1 /*
   2  * Copyright (c) 1999, 2014, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "interpreter/interpreter.hpp"
  33 #include "jvm_linux.h"
  34 #include "memory/allocation.inline.hpp"
  35 #include "memory/filemap.hpp"
  36 #include "mutex_linux.inline.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "os_share_linux.hpp"
  39 #include "prims/jniFastGetField.hpp"
  40 #include "prims/jvm.h"
  41 #include "prims/jvm_misc.hpp"
  42 #include "runtime/arguments.hpp"
  43 #include "runtime/extendedPC.hpp"
  44 #include "runtime/globals.hpp"
  45 #include "runtime/interfaceSupport.hpp"
  46 #include "runtime/init.hpp"
  47 #include "runtime/java.hpp"
  48 #include "runtime/javaCalls.hpp"
  49 #include "runtime/mutexLocker.hpp"
  50 #include "runtime/objectMonitor.hpp"
  51 #include "runtime/osThread.hpp"
  52 #include "runtime/perfMemory.hpp"
  53 #include "runtime/sharedRuntime.hpp"
  54 #include "runtime/statSampler.hpp"
  55 #include "runtime/stubRoutines.hpp"
  56 #include "runtime/threadCritical.hpp"
  57 #include "runtime/timer.hpp"
  58 #include "services/attachListener.hpp"
  59 #include "services/memTracker.hpp"
  60 #include "services/runtimeService.hpp"
  61 #include "thread_linux.inline.hpp"
  62 #include "utilities/decoder.hpp"
  63 #include "utilities/defaultStream.hpp"
  64 #include "utilities/events.hpp"
  65 #include "utilities/elfFile.hpp"
  66 #include "utilities/growableArray.hpp"
  67 #include "utilities/vmError.hpp"
  68 #ifdef TARGET_ARCH_x86
  69 # include "assembler_x86.inline.hpp"
  70 # include "nativeInst_x86.hpp"
  71 #endif
  72 #ifdef TARGET_ARCH_aarch64
  73 # include "assembler_aarch64.inline.hpp"
  74 # include "nativeInst_aarch64.hpp"
  75 #endif
  76 #ifdef TARGET_ARCH_sparc
  77 # include "assembler_sparc.inline.hpp"
  78 # include "nativeInst_sparc.hpp"
  79 #endif
  80 #ifdef TARGET_ARCH_zero
  81 # include "assembler_zero.inline.hpp"
  82 # include "nativeInst_zero.hpp"
  83 #endif
  84 #ifdef TARGET_ARCH_arm
  85 # include "assembler_arm.inline.hpp"
  86 # include "nativeInst_arm.hpp"
  87 #endif
  88 #ifdef TARGET_ARCH_ppc
  89 # include "assembler_ppc.inline.hpp"
  90 # include "nativeInst_ppc.hpp"
  91 #endif
  92 
  93 // put OS-includes here
  94 # include <sys/types.h>
  95 # include <sys/mman.h>
  96 # include <sys/stat.h>
  97 # include <sys/select.h>
  98 # include <pthread.h>
  99 # include <signal.h>
 100 # include <errno.h>
 101 # include <dlfcn.h>
 102 # include <stdio.h>
 103 # include <unistd.h>
 104 # include <sys/resource.h>
 105 # include <pthread.h>
 106 # include <sys/stat.h>
 107 # include <sys/time.h>
 108 # include <sys/times.h>
 109 # include <sys/utsname.h>
 110 # include <sys/socket.h>
 111 # include <sys/wait.h>
 112 # include <pwd.h>
 113 # include <poll.h>
 114 # include <semaphore.h>
 115 # include <fcntl.h>
 116 # include <string.h>
 117 # include <syscall.h>
 118 # include <sys/sysinfo.h>
 119 # include <gnu/libc-version.h>
 120 # include <sys/ipc.h>
 121 # include <sys/shm.h>
 122 # include <link.h>
 123 # include <stdint.h>
 124 # include <inttypes.h>
 125 # include <sys/ioctl.h>
 126 
 127 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 128 // getrusage() is prepared to handle the associated failure.
 129 #ifndef RUSAGE_THREAD
 130 #define RUSAGE_THREAD   (1)               /* only the calling thread */
 131 #endif
 132 
 133 #define MAX_PATH    (2 * K)
 134 
 135 #define MAX_SECS 100000000
 136 
 137 // for timer info max values which include all bits
 138 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 139 
 140 #define LARGEPAGES_BIT (1 << 6)
 141 
 142 #ifndef EM_AARCH64
 143 #define EM_AARCH64      183             /* ARM AARCH64 */
 144 #endif
 145 
 146 ////////////////////////////////////////////////////////////////////////////////
 147 // global variables
 148 julong os::Linux::_physical_memory = 0;
 149 
 150 address   os::Linux::_initial_thread_stack_bottom = NULL;
 151 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 152 
 153 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 154 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 155 Mutex* os::Linux::_createThread_lock = NULL;
 156 pthread_t os::Linux::_main_thread;
 157 int os::Linux::_page_size = -1;
 158 const int os::Linux::_vm_default_page_size = (8 * K);
 159 bool os::Linux::_is_floating_stack = false;
 160 bool os::Linux::_is_NPTL = false;
 161 bool os::Linux::_supports_fast_thread_cpu_time = false;
 162 const char * os::Linux::_glibc_version = NULL;
 163 const char * os::Linux::_libpthread_version = NULL;
 164 pthread_condattr_t os::Linux::_condattr[1];
 165 
 166 static jlong initial_time_count=0;
 167 
 168 static int clock_tics_per_sec = 100;
 169 
 170 // For diagnostics to print a message once. see run_periodic_checks
 171 static sigset_t check_signal_done;
 172 static bool check_signals = true;
 173 
 174 static pid_t _initial_pid = 0;
 175 
 176 /* Signal number used to suspend/resume a thread */
 177 
 178 /* do not use any signal number less than SIGSEGV, see 4355769 */
 179 static int SR_signum = SIGUSR2;
 180 sigset_t SR_sigset;
 181 
 182 /* Used to protect dlsym() calls */
 183 static pthread_mutex_t dl_mutex;
 184 
 185 // Declarations
 186 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 187 
 188 #ifdef JAVASE_EMBEDDED
 189 class MemNotifyThread: public Thread {
 190   friend class VMStructs;
 191  public:
 192   virtual void run();
 193 
 194  private:
 195   static MemNotifyThread* _memnotify_thread;
 196   int _fd;
 197 
 198  public:
 199 
 200   // Constructor
 201   MemNotifyThread(int fd);
 202 
 203   // Tester
 204   bool is_memnotify_thread() const { return true; }
 205 
 206   // Printing
 207   char* name() const { return (char*)"Linux MemNotify Thread"; }
 208 
 209   // Returns the single instance of the MemNotifyThread
 210   static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
 211 
 212   // Create and start the single instance of MemNotifyThread
 213   static void start();
 214 };
 215 #endif // JAVASE_EMBEDDED
 216 
 217 // utility functions
 218 
 219 static int SR_initialize();
 220 static int SR_finalize();
 221 
 222 julong os::available_memory() {
 223   return Linux::available_memory();
 224 }
 225 
 226 julong os::Linux::available_memory() {
 227   // values in struct sysinfo are "unsigned long"
 228   struct sysinfo si;
 229   sysinfo(&si);
 230 
 231   return (julong)si.freeram * si.mem_unit;
 232 }
 233 
 234 julong os::physical_memory() {
 235   return Linux::physical_memory();
 236 }
 237 
 238 julong os::allocatable_physical_memory(julong size) {
 239 #ifdef _LP64
 240   return size;
 241 #else
 242   julong result = MIN2(size, (julong)3800*M);
 243    if (!is_allocatable(result)) {
 244      // See comments under solaris for alignment considerations
 245      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 246      result =  MIN2(size, reasonable_size);
 247    }
 248    return result;
 249 #endif // _LP64
 250 }
 251 
 252 ////////////////////////////////////////////////////////////////////////////////
 253 // environment support
 254 
 255 bool os::getenv(const char* name, char* buf, int len) {
 256   const char* val = ::getenv(name);
 257   if (val != NULL && strlen(val) < (size_t)len) {
 258     strcpy(buf, val);
 259     return true;
 260   }
 261   if (len > 0) buf[0] = 0;  // return a null string
 262   return false;
 263 }
 264 
 265 
 266 // Return true if user is running as root.
 267 
 268 bool os::have_special_privileges() {
 269   static bool init = false;
 270   static bool privileges = false;
 271   if (!init) {
 272     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 273     init = true;
 274   }
 275   return privileges;
 276 }
 277 
 278 
 279 #ifndef SYS_gettid
 280 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 281 #ifdef __ia64__
 282 #define SYS_gettid 1105
 283 #elif __i386__
 284 #define SYS_gettid 224
 285 #elif __amd64__
 286 #define SYS_gettid 186
 287 #elif __sparc__
 288 #define SYS_gettid 143
 289 #else
 290 #error define gettid for the arch
 291 #endif
 292 #endif
 293 
 294 // Cpu architecture string
 295 #if   defined(ZERO)
 296 static char cpu_arch[] = ZERO_LIBARCH;
 297 #elif defined(IA64)
 298 static char cpu_arch[] = "ia64";
 299 #elif defined(IA32)
 300 static char cpu_arch[] = "i386";
 301 #elif defined(AMD64)
 302 static char cpu_arch[] = "amd64";
 303 #elif defined(ARM)
 304 static char cpu_arch[] = "arm";
 305 #elif defined(PPC32)
 306 static char cpu_arch[] = "ppc";
 307 #elif defined(PPC64)
 308 static char cpu_arch[] = "ppc64";
 309 #elif defined(SPARC)
 310 #  ifdef _LP64
 311 static char cpu_arch[] = "sparcv9";
 312 #  else
 313 static char cpu_arch[] = "sparc";
 314 #  endif
 315 #elif defined(AARCH64)
 316 static char cpu_arch[] = "aarch64";
 317 #else
 318 #error Add appropriate cpu_arch setting
 319 #endif
 320 
 321 
 322 // pid_t gettid()
 323 //
 324 // Returns the kernel thread id of the currently running thread. Kernel
 325 // thread id is used to access /proc.
 326 //
 327 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 328 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 329 //
 330 pid_t os::Linux::gettid() {
 331   int rslt = syscall(SYS_gettid);
 332   if (rslt == -1) {
 333      // old kernel, no NPTL support
 334      return getpid();
 335   } else {
 336      return (pid_t)rslt;
 337   }
 338 }
 339 
 340 // Most versions of linux have a bug where the number of processors are
 341 // determined by looking at the /proc file system.  In a chroot environment,
 342 // the system call returns 1.  This causes the VM to act as if it is
 343 // a single processor and elide locking (see is_MP() call).
 344 static bool unsafe_chroot_detected = false;
 345 static const char *unstable_chroot_error = "/proc file system not found.\n"
 346                      "Java may be unstable running multithreaded in a chroot "
 347                      "environment on Linux when /proc filesystem is not mounted.";
 348 
 349 void os::Linux::initialize_system_info() {
 350   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 351   if (processor_count() == 1) {
 352     pid_t pid = os::Linux::gettid();
 353     char fname[32];
 354     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 355     FILE *fp = fopen(fname, "r");
 356     if (fp == NULL) {
 357       unsafe_chroot_detected = true;
 358     } else {
 359       fclose(fp);
 360     }
 361   }
 362   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 363   assert(processor_count() > 0, "linux error");
 364 }
 365 
 366 void os::init_system_properties_values() {
 367 //  char arch[12];
 368 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 369 
 370   // The next steps are taken in the product version:
 371   //
 372   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 373   // This library should be located at:
 374   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 375   //
 376   // If "/jre/lib/" appears at the right place in the path, then we
 377   // assume libjvm[_g].so is installed in a JDK and we use this path.
 378   //
 379   // Otherwise exit with message: "Could not create the Java virtual machine."
 380   //
 381   // The following extra steps are taken in the debugging version:
 382   //
 383   // If "/jre/lib/" does NOT appear at the right place in the path
 384   // instead of exit check for $JAVA_HOME environment variable.
 385   //
 386   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 387   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 388   // it looks like libjvm[_g].so is installed there
 389   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 390   //
 391   // Otherwise exit.
 392   //
 393   // Important note: if the location of libjvm.so changes this
 394   // code needs to be changed accordingly.
 395 
 396   // The next few definitions allow the code to be verbatim:
 397 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
 398 #define getenv(n) ::getenv(n)
 399 
 400 /*
 401  * See ld(1):
 402  *      The linker uses the following search paths to locate required
 403  *      shared libraries:
 404  *        1: ...
 405  *        ...
 406  *        7: The default directories, normally /lib and /usr/lib.
 407  */
 408 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390) || defined(AARCH64)) || defined(BUILTIN_SIM)
 409 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 410 #else
 411 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 412 #endif
 413 
 414 #define EXTENSIONS_DIR  "/lib/ext"
 415 #define ENDORSED_DIR    "/lib/endorsed"
 416 #define REG_DIR         "/usr/java/packages"
 417 
 418   {
 419     /* sysclasspath, java_home, dll_dir */
 420     {
 421         char *home_path;
 422         char *dll_path;
 423         char *pslash;
 424         char buf[MAXPATHLEN];
 425         os::jvm_path(buf, sizeof(buf));
 426 
 427         // Found the full path to libjvm.so.
 428         // Now cut the path to <java_home>/jre if we can.
 429         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 430         pslash = strrchr(buf, '/');
 431         if (pslash != NULL)
 432             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 433         dll_path = malloc(strlen(buf) + 1);
 434         if (dll_path == NULL)
 435             return;
 436         strcpy(dll_path, buf);
 437         Arguments::set_dll_dir(dll_path);
 438 
 439         if (pslash != NULL) {
 440             pslash = strrchr(buf, '/');
 441             if (pslash != NULL) {
 442                 *pslash = '\0';       /* get rid of /<arch> */
 443                 pslash = strrchr(buf, '/');
 444                 if (pslash != NULL)
 445                     *pslash = '\0';   /* get rid of /lib */
 446             }
 447         }
 448 
 449         home_path = malloc(strlen(buf) + 1);
 450         if (home_path == NULL)
 451             return;
 452         strcpy(home_path, buf);
 453         Arguments::set_java_home(home_path);
 454 
 455         if (!set_boot_path('/', ':'))
 456             return;
 457     }
 458 
 459     /*
 460      * Where to look for native libraries
 461      *
 462      * Note: Due to a legacy implementation, most of the library path
 463      * is set in the launcher.  This was to accomodate linking restrictions
 464      * on legacy Linux implementations (which are no longer supported).
 465      * Eventually, all the library path setting will be done here.
 466      *
 467      * However, to prevent the proliferation of improperly built native
 468      * libraries, the new path component /usr/java/packages is added here.
 469      * Eventually, all the library path setting will be done here.
 470      */
 471     {
 472         char *ld_library_path;
 473 
 474         /*
 475          * Construct the invariant part of ld_library_path. Note that the
 476          * space for the colon and the trailing null are provided by the
 477          * nulls included by the sizeof operator (so actually we allocate
 478          * a byte more than necessary).
 479          */
 480         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 481             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 482         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 483 
 484         /*
 485          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 486          * should always exist (until the legacy problem cited above is
 487          * addressed).
 488          */
 489         char *v = getenv("LD_LIBRARY_PATH");
 490         if (v != NULL) {
 491             char *t = ld_library_path;
 492             /* That's +1 for the colon and +1 for the trailing '\0' */
 493             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 494             sprintf(ld_library_path, "%s:%s", v, t);
 495         }
 496         Arguments::set_library_path(ld_library_path);
 497     }
 498 
 499     /*
 500      * Extensions directories.
 501      *
 502      * Note that the space for the colon and the trailing null are provided
 503      * by the nulls included by the sizeof operator (so actually one byte more
 504      * than necessary is allocated).
 505      */
 506     {
 507         char *buf = malloc(strlen(Arguments::get_java_home()) +
 508             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 509         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 510             Arguments::get_java_home());
 511         Arguments::set_ext_dirs(buf);
 512     }
 513 
 514     /* Endorsed standards default directory. */
 515     {
 516         char * buf;
 517         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 518         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 519         Arguments::set_endorsed_dirs(buf);
 520     }
 521   }
 522 
 523 #undef malloc
 524 #undef getenv
 525 #undef EXTENSIONS_DIR
 526 #undef ENDORSED_DIR
 527 
 528   // Done
 529   return;
 530 }
 531 
 532 ////////////////////////////////////////////////////////////////////////////////
 533 // breakpoint support
 534 
 535 void os::breakpoint() {
 536   BREAKPOINT;
 537 }
 538 
 539 extern "C" void breakpoint() {
 540   // use debugger to set breakpoint here
 541 }
 542 
 543 ////////////////////////////////////////////////////////////////////////////////
 544 // signal support
 545 
 546 debug_only(static bool signal_sets_initialized = false);
 547 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 548 
 549 bool os::Linux::is_sig_ignored(int sig) {
 550       struct sigaction oact;
 551       sigaction(sig, (struct sigaction*)NULL, &oact);
 552       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 553                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 554       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 555            return true;
 556       else
 557            return false;
 558 }
 559 
 560 void os::Linux::signal_sets_init() {
 561   // Should also have an assertion stating we are still single-threaded.
 562   assert(!signal_sets_initialized, "Already initialized");
 563   // Fill in signals that are necessarily unblocked for all threads in
 564   // the VM. Currently, we unblock the following signals:
 565   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 566   //                         by -Xrs (=ReduceSignalUsage));
 567   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 568   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 569   // the dispositions or masks wrt these signals.
 570   // Programs embedding the VM that want to use the above signals for their
 571   // own purposes must, at this time, use the "-Xrs" option to prevent
 572   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 573   // (See bug 4345157, and other related bugs).
 574   // In reality, though, unblocking these signals is really a nop, since
 575   // these signals are not blocked by default.
 576   sigemptyset(&unblocked_sigs);
 577   sigemptyset(&allowdebug_blocked_sigs);
 578   sigaddset(&unblocked_sigs, SIGILL);
 579   sigaddset(&unblocked_sigs, SIGSEGV);
 580   sigaddset(&unblocked_sigs, SIGBUS);
 581   sigaddset(&unblocked_sigs, SIGFPE);
 582 #if defined(PPC64)
 583   sigaddset(&unblocked_sigs, SIGTRAP);
 584 #endif
 585   sigaddset(&unblocked_sigs, SR_signum);
 586 
 587   if (!ReduceSignalUsage) {
 588    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 589       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 590       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 591    }
 592    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 593       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 594       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 595    }
 596    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 597       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 598       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 599    }
 600   }
 601   // Fill in signals that are blocked by all but the VM thread.
 602   sigemptyset(&vm_sigs);
 603   if (!ReduceSignalUsage)
 604     sigaddset(&vm_sigs, BREAK_SIGNAL);
 605   debug_only(signal_sets_initialized = true);
 606 
 607 }
 608 
 609 // These are signals that are unblocked while a thread is running Java.
 610 // (For some reason, they get blocked by default.)
 611 sigset_t* os::Linux::unblocked_signals() {
 612   assert(signal_sets_initialized, "Not initialized");
 613   return &unblocked_sigs;
 614 }
 615 
 616 // These are the signals that are blocked while a (non-VM) thread is
 617 // running Java. Only the VM thread handles these signals.
 618 sigset_t* os::Linux::vm_signals() {
 619   assert(signal_sets_initialized, "Not initialized");
 620   return &vm_sigs;
 621 }
 622 
 623 // These are signals that are blocked during cond_wait to allow debugger in
 624 sigset_t* os::Linux::allowdebug_blocked_signals() {
 625   assert(signal_sets_initialized, "Not initialized");
 626   return &allowdebug_blocked_sigs;
 627 }
 628 
 629 void os::Linux::hotspot_sigmask(Thread* thread) {
 630 
 631   //Save caller's signal mask before setting VM signal mask
 632   sigset_t caller_sigmask;
 633   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 634 
 635   OSThread* osthread = thread->osthread();
 636   osthread->set_caller_sigmask(caller_sigmask);
 637 
 638   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 639 
 640   if (!ReduceSignalUsage) {
 641     if (thread->is_VM_thread()) {
 642       // Only the VM thread handles BREAK_SIGNAL ...
 643       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 644     } else {
 645       // ... all other threads block BREAK_SIGNAL
 646       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 647     }
 648   }
 649 }
 650 
 651 //////////////////////////////////////////////////////////////////////////////
 652 // detecting pthread library
 653 
 654 void os::Linux::libpthread_init() {
 655   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 656   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 657   // generic name for earlier versions.
 658   // Define macros here so we can build HotSpot on old systems.
 659 # ifndef _CS_GNU_LIBC_VERSION
 660 # define _CS_GNU_LIBC_VERSION 2
 661 # endif
 662 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 663 # define _CS_GNU_LIBPTHREAD_VERSION 3
 664 # endif
 665 
 666   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 667   if (n > 0) {
 668      char *str = (char *)malloc(n, mtInternal);
 669      confstr(_CS_GNU_LIBC_VERSION, str, n);
 670      os::Linux::set_glibc_version(str);
 671   } else {
 672      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 673      static char _gnu_libc_version[32];
 674      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 675               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 676      os::Linux::set_glibc_version(_gnu_libc_version);
 677   }
 678 
 679   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 680   if (n > 0) {
 681      char *str = (char *)malloc(n, mtInternal);
 682      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 683      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 684      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 685      // is the case. LinuxThreads has a hard limit on max number of threads.
 686      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 687      // On the other hand, NPTL does not have such a limit, sysconf()
 688      // will return -1 and errno is not changed. Check if it is really NPTL.
 689      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 690          strstr(str, "NPTL") &&
 691          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 692        free(str);
 693        os::Linux::set_libpthread_version("linuxthreads");
 694      } else {
 695        os::Linux::set_libpthread_version(str);
 696      }
 697   } else {
 698     // glibc before 2.3.2 only has LinuxThreads.
 699     os::Linux::set_libpthread_version("linuxthreads");
 700   }
 701 
 702   if (strstr(libpthread_version(), "NPTL")) {
 703      os::Linux::set_is_NPTL();
 704   } else {
 705      os::Linux::set_is_LinuxThreads();
 706   }
 707 
 708   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 709   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 710   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 711      os::Linux::set_is_floating_stack();
 712   }
 713 }
 714 
 715 /////////////////////////////////////////////////////////////////////////////
 716 // thread stack
 717 
 718 // Force Linux kernel to expand current thread stack. If "bottom" is close
 719 // to the stack guard, caller should block all signals.
 720 //
 721 // MAP_GROWSDOWN:
 722 //   A special mmap() flag that is used to implement thread stacks. It tells
 723 //   kernel that the memory region should extend downwards when needed. This
 724 //   allows early versions of LinuxThreads to only mmap the first few pages
 725 //   when creating a new thread. Linux kernel will automatically expand thread
 726 //   stack as needed (on page faults).
 727 //
 728 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 729 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 730 //   region, it's hard to tell if the fault is due to a legitimate stack
 731 //   access or because of reading/writing non-exist memory (e.g. buffer
 732 //   overrun). As a rule, if the fault happens below current stack pointer,
 733 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 734 //   application (see Linux kernel fault.c).
 735 //
 736 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 737 //   stack overflow detection.
 738 //
 739 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 740 //   not use this flag. However, the stack of initial thread is not created
 741 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 742 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 743 //   and then attach the thread to JVM.
 744 //
 745 // To get around the problem and allow stack banging on Linux, we need to
 746 // manually expand thread stack after receiving the SIGSEGV.
 747 //
 748 // There are two ways to expand thread stack to address "bottom", we used
 749 // both of them in JVM before 1.5:
 750 //   1. adjust stack pointer first so that it is below "bottom", and then
 751 //      touch "bottom"
 752 //   2. mmap() the page in question
 753 //
 754 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 755 // if current sp is already near the lower end of page 101, and we need to
 756 // call mmap() to map page 100, it is possible that part of the mmap() frame
 757 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 758 // That will destroy the mmap() frame and cause VM to crash.
 759 //
 760 // The following code works by adjusting sp first, then accessing the "bottom"
 761 // page to force a page fault. Linux kernel will then automatically expand the
 762 // stack mapping.
 763 //
 764 // _expand_stack_to() assumes its frame size is less than page size, which
 765 // should always be true if the function is not inlined.
 766 
 767 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 768 #define NOINLINE
 769 #else
 770 #define NOINLINE __attribute__ ((noinline))
 771 #endif
 772 
 773 static void _expand_stack_to(address bottom) NOINLINE;
 774 
 775 static void _expand_stack_to(address bottom) {
 776   address sp;
 777   size_t size;
 778   volatile char *p;
 779 
 780   // Adjust bottom to point to the largest address within the same page, it
 781   // gives us a one-page buffer if alloca() allocates slightly more memory.
 782   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 783   bottom += os::Linux::page_size() - 1;
 784 
 785   // sp might be slightly above current stack pointer; if that's the case, we
 786   // will alloca() a little more space than necessary, which is OK. Don't use
 787   // os::current_stack_pointer(), as its result can be slightly below current
 788   // stack pointer, causing us to not alloca enough to reach "bottom".
