1 /*
   2  * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "memory/allocation.inline.hpp"
  36 #include "memory/filemap.hpp"
  37 #include "mutex_linux.inline.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_share_linux.hpp"
  40 #include "prims/jniFastGetField.hpp"
  41 #include "prims/jvm.h"
  42 #include "prims/jvm_misc.hpp"
  43 #include "runtime/arguments.hpp"
  44 #include "runtime/extendedPC.hpp"
  45 #include "runtime/globals.hpp"
  46 #include "runtime/interfaceSupport.hpp"
  47 #include "runtime/init.hpp"
  48 #include "runtime/java.hpp"
  49 #include "runtime/javaCalls.hpp"
  50 #include "runtime/mutexLocker.hpp"
  51 #include "runtime/objectMonitor.hpp"
  52 #include "runtime/osThread.hpp"
  53 #include "runtime/perfMemory.hpp"
  54 #include "runtime/sharedRuntime.hpp"
  55 #include "runtime/statSampler.hpp"
  56 #include "runtime/stubRoutines.hpp"
  57 #include "runtime/thread.inline.hpp"
  58 #include "runtime/threadCritical.hpp"
  59 #include "runtime/timer.hpp"
  60 #include "services/attachListener.hpp"
  61 #include "services/memTracker.hpp"
  62 #include "services/runtimeService.hpp"
  63 #include "utilities/decoder.hpp"
  64 #include "utilities/defaultStream.hpp"
  65 #include "utilities/events.hpp"
  66 #include "utilities/elfFile.hpp"
  67 #include "utilities/growableArray.hpp"
  68 #include "utilities/vmError.hpp"
  69 
  70 // put OS-includes here
  71 # include <sys/types.h>
  72 # include <sys/mman.h>
  73 # include <sys/stat.h>
  74 # include <sys/select.h>
  75 # include <pthread.h>
  76 # include <signal.h>
  77 # include <errno.h>
  78 # include <dlfcn.h>
  79 # include <stdio.h>
  80 # include <unistd.h>
  81 # include <sys/resource.h>
  82 # include <pthread.h>
  83 # include <sys/stat.h>
  84 # include <sys/time.h>
  85 # include <sys/times.h>
  86 # include <sys/utsname.h>
  87 # include <sys/socket.h>
  88 # include <sys/wait.h>
  89 # include <pwd.h>
  90 # include <poll.h>
  91 # include <semaphore.h>
  92 # include <fcntl.h>
  93 # include <string.h>
  94 # include <syscall.h>
  95 # include <sys/sysinfo.h>
  96 # include <gnu/libc-version.h>
  97 # include <sys/ipc.h>
  98 # include <sys/shm.h>
  99 # include <link.h>
 100 # include <stdint.h>
 101 # include <inttypes.h>
 102 # include <sys/ioctl.h>
 103 
 104 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 105 // getrusage() is prepared to handle the associated failure.
 106 #ifndef RUSAGE_THREAD
 107 #define RUSAGE_THREAD   (1)               /* only the calling thread */
 108 #endif
 109 
 110 #define MAX_PATH    (2 * K)
 111 
 112 // for timer info max values which include all bits
 113 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 114 
 115 #define LARGEPAGES_BIT (1 << 6)
 116 ////////////////////////////////////////////////////////////////////////////////
 117 // global variables
 118 julong os::Linux::_physical_memory = 0;
 119 
 120 address   os::Linux::_initial_thread_stack_bottom = NULL;
 121 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 122 
 123 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 124 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 125 Mutex* os::Linux::_createThread_lock = NULL;
 126 pthread_t os::Linux::_main_thread;
 127 int os::Linux::_page_size = -1;
 128 const int os::Linux::_vm_default_page_size = (8 * K);
 129 bool os::Linux::_is_floating_stack = false;
 130 bool os::Linux::_is_NPTL = false;
 131 bool os::Linux::_supports_fast_thread_cpu_time = false;
 132 const char * os::Linux::_glibc_version = NULL;
 133 const char * os::Linux::_libpthread_version = NULL;
 134 
 135 static jlong initial_time_count=0;
 136 
 137 static int clock_tics_per_sec = 100;
 138 
 139 // For diagnostics to print a message once. see run_periodic_checks
 140 static sigset_t check_signal_done;
 141 static bool check_signals = true;;
 142 
 143 static pid_t _initial_pid = 0;
 144 
 145 /* Signal number used to suspend/resume a thread */
 146 
 147 /* do not use any signal number less than SIGSEGV, see 4355769 */
 148 static int SR_signum = SIGUSR2;
 149 sigset_t SR_sigset;
 150 
 151 /* Used to protect dlsym() calls */
 152 static pthread_mutex_t dl_mutex;
 153 
 154 #ifdef JAVASE_EMBEDDED
 155 class MemNotifyThread: public Thread {
 156   friend class VMStructs;
 157  public:
 158   virtual void run();
 159 
 160  private:
 161   static MemNotifyThread* _memnotify_thread;
 162   int _fd;
 163 
 164  public:
 165 
 166   // Constructor
 167   MemNotifyThread(int fd);
 168 
 169   // Tester
 170   bool is_memnotify_thread() const { return true; }
 171 
 172   // Printing
 173   char* name() const { return (char*)"Linux MemNotify Thread"; }
 174 
 175   // Returns the single instance of the MemNotifyThread
 176   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 177 
 178   // Create and start the single instance of MemNotifyThread
 179   static void start();
 180 };
 181 #endif // JAVASE_EMBEDDED
 182 
 183 // utility functions
 184 
 185 static int SR_initialize();
 186 
 187 julong os::available_memory() {
 188   return Linux::available_memory();
 189 }
 190 
 191 julong os::Linux::available_memory() {
 192   // values in struct sysinfo are "unsigned long"
 193   struct sysinfo si;
 194   sysinfo(&si);
 195 
 196   return (julong)si.freeram * si.mem_unit;
 197 }
 198 
 199 julong os::physical_memory() {
 200   return Linux::physical_memory();
 201 }
 202 
 203 ////////////////////////////////////////////////////////////////////////////////
 204 // environment support
 205 
 206 bool os::getenv(const char* name, char* buf, int len) {
 207   const char* val = ::getenv(name);
 208   if (val != NULL && strlen(val) < (size_t)len) {
 209     strcpy(buf, val);
 210     return true;
 211   }
 212   if (len > 0) buf[0] = 0;  // return a null string
 213   return false;
 214 }
 215 
 216 
 217 // Return true if user is running as root.
 218 
 219 bool os::have_special_privileges() {
 220   static bool init = false;
 221   static bool privileges = false;
 222   if (!init) {
 223     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 224     init = true;
 225   }
 226   return privileges;
 227 }
 228 
 229 
 230 #ifndef SYS_gettid
 231 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 232 #ifdef __ia64__
 233 #define SYS_gettid 1105
 234 #elif __i386__
 235 #define SYS_gettid 224
 236 #elif __amd64__
 237 #define SYS_gettid 186
 238 #elif __sparc__
 239 #define SYS_gettid 143
 240 #else
 241 #error define gettid for the arch
 242 #endif
 243 #endif
 244 
 245 // Cpu architecture string
 246 #if   defined(ZERO)
 247 static char cpu_arch[] = ZERO_LIBARCH;
 248 #elif defined(IA64)
 249 static char cpu_arch[] = "ia64";
 250 #elif defined(IA32)
 251 static char cpu_arch[] = "i386";
 252 #elif defined(AMD64)
 253 static char cpu_arch[] = "amd64";
 254 #elif defined(ARM)
 255 static char cpu_arch[] = "arm";
 256 #elif defined(PPC)
 257 static char cpu_arch[] = "ppc";
 258 #elif defined(SPARC)
 259 #  ifdef _LP64
 260 static char cpu_arch[] = "sparcv9";
 261 #  else
 262 static char cpu_arch[] = "sparc";
 263 #  endif
 264 #else
 265 #error Add appropriate cpu_arch setting
 266 #endif
 267 
 268 
 269 // pid_t gettid()
 270 //
 271 // Returns the kernel thread id of the currently running thread. Kernel
 272 // thread id is used to access /proc.
 273 //
 274 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 275 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 276 //
 277 pid_t os::Linux::gettid() {
 278   int rslt = syscall(SYS_gettid);
 279   if (rslt == -1) {
 280      // old kernel, no NPTL support
 281      return getpid();
 282   } else {
 283      return (pid_t)rslt;
 284   }
 285 }
 286 
 287 // Most versions of linux have a bug where the number of processors are
 288 // determined by looking at the /proc file system.  In a chroot environment,
 289 // the system call returns 1.  This causes the VM to act as if it is
 290 // a single processor and elide locking (see is_MP() call).
 291 static bool unsafe_chroot_detected = false;
 292 static const char *unstable_chroot_error = "/proc file system not found.\n"
 293                      "Java may be unstable running multithreaded in a chroot "
 294                      "environment on Linux when /proc filesystem is not mounted.";
 295 
 296 void os::Linux::initialize_system_info() {
 297   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 298   if (processor_count() == 1) {
 299     pid_t pid = os::Linux::gettid();
 300     char fname[32];
 301     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 302     FILE *fp = fopen(fname, "r");
 303     if (fp == NULL) {
 304       unsafe_chroot_detected = true;
 305     } else {
 306       fclose(fp);
 307     }
 308   }
 309   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 310   assert(processor_count() > 0, "linux error");
 311 }
 312 
 313 void os::init_system_properties_values() {
 314 //  char arch[12];
 315 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 316 
 317   // The next steps are taken in the product version:
 318   //
 319   // Obtain the JAVA_HOME value from the location of libjvm.so.
 320   // This library should be located at:
 321   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 322   //
 323   // If "/jre/lib/" appears at the right place in the path, then we
 324   // assume libjvm.so is installed in a JDK and we use this path.
 325   //
 326   // Otherwise exit with message: "Could not create the Java virtual machine."
 327   //
 328   // The following extra steps are taken in the debugging version:
 329   //
 330   // If "/jre/lib/" does NOT appear at the right place in the path
 331   // instead of exit check for $JAVA_HOME environment variable.
 332   //
 333   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 334   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 335   // it looks like libjvm.so is installed there
 336   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 337   //
 338   // Otherwise exit.
 339   //
 340   // Important note: if the location of libjvm.so changes this
 341   // code needs to be changed accordingly.
 342 
 343   // The next few definitions allow the code to be verbatim:
 344 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
 345 #define getenv(n) ::getenv(n)
 346 
 347 /*
 348  * See ld(1):
 349  *      The linker uses the following search paths to locate required
 350  *      shared libraries:
 351  *        1: ...
 352  *        ...
 353  *        7: The default directories, normally /lib and /usr/lib.
 354  */
 355 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 356 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 357 #else
 358 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 359 #endif
 360 
 361 #define EXTENSIONS_DIR  "/lib/ext"
 362 #define ENDORSED_DIR    "/lib/endorsed"
 363 #define REG_DIR         "/usr/java/packages"
 364 
 365   {
 366     /* sysclasspath, java_home, dll_dir */
 367     {
 368         char *home_path;
 369         char *dll_path;
 370         char *pslash;
 371         char buf[MAXPATHLEN];
 372         os::jvm_path(buf, sizeof(buf));
 373 
 374         // Found the full path to libjvm.so.
 375         // Now cut the path to <java_home>/jre if we can.
 376         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 377         pslash = strrchr(buf, '/');
 378         if (pslash != NULL)
 379             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 380         dll_path = malloc(strlen(buf) + 1);
 381         if (dll_path == NULL)
 382             return;
 383         strcpy(dll_path, buf);
 384         Arguments::set_dll_dir(dll_path);
 385 
 386         if (pslash != NULL) {
 387             pslash = strrchr(buf, '/');
 388             if (pslash != NULL) {
 389                 *pslash = '\0';       /* get rid of /<arch> */
 390                 pslash = strrchr(buf, '/');
 391                 if (pslash != NULL)
 392                     *pslash = '\0';   /* get rid of /lib */
 393             }
 394         }
 395 
 396         home_path = malloc(strlen(buf) + 1);
 397         if (home_path == NULL)
 398             return;
 399         strcpy(home_path, buf);
 400         Arguments::set_java_home(home_path);
 401 
 402         if (!set_boot_path('/', ':'))
 403             return;
 404     }
 405 
 406     /*
 407      * Where to look for native libraries
 408      *
 409      * Note: Due to a legacy implementation, most of the library path
 410      * is set in the launcher.  This was to accomodate linking restrictions
 411      * on legacy Linux implementations (which are no longer supported).
 412      * Eventually, all the library path setting will be done here.
 413      *
 414      * However, to prevent the proliferation of improperly built native
 415      * libraries, the new path component /usr/java/packages is added here.
 416      * Eventually, all the library path setting will be done here.
 417      */
 418     {
 419         char *ld_library_path;
 420 
 421         /*
 422          * Construct the invariant part of ld_library_path. Note that the
 423          * space for the colon and the trailing null are provided by the
 424          * nulls included by the sizeof operator (so actually we allocate
 425          * a byte more than necessary).
 426          */
 427         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 428             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 429         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 430 
 431         /*
 432          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 433          * should always exist (until the legacy problem cited above is
 434          * addressed).
 435          */
 436         char *v = getenv("LD_LIBRARY_PATH");
 437         if (v != NULL) {
 438             char *t = ld_library_path;
 439             /* That's +1 for the colon and +1 for the trailing '\0' */
 440             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 441             sprintf(ld_library_path, "%s:%s", v, t);
 442         }
 443         Arguments::set_library_path(ld_library_path);
 444     }
 445 
 446     /*
 447      * Extensions directories.
 448      *
 449      * Note that the space for the colon and the trailing null are provided
 450      * by the nulls included by the sizeof operator (so actually one byte more
 451      * than necessary is allocated).
 452      */
 453     {
 454         char *buf = malloc(strlen(Arguments::get_java_home()) +
 455             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 456         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 457             Arguments::get_java_home());
 458         Arguments::set_ext_dirs(buf);
 459     }
 460 
 461     /* Endorsed standards default directory. */
 462     {
 463         char * buf;
 464         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 465         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 466         Arguments::set_endorsed_dirs(buf);
 467     }
 468   }
 469 
 470 #undef malloc
 471 #undef getenv
 472 #undef EXTENSIONS_DIR
 473 #undef ENDORSED_DIR
 474 
 475   // Done
 476   return;
 477 }
 478 
 479 ////////////////////////////////////////////////////////////////////////////////
 480 // breakpoint support
 481 
 482 void os::breakpoint() {
 483   BREAKPOINT;
 484 }
 485 
 486 extern "C" void breakpoint() {
 487   // use debugger to set breakpoint here
 488 }
 489 
 490 ////////////////////////////////////////////////////////////////////////////////
 491 // signal support
 492 
 493 debug_only(static bool signal_sets_initialized = false);
 494 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 495 
 496 bool os::Linux::is_sig_ignored(int sig) {
 497       struct sigaction oact;
 498       sigaction(sig, (struct sigaction*)NULL, &oact);
 499       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 500                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 501       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 502            return true;
 503       else
 504            return false;
 505 }
 506 
 507 void os::Linux::signal_sets_init() {
 508   // Should also have an assertion stating we are still single-threaded.
 509   assert(!signal_sets_initialized, "Already initialized");
 510   // Fill in signals that are necessarily unblocked for all threads in
 511   // the VM. Currently, we unblock the following signals:
 512   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 513   //                         by -Xrs (=ReduceSignalUsage));
 514   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 515   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 516   // the dispositions or masks wrt these signals.
 517   // Programs embedding the VM that want to use the above signals for their
 518   // own purposes must, at this time, use the "-Xrs" option to prevent
 519   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 520   // (See bug 4345157, and other related bugs).
 521   // In reality, though, unblocking these signals is really a nop, since
 522   // these signals are not blocked by default.
 523   sigemptyset(&unblocked_sigs);
 524   sigemptyset(&allowdebug_blocked_sigs);
 525   sigaddset(&unblocked_sigs, SIGILL);
 526   sigaddset(&unblocked_sigs, SIGSEGV);
 527   sigaddset(&unblocked_sigs, SIGBUS);
 528   sigaddset(&unblocked_sigs, SIGFPE);
 529   sigaddset(&unblocked_sigs, SR_signum);
 530 
 531   if (!ReduceSignalUsage) {
 532    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 533       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 534       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 535    }
 536    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 537       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 538       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 539    }
 540    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 541       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 542       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 543    }
 544   }
 545   // Fill in signals that are blocked by all but the VM thread.
 546   sigemptyset(&vm_sigs);
 547   if (!ReduceSignalUsage)
 548     sigaddset(&vm_sigs, BREAK_SIGNAL);
 549   debug_only(signal_sets_initialized = true);
 550 
 551 }
 552 
 553 // These are signals that are unblocked while a thread is running Java.
 554 // (For some reason, they get blocked by default.)
 555 sigset_t* os::Linux::unblocked_signals() {
 556   assert(signal_sets_initialized, "Not initialized");
 557   return &unblocked_sigs;
 558 }
 559 
 560 // These are the signals that are blocked while a (non-VM) thread is
 561 // running Java. Only the VM thread handles these signals.
 562 sigset_t* os::Linux::vm_signals() {
 563   assert(signal_sets_initialized, "Not initialized");
 564   return &vm_sigs;
 565 }
 566 
 567 // These are signals that are blocked during cond_wait to allow debugger in
 568 sigset_t* os::Linux::allowdebug_blocked_signals() {
 569   assert(signal_sets_initialized, "Not initialized");
 570   return &allowdebug_blocked_sigs;
 571 }
 572 
 573 void os::Linux::hotspot_sigmask(Thread* thread) {
 574 
 575   //Save caller's signal mask before setting VM signal mask
 576   sigset_t caller_sigmask;
 577   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 578 
 579   OSThread* osthread = thread->osthread();
 580   osthread->set_caller_sigmask(caller_sigmask);
 581 
 582   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 583 
 584   if (!ReduceSignalUsage) {
 585     if (thread->is_VM_thread()) {
 586       // Only the VM thread handles BREAK_SIGNAL ...
 587       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 588     } else {
 589       // ... all other threads block BREAK_SIGNAL
 590       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 591     }
 592   }
 593 }
 594 
 595 //////////////////////////////////////////////////////////////////////////////
 596 // detecting pthread library
 597 
 598 void os::Linux::libpthread_init() {
 599   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 600   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 601   // generic name for earlier versions.
 602   // Define macros here so we can build HotSpot on old systems.
 603 # ifndef _CS_GNU_LIBC_VERSION
 604 # define _CS_GNU_LIBC_VERSION 2
 605 # endif
 606 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 607 # define _CS_GNU_LIBPTHREAD_VERSION 3
 608 # endif
 609 
 610   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 611   if (n > 0) {
 612      char *str = (char *)malloc(n, mtInternal);
 613      confstr(_CS_GNU_LIBC_VERSION, str, n);
 614      os::Linux::set_glibc_version(str);
 615   } else {
 616      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 617      static char _gnu_libc_version[32];
 618      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 619               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 620      os::Linux::set_glibc_version(_gnu_libc_version);
 621   }
 622 
 623   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 624   if (n > 0) {
 625      char *str = (char *)malloc(n, mtInternal);
 626      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 627      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 628      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 629      // is the case. LinuxThreads has a hard limit on max number of threads.
 630      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 631      // On the other hand, NPTL does not have such a limit, sysconf()
 632      // will return -1 and errno is not changed. Check if it is really NPTL.
 633      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 634          strstr(str, "NPTL") &&
 635          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 636        free(str);
 637        os::Linux::set_libpthread_version("linuxthreads");
 638      } else {
 639        os::Linux::set_libpthread_version(str);
 640      }
 641   } else {
 642     // glibc before 2.3.2 only has LinuxThreads.
 643     os::Linux::set_libpthread_version("linuxthreads");
 644   }
 645 
 646   if (strstr(libpthread_version(), "NPTL")) {
 647      os::Linux::set_is_NPTL();
 648   } else {
 649      os::Linux::set_is_LinuxThreads();
 650   }
 651 
 652   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 653   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 654   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 655      os::Linux::set_is_floating_stack();
 656   }
 657 }
 658 
 659 /////////////////////////////////////////////////////////////////////////////
 660 // thread stack
 661 
 662 // Force Linux kernel to expand current thread stack. If "bottom" is close
 663 // to the stack guard, caller should block all signals.
 664 //
 665 // MAP_GROWSDOWN:
 666 //   A special mmap() flag that is used to implement thread stacks. It tells
 667 //   kernel that the memory region should extend downwards when needed. This
 668 //   allows early versions of LinuxThreads to only mmap the first few pages
 669 //   when creating a new thread. Linux kernel will automatically expand thread
 670 //   stack as needed (on page faults).
 671 //
 672 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 673 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 674 //   region, it's hard to tell if the fault is due to a legitimate stack
 675 //   access or because of reading/writing non-exist memory (e.g. buffer
 676 //   overrun). As a rule, if the fault happens below current stack pointer,
 677 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 678 //   application (see Linux kernel fault.c).
 679 //
 680 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 681 //   stack overflow detection.
 682 //
 683 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 684 //   not use this flag. However, the stack of initial thread is not created
 685 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 686 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 687 //   and then attach the thread to JVM.
 688 //
 689 // To get around the problem and allow stack banging on Linux, we need to
 690 // manually expand thread stack after receiving the SIGSEGV.
 691 //
 692 // There are two ways to expand thread stack to address "bottom", we used
 693 // both of them in JVM before 1.5:
 694 //   1. adjust stack pointer first so that it is below "bottom", and then
 695 //      touch "bottom"
 696 //   2. mmap() the page in question
 697 //
 698 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 699 // if current sp is already near the lower end of page 101, and we need to
 700 // call mmap() to map page 100, it is possible that part of the mmap() frame
 701 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 702 // That will destroy the mmap() frame and cause VM to crash.
