1 /*
   2  * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 # define __STDC_FORMAT_MACROS
  26 
  27 // no precompiled headers
  28 #include "classfile/classLoader.hpp"
  29 #include "classfile/systemDictionary.hpp"
  30 #include "classfile/vmSymbols.hpp"
  31 #include "code/icBuffer.hpp"
  32 #include "code/vtableStubs.hpp"
  33 #include "compiler/compileBroker.hpp"
  34 #include "interpreter/interpreter.hpp"
  35 #include "jvm_linux.h"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "mutex_linux.inline.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "prims/jniFastGetField.hpp"
  42 #include "prims/jvm.h"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/extendedPC.hpp"
  46 #include "runtime/globals.hpp"
  47 #include "runtime/interfaceSupport.hpp"
  48 #include "runtime/java.hpp"
  49 #include "runtime/javaCalls.hpp"
  50 #include "runtime/mutexLocker.hpp"
  51 #include "runtime/objectMonitor.hpp"
  52 #include "runtime/osThread.hpp"
  53 #include "runtime/perfMemory.hpp"
  54 #include "runtime/sharedRuntime.hpp"
  55 #include "runtime/statSampler.hpp"
  56 #include "runtime/stubRoutines.hpp"
  57 #include "runtime/threadCritical.hpp"
  58 #include "runtime/timer.hpp"
  59 #include "services/attachListener.hpp"
  60 #include "services/runtimeService.hpp"
  61 #include "thread_linux.inline.hpp"
  62 #include "utilities/decoder.hpp"
  63 #include "utilities/defaultStream.hpp"
  64 #include "utilities/events.hpp"
  65 #include "utilities/growableArray.hpp"
  66 #include "utilities/vmError.hpp"
  67 #ifdef TARGET_ARCH_x86
  68 # include "assembler_x86.inline.hpp"
  69 # include "nativeInst_x86.hpp"
  70 #endif
  71 #ifdef TARGET_ARCH_sparc
  72 # include "assembler_sparc.inline.hpp"
  73 # include "nativeInst_sparc.hpp"
  74 #endif
  75 #ifdef TARGET_ARCH_zero
  76 # include "assembler_zero.inline.hpp"
  77 # include "nativeInst_zero.hpp"
  78 #endif
  79 #ifdef TARGET_ARCH_arm
  80 # include "assembler_arm.inline.hpp"
  81 # include "nativeInst_arm.hpp"
  82 #endif
  83 #ifdef TARGET_ARCH_ppc
  84 # include "assembler_ppc.inline.hpp"
  85 # include "nativeInst_ppc.hpp"
  86 #endif
  87 #ifdef COMPILER1
  88 #include "c1/c1_Runtime1.hpp"
  89 #endif
  90 #ifdef COMPILER2
  91 #include "opto/runtime.hpp"
  92 #endif
  93 
  94 // put OS-includes here
  95 # include <sys/types.h>
  96 # include <sys/mman.h>
  97 # include <sys/stat.h>
  98 # include <sys/select.h>
  99 # include <pthread.h>
 100 # include <signal.h>
 101 # include <errno.h>
 102 # include <dlfcn.h>
 103 # include <stdio.h>
 104 # include <unistd.h>
 105 # include <sys/resource.h>
 106 # include <pthread.h>
 107 # include <sys/stat.h>
 108 # include <sys/time.h>
 109 # include <sys/times.h>
 110 # include <sys/utsname.h>
 111 # include <sys/socket.h>
 112 # include <sys/wait.h>
 113 # include <pwd.h>
 114 # include <poll.h>
 115 # include <semaphore.h>
 116 # include <fcntl.h>
 117 # include <string.h>
 118 # include <syscall.h>
 119 # include <sys/sysinfo.h>
 120 # include <gnu/libc-version.h>
 121 # include <sys/ipc.h>
 122 # include <sys/shm.h>
 123 # include <link.h>
 124 # include <stdint.h>
 125 # include <inttypes.h>
 126 # include <sys/ioctl.h>
 127 
 128 #define MAX_PATH    (2 * K)
 129 
 130 // for timer info max values which include all bits
 131 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 132 #define SEC_IN_NANOSECS  1000000000LL
 133 
 134 #define LARGEPAGES_BIT (1 << 6)
 135 ////////////////////////////////////////////////////////////////////////////////
 136 // global variables
 137 julong os::Linux::_physical_memory = 0;
 138 
 139 address   os::Linux::_initial_thread_stack_bottom = NULL;
 140 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 141 
 142 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 143 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 144 Mutex* os::Linux::_createThread_lock = NULL;
 145 pthread_t os::Linux::_main_thread;
 146 int os::Linux::_page_size = -1;
 147 bool os::Linux::_is_floating_stack = false;
 148 bool os::Linux::_is_NPTL = false;
 149 bool os::Linux::_supports_fast_thread_cpu_time = false;
 150 const char * os::Linux::_glibc_version = NULL;
 151 const char * os::Linux::_libpthread_version = NULL;
 152 
 153 static jlong initial_time_count=0;
 154 
 155 static int clock_tics_per_sec = 100;
 156 
 157 // For diagnostics to print a message once. see run_periodic_checks
 158 static sigset_t check_signal_done;
 159 static bool check_signals = true;;
 160 
 161 static pid_t _initial_pid = 0;
 162 
 163 /* Signal number used to suspend/resume a thread */
 164 
 165 /* do not use any signal number less than SIGSEGV, see 4355769 */
 166 static int SR_signum = SIGUSR2;
 167 sigset_t SR_sigset;
 168 
 169 /* Used to protect dlsym() calls */
 170 static pthread_mutex_t dl_mutex;
 171 
 172 #ifdef JAVASE_EMBEDDED
 173 class MemNotifyThread: public Thread {
 174   friend class VMStructs;
 175  public:
 176   virtual void run();
 177 
 178  private:
 179   static MemNotifyThread* _memnotify_thread;
 180   int _fd;
 181 
 182  public:
 183 
 184   // Constructor
 185   MemNotifyThread(int fd);
 186 
 187   // Tester
 188   bool is_memnotify_thread() const { return true; }
 189 
 190   // Printing
 191   char* name() const { return (char*)"Linux MemNotify Thread"; }
 192 
 193   // Returns the single instance of the MemNotifyThread
 194   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 195 
 196   // Create and start the single instance of MemNotifyThread
 197   static void start();
 198 };
 199 #endif // JAVASE_EMBEDDED
 200 
 201 // utility functions
 202 
 203 static int SR_initialize();
 204 static int SR_finalize();
 205 
 206 julong os::available_memory() {
 207   return Linux::available_memory();
 208 }
 209 
 210 julong os::Linux::available_memory() {
 211   // values in struct sysinfo are "unsigned long"
 212   struct sysinfo si;
 213   sysinfo(&si);
 214 
 215   return (julong)si.freeram * si.mem_unit;
 216 }
 217 
 218 julong os::physical_memory() {
 219   return Linux::physical_memory();
 220 }
 221 
 222 julong os::allocatable_physical_memory(julong size) {
 223 #ifdef _LP64
 224   return size;
 225 #else
 226   julong result = MIN2(size, (julong)3800*M);
 227    if (!is_allocatable(result)) {
 228      // See comments under solaris for alignment considerations
 229      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 230      result =  MIN2(size, reasonable_size);
 231    }
 232    return result;
 233 #endif // _LP64
 234 }
 235 
 236 ////////////////////////////////////////////////////////////////////////////////
 237 // environment support
 238 
 239 bool os::getenv(const char* name, char* buf, int len) {
 240   const char* val = ::getenv(name);
 241   if (val != NULL && strlen(val) < (size_t)len) {
 242     strcpy(buf, val);
 243     return true;
 244   }
 245   if (len > 0) buf[0] = 0;  // return a null string
 246   return false;
 247 }
 248 
 249 
 250 // Return true if user is running as root.
 251 
 252 bool os::have_special_privileges() {
 253   static bool init = false;
 254   static bool privileges = false;
 255   if (!init) {
 256     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 257     init = true;
 258   }
 259   return privileges;
 260 }
 261 
 262 
 263 #ifndef SYS_gettid
 264 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 265 #ifdef __ia64__
 266 #define SYS_gettid 1105
 267 #elif __i386__
 268 #define SYS_gettid 224
 269 #elif __amd64__
 270 #define SYS_gettid 186
 271 #elif __sparc__
 272 #define SYS_gettid 143
 273 #else
 274 #error define gettid for the arch
 275 #endif
 276 #endif
 277 
 278 // Cpu architecture string
 279 #if   defined(ZERO)
 280 static char cpu_arch[] = ZERO_LIBARCH;
 281 #elif defined(IA64)
 282 static char cpu_arch[] = "ia64";
 283 #elif defined(IA32)
 284 static char cpu_arch[] = "i386";
 285 #elif defined(AMD64)
 286 static char cpu_arch[] = "amd64";
 287 #elif defined(ARM)
 288 static char cpu_arch[] = "arm";
 289 #elif defined(PPC)
 290 static char cpu_arch[] = "ppc";
 291 #elif defined(SPARC)
 292 #  ifdef _LP64
 293 static char cpu_arch[] = "sparcv9";
 294 #  else
 295 static char cpu_arch[] = "sparc";
 296 #  endif
 297 #else
 298 #error Add appropriate cpu_arch setting
 299 #endif
 300 
 301 
 302 // pid_t gettid()
 303 //
 304 // Returns the kernel thread id of the currently running thread. Kernel
 305 // thread id is used to access /proc.
 306 //
 307 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 308 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 309 //
 310 pid_t os::Linux::gettid() {
 311   int rslt = syscall(SYS_gettid);
 312   if (rslt == -1) {
 313      // old kernel, no NPTL support
 314      return getpid();
 315   } else {
 316      return (pid_t)rslt;
 317   }
 318 }
 319 
 320 // Most versions of linux have a bug where the number of processors are
 321 // determined by looking at the /proc file system.  In a chroot environment,
 322 // the system call returns 1.  This causes the VM to act as if it is
 323 // a single processor and elide locking (see is_MP() call).
 324 static bool unsafe_chroot_detected = false;
 325 static const char *unstable_chroot_error = "/proc file system not found.\n"
 326                      "Java may be unstable running multithreaded in a chroot "
 327                      "environment on Linux when /proc filesystem is not mounted.";
 328 
 329 void os::Linux::initialize_system_info() {
 330   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 331   if (processor_count() == 1) {
 332     pid_t pid = os::Linux::gettid();
 333     char fname[32];
 334     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 335     FILE *fp = fopen(fname, "r");
 336     if (fp == NULL) {
 337       unsafe_chroot_detected = true;
 338     } else {
 339       fclose(fp);
 340     }
 341   }
 342   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 343   assert(processor_count() > 0, "linux error");
 344 }
 345 
 346 void os::init_system_properties_values() {
 347 //  char arch[12];
 348 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 349 
 350   // The next steps are taken in the product version:
 351   //
 352   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 353   // This library should be located at:
 354   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 355   //
 356   // If "/jre/lib/" appears at the right place in the path, then we
 357   // assume libjvm[_g].so is installed in a JDK and we use this path.
 358   //
 359   // Otherwise exit with message: "Could not create the Java virtual machine."
 360   //
 361   // The following extra steps are taken in the debugging version:
 362   //
 363   // If "/jre/lib/" does NOT appear at the right place in the path
 364   // instead of exit check for $JAVA_HOME environment variable.
 365   //
 366   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 367   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 368   // it looks like libjvm[_g].so is installed there
 369   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 370   //
 371   // Otherwise exit.
 372   //
 373   // Important note: if the location of libjvm.so changes this
 374   // code needs to be changed accordingly.
 375 
 376   // The next few definitions allow the code to be verbatim:
 377 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 378 #define getenv(n) ::getenv(n)
 379 
 380 /*
 381  * See ld(1):
 382  *      The linker uses the following search paths to locate required
 383  *      shared libraries:
 384  *        1: ...
 385  *        ...
 386  *        7: The default directories, normally /lib and /usr/lib.
 387  */
 388 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 389 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 390 #else
 391 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 392 #endif
 393 
 394 #define EXTENSIONS_DIR  "/lib/ext"
 395 #define ENDORSED_DIR    "/lib/endorsed"
 396 #define REG_DIR         "/usr/java/packages"
 397 
 398   {
 399     /* sysclasspath, java_home, dll_dir */
 400     {
 401         char *home_path;
 402         char *dll_path;
 403         char *pslash;
 404         char buf[MAXPATHLEN];
 405         os::jvm_path(buf, sizeof(buf));
 406 
 407         // Found the full path to libjvm.so.
 408         // Now cut the path to <java_home>/jre if we can.
 409         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 410         pslash = strrchr(buf, '/');
 411         if (pslash != NULL)
 412             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 413         dll_path = malloc(strlen(buf) + 1);
 414         if (dll_path == NULL)
 415             return;
 416         strcpy(dll_path, buf);
 417         Arguments::set_dll_dir(dll_path);
 418 
 419         if (pslash != NULL) {
 420             pslash = strrchr(buf, '/');
 421             if (pslash != NULL) {
 422                 *pslash = '\0';       /* get rid of /<arch> */
 423                 pslash = strrchr(buf, '/');
 424                 if (pslash != NULL)
 425                     *pslash = '\0';   /* get rid of /lib */
 426             }
 427         }
 428 
 429         home_path = malloc(strlen(buf) + 1);
 430         if (home_path == NULL)
 431             return;
 432         strcpy(home_path, buf);
 433         Arguments::set_java_home(home_path);
 434 
 435         if (!set_boot_path('/', ':'))
 436             return;
 437     }
 438 
 439     /*
 440      * Where to look for native libraries
 441      *
 442      * Note: Due to a legacy implementation, most of the library path
 443      * is set in the launcher.  This was to accomodate linking restrictions
 444      * on legacy Linux implementations (which are no longer supported).
 445      * Eventually, all the library path setting will be done here.
 446      *
 447      * However, to prevent the proliferation of improperly built native
 448      * libraries, the new path component /usr/java/packages is added here.
 449      * Eventually, all the library path setting will be done here.
 450      */
 451     {
 452         char *ld_library_path;
 453 
 454         /*
 455          * Construct the invariant part of ld_library_path. Note that the
 456          * space for the colon and the trailing null are provided by the
 457          * nulls included by the sizeof operator (so actually we allocate
 458          * a byte more than necessary).
 459          */
 460         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 461             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 462         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 463 
 464         /*
 465          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 466          * should always exist (until the legacy problem cited above is
 467          * addressed).
 468          */
 469         char *v = getenv("LD_LIBRARY_PATH");
 470         if (v != NULL) {
 471             char *t = ld_library_path;
 472             /* That's +1 for the colon and +1 for the trailing '\0' */
 473             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 474             sprintf(ld_library_path, "%s:%s", v, t);
 475         }
 476         Arguments::set_library_path(ld_library_path);
 477     }
 478 
 479     /*
 480      * Extensions directories.
 481      *
 482      * Note that the space for the colon and the trailing null are provided
 483      * by the nulls included by the sizeof operator (so actually one byte more
 484      * than necessary is allocated).
 485      */
 486     {
 487         char *buf = malloc(strlen(Arguments::get_java_home()) +
 488             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 489         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 490             Arguments::get_java_home());
 491         Arguments::set_ext_dirs(buf);
 492     }
 493 
 494     /* Endorsed standards default directory. */
 495     {
 496         char * buf;
 497         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 498         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 499         Arguments::set_endorsed_dirs(buf);
 500     }
 501   }
 502 
 503 #undef malloc
 504 #undef getenv
 505 #undef EXTENSIONS_DIR
 506 #undef ENDORSED_DIR
 507 
 508   // Done
 509   return;
 510 }
 511 
 512 ////////////////////////////////////////////////////////////////////////////////
 513 // breakpoint support
 514 
 515 void os::breakpoint() {
 516   BREAKPOINT;
 517 }
 518 
 519 extern "C" void breakpoint() {
 520   // use debugger to set breakpoint here
 521 }
 522 
 523 ////////////////////////////////////////////////////////////////////////////////
 524 // signal support
 525 
 526 debug_only(static bool signal_sets_initialized = false);
 527 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 528 
 529 bool os::Linux::is_sig_ignored(int sig) {
 530       struct sigaction oact;
 531       sigaction(sig, (struct sigaction*)NULL, &oact);
 532       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 533                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 534       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 535            return true;
 536       else
 537            return false;
 538 }
 539 
 540 void os::Linux::signal_sets_init() {
 541   // Should also have an assertion stating we are still single-threaded.
 542   assert(!signal_sets_initialized, "Already initialized");
 543   // Fill in signals that are necessarily unblocked for all threads in
 544   // the VM. Currently, we unblock the following signals:
 545   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 546   //                         by -Xrs (=ReduceSignalUsage));
 547   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 548   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 549   // the dispositions or masks wrt these signals.
 550   // Programs embedding the VM that want to use the above signals for their
 551   // own purposes must, at this time, use the "-Xrs" option to prevent
 552   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 553   // (See bug 4345157, and other related bugs).
 554   // In reality, though, unblocking these signals is really a nop, since
 555   // these signals are not blocked by default.
 556   sigemptyset(&unblocked_sigs);
 557   sigemptyset(&allowdebug_blocked_sigs);
 558   sigaddset(&unblocked_sigs, SIGILL);
 559   sigaddset(&unblocked_sigs, SIGSEGV);
 560   sigaddset(&unblocked_sigs, SIGBUS);
 561   sigaddset(&unblocked_sigs, SIGFPE);
 562   sigaddset(&unblocked_sigs, SR_signum);
 563 
 564   if (!ReduceSignalUsage) {
 565    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 566       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 567       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 568    }
 569    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 570       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 571       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 572    }
 573    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 574       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 575       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 576    }
 577   }
 578   // Fill in signals that are blocked by all but the VM thread.
 579   sigemptyset(&vm_sigs);
 580   if (!ReduceSignalUsage)
 581     sigaddset(&vm_sigs, BREAK_SIGNAL);
 582   debug_only(signal_sets_initialized = true);
 583 
 584 }
 585 
 586 // These are signals that are unblocked while a thread is running Java.
 587 // (For some reason, they get blocked by default.)
 588 sigset_t* os::Linux::unblocked_signals() {
 589   assert(signal_sets_initialized, "Not initialized");
 590   return &unblocked_sigs;
 591 }
 592 
 593 // These are the signals that are blocked while a (non-VM) thread is
 594 // running Java. Only the VM thread handles these signals.
 595 sigset_t* os::Linux::vm_signals() {
 596   assert(signal_sets_initialized, "Not initialized");
 597   return &vm_sigs;
 598 }
 599 
 600 // These are signals that are blocked during cond_wait to allow debugger in
 601 sigset_t* os::Linux::allowdebug_blocked_signals() {
 602   assert(signal_sets_initialized, "Not initialized");
 603   return &allowdebug_blocked_sigs;
 604 }
 605 
 606 void os::Linux::hotspot_sigmask(Thread* thread) {
 607 
 608   //Save caller's signal mask before setting VM signal mask
 609   sigset_t caller_sigmask;
 610   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 611 
 612   OSThread* osthread = thread->osthread();
 613   osthread->set_caller_sigmask(caller_sigmask);
 614 
 615   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 616 
 617   if (!ReduceSignalUsage) {
 618     if (thread->is_VM_thread()) {
 619       // Only the VM thread handles BREAK_SIGNAL ...
 620       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 621     } else {
 622       // ... all other threads block BREAK_SIGNAL
 623       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 624     }
 625   }
 626 }
 627 
 628 //////////////////////////////////////////////////////////////////////////////
 629 // detecting pthread library
 630 
 631 void os::Linux::libpthread_init() {
 632   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 633   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 634   // generic name for earlier versions.
 635   // Define macros here so we can build HotSpot on old systems.
 636 # ifndef _CS_GNU_LIBC_VERSION
 637 # define _CS_GNU_LIBC_VERSION 2
 638 # endif
 639 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 640 # define _CS_GNU_LIBPTHREAD_VERSION 3
 641 # endif
 642 
 643   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 644   if (n > 0) {
 645      char *str = (char *)malloc(n);
 646      confstr(_CS_GNU_LIBC_VERSION, str, n);
 647      os::Linux::set_glibc_version(str);
 648   } else {
 649      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 650      static char _gnu_libc_version[32];
 651      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 652               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 653      os::Linux::set_glibc_version(_gnu_libc_version);
 654   }
 655 
 656   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 657   if (n > 0) {
 658      char *str = (char *)malloc(n);
 659      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 660      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 661      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 662      // is the case. LinuxThreads has a hard limit on max number of threads.
 663      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 664      // On the other hand, NPTL does not have such a limit, sysconf()
 665      // will return -1 and errno is not changed. Check if it is really NPTL.
 666      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 667          strstr(str, "NPTL") &&
 668          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 669        free(str);
 670        os::Linux::set_libpthread_version("linuxthreads");
 671      } else {
 672        os::Linux::set_libpthread_version(str);
 673      }
 674   } else {
 675     // glibc before 2.3.2 only has LinuxThreads.
 676     os::Linux::set_libpthread_version("linuxthreads");
 677   }
 678 
 679   if (strstr(libpthread_version(), "NPTL")) {
 680      os::Linux::set_is_NPTL();
 681   } else {
 682      os::Linux::set_is_LinuxThreads();
 683   }
 684 
 685   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 686   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 687   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 688      os::Linux::set_is_floating_stack();
 689   }
 690 }
 691 
 692 /////////////////////////////////////////////////////////////////////////////
 693 // thread stack
 694 
 695 // Force Linux kernel to expand current thread stack. If "bottom" is close
 696 // to the stack guard, caller should block all signals.
