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4958 // and only enabling the work-around for vulnerable environments.
4959 
4960 // utility to compute the abstime argument to timedwait:
4961 // millis is the relative timeout time
4962 // abstime will be the absolute timeout time
4963 // TODO: replace compute_abstime() with unpackTime()
4964 
4965 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4966   if (millis < 0)  millis = 0;
4967   struct timeval now;
4968   int status = gettimeofday(&now, NULL);
4969   assert(status == 0, "gettimeofday");
4970   jlong seconds = millis / 1000;
4971   millis %= 1000;
4972   if (seconds > 50000000) { // see man cond_timedwait(3T)
4973     seconds = 50000000;
4974   }
4975   abstime->tv_sec = now.tv_sec  + seconds;
4976   long       usec = now.tv_usec + millis * 1000;
4977   if (usec >= 1000000) {
4978     abstime->tv_sec += 1;
4979     usec -= 1000000;
4980   }
4981   abstime->tv_nsec = usec * 1000;
4982   return abstime;
4983 }
4984 
4985 
4986 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4987 // Conceptually TryPark() should be equivalent to park(0).
4988 
4989 int os::PlatformEvent::TryPark() {
4990   for (;;) {
4991     const int v = _Event ;
4992     guarantee ((v == 0) || (v == 1), "invariant") ;
4993     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
4994   }
4995 }
4996 
4997 void os::PlatformEvent::park() {       // AKA "down()"
4998   // Invariant: Only the thread associated with the Event/PlatformEvent
4999   // may call park().
5000   // TODO: assert that _Assoc != NULL or _Assoc == Self
5001   int v ;
5002   for (;;) {
5003       v = _Event ;
5004       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5005   }
5006   guarantee (v >= 0, "invariant") ;
5007   if (v == 0) {
5008      // Do this the hard way by blocking ...
5009      int status = pthread_mutex_lock(_mutex);
5010      assert_status(status == 0, status, "mutex_lock");
5011      guarantee (_nParked == 0, "invariant") ;
5012      ++ _nParked ;
5013      while (_Event < 0) {
5014         status = pthread_cond_wait(_cond, _mutex);
5015         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5016         // Treat this the same as if the wait was interrupted
5017         if (status == ETIME) { status = EINTR; }
5018         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5019      }
5020      -- _nParked ;
5021 
5022     // In theory we could move the ST of 0 into _Event past the unlock(),
5023     // but then we'd need a MEMBAR after the ST.
5024     _Event = 0 ;
5025      status = pthread_mutex_unlock(_mutex);
5026      assert_status(status == 0, status, "mutex_unlock");
5027   }
5028   guarantee (_Event >= 0, "invariant") ;
5029 }
5030 
5031 int os::PlatformEvent::park(jlong millis) {
5032   guarantee (_nParked == 0, "invariant") ;
5033 
5034   int v ;
5035   for (;;) {
5036       v = _Event ;
5037       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5038   }
5039   guarantee (v >= 0, "invariant") ;
5040   if (v != 0) return OS_OK ;
5041 
5042   // We do this the hard way, by blocking the thread.
5043   // Consider enforcing a minimum timeout value.
5044   struct timespec abst;
5045   compute_abstime(&abst, millis);
5046 
5047   int ret = OS_TIMEOUT;
5048   int status = pthread_mutex_lock(_mutex);
5049   assert_status(status == 0, status, "mutex_lock");
5050   guarantee (_nParked == 0, "invariant") ;
5051   ++_nParked ;
5052 
5053   // Object.wait(timo) will return because of
5054   // (a) notification
5055   // (b) timeout
5056   // (c) thread.interrupt
5057   //
5058   // Thread.interrupt and object.notify{All} both call Event::set.
5059   // That is, we treat thread.interrupt as a special case of notification.
5060   // The underlying Solaris implementation, cond_timedwait, admits
5061   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5062   // JVM from making those visible to Java code.  As such, we must
5063   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5064   //
5065   // TODO: properly differentiate simultaneous notify+interrupt.
5066   // In that case, we should propagate the notify to another waiter.
5067 
5068   while (_Event < 0) {
5069     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5070     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5071       pthread_cond_destroy (_cond);
5072       pthread_cond_init (_cond, NULL) ;
5073     }
5074     assert_status(status == 0 || status == EINTR ||
5075                   status == ETIME || status == ETIMEDOUT,
5076                   status, "cond_timedwait");
5077     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5078     if (status == ETIME || status == ETIMEDOUT) break ;
5079     // We consume and ignore EINTR and spurious wakeups.