 703 //
 704 // The following code works by adjusting sp first, then accessing the "bottom"
 705 // page to force a page fault. Linux kernel will then automatically expand the
 706 // stack mapping.
 707 //
 708 // _expand_stack_to() assumes its frame size is less than page size, which
 709 // should always be true if the function is not inlined.
 710 
 711 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 712 #define NOINLINE
 713 #else
 714 #define NOINLINE __attribute__ ((noinline))
 715 #endif
 716 
 717 static void _expand_stack_to(address bottom) NOINLINE;
 718 
 719 static void _expand_stack_to(address bottom) {
 720   address sp;
 721   size_t size;
 722   volatile char *p;
 723 
 724   // Adjust bottom to point to the largest address within the same page, it
 725   // gives us a one-page buffer if alloca() allocates slightly more memory.
 726   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 727   bottom += os::Linux::page_size() - 1;
 728 
 729   // sp might be slightly above current stack pointer; if that's the case, we
 730   // will alloca() a little more space than necessary, which is OK. Don't use
 731   // os::current_stack_pointer(), as its result can be slightly below current
 732   // stack pointer, causing us to not alloca enough to reach "bottom".
 733   sp = (address)&sp;
 734 
 735   if (sp > bottom) {
 736     size = sp - bottom;
 737     p = (volatile char *)alloca(size);
 738     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 739     p[0] = '\0';
 740   }
 741 }
 742 
 743 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 744   assert(t!=NULL, "just checking");
 745   assert(t->osthread()->expanding_stack(), "expand should be set");
 746   assert(t->stack_base() != NULL, "stack_base was not initialized");
 747 
 748   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 749     sigset_t mask_all, old_sigset;
 750     sigfillset(&mask_all);
 751     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 752     _expand_stack_to(addr);
 753     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 754     return true;
 755   }
 756   return false;
 757 }
 758 
 759 //////////////////////////////////////////////////////////////////////////////
 760 // create new thread
 761 
 762 static address highest_vm_reserved_address();
 763 
 764 // check if it's safe to start a new thread
 765 static bool _thread_safety_check(Thread* thread) {
 766   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 767     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 768     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 769     //   allocated (MAP_FIXED) from high address space. Every thread stack
 770     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 771     //   it to other values if they rebuild LinuxThreads).
 772     //
 773     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 774     // the memory region has already been mmap'ed. That means if we have too
 775     // many threads and/or very large heap, eventually thread stack will
 776     // collide with heap.
 777     //
 778     // Here we try to prevent heap/stack collision by comparing current
 779     // stack bottom with the highest address that has been mmap'ed by JVM
 780     // plus a safety margin for memory maps created by native code.
 781     //
 782     // This feature can be disabled by setting ThreadSafetyMargin to 0
 783     //
 784     if (ThreadSafetyMargin > 0) {
 785       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 786 
 787       // not safe if our stack extends below the safety margin
 788       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 789     } else {
 790       return true;
 791     }
 792   } else {
 793     // Floating stack LinuxThreads or NPTL:
 794     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 795     //   there's not enough space left, pthread_create() will fail. If we come
 796     //   here, that means enough space has been reserved for stack.
 797     return true;
 798   }
 799 }
 800 
 801 // Thread start routine for all newly created threads
 802 static void *java_start(Thread *thread) {
 803   // Try to randomize the cache line index of hot stack frames.
 804   // This helps when threads of the same stack traces evict each other's
 805   // cache lines. The threads can be either from the same JVM instance, or
 806   // from different JVM instances. The benefit is especially true for
 807   // processors with hyperthreading technology.
 808   static int counter = 0;
 809   int pid = os::current_process_id();
 810   alloca(((pid ^ counter++) & 7) * 128);
 811 
 812   ThreadLocalStorage::set_thread(thread);
 813 
 814   OSThread* osthread = thread->osthread();
 815   Monitor* sync = osthread->startThread_lock();
 816 
 817   // non floating stack LinuxThreads needs extra check, see above
 818   if (!_thread_safety_check(thread)) {
 819     // notify parent thread
 820     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 821     osthread->set_state(ZOMBIE);
 822     sync->notify_all();
 823     return NULL;
 824   }
 825 
 826   // thread_id is kernel thread id (similar to Solaris LWP id)
 827   osthread->set_thread_id(os::Linux::gettid());
 828 
 829   if (UseNUMA) {
 830     int lgrp_id = os::numa_get_group_id();
 831     if (lgrp_id != -1) {
 832       thread->set_lgrp_id(lgrp_id);
 833     }
 834   }
 835   // initialize signal mask for this thread
 836   os::Linux::hotspot_sigmask(thread);
 837 
 838   // initialize floating point control register
 839   os::Linux::init_thread_fpu_state();
 840 
 841   // handshaking with parent thread
 842   {
 843     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 844 
 845     // notify parent thread
 846     osthread->set_state(INITIALIZED);
 847     sync->notify_all();
 848 
 849     // wait until os::start_thread()
 850     while (osthread->get_state() == INITIALIZED) {
 851       sync->wait(Mutex::_no_safepoint_check_flag);
 852     }
 853   }
 854 
 855   // call one more level start routine
 856   thread->run();
 857 
 858   return 0;
 859 }
 860 
 861 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 862   assert(thread->osthread() == NULL, "caller responsible");
 863 
 864   // Allocate the OSThread object
 865   OSThread* osthread = new OSThread(NULL, NULL);
 866   if (osthread == NULL) {
 867     return false;
 868   }
 869 
 870   // set the correct thread state
 871   osthread->set_thread_type(thr_type);
 872 
 873   // Initial state is ALLOCATED but not INITIALIZED
 874   osthread->set_state(ALLOCATED);
 875 
 876   thread->set_osthread(osthread);
 877 
 878   // init thread attributes
 879   pthread_attr_t attr;
 880   pthread_attr_init(&attr);
 881   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 882 
 883   // stack size
 884   if (os::Linux::supports_variable_stack_size()) {
 885     // calculate stack size if it's not specified by caller
 886     if (stack_size == 0) {
 887       stack_size = os::Linux::default_stack_size(thr_type);
 888 
 889       switch (thr_type) {
 890       case os::java_thread:
 891         // Java threads use ThreadStackSize which default value can be
 892         // changed with the flag -Xss
 893         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 894         stack_size = JavaThread::stack_size_at_create();
 895         break;
 896       case os::compiler_thread:
 897         if (CompilerThreadStackSize > 0) {
 898           stack_size = (size_t)(CompilerThreadStackSize * K);
 899           break;
 900         } // else fall through:
 901           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 902       case os::vm_thread:
 903       case os::pgc_thread:
 904       case os::cgc_thread:
 905       case os::watcher_thread:
 906         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 907         break;
 908       }
 909     }
 910 
 911     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 912     pthread_attr_setstacksize(&attr, stack_size);
 913   } else {
 914     // let pthread_create() pick the default value.
 915   }
 916 
 917   // glibc guard page
 918   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 919 
 920   ThreadState state;
 921 
 922   {
 923     // Serialize thread creation if we are running with fixed stack LinuxThreads
 924     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 925     if (lock) {
 926       os::Linux::createThread_lock()->lock_without_safepoint_check();
 927     }
 928 
 929     pthread_t tid;
 930     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 931 
 932     pthread_attr_destroy(&attr);
 933 
 934     if (ret != 0) {
 935       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 936         perror("pthread_create()");
 937       }
 938       // Need to clean up stuff we've allocated so far
 939       thread->set_osthread(NULL);
 940       delete osthread;
 941       if (lock) os::Linux::createThread_lock()->unlock();
 942       return false;
 943     }
 944 
 945     // Store pthread info into the OSThread
 946     osthread->set_pthread_id(tid);
 947 
 948     // Wait until child thread is either initialized or aborted
 949     {
 950       Monitor* sync_with_child = osthread->startThread_lock();
 951       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 952       while ((state = osthread->get_state()) == ALLOCATED) {
 953         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 954       }
 955     }
 956 
 957     if (lock) {
 958       os::Linux::createThread_lock()->unlock();
 959     }
 960   }
 961 
 962   // Aborted due to thread limit being reached
 963   if (state == ZOMBIE) {
 964       thread->set_osthread(NULL);
 965       delete osthread;
 966       return false;
 967   }
 968 
 969   // The thread is returned suspended (in state INITIALIZED),
 970   // and is started higher up in the call chain
 971   assert(state == INITIALIZED, "race condition");
 972   return true;
 973 }
 974 
 975 /////////////////////////////////////////////////////////////////////////////
 976 // attach existing thread
 977 
 978 // bootstrap the main thread
 979 bool os::create_main_thread(JavaThread* thread) {
 980   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 981   return create_attached_thread(thread);
 982 }
 983 
 984 bool os::create_attached_thread(JavaThread* thread) {
 985 #ifdef ASSERT
 986     thread->verify_not_published();
 987 #endif
 988 
 989   // Allocate the OSThread object
 990   OSThread* osthread = new OSThread(NULL, NULL);
 991 
 992   if (osthread == NULL) {
 993     return false;
 994   }
 995 
 996   // Store pthread info into the OSThread
 997   osthread->set_thread_id(os::Linux::gettid());
 998   osthread->set_pthread_id(::pthread_self());
 999 
1000   // initialize floating point control register
1001   os::Linux::init_thread_fpu_state();
1002 
1003   // Initial thread state is RUNNABLE
1004   osthread->set_state(RUNNABLE);
1005 
1006   thread->set_osthread(osthread);
1007 
1008   if (UseNUMA) {
1009     int lgrp_id = os::numa_get_group_id();
1010     if (lgrp_id != -1) {
1011       thread->set_lgrp_id(lgrp_id);
1012     }
1013   }
1014 
1015   if (os::Linux::is_initial_thread()) {
1016     // If current thread is initial thread, its stack is mapped on demand,
1017     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1018     // the entire stack region to avoid SEGV in stack banging.
1019     // It is also useful to get around the heap-stack-gap problem on SuSE
1020     // kernel (see 4821821 for details). We first expand stack to the top
1021     // of yellow zone, then enable stack yellow zone (order is significant,
1022     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1023     // is no gap between the last two virtual memory regions.
1024 
1025     JavaThread *jt = (JavaThread *)thread;
1026     address addr = jt->stack_yellow_zone_base();
1027     assert(addr != NULL, "initialization problem?");
1028     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1029 
1030     osthread->set_expanding_stack();
1031     os::Linux::manually_expand_stack(jt, addr);
1032     osthread->clear_expanding_stack();
1033   }
1034 
1035   // initialize signal mask for this thread
1036   // and save the caller's signal mask
1037   os::Linux::hotspot_sigmask(thread);
1038 
1039   return true;
1040 }
1041 
1042 void os::pd_start_thread(Thread* thread) {
1043   OSThread * osthread = thread->osthread();
1044   assert(osthread->get_state() != INITIALIZED, "just checking");
1045   Monitor* sync_with_child = osthread->startThread_lock();
1046   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1047   sync_with_child->notify();
1048 }
1049 
1050 // Free Linux resources related to the OSThread
1051 void os::free_thread(OSThread* osthread) {
1052   assert(osthread != NULL, "osthread not set");
1053 
1054   if (Thread::current()->osthread() == osthread) {
1055     // Restore caller's signal mask
1056     sigset_t sigmask = osthread->caller_sigmask();
1057     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1058    }
1059 
1060   delete osthread;
1061 }
1062 
1063 //////////////////////////////////////////////////////////////////////////////
1064 // thread local storage
1065 
1066 int os::allocate_thread_local_storage() {
1067   pthread_key_t key;
1068   int rslt = pthread_key_create(&key, NULL);
1069   assert(rslt == 0, "cannot allocate thread local storage");
1070   return (int)key;
1071 }
1072 
1073 // Note: This is currently not used by VM, as we don't destroy TLS key
1074 // on VM exit.
1075 void os::free_thread_local_storage(int index) {
1076   int rslt = pthread_key_delete((pthread_key_t)index);
1077   assert(rslt == 0, "invalid index");
1078 }
1079 
1080 void os::thread_local_storage_at_put(int index, void* value) {
1081   int rslt = pthread_setspecific((pthread_key_t)index, value);
1082   assert(rslt == 0, "pthread_setspecific failed");
1083 }
1084 
1085 extern "C" Thread* get_thread() {
1086   return ThreadLocalStorage::thread();
1087 }
1088 
1089 //////////////////////////////////////////////////////////////////////////////
1090 // initial thread
1091 
1092 // Check if current thread is the initial thread, similar to Solaris thr_main.
1093 bool os::Linux::is_initial_thread(void) {
1094   char dummy;
1095   // If called before init complete, thread stack bottom will be null.
1096   // Can be called if fatal error occurs before initialization.
1097   if (initial_thread_stack_bottom() == NULL) return false;
1098   assert(initial_thread_stack_bottom() != NULL &&
1099          initial_thread_stack_size()   != 0,
1100          "os::init did not locate initial thread's stack region");
1101   if ((address)&dummy >= initial_thread_stack_bottom() &&
1102       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1103        return true;
1104   else return false;
1105 }
1106 
1107 // Find the virtual memory area that contains addr
1108 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1109   FILE *fp = fopen("/proc/self/maps", "r");
1110   if (fp) {
1111     address low, high;
1112     while (!feof(fp)) {
1113       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1114         if (low <= addr && addr < high) {
1115            if (vma_low)  *vma_low  = low;
1116            if (vma_high) *vma_high = high;
1117            fclose (fp);
1118            return true;
1119         }
1120       }
1121       for (;;) {
1122         int ch = fgetc(fp);
1123         if (ch == EOF || ch == (int)'\n') break;
1124       }
1125     }
1126     fclose(fp);
1127   }
1128   return false;
1129 }
1130 
1131 // Locate initial thread stack. This special handling of initial thread stack
1132 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1133 // bogus value for initial thread.
1134 void os::Linux::capture_initial_stack(size_t max_size) {
1135   // stack size is the easy part, get it from RLIMIT_STACK
1136   size_t stack_size;
1137   struct rlimit rlim;
1138   getrlimit(RLIMIT_STACK, &rlim);
1139   stack_size = rlim.rlim_cur;
1140 
1141   // 6308388: a bug in ld.so will relocate its own .data section to the
1142   //   lower end of primordial stack; reduce ulimit -s value a little bit
1143   //   so we won't install guard page on ld.so's data section.
1144   stack_size -= 2 * page_size();
1145 
1146   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1147   //   7.1, in both cases we will get 2G in return value.
1148   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1149   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1150   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1151   //   in case other parts in glibc still assumes 2M max stack size.
1152   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1153   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1154   if (stack_size > 2 * K * K IA64_ONLY(*2))
1155       stack_size = 2 * K * K IA64_ONLY(*2);
1156   // Try to figure out where the stack base (top) is. This is harder.
1157   //
1158   // When an application is started, glibc saves the initial stack pointer in
1159   // a global variable "__libc_stack_end", which is then used by system
1160   // libraries. __libc_stack_end should be pretty close to stack top. The
1161   // variable is available since the very early days. However, because it is
1162   // a private interface, it could disappear in the future.
1163   //
1164   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1165   // to __libc_stack_end, it is very close to stack top, but isn't the real
1166   // stack top. Note that /proc may not exist if VM is running as a chroot
1167   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1168   // /proc/<pid>/stat could change in the future (though unlikely).
1169   //
1170   // We try __libc_stack_end first. If that doesn't work, look for
1171   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1172   // as a hint, which should work well in most cases.
1173 
1174   uintptr_t stack_start;
1175 
1176   // try __libc_stack_end first
1177   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1178   if (p && *p) {
1179     stack_start = *p;
1180   } else {
1181     // see if we can get the start_stack field from /proc/self/stat
1182     FILE *fp;
1183     int pid;
1184     char state;
1185     int ppid;
1186     int pgrp;
1187     int session;
1188     int nr;
1189     int tpgrp;
1190     unsigned long flags;
1191     unsigned long minflt;
1192     unsigned long cminflt;
1193     unsigned long majflt;
1194     unsigned long cmajflt;
1195     unsigned long utime;
1196     unsigned long stime;
1197     long cutime;
1198     long cstime;
1199     long prio;
1200     long nice;
1201     long junk;
1202     long it_real;
1203     uintptr_t start;
1204     uintptr_t vsize;
1205     intptr_t rss;
1206     uintptr_t rsslim;
1207     uintptr_t scodes;
1208     uintptr_t ecode;
1209     int i;
1210 
1211     // Figure what the primordial thread stack base is. Code is inspired
1212     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1213     // followed by command name surrounded by parentheses, state, etc.
1214     char stat[2048];
1215     int statlen;
1216 
1217     fp = fopen("/proc/self/stat", "r");
1218     if (fp) {
1219       statlen = fread(stat, 1, 2047, fp);
1220       stat[statlen] = '\0';
1221       fclose(fp);
1222 
1223       // Skip pid and the command string. Note that we could be dealing with
1224       // weird command names, e.g. user could decide to rename java launcher
1225       // to "java 1.4.2 :)", then the stat file would look like
1226       //                1234 (java 1.4.2 :)) R ... ...
1227       // We don't really need to know the command string, just find the last
1228       // occurrence of ")" and then start parsing from there. See bug 4726580.
1229       char * s = strrchr(stat, ')');
1230 
1231       i = 0;
1232       if (s) {
1233         // Skip blank chars
1234         do s++; while (isspace(*s));
1235 
1236 #define _UFM UINTX_FORMAT
1237 #define _DFM INTX_FORMAT
1238 
1239         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1240         /*              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 */
1241         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,
1242              &state,          /* 3  %c  */
1243              &ppid,           /* 4  %d  */
1244              &pgrp,           /* 5  %d  */
1245              &session,        /* 6  %d  */
1246              &nr,             /* 7  %d  */
1247              &tpgrp,          /* 8  %d  */
1248              &flags,          /* 9  %lu  */
1249              &minflt,         /* 10 %lu  */
1250              &cminflt,        /* 11 %lu  */
1251              &majflt,         /* 12 %lu  */
1252              &cmajflt,        /* 13 %lu  */
1253              &utime,          /* 14 %lu  */
1254              &stime,          /* 15 %lu  */
1255              &cutime,         /* 16 %ld  */
1256              &cstime,         /* 17 %ld  */
1257              &prio,           /* 18 %ld  */
1258              &nice,           /* 19 %ld  */
1259              &junk,           /* 20 %ld  */
1260              &it_real,        /* 21 %ld  */
1261              &start,          /* 22 UINTX_FORMAT */
1262              &vsize,          /* 23 UINTX_FORMAT */
1263              &rss,            /* 24 INTX_FORMAT  */
1264              &rsslim,         /* 25 UINTX_FORMAT */
1265              &scodes,         /* 26 UINTX_FORMAT */
1266              &ecode,          /* 27 UINTX_FORMAT */
1267              &stack_start);   /* 28 UINTX_FORMAT */
1268       }
1269 
1270 #undef _UFM
1271 #undef _DFM
1272 
1273       if (i != 28 - 2) {
1274          assert(false, "Bad conversion from /proc/self/stat");
1275          // product mode - assume we are the initial thread, good luck in the
1276          // embedded case.
1277          warning("Can't detect initial thread stack location - bad conversion");
1278          stack_start = (uintptr_t) &rlim;
1279       }
1280     } else {
1281       // For some reason we can't open /proc/self/stat (for example, running on
1282       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1283       // most cases, so don't abort:
1284       warning("Can't detect initial thread stack location - no /proc/self/stat");
1285       stack_start = (uintptr_t) &rlim;
1286     }
1287   }
1288 
1289   // Now we have a pointer (stack_start) very close to the stack top, the
1290   // next thing to do is to figure out the exact location of stack top. We
1291   // can find out the virtual memory area that contains stack_start by
1292   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1293   // and its upper limit is the real stack top. (again, this would fail if
1294   // running inside chroot, because /proc may not exist.)
1295 
1296   uintptr_t stack_top;
1297   address low, high;
1298   if (find_vma((address)stack_start, &low, &high)) {
1299     // success, "high" is the true stack top. (ignore "low", because initial
1300     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1301     stack_top = (uintptr_t)high;
1302   } else {
1303     // failed, likely because /proc/self/maps does not exist
1304     warning("Can't detect initial thread stack location - find_vma failed");
1305     // best effort: stack_start is normally within a few pages below the real
1306     // stack top, use it as stack top, and reduce stack size so we won't put
1307     // guard page outside stack.
1308     stack_top = stack_start;
1309     stack_size -= 16 * page_size();
1310   }
1311 
1312   // stack_top could be partially down the page so align it
1313   stack_top = align_size_up(stack_top, page_size());
1314 
1315   if (max_size && stack_size > max_size) {
1316      _initial_thread_stack_size = max_size;
1317   } else {
1318      _initial_thread_stack_size = stack_size;
1319   }
1320 
1321   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1322   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1323 }
1324 
1325 ////////////////////////////////////////////////////////////////////////////////
1326 // time support
1327 
1328 // Time since start-up in seconds to a fine granularity.
1329 // Used by VMSelfDestructTimer and the MemProfiler.