 697 //
 698 // MAP_GROWSDOWN:
 699 //   A special mmap() flag that is used to implement thread stacks. It tells
 700 //   kernel that the memory region should extend downwards when needed. This
 701 //   allows early versions of LinuxThreads to only mmap the first few pages
 702 //   when creating a new thread. Linux kernel will automatically expand thread
 703 //   stack as needed (on page faults).
 704 //
 705 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 706 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 707 //   region, it's hard to tell if the fault is due to a legitimate stack
 708 //   access or because of reading/writing non-exist memory (e.g. buffer
 709 //   overrun). As a rule, if the fault happens below current stack pointer,
 710 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 711 //   application (see Linux kernel fault.c).
 712 //
 713 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 714 //   stack overflow detection.
 715 //
 716 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 717 //   not use this flag. However, the stack of initial thread is not created
 718 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 719 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 720 //   and then attach the thread to JVM.
 721 //
 722 // To get around the problem and allow stack banging on Linux, we need to
 723 // manually expand thread stack after receiving the SIGSEGV.
 724 //
 725 // There are two ways to expand thread stack to address "bottom", we used
 726 // both of them in JVM before 1.5:
 727 //   1. adjust stack pointer first so that it is below "bottom", and then
 728 //      touch "bottom"
 729 //   2. mmap() the page in question
 730 //
 731 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 732 // if current sp is already near the lower end of page 101, and we need to
 733 // call mmap() to map page 100, it is possible that part of the mmap() frame
 734 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 735 // That will destroy the mmap() frame and cause VM to crash.
 736 //
 737 // The following code works by adjusting sp first, then accessing the "bottom"
 738 // page to force a page fault. Linux kernel will then automatically expand the
 739 // stack mapping.
 740 //
 741 // _expand_stack_to() assumes its frame size is less than page size, which
 742 // should always be true if the function is not inlined.
 743 
 744 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 745 #define NOINLINE
 746 #else
 747 #define NOINLINE __attribute__ ((noinline))
 748 #endif
 749 
 750 static void _expand_stack_to(address bottom) NOINLINE;
 751 
 752 static void _expand_stack_to(address bottom) {
 753   address sp;
 754   size_t size;
 755   volatile char *p;
 756 
 757   // Adjust bottom to point to the largest address within the same page, it
 758   // gives us a one-page buffer if alloca() allocates slightly more memory.
 759   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 760   bottom += os::Linux::page_size() - 1;
 761 
 762   // sp might be slightly above current stack pointer; if that's the case, we
 763   // will alloca() a little more space than necessary, which is OK. Don't use
 764   // os::current_stack_pointer(), as its result can be slightly below current
 765   // stack pointer, causing us to not alloca enough to reach "bottom".
 766   sp = (address)&sp;
 767 
 768   if (sp > bottom) {
 769     size = sp - bottom;
 770     p = (volatile char *)alloca(size);
 771     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 772     p[0] = '\0';
 773   }
 774 }
 775 
 776 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 777   assert(t!=NULL, "just checking");
 778   assert(t->osthread()->expanding_stack(), "expand should be set");
 779   assert(t->stack_base() != NULL, "stack_base was not initialized");
 780 
 781   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 782     sigset_t mask_all, old_sigset;
 783     sigfillset(&mask_all);
 784     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 785     _expand_stack_to(addr);
 786     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 787     return true;
 788   }
 789   return false;
 790 }
 791 
 792 //////////////////////////////////////////////////////////////////////////////
 793 // create new thread
 794 
 795 static address highest_vm_reserved_address();
 796 
 797 // check if it's safe to start a new thread
 798 static bool _thread_safety_check(Thread* thread) {
 799   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 800     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 801     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 802     //   allocated (MAP_FIXED) from high address space. Every thread stack
 803     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 804     //   it to other values if they rebuild LinuxThreads).
 805     //
 806     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 807     // the memory region has already been mmap'ed. That means if we have too
 808     // many threads and/or very large heap, eventually thread stack will
 809     // collide with heap.
 810     //
 811     // Here we try to prevent heap/stack collision by comparing current
 812     // stack bottom with the highest address that has been mmap'ed by JVM
 813     // plus a safety margin for memory maps created by native code.
 814     //
 815     // This feature can be disabled by setting ThreadSafetyMargin to 0
 816     //
 817     if (ThreadSafetyMargin > 0) {
 818       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 819 
 820       // not safe if our stack extends below the safety margin
 821       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 822     } else {
 823       return true;
 824     }
 825   } else {
 826     // Floating stack LinuxThreads or NPTL:
 827     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 828     //   there's not enough space left, pthread_create() will fail. If we come
 829     //   here, that means enough space has been reserved for stack.
 830     return true;
 831   }
 832 }
 833 
 834 // Thread start routine for all newly created threads
 835 static void *java_start(Thread *thread) {
 836   // Try to randomize the cache line index of hot stack frames.
 837   // This helps when threads of the same stack traces evict each other's
 838   // cache lines. The threads can be either from the same JVM instance, or
 839   // from different JVM instances. The benefit is especially true for
 840   // processors with hyperthreading technology.
 841   static int counter = 0;
 842   int pid = os::current_process_id();
 843   alloca(((pid ^ counter++) & 7) * 128);
 844 
 845   ThreadLocalStorage::set_thread(thread);
 846 
 847   OSThread* osthread = thread->osthread();
 848   Monitor* sync = osthread->startThread_lock();
 849 
 850   // non floating stack LinuxThreads needs extra check, see above
 851   if (!_thread_safety_check(thread)) {
 852     // notify parent thread
 853     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 854     osthread->set_state(ZOMBIE);
 855     sync->notify_all();
 856     return NULL;
 857   }
 858 
 859   // thread_id is kernel thread id (similar to Solaris LWP id)
 860   osthread->set_thread_id(os::Linux::gettid());
 861 
 862   if (UseNUMA) {
 863     int lgrp_id = os::numa_get_group_id();
 864     if (lgrp_id != -1) {
 865       thread->set_lgrp_id(lgrp_id);
 866     }
 867   }
 868   // initialize signal mask for this thread
 869   os::Linux::hotspot_sigmask(thread);
 870 
 871   // initialize floating point control register
 872   os::Linux::init_thread_fpu_state();
 873 
 874   // handshaking with parent thread
 875   {
 876     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 877 
 878     // notify parent thread
 879     osthread->set_state(INITIALIZED);
 880     sync->notify_all();
 881 
 882     // wait until os::start_thread()
 883     while (osthread->get_state() == INITIALIZED) {
 884       sync->wait(Mutex::_no_safepoint_check_flag);
 885     }
 886   }
 887 
 888   // call one more level start routine
 889   thread->run();
 890 
 891   return 0;
 892 }
 893 
 894 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 895   assert(thread->osthread() == NULL, "caller responsible");
 896 
 897   // Allocate the OSThread object
 898   OSThread* osthread = new OSThread(NULL, NULL);
 899   if (osthread == NULL) {
 900     return false;
 901   }
 902 
 903   // set the correct thread state
 904   osthread->set_thread_type(thr_type);
 905 
 906   // Initial state is ALLOCATED but not INITIALIZED
 907   osthread->set_state(ALLOCATED);
 908 
 909   thread->set_osthread(osthread);
 910 
 911   // init thread attributes
 912   pthread_attr_t attr;
 913   pthread_attr_init(&attr);
 914   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 915 
 916   // stack size
 917   if (os::Linux::supports_variable_stack_size()) {
 918     // calculate stack size if it's not specified by caller
 919     if (stack_size == 0) {
 920       stack_size = os::Linux::default_stack_size(thr_type);
 921 
 922       switch (thr_type) {
 923       case os::java_thread:
 924         // Java threads use ThreadStackSize which default value can be
 925         // changed with the flag -Xss
 926         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 927         stack_size = JavaThread::stack_size_at_create();
 928         break;
 929       case os::compiler_thread:
 930         if (CompilerThreadStackSize > 0) {
 931           stack_size = (size_t)(CompilerThreadStackSize * K);
 932           break;
 933         } // else fall through:
 934           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 935       case os::vm_thread:
 936       case os::pgc_thread:
 937       case os::cgc_thread:
 938       case os::watcher_thread:
 939         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 940         break;
 941       }
 942     }
 943 
 944     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 945     pthread_attr_setstacksize(&attr, stack_size);
 946   } else {
 947     // let pthread_create() pick the default value.
 948   }
 949 
 950   // glibc guard page
 951   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 952 
 953   ThreadState state;
 954 
 955   {
 956     // Serialize thread creation if we are running with fixed stack LinuxThreads
 957     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 958     if (lock) {
 959       os::Linux::createThread_lock()->lock_without_safepoint_check();
 960     }
 961 
 962     pthread_t tid;
 963     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 964 
 965     pthread_attr_destroy(&attr);
 966 
 967     if (ret != 0) {
 968       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 969         perror("pthread_create()");
 970       }
 971       // Need to clean up stuff we've allocated so far
 972       thread->set_osthread(NULL);
 973       delete osthread;
 974       if (lock) os::Linux::createThread_lock()->unlock();
 975       return false;
 976     }
 977 
 978     // Store pthread info into the OSThread
 979     osthread->set_pthread_id(tid);
 980 
 981     // Wait until child thread is either initialized or aborted
 982     {
 983       Monitor* sync_with_child = osthread->startThread_lock();
 984       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 985       while ((state = osthread->get_state()) == ALLOCATED) {
 986         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 987       }
 988     }
 989 
 990     if (lock) {
 991       os::Linux::createThread_lock()->unlock();
 992     }
 993   }
 994 
 995   // Aborted due to thread limit being reached
 996   if (state == ZOMBIE) {
 997       thread->set_osthread(NULL);
 998       delete osthread;
 999       return false;
1000   }
1001 
1002   // The thread is returned suspended (in state INITIALIZED),
1003   // and is started higher up in the call chain
1004   assert(state == INITIALIZED, "race condition");
1005   return true;
1006 }
1007 
1008 /////////////////////////////////////////////////////////////////////////////
1009 // attach existing thread
1010 
1011 // bootstrap the main thread
1012 bool os::create_main_thread(JavaThread* thread) {
1013   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
1014   return create_attached_thread(thread);
1015 }
1016 
1017 bool os::create_attached_thread(JavaThread* thread) {
1018 #ifdef ASSERT
1019     thread->verify_not_published();
1020 #endif
1021 
1022   // Allocate the OSThread object
1023   OSThread* osthread = new OSThread(NULL, NULL);
1024 
1025   if (osthread == NULL) {
1026     return false;
1027   }
1028 
1029   // Store pthread info into the OSThread
1030   osthread->set_thread_id(os::Linux::gettid());
1031   osthread->set_pthread_id(::pthread_self());
1032 
1033   // initialize floating point control register
1034   os::Linux::init_thread_fpu_state();
1035 
1036   // Initial thread state is RUNNABLE
1037   osthread->set_state(RUNNABLE);
1038 
1039   thread->set_osthread(osthread);
1040 
1041   if (UseNUMA) {
1042     int lgrp_id = os::numa_get_group_id();
1043     if (lgrp_id != -1) {
1044       thread->set_lgrp_id(lgrp_id);
1045     }
1046   }
1047 
1048   if (os::Linux::is_initial_thread()) {
1049     // If current thread is initial thread, its stack is mapped on demand,
1050     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1051     // the entire stack region to avoid SEGV in stack banging.
1052     // It is also useful to get around the heap-stack-gap problem on SuSE
1053     // kernel (see 4821821 for details). We first expand stack to the top
1054     // of yellow zone, then enable stack yellow zone (order is significant,
1055     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1056     // is no gap between the last two virtual memory regions.
1057 
1058     JavaThread *jt = (JavaThread *)thread;
1059     address addr = jt->stack_yellow_zone_base();
1060     assert(addr != NULL, "initialization problem?");
1061     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1062 
1063     osthread->set_expanding_stack();
1064     os::Linux::manually_expand_stack(jt, addr);
1065     osthread->clear_expanding_stack();
1066   }
1067 
1068   // initialize signal mask for this thread
1069   // and save the caller's signal mask
1070   os::Linux::hotspot_sigmask(thread);
1071 
1072   return true;
1073 }
1074 
1075 void os::pd_start_thread(Thread* thread) {
1076   OSThread * osthread = thread->osthread();
1077   assert(osthread->get_state() != INITIALIZED, "just checking");
1078   Monitor* sync_with_child = osthread->startThread_lock();
1079   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1080   sync_with_child->notify();
1081 }
1082 
1083 // Free Linux resources related to the OSThread
1084 void os::free_thread(OSThread* osthread) {
1085   assert(osthread != NULL, "osthread not set");
1086 
1087   if (Thread::current()->osthread() == osthread) {
1088     // Restore caller's signal mask
1089     sigset_t sigmask = osthread->caller_sigmask();
1090     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1091    }
1092 
1093   delete osthread;
1094 }
1095 
1096 //////////////////////////////////////////////////////////////////////////////
1097 // thread local storage
1098 
1099 int os::allocate_thread_local_storage() {
1100   pthread_key_t key;
1101   int rslt = pthread_key_create(&key, NULL);
1102   assert(rslt == 0, "cannot allocate thread local storage");
1103   return (int)key;
1104 }
1105 
1106 // Note: This is currently not used by VM, as we don't destroy TLS key
1107 // on VM exit.
1108 void os::free_thread_local_storage(int index) {
1109   int rslt = pthread_key_delete((pthread_key_t)index);
1110   assert(rslt == 0, "invalid index");
1111 }
1112 
1113 void os::thread_local_storage_at_put(int index, void* value) {
1114   int rslt = pthread_setspecific((pthread_key_t)index, value);
1115   assert(rslt == 0, "pthread_setspecific failed");
1116 }
1117 
1118 extern "C" Thread* get_thread() {
1119   return ThreadLocalStorage::thread();
1120 }
1121 
1122 //////////////////////////////////////////////////////////////////////////////
1123 // initial thread
1124 
1125 // Check if current thread is the initial thread, similar to Solaris thr_main.
1126 bool os::Linux::is_initial_thread(void) {
1127   char dummy;
1128   // If called before init complete, thread stack bottom will be null.
1129   // Can be called if fatal error occurs before initialization.
1130   if (initial_thread_stack_bottom() == NULL) return false;
1131   assert(initial_thread_stack_bottom() != NULL &&
1132          initial_thread_stack_size()   != 0,
1133          "os::init did not locate initial thread's stack region");
1134   if ((address)&dummy >= initial_thread_stack_bottom() &&
1135       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1136        return true;
1137   else return false;
1138 }
1139 
1140 // Find the virtual memory area that contains addr
1141 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1142   FILE *fp = fopen("/proc/self/maps", "r");
1143   if (fp) {
1144     address low, high;
1145     while (!feof(fp)) {
1146       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1147         if (low <= addr && addr < high) {
1148            if (vma_low)  *vma_low  = low;
1149            if (vma_high) *vma_high = high;
1150            fclose (fp);
1151            return true;
1152         }
1153       }
1154       for (;;) {
1155         int ch = fgetc(fp);
1156         if (ch == EOF || ch == (int)'\n') break;
1157       }
1158     }
1159     fclose(fp);
1160   }
1161   return false;
1162 }
1163 
1164 // Locate initial thread stack. This special handling of initial thread stack
1165 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1166 // bogus value for initial thread.
1167 void os::Linux::capture_initial_stack(size_t max_size) {
1168   // stack size is the easy part, get it from RLIMIT_STACK
1169   size_t stack_size;
1170   struct rlimit rlim;
1171   getrlimit(RLIMIT_STACK, &rlim);
1172   stack_size = rlim.rlim_cur;
1173 
1174   // 6308388: a bug in ld.so will relocate its own .data section to the
1175   //   lower end of primordial stack; reduce ulimit -s value a little bit
1176   //   so we won't install guard page on ld.so's data section.
1177   stack_size -= 2 * page_size();
1178 
1179   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1180   //   7.1, in both cases we will get 2G in return value.
1181   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1182   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1183   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1184   //   in case other parts in glibc still assumes 2M max stack size.
1185   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1186 #ifndef IA64
1187   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1188 #else
1189   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1190   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1191 #endif
1192 
1193   // Try to figure out where the stack base (top) is. This is harder.
1194   //
1195   // When an application is started, glibc saves the initial stack pointer in
1196   // a global variable "__libc_stack_end", which is then used by system
1197   // libraries. __libc_stack_end should be pretty close to stack top. The
1198   // variable is available since the very early days. However, because it is
1199   // a private interface, it could disappear in the future.
1200   //
1201   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1202   // to __libc_stack_end, it is very close to stack top, but isn't the real
1203   // stack top. Note that /proc may not exist if VM is running as a chroot
1204   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1205   // /proc/<pid>/stat could change in the future (though unlikely).
1206   //
1207   // We try __libc_stack_end first. If that doesn't work, look for
1208   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1209   // as a hint, which should work well in most cases.
1210 
1211   uintptr_t stack_start;
1212 
1213   // try __libc_stack_end first
1214   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1215   if (p && *p) {
1216     stack_start = *p;
1217   } else {
1218     // see if we can get the start_stack field from /proc/self/stat
1219     FILE *fp;
1220     int pid;
1221     char state;
1222     int ppid;
1223     int pgrp;
1224     int session;
1225     int nr;
1226     int tpgrp;
1227     unsigned long flags;
1228     unsigned long minflt;
1229     unsigned long cminflt;
1230     unsigned long majflt;
1231     unsigned long cmajflt;
1232     unsigned long utime;
1233     unsigned long stime;
1234     long cutime;
1235     long cstime;
1236     long prio;
1237     long nice;
1238     long junk;
1239     long it_real;
1240     uintptr_t start;
1241     uintptr_t vsize;
1242     intptr_t rss;
1243     uintptr_t rsslim;
1244     uintptr_t scodes;
1245     uintptr_t ecode;
1246     int i;
1247 
1248     // Figure what the primordial thread stack base is. Code is inspired
1249     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1250     // followed by command name surrounded by parentheses, state, etc.
1251     char stat[2048];
1252     int statlen;
1253 
1254     fp = fopen("/proc/self/stat", "r");
1255     if (fp) {
1256       statlen = fread(stat, 1, 2047, fp);
1257       stat[statlen] = '\0';
1258       fclose(fp);
1259 
1260       // Skip pid and the command string. Note that we could be dealing with
1261       // weird command names, e.g. user could decide to rename java launcher
1262       // to "java 1.4.2 :)", then the stat file would look like
1263       //                1234 (java 1.4.2 :)) R ... ...
1264       // We don't really need to know the command string, just find the last
1265       // occurrence of ")" and then start parsing from there. See bug 4726580.
1266       char * s = strrchr(stat, ')');
1267 
1268       i = 0;
1269       if (s) {
1270         // Skip blank chars
1271         do s++; while (isspace(*s));
1272 
1273 #define _UFM UINTX_FORMAT
1274 #define _DFM INTX_FORMAT
1275 
1276         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1277         /*              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 */
1278         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,
1279              &state,          /* 3  %c  */
1280              &ppid,           /* 4  %d  */
1281              &pgrp,           /* 5  %d  */
1282              &session,        /* 6  %d  */
1283              &nr,             /* 7  %d  */
1284              &tpgrp,          /* 8  %d  */
1285              &flags,          /* 9  %lu  */
1286              &minflt,         /* 10 %lu  */
1287              &cminflt,        /* 11 %lu  */
1288              &majflt,         /* 12 %lu  */
1289              &cmajflt,        /* 13 %lu  */
1290              &utime,          /* 14 %lu  */
1291              &stime,          /* 15 %lu  */
1292              &cutime,         /* 16 %ld  */
1293              &cstime,         /* 17 %ld  */
1294              &prio,           /* 18 %ld  */
1295              &nice,           /* 19 %ld  */
1296              &junk,           /* 20 %ld  */
1297              &it_real,        /* 21 %ld  */
1298              &start,          /* 22 UINTX_FORMAT */
1299              &vsize,          /* 23 UINTX_FORMAT */
1300              &rss,            /* 24 INTX_FORMAT  */
1301              &rsslim,         /* 25 UINTX_FORMAT */
1302              &scodes,         /* 26 UINTX_FORMAT */
1303              &ecode,          /* 27 UINTX_FORMAT */
1304              &stack_start);   /* 28 UINTX_FORMAT */
1305       }
1306 
1307 #undef _UFM
1308 #undef _DFM
1309 
1310       if (i != 28 - 2) {
1311          assert(false, "Bad conversion from /proc/self/stat");
1312          // product mode - assume we are the initial thread, good luck in the
1313          // embedded case.
1314          warning("Can't detect initial thread stack location - bad conversion");
1315          stack_start = (uintptr_t) &rlim;
1316       }
1317     } else {
1318       // For some reason we can't open /proc/self/stat (for example, running on
1319       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1320       // most cases, so don't abort:
1321       warning("Can't detect initial thread stack location - no /proc/self/stat");
1322       stack_start = (uintptr_t) &rlim;
1323     }
1324   }
1325 
1326   // Now we have a pointer (stack_start) very close to the stack top, the
1327   // next thing to do is to figure out the exact location of stack top. We
1328   // can find out the virtual memory area that contains stack_start by
1329   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1330   // and its upper limit is the real stack top. (again, this would fail if
1331   // running inside chroot, because /proc may not exist.)
