1 /*
   2  * Copyright (c) 1999, 2012, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "interpreter/interpreter.hpp"
  33 #include "jvm_linux.h"
  34 #include "memory/allocation.inline.hpp"
  35 #include "memory/filemap.hpp"
  36 #include "mutex_linux.inline.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "os_share_linux.hpp"
  39 #include "prims/jniFastGetField.hpp"
  40 #include "prims/jvm.h"
  41 #include "prims/jvm_misc.hpp"
  42 #include "runtime/arguments.hpp"
  43 #include "runtime/extendedPC.hpp"
  44 #include "runtime/globals.hpp"
  45 #include "runtime/interfaceSupport.hpp"
  46 #include "runtime/java.hpp"
  47 #include "runtime/javaCalls.hpp"
  48 #include "runtime/mutexLocker.hpp"
  49 #include "runtime/objectMonitor.hpp"
  50 #include "runtime/osThread.hpp"
  51 #include "runtime/perfMemory.hpp"
  52 #include "runtime/sharedRuntime.hpp"
  53 #include "runtime/statSampler.hpp"
  54 #include "runtime/stubRoutines.hpp"
  55 #include "runtime/threadCritical.hpp"
  56 #include "runtime/timer.hpp"
  57 #include "services/attachListener.hpp"
  58 #include "services/runtimeService.hpp"
  59 #include "thread_linux.inline.hpp"
  60 #include "utilities/decoder.hpp"
  61 #include "utilities/defaultStream.hpp"
  62 #include "utilities/events.hpp"
  63 #include "utilities/growableArray.hpp"
  64 #include "utilities/vmError.hpp"
  65 #ifdef TARGET_ARCH_x86
  66 # include "assembler_x86.inline.hpp"
  67 # include "nativeInst_x86.hpp"
  68 #endif
  69 #ifdef TARGET_ARCH_sparc
  70 # include "assembler_sparc.inline.hpp"
  71 # include "nativeInst_sparc.hpp"
  72 #endif
  73 #ifdef TARGET_ARCH_zero
  74 # include "assembler_zero.inline.hpp"
  75 # include "nativeInst_zero.hpp"
  76 #endif
  77 #ifdef TARGET_ARCH_arm
  78 # include "assembler_arm.inline.hpp"
  79 # include "nativeInst_arm.hpp"
  80 #endif
  81 #ifdef TARGET_ARCH_ppc
  82 # include "assembler_ppc.inline.hpp"
  83 # include "nativeInst_ppc.hpp"
  84 #endif
  85 #ifdef COMPILER1
  86 #include "c1/c1_Runtime1.hpp"
  87 #endif
  88 #ifdef COMPILER2
  89 #include "opto/runtime.hpp"
  90 #endif
  91 
  92 // put OS-includes here
  93 # include <sys/types.h>
  94 # include <sys/mman.h>
  95 # include <sys/stat.h>
  96 # include <sys/select.h>
  97 # include <pthread.h>
  98 # include <signal.h>
  99 # include <errno.h>
 100 # include <dlfcn.h>
 101 # include <stdio.h>
 102 # include <unistd.h>
 103 # include <sys/resource.h>
 104 # include <pthread.h>
 105 # include <sys/stat.h>
 106 # include <sys/time.h>
 107 # include <sys/times.h>
 108 # include <sys/utsname.h>
 109 # include <sys/socket.h>
 110 # include <sys/wait.h>
 111 # include <pwd.h>
 112 # include <poll.h>
 113 # include <semaphore.h>
 114 # include <fcntl.h>
 115 # include <string.h>
 116 # include <syscall.h>
 117 # include <sys/sysinfo.h>
 118 # include <gnu/libc-version.h>
 119 # include <sys/ipc.h>
 120 # include <sys/shm.h>
 121 # include <link.h>
 122 # include <stdint.h>
 123 # include <inttypes.h>
 124 # include <sys/ioctl.h>
 125 
 126 #define MAX_PATH    (2 * K)
 127 
 128 // for timer info max values which include all bits
 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 130 
 131 #define LARGEPAGES_BIT (1 << 6)
 132 ////////////////////////////////////////////////////////////////////////////////
 133 // global variables
 134 julong os::Linux::_physical_memory = 0;
 135 
 136 address   os::Linux::_initial_thread_stack_bottom = NULL;
 137 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 138 
 139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 141 Mutex* os::Linux::_createThread_lock = NULL;
 142 pthread_t os::Linux::_main_thread;
 143 int os::Linux::_page_size = -1;
 144 bool os::Linux::_is_floating_stack = false;
 145 bool os::Linux::_is_NPTL = false;
 146 bool os::Linux::_supports_fast_thread_cpu_time = false;
 147 const char * os::Linux::_glibc_version = NULL;
 148 const char * os::Linux::_libpthread_version = NULL;
 149 
 150 static jlong initial_time_count=0;
 151 
 152 static int clock_tics_per_sec = 100;
 153 
 154 // For diagnostics to print a message once. see run_periodic_checks
 155 static sigset_t check_signal_done;
 156 static bool check_signals = true;;
 157 
 158 static pid_t _initial_pid = 0;
 159 
 160 /* Signal number used to suspend/resume a thread */
 161 
 162 /* do not use any signal number less than SIGSEGV, see 4355769 */
 163 static int SR_signum = SIGUSR2;
 164 sigset_t SR_sigset;
 165 
 166 /* Used to protect dlsym() calls */
 167 static pthread_mutex_t dl_mutex;
 168 
 169 #ifdef JAVASE_EMBEDDED
 170 class MemNotifyThread: public Thread {
 171   friend class VMStructs;
 172  public:
 173   virtual void run();
 174 
 175  private:
 176   static MemNotifyThread* _memnotify_thread;
 177   int _fd;
 178 
 179  public:
 180 
 181   // Constructor
 182   MemNotifyThread(int fd);
 183 
 184   // Tester
 185   bool is_memnotify_thread() const { return true; }
 186 
 187   // Printing
 188   char* name() const { return (char*)"Linux MemNotify Thread"; }
 189 
 190   // Returns the single instance of the MemNotifyThread
 191   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 192 
 193   // Create and start the single instance of MemNotifyThread
 194   static void start();
 195 };
 196 #endif // JAVASE_EMBEDDED
 197 
 198 // utility functions
 199 
 200 static int SR_initialize();
 201 static int SR_finalize();
 202 
 203 julong os::available_memory() {
 204   return Linux::available_memory();
 205 }
 206 
 207 julong os::Linux::available_memory() {
 208   // values in struct sysinfo are "unsigned long"
 209   struct sysinfo si;
 210   sysinfo(&si);
 211 
 212   return (julong)si.freeram * si.mem_unit;
 213 }
 214 
 215 julong os::physical_memory() {
 216   return Linux::physical_memory();
 217 }
 218 
 219 julong os::allocatable_physical_memory(julong size) {
 220 #ifdef _LP64
 221   return size;
 222 #else
 223   julong result = MIN2(size, (julong)3800*M);
 224    if (!is_allocatable(result)) {
 225      // See comments under solaris for alignment considerations
 226      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 227      result =  MIN2(size, reasonable_size);
 228    }
 229    return result;
 230 #endif // _LP64
 231 }
 232 
 233 ////////////////////////////////////////////////////////////////////////////////
 234 // environment support
 235 
 236 bool os::getenv(const char* name, char* buf, int len) {
 237   const char* val = ::getenv(name);
 238   if (val != NULL && strlen(val) < (size_t)len) {
 239     strcpy(buf, val);
 240     return true;
 241   }
 242   if (len > 0) buf[0] = 0;  // return a null string
 243   return false;
 244 }
 245 
 246 
 247 // Return true if user is running as root.
 248 
 249 bool os::have_special_privileges() {
 250   static bool init = false;
 251   static bool privileges = false;
 252   if (!init) {
 253     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 254     init = true;
 255   }
 256   return privileges;
 257 }
 258 
 259 
 260 #ifndef SYS_gettid
 261 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 262 #ifdef __ia64__
 263 #define SYS_gettid 1105
 264 #elif __i386__
 265 #define SYS_gettid 224
 266 #elif __amd64__
 267 #define SYS_gettid 186
 268 #elif __sparc__
 269 #define SYS_gettid 143
 270 #else
 271 #error define gettid for the arch
 272 #endif
 273 #endif
 274 
 275 // Cpu architecture string
 276 #if   defined(ZERO)
 277 static char cpu_arch[] = ZERO_LIBARCH;
 278 #elif defined(IA64)
 279 static char cpu_arch[] = "ia64";
 280 #elif defined(IA32)
 281 static char cpu_arch[] = "i386";
 282 #elif defined(AMD64)
 283 static char cpu_arch[] = "amd64";
 284 #elif defined(ARM)
 285 static char cpu_arch[] = "arm";
 286 #elif defined(PPC)
 287 static char cpu_arch[] = "ppc";
 288 #elif defined(SPARC)
 289 #  ifdef _LP64
 290 static char cpu_arch[] = "sparcv9";
 291 #  else
 292 static char cpu_arch[] = "sparc";
 293 #  endif
 294 #else
 295 #error Add appropriate cpu_arch setting
 296 #endif
 297 
 298 
 299 // pid_t gettid()
 300 //
 301 // Returns the kernel thread id of the currently running thread. Kernel
 302 // thread id is used to access /proc.
 303 //
 304 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 305 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 306 //
 307 pid_t os::Linux::gettid() {
 308   int rslt = syscall(SYS_gettid);
 309   if (rslt == -1) {
 310      // old kernel, no NPTL support
 311      return getpid();
 312   } else {
 313      return (pid_t)rslt;
 314   }
 315 }
 316 
 317 // Most versions of linux have a bug where the number of processors are
 318 // determined by looking at the /proc file system.  In a chroot environment,
 319 // the system call returns 1.  This causes the VM to act as if it is
 320 // a single processor and elide locking (see is_MP() call).
 321 static bool unsafe_chroot_detected = false;
 322 static const char *unstable_chroot_error = "/proc file system not found.\n"
 323                      "Java may be unstable running multithreaded in a chroot "
 324                      "environment on Linux when /proc filesystem is not mounted.";
 325 
 326 void os::Linux::initialize_system_info() {
 327   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 328   if (processor_count() == 1) {
 329     pid_t pid = os::Linux::gettid();
 330     char fname[32];
 331     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 332     FILE *fp = fopen(fname, "r");
 333     if (fp == NULL) {
 334       unsafe_chroot_detected = true;
 335     } else {
 336       fclose(fp);
 337     }
 338   }
 339   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 340   assert(processor_count() > 0, "linux error");
 341 }
 342 
 343 void os::init_system_properties_values() {
 344 //  char arch[12];
 345 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 346 
 347   // The next steps are taken in the product version:
 348   //
 349   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 350   // This library should be located at:
 351   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 352   //
 353   // If "/jre/lib/" appears at the right place in the path, then we
 354   // assume libjvm[_g].so is installed in a JDK and we use this path.
 355   //
 356   // Otherwise exit with message: "Could not create the Java virtual machine."
 357   //
 358   // The following extra steps are taken in the debugging version:
 359   //
 360   // If "/jre/lib/" does NOT appear at the right place in the path
 361   // instead of exit check for $JAVA_HOME environment variable.
 362   //
 363   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 364   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 365   // it looks like libjvm[_g].so is installed there
 366   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 367   //
 368   // Otherwise exit.
 369   //
 370   // Important note: if the location of libjvm.so changes this
 371   // code needs to be changed accordingly.
 372 
 373   // The next few definitions allow the code to be verbatim:
 374 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 375 #define getenv(n) ::getenv(n)
 376 
 377 /*
 378  * See ld(1):
 379  *      The linker uses the following search paths to locate required
 380  *      shared libraries:
 381  *        1: ...
 382  *        ...
 383  *        7: The default directories, normally /lib and /usr/lib.
 384  */
 385 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 386 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 387 #else
 388 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 389 #endif
 390 
 391 #define EXTENSIONS_DIR  "/lib/ext"
 392 #define ENDORSED_DIR    "/lib/endorsed"
 393 #define REG_DIR         "/usr/java/packages"
 394 
 395   {
 396     /* sysclasspath, java_home, dll_dir */
 397     {
 398         char *home_path;
 399         char *dll_path;
 400         char *pslash;
 401         char buf[MAXPATHLEN];
 402         os::jvm_path(buf, sizeof(buf));
 403 
 404         // Found the full path to libjvm.so.
 405         // Now cut the path to <java_home>/jre if we can.
 406         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 407         pslash = strrchr(buf, '/');
 408         if (pslash != NULL)
 409             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 410         dll_path = malloc(strlen(buf) + 1);
 411         if (dll_path == NULL)
 412             return;
 413         strcpy(dll_path, buf);
 414         Arguments::set_dll_dir(dll_path);
 415 
 416         if (pslash != NULL) {
 417             pslash = strrchr(buf, '/');
 418             if (pslash != NULL) {
 419                 *pslash = '\0';       /* get rid of /<arch> */
 420                 pslash = strrchr(buf, '/');
 421                 if (pslash != NULL)
 422                     *pslash = '\0';   /* get rid of /lib */
 423             }
 424         }
 425 
 426         home_path = malloc(strlen(buf) + 1);
 427         if (home_path == NULL)
 428             return;
 429         strcpy(home_path, buf);
 430         Arguments::set_java_home(home_path);
 431 
 432         if (!set_boot_path('/', ':'))
 433             return;
 434     }
 435 
 436     /*
 437      * Where to look for native libraries
 438      *
 439      * Note: Due to a legacy implementation, most of the library path
 440      * is set in the launcher.  This was to accomodate linking restrictions
 441      * on legacy Linux implementations (which are no longer supported).
 442      * Eventually, all the library path setting will be done here.
 443      *
 444      * However, to prevent the proliferation of improperly built native
 445      * libraries, the new path component /usr/java/packages is added here.
 446      * Eventually, all the library path setting will be done here.
 447      */
 448     {
 449         char *ld_library_path;
 450 
 451         /*
 452          * Construct the invariant part of ld_library_path. Note that the
 453          * space for the colon and the trailing null are provided by the
 454          * nulls included by the sizeof operator (so actually we allocate
 455          * a byte more than necessary).
 456          */
 457         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 458             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 459         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 460 
 461         /*
 462          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 463          * should always exist (until the legacy problem cited above is
 464          * addressed).
 465          */
 466         char *v = getenv("LD_LIBRARY_PATH");
 467         if (v != NULL) {
 468             char *t = ld_library_path;
 469             /* That's +1 for the colon and +1 for the trailing '\0' */
 470             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 471             sprintf(ld_library_path, "%s:%s", v, t);
 472         }
 473         Arguments::set_library_path(ld_library_path);
 474     }
 475 
 476     /*
 477      * Extensions directories.
 478      *
 479      * Note that the space for the colon and the trailing null are provided
 480      * by the nulls included by the sizeof operator (so actually one byte more
 481      * than necessary is allocated).
 482      */
 483     {
 484         char *buf = malloc(strlen(Arguments::get_java_home()) +
 485             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 486         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 487             Arguments::get_java_home());
 488         Arguments::set_ext_dirs(buf);
 489     }
 490 
 491     /* Endorsed standards default directory. */
 492     {
 493         char * buf;
 494         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 495         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 496         Arguments::set_endorsed_dirs(buf);
 497     }
 498   }
 499 
 500 #undef malloc
 501 #undef getenv
 502 #undef EXTENSIONS_DIR
 503 #undef ENDORSED_DIR
 504 
 505   // Done
 506   return;
 507 }
 508 
 509 ////////////////////////////////////////////////////////////////////////////////
 510 // breakpoint support
 511 
 512 void os::breakpoint() {
 513   BREAKPOINT;
 514 }
 515 
 516 extern "C" void breakpoint() {
 517   // use debugger to set breakpoint here
 518 }
 519 
 520 ////////////////////////////////////////////////////////////////////////////////
 521 // signal support
 522 
 523 debug_only(static bool signal_sets_initialized = false);
 524 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 525 
 526 bool os::Linux::is_sig_ignored(int sig) {
 527       struct sigaction oact;
 528       sigaction(sig, (struct sigaction*)NULL, &oact);
 529       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 530                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 531       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 532            return true;
 533       else
 534            return false;
 535 }
 536 
 537 void os::Linux::signal_sets_init() {
 538   // Should also have an assertion stating we are still single-threaded.
 539   assert(!signal_sets_initialized, "Already initialized");
 540   // Fill in signals that are necessarily unblocked for all threads in
 541   // the VM. Currently, we unblock the following signals:
 542   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 543   //                         by -Xrs (=ReduceSignalUsage));
 544   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 545   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 546   // the dispositions or masks wrt these signals.
 547   // Programs embedding the VM that want to use the above signals for their
 548   // own purposes must, at this time, use the "-Xrs" option to prevent
 549   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 550   // (See bug 4345157, and other related bugs).
 551   // In reality, though, unblocking these signals is really a nop, since
 552   // these signals are not blocked by default.
 553   sigemptyset(&unblocked_sigs);
 554   sigemptyset(&allowdebug_blocked_sigs);
 555   sigaddset(&unblocked_sigs, SIGILL);
 556   sigaddset(&unblocked_sigs, SIGSEGV);
 557   sigaddset(&unblocked_sigs, SIGBUS);
 558   sigaddset(&unblocked_sigs, SIGFPE);
 559   sigaddset(&unblocked_sigs, SR_signum);
 560 
 561   if (!ReduceSignalUsage) {
 562    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 563       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 564       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 565    }
 566    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 567       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 568       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 569    }
 570    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 571       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 572       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 573    }
 574   }
 575   // Fill in signals that are blocked by all but the VM thread.
 576   sigemptyset(&vm_sigs);
 577   if (!ReduceSignalUsage)
 578     sigaddset(&vm_sigs, BREAK_SIGNAL);
 579   debug_only(signal_sets_initialized = true);
 580 
 581 }
 582 
 583 // These are signals that are unblocked while a thread is running Java.
 584 // (For some reason, they get blocked by default.)
 585 sigset_t* os::Linux::unblocked_signals() {
 586   assert(signal_sets_initialized, "Not initialized");
 587   return &unblocked_sigs;
 588 }
 589 
 590 // These are the signals that are blocked while a (non-VM) thread is
 591 // running Java. Only the VM thread handles these signals.
 592 sigset_t* os::Linux::vm_signals() {
 593   assert(signal_sets_initialized, "Not initialized");
 594   return &vm_sigs;
 595 }
 596 
 597 // These are signals that are blocked during cond_wait to allow debugger in
 598 sigset_t* os::Linux::allowdebug_blocked_signals() {
 599   assert(signal_sets_initialized, "Not initialized");
 600   return &allowdebug_blocked_sigs;
 601 }
 602 
 603 void os::Linux::hotspot_sigmask(Thread* thread) {
 604 
 605   //Save caller's signal mask before setting VM signal mask
 606   sigset_t caller_sigmask;
 607   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 608 
 609   OSThread* osthread = thread->osthread();
 610   osthread->set_caller_sigmask(caller_sigmask);
 611 
 612   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 613 
 614   if (!ReduceSignalUsage) {
 615     if (thread->is_VM_thread()) {
 616       // Only the VM thread handles BREAK_SIGNAL ...
 617       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 618     } else {
 619       // ... all other threads block BREAK_SIGNAL
 620       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 621     }
 622   }
 623 }
 624 
 625 //////////////////////////////////////////////////////////////////////////////
 626 // detecting pthread library
 627 
 628 void os::Linux::libpthread_init() {
 629   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 630   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 631   // generic name for earlier versions.
 632   // Define macros here so we can build HotSpot on old systems.
 633 # ifndef _CS_GNU_LIBC_VERSION
 634 # define _CS_GNU_LIBC_VERSION 2
 635 # endif
 636 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 637 # define _CS_GNU_LIBPTHREAD_VERSION 3
 638 # endif
 639 
 640   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 641   if (n > 0) {
 642      char *str = (char *)malloc(n);
 643      confstr(_CS_GNU_LIBC_VERSION, str, n);
 644      os::Linux::set_glibc_version(str);
 645   } else {
 646      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 647      static char _gnu_libc_version[32];
 648      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 649               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 650      os::Linux::set_glibc_version(_gnu_libc_version);
 651   }
 652 
 653   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 654   if (n > 0) {
 655      char *str = (char *)malloc(n);
 656      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 657      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 658      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 659      // is the case. LinuxThreads has a hard limit on max number of threads.
 660      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 661      // On the other hand, NPTL does not have such a limit, sysconf()
 662      // will return -1 and errno is not changed. Check if it is really NPTL.
 663      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 664          strstr(str, "NPTL") &&
 665          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 666        free(str);
 667        os::Linux::set_libpthread_version("linuxthreads");
 668      } else {
 669        os::Linux::set_libpthread_version(str);
 670      }
 671   } else {
 672     // glibc before 2.3.2 only has LinuxThreads.
 673     os::Linux::set_libpthread_version("linuxthreads");
 674   }
 675 
 676   if (strstr(libpthread_version(), "NPTL")) {
 677      os::Linux::set_is_NPTL();
 678   } else {
 679      os::Linux::set_is_LinuxThreads();
 680   }
 681 
 682   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 683   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 684   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 685      os::Linux::set_is_floating_stack();
 686   }
 687 }
 688 
 689 /////////////////////////////////////////////////////////////////////////////
 690 // thread stack
 691 
 692 // Force Linux kernel to expand current thread stack. If "bottom" is close
 693 // to the stack guard, caller should block all signals.