 789   sp = (address)&sp;
 790 
 791   if (sp > bottom) {
 792     size = sp - bottom;
 793     p = (volatile char *)alloca(size);
 794     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 795     p[0] = '\0';
 796   }
 797 }
 798 
 799 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 800   assert(t!=NULL, "just checking");
 801   assert(t->osthread()->expanding_stack(), "expand should be set");
 802   assert(t->stack_base() != NULL, "stack_base was not initialized");
 803 
 804   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 805     sigset_t mask_all, old_sigset;
 806     sigfillset(&mask_all);
 807     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 808     _expand_stack_to(addr);
 809     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 810     return true;
 811   }
 812   return false;
 813 }
 814 
 815 //////////////////////////////////////////////////////////////////////////////
 816 // create new thread
 817 
 818 static address highest_vm_reserved_address();
 819 
 820 // check if it's safe to start a new thread
 821 static bool _thread_safety_check(Thread* thread) {
 822   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 823     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 824     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 825     //   allocated (MAP_FIXED) from high address space. Every thread stack
 826     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 827     //   it to other values if they rebuild LinuxThreads).
 828     //
 829     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 830     // the memory region has already been mmap'ed. That means if we have too
 831     // many threads and/or very large heap, eventually thread stack will
 832     // collide with heap.
 833     //
 834     // Here we try to prevent heap/stack collision by comparing current
 835     // stack bottom with the highest address that has been mmap'ed by JVM
 836     // plus a safety margin for memory maps created by native code.
 837     //
 838     // This feature can be disabled by setting ThreadSafetyMargin to 0
 839     //
 840     if (ThreadSafetyMargin > 0) {
 841       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 842 
 843       // not safe if our stack extends below the safety margin
 844       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 845     } else {
 846       return true;
 847     }
 848   } else {
 849     // Floating stack LinuxThreads or NPTL:
 850     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 851     //   there's not enough space left, pthread_create() will fail. If we come
 852     //   here, that means enough space has been reserved for stack.
 853     return true;
 854   }
 855 }
 856 
 857 // Thread start routine for all newly created threads
 858 static void *java_start(Thread *thread) {
 859   // Try to randomize the cache line index of hot stack frames.
 860   // This helps when threads of the same stack traces evict each other's
 861   // cache lines. The threads can be either from the same JVM instance, or
 862   // from different JVM instances. The benefit is especially true for
 863   // processors with hyperthreading technology.
 864   static int counter = 0;
 865   int pid = os::current_process_id();
 866   alloca(((pid ^ counter++) & 7) * 128);
 867 
 868   ThreadLocalStorage::set_thread(thread);
 869 
 870   OSThread* osthread = thread->osthread();
 871   Monitor* sync = osthread->startThread_lock();
 872 
 873   // non floating stack LinuxThreads needs extra check, see above
 874   if (!_thread_safety_check(thread)) {
 875     // notify parent thread
 876     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 877     osthread->set_state(ZOMBIE);
 878     sync->notify_all();
 879     return NULL;
 880   }
 881 
 882   // thread_id is kernel thread id (similar to Solaris LWP id)
 883   osthread->set_thread_id(os::Linux::gettid());
 884 
 885   if (UseNUMA) {
 886     int lgrp_id = os::numa_get_group_id();
 887     if (lgrp_id != -1) {
 888       thread->set_lgrp_id(lgrp_id);
 889     }
 890   }
 891   // initialize signal mask for this thread
 892   os::Linux::hotspot_sigmask(thread);
 893 
 894   // initialize floating point control register
 895   os::Linux::init_thread_fpu_state();
 896 
 897   // handshaking with parent thread
 898   {
 899     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 900 
 901     // notify parent thread
 902     osthread->set_state(INITIALIZED);
 903     sync->notify_all();
 904 
 905     // wait until os::start_thread()
 906     while (osthread->get_state() == INITIALIZED) {
 907       sync->wait(Mutex::_no_safepoint_check_flag);
 908     }
 909   }
 910 
 911   // call one more level start routine
 912   thread->run();
 913 
 914   return 0;
 915 }
 916 
 917 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
 918   assert(thread->osthread() == NULL, "caller responsible");
 919 
 920   // Allocate the OSThread object
 921   OSThread* osthread = new OSThread(NULL, NULL);
 922   if (osthread == NULL) {
 923     return false;
 924   }
 925 
 926   // set the correct thread state
 927   osthread->set_thread_type(thr_type);
 928 
 929   // Initial state is ALLOCATED but not INITIALIZED
 930   osthread->set_state(ALLOCATED);
 931 
 932   thread->set_osthread(osthread);
 933 
 934   // init thread attributes
 935   pthread_attr_t attr;
 936   pthread_attr_init(&attr);
 937   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 938 
 939   // stack size
 940   if (os::Linux::supports_variable_stack_size()) {
 941     // calculate stack size if it's not specified by caller
 942     if (stack_size == 0) {
 943       stack_size = os::Linux::default_stack_size(thr_type);
 944 
 945       switch (thr_type) {
 946       case os::java_thread:
 947         // Java threads use ThreadStackSize which default value can be
 948         // changed with the flag -Xss
 949         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
 950         stack_size = JavaThread::stack_size_at_create();
 951         break;
 952       case os::compiler_thread:
 953         if (CompilerThreadStackSize > 0) {
 954           stack_size = (size_t)(CompilerThreadStackSize * K);
 955           break;
 956         } // else fall through:
 957           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 958       case os::vm_thread:
 959       case os::pgc_thread:
 960       case os::cgc_thread:
 961       case os::watcher_thread:
 962         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 963         break;
 964       }
 965     }
 966 
 967     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 968     pthread_attr_setstacksize(&attr, stack_size);
 969   } else {
 970     // let pthread_create() pick the default value.
 971   }
 972 
 973   // glibc guard page
 974   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 975 
 976   ThreadState state;
 977 
 978   {
 979     // Serialize thread creation if we are running with fixed stack LinuxThreads
 980     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 981     if (lock) {
 982       os::Linux::createThread_lock()->lock_without_safepoint_check();
 983     }
 984 
 985     pthread_t tid;
 986     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 987 
 988     pthread_attr_destroy(&attr);
 989 
 990     if (ret != 0) {
 991       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 992         perror("pthread_create()");
 993       }
 994       // Need to clean up stuff we've allocated so far
 995       thread->set_osthread(NULL);
 996       delete osthread;
 997       if (lock) os::Linux::createThread_lock()->unlock();
 998       return false;
 999     }
1000 
1001     // Store pthread info into the OSThread
1002     osthread->set_pthread_id(tid);
1003 
1004     // Wait until child thread is either initialized or aborted
1005     {
1006       Monitor* sync_with_child = osthread->startThread_lock();
1007       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1008       while ((state = osthread->get_state()) == ALLOCATED) {
1009         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
1010       }
1011     }
1012 
1013     if (lock) {
1014       os::Linux::createThread_lock()->unlock();
1015     }
1016   }
1017 
1018   // Aborted due to thread limit being reached
1019   if (state == ZOMBIE) {
1020       thread->set_osthread(NULL);
1021       delete osthread;
1022       return false;
1023   }
1024 
1025   // The thread is returned suspended (in state INITIALIZED),
1026   // and is started higher up in the call chain
1027   assert(state == INITIALIZED, "race condition");
1028   return true;
1029 }
1030 
1031 /////////////////////////////////////////////////////////////////////////////
1032 // attach existing thread
1033 
1034 // bootstrap the main thread
1035 bool os::create_main_thread(JavaThread* thread) {
1036   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
1037   return create_attached_thread(thread);
1038 }
1039 
1040 bool os::create_attached_thread(JavaThread* thread) {
1041 #ifdef ASSERT
1042     thread->verify_not_published();
1043 #endif
1044 
1045   // Allocate the OSThread object
1046   OSThread* osthread = new OSThread(NULL, NULL);
1047 
1048   if (osthread == NULL) {
1049     return false;
1050   }
1051 
1052   // Store pthread info into the OSThread
1053   osthread->set_thread_id(os::Linux::gettid());
1054   osthread->set_pthread_id(::pthread_self());
1055 
1056   // initialize floating point control register
1057   os::Linux::init_thread_fpu_state();
1058 
1059   // Initial thread state is RUNNABLE
1060   osthread->set_state(RUNNABLE);
1061 
1062   thread->set_osthread(osthread);
1063 
1064   if (UseNUMA) {
1065     int lgrp_id = os::numa_get_group_id();
1066     if (lgrp_id != -1) {
1067       thread->set_lgrp_id(lgrp_id);
1068     }
1069   }
1070 
1071   if (os::Linux::is_initial_thread()) {
1072     // If current thread is initial thread, its stack is mapped on demand,
1073     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1074     // the entire stack region to avoid SEGV in stack banging.
1075     // It is also useful to get around the heap-stack-gap problem on SuSE
1076     // kernel (see 4821821 for details). We first expand stack to the top
1077     // of yellow zone, then enable stack yellow zone (order is significant,
1078     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1079     // is no gap between the last two virtual memory regions.
1080 
1081     JavaThread *jt = (JavaThread *)thread;
1082     address addr = jt->stack_yellow_zone_base();
1083     assert(addr != NULL, "initialization problem?");
1084     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1085 
1086     osthread->set_expanding_stack();
1087     os::Linux::manually_expand_stack(jt, addr);
1088     osthread->clear_expanding_stack();
1089   }
1090 
1091   // initialize signal mask for this thread
1092   // and save the caller's signal mask
1093   os::Linux::hotspot_sigmask(thread);
1094 
1095   return true;
1096 }
1097 
1098 void os::pd_start_thread(Thread* thread) {
1099   OSThread * osthread = thread->osthread();
1100   assert(osthread->get_state() != INITIALIZED, "just checking");
1101   Monitor* sync_with_child = osthread->startThread_lock();
1102   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1103   sync_with_child->notify();
1104 }
1105 
1106 // Free Linux resources related to the OSThread
1107 void os::free_thread(OSThread* osthread) {
1108   assert(osthread != NULL, "osthread not set");
1109 
1110   if (Thread::current()->osthread() == osthread) {
1111     // Restore caller's signal mask
1112     sigset_t sigmask = osthread->caller_sigmask();
1113     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1114    }
1115 
1116   delete osthread;
1117 }
1118 
1119 //////////////////////////////////////////////////////////////////////////////
1120 // thread local storage
1121 
1122 // Restore the thread pointer if the destructor is called. This is in case
1123 // someone from JNI code sets up a destructor with pthread_key_create to run
1124 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1125 // will hang or crash. When detachCurrentThread is called the key will be set
1126 // to null and we will not be called again. If detachCurrentThread is never
1127 // called we could loop forever depending on the pthread implementation.
1128 static void restore_thread_pointer(void* p) {
1129   Thread* thread = (Thread*) p;
1130   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1131 }
1132 
1133 int os::allocate_thread_local_storage() {
1134   pthread_key_t key;
1135   int rslt = pthread_key_create(&key, restore_thread_pointer);
1136   assert(rslt == 0, "cannot allocate thread local storage");
1137   return (int)key;
1138 }
1139 
1140 // Note: This is currently not used by VM, as we don't destroy TLS key
1141 // on VM exit.
1142 void os::free_thread_local_storage(int index) {
1143   int rslt = pthread_key_delete((pthread_key_t)index);
1144   assert(rslt == 0, "invalid index");
1145 }
1146 
1147 void os::thread_local_storage_at_put(int index, void* value) {
1148   int rslt = pthread_setspecific((pthread_key_t)index, value);
1149   assert(rslt == 0, "pthread_setspecific failed");
1150 }
1151 
1152 extern "C" Thread* get_thread() {
1153   return ThreadLocalStorage::thread();
1154 }
1155 
1156 //////////////////////////////////////////////////////////////////////////////
1157 // initial thread
1158 
1159 // Check if current thread is the initial thread, similar to Solaris thr_main.
1160 bool os::Linux::is_initial_thread(void) {
1161   char dummy;
1162   // If called before init complete, thread stack bottom will be null.
1163   // Can be called if fatal error occurs before initialization.
1164   if (initial_thread_stack_bottom() == NULL) return false;
1165   assert(initial_thread_stack_bottom() != NULL &&
1166          initial_thread_stack_size()   != 0,
1167          "os::init did not locate initial thread's stack region");
1168   if ((address)&dummy >= initial_thread_stack_bottom() &&
1169       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1170        return true;
1171   else return false;
1172 }
1173 
1174 // Find the virtual memory area that contains addr
1175 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1176   FILE *fp = fopen("/proc/self/maps", "r");
1177   if (fp) {
1178     address low, high;
1179     while (!feof(fp)) {
1180       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1181         if (low <= addr && addr < high) {
1182            if (vma_low)  *vma_low  = low;
1183            if (vma_high) *vma_high = high;
1184            fclose (fp);
1185            return true;
1186         }
1187       }
1188       for (;;) {
1189         int ch = fgetc(fp);
1190         if (ch == EOF || ch == (int)'\n') break;
1191       }
1192     }
1193     fclose(fp);
1194   }
1195   return false;
1196 }
1197 
1198 // Locate initial thread stack. This special handling of initial thread stack
1199 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1200 // bogus value for initial thread.
1201 void os::Linux::capture_initial_stack(size_t max_size) {
1202   // stack size is the easy part, get it from RLIMIT_STACK
1203   size_t stack_size;
1204   struct rlimit rlim;
1205   getrlimit(RLIMIT_STACK, &rlim);
1206   stack_size = rlim.rlim_cur;
1207 
1208   // 6308388: a bug in ld.so will relocate its own .data section to the
1209   //   lower end of primordial stack; reduce ulimit -s value a little bit
1210   //   so we won't install guard page on ld.so's data section.
1211   stack_size -= 2 * page_size();
1212 
1213   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1214   //   7.1, in both cases we will get 2G in return value.
1215   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1216   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1217   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1218   //   in case other parts in glibc still assumes 2M max stack size.
1219   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1220 #ifndef IA64
1221   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1222 #else
1223   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1224   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1225 #endif
1226 
1227   // Try to figure out where the stack base (top) is. This is harder.
1228   //
1229   // When an application is started, glibc saves the initial stack pointer in
1230   // a global variable "__libc_stack_end", which is then used by system
1231   // libraries. __libc_stack_end should be pretty close to stack top. The
1232   // variable is available since the very early days. However, because it is
1233   // a private interface, it could disappear in the future.
1234   //
1235   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1236   // to __libc_stack_end, it is very close to stack top, but isn't the real
1237   // stack top. Note that /proc may not exist if VM is running as a chroot
1238   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1239   // /proc/<pid>/stat could change in the future (though unlikely).
1240   //
1241   // We try __libc_stack_end first. If that doesn't work, look for
1242   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1243   // as a hint, which should work well in most cases.
1244 
1245   uintptr_t stack_start;
1246 
1247   // try __libc_stack_end first
1248   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1249   if (p && *p) {
1250     stack_start = *p;
1251   } else {
1252     // see if we can get the start_stack field from /proc/self/stat
1253     FILE *fp;
1254     int pid;
1255     char state;
1256     int ppid;
1257     int pgrp;
1258     int session;
1259     int nr;
1260     int tpgrp;
1261     unsigned long flags;
1262     unsigned long minflt;
1263     unsigned long cminflt;
1264     unsigned long majflt;
1265     unsigned long cmajflt;
1266     unsigned long utime;
1267     unsigned long stime;
1268     long cutime;
1269     long cstime;
1270     long prio;
1271     long nice;
1272     long junk;
1273     long it_real;
1274     uintptr_t start;
1275     uintptr_t vsize;
1276     intptr_t rss;
1277     uintptr_t rsslim;
1278     uintptr_t scodes;
1279     uintptr_t ecode;
1280     int i;
1281 
1282     // Figure what the primordial thread stack base is. Code is inspired
1283     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1284     // followed by command name surrounded by parentheses, state, etc.
1285     char stat[2048];
1286     int statlen;
1287 
1288     fp = fopen("/proc/self/stat", "r");
1289     if (fp) {
1290       statlen = fread(stat, 1, 2047, fp);
1291       stat[statlen] = '\0';
1292       fclose(fp);
1293 
1294       // Skip pid and the command string. Note that we could be dealing with
1295       // weird command names, e.g. user could decide to rename java launcher
1296       // to "java 1.4.2 :)", then the stat file would look like
1297       //                1234 (java 1.4.2 :)) R ... ...
1298       // We don't really need to know the command string, just find the last
1299       // occurrence of ")" and then start parsing from there. See bug 4726580.
1300       char * s = strrchr(stat, ')');
1301 
1302       i = 0;
1303       if (s) {
1304         // Skip blank chars
1305         do s++; while (isspace(*s));
1306 
1307 #define _UFM UINTX_FORMAT
1308 #define _DFM INTX_FORMAT
1309 
1310         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1311         /*              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 */
1312         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,
1313              &state,          /* 3  %c  */
1314              &ppid,           /* 4  %d  */
1315              &pgrp,           /* 5  %d  */
1316              &session,        /* 6  %d  */
1317              &nr,             /* 7  %d  */
1318              &tpgrp,          /* 8  %d  */
1319              &flags,          /* 9  %lu  */
1320              &minflt,         /* 10 %lu  */
1321              &cminflt,        /* 11 %lu  */
1322              &majflt,         /* 12 %lu  */
1323              &cmajflt,        /* 13 %lu  */
1324              &utime,          /* 14 %lu  */
1325              &stime,          /* 15 %lu  */
1326              &cutime,         /* 16 %ld  */
1327              &cstime,         /* 17 %ld  */
1328              &prio,           /* 18 %ld  */
1329              &nice,           /* 19 %ld  */
1330              &junk,           /* 20 %ld  */
1331              &it_real,        /* 21 %ld  */
1332              &start,          /* 22 UINTX_FORMAT */
1333              &vsize,          /* 23 UINTX_FORMAT */
1334              &rss,            /* 24 INTX_FORMAT  */
1335              &rsslim,         /* 25 UINTX_FORMAT */
1336              &scodes,         /* 26 UINTX_FORMAT */
1337              &ecode,          /* 27 UINTX_FORMAT */
1338              &stack_start);   /* 28 UINTX_FORMAT */
1339       }
1340 
1341 #undef _UFM
1342 #undef _DFM
1343 
1344       if (i != 28 - 2) {
1345          assert(false, "Bad conversion from /proc/self/stat");
1346          // product mode - assume we are the initial thread, good luck in the
1347          // embedded case.
1348          warning("Can't detect initial thread stack location - bad conversion");
1349          stack_start = (uintptr_t) &rlim;
1350       }
1351     } else {
1352       // For some reason we can't open /proc/self/stat (for example, running on
1353       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1354       // most cases, so don't abort:
1355       warning("Can't detect initial thread stack location - no /proc/self/stat");
1356       stack_start = (uintptr_t) &rlim;
1357     }
1358   }
1359 
1360   // Now we have a pointer (stack_start) very close to the stack top, the
1361   // next thing to do is to figure out the exact location of stack top. We
1362   // can find out the virtual memory area that contains stack_start by
1363   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1364   // and its upper limit is the real stack top. (again, this would fail if
1365   // running inside chroot, because /proc may not exist.)
1366 
1367   uintptr_t stack_top;
1368   address low, high;
1369   if (find_vma((address)stack_start, &low, &high)) {
1370     // success, "high" is the true stack top. (ignore "low", because initial
1371     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1372     stack_top = (uintptr_t)high;
1373   } else {
1374     // failed, likely because /proc/self/maps does not exist
1375     warning("Can't detect initial thread stack location - find_vma failed");
1376     // best effort: stack_start is normally within a few pages below the real
1377     // stack top, use it as stack top, and reduce stack size so we won't put
1378     // guard page outside stack.
1379     stack_top = stack_start;
1380     stack_size -= 16 * page_size();
1381   }
1382 
1383   // stack_top could be partially down the page so align it
1384   stack_top = align_size_up(stack_top, page_size());
1385 
1386   if (max_size && stack_size > max_size) {
1387      _initial_thread_stack_size = max_size;
1388   } else {
1389      _initial_thread_stack_size = stack_size;
1390   }
1391 
1392   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1393   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1394 }
1395 
1396 ////////////////////////////////////////////////////////////////////////////////
1397 // time support
1398 
1399 // Time since start-up in seconds to a fine granularity.
1400 // Used by VMSelfDestructTimer and the MemProfiler.
1401 double os::elapsedTime() {
1402 
1403   return (double)(os::elapsed_counter()) * 0.000001;
1404 }
1405 
1406 jlong os::elapsed_counter() {
1407   timeval time;
1408   int status = gettimeofday(&time, NULL);
1409   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1410 }
1411 
1412 jlong os::elapsed_frequency() {
1413   return (1000 * 1000);
1414 }
1415 
1416 bool os::supports_vtime() { return true; }
1417 bool os::enable_vtime()   { return false; }
1418 bool os::vtime_enabled()  { return false; }
1419 
1420 double os::elapsedVTime() {
1421   struct rusage usage;
1422   int retval = getrusage(RUSAGE_THREAD, &usage);
1423   if (retval == 0) {
1424     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);
1425   } else {
1426     // better than nothing, but not much
1427     return elapsedTime();
1428   }
1429 }
1430 
1431 jlong os::javaTimeMillis() {
1432   timeval time;
1433   int status = gettimeofday(&time, NULL);
1434   assert(status != -1, "linux error");
1435   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1436 }
1437 
1438 #ifndef CLOCK_MONOTONIC
1439 #define CLOCK_MONOTONIC (1)
1440 #endif
1441 
1442 void os::Linux::clock_init() {
1443   // we do dlopen's in this particular order due to bug in linux
1444   // dynamical loader (see 6348968) leading to crash on exit
1445   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1446   if (handle == NULL) {
1447     handle = dlopen("librt.so", RTLD_LAZY);
1448   }
1449 
1450   if (handle) {
1451     int (*clock_getres_func)(clockid_t, struct timespec*) =
1452            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1453     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1454            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1455     if (clock_getres_func && clock_gettime_func) {
1456       // See if monotonic clock is supported by the kernel. Note that some
1457       // early implementations simply return kernel jiffies (updated every
1458       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1459       // for nano time (though the monotonic property is still nice to have).
1460       // It's fixed in newer kernels, however clock_getres() still returns
1461       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1462       // resolution for now. Hopefully as people move to new kernels, this
1463       // won't be a problem.
1464       struct timespec res;
1465       struct timespec tp;
1466       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1467           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1468         // yes, monotonic clock is supported
1469         _clock_gettime = clock_gettime_func;
1470         return;
1471       } else {
1472         // close librt if there is no monotonic clock
1473         dlclose(handle);
1474       }
1475     }
1476   }
1477   warning("No monotonic clock was available - timed services may " \
1478           "be adversely affected if the time-of-day clock changes");
1479 }
1480 
1481 #ifndef SYS_clock_getres
1482 
1483 #if defined(IA32) || defined(AMD64) || defined(AARCH64)
1484 #ifdef BUILTIN_SIM
1485 #define SYS_clock_getres 229
1486 #else
1487 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229) AARCH64_ONLY(114)
1488 #endif
1489 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1490 #else
1491 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1492 #define sys_clock_getres(x,y)  -1
1493 #endif
1494 
1495 #else
1496 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1497 #endif
1498 
1499 void os::Linux::fast_thread_clock_init() {
1500   if (!UseLinuxPosixThreadCPUClocks) {
1501     return;
1502   }
1503   clockid_t clockid;
1504   struct timespec tp;
1505   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1506       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1507 
1508   // Switch to using fast clocks for thread cpu time if
1509   // the sys_clock_getres() returns 0 error code.
1510   // Note, that some kernels may support the current thread
1511   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1512   // returned by the pthread_getcpuclockid().
1513   // If the fast Posix clocks are supported then the sys_clock_getres()
1514   // must return at least tp.tv_sec == 0 which means a resolution
1515   // better than 1 sec. This is extra check for reliability.
1516 
1517   if(pthread_getcpuclockid_func &&
1518      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1519      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1520 
1521     _supports_fast_thread_cpu_time = true;
1522     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1523   }
1524 }
1525 
1526 jlong os::javaTimeNanos() {
1527   if (Linux::supports_monotonic_clock()) {
1528     struct timespec tp;
1529     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1530     assert(status == 0, "gettime error");
1531     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1532     return result;
1533   } else {
1534     timeval time;
1535     int status = gettimeofday(&time, NULL);
1536     assert(status != -1, "linux error");
1537     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1538     return 1000 * usecs;
1539   }
1540 }
1541 
1542 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1543   if (Linux::supports_monotonic_clock()) {
1544     info_ptr->max_value = ALL_64_BITS;
1545 
1546     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1547     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1548     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1549   } else {
1550     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1551     info_ptr->max_value = ALL_64_BITS;
1552 
1553     // gettimeofday is a real time clock so it skips
1554     info_ptr->may_skip_backward = true;
1555     info_ptr->may_skip_forward = true;
1556   }
1557 
1558   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1559 }
1560 
1561 // Return the real, user, and system times in seconds from an
1562 // arbitrary fixed point in the past.