1330 double os::elapsedTime() {
1331 
1332   return (double)(os::elapsed_counter()) * 0.000001;
1333 }
1334 
1335 jlong os::elapsed_counter() {
1336   timeval time;
1337   int status = gettimeofday(&time, NULL);
1338   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1339 }
1340 
1341 jlong os::elapsed_frequency() {
1342   return (1000 * 1000);
1343 }
1344 
1345 bool os::supports_vtime() { return true; }
1346 bool os::enable_vtime()   { return false; }
1347 bool os::vtime_enabled()  { return false; }
1348 
1349 double os::elapsedVTime() {
1350   struct rusage usage;
1351   int retval = getrusage(RUSAGE_THREAD, &usage);
1352   if (retval == 0) {
1353     return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1354   } else {
1355     // better than nothing, but not much
1356     return elapsedTime();
1357   }
1358 }
1359 
1360 jlong os::javaTimeMillis() {
1361   timeval time;
1362   int status = gettimeofday(&time, NULL);
1363   assert(status != -1, "linux error");
1364   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1365 }
1366 
1367 #ifndef CLOCK_MONOTONIC
1368 #define CLOCK_MONOTONIC (1)
1369 #endif
1370 
1371 void os::Linux::clock_init() {
1372   // we do dlopen's in this particular order due to bug in linux
1373   // dynamical loader (see 6348968) leading to crash on exit
1374   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1375   if (handle == NULL) {
1376     handle = dlopen("librt.so", RTLD_LAZY);
1377   }
1378 
1379   if (handle) {
1380     int (*clock_getres_func)(clockid_t, struct timespec*) =
1381            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1382     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1383            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1384     if (clock_getres_func && clock_gettime_func) {
1385       // See if monotonic clock is supported by the kernel. Note that some
1386       // early implementations simply return kernel jiffies (updated every
1387       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1388       // for nano time (though the monotonic property is still nice to have).
1389       // It's fixed in newer kernels, however clock_getres() still returns
1390       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1391       // resolution for now. Hopefully as people move to new kernels, this
1392       // won't be a problem.
1393       struct timespec res;
1394       struct timespec tp;
1395       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1396           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1397         // yes, monotonic clock is supported
1398         _clock_gettime = clock_gettime_func;
1399       } else {
1400         // close librt if there is no monotonic clock
1401         dlclose(handle);
1402       }
1403     }
1404   }
1405 }
1406 
1407 #ifndef SYS_clock_getres
1408 
1409 #if defined(IA32) || defined(AMD64)
1410 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1411 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1412 #else
1413 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1414 #define sys_clock_getres(x,y)  -1
1415 #endif
1416 
1417 #else
1418 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1419 #endif
1420 
1421 void os::Linux::fast_thread_clock_init() {
1422   if (!UseLinuxPosixThreadCPUClocks) {
1423     return;
1424   }
1425   clockid_t clockid;
1426   struct timespec tp;
1427   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1428       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1429 
1430   // Switch to using fast clocks for thread cpu time if
1431   // the sys_clock_getres() returns 0 error code.
1432   // Note, that some kernels may support the current thread
1433   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1434   // returned by the pthread_getcpuclockid().
1435   // If the fast Posix clocks are supported then the sys_clock_getres()
1436   // must return at least tp.tv_sec == 0 which means a resolution
1437   // better than 1 sec. This is extra check for reliability.
1438 
1439   if(pthread_getcpuclockid_func &&
1440      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1441      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1442 
1443     _supports_fast_thread_cpu_time = true;
1444     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1445   }
1446 }
1447 
1448 jlong os::javaTimeNanos() {
1449   if (Linux::supports_monotonic_clock()) {
1450     struct timespec tp;
1451     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1452     assert(status == 0, "gettime error");
1453     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1454     return result;
1455   } else {
1456     timeval time;
1457     int status = gettimeofday(&time, NULL);
1458     assert(status != -1, "linux error");
1459     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1460     return 1000 * usecs;
1461   }
1462 }
1463 
1464 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1465   if (Linux::supports_monotonic_clock()) {
1466     info_ptr->max_value = ALL_64_BITS;
1467 
1468     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1469     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1470     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1471   } else {
1472     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1473     info_ptr->max_value = ALL_64_BITS;
1474 
1475     // gettimeofday is a real time clock so it skips
1476     info_ptr->may_skip_backward = true;
1477     info_ptr->may_skip_forward = true;
1478   }
1479 
1480   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1481 }
1482 
1483 // Return the real, user, and system times in seconds from an
1484 // arbitrary fixed point in the past.
1485 bool os::getTimesSecs(double* process_real_time,
1486                       double* process_user_time,
1487                       double* process_system_time) {
1488   struct tms ticks;
1489   clock_t real_ticks = times(&ticks);
1490 
1491   if (real_ticks == (clock_t) (-1)) {
1492     return false;
1493   } else {
1494     double ticks_per_second = (double) clock_tics_per_sec;
1495     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1496     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1497     *process_real_time = ((double) real_ticks) / ticks_per_second;
1498 
1499     return true;
1500   }
1501 }
1502 
1503 
1504 char * os::local_time_string(char *buf, size_t buflen) {
1505   struct tm t;
1506   time_t long_time;
1507   time(&long_time);
1508   localtime_r(&long_time, &t);
1509   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1510                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1511                t.tm_hour, t.tm_min, t.tm_sec);
1512   return buf;
1513 }
1514 
1515 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1516   return localtime_r(clock, res);
1517 }
1518 
1519 ////////////////////////////////////////////////////////////////////////////////
1520 // runtime exit support
1521 
1522 // Note: os::shutdown() might be called very early during initialization, or
1523 // called from signal handler. Before adding something to os::shutdown(), make
1524 // sure it is async-safe and can handle partially initialized VM.
1525 void os::shutdown() {
1526 
1527   // allow PerfMemory to attempt cleanup of any persistent resources
1528   perfMemory_exit();
1529 
1530   // needs to remove object in file system
1531   AttachListener::abort();
1532 
1533   // flush buffered output, finish log files
1534   ostream_abort();
1535 
1536   // Check for abort hook
1537   abort_hook_t abort_hook = Arguments::abort_hook();
1538   if (abort_hook != NULL) {
1539     abort_hook();
1540   }
1541 
1542 }
1543 
1544 // Note: os::abort() might be called very early during initialization, or
1545 // called from signal handler. Before adding something to os::abort(), make
1546 // sure it is async-safe and can handle partially initialized VM.
1547 void os::abort(bool dump_core) {
1548   os::shutdown();
1549   if (dump_core) {
1550 #ifndef PRODUCT
1551     fdStream out(defaultStream::output_fd());
1552     out.print_raw("Current thread is ");
1553     char buf[16];
1554     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1555     out.print_raw_cr(buf);
1556     out.print_raw_cr("Dumping core ...");
1557 #endif
1558     ::abort(); // dump core
1559   }
1560 
1561   ::exit(1);
1562 }
1563 
1564 // Die immediately, no exit hook, no abort hook, no cleanup.
1565 void os::die() {
1566   // _exit() on LinuxThreads only kills current thread
1567   ::abort();
1568 }
1569 
1570 // unused on linux for now.
1571 void os::set_error_file(const char *logfile) {}
1572 
1573 
1574 // This method is a copy of JDK's sysGetLastErrorString
1575 // from src/solaris/hpi/src/system_md.c
1576 
1577 size_t os::lasterror(char *buf, size_t len) {
1578 
1579   if (errno == 0)  return 0;
1580 
1581   const char *s = ::strerror(errno);
1582   size_t n = ::strlen(s);
1583   if (n >= len) {
1584     n = len - 1;
1585   }
1586   ::strncpy(buf, s, n);
1587   buf[n] = '\0';
1588   return n;
1589 }
1590 
1591 intx os::current_thread_id() { return (intx)pthread_self(); }
1592 int os::current_process_id() {
1593 
1594   // Under the old linux thread library, linux gives each thread
1595   // its own process id. Because of this each thread will return
1596   // a different pid if this method were to return the result
1597   // of getpid(2). Linux provides no api that returns the pid
1598   // of the launcher thread for the vm. This implementation
1599   // returns a unique pid, the pid of the launcher thread
1600   // that starts the vm 'process'.
1601 
1602   // Under the NPTL, getpid() returns the same pid as the
1603   // launcher thread rather than a unique pid per thread.
1604   // Use gettid() if you want the old pre NPTL behaviour.
1605 
1606   // if you are looking for the result of a call to getpid() that
1607   // returns a unique pid for the calling thread, then look at the
1608   // OSThread::thread_id() method in osThread_linux.hpp file
1609 
1610   return (int)(_initial_pid ? _initial_pid : getpid());
1611 }
1612 
1613 // DLL functions
1614 
1615 const char* os::dll_file_extension() { return ".so"; }
1616 
1617 // This must be hard coded because it's the system's temporary
1618 // directory not the java application's temp directory, ala java.io.tmpdir.
1619 const char* os::get_temp_directory() { return "/tmp"; }
1620 
1621 static bool file_exists(const char* filename) {
1622   struct stat statbuf;
1623   if (filename == NULL || strlen(filename) == 0) {
1624     return false;
1625   }
1626   return os::stat(filename, &statbuf) == 0;
1627 }
1628 
1629 bool os::dll_build_name(char* buffer, size_t buflen,
1630                         const char* pname, const char* fname) {
1631   bool retval = false;
1632   // Copied from libhpi
1633   const size_t pnamelen = pname ? strlen(pname) : 0;
1634 
1635   // Return error on buffer overflow.
1636   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1637     return retval;
1638   }
1639 
1640   if (pnamelen == 0) {
1641     snprintf(buffer, buflen, "lib%s.so", fname);
1642     retval = true;
1643   } else if (strchr(pname, *os::path_separator()) != NULL) {
1644     int n;
1645     char** pelements = split_path(pname, &n);
1646     if (pelements == NULL) {
1647       return false;
1648     }
1649     for (int i = 0 ; i < n ; i++) {
1650       // Really shouldn't be NULL, but check can't hurt
1651       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1652         continue; // skip the empty path values
1653       }
1654       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1655       if (file_exists(buffer)) {
1656         retval = true;
1657         break;
1658       }
1659     }
1660     // release the storage
1661     for (int i = 0 ; i < n ; i++) {
1662       if (pelements[i] != NULL) {
1663         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1664       }
1665     }
1666     if (pelements != NULL) {
1667       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1668     }
1669   } else {
1670     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1671     retval = true;
1672   }
1673   return retval;
1674 }
1675 
1676 // check if addr is inside libjvm.so
1677 bool os::address_is_in_vm(address addr) {
1678   static address libjvm_base_addr;
1679   Dl_info dlinfo;
1680 
1681   if (libjvm_base_addr == NULL) {
1682     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1683     libjvm_base_addr = (address)dlinfo.dli_fbase;
1684     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1685   }
1686 
1687   if (dladdr((void *)addr, &dlinfo)) {
1688     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1689   }
1690 
1691   return false;
1692 }
1693 
1694 bool os::dll_address_to_function_name(address addr, char *buf,
1695                                       int buflen, int *offset) {
1696   Dl_info dlinfo;
1697 
1698   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1699     if (buf != NULL) {
1700       if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1701         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1702       }
1703     }
1704     if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1705     return true;
1706   } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1707     if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1708         buf, buflen, offset, dlinfo.dli_fname)) {
1709        return true;
1710     }
1711   }
1712 
1713   if (buf != NULL) buf[0] = '\0';
1714   if (offset != NULL) *offset = -1;
1715   return false;
1716 }
1717 
1718 struct _address_to_library_name {
1719   address addr;          // input : memory address
1720   size_t  buflen;        //         size of fname
1721   char*   fname;         // output: library name
1722   address base;          //         library base addr
1723 };
1724 
1725 static int address_to_library_name_callback(struct dl_phdr_info *info,
1726                                             size_t size, void *data) {
1727   int i;
1728   bool found = false;
1729   address libbase = NULL;
1730   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1731 
1732   // iterate through all loadable segments
1733   for (i = 0; i < info->dlpi_phnum; i++) {
1734     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1735     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1736       // base address of a library is the lowest address of its loaded
1737       // segments.
1738       if (libbase == NULL || libbase > segbase) {
1739         libbase = segbase;
1740       }
1741       // see if 'addr' is within current segment
1742       if (segbase <= d->addr &&
1743           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1744         found = true;
1745       }
1746     }
1747   }
1748 
1749   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1750   // so dll_address_to_library_name() can fall through to use dladdr() which
1751   // can figure out executable name from argv[0].
1752   if (found && info->dlpi_name && info->dlpi_name[0]) {
1753     d->base = libbase;
1754     if (d->fname) {
1755       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1756     }
1757     return 1;
1758   }
1759   return 0;
1760 }
1761 
1762 bool os::dll_address_to_library_name(address addr, char* buf,
1763                                      int buflen, int* offset) {
1764   Dl_info dlinfo;
1765   struct _address_to_library_name data;
1766 
1767   // There is a bug in old glibc dladdr() implementation that it could resolve
1768   // to wrong library name if the .so file has a base address != NULL. Here
1769   // we iterate through the program headers of all loaded libraries to find
1770   // out which library 'addr' really belongs to. This workaround can be
1771   // removed once the minimum requirement for glibc is moved to 2.3.x.
1772   data.addr = addr;
1773   data.fname = buf;
1774   data.buflen = buflen;
1775   data.base = NULL;
1776   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1777 
1778   if (rslt) {
1779      // buf already contains library name
1780      if (offset) *offset = addr - data.base;
1781      return true;
1782   } else if (dladdr((void*)addr, &dlinfo)){
1783      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1784      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1785      return true;
1786   } else {
1787      if (buf) buf[0] = '\0';
1788      if (offset) *offset = -1;
1789      return false;
1790   }
1791 }
1792 
1793   // Loads .dll/.so and
1794   // in case of error it checks if .dll/.so was built for the
1795   // same architecture as Hotspot is running on
1796 
1797 
1798 // Remember the stack's state. The Linux dynamic linker will change
1799 // the stack to 'executable' at most once, so we must safepoint only once.
1800 bool os::Linux::_stack_is_executable = false;
1801 
1802 // VM operation that loads a library.  This is necessary if stack protection
1803 // of the Java stacks can be lost during loading the library.  If we
1804 // do not stop the Java threads, they can stack overflow before the stacks
1805 // are protected again.
1806 class VM_LinuxDllLoad: public VM_Operation {
1807  private:
1808   const char *_filename;
1809   char *_ebuf;
1810   int _ebuflen;
1811   void *_lib;
1812  public:
1813   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1814     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1815   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1816   void doit() {
1817     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1818     os::Linux::_stack_is_executable = true;
1819   }
1820   void* loaded_library() { return _lib; }
1821 };
1822 
1823 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1824 {
1825   void * result = NULL;
1826   bool load_attempted = false;
1827 
1828   // Check whether the library to load might change execution rights
1829   // of the stack. If they are changed, the protection of the stack
1830   // guard pages will be lost. We need a safepoint to fix this.
1831   //
1832   // See Linux man page execstack(8) for more info.
1833   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1834     ElfFile ef(filename);
1835     if (!ef.specifies_noexecstack()) {
1836       if (!is_init_completed()) {
1837         os::Linux::_stack_is_executable = true;
1838         // This is OK - No Java threads have been created yet, and hence no
1839         // stack guard pages to fix.
1840         //
1841         // This should happen only when you are building JDK7 using a very
1842         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1843         //
1844         // Dynamic loader will make all stacks executable after
1845         // this function returns, and will not do that again.
1846         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1847       } else {
1848         warning("You have loaded library %s which might have disabled stack guard. "
1849                 "The VM will try to fix the stack guard now.\n"
1850                 "It's highly recommended that you fix the library with "
1851                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1852                 filename);
1853 
1854         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1855         JavaThread *jt = JavaThread::current();
1856         if (jt->thread_state() != _thread_in_native) {
1857           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1858           // that requires ExecStack. Cannot enter safe point. Let's give up.
1859           warning("Unable to fix stack guard. Giving up.");
1860         } else {
1861           if (!LoadExecStackDllInVMThread) {
1862             // This is for the case where the DLL has an static
1863             // constructor function that executes JNI code. We cannot
1864             // load such DLLs in the VMThread.
1865             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1866           }
1867 
1868           ThreadInVMfromNative tiv(jt);
1869           debug_only(VMNativeEntryWrapper vew;)
1870 
1871           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1872           VMThread::execute(&op);
1873           if (LoadExecStackDllInVMThread) {
1874             result = op.loaded_library();
1875           }
1876           load_attempted = true;
1877         }
1878       }
1879     }
1880   }
1881 
1882   if (!load_attempted) {
1883     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1884   }
1885 
1886   if (result != NULL) {
1887     // Successful loading
1888     return result;
1889   }
1890 
1891   Elf32_Ehdr elf_head;
1892   int diag_msg_max_length=ebuflen-strlen(ebuf);
1893   char* diag_msg_buf=ebuf+strlen(ebuf);
1894 
1895   if (diag_msg_max_length==0) {
1896     // No more space in ebuf for additional diagnostics message
1897     return NULL;
1898   }
1899 
1900 
1901   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1902 
1903   if (file_descriptor < 0) {
1904     // Can't open library, report dlerror() message
1905     return NULL;
1906   }
1907 
1908   bool failed_to_read_elf_head=
1909     (sizeof(elf_head)!=
1910         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1911 
1912   ::close(file_descriptor);
1913   if (failed_to_read_elf_head) {
1914     // file i/o error - report dlerror() msg
1915     return NULL;
1916   }
1917 
1918   typedef struct {
1919     Elf32_Half  code;         // Actual value as defined in elf.h
1920     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1921     char        elf_class;    // 32 or 64 bit
1922     char        endianess;    // MSB or LSB
1923     char*       name;         // String representation
1924   } arch_t;
1925 
1926   #ifndef EM_486
1927   #define EM_486          6               /* Intel 80486 */
1928   #endif
1929 
1930   static const arch_t arch_array[]={
1931     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1932     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1933     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1934     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1935     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1936     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1937     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1938     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1939     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1940     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1941     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1942     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1943     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1944     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1945     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1946     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1947   };
1948 
1949   #if  (defined IA32)
1950     static  Elf32_Half running_arch_code=EM_386;
1951   #elif   (defined AMD64)
1952     static  Elf32_Half running_arch_code=EM_X86_64;
1953   #elif  (defined IA64)
1954     static  Elf32_Half running_arch_code=EM_IA_64;
1955   #elif  (defined __sparc) && (defined _LP64)
1956     static  Elf32_Half running_arch_code=EM_SPARCV9;
1957   #elif  (defined __sparc) && (!defined _LP64)
1958     static  Elf32_Half running_arch_code=EM_SPARC;
1959   #elif  (defined __powerpc64__)
1960     static  Elf32_Half running_arch_code=EM_PPC64;
1961   #elif  (defined __powerpc__)
1962     static  Elf32_Half running_arch_code=EM_PPC;
1963   #elif  (defined ARM)
1964     static  Elf32_Half running_arch_code=EM_ARM;
1965   #elif  (defined S390)
1966     static  Elf32_Half running_arch_code=EM_S390;
1967   #elif  (defined ALPHA)
1968     static  Elf32_Half running_arch_code=EM_ALPHA;
1969   #elif  (defined MIPSEL)
1970     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1971   #elif  (defined PARISC)
1972     static  Elf32_Half running_arch_code=EM_PARISC;
1973   #elif  (defined MIPS)
1974     static  Elf32_Half running_arch_code=EM_MIPS;
1975   #elif  (defined M68K)
1976     static  Elf32_Half running_arch_code=EM_68K;
1977   #else
1978     #error Method os::dll_load requires that one of following is defined:\
1979          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1980   #endif
1981 
1982   // Identify compatability class for VM's architecture and library's architecture
1983   // Obtain string descriptions for architectures
1984 
1985   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1986   int running_arch_index=-1;
1987 
1988   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1989     if (running_arch_code == arch_array[i].code) {
1990       running_arch_index    = i;
1991     }
1992     if (lib_arch.code == arch_array[i].code) {
1993       lib_arch.compat_class = arch_array[i].compat_class;
1994       lib_arch.name         = arch_array[i].name;
1995     }
1996   }
1997 
1998   assert(running_arch_index != -1,
1999     "Didn't find running architecture code (running_arch_code) in arch_array");
2000   if (running_arch_index == -1) {
2001     // Even though running architecture detection failed
2002     // we may still continue with reporting dlerror() message
2003     return NULL;
2004   }
2005 
2006   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2007     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2008     return NULL;
2009   }
2010 
2011 #ifndef S390
2012   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2013     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2014     return NULL;
2015   }
2016 #endif // !S390
2017 
2018   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2019     if ( lib_arch.name!=NULL ) {
2020       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2021         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2022         lib_arch.name, arch_array[running_arch_index].name);
2023     } else {
2024       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2025       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2026         lib_arch.code,
2027         arch_array[running_arch_index].name);
2028     }
2029   }
2030 
2031   return NULL;
2032 }
2033 
2034 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2035   void * result = ::dlopen(filename, RTLD_LAZY);
2036   if (result == NULL) {
2037     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2038     ebuf[ebuflen-1] = '\0';
2039   }
2040   return result;
2041 }
2042 
2043 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2044   void * result = NULL;
2045   if (LoadExecStackDllInVMThread) {
2046     result = dlopen_helper(filename, ebuf, ebuflen);
2047   }
2048 
2049   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2050   // library that requires an executable stack, or which does not have this
2051   // stack attribute set, dlopen changes the stack attribute to executable. The
2052   // read protection of the guard pages gets lost.
2053   //
2054   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2055   // may have been queued at the same time.
2056 
2057   if (!_stack_is_executable) {
2058     JavaThread *jt = Threads::first();
2059 
2060     while (jt) {
2061       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2062           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2063         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2064                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2065           warning("Attempt to reguard stack yellow zone failed.");
2066         }
2067       }
2068       jt = jt->next();
2069     }
2070   }
2071 
2072   return result;
2073 }
2074 
2075 /*
2076  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2077  * chances are you might want to run the generated bits against glibc-2.0
2078  * libdl.so, so always use locking for any version of glibc.