 694 //
 695 // MAP_GROWSDOWN:
 696 //   A special mmap() flag that is used to implement thread stacks. It tells
 697 //   kernel that the memory region should extend downwards when needed. This
 698 //   allows early versions of LinuxThreads to only mmap the first few pages
 699 //   when creating a new thread. Linux kernel will automatically expand thread
 700 //   stack as needed (on page faults).
 701 //
 702 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 703 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 704 //   region, it's hard to tell if the fault is due to a legitimate stack
 705 //   access or because of reading/writing non-exist memory (e.g. buffer
 706 //   overrun). As a rule, if the fault happens below current stack pointer,
 707 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 708 //   application (see Linux kernel fault.c).
 709 //
 710 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 711 //   stack overflow detection.
 712 //
 713 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 714 //   not use this flag. However, the stack of initial thread is not created
 715 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 716 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 717 //   and then attach the thread to JVM.
 718 //
 719 // To get around the problem and allow stack banging on Linux, we need to
 720 // manually expand thread stack after receiving the SIGSEGV.
 721 //
 722 // There are two ways to expand thread stack to address "bottom", we used
 723 // both of them in JVM before 1.5:
 724 //   1. adjust stack pointer first so that it is below "bottom", and then
 725 //      touch "bottom"
 726 //   2. mmap() the page in question
 727 //
 728 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 729 // if current sp is already near the lower end of page 101, and we need to
 730 // call mmap() to map page 100, it is possible that part of the mmap() frame
 731 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 732 // That will destroy the mmap() frame and cause VM to crash.
 733 //
 734 // The following code works by adjusting sp first, then accessing the "bottom"
 735 // page to force a page fault. Linux kernel will then automatically expand the
 736 // stack mapping.
 737 //
 738 // _expand_stack_to() assumes its frame size is less than page size, which
 739 // should always be true if the function is not inlined.
 740 
 741 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 742 #define NOINLINE
 743 #else
 744 #define NOINLINE __attribute__ ((noinline))
 745 #endif
 746 
 747 static void _expand_stack_to(address bottom) NOINLINE;
 748 
 749 static void _expand_stack_to(address bottom) {
 750   address sp;
 751   size_t size;
 752   volatile char *p;
 753 
 754   // Adjust bottom to point to the largest address within the same page, it
 755   // gives us a one-page buffer if alloca() allocates slightly more memory.
 756   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 757   bottom += os::Linux::page_size() - 1;
 758 
 759   // sp might be slightly above current stack pointer; if that's the case, we
 760   // will alloca() a little more space than necessary, which is OK. Don't use
 761   // os::current_stack_pointer(), as its result can be slightly below current
 762   // stack pointer, causing us to not alloca enough to reach "bottom".
 763   sp = (address)&sp;
 764 
 765   if (sp > bottom) {
 766     size = sp - bottom;
 767     p = (volatile char *)alloca(size);
 768     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 769     p[0] = '\0';
 770   }
 771 }
 772 
 773 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 774   assert(t!=NULL, "just checking");
 775   assert(t->osthread()->expanding_stack(), "expand should be set");
 776   assert(t->stack_base() != NULL, "stack_base was not initialized");
 777 
 778   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 779     sigset_t mask_all, old_sigset;
 780     sigfillset(&mask_all);
 781     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 782     _expand_stack_to(addr);
 783     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 784     return true;
 785   }
 786   return false;
 787 }
 788 
 789 //////////////////////////////////////////////////////////////////////////////
 790 // create new thread
 791 
 792 static address highest_vm_reserved_address();
 793 
 794 // check if it's safe to start a new thread
 795 static bool _thread_safety_check(Thread* thread) {
 796   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 797     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 798     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 799     //   allocated (MAP_FIXED) from high address space. Every thread stack
 800     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 801     //   it to other values if they rebuild LinuxThreads).
 802     //
 803     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 804     // the memory region has already been mmap'ed. That means if we have too
 805     // many threads and/or very large heap, eventually thread stack will
 806     // collide with heap.
 807     //
 808     // Here we try to prevent heap/stack collision by comparing current
 809     // stack bottom with the highest address that has been mmap'ed by JVM
 810     // plus a safety margin for memory maps created by native code.
 811     //
 812     // This feature can be disabled by setting ThreadSafetyMargin to 0
 813     //
 814     if (ThreadSafetyMargin > 0) {
 815       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 816 
 817       // not safe if our stack extends below the safety margin
 818       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 819     } else {
 820       return true;
 821     }
 822   } else {
 823     // Floating stack LinuxThreads or NPTL:
 824     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 825     //   there's not enough space left, pthread_create() will fail. If we come
 826     //   here, that means enough space has been reserved for stack.
 827     return true;
 828   }
 829 }
 830 
 831 // Thread start routine for all newly created threads
 832 static void *java_start(Thread *thread) {
 833   // Try to randomize the cache line index of hot stack frames.
 834   // This helps when threads of the same stack traces evict each other's
 835   // cache lines. The threads can be either from the same JVM instance, or
 836   // from different JVM instances. The benefit is especially true for
 837   // processors with hyperthreading technology.
 838   static int counter = 0;
 839   int pid = os::current_process_id();
 840   alloca(((pid ^ counter++) & 7) * 128);
 841 
 842   ThreadLocalStorage::set_thread(thread);
 843 
 844   OSThread* osthread = thread->osthread();
 845   Monitor* sync = osthread->startThread_lock();
 846 
 847   // non floating stack LinuxThreads needs extra check, see above
 848   if (!_thread_safety_check(thread)) {
 849     // notify parent thread
 850     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 851     osthread->set_state(ZOMBIE);
 852     sync->notify_all();
 853     return NULL;
 854   }
 855 
 856   // thread_id is kernel thread id (similar to Solaris LWP id)
 857   osthread->set_thread_id(os::Linux::gettid());
 858 
 859   if (UseNUMA) {
 860     int lgrp_id = os::numa_get_group_id();
 861     if (lgrp_id != -1) {
 862       thread->set_lgrp_id(lgrp_id);
 863     }
 864   }
 865   // initialize signal mask for this thread
 866   os::Linux::hotspot_sigmask(thread);
 867 
 868   // initialize floating point control register
 869   os::Linux::init_thread_fpu_state();
 870 
 871   // handshaking with parent thread
 872   {
 873     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 874 
 875     // notify parent thread
 876     osthread->set_state(INITIALIZED);
 877     sync->notify_all();
 878 
 879     // wait until os::start_thread()
 880     while (osthread->get_state() == INITIALIZED) {
 881       sync->wait(Mutex::_no_safepoint_check_flag);
 882     }
 883   }
 884 
 885   // call one more level start routine
 886   thread->run();
 887 
 888   return 0;
 889 }
 890 
 891 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 892   assert(thread->osthread() == NULL, "caller responsible");
 893 
 894   // Allocate the OSThread object
 895   OSThread* osthread = new OSThread(NULL, NULL);
 896   if (osthread == NULL) {
 897     return false;
 898   }
 899 
 900   // set the correct thread state
 901   osthread->set_thread_type(thr_type);
 902 
 903   // Initial state is ALLOCATED but not INITIALIZED
 904   osthread->set_state(ALLOCATED);
 905 
 906   thread->set_osthread(osthread);
 907 
 908   // init thread attributes
 909   pthread_attr_t attr;
 910   pthread_attr_init(&attr);
 911   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 912 
 913   // stack size
 914   if (os::Linux::supports_variable_stack_size()) {
 915     // calculate stack size if it's not specified by caller
 916     if (stack_size == 0) {
 917       stack_size = os::Linux::default_stack_size(thr_type);
 918 
 919       switch (thr_type) {
 920       case os::java_thread:
 921         // Java threads use ThreadStackSize which default value can be
 922         // changed with the flag -Xss
 923         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 924         stack_size = JavaThread::stack_size_at_create();
 925         break;
 926       case os::compiler_thread:
 927         if (CompilerThreadStackSize > 0) {
 928           stack_size = (size_t)(CompilerThreadStackSize * K);
 929           break;
 930         } // else fall through:
 931           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 932       case os::vm_thread:
 933       case os::pgc_thread:
 934       case os::cgc_thread:
 935       case os::watcher_thread:
 936         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 937         break;
 938       }
 939     }
 940 
 941     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 942     pthread_attr_setstacksize(&attr, stack_size);
 943   } else {
 944     // let pthread_create() pick the default value.
 945   }
 946 
 947   // glibc guard page
 948   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 949 
 950   ThreadState state;
 951 
 952   {
 953     // Serialize thread creation if we are running with fixed stack LinuxThreads
 954     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 955     if (lock) {
 956       os::Linux::createThread_lock()->lock_without_safepoint_check();
 957     }
 958 
 959     pthread_t tid;
 960     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 961 
 962     pthread_attr_destroy(&attr);
 963 
 964     if (ret != 0) {
 965       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 966         perror("pthread_create()");
 967       }
 968       // Need to clean up stuff we've allocated so far
 969       thread->set_osthread(NULL);
 970       delete osthread;
 971       if (lock) os::Linux::createThread_lock()->unlock();
 972       return false;
 973     }
 974 
 975     // Store pthread info into the OSThread
 976     osthread->set_pthread_id(tid);
 977 
 978     // Wait until child thread is either initialized or aborted
 979     {
 980       Monitor* sync_with_child = osthread->startThread_lock();
 981       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 982       while ((state = osthread->get_state()) == ALLOCATED) {
 983         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 984       }
 985     }
 986 
 987     if (lock) {
 988       os::Linux::createThread_lock()->unlock();
 989     }
 990   }
 991 
 992   // Aborted due to thread limit being reached
 993   if (state == ZOMBIE) {
 994       thread->set_osthread(NULL);
 995       delete osthread;
 996       return false;
 997   }
 998 
 999   // The thread is returned suspended (in state INITIALIZED),
1000   // and is started higher up in the call chain
1001   assert(state == INITIALIZED, "race condition");
1002   return true;
1003 }
1004 
1005 /////////////////////////////////////////////////////////////////////////////
1006 // attach existing thread
1007 
1008 // bootstrap the main thread
1009 bool os::create_main_thread(JavaThread* thread) {
1010   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
1011   return create_attached_thread(thread);
1012 }
1013 
1014 bool os::create_attached_thread(JavaThread* thread) {
1015 #ifdef ASSERT
1016     thread->verify_not_published();
1017 #endif
1018 
1019   // Allocate the OSThread object
1020   OSThread* osthread = new OSThread(NULL, NULL);
1021 
1022   if (osthread == NULL) {
1023     return false;
1024   }
1025 
1026   // Store pthread info into the OSThread
1027   osthread->set_thread_id(os::Linux::gettid());
1028   osthread->set_pthread_id(::pthread_self());
1029 
1030   // initialize floating point control register
1031   os::Linux::init_thread_fpu_state();
1032 
1033   // Initial thread state is RUNNABLE
1034   osthread->set_state(RUNNABLE);
1035 
1036   thread->set_osthread(osthread);
1037 
1038   if (UseNUMA) {
1039     int lgrp_id = os::numa_get_group_id();
1040     if (lgrp_id != -1) {
1041       thread->set_lgrp_id(lgrp_id);
1042     }
1043   }
1044 
1045   if (os::Linux::is_initial_thread()) {
1046     // If current thread is initial thread, its stack is mapped on demand,
1047     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1048     // the entire stack region to avoid SEGV in stack banging.
1049     // It is also useful to get around the heap-stack-gap problem on SuSE
1050     // kernel (see 4821821 for details). We first expand stack to the top
1051     // of yellow zone, then enable stack yellow zone (order is significant,
1052     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1053     // is no gap between the last two virtual memory regions.
1054 
1055     JavaThread *jt = (JavaThread *)thread;
1056     address addr = jt->stack_yellow_zone_base();
1057     assert(addr != NULL, "initialization problem?");
1058     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1059 
1060     osthread->set_expanding_stack();
1061     os::Linux::manually_expand_stack(jt, addr);
1062     osthread->clear_expanding_stack();
1063   }
1064 
1065   // initialize signal mask for this thread
1066   // and save the caller's signal mask
1067   os::Linux::hotspot_sigmask(thread);
1068 
1069   return true;
1070 }
1071 
1072 void os::pd_start_thread(Thread* thread) {
1073   OSThread * osthread = thread->osthread();
1074   assert(osthread->get_state() != INITIALIZED, "just checking");
1075   Monitor* sync_with_child = osthread->startThread_lock();
1076   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1077   sync_with_child->notify();
1078 }
1079 
1080 // Free Linux resources related to the OSThread
1081 void os::free_thread(OSThread* osthread) {
1082   assert(osthread != NULL, "osthread not set");
1083 
1084   if (Thread::current()->osthread() == osthread) {
1085     // Restore caller's signal mask
1086     sigset_t sigmask = osthread->caller_sigmask();
1087     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1088    }
1089 
1090   delete osthread;
1091 }
1092 
1093 //////////////////////////////////////////////////////////////////////////////
1094 // thread local storage
1095 
1096 int os::allocate_thread_local_storage() {
1097   pthread_key_t key;
1098   int rslt = pthread_key_create(&key, NULL);
1099   assert(rslt == 0, "cannot allocate thread local storage");
1100   return (int)key;
1101 }
1102 
1103 // Note: This is currently not used by VM, as we don't destroy TLS key
1104 // on VM exit.
1105 void os::free_thread_local_storage(int index) {
1106   int rslt = pthread_key_delete((pthread_key_t)index);
1107   assert(rslt == 0, "invalid index");
1108 }
1109 
1110 void os::thread_local_storage_at_put(int index, void* value) {
1111   int rslt = pthread_setspecific((pthread_key_t)index, value);
1112   assert(rslt == 0, "pthread_setspecific failed");
1113 }
1114 
1115 extern "C" Thread* get_thread() {
1116   return ThreadLocalStorage::thread();
1117 }
1118 
1119 //////////////////////////////////////////////////////////////////////////////
1120 // initial thread
1121 
1122 // Check if current thread is the initial thread, similar to Solaris thr_main.
1123 bool os::Linux::is_initial_thread(void) {
1124   char dummy;
1125   // If called before init complete, thread stack bottom will be null.
1126   // Can be called if fatal error occurs before initialization.
1127   if (initial_thread_stack_bottom() == NULL) return false;
1128   assert(initial_thread_stack_bottom() != NULL &&
1129          initial_thread_stack_size()   != 0,
1130          "os::init did not locate initial thread's stack region");
1131   if ((address)&dummy >= initial_thread_stack_bottom() &&
1132       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1133        return true;
1134   else return false;
1135 }
1136 
1137 // Find the virtual memory area that contains addr
1138 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1139   FILE *fp = fopen("/proc/self/maps", "r");
1140   if (fp) {
1141     address low, high;
1142     while (!feof(fp)) {
1143       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1144         if (low <= addr && addr < high) {
1145            if (vma_low)  *vma_low  = low;
1146            if (vma_high) *vma_high = high;
1147            fclose (fp);
1148            return true;
1149         }
1150       }
1151       for (;;) {
1152         int ch = fgetc(fp);
1153         if (ch == EOF || ch == (int)'\n') break;
1154       }
1155     }
1156     fclose(fp);
1157   }
1158   return false;
1159 }
1160 
1161 // Locate initial thread stack. This special handling of initial thread stack
1162 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1163 // bogus value for initial thread.
1164 void os::Linux::capture_initial_stack(size_t max_size) {
1165   // stack size is the easy part, get it from RLIMIT_STACK
1166   size_t stack_size;
1167   struct rlimit rlim;
1168   getrlimit(RLIMIT_STACK, &rlim);
1169   stack_size = rlim.rlim_cur;
1170 
1171   // 6308388: a bug in ld.so will relocate its own .data section to the
1172   //   lower end of primordial stack; reduce ulimit -s value a little bit
1173   //   so we won't install guard page on ld.so's data section.
1174   stack_size -= 2 * page_size();
1175 
1176   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1177   //   7.1, in both cases we will get 2G in return value.
1178   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1179   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1180   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1181   //   in case other parts in glibc still assumes 2M max stack size.
1182   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1183 #ifndef IA64
1184   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1185 #else
1186   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1187   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1188 #endif
1189 
1190   // Try to figure out where the stack base (top) is. This is harder.
1191   //
1192   // When an application is started, glibc saves the initial stack pointer in
1193   // a global variable "__libc_stack_end", which is then used by system
1194   // libraries. __libc_stack_end should be pretty close to stack top. The
1195   // variable is available since the very early days. However, because it is
1196   // a private interface, it could disappear in the future.
1197   //
1198   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1199   // to __libc_stack_end, it is very close to stack top, but isn't the real
1200   // stack top. Note that /proc may not exist if VM is running as a chroot
1201   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1202   // /proc/<pid>/stat could change in the future (though unlikely).
1203   //
1204   // We try __libc_stack_end first. If that doesn't work, look for
1205   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1206   // as a hint, which should work well in most cases.
1207 
1208   uintptr_t stack_start;
1209 
1210   // try __libc_stack_end first
1211   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1212   if (p && *p) {
1213     stack_start = *p;
1214   } else {
1215     // see if we can get the start_stack field from /proc/self/stat
1216     FILE *fp;
1217     int pid;
1218     char state;
1219     int ppid;
1220     int pgrp;
1221     int session;
1222     int nr;
1223     int tpgrp;
1224     unsigned long flags;
1225     unsigned long minflt;
1226     unsigned long cminflt;
1227     unsigned long majflt;
1228     unsigned long cmajflt;
1229     unsigned long utime;
1230     unsigned long stime;
1231     long cutime;
1232     long cstime;
1233     long prio;
1234     long nice;
1235     long junk;
1236     long it_real;
1237     uintptr_t start;
1238     uintptr_t vsize;
1239     intptr_t rss;
1240     uintptr_t rsslim;
1241     uintptr_t scodes;
1242     uintptr_t ecode;
1243     int i;
1244 
1245     // Figure what the primordial thread stack base is. Code is inspired
1246     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1247     // followed by command name surrounded by parentheses, state, etc.
1248     char stat[2048];
1249     int statlen;
1250 
1251     fp = fopen("/proc/self/stat", "r");
1252     if (fp) {
1253       statlen = fread(stat, 1, 2047, fp);
1254       stat[statlen] = '\0';
1255       fclose(fp);
1256 
1257       // Skip pid and the command string. Note that we could be dealing with
1258       // weird command names, e.g. user could decide to rename java launcher
1259       // to "java 1.4.2 :)", then the stat file would look like
1260       //                1234 (java 1.4.2 :)) R ... ...
1261       // We don't really need to know the command string, just find the last
1262       // occurrence of ")" and then start parsing from there. See bug 4726580.
1263       char * s = strrchr(stat, ')');
1264 
1265       i = 0;
1266       if (s) {
1267         // Skip blank chars
1268         do s++; while (isspace(*s));
1269 
1270 #define _UFM UINTX_FORMAT
1271 #define _DFM INTX_FORMAT
1272 
1273         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1274         /*              3  4  5  6  7  8   9   0   1   2   3   4   5   6   7   8   9   0   1    2    3    4    5    6    7    8 */
1275         i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1276              &state,          /* 3  %c  */
1277              &ppid,           /* 4  %d  */
1278              &pgrp,           /* 5  %d  */
1279              &session,        /* 6  %d  */
1280              &nr,             /* 7  %d  */
1281              &tpgrp,          /* 8  %d  */
1282              &flags,          /* 9  %lu  */
1283              &minflt,         /* 10 %lu  */
1284              &cminflt,        /* 11 %lu  */
1285              &majflt,         /* 12 %lu  */
1286              &cmajflt,        /* 13 %lu  */
1287              &utime,          /* 14 %lu  */
1288              &stime,          /* 15 %lu  */
1289              &cutime,         /* 16 %ld  */
1290              &cstime,         /* 17 %ld  */
1291              &prio,           /* 18 %ld  */
1292              &nice,           /* 19 %ld  */
1293              &junk,           /* 20 %ld  */
1294              &it_real,        /* 21 %ld  */
1295              &start,          /* 22 UINTX_FORMAT */
1296              &vsize,          /* 23 UINTX_FORMAT */
1297              &rss,            /* 24 INTX_FORMAT  */
1298              &rsslim,         /* 25 UINTX_FORMAT */
1299              &scodes,         /* 26 UINTX_FORMAT */
1300              &ecode,          /* 27 UINTX_FORMAT */
1301              &stack_start);   /* 28 UINTX_FORMAT */
1302       }
1303 
1304 #undef _UFM
1305 #undef _DFM
1306 
1307       if (i != 28 - 2) {
1308          assert(false, "Bad conversion from /proc/self/stat");
1309          // product mode - assume we are the initial thread, good luck in the
1310          // embedded case.
1311          warning("Can't detect initial thread stack location - bad conversion");
1312          stack_start = (uintptr_t) &rlim;
1313       }
1314     } else {
1315       // For some reason we can't open /proc/self/stat (for example, running on
1316       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1317       // most cases, so don't abort:
1318       warning("Can't detect initial thread stack location - no /proc/self/stat");
1319       stack_start = (uintptr_t) &rlim;
1320     }
1321   }
1322 
1323   // Now we have a pointer (stack_start) very close to the stack top, the
1324   // next thing to do is to figure out the exact location of stack top. We
1325   // can find out the virtual memory area that contains stack_start by
1326   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1327   // and its upper limit is the real stack top. (again, this would fail if
1328   // running inside chroot, because /proc may not exist.)