1563 bool os::getTimesSecs(double* process_real_time,
1564                       double* process_user_time,
1565                       double* process_system_time) {
1566   struct tms ticks;
1567   clock_t real_ticks = times(&ticks);
1568 
1569   if (real_ticks == (clock_t) (-1)) {
1570     return false;
1571   } else {
1572     double ticks_per_second = (double) clock_tics_per_sec;
1573     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1574     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1575     *process_real_time = ((double) real_ticks) / ticks_per_second;
1576 
1577     return true;
1578   }
1579 }
1580 
1581 
1582 char * os::local_time_string(char *buf, size_t buflen) {
1583   struct tm t;
1584   time_t long_time;
1585   time(&long_time);
1586   localtime_r(&long_time, &t);
1587   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1588                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1589                t.tm_hour, t.tm_min, t.tm_sec);
1590   return buf;
1591 }
1592 
1593 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1594   return localtime_r(clock, res);
1595 }
1596 
1597 ////////////////////////////////////////////////////////////////////////////////
1598 // runtime exit support
1599 
1600 // Note: os::shutdown() might be called very early during initialization, or
1601 // called from signal handler. Before adding something to os::shutdown(), make
1602 // sure it is async-safe and can handle partially initialized VM.
1603 void os::shutdown() {
1604 
1605   // allow PerfMemory to attempt cleanup of any persistent resources
1606   perfMemory_exit();
1607 
1608   // needs to remove object in file system
1609   AttachListener::abort();
1610 
1611   // flush buffered output, finish log files
1612   ostream_abort();
1613 
1614   // Check for abort hook
1615   abort_hook_t abort_hook = Arguments::abort_hook();
1616   if (abort_hook != NULL) {
1617     abort_hook();
1618   }
1619 
1620 }
1621 
1622 // Note: os::abort() might be called very early during initialization, or
1623 // called from signal handler. Before adding something to os::abort(), make
1624 // sure it is async-safe and can handle partially initialized VM.
1625 void os::abort(bool dump_core) {
1626   os::shutdown();
1627   if (dump_core) {
1628 #ifndef PRODUCT
1629     fdStream out(defaultStream::output_fd());
1630     out.print_raw("Current thread is ");
1631     char buf[16];
1632     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1633     out.print_raw_cr(buf);
1634     out.print_raw_cr("Dumping core ...");
1635 #endif
1636     ::abort(); // dump core
1637   }
1638 
1639   ::exit(1);
1640 }
1641 
1642 // Die immediately, no exit hook, no abort hook, no cleanup.
1643 void os::die() {
1644   // _exit() on LinuxThreads only kills current thread
1645   ::abort();
1646 }
1647 
1648 
1649 // This method is a copy of JDK's sysGetLastErrorString
1650 // from src/solaris/hpi/src/system_md.c
1651 
1652 size_t os::lasterror(char *buf, size_t len) {
1653 
1654   if (errno == 0)  return 0;
1655 
1656   const char *s = ::strerror(errno);
1657   size_t n = ::strlen(s);
1658   if (n >= len) {
1659     n = len - 1;
1660   }
1661   ::strncpy(buf, s, n);
1662   buf[n] = '\0';
1663   return n;
1664 }
1665 
1666 intx os::current_thread_id() { return (intx)pthread_self(); }
1667 int os::current_process_id() {
1668 
1669   // Under the old linux thread library, linux gives each thread
1670   // its own process id. Because of this each thread will return
1671   // a different pid if this method were to return the result
1672   // of getpid(2). Linux provides no api that returns the pid
1673   // of the launcher thread for the vm. This implementation
1674   // returns a unique pid, the pid of the launcher thread
1675   // that starts the vm 'process'.
1676 
1677   // Under the NPTL, getpid() returns the same pid as the
1678   // launcher thread rather than a unique pid per thread.
1679   // Use gettid() if you want the old pre NPTL behaviour.
1680 
1681   // if you are looking for the result of a call to getpid() that
1682   // returns a unique pid for the calling thread, then look at the
1683   // OSThread::thread_id() method in osThread_linux.hpp file
1684 
1685   return (int)(_initial_pid ? _initial_pid : getpid());
1686 }
1687 
1688 // DLL functions
1689 
1690 const char* os::dll_file_extension() { return ".so"; }
1691 
1692 // This must be hard coded because it's the system's temporary
1693 // directory not the java application's temp directory, ala java.io.tmpdir.
1694 const char* os::get_temp_directory() { return "/tmp"; }
1695 
1696 static bool file_exists(const char* filename) {
1697   struct stat statbuf;
1698   if (filename == NULL || strlen(filename) == 0) {
1699     return false;
1700   }
1701   return os::stat(filename, &statbuf) == 0;
1702 }
1703 
1704 void os::dll_build_name(char* buffer, size_t buflen,
1705                         const char* pname, const char* fname) {
1706   // Copied from libhpi
1707   const size_t pnamelen = pname ? strlen(pname) : 0;
1708 
1709   // Quietly truncate on buffer overflow.  Should be an error.
1710   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1711       *buffer = '\0';
1712       return;
1713   }
1714 
1715   if (pnamelen == 0) {
1716     snprintf(buffer, buflen, "lib%s.so", fname);
1717   } else if (strchr(pname, *os::path_separator()) != NULL) {
1718     int n;
1719     char** pelements = split_path(pname, &n);
1720     for (int i = 0 ; i < n ; i++) {
1721       // Really shouldn't be NULL, but check can't hurt
1722       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1723         continue; // skip the empty path values
1724       }
1725       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1726       if (file_exists(buffer)) {
1727         break;
1728       }
1729     }
1730     // release the storage
1731     for (int i = 0 ; i < n ; i++) {
1732       if (pelements[i] != NULL) {
1733         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1734       }
1735     }
1736     if (pelements != NULL) {
1737       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1738     }
1739   } else {
1740     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1741   }
1742 }
1743 
1744 const char* os::get_current_directory(char *buf, int buflen) {
1745   return getcwd(buf, buflen);
1746 }
1747 
1748 // check if addr is inside libjvm[_g].so
1749 bool os::address_is_in_vm(address addr) {
1750   static address libjvm_base_addr;
1751   Dl_info dlinfo;
1752 
1753   if (libjvm_base_addr == NULL) {
1754     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1755       libjvm_base_addr = (address)dlinfo.dli_fbase;
1756     }
1757     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1758   }
1759 
1760   if (dladdr((void *)addr, &dlinfo) != 0) {
1761     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1762   }
1763 
1764   return false;
1765 }
1766 
1767 bool os::dll_address_to_function_name(address addr, char *buf,
1768                                       int buflen, int *offset) {
1769   // buf is not optional, but offset is optional
1770   assert(buf != NULL, "sanity check");
1771 
1772   Dl_info dlinfo;
1773 
1774   if (dladdr((void*)addr, &dlinfo) != 0) {
1775     // see if we have a matching symbol
1776     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1777       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1778         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1779       }
1780       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1781       return true;
1782     }
1783     // no matching symbol so try for just file info
1784     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1785       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1786                           buf, buflen, offset, dlinfo.dli_fname)) {
1787         return true;
1788       }
1789     }
1790   }
1791 
1792   buf[0] = '\0';
1793   if (offset != NULL) *offset = -1;
1794   return false;
1795 }
1796 
1797 struct _address_to_library_name {
1798   address addr;          // input : memory address
1799   size_t  buflen;        //         size of fname
1800   char*   fname;         // output: library name
1801   address base;          //         library base addr
1802 };
1803 
1804 static int address_to_library_name_callback(struct dl_phdr_info *info,
1805                                             size_t size, void *data) {
1806   int i;
1807   bool found = false;
1808   address libbase = NULL;
1809   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1810 
1811   // iterate through all loadable segments
1812   for (i = 0; i < info->dlpi_phnum; i++) {
1813     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1814     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1815       // base address of a library is the lowest address of its loaded
1816       // segments.
1817       if (libbase == NULL || libbase > segbase) {
1818         libbase = segbase;
1819       }
1820       // see if 'addr' is within current segment
1821       if (segbase <= d->addr &&
1822           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1823         found = true;
1824       }
1825     }
1826   }
1827 
1828   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1829   // so dll_address_to_library_name() can fall through to use dladdr() which
1830   // can figure out executable name from argv[0].
1831   if (found && info->dlpi_name && info->dlpi_name[0]) {
1832     d->base = libbase;
1833     if (d->fname) {
1834       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1835     }
1836     return 1;
1837   }
1838   return 0;
1839 }
1840 
1841 bool os::dll_address_to_library_name(address addr, char* buf,
1842                                      int buflen, int* offset) {
1843   // buf is not optional, but offset is optional
1844   assert(buf != NULL, "sanity check");
1845 
1846   Dl_info dlinfo;
1847   struct _address_to_library_name data;
1848 
1849   // There is a bug in old glibc dladdr() implementation that it could resolve
1850   // to wrong library name if the .so file has a base address != NULL. Here
1851   // we iterate through the program headers of all loaded libraries to find
1852   // out which library 'addr' really belongs to. This workaround can be
1853   // removed once the minimum requirement for glibc is moved to 2.3.x.
1854   data.addr = addr;
1855   data.fname = buf;
1856   data.buflen = buflen;
1857   data.base = NULL;
1858   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1859 
1860   if (rslt) {
1861      // buf already contains library name
1862      if (offset) *offset = addr - data.base;
1863      return true;
1864   }
1865   if (dladdr((void*)addr, &dlinfo) != 0) {
1866     if (dlinfo.dli_fname != NULL) {
1867       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1868     }
1869     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1870       *offset = addr - (address)dlinfo.dli_fbase;
1871     }
1872     return true;
1873   }
1874 
1875   buf[0] = '\0';
1876   if (offset) *offset = -1;
1877   return false;
1878 }
1879 
1880   // Loads .dll/.so and
1881   // in case of error it checks if .dll/.so was built for the
1882   // same architecture as Hotspot is running on
1883 
1884 
1885 // Remember the stack's state. The Linux dynamic linker will change
1886 // the stack to 'executable' at most once, so we must safepoint only once.
1887 bool os::Linux::_stack_is_executable = false;
1888 
1889 // VM operation that loads a library.  This is necessary if stack protection
1890 // of the Java stacks can be lost during loading the library.  If we
1891 // do not stop the Java threads, they can stack overflow before the stacks
1892 // are protected again.
1893 class VM_LinuxDllLoad: public VM_Operation {
1894  private:
1895   const char *_filename;
1896   char *_ebuf;
1897   int _ebuflen;
1898   void *_lib;
1899  public:
1900   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1901     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1902   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1903   void doit() {
1904     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1905     os::Linux::_stack_is_executable = true;
1906   }
1907   void* loaded_library() { return _lib; }
1908 };
1909 
1910 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1911 {
1912   void * result = NULL;
1913   bool load_attempted = false;
1914 
1915   // Check whether the library to load might change execution rights
1916   // of the stack. If they are changed, the protection of the stack
1917   // guard pages will be lost. We need a safepoint to fix this.
1918   //
1919   // See Linux man page execstack(8) for more info.
1920   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1921     ElfFile ef(filename);
1922     if (!ef.specifies_noexecstack()) {
1923       if (!is_init_completed()) {
1924         os::Linux::_stack_is_executable = true;
1925         // This is OK - No Java threads have been created yet, and hence no
1926         // stack guard pages to fix.
1927         //
1928         // This should happen only when you are building JDK7 using a very
1929         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1930         //
1931         // Dynamic loader will make all stacks executable after
1932         // this function returns, and will not do that again.
1933         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1934       } else {
1935         warning("You have loaded library %s which might have disabled stack guard. "
1936                 "The VM will try to fix the stack guard now.\n"
1937                 "It's highly recommended that you fix the library with "
1938                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1939                 filename);
1940 
1941         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1942         JavaThread *jt = JavaThread::current();
1943         if (jt->thread_state() != _thread_in_native) {
1944           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1945           // that requires ExecStack. Cannot enter safe point. Let's give up.
1946           warning("Unable to fix stack guard. Giving up.");
1947         } else {
1948           if (!LoadExecStackDllInVMThread) {
1949             // This is for the case where the DLL has an static
1950             // constructor function that executes JNI code. We cannot
1951             // load such DLLs in the VMThread.
1952             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1953           }
1954 
1955           ThreadInVMfromNative tiv(jt);
1956           debug_only(VMNativeEntryWrapper vew;)
1957 
1958           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1959           VMThread::execute(&op);
1960           if (LoadExecStackDllInVMThread) {
1961             result = op.loaded_library();
1962           }
1963           load_attempted = true;
1964         }
1965       }
1966     }
1967   }
1968 
1969   if (!load_attempted) {
1970     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1971   }
1972 
1973   if (result != NULL) {
1974     // Successful loading
1975     return result;
1976   }
1977 
1978   Elf32_Ehdr elf_head;
1979   int diag_msg_max_length=ebuflen-strlen(ebuf);
1980   char* diag_msg_buf=ebuf+strlen(ebuf);
1981 
1982   if (diag_msg_max_length==0) {
1983     // No more space in ebuf for additional diagnostics message
1984     return NULL;
1985   }
1986 
1987 
1988   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1989 
1990   if (file_descriptor < 0) {
1991     // Can't open library, report dlerror() message
1992     return NULL;
1993   }
1994 
1995   bool failed_to_read_elf_head=
1996     (sizeof(elf_head)!=
1997         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1998 
1999   ::close(file_descriptor);
2000   if (failed_to_read_elf_head) {
2001     // file i/o error - report dlerror() msg
2002     return NULL;
2003   }
2004 
2005   typedef struct {
2006     Elf32_Half  code;         // Actual value as defined in elf.h
2007     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
2008     char        elf_class;    // 32 or 64 bit
2009     char        endianess;    // MSB or LSB
2010     char*       name;         // String representation
2011   } arch_t;
2012 
2013   #ifndef EM_486
2014   #define EM_486          6               /* Intel 80486 */
2015   #endif
2016 
2017   #ifndef EM_AARCH64
2018   #define EM_AARCH64    183
2019   #endif
2020 
2021   static const arch_t arch_array[]={
2022     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2023     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2024     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2025     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2026     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2027     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2028     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2029     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2030 #if defined(VM_LITTLE_ENDIAN)
2031     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
2032 #else
2033     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
2034 #endif
2035     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
2036     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
2037     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
2038     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
2039     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
2040     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
2041     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
2042     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2LSB, (char*)"SH"}, /* Support little endian only*/
2043     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"} /* Support little endian only*/
2044   };
2045 
2046   #if  (defined IA32)
2047     static  Elf32_Half running_arch_code=EM_386;
2048   #elif   (defined AMD64)
2049     static  Elf32_Half running_arch_code=EM_X86_64;
2050   #elif  (defined IA64)
2051     static  Elf32_Half running_arch_code=EM_IA_64;
2052   #elif  (defined __sparc) && (defined _LP64)
2053     static  Elf32_Half running_arch_code=EM_SPARCV9;
2054   #elif  (defined __sparc) && (!defined _LP64)
2055     static  Elf32_Half running_arch_code=EM_SPARC;
2056   #elif  (defined __powerpc64__)
2057     static  Elf32_Half running_arch_code=EM_PPC64;
2058   #elif  (defined __powerpc__)
2059     static  Elf32_Half running_arch_code=EM_PPC;
2060   #elif  (defined ARM)
2061     static  Elf32_Half running_arch_code=EM_ARM;
2062   #elif  (defined S390)
2063     static  Elf32_Half running_arch_code=EM_S390;
2064   #elif  (defined ALPHA)
2065     static  Elf32_Half running_arch_code=EM_ALPHA;
2066   #elif  (defined MIPSEL)
2067     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
2068   #elif  (defined PARISC)
2069     static  Elf32_Half running_arch_code=EM_PARISC;
2070   #elif  (defined MIPS)
2071     static  Elf32_Half running_arch_code=EM_MIPS;
2072   #elif  (defined M68K)
2073     static  Elf32_Half running_arch_code=EM_68K;
2074   #elif  (defined SH)
2075     static  Elf32_Half running_arch_code=EM_SH;
2076   #elif  (defined AARCH64)
2077     static  Elf32_Half running_arch_code=EM_AARCH64;
2078   #else
2079     #error Method os::dll_load requires that one of following is defined:\
2080       IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, SH, AARCH64
2081   #endif
2082 
2083   // Identify compatability class for VM's architecture and library's architecture
2084   // Obtain string descriptions for architectures
2085 
2086   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2087   int running_arch_index=-1;
2088 
2089   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2090     if (running_arch_code == arch_array[i].code) {
2091       running_arch_index    = i;
2092     }
2093     if (lib_arch.code == arch_array[i].code) {
2094       lib_arch.compat_class = arch_array[i].compat_class;
2095       lib_arch.name         = arch_array[i].name;
2096     }
2097   }
2098 
2099   assert(running_arch_index != -1,
2100     "Didn't find running architecture code (running_arch_code) in arch_array");
2101   if (running_arch_index == -1) {
2102     // Even though running architecture detection failed
2103     // we may still continue with reporting dlerror() message
2104     return NULL;
2105   }
2106 
2107   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2108     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2109     return NULL;
2110   }
2111 
2112 #ifndef S390
2113   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2114     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2115     return NULL;
2116   }
2117 #endif // !S390
2118 
2119   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2120     if ( lib_arch.name!=NULL ) {
2121       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2122         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2123         lib_arch.name, arch_array[running_arch_index].name);
2124     } else {
2125       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2126       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2127         lib_arch.code,
2128         arch_array[running_arch_index].name);
2129     }
2130   }
2131 
2132   return NULL;
2133 }
2134 
2135 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2136   void * result = ::dlopen(filename, RTLD_LAZY);
2137   if (result == NULL) {
2138     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2139     ebuf[ebuflen-1] = '\0';
2140   }
2141   return result;
2142 }
2143 
2144 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2145   void * result = NULL;
2146   if (LoadExecStackDllInVMThread) {
2147     result = dlopen_helper(filename, ebuf, ebuflen);
2148   }
2149 
2150   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2151   // library that requires an executable stack, or which does not have this
2152   // stack attribute set, dlopen changes the stack attribute to executable. The
2153   // read protection of the guard pages gets lost.
2154   //
2155   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2156   // may have been queued at the same time.
2157 
2158   if (!_stack_is_executable) {
2159     JavaThread *jt = Threads::first();
2160 
2161     while (jt) {
2162       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2163           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2164         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2165                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2166           warning("Attempt to reguard stack yellow zone failed.");
2167         }
2168       }
2169       jt = jt->next();
2170     }
2171   }
2172 
2173   return result;
2174 }
2175 
2176 /*
2177  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2178  * chances are you might want to run the generated bits against glibc-2.0
2179  * libdl.so, so always use locking for any version of glibc.
2180  */
2181 void* os::dll_lookup(void* handle, const char* name) {
2182   pthread_mutex_lock(&dl_mutex);
2183   void* res = dlsym(handle, name);
2184   pthread_mutex_unlock(&dl_mutex);
2185   return res;
2186 }
2187 
2188 
2189 static bool _print_ascii_file(const char* filename, outputStream* st) {
2190   int fd = ::open(filename, O_RDONLY);
2191   if (fd == -1) {
2192      return false;
2193   }
2194 
2195   char buf[32];
2196   int bytes;
2197   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2198     st->print_raw(buf, bytes);
2199   }
2200 
2201   ::close(fd);
2202 
2203   return true;
2204 }
2205 
2206 bool _print_lsb_file(const char* filename, outputStream* st) {
2207   int fd = open(filename, O_RDONLY);
2208   if (fd == -1) {
2209      return false;
2210   }
2211 
2212   char buf[512], *d_i, *d_r, *d_c;
2213   int bytes;
2214 
2215   if ((bytes = read(fd, buf, sizeof(buf)-1)) == sizeof(buf)-1) {
2216      close(fd);
2217      return false;
2218   }
2219   close(fd);
2220 
2221   buf[bytes] = '\n';
2222   buf[bytes+1] = '\0';
2223   d_i = strstr(buf, "DISTRIB_ID=");
2224   d_r = strstr(buf, "DISTRIB_RELEASE=");
2225   d_c = strstr(buf, "DISTRIB_CODENAME=");
2226   if (!d_i || !d_r || !d_c) {
2227      return false;
2228   }
2229   d_i = strchr(d_i, '=') + 1;  *strchrnul(d_i, '\n') = '\0';
2230   d_r = strchr(d_r, '=') + 1;  *strchrnul(d_r, '\n') = '\0';
2231   d_c = strchr(d_c, '=') + 1;  *strchrnul(d_c, '\n') = '\0';
2232   st->print("%s %s (%s)", d_i, d_r, d_c);
2233 
2234   return true;
2235 }
2236 
2237 void os::print_dll_info(outputStream *st) {
2238    st->print_cr("Dynamic libraries:");
2239 
2240    char fname[32];
2241    pid_t pid = os::Linux::gettid();
2242 
2243    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2244 
2245    if (!_print_ascii_file(fname, st)) {
2246      st->print("Can not get library information for pid = %d\n", pid);
2247    }
2248 }
2249 
2250 void os::print_os_info_brief(outputStream* st) {
2251   os::Linux::print_distro_info(st);
2252 
2253   os::Posix::print_uname_info(st);
2254 
2255   os::Linux::print_libversion_info(st);
2256 
2257 }
2258 
2259 void os::print_os_info(outputStream* st) {
2260   st->print("OS:");
2261 
2262   os::Linux::print_distro_info(st);
2263 
2264   os::Posix::print_uname_info(st);
2265 
2266   // Print warning if unsafe chroot environment detected
2267   if (unsafe_chroot_detected) {
2268     st->print("WARNING!! ");
2269     st->print_cr(unstable_chroot_error);
2270   }
2271 
2272   os::Linux::print_libversion_info(st);
2273 
2274   os::Posix::print_rlimit_info(st);
2275 
2276   os::Posix::print_load_average(st);
2277 
2278   os::Linux::print_full_memory_info(st);
2279 }
2280 
2281 // Try to identify popular distros.
2282 // Most Linux distributions have /etc/XXX-release file, which contains
2283 // the OS version string. Some have more than one /etc/XXX-release file
2284 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2285 // so the order is important.
2286 void os::Linux::print_distro_info(outputStream* st) {
2287   if (!_print_ascii_file("/etc/mandrake-release", st) &&
2288       !_print_ascii_file("/etc/sun-release", st) &&
2289       !_print_ascii_file("/etc/redhat-release", st) &&
2290       !_print_ascii_file("/etc/SuSE-release", st) &&
2291       !_print_ascii_file("/etc/turbolinux-release", st) &&
2292       !_print_ascii_file("/etc/gentoo-release", st) &&
2293       !_print_lsb_file("/etc/lsb-release", st) &&
2294       !_print_ascii_file("/etc/debian_version", st) &&
2295       !_print_ascii_file("/etc/ltib-release", st) &&
2296       !_print_ascii_file("/etc/angstrom-version", st)) {
2297       st->print("Linux");
2298   }
2299   st->cr();
2300 }
2301 
2302 void os::Linux::print_libversion_info(outputStream* st) {
2303   // libc, pthread
2304   st->print("libc:");
2305   st->print(os::Linux::glibc_version()); st->print(" ");
2306   st->print(os::Linux::libpthread_version()); st->print(" ");
2307   if (os::Linux::is_LinuxThreads()) {
2308      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2309   }
2310   st->cr();
2311 }
2312 
2313 void os::Linux::print_full_memory_info(outputStream* st) {
2314    st->print("\n/proc/meminfo:\n");
2315    _print_ascii_file("/proc/meminfo", st);
2316    st->cr();
2317 }
2318 
2319 void os::print_memory_info(outputStream* st) {
2320 
2321   st->print("Memory:");
2322   st->print(" %dk page", os::vm_page_size()>>10);
2323 
2324   // values in struct sysinfo are "unsigned long"
2325   struct sysinfo si;
2326   sysinfo(&si);
2327 
2328   st->print(", physical " UINT64_FORMAT "k",
2329             os::physical_memory() >> 10);
2330   st->print("(" UINT64_FORMAT "k free)",
2331             os::available_memory() >> 10);
2332   st->print(", swap " UINT64_FORMAT "k",
2333             ((jlong)si.totalswap * si.mem_unit) >> 10);
2334   st->print("(" UINT64_FORMAT "k free)",
2335             ((jlong)si.freeswap * si.mem_unit) >> 10);
2336   st->cr();
2337 }
2338 
2339 void os::pd_print_cpu_info(outputStream* st) {
2340   st->print("\n/proc/cpuinfo:\n");
2341   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2342     st->print("  <Not Available>");
2343   }
2344   st->cr();
2345 }
2346 
2347 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2348 // but they're the same for all the linux arch that we support
2349 // and they're the same for solaris but there's no common place to put this.