5080   }
5081   --_nParked ;
5082   if (_Event >= 0) {
5083      ret = OS_OK;
5084   }
5085   _Event = 0 ;
5086   status = pthread_mutex_unlock(_mutex);
5087   assert_status(status == 0, status, "mutex_unlock");
5088   assert (_nParked == 0, "invariant") ;
5089   return ret;
5090 }
5091 
5092 void os::PlatformEvent::unpark() {
5093   int v, AnyWaiters ;
5094   for (;;) {
5095       v = _Event ;
5096       if (v > 0) {
5097          // The LD of _Event could have reordered or be satisfied
5098          // by a read-aside from this processor's write buffer.
5099          // To avoid problems execute a barrier and then
5100          // ratify the value.
5101          OrderAccess::fence() ;
5102          if (_Event == v) return ;
5103          continue ;
5104       }
5105       if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5106   }
5107   if (v < 0) {
5108      // Wait for the thread associated with the event to vacate
5109      int status = pthread_mutex_lock(_mutex);
5110      assert_status(status == 0, status, "mutex_lock");
5111      AnyWaiters = _nParked ;
5112      assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5113      if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5114         AnyWaiters = 0 ;
5115         pthread_cond_signal (_cond);
5116      }
5117      status = pthread_mutex_unlock(_mutex);
5118      assert_status(status == 0, status, "mutex_unlock");
5119      if (AnyWaiters != 0) {
5120         status = pthread_cond_signal(_cond);
5121         assert_status(status == 0, status, "cond_signal");
5122      }
5123   }
5124 
5125   // Note that we signal() _after dropping the lock for "immortal" Events.
5126   // This is safe and avoids a common class of  futile wakeups.  In rare
5127   // circumstances this can cause a thread to return prematurely from
5128   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5129   // simply re-test the condition and re-park itself.
5130 }
5131 
5132 
5133 // JSR166
5134 // -------------------------------------------------------
5135 
5136 /*
5137  * The solaris and linux implementations of park/unpark are fairly
5138  * conservative for now, but can be improved. They currently use a
5139  * mutex/condvar pair, plus a a count.
5140  * Park decrements count if > 0, else does a condvar wait.  Unpark
5141  * sets count to 1 and signals condvar.  Only one thread ever waits
5142  * on the condvar. Contention seen when trying to park implies that someone
5143  * is unparking you, so don't wait. And spurious returns are fine, so there
5144  * is no need to track notifications.
5145  */
5146 
5147 #define MAX_SECS 100000000
5148 /*
5149  * This code is common to linux and solaris and will be moved to a
5150  * common place in dolphin.
5151  *
5152  * The passed in time value is either a relative time in nanoseconds
5153  * or an absolute time in milliseconds. Either way it has to be unpacked
5154  * into suitable seconds and nanoseconds components and stored in the
5155  * given timespec structure.
5156  * Given time is a 64-bit value and the time_t used in the timespec is only
5157  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5158  * overflow if times way in the future are given. Further on Solaris versions
5159  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5160  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5161  * As it will be 28 years before "now + 100000000" will overflow we can
5162  * ignore overflow and just impose a hard-limit on seconds using the value
5163  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5164  * years from "now".
5165  */
5166 
5167 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5168   assert (time > 0, "convertTime");
5169 
5170   struct timeval now;
5171   int status = gettimeofday(&now, NULL);
5172   assert(status == 0, "gettimeofday");
5173 
5174   time_t max_secs = now.tv_sec + MAX_SECS;
5175 
5176   if (isAbsolute) {
5177     jlong secs = time / 1000;
5178     if (secs > max_secs) {
5179       absTime->tv_sec = max_secs;
5180     }
5181     else {
5182       absTime->tv_sec = secs;
5183     }
5184     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5185   }
5186   else {
5187     jlong secs = time / NANOSECS_PER_SEC;
5188     if (secs >= MAX_SECS) {
5189       absTime->tv_sec = max_secs;
5190       absTime->tv_nsec = 0;
5191     }
5192     else {
5193       absTime->tv_sec = now.tv_sec + secs;
5194       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5195       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5196         absTime->tv_nsec -= NANOSECS_PER_SEC;
5197         ++absTime->tv_sec; // note: this must be <= max_secs
5198       }
5199     }
5200   }
5201   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5202   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5203   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5204   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5205 }
5206 
5207 void Parker::park(bool isAbsolute, jlong time) {
5208   // Optional fast-path check:
5209   // Return immediately if a permit is available.
5210   if (_counter > 0) {
5211       _counter = 0 ;
5212       OrderAccess::fence();
5213       return ;
5214   }
5215 
5216   Thread* thread = Thread::current();
5217   assert(thread->is_Java_thread(), "Must be JavaThread");
5218   JavaThread *jt = (JavaThread *)thread;
5219 
5220   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5221   // Check interrupt before trying to wait
5222   if (Thread::is_interrupted(thread, false)) {
5223     return;
5224   }
5225 
5226   // Next, demultiplex/decode time arguments
5227   timespec absTime;
5228   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5229     return;
5230   }
5231   if (time > 0) {
5232     unpackTime(&absTime, isAbsolute, time);
5233   }
5234 
5235 
5236   // Enter safepoint region
5237   // Beware of deadlocks such as 6317397.