1329 
1330   uintptr_t stack_top;
1331   address low, high;
1332   if (find_vma((address)stack_start, &low, &high)) {
1333     // success, "high" is the true stack top. (ignore "low", because initial
1334     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1335     stack_top = (uintptr_t)high;
1336   } else {
1337     // failed, likely because /proc/self/maps does not exist
1338     warning("Can't detect initial thread stack location - find_vma failed");
1339     // best effort: stack_start is normally within a few pages below the real
1340     // stack top, use it as stack top, and reduce stack size so we won't put
1341     // guard page outside stack.
1342     stack_top = stack_start;
1343     stack_size -= 16 * page_size();
1344   }
1345 
1346   // stack_top could be partially down the page so align it
1347   stack_top = align_size_up(stack_top, page_size());
1348 
1349   if (max_size && stack_size > max_size) {
1350      _initial_thread_stack_size = max_size;
1351   } else {
1352      _initial_thread_stack_size = stack_size;
1353   }
1354 
1355   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1356   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1357 }
1358 
1359 ////////////////////////////////////////////////////////////////////////////////
1360 // time support
1361 
1362 // Time since start-up in seconds to a fine granularity.
1363 // Used by VMSelfDestructTimer and the MemProfiler.
1364 double os::elapsedTime() {
1365 
1366   return (double)(os::elapsed_counter()) * 0.000001;
1367 }
1368 
1369 jlong os::elapsed_counter() {
1370   timeval time;
1371   int status = gettimeofday(&time, NULL);
1372   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1373 }
1374 
1375 jlong os::elapsed_frequency() {
1376   return (1000 * 1000);
1377 }
1378 
1379 // For now, we say that linux does not support vtime.  I have no idea
1380 // whether it can actually be made to (DLD, 9/13/05).
1381 
1382 bool os::supports_vtime() { return false; }
1383 bool os::enable_vtime()   { return false; }
1384 bool os::vtime_enabled()  { return false; }
1385 double os::elapsedVTime() {
1386   // better than nothing, but not much
1387   return elapsedTime();
1388 }
1389 
1390 jlong os::javaTimeMillis() {
1391   timeval time;
1392   int status = gettimeofday(&time, NULL);
1393   assert(status != -1, "linux error");
1394   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1395 }
1396 
1397 #ifndef CLOCK_MONOTONIC
1398 #define CLOCK_MONOTONIC (1)
1399 #endif
1400 
1401 void os::Linux::clock_init() {
1402   // we do dlopen's in this particular order due to bug in linux
1403   // dynamical loader (see 6348968) leading to crash on exit
1404   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1405   if (handle == NULL) {
1406     handle = dlopen("librt.so", RTLD_LAZY);
1407   }
1408 
1409   if (handle) {
1410     int (*clock_getres_func)(clockid_t, struct timespec*) =
1411            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1412     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1413            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1414     if (clock_getres_func && clock_gettime_func) {
1415       // See if monotonic clock is supported by the kernel. Note that some
1416       // early implementations simply return kernel jiffies (updated every
1417       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1418       // for nano time (though the monotonic property is still nice to have).
1419       // It's fixed in newer kernels, however clock_getres() still returns
1420       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1421       // resolution for now. Hopefully as people move to new kernels, this
1422       // won't be a problem.
1423       struct timespec res;
1424       struct timespec tp;
1425       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1426           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1427         // yes, monotonic clock is supported
1428         _clock_gettime = clock_gettime_func;
1429       } else {
1430         // close librt if there is no monotonic clock
1431         dlclose(handle);
1432       }
1433     }
1434   }
1435 }
1436 
1437 #ifndef SYS_clock_getres
1438 
1439 #if defined(IA32) || defined(AMD64)
1440 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1441 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1442 #else
1443 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1444 #define sys_clock_getres(x,y)  -1
1445 #endif
1446 
1447 #else
1448 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1449 #endif
1450 
1451 void os::Linux::fast_thread_clock_init() {
1452   if (!UseLinuxPosixThreadCPUClocks) {
1453     return;
1454   }
1455   clockid_t clockid;
1456   struct timespec tp;
1457   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1458       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1459 
1460   // Switch to using fast clocks for thread cpu time if
1461   // the sys_clock_getres() returns 0 error code.
1462   // Note, that some kernels may support the current thread
1463   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1464   // returned by the pthread_getcpuclockid().
1465   // If the fast Posix clocks are supported then the sys_clock_getres()
1466   // must return at least tp.tv_sec == 0 which means a resolution
1467   // better than 1 sec. This is extra check for reliability.
1468 
1469   if(pthread_getcpuclockid_func &&
1470      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1471      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1472 
1473     _supports_fast_thread_cpu_time = true;
1474     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1475   }
1476 }
1477 
1478 jlong os::javaTimeNanos() {
1479   if (Linux::supports_monotonic_clock()) {
1480     struct timespec tp;
1481     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1482     assert(status == 0, "gettime error");
1483     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1484     return result;
1485   } else {
1486     timeval time;
1487     int status = gettimeofday(&time, NULL);
1488     assert(status != -1, "linux error");
1489     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1490     return 1000 * usecs;
1491   }
1492 }
1493 
1494 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1495   if (Linux::supports_monotonic_clock()) {
1496     info_ptr->max_value = ALL_64_BITS;
1497 
1498     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1499     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1500     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1501   } else {
1502     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1503     info_ptr->max_value = ALL_64_BITS;
1504 
1505     // gettimeofday is a real time clock so it skips
1506     info_ptr->may_skip_backward = true;
1507     info_ptr->may_skip_forward = true;
1508   }
1509 
1510   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1511 }
1512 
1513 // Return the real, user, and system times in seconds from an
1514 // arbitrary fixed point in the past.
1515 bool os::getTimesSecs(double* process_real_time,
1516                       double* process_user_time,
1517                       double* process_system_time) {
1518   struct tms ticks;
1519   clock_t real_ticks = times(&ticks);
1520 
1521   if (real_ticks == (clock_t) (-1)) {
1522     return false;
1523   } else {
1524     double ticks_per_second = (double) clock_tics_per_sec;
1525     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1526     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1527     *process_real_time = ((double) real_ticks) / ticks_per_second;
1528 
1529     return true;
1530   }
1531 }
1532 
1533 
1534 char * os::local_time_string(char *buf, size_t buflen) {
1535   struct tm t;
1536   time_t long_time;
1537   time(&long_time);
1538   localtime_r(&long_time, &t);
1539   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1540                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1541                t.tm_hour, t.tm_min, t.tm_sec);
1542   return buf;
1543 }
1544 
1545 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1546   return localtime_r(clock, res);
1547 }
1548 
1549 ////////////////////////////////////////////////////////////////////////////////
1550 // runtime exit support
1551 
1552 // Note: os::shutdown() might be called very early during initialization, or
1553 // called from signal handler. Before adding something to os::shutdown(), make
1554 // sure it is async-safe and can handle partially initialized VM.
1555 void os::shutdown() {
1556 
1557   // allow PerfMemory to attempt cleanup of any persistent resources
1558   perfMemory_exit();
1559 
1560   // needs to remove object in file system
1561   AttachListener::abort();
1562 
1563   // flush buffered output, finish log files
1564   ostream_abort();
1565 
1566   // Check for abort hook
1567   abort_hook_t abort_hook = Arguments::abort_hook();
1568   if (abort_hook != NULL) {
1569     abort_hook();
1570   }
1571 
1572 }
1573 
1574 // Note: os::abort() might be called very early during initialization, or
1575 // called from signal handler. Before adding something to os::abort(), make
1576 // sure it is async-safe and can handle partially initialized VM.
1577 void os::abort(bool dump_core) {
1578   os::shutdown();
1579   if (dump_core) {
1580 #ifndef PRODUCT
1581     fdStream out(defaultStream::output_fd());
1582     out.print_raw("Current thread is ");
1583     char buf[16];
1584     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1585     out.print_raw_cr(buf);
1586     out.print_raw_cr("Dumping core ...");
1587 #endif
1588     ::abort(); // dump core
1589   }
1590 
1591   ::exit(1);
1592 }
1593 
1594 // Die immediately, no exit hook, no abort hook, no cleanup.
1595 void os::die() {
1596   // _exit() on LinuxThreads only kills current thread
1597   ::abort();
1598 }
1599 
1600 // unused on linux for now.
1601 void os::set_error_file(const char *logfile) {}
1602 
1603 
1604 // This method is a copy of JDK's sysGetLastErrorString
1605 // from src/solaris/hpi/src/system_md.c
1606 
1607 size_t os::lasterror(char *buf, size_t len) {
1608 
1609   if (errno == 0)  return 0;
1610 
1611   const char *s = ::strerror(errno);
1612   size_t n = ::strlen(s);
1613   if (n >= len) {
1614     n = len - 1;
1615   }
1616   ::strncpy(buf, s, n);
1617   buf[n] = '\0';
1618   return n;
1619 }
1620 
1621 intx os::current_thread_id() { return (intx)pthread_self(); }
1622 int os::current_process_id() {
1623 
1624   // Under the old linux thread library, linux gives each thread
1625   // its own process id. Because of this each thread will return
1626   // a different pid if this method were to return the result
1627   // of getpid(2). Linux provides no api that returns the pid
1628   // of the launcher thread for the vm. This implementation
1629   // returns a unique pid, the pid of the launcher thread
1630   // that starts the vm 'process'.
1631 
1632   // Under the NPTL, getpid() returns the same pid as the
1633   // launcher thread rather than a unique pid per thread.
1634   // Use gettid() if you want the old pre NPTL behaviour.
1635 
1636   // if you are looking for the result of a call to getpid() that
1637   // returns a unique pid for the calling thread, then look at the
1638   // OSThread::thread_id() method in osThread_linux.hpp file
1639 
1640   return (int)(_initial_pid ? _initial_pid : getpid());
1641 }
1642 
1643 // DLL functions
1644 
1645 const char* os::dll_file_extension() { return ".so"; }
1646 
1647 // This must be hard coded because it's the system's temporary
1648 // directory not the java application's temp directory, ala java.io.tmpdir.
1649 const char* os::get_temp_directory() { return "/tmp"; }
1650 
1651 static bool file_exists(const char* filename) {
1652   struct stat statbuf;
1653   if (filename == NULL || strlen(filename) == 0) {
1654     return false;
1655   }
1656   return os::stat(filename, &statbuf) == 0;
1657 }
1658 
1659 void os::dll_build_name(char* buffer, size_t buflen,
1660                         const char* pname, const char* fname) {
1661   // Copied from libhpi
1662   const size_t pnamelen = pname ? strlen(pname) : 0;
1663 
1664   // Quietly truncate on buffer overflow.  Should be an error.
1665   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1666       *buffer = '\0';
1667       return;
1668   }
1669 
1670   if (pnamelen == 0) {
1671     snprintf(buffer, buflen, "lib%s.so", fname);
1672   } else if (strchr(pname, *os::path_separator()) != NULL) {
1673     int n;
1674     char** pelements = split_path(pname, &n);
1675     for (int i = 0 ; i < n ; i++) {
1676       // Really shouldn't be NULL, but check can't hurt
1677       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1678         continue; // skip the empty path values
1679       }
1680       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1681       if (file_exists(buffer)) {
1682         break;
1683       }
1684     }
1685     // release the storage
1686     for (int i = 0 ; i < n ; i++) {
1687       if (pelements[i] != NULL) {
1688         FREE_C_HEAP_ARRAY(char, pelements[i]);
1689       }
1690     }
1691     if (pelements != NULL) {
1692       FREE_C_HEAP_ARRAY(char*, pelements);
1693     }
1694   } else {
1695     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1696   }
1697 }
1698 
1699 const char* os::get_current_directory(char *buf, int buflen) {
1700   return getcwd(buf, buflen);
1701 }
1702 
1703 // check if addr is inside libjvm[_g].so
1704 bool os::address_is_in_vm(address addr) {
1705   static address libjvm_base_addr;
1706   Dl_info dlinfo;
1707 
1708   if (libjvm_base_addr == NULL) {
1709     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1710     libjvm_base_addr = (address)dlinfo.dli_fbase;
1711     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1712   }
1713 
1714   if (dladdr((void *)addr, &dlinfo)) {
1715     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1716   }
1717 
1718   return false;
1719 }
1720 
1721 bool os::dll_address_to_function_name(address addr, char *buf,
1722                                       int buflen, int *offset) {
1723   Dl_info dlinfo;
1724 
1725   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1726     if (buf != NULL) {
1727       if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1728         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1729       }
1730     }
1731     if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1732     return true;
1733   } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1734     if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1735         buf, buflen, offset, dlinfo.dli_fname)) {
1736        return true;
1737     }
1738   }
1739 
1740   if (buf != NULL) buf[0] = '\0';
1741   if (offset != NULL) *offset = -1;
1742   return false;
1743 }
1744 
1745 struct _address_to_library_name {
1746   address addr;          // input : memory address
1747   size_t  buflen;        //         size of fname
1748   char*   fname;         // output: library name
1749   address base;          //         library base addr
1750 };
1751 
1752 static int address_to_library_name_callback(struct dl_phdr_info *info,
1753                                             size_t size, void *data) {
1754   int i;
1755   bool found = false;
1756   address libbase = NULL;
1757   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1758 
1759   // iterate through all loadable segments
1760   for (i = 0; i < info->dlpi_phnum; i++) {
1761     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1762     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1763       // base address of a library is the lowest address of its loaded
1764       // segments.
1765       if (libbase == NULL || libbase > segbase) {
1766         libbase = segbase;
1767       }
1768       // see if 'addr' is within current segment
1769       if (segbase <= d->addr &&
1770           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1771         found = true;
1772       }
1773     }
1774   }
1775 
1776   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1777   // so dll_address_to_library_name() can fall through to use dladdr() which
1778   // can figure out executable name from argv[0].
1779   if (found && info->dlpi_name && info->dlpi_name[0]) {
1780     d->base = libbase;
1781     if (d->fname) {
1782       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1783     }
1784     return 1;
1785   }
1786   return 0;
1787 }
1788 
1789 bool os::dll_address_to_library_name(address addr, char* buf,
1790                                      int buflen, int* offset) {
1791   Dl_info dlinfo;
1792   struct _address_to_library_name data;
1793 
1794   // There is a bug in old glibc dladdr() implementation that it could resolve
1795   // to wrong library name if the .so file has a base address != NULL. Here
1796   // we iterate through the program headers of all loaded libraries to find
1797   // out which library 'addr' really belongs to. This workaround can be
1798   // removed once the minimum requirement for glibc is moved to 2.3.x.
1799   data.addr = addr;
1800   data.fname = buf;
1801   data.buflen = buflen;
1802   data.base = NULL;
1803   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1804 
1805   if (rslt) {
1806      // buf already contains library name
1807      if (offset) *offset = addr - data.base;
1808      return true;
1809   } else if (dladdr((void*)addr, &dlinfo)){
1810      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1811      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1812      return true;
1813   } else {
1814      if (buf) buf[0] = '\0';
1815      if (offset) *offset = -1;
1816      return false;
1817   }
1818 }
1819 
1820   // Loads .dll/.so and
1821   // in case of error it checks if .dll/.so was built for the
1822   // same architecture as Hotspot is running on
1823 
1824 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1825 {
1826   void * result= ::dlopen(filename, RTLD_LAZY);
1827   if (result != NULL) {
1828     // Successful loading
1829     return result;
1830   }
1831 
1832   Elf32_Ehdr elf_head;
1833 
1834   // Read system error message into ebuf
1835   // It may or may not be overwritten below
1836   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1837   ebuf[ebuflen-1]='\0';
1838   int diag_msg_max_length=ebuflen-strlen(ebuf);
1839   char* diag_msg_buf=ebuf+strlen(ebuf);
1840 
1841   if (diag_msg_max_length==0) {
1842     // No more space in ebuf for additional diagnostics message
1843     return NULL;
1844   }
1845 
1846 
1847   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1848 
1849   if (file_descriptor < 0) {
1850     // Can't open library, report dlerror() message
1851     return NULL;
1852   }
1853 
1854   bool failed_to_read_elf_head=
1855     (sizeof(elf_head)!=
1856         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1857 
1858   ::close(file_descriptor);
1859   if (failed_to_read_elf_head) {
1860     // file i/o error - report dlerror() msg
1861     return NULL;
1862   }
1863 
1864   typedef struct {
1865     Elf32_Half  code;         // Actual value as defined in elf.h
1866     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1867     char        elf_class;    // 32 or 64 bit
1868     char        endianess;    // MSB or LSB
1869     char*       name;         // String representation
1870   } arch_t;
1871 
1872   #ifndef EM_486
1873   #define EM_486          6               /* Intel 80486 */
1874   #endif
1875 
1876   static const arch_t arch_array[]={
1877     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1878     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1879     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1880     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1881     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1882     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1883     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1884     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1885     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1886     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1887     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1888     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1889     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1890     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1891     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1892     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1893   };
1894 
1895   #if  (defined IA32)
1896     static  Elf32_Half running_arch_code=EM_386;
1897   #elif   (defined AMD64)
1898     static  Elf32_Half running_arch_code=EM_X86_64;
1899   #elif  (defined IA64)
1900     static  Elf32_Half running_arch_code=EM_IA_64;
1901   #elif  (defined __sparc) && (defined _LP64)
1902     static  Elf32_Half running_arch_code=EM_SPARCV9;
1903   #elif  (defined __sparc) && (!defined _LP64)
1904     static  Elf32_Half running_arch_code=EM_SPARC;
1905   #elif  (defined __powerpc64__)
1906     static  Elf32_Half running_arch_code=EM_PPC64;
1907   #elif  (defined __powerpc__)
1908     static  Elf32_Half running_arch_code=EM_PPC;
1909   #elif  (defined ARM)
1910     static  Elf32_Half running_arch_code=EM_ARM;
1911   #elif  (defined S390)
1912     static  Elf32_Half running_arch_code=EM_S390;
1913   #elif  (defined ALPHA)
1914     static  Elf32_Half running_arch_code=EM_ALPHA;
1915   #elif  (defined MIPSEL)
1916     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1917   #elif  (defined PARISC)
1918     static  Elf32_Half running_arch_code=EM_PARISC;
1919   #elif  (defined MIPS)
1920     static  Elf32_Half running_arch_code=EM_MIPS;
1921   #elif  (defined M68K)
1922     static  Elf32_Half running_arch_code=EM_68K;
1923   #else
1924     #error Method os::dll_load requires that one of following is defined:\
1925          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1926   #endif
1927 
1928   // Identify compatability class for VM's architecture and library's architecture
1929   // Obtain string descriptions for architectures
1930 
1931   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1932   int running_arch_index=-1;
1933 
1934   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1935     if (running_arch_code == arch_array[i].code) {
1936       running_arch_index    = i;
1937     }
1938     if (lib_arch.code == arch_array[i].code) {
1939       lib_arch.compat_class = arch_array[i].compat_class;
1940       lib_arch.name         = arch_array[i].name;
1941     }
1942   }
1943 
1944   assert(running_arch_index != -1,
1945     "Didn't find running architecture code (running_arch_code) in arch_array");
1946   if (running_arch_index == -1) {
1947     // Even though running architecture detection failed
1948     // we may still continue with reporting dlerror() message
1949     return NULL;
1950   }
1951 
1952   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1953     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1954     return NULL;
1955   }
1956 
1957 #ifndef S390
1958   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1959     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1960     return NULL;
1961   }
1962 #endif // !S390
1963 
1964   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1965     if ( lib_arch.name!=NULL ) {
1966       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1967         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1968         lib_arch.name, arch_array[running_arch_index].name);
1969     } else {
1970       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1971       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1972         lib_arch.code,
1973         arch_array[running_arch_index].name);
1974     }
1975   }
1976 
1977   return NULL;
1978 }
1979 
1980 /*
1981  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
1982  * chances are you might want to run the generated bits against glibc-2.0
1983  * libdl.so, so always use locking for any version of glibc.
1984  */
1985 void* os::dll_lookup(void* handle, const char* name) {
1986   pthread_mutex_lock(&dl_mutex);
1987   void* res = dlsym(handle, name);
1988   pthread_mutex_unlock(&dl_mutex);
1989   return res;
1990 }
1991 
1992 
1993 static bool _print_ascii_file(const char* filename, outputStream* st) {
1994   int fd = ::open(filename, O_RDONLY);
1995   if (fd == -1) {
1996      return false;
1997   }
1998 
1999   char buf[32];
2000   int bytes;
2001   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2002     st->print_raw(buf, bytes);
2003   }
2004 
2005   ::close(fd);
2006 
2007   return true;
2008 }
2009 
2010 void os::print_dll_info(outputStream *st) {
2011    st->print_cr("Dynamic libraries:");
2012 
2013    char fname[32];
2014    pid_t pid = os::Linux::gettid();
2015 
2016    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2017 
2018    if (!_print_ascii_file(fname, st)) {
2019      st->print("Can not get library information for pid = %d\n", pid);
2020    }
2021 }
2022 
2023 void os::print_os_info_brief(outputStream* st) {
2024   os::Linux::print_distro_info(st);
2025 
2026   os::Posix::print_uname_info(st);
2027 
2028   os::Linux::print_libversion_info(st);
2029 
2030 }
2031 
2032 void os::print_os_info(outputStream* st) {
2033   st->print("OS:");
2034 
2035   os::Linux::print_distro_info(st);
2036 
2037   os::Posix::print_uname_info(st);
2038 
2039   // Print warning if unsafe chroot environment detected
2040   if (unsafe_chroot_detected) {
2041     st->print("WARNING!! ");
2042     st->print_cr(unstable_chroot_error);
2043   }
2044 
2045   os::Linux::print_libversion_info(st);
2046 
2047   os::Posix::print_rlimit_info(st);
2048 
2049   os::Posix::print_load_average(st);
2050 
2051   os::Linux::print_full_memory_info(st);
2052 }
2053 
2054 // Try to identify popular distros.