2350 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2351                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2352                           "ILL_COPROC", "ILL_BADSTK" };
2353 
2354 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2355                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2356                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2357 
2358 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2359 
2360 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2361 
2362 void os::print_siginfo(outputStream* st, void* siginfo) {
2363   st->print("siginfo:");
2364 
2365   const int buflen = 100;
2366   char buf[buflen];
2367   siginfo_t *si = (siginfo_t*)siginfo;
2368   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2369   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2370     st->print("si_errno=%s", buf);
2371   } else {
2372     st->print("si_errno=%d", si->si_errno);
2373   }
2374   const int c = si->si_code;
2375   assert(c > 0, "unexpected si_code");
2376   switch (si->si_signo) {
2377   case SIGILL:
2378     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2379     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2380     break;
2381   case SIGFPE:
2382     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2383     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2384     break;
2385   case SIGSEGV:
2386     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2387     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2388     break;
2389   case SIGBUS:
2390     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2391     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2392     break;
2393   default:
2394     st->print(", si_code=%d", si->si_code);
2395     // no si_addr
2396   }
2397 
2398   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2399       UseSharedSpaces) {
2400     FileMapInfo* mapinfo = FileMapInfo::current_info();
2401     if (mapinfo->is_in_shared_space(si->si_addr)) {
2402       st->print("\n\nError accessing class data sharing archive."   \
2403                 " Mapped file inaccessible during execution, "      \
2404                 " possible disk/network problem.");
2405     }
2406   }
2407   st->cr();
2408 }
2409 
2410 
2411 static void print_signal_handler(outputStream* st, int sig,
2412                                  char* buf, size_t buflen);
2413 
2414 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2415   st->print_cr("Signal Handlers:");
2416   print_signal_handler(st, SIGSEGV, buf, buflen);
2417   print_signal_handler(st, SIGBUS , buf, buflen);
2418   print_signal_handler(st, SIGFPE , buf, buflen);
2419   print_signal_handler(st, SIGPIPE, buf, buflen);
2420   print_signal_handler(st, SIGXFSZ, buf, buflen);
2421   print_signal_handler(st, SIGILL , buf, buflen);
2422   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2423   print_signal_handler(st, SR_signum, buf, buflen);
2424   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2425   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2426   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2427   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2428   print_signal_handler(st, SIGTRAP, buf, buflen);
2429 }
2430 
2431 static char saved_jvm_path[MAXPATHLEN] = {0};
2432 
2433 // Find the full path to the current module, libjvm.so or libjvm_g.so
2434 void os::jvm_path(char *buf, jint buflen) {
2435   // Error checking.
2436   if (buflen < MAXPATHLEN) {
2437     assert(false, "must use a large-enough buffer");
2438     buf[0] = '\0';
2439     return;
2440   }
2441   // Lazy resolve the path to current module.
2442   if (saved_jvm_path[0] != 0) {
2443     strcpy(buf, saved_jvm_path);
2444     return;
2445   }
2446 
2447   char dli_fname[MAXPATHLEN];
2448   bool ret = dll_address_to_library_name(
2449                 CAST_FROM_FN_PTR(address, os::jvm_path),
2450                 dli_fname, sizeof(dli_fname), NULL);
2451   assert(ret, "cannot locate libjvm");
2452   char *rp = NULL;
2453   if (ret && dli_fname[0] != '\0') {
2454     rp = realpath(dli_fname, buf);
2455   }
2456   if (rp == NULL)
2457     return;
2458 
2459   if (Arguments::created_by_gamma_launcher()) {
2460     // Support for the gamma launcher.  Typical value for buf is
2461     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2462     // the right place in the string, then assume we are installed in a JDK and
2463     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2464     // up the path so it looks like libjvm.so is installed there (append a
2465     // fake suffix hotspot/libjvm.so).
2466     const char *p = buf + strlen(buf) - 1;
2467     for (int count = 0; p > buf && count < 5; ++count) {
2468       for (--p; p > buf && *p != '/'; --p)
2469         /* empty */ ;
2470     }
2471 
2472     if (strncmp(p, "/jre/lib/", 9) != 0) {
2473       // Look for JAVA_HOME in the environment.
2474       char* java_home_var = ::getenv("JAVA_HOME");
2475       if (java_home_var != NULL && java_home_var[0] != 0) {
2476         char* jrelib_p;
2477         int len;
2478 
2479         // Check the current module name "libjvm.so" or "libjvm_g.so".
2480         p = strrchr(buf, '/');
2481         assert(strstr(p, "/libjvm") == p, "invalid library name");
2482         p = strstr(p, "_g") ? "_g" : "";
2483 
2484         rp = realpath(java_home_var, buf);
2485         if (rp == NULL)
2486           return;
2487 
2488         // determine if this is a legacy image or modules image
2489         // modules image doesn't have "jre" subdirectory
2490         len = strlen(buf);
2491         assert(len < buflen, "Ran out of buffer room");
2492         jrelib_p = buf + len;
2493         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2494         if (0 != access(buf, F_OK)) {
2495           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2496         }
2497 
2498         if (0 == access(buf, F_OK)) {
2499           // Use current module name "libjvm[_g].so" instead of
2500           // "libjvm"debug_only("_g")".so" since for fastdebug version
2501           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2502           len = strlen(buf);
2503           snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2504         } else {
2505           // Go back to path of .so
2506           rp = realpath(dli_fname, buf);
2507           if (rp == NULL)
2508             return;
2509         }
2510       }
2511     }
2512   }
2513 
2514   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2515 }
2516 
2517 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2518   // no prefix required, not even "_"
2519 }
2520 
2521 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2522   // no suffix required
2523 }
2524 
2525 ////////////////////////////////////////////////////////////////////////////////
2526 // sun.misc.Signal support
2527 
2528 static volatile jint sigint_count = 0;
2529 
2530 static void
2531 UserHandler(int sig, void *siginfo, void *context) {
2532   // 4511530 - sem_post is serialized and handled by the manager thread. When
2533   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2534   // don't want to flood the manager thread with sem_post requests.
2535   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2536       return;
2537 
2538   // Ctrl-C is pressed during error reporting, likely because the error
2539   // handler fails to abort. Let VM die immediately.
2540   if (sig == SIGINT && is_error_reported()) {
2541      os::die();
2542   }
2543 
2544   os::signal_notify(sig);
2545 }
2546 
2547 void* os::user_handler() {
2548   return CAST_FROM_FN_PTR(void*, UserHandler);
2549 }
2550 
2551 class Semaphore : public StackObj {
2552   public:
2553     Semaphore();
2554     ~Semaphore();
2555     void signal();
2556     void wait();
2557     bool trywait();
2558     bool timedwait(unsigned int sec, int nsec);
2559   private:
2560     sem_t _semaphore;
2561 };
2562 
2563 Semaphore::Semaphore() {
2564   sem_init(&_semaphore, 0, 0);
2565 }
2566 
2567 Semaphore::~Semaphore() {
2568   sem_destroy(&_semaphore);
2569 }
2570 
2571 void Semaphore::signal() {
2572   sem_post(&_semaphore);
2573 }
2574 
2575 void Semaphore::wait() {
2576   sem_wait(&_semaphore);
2577 }
2578 
2579 bool Semaphore::trywait() {
2580   return sem_trywait(&_semaphore) == 0;
2581 }
2582 
2583 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2584 
2585   struct timespec ts;
2586   // Semaphore's are always associated with CLOCK_REALTIME
2587   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2588   // see unpackTime for discussion on overflow checking
2589   if (sec >= MAX_SECS) {
2590     ts.tv_sec += MAX_SECS;
2591     ts.tv_nsec = 0;
2592   } else {
2593     ts.tv_sec += sec;
2594     ts.tv_nsec += nsec;
2595     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2596       ts.tv_nsec -= NANOSECS_PER_SEC;
2597       ++ts.tv_sec; // note: this must be <= max_secs
2598     }
2599   }
2600 
2601   while (1) {
2602     int result = sem_timedwait(&_semaphore, &ts);
2603     if (result == 0) {
2604       return true;
2605     } else if (errno == EINTR) {
2606       continue;
2607     } else if (errno == ETIMEDOUT) {
2608       return false;
2609     } else {
2610       return false;
2611     }
2612   }
2613 }
2614 
2615 extern "C" {
2616   typedef void (*sa_handler_t)(int);
2617   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2618 }
2619 
2620 void* os::signal(int signal_number, void* handler) {
2621   struct sigaction sigAct, oldSigAct;
2622 
2623   sigfillset(&(sigAct.sa_mask));
2624   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2625   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2626 
2627   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2628     // -1 means registration failed
2629     return (void *)-1;
2630   }
2631 
2632   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2633 }
2634 
2635 void os::signal_raise(int signal_number) {
2636   ::raise(signal_number);
2637 }
2638 
2639 /*
2640  * The following code is moved from os.cpp for making this
2641  * code platform specific, which it is by its very nature.
2642  */
2643 
2644 // Will be modified when max signal is changed to be dynamic
2645 int os::sigexitnum_pd() {
2646   return NSIG;
2647 }
2648 
2649 // a counter for each possible signal value
2650 static volatile jint pending_signals[NSIG+1] = { 0 };
2651 
2652 // Linux(POSIX) specific hand shaking semaphore.
2653 static sem_t sig_sem;
2654 static Semaphore sr_semaphore;
2655 
2656 void os::signal_init_pd() {
2657   // Initialize signal structures
2658   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2659 
2660   // Initialize signal semaphore
2661   ::sem_init(&sig_sem, 0, 0);
2662 }
2663 
2664 void os::signal_notify(int sig) {
2665   Atomic::inc(&pending_signals[sig]);
2666   ::sem_post(&sig_sem);
2667 }
2668 
2669 static int check_pending_signals(bool wait) {
2670   Atomic::store(0, &sigint_count);
2671   for (;;) {
2672     for (int i = 0; i < NSIG + 1; i++) {
2673       jint n = pending_signals[i];
2674       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2675         return i;
2676       }
2677     }
2678     if (!wait) {
2679       return -1;
2680     }
2681     JavaThread *thread = JavaThread::current();
2682     ThreadBlockInVM tbivm(thread);
2683 
2684     bool threadIsSuspended;
2685     do {
2686       thread->set_suspend_equivalent();
2687       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2688       ::sem_wait(&sig_sem);
2689 
2690       // were we externally suspended while we were waiting?
2691       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2692       if (threadIsSuspended) {
2693         //
2694         // The semaphore has been incremented, but while we were waiting
2695         // another thread suspended us. We don't want to continue running
2696         // while suspended because that would surprise the thread that
2697         // suspended us.
2698         //
2699         ::sem_post(&sig_sem);
2700 
2701         thread->java_suspend_self();
2702       }
2703     } while (threadIsSuspended);
2704   }
2705 }
2706 
2707 int os::signal_lookup() {
2708   return check_pending_signals(false);
2709 }
2710 
2711 int os::signal_wait() {
2712   return check_pending_signals(true);
2713 }
2714 
2715 ////////////////////////////////////////////////////////////////////////////////
2716 // Virtual Memory
2717 
2718 int os::vm_page_size() {
2719   // Seems redundant as all get out
2720   assert(os::Linux::page_size() != -1, "must call os::init");
2721   return os::Linux::page_size();
2722 }
2723 
2724 // Solaris allocates memory by pages.
2725 int os::vm_allocation_granularity() {
2726   assert(os::Linux::page_size() != -1, "must call os::init");
2727   return os::Linux::page_size();
2728 }
2729 
2730 // Rationale behind this function:
2731 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2732 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2733 //  samples for JITted code. Here we create private executable mapping over the code cache
2734 //  and then we can use standard (well, almost, as mapping can change) way to provide
2735 //  info for the reporting script by storing timestamp and location of symbol
2736 void linux_wrap_code(char* base, size_t size) {
2737   static volatile jint cnt = 0;
2738 
2739   if (!UseOprofile) {
2740     return;
2741   }
2742 
2743   char buf[PATH_MAX+1];
2744   int num = Atomic::add(1, &cnt);
2745 
2746   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2747            os::get_temp_directory(), os::current_process_id(), num);
2748   unlink(buf);
2749 
2750   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2751 
2752   if (fd != -1) {
2753     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2754     if (rv != (off_t)-1) {
2755       if (::write(fd, "", 1) == 1) {
2756         mmap(base, size,
2757              PROT_READ|PROT_WRITE|PROT_EXEC,
2758              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2759       }
2760     }
2761     ::close(fd);
2762     unlink(buf);
2763   }
2764 }
2765 
2766 static bool recoverable_mmap_error(int err) {
2767   // See if the error is one we can let the caller handle. This
2768   // list of errno values comes from JBS-6843484. I can't find a
2769   // Linux man page that documents this specific set of errno
2770   // values so while this list currently matches Solaris, it may
2771   // change as we gain experience with this failure mode.
2772   switch (err) {
2773   case EBADF:
2774   case EINVAL:
2775   case ENOTSUP:
2776     // let the caller deal with these errors
2777     return true;
2778 
2779   default:
2780     // Any remaining errors on this OS can cause our reserved mapping
2781     // to be lost. That can cause confusion where different data
2782     // structures think they have the same memory mapped. The worst
2783     // scenario is if both the VM and a library think they have the
2784     // same memory mapped.
2785     return false;
2786   }
2787 }
2788 
2789 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2790                                     int err) {
2791   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2792           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2793           strerror(err), err);
2794 }
2795 
2796 static void warn_fail_commit_memory(char* addr, size_t size,
2797                                     size_t alignment_hint, bool exec,
2798                                     int err) {
2799   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2800           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2801           alignment_hint, exec, strerror(err), err);
2802 }
2803 
2804 // NOTE: Linux kernel does not really reserve the pages for us.
2805 //       All it does is to check if there are enough free pages
2806 //       left at the time of mmap(). This could be a potential
2807 //       problem.
2808 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2809   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2810   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2811                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2812   if (res != (uintptr_t) MAP_FAILED) {
2813     if (UseNUMAInterleaving) {
2814       numa_make_global(addr, size);
2815     }
2816     return 0;
2817   }
2818 
2819   int err = errno;  // save errno from mmap() call above
2820 
2821   if (!recoverable_mmap_error(err)) {
2822     warn_fail_commit_memory(addr, size, exec, err);
2823     vm_exit_out_of_memory(size, "committing reserved memory.");
2824   }
2825 
2826   return err;
2827 }
2828 
2829 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2830   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2831 }
2832 
2833 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2834                                   const char* mesg) {
2835   assert(mesg != NULL, "mesg must be specified");
2836   int err = os::Linux::commit_memory_impl(addr, size, exec);
2837   if (err != 0) {
2838     // the caller wants all commit errors to exit with the specified mesg:
2839     warn_fail_commit_memory(addr, size, exec, err);
2840     vm_exit_out_of_memory(size, mesg);
2841   }
2842 }
2843 
2844 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2845 #ifndef MAP_HUGETLB
2846 #define MAP_HUGETLB 0x40000
2847 #endif
2848 
2849 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2850 #ifndef MADV_HUGEPAGE
2851 #define MADV_HUGEPAGE 14
2852 #endif
2853 
2854 int os::Linux::commit_memory_impl(char* addr, size_t size,
2855                                   size_t alignment_hint, bool exec) {
2856   int err = os::Linux::commit_memory_impl(addr, size, exec);
2857   if (err == 0) {
2858     realign_memory(addr, size, alignment_hint);
2859   }
2860   return err;
2861 }
2862 
2863 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2864                           bool exec) {
2865   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2866 }
2867 
2868 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2869                                   size_t alignment_hint, bool exec,
2870                                   const char* mesg) {
2871   assert(mesg != NULL, "mesg must be specified");
2872   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2873   if (err != 0) {
2874     // the caller wants all commit errors to exit with the specified mesg:
2875     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2876     vm_exit_out_of_memory(size, mesg);
2877   }
2878 }
2879 
2880 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2881   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2882     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2883     // be supported or the memory may already be backed by huge pages.
2884     ::madvise(addr, bytes, MADV_HUGEPAGE);
2885   }
2886 }
2887 
2888 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2889   // This method works by doing an mmap over an existing mmaping and effectively discarding
2890   // the existing pages. However it won't work for SHM-based large pages that cannot be
2891   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2892   // small pages on top of the SHM segment. This method always works for small pages, so we
2893   // allow that in any case.
2894   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2895     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2896   }
2897 }
2898 
2899 void os::numa_make_global(char *addr, size_t bytes) {
2900   Linux::numa_interleave_memory(addr, bytes);
2901 }
2902 
2903 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2904   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2905 }
2906 
2907 bool os::numa_topology_changed()   { return false; }
2908 
2909 size_t os::numa_get_groups_num() {
2910   int max_node = Linux::numa_max_node();
2911   return max_node > 0 ? max_node + 1 : 1;
2912 }
2913 
2914 int os::numa_get_group_id() {
2915   int cpu_id = Linux::sched_getcpu();
2916   if (cpu_id != -1) {
2917     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2918     if (lgrp_id != -1) {
2919       return lgrp_id;
2920     }
2921   }
2922   return 0;
2923 }
2924 
2925 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2926   for (size_t i = 0; i < size; i++) {
2927     ids[i] = i;
2928   }
2929   return size;
2930 }
2931 
2932 bool os::get_page_info(char *start, page_info* info) {
2933   return false;
2934 }
2935 
2936 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2937   return end;
2938 }
2939 
2940 
2941 int os::Linux::sched_getcpu_syscall(void) {
2942   unsigned int cpu;
2943   int retval = -1;
2944 
2945 #if defined(AMD64) || defined(BUILTIN_SIM)
2946 // Unfortunately we have to bring all these macros here from vsyscall.h
2947 // to be able to compile on old linuxes.
2948 # define __NR_vgetcpu 2
2949 # define VSYSCALL_START (-10UL << 20)
2950 # define VSYSCALL_SIZE 1024
2951 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2952   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2953   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2954   retval = vgetcpu(&cpu, NULL, NULL);
2955 #elif defined(IA32) || defined(AARCH64)
2956 # ifndef SYS_getcpu
2957 # define SYS_getcpu AARCH64_ONLY(168) NOT_AARCH64(318)
2958 # endif
2959   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2960 #endif
2961 
2962   return (retval == -1) ? retval : cpu;
2963 }
2964 
2965 // Something to do with the numa-aware allocator needs these symbols
2966 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2967 extern "C" JNIEXPORT void numa_error(char *where) { }
2968 extern "C" JNIEXPORT int fork1() { return fork(); }
2969 
2970 
2971 // If we are running with libnuma version > 2, then we should
2972 // be trying to use symbols with versions 1.1
2973 // If we are running with earlier version, which did not have symbol versions,
2974 // we should use the base version.
2975 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2976   void *f = dlvsym(handle, name, "libnuma_1.1");
2977   if (f == NULL) {
2978     f = dlsym(handle, name);
2979   }
2980   return f;
2981 }
2982 
2983 bool os::Linux::libnuma_init() {
2984   // sched_getcpu() should be in libc.
2985   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2986                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2987 
2988   // If it's not, try a direct syscall.
2989   if (sched_getcpu() == -1)
2990     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2991 
2992   if (sched_getcpu() != -1) { // Does it work?
2993     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2994     if (handle != NULL) {
2995       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2996                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2997       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2998                                        libnuma_dlsym(handle, "numa_max_node")));
2999       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3000                                         libnuma_dlsym(handle, "numa_available")));
3001       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3002                                             libnuma_dlsym(handle, "numa_tonode_memory")));
3003       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3004                                             libnuma_dlsym(handle, "numa_interleave_memory")));
3005 
3006 
3007       if (numa_available() != -1) {
3008         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3009         // Create a cpu -> node mapping
3010         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3011         rebuild_cpu_to_node_map();
3012         return true;
3013       }
3014     }
3015   }
3016   return false;
3017 }
3018 
3019 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3020 // The table is later used in get_node_by_cpu().
3021 void os::Linux::rebuild_cpu_to_node_map() {
3022   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3023                               // in libnuma (possible values are starting from 16,
3024                               // and continuing up with every other power of 2, but less
3025                               // than the maximum number of CPUs supported by kernel), and
3026                               // is a subject to change (in libnuma version 2 the requirements
3027                               // are more reasonable) we'll just hardcode the number they use
3028                               // in the library.
3029   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3030 
3031   size_t cpu_num = os::active_processor_count();
3032   size_t cpu_map_size = NCPUS / BitsPerCLong;
3033   size_t cpu_map_valid_size =
3034     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3035 
3036   cpu_to_node()->clear();
3037   cpu_to_node()->at_grow(cpu_num - 1);
3038   size_t node_num = numa_get_groups_num();
3039 
3040   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3041   for (size_t i = 0; i < node_num; i++) {
3042     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3043       for (size_t j = 0; j < cpu_map_valid_size; j++) {
3044         if (cpu_map[j] != 0) {
3045           for (size_t k = 0; k < BitsPerCLong; k++) {
3046             if (cpu_map[j] & (1UL << k)) {
3047               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
3048             }
3049           }
3050         }
3051       }
3052     }
3053   }
3054   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
3055 }
3056 
3057 int os::Linux::get_node_by_cpu(int cpu_id) {
3058   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3059     return cpu_to_node()->at(cpu_id);
3060   }
3061   return -1;
3062 }
3063 
3064 GrowableArray<int>* os::Linux::_cpu_to_node;
3065 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3066 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3067 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3068 os::Linux::numa_available_func_t os::Linux::_numa_available;
3069 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3070 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3071 unsigned long* os::Linux::_numa_all_nodes;
3072 
3073 bool os::pd_uncommit_memory(char* addr, size_t size) {
3074   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3075                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3076   return res  != (uintptr_t) MAP_FAILED;
3077 }
3078 
3079 static
3080 address get_stack_commited_bottom(address bottom, size_t size) {
3081   address nbot = bottom;
3082   address ntop = bottom + size;
3083 
3084   size_t page_sz = os::vm_page_size();
3085   unsigned pages = size / page_sz;
3086 
3087   unsigned char vec[1];
3088   unsigned imin = 1, imax = pages + 1, imid;
3089   int mincore_return_value;
3090 
3091   while (imin < imax) {
3092     imid = (imax + imin) / 2;
3093     nbot = ntop - (imid * page_sz);
3094 
3095     // Use a trick with mincore to check whether the page is mapped or not.
3096     // mincore sets vec to 1 if page resides in memory and to 0 if page
3097     // is swapped output but if page we are asking for is unmapped
3098     // it returns -1,ENOMEM
3099     mincore_return_value = mincore(nbot, page_sz, vec);
3100 
3101     if (mincore_return_value == -1) {
3102       // Page is not mapped go up
3103       // to find first mapped page
3104       if (errno != EAGAIN) {
3105         assert(errno == ENOMEM, "Unexpected mincore errno");
3106         imax = imid;
3107       }
3108     } else {
3109       // Page is mapped go down
3110       // to find first not mapped page
3111       imin = imid + 1;
3112     }
3113   }
3114 
3115   nbot = nbot + page_sz;
3116 
3117   // Adjust stack bottom one page up if last checked page is not mapped
3118   if (mincore_return_value == -1) {
3119     nbot = nbot + page_sz;
3120   }
3121 
3122   return nbot;
3123 }
3124 
3125 
3126 // Linux uses a growable mapping for the stack, and if the mapping for
3127 // the stack guard pages is not removed when we detach a thread the
3128 // stack cannot grow beyond the pages where the stack guard was
3129 // mapped.  If at some point later in the process the stack expands to
3130 // that point, the Linux kernel cannot expand the stack any further
3131 // because the guard pages are in the way, and a segfault occurs.
3132 //
3133 // However, it's essential not to split the stack region by unmapping
3134 // a region (leaving a hole) that's already part of the stack mapping,
3135 // so if the stack mapping has already grown beyond the guard pages at
3136 // the time we create them, we have to truncate the stack mapping.
3137 // So, we need to know the extent of the stack mapping when
3138 // create_stack_guard_pages() is called.
3139 
3140 // We only need this for stacks that are growable: at the time of
3141 // writing thread stacks don't use growable mappings (i.e. those
3142 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3143 // only applies to the main thread.
3144 
3145 // If the (growable) stack mapping already extends beyond the point
3146 // where we're going to put our guard pages, truncate the mapping at
3147 // that point by munmap()ping it.  This ensures that when we later
3148 // munmap() the guard pages we don't leave a hole in the stack
3149 // mapping. This only affects the main/initial thread
3150 
3151 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3152 
3153   if (os::Linux::is_initial_thread()) {
3154     // As we manually grow stack up to bottom inside create_attached_thread(),
3155     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3156     // we don't need to do anything special.
3157     // Check it first, before calling heavy function.