1332 
1333   uintptr_t stack_top;
1334   address low, high;
1335   if (find_vma((address)stack_start, &low, &high)) {
1336     // success, "high" is the true stack top. (ignore "low", because initial
1337     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1338     stack_top = (uintptr_t)high;
1339   } else {
1340     // failed, likely because /proc/self/maps does not exist
1341     warning("Can't detect initial thread stack location - find_vma failed");
1342     // best effort: stack_start is normally within a few pages below the real
1343     // stack top, use it as stack top, and reduce stack size so we won't put
1344     // guard page outside stack.
1345     stack_top = stack_start;
1346     stack_size -= 16 * page_size();
1347   }
1348 
1349   // stack_top could be partially down the page so align it
1350   stack_top = align_size_up(stack_top, page_size());
1351 
1352   if (max_size && stack_size > max_size) {
1353      _initial_thread_stack_size = max_size;
1354   } else {
1355      _initial_thread_stack_size = stack_size;
1356   }
1357 
1358   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1359   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1360 }
1361 
1362 ////////////////////////////////////////////////////////////////////////////////
1363 // time support
1364 
1365 // Time since start-up in seconds to a fine granularity.
1366 // Used by VMSelfDestructTimer and the MemProfiler.
1367 double os::elapsedTime() {
1368 
1369   return (double)(os::elapsed_counter()) * 0.000001;
1370 }
1371 
1372 jlong os::elapsed_counter() {
1373   timeval time;
1374   int status = gettimeofday(&time, NULL);
1375   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1376 }
1377 
1378 jlong os::elapsed_frequency() {
1379   return (1000 * 1000);
1380 }
1381 
1382 // For now, we say that linux does not support vtime.  I have no idea
1383 // whether it can actually be made to (DLD, 9/13/05).
1384 
1385 bool os::supports_vtime() { return false; }
1386 bool os::enable_vtime()   { return false; }
1387 bool os::vtime_enabled()  { return false; }
1388 double os::elapsedVTime() {
1389   // better than nothing, but not much
1390   return elapsedTime();
1391 }
1392 
1393 jlong os::javaTimeMillis() {
1394   timeval time;
1395   int status = gettimeofday(&time, NULL);
1396   assert(status != -1, "linux error");
1397   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1398 }
1399 
1400 #ifndef CLOCK_MONOTONIC
1401 #define CLOCK_MONOTONIC (1)
1402 #endif
1403 
1404 void os::Linux::clock_init() {
1405   // we do dlopen's in this particular order due to bug in linux
1406   // dynamical loader (see 6348968) leading to crash on exit
1407   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1408   if (handle == NULL) {
1409     handle = dlopen("librt.so", RTLD_LAZY);
1410   }
1411 
1412   if (handle) {
1413     int (*clock_getres_func)(clockid_t, struct timespec*) =
1414            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1415     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1416            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1417     if (clock_getres_func && clock_gettime_func) {
1418       // See if monotonic clock is supported by the kernel. Note that some
1419       // early implementations simply return kernel jiffies (updated every
1420       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1421       // for nano time (though the monotonic property is still nice to have).
1422       // It's fixed in newer kernels, however clock_getres() still returns
1423       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1424       // resolution for now. Hopefully as people move to new kernels, this
1425       // won't be a problem.
1426       struct timespec res;
1427       struct timespec tp;
1428       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1429           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1430         // yes, monotonic clock is supported
1431         _clock_gettime = clock_gettime_func;
1432       } else {
1433         // close librt if there is no monotonic clock
1434         dlclose(handle);
1435       }
1436     }
1437   }
1438 }
1439 
1440 #ifndef SYS_clock_getres
1441 
1442 #if defined(IA32) || defined(AMD64)
1443 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1444 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1445 #else
1446 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1447 #define sys_clock_getres(x,y)  -1
1448 #endif
1449 
1450 #else
1451 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1452 #endif
1453 
1454 void os::Linux::fast_thread_clock_init() {
1455   if (!UseLinuxPosixThreadCPUClocks) {
1456     return;
1457   }
1458   clockid_t clockid;
1459   struct timespec tp;
1460   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1461       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1462 
1463   // Switch to using fast clocks for thread cpu time if
1464   // the sys_clock_getres() returns 0 error code.
1465   // Note, that some kernels may support the current thread
1466   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1467   // returned by the pthread_getcpuclockid().
1468   // If the fast Posix clocks are supported then the sys_clock_getres()
1469   // must return at least tp.tv_sec == 0 which means a resolution
1470   // better than 1 sec. This is extra check for reliability.
1471 
1472   if(pthread_getcpuclockid_func &&
1473      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1474      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1475 
1476     _supports_fast_thread_cpu_time = true;
1477     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1478   }
1479 }
1480 
1481 jlong os::javaTimeNanos() {
1482   if (Linux::supports_monotonic_clock()) {
1483     struct timespec tp;
1484     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1485     assert(status == 0, "gettime error");
1486     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1487     return result;
1488   } else {
1489     timeval time;
1490     int status = gettimeofday(&time, NULL);
1491     assert(status != -1, "linux error");
1492     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1493     return 1000 * usecs;
1494   }
1495 }
1496 
1497 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1498   if (Linux::supports_monotonic_clock()) {
1499     info_ptr->max_value = ALL_64_BITS;
1500 
1501     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1502     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1503     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1504   } else {
1505     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1506     info_ptr->max_value = ALL_64_BITS;
1507 
1508     // gettimeofday is a real time clock so it skips
1509     info_ptr->may_skip_backward = true;
1510     info_ptr->may_skip_forward = true;
1511   }
1512 
1513   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1514 }
1515 
1516 // Return the real, user, and system times in seconds from an
1517 // arbitrary fixed point in the past.
1518 bool os::getTimesSecs(double* process_real_time,
1519                       double* process_user_time,
1520                       double* process_system_time) {
1521   struct tms ticks;
1522   clock_t real_ticks = times(&ticks);
1523 
1524   if (real_ticks == (clock_t) (-1)) {
1525     return false;
1526   } else {
1527     double ticks_per_second = (double) clock_tics_per_sec;
1528     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1529     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1530     *process_real_time = ((double) real_ticks) / ticks_per_second;
1531 
1532     return true;
1533   }
1534 }
1535 
1536 
1537 char * os::local_time_string(char *buf, size_t buflen) {
1538   struct tm t;
1539   time_t long_time;
1540   time(&long_time);
1541   localtime_r(&long_time, &t);
1542   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1543                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1544                t.tm_hour, t.tm_min, t.tm_sec);
1545   return buf;
1546 }
1547 
1548 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1549   return localtime_r(clock, res);
1550 }
1551 
1552 ////////////////////////////////////////////////////////////////////////////////
1553 // runtime exit support
1554 
1555 // Note: os::shutdown() might be called very early during initialization, or
1556 // called from signal handler. Before adding something to os::shutdown(), make
1557 // sure it is async-safe and can handle partially initialized VM.
1558 void os::shutdown() {
1559 
1560   // allow PerfMemory to attempt cleanup of any persistent resources
1561   perfMemory_exit();
1562 
1563   // needs to remove object in file system
1564   AttachListener::abort();
1565 
1566   // flush buffered output, finish log files
1567   ostream_abort();
1568 
1569   // Check for abort hook
1570   abort_hook_t abort_hook = Arguments::abort_hook();
1571   if (abort_hook != NULL) {
1572     abort_hook();
1573   }
1574 
1575 }
1576 
1577 // Note: os::abort() might be called very early during initialization, or
1578 // called from signal handler. Before adding something to os::abort(), make
1579 // sure it is async-safe and can handle partially initialized VM.
1580 void os::abort(bool dump_core) {
1581   os::shutdown();
1582   if (dump_core) {
1583 #ifndef PRODUCT
1584     fdStream out(defaultStream::output_fd());
1585     out.print_raw("Current thread is ");
1586     char buf[16];
1587     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1588     out.print_raw_cr(buf);
1589     out.print_raw_cr("Dumping core ...");
1590 #endif
1591     ::abort(); // dump core
1592   }
1593 
1594   ::exit(1);
1595 }
1596 
1597 // Die immediately, no exit hook, no abort hook, no cleanup.
1598 void os::die() {
1599   // _exit() on LinuxThreads only kills current thread
1600   ::abort();
1601 }
1602 
1603 // unused on linux for now.
1604 void os::set_error_file(const char *logfile) {}
1605 
1606 
1607 // This method is a copy of JDK's sysGetLastErrorString
1608 // from src/solaris/hpi/src/system_md.c
1609 
1610 size_t os::lasterror(char *buf, size_t len) {
1611 
1612   if (errno == 0)  return 0;
1613 
1614   const char *s = ::strerror(errno);
1615   size_t n = ::strlen(s);
1616   if (n >= len) {
1617     n = len - 1;
1618   }
1619   ::strncpy(buf, s, n);
1620   buf[n] = '\0';
1621   return n;
1622 }
1623 
1624 intx os::current_thread_id() { return (intx)pthread_self(); }
1625 int os::current_process_id() {
1626 
1627   // Under the old linux thread library, linux gives each thread
1628   // its own process id. Because of this each thread will return
1629   // a different pid if this method were to return the result
1630   // of getpid(2). Linux provides no api that returns the pid
1631   // of the launcher thread for the vm. This implementation
1632   // returns a unique pid, the pid of the launcher thread
1633   // that starts the vm 'process'.
1634 
1635   // Under the NPTL, getpid() returns the same pid as the
1636   // launcher thread rather than a unique pid per thread.
1637   // Use gettid() if you want the old pre NPTL behaviour.
1638 
1639   // if you are looking for the result of a call to getpid() that
1640   // returns a unique pid for the calling thread, then look at the
1641   // OSThread::thread_id() method in osThread_linux.hpp file
1642 
1643   return (int)(_initial_pid ? _initial_pid : getpid());
1644 }
1645 
1646 // DLL functions
1647 
1648 const char* os::dll_file_extension() { return ".so"; }
1649 
1650 // This must be hard coded because it's the system's temporary
1651 // directory not the java application's temp directory, ala java.io.tmpdir.
1652 const char* os::get_temp_directory() { return "/tmp"; }
1653 
1654 static bool file_exists(const char* filename) {
1655   struct stat statbuf;
1656   if (filename == NULL || strlen(filename) == 0) {
1657     return false;
1658   }
1659   return os::stat(filename, &statbuf) == 0;
1660 }
1661 
1662 void os::dll_build_name(char* buffer, size_t buflen,
1663                         const char* pname, const char* fname) {
1664   // Copied from libhpi
1665   const size_t pnamelen = pname ? strlen(pname) : 0;
1666 
1667   // Quietly truncate on buffer overflow.  Should be an error.
1668   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1669       *buffer = '\0';
1670       return;
1671   }
1672 
1673   if (pnamelen == 0) {
1674     snprintf(buffer, buflen, "lib%s.so", fname);
1675   } else if (strchr(pname, *os::path_separator()) != NULL) {
1676     int n;
1677     char** pelements = split_path(pname, &n);
1678     for (int i = 0 ; i < n ; i++) {
1679       // Really shouldn't be NULL, but check can't hurt
1680       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1681         continue; // skip the empty path values
1682       }
1683       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1684       if (file_exists(buffer)) {
1685         break;
1686       }
1687     }
1688     // release the storage
1689     for (int i = 0 ; i < n ; i++) {
1690       if (pelements[i] != NULL) {
1691         FREE_C_HEAP_ARRAY(char, pelements[i]);
1692       }
1693     }
1694     if (pelements != NULL) {
1695       FREE_C_HEAP_ARRAY(char*, pelements);
1696     }
1697   } else {
1698     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1699   }
1700 }
1701 
1702 const char* os::get_current_directory(char *buf, int buflen) {
1703   return getcwd(buf, buflen);
1704 }
1705 
1706 // check if addr is inside libjvm[_g].so
1707 bool os::address_is_in_vm(address addr) {
1708   static address libjvm_base_addr;
1709   Dl_info dlinfo;
1710 
1711   if (libjvm_base_addr == NULL) {
1712     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1713     libjvm_base_addr = (address)dlinfo.dli_fbase;
1714     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1715   }
1716 
1717   if (dladdr((void *)addr, &dlinfo)) {
1718     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1719   }
1720 
1721   return false;
1722 }
1723 
1724 bool os::dll_address_to_function_name(address addr, char *buf,
1725                                       int buflen, int *offset) {
1726   Dl_info dlinfo;
1727 
1728   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1729     if (buf != NULL) {
1730       if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1731         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1732       }
1733     }
1734     if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1735     return true;
1736   } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1737     if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1738        dlinfo.dli_fname, buf, buflen, offset) == Decoder::no_error) {
1739        return true;
1740     }
1741   }
1742 
1743   if (buf != NULL) buf[0] = '\0';
1744   if (offset != NULL) *offset = -1;
1745   return false;
1746 }
1747 
1748 struct _address_to_library_name {
1749   address addr;          // input : memory address
1750   size_t  buflen;        //         size of fname
1751   char*   fname;         // output: library name
1752   address base;          //         library base addr
1753 };
1754 
1755 static int address_to_library_name_callback(struct dl_phdr_info *info,
1756                                             size_t size, void *data) {
1757   int i;
1758   bool found = false;
1759   address libbase = NULL;
1760   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1761 
1762   // iterate through all loadable segments
1763   for (i = 0; i < info->dlpi_phnum; i++) {
1764     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1765     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1766       // base address of a library is the lowest address of its loaded
1767       // segments.
1768       if (libbase == NULL || libbase > segbase) {
1769         libbase = segbase;
1770       }
1771       // see if 'addr' is within current segment
1772       if (segbase <= d->addr &&
1773           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1774         found = true;
1775       }
1776     }
1777   }
1778 
1779   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1780   // so dll_address_to_library_name() can fall through to use dladdr() which
1781   // can figure out executable name from argv[0].
1782   if (found && info->dlpi_name && info->dlpi_name[0]) {
1783     d->base = libbase;
1784     if (d->fname) {
1785       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1786     }
1787     return 1;
1788   }
1789   return 0;
1790 }
1791 
1792 bool os::dll_address_to_library_name(address addr, char* buf,
1793                                      int buflen, int* offset) {
1794   Dl_info dlinfo;
1795   struct _address_to_library_name data;
1796 
1797   // There is a bug in old glibc dladdr() implementation that it could resolve
1798   // to wrong library name if the .so file has a base address != NULL. Here
1799   // we iterate through the program headers of all loaded libraries to find
1800   // out which library 'addr' really belongs to. This workaround can be
1801   // removed once the minimum requirement for glibc is moved to 2.3.x.
1802   data.addr = addr;
1803   data.fname = buf;
1804   data.buflen = buflen;
1805   data.base = NULL;
1806   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1807 
1808   if (rslt) {
1809      // buf already contains library name
1810      if (offset) *offset = addr - data.base;
1811      return true;
1812   } else if (dladdr((void*)addr, &dlinfo)){
1813      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1814      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1815      return true;
1816   } else {
1817      if (buf) buf[0] = '\0';
1818      if (offset) *offset = -1;
1819      return false;
1820   }
1821 }
1822 
1823   // Loads .dll/.so and
1824   // in case of error it checks if .dll/.so was built for the
1825   // same architecture as Hotspot is running on
1826 
1827 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1828 {
1829   void * result= ::dlopen(filename, RTLD_LAZY);
1830   if (result != NULL) {
1831     // Successful loading
1832     return result;
1833   }
1834 
1835   Elf32_Ehdr elf_head;
1836 
1837   // Read system error message into ebuf
1838   // It may or may not be overwritten below
1839   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1840   ebuf[ebuflen-1]='\0';
1841   int diag_msg_max_length=ebuflen-strlen(ebuf);
1842   char* diag_msg_buf=ebuf+strlen(ebuf);
1843 
1844   if (diag_msg_max_length==0) {
1845     // No more space in ebuf for additional diagnostics message
1846     return NULL;
1847   }
1848 
1849 
1850   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1851 
1852   if (file_descriptor < 0) {
1853     // Can't open library, report dlerror() message
1854     return NULL;
1855   }
1856 
1857   bool failed_to_read_elf_head=
1858     (sizeof(elf_head)!=
1859         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1860 
1861   ::close(file_descriptor);
1862   if (failed_to_read_elf_head) {
1863     // file i/o error - report dlerror() msg
1864     return NULL;
1865   }
1866 
1867   typedef struct {
1868     Elf32_Half  code;         // Actual value as defined in elf.h
1869     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1870     char        elf_class;    // 32 or 64 bit
1871     char        endianess;    // MSB or LSB
1872     char*       name;         // String representation
1873   } arch_t;
1874 
1875   #ifndef EM_486
1876   #define EM_486          6               /* Intel 80486 */
1877   #endif
1878 
1879   static const arch_t arch_array[]={
1880     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1881     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1882     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1883     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1884     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1885     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1886     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1887     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1888     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1889     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1890     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1891     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1892     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1893     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1894     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1895     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1896   };
1897 
1898   #if  (defined IA32)
1899     static  Elf32_Half running_arch_code=EM_386;
1900   #elif   (defined AMD64)
1901     static  Elf32_Half running_arch_code=EM_X86_64;
1902   #elif  (defined IA64)
1903     static  Elf32_Half running_arch_code=EM_IA_64;
1904   #elif  (defined __sparc) && (defined _LP64)
1905     static  Elf32_Half running_arch_code=EM_SPARCV9;
1906   #elif  (defined __sparc) && (!defined _LP64)
1907     static  Elf32_Half running_arch_code=EM_SPARC;
1908   #elif  (defined __powerpc64__)
1909     static  Elf32_Half running_arch_code=EM_PPC64;
1910   #elif  (defined __powerpc__)
1911     static  Elf32_Half running_arch_code=EM_PPC;
1912   #elif  (defined ARM)
1913     static  Elf32_Half running_arch_code=EM_ARM;
1914   #elif  (defined S390)
1915     static  Elf32_Half running_arch_code=EM_S390;
1916   #elif  (defined ALPHA)
1917     static  Elf32_Half running_arch_code=EM_ALPHA;
1918   #elif  (defined MIPSEL)
1919     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1920   #elif  (defined PARISC)
1921     static  Elf32_Half running_arch_code=EM_PARISC;
1922   #elif  (defined MIPS)
1923     static  Elf32_Half running_arch_code=EM_MIPS;
1924   #elif  (defined M68K)
1925     static  Elf32_Half running_arch_code=EM_68K;
1926   #else
1927     #error Method os::dll_load requires that one of following is defined:\
1928          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1929   #endif
1930 
1931   // Identify compatability class for VM's architecture and library's architecture
1932   // Obtain string descriptions for architectures
1933 
1934   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1935   int running_arch_index=-1;
1936 
1937   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1938     if (running_arch_code == arch_array[i].code) {
1939       running_arch_index    = i;
1940     }
1941     if (lib_arch.code == arch_array[i].code) {
1942       lib_arch.compat_class = arch_array[i].compat_class;
1943       lib_arch.name         = arch_array[i].name;
1944     }
1945   }
1946 
1947   assert(running_arch_index != -1,
1948     "Didn't find running architecture code (running_arch_code) in arch_array");
1949   if (running_arch_index == -1) {
1950     // Even though running architecture detection failed
1951     // we may still continue with reporting dlerror() message
1952     return NULL;
1953   }
1954 
1955   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1956     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1957     return NULL;
1958   }
1959 
1960 #ifndef S390
1961   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1962     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1963     return NULL;
1964   }
1965 #endif // !S390
1966 
1967   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1968     if ( lib_arch.name!=NULL ) {
1969       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1970         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1971         lib_arch.name, arch_array[running_arch_index].name);
1972     } else {
1973       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1974       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1975         lib_arch.code,
1976         arch_array[running_arch_index].name);
1977     }
1978   }
1979 
1980   return NULL;
1981 }
1982 
1983 /*
1984  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
1985  * chances are you might want to run the generated bits against glibc-2.0
1986  * libdl.so, so always use locking for any version of glibc.
1987  */
1988 void* os::dll_lookup(void* handle, const char* name) {
1989   pthread_mutex_lock(&dl_mutex);
1990   void* res = dlsym(handle, name);
1991   pthread_mutex_unlock(&dl_mutex);
1992   return res;
1993 }
1994 
1995 
1996 static bool _print_ascii_file(const char* filename, outputStream* st) {
1997   int fd = ::open(filename, O_RDONLY);
1998   if (fd == -1) {
1999      return false;
2000   }
2001 
2002   char buf[32];
2003   int bytes;
2004   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2005     st->print_raw(buf, bytes);
2006   }
2007 
2008   ::close(fd);
2009 
2010   return true;
2011 }
2012 
2013 void os::print_dll_info(outputStream *st) {
2014    st->print_cr("Dynamic libraries:");
2015 
2016    char fname[32];
2017    pid_t pid = os::Linux::gettid();
2018 
2019    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2020 
2021    if (!_print_ascii_file(fname, st)) {
2022      st->print("Can not get library information for pid = %d\n", pid);
2023    }
2024 }
2025 
2026 
2027 void os::print_os_info(outputStream* st) {
2028   st->print("OS:");
2029 
2030   // Try to identify popular distros.
2031   // Most Linux distributions have /etc/XXX-release file, which contains
2032   // the OS version string. Some have more than one /etc/XXX-release file
2033   // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2034   // so the order is important.