2079  */
2080 void* os::dll_lookup(void* handle, const char* name) {
2081   pthread_mutex_lock(&dl_mutex);
2082   void* res = dlsym(handle, name);
2083   pthread_mutex_unlock(&dl_mutex);
2084   return res;
2085 }
2086 
2087 
2088 static bool _print_ascii_file(const char* filename, outputStream* st) {
2089   int fd = ::open(filename, O_RDONLY);
2090   if (fd == -1) {
2091      return false;
2092   }
2093 
2094   char buf[32];
2095   int bytes;
2096   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2097     st->print_raw(buf, bytes);
2098   }
2099 
2100   ::close(fd);
2101 
2102   return true;
2103 }
2104 
2105 void os::print_dll_info(outputStream *st) {
2106    st->print_cr("Dynamic libraries:");
2107 
2108    char fname[32];
2109    pid_t pid = os::Linux::gettid();
2110 
2111    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2112 
2113    if (!_print_ascii_file(fname, st)) {
2114      st->print("Can not get library information for pid = %d\n", pid);
2115    }
2116 }
2117 
2118 void os::print_os_info_brief(outputStream* st) {
2119   os::Linux::print_distro_info(st);
2120 
2121   os::Posix::print_uname_info(st);
2122 
2123   os::Linux::print_libversion_info(st);
2124 
2125 }
2126 
2127 void os::print_os_info(outputStream* st) {
2128   st->print("OS:");
2129 
2130   os::Linux::print_distro_info(st);
2131 
2132   os::Posix::print_uname_info(st);
2133 
2134   // Print warning if unsafe chroot environment detected
2135   if (unsafe_chroot_detected) {
2136     st->print("WARNING!! ");
2137     st->print_cr(unstable_chroot_error);
2138   }
2139 
2140   os::Linux::print_libversion_info(st);
2141 
2142   os::Posix::print_rlimit_info(st);
2143 
2144   os::Posix::print_load_average(st);
2145 
2146   os::Linux::print_full_memory_info(st);
2147 }
2148 
2149 // Try to identify popular distros.
2150 // Most Linux distributions have /etc/XXX-release file, which contains
2151 // the OS version string. Some have more than one /etc/XXX-release file
2152 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2153 // so the order is important.
2154 void os::Linux::print_distro_info(outputStream* st) {
2155   if (!_print_ascii_file("/etc/mandrake-release", st) &&
2156       !_print_ascii_file("/etc/sun-release", st) &&
2157       !_print_ascii_file("/etc/redhat-release", st) &&
2158       !_print_ascii_file("/etc/SuSE-release", st) &&
2159       !_print_ascii_file("/etc/turbolinux-release", st) &&
2160       !_print_ascii_file("/etc/gentoo-release", st) &&
2161       !_print_ascii_file("/etc/debian_version", st) &&
2162       !_print_ascii_file("/etc/ltib-release", st) &&
2163       !_print_ascii_file("/etc/angstrom-version", st)) {
2164       st->print("Linux");
2165   }
2166   st->cr();
2167 }
2168 
2169 void os::Linux::print_libversion_info(outputStream* st) {
2170   // libc, pthread
2171   st->print("libc:");
2172   st->print(os::Linux::glibc_version()); st->print(" ");
2173   st->print(os::Linux::libpthread_version()); st->print(" ");
2174   if (os::Linux::is_LinuxThreads()) {
2175      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2176   }
2177   st->cr();
2178 }
2179 
2180 void os::Linux::print_full_memory_info(outputStream* st) {
2181    st->print("\n/proc/meminfo:\n");
2182    _print_ascii_file("/proc/meminfo", st);
2183    st->cr();
2184 }
2185 
2186 void os::print_memory_info(outputStream* st) {
2187 
2188   st->print("Memory:");
2189   st->print(" %dk page", os::vm_page_size()>>10);
2190 
2191   // values in struct sysinfo are "unsigned long"
2192   struct sysinfo si;
2193   sysinfo(&si);
2194 
2195   st->print(", physical " UINT64_FORMAT "k",
2196             os::physical_memory() >> 10);
2197   st->print("(" UINT64_FORMAT "k free)",
2198             os::available_memory() >> 10);
2199   st->print(", swap " UINT64_FORMAT "k",
2200             ((jlong)si.totalswap * si.mem_unit) >> 10);
2201   st->print("(" UINT64_FORMAT "k free)",
2202             ((jlong)si.freeswap * si.mem_unit) >> 10);
2203   st->cr();
2204 }
2205 
2206 void os::pd_print_cpu_info(outputStream* st) {
2207   st->print("\n/proc/cpuinfo:\n");
2208   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2209     st->print("  <Not Available>");
2210   }
2211   st->cr();
2212 }
2213 
2214 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2215 // but they're the same for all the linux arch that we support
2216 // and they're the same for solaris but there's no common place to put this.
2217 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2218                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2219                           "ILL_COPROC", "ILL_BADSTK" };
2220 
2221 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2222                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2223                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2224 
2225 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2226 
2227 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2228 
2229 void os::print_siginfo(outputStream* st, void* siginfo) {
2230   st->print("siginfo:");
2231 
2232   const int buflen = 100;
2233   char buf[buflen];
2234   siginfo_t *si = (siginfo_t*)siginfo;
2235   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2236   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2237     st->print("si_errno=%s", buf);
2238   } else {
2239     st->print("si_errno=%d", si->si_errno);
2240   }
2241   const int c = si->si_code;
2242   assert(c > 0, "unexpected si_code");
2243   switch (si->si_signo) {
2244   case SIGILL:
2245     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2246     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2247     break;
2248   case SIGFPE:
2249     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2250     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2251     break;
2252   case SIGSEGV:
2253     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2254     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2255     break;
2256   case SIGBUS:
2257     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2258     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2259     break;
2260   default:
2261     st->print(", si_code=%d", si->si_code);
2262     // no si_addr
2263   }
2264 
2265   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2266       UseSharedSpaces) {
2267     FileMapInfo* mapinfo = FileMapInfo::current_info();
2268     if (mapinfo->is_in_shared_space(si->si_addr)) {
2269       st->print("\n\nError accessing class data sharing archive."   \
2270                 " Mapped file inaccessible during execution, "      \
2271                 " possible disk/network problem.");
2272     }
2273   }
2274   st->cr();
2275 }
2276 
2277 
2278 static void print_signal_handler(outputStream* st, int sig,
2279                                  char* buf, size_t buflen);
2280 
2281 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2282   st->print_cr("Signal Handlers:");
2283   print_signal_handler(st, SIGSEGV, buf, buflen);
2284   print_signal_handler(st, SIGBUS , buf, buflen);
2285   print_signal_handler(st, SIGFPE , buf, buflen);
2286   print_signal_handler(st, SIGPIPE, buf, buflen);
2287   print_signal_handler(st, SIGXFSZ, buf, buflen);
2288   print_signal_handler(st, SIGILL , buf, buflen);
2289   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2290   print_signal_handler(st, SR_signum, buf, buflen);
2291   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2292   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2293   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2294   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2295 }
2296 
2297 static char saved_jvm_path[MAXPATHLEN] = {0};
2298 
2299 // Find the full path to the current module, libjvm.so
2300 void os::jvm_path(char *buf, jint buflen) {
2301   // Error checking.
2302   if (buflen < MAXPATHLEN) {
2303     assert(false, "must use a large-enough buffer");
2304     buf[0] = '\0';
2305     return;
2306   }
2307   // Lazy resolve the path to current module.
2308   if (saved_jvm_path[0] != 0) {
2309     strcpy(buf, saved_jvm_path);
2310     return;
2311   }
2312 
2313   char dli_fname[MAXPATHLEN];
2314   bool ret = dll_address_to_library_name(
2315                 CAST_FROM_FN_PTR(address, os::jvm_path),
2316                 dli_fname, sizeof(dli_fname), NULL);
2317   assert(ret != 0, "cannot locate libjvm");
2318   char *rp = realpath(dli_fname, buf);
2319   if (rp == NULL)
2320     return;
2321 
2322   if (Arguments::created_by_gamma_launcher()) {
2323     // Support for the gamma launcher.  Typical value for buf is
2324     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2325     // the right place in the string, then assume we are installed in a JDK and
2326     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2327     // up the path so it looks like libjvm.so is installed there (append a
2328     // fake suffix hotspot/libjvm.so).
2329     const char *p = buf + strlen(buf) - 1;
2330     for (int count = 0; p > buf && count < 5; ++count) {
2331       for (--p; p > buf && *p != '/'; --p)
2332         /* empty */ ;
2333     }
2334 
2335     if (strncmp(p, "/jre/lib/", 9) != 0) {
2336       // Look for JAVA_HOME in the environment.
2337       char* java_home_var = ::getenv("JAVA_HOME");
2338       if (java_home_var != NULL && java_home_var[0] != 0) {
2339         char* jrelib_p;
2340         int len;
2341 
2342         // Check the current module name "libjvm.so".
2343         p = strrchr(buf, '/');
2344         assert(strstr(p, "/libjvm") == p, "invalid library name");
2345 
2346         rp = realpath(java_home_var, buf);
2347         if (rp == NULL)
2348           return;
2349 
2350         // determine if this is a legacy image or modules image
2351         // modules image doesn't have "jre" subdirectory
2352         len = strlen(buf);
2353         jrelib_p = buf + len;
2354         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2355         if (0 != access(buf, F_OK)) {
2356           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2357         }
2358 
2359         if (0 == access(buf, F_OK)) {
2360           // Use current module name "libjvm.so"
2361           len = strlen(buf);
2362           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2363         } else {
2364           // Go back to path of .so
2365           rp = realpath(dli_fname, buf);
2366           if (rp == NULL)
2367             return;
2368         }
2369       }
2370     }
2371   }
2372 
2373   strcpy(saved_jvm_path, buf);
2374 }
2375 
2376 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2377   // no prefix required, not even "_"
2378 }
2379 
2380 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2381   // no suffix required
2382 }
2383 
2384 ////////////////////////////////////////////////////////////////////////////////
2385 // sun.misc.Signal support
2386 
2387 static volatile jint sigint_count = 0;
2388 
2389 static void
2390 UserHandler(int sig, void *siginfo, void *context) {
2391   // 4511530 - sem_post is serialized and handled by the manager thread. When
2392   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2393   // don't want to flood the manager thread with sem_post requests.
2394   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2395       return;
2396 
2397   // Ctrl-C is pressed during error reporting, likely because the error
2398   // handler fails to abort. Let VM die immediately.
2399   if (sig == SIGINT && is_error_reported()) {
2400      os::die();
2401   }
2402 
2403   os::signal_notify(sig);
2404 }
2405 
2406 void* os::user_handler() {
2407   return CAST_FROM_FN_PTR(void*, UserHandler);
2408 }
2409 
2410 extern "C" {
2411   typedef void (*sa_handler_t)(int);
2412   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2413 }
2414 
2415 void* os::signal(int signal_number, void* handler) {
2416   struct sigaction sigAct, oldSigAct;
2417 
2418   sigfillset(&(sigAct.sa_mask));
2419   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2420   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2421 
2422   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2423     // -1 means registration failed
2424     return (void *)-1;
2425   }
2426 
2427   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2428 }
2429 
2430 void os::signal_raise(int signal_number) {
2431   ::raise(signal_number);
2432 }
2433 
2434 /*
2435  * The following code is moved from os.cpp for making this
2436  * code platform specific, which it is by its very nature.
2437  */
2438 
2439 // Will be modified when max signal is changed to be dynamic
2440 int os::sigexitnum_pd() {
2441   return NSIG;
2442 }
2443 
2444 // a counter for each possible signal value
2445 static volatile jint pending_signals[NSIG+1] = { 0 };
2446 
2447 // Linux(POSIX) specific hand shaking semaphore.
2448 static sem_t sig_sem;
2449 
2450 void os::signal_init_pd() {
2451   // Initialize signal structures
2452   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2453 
2454   // Initialize signal semaphore
2455   ::sem_init(&sig_sem, 0, 0);
2456 }
2457 
2458 void os::signal_notify(int sig) {
2459   Atomic::inc(&pending_signals[sig]);
2460   ::sem_post(&sig_sem);
2461 }
2462 
2463 static int check_pending_signals(bool wait) {
2464   Atomic::store(0, &sigint_count);
2465   for (;;) {
2466     for (int i = 0; i < NSIG + 1; i++) {
2467       jint n = pending_signals[i];
2468       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2469         return i;
2470       }
2471     }
2472     if (!wait) {
2473       return -1;
2474     }
2475     JavaThread *thread = JavaThread::current();
2476     ThreadBlockInVM tbivm(thread);
2477 
2478     bool threadIsSuspended;
2479     do {
2480       thread->set_suspend_equivalent();
2481       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2482       ::sem_wait(&sig_sem);
2483 
2484       // were we externally suspended while we were waiting?
2485       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2486       if (threadIsSuspended) {
2487         //
2488         // The semaphore has been incremented, but while we were waiting
2489         // another thread suspended us. We don't want to continue running
2490         // while suspended because that would surprise the thread that
2491         // suspended us.
2492         //
2493         ::sem_post(&sig_sem);
2494 
2495         thread->java_suspend_self();
2496       }
2497     } while (threadIsSuspended);
2498   }
2499 }
2500 
2501 int os::signal_lookup() {
2502   return check_pending_signals(false);
2503 }
2504 
2505 int os::signal_wait() {
2506   return check_pending_signals(true);
2507 }
2508 
2509 ////////////////////////////////////////////////////////////////////////////////
2510 // Virtual Memory
2511 
2512 int os::vm_page_size() {
2513   // Seems redundant as all get out
2514   assert(os::Linux::page_size() != -1, "must call os::init");
2515   return os::Linux::page_size();
2516 }
2517 
2518 // Solaris allocates memory by pages.
2519 int os::vm_allocation_granularity() {
2520   assert(os::Linux::page_size() != -1, "must call os::init");
2521   return os::Linux::page_size();
2522 }
2523 
2524 // Rationale behind this function:
2525 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2526 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2527 //  samples for JITted code. Here we create private executable mapping over the code cache
2528 //  and then we can use standard (well, almost, as mapping can change) way to provide
2529 //  info for the reporting script by storing timestamp and location of symbol
2530 void linux_wrap_code(char* base, size_t size) {
2531   static volatile jint cnt = 0;
2532 
2533   if (!UseOprofile) {
2534     return;
2535   }
2536 
2537   char buf[PATH_MAX+1];
2538   int num = Atomic::add(1, &cnt);
2539 
2540   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2541            os::get_temp_directory(), os::current_process_id(), num);
2542   unlink(buf);
2543 
2544   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2545 
2546   if (fd != -1) {
2547     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2548     if (rv != (off_t)-1) {
2549       if (::write(fd, "", 1) == 1) {
2550         mmap(base, size,
2551              PROT_READ|PROT_WRITE|PROT_EXEC,
2552              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2553       }
2554     }
2555     ::close(fd);
2556     unlink(buf);
2557   }
2558 }
2559 
2560 // NOTE: Linux kernel does not really reserve the pages for us.
2561 //       All it does is to check if there are enough free pages
2562 //       left at the time of mmap(). This could be a potential
2563 //       problem.
2564 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2565   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2566   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2567                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2568   if (res != (uintptr_t) MAP_FAILED) {
2569     if (UseNUMAInterleaving) {
2570       numa_make_global(addr, size);
2571     }
2572     return true;
2573   }
2574   return false;
2575 }
2576 
2577 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2578 #ifndef MAP_HUGETLB
2579 #define MAP_HUGETLB 0x40000
2580 #endif
2581 
2582 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2583 #ifndef MADV_HUGEPAGE
2584 #define MADV_HUGEPAGE 14
2585 #endif
2586 
2587 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2588                        bool exec) {
2589   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2590     int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2591     uintptr_t res =
2592       (uintptr_t) ::mmap(addr, size, prot,
2593                          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2594                          -1, 0);
2595     if (res != (uintptr_t) MAP_FAILED) {
2596       if (UseNUMAInterleaving) {
2597         numa_make_global(addr, size);
2598       }
2599       return true;
2600     }
2601     // Fall through and try to use small pages
2602   }
2603 
2604   if (commit_memory(addr, size, exec)) {
2605     realign_memory(addr, size, alignment_hint);
2606     return true;
2607   }
2608   return false;
2609 }
2610 
2611 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2612   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2613     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2614     // be supported or the memory may already be backed by huge pages.
2615     ::madvise(addr, bytes, MADV_HUGEPAGE);
2616   }
2617 }
2618 
2619 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2620   // This method works by doing an mmap over an existing mmaping and effectively discarding
2621   // the existing pages. However it won't work for SHM-based large pages that cannot be
2622   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2623   // small pages on top of the SHM segment. This method always works for small pages, so we
2624   // allow that in any case.
2625   if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2626     commit_memory(addr, bytes, alignment_hint, false);
2627   }
2628 }
2629 
2630 void os::numa_make_global(char *addr, size_t bytes) {
2631   Linux::numa_interleave_memory(addr, bytes);
2632 }
2633 
2634 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2635   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2636 }
2637 
2638 bool os::numa_topology_changed()   { return false; }
2639 
2640 size_t os::numa_get_groups_num() {
2641   int max_node = Linux::numa_max_node();
2642   return max_node > 0 ? max_node + 1 : 1;
2643 }
2644 
2645 int os::numa_get_group_id() {
2646   int cpu_id = Linux::sched_getcpu();
2647   if (cpu_id != -1) {
2648     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2649     if (lgrp_id != -1) {
2650       return lgrp_id;
2651     }
2652   }
2653   return 0;
2654 }
2655 
2656 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2657   for (size_t i = 0; i < size; i++) {
2658     ids[i] = i;
2659   }
2660   return size;
2661 }
2662 
2663 bool os::get_page_info(char *start, page_info* info) {
2664   return false;
2665 }
2666 
2667 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2668   return end;
2669 }
2670 
2671 
2672 int os::Linux::sched_getcpu_syscall(void) {
2673   unsigned int cpu;
2674   int retval = -1;
2675 
2676 #if defined(IA32)
2677 # ifndef SYS_getcpu
2678 # define SYS_getcpu 318
2679 # endif
2680   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2681 #elif defined(AMD64)
2682 // Unfortunately we have to bring all these macros here from vsyscall.h
2683 // to be able to compile on old linuxes.
2684 # define __NR_vgetcpu 2
2685 # define VSYSCALL_START (-10UL << 20)
2686 # define VSYSCALL_SIZE 1024
2687 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2688   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2689   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2690   retval = vgetcpu(&cpu, NULL, NULL);
2691 #endif
2692 
2693   return (retval == -1) ? retval : cpu;
2694 }
2695 
2696 // Something to do with the numa-aware allocator needs these symbols
2697 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2698 extern "C" JNIEXPORT void numa_error(char *where) { }
2699 extern "C" JNIEXPORT int fork1() { return fork(); }
2700 
2701 
2702 // If we are running with libnuma version > 2, then we should
2703 // be trying to use symbols with versions 1.1
2704 // If we are running with earlier version, which did not have symbol versions,
2705 // we should use the base version.
2706 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2707   void *f = dlvsym(handle, name, "libnuma_1.1");
2708   if (f == NULL) {
2709     f = dlsym(handle, name);
2710   }
2711   return f;
2712 }
2713 
2714 bool os::Linux::libnuma_init() {
2715   // sched_getcpu() should be in libc.
2716   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2717                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2718 
2719   // If it's not, try a direct syscall.
2720   if (sched_getcpu() == -1)
2721     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2722 
2723   if (sched_getcpu() != -1) { // Does it work?
2724     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2725     if (handle != NULL) {
2726       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2727                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2728       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2729                                        libnuma_dlsym(handle, "numa_max_node")));
2730       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2731                                         libnuma_dlsym(handle, "numa_available")));
2732       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2733                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2734       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2735                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2736 
2737 
2738       if (numa_available() != -1) {
2739         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2740         // Create a cpu -> node mapping
2741         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2742         rebuild_cpu_to_node_map();
2743         return true;
2744       }
2745     }
2746   }
2747   return false;
2748 }
2749 
2750 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2751 // The table is later used in get_node_by_cpu().
2752 void os::Linux::rebuild_cpu_to_node_map() {
2753   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2754                               // in libnuma (possible values are starting from 16,
2755                               // and continuing up with every other power of 2, but less
2756                               // than the maximum number of CPUs supported by kernel), and
2757                               // is a subject to change (in libnuma version 2 the requirements
2758                               // are more reasonable) we'll just hardcode the number they use
2759                               // in the library.
2760   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2761 
2762   size_t cpu_num = os::active_processor_count();
2763   size_t cpu_map_size = NCPUS / BitsPerCLong;
2764   size_t cpu_map_valid_size =
2765     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2766 
2767   cpu_to_node()->clear();
2768   cpu_to_node()->at_grow(cpu_num - 1);
2769   size_t node_num = numa_get_groups_num();
2770 
2771   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2772   for (size_t i = 0; i < node_num; i++) {
2773     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2774       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2775         if (cpu_map[j] != 0) {
2776           for (size_t k = 0; k < BitsPerCLong; k++) {
2777             if (cpu_map[j] & (1UL << k)) {
2778               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2779             }
2780           }
2781         }
2782       }
2783     }
2784   }
2785   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2786 }
2787 
2788 int os::Linux::get_node_by_cpu(int cpu_id) {
2789   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2790     return cpu_to_node()->at(cpu_id);
2791   }
2792   return -1;
2793 }
2794 
2795 GrowableArray<int>* os::Linux::_cpu_to_node;
2796 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2797 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2798 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2799 os::Linux::numa_available_func_t os::Linux::_numa_available;
2800 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2801 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2802 unsigned long* os::Linux::_numa_all_nodes;
2803 
2804 bool os::pd_uncommit_memory(char* addr, size_t size) {
2805   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2806                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2807   return res  != (uintptr_t) MAP_FAILED;
2808 }
2809 
2810 // Linux uses a growable mapping for the stack, and if the mapping for
2811 // the stack guard pages is not removed when we detach a thread the
2812 // stack cannot grow beyond the pages where the stack guard was
2813 // mapped.  If at some point later in the process the stack expands to
2814 // that point, the Linux kernel cannot expand the stack any further
2815 // because the guard pages are in the way, and a segfault occurs.