2055 // Most Linux distributions have /etc/XXX-release file, which contains
2056 // the OS version string. Some have more than one /etc/XXX-release file
2057 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2058 // so the order is important.
2059 void os::Linux::print_distro_info(outputStream* st) {
2060   if (!_print_ascii_file("/etc/mandrake-release", st) &&
2061       !_print_ascii_file("/etc/sun-release", st) &&
2062       !_print_ascii_file("/etc/redhat-release", st) &&
2063       !_print_ascii_file("/etc/SuSE-release", st) &&
2064       !_print_ascii_file("/etc/turbolinux-release", st) &&
2065       !_print_ascii_file("/etc/gentoo-release", st) &&
2066       !_print_ascii_file("/etc/debian_version", st) &&
2067       !_print_ascii_file("/etc/ltib-release", st) &&
2068       !_print_ascii_file("/etc/angstrom-version", st)) {
2069       st->print("Linux");
2070   }
2071   st->cr();
2072 }
2073 
2074 void os::Linux::print_libversion_info(outputStream* st) {
2075   // libc, pthread
2076   st->print("libc:");
2077   st->print(os::Linux::glibc_version()); st->print(" ");
2078   st->print(os::Linux::libpthread_version()); st->print(" ");
2079   if (os::Linux::is_LinuxThreads()) {
2080      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2081   }
2082   st->cr();
2083 }
2084 
2085 void os::Linux::print_full_memory_info(outputStream* st) {
2086    st->print("\n/proc/meminfo:\n");
2087    _print_ascii_file("/proc/meminfo", st);
2088    st->cr();
2089 }
2090 
2091 void os::print_memory_info(outputStream* st) {
2092 
2093   st->print("Memory:");
2094   st->print(" %dk page", os::vm_page_size()>>10);
2095 
2096   // values in struct sysinfo are "unsigned long"
2097   struct sysinfo si;
2098   sysinfo(&si);
2099 
2100   st->print(", physical " UINT64_FORMAT "k",
2101             os::physical_memory() >> 10);
2102   st->print("(" UINT64_FORMAT "k free)",
2103             os::available_memory() >> 10);
2104   st->print(", swap " UINT64_FORMAT "k",
2105             ((jlong)si.totalswap * si.mem_unit) >> 10);
2106   st->print("(" UINT64_FORMAT "k free)",
2107             ((jlong)si.freeswap * si.mem_unit) >> 10);
2108   st->cr();
2109 }
2110 
2111 void os::pd_print_cpu_info(outputStream* st) {
2112   st->print("\n/proc/cpuinfo:\n");
2113   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2114     st->print("  <Not Available>");
2115   }
2116   st->cr();
2117 }
2118 
2119 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2120 // but they're the same for all the linux arch that we support
2121 // and they're the same for solaris but there's no common place to put this.
2122 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2123                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2124                           "ILL_COPROC", "ILL_BADSTK" };
2125 
2126 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2127                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2128                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2129 
2130 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2131 
2132 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2133 
2134 void os::print_siginfo(outputStream* st, void* siginfo) {
2135   st->print("siginfo:");
2136 
2137   const int buflen = 100;
2138   char buf[buflen];
2139   siginfo_t *si = (siginfo_t*)siginfo;
2140   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2141   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2142     st->print("si_errno=%s", buf);
2143   } else {
2144     st->print("si_errno=%d", si->si_errno);
2145   }
2146   const int c = si->si_code;
2147   assert(c > 0, "unexpected si_code");
2148   switch (si->si_signo) {
2149   case SIGILL:
2150     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2151     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2152     break;
2153   case SIGFPE:
2154     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2155     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2156     break;
2157   case SIGSEGV:
2158     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2159     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2160     break;
2161   case SIGBUS:
2162     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2163     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2164     break;
2165   default:
2166     st->print(", si_code=%d", si->si_code);
2167     // no si_addr
2168   }
2169 
2170   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2171       UseSharedSpaces) {
2172     FileMapInfo* mapinfo = FileMapInfo::current_info();
2173     if (mapinfo->is_in_shared_space(si->si_addr)) {
2174       st->print("\n\nError accessing class data sharing archive."   \
2175                 " Mapped file inaccessible during execution, "      \
2176                 " possible disk/network problem.");
2177     }
2178   }
2179   st->cr();
2180 }
2181 
2182 
2183 static void print_signal_handler(outputStream* st, int sig,
2184                                  char* buf, size_t buflen);
2185 
2186 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2187   st->print_cr("Signal Handlers:");
2188   print_signal_handler(st, SIGSEGV, buf, buflen);
2189   print_signal_handler(st, SIGBUS , buf, buflen);
2190   print_signal_handler(st, SIGFPE , buf, buflen);
2191   print_signal_handler(st, SIGPIPE, buf, buflen);
2192   print_signal_handler(st, SIGXFSZ, buf, buflen);
2193   print_signal_handler(st, SIGILL , buf, buflen);
2194   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2195   print_signal_handler(st, SR_signum, buf, buflen);
2196   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2197   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2198   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2199   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2200 }
2201 
2202 static char saved_jvm_path[MAXPATHLEN] = {0};
2203 
2204 // Find the full path to the current module, libjvm.so or libjvm_g.so
2205 void os::jvm_path(char *buf, jint buflen) {
2206   // Error checking.
2207   if (buflen < MAXPATHLEN) {
2208     assert(false, "must use a large-enough buffer");
2209     buf[0] = '\0';
2210     return;
2211   }
2212   // Lazy resolve the path to current module.
2213   if (saved_jvm_path[0] != 0) {
2214     strcpy(buf, saved_jvm_path);
2215     return;
2216   }
2217 
2218   char dli_fname[MAXPATHLEN];
2219   bool ret = dll_address_to_library_name(
2220                 CAST_FROM_FN_PTR(address, os::jvm_path),
2221                 dli_fname, sizeof(dli_fname), NULL);
2222   assert(ret != 0, "cannot locate libjvm");
2223   char *rp = realpath(dli_fname, buf);
2224   if (rp == NULL)
2225     return;
2226 
2227   if (Arguments::created_by_gamma_launcher()) {
2228     // Support for the gamma launcher.  Typical value for buf is
2229     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2230     // the right place in the string, then assume we are installed in a JDK and
2231     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2232     // up the path so it looks like libjvm.so is installed there (append a
2233     // fake suffix hotspot/libjvm.so).
2234     const char *p = buf + strlen(buf) - 1;
2235     for (int count = 0; p > buf && count < 5; ++count) {
2236       for (--p; p > buf && *p != '/'; --p)
2237         /* empty */ ;
2238     }
2239 
2240     if (strncmp(p, "/jre/lib/", 9) != 0) {
2241       // Look for JAVA_HOME in the environment.
2242       char* java_home_var = ::getenv("JAVA_HOME");
2243       if (java_home_var != NULL && java_home_var[0] != 0) {
2244         char* jrelib_p;
2245         int len;
2246 
2247         // Check the current module name "libjvm.so" or "libjvm_g.so".
2248         p = strrchr(buf, '/');
2249         assert(strstr(p, "/libjvm") == p, "invalid library name");
2250         p = strstr(p, "_g") ? "_g" : "";
2251 
2252         rp = realpath(java_home_var, buf);
2253         if (rp == NULL)
2254           return;
2255 
2256         // determine if this is a legacy image or modules image
2257         // modules image doesn't have "jre" subdirectory
2258         len = strlen(buf);
2259         jrelib_p = buf + len;
2260         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2261         if (0 != access(buf, F_OK)) {
2262           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2263         }
2264 
2265         if (0 == access(buf, F_OK)) {
2266           // Use current module name "libjvm[_g].so" instead of
2267           // "libjvm"debug_only("_g")".so" since for fastdebug version
2268           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2269           len = strlen(buf);
2270           snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2271         } else {
2272           // Go back to path of .so
2273           rp = realpath(dli_fname, buf);
2274           if (rp == NULL)
2275             return;
2276         }
2277       }
2278     }
2279   }
2280 
2281   strcpy(saved_jvm_path, buf);
2282 }
2283 
2284 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2285   // no prefix required, not even "_"
2286 }
2287 
2288 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2289   // no suffix required
2290 }
2291 
2292 ////////////////////////////////////////////////////////////////////////////////
2293 // sun.misc.Signal support
2294 
2295 static volatile jint sigint_count = 0;
2296 
2297 static void
2298 UserHandler(int sig, void *siginfo, void *context) {
2299   // 4511530 - sem_post is serialized and handled by the manager thread. When
2300   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2301   // don't want to flood the manager thread with sem_post requests.
2302   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2303       return;
2304 
2305   // Ctrl-C is pressed during error reporting, likely because the error
2306   // handler fails to abort. Let VM die immediately.
2307   if (sig == SIGINT && is_error_reported()) {
2308      os::die();
2309   }
2310 
2311   os::signal_notify(sig);
2312 }
2313 
2314 void* os::user_handler() {
2315   return CAST_FROM_FN_PTR(void*, UserHandler);
2316 }
2317 
2318 extern "C" {
2319   typedef void (*sa_handler_t)(int);
2320   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2321 }
2322 
2323 void* os::signal(int signal_number, void* handler) {
2324   struct sigaction sigAct, oldSigAct;
2325 
2326   sigfillset(&(sigAct.sa_mask));
2327   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2328   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2329 
2330   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2331     // -1 means registration failed
2332     return (void *)-1;
2333   }
2334 
2335   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2336 }
2337 
2338 void os::signal_raise(int signal_number) {
2339   ::raise(signal_number);
2340 }
2341 
2342 /*
2343  * The following code is moved from os.cpp for making this
2344  * code platform specific, which it is by its very nature.
2345  */
2346 
2347 // Will be modified when max signal is changed to be dynamic
2348 int os::sigexitnum_pd() {
2349   return NSIG;
2350 }
2351 
2352 // a counter for each possible signal value
2353 static volatile jint pending_signals[NSIG+1] = { 0 };
2354 
2355 // Linux(POSIX) specific hand shaking semaphore.
2356 static sem_t sig_sem;
2357 
2358 void os::signal_init_pd() {
2359   // Initialize signal structures
2360   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2361 
2362   // Initialize signal semaphore
2363   ::sem_init(&sig_sem, 0, 0);
2364 }
2365 
2366 void os::signal_notify(int sig) {
2367   Atomic::inc(&pending_signals[sig]);
2368   ::sem_post(&sig_sem);
2369 }
2370 
2371 static int check_pending_signals(bool wait) {
2372   Atomic::store(0, &sigint_count);
2373   for (;;) {
2374     for (int i = 0; i < NSIG + 1; i++) {
2375       jint n = pending_signals[i];
2376       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2377         return i;
2378       }
2379     }
2380     if (!wait) {
2381       return -1;
2382     }
2383     JavaThread *thread = JavaThread::current();
2384     ThreadBlockInVM tbivm(thread);
2385 
2386     bool threadIsSuspended;
2387     do {
2388       thread->set_suspend_equivalent();
2389       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2390       ::sem_wait(&sig_sem);
2391 
2392       // were we externally suspended while we were waiting?
2393       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2394       if (threadIsSuspended) {
2395         //
2396         // The semaphore has been incremented, but while we were waiting
2397         // another thread suspended us. We don't want to continue running
2398         // while suspended because that would surprise the thread that
2399         // suspended us.
2400         //
2401         ::sem_post(&sig_sem);
2402 
2403         thread->java_suspend_self();
2404       }
2405     } while (threadIsSuspended);
2406   }
2407 }
2408 
2409 int os::signal_lookup() {
2410   return check_pending_signals(false);
2411 }
2412 
2413 int os::signal_wait() {
2414   return check_pending_signals(true);
2415 }
2416 
2417 ////////////////////////////////////////////////////////////////////////////////
2418 // Virtual Memory
2419 
2420 int os::vm_page_size() {
2421   // Seems redundant as all get out
2422   assert(os::Linux::page_size() != -1, "must call os::init");
2423   return os::Linux::page_size();
2424 }
2425 
2426 // Solaris allocates memory by pages.
2427 int os::vm_allocation_granularity() {
2428   assert(os::Linux::page_size() != -1, "must call os::init");
2429   return os::Linux::page_size();
2430 }
2431 
2432 // Rationale behind this function:
2433 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2434 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2435 //  samples for JITted code. Here we create private executable mapping over the code cache
2436 //  and then we can use standard (well, almost, as mapping can change) way to provide
2437 //  info for the reporting script by storing timestamp and location of symbol
2438 void linux_wrap_code(char* base, size_t size) {
2439   static volatile jint cnt = 0;
2440 
2441   if (!UseOprofile) {
2442     return;
2443   }
2444 
2445   char buf[PATH_MAX+1];
2446   int num = Atomic::add(1, &cnt);
2447 
2448   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2449            os::get_temp_directory(), os::current_process_id(), num);
2450   unlink(buf);
2451 
2452   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2453 
2454   if (fd != -1) {
2455     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2456     if (rv != (off_t)-1) {
2457       if (::write(fd, "", 1) == 1) {
2458         mmap(base, size,
2459              PROT_READ|PROT_WRITE|PROT_EXEC,
2460              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2461       }
2462     }
2463     ::close(fd);
2464     unlink(buf);
2465   }
2466 }
2467 
2468 // NOTE: Linux kernel does not really reserve the pages for us.
2469 //       All it does is to check if there are enough free pages
2470 //       left at the time of mmap(). This could be a potential
2471 //       problem.
2472 bool os::commit_memory(char* addr, size_t size, bool exec) {
2473   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2474   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2475                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2476   if (res != (uintptr_t) MAP_FAILED) {
2477     if (UseNUMAInterleaving) {
2478       numa_make_global(addr, size);
2479     }
2480     return true;
2481   }
2482   return false;
2483 }
2484 
2485 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2486 #ifndef MAP_HUGETLB
2487 #define MAP_HUGETLB 0x40000
2488 #endif
2489 
2490 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2491 #ifndef MADV_HUGEPAGE
2492 #define MADV_HUGEPAGE 14
2493 #endif
2494 
2495 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2496                        bool exec) {
2497   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2498     int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2499     uintptr_t res =
2500       (uintptr_t) ::mmap(addr, size, prot,
2501                          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2502                          -1, 0);
2503     if (res != (uintptr_t) MAP_FAILED) {
2504       if (UseNUMAInterleaving) {
2505         numa_make_global(addr, size);
2506       }
2507       return true;
2508     }
2509     // Fall through and try to use small pages
2510   }
2511 
2512   if (commit_memory(addr, size, exec)) {
2513     realign_memory(addr, size, alignment_hint);
2514     return true;
2515   }
2516   return false;
2517 }
2518 
2519 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2520   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2521     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2522     // be supported or the memory may already be backed by huge pages.
2523     ::madvise(addr, bytes, MADV_HUGEPAGE);
2524   }
2525 }
2526 
2527 void os::free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2528   // This method works by doing an mmap over an existing mmaping and effectively discarding
2529   // the existing pages. However it won't work for SHM-based large pages that cannot be
2530   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2531   // small pages on top of the SHM segment. This method always works for small pages, so we
2532   // allow that in any case.
2533   if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2534     commit_memory(addr, bytes, alignment_hint, false);
2535   }
2536 }
2537 
2538 void os::numa_make_global(char *addr, size_t bytes) {
2539   Linux::numa_interleave_memory(addr, bytes);
2540 }
2541 
2542 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2543   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2544 }
2545 
2546 bool os::numa_topology_changed()   { return false; }
2547 
2548 size_t os::numa_get_groups_num() {
2549   int max_node = Linux::numa_max_node();
2550   return max_node > 0 ? max_node + 1 : 1;
2551 }
2552 
2553 int os::numa_get_group_id() {
2554   int cpu_id = Linux::sched_getcpu();
2555   if (cpu_id != -1) {
2556     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2557     if (lgrp_id != -1) {
2558       return lgrp_id;
2559     }
2560   }
2561   return 0;
2562 }
2563 
2564 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2565   for (size_t i = 0; i < size; i++) {
2566     ids[i] = i;
2567   }
2568   return size;
2569 }
2570 
2571 bool os::get_page_info(char *start, page_info* info) {
2572   return false;
2573 }
2574 
2575 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2576   return end;
2577 }
2578 
2579 
2580 int os::Linux::sched_getcpu_syscall(void) {
2581   unsigned int cpu;
2582   int retval = -1;
2583 
2584 #if defined(IA32)
2585 # ifndef SYS_getcpu
2586 # define SYS_getcpu 318
2587 # endif
2588   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2589 #elif defined(AMD64)
2590 // Unfortunately we have to bring all these macros here from vsyscall.h
2591 // to be able to compile on old linuxes.
2592 # define __NR_vgetcpu 2
2593 # define VSYSCALL_START (-10UL << 20)
2594 # define VSYSCALL_SIZE 1024
2595 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2596   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2597   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2598   retval = vgetcpu(&cpu, NULL, NULL);
2599 #endif
2600 
2601   return (retval == -1) ? retval : cpu;
2602 }
2603 
2604 // Something to do with the numa-aware allocator needs these symbols
2605 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2606 extern "C" JNIEXPORT void numa_error(char *where) { }
2607 extern "C" JNIEXPORT int fork1() { return fork(); }
2608 
2609 
2610 // If we are running with libnuma version > 2, then we should
2611 // be trying to use symbols with versions 1.1
2612 // If we are running with earlier version, which did not have symbol versions,
2613 // we should use the base version.
2614 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2615   void *f = dlvsym(handle, name, "libnuma_1.1");
2616   if (f == NULL) {
2617     f = dlsym(handle, name);
2618   }
2619   return f;
2620 }
2621 
2622 bool os::Linux::libnuma_init() {
2623   // sched_getcpu() should be in libc.
2624   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2625                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2626 
2627   // If it's not, try a direct syscall.
2628   if (sched_getcpu() == -1)
2629     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2630 
2631   if (sched_getcpu() != -1) { // Does it work?
2632     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2633     if (handle != NULL) {
2634       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2635                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2636       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2637                                        libnuma_dlsym(handle, "numa_max_node")));
2638       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2639                                         libnuma_dlsym(handle, "numa_available")));
2640       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2641                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2642       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2643                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2644 
2645 
2646       if (numa_available() != -1) {
2647         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2648         // Create a cpu -> node mapping
2649         _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2650         rebuild_cpu_to_node_map();
2651         return true;
2652       }
2653     }
2654   }
2655   return false;
2656 }
2657 
2658 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2659 // The table is later used in get_node_by_cpu().
2660 void os::Linux::rebuild_cpu_to_node_map() {
2661   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2662                               // in libnuma (possible values are starting from 16,
2663                               // and continuing up with every other power of 2, but less
2664                               // than the maximum number of CPUs supported by kernel), and
2665                               // is a subject to change (in libnuma version 2 the requirements
2666                               // are more reasonable) we'll just hardcode the number they use
2667                               // in the library.
2668   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2669 
2670   size_t cpu_num = os::active_processor_count();
2671   size_t cpu_map_size = NCPUS / BitsPerCLong;
2672   size_t cpu_map_valid_size =
2673     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2674 
2675   cpu_to_node()->clear();
2676   cpu_to_node()->at_grow(cpu_num - 1);
2677   size_t node_num = numa_get_groups_num();
2678 
2679   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2680   for (size_t i = 0; i < node_num; i++) {
2681     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2682       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2683         if (cpu_map[j] != 0) {
2684           for (size_t k = 0; k < BitsPerCLong; k++) {
2685             if (cpu_map[j] & (1UL << k)) {
2686               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2687             }
2688           }
2689         }
2690       }
2691     }
2692   }
2693   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2694 }
2695 
2696 int os::Linux::get_node_by_cpu(int cpu_id) {
2697   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2698     return cpu_to_node()->at(cpu_id);
2699   }
2700   return -1;
2701 }
2702 
2703 GrowableArray<int>* os::Linux::_cpu_to_node;
2704 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2705 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2706 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2707 os::Linux::numa_available_func_t os::Linux::_numa_available;
2708 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2709 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2710 unsigned long* os::Linux::_numa_all_nodes;
2711 
2712 bool os::uncommit_memory(char* addr, size_t size) {
2713   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2714                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2715   return res  != (uintptr_t) MAP_FAILED;
2716 }
2717 
2718 // Linux uses a growable mapping for the stack, and if the mapping for
2719 // the stack guard pages is not removed when we detach a thread the
2720 // stack cannot grow beyond the pages where the stack guard was
2721 // mapped.  If at some point later in the process the stack expands to
2722 // that point, the Linux kernel cannot expand the stack any further
2723 // because the guard pages are in the way, and a segfault occurs.
2724 //
2725 // However, it's essential not to split the stack region by unmapping
2726 // a region (leaving a hole) that's already part of the stack mapping,
2727 // so if the stack mapping has already grown beyond the guard pages at
2728 // the time we create them, we have to truncate the stack mapping.
2729 // So, we need to know the extent of the stack mapping when
2730 // create_stack_guard_pages() is called.
2731 
2732 // Find the bounds of the stack mapping.  Return true for success.
2733 //
2734 // We only need this for stacks that are growable: at the time of
2735 // writing thread stacks don't use growable mappings (i.e. those
2736 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2737 // only applies to the main thread.