2035   if (!_print_ascii_file("/etc/mandrake-release", st) &&
2036       !_print_ascii_file("/etc/sun-release", st) &&
2037       !_print_ascii_file("/etc/redhat-release", st) &&
2038       !_print_ascii_file("/etc/SuSE-release", st) &&
2039       !_print_ascii_file("/etc/turbolinux-release", st) &&
2040       !_print_ascii_file("/etc/gentoo-release", st) &&
2041       !_print_ascii_file("/etc/debian_version", st) &&
2042       !_print_ascii_file("/etc/ltib-release", st) &&
2043       !_print_ascii_file("/etc/angstrom-version", st)) {
2044       st->print("Linux");
2045   }
2046   st->cr();
2047 
2048   // kernel
2049   st->print("uname:");
2050   struct utsname name;
2051   uname(&name);
2052   st->print(name.sysname); st->print(" ");
2053   st->print(name.release); st->print(" ");
2054   st->print(name.version); st->print(" ");
2055   st->print(name.machine);
2056   st->cr();
2057 
2058   // Print warning if unsafe chroot environment detected
2059   if (unsafe_chroot_detected) {
2060     st->print("WARNING!! ");
2061     st->print_cr(unstable_chroot_error);
2062   }
2063 
2064   // libc, pthread
2065   st->print("libc:");
2066   st->print(os::Linux::glibc_version()); st->print(" ");
2067   st->print(os::Linux::libpthread_version()); st->print(" ");
2068   if (os::Linux::is_LinuxThreads()) {
2069      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2070   }
2071   st->cr();
2072 
2073   // rlimit
2074   st->print("rlimit:");
2075   struct rlimit rlim;
2076 
2077   st->print(" STACK ");
2078   getrlimit(RLIMIT_STACK, &rlim);
2079   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2080   else st->print("%uk", rlim.rlim_cur >> 10);
2081 
2082   st->print(", CORE ");
2083   getrlimit(RLIMIT_CORE, &rlim);
2084   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2085   else st->print("%uk", rlim.rlim_cur >> 10);
2086 
2087   st->print(", NPROC ");
2088   getrlimit(RLIMIT_NPROC, &rlim);
2089   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2090   else st->print("%d", rlim.rlim_cur);
2091 
2092   st->print(", NOFILE ");
2093   getrlimit(RLIMIT_NOFILE, &rlim);
2094   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2095   else st->print("%d", rlim.rlim_cur);
2096 
2097   st->print(", AS ");
2098   getrlimit(RLIMIT_AS, &rlim);
2099   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2100   else st->print("%uk", rlim.rlim_cur >> 10);
2101   st->cr();
2102 
2103   // load average
2104   st->print("load average:");
2105   double loadavg[3];
2106   os::loadavg(loadavg, 3);
2107   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2108   st->cr();
2109 
2110   // meminfo
2111   st->print("\n/proc/meminfo:\n");
2112   _print_ascii_file("/proc/meminfo", st);
2113   st->cr();
2114 }
2115 
2116 void os::pd_print_cpu_info(outputStream* st) {
2117   st->print("\n/proc/cpuinfo:\n");
2118   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2119     st->print("  <Not Available>");
2120   }
2121   st->cr();
2122 }
2123 
2124 void os::print_memory_info(outputStream* st) {
2125 
2126   st->print("Memory:");
2127   st->print(" %dk page", os::vm_page_size()>>10);
2128 
2129   // values in struct sysinfo are "unsigned long"
2130   struct sysinfo si;
2131   sysinfo(&si);
2132 
2133   st->print(", physical " UINT64_FORMAT "k",
2134             os::physical_memory() >> 10);
2135   st->print("(" UINT64_FORMAT "k free)",
2136             os::available_memory() >> 10);
2137   st->print(", swap " UINT64_FORMAT "k",
2138             ((jlong)si.totalswap * si.mem_unit) >> 10);
2139   st->print("(" UINT64_FORMAT "k free)",
2140             ((jlong)si.freeswap * si.mem_unit) >> 10);
2141   st->cr();
2142 }
2143 
2144 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2145 // but they're the same for all the linux arch that we support
2146 // and they're the same for solaris but there's no common place to put this.
2147 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2148                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2149                           "ILL_COPROC", "ILL_BADSTK" };
2150 
2151 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2152                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2153                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2154 
2155 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2156 
2157 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2158 
2159 void os::print_siginfo(outputStream* st, void* siginfo) {
2160   st->print("siginfo:");
2161 
2162   const int buflen = 100;
2163   char buf[buflen];
2164   siginfo_t *si = (siginfo_t*)siginfo;
2165   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2166   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2167     st->print("si_errno=%s", buf);
2168   } else {
2169     st->print("si_errno=%d", si->si_errno);
2170   }
2171   const int c = si->si_code;
2172   assert(c > 0, "unexpected si_code");
2173   switch (si->si_signo) {
2174   case SIGILL:
2175     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2176     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2177     break;
2178   case SIGFPE:
2179     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2180     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2181     break;
2182   case SIGSEGV:
2183     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2184     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2185     break;
2186   case SIGBUS:
2187     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2188     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2189     break;
2190   default:
2191     st->print(", si_code=%d", si->si_code);
2192     // no si_addr
2193   }
2194 
2195   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2196       UseSharedSpaces) {
2197     FileMapInfo* mapinfo = FileMapInfo::current_info();
2198     if (mapinfo->is_in_shared_space(si->si_addr)) {
2199       st->print("\n\nError accessing class data sharing archive."   \
2200                 " Mapped file inaccessible during execution, "      \
2201                 " possible disk/network problem.");
2202     }
2203   }
2204   st->cr();
2205 }
2206 
2207 
2208 static void print_signal_handler(outputStream* st, int sig,
2209                                  char* buf, size_t buflen);
2210 
2211 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2212   st->print_cr("Signal Handlers:");
2213   print_signal_handler(st, SIGSEGV, buf, buflen);
2214   print_signal_handler(st, SIGBUS , buf, buflen);
2215   print_signal_handler(st, SIGFPE , buf, buflen);
2216   print_signal_handler(st, SIGPIPE, buf, buflen);
2217   print_signal_handler(st, SIGXFSZ, buf, buflen);
2218   print_signal_handler(st, SIGILL , buf, buflen);
2219   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2220   print_signal_handler(st, SR_signum, buf, buflen);
2221   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2222   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2223   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2224   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2225 }
2226 
2227 static char saved_jvm_path[MAXPATHLEN] = {0};
2228 
2229 // Find the full path to the current module, libjvm.so or libjvm_g.so
2230 void os::jvm_path(char *buf, jint buflen) {
2231   // Error checking.
2232   if (buflen < MAXPATHLEN) {
2233     assert(false, "must use a large-enough buffer");
2234     buf[0] = '\0';
2235     return;
2236   }
2237   // Lazy resolve the path to current module.
2238   if (saved_jvm_path[0] != 0) {
2239     strcpy(buf, saved_jvm_path);
2240     return;
2241   }
2242 
2243   char dli_fname[MAXPATHLEN];
2244   bool ret = dll_address_to_library_name(
2245                 CAST_FROM_FN_PTR(address, os::jvm_path),
2246                 dli_fname, sizeof(dli_fname), NULL);
2247   assert(ret != 0, "cannot locate libjvm");
2248   char *rp = realpath(dli_fname, buf);
2249   if (rp == NULL)
2250     return;
2251 
2252   if (Arguments::created_by_gamma_launcher()) {
2253     // Support for the gamma launcher.  Typical value for buf is
2254     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2255     // the right place in the string, then assume we are installed in a JDK and
2256     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2257     // up the path so it looks like libjvm.so is installed there (append a
2258     // fake suffix hotspot/libjvm.so).
2259     const char *p = buf + strlen(buf) - 1;
2260     for (int count = 0; p > buf && count < 5; ++count) {
2261       for (--p; p > buf && *p != '/'; --p)
2262         /* empty */ ;
2263     }
2264 
2265     if (strncmp(p, "/jre/lib/", 9) != 0) {
2266       // Look for JAVA_HOME in the environment.
2267       char* java_home_var = ::getenv("JAVA_HOME");
2268       if (java_home_var != NULL && java_home_var[0] != 0) {
2269         char* jrelib_p;
2270         int len;
2271 
2272         // Check the current module name "libjvm.so" or "libjvm_g.so".
2273         p = strrchr(buf, '/');
2274         assert(strstr(p, "/libjvm") == p, "invalid library name");
2275         p = strstr(p, "_g") ? "_g" : "";
2276 
2277         rp = realpath(java_home_var, buf);
2278         if (rp == NULL)
2279           return;
2280 
2281         // determine if this is a legacy image or modules image
2282         // modules image doesn't have "jre" subdirectory
2283         len = strlen(buf);
2284         jrelib_p = buf + len;
2285         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2286         if (0 != access(buf, F_OK)) {
2287           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2288         }
2289 
2290         if (0 == access(buf, F_OK)) {
2291           // Use current module name "libjvm[_g].so" instead of
2292           // "libjvm"debug_only("_g")".so" since for fastdebug version
2293           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2294           len = strlen(buf);
2295           snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2296         } else {
2297           // Go back to path of .so
2298           rp = realpath(dli_fname, buf);
2299           if (rp == NULL)
2300             return;
2301         }
2302       }
2303     }
2304   }
2305 
2306   strcpy(saved_jvm_path, buf);
2307 }
2308 
2309 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2310   // no prefix required, not even "_"
2311 }
2312 
2313 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2314   // no suffix required
2315 }
2316 
2317 ////////////////////////////////////////////////////////////////////////////////
2318 // sun.misc.Signal support
2319 
2320 static volatile jint sigint_count = 0;
2321 
2322 static void
2323 UserHandler(int sig, void *siginfo, void *context) {
2324   // 4511530 - sem_post is serialized and handled by the manager thread. When
2325   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2326   // don't want to flood the manager thread with sem_post requests.
2327   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2328       return;
2329 
2330   // Ctrl-C is pressed during error reporting, likely because the error
2331   // handler fails to abort. Let VM die immediately.
2332   if (sig == SIGINT && is_error_reported()) {
2333      os::die();
2334   }
2335 
2336   os::signal_notify(sig);
2337 }
2338 
2339 void* os::user_handler() {
2340   return CAST_FROM_FN_PTR(void*, UserHandler);
2341 }
2342 
2343 extern "C" {
2344   typedef void (*sa_handler_t)(int);
2345   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2346 }
2347 
2348 void* os::signal(int signal_number, void* handler) {
2349   struct sigaction sigAct, oldSigAct;
2350 
2351   sigfillset(&(sigAct.sa_mask));
2352   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2353   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2354 
2355   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2356     // -1 means registration failed
2357     return (void *)-1;
2358   }
2359 
2360   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2361 }
2362 
2363 void os::signal_raise(int signal_number) {
2364   ::raise(signal_number);
2365 }
2366 
2367 /*
2368  * The following code is moved from os.cpp for making this
2369  * code platform specific, which it is by its very nature.
2370  */
2371 
2372 // Will be modified when max signal is changed to be dynamic
2373 int os::sigexitnum_pd() {
2374   return NSIG;
2375 }
2376 
2377 // a counter for each possible signal value
2378 static volatile jint pending_signals[NSIG+1] = { 0 };
2379 
2380 // Linux(POSIX) specific hand shaking semaphore.
2381 static sem_t sig_sem;
2382 
2383 void os::signal_init_pd() {
2384   // Initialize signal structures
2385   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2386 
2387   // Initialize signal semaphore
2388   ::sem_init(&sig_sem, 0, 0);
2389 }
2390 
2391 void os::signal_notify(int sig) {
2392   Atomic::inc(&pending_signals[sig]);
2393   ::sem_post(&sig_sem);
2394 }
2395 
2396 static int check_pending_signals(bool wait) {
2397   Atomic::store(0, &sigint_count);
2398   for (;;) {
2399     for (int i = 0; i < NSIG + 1; i++) {
2400       jint n = pending_signals[i];
2401       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2402         return i;
2403       }
2404     }
2405     if (!wait) {
2406       return -1;
2407     }
2408     JavaThread *thread = JavaThread::current();
2409     ThreadBlockInVM tbivm(thread);
2410 
2411     bool threadIsSuspended;
2412     do {
2413       thread->set_suspend_equivalent();
2414       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2415       ::sem_wait(&sig_sem);
2416 
2417       // were we externally suspended while we were waiting?
2418       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2419       if (threadIsSuspended) {
2420         //
2421         // The semaphore has been incremented, but while we were waiting
2422         // another thread suspended us. We don't want to continue running
2423         // while suspended because that would surprise the thread that
2424         // suspended us.
2425         //
2426         ::sem_post(&sig_sem);
2427 
2428         thread->java_suspend_self();
2429       }
2430     } while (threadIsSuspended);
2431   }
2432 }
2433 
2434 int os::signal_lookup() {
2435   return check_pending_signals(false);
2436 }
2437 
2438 int os::signal_wait() {
2439   return check_pending_signals(true);
2440 }
2441 
2442 ////////////////////////////////////////////////////////////////////////////////
2443 // Virtual Memory
2444 
2445 int os::vm_page_size() {
2446   // Seems redundant as all get out
2447   assert(os::Linux::page_size() != -1, "must call os::init");
2448   return os::Linux::page_size();
2449 }
2450 
2451 // Solaris allocates memory by pages.
2452 int os::vm_allocation_granularity() {
2453   assert(os::Linux::page_size() != -1, "must call os::init");
2454   return os::Linux::page_size();
2455 }
2456 
2457 // Rationale behind this function:
2458 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2459 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2460 //  samples for JITted code. Here we create private executable mapping over the code cache
2461 //  and then we can use standard (well, almost, as mapping can change) way to provide
2462 //  info for the reporting script by storing timestamp and location of symbol
2463 void linux_wrap_code(char* base, size_t size) {
2464   static volatile jint cnt = 0;
2465 
2466   if (!UseOprofile) {
2467     return;
2468   }
2469 
2470   char buf[PATH_MAX+1];
2471   int num = Atomic::add(1, &cnt);
2472 
2473   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2474            os::get_temp_directory(), os::current_process_id(), num);
2475   unlink(buf);
2476 
2477   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2478 
2479   if (fd != -1) {
2480     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2481     if (rv != (off_t)-1) {
2482       if (::write(fd, "", 1) == 1) {
2483         mmap(base, size,
2484              PROT_READ|PROT_WRITE|PROT_EXEC,
2485              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2486       }
2487     }
2488     ::close(fd);
2489     unlink(buf);
2490   }
2491 }
2492 
2493 // NOTE: Linux kernel does not really reserve the pages for us.
2494 //       All it does is to check if there are enough free pages
2495 //       left at the time of mmap(). This could be a potential
2496 //       problem.
2497 bool os::commit_memory(char* addr, size_t size, bool exec) {
2498   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2499   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2500                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2501   if (res != (uintptr_t) MAP_FAILED) {
2502     if (UseNUMAInterleaving) {
2503       numa_make_global(addr, size);
2504     }
2505     return true;
2506   }
2507   return false;
2508 }
2509 
2510 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2511 #ifndef MAP_HUGETLB
2512 #define MAP_HUGETLB 0x40000
2513 #endif
2514 
2515 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2516 #ifndef MADV_HUGEPAGE
2517 #define MADV_HUGEPAGE 14
2518 #endif
2519 
2520 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2521                        bool exec) {
2522   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2523     int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2524     uintptr_t res =
2525       (uintptr_t) ::mmap(addr, size, prot,
2526                          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2527                          -1, 0);
2528     if (res != (uintptr_t) MAP_FAILED) {
2529       if (UseNUMAInterleaving) {
2530         numa_make_global(addr, size);
2531       }
2532       return true;
2533     }
2534     // Fall through and try to use small pages
2535   }
2536 
2537   if (commit_memory(addr, size, exec)) {
2538     realign_memory(addr, size, alignment_hint);
2539     return true;
2540   }
2541   return false;
2542 }
2543 
2544 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2545   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2546     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2547     // be supported or the memory may already be backed by huge pages.
2548     ::madvise(addr, bytes, MADV_HUGEPAGE);
2549   }
2550 }
2551 
2552 void os::free_memory(char *addr, size_t bytes) {
2553   commit_memory(addr, bytes, false);
2554 }
2555 
2556 void os::numa_make_global(char *addr, size_t bytes) {
2557   Linux::numa_interleave_memory(addr, bytes);
2558 }
2559 
2560 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2561   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2562 }
2563 
2564 bool os::numa_topology_changed()   { return false; }
2565 
2566 size_t os::numa_get_groups_num() {
2567   int max_node = Linux::numa_max_node();
2568   return max_node > 0 ? max_node + 1 : 1;
2569 }
2570 
2571 int os::numa_get_group_id() {
2572   int cpu_id = Linux::sched_getcpu();
2573   if (cpu_id != -1) {
2574     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2575     if (lgrp_id != -1) {
2576       return lgrp_id;
2577     }
2578   }
2579   return 0;
2580 }
2581 
2582 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2583   for (size_t i = 0; i < size; i++) {
2584     ids[i] = i;
2585   }
2586   return size;
2587 }
2588 
2589 bool os::get_page_info(char *start, page_info* info) {
2590   return false;
2591 }
2592 
2593 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2594   return end;
2595 }
2596 
2597 
2598 int os::Linux::sched_getcpu_syscall(void) {
2599   unsigned int cpu;
2600   int retval = -1;
2601 
2602 #if defined(IA32)
2603 # ifndef SYS_getcpu
2604 # define SYS_getcpu 318
2605 # endif
2606   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2607 #elif defined(AMD64)
2608 // Unfortunately we have to bring all these macros here from vsyscall.h
2609 // to be able to compile on old linuxes.
2610 # define __NR_vgetcpu 2
2611 # define VSYSCALL_START (-10UL << 20)
2612 # define VSYSCALL_SIZE 1024
2613 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2614   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2615   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2616   retval = vgetcpu(&cpu, NULL, NULL);
2617 #endif
2618 
2619   return (retval == -1) ? retval : cpu;
2620 }
2621 
2622 // Something to do with the numa-aware allocator needs these symbols
2623 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2624 extern "C" JNIEXPORT void numa_error(char *where) { }
2625 extern "C" JNIEXPORT int fork1() { return fork(); }
2626 
2627 
2628 // If we are running with libnuma version > 2, then we should
2629 // be trying to use symbols with versions 1.1
2630 // If we are running with earlier version, which did not have symbol versions,
2631 // we should use the base version.
2632 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2633   void *f = dlvsym(handle, name, "libnuma_1.1");
2634   if (f == NULL) {
2635     f = dlsym(handle, name);
2636   }
2637   return f;
2638 }
2639 
2640 bool os::Linux::libnuma_init() {
2641   // sched_getcpu() should be in libc.
2642   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2643                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2644 
2645   // If it's not, try a direct syscall.
2646   if (sched_getcpu() == -1)
2647     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2648 
2649   if (sched_getcpu() != -1) { // Does it work?
2650     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2651     if (handle != NULL) {
2652       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2653                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2654       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2655                                        libnuma_dlsym(handle, "numa_max_node")));
2656       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2657                                         libnuma_dlsym(handle, "numa_available")));
2658       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2659                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2660       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2661                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2662 
2663 
2664       if (numa_available() != -1) {
2665         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2666         // Create a cpu -> node mapping
2667         _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2668         rebuild_cpu_to_node_map();
2669         return true;
2670       }
2671     }
2672   }
2673   return false;
2674 }
2675 
2676 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2677 // The table is later used in get_node_by_cpu().
2678 void os::Linux::rebuild_cpu_to_node_map() {
2679   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2680                               // in libnuma (possible values are starting from 16,
2681                               // and continuing up with every other power of 2, but less
2682                               // than the maximum number of CPUs supported by kernel), and
2683                               // is a subject to change (in libnuma version 2 the requirements
2684                               // are more reasonable) we'll just hardcode the number they use
2685                               // in the library.
2686   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2687 
2688   size_t cpu_num = os::active_processor_count();
2689   size_t cpu_map_size = NCPUS / BitsPerCLong;
2690   size_t cpu_map_valid_size =
2691     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2692 
2693   cpu_to_node()->clear();
2694   cpu_to_node()->at_grow(cpu_num - 1);
2695   size_t node_num = numa_get_groups_num();
2696 
2697   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2698   for (size_t i = 0; i < node_num; i++) {
2699     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2700       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2701         if (cpu_map[j] != 0) {
2702           for (size_t k = 0; k < BitsPerCLong; k++) {
2703             if (cpu_map[j] & (1UL << k)) {
2704               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2705             }
2706           }
2707         }
2708       }
2709     }
2710   }
2711   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2712 }
2713 
2714 int os::Linux::get_node_by_cpu(int cpu_id) {
2715   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2716     return cpu_to_node()->at(cpu_id);
2717   }
2718   return -1;
2719 }
2720 
2721 GrowableArray<int>* os::Linux::_cpu_to_node;
2722 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2723 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2724 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2725 os::Linux::numa_available_func_t os::Linux::_numa_available;
2726 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2727 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2728 unsigned long* os::Linux::_numa_all_nodes;
2729 
2730 bool os::uncommit_memory(char* addr, size_t size) {
2731   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2732                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2733   return res  != (uintptr_t) MAP_FAILED;
2734 }
2735 
2736 // Linux uses a growable mapping for the stack, and if the mapping for
2737 // the stack guard pages is not removed when we detach a thread the
2738 // stack cannot grow beyond the pages where the stack guard was
2739 // mapped.  If at some point later in the process the stack expands to
2740 // that point, the Linux kernel cannot expand the stack any further
2741 // because the guard pages are in the way, and a segfault occurs.