2816 //
2817 // However, it's essential not to split the stack region by unmapping
2818 // a region (leaving a hole) that's already part of the stack mapping,
2819 // so if the stack mapping has already grown beyond the guard pages at
2820 // the time we create them, we have to truncate the stack mapping.
2821 // So, we need to know the extent of the stack mapping when
2822 // create_stack_guard_pages() is called.
2823 
2824 // Find the bounds of the stack mapping.  Return true for success.
2825 //
2826 // We only need this for stacks that are growable: at the time of
2827 // writing thread stacks don't use growable mappings (i.e. those
2828 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2829 // only applies to the main thread.
2830 
2831 static
2832 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2833 
2834   char buf[128];
2835   int fd, sz;
2836 
2837   if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2838     return false;
2839   }
2840 
2841   const char kw[] = "[stack]";
2842   const int kwlen = sizeof(kw)-1;
2843 
2844   // Address part of /proc/self/maps couldn't be more than 128 bytes
2845   while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2846      if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2847         // Extract addresses
2848         if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2849            uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2850            if (sp >= *bottom && sp <= *top) {
2851               ::close(fd);
2852               return true;
2853            }
2854         }
2855      }
2856   }
2857 
2858  ::close(fd);
2859   return false;
2860 }
2861 
2862 
2863 // If the (growable) stack mapping already extends beyond the point
2864 // where we're going to put our guard pages, truncate the mapping at
2865 // that point by munmap()ping it.  This ensures that when we later
2866 // munmap() the guard pages we don't leave a hole in the stack
2867 // mapping. This only affects the main/initial thread, but guard
2868 // against future OS changes
2869 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2870   uintptr_t stack_extent, stack_base;
2871   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2872   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2873       assert(os::Linux::is_initial_thread(),
2874            "growable stack in non-initial thread");
2875     if (stack_extent < (uintptr_t)addr)
2876       ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2877   }
2878 
2879   return os::commit_memory(addr, size);
2880 }
2881 
2882 // If this is a growable mapping, remove the guard pages entirely by
2883 // munmap()ping them.  If not, just call uncommit_memory(). This only
2884 // affects the main/initial thread, but guard against future OS changes
2885 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2886   uintptr_t stack_extent, stack_base;
2887   bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2888   if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2889       assert(os::Linux::is_initial_thread(),
2890            "growable stack in non-initial thread");
2891 
2892     return ::munmap(addr, size) == 0;
2893   }
2894 
2895   return os::uncommit_memory(addr, size);
2896 }
2897 
2898 static address _highest_vm_reserved_address = NULL;
2899 
2900 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2901 // at 'requested_addr'. If there are existing memory mappings at the same
2902 // location, however, they will be overwritten. If 'fixed' is false,
2903 // 'requested_addr' is only treated as a hint, the return value may or
2904 // may not start from the requested address. Unlike Linux mmap(), this
2905 // function returns NULL to indicate failure.
2906 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2907   char * addr;
2908   int flags;
2909 
2910   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2911   if (fixed) {
2912     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2913     flags |= MAP_FIXED;
2914   }
2915 
2916   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
2917   // touch an uncommitted page. Otherwise, the read/write might
2918   // succeed if we have enough swap space to back the physical page.
2919   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
2920                        flags, -1, 0);
2921 
2922   if (addr != MAP_FAILED) {
2923     // anon_mmap() should only get called during VM initialization,
2924     // don't need lock (actually we can skip locking even it can be called
2925     // from multiple threads, because _highest_vm_reserved_address is just a
2926     // hint about the upper limit of non-stack memory regions.)
2927     if ((address)addr + bytes > _highest_vm_reserved_address) {
2928       _highest_vm_reserved_address = (address)addr + bytes;
2929     }
2930   }
2931 
2932   return addr == MAP_FAILED ? NULL : addr;
2933 }
2934 
2935 // Don't update _highest_vm_reserved_address, because there might be memory
2936 // regions above addr + size. If so, releasing a memory region only creates
2937 // a hole in the address space, it doesn't help prevent heap-stack collision.
2938 //
2939 static int anon_munmap(char * addr, size_t size) {
2940   return ::munmap(addr, size) == 0;
2941 }
2942 
2943 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2944                          size_t alignment_hint) {
2945   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2946 }
2947 
2948 bool os::pd_release_memory(char* addr, size_t size) {
2949   return anon_munmap(addr, size);
2950 }
2951 
2952 static address highest_vm_reserved_address() {
2953   return _highest_vm_reserved_address;
2954 }
2955 
2956 static bool linux_mprotect(char* addr, size_t size, int prot) {
2957   // Linux wants the mprotect address argument to be page aligned.
2958   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2959 
2960   // According to SUSv3, mprotect() should only be used with mappings
2961   // established by mmap(), and mmap() always maps whole pages. Unaligned
2962   // 'addr' likely indicates problem in the VM (e.g. trying to change
2963   // protection of malloc'ed or statically allocated memory). Check the
2964   // caller if you hit this assert.
2965   assert(addr == bottom, "sanity check");
2966 
2967   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2968   return ::mprotect(bottom, size, prot) == 0;
2969 }
2970 
2971 // Set protections specified
2972 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2973                         bool is_committed) {
2974   unsigned int p = 0;
2975   switch (prot) {
2976   case MEM_PROT_NONE: p = PROT_NONE; break;
2977   case MEM_PROT_READ: p = PROT_READ; break;
2978   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2979   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2980   default:
2981     ShouldNotReachHere();
2982   }
2983   // is_committed is unused.
2984   return linux_mprotect(addr, bytes, p);
2985 }
2986 
2987 bool os::guard_memory(char* addr, size_t size) {
2988   return linux_mprotect(addr, size, PROT_NONE);
2989 }
2990 
2991 bool os::unguard_memory(char* addr, size_t size) {
2992   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2993 }
2994 
2995 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2996   bool result = false;
2997   void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2998                   MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2999                   -1, 0);
3000 
3001   if (p != (void *) -1) {
3002     // We don't know if this really is a huge page or not.
3003     FILE *fp = fopen("/proc/self/maps", "r");
3004     if (fp) {
3005       while (!feof(fp)) {
3006         char chars[257];
3007         long x = 0;
3008         if (fgets(chars, sizeof(chars), fp)) {
3009           if (sscanf(chars, "%lx-%*x", &x) == 1
3010               && x == (long)p) {
3011             if (strstr (chars, "hugepage")) {
3012               result = true;
3013               break;
3014             }
3015           }
3016         }
3017       }
3018       fclose(fp);
3019     }
3020     munmap (p, page_size);
3021     if (result)
3022       return true;
3023   }
3024 
3025   if (warn) {
3026     warning("HugeTLBFS is not supported by the operating system.");
3027   }
3028 
3029   return result;
3030 }
3031 
3032 /*
3033 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3034 *
3035 * From the coredump_filter documentation:
3036 *
3037 * - (bit 0) anonymous private memory
3038 * - (bit 1) anonymous shared memory
3039 * - (bit 2) file-backed private memory
3040 * - (bit 3) file-backed shared memory
3041 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3042 *           effective only if the bit 2 is cleared)
3043 * - (bit 5) hugetlb private memory
3044 * - (bit 6) hugetlb shared memory
3045 */
3046 static void set_coredump_filter(void) {
3047   FILE *f;
3048   long cdm;
3049 
3050   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3051     return;
3052   }
3053 
3054   if (fscanf(f, "%lx", &cdm) != 1) {
3055     fclose(f);
3056     return;
3057   }
3058 
3059   rewind(f);
3060 
3061   if ((cdm & LARGEPAGES_BIT) == 0) {
3062     cdm |= LARGEPAGES_BIT;
3063     fprintf(f, "%#lx", cdm);
3064   }
3065 
3066   fclose(f);
3067 }
3068 
3069 // Large page support
3070 
3071 static size_t _large_page_size = 0;
3072 
3073 void os::large_page_init() {
3074   if (!UseLargePages) {
3075     UseHugeTLBFS = false;
3076     UseSHM = false;
3077     return;
3078   }
3079 
3080   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3081     // If UseLargePages is specified on the command line try both methods,
3082     // if it's default, then try only HugeTLBFS.
3083     if (FLAG_IS_DEFAULT(UseLargePages)) {
3084       UseHugeTLBFS = true;
3085     } else {
3086       UseHugeTLBFS = UseSHM = true;
3087     }
3088   }
3089 
3090   if (LargePageSizeInBytes) {
3091     _large_page_size = LargePageSizeInBytes;
3092   } else {
3093     // large_page_size on Linux is used to round up heap size. x86 uses either
3094     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3095     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3096     // page as large as 256M.
3097     //
3098     // Here we try to figure out page size by parsing /proc/meminfo and looking
3099     // for a line with the following format:
3100     //    Hugepagesize:     2048 kB
3101     //
3102     // If we can't determine the value (e.g. /proc is not mounted, or the text
3103     // format has been changed), we'll use the largest page size supported by
3104     // the processor.
3105 
3106 #ifndef ZERO
3107     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3108                        ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3109 #endif // ZERO
3110 
3111     FILE *fp = fopen("/proc/meminfo", "r");
3112     if (fp) {
3113       while (!feof(fp)) {
3114         int x = 0;
3115         char buf[16];
3116         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3117           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3118             _large_page_size = x * K;
3119             break;
3120           }
3121         } else {
3122           // skip to next line
3123           for (;;) {
3124             int ch = fgetc(fp);
3125             if (ch == EOF || ch == (int)'\n') break;
3126           }
3127         }
3128       }
3129       fclose(fp);
3130     }
3131   }
3132 
3133   // print a warning if any large page related flag is specified on command line
3134   bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3135 
3136   const size_t default_page_size = (size_t)Linux::page_size();
3137   if (_large_page_size > default_page_size) {
3138     _page_sizes[0] = _large_page_size;
3139     _page_sizes[1] = default_page_size;
3140     _page_sizes[2] = 0;
3141   }
3142   UseHugeTLBFS = UseHugeTLBFS &&
3143                  Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3144 
3145   if (UseHugeTLBFS)
3146     UseSHM = false;
3147 
3148   UseLargePages = UseHugeTLBFS || UseSHM;
3149 
3150   set_coredump_filter();
3151 }
3152 
3153 #ifndef SHM_HUGETLB
3154 #define SHM_HUGETLB 04000
3155 #endif
3156 
3157 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3158   // "exec" is passed in but not used.  Creating the shared image for
3159   // the code cache doesn't have an SHM_X executable permission to check.
3160   assert(UseLargePages && UseSHM, "only for SHM large pages");
3161 
3162   key_t key = IPC_PRIVATE;
3163   char *addr;
3164 
3165   bool warn_on_failure = UseLargePages &&
3166                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3167                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3168                         );
3169   char msg[128];
3170 
3171   // Create a large shared memory region to attach to based on size.
3172   // Currently, size is the total size of the heap
3173   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3174   if (shmid == -1) {
3175      // Possible reasons for shmget failure:
3176      // 1. shmmax is too small for Java heap.
3177      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3178      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3179      // 2. not enough large page memory.
3180      //    > check available large pages: cat /proc/meminfo
3181      //    > increase amount of large pages:
3182      //          echo new_value > /proc/sys/vm/nr_hugepages
3183      //      Note 1: different Linux may use different name for this property,
3184      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3185      //      Note 2: it's possible there's enough physical memory available but
3186      //            they are so fragmented after a long run that they can't
3187      //            coalesce into large pages. Try to reserve large pages when
3188      //            the system is still "fresh".
3189      if (warn_on_failure) {
3190        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3191        warning(msg);
3192      }
3193      return NULL;
3194   }
3195 
3196   // attach to the region
3197   addr = (char*)shmat(shmid, req_addr, 0);
3198   int err = errno;
3199 
3200   // Remove shmid. If shmat() is successful, the actual shared memory segment
3201   // will be deleted when it's detached by shmdt() or when the process
3202   // terminates. If shmat() is not successful this will remove the shared
3203   // segment immediately.
3204   shmctl(shmid, IPC_RMID, NULL);
3205 
3206   if ((intptr_t)addr == -1) {
3207      if (warn_on_failure) {
3208        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3209        warning(msg);
3210      }
3211      return NULL;
3212   }
3213 
3214   if ((addr != NULL) && UseNUMAInterleaving) {
3215     numa_make_global(addr, bytes);
3216   }
3217 
3218   // The memory is committed
3219   address pc = CALLER_PC;
3220   MemTracker::record_virtual_memory_reserve((address)addr, bytes, pc);
3221   MemTracker::record_virtual_memory_commit((address)addr, bytes, pc);
3222 
3223   return addr;
3224 }
3225 
3226 bool os::release_memory_special(char* base, size_t bytes) {
3227   // detaching the SHM segment will also delete it, see reserve_memory_special()
3228   int rslt = shmdt(base);
3229   if (rslt == 0) {
3230     MemTracker::record_virtual_memory_uncommit((address)base, bytes);
3231     MemTracker::record_virtual_memory_release((address)base, bytes);
3232     return true;
3233   } else {
3234    return false;
3235   }
3236 }
3237 
3238 size_t os::large_page_size() {
3239   return _large_page_size;
3240 }
3241 
3242 // HugeTLBFS allows application to commit large page memory on demand;
3243 // with SysV SHM the entire memory region must be allocated as shared
3244 // memory.
3245 bool os::can_commit_large_page_memory() {
3246   return UseHugeTLBFS;
3247 }
3248 
3249 bool os::can_execute_large_page_memory() {
3250   return UseHugeTLBFS;
3251 }
3252 
3253 // Reserve memory at an arbitrary address, only if that area is
3254 // available (and not reserved for something else).
3255 
3256 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3257   const int max_tries = 10;
3258   char* base[max_tries];
3259   size_t size[max_tries];
3260   const size_t gap = 0x000000;
3261 
3262   // Assert only that the size is a multiple of the page size, since
3263   // that's all that mmap requires, and since that's all we really know
3264   // about at this low abstraction level.  If we need higher alignment,
3265   // we can either pass an alignment to this method or verify alignment
3266   // in one of the methods further up the call chain.  See bug 5044738.
3267   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3268 
3269   // Repeatedly allocate blocks until the block is allocated at the
3270   // right spot. Give up after max_tries. Note that reserve_memory() will
3271   // automatically update _highest_vm_reserved_address if the call is
3272   // successful. The variable tracks the highest memory address every reserved
3273   // by JVM. It is used to detect heap-stack collision if running with
3274   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3275   // space than needed, it could confuse the collision detecting code. To
3276   // solve the problem, save current _highest_vm_reserved_address and
3277   // calculate the correct value before return.
3278   address old_highest = _highest_vm_reserved_address;
3279 
3280   // Linux mmap allows caller to pass an address as hint; give it a try first,
3281   // if kernel honors the hint then we can return immediately.
3282   char * addr = anon_mmap(requested_addr, bytes, false);
3283   if (addr == requested_addr) {
3284      return requested_addr;
3285   }
3286 
3287   if (addr != NULL) {
3288      // mmap() is successful but it fails to reserve at the requested address
3289      anon_munmap(addr, bytes);
3290   }
3291 
3292   int i;
3293   for (i = 0; i < max_tries; ++i) {
3294     base[i] = reserve_memory(bytes);
3295 
3296     if (base[i] != NULL) {
3297       // Is this the block we wanted?
3298       if (base[i] == requested_addr) {
3299         size[i] = bytes;
3300         break;
3301       }
3302 
3303       // Does this overlap the block we wanted? Give back the overlapped
3304       // parts and try again.
3305 
3306       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3307       if (top_overlap >= 0 && top_overlap < bytes) {
3308         unmap_memory(base[i], top_overlap);
3309         base[i] += top_overlap;
3310         size[i] = bytes - top_overlap;
3311       } else {
3312         size_t bottom_overlap = base[i] + bytes - requested_addr;
3313         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3314           unmap_memory(requested_addr, bottom_overlap);
3315           size[i] = bytes - bottom_overlap;
3316         } else {
3317           size[i] = bytes;
3318         }
3319       }
3320     }
3321   }
3322 
3323   // Give back the unused reserved pieces.
3324 
3325   for (int j = 0; j < i; ++j) {
3326     if (base[j] != NULL) {
3327       unmap_memory(base[j], size[j]);
3328     }
3329   }
3330 
3331   if (i < max_tries) {
3332     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3333     return requested_addr;
3334   } else {
3335     _highest_vm_reserved_address = old_highest;
3336     return NULL;
3337   }
3338 }
3339 
3340 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3341   return ::read(fd, buf, nBytes);
3342 }
3343 
3344 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3345 // Solaris uses poll(), linux uses park().
3346 // Poll() is likely a better choice, assuming that Thread.interrupt()
3347 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3348 // SIGSEGV, see 4355769.
3349 
3350 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3351   assert(thread == Thread::current(),  "thread consistency check");
3352 
3353   ParkEvent * const slp = thread->_SleepEvent ;
3354   slp->reset() ;
3355   OrderAccess::fence() ;
3356 
3357   if (interruptible) {
3358     jlong prevtime = javaTimeNanos();
3359 
3360     for (;;) {
3361       if (os::is_interrupted(thread, true)) {
3362         return OS_INTRPT;
3363       }
3364 
3365       jlong newtime = javaTimeNanos();
3366 
3367       if (newtime - prevtime < 0) {
3368         // time moving backwards, should only happen if no monotonic clock
3369         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3370         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3371       } else {
3372         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3373       }
3374 
3375       if(millis <= 0) {
3376         return OS_OK;
3377       }
3378 
3379       prevtime = newtime;
3380 
3381       {
3382         assert(thread->is_Java_thread(), "sanity check");
3383         JavaThread *jt = (JavaThread *) thread;
3384         ThreadBlockInVM tbivm(jt);
3385         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3386 
3387         jt->set_suspend_equivalent();
3388         // cleared by handle_special_suspend_equivalent_condition() or
3389         // java_suspend_self() via check_and_wait_while_suspended()
3390 
3391         slp->park(millis);
3392 
3393         // were we externally suspended while we were waiting?
3394         jt->check_and_wait_while_suspended();
3395       }
3396     }
3397   } else {
3398     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3399     jlong prevtime = javaTimeNanos();
3400 
3401     for (;;) {
3402       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3403       // the 1st iteration ...
3404       jlong newtime = javaTimeNanos();
3405 
3406       if (newtime - prevtime < 0) {
3407         // time moving backwards, should only happen if no monotonic clock
3408         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3409         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3410       } else {
3411         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3412       }
3413 
3414       if(millis <= 0) break ;
3415 
3416       prevtime = newtime;
3417       slp->park(millis);
3418     }
3419     return OS_OK ;
3420   }
3421 }
3422 
3423 int os::naked_sleep() {
3424   // %% make the sleep time an integer flag. for now use 1 millisec.
3425   return os::sleep(Thread::current(), 1, false);
3426 }
3427 
3428 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3429 void os::infinite_sleep() {
3430   while (true) {    // sleep forever ...
3431     ::sleep(100);   // ... 100 seconds at a time
3432   }
3433 }
3434 
3435 // Used to convert frequent JVM_Yield() to nops
3436 bool os::dont_yield() {
3437   return DontYieldALot;
3438 }
3439 
3440 void os::yield() {
3441   sched_yield();
3442 }
3443 
3444 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3445 
3446 void os::yield_all(int attempts) {
3447   // Yields to all threads, including threads with lower priorities
3448   // Threads on Linux are all with same priority. The Solaris style
3449   // os::yield_all() with nanosleep(1ms) is not necessary.
3450   sched_yield();
3451 }
3452 
3453 // Called from the tight loops to possibly influence time-sharing heuristics
3454 void os::loop_breaker(int attempts) {
3455   os::yield_all(attempts);
3456 }
3457 
3458 ////////////////////////////////////////////////////////////////////////////////
3459 // thread priority support
3460 
3461 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3462 // only supports dynamic priority, static priority must be zero. For real-time
3463 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3464 // However, for large multi-threaded applications, SCHED_RR is not only slower
3465 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3466 // of 5 runs - Sep 2005).
3467 //
3468 // The following code actually changes the niceness of kernel-thread/LWP. It
3469 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3470 // not the entire user process, and user level threads are 1:1 mapped to kernel
3471 // threads. It has always been the case, but could change in the future. For
3472 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3473 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3474 
3475 int os::java_to_os_priority[CriticalPriority + 1] = {
3476   19,              // 0 Entry should never be used
3477 
3478    4,              // 1 MinPriority
3479    3,              // 2
3480    2,              // 3
3481 
3482    1,              // 4
3483    0,              // 5 NormPriority
3484   -1,              // 6
3485 
3486   -2,              // 7
3487   -3,              // 8
3488   -4,              // 9 NearMaxPriority
3489 
3490   -5,              // 10 MaxPriority
3491 
3492   -5               // 11 CriticalPriority
3493 };
3494 
3495 static int prio_init() {
3496   if (ThreadPriorityPolicy == 1) {
3497     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3498     // if effective uid is not root. Perhaps, a more elegant way of doing
3499     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3500     if (geteuid() != 0) {
3501       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3502         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3503       }
3504       ThreadPriorityPolicy = 0;
3505     }
3506   }
3507   if (UseCriticalJavaThreadPriority) {
3508     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3509   }
3510   return 0;
3511 }
3512 
3513 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3514   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3515 
3516   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3517   return (ret == 0) ? OS_OK : OS_ERR;
3518 }
3519 
3520 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3521   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3522     *priority_ptr = java_to_os_priority[NormPriority];
3523     return OS_OK;
3524   }
3525 
3526   errno = 0;
3527   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3528   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3529 }
3530 
3531 // Hint to the underlying OS that a task switch would not be good.