5238   // The per-thread Parker:: mutex is a classic leaf-lock.
5239   // In particular a thread must never block on the Threads_lock while
5240   // holding the Parker:: mutex.  If safepoints are pending both the
5241   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5242   ThreadBlockInVM tbivm(jt);
5243 
5244   // Don't wait if cannot get lock since interference arises from
5245   // unblocking.  Also. check interrupt before trying wait
5246   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5247     return;
5248   }
5249 
5250   int status ;
5251   if (_counter > 0)  { // no wait needed
5252     _counter = 0;
5253     status = pthread_mutex_unlock(_mutex);
5254     assert (status == 0, "invariant") ;
5255     OrderAccess::fence();
5256     return;
5257   }
5258 
5259 #ifdef ASSERT
5260   // Don't catch signals while blocked; let the running threads have the signals.
5261   // (This allows a debugger to break into the running thread.)
5262   sigset_t oldsigs;
5263   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5264   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5265 #endif
5266 
5267   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5268   jt->set_suspend_equivalent();
5269   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5270 
5271   if (time == 0) {
5272     status = pthread_cond_wait (_cond, _mutex) ;
5273   } else {
5274     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5275     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5276       pthread_cond_destroy (_cond) ;
5277       pthread_cond_init    (_cond, NULL);
5278     }
5279   }
5280   assert_status(status == 0 || status == EINTR ||
5281                 status == ETIME || status == ETIMEDOUT,
5282                 status, "cond_timedwait");
5283 
5284 #ifdef ASSERT
5285   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5286 #endif
5287 
5288   _counter = 0 ;
5289   status = pthread_mutex_unlock(_mutex) ;
5290   assert_status(status == 0, status, "invariant") ;
5291   // If externally suspended while waiting, re-suspend
5292   if (jt->handle_special_suspend_equivalent_condition()) {
5293     jt->java_suspend_self();
5294   }
5295 
5296   OrderAccess::fence();
5297 }
5298 
5299 void Parker::unpark() {
5300   int s, status ;
5301   status = pthread_mutex_lock(_mutex);
5302   assert (status == 0, "invariant") ;
5303   s = _counter;
5304   _counter = 1;
5305   if (s < 1) {
5306      if (WorkAroundNPTLTimedWaitHang) {
5307         status = pthread_cond_signal (_cond) ;
5308         assert (status == 0, "invariant") ;
5309         status = pthread_mutex_unlock(_mutex);
5310         assert (status == 0, "invariant") ;
5311      } else {
5312         status = pthread_mutex_unlock(_mutex);
5313         assert (status == 0, "invariant") ;
5314         status = pthread_cond_signal (_cond) ;
5315         assert (status == 0, "invariant") ;
5316      }
5317   } else {
5318     pthread_mutex_unlock(_mutex);
5319     assert (status == 0, "invariant") ;
5320   }
5321 }
5322 
5323 
5324 extern char** environ;
5325 
5326 #ifndef __NR_fork
5327 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5328 #endif
5329 
5330 #ifndef __NR_execve
5331 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5332 #endif
5333 
5334 // Run the specified command in a separate process. Return its exit value,
5335 // or -1 on failure (e.g. can't fork a new process).
5336 // Unlike system(), this function can be called from signal handler. It
5337 // doesn't block SIGINT et al.
5338 int os::fork_and_exec(char* cmd) {
5339   const char * argv[4] = {"sh", "-c", cmd, NULL};
5340 
5341   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5342   // pthread_atfork handlers and reset pthread library. All we need is a
5343   // separate process to execve. Make a direct syscall to fork process.
5344   // On IA64 there's no fork syscall, we have to use fork() and hope for
5345   // the best...
5346   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5347               IA64_ONLY(fork();)
5348 
5349   if (pid < 0) {
5350     // fork failed
5351     return -1;
5352 
5353   } else if (pid == 0) {
5354     // child process
5355 
5356     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5357     // first to kill every thread on the thread list. Because this list is
5358     // not reset by fork() (see notes above), execve() will instead kill
5359     // every thread in the parent process. We know this is the only thread
5360     // in the new process, so make a system call directly.
5361     // IA64 should use normal execve() from glibc to match the glibc fork()
5362     // above.
5363     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5364     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5365 
5366     // execve failed
5367     _exit(-1);
5368 
5369   } else  {
5370     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5371     // care about the actual exit code, for now.