2742 //
2743 // However, it's essential not to split the stack region by unmapping
2744 // a region (leaving a hole) that's already part of the stack mapping,
2745 // so if the stack mapping has already grown beyond the guard pages at
2746 // the time we create them, we have to truncate the stack mapping.
2747 // So, we need to know the extent of the stack mapping when
2748 // create_stack_guard_pages() is called.
2749 
2750 // Find the bounds of the stack mapping.  Return true for success.
2751 //
2752 // We only need this for stacks that are growable: at the time of
2753 // writing thread stacks don't use growable mappings (i.e. those
2754 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2755 // only applies to the main thread.
2756 
2757 static
2758 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2759 
2760   char buf[128];
2761   int fd, sz;
2762 
2763   if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2764     return false;
2765   }
2766 
2767   const char kw[] = "[stack]";
2768   const int kwlen = sizeof(kw)-1;
2769 
2770   // Address part of /proc/self/maps couldn't be more than 128 bytes
2771   while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2772      if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2773         // Extract addresses
2774         if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2775            uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2776            if (sp >= *bottom && sp <= *top) {
2777               ::close(fd);
2778               return true;
2779            }
2780         }
2781      }
2782   }
2783 
2784  ::close(fd);
2785   return false;
2786 }
2787 
2788 
2789 // If the (growable) stack mapping already extends beyond the point
2790 // where we're going to put our guard pages, truncate the mapping at
2791 // that point by munmap()ping it.  This ensures that when we later
2792 // munmap() the guard pages we don't leave a hole in the stack
2793 // mapping. This only affects the main/initial thread, but guard
2794 // against future OS changes
2795 bool os::create_stack_guard_pages(char* addr, size_t size) {
2796   uintptr_t stack_extent, stack_base;
2797   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2798   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2799       assert(os::Linux::is_initial_thread(),
2800            "growable stack in non-initial thread");
2801     if (stack_extent < (uintptr_t)addr)
2802       ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2803   }
2804 
2805   return os::commit_memory(addr, size);
2806 }
2807 
2808 // If this is a growable mapping, remove the guard pages entirely by
2809 // munmap()ping them.  If not, just call uncommit_memory(). This only
2810 // affects the main/initial thread, but guard against future OS changes
2811 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2812   uintptr_t stack_extent, stack_base;
2813   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2814   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2815       assert(os::Linux::is_initial_thread(),
2816            "growable stack in non-initial thread");
2817 
2818     return ::munmap(addr, size) == 0;
2819   }
2820 
2821   return os::uncommit_memory(addr, size);
2822 }
2823 
2824 static address _highest_vm_reserved_address = NULL;
2825 
2826 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2827 // at 'requested_addr'. If there are existing memory mappings at the same
2828 // location, however, they will be overwritten. If 'fixed' is false,
2829 // 'requested_addr' is only treated as a hint, the return value may or
2830 // may not start from the requested address. Unlike Linux mmap(), this
2831 // function returns NULL to indicate failure.
2832 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2833   char * addr;
2834   int flags;
2835 
2836   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2837   if (fixed) {
2838     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2839     flags |= MAP_FIXED;
2840   }
2841 
2842   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2843   // to PROT_EXEC if executable when we commit the page.
2844   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2845                        flags, -1, 0);
2846 
2847   if (addr != MAP_FAILED) {
2848     // anon_mmap() should only get called during VM initialization,
2849     // don't need lock (actually we can skip locking even it can be called
2850     // from multiple threads, because _highest_vm_reserved_address is just a
2851     // hint about the upper limit of non-stack memory regions.)
2852     if ((address)addr + bytes > _highest_vm_reserved_address) {
2853       _highest_vm_reserved_address = (address)addr + bytes;
2854     }
2855   }
2856 
2857   return addr == MAP_FAILED ? NULL : addr;
2858 }
2859 
2860 // Don't update _highest_vm_reserved_address, because there might be memory
2861 // regions above addr + size. If so, releasing a memory region only creates
2862 // a hole in the address space, it doesn't help prevent heap-stack collision.
2863 //
2864 static int anon_munmap(char * addr, size_t size) {
2865   return ::munmap(addr, size) == 0;
2866 }
2867 
2868 char* os::reserve_memory(size_t bytes, char* requested_addr,
2869                          size_t alignment_hint) {
2870   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2871 }
2872 
2873 bool os::release_memory(char* addr, size_t size) {
2874   return anon_munmap(addr, size);
2875 }
2876 
2877 static address highest_vm_reserved_address() {
2878   return _highest_vm_reserved_address;
2879 }
2880 
2881 static bool linux_mprotect(char* addr, size_t size, int prot) {
2882   // Linux wants the mprotect address argument to be page aligned.
2883   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2884 
2885   // According to SUSv3, mprotect() should only be used with mappings
2886   // established by mmap(), and mmap() always maps whole pages. Unaligned
2887   // 'addr' likely indicates problem in the VM (e.g. trying to change
2888   // protection of malloc'ed or statically allocated memory). Check the
2889   // caller if you hit this assert.
2890   assert(addr == bottom, "sanity check");
2891 
2892   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2893   return ::mprotect(bottom, size, prot) == 0;
2894 }
2895 
2896 // Set protections specified
2897 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2898                         bool is_committed) {
2899   unsigned int p = 0;
2900   switch (prot) {
2901   case MEM_PROT_NONE: p = PROT_NONE; break;
2902   case MEM_PROT_READ: p = PROT_READ; break;
2903   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2904   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2905   default:
2906     ShouldNotReachHere();
2907   }
2908   // is_committed is unused.
2909   return linux_mprotect(addr, bytes, p);
2910 }
2911 
2912 bool os::guard_memory(char* addr, size_t size) {
2913   return linux_mprotect(addr, size, PROT_NONE);
2914 }
2915 
2916 bool os::unguard_memory(char* addr, size_t size) {
2917   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2918 }
2919 
2920 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2921   bool result = false;
2922   void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2923                   MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2924                   -1, 0);
2925 
2926   if (p != (void *) -1) {
2927     // We don't know if this really is a huge page or not.
2928     FILE *fp = fopen("/proc/self/maps", "r");
2929     if (fp) {
2930       while (!feof(fp)) {
2931         char chars[257];
2932         long x = 0;
2933         if (fgets(chars, sizeof(chars), fp)) {
2934           if (sscanf(chars, "%lx-%*x", &x) == 1
2935               && x == (long)p) {
2936             if (strstr (chars, "hugepage")) {
2937               result = true;
2938               break;
2939             }
2940           }
2941         }
2942       }
2943       fclose(fp);
2944     }
2945     munmap (p, page_size);
2946     if (result)
2947       return true;
2948   }
2949 
2950   if (warn) {
2951     warning("HugeTLBFS is not supported by the operating system.");
2952   }
2953 
2954   return result;
2955 }
2956 
2957 /*
2958 * Set the coredump_filter bits to include largepages in core dump (bit 6)
2959 *
2960 * From the coredump_filter documentation:
2961 *
2962 * - (bit 0) anonymous private memory
2963 * - (bit 1) anonymous shared memory
2964 * - (bit 2) file-backed private memory
2965 * - (bit 3) file-backed shared memory
2966 * - (bit 4) ELF header pages in file-backed private memory areas (it is
2967 *           effective only if the bit 2 is cleared)
2968 * - (bit 5) hugetlb private memory
2969 * - (bit 6) hugetlb shared memory
2970 */
2971 static void set_coredump_filter(void) {
2972   FILE *f;
2973   long cdm;
2974 
2975   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
2976     return;
2977   }
2978 
2979   if (fscanf(f, "%lx", &cdm) != 1) {
2980     fclose(f);
2981     return;
2982   }
2983 
2984   rewind(f);
2985 
2986   if ((cdm & LARGEPAGES_BIT) == 0) {
2987     cdm |= LARGEPAGES_BIT;
2988     fprintf(f, "%#lx", cdm);
2989   }
2990 
2991   fclose(f);
2992 }
2993 
2994 // Large page support
2995 
2996 static size_t _large_page_size = 0;
2997 
2998 void os::large_page_init() {
2999   if (!UseLargePages) {
3000     UseHugeTLBFS = false;
3001     UseSHM = false;
3002     return;
3003   }
3004 
3005   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3006     // If UseLargePages is specified on the command line try both methods,
3007     // if it's default, then try only HugeTLBFS.
3008     if (FLAG_IS_DEFAULT(UseLargePages)) {
3009       UseHugeTLBFS = true;
3010     } else {
3011       UseHugeTLBFS = UseSHM = true;
3012     }
3013   }
3014 
3015   if (LargePageSizeInBytes) {
3016     _large_page_size = LargePageSizeInBytes;
3017   } else {
3018     // large_page_size on Linux is used to round up heap size. x86 uses either
3019     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3020     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3021     // page as large as 256M.
3022     //
3023     // Here we try to figure out page size by parsing /proc/meminfo and looking
3024     // for a line with the following format:
3025     //    Hugepagesize:     2048 kB
3026     //
3027     // If we can't determine the value (e.g. /proc is not mounted, or the text
3028     // format has been changed), we'll use the largest page size supported by
3029     // the processor.
3030 
3031 #ifndef ZERO
3032     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3033                        ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3034 #endif // ZERO
3035 
3036     FILE *fp = fopen("/proc/meminfo", "r");
3037     if (fp) {
3038       while (!feof(fp)) {
3039         int x = 0;
3040         char buf[16];
3041         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3042           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3043             _large_page_size = x * K;
3044             break;
3045           }
3046         } else {
3047           // skip to next line
3048           for (;;) {
3049             int ch = fgetc(fp);
3050             if (ch == EOF || ch == (int)'\n') break;
3051           }
3052         }
3053       }
3054       fclose(fp);
3055     }
3056   }
3057 
3058   // print a warning if any large page related flag is specified on command line
3059   bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3060 
3061   const size_t default_page_size = (size_t)Linux::page_size();
3062   if (_large_page_size > default_page_size) {
3063     _page_sizes[0] = _large_page_size;
3064     _page_sizes[1] = default_page_size;
3065     _page_sizes[2] = 0;
3066   }
3067   UseHugeTLBFS = UseHugeTLBFS &&
3068                  Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3069 
3070   if (UseHugeTLBFS)
3071     UseSHM = false;
3072 
3073   UseLargePages = UseHugeTLBFS || UseSHM;
3074 
3075   set_coredump_filter();
3076 }
3077 
3078 #ifndef SHM_HUGETLB
3079 #define SHM_HUGETLB 04000
3080 #endif
3081 
3082 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3083   // "exec" is passed in but not used.  Creating the shared image for
3084   // the code cache doesn't have an SHM_X executable permission to check.
3085   assert(UseLargePages && UseSHM, "only for SHM large pages");
3086 
3087   key_t key = IPC_PRIVATE;
3088   char *addr;
3089 
3090   bool warn_on_failure = UseLargePages &&
3091                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3092                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3093                         );
3094   char msg[128];
3095 
3096   // Create a large shared memory region to attach to based on size.
3097   // Currently, size is the total size of the heap
3098   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3099   if (shmid == -1) {
3100      // Possible reasons for shmget failure:
3101      // 1. shmmax is too small for Java heap.
3102      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3103      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3104      // 2. not enough large page memory.
3105      //    > check available large pages: cat /proc/meminfo
3106      //    > increase amount of large pages:
3107      //          echo new_value > /proc/sys/vm/nr_hugepages
3108      //      Note 1: different Linux may use different name for this property,
3109      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3110      //      Note 2: it's possible there's enough physical memory available but
3111      //            they are so fragmented after a long run that they can't
3112      //            coalesce into large pages. Try to reserve large pages when
3113      //            the system is still "fresh".
3114      if (warn_on_failure) {
3115        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3116        warning(msg);
3117      }
3118      return NULL;
3119   }
3120 
3121   // attach to the region
3122   addr = (char*)shmat(shmid, req_addr, 0);
3123   int err = errno;
3124 
3125   // Remove shmid. If shmat() is successful, the actual shared memory segment
3126   // will be deleted when it's detached by shmdt() or when the process
3127   // terminates. If shmat() is not successful this will remove the shared
3128   // segment immediately.
3129   shmctl(shmid, IPC_RMID, NULL);
3130 
3131   if ((intptr_t)addr == -1) {
3132      if (warn_on_failure) {
3133        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3134        warning(msg);
3135      }
3136      return NULL;
3137   }
3138 
3139   if ((addr != NULL) && UseNUMAInterleaving) {
3140     numa_make_global(addr, bytes);
3141   }
3142 
3143   return addr;
3144 }
3145 
3146 bool os::release_memory_special(char* base, size_t bytes) {
3147   // detaching the SHM segment will also delete it, see reserve_memory_special()
3148   int rslt = shmdt(base);
3149   return rslt == 0;
3150 }
3151 
3152 size_t os::large_page_size() {
3153   return _large_page_size;
3154 }
3155 
3156 // HugeTLBFS allows application to commit large page memory on demand;
3157 // with SysV SHM the entire memory region must be allocated as shared
3158 // memory.
3159 bool os::can_commit_large_page_memory() {
3160   return UseHugeTLBFS;
3161 }
3162 
3163 bool os::can_execute_large_page_memory() {
3164   return UseHugeTLBFS;
3165 }
3166 
3167 // Reserve memory at an arbitrary address, only if that area is
3168 // available (and not reserved for something else).
3169 
3170 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3171   const int max_tries = 10;
3172   char* base[max_tries];
3173   size_t size[max_tries];
3174   const size_t gap = 0x000000;
3175 
3176   // Assert only that the size is a multiple of the page size, since
3177   // that's all that mmap requires, and since that's all we really know
3178   // about at this low abstraction level.  If we need higher alignment,
3179   // we can either pass an alignment to this method or verify alignment
3180   // in one of the methods further up the call chain.  See bug 5044738.
3181   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3182 
3183   // Repeatedly allocate blocks until the block is allocated at the
3184   // right spot. Give up after max_tries. Note that reserve_memory() will
3185   // automatically update _highest_vm_reserved_address if the call is
3186   // successful. The variable tracks the highest memory address every reserved
3187   // by JVM. It is used to detect heap-stack collision if running with
3188   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3189   // space than needed, it could confuse the collision detecting code. To
3190   // solve the problem, save current _highest_vm_reserved_address and
3191   // calculate the correct value before return.
3192   address old_highest = _highest_vm_reserved_address;
3193 
3194   // Linux mmap allows caller to pass an address as hint; give it a try first,
3195   // if kernel honors the hint then we can return immediately.
3196   char * addr = anon_mmap(requested_addr, bytes, false);
3197   if (addr == requested_addr) {
3198      return requested_addr;
3199   }
3200 
3201   if (addr != NULL) {
3202      // mmap() is successful but it fails to reserve at the requested address
3203      anon_munmap(addr, bytes);
3204   }
3205 
3206   int i;
3207   for (i = 0; i < max_tries; ++i) {
3208     base[i] = reserve_memory(bytes);
3209 
3210     if (base[i] != NULL) {
3211       // Is this the block we wanted?
3212       if (base[i] == requested_addr) {
3213         size[i] = bytes;
3214         break;
3215       }
3216 
3217       // Does this overlap the block we wanted? Give back the overlapped
3218       // parts and try again.
3219 
3220       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3221       if (top_overlap >= 0 && top_overlap < bytes) {
3222         unmap_memory(base[i], top_overlap);
3223         base[i] += top_overlap;
3224         size[i] = bytes - top_overlap;
3225       } else {
3226         size_t bottom_overlap = base[i] + bytes - requested_addr;
3227         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3228           unmap_memory(requested_addr, bottom_overlap);
3229           size[i] = bytes - bottom_overlap;
3230         } else {
3231           size[i] = bytes;
3232         }
3233       }
3234     }
3235   }
3236 
3237   // Give back the unused reserved pieces.
3238 
3239   for (int j = 0; j < i; ++j) {
3240     if (base[j] != NULL) {
3241       unmap_memory(base[j], size[j]);
3242     }
3243   }
3244 
3245   if (i < max_tries) {
3246     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3247     return requested_addr;
3248   } else {
3249     _highest_vm_reserved_address = old_highest;
3250     return NULL;
3251   }
3252 }
3253 
3254 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3255   return ::read(fd, buf, nBytes);
3256 }
3257 
3258 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3259 // Solaris uses poll(), linux uses park().
3260 // Poll() is likely a better choice, assuming that Thread.interrupt()
3261 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3262 // SIGSEGV, see 4355769.
3263 
3264 const int NANOSECS_PER_MILLISECS = 1000000;
3265 
3266 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3267   assert(thread == Thread::current(),  "thread consistency check");
3268 
3269   ParkEvent * const slp = thread->_SleepEvent ;
3270   slp->reset() ;
3271   OrderAccess::fence() ;
3272 
3273   if (interruptible) {
3274     jlong prevtime = javaTimeNanos();
3275 
3276     for (;;) {
3277       if (os::is_interrupted(thread, true)) {
3278         return OS_INTRPT;
3279       }
3280 
3281       jlong newtime = javaTimeNanos();
3282 
3283       if (newtime - prevtime < 0) {
3284         // time moving backwards, should only happen if no monotonic clock
3285         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3286         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3287       } else {
3288         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3289       }
3290 
3291       if(millis <= 0) {
3292         return OS_OK;
3293       }
3294 
3295       prevtime = newtime;
3296 
3297       {
3298         assert(thread->is_Java_thread(), "sanity check");
3299         JavaThread *jt = (JavaThread *) thread;
3300         ThreadBlockInVM tbivm(jt);
3301         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3302 
3303         jt->set_suspend_equivalent();
3304         // cleared by handle_special_suspend_equivalent_condition() or
3305         // java_suspend_self() via check_and_wait_while_suspended()
3306 
3307         slp->park(millis);
3308 
3309         // were we externally suspended while we were waiting?
3310         jt->check_and_wait_while_suspended();
3311       }
3312     }
3313   } else {
3314     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3315     jlong prevtime = javaTimeNanos();
3316 
3317     for (;;) {
3318       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3319       // the 1st iteration ...
3320       jlong newtime = javaTimeNanos();
3321 
3322       if (newtime - prevtime < 0) {
3323         // time moving backwards, should only happen if no monotonic clock
3324         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3325         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3326       } else {
3327         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3328       }
3329 
3330       if(millis <= 0) break ;
3331 
3332       prevtime = newtime;
3333       slp->park(millis);
3334     }
3335     return OS_OK ;
3336   }
3337 }
3338 
3339 int os::naked_sleep() {
3340   // %% make the sleep time an integer flag. for now use 1 millisec.
3341   return os::sleep(Thread::current(), 1, false);
3342 }
3343 
3344 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3345 void os::infinite_sleep() {
3346   while (true) {    // sleep forever ...
3347     ::sleep(100);   // ... 100 seconds at a time
3348   }
3349 }
3350 
3351 // Used to convert frequent JVM_Yield() to nops
3352 bool os::dont_yield() {
3353   return DontYieldALot;
3354 }
3355 
3356 void os::yield() {
3357   sched_yield();
3358 }
3359 
3360 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3361 
3362 void os::yield_all(int attempts) {
3363   // Yields to all threads, including threads with lower priorities
3364   // Threads on Linux are all with same priority. The Solaris style
3365   // os::yield_all() with nanosleep(1ms) is not necessary.
3366   sched_yield();
3367 }
3368 
3369 // Called from the tight loops to possibly influence time-sharing heuristics
3370 void os::loop_breaker(int attempts) {
3371   os::yield_all(attempts);
3372 }
3373 
3374 ////////////////////////////////////////////////////////////////////////////////
3375 // thread priority support
3376 
3377 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3378 // only supports dynamic priority, static priority must be zero. For real-time
3379 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3380 // However, for large multi-threaded applications, SCHED_RR is not only slower
3381 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3382 // of 5 runs - Sep 2005).
3383 //
3384 // The following code actually changes the niceness of kernel-thread/LWP. It
3385 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3386 // not the entire user process, and user level threads are 1:1 mapped to kernel
3387 // threads. It has always been the case, but could change in the future. For
3388 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3389 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3390 
3391 int os::java_to_os_priority[MaxPriority + 1] = {
3392   19,              // 0 Entry should never be used
3393 
3394    4,              // 1 MinPriority
3395    3,              // 2
3396    2,              // 3
3397 
3398    1,              // 4
3399    0,              // 5 NormPriority
3400   -1,              // 6
3401 
3402   -2,              // 7
3403   -3,              // 8
3404   -4,              // 9 NearMaxPriority
3405 
3406   -5               // 10 MaxPriority
3407 };
3408 
3409 static int prio_init() {
3410   if (ThreadPriorityPolicy == 1) {
3411     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3412     // if effective uid is not root. Perhaps, a more elegant way of doing
3413     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3414     if (geteuid() != 0) {
3415       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3416         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3417       }
3418       ThreadPriorityPolicy = 0;
3419     }
3420   }
3421   return 0;
3422 }
3423 
3424 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3425   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3426 
3427   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3428   return (ret == 0) ? OS_OK : OS_ERR;
3429 }
3430 
3431 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3432   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3433     *priority_ptr = java_to_os_priority[NormPriority];
3434     return OS_OK;
3435   }
3436 
3437   errno = 0;
3438   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3439   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3440 }
3441 
3442 // Hint to the underlying OS that a task switch would not be good.
3443 // Void return because it's a hint and can fail.