2738 
2739 static
2740 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2741 
2742   char buf[128];
2743   int fd, sz;
2744 
2745   if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2746     return false;
2747   }
2748 
2749   const char kw[] = "[stack]";
2750   const int kwlen = sizeof(kw)-1;
2751 
2752   // Address part of /proc/self/maps couldn't be more than 128 bytes
2753   while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2754      if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2755         // Extract addresses
2756         if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2757            uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2758            if (sp >= *bottom && sp <= *top) {
2759               ::close(fd);
2760               return true;
2761            }
2762         }
2763      }
2764   }
2765 
2766  ::close(fd);
2767   return false;
2768 }
2769 
2770 
2771 // If the (growable) stack mapping already extends beyond the point
2772 // where we're going to put our guard pages, truncate the mapping at
2773 // that point by munmap()ping it.  This ensures that when we later
2774 // munmap() the guard pages we don't leave a hole in the stack
2775 // mapping. This only affects the main/initial thread, but guard
2776 // against future OS changes
2777 bool os::create_stack_guard_pages(char* addr, size_t size) {
2778   uintptr_t stack_extent, stack_base;
2779   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2780   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2781       assert(os::Linux::is_initial_thread(),
2782            "growable stack in non-initial thread");
2783     if (stack_extent < (uintptr_t)addr)
2784       ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2785   }
2786 
2787   return os::commit_memory(addr, size);
2788 }
2789 
2790 // If this is a growable mapping, remove the guard pages entirely by
2791 // munmap()ping them.  If not, just call uncommit_memory(). This only
2792 // affects the main/initial thread, but guard against future OS changes
2793 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2794   uintptr_t stack_extent, stack_base;
2795   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2796   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2797       assert(os::Linux::is_initial_thread(),
2798            "growable stack in non-initial thread");
2799 
2800     return ::munmap(addr, size) == 0;
2801   }
2802 
2803   return os::uncommit_memory(addr, size);
2804 }
2805 
2806 static address _highest_vm_reserved_address = NULL;
2807 
2808 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2809 // at 'requested_addr'. If there are existing memory mappings at the same
2810 // location, however, they will be overwritten. If 'fixed' is false,
2811 // 'requested_addr' is only treated as a hint, the return value may or
2812 // may not start from the requested address. Unlike Linux mmap(), this
2813 // function returns NULL to indicate failure.
2814 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2815   char * addr;
2816   int flags;
2817 
2818   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2819   if (fixed) {
2820     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2821     flags |= MAP_FIXED;
2822   }
2823 
2824   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2825   // to PROT_EXEC if executable when we commit the page.
2826   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2827                        flags, -1, 0);
2828 
2829   if (addr != MAP_FAILED) {
2830     // anon_mmap() should only get called during VM initialization,
2831     // don't need lock (actually we can skip locking even it can be called
2832     // from multiple threads, because _highest_vm_reserved_address is just a
2833     // hint about the upper limit of non-stack memory regions.)
2834     if ((address)addr + bytes > _highest_vm_reserved_address) {
2835       _highest_vm_reserved_address = (address)addr + bytes;
2836     }
2837   }
2838 
2839   return addr == MAP_FAILED ? NULL : addr;
2840 }
2841 
2842 // Don't update _highest_vm_reserved_address, because there might be memory
2843 // regions above addr + size. If so, releasing a memory region only creates
2844 // a hole in the address space, it doesn't help prevent heap-stack collision.
2845 //
2846 static int anon_munmap(char * addr, size_t size) {
2847   return ::munmap(addr, size) == 0;
2848 }
2849 
2850 char* os::reserve_memory(size_t bytes, char* requested_addr,
2851                          size_t alignment_hint) {
2852   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2853 }
2854 
2855 bool os::release_memory(char* addr, size_t size) {
2856   return anon_munmap(addr, size);
2857 }
2858 
2859 static address highest_vm_reserved_address() {
2860   return _highest_vm_reserved_address;
2861 }
2862 
2863 static bool linux_mprotect(char* addr, size_t size, int prot) {
2864   // Linux wants the mprotect address argument to be page aligned.
2865   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2866 
2867   // According to SUSv3, mprotect() should only be used with mappings
2868   // established by mmap(), and mmap() always maps whole pages. Unaligned
2869   // 'addr' likely indicates problem in the VM (e.g. trying to change
2870   // protection of malloc'ed or statically allocated memory). Check the
2871   // caller if you hit this assert.
2872   assert(addr == bottom, "sanity check");
2873 
2874   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2875   return ::mprotect(bottom, size, prot) == 0;
2876 }
2877 
2878 // Set protections specified
2879 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2880                         bool is_committed) {
2881   unsigned int p = 0;
2882   switch (prot) {
2883   case MEM_PROT_NONE: p = PROT_NONE; break;
2884   case MEM_PROT_READ: p = PROT_READ; break;
2885   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2886   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2887   default:
2888     ShouldNotReachHere();
2889   }
2890   // is_committed is unused.
2891   return linux_mprotect(addr, bytes, p);
2892 }
2893 
2894 bool os::guard_memory(char* addr, size_t size) {
2895   return linux_mprotect(addr, size, PROT_NONE);
2896 }
2897 
2898 bool os::unguard_memory(char* addr, size_t size) {
2899   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2900 }
2901 
2902 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2903   bool result = false;
2904   void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2905                   MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2906                   -1, 0);
2907 
2908   if (p != (void *) -1) {
2909     // We don't know if this really is a huge page or not.
2910     FILE *fp = fopen("/proc/self/maps", "r");
2911     if (fp) {
2912       while (!feof(fp)) {
2913         char chars[257];
2914         long x = 0;
2915         if (fgets(chars, sizeof(chars), fp)) {
2916           if (sscanf(chars, "%lx-%*x", &x) == 1
2917               && x == (long)p) {
2918             if (strstr (chars, "hugepage")) {
2919               result = true;
2920               break;
2921             }
2922           }
2923         }
2924       }
2925       fclose(fp);
2926     }
2927     munmap (p, page_size);
2928     if (result)
2929       return true;
2930   }
2931 
2932   if (warn) {
2933     warning("HugeTLBFS is not supported by the operating system.");
2934   }
2935 
2936   return result;
2937 }
2938 
2939 /*
2940 * Set the coredump_filter bits to include largepages in core dump (bit 6)
2941 *
2942 * From the coredump_filter documentation:
2943 *
2944 * - (bit 0) anonymous private memory
2945 * - (bit 1) anonymous shared memory
2946 * - (bit 2) file-backed private memory
2947 * - (bit 3) file-backed shared memory
2948 * - (bit 4) ELF header pages in file-backed private memory areas (it is
2949 *           effective only if the bit 2 is cleared)
2950 * - (bit 5) hugetlb private memory
2951 * - (bit 6) hugetlb shared memory
2952 */
2953 static void set_coredump_filter(void) {
2954   FILE *f;
2955   long cdm;
2956 
2957   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
2958     return;
2959   }
2960 
2961   if (fscanf(f, "%lx", &cdm) != 1) {
2962     fclose(f);
2963     return;
2964   }
2965 
2966   rewind(f);
2967 
2968   if ((cdm & LARGEPAGES_BIT) == 0) {
2969     cdm |= LARGEPAGES_BIT;
2970     fprintf(f, "%#lx", cdm);
2971   }
2972 
2973   fclose(f);
2974 }
2975 
2976 // Large page support
2977 
2978 static size_t _large_page_size = 0;
2979 
2980 void os::large_page_init() {
2981   if (!UseLargePages) {
2982     UseHugeTLBFS = false;
2983     UseSHM = false;
2984     return;
2985   }
2986 
2987   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
2988     // If UseLargePages is specified on the command line try both methods,
2989     // if it's default, then try only HugeTLBFS.
2990     if (FLAG_IS_DEFAULT(UseLargePages)) {
2991       UseHugeTLBFS = true;
2992     } else {
2993       UseHugeTLBFS = UseSHM = true;
2994     }
2995   }
2996 
2997   if (LargePageSizeInBytes) {
2998     _large_page_size = LargePageSizeInBytes;
2999   } else {
3000     // large_page_size on Linux is used to round up heap size. x86 uses either
3001     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3002     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3003     // page as large as 256M.
3004     //
3005     // Here we try to figure out page size by parsing /proc/meminfo and looking
3006     // for a line with the following format:
3007     //    Hugepagesize:     2048 kB
3008     //
3009     // If we can't determine the value (e.g. /proc is not mounted, or the text
3010     // format has been changed), we'll use the largest page size supported by
3011     // the processor.
3012 
3013 #ifndef ZERO
3014     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3015                        ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3016 #endif // ZERO
3017 
3018     FILE *fp = fopen("/proc/meminfo", "r");
3019     if (fp) {
3020       while (!feof(fp)) {
3021         int x = 0;
3022         char buf[16];
3023         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3024           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3025             _large_page_size = x * K;
3026             break;
3027           }
3028         } else {
3029           // skip to next line
3030           for (;;) {
3031             int ch = fgetc(fp);
3032             if (ch == EOF || ch == (int)'\n') break;
3033           }
3034         }
3035       }
3036       fclose(fp);
3037     }
3038   }
3039 
3040   // print a warning if any large page related flag is specified on command line
3041   bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3042 
3043   const size_t default_page_size = (size_t)Linux::page_size();
3044   if (_large_page_size > default_page_size) {
3045     _page_sizes[0] = _large_page_size;
3046     _page_sizes[1] = default_page_size;
3047     _page_sizes[2] = 0;
3048   }
3049   UseHugeTLBFS = UseHugeTLBFS &&
3050                  Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3051 
3052   if (UseHugeTLBFS)
3053     UseSHM = false;
3054 
3055   UseLargePages = UseHugeTLBFS || UseSHM;
3056 
3057   set_coredump_filter();
3058 }
3059 
3060 #ifndef SHM_HUGETLB
3061 #define SHM_HUGETLB 04000
3062 #endif
3063 
3064 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3065   // "exec" is passed in but not used.  Creating the shared image for
3066   // the code cache doesn't have an SHM_X executable permission to check.
3067   assert(UseLargePages && UseSHM, "only for SHM large pages");
3068 
3069   key_t key = IPC_PRIVATE;
3070   char *addr;
3071 
3072   bool warn_on_failure = UseLargePages &&
3073                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3074                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3075                         );
3076   char msg[128];
3077 
3078   // Create a large shared memory region to attach to based on size.
3079   // Currently, size is the total size of the heap
3080   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3081   if (shmid == -1) {
3082      // Possible reasons for shmget failure:
3083      // 1. shmmax is too small for Java heap.
3084      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3085      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3086      // 2. not enough large page memory.
3087      //    > check available large pages: cat /proc/meminfo
3088      //    > increase amount of large pages:
3089      //          echo new_value > /proc/sys/vm/nr_hugepages
3090      //      Note 1: different Linux may use different name for this property,
3091      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3092      //      Note 2: it's possible there's enough physical memory available but
3093      //            they are so fragmented after a long run that they can't
3094      //            coalesce into large pages. Try to reserve large pages when
3095      //            the system is still "fresh".
3096      if (warn_on_failure) {
3097        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3098        warning(msg);
3099      }
3100      return NULL;
3101   }
3102 
3103   // attach to the region
3104   addr = (char*)shmat(shmid, req_addr, 0);
3105   int err = errno;
3106 
3107   // Remove shmid. If shmat() is successful, the actual shared memory segment
3108   // will be deleted when it's detached by shmdt() or when the process
3109   // terminates. If shmat() is not successful this will remove the shared
3110   // segment immediately.
3111   shmctl(shmid, IPC_RMID, NULL);
3112 
3113   if ((intptr_t)addr == -1) {
3114      if (warn_on_failure) {
3115        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3116        warning(msg);
3117      }
3118      return NULL;
3119   }
3120 
3121   if ((addr != NULL) && UseNUMAInterleaving) {
3122     numa_make_global(addr, bytes);
3123   }
3124 
3125   return addr;
3126 }
3127 
3128 bool os::release_memory_special(char* base, size_t bytes) {
3129   // detaching the SHM segment will also delete it, see reserve_memory_special()
3130   int rslt = shmdt(base);
3131   return rslt == 0;
3132 }
3133 
3134 size_t os::large_page_size() {
3135   return _large_page_size;
3136 }
3137 
3138 // HugeTLBFS allows application to commit large page memory on demand;
3139 // with SysV SHM the entire memory region must be allocated as shared
3140 // memory.
3141 bool os::can_commit_large_page_memory() {
3142   return UseHugeTLBFS;
3143 }
3144 
3145 bool os::can_execute_large_page_memory() {
3146   return UseHugeTLBFS;
3147 }
3148 
3149 // Reserve memory at an arbitrary address, only if that area is
3150 // available (and not reserved for something else).
3151 
3152 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3153   const int max_tries = 10;
3154   char* base[max_tries];
3155   size_t size[max_tries];
3156   const size_t gap = 0x000000;
3157 
3158   // Assert only that the size is a multiple of the page size, since
3159   // that's all that mmap requires, and since that's all we really know
3160   // about at this low abstraction level.  If we need higher alignment,
3161   // we can either pass an alignment to this method or verify alignment
3162   // in one of the methods further up the call chain.  See bug 5044738.
3163   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3164 
3165   // Repeatedly allocate blocks until the block is allocated at the
3166   // right spot. Give up after max_tries. Note that reserve_memory() will
3167   // automatically update _highest_vm_reserved_address if the call is
3168   // successful. The variable tracks the highest memory address every reserved
3169   // by JVM. It is used to detect heap-stack collision if running with
3170   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3171   // space than needed, it could confuse the collision detecting code. To
3172   // solve the problem, save current _highest_vm_reserved_address and
3173   // calculate the correct value before return.
3174   address old_highest = _highest_vm_reserved_address;
3175 
3176   // Linux mmap allows caller to pass an address as hint; give it a try first,
3177   // if kernel honors the hint then we can return immediately.
3178   char * addr = anon_mmap(requested_addr, bytes, false);
3179   if (addr == requested_addr) {
3180      return requested_addr;
3181   }
3182 
3183   if (addr != NULL) {
3184      // mmap() is successful but it fails to reserve at the requested address
3185      anon_munmap(addr, bytes);
3186   }
3187 
3188   int i;
3189   for (i = 0; i < max_tries; ++i) {
3190     base[i] = reserve_memory(bytes);
3191 
3192     if (base[i] != NULL) {
3193       // Is this the block we wanted?
3194       if (base[i] == requested_addr) {
3195         size[i] = bytes;
3196         break;
3197       }
3198 
3199       // Does this overlap the block we wanted? Give back the overlapped
3200       // parts and try again.
3201 
3202       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3203       if (top_overlap >= 0 && top_overlap < bytes) {
3204         unmap_memory(base[i], top_overlap);
3205         base[i] += top_overlap;
3206         size[i] = bytes - top_overlap;
3207       } else {
3208         size_t bottom_overlap = base[i] + bytes - requested_addr;
3209         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3210           unmap_memory(requested_addr, bottom_overlap);
3211           size[i] = bytes - bottom_overlap;
3212         } else {
3213           size[i] = bytes;
3214         }
3215       }
3216     }
3217   }
3218 
3219   // Give back the unused reserved pieces.
3220 
3221   for (int j = 0; j < i; ++j) {
3222     if (base[j] != NULL) {
3223       unmap_memory(base[j], size[j]);
3224     }
3225   }
3226 
3227   if (i < max_tries) {
3228     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3229     return requested_addr;
3230   } else {
3231     _highest_vm_reserved_address = old_highest;
3232     return NULL;
3233   }
3234 }
3235 
3236 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3237   return ::read(fd, buf, nBytes);
3238 }
3239 
3240 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3241 // Solaris uses poll(), linux uses park().
3242 // Poll() is likely a better choice, assuming that Thread.interrupt()
3243 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3244 // SIGSEGV, see 4355769.
3245 
3246 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3247   assert(thread == Thread::current(),  "thread consistency check");
3248 
3249   ParkEvent * const slp = thread->_SleepEvent ;
3250   slp->reset() ;
3251   OrderAccess::fence() ;
3252 
3253   if (interruptible) {
3254     jlong prevtime = javaTimeNanos();
3255 
3256     for (;;) {
3257       if (os::is_interrupted(thread, true)) {
3258         return OS_INTRPT;
3259       }
3260 
3261       jlong newtime = javaTimeNanos();
3262 
3263       if (newtime - prevtime < 0) {
3264         // time moving backwards, should only happen if no monotonic clock
3265         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3266         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3267       } else {
3268         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3269       }
3270 
3271       if(millis <= 0) {
3272         return OS_OK;
3273       }
3274 
3275       prevtime = newtime;
3276 
3277       {
3278         assert(thread->is_Java_thread(), "sanity check");
3279         JavaThread *jt = (JavaThread *) thread;
3280         ThreadBlockInVM tbivm(jt);
3281         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3282 
3283         jt->set_suspend_equivalent();
3284         // cleared by handle_special_suspend_equivalent_condition() or
3285         // java_suspend_self() via check_and_wait_while_suspended()
3286 
3287         slp->park(millis);
3288 
3289         // were we externally suspended while we were waiting?
3290         jt->check_and_wait_while_suspended();
3291       }
3292     }
3293   } else {
3294     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3295     jlong prevtime = javaTimeNanos();
3296 
3297     for (;;) {
3298       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3299       // the 1st iteration ...
3300       jlong newtime = javaTimeNanos();
3301 
3302       if (newtime - prevtime < 0) {
3303         // time moving backwards, should only happen if no monotonic clock
3304         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3305         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3306       } else {
3307         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3308       }
3309 
3310       if(millis <= 0) break ;
3311 
3312       prevtime = newtime;
3313       slp->park(millis);
3314     }
3315     return OS_OK ;
3316   }
3317 }
3318 
3319 int os::naked_sleep() {
3320   // %% make the sleep time an integer flag. for now use 1 millisec.
3321   return os::sleep(Thread::current(), 1, false);
3322 }
3323 
3324 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3325 void os::infinite_sleep() {
3326   while (true) {    // sleep forever ...
3327     ::sleep(100);   // ... 100 seconds at a time
3328   }
3329 }
3330 
3331 // Used to convert frequent JVM_Yield() to nops
3332 bool os::dont_yield() {
3333   return DontYieldALot;
3334 }
3335 
3336 void os::yield() {
3337   sched_yield();
3338 }
3339 
3340 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3341 
3342 void os::yield_all(int attempts) {
3343   // Yields to all threads, including threads with lower priorities
3344   // Threads on Linux are all with same priority. The Solaris style
3345   // os::yield_all() with nanosleep(1ms) is not necessary.
3346   sched_yield();
3347 }
3348 
3349 // Called from the tight loops to possibly influence time-sharing heuristics
3350 void os::loop_breaker(int attempts) {
3351   os::yield_all(attempts);
3352 }
3353 
3354 ////////////////////////////////////////////////////////////////////////////////
3355 // thread priority support
3356 
3357 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3358 // only supports dynamic priority, static priority must be zero. For real-time
3359 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3360 // However, for large multi-threaded applications, SCHED_RR is not only slower
3361 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3362 // of 5 runs - Sep 2005).
3363 //
3364 // The following code actually changes the niceness of kernel-thread/LWP. It
3365 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3366 // not the entire user process, and user level threads are 1:1 mapped to kernel
3367 // threads. It has always been the case, but could change in the future. For
3368 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3369 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3370 
3371 int os::java_to_os_priority[CriticalPriority + 1] = {
3372   19,              // 0 Entry should never be used
3373 
3374    4,              // 1 MinPriority
3375    3,              // 2
3376    2,              // 3
3377 
3378    1,              // 4
3379    0,              // 5 NormPriority
3380   -1,              // 6
3381 
3382   -2,              // 7
3383   -3,              // 8
3384   -4,              // 9 NearMaxPriority
3385 
3386   -5,              // 10 MaxPriority
3387 
3388   -5               // 11 CriticalPriority
3389 };
3390 
3391 static int prio_init() {
3392   if (ThreadPriorityPolicy == 1) {
3393     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3394     // if effective uid is not root. Perhaps, a more elegant way of doing
3395     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3396     if (geteuid() != 0) {
3397       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3398         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3399       }
3400       ThreadPriorityPolicy = 0;
3401     }
3402   }
3403   if (UseCriticalJavaThreadPriority) {
3404     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3405   }
3406   return 0;
3407 }
3408 
3409 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3410   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3411 
3412   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3413   return (ret == 0) ? OS_OK : OS_ERR;
3414 }
3415 
3416 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3417   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3418     *priority_ptr = java_to_os_priority[NormPriority];
3419     return OS_OK;
3420   }
3421 
3422   errno = 0;
3423   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3424   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3425 }
3426 
3427 // Hint to the underlying OS that a task switch would not be good.
3428 // Void return because it's a hint and can fail.
3429 void os::hint_no_preempt() {}
3430 
3431 ////////////////////////////////////////////////////////////////////////////////
3432 // suspend/resume support
3433 
3434 //  the low-level signal-based suspend/resume support is a remnant from the
3435 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3436 //  within hotspot. Now there is a single use-case for this:
3437 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3438 //      that runs in the watcher thread.
3439 //  The remaining code is greatly simplified from the more general suspension
3440 //  code that used to be used.