3532 // Void return because it's a hint and can fail.
3533 void os::hint_no_preempt() {}
3534 
3535 ////////////////////////////////////////////////////////////////////////////////
3536 // suspend/resume support
3537 
3538 //  the low-level signal-based suspend/resume support is a remnant from the
3539 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3540 //  within hotspot. Now there is a single use-case for this:
3541 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3542 //      that runs in the watcher thread.
3543 //  The remaining code is greatly simplified from the more general suspension
3544 //  code that used to be used.
3545 //
3546 //  The protocol is quite simple:
3547 //  - suspend:
3548 //      - sends a signal to the target thread
3549 //      - polls the suspend state of the osthread using a yield loop
3550 //      - target thread signal handler (SR_handler) sets suspend state
3551 //        and blocks in sigsuspend until continued
3552 //  - resume:
3553 //      - sets target osthread state to continue
3554 //      - sends signal to end the sigsuspend loop in the SR_handler
3555 //
3556 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3557 //
3558 
3559 static void resume_clear_context(OSThread *osthread) {
3560   osthread->set_ucontext(NULL);
3561   osthread->set_siginfo(NULL);
3562 
3563   // notify the suspend action is completed, we have now resumed
3564   osthread->sr.clear_suspended();
3565 }
3566 
3567 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3568   osthread->set_ucontext(context);
3569   osthread->set_siginfo(siginfo);
3570 }
3571 
3572 //
3573 // Handler function invoked when a thread's execution is suspended or
3574 // resumed. We have to be careful that only async-safe functions are
3575 // called here (Note: most pthread functions are not async safe and
3576 // should be avoided.)
3577 //
3578 // Note: sigwait() is a more natural fit than sigsuspend() from an
3579 // interface point of view, but sigwait() prevents the signal hander
3580 // from being run. libpthread would get very confused by not having
3581 // its signal handlers run and prevents sigwait()'s use with the
3582 // mutex granting granting signal.
3583 //
3584 // Currently only ever called on the VMThread
3585 //
3586 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3587   // Save and restore errno to avoid confusing native code with EINTR
3588   // after sigsuspend.
3589   int old_errno = errno;
3590 
3591   Thread* thread = Thread::current();
3592   OSThread* osthread = thread->osthread();
3593   assert(thread->is_VM_thread(), "Must be VMThread");
3594   // read current suspend action
3595   int action = osthread->sr.suspend_action();
3596   if (action == os::Linux::SuspendResume::SR_SUSPEND) {
3597     suspend_save_context(osthread, siginfo, context);
3598 
3599     // Notify the suspend action is about to be completed. do_suspend()
3600     // waits until SR_SUSPENDED is set and then returns. We will wait
3601     // here for a resume signal and that completes the suspend-other
3602     // action. do_suspend/do_resume is always called as a pair from
3603     // the same thread - so there are no races
3604 
3605     // notify the caller
3606     osthread->sr.set_suspended();
3607 
3608     sigset_t suspend_set;  // signals for sigsuspend()
3609 
3610     // get current set of blocked signals and unblock resume signal
3611     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3612     sigdelset(&suspend_set, SR_signum);
3613 
3614     // wait here until we are resumed
3615     do {
3616       sigsuspend(&suspend_set);
3617       // ignore all returns until we get a resume signal
3618     } while (osthread->sr.suspend_action() != os::Linux::SuspendResume::SR_CONTINUE);
3619 
3620     resume_clear_context(osthread);
3621 
3622   } else {
3623     assert(action == os::Linux::SuspendResume::SR_CONTINUE, "unexpected sr action");
3624     // nothing special to do - just leave the handler
3625   }
3626 
3627   errno = old_errno;
3628 }
3629 
3630 
3631 static int SR_initialize() {
3632   struct sigaction act;
3633   char *s;
3634   /* Get signal number to use for suspend/resume */
3635   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3636     int sig = ::strtol(s, 0, 10);
3637     if (sig > 0 || sig < _NSIG) {
3638         SR_signum = sig;
3639     }
3640   }
3641 
3642   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3643         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3644 
3645   sigemptyset(&SR_sigset);
3646   sigaddset(&SR_sigset, SR_signum);
3647 
3648   /* Set up signal handler for suspend/resume */
3649   act.sa_flags = SA_RESTART|SA_SIGINFO;
3650   act.sa_handler = (void (*)(int)) SR_handler;
3651 
3652   // SR_signum is blocked by default.
3653   // 4528190 - We also need to block pthread restart signal (32 on all
3654   // supported Linux platforms). Note that LinuxThreads need to block
3655   // this signal for all threads to work properly. So we don't have
3656   // to use hard-coded signal number when setting up the mask.
3657   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3658 
3659   if (sigaction(SR_signum, &act, 0) == -1) {
3660     return -1;
3661   }
3662 
3663   // Save signal flag
3664   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3665   return 0;
3666 }
3667 
3668 
3669 // returns true on success and false on error - really an error is fatal
3670 // but this seems the normal response to library errors
3671 static bool do_suspend(OSThread* osthread) {
3672   // mark as suspended and send signal
3673   osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_SUSPEND);
3674   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3675   assert_status(status == 0, status, "pthread_kill");
3676 
3677   // check status and wait until notified of suspension
3678   if (status == 0) {
3679     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3680       os::yield_all(i);
3681     }
3682     osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3683     return true;
3684   }
3685   else {
3686     osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3687     return false;
3688   }
3689 }
3690 
3691 static void do_resume(OSThread* osthread) {
3692   assert(osthread->sr.is_suspended(), "thread should be suspended");
3693   osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_CONTINUE);
3694 
3695   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3696   assert_status(status == 0, status, "pthread_kill");
3697   // check status and wait unit notified of resumption
3698   if (status == 0) {
3699     for (int i = 0; osthread->sr.is_suspended(); i++) {
3700       os::yield_all(i);
3701     }
3702   }
3703   osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3704 }
3705 
3706 ////////////////////////////////////////////////////////////////////////////////
3707 // interrupt support
3708 
3709 void os::interrupt(Thread* thread) {
3710   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3711     "possibility of dangling Thread pointer");
3712 
3713   OSThread* osthread = thread->osthread();
3714 
3715   if (!osthread->interrupted()) {
3716     osthread->set_interrupted(true);
3717     // More than one thread can get here with the same value of osthread,
3718     // resulting in multiple notifications.  We do, however, want the store
3719     // to interrupted() to be visible to other threads before we execute unpark().
3720     OrderAccess::fence();
3721     ParkEvent * const slp = thread->_SleepEvent ;
3722     if (slp != NULL) slp->unpark() ;
3723   }
3724 
3725   // For JSR166. Unpark even if interrupt status already was set
3726   if (thread->is_Java_thread())
3727     ((JavaThread*)thread)->parker()->unpark();
3728 
3729   ParkEvent * ev = thread->_ParkEvent ;
3730   if (ev != NULL) ev->unpark() ;
3731 
3732 }
3733 
3734 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3735   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3736     "possibility of dangling Thread pointer");
3737 
3738   OSThread* osthread = thread->osthread();
3739 
3740   bool interrupted = osthread->interrupted();
3741 
3742   if (interrupted && clear_interrupted) {
3743     osthread->set_interrupted(false);
3744     // consider thread->_SleepEvent->reset() ... optional optimization
3745   }
3746 
3747   return interrupted;
3748 }
3749 
3750 ///////////////////////////////////////////////////////////////////////////////////
3751 // signal handling (except suspend/resume)
3752 
3753 // This routine may be used by user applications as a "hook" to catch signals.
3754 // The user-defined signal handler must pass unrecognized signals to this
3755 // routine, and if it returns true (non-zero), then the signal handler must
3756 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3757 // routine will never retun false (zero), but instead will execute a VM panic
3758 // routine kill the process.
3759 //
3760 // If this routine returns false, it is OK to call it again.  This allows
3761 // the user-defined signal handler to perform checks either before or after
3762 // the VM performs its own checks.  Naturally, the user code would be making
3763 // a serious error if it tried to handle an exception (such as a null check
3764 // or breakpoint) that the VM was generating for its own correct operation.
3765 //
3766 // This routine may recognize any of the following kinds of signals:
3767 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3768 // It should be consulted by handlers for any of those signals.
3769 //
3770 // The caller of this routine must pass in the three arguments supplied
3771 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3772 // field of the structure passed to sigaction().  This routine assumes that
3773 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3774 //
3775 // Note that the VM will print warnings if it detects conflicting signal
3776 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3777 //
3778 extern "C" JNIEXPORT int
3779 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3780                         void* ucontext, int abort_if_unrecognized);
3781 
3782 void signalHandler(int sig, siginfo_t* info, void* uc) {
3783   assert(info != NULL && uc != NULL, "it must be old kernel");
3784   int orig_errno = errno;  // Preserve errno value over signal handler.
3785   JVM_handle_linux_signal(sig, info, uc, true);
3786   errno = orig_errno;
3787 }
3788 
3789 
3790 // This boolean allows users to forward their own non-matching signals
3791 // to JVM_handle_linux_signal, harmlessly.
3792 bool os::Linux::signal_handlers_are_installed = false;
3793 
3794 // For signal-chaining
3795 struct sigaction os::Linux::sigact[MAXSIGNUM];
3796 unsigned int os::Linux::sigs = 0;
3797 bool os::Linux::libjsig_is_loaded = false;
3798 typedef struct sigaction *(*get_signal_t)(int);
3799 get_signal_t os::Linux::get_signal_action = NULL;
3800 
3801 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3802   struct sigaction *actp = NULL;
3803 
3804   if (libjsig_is_loaded) {
3805     // Retrieve the old signal handler from libjsig
3806     actp = (*get_signal_action)(sig);
3807   }
3808   if (actp == NULL) {
3809     // Retrieve the preinstalled signal handler from jvm
3810     actp = get_preinstalled_handler(sig);
3811   }
3812 
3813   return actp;
3814 }
3815 
3816 static bool call_chained_handler(struct sigaction *actp, int sig,
3817                                  siginfo_t *siginfo, void *context) {
3818   // Call the old signal handler
3819   if (actp->sa_handler == SIG_DFL) {
3820     // It's more reasonable to let jvm treat it as an unexpected exception
3821     // instead of taking the default action.
3822     return false;
3823   } else if (actp->sa_handler != SIG_IGN) {
3824     if ((actp->sa_flags & SA_NODEFER) == 0) {
3825       // automaticlly block the signal
3826       sigaddset(&(actp->sa_mask), sig);
3827     }
3828 
3829     sa_handler_t hand;
3830     sa_sigaction_t sa;
3831     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3832     // retrieve the chained handler
3833     if (siginfo_flag_set) {
3834       sa = actp->sa_sigaction;
3835     } else {
3836       hand = actp->sa_handler;
3837     }
3838 
3839     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3840       actp->sa_handler = SIG_DFL;
3841     }
3842 
3843     // try to honor the signal mask
3844     sigset_t oset;
3845     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3846 
3847     // call into the chained handler
3848     if (siginfo_flag_set) {
3849       (*sa)(sig, siginfo, context);
3850     } else {
3851       (*hand)(sig);
3852     }
3853 
3854     // restore the signal mask
3855     pthread_sigmask(SIG_SETMASK, &oset, 0);
3856   }
3857   // Tell jvm's signal handler the signal is taken care of.
3858   return true;
3859 }
3860 
3861 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3862   bool chained = false;
3863   // signal-chaining
3864   if (UseSignalChaining) {
3865     struct sigaction *actp = get_chained_signal_action(sig);
3866     if (actp != NULL) {
3867       chained = call_chained_handler(actp, sig, siginfo, context);
3868     }
3869   }
3870   return chained;
3871 }
3872 
3873 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3874   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3875     return &sigact[sig];
3876   }
3877   return NULL;
3878 }
3879 
3880 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3881   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3882   sigact[sig] = oldAct;
3883   sigs |= (unsigned int)1 << sig;
3884 }
3885 
3886 // for diagnostic
3887 int os::Linux::sigflags[MAXSIGNUM];
3888 
3889 int os::Linux::get_our_sigflags(int sig) {
3890   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3891   return sigflags[sig];
3892 }
3893 
3894 void os::Linux::set_our_sigflags(int sig, int flags) {
3895   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3896   sigflags[sig] = flags;
3897 }
3898 
3899 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3900   // Check for overwrite.
3901   struct sigaction oldAct;
3902   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3903 
3904   void* oldhand = oldAct.sa_sigaction
3905                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3906                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3907   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3908       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3909       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3910     if (AllowUserSignalHandlers || !set_installed) {
3911       // Do not overwrite; user takes responsibility to forward to us.
3912       return;
3913     } else if (UseSignalChaining) {
3914       // save the old handler in jvm
3915       save_preinstalled_handler(sig, oldAct);
3916       // libjsig also interposes the sigaction() call below and saves the
3917       // old sigaction on it own.
3918     } else {
3919       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3920                     "%#lx for signal %d.", (long)oldhand, sig));
3921     }
3922   }
3923 
3924   struct sigaction sigAct;
3925   sigfillset(&(sigAct.sa_mask));
3926   sigAct.sa_handler = SIG_DFL;
3927   if (!set_installed) {
3928     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3929   } else {
3930     sigAct.sa_sigaction = signalHandler;
3931     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3932   }
3933   // Save flags, which are set by ours
3934   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3935   sigflags[sig] = sigAct.sa_flags;
3936 
3937   int ret = sigaction(sig, &sigAct, &oldAct);
3938   assert(ret == 0, "check");
3939 
3940   void* oldhand2  = oldAct.sa_sigaction
3941                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3942                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3943   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3944 }
3945 
3946 // install signal handlers for signals that HotSpot needs to
3947 // handle in order to support Java-level exception handling.
3948 
3949 void os::Linux::install_signal_handlers() {
3950   if (!signal_handlers_are_installed) {
3951     signal_handlers_are_installed = true;
3952 
3953     // signal-chaining
3954     typedef void (*signal_setting_t)();
3955     signal_setting_t begin_signal_setting = NULL;
3956     signal_setting_t end_signal_setting = NULL;
3957     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3958                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3959     if (begin_signal_setting != NULL) {
3960       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3961                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3962       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3963                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3964       libjsig_is_loaded = true;
3965       assert(UseSignalChaining, "should enable signal-chaining");
3966     }
3967     if (libjsig_is_loaded) {
3968       // Tell libjsig jvm is setting signal handlers
3969       (*begin_signal_setting)();
3970     }
3971 
3972     set_signal_handler(SIGSEGV, true);
3973     set_signal_handler(SIGPIPE, true);
3974     set_signal_handler(SIGBUS, true);
3975     set_signal_handler(SIGILL, true);
3976     set_signal_handler(SIGFPE, true);
3977     set_signal_handler(SIGXFSZ, true);
3978 
3979     if (libjsig_is_loaded) {
3980       // Tell libjsig jvm finishes setting signal handlers
3981       (*end_signal_setting)();
3982     }
3983 
3984     // We don't activate signal checker if libjsig is in place, we trust ourselves
3985     // and if UserSignalHandler is installed all bets are off.
3986     // Log that signal checking is off only if -verbose:jni is specified.
3987     if (CheckJNICalls) {
3988       if (libjsig_is_loaded) {
3989         if (PrintJNIResolving) {
3990           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3991         }
3992         check_signals = false;
3993       }
3994       if (AllowUserSignalHandlers) {
3995         if (PrintJNIResolving) {
3996           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3997         }
3998         check_signals = false;
3999       }
4000     }
4001   }
4002 }
4003 
4004 // This is the fastest way to get thread cpu time on Linux.
4005 // Returns cpu time (user+sys) for any thread, not only for current.
4006 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4007 // It might work on 2.6.10+ with a special kernel/glibc patch.
4008 // For reference, please, see IEEE Std 1003.1-2004:
4009 //   http://www.unix.org/single_unix_specification
4010 
4011 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4012   struct timespec tp;
4013   int rc = os::Linux::clock_gettime(clockid, &tp);
4014   assert(rc == 0, "clock_gettime is expected to return 0 code");
4015 
4016   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4017 }
4018 
4019 /////
4020 // glibc on Linux platform uses non-documented flag
4021 // to indicate, that some special sort of signal
4022 // trampoline is used.
4023 // We will never set this flag, and we should
4024 // ignore this flag in our diagnostic
4025 #ifdef SIGNIFICANT_SIGNAL_MASK
4026 #undef SIGNIFICANT_SIGNAL_MASK
4027 #endif
4028 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4029 
4030 static const char* get_signal_handler_name(address handler,
4031                                            char* buf, int buflen) {
4032   int offset;
4033   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4034   if (found) {
4035     // skip directory names
4036     const char *p1, *p2;
4037     p1 = buf;
4038     size_t len = strlen(os::file_separator());
4039     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4040     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4041   } else {
4042     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4043   }
4044   return buf;
4045 }
4046 
4047 static void print_signal_handler(outputStream* st, int sig,
4048                                  char* buf, size_t buflen) {
4049   struct sigaction sa;
4050 
4051   sigaction(sig, NULL, &sa);
4052 
4053   // See comment for SIGNIFICANT_SIGNAL_MASK define
4054   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4055 
4056   st->print("%s: ", os::exception_name(sig, buf, buflen));
4057 
4058   address handler = (sa.sa_flags & SA_SIGINFO)
4059     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4060     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4061 
4062   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4063     st->print("SIG_DFL");
4064   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4065     st->print("SIG_IGN");
4066   } else {
4067     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4068   }
4069 
4070   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4071 
4072   address rh = VMError::get_resetted_sighandler(sig);
4073   // May be, handler was resetted by VMError?
4074   if(rh != NULL) {
4075     handler = rh;
4076     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4077   }
4078 
4079   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
4080 
4081   // Check: is it our handler?
4082   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4083      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4084     // It is our signal handler
4085     // check for flags, reset system-used one!
4086     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4087       st->print(
4088                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4089                 os::Linux::get_our_sigflags(sig));
4090     }
4091   }
4092   st->cr();
4093 }
4094 
4095 
4096 #define DO_SIGNAL_CHECK(sig) \
4097   if (!sigismember(&check_signal_done, sig)) \
4098     os::Linux::check_signal_handler(sig)
4099 
4100 // This method is a periodic task to check for misbehaving JNI applications
4101 // under CheckJNI, we can add any periodic checks here
4102 
4103 void os::run_periodic_checks() {
4104 
4105   if (check_signals == false) return;
4106 
4107   // SEGV and BUS if overridden could potentially prevent
4108   // generation of hs*.log in the event of a crash, debugging
4109   // such a case can be very challenging, so we absolutely
4110   // check the following for a good measure:
4111   DO_SIGNAL_CHECK(SIGSEGV);
4112   DO_SIGNAL_CHECK(SIGILL);
4113   DO_SIGNAL_CHECK(SIGFPE);
4114   DO_SIGNAL_CHECK(SIGBUS);
4115   DO_SIGNAL_CHECK(SIGPIPE);
4116   DO_SIGNAL_CHECK(SIGXFSZ);
4117 
4118 
4119   // ReduceSignalUsage allows the user to override these handlers
4120   // see comments at the very top and jvm_solaris.h
4121   if (!ReduceSignalUsage) {
4122     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4123     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4124     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4125     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4126   }
4127 
4128   DO_SIGNAL_CHECK(SR_signum);
4129   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4130 }
4131 
4132 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4133 
4134 static os_sigaction_t os_sigaction = NULL;
4135 
4136 void os::Linux::check_signal_handler(int sig) {
4137   char buf[O_BUFLEN];
4138   address jvmHandler = NULL;
4139 
4140 
4141   struct sigaction act;
4142   if (os_sigaction == NULL) {
4143     // only trust the default sigaction, in case it has been interposed
4144     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4145     if (os_sigaction == NULL) return;
4146   }
4147 
4148   os_sigaction(sig, (struct sigaction*)NULL, &act);
4149 
4150 
4151   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4152 
4153   address thisHandler = (act.sa_flags & SA_SIGINFO)
4154     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4155     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4156 
4157 
4158   switch(sig) {
4159   case SIGSEGV:
4160   case SIGBUS:
4161   case SIGFPE:
4162   case SIGPIPE:
4163   case SIGILL:
4164   case SIGXFSZ:
4165     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4166     break;
4167 
4168   case SHUTDOWN1_SIGNAL:
4169   case SHUTDOWN2_SIGNAL:
4170   case SHUTDOWN3_SIGNAL:
4171   case BREAK_SIGNAL:
4172     jvmHandler = (address)user_handler();
4173     break;
4174 
4175   case INTERRUPT_SIGNAL:
4176     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4177     break;
4178 
4179   default:
4180     if (sig == SR_signum) {
4181       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4182     } else {
4183       return;
4184     }
4185     break;
4186   }
4187 
4188   if (thisHandler != jvmHandler) {
4189     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4190     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4191     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4192     // No need to check this sig any longer
4193     sigaddset(&check_signal_done, sig);
4194   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4195     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4196     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4197     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4198     // No need to check this sig any longer
4199     sigaddset(&check_signal_done, sig);
4200   }
4201 
4202   // Dump all the signal
4203   if (sigismember(&check_signal_done, sig)) {
4204     print_signal_handlers(tty, buf, O_BUFLEN);
4205   }
4206 }
4207 
4208 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4209 
4210 extern bool signal_name(int signo, char* buf, size_t len);
4211 
4212 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4213   if (0 < exception_code && exception_code <= SIGRTMAX) {
4214     // signal
4215     if (!signal_name(exception_code, buf, size)) {
4216       jio_snprintf(buf, size, "SIG%d", exception_code);
4217     }
4218     return buf;
4219   } else {
4220     return NULL;
4221   }
4222 }
4223 
4224 // this is called _before_ the most of global arguments have been parsed
4225 void os::init(void) {
4226   char dummy;   /* used to get a guess on initial stack address */
4227 //  first_hrtime = gethrtime();
4228 
4229   // With LinuxThreads the JavaMain thread pid (primordial thread)
4230   // is different than the pid of the java launcher thread.