3444 void os::hint_no_preempt() {}
3445 
3446 ////////////////////////////////////////////////////////////////////////////////
3447 // suspend/resume support
3448 
3449 //  the low-level signal-based suspend/resume support is a remnant from the
3450 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3451 //  within hotspot. Now there is a single use-case for this:
3452 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3453 //      that runs in the watcher thread.
3454 //  The remaining code is greatly simplified from the more general suspension
3455 //  code that used to be used.
3456 //
3457 //  The protocol is quite simple:
3458 //  - suspend:
3459 //      - sends a signal to the target thread
3460 //      - polls the suspend state of the osthread using a yield loop
3461 //      - target thread signal handler (SR_handler) sets suspend state
3462 //        and blocks in sigsuspend until continued
3463 //  - resume:
3464 //      - sets target osthread state to continue
3465 //      - sends signal to end the sigsuspend loop in the SR_handler
3466 //
3467 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3468 //
3469 
3470 static void resume_clear_context(OSThread *osthread) {
3471   osthread->set_ucontext(NULL);
3472   osthread->set_siginfo(NULL);
3473 
3474   // notify the suspend action is completed, we have now resumed
3475   osthread->sr.clear_suspended();
3476 }
3477 
3478 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3479   osthread->set_ucontext(context);
3480   osthread->set_siginfo(siginfo);
3481 }
3482 
3483 //
3484 // Handler function invoked when a thread's execution is suspended or
3485 // resumed. We have to be careful that only async-safe functions are
3486 // called here (Note: most pthread functions are not async safe and
3487 // should be avoided.)
3488 //
3489 // Note: sigwait() is a more natural fit than sigsuspend() from an
3490 // interface point of view, but sigwait() prevents the signal hander
3491 // from being run. libpthread would get very confused by not having
3492 // its signal handlers run and prevents sigwait()'s use with the
3493 // mutex granting granting signal.
3494 //
3495 // Currently only ever called on the VMThread
3496 //
3497 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3498   // Save and restore errno to avoid confusing native code with EINTR
3499   // after sigsuspend.
3500   int old_errno = errno;
3501 
3502   Thread* thread = Thread::current();
3503   OSThread* osthread = thread->osthread();
3504   assert(thread->is_VM_thread(), "Must be VMThread");
3505   // read current suspend action
3506   int action = osthread->sr.suspend_action();
3507   if (action == SR_SUSPEND) {
3508     suspend_save_context(osthread, siginfo, context);
3509 
3510     // Notify the suspend action is about to be completed. do_suspend()
3511     // waits until SR_SUSPENDED is set and then returns. We will wait
3512     // here for a resume signal and that completes the suspend-other
3513     // action. do_suspend/do_resume is always called as a pair from
3514     // the same thread - so there are no races
3515 
3516     // notify the caller
3517     osthread->sr.set_suspended();
3518 
3519     sigset_t suspend_set;  // signals for sigsuspend()
3520 
3521     // get current set of blocked signals and unblock resume signal
3522     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3523     sigdelset(&suspend_set, SR_signum);
3524 
3525     // wait here until we are resumed
3526     do {
3527       sigsuspend(&suspend_set);
3528       // ignore all returns until we get a resume signal
3529     } while (osthread->sr.suspend_action() != SR_CONTINUE);
3530 
3531     resume_clear_context(osthread);
3532 
3533   } else {
3534     assert(action == SR_CONTINUE, "unexpected sr action");
3535     // nothing special to do - just leave the handler
3536   }
3537 
3538   errno = old_errno;
3539 }
3540 
3541 
3542 static int SR_initialize() {
3543   struct sigaction act;
3544   char *s;
3545   /* Get signal number to use for suspend/resume */
3546   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3547     int sig = ::strtol(s, 0, 10);
3548     if (sig > 0 || sig < _NSIG) {
3549         SR_signum = sig;
3550     }
3551   }
3552 
3553   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3554         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3555 
3556   sigemptyset(&SR_sigset);
3557   sigaddset(&SR_sigset, SR_signum);
3558 
3559   /* Set up signal handler for suspend/resume */
3560   act.sa_flags = SA_RESTART|SA_SIGINFO;
3561   act.sa_handler = (void (*)(int)) SR_handler;
3562 
3563   // SR_signum is blocked by default.
3564   // 4528190 - We also need to block pthread restart signal (32 on all
3565   // supported Linux platforms). Note that LinuxThreads need to block
3566   // this signal for all threads to work properly. So we don't have
3567   // to use hard-coded signal number when setting up the mask.
3568   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3569 
3570   if (sigaction(SR_signum, &act, 0) == -1) {
3571     return -1;
3572   }
3573 
3574   // Save signal flag
3575   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3576   return 0;
3577 }
3578 
3579 static int SR_finalize() {
3580   return 0;
3581 }
3582 
3583 
3584 // returns true on success and false on error - really an error is fatal
3585 // but this seems the normal response to library errors
3586 static bool do_suspend(OSThread* osthread) {
3587   // mark as suspended and send signal
3588   osthread->sr.set_suspend_action(SR_SUSPEND);
3589   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3590   assert_status(status == 0, status, "pthread_kill");
3591 
3592   // check status and wait until notified of suspension
3593   if (status == 0) {
3594     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3595       os::yield_all(i);
3596     }
3597     osthread->sr.set_suspend_action(SR_NONE);
3598     return true;
3599   }
3600   else {
3601     osthread->sr.set_suspend_action(SR_NONE);
3602     return false;
3603   }
3604 }
3605 
3606 static void do_resume(OSThread* osthread) {
3607   assert(osthread->sr.is_suspended(), "thread should be suspended");
3608   osthread->sr.set_suspend_action(SR_CONTINUE);
3609 
3610   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3611   assert_status(status == 0, status, "pthread_kill");
3612   // check status and wait unit notified of resumption
3613   if (status == 0) {
3614     for (int i = 0; osthread->sr.is_suspended(); i++) {
3615       os::yield_all(i);
3616     }
3617   }
3618   osthread->sr.set_suspend_action(SR_NONE);
3619 }
3620 
3621 ////////////////////////////////////////////////////////////////////////////////
3622 // interrupt support
3623 
3624 void os::interrupt(Thread* thread) {
3625   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3626     "possibility of dangling Thread pointer");
3627 
3628   OSThread* osthread = thread->osthread();
3629 
3630   if (!osthread->interrupted()) {
3631     osthread->set_interrupted(true);
3632     // More than one thread can get here with the same value of osthread,
3633     // resulting in multiple notifications.  We do, however, want the store
3634     // to interrupted() to be visible to other threads before we execute unpark().
3635     OrderAccess::fence();
3636     ParkEvent * const slp = thread->_SleepEvent ;
3637     if (slp != NULL) slp->unpark() ;
3638   }
3639 
3640   // For JSR166. Unpark even if interrupt status already was set
3641   if (thread->is_Java_thread())
3642     ((JavaThread*)thread)->parker()->unpark();
3643 
3644   ParkEvent * ev = thread->_ParkEvent ;
3645   if (ev != NULL) ev->unpark() ;
3646 
3647 }
3648 
3649 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3650   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3651     "possibility of dangling Thread pointer");
3652 
3653   OSThread* osthread = thread->osthread();
3654 
3655   bool interrupted = osthread->interrupted();
3656 
3657   if (interrupted && clear_interrupted) {
3658     osthread->set_interrupted(false);
3659     // consider thread->_SleepEvent->reset() ... optional optimization
3660   }
3661 
3662   return interrupted;
3663 }
3664 
3665 ///////////////////////////////////////////////////////////////////////////////////
3666 // signal handling (except suspend/resume)
3667 
3668 // This routine may be used by user applications as a "hook" to catch signals.
3669 // The user-defined signal handler must pass unrecognized signals to this
3670 // routine, and if it returns true (non-zero), then the signal handler must
3671 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3672 // routine will never retun false (zero), but instead will execute a VM panic
3673 // routine kill the process.
3674 //
3675 // If this routine returns false, it is OK to call it again.  This allows
3676 // the user-defined signal handler to perform checks either before or after
3677 // the VM performs its own checks.  Naturally, the user code would be making
3678 // a serious error if it tried to handle an exception (such as a null check
3679 // or breakpoint) that the VM was generating for its own correct operation.
3680 //
3681 // This routine may recognize any of the following kinds of signals:
3682 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3683 // It should be consulted by handlers for any of those signals.
3684 //
3685 // The caller of this routine must pass in the three arguments supplied
3686 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3687 // field of the structure passed to sigaction().  This routine assumes that
3688 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3689 //
3690 // Note that the VM will print warnings if it detects conflicting signal
3691 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3692 //
3693 extern "C" JNIEXPORT int
3694 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3695                         void* ucontext, int abort_if_unrecognized);
3696 
3697 void signalHandler(int sig, siginfo_t* info, void* uc) {
3698   assert(info != NULL && uc != NULL, "it must be old kernel");
3699   JVM_handle_linux_signal(sig, info, uc, true);
3700 }
3701 
3702 
3703 // This boolean allows users to forward their own non-matching signals
3704 // to JVM_handle_linux_signal, harmlessly.
3705 bool os::Linux::signal_handlers_are_installed = false;
3706 
3707 // For signal-chaining
3708 struct sigaction os::Linux::sigact[MAXSIGNUM];
3709 unsigned int os::Linux::sigs = 0;
3710 bool os::Linux::libjsig_is_loaded = false;
3711 typedef struct sigaction *(*get_signal_t)(int);
3712 get_signal_t os::Linux::get_signal_action = NULL;
3713 
3714 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3715   struct sigaction *actp = NULL;
3716 
3717   if (libjsig_is_loaded) {
3718     // Retrieve the old signal handler from libjsig
3719     actp = (*get_signal_action)(sig);
3720   }
3721   if (actp == NULL) {
3722     // Retrieve the preinstalled signal handler from jvm
3723     actp = get_preinstalled_handler(sig);
3724   }
3725 
3726   return actp;
3727 }
3728 
3729 static bool call_chained_handler(struct sigaction *actp, int sig,
3730                                  siginfo_t *siginfo, void *context) {
3731   // Call the old signal handler
3732   if (actp->sa_handler == SIG_DFL) {
3733     // It's more reasonable to let jvm treat it as an unexpected exception
3734     // instead of taking the default action.
3735     return false;
3736   } else if (actp->sa_handler != SIG_IGN) {
3737     if ((actp->sa_flags & SA_NODEFER) == 0) {
3738       // automaticlly block the signal
3739       sigaddset(&(actp->sa_mask), sig);
3740     }
3741 
3742     sa_handler_t hand;
3743     sa_sigaction_t sa;
3744     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3745     // retrieve the chained handler
3746     if (siginfo_flag_set) {
3747       sa = actp->sa_sigaction;
3748     } else {
3749       hand = actp->sa_handler;
3750     }
3751 
3752     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3753       actp->sa_handler = SIG_DFL;
3754     }
3755 
3756     // try to honor the signal mask
3757     sigset_t oset;
3758     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3759 
3760     // call into the chained handler
3761     if (siginfo_flag_set) {
3762       (*sa)(sig, siginfo, context);
3763     } else {
3764       (*hand)(sig);
3765     }
3766 
3767     // restore the signal mask
3768     pthread_sigmask(SIG_SETMASK, &oset, 0);
3769   }
3770   // Tell jvm's signal handler the signal is taken care of.
3771   return true;
3772 }
3773 
3774 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3775   bool chained = false;
3776   // signal-chaining
3777   if (UseSignalChaining) {
3778     struct sigaction *actp = get_chained_signal_action(sig);
3779     if (actp != NULL) {
3780       chained = call_chained_handler(actp, sig, siginfo, context);
3781     }
3782   }
3783   return chained;
3784 }
3785 
3786 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3787   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3788     return &sigact[sig];
3789   }
3790   return NULL;
3791 }
3792 
3793 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3794   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3795   sigact[sig] = oldAct;
3796   sigs |= (unsigned int)1 << sig;
3797 }
3798 
3799 // for diagnostic
3800 int os::Linux::sigflags[MAXSIGNUM];
3801 
3802 int os::Linux::get_our_sigflags(int sig) {
3803   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3804   return sigflags[sig];
3805 }
3806 
3807 void os::Linux::set_our_sigflags(int sig, int flags) {
3808   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3809   sigflags[sig] = flags;
3810 }
3811 
3812 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3813   // Check for overwrite.
3814   struct sigaction oldAct;
3815   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3816 
3817   void* oldhand = oldAct.sa_sigaction
3818                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3819                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3820   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3821       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3822       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3823     if (AllowUserSignalHandlers || !set_installed) {
3824       // Do not overwrite; user takes responsibility to forward to us.
3825       return;
3826     } else if (UseSignalChaining) {
3827       // save the old handler in jvm
3828       save_preinstalled_handler(sig, oldAct);
3829       // libjsig also interposes the sigaction() call below and saves the
3830       // old sigaction on it own.
3831     } else {
3832       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3833                     "%#lx for signal %d.", (long)oldhand, sig));
3834     }
3835   }
3836 
3837   struct sigaction sigAct;
3838   sigfillset(&(sigAct.sa_mask));
3839   sigAct.sa_handler = SIG_DFL;
3840   if (!set_installed) {
3841     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3842   } else {
3843     sigAct.sa_sigaction = signalHandler;
3844     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3845   }
3846   // Save flags, which are set by ours
3847   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3848   sigflags[sig] = sigAct.sa_flags;
3849 
3850   int ret = sigaction(sig, &sigAct, &oldAct);
3851   assert(ret == 0, "check");
3852 
3853   void* oldhand2  = oldAct.sa_sigaction
3854                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3855                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3856   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3857 }
3858 
3859 // install signal handlers for signals that HotSpot needs to
3860 // handle in order to support Java-level exception handling.
3861 
3862 void os::Linux::install_signal_handlers() {
3863   if (!signal_handlers_are_installed) {
3864     signal_handlers_are_installed = true;
3865 
3866     // signal-chaining
3867     typedef void (*signal_setting_t)();
3868     signal_setting_t begin_signal_setting = NULL;
3869     signal_setting_t end_signal_setting = NULL;
3870     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3871                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3872     if (begin_signal_setting != NULL) {
3873       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3874                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3875       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3876                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3877       libjsig_is_loaded = true;
3878       assert(UseSignalChaining, "should enable signal-chaining");
3879     }
3880     if (libjsig_is_loaded) {
3881       // Tell libjsig jvm is setting signal handlers
3882       (*begin_signal_setting)();
3883     }
3884 
3885     set_signal_handler(SIGSEGV, true);
3886     set_signal_handler(SIGPIPE, true);
3887     set_signal_handler(SIGBUS, true);
3888     set_signal_handler(SIGILL, true);
3889     set_signal_handler(SIGFPE, true);
3890     set_signal_handler(SIGXFSZ, true);
3891 
3892     if (libjsig_is_loaded) {
3893       // Tell libjsig jvm finishes setting signal handlers
3894       (*end_signal_setting)();
3895     }
3896 
3897     // We don't activate signal checker if libjsig is in place, we trust ourselves
3898     // and if UserSignalHandler is installed all bets are off
3899     if (CheckJNICalls) {
3900       if (libjsig_is_loaded) {
3901         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3902         check_signals = false;
3903       }
3904       if (AllowUserSignalHandlers) {
3905         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3906         check_signals = false;
3907       }
3908     }
3909   }
3910 }
3911 
3912 // This is the fastest way to get thread cpu time on Linux.
3913 // Returns cpu time (user+sys) for any thread, not only for current.
3914 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3915 // It might work on 2.6.10+ with a special kernel/glibc patch.
3916 // For reference, please, see IEEE Std 1003.1-2004:
3917 //   http://www.unix.org/single_unix_specification
3918 
3919 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3920   struct timespec tp;
3921   int rc = os::Linux::clock_gettime(clockid, &tp);
3922   assert(rc == 0, "clock_gettime is expected to return 0 code");
3923 
3924   return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3925 }
3926 
3927 /////
3928 // glibc on Linux platform uses non-documented flag
3929 // to indicate, that some special sort of signal
3930 // trampoline is used.
3931 // We will never set this flag, and we should
3932 // ignore this flag in our diagnostic
3933 #ifdef SIGNIFICANT_SIGNAL_MASK
3934 #undef SIGNIFICANT_SIGNAL_MASK
3935 #endif
3936 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3937 
3938 static const char* get_signal_handler_name(address handler,
3939                                            char* buf, int buflen) {
3940   int offset;
3941   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3942   if (found) {
3943     // skip directory names
3944     const char *p1, *p2;
3945     p1 = buf;
3946     size_t len = strlen(os::file_separator());
3947     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3948     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3949   } else {
3950     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3951   }
3952   return buf;
3953 }
3954 
3955 static void print_signal_handler(outputStream* st, int sig,
3956                                  char* buf, size_t buflen) {
3957   struct sigaction sa;
3958 
3959   sigaction(sig, NULL, &sa);
3960 
3961   // See comment for SIGNIFICANT_SIGNAL_MASK define
3962   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3963 
3964   st->print("%s: ", os::exception_name(sig, buf, buflen));
3965 
3966   address handler = (sa.sa_flags & SA_SIGINFO)
3967     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3968     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3969 
3970   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3971     st->print("SIG_DFL");
3972   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3973     st->print("SIG_IGN");
3974   } else {
3975     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3976   }
3977 
3978   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3979 
3980   address rh = VMError::get_resetted_sighandler(sig);
3981   // May be, handler was resetted by VMError?
3982   if(rh != NULL) {
3983     handler = rh;
3984     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3985   }
3986 
3987   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
3988 
3989   // Check: is it our handler?
3990   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3991      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3992     // It is our signal handler
3993     // check for flags, reset system-used one!
3994     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3995       st->print(
3996                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3997                 os::Linux::get_our_sigflags(sig));
3998     }
3999   }
4000   st->cr();
4001 }
4002 
4003 
4004 #define DO_SIGNAL_CHECK(sig) \
4005   if (!sigismember(&check_signal_done, sig)) \
4006     os::Linux::check_signal_handler(sig)
4007 
4008 // This method is a periodic task to check for misbehaving JNI applications
4009 // under CheckJNI, we can add any periodic checks here
4010 
4011 void os::run_periodic_checks() {
4012 
4013   if (check_signals == false) return;
4014 
4015   // SEGV and BUS if overridden could potentially prevent
4016   // generation of hs*.log in the event of a crash, debugging
4017   // such a case can be very challenging, so we absolutely
4018   // check the following for a good measure:
4019   DO_SIGNAL_CHECK(SIGSEGV);
4020   DO_SIGNAL_CHECK(SIGILL);
4021   DO_SIGNAL_CHECK(SIGFPE);
4022   DO_SIGNAL_CHECK(SIGBUS);
4023   DO_SIGNAL_CHECK(SIGPIPE);
4024   DO_SIGNAL_CHECK(SIGXFSZ);
4025 
4026 
4027   // ReduceSignalUsage allows the user to override these handlers
4028   // see comments at the very top and jvm_solaris.h
4029   if (!ReduceSignalUsage) {
4030     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4031     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4032     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4033     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4034   }
4035 
4036   DO_SIGNAL_CHECK(SR_signum);
4037   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4038 }
4039 
4040 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4041 
4042 static os_sigaction_t os_sigaction = NULL;
4043 
4044 void os::Linux::check_signal_handler(int sig) {
4045   char buf[O_BUFLEN];
4046   address jvmHandler = NULL;
4047 
4048 
4049   struct sigaction act;
4050   if (os_sigaction == NULL) {
4051     // only trust the default sigaction, in case it has been interposed
4052     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4053     if (os_sigaction == NULL) return;
4054   }
4055 
4056   os_sigaction(sig, (struct sigaction*)NULL, &act);
4057 
4058 
4059   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4060 
4061   address thisHandler = (act.sa_flags & SA_SIGINFO)
4062     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4063     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4064 
4065 
4066   switch(sig) {
4067   case SIGSEGV:
4068   case SIGBUS:
4069   case SIGFPE:
4070   case SIGPIPE:
4071   case SIGILL:
4072   case SIGXFSZ:
4073     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4074     break;
4075 
4076   case SHUTDOWN1_SIGNAL:
4077   case SHUTDOWN2_SIGNAL:
4078   case SHUTDOWN3_SIGNAL:
4079   case BREAK_SIGNAL:
4080     jvmHandler = (address)user_handler();
4081     break;
4082 
4083   case INTERRUPT_SIGNAL:
4084     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4085     break;
4086 
4087   default:
4088     if (sig == SR_signum) {
4089       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4090     } else {
4091       return;
4092     }
4093     break;
4094   }
4095 
4096   if (thisHandler != jvmHandler) {
4097     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4098     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4099     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4100     // No need to check this sig any longer
4101     sigaddset(&check_signal_done, sig);
4102   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4103     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4104     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4105     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4106     // No need to check this sig any longer
4107     sigaddset(&check_signal_done, sig);
4108   }
4109 
4110   // Dump all the signal
4111   if (sigismember(&check_signal_done, sig)) {
4112     print_signal_handlers(tty, buf, O_BUFLEN);
4113   }
4114 }
4115 
4116 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4117 
4118 extern bool signal_name(int signo, char* buf, size_t len);
4119 
4120 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4121   if (0 < exception_code && exception_code <= SIGRTMAX) {
4122     // signal
4123     if (!signal_name(exception_code, buf, size)) {
4124       jio_snprintf(buf, size, "SIG%d", exception_code);
4125     }
4126     return buf;
4127   } else {
4128     return NULL;
4129   }
4130 }
4131 
4132 // this is called _before_ the most of global arguments have been parsed
4133 void os::init(void) {
4134   char dummy;   /* used to get a guess on initial stack address */
4135 //  first_hrtime = gethrtime();
4136 
4137   // With LinuxThreads the JavaMain thread pid (primordial thread)
4138   // is different than the pid of the java launcher thread.