3158     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3159     unsigned char vec[1];
3160 
3161     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3162       // Fallback to slow path on all errors, including EAGAIN
3163       stack_extent = (uintptr_t) get_stack_commited_bottom(
3164                                     os::Linux::initial_thread_stack_bottom(),
3165                                     (size_t)addr - stack_extent);
3166     }
3167 
3168     if (stack_extent < (uintptr_t)addr) {
3169       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3170     }
3171   }
3172 
3173   return os::commit_memory(addr, size, !ExecMem);
3174 }
3175 
3176 // If this is a growable mapping, remove the guard pages entirely by
3177 // munmap()ping them.  If not, just call uncommit_memory(). This only
3178 // affects the main/initial thread, but guard against future OS changes
3179 // It's safe to always unmap guard pages for initial thread because we
3180 // always place it right after end of the mapped region
3181 
3182 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3183   uintptr_t stack_extent, stack_base;
3184 
3185   if (os::Linux::is_initial_thread()) {
3186     return ::munmap(addr, size) == 0;
3187   }
3188 
3189   return os::uncommit_memory(addr, size);
3190 }
3191 
3192 static address _highest_vm_reserved_address = NULL;
3193 
3194 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3195 // at 'requested_addr'. If there are existing memory mappings at the same
3196 // location, however, they will be overwritten. If 'fixed' is false,
3197 // 'requested_addr' is only treated as a hint, the return value may or
3198 // may not start from the requested address. Unlike Linux mmap(), this
3199 // function returns NULL to indicate failure.
3200 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3201   char * addr;
3202   int flags;
3203 
3204   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3205   if (fixed) {
3206     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3207     flags |= MAP_FIXED;
3208   }
3209 
3210   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
3211   // to PROT_EXEC if executable when we commit the page.
3212   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
3213                        flags, -1, 0);
3214 
3215   if (addr != MAP_FAILED) {
3216     // anon_mmap() should only get called during VM initialization,
3217     // don't need lock (actually we can skip locking even it can be called
3218     // from multiple threads, because _highest_vm_reserved_address is just a
3219     // hint about the upper limit of non-stack memory regions.)
3220     if ((address)addr + bytes > _highest_vm_reserved_address) {
3221       _highest_vm_reserved_address = (address)addr + bytes;
3222     }
3223   }
3224 
3225   return addr == MAP_FAILED ? NULL : addr;
3226 }
3227 
3228 // Don't update _highest_vm_reserved_address, because there might be memory
3229 // regions above addr + size. If so, releasing a memory region only creates
3230 // a hole in the address space, it doesn't help prevent heap-stack collision.
3231 //
3232 static int anon_munmap(char * addr, size_t size) {
3233   return ::munmap(addr, size) == 0;
3234 }
3235 
3236 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3237                          size_t alignment_hint) {
3238   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3239 }
3240 
3241 bool os::pd_release_memory(char* addr, size_t size) {
3242   return anon_munmap(addr, size);
3243 }
3244 
3245 static address highest_vm_reserved_address() {
3246   return _highest_vm_reserved_address;
3247 }
3248 
3249 static bool linux_mprotect(char* addr, size_t size, int prot) {
3250   // Linux wants the mprotect address argument to be page aligned.
3251   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3252 
3253   // According to SUSv3, mprotect() should only be used with mappings
3254   // established by mmap(), and mmap() always maps whole pages. Unaligned
3255   // 'addr' likely indicates problem in the VM (e.g. trying to change
3256   // protection of malloc'ed or statically allocated memory). Check the
3257   // caller if you hit this assert.
3258   assert(addr == bottom, "sanity check");
3259 
3260   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3261   return ::mprotect(bottom, size, prot) == 0;
3262 }
3263 
3264 // Set protections specified
3265 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3266                         bool is_committed) {
3267   unsigned int p = 0;
3268   switch (prot) {
3269   case MEM_PROT_NONE: p = PROT_NONE; break;
3270   case MEM_PROT_READ: p = PROT_READ; break;
3271   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3272   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3273   default:
3274     ShouldNotReachHere();
3275   }
3276   // is_committed is unused.
3277   return linux_mprotect(addr, bytes, p);
3278 }
3279 
3280 bool os::guard_memory(char* addr, size_t size) {
3281   return linux_mprotect(addr, size, PROT_NONE);
3282 }
3283 
3284 bool os::unguard_memory(char* addr, size_t size) {
3285   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3286 }
3287 
3288 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3289   bool result = false;
3290   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3291                  MAP_ANONYMOUS|MAP_PRIVATE,
3292                  -1, 0);
3293   if (p != MAP_FAILED) {
3294     void *aligned_p = align_ptr_up(p, page_size);
3295 
3296     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3297 
3298     munmap(p, page_size * 2);
3299   }
3300 
3301   if (warn && !result) {
3302     warning("TransparentHugePages is not supported by the operating system.");
3303   }
3304 
3305   return result;
3306 }
3307 
3308 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3309   bool result = false;
3310   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3311                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3312                  -1, 0);
3313 
3314   if (p != MAP_FAILED) {
3315     // We don't know if this really is a huge page or not.
3316     FILE *fp = fopen("/proc/self/maps", "r");
3317     if (fp) {
3318       while (!feof(fp)) {
3319         char chars[257];
3320         long x = 0;
3321         if (fgets(chars, sizeof(chars), fp)) {
3322           if (sscanf(chars, "%lx-%*x", &x) == 1
3323               && x == (long)p) {
3324             if (strstr (chars, "hugepage")) {
3325               result = true;
3326               break;
3327             }
3328           }
3329         }
3330       }
3331       fclose(fp);
3332     }
3333     munmap(p, page_size);
3334   }
3335 
3336   if (warn && !result) {
3337     warning("HugeTLBFS is not supported by the operating system.");
3338   }
3339 
3340   return result;
3341 }
3342 
3343 /*
3344 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3345 *
3346 * From the coredump_filter documentation:
3347 *
3348 * - (bit 0) anonymous private memory
3349 * - (bit 1) anonymous shared memory
3350 * - (bit 2) file-backed private memory
3351 * - (bit 3) file-backed shared memory
3352 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3353 *           effective only if the bit 2 is cleared)
3354 * - (bit 5) hugetlb private memory
3355 * - (bit 6) hugetlb shared memory
3356 */
3357 static void set_coredump_filter(void) {
3358   FILE *f;
3359   long cdm;
3360 
3361   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3362     return;
3363   }
3364 
3365   if (fscanf(f, "%lx", &cdm) != 1) {
3366     fclose(f);
3367     return;
3368   }
3369 
3370   rewind(f);
3371 
3372   if ((cdm & LARGEPAGES_BIT) == 0) {
3373     cdm |= LARGEPAGES_BIT;
3374     fprintf(f, "%#lx", cdm);
3375   }
3376 
3377   fclose(f);
3378 }
3379 
3380 // Large page support
3381 
3382 static size_t _large_page_size = 0;
3383 
3384 size_t os::Linux::find_large_page_size() {
3385   size_t large_page_size = 0;
3386 
3387   // large_page_size on Linux is used to round up heap size. x86 uses either
3388   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3389   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3390   // page as large as 256M.
3391   //
3392   // Here we try to figure out page size by parsing /proc/meminfo and looking
3393   // for a line with the following format:
3394   //    Hugepagesize:     2048 kB
3395   //
3396   // If we can't determine the value (e.g. /proc is not mounted, or the text
3397   // format has been changed), we'll use the largest page size supported by
3398   // the processor.
3399 
3400 #ifndef ZERO
3401   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3402                      ARM_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3403 #endif // ZERO
3404 
3405   FILE *fp = fopen("/proc/meminfo", "r");
3406   if (fp) {
3407     while (!feof(fp)) {
3408       int x = 0;
3409       char buf[16];
3410       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3411         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3412           large_page_size = x * K;
3413           break;
3414         }
3415       } else {
3416         // skip to next line
3417         for (;;) {
3418           int ch = fgetc(fp);
3419           if (ch == EOF || ch == (int)'\n') break;
3420         }
3421       }
3422     }
3423     fclose(fp);
3424   }
3425 
3426   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3427     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3428         SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3429         proper_unit_for_byte_size(large_page_size));
3430   }
3431 
3432   return large_page_size;
3433 }
3434 
3435 size_t os::Linux::setup_large_page_size() {
3436   _large_page_size = Linux::find_large_page_size();
3437   const size_t default_page_size = (size_t)Linux::page_size();
3438   if (_large_page_size > default_page_size) {
3439     _page_sizes[0] = _large_page_size;
3440     _page_sizes[1] = default_page_size;
3441     _page_sizes[2] = 0;
3442   }
3443 
3444   return _large_page_size;
3445 }
3446 
3447 bool os::Linux::setup_large_page_type(size_t page_size) {
3448   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3449       FLAG_IS_DEFAULT(UseSHM) &&
3450       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3451 
3452     // The type of large pages has not been specified by the user.
3453 
3454     // Try UseHugeTLBFS and then UseSHM.
3455     UseHugeTLBFS = UseSHM = true;
3456 
3457     // Don't try UseTransparentHugePages since there are known
3458     // performance issues with it turned on. This might change in the future.
3459     UseTransparentHugePages = false;
3460   }
3461 
3462   if (UseTransparentHugePages) {
3463     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3464     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3465       UseHugeTLBFS = false;
3466       UseSHM = false;
3467       return true;
3468     }
3469     UseTransparentHugePages = false;
3470   }
3471 
3472   if (UseHugeTLBFS) {
3473     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3474     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3475       UseSHM = false;
3476       return true;
3477     }
3478     UseHugeTLBFS = false;
3479   }
3480 
3481   return UseSHM;
3482 }
3483 
3484 void os::large_page_init() {
3485   if (!UseLargePages &&
3486       !UseTransparentHugePages &&
3487       !UseHugeTLBFS &&
3488       !UseSHM) {
3489     // Not using large pages.
3490     return;
3491   }
3492 
3493   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3494     // The user explicitly turned off large pages.
3495     // Ignore the rest of the large pages flags.
3496     UseTransparentHugePages = false;
3497     UseHugeTLBFS = false;
3498     UseSHM = false;
3499     return;
3500   }
3501 
3502   size_t large_page_size = Linux::setup_large_page_size();
3503   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3504 
3505   set_coredump_filter();
3506 }
3507 
3508 #ifndef SHM_HUGETLB
3509 #define SHM_HUGETLB 04000
3510 #endif
3511 
3512 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3513   // "exec" is passed in but not used.  Creating the shared image for
3514   // the code cache doesn't have an SHM_X executable permission to check.
3515   assert(UseLargePages && UseSHM, "only for SHM large pages");
3516   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3517 
3518   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3519     return NULL; // Fallback to small pages.
3520   }
3521 
3522   key_t key = IPC_PRIVATE;
3523   char *addr;
3524 
3525   bool warn_on_failure = UseLargePages &&
3526                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3527                          !FLAG_IS_DEFAULT(UseSHM) ||
3528                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3529                         );
3530   char msg[128];
3531 
3532   // Create a large shared memory region to attach to based on size.
3533   // Currently, size is the total size of the heap
3534   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3535   if (shmid == -1) {
3536      // Possible reasons for shmget failure:
3537      // 1. shmmax is too small for Java heap.
3538      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3539      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3540      // 2. not enough large page memory.
3541      //    > check available large pages: cat /proc/meminfo
3542      //    > increase amount of large pages:
3543      //          echo new_value > /proc/sys/vm/nr_hugepages
3544      //      Note 1: different Linux may use different name for this property,
3545      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3546      //      Note 2: it's possible there's enough physical memory available but
3547      //            they are so fragmented after a long run that they can't
3548      //            coalesce into large pages. Try to reserve large pages when
3549      //            the system is still "fresh".
3550      if (warn_on_failure) {
3551        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3552        warning(msg);
3553      }
3554      return NULL;
3555   }
3556 
3557   // attach to the region
3558   addr = (char*)shmat(shmid, req_addr, 0);
3559   int err = errno;
3560 
3561   // Remove shmid. If shmat() is successful, the actual shared memory segment
3562   // will be deleted when it's detached by shmdt() or when the process
3563   // terminates. If shmat() is not successful this will remove the shared
3564   // segment immediately.
3565   shmctl(shmid, IPC_RMID, NULL);
3566 
3567   if ((intptr_t)addr == -1) {
3568      if (warn_on_failure) {
3569        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3570        warning(msg);
3571      }
3572      return NULL;
3573   }
3574 
3575   return addr;
3576 }
3577 
3578 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3579   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3580 
3581   bool warn_on_failure = UseLargePages &&
3582       (!FLAG_IS_DEFAULT(UseLargePages) ||
3583        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3584        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3585 
3586   if (warn_on_failure) {
3587     char msg[128];
3588     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3589         PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3590     warning(msg);
3591   }
3592 }
3593 
3594 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3595   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3596   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3597   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3598 
3599   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3600   char* addr = (char*)::mmap(req_addr, bytes, prot,
3601                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3602                              -1, 0);
3603 
3604   if (addr == MAP_FAILED) {
3605     warn_on_large_pages_failure(req_addr, bytes, errno);
3606     return NULL;
3607   }
3608 
3609   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3610 
3611   return addr;
3612 }
3613 
3614 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3615   size_t large_page_size = os::large_page_size();
3616 
3617   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3618 
3619   // Allocate small pages.
3620 
3621   char* start;
3622   if (req_addr != NULL) {
3623     assert(is_ptr_aligned(req_addr, alignment), "Must be");
3624     assert(is_size_aligned(bytes, alignment), "Must be");
3625     start = os::reserve_memory(bytes, req_addr);
3626     assert(start == NULL || start == req_addr, "Must be");
3627   } else {
3628     start = os::reserve_memory_aligned(bytes, alignment);
3629   }
3630 
3631   if (start == NULL) {
3632     return NULL;
3633   }
3634 
3635   assert(is_ptr_aligned(start, alignment), "Must be");
3636 
3637   // os::reserve_memory_special will record this memory area.
3638   // Need to release it here to prevent overlapping reservations.
3639   MemTracker::record_virtual_memory_release((address)start, bytes);
3640 
3641   char* end = start + bytes;
3642 
3643   // Find the regions of the allocated chunk that can be promoted to large pages.
3644   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3645   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3646 
3647   size_t lp_bytes = lp_end - lp_start;
3648 
3649   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3650 
3651   if (lp_bytes == 0) {
3652     // The mapped region doesn't even span the start and the end of a large page.
3653     // Fall back to allocate a non-special area.
3654     ::munmap(start, end - start);
3655     return NULL;
3656   }
3657 
3658   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3659 
3660 
3661   void* result;
3662 
3663   if (start != lp_start) {
3664     result = ::mmap(start, lp_start - start, prot,
3665                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3666                     -1, 0);
3667     if (result == MAP_FAILED) {
3668       ::munmap(lp_start, end - lp_start);
3669       return NULL;
3670     }
3671   }
3672 
3673   result = ::mmap(lp_start, lp_bytes, prot,
3674                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3675                   -1, 0);
3676   if (result == MAP_FAILED) {
3677     warn_on_large_pages_failure(req_addr, bytes, errno);
3678     // If the mmap above fails, the large pages region will be unmapped and we
3679     // have regions before and after with small pages. Release these regions.
3680     //
3681     // |  mapped  |  unmapped  |  mapped  |
3682     // ^          ^            ^          ^
3683     // start      lp_start     lp_end     end
3684     //
3685     ::munmap(start, lp_start - start);
3686     ::munmap(lp_end, end - lp_end);
3687     return NULL;
3688   }
3689 
3690   if (lp_end != end) {
3691       result = ::mmap(lp_end, end - lp_end, prot,
3692                       MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3693                       -1, 0);
3694     if (result == MAP_FAILED) {
3695       ::munmap(start, lp_end - start);
3696       return NULL;
3697     }
3698   }
3699 
3700   return start;
3701 }
3702 
3703 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3704   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3705   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3706   assert(is_power_of_2(alignment), "Must be");
3707   assert(is_power_of_2(os::large_page_size()), "Must be");
3708   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3709 
3710   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3711     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3712   } else {
3713     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3714   }
3715 }
3716 
3717 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3718   assert(UseLargePages, "only for large pages");
3719 
3720   char* addr;
3721   if (UseSHM) {
3722     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3723   } else {
3724     assert(UseHugeTLBFS, "must be");
3725     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3726   }
3727 
3728   if (addr != NULL) {
3729     if (UseNUMAInterleaving) {
3730       numa_make_global(addr, bytes);
3731     }
3732 
3733     // The memory is committed
3734     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3735   }
3736 
3737   return addr;
3738 }
3739 
3740 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3741   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3742   return shmdt(base) == 0;
3743 }
3744 
3745 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3746   return pd_release_memory(base, bytes);
3747 }
3748 
3749 bool os::release_memory_special(char* base, size_t bytes) {
3750   assert(UseLargePages, "only for large pages");
3751 
3752   MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3753 
3754   bool res;
3755   if (UseSHM) {
3756     res = os::Linux::release_memory_special_shm(base, bytes);
3757   } else {
3758     assert(UseHugeTLBFS, "must be");
3759     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3760   }
3761 
3762   if (res) {
3763     tkr.record((address)base, bytes);
3764   } else {
3765     tkr.discard();
3766   }
3767 
3768   return res;
3769 }
3770 
3771 
3772 size_t os::large_page_size() {
3773   return _large_page_size;
3774 }
3775 
3776 // With SysV SHM the entire memory region must be allocated as shared
3777 // memory.
3778 // HugeTLBFS allows application to commit large page memory on demand.
3779 // However, when committing memory with HugeTLBFS fails, the region
3780 // that was supposed to be committed will lose the old reservation
3781 // and allow other threads to steal that memory region. Because of this
3782 // behavior we can't commit HugeTLBFS memory.
3783 bool os::can_commit_large_page_memory() {
3784   return UseTransparentHugePages;
3785 }
3786 
3787 bool os::can_execute_large_page_memory() {
3788   return UseTransparentHugePages || UseHugeTLBFS;
3789 }
3790 
3791 // Reserve memory at an arbitrary address, only if that area is
3792 // available (and not reserved for something else).
3793 
3794 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3795   const int max_tries = 10;
3796   char* base[max_tries];
3797   size_t size[max_tries];
3798   const size_t gap = 0x000000;
3799 
3800   // Assert only that the size is a multiple of the page size, since
3801   // that's all that mmap requires, and since that's all we really know
3802   // about at this low abstraction level.  If we need higher alignment,
3803   // we can either pass an alignment to this method or verify alignment
3804   // in one of the methods further up the call chain.  See bug 5044738.
3805   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3806 
3807   // Repeatedly allocate blocks until the block is allocated at the
3808   // right spot. Give up after max_tries. Note that reserve_memory() will
3809   // automatically update _highest_vm_reserved_address if the call is
3810   // successful. The variable tracks the highest memory address every reserved
3811   // by JVM. It is used to detect heap-stack collision if running with
3812   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3813   // space than needed, it could confuse the collision detecting code. To
3814   // solve the problem, save current _highest_vm_reserved_address and
3815   // calculate the correct value before return.
3816   address old_highest = _highest_vm_reserved_address;
3817 
3818   // Linux mmap allows caller to pass an address as hint; give it a try first,
3819   // if kernel honors the hint then we can return immediately.
3820   char * addr = anon_mmap(requested_addr, bytes, false);
3821   if (addr == requested_addr) {
3822      return requested_addr;
3823   }
3824 
3825   if (addr != NULL) {
3826      // mmap() is successful but it fails to reserve at the requested address
3827      anon_munmap(addr, bytes);
3828   }
3829 
3830   int i;
3831   for (i = 0; i < max_tries; ++i) {
3832     base[i] = reserve_memory(bytes);
3833 
3834     if (base[i] != NULL) {
3835       // Is this the block we wanted?
3836       if (base[i] == requested_addr) {
3837         size[i] = bytes;
3838         break;
3839       }
3840 
3841       // Does this overlap the block we wanted? Give back the overlapped
3842       // parts and try again.
3843 
3844       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3845       if (top_overlap >= 0 && top_overlap < bytes) {
3846         unmap_memory(base[i], top_overlap);
3847         base[i] += top_overlap;
3848         size[i] = bytes - top_overlap;
3849       } else {
3850         size_t bottom_overlap = base[i] + bytes - requested_addr;
3851         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3852           unmap_memory(requested_addr, bottom_overlap);
3853           size[i] = bytes - bottom_overlap;
3854         } else {
3855           size[i] = bytes;
3856         }
3857       }
3858     }
3859   }
3860 
3861   // Give back the unused reserved pieces.
3862 
3863   for (int j = 0; j < i; ++j) {
3864     if (base[j] != NULL) {
3865       unmap_memory(base[j], size[j]);
3866     }
3867   }
3868 
3869   if (i < max_tries) {
3870     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3871     return requested_addr;
3872   } else {
3873     _highest_vm_reserved_address = old_highest;
3874     return NULL;
3875   }
3876 }
3877 
3878 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3879   return ::read(fd, buf, nBytes);
3880 }
3881 
3882 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3883 // Solaris uses poll(), linux uses park().
3884 // Poll() is likely a better choice, assuming that Thread.interrupt()
3885 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3886 // SIGSEGV, see 4355769.
3887 
3888 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3889   assert(thread == Thread::current(),  "thread consistency check");
3890 
3891   ParkEvent * const slp = thread->_SleepEvent ;
3892   slp->reset() ;
3893   OrderAccess::fence() ;
3894 
3895   if (interruptible) {
3896     jlong prevtime = javaTimeNanos();
3897 
3898     for (;;) {
3899       if (os::is_interrupted(thread, true)) {
3900         return OS_INTRPT;
3901       }
3902 
3903       jlong newtime = javaTimeNanos();
3904 
3905       if (newtime - prevtime < 0) {
3906         // time moving backwards, should only happen if no monotonic clock
3907         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3908         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3909       } else {
3910         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3911       }
3912 
3913       if(millis <= 0) {
3914         return OS_OK;
3915       }
3916 
3917       prevtime = newtime;
3918 
3919       {
3920         assert(thread->is_Java_thread(), "sanity check");
3921         JavaThread *jt = (JavaThread *) thread;
3922         ThreadBlockInVM tbivm(jt);
3923         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3924 
3925         jt->set_suspend_equivalent();
3926         // cleared by handle_special_suspend_equivalent_condition() or
3927         // java_suspend_self() via check_and_wait_while_suspended()
3928 
3929         slp->park(millis);
3930 
3931         // were we externally suspended while we were waiting?
3932         jt->check_and_wait_while_suspended();
3933       }
3934     }
3935   } else {
3936     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3937     jlong prevtime = javaTimeNanos();
3938 
3939     for (;;) {
3940       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3941       // the 1st iteration ...
3942       jlong newtime = javaTimeNanos();
3943 
3944       if (newtime - prevtime < 0) {
3945         // time moving backwards, should only happen if no monotonic clock
3946         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3947         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3948       } else {
3949         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3950       }
3951 
3952       if(millis <= 0) break ;
3953 
3954       prevtime = newtime;
3955       slp->park(millis);
3956     }
3957     return OS_OK ;
3958   }
3959 }
3960 
3961 //
3962 // Short sleep, direct OS call.
3963 //
3964 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3965 // sched_yield(2) will actually give up the CPU:
3966 //
3967 //   * Alone on this pariticular CPU, keeps running.
3968 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3969 //     (pre 2.6.39).
3970 //
3971 // So calling this with 0 is an alternative.
3972 //
3973 void os::naked_short_sleep(jlong ms) {
3974   struct timespec req;
3975 
3976   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3977   req.tv_sec = 0;
3978   if (ms > 0) {
3979     req.tv_nsec = (ms % 1000) * 1000000;
3980   }
3981   else {
3982     req.tv_nsec = 1;
3983   }
3984 
3985   nanosleep(&req, NULL);
3986 
3987   return;
3988 }
3989 
3990 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3991 void os::infinite_sleep() {
3992   while (true) {    // sleep forever ...
3993     ::sleep(100);   // ... 100 seconds at a time
3994   }
3995 }
3996 
3997 // Used to convert frequent JVM_Yield() to nops
3998 bool os::dont_yield() {
3999   return DontYieldALot;
4000 }
4001 
4002 void os::yield() {
4003   sched_yield();
4004 }
4005 
4006 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
4007 
4008 void os::yield_all(int attempts) {
4009   // Yields to all threads, including threads with lower priorities
4010   // Threads on Linux are all with same priority. The Solaris style
4011   // os::yield_all() with nanosleep(1ms) is not necessary.
4012   sched_yield();
4013 }
4014 
4015 // Called from the tight loops to possibly influence time-sharing heuristics
4016 void os::loop_breaker(int attempts) {
4017   os::yield_all(attempts);
4018 }
4019 
4020 ////////////////////////////////////////////////////////////////////////////////
4021 // thread priority support
4022 
4023 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4024 // only supports dynamic priority, static priority must be zero. For real-time
4025 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4026 // However, for large multi-threaded applications, SCHED_RR is not only slower
4027 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4028 // of 5 runs - Sep 2005).