4231   // So, on Linux, the launcher thread pid is passed to the VM
4232   // via the sun.java.launcher.pid property.
4233   // Use this property instead of getpid() if it was correctly passed.
4234   // See bug 6351349.
4235   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4236 
4237   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4238 
4239   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4240 
4241   init_random(1234567);
4242 
4243   ThreadCritical::initialize();
4244 
4245   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4246   if (Linux::page_size() == -1) {
4247     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4248                   strerror(errno)));
4249   }
4250   init_page_sizes((size_t) Linux::page_size());
4251 
4252   Linux::initialize_system_info();
4253 
4254   // main_thread points to the aboriginal thread
4255   Linux::_main_thread = pthread_self();
4256 
4257   Linux::clock_init();
4258   initial_time_count = os::elapsed_counter();
4259   pthread_mutex_init(&dl_mutex, NULL);
4260 
4261   // If the pagesize of the VM is greater than 8K determine the appropriate
4262   // number of initial guard pages.  The user can change this with the
4263   // command line arguments, if needed.
4264   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4265     StackYellowPages = 1;
4266     StackRedPages = 1;
4267     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4268   }
4269 }
4270 
4271 // To install functions for atexit system call
4272 extern "C" {
4273   static void perfMemory_exit_helper() {
4274     perfMemory_exit();
4275   }
4276 }
4277 
4278 // this is called _after_ the global arguments have been parsed
4279 jint os::init_2(void)
4280 {
4281   Linux::fast_thread_clock_init();
4282 
4283   // Allocate a single page and mark it as readable for safepoint polling
4284   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4285   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4286 
4287   os::set_polling_page( polling_page );
4288 
4289 #ifndef PRODUCT
4290   if(Verbose && PrintMiscellaneous)
4291     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4292 #endif
4293 
4294   if (!UseMembar) {
4295     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4296     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4297     os::set_memory_serialize_page( mem_serialize_page );
4298 
4299 #ifndef PRODUCT
4300     if(Verbose && PrintMiscellaneous)
4301       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4302 #endif
4303   }
4304 
4305   // initialize suspend/resume support - must do this before signal_sets_init()
4306   if (SR_initialize() != 0) {
4307     perror("SR_initialize failed");
4308     return JNI_ERR;
4309   }
4310 
4311   Linux::signal_sets_init();
4312   Linux::install_signal_handlers();
4313 
4314   // Check minimum allowable stack size for thread creation and to initialize
4315   // the java system classes, including StackOverflowError - depends on page
4316   // size.  Add a page for compiler2 recursion in main thread.
4317   // Add in 2*BytesPerWord times page size to account for VM stack during
4318   // class initialization depending on 32 or 64 bit VM.
4319   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4320             (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4321                     (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4322 
4323   size_t threadStackSizeInBytes = ThreadStackSize * K;
4324   if (threadStackSizeInBytes != 0 &&
4325       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4326         tty->print_cr("\nThe stack size specified is too small, "
4327                       "Specify at least %dk",
4328                       os::Linux::min_stack_allowed/ K);
4329         return JNI_ERR;
4330   }
4331 
4332   // Make the stack size a multiple of the page size so that
4333   // the yellow/red zones can be guarded.
4334   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4335         vm_page_size()));
4336 
4337   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4338 
4339   Linux::libpthread_init();
4340   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4341      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4342           Linux::glibc_version(), Linux::libpthread_version(),
4343           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4344   }
4345 
4346   if (UseNUMA) {
4347     if (!Linux::libnuma_init()) {
4348       UseNUMA = false;
4349     } else {
4350       if ((Linux::numa_max_node() < 1)) {
4351         // There's only one node(they start from 0), disable NUMA.
4352         UseNUMA = false;
4353       }
4354     }
4355     // With SHM large pages we cannot uncommit a page, so there's not way
4356     // we can make the adaptive lgrp chunk resizing work. If the user specified
4357     // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4358     // disable adaptive resizing.
4359     if (UseNUMA && UseLargePages && UseSHM) {
4360       if (!FLAG_IS_DEFAULT(UseNUMA)) {
4361         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4362           UseLargePages = false;
4363         } else {
4364           warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4365           UseAdaptiveSizePolicy = false;
4366           UseAdaptiveNUMAChunkSizing = false;
4367         }
4368       } else {
4369         UseNUMA = false;
4370       }
4371     }
4372     if (!UseNUMA && ForceNUMA) {
4373       UseNUMA = true;
4374     }
4375   }
4376 
4377   if (MaxFDLimit) {
4378     // set the number of file descriptors to max. print out error
4379     // if getrlimit/setrlimit fails but continue regardless.
4380     struct rlimit nbr_files;
4381     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4382     if (status != 0) {
4383       if (PrintMiscellaneous && (Verbose || WizardMode))
4384         perror("os::init_2 getrlimit failed");
4385     } else {
4386       nbr_files.rlim_cur = nbr_files.rlim_max;
4387       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4388       if (status != 0) {
4389         if (PrintMiscellaneous && (Verbose || WizardMode))
4390           perror("os::init_2 setrlimit failed");
4391       }
4392     }
4393   }
4394 
4395   // Initialize lock used to serialize thread creation (see os::create_thread)
4396   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4397 
4398   // at-exit methods are called in the reverse order of their registration.
4399   // atexit functions are called on return from main or as a result of a
4400   // call to exit(3C). There can be only 32 of these functions registered
4401   // and atexit() does not set errno.
4402 
4403   if (PerfAllowAtExitRegistration) {
4404     // only register atexit functions if PerfAllowAtExitRegistration is set.
4405     // atexit functions can be delayed until process exit time, which
4406     // can be problematic for embedded VM situations. Embedded VMs should
4407     // call DestroyJavaVM() to assure that VM resources are released.
4408 
4409     // note: perfMemory_exit_helper atexit function may be removed in
4410     // the future if the appropriate cleanup code can be added to the
4411     // VM_Exit VMOperation's doit method.
4412     if (atexit(perfMemory_exit_helper) != 0) {
4413       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4414     }
4415   }
4416 
4417   // initialize thread priority policy
4418   prio_init();
4419 
4420   return JNI_OK;
4421 }
4422 
4423 // this is called at the end of vm_initialization
4424 void os::init_3(void)
4425 {
4426 #ifdef JAVASE_EMBEDDED
4427   // Start the MemNotifyThread
4428   if (LowMemoryProtection) {
4429     MemNotifyThread::start();
4430   }
4431   return;
4432 #endif
4433 }
4434 
4435 // Mark the polling page as unreadable
4436 void os::make_polling_page_unreadable(void) {
4437   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4438     fatal("Could not disable polling page");
4439 };
4440 
4441 // Mark the polling page as readable
4442 void os::make_polling_page_readable(void) {
4443   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4444     fatal("Could not enable polling page");
4445   }
4446 };
4447 
4448 int os::active_processor_count() {
4449   // Linux doesn't yet have a (official) notion of processor sets,
4450   // so just return the number of online processors.
4451   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4452   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4453   return online_cpus;
4454 }
4455 
4456 void os::set_native_thread_name(const char *name) {
4457   // Not yet implemented.
4458   return;
4459 }
4460 
4461 bool os::distribute_processes(uint length, uint* distribution) {
4462   // Not yet implemented.
4463   return false;
4464 }
4465 
4466 bool os::bind_to_processor(uint processor_id) {
4467   // Not yet implemented.
4468   return false;
4469 }
4470 
4471 ///
4472 
4473 // Suspends the target using the signal mechanism and then grabs the PC before
4474 // resuming the target. Used by the flat-profiler only
4475 ExtendedPC os::get_thread_pc(Thread* thread) {
4476   // Make sure that it is called by the watcher for the VMThread
4477   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4478   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4479 
4480   ExtendedPC epc;
4481 
4482   OSThread* osthread = thread->osthread();
4483   if (do_suspend(osthread)) {
4484     if (osthread->ucontext() != NULL) {
4485       epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4486     } else {
4487       // NULL context is unexpected, double-check this is the VMThread
4488       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4489     }
4490     do_resume(osthread);
4491   }
4492   // failure means pthread_kill failed for some reason - arguably this is
4493   // a fatal problem, but such problems are ignored elsewhere
4494 
4495   return epc;
4496 }
4497 
4498 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4499 {
4500    if (is_NPTL()) {
4501       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4502    } else {
4503       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4504       // word back to default 64bit precision if condvar is signaled. Java
4505       // wants 53bit precision.  Save and restore current value.
4506       int fpu = get_fpu_control_word();
4507       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4508       set_fpu_control_word(fpu);
4509       return status;
4510    }
4511 }
4512 
4513 ////////////////////////////////////////////////////////////////////////////////
4514 // debug support
4515 
4516 bool os::find(address addr, outputStream* st) {
4517   Dl_info dlinfo;
4518   memset(&dlinfo, 0, sizeof(dlinfo));
4519   if (dladdr(addr, &dlinfo)) {
4520     st->print(PTR_FORMAT ": ", addr);
4521     if (dlinfo.dli_sname != NULL) {
4522       st->print("%s+%#x", dlinfo.dli_sname,
4523                  addr - (intptr_t)dlinfo.dli_saddr);
4524     } else if (dlinfo.dli_fname) {
4525       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4526     } else {
4527       st->print("<absolute address>");
4528     }
4529     if (dlinfo.dli_fname) {
4530       st->print(" in %s", dlinfo.dli_fname);
4531     }
4532     if (dlinfo.dli_fbase) {
4533       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4534     }
4535     st->cr();
4536 
4537     if (Verbose) {
4538       // decode some bytes around the PC
4539       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4540       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4541       address       lowest = (address) dlinfo.dli_sname;
4542       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4543       if (begin < lowest)  begin = lowest;
4544       Dl_info dlinfo2;
4545       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4546           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4547         end = (address) dlinfo2.dli_saddr;
4548       Disassembler::decode(begin, end, st);
4549     }
4550     return true;
4551   }
4552   return false;
4553 }
4554 
4555 ////////////////////////////////////////////////////////////////////////////////
4556 // misc
4557 
4558 // This does not do anything on Linux. This is basically a hook for being
4559 // able to use structured exception handling (thread-local exception filters)
4560 // on, e.g., Win32.
4561 void
4562 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4563                          JavaCallArguments* args, Thread* thread) {
4564   f(value, method, args, thread);
4565 }
4566 
4567 void os::print_statistics() {
4568 }
4569 
4570 int os::message_box(const char* title, const char* message) {
4571   int i;
4572   fdStream err(defaultStream::error_fd());
4573   for (i = 0; i < 78; i++) err.print_raw("=");
4574   err.cr();
4575   err.print_raw_cr(title);
4576   for (i = 0; i < 78; i++) err.print_raw("-");
4577   err.cr();
4578   err.print_raw_cr(message);
4579   for (i = 0; i < 78; i++) err.print_raw("=");
4580   err.cr();
4581 
4582   char buf[16];
4583   // Prevent process from exiting upon "read error" without consuming all CPU
4584   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4585 
4586   return buf[0] == 'y' || buf[0] == 'Y';
4587 }
4588 
4589 int os::stat(const char *path, struct stat *sbuf) {
4590   char pathbuf[MAX_PATH];
4591   if (strlen(path) > MAX_PATH - 1) {
4592     errno = ENAMETOOLONG;
4593     return -1;
4594   }
4595   os::native_path(strcpy(pathbuf, path));
4596   return ::stat(pathbuf, sbuf);
4597 }
4598 
4599 bool os::check_heap(bool force) {
4600   return true;
4601 }
4602 
4603 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4604   return ::vsnprintf(buf, count, format, args);
4605 }
4606 
4607 // Is a (classpath) directory empty?
4608 bool os::dir_is_empty(const char* path) {
4609   DIR *dir = NULL;
4610   struct dirent *ptr;
4611 
4612   dir = opendir(path);
4613   if (dir == NULL) return true;
4614 
4615   /* Scan the directory */
4616   bool result = true;
4617   char buf[sizeof(struct dirent) + MAX_PATH];
4618   while (result && (ptr = ::readdir(dir)) != NULL) {
4619     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4620       result = false;
4621     }
4622   }
4623   closedir(dir);
4624   return result;
4625 }
4626 
4627 // This code originates from JDK's sysOpen and open64_w
4628 // from src/solaris/hpi/src/system_md.c
4629 
4630 #ifndef O_DELETE
4631 #define O_DELETE 0x10000
4632 #endif
4633 
4634 // Open a file. Unlink the file immediately after open returns
4635 // if the specified oflag has the O_DELETE flag set.
4636 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4637 
4638 int os::open(const char *path, int oflag, int mode) {
4639 
4640   if (strlen(path) > MAX_PATH - 1) {
4641     errno = ENAMETOOLONG;
4642     return -1;
4643   }
4644   int fd;
4645   int o_delete = (oflag & O_DELETE);
4646   oflag = oflag & ~O_DELETE;
4647 
4648   fd = ::open64(path, oflag, mode);
4649   if (fd == -1) return -1;
4650 
4651   //If the open succeeded, the file might still be a directory
4652   {
4653     struct stat64 buf64;
4654     int ret = ::fstat64(fd, &buf64);
4655     int st_mode = buf64.st_mode;
4656 
4657     if (ret != -1) {
4658       if ((st_mode & S_IFMT) == S_IFDIR) {
4659         errno = EISDIR;
4660         ::close(fd);
4661         return -1;
4662       }
4663     } else {
4664       ::close(fd);
4665       return -1;
4666     }
4667   }
4668 
4669     /*
4670      * All file descriptors that are opened in the JVM and not
4671      * specifically destined for a subprocess should have the
4672      * close-on-exec flag set.  If we don't set it, then careless 3rd
4673      * party native code might fork and exec without closing all
4674      * appropriate file descriptors (e.g. as we do in closeDescriptors in
4675      * UNIXProcess.c), and this in turn might:
4676      *
4677      * - cause end-of-file to fail to be detected on some file
4678      *   descriptors, resulting in mysterious hangs, or
4679      *
4680      * - might cause an fopen in the subprocess to fail on a system
4681      *   suffering from bug 1085341.
4682      *
4683      * (Yes, the default setting of the close-on-exec flag is a Unix
4684      * design flaw)
4685      *
4686      * See:
4687      * 1085341: 32-bit stdio routines should support file descriptors >255
4688      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4689      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4690      */
4691 #ifdef FD_CLOEXEC
4692     {
4693         int flags = ::fcntl(fd, F_GETFD);
4694         if (flags != -1)
4695             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4696     }
4697 #endif
4698 
4699   if (o_delete != 0) {
4700     ::unlink(path);
4701   }
4702   return fd;
4703 }
4704 
4705 
4706 // create binary file, rewriting existing file if required
4707 int os::create_binary_file(const char* path, bool rewrite_existing) {
4708   int oflags = O_WRONLY | O_CREAT;
4709   if (!rewrite_existing) {
4710     oflags |= O_EXCL;
4711   }
4712   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4713 }
4714 
4715 // return current position of file pointer
4716 jlong os::current_file_offset(int fd) {
4717   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4718 }
4719 
4720 // move file pointer to the specified offset
4721 jlong os::seek_to_file_offset(int fd, jlong offset) {
4722   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4723 }
4724 
4725 // This code originates from JDK's sysAvailable
4726 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4727 
4728 int os::available(int fd, jlong *bytes) {
4729   jlong cur, end;
4730   int mode;
4731   struct stat64 buf64;
4732 
4733   if (::fstat64(fd, &buf64) >= 0) {
4734     mode = buf64.st_mode;
4735     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4736       /*
4737       * XXX: is the following call interruptible? If so, this might
4738       * need to go through the INTERRUPT_IO() wrapper as for other
4739       * blocking, interruptible calls in this file.
4740       */
4741       int n;
4742       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4743         *bytes = n;
4744         return 1;
4745       }
4746     }
4747   }
4748   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4749     return 0;
4750   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4751     return 0;
4752   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4753     return 0;
4754   }
4755   *bytes = end - cur;
4756   return 1;
4757 }
4758 
4759 int os::socket_available(int fd, jint *pbytes) {
4760   // Linux doc says EINTR not returned, unlike Solaris
4761   int ret = ::ioctl(fd, FIONREAD, pbytes);
4762 
4763   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4764   // is expected to return 0 on failure and 1 on success to the jdk.
4765   return (ret < 0) ? 0 : 1;
4766 }
4767 
4768 // Map a block of memory.
4769 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4770                      char *addr, size_t bytes, bool read_only,
4771                      bool allow_exec) {
4772   int prot;
4773   int flags = MAP_PRIVATE;
4774 
4775   if (read_only) {
4776     prot = PROT_READ;
4777   } else {
4778     prot = PROT_READ | PROT_WRITE;
4779   }
4780 
4781   if (allow_exec) {
4782     prot |= PROT_EXEC;
4783   }
4784 
4785   if (addr != NULL) {
4786     flags |= MAP_FIXED;
4787   }
4788 
4789   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4790                                      fd, file_offset);
4791   if (mapped_address == MAP_FAILED) {
4792     return NULL;
4793   }
4794   return mapped_address;
4795 }
4796 
4797 
4798 // Remap a block of memory.
4799 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4800                        char *addr, size_t bytes, bool read_only,
4801                        bool allow_exec) {
4802   // same as map_memory() on this OS
4803   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4804                         allow_exec);
4805 }
4806 
4807 
4808 // Unmap a block of memory.
4809 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4810   return munmap(addr, bytes) == 0;
4811 }
4812 
4813 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4814 
4815 static clockid_t thread_cpu_clockid(Thread* thread) {
4816   pthread_t tid = thread->osthread()->pthread_id();
4817   clockid_t clockid;
4818 
4819   // Get thread clockid
4820   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4821   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4822   return clockid;
4823 }
4824 
4825 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4826 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4827 // of a thread.
4828 //
4829 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4830 // the fast estimate available on the platform.
4831 
4832 jlong os::current_thread_cpu_time() {
4833   if (os::Linux::supports_fast_thread_cpu_time()) {
4834     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4835   } else {
4836     // return user + sys since the cost is the same
4837     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4838   }
4839 }
4840 
4841 jlong os::thread_cpu_time(Thread* thread) {
4842   // consistent with what current_thread_cpu_time() returns
4843   if (os::Linux::supports_fast_thread_cpu_time()) {
4844     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4845   } else {
4846     return slow_thread_cpu_time(thread, true /* user + sys */);
4847   }
4848 }
4849 
4850 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4851   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4852     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4853   } else {
4854     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4855   }
4856 }
4857 
4858 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4859   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4860     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4861   } else {
4862     return slow_thread_cpu_time(thread, user_sys_cpu_time);
4863   }
4864 }
4865 
4866 //
4867 //  -1 on error.
4868 //
4869 
4870 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4871   static bool proc_task_unchecked = true;
4872   static const char *proc_stat_path = "/proc/%d/stat";
4873   pid_t  tid = thread->osthread()->thread_id();
4874   char *s;
4875   char stat[2048];
4876   int statlen;
4877   char proc_name[64];
4878   int count;
4879   long sys_time, user_time;
4880   char cdummy;
4881   int idummy;
4882   long ldummy;
4883   FILE *fp;
4884 
4885   // The /proc/<tid>/stat aggregates per-process usage on
4886   // new Linux kernels 2.6+ where NPTL is supported.
4887   // The /proc/self/task/<tid>/stat still has the per-thread usage.
4888   // See bug 6328462.
4889   // There possibly can be cases where there is no directory
4890   // /proc/self/task, so we check its availability.
4891   if (proc_task_unchecked && os::Linux::is_NPTL()) {
4892     // This is executed only once
4893     proc_task_unchecked = false;
4894     fp = fopen("/proc/self/task", "r");
4895     if (fp != NULL) {
4896       proc_stat_path = "/proc/self/task/%d/stat";
4897       fclose(fp);
4898     }
4899   }
4900 
4901   sprintf(proc_name, proc_stat_path, tid);
4902   fp = fopen(proc_name, "r");
4903   if ( fp == NULL ) return -1;
4904   statlen = fread(stat, 1, 2047, fp);
4905   stat[statlen] = '\0';
4906   fclose(fp);
4907 
4908   // Skip pid and the command string. Note that we could be dealing with
4909   // weird command names, e.g. user could decide to rename java launcher
4910   // to "java 1.4.2 :)", then the stat file would look like
4911   //                1234 (java 1.4.2 :)) R ... ...
4912   // We don't really need to know the command string, just find the last
4913   // occurrence of ")" and then start parsing from there. See bug 4726580.
4914   s = strrchr(stat, ')');
4915   if (s == NULL ) return -1;
4916 
4917   // Skip blank chars
4918   do s++; while (isspace(*s));
4919 
4920   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4921                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4922                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4923                  &user_time, &sys_time);
4924   if ( count != 13 ) return -1;
4925   if (user_sys_cpu_time) {
4926     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4927   } else {
4928     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4929   }
4930 }
4931 
4932 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4933   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4934   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4935   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4936   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4937 }
4938 
4939 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4940   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
4941   info_ptr->may_skip_backward = false;     // elapsed time not wall time
4942   info_ptr->may_skip_forward = false;      // elapsed time not wall time
4943   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
4944 }
4945 
4946 bool os::is_thread_cpu_time_supported() {
4947   return true;
4948 }
4949 
4950 // System loadavg support.  Returns -1 if load average cannot be obtained.