4139   // So, on Linux, the launcher thread pid is passed to the VM
4140   // via the sun.java.launcher.pid property.
4141   // Use this property instead of getpid() if it was correctly passed.
4142   // See bug 6351349.
4143   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4144 
4145   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4146 
4147   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4148 
4149   init_random(1234567);
4150 
4151   ThreadCritical::initialize();
4152 
4153   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4154   if (Linux::page_size() == -1) {
4155     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4156                   strerror(errno)));
4157   }
4158   init_page_sizes((size_t) Linux::page_size());
4159 
4160   Linux::initialize_system_info();
4161 
4162   // main_thread points to the aboriginal thread
4163   Linux::_main_thread = pthread_self();
4164 
4165   Linux::clock_init();
4166   initial_time_count = os::elapsed_counter();
4167   pthread_mutex_init(&dl_mutex, NULL);
4168 }
4169 
4170 // To install functions for atexit system call
4171 extern "C" {
4172   static void perfMemory_exit_helper() {
4173     perfMemory_exit();
4174   }
4175 }
4176 
4177 // this is called _after_ the global arguments have been parsed
4178 jint os::init_2(void)
4179 {
4180   Linux::fast_thread_clock_init();
4181 
4182   // Allocate a single page and mark it as readable for safepoint polling
4183   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4184   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4185 
4186   os::set_polling_page( polling_page );
4187 
4188 #ifndef PRODUCT
4189   if(Verbose && PrintMiscellaneous)
4190     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4191 #endif
4192 
4193   if (!UseMembar) {
4194     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4195     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4196     os::set_memory_serialize_page( mem_serialize_page );
4197 
4198 #ifndef PRODUCT
4199     if(Verbose && PrintMiscellaneous)
4200       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4201 #endif
4202   }
4203 
4204   os::large_page_init();
4205 
4206   // initialize suspend/resume support - must do this before signal_sets_init()
4207   if (SR_initialize() != 0) {
4208     perror("SR_initialize failed");
4209     return JNI_ERR;
4210   }
4211 
4212   Linux::signal_sets_init();
4213   Linux::install_signal_handlers();
4214 
4215   // Check minimum allowable stack size for thread creation and to initialize
4216   // the java system classes, including StackOverflowError - depends on page
4217   // size.  Add a page for compiler2 recursion in main thread.
4218   // Add in 2*BytesPerWord times page size to account for VM stack during
4219   // class initialization depending on 32 or 64 bit VM.
4220   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4221             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4222                     2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4223 
4224   size_t threadStackSizeInBytes = ThreadStackSize * K;
4225   if (threadStackSizeInBytes != 0 &&
4226       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4227         tty->print_cr("\nThe stack size specified is too small, "
4228                       "Specify at least %dk",
4229                       os::Linux::min_stack_allowed/ K);
4230         return JNI_ERR;
4231   }
4232 
4233   // Make the stack size a multiple of the page size so that
4234   // the yellow/red zones can be guarded.
4235   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4236         vm_page_size()));
4237 
4238   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4239 
4240   Linux::libpthread_init();
4241   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4242      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4243           Linux::glibc_version(), Linux::libpthread_version(),
4244           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4245   }
4246 
4247   if (UseNUMA) {
4248     if (!Linux::libnuma_init()) {
4249       UseNUMA = false;
4250     } else {
4251       if ((Linux::numa_max_node() < 1)) {
4252         // There's only one node(they start from 0), disable NUMA.
4253         UseNUMA = false;
4254       }
4255     }
4256     // With SHM large pages we cannot uncommit a page, so there's not way
4257     // we can make the adaptive lgrp chunk resizing work. If the user specified
4258     // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4259     // disable adaptive resizing.
4260     if (UseNUMA && UseLargePages && UseSHM) {
4261       if (!FLAG_IS_DEFAULT(UseNUMA)) {
4262         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4263           UseLargePages = false;
4264         } else {
4265           warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4266           UseAdaptiveSizePolicy = false;
4267           UseAdaptiveNUMAChunkSizing = false;
4268         }
4269       } else {
4270         UseNUMA = false;
4271       }
4272     }
4273     if (!UseNUMA && ForceNUMA) {
4274       UseNUMA = true;
4275     }
4276   }
4277 
4278   if (MaxFDLimit) {
4279     // set the number of file descriptors to max. print out error
4280     // if getrlimit/setrlimit fails but continue regardless.
4281     struct rlimit nbr_files;
4282     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4283     if (status != 0) {
4284       if (PrintMiscellaneous && (Verbose || WizardMode))
4285         perror("os::init_2 getrlimit failed");
4286     } else {
4287       nbr_files.rlim_cur = nbr_files.rlim_max;
4288       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4289       if (status != 0) {
4290         if (PrintMiscellaneous && (Verbose || WizardMode))
4291           perror("os::init_2 setrlimit failed");
4292       }
4293     }
4294   }
4295 
4296   // Initialize lock used to serialize thread creation (see os::create_thread)
4297   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4298 
4299   // at-exit methods are called in the reverse order of their registration.
4300   // atexit functions are called on return from main or as a result of a
4301   // call to exit(3C). There can be only 32 of these functions registered
4302   // and atexit() does not set errno.
4303 
4304   if (PerfAllowAtExitRegistration) {
4305     // only register atexit functions if PerfAllowAtExitRegistration is set.
4306     // atexit functions can be delayed until process exit time, which
4307     // can be problematic for embedded VM situations. Embedded VMs should
4308     // call DestroyJavaVM() to assure that VM resources are released.
4309 
4310     // note: perfMemory_exit_helper atexit function may be removed in
4311     // the future if the appropriate cleanup code can be added to the
4312     // VM_Exit VMOperation's doit method.
4313     if (atexit(perfMemory_exit_helper) != 0) {
4314       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4315     }
4316   }
4317 
4318   // initialize thread priority policy
4319   prio_init();
4320 
4321   return JNI_OK;
4322 }
4323 
4324 // this is called at the end of vm_initialization
4325 void os::init_3(void)
4326 {
4327 #ifdef JAVASE_EMBEDDED
4328   // Start the MemNotifyThread
4329   if (LowMemoryProtection) {
4330     MemNotifyThread::start();
4331   }
4332   return;
4333 #endif
4334 }
4335 
4336 // Mark the polling page as unreadable
4337 void os::make_polling_page_unreadable(void) {
4338   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4339     fatal("Could not disable polling page");
4340 };
4341 
4342 // Mark the polling page as readable
4343 void os::make_polling_page_readable(void) {
4344   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4345     fatal("Could not enable polling page");
4346   }
4347 };
4348 
4349 int os::active_processor_count() {
4350   // Linux doesn't yet have a (official) notion of processor sets,
4351   // so just return the number of online processors.
4352   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4353   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4354   return online_cpus;
4355 }
4356 
4357 bool os::distribute_processes(uint length, uint* distribution) {
4358   // Not yet implemented.
4359   return false;
4360 }
4361 
4362 bool os::bind_to_processor(uint processor_id) {
4363   // Not yet implemented.
4364   return false;
4365 }
4366 
4367 ///
4368 
4369 // Suspends the target using the signal mechanism and then grabs the PC before
4370 // resuming the target. Used by the flat-profiler only
4371 ExtendedPC os::get_thread_pc(Thread* thread) {
4372   // Make sure that it is called by the watcher for the VMThread
4373   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4374   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4375 
4376   ExtendedPC epc;
4377 
4378   OSThread* osthread = thread->osthread();
4379   if (do_suspend(osthread)) {
4380     if (osthread->ucontext() != NULL) {
4381       epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4382     } else {
4383       // NULL context is unexpected, double-check this is the VMThread
4384       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4385     }
4386     do_resume(osthread);
4387   }
4388   // failure means pthread_kill failed for some reason - arguably this is
4389   // a fatal problem, but such problems are ignored elsewhere
4390 
4391   return epc;
4392 }
4393 
4394 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4395 {
4396    if (is_NPTL()) {
4397       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4398    } else {
4399 #ifndef IA64
4400       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4401       // word back to default 64bit precision if condvar is signaled. Java
4402       // wants 53bit precision.  Save and restore current value.
4403       int fpu = get_fpu_control_word();
4404 #endif // IA64
4405       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4406 #ifndef IA64
4407       set_fpu_control_word(fpu);
4408 #endif // IA64
4409       return status;
4410    }
4411 }
4412 
4413 ////////////////////////////////////////////////////////////////////////////////
4414 // debug support
4415 
4416 static address same_page(address x, address y) {
4417   int page_bits = -os::vm_page_size();
4418   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4419     return x;
4420   else if (x > y)
4421     return (address)(intptr_t(y) | ~page_bits) + 1;
4422   else
4423     return (address)(intptr_t(y) & page_bits);
4424 }
4425 
4426 bool os::find(address addr, outputStream* st) {
4427   Dl_info dlinfo;
4428   memset(&dlinfo, 0, sizeof(dlinfo));
4429   if (dladdr(addr, &dlinfo)) {
4430     st->print(PTR_FORMAT ": ", addr);
4431     if (dlinfo.dli_sname != NULL) {
4432       st->print("%s+%#x", dlinfo.dli_sname,
4433                  addr - (intptr_t)dlinfo.dli_saddr);
4434     } else if (dlinfo.dli_fname) {
4435       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4436     } else {
4437       st->print("<absolute address>");
4438     }
4439     if (dlinfo.dli_fname) {
4440       st->print(" in %s", dlinfo.dli_fname);
4441     }
4442     if (dlinfo.dli_fbase) {
4443       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4444     }
4445     st->cr();
4446 
4447     if (Verbose) {
4448       // decode some bytes around the PC
4449       address begin = same_page(addr-40, addr);
4450       address end   = same_page(addr+40, addr);
4451       address       lowest = (address) dlinfo.dli_sname;
4452       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4453       if (begin < lowest)  begin = lowest;
4454       Dl_info dlinfo2;
4455       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4456           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4457         end = (address) dlinfo2.dli_saddr;
4458       Disassembler::decode(begin, end, st);
4459     }
4460     return true;
4461   }
4462   return false;
4463 }
4464 
4465 ////////////////////////////////////////////////////////////////////////////////
4466 // misc
4467 
4468 // This does not do anything on Linux. This is basically a hook for being
4469 // able to use structured exception handling (thread-local exception filters)
4470 // on, e.g., Win32.
4471 void
4472 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4473                          JavaCallArguments* args, Thread* thread) {
4474   f(value, method, args, thread);
4475 }
4476 
4477 void os::print_statistics() {
4478 }
4479 
4480 int os::message_box(const char* title, const char* message) {
4481   int i;
4482   fdStream err(defaultStream::error_fd());
4483   for (i = 0; i < 78; i++) err.print_raw("=");
4484   err.cr();
4485   err.print_raw_cr(title);
4486   for (i = 0; i < 78; i++) err.print_raw("-");
4487   err.cr();
4488   err.print_raw_cr(message);
4489   for (i = 0; i < 78; i++) err.print_raw("=");
4490   err.cr();
4491 
4492   char buf[16];
4493   // Prevent process from exiting upon "read error" without consuming all CPU
4494   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4495 
4496   return buf[0] == 'y' || buf[0] == 'Y';
4497 }
4498 
4499 int os::stat(const char *path, struct stat *sbuf) {
4500   char pathbuf[MAX_PATH];
4501   if (strlen(path) > MAX_PATH - 1) {
4502     errno = ENAMETOOLONG;
4503     return -1;
4504   }
4505   os::native_path(strcpy(pathbuf, path));
4506   return ::stat(pathbuf, sbuf);
4507 }
4508 
4509 bool os::check_heap(bool force) {
4510   return true;
4511 }
4512 
4513 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4514   return ::vsnprintf(buf, count, format, args);
4515 }
4516 
4517 // Is a (classpath) directory empty?
4518 bool os::dir_is_empty(const char* path) {
4519   DIR *dir = NULL;
4520   struct dirent *ptr;
4521 
4522   dir = opendir(path);
4523   if (dir == NULL) return true;
4524 
4525   /* Scan the directory */
4526   bool result = true;
4527   char buf[sizeof(struct dirent) + MAX_PATH];
4528   while (result && (ptr = ::readdir(dir)) != NULL) {
4529     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4530       result = false;
4531     }
4532   }
4533   closedir(dir);
4534   return result;
4535 }
4536 
4537 // This code originates from JDK's sysOpen and open64_w
4538 // from src/solaris/hpi/src/system_md.c
4539 
4540 #ifndef O_DELETE
4541 #define O_DELETE 0x10000
4542 #endif
4543 
4544 // Open a file. Unlink the file immediately after open returns
4545 // if the specified oflag has the O_DELETE flag set.
4546 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4547 
4548 int os::open(const char *path, int oflag, int mode) {
4549 
4550   if (strlen(path) > MAX_PATH - 1) {
4551     errno = ENAMETOOLONG;
4552     return -1;
4553   }
4554   int fd;
4555   int o_delete = (oflag & O_DELETE);
4556   oflag = oflag & ~O_DELETE;
4557 
4558   fd = ::open64(path, oflag, mode);
4559   if (fd == -1) return -1;
4560 
4561   //If the open succeeded, the file might still be a directory
4562   {
4563     struct stat64 buf64;
4564     int ret = ::fstat64(fd, &buf64);
4565     int st_mode = buf64.st_mode;
4566 
4567     if (ret != -1) {
4568       if ((st_mode & S_IFMT) == S_IFDIR) {
4569         errno = EISDIR;
4570         ::close(fd);
4571         return -1;
4572       }
4573     } else {
4574       ::close(fd);
4575       return -1;
4576     }
4577   }
4578 
4579     /*
4580      * All file descriptors that are opened in the JVM and not
4581      * specifically destined for a subprocess should have the
4582      * close-on-exec flag set.  If we don't set it, then careless 3rd
4583      * party native code might fork and exec without closing all
4584      * appropriate file descriptors (e.g. as we do in closeDescriptors in
4585      * UNIXProcess.c), and this in turn might:
4586      *
4587      * - cause end-of-file to fail to be detected on some file
4588      *   descriptors, resulting in mysterious hangs, or
4589      *
4590      * - might cause an fopen in the subprocess to fail on a system
4591      *   suffering from bug 1085341.
4592      *
4593      * (Yes, the default setting of the close-on-exec flag is a Unix
4594      * design flaw)
4595      *
4596      * See:
4597      * 1085341: 32-bit stdio routines should support file descriptors >255
4598      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4599      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4600      */
4601 #ifdef FD_CLOEXEC
4602     {
4603         int flags = ::fcntl(fd, F_GETFD);
4604         if (flags != -1)
4605             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4606     }
4607 #endif
4608 
4609   if (o_delete != 0) {
4610     ::unlink(path);
4611   }
4612   return fd;
4613 }
4614 
4615 
4616 // create binary file, rewriting existing file if required
4617 int os::create_binary_file(const char* path, bool rewrite_existing) {
4618   int oflags = O_WRONLY | O_CREAT;
4619   if (!rewrite_existing) {
4620     oflags |= O_EXCL;
4621   }
4622   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4623 }
4624 
4625 // return current position of file pointer
4626 jlong os::current_file_offset(int fd) {
4627   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4628 }
4629 
4630 // move file pointer to the specified offset
4631 jlong os::seek_to_file_offset(int fd, jlong offset) {
4632   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4633 }
4634 
4635 // This code originates from JDK's sysAvailable
4636 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4637 
4638 int os::available(int fd, jlong *bytes) {
4639   jlong cur, end;
4640   int mode;
4641   struct stat64 buf64;
4642 
4643   if (::fstat64(fd, &buf64) >= 0) {
4644     mode = buf64.st_mode;
4645     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4646       /*
4647       * XXX: is the following call interruptible? If so, this might
4648       * need to go through the INTERRUPT_IO() wrapper as for other
4649       * blocking, interruptible calls in this file.
4650       */
4651       int n;
4652       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4653         *bytes = n;
4654         return 1;
4655       }
4656     }
4657   }
4658   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4659     return 0;
4660   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4661     return 0;
4662   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4663     return 0;
4664   }
4665   *bytes = end - cur;
4666   return 1;
4667 }
4668 
4669 int os::socket_available(int fd, jint *pbytes) {
4670   // Linux doc says EINTR not returned, unlike Solaris
4671   int ret = ::ioctl(fd, FIONREAD, pbytes);
4672 
4673   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4674   // is expected to return 0 on failure and 1 on success to the jdk.
4675   return (ret < 0) ? 0 : 1;
4676 }
4677 
4678 // Map a block of memory.
4679 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4680                      char *addr, size_t bytes, bool read_only,
4681                      bool allow_exec) {
4682   int prot;
4683   int flags;
4684 
4685   if (read_only) {
4686     prot = PROT_READ;
4687     flags = MAP_SHARED;
4688   } else {
4689     prot = PROT_READ | PROT_WRITE;
4690     flags = MAP_PRIVATE;
4691   }
4692 
4693   if (allow_exec) {
4694     prot |= PROT_EXEC;
4695   }
4696 
4697   if (addr != NULL) {
4698     flags |= MAP_FIXED;
4699   }
4700 
4701   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4702                                      fd, file_offset);
4703   if (mapped_address == MAP_FAILED) {
4704     return NULL;
4705   }
4706   return mapped_address;
4707 }
4708 
4709 
4710 // Remap a block of memory.
4711 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4712                        char *addr, size_t bytes, bool read_only,
4713                        bool allow_exec) {
4714   // same as map_memory() on this OS
4715   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4716                         allow_exec);
4717 }
4718 
4719 
4720 // Unmap a block of memory.
4721 bool os::unmap_memory(char* addr, size_t bytes) {
4722   return munmap(addr, bytes) == 0;
4723 }
4724 
4725 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4726 
4727 static clockid_t thread_cpu_clockid(Thread* thread) {
4728   pthread_t tid = thread->osthread()->pthread_id();
4729   clockid_t clockid;
4730 
4731   // Get thread clockid
4732   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4733   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4734   return clockid;
4735 }
4736 
4737 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4738 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4739 // of a thread.
4740 //
4741 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4742 // the fast estimate available on the platform.
4743 
4744 jlong os::current_thread_cpu_time() {
4745   if (os::Linux::supports_fast_thread_cpu_time()) {
4746     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4747   } else {
4748     // return user + sys since the cost is the same
4749     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4750   }
4751 }
4752 
4753 jlong os::thread_cpu_time(Thread* thread) {
4754   // consistent with what current_thread_cpu_time() returns
4755   if (os::Linux::supports_fast_thread_cpu_time()) {
4756     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4757   } else {
4758     return slow_thread_cpu_time(thread, true /* user + sys */);
4759   }
4760 }
4761 
4762 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4763   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4764     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4765   } else {
4766     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4767   }
4768 }
4769 
4770 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4771   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4772     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4773   } else {
4774     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4775   }
4776 }
4777 
4778 //
4779 //  -1 on error.
4780 //
4781 
4782 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4783   static bool proc_pid_cpu_avail = true;
4784   static bool proc_task_unchecked = true;
4785   static const char *proc_stat_path = "/proc/%d/stat";
4786   pid_t  tid = thread->osthread()->thread_id();
4787   int i;
4788   char *s;
4789   char stat[2048];
4790   int statlen;
4791   char proc_name[64];
4792   int count;
4793   long sys_time, user_time;
4794   char string[64];
4795   char cdummy;
4796   int idummy;
4797   long ldummy;
4798   FILE *fp;
4799 
4800   // We first try accessing /proc/<pid>/cpu since this is faster to
4801   // process.  If this file is not present (linux kernels 2.5 and above)
4802   // then we open /proc/<pid>/stat.
4803   if ( proc_pid_cpu_avail ) {
4804     sprintf(proc_name, "/proc/%d/cpu", tid);
4805     fp =  fopen(proc_name, "r");
4806     if ( fp != NULL ) {
4807       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4808       fclose(fp);
4809       if ( count != 3 ) return -1;
4810 
4811       if (user_sys_cpu_time) {
4812         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4813       } else {
4814         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4815       }
4816     }
4817     else proc_pid_cpu_avail = false;
4818   }
4819 
4820   // The /proc/<tid>/stat aggregates per-process usage on
4821   // new Linux kernels 2.6+ where NPTL is supported.
4822   // The /proc/self/task/<tid>/stat still has the per-thread usage.
4823   // See bug 6328462.
4824   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4825   // and possibly in some other cases, so we check its availability.
4826   if (proc_task_unchecked && os::Linux::is_NPTL()) {
4827     // This is executed only once
4828     proc_task_unchecked = false;
4829     fp = fopen("/proc/self/task", "r");
4830     if (fp != NULL) {
4831       proc_stat_path = "/proc/self/task/%d/stat";
4832       fclose(fp);
4833     }
4834   }
4835 
4836   sprintf(proc_name, proc_stat_path, tid);
4837   fp = fopen(proc_name, "r");
4838   if ( fp == NULL ) return -1;
4839   statlen = fread(stat, 1, 2047, fp);
4840   stat[statlen] = '\0';
4841   fclose(fp);
4842 
4843   // Skip pid and the command string. Note that we could be dealing with
4844   // weird command names, e.g. user could decide to rename java launcher
4845   // to "java 1.4.2 :)", then the stat file would look like
4846   //                1234 (java 1.4.2 :)) R ... ...