3441 //
3442 //  The protocol is quite simple:
3443 //  - suspend:
3444 //      - sends a signal to the target thread
3445 //      - polls the suspend state of the osthread using a yield loop
3446 //      - target thread signal handler (SR_handler) sets suspend state
3447 //        and blocks in sigsuspend until continued
3448 //  - resume:
3449 //      - sets target osthread state to continue
3450 //      - sends signal to end the sigsuspend loop in the SR_handler
3451 //
3452 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3453 //
3454 
3455 static void resume_clear_context(OSThread *osthread) {
3456   osthread->set_ucontext(NULL);
3457   osthread->set_siginfo(NULL);
3458 
3459   // notify the suspend action is completed, we have now resumed
3460   osthread->sr.clear_suspended();
3461 }
3462 
3463 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3464   osthread->set_ucontext(context);
3465   osthread->set_siginfo(siginfo);
3466 }
3467 
3468 //
3469 // Handler function invoked when a thread's execution is suspended or
3470 // resumed. We have to be careful that only async-safe functions are
3471 // called here (Note: most pthread functions are not async safe and
3472 // should be avoided.)
3473 //
3474 // Note: sigwait() is a more natural fit than sigsuspend() from an
3475 // interface point of view, but sigwait() prevents the signal hander
3476 // from being run. libpthread would get very confused by not having
3477 // its signal handlers run and prevents sigwait()'s use with the
3478 // mutex granting granting signal.
3479 //
3480 // Currently only ever called on the VMThread
3481 //
3482 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3483   // Save and restore errno to avoid confusing native code with EINTR
3484   // after sigsuspend.
3485   int old_errno = errno;
3486 
3487   Thread* thread = Thread::current();
3488   OSThread* osthread = thread->osthread();
3489   assert(thread->is_VM_thread(), "Must be VMThread");
3490   // read current suspend action
3491   int action = osthread->sr.suspend_action();
3492   if (action == SR_SUSPEND) {
3493     suspend_save_context(osthread, siginfo, context);
3494 
3495     // Notify the suspend action is about to be completed. do_suspend()
3496     // waits until SR_SUSPENDED is set and then returns. We will wait
3497     // here for a resume signal and that completes the suspend-other
3498     // action. do_suspend/do_resume is always called as a pair from
3499     // the same thread - so there are no races
3500 
3501     // notify the caller
3502     osthread->sr.set_suspended();
3503 
3504     sigset_t suspend_set;  // signals for sigsuspend()
3505 
3506     // get current set of blocked signals and unblock resume signal
3507     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3508     sigdelset(&suspend_set, SR_signum);
3509 
3510     // wait here until we are resumed
3511     do {
3512       sigsuspend(&suspend_set);
3513       // ignore all returns until we get a resume signal
3514     } while (osthread->sr.suspend_action() != SR_CONTINUE);
3515 
3516     resume_clear_context(osthread);
3517 
3518   } else {
3519     assert(action == SR_CONTINUE, "unexpected sr action");
3520     // nothing special to do - just leave the handler
3521   }
3522 
3523   errno = old_errno;
3524 }
3525 
3526 
3527 static int SR_initialize() {
3528   struct sigaction act;
3529   char *s;
3530   /* Get signal number to use for suspend/resume */
3531   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3532     int sig = ::strtol(s, 0, 10);
3533     if (sig > 0 || sig < _NSIG) {
3534         SR_signum = sig;
3535     }
3536   }
3537 
3538   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3539         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3540 
3541   sigemptyset(&SR_sigset);
3542   sigaddset(&SR_sigset, SR_signum);
3543 
3544   /* Set up signal handler for suspend/resume */
3545   act.sa_flags = SA_RESTART|SA_SIGINFO;
3546   act.sa_handler = (void (*)(int)) SR_handler;
3547 
3548   // SR_signum is blocked by default.
3549   // 4528190 - We also need to block pthread restart signal (32 on all
3550   // supported Linux platforms). Note that LinuxThreads need to block
3551   // this signal for all threads to work properly. So we don't have
3552   // to use hard-coded signal number when setting up the mask.
3553   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3554 
3555   if (sigaction(SR_signum, &act, 0) == -1) {
3556     return -1;
3557   }
3558 
3559   // Save signal flag
3560   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3561   return 0;
3562 }
3563 
3564 static int SR_finalize() {
3565   return 0;
3566 }
3567 
3568 
3569 // returns true on success and false on error - really an error is fatal
3570 // but this seems the normal response to library errors
3571 static bool do_suspend(OSThread* osthread) {
3572   // mark as suspended and send signal
3573   osthread->sr.set_suspend_action(SR_SUSPEND);
3574   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3575   assert_status(status == 0, status, "pthread_kill");
3576 
3577   // check status and wait until notified of suspension
3578   if (status == 0) {
3579     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3580       os::yield_all(i);
3581     }
3582     osthread->sr.set_suspend_action(SR_NONE);
3583     return true;
3584   }
3585   else {
3586     osthread->sr.set_suspend_action(SR_NONE);
3587     return false;
3588   }
3589 }
3590 
3591 static void do_resume(OSThread* osthread) {
3592   assert(osthread->sr.is_suspended(), "thread should be suspended");
3593   osthread->sr.set_suspend_action(SR_CONTINUE);
3594 
3595   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3596   assert_status(status == 0, status, "pthread_kill");
3597   // check status and wait unit notified of resumption
3598   if (status == 0) {
3599     for (int i = 0; osthread->sr.is_suspended(); i++) {
3600       os::yield_all(i);
3601     }
3602   }
3603   osthread->sr.set_suspend_action(SR_NONE);
3604 }
3605 
3606 ////////////////////////////////////////////////////////////////////////////////
3607 // interrupt support
3608 
3609 void os::interrupt(Thread* thread) {
3610   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3611     "possibility of dangling Thread pointer");
3612 
3613   OSThread* osthread = thread->osthread();
3614 
3615   if (!osthread->interrupted()) {
3616     osthread->set_interrupted(true);
3617     // More than one thread can get here with the same value of osthread,
3618     // resulting in multiple notifications.  We do, however, want the store
3619     // to interrupted() to be visible to other threads before we execute unpark().
3620     OrderAccess::fence();
3621     ParkEvent * const slp = thread->_SleepEvent ;
3622     if (slp != NULL) slp->unpark() ;
3623   }
3624 
3625   // For JSR166. Unpark even if interrupt status already was set
3626   if (thread->is_Java_thread())
3627     ((JavaThread*)thread)->parker()->unpark();
3628 
3629   ParkEvent * ev = thread->_ParkEvent ;
3630   if (ev != NULL) ev->unpark() ;
3631 
3632 }
3633 
3634 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3635   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3636     "possibility of dangling Thread pointer");
3637 
3638   OSThread* osthread = thread->osthread();
3639 
3640   bool interrupted = osthread->interrupted();
3641 
3642   if (interrupted && clear_interrupted) {
3643     osthread->set_interrupted(false);
3644     // consider thread->_SleepEvent->reset() ... optional optimization
3645   }
3646 
3647   return interrupted;
3648 }
3649 
3650 ///////////////////////////////////////////////////////////////////////////////////
3651 // signal handling (except suspend/resume)
3652 
3653 // This routine may be used by user applications as a "hook" to catch signals.
3654 // The user-defined signal handler must pass unrecognized signals to this
3655 // routine, and if it returns true (non-zero), then the signal handler must
3656 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3657 // routine will never retun false (zero), but instead will execute a VM panic
3658 // routine kill the process.
3659 //
3660 // If this routine returns false, it is OK to call it again.  This allows
3661 // the user-defined signal handler to perform checks either before or after
3662 // the VM performs its own checks.  Naturally, the user code would be making
3663 // a serious error if it tried to handle an exception (such as a null check
3664 // or breakpoint) that the VM was generating for its own correct operation.
3665 //
3666 // This routine may recognize any of the following kinds of signals:
3667 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3668 // It should be consulted by handlers for any of those signals.
3669 //
3670 // The caller of this routine must pass in the three arguments supplied
3671 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3672 // field of the structure passed to sigaction().  This routine assumes that
3673 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3674 //
3675 // Note that the VM will print warnings if it detects conflicting signal
3676 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3677 //
3678 extern "C" JNIEXPORT int
3679 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3680                         void* ucontext, int abort_if_unrecognized);
3681 
3682 void signalHandler(int sig, siginfo_t* info, void* uc) {
3683   assert(info != NULL && uc != NULL, "it must be old kernel");
3684   JVM_handle_linux_signal(sig, info, uc, true);
3685 }
3686 
3687 
3688 // This boolean allows users to forward their own non-matching signals
3689 // to JVM_handle_linux_signal, harmlessly.
3690 bool os::Linux::signal_handlers_are_installed = false;
3691 
3692 // For signal-chaining
3693 struct sigaction os::Linux::sigact[MAXSIGNUM];
3694 unsigned int os::Linux::sigs = 0;
3695 bool os::Linux::libjsig_is_loaded = false;
3696 typedef struct sigaction *(*get_signal_t)(int);
3697 get_signal_t os::Linux::get_signal_action = NULL;
3698 
3699 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3700   struct sigaction *actp = NULL;
3701 
3702   if (libjsig_is_loaded) {
3703     // Retrieve the old signal handler from libjsig
3704     actp = (*get_signal_action)(sig);
3705   }
3706   if (actp == NULL) {
3707     // Retrieve the preinstalled signal handler from jvm
3708     actp = get_preinstalled_handler(sig);
3709   }
3710 
3711   return actp;
3712 }
3713 
3714 static bool call_chained_handler(struct sigaction *actp, int sig,
3715                                  siginfo_t *siginfo, void *context) {
3716   // Call the old signal handler
3717   if (actp->sa_handler == SIG_DFL) {
3718     // It's more reasonable to let jvm treat it as an unexpected exception
3719     // instead of taking the default action.
3720     return false;
3721   } else if (actp->sa_handler != SIG_IGN) {
3722     if ((actp->sa_flags & SA_NODEFER) == 0) {
3723       // automaticlly block the signal
3724       sigaddset(&(actp->sa_mask), sig);
3725     }
3726 
3727     sa_handler_t hand;
3728     sa_sigaction_t sa;
3729     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3730     // retrieve the chained handler
3731     if (siginfo_flag_set) {
3732       sa = actp->sa_sigaction;
3733     } else {
3734       hand = actp->sa_handler;
3735     }
3736 
3737     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3738       actp->sa_handler = SIG_DFL;
3739     }
3740 
3741     // try to honor the signal mask
3742     sigset_t oset;
3743     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3744 
3745     // call into the chained handler
3746     if (siginfo_flag_set) {
3747       (*sa)(sig, siginfo, context);
3748     } else {
3749       (*hand)(sig);
3750     }
3751 
3752     // restore the signal mask
3753     pthread_sigmask(SIG_SETMASK, &oset, 0);
3754   }
3755   // Tell jvm's signal handler the signal is taken care of.
3756   return true;
3757 }
3758 
3759 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3760   bool chained = false;
3761   // signal-chaining
3762   if (UseSignalChaining) {
3763     struct sigaction *actp = get_chained_signal_action(sig);
3764     if (actp != NULL) {
3765       chained = call_chained_handler(actp, sig, siginfo, context);
3766     }
3767   }
3768   return chained;
3769 }
3770 
3771 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3772   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3773     return &sigact[sig];
3774   }
3775   return NULL;
3776 }
3777 
3778 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3779   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3780   sigact[sig] = oldAct;
3781   sigs |= (unsigned int)1 << sig;
3782 }
3783 
3784 // for diagnostic
3785 int os::Linux::sigflags[MAXSIGNUM];
3786 
3787 int os::Linux::get_our_sigflags(int sig) {
3788   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3789   return sigflags[sig];
3790 }
3791 
3792 void os::Linux::set_our_sigflags(int sig, int flags) {
3793   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3794   sigflags[sig] = flags;
3795 }
3796 
3797 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3798   // Check for overwrite.
3799   struct sigaction oldAct;
3800   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3801 
3802   void* oldhand = oldAct.sa_sigaction
3803                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3804                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3805   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3806       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3807       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3808     if (AllowUserSignalHandlers || !set_installed) {
3809       // Do not overwrite; user takes responsibility to forward to us.
3810       return;
3811     } else if (UseSignalChaining) {
3812       // save the old handler in jvm
3813       save_preinstalled_handler(sig, oldAct);
3814       // libjsig also interposes the sigaction() call below and saves the
3815       // old sigaction on it own.
3816     } else {
3817       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3818                     "%#lx for signal %d.", (long)oldhand, sig));
3819     }
3820   }
3821 
3822   struct sigaction sigAct;
3823   sigfillset(&(sigAct.sa_mask));
3824   sigAct.sa_handler = SIG_DFL;
3825   if (!set_installed) {
3826     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3827   } else {
3828     sigAct.sa_sigaction = signalHandler;
3829     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3830   }
3831   // Save flags, which are set by ours
3832   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3833   sigflags[sig] = sigAct.sa_flags;
3834 
3835   int ret = sigaction(sig, &sigAct, &oldAct);
3836   assert(ret == 0, "check");
3837 
3838   void* oldhand2  = oldAct.sa_sigaction
3839                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3840                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3841   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3842 }
3843 
3844 // install signal handlers for signals that HotSpot needs to
3845 // handle in order to support Java-level exception handling.
3846 
3847 void os::Linux::install_signal_handlers() {
3848   if (!signal_handlers_are_installed) {
3849     signal_handlers_are_installed = true;
3850 
3851     // signal-chaining
3852     typedef void (*signal_setting_t)();
3853     signal_setting_t begin_signal_setting = NULL;
3854     signal_setting_t end_signal_setting = NULL;
3855     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3856                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3857     if (begin_signal_setting != NULL) {
3858       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3859                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3860       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3861                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3862       libjsig_is_loaded = true;
3863       assert(UseSignalChaining, "should enable signal-chaining");
3864     }
3865     if (libjsig_is_loaded) {
3866       // Tell libjsig jvm is setting signal handlers
3867       (*begin_signal_setting)();
3868     }
3869 
3870     set_signal_handler(SIGSEGV, true);
3871     set_signal_handler(SIGPIPE, true);
3872     set_signal_handler(SIGBUS, true);
3873     set_signal_handler(SIGILL, true);
3874     set_signal_handler(SIGFPE, true);
3875     set_signal_handler(SIGXFSZ, true);
3876 
3877     if (libjsig_is_loaded) {
3878       // Tell libjsig jvm finishes setting signal handlers
3879       (*end_signal_setting)();
3880     }
3881 
3882     // We don't activate signal checker if libjsig is in place, we trust ourselves
3883     // and if UserSignalHandler is installed all bets are off.
3884     // Log that signal checking is off only if -verbose:jni is specified.
3885     if (CheckJNICalls) {
3886       if (libjsig_is_loaded) {
3887         if (PrintJNIResolving) {
3888           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3889         }
3890         check_signals = false;
3891       }
3892       if (AllowUserSignalHandlers) {
3893         if (PrintJNIResolving) {
3894           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3895         }
3896         check_signals = false;
3897       }
3898     }
3899   }
3900 }
3901 
3902 // This is the fastest way to get thread cpu time on Linux.
3903 // Returns cpu time (user+sys) for any thread, not only for current.
3904 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3905 // It might work on 2.6.10+ with a special kernel/glibc patch.
3906 // For reference, please, see IEEE Std 1003.1-2004:
3907 //   http://www.unix.org/single_unix_specification
3908 
3909 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3910   struct timespec tp;
3911   int rc = os::Linux::clock_gettime(clockid, &tp);
3912   assert(rc == 0, "clock_gettime is expected to return 0 code");
3913 
3914   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
3915 }
3916 
3917 /////
3918 // glibc on Linux platform uses non-documented flag
3919 // to indicate, that some special sort of signal
3920 // trampoline is used.
3921 // We will never set this flag, and we should
3922 // ignore this flag in our diagnostic
3923 #ifdef SIGNIFICANT_SIGNAL_MASK
3924 #undef SIGNIFICANT_SIGNAL_MASK
3925 #endif
3926 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3927 
3928 static const char* get_signal_handler_name(address handler,
3929                                            char* buf, int buflen) {
3930   int offset;
3931   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3932   if (found) {
3933     // skip directory names
3934     const char *p1, *p2;
3935     p1 = buf;
3936     size_t len = strlen(os::file_separator());
3937     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3938     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3939   } else {
3940     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3941   }
3942   return buf;
3943 }
3944 
3945 static void print_signal_handler(outputStream* st, int sig,
3946                                  char* buf, size_t buflen) {
3947   struct sigaction sa;
3948 
3949   sigaction(sig, NULL, &sa);
3950 
3951   // See comment for SIGNIFICANT_SIGNAL_MASK define
3952   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3953 
3954   st->print("%s: ", os::exception_name(sig, buf, buflen));
3955 
3956   address handler = (sa.sa_flags & SA_SIGINFO)
3957     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3958     : CAST_FROM_FN_PTR(address, sa.sa_handler);
3959 
3960   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3961     st->print("SIG_DFL");
3962   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3963     st->print("SIG_IGN");
3964   } else {
3965     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3966   }
3967 
3968   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3969 
3970   address rh = VMError::get_resetted_sighandler(sig);
3971   // May be, handler was resetted by VMError?
3972   if(rh != NULL) {
3973     handler = rh;
3974     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3975   }
3976 
3977   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
3978 
3979   // Check: is it our handler?
3980   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3981      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3982     // It is our signal handler
3983     // check for flags, reset system-used one!
3984     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3985       st->print(
3986                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3987                 os::Linux::get_our_sigflags(sig));
3988     }
3989   }
3990   st->cr();
3991 }
3992 
3993 
3994 #define DO_SIGNAL_CHECK(sig) \
3995   if (!sigismember(&check_signal_done, sig)) \
3996     os::Linux::check_signal_handler(sig)
3997 
3998 // This method is a periodic task to check for misbehaving JNI applications
3999 // under CheckJNI, we can add any periodic checks here
4000 
4001 void os::run_periodic_checks() {
4002 
4003   if (check_signals == false) return;
4004 
4005   // SEGV and BUS if overridden could potentially prevent
4006   // generation of hs*.log in the event of a crash, debugging
4007   // such a case can be very challenging, so we absolutely
4008   // check the following for a good measure:
4009   DO_SIGNAL_CHECK(SIGSEGV);
4010   DO_SIGNAL_CHECK(SIGILL);
4011   DO_SIGNAL_CHECK(SIGFPE);
4012   DO_SIGNAL_CHECK(SIGBUS);
4013   DO_SIGNAL_CHECK(SIGPIPE);
4014   DO_SIGNAL_CHECK(SIGXFSZ);
4015 
4016 
4017   // ReduceSignalUsage allows the user to override these handlers
4018   // see comments at the very top and jvm_solaris.h
4019   if (!ReduceSignalUsage) {
4020     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4021     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4022     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4023     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4024   }
4025 
4026   DO_SIGNAL_CHECK(SR_signum);
4027   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4028 }
4029 
4030 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4031 
4032 static os_sigaction_t os_sigaction = NULL;
4033 
4034 void os::Linux::check_signal_handler(int sig) {
4035   char buf[O_BUFLEN];
4036   address jvmHandler = NULL;
4037 
4038 
4039   struct sigaction act;
4040   if (os_sigaction == NULL) {
4041     // only trust the default sigaction, in case it has been interposed
4042     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4043     if (os_sigaction == NULL) return;
4044   }
4045 
4046   os_sigaction(sig, (struct sigaction*)NULL, &act);
4047 
4048 
4049   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4050 
4051   address thisHandler = (act.sa_flags & SA_SIGINFO)
4052     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4053     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4054 
4055 
4056   switch(sig) {
4057   case SIGSEGV:
4058   case SIGBUS:
4059   case SIGFPE:
4060   case SIGPIPE:
4061   case SIGILL:
4062   case SIGXFSZ:
4063     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4064     break;
4065 
4066   case SHUTDOWN1_SIGNAL:
4067   case SHUTDOWN2_SIGNAL:
4068   case SHUTDOWN3_SIGNAL:
4069   case BREAK_SIGNAL:
4070     jvmHandler = (address)user_handler();
4071     break;
4072 
4073   case INTERRUPT_SIGNAL:
4074     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4075     break;
4076 
4077   default:
4078     if (sig == SR_signum) {
4079       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4080     } else {
4081       return;
4082     }
4083     break;
4084   }
4085 
4086   if (thisHandler != jvmHandler) {
4087     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4088     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4089     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4090     // No need to check this sig any longer
4091     sigaddset(&check_signal_done, sig);
4092   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4093     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4094     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4095     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4096     // No need to check this sig any longer
4097     sigaddset(&check_signal_done, sig);
4098   }
4099 
4100   // Dump all the signal
4101   if (sigismember(&check_signal_done, sig)) {
4102     print_signal_handlers(tty, buf, O_BUFLEN);
4103   }
4104 }
4105 
4106 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4107 
4108 extern bool signal_name(int signo, char* buf, size_t len);
4109 
4110 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4111   if (0 < exception_code && exception_code <= SIGRTMAX) {
4112     // signal
4113     if (!signal_name(exception_code, buf, size)) {
4114       jio_snprintf(buf, size, "SIG%d", exception_code);
4115     }
4116     return buf;
4117   } else {
4118     return NULL;
4119   }
4120 }
4121 
4122 // this is called _before_ the most of global arguments have been parsed
4123 void os::init(void) {
4124   char dummy;   /* used to get a guess on initial stack address */
4125 //  first_hrtime = gethrtime();
4126 
4127   // With LinuxThreads the JavaMain thread pid (primordial thread)
4128   // is different than the pid of the java launcher thread.
4129   // So, on Linux, the launcher thread pid is passed to the VM
4130   // via the sun.java.launcher.pid property.
4131   // Use this property instead of getpid() if it was correctly passed.