5372 
5373     int status;
5374 
5375     // Wait for the child process to exit.  This returns immediately if
5376     // the child has already exited. */
5377     while (waitpid(pid, &status, 0) < 0) {
5378         switch (errno) {
5379         case ECHILD: return 0;
5380         case EINTR: break;
5381         default: return -1;
5382         }
5383     }
5384 
5385     if (WIFEXITED(status)) {
5386        // The child exited normally; get its exit code.
5387        return WEXITSTATUS(status);
5388     } else if (WIFSIGNALED(status)) {
5389        // The child exited because of a signal
5390        // The best value to return is 0x80 + signal number,
5391        // because that is what all Unix shells do, and because
5392        // it allows callers to distinguish between process exit and
5393        // process death by signal.
5394        return 0x80 + WTERMSIG(status);
5395     } else {
5396        // Unknown exit code; pass it through
5397        return status;
5398     }
5399   }
5400 }
5401 
5402 // is_headless_jre()
5403 //
5404 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5405 // in order to report if we are running in a headless jre
5406 //
5407 // Since JDK8 xawt/libmawt.so was moved into the same directory
5408 // as libawt.so, and renamed libawt_xawt.so
5409 //
5410 bool os::is_headless_jre() {
5411     struct stat statbuf;
5412     char buf[MAXPATHLEN];
5413     char libmawtpath[MAXPATHLEN];
5414     const char *xawtstr  = "/xawt/libmawt.so";
5415     const char *new_xawtstr = "/libawt_xawt.so";
5416     char *p;
5417 
5418     // Get path to libjvm.so
5419     os::jvm_path(buf, sizeof(buf));
5420 
5421     // Get rid of libjvm.so
5422     p = strrchr(buf, '/');
5423     if (p == NULL) return false;
5424     else *p = '\0';
5425 
5426     // Get rid of client or server
5427     p = strrchr(buf, '/');
5428     if (p == NULL) return false;
5429     else *p = '\0';
5430 
5431     // check xawt/libmawt.so
5432     strcpy(libmawtpath, buf);
5433     strcat(libmawtpath, xawtstr);
5434     if (::stat(libmawtpath, &statbuf) == 0) return false;
5435 
5436     // check libawt_xawt.so
5437     strcpy(libmawtpath, buf);
5438     strcat(libmawtpath, new_xawtstr);
5439     if (::stat(libmawtpath, &statbuf) == 0) return false;
5440 
5441     return true;
5442 }
5443 
5444 // Get the default path to the core file
5445 // Returns the length of the string
5446 int os::get_core_path(char* buffer, size_t bufferSize) {
5447   const char* p = get_current_directory(buffer, bufferSize);
5448 
5449   if (p == NULL) {
5450     assert(p != NULL, "failed to get current directory");
5451     return 0;
5452   }
5453 
5454   return strlen(buffer);
5455 }
5456 
5457 #ifdef JAVASE_EMBEDDED
5458 //
5459 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5460 //
5461 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5462 
5463 // ctor
5464 //
5465 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5466   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5467   _fd = fd;
5468 
5469   if (os::create_thread(this, os::os_thread)) {
5470     _memnotify_thread = this;
5471     os::set_priority(this, NearMaxPriority);
5472     os::start_thread(this);
5473   }
5474 }
5475 
5476 // Where all the work gets done
5477 //
5478 void MemNotifyThread::run() {
5479   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5480 
5481   // Set up the select arguments
5482   fd_set rfds;
5483   if (_fd != -1) {
5484     FD_ZERO(&rfds);
5485     FD_SET(_fd, &rfds);
5486   }
5487 
5488   // Now wait for the mem_notify device to wake up
5489   while (1) {
5490     // Wait for the mem_notify device to signal us..
5491     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5492     if (rc == -1) {
5493       perror("select!\n");
5494       break;
5495     } else if (rc) {
5496       //ssize_t free_before = os::available_memory();
5497       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5498 
5499       // The kernel is telling us there is not much memory left...
5500       // try to do something about that
5501 
5502       // If we are not already in a GC, try one.
5503       if (!Universe::heap()->is_gc_active()) {
5504         Universe::heap()->collect(GCCause::_allocation_failure);
5505 
5506         //ssize_t free_after = os::available_memory();
5507         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5508         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5509       }
5510       // We might want to do something like the following if we find the GC's are not helping...
5511       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5512     }
5513   }
5514 }
5515 
5516 //
5517 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5518 //
5519 void MemNotifyThread::start() {
5520   int    fd;
5521   fd = open ("/dev/mem_notify", O_RDONLY, 0);
5522   if (fd < 0) {
5523       return;
5524   }
5525 
5526   if (memnotify_thread() == NULL) {
5527     new MemNotifyThread(fd);
5528   }
5529 }
5530 #endif // JAVASE_EMBEDDED
--- EOF ---