4847   // We don't really need to know the command string, just find the last
4848   // occurrence of ")" and then start parsing from there. See bug 4726580.
4849   s = strrchr(stat, ')');
4850   i = 0;
4851   if (s == NULL ) return -1;
4852 
4853   // Skip blank chars
4854   do s++; while (isspace(*s));
4855 
4856   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4857                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4858                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4859                  &user_time, &sys_time);
4860   if ( count != 13 ) return -1;
4861   if (user_sys_cpu_time) {
4862     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4863   } else {
4864     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4865   }
4866 }
4867 
4868 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4869   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4870   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4871   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4872   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4873 }
4874 
4875 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4876   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4877   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4878   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4879   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4880 }
4881 
4882 bool os::is_thread_cpu_time_supported() {
4883   return true;
4884 }
4885 
4886 // System loadavg support.  Returns -1 if load average cannot be obtained.
4887 // Linux doesn't yet have a (official) notion of processor sets,
4888 // so just return the system wide load average.
4889 int os::loadavg(double loadavg[], int nelem) {
4890   return ::getloadavg(loadavg, nelem);
4891 }
4892 
4893 void os::pause() {
4894   char filename[MAX_PATH];
4895   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4896     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4897   } else {
4898     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4899   }
4900 
4901   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4902   if (fd != -1) {
4903     struct stat buf;
4904     ::close(fd);
4905     while (::stat(filename, &buf) == 0) {
4906       (void)::poll(NULL, 0, 100);
4907     }
4908   } else {
4909     jio_fprintf(stderr,
4910       "Could not open pause file '%s', continuing immediately.\n", filename);
4911   }
4912 }
4913 
4914 
4915 // Refer to the comments in os_solaris.cpp park-unpark.
4916 //
4917 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4918 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4919 // For specifics regarding the bug see GLIBC BUGID 261237 :
4920 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4921 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4922 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4923 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4924 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4925 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4926 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4927 // of libpthread avoids the problem, but isn't practical.
4928 //
4929 // Possible remedies:
4930 //
4931 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4932 //      This is palliative and probabilistic, however.  If the thread is preempted
4933 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4934 //      than the minimum period may have passed, and the abstime may be stale (in the
4935 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
4936 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4937 //
4938 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4939 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
4940 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4941 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
4942 //      thread.
4943 //
4944 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4945 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
4946 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4947 //      This also works well.  In fact it avoids kernel-level scalability impediments
4948 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4949 //      timers in a graceful fashion.
4950 //
4951 // 4.   When the abstime value is in the past it appears that control returns
4952 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4953 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
4954 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
4955 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
4956 //      It may be possible to avoid reinitialization by checking the return
4957 //      value from pthread_cond_timedwait().  In addition to reinitializing the
4958 //      condvar we must establish the invariant that cond_signal() is only called
4959 //      within critical sections protected by the adjunct mutex.  This prevents
4960 //      cond_signal() from "seeing" a condvar that's in the midst of being
4961 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
4962 //      desirable signal-after-unlock optimization that avoids futile context switching.
4963 //
4964 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
4965 //      structure when a condvar is used or initialized.  cond_destroy()  would
4966 //      release the helper structure.  Our reinitialize-after-timedwait fix
4967 //      put excessive stress on malloc/free and locks protecting the c-heap.
4968 //
4969 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
4970 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4971 // and only enabling the work-around for vulnerable environments.
4972 
4973 // utility to compute the abstime argument to timedwait:
4974 // millis is the relative timeout time
4975 // abstime will be the absolute timeout time
4976 // TODO: replace compute_abstime() with unpackTime()
4977 
4978 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4979   if (millis < 0)  millis = 0;
4980   struct timeval now;
4981   int status = gettimeofday(&now, NULL);
4982   assert(status == 0, "gettimeofday");
4983   jlong seconds = millis / 1000;
4984   millis %= 1000;
4985   if (seconds > 50000000) { // see man cond_timedwait(3T)
4986     seconds = 50000000;
4987   }
4988   abstime->tv_sec = now.tv_sec  + seconds;
4989   long       usec = now.tv_usec + millis * 1000;
4990   if (usec >= 1000000) {
4991     abstime->tv_sec += 1;
4992     usec -= 1000000;
4993   }
4994   abstime->tv_nsec = usec * 1000;
4995   return abstime;
4996 }
4997 
4998 
4999 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5000 // Conceptually TryPark() should be equivalent to park(0).
5001 
5002 int os::PlatformEvent::TryPark() {
5003   for (;;) {
5004     const int v = _Event ;
5005     guarantee ((v == 0) || (v == 1), "invariant") ;
5006     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5007   }
5008 }
5009 
5010 void os::PlatformEvent::park() {       // AKA "down()"
5011   // Invariant: Only the thread associated with the Event/PlatformEvent
5012   // may call park().
5013   // TODO: assert that _Assoc != NULL or _Assoc == Self
5014   int v ;
5015   for (;;) {
5016       v = _Event ;
5017       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5018   }
5019   guarantee (v >= 0, "invariant") ;
5020   if (v == 0) {
5021      // Do this the hard way by blocking ...
5022      int status = pthread_mutex_lock(_mutex);
5023      assert_status(status == 0, status, "mutex_lock");
5024      guarantee (_nParked == 0, "invariant") ;
5025      ++ _nParked ;
5026      while (_Event < 0) {
5027         status = pthread_cond_wait(_cond, _mutex);
5028         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5029         // Treat this the same as if the wait was interrupted
5030         if (status == ETIME) { status = EINTR; }
5031         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5032      }
5033      -- _nParked ;
5034 
5035     // In theory we could move the ST of 0 into _Event past the unlock(),
5036     // but then we'd need a MEMBAR after the ST.
5037     _Event = 0 ;
5038      status = pthread_mutex_unlock(_mutex);
5039      assert_status(status == 0, status, "mutex_unlock");
5040   }
5041   guarantee (_Event >= 0, "invariant") ;
5042 }
5043 
5044 int os::PlatformEvent::park(jlong millis) {
5045   guarantee (_nParked == 0, "invariant") ;
5046 
5047   int v ;
5048   for (;;) {
5049       v = _Event ;
5050       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5051   }
5052   guarantee (v >= 0, "invariant") ;
5053   if (v != 0) return OS_OK ;
5054 
5055   // We do this the hard way, by blocking the thread.
5056   // Consider enforcing a minimum timeout value.
5057   struct timespec abst;
5058   compute_abstime(&abst, millis);
5059 
5060   int ret = OS_TIMEOUT;
5061   int status = pthread_mutex_lock(_mutex);
5062   assert_status(status == 0, status, "mutex_lock");
5063   guarantee (_nParked == 0, "invariant") ;
5064   ++_nParked ;
5065 
5066   // Object.wait(timo) will return because of
5067   // (a) notification
5068   // (b) timeout
5069   // (c) thread.interrupt
5070   //
5071   // Thread.interrupt and object.notify{All} both call Event::set.
5072   // That is, we treat thread.interrupt as a special case of notification.
5073   // The underlying Solaris implementation, cond_timedwait, admits
5074   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5075   // JVM from making those visible to Java code.  As such, we must
5076   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5077   //
5078   // TODO: properly differentiate simultaneous notify+interrupt.
5079   // In that case, we should propagate the notify to another waiter.
5080 
5081   while (_Event < 0) {
5082     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5083     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5084       pthread_cond_destroy (_cond);
5085       pthread_cond_init (_cond, NULL) ;
5086     }
5087     assert_status(status == 0 || status == EINTR ||
5088                   status == ETIME || status == ETIMEDOUT,
5089                   status, "cond_timedwait");
5090     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5091     if (status == ETIME || status == ETIMEDOUT) break ;
5092     // We consume and ignore EINTR and spurious wakeups.
5093   }
5094   --_nParked ;
5095   if (_Event >= 0) {
5096      ret = OS_OK;
5097   }
5098   _Event = 0 ;
5099   status = pthread_mutex_unlock(_mutex);
5100   assert_status(status == 0, status, "mutex_unlock");
5101   assert (_nParked == 0, "invariant") ;
5102   return ret;
5103 }
5104 
5105 void os::PlatformEvent::unpark() {
5106   int v, AnyWaiters ;
5107   for (;;) {
5108       v = _Event ;
5109       if (v > 0) {
5110          // The LD of _Event could have reordered or be satisfied
5111          // by a read-aside from this processor's write buffer.
5112          // To avoid problems execute a barrier and then
5113          // ratify the value.
5114          OrderAccess::fence() ;
5115          if (_Event == v) return ;
5116          continue ;
5117       }
5118       if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5119   }
5120   if (v < 0) {
5121      // Wait for the thread associated with the event to vacate
5122      int status = pthread_mutex_lock(_mutex);
5123      assert_status(status == 0, status, "mutex_lock");
5124      AnyWaiters = _nParked ;
5125      assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5126      if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5127         AnyWaiters = 0 ;
5128         pthread_cond_signal (_cond);
5129      }
5130      status = pthread_mutex_unlock(_mutex);
5131      assert_status(status == 0, status, "mutex_unlock");
5132      if (AnyWaiters != 0) {
5133         status = pthread_cond_signal(_cond);
5134         assert_status(status == 0, status, "cond_signal");
5135      }
5136   }
5137 
5138   // Note that we signal() _after dropping the lock for "immortal" Events.
5139   // This is safe and avoids a common class of  futile wakeups.  In rare
5140   // circumstances this can cause a thread to return prematurely from
5141   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5142   // simply re-test the condition and re-park itself.
5143 }
5144 
5145 
5146 // JSR166
5147 // -------------------------------------------------------
5148 
5149 /*
5150  * The solaris and linux implementations of park/unpark are fairly
5151  * conservative for now, but can be improved. They currently use a
5152  * mutex/condvar pair, plus a a count.
5153  * Park decrements count if > 0, else does a condvar wait.  Unpark
5154  * sets count to 1 and signals condvar.  Only one thread ever waits
5155  * on the condvar. Contention seen when trying to park implies that someone
5156  * is unparking you, so don't wait. And spurious returns are fine, so there
5157  * is no need to track notifications.
5158  */
5159 
5160 
5161 #define NANOSECS_PER_SEC 1000000000
5162 #define NANOSECS_PER_MILLISEC 1000000
5163 #define MAX_SECS 100000000
5164 /*
5165  * This code is common to linux and solaris and will be moved to a
5166  * common place in dolphin.
5167  *
5168  * The passed in time value is either a relative time in nanoseconds
5169  * or an absolute time in milliseconds. Either way it has to be unpacked
5170  * into suitable seconds and nanoseconds components and stored in the
5171  * given timespec structure.
5172  * Given time is a 64-bit value and the time_t used in the timespec is only
5173  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5174  * overflow if times way in the future are given. Further on Solaris versions
5175  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5176  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5177  * As it will be 28 years before "now + 100000000" will overflow we can
5178  * ignore overflow and just impose a hard-limit on seconds using the value
5179  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5180  * years from "now".
5181  */
5182 
5183 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5184   assert (time > 0, "convertTime");
5185 
5186   struct timeval now;
5187   int status = gettimeofday(&now, NULL);
5188   assert(status == 0, "gettimeofday");
5189 
5190   time_t max_secs = now.tv_sec + MAX_SECS;
5191 
5192   if (isAbsolute) {
5193     jlong secs = time / 1000;
5194     if (secs > max_secs) {
5195       absTime->tv_sec = max_secs;
5196     }
5197     else {
5198       absTime->tv_sec = secs;
5199     }
5200     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5201   }
5202   else {
5203     jlong secs = time / NANOSECS_PER_SEC;
5204     if (secs >= MAX_SECS) {
5205       absTime->tv_sec = max_secs;
5206       absTime->tv_nsec = 0;
5207     }
5208     else {
5209       absTime->tv_sec = now.tv_sec + secs;
5210       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5211       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5212         absTime->tv_nsec -= NANOSECS_PER_SEC;
5213         ++absTime->tv_sec; // note: this must be <= max_secs
5214       }
5215     }
5216   }
5217   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5218   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5219   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5220   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5221 }
5222 
5223 void Parker::park(bool isAbsolute, jlong time) {
5224   // Optional fast-path check:
5225   // Return immediately if a permit is available.
5226   if (_counter > 0) {
5227       _counter = 0 ;
5228       OrderAccess::fence();
5229       return ;
5230   }
5231 
5232   Thread* thread = Thread::current();
5233   assert(thread->is_Java_thread(), "Must be JavaThread");
5234   JavaThread *jt = (JavaThread *)thread;
5235 
5236   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5237   // Check interrupt before trying to wait
5238   if (Thread::is_interrupted(thread, false)) {
5239     return;
5240   }
5241 
5242   // Next, demultiplex/decode time arguments
5243   timespec absTime;
5244   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5245     return;
5246   }
5247   if (time > 0) {
5248     unpackTime(&absTime, isAbsolute, time);
5249   }
5250 
5251 
5252   // Enter safepoint region
5253   // Beware of deadlocks such as 6317397.
5254   // The per-thread Parker:: mutex is a classic leaf-lock.
5255   // In particular a thread must never block on the Threads_lock while
5256   // holding the Parker:: mutex.  If safepoints are pending both the
5257   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5258   ThreadBlockInVM tbivm(jt);
5259 
5260   // Don't wait if cannot get lock since interference arises from
5261   // unblocking.  Also. check interrupt before trying wait
5262   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5263     return;
5264   }
5265 
5266   int status ;
5267   if (_counter > 0)  { // no wait needed
5268     _counter = 0;
5269     status = pthread_mutex_unlock(_mutex);
5270     assert (status == 0, "invariant") ;
5271     OrderAccess::fence();
5272     return;
5273   }
5274 
5275 #ifdef ASSERT
5276   // Don't catch signals while blocked; let the running threads have the signals.
5277   // (This allows a debugger to break into the running thread.)
5278   sigset_t oldsigs;
5279   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5280   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5281 #endif
5282 
5283   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5284   jt->set_suspend_equivalent();
5285   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5286 
5287   if (time == 0) {
5288     status = pthread_cond_wait (_cond, _mutex) ;
5289   } else {
5290     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5291     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5292       pthread_cond_destroy (_cond) ;
5293       pthread_cond_init    (_cond, NULL);
5294     }
5295   }
5296   assert_status(status == 0 || status == EINTR ||
5297                 status == ETIME || status == ETIMEDOUT,
5298                 status, "cond_timedwait");
5299 
5300 #ifdef ASSERT
5301   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5302 #endif
5303 
5304   _counter = 0 ;
5305   status = pthread_mutex_unlock(_mutex) ;
5306   assert_status(status == 0, status, "invariant") ;
5307   // If externally suspended while waiting, re-suspend
5308   if (jt->handle_special_suspend_equivalent_condition()) {
5309     jt->java_suspend_self();
5310   }
5311 
5312   OrderAccess::fence();
5313 }
5314 
5315 void Parker::unpark() {
5316   int s, status ;
5317   status = pthread_mutex_lock(_mutex);
5318   assert (status == 0, "invariant") ;
5319   s = _counter;
5320   _counter = 1;
5321   if (s < 1) {
5322      if (WorkAroundNPTLTimedWaitHang) {
5323         status = pthread_cond_signal (_cond) ;
5324         assert (status == 0, "invariant") ;
5325         status = pthread_mutex_unlock(_mutex);
5326         assert (status == 0, "invariant") ;
5327      } else {
5328         status = pthread_mutex_unlock(_mutex);
5329         assert (status == 0, "invariant") ;
5330         status = pthread_cond_signal (_cond) ;
5331         assert (status == 0, "invariant") ;
5332      }
5333   } else {
5334     pthread_mutex_unlock(_mutex);
5335     assert (status == 0, "invariant") ;
5336   }
5337 }
5338 
5339 
5340 extern char** environ;
5341 
5342 #ifndef __NR_fork
5343 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5344 #endif
5345 
5346 #ifndef __NR_execve
5347 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5348 #endif
5349 
5350 // Run the specified command in a separate process. Return its exit value,
5351 // or -1 on failure (e.g. can't fork a new process).
5352 // Unlike system(), this function can be called from signal handler. It
5353 // doesn't block SIGINT et al.
5354 int os::fork_and_exec(char* cmd) {
5355   const char * argv[4] = {"sh", "-c", cmd, NULL};
5356 
5357   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5358   // pthread_atfork handlers and reset pthread library. All we need is a
5359   // separate process to execve. Make a direct syscall to fork process.
5360   // On IA64 there's no fork syscall, we have to use fork() and hope for
5361   // the best...
5362   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5363               IA64_ONLY(fork();)
5364 
5365   if (pid < 0) {
5366     // fork failed
5367     return -1;
5368 
5369   } else if (pid == 0) {
5370     // child process
5371 
5372     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5373     // first to kill every thread on the thread list. Because this list is
5374     // not reset by fork() (see notes above), execve() will instead kill
5375     // every thread in the parent process. We know this is the only thread
5376     // in the new process, so make a system call directly.
5377     // IA64 should use normal execve() from glibc to match the glibc fork()
5378     // above.
5379     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5380     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5381 
5382     // execve failed
5383     _exit(-1);
5384 
5385   } else  {
5386     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5387     // care about the actual exit code, for now.
5388 
5389     int status;
5390 
5391     // Wait for the child process to exit.  This returns immediately if
5392     // the child has already exited. */
5393     while (waitpid(pid, &status, 0) < 0) {
5394         switch (errno) {
5395         case ECHILD: return 0;
5396         case EINTR: break;
5397         default: return -1;
5398         }
5399     }
5400 
5401     if (WIFEXITED(status)) {
5402        // The child exited normally; get its exit code.
5403        return WEXITSTATUS(status);
5404     } else if (WIFSIGNALED(status)) {
5405        // The child exited because of a signal
5406        // The best value to return is 0x80 + signal number,
5407        // because that is what all Unix shells do, and because
5408        // it allows callers to distinguish between process exit and
5409        // process death by signal.
5410        return 0x80 + WTERMSIG(status);
5411     } else {
5412        // Unknown exit code; pass it through
5413        return status;
5414     }
5415   }
5416 }
5417 
5418 // is_headless_jre()
5419 //
5420 // Test for the existence of libmawt in motif21 or xawt directories
5421 // in order to report if we are running in a headless jre
5422 //
5423 bool os::is_headless_jre() {
5424     struct stat statbuf;
5425     char buf[MAXPATHLEN];
5426     char libmawtpath[MAXPATHLEN];
5427     const char *xawtstr  = "/xawt/libmawt.so";
5428     const char *motifstr = "/motif21/libmawt.so";
5429     char *p;
5430 
5431     // Get path to libjvm.so
5432     os::jvm_path(buf, sizeof(buf));
5433 
5434     // Get rid of libjvm.so
5435     p = strrchr(buf, '/');
5436     if (p == NULL) return false;
5437     else *p = '\0';
5438 
5439     // Get rid of client or server
5440     p = strrchr(buf, '/');
5441     if (p == NULL) return false;
5442     else *p = '\0';
5443 
5444     // check xawt/libmawt.so
5445     strcpy(libmawtpath, buf);
5446     strcat(libmawtpath, xawtstr);
5447     if (::stat(libmawtpath, &statbuf) == 0) return false;
5448 
5449     // check motif21/libmawt.so
5450     strcpy(libmawtpath, buf);
5451     strcat(libmawtpath, motifstr);
5452     if (::stat(libmawtpath, &statbuf) == 0) return false;
5453 
5454     return true;
5455 }
5456 
5457 
5458 #ifdef JAVASE_EMBEDDED
5459 //
5460 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5461 //
5462 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5463 
5464 // ctor
5465 //
5466 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5467   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5468   _fd = fd;
5469 
5470   if (os::create_thread(this, os::os_thread)) {
5471     _memnotify_thread = this;
5472     os::set_priority(this, NearMaxPriority);
5473     os::start_thread(this);
5474   }
5475 }
5476 
5477 // Where all the work gets done
5478 //
5479 void MemNotifyThread::run() {
5480   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5481 
5482   // Set up the select arguments
5483   fd_set rfds;
5484   if (_fd != -1) {
5485     FD_ZERO(&rfds);
5486     FD_SET(_fd, &rfds);
5487   }
5488 
5489   // Now wait for the mem_notify device to wake up
5490   while (1) {
5491     // Wait for the mem_notify device to signal us..
5492     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5493     if (rc == -1) {
5494       perror("select!\n");
5495       break;
5496     } else if (rc) {
5497       //ssize_t free_before = os::available_memory();
5498       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5499 
5500       // The kernel is telling us there is not much memory left...
5501       // try to do something about that
5502 
5503       // If we are not already in a GC, try one.
5504       if (!Universe::heap()->is_gc_active()) {
5505         Universe::heap()->collect(GCCause::_allocation_failure);
5506 
5507         //ssize_t free_after = os::available_memory();
5508         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5509         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5510       }
5511       // We might want to do something like the following if we find the GC's are not helping...
5512       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5513     }
5514   }
5515 }
5516 
5517 //
5518 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5519 //
5520 void MemNotifyThread::start() {
5521   int    fd;
5522   fd = open ("/dev/mem_notify", O_RDONLY, 0);
5523   if (fd < 0) {
5524       return;
5525   }
5526 
5527   if (memnotify_thread() == NULL) {
5528     new MemNotifyThread(fd);
5529   }
5530 }
5531 #endif // JAVASE_EMBEDDED
--- EOF ---