4029 //
4030 // The following code actually changes the niceness of kernel-thread/LWP. It
4031 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4032 // not the entire user process, and user level threads are 1:1 mapped to kernel
4033 // threads. It has always been the case, but could change in the future. For
4034 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4035 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4036 
4037 int os::java_to_os_priority[CriticalPriority + 1] = {
4038   19,              // 0 Entry should never be used
4039 
4040    4,              // 1 MinPriority
4041    3,              // 2
4042    2,              // 3
4043 
4044    1,              // 4
4045    0,              // 5 NormPriority
4046   -1,              // 6
4047 
4048   -2,              // 7
4049   -3,              // 8
4050   -4,              // 9 NearMaxPriority
4051 
4052   -5,              // 10 MaxPriority
4053 
4054   -5               // 11 CriticalPriority
4055 };
4056 
4057 static int prio_init() {
4058   if (ThreadPriorityPolicy == 1) {
4059     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4060     // if effective uid is not root. Perhaps, a more elegant way of doing
4061     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4062     if (geteuid() != 0) {
4063       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4064         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4065       }
4066       ThreadPriorityPolicy = 0;
4067     }
4068   }
4069   if (UseCriticalJavaThreadPriority) {
4070     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4071   }
4072   return 0;
4073 }
4074 
4075 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4076   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4077 
4078   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4079   return (ret == 0) ? OS_OK : OS_ERR;
4080 }
4081 
4082 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4083   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4084     *priority_ptr = java_to_os_priority[NormPriority];
4085     return OS_OK;
4086   }
4087 
4088   errno = 0;
4089   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4090   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4091 }
4092 
4093 // Hint to the underlying OS that a task switch would not be good.
4094 // Void return because it's a hint and can fail.
4095 void os::hint_no_preempt() {}
4096 
4097 ////////////////////////////////////////////////////////////////////////////////
4098 // suspend/resume support
4099 
4100 //  the low-level signal-based suspend/resume support is a remnant from the
4101 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4102 //  within hotspot. Now there is a single use-case for this:
4103 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
4104 //      that runs in the watcher thread.
4105 //  The remaining code is greatly simplified from the more general suspension
4106 //  code that used to be used.
4107 //
4108 //  The protocol is quite simple:
4109 //  - suspend:
4110 //      - sends a signal to the target thread
4111 //      - polls the suspend state of the osthread using a yield loop
4112 //      - target thread signal handler (SR_handler) sets suspend state
4113 //        and blocks in sigsuspend until continued
4114 //  - resume:
4115 //      - sets target osthread state to continue
4116 //      - sends signal to end the sigsuspend loop in the SR_handler
4117 //
4118 //  Note that the SR_lock plays no role in this suspend/resume protocol.
4119 //
4120 
4121 static void resume_clear_context(OSThread *osthread) {
4122   osthread->set_ucontext(NULL);
4123   osthread->set_siginfo(NULL);
4124 }
4125 
4126 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4127   osthread->set_ucontext(context);
4128   osthread->set_siginfo(siginfo);
4129 }
4130 
4131 //
4132 // Handler function invoked when a thread's execution is suspended or
4133 // resumed. We have to be careful that only async-safe functions are
4134 // called here (Note: most pthread functions are not async safe and
4135 // should be avoided.)
4136 //
4137 // Note: sigwait() is a more natural fit than sigsuspend() from an
4138 // interface point of view, but sigwait() prevents the signal hander
4139 // from being run. libpthread would get very confused by not having
4140 // its signal handlers run and prevents sigwait()'s use with the
4141 // mutex granting granting signal.
4142 //
4143 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4144 //
4145 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4146   // Save and restore errno to avoid confusing native code with EINTR
4147   // after sigsuspend.
4148   int old_errno = errno;
4149 
4150   Thread* thread = Thread::current();
4151   OSThread* osthread = thread->osthread();
4152   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4153 
4154   os::SuspendResume::State current = osthread->sr.state();
4155   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4156     suspend_save_context(osthread, siginfo, context);
4157 
4158     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4159     os::SuspendResume::State state = osthread->sr.suspended();
4160     if (state == os::SuspendResume::SR_SUSPENDED) {
4161       sigset_t suspend_set;  // signals for sigsuspend()
4162 
4163       // get current set of blocked signals and unblock resume signal
4164       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4165       sigdelset(&suspend_set, SR_signum);
4166 
4167       sr_semaphore.signal();
4168       // wait here until we are resumed
4169       while (1) {
4170         sigsuspend(&suspend_set);
4171 
4172         os::SuspendResume::State result = osthread->sr.running();
4173         if (result == os::SuspendResume::SR_RUNNING) {
4174           sr_semaphore.signal();
4175           break;
4176         }
4177       }
4178 
4179     } else if (state == os::SuspendResume::SR_RUNNING) {
4180       // request was cancelled, continue
4181     } else {
4182       ShouldNotReachHere();
4183     }
4184 
4185     resume_clear_context(osthread);
4186   } else if (current == os::SuspendResume::SR_RUNNING) {
4187     // request was cancelled, continue
4188   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4189     // ignore
4190   } else {
4191     // ignore
4192   }
4193 
4194   errno = old_errno;
4195 }
4196 
4197 
4198 static int SR_initialize() {
4199   struct sigaction act;
4200   char *s;
4201   /* Get signal number to use for suspend/resume */
4202   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4203     int sig = ::strtol(s, 0, 10);
4204     if (sig > 0 || sig < _NSIG) {
4205         SR_signum = sig;
4206     }
4207   }
4208 
4209   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4210         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4211 
4212   sigemptyset(&SR_sigset);
4213   sigaddset(&SR_sigset, SR_signum);
4214 
4215   /* Set up signal handler for suspend/resume */
4216   act.sa_flags = SA_RESTART|SA_SIGINFO;
4217   act.sa_handler = (void (*)(int)) SR_handler;
4218 
4219   // SR_signum is blocked by default.
4220   // 4528190 - We also need to block pthread restart signal (32 on all
4221   // supported Linux platforms). Note that LinuxThreads need to block
4222   // this signal for all threads to work properly. So we don't have
4223   // to use hard-coded signal number when setting up the mask.
4224   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4225 
4226   if (sigaction(SR_signum, &act, 0) == -1) {
4227     return -1;
4228   }
4229 
4230   // Save signal flag
4231   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4232   return 0;
4233 }
4234 
4235 static int SR_finalize() {
4236   return 0;
4237 }
4238 
4239 static int sr_notify(OSThread* osthread) {
4240   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4241   assert_status(status == 0, status, "pthread_kill");
4242   return status;
4243 }
4244 
4245 // "Randomly" selected value for how long we want to spin
4246 // before bailing out on suspending a thread, also how often
4247 // we send a signal to a thread we want to resume
4248 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4249 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4250 
4251 // returns true on success and false on error - really an error is fatal
4252 // but this seems the normal response to library errors
4253 static bool do_suspend(OSThread* osthread) {
4254   assert(osthread->sr.is_running(), "thread should be running");
4255   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4256 
4257   // mark as suspended and send signal
4258   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4259     // failed to switch, state wasn't running?
4260     ShouldNotReachHere();
4261     return false;
4262   }
4263 
4264   if (sr_notify(osthread) != 0) {
4265     ShouldNotReachHere();
4266   }
4267 
4268   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4269   while (true) {
4270     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4271       break;
4272     } else {
4273       // timeout
4274       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4275       if (cancelled == os::SuspendResume::SR_RUNNING) {
4276         return false;
4277       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4278         // make sure that we consume the signal on the semaphore as well
4279         sr_semaphore.wait();
4280         break;
4281       } else {
4282         ShouldNotReachHere();
4283         return false;
4284       }
4285     }
4286   }
4287 
4288   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4289   return true;
4290 }
4291 
4292 static void do_resume(OSThread* osthread) {
4293   assert(osthread->sr.is_suspended(), "thread should be suspended");
4294   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4295 
4296   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4297     // failed to switch to WAKEUP_REQUEST
4298     ShouldNotReachHere();
4299     return;
4300   }
4301 
4302   while (true) {
4303     if (sr_notify(osthread) == 0) {
4304       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4305         if (osthread->sr.is_running()) {
4306           return;
4307         }
4308       }
4309     } else {
4310       ShouldNotReachHere();
4311     }
4312   }
4313 
4314   guarantee(osthread->sr.is_running(), "Must be running!");
4315 }
4316 
4317 ////////////////////////////////////////////////////////////////////////////////
4318 // interrupt support
4319 
4320 void os::interrupt(Thread* thread) {
4321   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4322     "possibility of dangling Thread pointer");
4323 
4324   OSThread* osthread = thread->osthread();
4325 
4326   if (!osthread->interrupted()) {
4327     osthread->set_interrupted(true);
4328     // More than one thread can get here with the same value of osthread,
4329     // resulting in multiple notifications.  We do, however, want the store
4330     // to interrupted() to be visible to other threads before we execute unpark().
4331     OrderAccess::fence();
4332     ParkEvent * const slp = thread->_SleepEvent ;
4333     if (slp != NULL) slp->unpark() ;
4334   }
4335 
4336   // For JSR166. Unpark even if interrupt status already was set
4337   if (thread->is_Java_thread())
4338     ((JavaThread*)thread)->parker()->unpark();
4339 
4340   ParkEvent * ev = thread->_ParkEvent ;
4341   if (ev != NULL) ev->unpark() ;
4342 
4343 }
4344 
4345 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4346   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4347     "possibility of dangling Thread pointer");
4348 
4349   OSThread* osthread = thread->osthread();
4350 
4351   bool interrupted = osthread->interrupted();
4352 
4353   if (interrupted && clear_interrupted) {
4354     osthread->set_interrupted(false);
4355     // consider thread->_SleepEvent->reset() ... optional optimization
4356   }
4357 
4358   return interrupted;
4359 }
4360 
4361 ///////////////////////////////////////////////////////////////////////////////////
4362 // signal handling (except suspend/resume)
4363 
4364 // This routine may be used by user applications as a "hook" to catch signals.
4365 // The user-defined signal handler must pass unrecognized signals to this
4366 // routine, and if it returns true (non-zero), then the signal handler must
4367 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4368 // routine will never retun false (zero), but instead will execute a VM panic
4369 // routine kill the process.
4370 //
4371 // If this routine returns false, it is OK to call it again.  This allows
4372 // the user-defined signal handler to perform checks either before or after
4373 // the VM performs its own checks.  Naturally, the user code would be making
4374 // a serious error if it tried to handle an exception (such as a null check
4375 // or breakpoint) that the VM was generating for its own correct operation.
4376 //
4377 // This routine may recognize any of the following kinds of signals:
4378 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4379 // It should be consulted by handlers for any of those signals.
4380 //
4381 // The caller of this routine must pass in the three arguments supplied
4382 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4383 // field of the structure passed to sigaction().  This routine assumes that
4384 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4385 //
4386 // Note that the VM will print warnings if it detects conflicting signal
4387 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4388 //
4389 extern "C" JNIEXPORT int
4390 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4391                         void* ucontext, int abort_if_unrecognized);
4392 
4393 void signalHandler(int sig, siginfo_t* info, void* uc) {
4394   assert(info != NULL && uc != NULL, "it must be old kernel");
4395   JVM_handle_linux_signal(sig, info, uc, true);
4396 }
4397 
4398 
4399 // This boolean allows users to forward their own non-matching signals
4400 // to JVM_handle_linux_signal, harmlessly.
4401 bool os::Linux::signal_handlers_are_installed = false;
4402 
4403 // For signal-chaining
4404 struct sigaction os::Linux::sigact[MAXSIGNUM];
4405 unsigned int os::Linux::sigs = 0;
4406 bool os::Linux::libjsig_is_loaded = false;
4407 typedef struct sigaction *(*get_signal_t)(int);
4408 get_signal_t os::Linux::get_signal_action = NULL;
4409 
4410 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4411   struct sigaction *actp = NULL;
4412 
4413   if (libjsig_is_loaded) {
4414     // Retrieve the old signal handler from libjsig
4415     actp = (*get_signal_action)(sig);
4416   }
4417   if (actp == NULL) {
4418     // Retrieve the preinstalled signal handler from jvm
4419     actp = get_preinstalled_handler(sig);
4420   }
4421 
4422   return actp;
4423 }
4424 
4425 static bool call_chained_handler(struct sigaction *actp, int sig,
4426                                  siginfo_t *siginfo, void *context) {
4427   // Call the old signal handler
4428   if (actp->sa_handler == SIG_DFL) {
4429     // It's more reasonable to let jvm treat it as an unexpected exception
4430     // instead of taking the default action.
4431     return false;
4432   } else if (actp->sa_handler != SIG_IGN) {
4433     if ((actp->sa_flags & SA_NODEFER) == 0) {
4434       // automaticlly block the signal
4435       sigaddset(&(actp->sa_mask), sig);
4436     }
4437 
4438     sa_handler_t hand;
4439     sa_sigaction_t sa;
4440     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4441     // retrieve the chained handler
4442     if (siginfo_flag_set) {
4443       sa = actp->sa_sigaction;
4444     } else {
4445       hand = actp->sa_handler;
4446     }
4447 
4448     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4449       actp->sa_handler = SIG_DFL;
4450     }
4451 
4452     // try to honor the signal mask
4453     sigset_t oset;
4454     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4455 
4456     // call into the chained handler
4457     if (siginfo_flag_set) {
4458       (*sa)(sig, siginfo, context);
4459     } else {
4460       (*hand)(sig);
4461     }
4462 
4463     // restore the signal mask
4464     pthread_sigmask(SIG_SETMASK, &oset, 0);
4465   }
4466   // Tell jvm's signal handler the signal is taken care of.
4467   return true;
4468 }
4469 
4470 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4471   bool chained = false;
4472   // signal-chaining
4473   if (UseSignalChaining) {
4474     struct sigaction *actp = get_chained_signal_action(sig);
4475     if (actp != NULL) {
4476       chained = call_chained_handler(actp, sig, siginfo, context);
4477     }
4478   }
4479   return chained;
4480 }
4481 
4482 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4483   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4484     return &sigact[sig];
4485   }
4486   return NULL;
4487 }
4488 
4489 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4490   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4491   sigact[sig] = oldAct;
4492   sigs |= (unsigned int)1 << sig;
4493 }
4494 
4495 // for diagnostic
4496 int os::Linux::sigflags[MAXSIGNUM];
4497 
4498 int os::Linux::get_our_sigflags(int sig) {
4499   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4500   return sigflags[sig];
4501 }
4502 
4503 void os::Linux::set_our_sigflags(int sig, int flags) {
4504   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4505   sigflags[sig] = flags;
4506 }
4507 
4508 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4509   // Check for overwrite.
4510   struct sigaction oldAct;
4511   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4512 
4513   void* oldhand = oldAct.sa_sigaction
4514                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4515                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4516   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4517       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4518       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4519     if (AllowUserSignalHandlers || !set_installed) {
4520       // Do not overwrite; user takes responsibility to forward to us.
4521       return;
4522     } else if (UseSignalChaining) {
4523       // save the old handler in jvm
4524       save_preinstalled_handler(sig, oldAct);
4525       // libjsig also interposes the sigaction() call below and saves the
4526       // old sigaction on it own.
4527     } else {
4528       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4529                     "%#lx for signal %d.", (long)oldhand, sig));
4530     }
4531   }
4532 
4533   struct sigaction sigAct;
4534   sigfillset(&(sigAct.sa_mask));
4535   sigAct.sa_handler = SIG_DFL;
4536   if (!set_installed) {
4537     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4538   } else {
4539     sigAct.sa_sigaction = signalHandler;
4540     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4541   }
4542   // Save flags, which are set by ours
4543   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4544   sigflags[sig] = sigAct.sa_flags;
4545 
4546   int ret = sigaction(sig, &sigAct, &oldAct);
4547   assert(ret == 0, "check");
4548 
4549   void* oldhand2  = oldAct.sa_sigaction
4550                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4551                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4552   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4553 }
4554 
4555 // install signal handlers for signals that HotSpot needs to
4556 // handle in order to support Java-level exception handling.
4557 
4558 void os::Linux::install_signal_handlers() {
4559   if (!signal_handlers_are_installed) {
4560     signal_handlers_are_installed = true;
4561 
4562     // signal-chaining
4563     typedef void (*signal_setting_t)();
4564     signal_setting_t begin_signal_setting = NULL;
4565     signal_setting_t end_signal_setting = NULL;
4566     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4567                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4568     if (begin_signal_setting != NULL) {
4569       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4570                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4571       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4572                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4573       libjsig_is_loaded = true;
4574       assert(UseSignalChaining, "should enable signal-chaining");
4575     }
4576     if (libjsig_is_loaded) {
4577       // Tell libjsig jvm is setting signal handlers
4578       (*begin_signal_setting)();
4579     }
4580 
4581     set_signal_handler(SIGSEGV, true);
4582     set_signal_handler(SIGPIPE, true);
4583     set_signal_handler(SIGBUS, true);
4584     set_signal_handler(SIGILL, true);
4585     set_signal_handler(SIGFPE, true);
4586 #if defined(PPC64)
4587     set_signal_handler(SIGTRAP, true);
4588 #endif
4589     set_signal_handler(SIGXFSZ, true);
4590 
4591     if (libjsig_is_loaded) {
4592       // Tell libjsig jvm finishes setting signal handlers
4593       (*end_signal_setting)();
4594     }
4595 
4596     // We don't activate signal checker if libjsig is in place, we trust ourselves
4597     // and if UserSignalHandler is installed all bets are off.
4598     // Log that signal checking is off only if -verbose:jni is specified.
4599     if (CheckJNICalls) {
4600       if (libjsig_is_loaded) {
4601         if (PrintJNIResolving) {
4602           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4603         }
4604         check_signals = false;
4605       }
4606       if (AllowUserSignalHandlers) {
4607         if (PrintJNIResolving) {
4608           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4609         }
4610         check_signals = false;
4611       }
4612     }
4613   }
4614 }
4615 
4616 // This is the fastest way to get thread cpu time on Linux.
4617 // Returns cpu time (user+sys) for any thread, not only for current.
4618 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4619 // It might work on 2.6.10+ with a special kernel/glibc patch.
4620 // For reference, please, see IEEE Std 1003.1-2004:
4621 //   http://www.unix.org/single_unix_specification
4622 
4623 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4624   struct timespec tp;
4625   int rc = os::Linux::clock_gettime(clockid, &tp);
4626   assert(rc == 0, "clock_gettime is expected to return 0 code");
4627 
4628   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4629 }
4630 
4631 /////
4632 // glibc on Linux platform uses non-documented flag
4633 // to indicate, that some special sort of signal
4634 // trampoline is used.
4635 // We will never set this flag, and we should
4636 // ignore this flag in our diagnostic
4637 #ifdef SIGNIFICANT_SIGNAL_MASK
4638 #undef SIGNIFICANT_SIGNAL_MASK
4639 #endif
4640 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4641 
4642 static const char* get_signal_handler_name(address handler,
4643                                            char* buf, int buflen) {
4644   int offset;
4645   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4646   if (found) {
4647     // skip directory names
4648     const char *p1, *p2;
4649     p1 = buf;
4650     size_t len = strlen(os::file_separator());
4651     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4652     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4653   } else {
4654     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4655   }
4656   return buf;
4657 }
4658 
4659 static void print_signal_handler(outputStream* st, int sig,
4660                                  char* buf, size_t buflen) {
4661   struct sigaction sa;
4662 
4663   sigaction(sig, NULL, &sa);
4664 
4665   // See comment for SIGNIFICANT_SIGNAL_MASK define
4666   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4667 
4668   st->print("%s: ", os::exception_name(sig, buf, buflen));
4669 
4670   address handler = (sa.sa_flags & SA_SIGINFO)
4671     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4672     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4673 
4674   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4675     st->print("SIG_DFL");
4676   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4677     st->print("SIG_IGN");
4678   } else {
4679     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4680   }
4681 
4682   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4683 
4684   address rh = VMError::get_resetted_sighandler(sig);
4685   // May be, handler was resetted by VMError?
4686   if(rh != NULL) {
4687     handler = rh;
4688     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4689   }
4690 
4691   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
4692 
4693   // Check: is it our handler?
4694   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4695      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4696     // It is our signal handler
4697     // check for flags, reset system-used one!
4698     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4699       st->print(
4700                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4701                 os::Linux::get_our_sigflags(sig));
4702     }
4703   }
4704   st->cr();
4705 }
4706 
4707 
4708 #define DO_SIGNAL_CHECK(sig) \
4709   if (!sigismember(&check_signal_done, sig)) \
4710     os::Linux::check_signal_handler(sig)
4711 
4712 // This method is a periodic task to check for misbehaving JNI applications
4713 // under CheckJNI, we can add any periodic checks here
4714 
4715 void os::run_periodic_checks() {
4716 
4717   if (check_signals == false) return;
4718 
4719   // SEGV and BUS if overridden could potentially prevent
4720   // generation of hs*.log in the event of a crash, debugging
4721   // such a case can be very challenging, so we absolutely
4722   // check the following for a good measure:
4723   DO_SIGNAL_CHECK(SIGSEGV);
4724   DO_SIGNAL_CHECK(SIGILL);
4725   DO_SIGNAL_CHECK(SIGFPE);
4726   DO_SIGNAL_CHECK(SIGBUS);
4727   DO_SIGNAL_CHECK(SIGPIPE);
4728   DO_SIGNAL_CHECK(SIGXFSZ);
4729 #if defined(PPC64)
4730   DO_SIGNAL_CHECK(SIGTRAP);
4731 #endif
4732 
4733   // ReduceSignalUsage allows the user to override these handlers
4734   // see comments at the very top and jvm_solaris.h
4735   if (!ReduceSignalUsage) {
4736     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4737     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4738     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4739     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4740   }
4741 
4742   DO_SIGNAL_CHECK(SR_signum);
4743   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4744 }
4745 
4746 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4747 
4748 static os_sigaction_t os_sigaction = NULL;
4749 
4750 void os::Linux::check_signal_handler(int sig) {
4751   char buf[O_BUFLEN];
4752   address jvmHandler = NULL;
4753 
4754 
4755   struct sigaction act;
4756   if (os_sigaction == NULL) {
4757     // only trust the default sigaction, in case it has been interposed
4758     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4759     if (os_sigaction == NULL) return;
4760   }
4761 
4762   os_sigaction(sig, (struct sigaction*)NULL, &act);
4763 
4764 
4765   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4766 
4767   address thisHandler = (act.sa_flags & SA_SIGINFO)
4768     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4769     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4770 
4771 
4772   switch(sig) {
4773   case SIGSEGV:
4774   case SIGBUS:
4775   case SIGFPE:
4776   case SIGPIPE:
4777   case SIGILL:
4778   case SIGXFSZ:
4779     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4780     break;
4781 
4782   case SHUTDOWN1_SIGNAL:
4783   case SHUTDOWN2_SIGNAL:
4784   case SHUTDOWN3_SIGNAL:
4785   case BREAK_SIGNAL:
4786     jvmHandler = (address)user_handler();
4787     break;
4788 
4789   case INTERRUPT_SIGNAL:
4790     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4791     break;
4792 
4793   default:
4794     if (sig == SR_signum) {
4795       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4796     } else {
4797       return;
4798     }
4799     break;
4800   }
4801 
4802   if (thisHandler != jvmHandler) {
4803     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4804     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4805     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4806     // No need to check this sig any longer
4807     sigaddset(&check_signal_done, sig);
4808   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4809     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4810     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4811     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4812     // No need to check this sig any longer
4813     sigaddset(&check_signal_done, sig);
4814   }
4815 
4816   // Dump all the signal
4817   if (sigismember(&check_signal_done, sig)) {
4818     print_signal_handlers(tty, buf, O_BUFLEN);
4819   }
4820 }
4821 
4822 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4823 
4824 extern bool signal_name(int signo, char* buf, size_t len);
4825 
4826 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4827   if (0 < exception_code && exception_code <= SIGRTMAX) {
4828     // signal
4829     if (!signal_name(exception_code, buf, size)) {
4830       jio_snprintf(buf, size, "SIG%d", exception_code);
4831     }
4832     return buf;
4833   } else {
4834     return NULL;
4835   }
4836 }
4837 
4838 // this is called _before_ the most of global arguments have been parsed
4839 void os::init(void) {
4840   char dummy;   /* used to get a guess on initial stack address */
4841 //  first_hrtime = gethrtime();
4842 
4843   // With LinuxThreads the JavaMain thread pid (primordial thread)
4844   // is different than the pid of the java launcher thread.
4845   // So, on Linux, the launcher thread pid is passed to the VM
4846   // via the sun.java.launcher.pid property.
4847   // Use this property instead of getpid() if it was correctly passed.
4848   // See bug 6351349.
4849   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4850 
4851   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4852 
4853   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4854 
4855   init_random(1234567);
4856 
4857   ThreadCritical::initialize();
4858 
4859   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4860   if (Linux::page_size() == -1) {
4861     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4862                   strerror(errno)));
4863   }
4864   init_page_sizes((size_t) Linux::page_size());
4865 
4866   Linux::initialize_system_info();
4867 
4868   // main_thread points to the aboriginal thread
4869   Linux::_main_thread = pthread_self();
4870 
4871   Linux::clock_init();
4872   initial_time_count = os::elapsed_counter();
4873 
4874   // pthread_condattr initialization for monotonic clock
4875   int status;
4876   pthread_condattr_t* _condattr = os::Linux::condAttr();
4877   if ((status = pthread_condattr_init(_condattr)) != 0) {
4878     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4879   }
4880   // Only set the clock if CLOCK_MONOTONIC is available
4881   if (Linux::supports_monotonic_clock()) {
4882     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4883       if (status == EINVAL) {
4884         warning("Unable to use monotonic clock with relative timed-waits" \
4885                 " - changes to the time-of-day clock may have adverse affects");
4886       } else {
4887         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4888       }
4889     }
4890   }
4891   // else it defaults to CLOCK_REALTIME
4892 
4893   pthread_mutex_init(&dl_mutex, NULL);
4894 
4895 NOT_ZERO (
4896   // If the pagesize of the VM is greater than 8K determine the appropriate
4897   // number of initial guard pages.  The user can change this with the
4898   // command line arguments, if needed.