4951 // Linux doesn't yet have a (official) notion of processor sets,
4952 // so just return the system wide load average.
4953 int os::loadavg(double loadavg[], int nelem) {
4954   return ::getloadavg(loadavg, nelem);
4955 }
4956 
4957 void os::pause() {
4958   char filename[MAX_PATH];
4959   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4960     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4961   } else {
4962     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4963   }
4964 
4965   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4966   if (fd != -1) {
4967     struct stat buf;
4968     ::close(fd);
4969     while (::stat(filename, &buf) == 0) {
4970       (void)::poll(NULL, 0, 100);
4971     }
4972   } else {
4973     jio_fprintf(stderr,
4974       "Could not open pause file '%s', continuing immediately.\n", filename);
4975   }
4976 }
4977 
4978 
4979 // Refer to the comments in os_solaris.cpp park-unpark.
4980 //
4981 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4982 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4983 // For specifics regarding the bug see GLIBC BUGID 261237 :
4984 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4985 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4986 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4987 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
4988 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4989 // and monitorenter when we're using 1-0 locking.  All those operations may result in
4990 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
4991 // of libpthread avoids the problem, but isn't practical.
4992 //
4993 // Possible remedies:
4994 //
4995 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
4996 //      This is palliative and probabilistic, however.  If the thread is preempted
4997 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
4998 //      than the minimum period may have passed, and the abstime may be stale (in the
4999 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5000 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5001 //
5002 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5003 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5004 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5005 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5006 //      thread.
5007 //
5008 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5009 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5010 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5011 //      This also works well.  In fact it avoids kernel-level scalability impediments
5012 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5013 //      timers in a graceful fashion.
5014 //
5015 // 4.   When the abstime value is in the past it appears that control returns
5016 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5017 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5018 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5019 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5020 //      It may be possible to avoid reinitialization by checking the return
5021 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5022 //      condvar we must establish the invariant that cond_signal() is only called
5023 //      within critical sections protected by the adjunct mutex.  This prevents
5024 //      cond_signal() from "seeing" a condvar that's in the midst of being
5025 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5026 //      desirable signal-after-unlock optimization that avoids futile context switching.
5027 //
5028 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5029 //      structure when a condvar is used or initialized.  cond_destroy()  would
5030 //      release the helper structure.  Our reinitialize-after-timedwait fix
5031 //      put excessive stress on malloc/free and locks protecting the c-heap.
5032 //
5033 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5034 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5035 // and only enabling the work-around for vulnerable environments.
5036 
5037 // utility to compute the abstime argument to timedwait:
5038 // millis is the relative timeout time
5039 // abstime will be the absolute timeout time
5040 // TODO: replace compute_abstime() with unpackTime()
5041 
5042 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5043   if (millis < 0)  millis = 0;
5044   struct timeval now;
5045   int status = gettimeofday(&now, NULL);
5046   assert(status == 0, "gettimeofday");
5047   jlong seconds = millis / 1000;
5048   millis %= 1000;
5049   if (seconds > 50000000) { // see man cond_timedwait(3T)
5050     seconds = 50000000;
5051   }
5052   abstime->tv_sec = now.tv_sec  + seconds;
5053   long       usec = now.tv_usec + millis * 1000;
5054   if (usec >= 1000000) {
5055     abstime->tv_sec += 1;
5056     usec -= 1000000;
5057   }
5058   abstime->tv_nsec = usec * 1000;
5059   return abstime;
5060 }
5061 
5062 
5063 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5064 // Conceptually TryPark() should be equivalent to park(0).
5065 
5066 int os::PlatformEvent::TryPark() {
5067   for (;;) {
5068     const int v = _Event ;
5069     guarantee ((v == 0) || (v == 1), "invariant") ;
5070     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5071   }
5072 }
5073 
5074 void os::PlatformEvent::park() {       // AKA "down()"
5075   // Invariant: Only the thread associated with the Event/PlatformEvent
5076   // may call park().
5077   // TODO: assert that _Assoc != NULL or _Assoc == Self
5078   int v ;
5079   for (;;) {
5080       v = _Event ;
5081       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5082   }
5083   guarantee (v >= 0, "invariant") ;
5084   if (v == 0) {
5085      // Do this the hard way by blocking ...
5086      int status = pthread_mutex_lock(_mutex);
5087      assert_status(status == 0, status, "mutex_lock");
5088      guarantee (_nParked == 0, "invariant") ;
5089      ++ _nParked ;
5090      while (_Event < 0) {
5091         status = pthread_cond_wait(_cond, _mutex);
5092         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5093         // Treat this the same as if the wait was interrupted
5094         if (status == ETIME) { status = EINTR; }
5095         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5096      }
5097      -- _nParked ;
5098 
5099     _Event = 0 ;
5100      status = pthread_mutex_unlock(_mutex);
5101      assert_status(status == 0, status, "mutex_unlock");
5102     // Paranoia to ensure our locked and lock-free paths interact
5103     // correctly with each other.
5104     OrderAccess::fence();
5105   }
5106   guarantee (_Event >= 0, "invariant") ;
5107 }
5108 
5109 int os::PlatformEvent::park(jlong millis) {
5110   guarantee (_nParked == 0, "invariant") ;
5111 
5112   int v ;
5113   for (;;) {
5114       v = _Event ;
5115       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5116   }
5117   guarantee (v >= 0, "invariant") ;
5118   if (v != 0) return OS_OK ;
5119 
5120   // We do this the hard way, by blocking the thread.
5121   // Consider enforcing a minimum timeout value.
5122   struct timespec abst;
5123   compute_abstime(&abst, millis);
5124 
5125   int ret = OS_TIMEOUT;
5126   int status = pthread_mutex_lock(_mutex);
5127   assert_status(status == 0, status, "mutex_lock");
5128   guarantee (_nParked == 0, "invariant") ;
5129   ++_nParked ;
5130 
5131   // Object.wait(timo) will return because of
5132   // (a) notification
5133   // (b) timeout
5134   // (c) thread.interrupt
5135   //
5136   // Thread.interrupt and object.notify{All} both call Event::set.
5137   // That is, we treat thread.interrupt as a special case of notification.
5138   // The underlying Solaris implementation, cond_timedwait, admits
5139   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5140   // JVM from making those visible to Java code.  As such, we must
5141   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5142   //
5143   // TODO: properly differentiate simultaneous notify+interrupt.
5144   // In that case, we should propagate the notify to another waiter.
5145 
5146   while (_Event < 0) {
5147     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5148     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5149       pthread_cond_destroy (_cond);
5150       pthread_cond_init (_cond, NULL) ;
5151     }
5152     assert_status(status == 0 || status == EINTR ||
5153                   status == ETIME || status == ETIMEDOUT,
5154                   status, "cond_timedwait");
5155     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5156     if (status == ETIME || status == ETIMEDOUT) break ;
5157     // We consume and ignore EINTR and spurious wakeups.
5158   }
5159   --_nParked ;
5160   if (_Event >= 0) {
5161      ret = OS_OK;
5162   }
5163   _Event = 0 ;
5164   status = pthread_mutex_unlock(_mutex);
5165   assert_status(status == 0, status, "mutex_unlock");
5166   assert (_nParked == 0, "invariant") ;
5167   // Paranoia to ensure our locked and lock-free paths interact
5168   // correctly with each other.
5169   OrderAccess::fence();
5170   return ret;
5171 }
5172 
5173 void os::PlatformEvent::unpark() {
5174   // Transitions for _Event:
5175   //    0 :=> 1
5176   //    1 :=> 1
5177   //   -1 :=> either 0 or 1; must signal target thread
5178   //          That is, we can safely transition _Event from -1 to either
5179   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
5180   //          unpark() calls.
5181   // See also: "Semaphores in Plan 9" by Mullender & Cox
5182   //
5183   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5184   // that it will take two back-to-back park() calls for the owning
5185   // thread to block. This has the benefit of forcing a spurious return
5186   // from the first park() call after an unpark() call which will help
5187   // shake out uses of park() and unpark() without condition variables.
5188 
5189   if (Atomic::xchg(1, &_Event) >= 0) return;
5190 
5191   // Wait for the thread associated with the event to vacate
5192   int status = pthread_mutex_lock(_mutex);
5193   assert_status(status == 0, status, "mutex_lock");
5194   int AnyWaiters = _nParked;
5195   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5196   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5197     AnyWaiters = 0;
5198     pthread_cond_signal(_cond);
5199   }
5200   status = pthread_mutex_unlock(_mutex);
5201   assert_status(status == 0, status, "mutex_unlock");
5202   if (AnyWaiters != 0) {
5203     status = pthread_cond_signal(_cond);
5204     assert_status(status == 0, status, "cond_signal");
5205   }
5206 
5207   // Note that we signal() _after dropping the lock for "immortal" Events.
5208   // This is safe and avoids a common class of  futile wakeups.  In rare
5209   // circumstances this can cause a thread to return prematurely from
5210   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5211   // simply re-test the condition and re-park itself.
5212 }
5213 
5214 
5215 // JSR166
5216 // -------------------------------------------------------
5217 
5218 /*
5219  * The solaris and linux implementations of park/unpark are fairly
5220  * conservative for now, but can be improved. They currently use a
5221  * mutex/condvar pair, plus a a count.
5222  * Park decrements count if > 0, else does a condvar wait.  Unpark
5223  * sets count to 1 and signals condvar.  Only one thread ever waits
5224  * on the condvar. Contention seen when trying to park implies that someone
5225  * is unparking you, so don't wait. And spurious returns are fine, so there
5226  * is no need to track notifications.
5227  */
5228 
5229 #define MAX_SECS 100000000
5230 /*
5231  * This code is common to linux and solaris and will be moved to a
5232  * common place in dolphin.
5233  *
5234  * The passed in time value is either a relative time in nanoseconds
5235  * or an absolute time in milliseconds. Either way it has to be unpacked
5236  * into suitable seconds and nanoseconds components and stored in the
5237  * given timespec structure.
5238  * Given time is a 64-bit value and the time_t used in the timespec is only
5239  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5240  * overflow if times way in the future are given. Further on Solaris versions
5241  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5242  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5243  * As it will be 28 years before "now + 100000000" will overflow we can
5244  * ignore overflow and just impose a hard-limit on seconds using the value
5245  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5246  * years from "now".
5247  */
5248 
5249 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5250   assert (time > 0, "convertTime");
5251 
5252   struct timeval now;
5253   int status = gettimeofday(&now, NULL);
5254   assert(status == 0, "gettimeofday");
5255 
5256   time_t max_secs = now.tv_sec + MAX_SECS;
5257 
5258   if (isAbsolute) {
5259     jlong secs = time / 1000;
5260     if (secs > max_secs) {
5261       absTime->tv_sec = max_secs;
5262     }
5263     else {
5264       absTime->tv_sec = secs;
5265     }
5266     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5267   }
5268   else {
5269     jlong secs = time / NANOSECS_PER_SEC;
5270     if (secs >= MAX_SECS) {
5271       absTime->tv_sec = max_secs;
5272       absTime->tv_nsec = 0;
5273     }
5274     else {
5275       absTime->tv_sec = now.tv_sec + secs;
5276       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5277       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5278         absTime->tv_nsec -= NANOSECS_PER_SEC;
5279         ++absTime->tv_sec; // note: this must be <= max_secs
5280       }
5281     }
5282   }
5283   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5284   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5285   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5286   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5287 }
5288 
5289 void Parker::park(bool isAbsolute, jlong time) {
5290   // Ideally we'd do something useful while spinning, such
5291   // as calling unpackTime().
5292 
5293   // Optional fast-path check:
5294   // Return immediately if a permit is available.
5295   // We depend on Atomic::xchg() having full barrier semantics
5296   // since we are doing a lock-free update to _counter.
5297   if (Atomic::xchg(0, &_counter) > 0) return;
5298 
5299   Thread* thread = Thread::current();
5300   assert(thread->is_Java_thread(), "Must be JavaThread");
5301   JavaThread *jt = (JavaThread *)thread;
5302 
5303   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5304   // Check interrupt before trying to wait
5305   if (Thread::is_interrupted(thread, false)) {
5306     return;
5307   }
5308 
5309   // Next, demultiplex/decode time arguments
5310   timespec absTime;
5311   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5312     return;
5313   }
5314   if (time > 0) {
5315     unpackTime(&absTime, isAbsolute, time);
5316   }
5317 
5318 
5319   // Enter safepoint region
5320   // Beware of deadlocks such as 6317397.
5321   // The per-thread Parker:: mutex is a classic leaf-lock.
5322   // In particular a thread must never block on the Threads_lock while
5323   // holding the Parker:: mutex.  If safepoints are pending both the
5324   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5325   ThreadBlockInVM tbivm(jt);
5326 
5327   // Don't wait if cannot get lock since interference arises from
5328   // unblocking.  Also. check interrupt before trying wait
5329   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5330     return;
5331   }
5332 
5333   int status ;
5334   if (_counter > 0)  { // no wait needed
5335     _counter = 0;
5336     status = pthread_mutex_unlock(_mutex);
5337     assert (status == 0, "invariant") ;
5338     // Paranoia to ensure our locked and lock-free paths interact
5339     // correctly with each other and Java-level accesses.
5340     OrderAccess::fence();
5341     return;
5342   }
5343 
5344 #ifdef ASSERT
5345   // Don't catch signals while blocked; let the running threads have the signals.
5346   // (This allows a debugger to break into the running thread.)
5347   sigset_t oldsigs;
5348   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5349   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5350 #endif
5351 
5352   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5353   jt->set_suspend_equivalent();
5354   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5355 
5356   if (time == 0) {
5357     status = pthread_cond_wait (_cond, _mutex) ;
5358   } else {
5359     status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5360     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5361       pthread_cond_destroy (_cond) ;
5362       pthread_cond_init    (_cond, NULL);
5363     }
5364   }
5365   assert_status(status == 0 || status == EINTR ||
5366                 status == ETIME || status == ETIMEDOUT,
5367                 status, "cond_timedwait");
5368 
5369 #ifdef ASSERT
5370   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5371 #endif
5372 
5373   _counter = 0 ;
5374   status = pthread_mutex_unlock(_mutex) ;
5375   assert_status(status == 0, status, "invariant") ;
5376   // Paranoia to ensure our locked and lock-free paths interact
5377   // correctly with each other and Java-level accesses.
5378   OrderAccess::fence();
5379 
5380   // If externally suspended while waiting, re-suspend
5381   if (jt->handle_special_suspend_equivalent_condition()) {
5382     jt->java_suspend_self();
5383   }
5384 }
5385 
5386 void Parker::unpark() {
5387   int s, status ;
5388   status = pthread_mutex_lock(_mutex);
5389   assert (status == 0, "invariant") ;
5390   s = _counter;
5391   _counter = 1;
5392   if (s < 1) {
5393      if (WorkAroundNPTLTimedWaitHang) {
5394         status = pthread_cond_signal (_cond) ;
5395         assert (status == 0, "invariant") ;
5396         status = pthread_mutex_unlock(_mutex);
5397         assert (status == 0, "invariant") ;
5398      } else {
5399         status = pthread_mutex_unlock(_mutex);
5400         assert (status == 0, "invariant") ;
5401         status = pthread_cond_signal (_cond) ;
5402         assert (status == 0, "invariant") ;
5403      }
5404   } else {
5405     pthread_mutex_unlock(_mutex);
5406     assert (status == 0, "invariant") ;
5407   }
5408 }
5409 
5410 
5411 extern char** environ;
5412 
5413 #ifndef __NR_fork
5414 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5415 #endif
5416 
5417 #ifndef __NR_execve
5418 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5419 #endif
5420 
5421 // Run the specified command in a separate process. Return its exit value,
5422 // or -1 on failure (e.g. can't fork a new process).
5423 // Unlike system(), this function can be called from signal handler. It
5424 // doesn't block SIGINT et al.
5425 int os::fork_and_exec(char* cmd) {
5426   const char * argv[4] = {"sh", "-c", cmd, NULL};
5427 
5428   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5429   // pthread_atfork handlers and reset pthread library. All we need is a
5430   // separate process to execve. Make a direct syscall to fork process.
5431   // On IA64 there's no fork syscall, we have to use fork() and hope for
5432   // the best...
5433   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5434               IA64_ONLY(fork();)
5435 
5436   if (pid < 0) {
5437     // fork failed
5438     return -1;
5439 
5440   } else if (pid == 0) {
5441     // child process
5442 
5443     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5444     // first to kill every thread on the thread list. Because this list is
5445     // not reset by fork() (see notes above), execve() will instead kill
5446     // every thread in the parent process. We know this is the only thread
5447     // in the new process, so make a system call directly.
5448     // IA64 should use normal execve() from glibc to match the glibc fork()
5449     // above.
5450     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5451     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5452 
5453     // execve failed
5454     _exit(-1);
5455 
5456   } else  {
5457     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5458     // care about the actual exit code, for now.
5459 
5460     int status;
5461 
5462     // Wait for the child process to exit.  This returns immediately if
5463     // the child has already exited. */
5464     while (waitpid(pid, &status, 0) < 0) {
5465         switch (errno) {
5466         case ECHILD: return 0;
5467         case EINTR: break;
5468         default: return -1;
5469         }
5470     }
5471 
5472     if (WIFEXITED(status)) {
5473        // The child exited normally; get its exit code.
5474        return WEXITSTATUS(status);
5475     } else if (WIFSIGNALED(status)) {
5476        // The child exited because of a signal
5477        // The best value to return is 0x80 + signal number,
5478        // because that is what all Unix shells do, and because
5479        // it allows callers to distinguish between process exit and
5480        // process death by signal.
5481        return 0x80 + WTERMSIG(status);
5482     } else {
5483        // Unknown exit code; pass it through
5484        return status;
5485     }
5486   }
5487 }
5488 
5489 // is_headless_jre()
5490 //
5491 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5492 // in order to report if we are running in a headless jre
5493 //
5494 // Since JDK8 xawt/libmawt.so was moved into the same directory
5495 // as libawt.so, and renamed libawt_xawt.so
5496 //
5497 bool os::is_headless_jre() {
5498     struct stat statbuf;
5499     char buf[MAXPATHLEN];
5500     char libmawtpath[MAXPATHLEN];
5501     const char *xawtstr  = "/xawt/libmawt.so";
5502     const char *new_xawtstr = "/libawt_xawt.so";
5503     char *p;
5504 
5505     // Get path to libjvm.so
5506     os::jvm_path(buf, sizeof(buf));
5507 
5508     // Get rid of libjvm.so
5509     p = strrchr(buf, '/');
5510     if (p == NULL) return false;
5511     else *p = '\0';
5512 
5513     // Get rid of client or server
5514     p = strrchr(buf, '/');
5515     if (p == NULL) return false;
5516     else *p = '\0';
5517 
5518     // check xawt/libmawt.so
5519     strcpy(libmawtpath, buf);
5520     strcat(libmawtpath, xawtstr);
5521     if (::stat(libmawtpath, &statbuf) == 0) return false;
5522 
5523     // check libawt_xawt.so
5524     strcpy(libmawtpath, buf);
5525     strcat(libmawtpath, new_xawtstr);
5526     if (::stat(libmawtpath, &statbuf) == 0) return false;
5527 
5528     return true;
5529 }
5530 
5531 // Get the default path to the core file
5532 // Returns the length of the string
5533 int os::get_core_path(char* buffer, size_t bufferSize) {
5534   const char* p = get_current_directory(buffer, bufferSize);
5535 
5536   if (p == NULL) {
5537     assert(p != NULL, "failed to get current directory");
5538     return 0;
5539   }
5540 
5541   return strlen(buffer);
5542 }
5543 
5544 #ifdef JAVASE_EMBEDDED
5545 //
5546 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5547 //
5548 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5549 
5550 // ctor
5551 //
5552 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5553   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5554   _fd = fd;
5555 
5556   if (os::create_thread(this, os::os_thread)) {
5557     _memnotify_thread = this;
5558     os::set_priority(this, NearMaxPriority);
5559     os::start_thread(this);
5560   }
5561 }
5562 
5563 // Where all the work gets done
5564 //
5565 void MemNotifyThread::run() {
5566   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5567 
5568   // Set up the select arguments
5569   fd_set rfds;
5570   if (_fd != -1) {
5571     FD_ZERO(&rfds);
5572     FD_SET(_fd, &rfds);
5573   }
5574 
5575   // Now wait for the mem_notify device to wake up
5576   while (1) {
5577     // Wait for the mem_notify device to signal us..
5578     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5579     if (rc == -1) {
5580       perror("select!\n");
5581       break;
5582     } else if (rc) {
5583       //ssize_t free_before = os::available_memory();
5584       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5585 
5586       // The kernel is telling us there is not much memory left...
5587       // try to do something about that
5588 
5589       // If we are not already in a GC, try one.
5590       if (!Universe::heap()->is_gc_active()) {
5591         Universe::heap()->collect(GCCause::_allocation_failure);
5592 
5593         //ssize_t free_after = os::available_memory();
5594         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5595         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5596       }
5597       // We might want to do something like the following if we find the GC's are not helping...
5598       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5599     }
5600   }
5601 }
5602 
5603 //
5604 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5605 //
5606 void MemNotifyThread::start() {
5607   int    fd;
5608   fd = open ("/dev/mem_notify", O_RDONLY, 0);
5609   if (fd < 0) {
5610       return;
5611   }
5612 
5613   if (memnotify_thread() == NULL) {
5614     new MemNotifyThread(fd);
5615   }
5616 }
5617 #endif // JAVASE_EMBEDDED