4132   // See bug 6351349.
4133   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4134 
4135   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4136 
4137   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4138 
4139   init_random(1234567);
4140 
4141   ThreadCritical::initialize();
4142 
4143   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4144   if (Linux::page_size() == -1) {
4145     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4146                   strerror(errno)));
4147   }
4148   init_page_sizes((size_t) Linux::page_size());
4149 
4150   Linux::initialize_system_info();
4151 
4152   // main_thread points to the aboriginal thread
4153   Linux::_main_thread = pthread_self();
4154 
4155   Linux::clock_init();
4156   initial_time_count = os::elapsed_counter();
4157   pthread_mutex_init(&dl_mutex, NULL);
4158 }
4159 
4160 // To install functions for atexit system call
4161 extern "C" {
4162   static void perfMemory_exit_helper() {
4163     perfMemory_exit();
4164   }
4165 }
4166 
4167 // this is called _after_ the global arguments have been parsed
4168 jint os::init_2(void)
4169 {
4170   Linux::fast_thread_clock_init();
4171 
4172   // Allocate a single page and mark it as readable for safepoint polling
4173   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4174   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4175 
4176   os::set_polling_page( polling_page );
4177 
4178 #ifndef PRODUCT
4179   if(Verbose && PrintMiscellaneous)
4180     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4181 #endif
4182 
4183   if (!UseMembar) {
4184     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4185     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4186     os::set_memory_serialize_page( mem_serialize_page );
4187 
4188 #ifndef PRODUCT
4189     if(Verbose && PrintMiscellaneous)
4190       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4191 #endif
4192   }
4193 
4194   os::large_page_init();
4195 
4196   // initialize suspend/resume support - must do this before signal_sets_init()
4197   if (SR_initialize() != 0) {
4198     perror("SR_initialize failed");
4199     return JNI_ERR;
4200   }
4201 
4202   Linux::signal_sets_init();
4203   Linux::install_signal_handlers();
4204 
4205   // Check minimum allowable stack size for thread creation and to initialize
4206   // the java system classes, including StackOverflowError - depends on page
4207   // size.  Add a page for compiler2 recursion in main thread.
4208   // Add in 2*BytesPerWord times page size to account for VM stack during
4209   // class initialization depending on 32 or 64 bit VM.
4210   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4211             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4212                     2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4213 
4214   size_t threadStackSizeInBytes = ThreadStackSize * K;
4215   if (threadStackSizeInBytes != 0 &&
4216       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4217         tty->print_cr("\nThe stack size specified is too small, "
4218                       "Specify at least %dk",
4219                       os::Linux::min_stack_allowed/ K);
4220         return JNI_ERR;
4221   }
4222 
4223   // Make the stack size a multiple of the page size so that
4224   // the yellow/red zones can be guarded.
4225   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4226         vm_page_size()));
4227 
4228   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4229 
4230   Linux::libpthread_init();
4231   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4232      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4233           Linux::glibc_version(), Linux::libpthread_version(),
4234           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4235   }
4236 
4237   if (UseNUMA) {
4238     if (!Linux::libnuma_init()) {
4239       UseNUMA = false;
4240     } else {
4241       if ((Linux::numa_max_node() < 1)) {
4242         // There's only one node(they start from 0), disable NUMA.
4243         UseNUMA = false;
4244       }
4245     }
4246     // With SHM large pages we cannot uncommit a page, so there's not way
4247     // we can make the adaptive lgrp chunk resizing work. If the user specified
4248     // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4249     // disable adaptive resizing.
4250     if (UseNUMA && UseLargePages && UseSHM) {
4251       if (!FLAG_IS_DEFAULT(UseNUMA)) {
4252         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4253           UseLargePages = false;
4254         } else {
4255           warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4256           UseAdaptiveSizePolicy = false;
4257           UseAdaptiveNUMAChunkSizing = false;
4258         }
4259       } else {
4260         UseNUMA = false;
4261       }
4262     }
4263     if (!UseNUMA && ForceNUMA) {
4264       UseNUMA = true;
4265     }
4266   }
4267 
4268   if (MaxFDLimit) {
4269     // set the number of file descriptors to max. print out error
4270     // if getrlimit/setrlimit fails but continue regardless.
4271     struct rlimit nbr_files;
4272     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4273     if (status != 0) {
4274       if (PrintMiscellaneous && (Verbose || WizardMode))
4275         perror("os::init_2 getrlimit failed");
4276     } else {
4277       nbr_files.rlim_cur = nbr_files.rlim_max;
4278       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4279       if (status != 0) {
4280         if (PrintMiscellaneous && (Verbose || WizardMode))
4281           perror("os::init_2 setrlimit failed");
4282       }
4283     }
4284   }
4285 
4286   // Initialize lock used to serialize thread creation (see os::create_thread)
4287   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4288 
4289   // at-exit methods are called in the reverse order of their registration.
4290   // atexit functions are called on return from main or as a result of a
4291   // call to exit(3C). There can be only 32 of these functions registered
4292   // and atexit() does not set errno.
4293 
4294   if (PerfAllowAtExitRegistration) {
4295     // only register atexit functions if PerfAllowAtExitRegistration is set.
4296     // atexit functions can be delayed until process exit time, which
4297     // can be problematic for embedded VM situations. Embedded VMs should
4298     // call DestroyJavaVM() to assure that VM resources are released.
4299 
4300     // note: perfMemory_exit_helper atexit function may be removed in
4301     // the future if the appropriate cleanup code can be added to the
4302     // VM_Exit VMOperation's doit method.
4303     if (atexit(perfMemory_exit_helper) != 0) {
4304       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4305     }
4306   }
4307 
4308   // initialize thread priority policy
4309   prio_init();
4310 
4311   return JNI_OK;
4312 }
4313 
4314 // this is called at the end of vm_initialization
4315 void os::init_3(void)
4316 {
4317 #ifdef JAVASE_EMBEDDED
4318   // Start the MemNotifyThread
4319   if (LowMemoryProtection) {
4320     MemNotifyThread::start();
4321   }
4322   return;
4323 #endif
4324 }
4325 
4326 // Mark the polling page as unreadable
4327 void os::make_polling_page_unreadable(void) {
4328   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4329     fatal("Could not disable polling page");
4330 };
4331 
4332 // Mark the polling page as readable
4333 void os::make_polling_page_readable(void) {
4334   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4335     fatal("Could not enable polling page");
4336   }
4337 };
4338 
4339 int os::active_processor_count() {
4340   // Linux doesn't yet have a (official) notion of processor sets,
4341   // so just return the number of online processors.
4342   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4343   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4344   return online_cpus;
4345 }
4346 
4347 void os::set_native_thread_name(const char *name) {
4348   // Not yet implemented.
4349   return;
4350 }
4351 
4352 bool os::distribute_processes(uint length, uint* distribution) {
4353   // Not yet implemented.
4354   return false;
4355 }
4356 
4357 bool os::bind_to_processor(uint processor_id) {
4358   // Not yet implemented.
4359   return false;
4360 }
4361 
4362 ///
4363 
4364 // Suspends the target using the signal mechanism and then grabs the PC before
4365 // resuming the target. Used by the flat-profiler only
4366 ExtendedPC os::get_thread_pc(Thread* thread) {
4367   // Make sure that it is called by the watcher for the VMThread
4368   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4369   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4370 
4371   ExtendedPC epc;
4372 
4373   OSThread* osthread = thread->osthread();
4374   if (do_suspend(osthread)) {
4375     if (osthread->ucontext() != NULL) {
4376       epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4377     } else {
4378       // NULL context is unexpected, double-check this is the VMThread
4379       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4380     }
4381     do_resume(osthread);
4382   }
4383   // failure means pthread_kill failed for some reason - arguably this is
4384   // a fatal problem, but such problems are ignored elsewhere
4385 
4386   return epc;
4387 }
4388 
4389 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4390 {
4391    if (is_NPTL()) {
4392       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4393    } else {
4394 #ifndef IA64
4395       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4396       // word back to default 64bit precision if condvar is signaled. Java
4397       // wants 53bit precision.  Save and restore current value.
4398       int fpu = get_fpu_control_word();
4399 #endif // IA64
4400       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4401 #ifndef IA64
4402       set_fpu_control_word(fpu);
4403 #endif // IA64
4404       return status;
4405    }
4406 }
4407 
4408 ////////////////////////////////////////////////////////////////////////////////
4409 // debug support
4410 
4411 static address same_page(address x, address y) {
4412   int page_bits = -os::vm_page_size();
4413   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4414     return x;
4415   else if (x > y)
4416     return (address)(intptr_t(y) | ~page_bits) + 1;
4417   else
4418     return (address)(intptr_t(y) & page_bits);
4419 }
4420 
4421 bool os::find(address addr, outputStream* st) {
4422   Dl_info dlinfo;
4423   memset(&dlinfo, 0, sizeof(dlinfo));
4424   if (dladdr(addr, &dlinfo)) {
4425     st->print(PTR_FORMAT ": ", addr);
4426     if (dlinfo.dli_sname != NULL) {
4427       st->print("%s+%#x", dlinfo.dli_sname,
4428                  addr - (intptr_t)dlinfo.dli_saddr);
4429     } else if (dlinfo.dli_fname) {
4430       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4431     } else {
4432       st->print("<absolute address>");
4433     }
4434     if (dlinfo.dli_fname) {
4435       st->print(" in %s", dlinfo.dli_fname);
4436     }
4437     if (dlinfo.dli_fbase) {
4438       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4439     }
4440     st->cr();
4441 
4442     if (Verbose) {
4443       // decode some bytes around the PC
4444       address begin = same_page(addr-40, addr);
4445       address end   = same_page(addr+40, addr);
4446       address       lowest = (address) dlinfo.dli_sname;
4447       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4448       if (begin < lowest)  begin = lowest;
4449       Dl_info dlinfo2;
4450       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4451           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4452         end = (address) dlinfo2.dli_saddr;
4453       Disassembler::decode(begin, end, st);
4454     }
4455     return true;
4456   }
4457   return false;
4458 }
4459 
4460 ////////////////////////////////////////////////////////////////////////////////
4461 // misc
4462 
4463 // This does not do anything on Linux. This is basically a hook for being
4464 // able to use structured exception handling (thread-local exception filters)
4465 // on, e.g., Win32.
4466 void
4467 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4468                          JavaCallArguments* args, Thread* thread) {
4469   f(value, method, args, thread);
4470 }
4471 
4472 void os::print_statistics() {
4473 }
4474 
4475 int os::message_box(const char* title, const char* message) {
4476   int i;
4477   fdStream err(defaultStream::error_fd());
4478   for (i = 0; i < 78; i++) err.print_raw("=");
4479   err.cr();
4480   err.print_raw_cr(title);
4481   for (i = 0; i < 78; i++) err.print_raw("-");
4482   err.cr();
4483   err.print_raw_cr(message);
4484   for (i = 0; i < 78; i++) err.print_raw("=");
4485   err.cr();
4486 
4487   char buf[16];
4488   // Prevent process from exiting upon "read error" without consuming all CPU
4489   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4490 
4491   return buf[0] == 'y' || buf[0] == 'Y';
4492 }
4493 
4494 int os::stat(const char *path, struct stat *sbuf) {
4495   char pathbuf[MAX_PATH];
4496   if (strlen(path) > MAX_PATH - 1) {
4497     errno = ENAMETOOLONG;
4498     return -1;
4499   }
4500   os::native_path(strcpy(pathbuf, path));
4501   return ::stat(pathbuf, sbuf);
4502 }
4503 
4504 bool os::check_heap(bool force) {
4505   return true;
4506 }
4507 
4508 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4509   return ::vsnprintf(buf, count, format, args);
4510 }
4511 
4512 // Is a (classpath) directory empty?
4513 bool os::dir_is_empty(const char* path) {
4514   DIR *dir = NULL;
4515   struct dirent *ptr;
4516 
4517   dir = opendir(path);
4518   if (dir == NULL) return true;
4519 
4520   /* Scan the directory */
4521   bool result = true;
4522   char buf[sizeof(struct dirent) + MAX_PATH];
4523   while (result && (ptr = ::readdir(dir)) != NULL) {
4524     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4525       result = false;
4526     }
4527   }
4528   closedir(dir);
4529   return result;
4530 }
4531 
4532 // This code originates from JDK's sysOpen and open64_w
4533 // from src/solaris/hpi/src/system_md.c
4534 
4535 #ifndef O_DELETE
4536 #define O_DELETE 0x10000
4537 #endif
4538 
4539 // Open a file. Unlink the file immediately after open returns
4540 // if the specified oflag has the O_DELETE flag set.
4541 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4542 
4543 int os::open(const char *path, int oflag, int mode) {
4544 
4545   if (strlen(path) > MAX_PATH - 1) {
4546     errno = ENAMETOOLONG;
4547     return -1;
4548   }
4549   int fd;
4550   int o_delete = (oflag & O_DELETE);
4551   oflag = oflag & ~O_DELETE;
4552 
4553   fd = ::open64(path, oflag, mode);
4554   if (fd == -1) return -1;
4555 
4556   //If the open succeeded, the file might still be a directory
4557   {
4558     struct stat64 buf64;
4559     int ret = ::fstat64(fd, &buf64);
4560     int st_mode = buf64.st_mode;
4561 
4562     if (ret != -1) {
4563       if ((st_mode & S_IFMT) == S_IFDIR) {
4564         errno = EISDIR;
4565         ::close(fd);
4566         return -1;
4567       }
4568     } else {
4569       ::close(fd);
4570       return -1;
4571     }
4572   }
4573 
4574     /*
4575      * All file descriptors that are opened in the JVM and not
4576      * specifically destined for a subprocess should have the
4577      * close-on-exec flag set.  If we don't set it, then careless 3rd
4578      * party native code might fork and exec without closing all
4579      * appropriate file descriptors (e.g. as we do in closeDescriptors in
4580      * UNIXProcess.c), and this in turn might:
4581      *
4582      * - cause end-of-file to fail to be detected on some file
4583      *   descriptors, resulting in mysterious hangs, or
4584      *
4585      * - might cause an fopen in the subprocess to fail on a system
4586      *   suffering from bug 1085341.
4587      *
4588      * (Yes, the default setting of the close-on-exec flag is a Unix
4589      * design flaw)
4590      *
4591      * See:
4592      * 1085341: 32-bit stdio routines should support file descriptors >255
4593      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4594      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4595      */
4596 #ifdef FD_CLOEXEC
4597     {
4598         int flags = ::fcntl(fd, F_GETFD);
4599         if (flags != -1)
4600             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4601     }
4602 #endif
4603 
4604   if (o_delete != 0) {
4605     ::unlink(path);
4606   }
4607   return fd;
4608 }
4609 
4610 
4611 // create binary file, rewriting existing file if required
4612 int os::create_binary_file(const char* path, bool rewrite_existing) {
4613   int oflags = O_WRONLY | O_CREAT;
4614   if (!rewrite_existing) {
4615     oflags |= O_EXCL;
4616   }
4617   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4618 }
4619 
4620 // return current position of file pointer
4621 jlong os::current_file_offset(int fd) {
4622   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4623 }
4624 
4625 // move file pointer to the specified offset
4626 jlong os::seek_to_file_offset(int fd, jlong offset) {
4627   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4628 }
4629 
4630 // This code originates from JDK's sysAvailable
4631 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4632 
4633 int os::available(int fd, jlong *bytes) {
4634   jlong cur, end;
4635   int mode;
4636   struct stat64 buf64;
4637 
4638   if (::fstat64(fd, &buf64) >= 0) {
4639     mode = buf64.st_mode;
4640     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4641       /*
4642       * XXX: is the following call interruptible? If so, this might
4643       * need to go through the INTERRUPT_IO() wrapper as for other
4644       * blocking, interruptible calls in this file.
4645       */
4646       int n;
4647       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4648         *bytes = n;
4649         return 1;
4650       }
4651     }
4652   }
4653   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4654     return 0;
4655   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4656     return 0;
4657   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4658     return 0;
4659   }
4660   *bytes = end - cur;
4661   return 1;
4662 }
4663 
4664 int os::socket_available(int fd, jint *pbytes) {
4665   // Linux doc says EINTR not returned, unlike Solaris
4666   int ret = ::ioctl(fd, FIONREAD, pbytes);
4667 
4668   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4669   // is expected to return 0 on failure and 1 on success to the jdk.
4670   return (ret < 0) ? 0 : 1;
4671 }
4672 
4673 // Map a block of memory.
4674 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4675                      char *addr, size_t bytes, bool read_only,
4676                      bool allow_exec) {
4677   int prot;
4678   int flags = MAP_PRIVATE;
4679 
4680   if (read_only) {
4681     prot = PROT_READ;
4682   } else {
4683     prot = PROT_READ | PROT_WRITE;
4684   }
4685 
4686   if (allow_exec) {
4687     prot |= PROT_EXEC;
4688   }
4689 
4690   if (addr != NULL) {
4691     flags |= MAP_FIXED;
4692   }
4693 
4694   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4695                                      fd, file_offset);
4696   if (mapped_address == MAP_FAILED) {
4697     return NULL;
4698   }
4699   return mapped_address;
4700 }
4701 
4702 
4703 // Remap a block of memory.
4704 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4705                        char *addr, size_t bytes, bool read_only,
4706                        bool allow_exec) {
4707   // same as map_memory() on this OS
4708   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4709                         allow_exec);
4710 }
4711 
4712 
4713 // Unmap a block of memory.
4714 bool os::unmap_memory(char* addr, size_t bytes) {
4715   return munmap(addr, bytes) == 0;
4716 }
4717 
4718 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4719 
4720 static clockid_t thread_cpu_clockid(Thread* thread) {
4721   pthread_t tid = thread->osthread()->pthread_id();
4722   clockid_t clockid;
4723 
4724   // Get thread clockid
4725   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4726   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4727   return clockid;
4728 }
4729 
4730 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4731 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4732 // of a thread.
4733 //
4734 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4735 // the fast estimate available on the platform.
4736 
4737 jlong os::current_thread_cpu_time() {
4738   if (os::Linux::supports_fast_thread_cpu_time()) {
4739     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4740   } else {
4741     // return user + sys since the cost is the same
4742     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4743   }
4744 }
4745 
4746 jlong os::thread_cpu_time(Thread* thread) {
4747   // consistent with what current_thread_cpu_time() returns
4748   if (os::Linux::supports_fast_thread_cpu_time()) {
4749     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4750   } else {
4751     return slow_thread_cpu_time(thread, true /* user + sys */);
4752   }
4753 }
4754 
4755 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4756   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4757     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4758   } else {
4759     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4760   }
4761 }
4762 
4763 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4764   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4765     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4766   } else {
4767     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4768   }
4769 }
4770 
4771 //
4772 //  -1 on error.
4773 //
4774 
4775 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4776   static bool proc_pid_cpu_avail = true;
4777   static bool proc_task_unchecked = true;
4778   static const char *proc_stat_path = "/proc/%d/stat";
4779   pid_t  tid = thread->osthread()->thread_id();
4780   int i;
4781   char *s;
4782   char stat[2048];
4783   int statlen;
4784   char proc_name[64];
4785   int count;
4786   long sys_time, user_time;
4787   char string[64];
4788   char cdummy;
4789   int idummy;
4790   long ldummy;
4791   FILE *fp;
4792 
4793   // We first try accessing /proc/<pid>/cpu since this is faster to
4794   // process.  If this file is not present (linux kernels 2.5 and above)
4795   // then we open /proc/<pid>/stat.
4796   if ( proc_pid_cpu_avail ) {
4797     sprintf(proc_name, "/proc/%d/cpu", tid);
4798     fp =  fopen(proc_name, "r");
4799     if ( fp != NULL ) {
4800       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4801       fclose(fp);
4802       if ( count != 3 ) return -1;
4803 
4804       if (user_sys_cpu_time) {
4805         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4806       } else {
4807         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4808       }
4809     }
4810     else proc_pid_cpu_avail = false;
4811   }
4812 
4813   // The /proc/<tid>/stat aggregates per-process usage on
4814   // new Linux kernels 2.6+ where NPTL is supported.
4815   // The /proc/self/task/<tid>/stat still has the per-thread usage.
4816   // See bug 6328462.
4817   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4818   // and possibly in some other cases, so we check its availability.
4819   if (proc_task_unchecked && os::Linux::is_NPTL()) {
4820     // This is executed only once
4821     proc_task_unchecked = false;
4822     fp = fopen("/proc/self/task", "r");
4823     if (fp != NULL) {
4824       proc_stat_path = "/proc/self/task/%d/stat";
4825       fclose(fp);
4826     }
4827   }
4828 
4829   sprintf(proc_name, proc_stat_path, tid);
4830   fp = fopen(proc_name, "r");
4831   if ( fp == NULL ) return -1;
4832   statlen = fread(stat, 1, 2047, fp);
4833   stat[statlen] = '\0';
4834   fclose(fp);
4835 
4836   // Skip pid and the command string. Note that we could be dealing with
4837   // weird command names, e.g. user could decide to rename java launcher
4838   // to "java 1.4.2 :)", then the stat file would look like
4839   //                1234 (java 1.4.2 :)) R ... ...
4840   // We don't really need to know the command string, just find the last
4841   // occurrence of ")" and then start parsing from there. See bug 4726580.
