1 /*
   2  * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "memory/allocation.inline.hpp"
  36 #include "memory/filemap.hpp"
  37 #include "mutex_linux.inline.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_linux.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "prims/jniFastGetField.hpp"
  42 #include "prims/jvm.h"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.inline.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.hpp"
  49 #include "runtime/init.hpp"
  50 #include "runtime/java.hpp"
  51 #include "runtime/javaCalls.hpp"
  52 #include "runtime/mutexLocker.hpp"
  53 #include "runtime/objectMonitor.hpp"
  54 #include "runtime/orderAccess.inline.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/perfMemory.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/statSampler.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "runtime/thread.inline.hpp"
  61 #include "runtime/threadCritical.hpp"
  62 #include "runtime/timer.hpp"
  63 #include "services/attachListener.hpp"
  64 #include "services/memTracker.hpp"
  65 #include "services/runtimeService.hpp"
  66 #include "utilities/decoder.hpp"
  67 #include "utilities/defaultStream.hpp"
  68 #include "utilities/events.hpp"
  69 #include "utilities/elfFile.hpp"
  70 #include "utilities/growableArray.hpp"
  71 #include "utilities/macros.hpp"
  72 #include "utilities/vmError.hpp"
  73 
  74 // put OS-includes here
  75 # include <sys/types.h>
  76 # include <sys/mman.h>
  77 # include <sys/stat.h>
  78 # include <sys/select.h>
  79 # include <pthread.h>
  80 # include <signal.h>
  81 # include <errno.h>
  82 # include <dlfcn.h>
  83 # include <stdio.h>
  84 # include <unistd.h>
  85 # include <sys/resource.h>
  86 # include <pthread.h>
  87 # include <sys/stat.h>
  88 # include <sys/time.h>
  89 # include <sys/times.h>
  90 # include <sys/utsname.h>
  91 # include <sys/socket.h>
  92 # include <sys/wait.h>
  93 # include <pwd.h>
  94 # include <poll.h>
  95 # include <semaphore.h>
  96 # include <fcntl.h>
  97 # include <string.h>
  98 # include <syscall.h>
  99 # include <sys/sysinfo.h>
 100 # include <gnu/libc-version.h>
 101 # include <sys/ipc.h>
 102 # include <sys/shm.h>
 103 # include <link.h>
 104 # include <stdint.h>
 105 # include <inttypes.h>
 106 # include <sys/ioctl.h>
 107 
 108 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
 109 
 110 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 111 // getrusage() is prepared to handle the associated failure.
 112 #ifndef RUSAGE_THREAD
 113   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 114 #endif
 115 
 116 #define MAX_PATH    (2 * K)
 117 
 118 #define MAX_SECS 100000000
 119 
 120 // for timer info max values which include all bits
 121 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 122 
 123 #define LARGEPAGES_BIT (1 << 6)
 124 ////////////////////////////////////////////////////////////////////////////////
 125 // global variables
 126 julong os::Linux::_physical_memory = 0;
 127 
 128 address   os::Linux::_initial_thread_stack_bottom = NULL;
 129 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 130 
 131 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 132 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 133 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 134 Mutex* os::Linux::_createThread_lock = NULL;
 135 pthread_t os::Linux::_main_thread;
 136 int os::Linux::_page_size = -1;
 137 const int os::Linux::_vm_default_page_size = (8 * K);
 138 bool os::Linux::_is_floating_stack = false;
 139 bool os::Linux::_is_NPTL = false;
 140 bool os::Linux::_supports_fast_thread_cpu_time = false;
 141 const char * os::Linux::_glibc_version = NULL;
 142 const char * os::Linux::_libpthread_version = NULL;
 143 pthread_condattr_t os::Linux::_condattr[1];
 144 
 145 static jlong initial_time_count=0;
 146 
 147 static int clock_tics_per_sec = 100;
 148 
 149 // For diagnostics to print a message once. see run_periodic_checks
 150 static sigset_t check_signal_done;
 151 static bool check_signals = true;
 152 
 153 static pid_t _initial_pid = 0;
 154 
 155 // Signal number used to suspend/resume a thread
 156 
 157 // do not use any signal number less than SIGSEGV, see 4355769
 158 static int SR_signum = SIGUSR2;
 159 sigset_t SR_sigset;
 160 
 161 // Used to protect dlsym() calls
 162 static pthread_mutex_t dl_mutex;
 163 
 164 // Declarations
 165 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 166 
 167 // utility functions
 168 
 169 static int SR_initialize();
 170 
 171 julong os::available_memory() {
 172   return Linux::available_memory();
 173 }
 174 
 175 julong os::Linux::available_memory() {
 176   // values in struct sysinfo are "unsigned long"
 177   struct sysinfo si;
 178   sysinfo(&si);
 179 
 180   return (julong)si.freeram * si.mem_unit;
 181 }
 182 
 183 julong os::physical_memory() {
 184   return Linux::physical_memory();
 185 }
 186 
 187 // Return true if user is running as root.