4899   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4900     StackYellowPages = 1;
4901     StackRedPages = 1;
4902     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4903   }
4904  )
4905 }
4906 
4907 // To install functions for atexit system call
4908 extern "C" {
4909   static void perfMemory_exit_helper() {
4910     perfMemory_exit();
4911   }
4912 }
4913 
4914 // this is called _after_ the global arguments have been parsed
4915 jint os::init_2(void)
4916 {
4917   Linux::fast_thread_clock_init();
4918 
4919   // Allocate a single page and mark it as readable for safepoint polling
4920   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4921   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4922 
4923   os::set_polling_page( polling_page );
4924 
4925 #ifndef PRODUCT
4926   if(Verbose && PrintMiscellaneous)
4927     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4928 #endif
4929 
4930   if (!UseMembar) {
4931     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4932     guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4933     os::set_memory_serialize_page( mem_serialize_page );
4934 
4935 #ifndef PRODUCT
4936     if(Verbose && PrintMiscellaneous)
4937       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4938 #endif
4939   }
4940 
4941   os::large_page_init();
4942 
4943   // initialize suspend/resume support - must do this before signal_sets_init()
4944   if (SR_initialize() != 0) {
4945     perror("SR_initialize failed");
4946     return JNI_ERR;
4947   }
4948 
4949   Linux::signal_sets_init();
4950   Linux::install_signal_handlers();
4951 
4952   // Check minimum allowable stack size for thread creation and to initialize
4953   // the java system classes, including StackOverflowError - depends on page
4954   // size.  Add a page for compiler2 recursion in main thread.
4955   // Add in 2*BytesPerWord times page size to account for VM stack during
4956   // class initialization depending on 32 or 64 bit VM.
4957   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4958             (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4959                     (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4960 
4961 #ifdef ZERO
4962   // If this is Zero, allow at the very minimum one page each for the
4963   // Zero stack and the native stack.  This won't make any difference
4964   // for 4k pages, but is significant for large pages.
4965   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4966              (size_t)(StackYellowPages+StackRedPages+StackShadowPages+2) * Linux::page_size());
4967 #endif
4968 
4969   size_t threadStackSizeInBytes = ThreadStackSize * K;
4970   if (threadStackSizeInBytes != 0 &&
4971       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4972         tty->print_cr("\nThe stack size specified is too small, "
4973                       "Specify at least %dk",
4974                       os::Linux::min_stack_allowed/ K);
4975         return JNI_ERR;
4976   }
4977 
4978   // Make the stack size a multiple of the page size so that
4979   // the yellow/red zones can be guarded.
4980   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4981         vm_page_size()));
4982 
4983   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4984 
4985 #if defined(IA32) && !defined(ZERO)
4986   workaround_expand_exec_shield_cs_limit();
4987 #endif
4988 
4989   Linux::libpthread_init();
4990   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4991      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4992           Linux::glibc_version(), Linux::libpthread_version(),
4993           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4994   }
4995 
4996   if (UseNUMA) {
4997     if (!Linux::libnuma_init()) {
4998       UseNUMA = false;
4999     } else {
5000       if ((Linux::numa_max_node() < 1)) {
5001         // There's only one node(they start from 0), disable NUMA.
5002         UseNUMA = false;
5003       }
5004     }
5005     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5006     // we can make the adaptive lgrp chunk resizing work. If the user specified
5007     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
5008     // disable adaptive resizing.
5009     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5010       if (FLAG_IS_DEFAULT(UseNUMA)) {
5011         UseNUMA = false;
5012       } else {
5013         if (FLAG_IS_DEFAULT(UseLargePages) &&
5014             FLAG_IS_DEFAULT(UseSHM) &&
5015             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
5016           UseLargePages = false;
5017         } else {
5018           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
5019           UseAdaptiveSizePolicy = false;
5020           UseAdaptiveNUMAChunkSizing = false;
5021         }
5022       }
5023     }
5024     if (!UseNUMA && ForceNUMA) {
5025       UseNUMA = true;
5026     }
5027   }
5028 
5029   if (MaxFDLimit) {
5030     // set the number of file descriptors to max. print out error
5031     // if getrlimit/setrlimit fails but continue regardless.
5032     struct rlimit nbr_files;
5033     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5034     if (status != 0) {
5035       if (PrintMiscellaneous && (Verbose || WizardMode))
5036         perror("os::init_2 getrlimit failed");
5037     } else {
5038       nbr_files.rlim_cur = nbr_files.rlim_max;
5039       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5040       if (status != 0) {
5041         if (PrintMiscellaneous && (Verbose || WizardMode))
5042           perror("os::init_2 setrlimit failed");
5043       }
5044     }
5045   }
5046 
5047   // Initialize lock used to serialize thread creation (see os::create_thread)
5048   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5049 
5050   // at-exit methods are called in the reverse order of their registration.
5051   // atexit functions are called on return from main or as a result of a
5052   // call to exit(3C). There can be only 32 of these functions registered
5053   // and atexit() does not set errno.
5054 
5055   if (PerfAllowAtExitRegistration) {
5056     // only register atexit functions if PerfAllowAtExitRegistration is set.
5057     // atexit functions can be delayed until process exit time, which
5058     // can be problematic for embedded VM situations. Embedded VMs should
5059     // call DestroyJavaVM() to assure that VM resources are released.
5060 
5061     // note: perfMemory_exit_helper atexit function may be removed in
5062     // the future if the appropriate cleanup code can be added to the
5063     // VM_Exit VMOperation's doit method.
5064     if (atexit(perfMemory_exit_helper) != 0) {
5065       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5066     }
5067   }
5068 
5069   // initialize thread priority policy
5070   prio_init();
5071 
5072   return JNI_OK;
5073 }
5074 
5075 // this is called at the end of vm_initialization
5076 void os::init_3(void) {
5077 #ifdef JAVASE_EMBEDDED
5078   // Start the MemNotifyThread
5079   if (LowMemoryProtection) {
5080     MemNotifyThread::start();
5081   }
5082   return;
5083 #endif
5084 }
5085 
5086 // Mark the polling page as unreadable
5087 void os::make_polling_page_unreadable(void) {
5088   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5089     fatal("Could not disable polling page");
5090 };
5091 
5092 // Mark the polling page as readable
5093 void os::make_polling_page_readable(void) {
5094   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5095     fatal("Could not enable polling page");
5096   }
5097 };
5098 
5099 int os::active_processor_count() {
5100   // Linux doesn't yet have a (official) notion of processor sets,
5101   // so just return the number of online processors.
5102   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5103   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
5104   return online_cpus;
5105 }
5106 
5107 void os::set_native_thread_name(const char *name) {
5108   // Not yet implemented.
5109   return;
5110 }
5111 
5112 bool os::distribute_processes(uint length, uint* distribution) {
5113   // Not yet implemented.
5114   return false;
5115 }
5116 
5117 bool os::bind_to_processor(uint processor_id) {
5118   // Not yet implemented.
5119   return false;
5120 }
5121 
5122 ///
5123 
5124 void os::SuspendedThreadTask::internal_do_task() {
5125   if (do_suspend(_thread->osthread())) {
5126     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5127     do_task(context);
5128     do_resume(_thread->osthread());
5129   }
5130 }
5131 
5132 class PcFetcher : public os::SuspendedThreadTask {
5133 public:
5134   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5135   ExtendedPC result();
5136 protected:
5137   void do_task(const os::SuspendedThreadTaskContext& context);
5138 private:
5139   ExtendedPC _epc;
5140 };
5141 
5142 ExtendedPC PcFetcher::result() {
5143   guarantee(is_done(), "task is not done yet.");
5144   return _epc;
5145 }
5146 
5147 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5148   Thread* thread = context.thread();
5149   OSThread* osthread = thread->osthread();
5150   if (osthread->ucontext() != NULL) {
5151     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5152   } else {
5153     // NULL context is unexpected, double-check this is the VMThread
5154     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5155   }
5156 }
5157 
5158 // Suspends the target using the signal mechanism and then grabs the PC before
5159 // resuming the target. Used by the flat-profiler only
5160 ExtendedPC os::get_thread_pc(Thread* thread) {
5161   // Make sure that it is called by the watcher for the VMThread
5162   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5163   assert(thread->is_VM_thread(), "Can only be called for VMThread");
5164 
5165   PcFetcher fetcher(thread);
5166   fetcher.run();
5167   return fetcher.result();
5168 }
5169 
5170 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5171 {
5172    if (is_NPTL()) {
5173       return pthread_cond_timedwait(_cond, _mutex, _abstime);
5174    } else {
5175 #ifndef IA64
5176       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5177       // word back to default 64bit precision if condvar is signaled. Java
5178       // wants 53bit precision.  Save and restore current value.
5179       int fpu = get_fpu_control_word();
5180 #endif // IA64
5181       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5182 #ifndef IA64
5183       set_fpu_control_word(fpu);
5184 #endif // IA64
5185       return status;
5186    }
5187 }
5188 
5189 ////////////////////////////////////////////////////////////////////////////////
5190 // debug support
5191 
5192 static address same_page(address x, address y) {
5193   intptr_t page_bits = -os::vm_page_size();
5194   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
5195     return x;
5196   else if (x > y)
5197     return (address)(intptr_t(y) | ~page_bits) + 1;
5198   else
5199     return (address)(intptr_t(y) & page_bits);
5200 }
5201 
5202 bool os::find(address addr, outputStream* st) {
5203   Dl_info dlinfo;
5204   memset(&dlinfo, 0, sizeof(dlinfo));
5205   if (dladdr(addr, &dlinfo) != 0) {
5206     st->print(PTR_FORMAT ": ", addr);
5207     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5208       st->print("%s+%#x", dlinfo.dli_sname,
5209                  addr - (intptr_t)dlinfo.dli_saddr);
5210     } else if (dlinfo.dli_fbase != NULL) {
5211       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5212     } else {
5213       st->print("<absolute address>");
5214     }
5215     if (dlinfo.dli_fname != NULL) {
5216       st->print(" in %s", dlinfo.dli_fname);
5217     }
5218     if (dlinfo.dli_fbase != NULL) {
5219       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5220     }
5221     st->cr();
5222 
5223     if (Verbose) {
5224       // decode some bytes around the PC
5225       address begin = same_page(addr-40, addr);
5226       address end   = same_page(addr+40, addr);
5227       address       lowest = (address) dlinfo.dli_sname;
5228       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5229       if (begin < lowest)  begin = lowest;
5230       Dl_info dlinfo2;
5231       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5232           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5233         end = (address) dlinfo2.dli_saddr;
5234       Disassembler::decode(begin, end, st);
5235     }
5236     return true;
5237   }
5238   return false;
5239 }
5240 
5241 ////////////////////////////////////////////////////////////////////////////////
5242 // misc
5243 
5244 // This does not do anything on Linux. This is basically a hook for being
5245 // able to use structured exception handling (thread-local exception filters)
5246 // on, e.g., Win32.
5247 void
5248 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5249                          JavaCallArguments* args, Thread* thread) {
5250   f(value, method, args, thread);
5251 }
5252 
5253 void os::print_statistics() {
5254 }
5255 
5256 int os::message_box(const char* title, const char* message) {
5257   int i;
5258   fdStream err(defaultStream::error_fd());
5259   for (i = 0; i < 78; i++) err.print_raw("=");
5260   err.cr();
5261   err.print_raw_cr(title);
5262   for (i = 0; i < 78; i++) err.print_raw("-");
5263   err.cr();
5264   err.print_raw_cr(message);
5265   for (i = 0; i < 78; i++) err.print_raw("=");
5266   err.cr();
5267 
5268   char buf[16];
5269   // Prevent process from exiting upon "read error" without consuming all CPU
5270   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5271 
5272   return buf[0] == 'y' || buf[0] == 'Y';
5273 }
5274 
5275 int os::stat(const char *path, struct stat *sbuf) {
5276   char pathbuf[MAX_PATH];
5277   if (strlen(path) > MAX_PATH - 1) {
5278     errno = ENAMETOOLONG;
5279     return -1;
5280   }
5281   os::native_path(strcpy(pathbuf, path));
5282   return ::stat(pathbuf, sbuf);
5283 }
5284 
5285 bool os::check_heap(bool force) {
5286   return true;
5287 }
5288 
5289 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5290   return ::vsnprintf(buf, count, format, args);
5291 }
5292 
5293 // Is a (classpath) directory empty?
5294 bool os::dir_is_empty(const char* path) {
5295   DIR *dir = NULL;
5296   struct dirent *ptr;
5297 
5298   dir = opendir(path);
5299   if (dir == NULL) return true;
5300 
5301   /* Scan the directory */
5302   bool result = true;
5303   char buf[sizeof(struct dirent) + MAX_PATH];
5304   while (result && (ptr = ::readdir(dir)) != NULL) {
5305     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5306       result = false;
5307     }
5308   }
5309   closedir(dir);
5310   return result;
5311 }
5312 
5313 // This code originates from JDK's sysOpen and open64_w
5314 // from src/solaris/hpi/src/system_md.c
5315 
5316 #ifndef O_DELETE
5317 #define O_DELETE 0x10000
5318 #endif
5319 
5320 // Open a file. Unlink the file immediately after open returns
5321 // if the specified oflag has the O_DELETE flag set.
5322 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5323 
5324 int os::open(const char *path, int oflag, int mode) {
5325 
5326   if (strlen(path) > MAX_PATH - 1) {
5327     errno = ENAMETOOLONG;
5328     return -1;
5329   }
5330   int fd;
5331   int o_delete = (oflag & O_DELETE);
5332   oflag = oflag & ~O_DELETE;
5333 
5334   fd = ::open64(path, oflag, mode);
5335   if (fd == -1) return -1;
5336 
5337   //If the open succeeded, the file might still be a directory
5338   {
5339     struct stat64 buf64;
5340     int ret = ::fstat64(fd, &buf64);
5341     int st_mode = buf64.st_mode;
5342 
5343     if (ret != -1) {
5344       if ((st_mode & S_IFMT) == S_IFDIR) {
5345         errno = EISDIR;
5346         ::close(fd);
5347         return -1;
5348       }
5349     } else {
5350       ::close(fd);
5351       return -1;
5352     }
5353   }
5354 
5355     /*
5356      * All file descriptors that are opened in the JVM and not
5357      * specifically destined for a subprocess should have the
5358      * close-on-exec flag set.  If we don't set it, then careless 3rd
5359      * party native code might fork and exec without closing all
5360      * appropriate file descriptors (e.g. as we do in closeDescriptors in
5361      * UNIXProcess.c), and this in turn might:
5362      *
5363      * - cause end-of-file to fail to be detected on some file
5364      *   descriptors, resulting in mysterious hangs, or
5365      *
5366      * - might cause an fopen in the subprocess to fail on a system
5367      *   suffering from bug 1085341.
5368      *
5369      * (Yes, the default setting of the close-on-exec flag is a Unix
5370      * design flaw)
5371      *
5372      * See:
5373      * 1085341: 32-bit stdio routines should support file descriptors >255
5374      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5375      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5376      */
5377 #ifdef FD_CLOEXEC
5378     {
5379         int flags = ::fcntl(fd, F_GETFD);
5380         if (flags != -1)
5381             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5382     }
5383 #endif
5384 
5385   if (o_delete != 0) {
5386     ::unlink(path);
5387   }
5388   return fd;
5389 }
5390 
5391 
5392 // create binary file, rewriting existing file if required
5393 int os::create_binary_file(const char* path, bool rewrite_existing) {
5394   int oflags = O_WRONLY | O_CREAT;
5395   if (!rewrite_existing) {
5396     oflags |= O_EXCL;
5397   }
5398   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5399 }
5400 
5401 // return current position of file pointer
5402 jlong os::current_file_offset(int fd) {
5403   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5404 }
5405 
5406 // move file pointer to the specified offset
5407 jlong os::seek_to_file_offset(int fd, jlong offset) {
5408   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5409 }
5410 
5411 // This code originates from JDK's sysAvailable
5412 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5413 
5414 int os::available(int fd, jlong *bytes) {
5415   jlong cur, end;
5416   int mode;
5417   struct stat64 buf64;
5418 
5419   if (::fstat64(fd, &buf64) >= 0) {
5420     mode = buf64.st_mode;
5421     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5422       /*
5423       * XXX: is the following call interruptible? If so, this might
5424       * need to go through the INTERRUPT_IO() wrapper as for other
5425       * blocking, interruptible calls in this file.
5426       */
5427       int n;
5428       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5429         *bytes = n;
5430         return 1;
5431       }
5432     }
5433   }
5434   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5435     return 0;
5436   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5437     return 0;
5438   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5439     return 0;
5440   }
5441   *bytes = end - cur;
5442   return 1;
5443 }
5444 
5445 int os::socket_available(int fd, jint *pbytes) {
5446   // Linux doc says EINTR not returned, unlike Solaris
5447   int ret = ::ioctl(fd, FIONREAD, pbytes);
5448 
5449   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5450   // is expected to return 0 on failure and 1 on success to the jdk.
5451   return (ret < 0) ? 0 : 1;
5452 }
5453 
5454 // Map a block of memory.
5455 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5456                      char *addr, size_t bytes, bool read_only,
5457                      bool allow_exec) {
5458   int prot;
5459   int flags = MAP_PRIVATE;
5460 
5461   if (read_only) {
5462     prot = PROT_READ;
5463   } else {
5464     prot = PROT_READ | PROT_WRITE;
5465   }
5466 
5467   if (allow_exec) {
5468     prot |= PROT_EXEC;
5469   }
5470 
5471   if (addr != NULL) {
5472     flags |= MAP_FIXED;
5473   }
5474 
5475   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5476                                      fd, file_offset);
5477   if (mapped_address == MAP_FAILED) {
5478     return NULL;
5479   }
5480   return mapped_address;
5481 }
5482 
5483 
5484 // Remap a block of memory.
5485 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5486                        char *addr, size_t bytes, bool read_only,
5487                        bool allow_exec) {
5488   // same as map_memory() on this OS
5489   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5490                         allow_exec);
5491 }
5492 
5493 
5494 // Unmap a block of memory.
5495 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5496   return munmap(addr, bytes) == 0;
5497 }
5498 
5499 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5500 
5501 static clockid_t thread_cpu_clockid(Thread* thread) {
5502   pthread_t tid = thread->osthread()->pthread_id();
5503   clockid_t clockid;
5504 
5505   // Get thread clockid
5506   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5507   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5508   return clockid;
5509 }
5510 
5511 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5512 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5513 // of a thread.
5514 //
5515 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5516 // the fast estimate available on the platform.
5517 
5518 jlong os::current_thread_cpu_time() {
5519   if (os::Linux::supports_fast_thread_cpu_time()) {
5520     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5521   } else {
5522     // return user + sys since the cost is the same
5523     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5524   }
5525 }
5526 
5527 jlong os::thread_cpu_time(Thread* thread) {
5528   // consistent with what current_thread_cpu_time() returns
5529   if (os::Linux::supports_fast_thread_cpu_time()) {
5530     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5531   } else {
5532     return slow_thread_cpu_time(thread, true /* user + sys */);
5533   }
5534 }
5535 
5536 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5537   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5538     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5539   } else {
5540     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5541   }
5542 }
5543 
5544 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5545   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5546     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5547   } else {
5548     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5549   }
5550 }
5551 
5552 //
5553 //  -1 on error.
5554 //
5555 
5556 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5557   static bool proc_pid_cpu_avail = true;
5558   static bool proc_task_unchecked = true;
5559   static const char *proc_stat_path = "/proc/%d/stat";
5560   pid_t  tid = thread->osthread()->thread_id();
5561   int i;
5562   char *s;
5563   char stat[2048];
5564   int statlen;
5565   char proc_name[64];
5566   int count;
5567   long sys_time, user_time;
5568   char string[64];
5569   char cdummy;
5570   int idummy;
5571   long ldummy;
5572   FILE *fp;
5573 
5574   // We first try accessing /proc/<pid>/cpu since this is faster to
5575   // process.  If this file is not present (linux kernels 2.5 and above)
5576   // then we open /proc/<pid>/stat.
5577   if ( proc_pid_cpu_avail ) {
5578     sprintf(proc_name, "/proc/%d/cpu", tid);
5579     fp =  fopen(proc_name, "r");
5580     if ( fp != NULL ) {
5581       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
5582       fclose(fp);
5583       if ( count != 3 ) return -1;
5584 
5585       if (user_sys_cpu_time) {
5586         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5587       } else {
5588         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5589       }
5590     }
5591     else proc_pid_cpu_avail = false;
5592   }
5593 
5594   // The /proc/<tid>/stat aggregates per-process usage on
5595   // new Linux kernels 2.6+ where NPTL is supported.
5596   // The /proc/self/task/<tid>/stat still has the per-thread usage.
5597   // See bug 6328462.
5598   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
5599   // and possibly in some other cases, so we check its availability.
5600   if (proc_task_unchecked && os::Linux::is_NPTL()) {
5601     // This is executed only once
5602     proc_task_unchecked = false;
5603     fp = fopen("/proc/self/task", "r");
5604     if (fp != NULL) {
5605       proc_stat_path = "/proc/self/task/%d/stat";
5606       fclose(fp);
5607     }
5608   }
5609 
5610   sprintf(proc_name, proc_stat_path, tid);
5611   fp = fopen(proc_name, "r");
5612   if ( fp == NULL ) return -1;
5613   statlen = fread(stat, 1, 2047, fp);
5614   stat[statlen] = '\0';
5615   fclose(fp);
5616 
5617   // Skip pid and the command string. Note that we could be dealing with
5618   // weird command names, e.g. user could decide to rename java launcher
5619   // to "java 1.4.2 :)", then the stat file would look like
5620   //                1234 (java 1.4.2 :)) R ... ...
5621   // We don't really need to know the command string, just find the last
5622   // occurrence of ")" and then start parsing from there. See bug 4726580.
5623   s = strrchr(stat, ')');
5624   i = 0;
5625   if (s == NULL ) return -1;
5626 
5627   // Skip blank chars
5628   do s++; while (isspace(*s));
5629 
5630   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5631                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5632                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5633                  &user_time, &sys_time);
5634   if ( count != 13 ) return -1;
5635   if (user_sys_cpu_time) {
5636     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5637   } else {
5638     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5639   }
5640 }
5641 
5642 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5643   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5644   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5645   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5646   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5647 }
5648 
5649 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5650   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5651   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5652   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5653   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5654 }
5655 
5656 bool os::is_thread_cpu_time_supported() {
5657   return true;
5658 }
5659 
5660 // System loadavg support.  Returns -1 if load average cannot be obtained.
5661 // Linux doesn't yet have a (official) notion of processor sets,
5662 // so just return the system wide load average.
5663 int os::loadavg(double loadavg[], int nelem) {
5664   return ::getloadavg(loadavg, nelem);
5665 }
5666 
5667 void os::pause() {
5668   char filename[MAX_PATH];
5669   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5670     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5671   } else {
5672     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5673   }
5674 
5675   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5676   if (fd != -1) {
5677     struct stat buf;
5678     ::close(fd);
5679     while (::stat(filename, &buf) == 0) {
5680       (void)::poll(NULL, 0, 100);
5681     }
5682   } else {
5683     jio_fprintf(stderr,
5684       "Could not open pause file '%s', continuing immediately.\n", filename);
5685   }
5686 }
5687 
5688 
5689 // Refer to the comments in os_solaris.cpp park-unpark.
5690 //
5691 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5692 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5693 // For specifics regarding the bug see GLIBC BUGID 261237 :
5694 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5695 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5696 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5697 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5698 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5699 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5700 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5701 // of libpthread avoids the problem, but isn't practical.
5702 //
5703 // Possible remedies:
5704 //
5705 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5706 //      This is palliative and probabilistic, however.  If the thread is preempted
5707 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5708 //      than the minimum period may have passed, and the abstime may be stale (in the
5709 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5710 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5711 //
5712 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5713 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5714 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5715 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5716 //      thread.
5717 //
5718 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5719 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5720 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5721 //      This also works well.  In fact it avoids kernel-level scalability impediments
5722 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5723 //      timers in a graceful fashion.
5724 //
5725 // 4.   When the abstime value is in the past it appears that control returns
5726 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5727 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5728 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5729 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5730 //      It may be possible to avoid reinitialization by checking the return
5731 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5732 //      condvar we must establish the invariant that cond_signal() is only called
5733 //      within critical sections protected by the adjunct mutex.  This prevents
5734 //      cond_signal() from "seeing" a condvar that's in the midst of being
5735 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5736 //      desirable signal-after-unlock optimization that avoids futile context switching.