4842   s = strrchr(stat, ')');
4843   i = 0;
4844   if (s == NULL ) return -1;
4845 
4846   // Skip blank chars
4847   do s++; while (isspace(*s));
4848 
4849   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4850                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4851                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4852                  &user_time, &sys_time);
4853   if ( count != 13 ) return -1;
4854   if (user_sys_cpu_time) {
4855     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4856   } else {
4857     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4858   }
4859 }
4860 
4861 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4862   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4863   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4864   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4865   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4866 }
4867 
4868 void os::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 bool os::is_thread_cpu_time_supported() {
4876   return true;
4877 }
4878 
4879 // System loadavg support.  Returns -1 if load average cannot be obtained.
4880 // Linux doesn't yet have a (official) notion of processor sets,
4881 // so just return the system wide load average.
4882 int os::loadavg(double loadavg[], int nelem) {
4883   return ::getloadavg(loadavg, nelem);
4884 }
4885 
4886 void os::pause() {
4887   char filename[MAX_PATH];
4888   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4889     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4890   } else {
4891     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4892   }
4893 
4894   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4895   if (fd != -1) {
4896     struct stat buf;
4897     ::close(fd);
4898     while (::stat(filename, &buf) == 0) {
4899       (void)::poll(NULL, 0, 100);
4900     }
4901   } else {
4902     jio_fprintf(stderr,
4903       "Could not open pause file '%s', continuing immediately.\n", filename);
4904   }
4905 }
4906 
4907 
4908 // Refer to the comments in os_solaris.cpp park-unpark.
4909 //
4910 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4911 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4912 // For specifics regarding the bug see GLIBC BUGID 261237 :
4913 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4914 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4915 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4916 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4917 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4918 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4919 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4920 // of libpthread avoids the problem, but isn't practical.
4921 //
4922 // Possible remedies:
4923 //
4924 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4925 //      This is palliative and probabilistic, however.  If the thread is preempted
4926 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4927 //      than the minimum period may have passed, and the abstime may be stale (in the
4928 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
4929 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
4930 //
4931 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4932 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
4933 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4934 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
4935 //      thread.
4936 //
4937 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
4938 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
4939 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
4940 //      This also works well.  In fact it avoids kernel-level scalability impediments
4941 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
4942 //      timers in a graceful fashion.
4943 //
4944 // 4.   When the abstime value is in the past it appears that control returns
4945 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4946 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
4947 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
4948 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
4949 //      It may be possible to avoid reinitialization by checking the return
4950 //      value from pthread_cond_timedwait().  In addition to reinitializing the
4951 //      condvar we must establish the invariant that cond_signal() is only called
4952 //      within critical sections protected by the adjunct mutex.  This prevents
4953 //      cond_signal() from "seeing" a condvar that's in the midst of being
4954 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
4955 //      desirable signal-after-unlock optimization that avoids futile context switching.
4956 //
4957 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
4958 //      structure when a condvar is used or initialized.  cond_destroy()  would
4959 //      release the helper structure.  Our reinitialize-after-timedwait fix
4960 //      put excessive stress on malloc/free and locks protecting the c-heap.
4961 //
4962 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
4963 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4964 // and only enabling the work-around for vulnerable environments.
4965 
4966 // utility to compute the abstime argument to timedwait:
4967 // millis is the relative timeout time
4968 // abstime will be the absolute timeout time
4969 // TODO: replace compute_abstime() with unpackTime()
4970 
4971 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4972   if (millis < 0)  millis = 0;
4973   struct timeval now;
4974   int status = gettimeofday(&now, NULL);
4975   assert(status == 0, "gettimeofday");
4976   jlong seconds = millis / 1000;
4977   millis %= 1000;
4978   if (seconds > 50000000) { // see man cond_timedwait(3T)
4979     seconds = 50000000;
4980   }
4981   abstime->tv_sec = now.tv_sec  + seconds;
4982   long       usec = now.tv_usec + millis * 1000;
4983   if (usec >= 1000000) {
4984     abstime->tv_sec += 1;
4985     usec -= 1000000;
4986   }
4987   abstime->tv_nsec = usec * 1000;
4988   return abstime;
4989 }
4990 
4991 
4992 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4993 // Conceptually TryPark() should be equivalent to park(0).
4994 
4995 int os::PlatformEvent::TryPark() {
4996   for (;;) {
4997     const int v = _Event ;
4998     guarantee ((v == 0) || (v == 1), "invariant") ;
4999     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5000   }
5001 }
5002 
5003 void os::PlatformEvent::park() {       // AKA "down()"
5004   // Invariant: Only the thread associated with the Event/PlatformEvent
5005   // may call park().
5006   // TODO: assert that _Assoc != NULL or _Assoc == Self
5007   int v ;
5008   for (;;) {
5009       v = _Event ;
5010       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5011   }
5012   guarantee (v >= 0, "invariant") ;
5013   if (v == 0) {
5014      // Do this the hard way by blocking ...
5015      int status = pthread_mutex_lock(_mutex);
5016      assert_status(status == 0, status, "mutex_lock");
5017      guarantee (_nParked == 0, "invariant") ;
5018      ++ _nParked ;
5019      while (_Event < 0) {
5020         status = pthread_cond_wait(_cond, _mutex);
5021         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5022         // Treat this the same as if the wait was interrupted
5023         if (status == ETIME) { status = EINTR; }
5024         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5025      }
5026      -- _nParked ;
5027 
5028     // In theory we could move the ST of 0 into _Event past the unlock(),
5029     // but then we'd need a MEMBAR after the ST.
5030     _Event = 0 ;
5031      status = pthread_mutex_unlock(_mutex);
5032      assert_status(status == 0, status, "mutex_unlock");
5033   }
5034   guarantee (_Event >= 0, "invariant") ;
5035 }
5036 
5037 int os::PlatformEvent::park(jlong millis) {
5038   guarantee (_nParked == 0, "invariant") ;
5039 
5040   int v ;
5041   for (;;) {
5042       v = _Event ;
5043       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5044   }
5045   guarantee (v >= 0, "invariant") ;
5046   if (v != 0) return OS_OK ;
5047 
5048   // We do this the hard way, by blocking the thread.
5049   // Consider enforcing a minimum timeout value.
5050   struct timespec abst;
5051   compute_abstime(&abst, millis);
5052 
5053   int ret = OS_TIMEOUT;
5054   int status = pthread_mutex_lock(_mutex);
5055   assert_status(status == 0, status, "mutex_lock");
5056   guarantee (_nParked == 0, "invariant") ;
5057   ++_nParked ;
5058 
5059   // Object.wait(timo) will return because of
5060   // (a) notification
5061   // (b) timeout
5062   // (c) thread.interrupt
5063   //
5064   // Thread.interrupt and object.notify{All} both call Event::set.
5065   // That is, we treat thread.interrupt as a special case of notification.
5066   // The underlying Solaris implementation, cond_timedwait, admits
5067   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5068   // JVM from making those visible to Java code.  As such, we must
5069   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5070   //
5071   // TODO: properly differentiate simultaneous notify+interrupt.
5072   // In that case, we should propagate the notify to another waiter.
5073 
5074   while (_Event < 0) {
5075     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5076     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5077       pthread_cond_destroy (_cond);
5078       pthread_cond_init (_cond, NULL) ;
5079     }
5080     assert_status(status == 0 || status == EINTR ||
5081                   status == ETIME || status == ETIMEDOUT,
5082                   status, "cond_timedwait");
5083     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5084     if (status == ETIME || status == ETIMEDOUT) break ;
5085     // We consume and ignore EINTR and spurious wakeups.
5086   }
5087   --_nParked ;
5088   if (_Event >= 0) {
5089      ret = OS_OK;
5090   }
5091   _Event = 0 ;
5092   status = pthread_mutex_unlock(_mutex);
5093   assert_status(status == 0, status, "mutex_unlock");
5094   assert (_nParked == 0, "invariant") ;
5095   return ret;
5096 }
5097 
5098 void os::PlatformEvent::unpark() {
5099   int v, AnyWaiters ;
5100   for (;;) {
5101       v = _Event ;
5102       if (v > 0) {
5103          // The LD of _Event could have reordered or be satisfied
5104          // by a read-aside from this processor's write buffer.
5105          // To avoid problems execute a barrier and then
5106          // ratify the value.
5107          OrderAccess::fence() ;
5108          if (_Event == v) return ;
5109          continue ;
5110       }
5111       if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5112   }
5113   if (v < 0) {
5114      // Wait for the thread associated with the event to vacate
5115      int status = pthread_mutex_lock(_mutex);
5116      assert_status(status == 0, status, "mutex_lock");
5117      AnyWaiters = _nParked ;
5118      assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5119      if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5120         AnyWaiters = 0 ;
5121         pthread_cond_signal (_cond);
5122      }
5123      status = pthread_mutex_unlock(_mutex);
5124      assert_status(status == 0, status, "mutex_unlock");
5125      if (AnyWaiters != 0) {
5126         status = pthread_cond_signal(_cond);
5127         assert_status(status == 0, status, "cond_signal");
5128      }
5129   }
5130 
5131   // Note that we signal() _after dropping the lock for "immortal" Events.
5132   // This is safe and avoids a common class of  futile wakeups.  In rare
5133   // circumstances this can cause a thread to return prematurely from
5134   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5135   // simply re-test the condition and re-park itself.
5136 }
5137 
5138 
5139 // JSR166
5140 // -------------------------------------------------------
5141 
5142 /*
5143  * The solaris and linux implementations of park/unpark are fairly
5144  * conservative for now, but can be improved. They currently use a
5145  * mutex/condvar pair, plus a a count.
5146  * Park decrements count if > 0, else does a condvar wait.  Unpark
5147  * sets count to 1 and signals condvar.  Only one thread ever waits
5148  * on the condvar. Contention seen when trying to park implies that someone
5149  * is unparking you, so don't wait. And spurious returns are fine, so there
5150  * is no need to track notifications.
5151  */
5152 
5153 #define MAX_SECS 100000000
5154 /*
5155  * This code is common to linux and solaris and will be moved to a
5156  * common place in dolphin.
5157  *
5158  * The passed in time value is either a relative time in nanoseconds
5159  * or an absolute time in milliseconds. Either way it has to be unpacked
5160  * into suitable seconds and nanoseconds components and stored in the
5161  * given timespec structure.
5162  * Given time is a 64-bit value and the time_t used in the timespec is only
5163  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5164  * overflow if times way in the future are given. Further on Solaris versions
5165  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5166  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5167  * As it will be 28 years before "now + 100000000" will overflow we can
5168  * ignore overflow and just impose a hard-limit on seconds using the value
5169  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5170  * years from "now".
5171  */
5172 
5173 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5174   assert (time > 0, "convertTime");
5175 
5176   struct timeval now;
5177   int status = gettimeofday(&now, NULL);
5178   assert(status == 0, "gettimeofday");
5179 
5180   time_t max_secs = now.tv_sec + MAX_SECS;
5181 
5182   if (isAbsolute) {
5183     jlong secs = time / 1000;
5184     if (secs > max_secs) {
5185       absTime->tv_sec = max_secs;
5186     }
5187     else {
5188       absTime->tv_sec = secs;
5189     }
5190     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5191   }
5192   else {
5193     jlong secs = time / NANOSECS_PER_SEC;
5194     if (secs >= MAX_SECS) {
5195       absTime->tv_sec = max_secs;
5196       absTime->tv_nsec = 0;
5197     }
5198     else {
5199       absTime->tv_sec = now.tv_sec + secs;
5200       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5201       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5202         absTime->tv_nsec -= NANOSECS_PER_SEC;
5203         ++absTime->tv_sec; // note: this must be <= max_secs
5204       }
5205     }
5206   }
5207   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5208   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5209   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5210   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5211 }
5212 
5213 void Parker::park(bool isAbsolute, jlong time) {
5214   // Optional fast-path check:
5215   // Return immediately if a permit is available.
5216   if (_counter > 0) {
5217       _counter = 0 ;
5218       OrderAccess::fence();
5219       return ;
5220   }
5221 
5222   Thread* thread = Thread::current();
5223   assert(thread->is_Java_thread(), "Must be JavaThread");
5224   JavaThread *jt = (JavaThread *)thread;
5225 
5226   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5227   // Check interrupt before trying to wait
5228   if (Thread::is_interrupted(thread, false)) {
5229     return;
5230   }
5231 
5232   // Next, demultiplex/decode time arguments
5233   timespec absTime;
5234   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5235     return;
5236   }
5237   if (time > 0) {
5238     unpackTime(&absTime, isAbsolute, time);
5239   }
5240 
5241 
5242   // Enter safepoint region
5243   // Beware of deadlocks such as 6317397.
5244   // The per-thread Parker:: mutex is a classic leaf-lock.
5245   // In particular a thread must never block on the Threads_lock while
5246   // holding the Parker:: mutex.  If safepoints are pending both the
5247   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5248   ThreadBlockInVM tbivm(jt);
5249 
5250   // Don't wait if cannot get lock since interference arises from
5251   // unblocking.  Also. check interrupt before trying wait
5252   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5253     return;
5254   }
5255 
5256   int status ;
5257   if (_counter > 0)  { // no wait needed
5258     _counter = 0;
5259     status = pthread_mutex_unlock(_mutex);
5260     assert (status == 0, "invariant") ;
5261     OrderAccess::fence();
5262     return;
5263   }
5264 
5265 #ifdef ASSERT
5266   // Don't catch signals while blocked; let the running threads have the signals.
5267   // (This allows a debugger to break into the running thread.)
5268   sigset_t oldsigs;
5269   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5270   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5271 #endif
5272 
5273   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5274   jt->set_suspend_equivalent();
5275   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5276 
5277   if (time == 0) {
5278     status = pthread_cond_wait (_cond, _mutex) ;
5279   } else {
5280     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5281     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5282       pthread_cond_destroy (_cond) ;
5283       pthread_cond_init    (_cond, NULL);
5284     }
5285   }
5286   assert_status(status == 0 || status == EINTR ||
5287                 status == ETIME || status == ETIMEDOUT,
5288                 status, "cond_timedwait");
5289 
5290 #ifdef ASSERT
5291   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5292 #endif
5293 
5294   _counter = 0 ;
5295   status = pthread_mutex_unlock(_mutex) ;
5296   assert_status(status == 0, status, "invariant") ;
5297   // If externally suspended while waiting, re-suspend
5298   if (jt->handle_special_suspend_equivalent_condition()) {
5299     jt->java_suspend_self();
5300   }
5301 
5302   OrderAccess::fence();
5303 }
5304 
5305 void Parker::unpark() {
5306   int s, status ;
5307   status = pthread_mutex_lock(_mutex);
5308   assert (status == 0, "invariant") ;
5309   s = _counter;
5310   _counter = 1;
5311   if (s < 1) {
5312      if (WorkAroundNPTLTimedWaitHang) {
5313         status = pthread_cond_signal (_cond) ;
5314         assert (status == 0, "invariant") ;
5315         status = pthread_mutex_unlock(_mutex);
5316         assert (status == 0, "invariant") ;
5317      } else {
5318         status = pthread_mutex_unlock(_mutex);
5319         assert (status == 0, "invariant") ;
5320         status = pthread_cond_signal (_cond) ;
5321         assert (status == 0, "invariant") ;
5322      }
5323   } else {
5324     pthread_mutex_unlock(_mutex);
5325     assert (status == 0, "invariant") ;
5326   }
5327 }
5328 
5329 
5330 extern char** environ;
5331 
5332 #ifndef __NR_fork
5333 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5334 #endif
5335 
5336 #ifndef __NR_execve
5337 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5338 #endif
5339 
5340 // Run the specified command in a separate process. Return its exit value,
5341 // or -1 on failure (e.g. can't fork a new process).
5342 // Unlike system(), this function can be called from signal handler. It
5343 // doesn't block SIGINT et al.
5344 int os::fork_and_exec(char* cmd) {
5345   const char * argv[4] = {"sh", "-c", cmd, NULL};
5346 
5347   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5348   // pthread_atfork handlers and reset pthread library. All we need is a
5349   // separate process to execve. Make a direct syscall to fork process.
5350   // On IA64 there's no fork syscall, we have to use fork() and hope for
5351   // the best...
5352   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5353               IA64_ONLY(fork();)
5354 
5355   if (pid < 0) {
5356     // fork failed
5357     return -1;
5358 
5359   } else if (pid == 0) {
5360     // child process
5361 
5362     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5363     // first to kill every thread on the thread list. Because this list is
5364     // not reset by fork() (see notes above), execve() will instead kill
5365     // every thread in the parent process. We know this is the only thread
5366     // in the new process, so make a system call directly.
5367     // IA64 should use normal execve() from glibc to match the glibc fork()
5368     // above.
5369     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5370     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5371 
5372     // execve failed
5373     _exit(-1);
5374 
5375   } else  {
5376     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5377     // care about the actual exit code, for now.
5378 
5379     int status;
5380 
5381     // Wait for the child process to exit.  This returns immediately if
5382     // the child has already exited. */
5383     while (waitpid(pid, &status, 0) < 0) {
5384         switch (errno) {
5385         case ECHILD: return 0;
5386         case EINTR: break;
5387         default: return -1;
5388         }
5389     }
5390 
5391     if (WIFEXITED(status)) {
5392        // The child exited normally; get its exit code.
5393        return WEXITSTATUS(status);
5394     } else if (WIFSIGNALED(status)) {
5395        // The child exited because of a signal
5396        // The best value to return is 0x80 + signal number,
5397        // because that is what all Unix shells do, and because
5398        // it allows callers to distinguish between process exit and
5399        // process death by signal.
5400        return 0x80 + WTERMSIG(status);
5401     } else {
5402        // Unknown exit code; pass it through
5403        return status;
5404     }
5405   }
5406 }
5407 
5408 // is_headless_jre()
5409 //
5410 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5411 // in order to report if we are running in a headless jre
5412 //
5413 // Since JDK8 xawt/libmawt.so was moved into the same directory
5414 // as libawt.so, and renamed libawt_xawt.so
5415 //
5416 bool os::is_headless_jre() {
5417     struct stat statbuf;
5418     char buf[MAXPATHLEN];
5419     char libmawtpath[MAXPATHLEN];
5420     const char *xawtstr  = "/xawt/libmawt.so";
5421     const char *new_xawtstr = "/libawt_xawt.so";
5422     char *p;
5423 
5424     // Get path to libjvm.so
5425     os::jvm_path(buf, sizeof(buf));
5426 
5427     // Get rid of libjvm.so
5428     p = strrchr(buf, '/');
5429     if (p == NULL) return false;
5430     else *p = '\0';
5431 
5432     // Get rid of client or server
5433     p = strrchr(buf, '/');
5434     if (p == NULL) return false;
5435     else *p = '\0';
5436 
5437     // check xawt/libmawt.so
5438     strcpy(libmawtpath, buf);
5439     strcat(libmawtpath, xawtstr);
5440     if (::stat(libmawtpath, &statbuf) == 0) return false;
5441 
5442     // check libawt_xawt.so
5443     strcpy(libmawtpath, buf);
5444     strcat(libmawtpath, new_xawtstr);
5445     if (::stat(libmawtpath, &statbuf) == 0) return false;
5446 
5447     return true;
5448 }
5449 
5450 // Generate the candidate core paths into the supplied buffer
5451 // Return the number of characters actually generated
5452 int os::get_core_paths(char* buffer, size_t bufferSize) {
5453   static char cwd[O_BUFLEN];
5454   int n;
5455 
5456   get_current_directory(cwd, sizeof(cwd));
5457 
5458   n = jio_snprintf(buffer, bufferSize, "%s/core or core.%d", cwd, current_process_id());
5459 
5460   // Truncate if theoretical string was longer than bufferSize
5461   n = MIN2(n, (int)bufferSize);
5462 
5463   return n;
5464 }
5465 
5466 #ifdef JAVASE_EMBEDDED
5467 //
5468 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5469 //
5470 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5471 
5472 // ctor
5473 //
5474 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5475   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5476   _fd = fd;
5477 
5478   if (os::create_thread(this, os::os_thread)) {
5479     _memnotify_thread = this;
5480     os::set_priority(this, NearMaxPriority);
5481     os::start_thread(this);
5482   }
5483 }
5484 
5485 // Where all the work gets done
5486 //
5487 void MemNotifyThread::run() {
5488   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5489 
5490   // Set up the select arguments
5491   fd_set rfds;
5492   if (_fd != -1) {
5493     FD_ZERO(&rfds);
5494     FD_SET(_fd, &rfds);
5495   }
5496 
5497   // Now wait for the mem_notify device to wake up
5498   while (1) {
5499     // Wait for the mem_notify device to signal us..
5500     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5501     if (rc == -1) {
5502       perror("select!\n");
5503       break;
5504     } else if (rc) {
5505       //ssize_t free_before = os::available_memory();
5506       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5507 
5508       // The kernel is telling us there is not much memory left...
5509       // try to do something about that
5510 
5511       // If we are not already in a GC, try one.
5512       if (!Universe::heap()->is_gc_active()) {
5513         Universe::heap()->collect(GCCause::_allocation_failure);
5514 
5515         //ssize_t free_after = os::available_memory();
5516         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5517         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5518       }
5519       // We might want to do something like the following if we find the GC's are not helping...
5520       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5521     }
5522   }
5523 }
5524 
5525 //
5526 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5527 //
5528 void MemNotifyThread::start() {
5529   int    fd;
5530   fd = open ("/dev/mem_notify", O_RDONLY, 0);
5531   if (fd < 0) {
5532       return;
5533   }
5534 
5535   if (memnotify_thread() == NULL) {
5536     new MemNotifyThread(fd);
5537   }
5538 }
5539 #endif // JAVASE_EMBEDDED