5737 //
5738 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5739 //      structure when a condvar is used or initialized.  cond_destroy()  would
5740 //      release the helper structure.  Our reinitialize-after-timedwait fix
5741 //      put excessive stress on malloc/free and locks protecting the c-heap.
5742 //
5743 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5744 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5745 // and only enabling the work-around for vulnerable environments.
5746 
5747 // utility to compute the abstime argument to timedwait:
5748 // millis is the relative timeout time
5749 // abstime will be the absolute timeout time
5750 // TODO: replace compute_abstime() with unpackTime()
5751 
5752 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5753   if (millis < 0)  millis = 0;
5754 
5755   jlong seconds = millis / 1000;
5756   millis %= 1000;
5757   if (seconds > 50000000) { // see man cond_timedwait(3T)
5758     seconds = 50000000;
5759   }
5760 
5761   if (os::Linux::supports_monotonic_clock()) {
5762     struct timespec now;
5763     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5764     assert_status(status == 0, status, "clock_gettime");
5765     abstime->tv_sec = now.tv_sec  + seconds;
5766     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5767     if (nanos >= NANOSECS_PER_SEC) {
5768       abstime->tv_sec += 1;
5769       nanos -= NANOSECS_PER_SEC;
5770     }
5771     abstime->tv_nsec = nanos;
5772   } else {
5773     struct timeval now;
5774     int status = gettimeofday(&now, NULL);
5775     assert(status == 0, "gettimeofday");
5776     abstime->tv_sec = now.tv_sec  + seconds;
5777     long usec = now.tv_usec + millis * 1000;
5778     if (usec >= 1000000) {
5779       abstime->tv_sec += 1;
5780       usec -= 1000000;
5781     }
5782     abstime->tv_nsec = usec * 1000;
5783   }
5784   return abstime;
5785 }
5786 
5787 
5788 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5789 // Conceptually TryPark() should be equivalent to park(0).
5790 
5791 int os::PlatformEvent::TryPark() {
5792   for (;;) {
5793     const int v = _Event ;
5794     guarantee ((v == 0) || (v == 1), "invariant") ;
5795     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5796   }
5797 }
5798 
5799 void os::PlatformEvent::park() {       // AKA "down()"
5800   // Invariant: Only the thread associated with the Event/PlatformEvent
5801   // may call park().
5802   // TODO: assert that _Assoc != NULL or _Assoc == Self
5803   int v ;
5804   for (;;) {
5805       v = _Event ;
5806       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5807   }
5808   guarantee (v >= 0, "invariant") ;
5809   if (v == 0) {
5810      // Do this the hard way by blocking ...
5811      int status = pthread_mutex_lock(_mutex);
5812      assert_status(status == 0, status, "mutex_lock");
5813      guarantee (_nParked == 0, "invariant") ;
5814      ++ _nParked ;
5815      while (_Event < 0) {
5816         status = pthread_cond_wait(_cond, _mutex);
5817         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5818         // Treat this the same as if the wait was interrupted
5819         if (status == ETIME) { status = EINTR; }
5820         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5821      }
5822      -- _nParked ;
5823 
5824     _Event = 0 ;
5825      status = pthread_mutex_unlock(_mutex);
5826      assert_status(status == 0, status, "mutex_unlock");
5827     // Paranoia to ensure our locked and lock-free paths interact
5828     // correctly with each other.
5829     OrderAccess::fence();
5830   }
5831   guarantee (_Event >= 0, "invariant") ;
5832 }
5833 
5834 int os::PlatformEvent::park(jlong millis) {
5835   guarantee (_nParked == 0, "invariant") ;
5836 
5837   int v ;
5838   for (;;) {
5839       v = _Event ;
5840       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5841   }
5842   guarantee (v >= 0, "invariant") ;
5843   if (v != 0) return OS_OK ;
5844 
5845   // We do this the hard way, by blocking the thread.
5846   // Consider enforcing a minimum timeout value.
5847   struct timespec abst;
5848   compute_abstime(&abst, millis);
5849 
5850   int ret = OS_TIMEOUT;
5851   int status = pthread_mutex_lock(_mutex);
5852   assert_status(status == 0, status, "mutex_lock");
5853   guarantee (_nParked == 0, "invariant") ;
5854   ++_nParked ;
5855 
5856   // Object.wait(timo) will return because of
5857   // (a) notification
5858   // (b) timeout
5859   // (c) thread.interrupt
5860   //
5861   // Thread.interrupt and object.notify{All} both call Event::set.
5862   // That is, we treat thread.interrupt as a special case of notification.
5863   // The underlying Solaris implementation, cond_timedwait, admits
5864   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5865   // JVM from making those visible to Java code.  As such, we must
5866   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5867   //
5868   // TODO: properly differentiate simultaneous notify+interrupt.
5869   // In that case, we should propagate the notify to another waiter.
5870 
5871   while (_Event < 0) {
5872     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5873     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5874       pthread_cond_destroy (_cond);
5875       pthread_cond_init (_cond, os::Linux::condAttr()) ;
5876     }
5877     assert_status(status == 0 || status == EINTR ||
5878                   status == ETIME || status == ETIMEDOUT,
5879                   status, "cond_timedwait");
5880     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5881     if (status == ETIME || status == ETIMEDOUT) break ;
5882     // We consume and ignore EINTR and spurious wakeups.
5883   }
5884   --_nParked ;
5885   if (_Event >= 0) {
5886      ret = OS_OK;
5887   }
5888   _Event = 0 ;
5889   status = pthread_mutex_unlock(_mutex);
5890   assert_status(status == 0, status, "mutex_unlock");
5891   assert (_nParked == 0, "invariant") ;
5892   // Paranoia to ensure our locked and lock-free paths interact
5893   // correctly with each other.
5894   OrderAccess::fence();
5895   return ret;
5896 }
5897 
5898 void os::PlatformEvent::unpark() {
5899   // Transitions for _Event:
5900   //    0 :=> 1
5901   //    1 :=> 1
5902   //   -1 :=> either 0 or 1; must signal target thread
5903   //          That is, we can safely transition _Event from -1 to either
5904   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
5905   //          unpark() calls.
5906   // See also: "Semaphores in Plan 9" by Mullender & Cox
5907   //
5908   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5909   // that it will take two back-to-back park() calls for the owning
5910   // thread to block. This has the benefit of forcing a spurious return
5911   // from the first park() call after an unpark() call which will help
5912   // shake out uses of park() and unpark() without condition variables.
5913 
5914   if (Atomic::xchg(1, &_Event) >= 0) return;
5915 
5916   // Wait for the thread associated with the event to vacate
5917   int status = pthread_mutex_lock(_mutex);
5918   assert_status(status == 0, status, "mutex_lock");
5919   int AnyWaiters = _nParked;
5920   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5921   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5922     AnyWaiters = 0;
5923     pthread_cond_signal(_cond);
5924   }
5925   status = pthread_mutex_unlock(_mutex);
5926   assert_status(status == 0, status, "mutex_unlock");
5927   if (AnyWaiters != 0) {
5928     status = pthread_cond_signal(_cond);
5929     assert_status(status == 0, status, "cond_signal");
5930   }
5931 
5932   // Note that we signal() _after dropping the lock for "immortal" Events.
5933   // This is safe and avoids a common class of  futile wakeups.  In rare
5934   // circumstances this can cause a thread to return prematurely from
5935   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5936   // simply re-test the condition and re-park itself.
5937 }
5938 
5939 
5940 // JSR166
5941 // -------------------------------------------------------
5942 
5943 /*
5944  * The solaris and linux implementations of park/unpark are fairly
5945  * conservative for now, but can be improved. They currently use a
5946  * mutex/condvar pair, plus a a count.
5947  * Park decrements count if > 0, else does a condvar wait.  Unpark
5948  * sets count to 1 and signals condvar.  Only one thread ever waits
5949  * on the condvar. Contention seen when trying to park implies that someone
5950  * is unparking you, so don't wait. And spurious returns are fine, so there
5951  * is no need to track notifications.
5952  */
5953 
5954 /*
5955  * This code is common to linux and solaris and will be moved to a
5956  * common place in dolphin.
5957  *
5958  * The passed in time value is either a relative time in nanoseconds
5959  * or an absolute time in milliseconds. Either way it has to be unpacked
5960  * into suitable seconds and nanoseconds components and stored in the
5961  * given timespec structure.
5962  * Given time is a 64-bit value and the time_t used in the timespec is only
5963  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5964  * overflow if times way in the future are given. Further on Solaris versions
5965  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5966  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5967  * As it will be 28 years before "now + 100000000" will overflow we can
5968  * ignore overflow and just impose a hard-limit on seconds using the value
5969  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5970  * years from "now".
5971  */
5972 
5973 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5974   assert (time > 0, "convertTime");
5975   time_t max_secs = 0;
5976 
5977   if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5978     struct timeval now;
5979     int status = gettimeofday(&now, NULL);
5980     assert(status == 0, "gettimeofday");
5981 
5982     max_secs = now.tv_sec + MAX_SECS;
5983 
5984     if (isAbsolute) {
5985       jlong secs = time / 1000;
5986       if (secs > max_secs) {
5987         absTime->tv_sec = max_secs;
5988       } else {
5989         absTime->tv_sec = secs;
5990       }
5991       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5992     } else {
5993       jlong secs = time / NANOSECS_PER_SEC;
5994       if (secs >= MAX_SECS) {
5995         absTime->tv_sec = max_secs;
5996         absTime->tv_nsec = 0;
5997       } else {
5998         absTime->tv_sec = now.tv_sec + secs;
5999         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6000         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6001           absTime->tv_nsec -= NANOSECS_PER_SEC;
6002           ++absTime->tv_sec; // note: this must be <= max_secs
6003         }
6004       }
6005     }
6006   } else {
6007     // must be relative using monotonic clock
6008     struct timespec now;
6009     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
6010     assert_status(status == 0, status, "clock_gettime");
6011     max_secs = now.tv_sec + MAX_SECS;
6012     jlong secs = time / NANOSECS_PER_SEC;
6013     if (secs >= MAX_SECS) {
6014       absTime->tv_sec = max_secs;
6015       absTime->tv_nsec = 0;
6016     } else {
6017       absTime->tv_sec = now.tv_sec + secs;
6018       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
6019       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6020         absTime->tv_nsec -= NANOSECS_PER_SEC;
6021         ++absTime->tv_sec; // note: this must be <= max_secs
6022       }
6023     }
6024   }
6025   assert(absTime->tv_sec >= 0, "tv_sec < 0");
6026   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6027   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6028   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6029 }
6030 
6031 void Parker::park(bool isAbsolute, jlong time) {
6032   // Ideally we'd do something useful while spinning, such
6033   // as calling unpackTime().
6034 
6035   // Optional fast-path check:
6036   // Return immediately if a permit is available.
6037   // We depend on Atomic::xchg() having full barrier semantics
6038   // since we are doing a lock-free update to _counter.
6039   if (Atomic::xchg(0, &_counter) > 0) return;
6040 
6041   Thread* thread = Thread::current();
6042   assert(thread->is_Java_thread(), "Must be JavaThread");
6043   JavaThread *jt = (JavaThread *)thread;
6044 
6045   // Optional optimization -- avoid state transitions if there's an interrupt pending.
6046   // Check interrupt before trying to wait
6047   if (Thread::is_interrupted(thread, false)) {
6048     return;
6049   }
6050 
6051   // Next, demultiplex/decode time arguments
6052   timespec absTime;
6053   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6054     return;
6055   }
6056   if (time > 0) {
6057     unpackTime(&absTime, isAbsolute, time);
6058   }
6059 
6060 
6061   // Enter safepoint region
6062   // Beware of deadlocks such as 6317397.
6063   // The per-thread Parker:: mutex is a classic leaf-lock.
6064   // In particular a thread must never block on the Threads_lock while
6065   // holding the Parker:: mutex.  If safepoints are pending both the
6066   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6067   ThreadBlockInVM tbivm(jt);
6068 
6069   // Don't wait if cannot get lock since interference arises from
6070   // unblocking.  Also. check interrupt before trying wait
6071   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
6072     return;
6073   }
6074 
6075   int status ;
6076   if (_counter > 0)  { // no wait needed
6077     _counter = 0;
6078     status = pthread_mutex_unlock(_mutex);
6079     assert (status == 0, "invariant") ;
6080     // Paranoia to ensure our locked and lock-free paths interact
6081     // correctly with each other and Java-level accesses.
6082     OrderAccess::fence();
6083     return;
6084   }
6085 
6086 #ifdef ASSERT
6087   // Don't catch signals while blocked; let the running threads have the signals.
6088   // (This allows a debugger to break into the running thread.)
6089   sigset_t oldsigs;
6090   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
6091   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6092 #endif
6093 
6094   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6095   jt->set_suspend_equivalent();
6096   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6097   assert(_cur_index == -1, "invariant");
6098   if (time == 0) {
6099     _cur_index = REL_INDEX; // arbitrary choice when not timed
6100     status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
6101   } else {
6102     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
6103     status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
6104     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6105       pthread_cond_destroy (&_cond[_cur_index]) ;
6106       pthread_cond_init    (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
6107     }
6108   }
6109   _cur_index = -1;
6110 
6111   assert_status(status == 0 || status == EINTR ||
6112                 status == ETIME || status == ETIMEDOUT,
6113                 status, "cond_timedwait");
6114 
6115 #ifdef ASSERT
6116   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6117 #endif
6118 
6119   _counter = 0 ;
6120   status = pthread_mutex_unlock(_mutex) ;
6121   assert_status(status == 0, status, "invariant") ;
6122   // Paranoia to ensure our locked and lock-free paths interact
6123   // correctly with each other and Java-level accesses.
6124   OrderAccess::fence();
6125 
6126   // If externally suspended while waiting, re-suspend
6127   if (jt->handle_special_suspend_equivalent_condition()) {
6128     jt->java_suspend_self();
6129   }
6130 }
6131 
6132 void Parker::unpark() {
6133   int s, status ;
6134   status = pthread_mutex_lock(_mutex);
6135   assert (status == 0, "invariant") ;
6136   s = _counter;
6137   _counter = 1;
6138   if (s < 1) {
6139     // thread might be parked
6140     if (_cur_index != -1) {
6141       // thread is definitely parked
6142       if (WorkAroundNPTLTimedWaitHang) {
6143         status = pthread_cond_signal (&_cond[_cur_index]);
6144         assert (status == 0, "invariant");
6145         status = pthread_mutex_unlock(_mutex);
6146         assert (status == 0, "invariant");
6147       } else {
6148         status = pthread_mutex_unlock(_mutex);
6149         assert (status == 0, "invariant");
6150         status = pthread_cond_signal (&_cond[_cur_index]);
6151         assert (status == 0, "invariant");
6152       }
6153     } else {
6154       pthread_mutex_unlock(_mutex);
6155       assert (status == 0, "invariant") ;
6156     }
6157   } else {
6158     pthread_mutex_unlock(_mutex);
6159     assert (status == 0, "invariant") ;
6160   }
6161 }
6162 
6163 
6164 extern char** environ;
6165 
6166 #ifndef __NR_fork
6167 #ifdef BUILTIN_SIM
6168 #define __NR_fork 57
6169 #else
6170 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) AARCH64_ONLY(1079)
6171 #endif
6172 #endif
6173 
6174 #ifndef __NR_execve
6175 #ifdef BUILTIN_SIM
6176 #define __NR_execve 59
6177 #else
6178 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) AARCH64_ONLY(221)
6179 #endif
6180 #endif
6181 
6182 // Run the specified command in a separate process. Return its exit value,
6183 // or -1 on failure (e.g. can't fork a new process).
6184 // Unlike system(), this function can be called from signal handler. It
6185 // doesn't block SIGINT et al.
6186 int os::fork_and_exec(char* cmd) {
6187   const char * argv[4] = {"sh", "-c", cmd, NULL};
6188 
6189   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
6190   // pthread_atfork handlers and reset pthread library. All we need is a
6191   // separate process to execve. Make a direct syscall to fork process.
6192   // On IA64 there's no fork syscall, we have to use fork() and hope for
6193   // the best...
6194   pid_t pid = NOT_IA64(syscall(__NR_fork);)
6195               IA64_ONLY(fork();)
6196 
6197   if (pid < 0) {
6198     // fork failed
6199     return -1;
6200 
6201   } else if (pid == 0) {
6202     // child process
6203 
6204     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
6205     // first to kill every thread on the thread list. Because this list is
6206     // not reset by fork() (see notes above), execve() will instead kill
6207     // every thread in the parent process. We know this is the only thread
6208     // in the new process, so make a system call directly.
6209     // IA64 should use normal execve() from glibc to match the glibc fork()
6210     // above.
6211     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
6212     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
6213 
6214     // execve failed
6215     _exit(-1);
6216 
6217   } else  {
6218     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6219     // care about the actual exit code, for now.
6220 
6221     int status;
6222 
6223     // Wait for the child process to exit.  This returns immediately if
6224     // the child has already exited. */
6225     while (waitpid(pid, &status, 0) < 0) {
6226         switch (errno) {
6227         case ECHILD: return 0;
6228         case EINTR: break;
6229         default: return -1;
6230         }
6231     }
6232 
6233     if (WIFEXITED(status)) {
6234        // The child exited normally; get its exit code.
6235        return WEXITSTATUS(status);
6236     } else if (WIFSIGNALED(status)) {
6237        // The child exited because of a signal
6238        // The best value to return is 0x80 + signal number,
6239        // because that is what all Unix shells do, and because
6240        // it allows callers to distinguish between process exit and
6241        // process death by signal.
6242        return 0x80 + WTERMSIG(status);
6243     } else {
6244        // Unknown exit code; pass it through
6245        return status;
6246     }
6247   }
6248 }
6249 
6250 // is_headless_jre()
6251 //
6252 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6253 // in order to report if we are running in a headless jre
6254 //
6255 // Since JDK8 xawt/libmawt.so was moved into the same directory
6256 // as libawt.so, and renamed libawt_xawt.so
6257 //
6258 bool os::is_headless_jre() {
6259     struct stat statbuf;
6260     char buf[MAXPATHLEN];
6261     char libmawtpath[MAXPATHLEN];
6262     const char *xawtstr  = "/xawt/libmawt.so";
6263     const char *new_xawtstr = "/libawt_xawt.so";
6264     char *p;
6265 
6266     // Get path to libjvm.so
6267     os::jvm_path(buf, sizeof(buf));
6268 
6269     // Get rid of libjvm.so
6270     p = strrchr(buf, '/');
6271     if (p == NULL) return false;
6272     else *p = '\0';
6273 
6274     // Get rid of client or server
6275     p = strrchr(buf, '/');
6276     if (p == NULL) return false;
6277     else *p = '\0';
6278 
6279     // check xawt/libmawt.so
6280     strcpy(libmawtpath, buf);
6281     strcat(libmawtpath, xawtstr);
6282     if (::stat(libmawtpath, &statbuf) == 0) return false;
6283 
6284     // check libawt_xawt.so
6285     strcpy(libmawtpath, buf);
6286     strcat(libmawtpath, new_xawtstr);
6287     if (::stat(libmawtpath, &statbuf) == 0) return false;
6288 
6289     return true;
6290 }
6291 
6292 // Get the default path to the core file
6293 // Returns the length of the string
6294 int os::get_core_path(char* buffer, size_t bufferSize) {
6295   const char* p = get_current_directory(buffer, bufferSize);
6296 
6297   if (p == NULL) {
6298     assert(p != NULL, "failed to get current directory");
6299     return 0;
6300   }
6301 
6302   return strlen(buffer);
6303 }
6304 
6305 #ifdef JAVASE_EMBEDDED
6306 //
6307 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6308 //
6309 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6310 
6311 // ctor
6312 //
6313 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6314   assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6315   _fd = fd;
6316 
6317   if (os::create_thread(this, os::os_thread)) {
6318     _memnotify_thread = this;
6319     os::set_priority(this, NearMaxPriority);
6320     os::start_thread(this);
6321   }
6322 }
6323 
6324 // Where all the work gets done
6325 //
6326 void MemNotifyThread::run() {
6327   assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6328 
6329   // Set up the select arguments
6330   fd_set rfds;
6331   if (_fd != -1) {
6332     FD_ZERO(&rfds);
6333     FD_SET(_fd, &rfds);
6334   }
6335 
6336   // Now wait for the mem_notify device to wake up
6337   while (1) {
6338     // Wait for the mem_notify device to signal us..
6339     int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6340     if (rc == -1) {
6341       perror("select!\n");
6342       break;
6343     } else if (rc) {
6344       //ssize_t free_before = os::available_memory();
6345       //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6346 
6347       // The kernel is telling us there is not much memory left...
6348       // try to do something about that
6349 
6350       // If we are not already in a GC, try one.
6351       if (!Universe::heap()->is_gc_active()) {
6352         Universe::heap()->collect(GCCause::_allocation_failure);
6353 
6354         //ssize_t free_after = os::available_memory();
6355         //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6356         //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6357       }
6358       // We might want to do something like the following if we find the GC's are not helping...
6359       // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6360     }
6361   }
6362 }
6363 
6364 //
6365 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6366 //
6367 void MemNotifyThread::start() {
6368   int    fd;
6369   fd = open ("/dev/mem_notify", O_RDONLY, 0);
6370   if (fd < 0) {
6371       return;
6372   }
6373 
6374   if (memnotify_thread() == NULL) {
6375     new MemNotifyThread(fd);
6376   }
6377 }
6378 
6379 #endif // JAVASE_EMBEDDED
6380 
6381 
6382 /////////////// Unit tests ///////////////
6383 
6384 #ifndef PRODUCT
6385 
6386 #define test_log(...) \
6387   do {\
6388     if (VerboseInternalVMTests) { \
6389       tty->print_cr(__VA_ARGS__); \
6390       tty->flush(); \
6391     }\
6392   } while (false)
6393 
6394 class TestReserveMemorySpecial : AllStatic {
6395  public:
6396   static void small_page_write(void* addr, size_t size) {
6397     size_t page_size = os::vm_page_size();
6398 
6399     char* end = (char*)addr + size;
6400     for (char* p = (char*)addr; p < end; p += page_size) {
6401       *p = 1;
6402     }
6403   }
6404 
6405   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6406     if (!UseHugeTLBFS) {
6407       return;
6408     }
6409 
6410     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6411 
6412     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6413 
6414     if (addr != NULL) {
6415       small_page_write(addr, size);
6416 
6417       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6418     }
6419   }
6420 
6421   static void test_reserve_memory_special_huge_tlbfs_only() {
6422     if (!UseHugeTLBFS) {
6423       return;
6424     }
6425 
6426     size_t lp = os::large_page_size();
6427 
6428     for (size_t size = lp; size <= lp * 10; size += lp) {
6429       test_reserve_memory_special_huge_tlbfs_only(size);
6430     }
6431   }
6432 
6433   static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6434     if (!UseHugeTLBFS) {
6435         return;
6436     }
6437 
6438     test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6439         size, alignment);
6440 
6441     assert(size >= os::large_page_size(), "Incorrect input to test");
6442 
6443     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6444 
6445     if (addr != NULL) {
6446       small_page_write(addr, size);
6447 
6448       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6449     }
6450   }
6451 
6452   static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6453     size_t lp = os::large_page_size();
6454     size_t ag = os::vm_allocation_granularity();
6455 
6456     for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6457       test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6458     }
6459   }
6460 
6461   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6462     size_t lp = os::large_page_size();
6463     size_t ag = os::vm_allocation_granularity();
6464 
6465     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6466     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6467     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6468     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6469     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6470     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6471     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6472     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6473     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6474   }
6475 
6476   static void test_reserve_memory_special_huge_tlbfs() {
6477     if (!UseHugeTLBFS) {
6478       return;
6479     }
6480 
6481     test_reserve_memory_special_huge_tlbfs_only();
6482     test_reserve_memory_special_huge_tlbfs_mixed();
6483   }
6484 
6485   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6486     if (!UseSHM) {
6487       return;
6488     }
6489 
6490     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6491 
6492     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6493 
6494     if (addr != NULL) {
6495       assert(is_ptr_aligned(addr, alignment), "Check");
6496       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6497 
6498       small_page_write(addr, size);
6499 
6500       os::Linux::release_memory_special_shm(addr, size);
6501     }
6502   }
6503 
6504   static void test_reserve_memory_special_shm() {
6505     size_t lp = os::large_page_size();
6506     size_t ag = os::vm_allocation_granularity();
6507 
6508     for (size_t size = ag; size < lp * 3; size += ag) {
6509       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6510         test_reserve_memory_special_shm(size, alignment);
6511       }
6512     }
6513   }
6514 
6515   static void test() {
6516     test_reserve_memory_special_huge_tlbfs();
6517     test_reserve_memory_special_shm();
6518   }
6519 };
6520 
6521 void TestReserveMemorySpecial_test() {
6522   TestReserveMemorySpecial::test();
6523 }
6524 
6525 #endif