 188 
 189 bool os::have_special_privileges() {
 190   static bool init = false;
 191   static bool privileges = false;
 192   if (!init) {
 193     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 194     init = true;
 195   }
 196   return privileges;
 197 }
 198 
 199 
 200 #ifndef SYS_gettid
 201 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 202   #ifdef __ia64__
 203     #define SYS_gettid 1105
 204   #elif __i386__
 205     #define SYS_gettid 224
 206   #elif __amd64__
 207     #define SYS_gettid 186
 208   #elif __sparc__
 209     #define SYS_gettid 143
 210   #else
 211     #error define gettid for the arch
 212   #endif
 213 #endif
 214 
 215 // Cpu architecture string
 216 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 217 
 218 
 219 // pid_t gettid()
 220 //
 221 // Returns the kernel thread id of the currently running thread. Kernel
 222 // thread id is used to access /proc.
 223 //
 224 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
 225 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 226 //
 227 pid_t os::Linux::gettid() {
 228   int rslt = syscall(SYS_gettid);
 229   if (rslt == -1) {
 230     // old kernel, no NPTL support
 231     return getpid();
 232   } else {
 233     return (pid_t)rslt;
 234   }
 235 }
 236 
 237 // Most versions of linux have a bug where the number of processors are
 238 // determined by looking at the /proc file system.  In a chroot environment,
 239 // the system call returns 1.  This causes the VM to act as if it is
 240 // a single processor and elide locking (see is_MP() call).
 241 static bool unsafe_chroot_detected = false;
 242 static const char *unstable_chroot_error = "/proc file system not found.\n"
 243                      "Java may be unstable running multithreaded in a chroot "
 244                      "environment on Linux when /proc filesystem is not mounted.";
 245 
 246 void os::Linux::initialize_system_info() {
 247   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 248   if (processor_count() == 1) {
 249     pid_t pid = os::Linux::gettid();
 250     char fname[32];
 251     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 252     FILE *fp = fopen(fname, "r");
 253     if (fp == NULL) {
 254       unsafe_chroot_detected = true;
 255     } else {
 256       fclose(fp);
 257     }
 258   }
 259   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 260   assert(processor_count() > 0, "linux error");
 261 }
 262 
 263 void os::init_system_properties_values() {
 264   // The next steps are taken in the product version:
 265   //
 266   // Obtain the JAVA_HOME value from the location of libjvm.so.
 267   // This library should be located at:
 268   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 269   //
 270   // If "/jre/lib/" appears at the right place in the path, then we
 271   // assume libjvm.so is installed in a JDK and we use this path.
 272   //
 273   // Otherwise exit with message: "Could not create the Java virtual machine."
 274   //
 275   // The following extra steps are taken in the debugging version:
 276   //
 277   // If "/jre/lib/" does NOT appear at the right place in the path
 278   // instead of exit check for $JAVA_HOME environment variable.
 279   //
 280   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 281   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 282   // it looks like libjvm.so is installed there
 283   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 284   //
 285   // Otherwise exit.
 286   //
 287   // Important note: if the location of libjvm.so changes this
 288   // code needs to be changed accordingly.
 289 
 290   // See ld(1):
 291   //      The linker uses the following search paths to locate required
 292   //      shared libraries:
 293   //        1: ...
 294   //        ...
 295   //        7: The default directories, normally /lib and /usr/lib.
 296 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 297   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 298 #else
 299   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 300 #endif
 301 
 302 // Base path of extensions installed on the system.
 303 #define SYS_EXT_DIR     "/usr/java/packages"
 304 #define EXTENSIONS_DIR  "/lib/ext"
 305 
 306   // Buffer that fits several sprintfs.
 307   // Note that the space for the colon and the trailing null are provided
 308   // by the nulls included by the sizeof operator.
 309   const size_t bufsize =
 310     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 311          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 312   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 313 
 314   // sysclasspath, java_home, dll_dir
 315   {
 316     char *pslash;
 317     os::jvm_path(buf, bufsize);
 318 
 319     // Found the full path to libjvm.so.
 320     // Now cut the path to <java_home>/jre if we can.
 321     pslash = strrchr(buf, '/');
 322     if (pslash != NULL) {
 323       *pslash = '\0';            // Get rid of /libjvm.so.
 324     }
 325     pslash = strrchr(buf, '/');
 326     if (pslash != NULL) {
 327       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 328     }
 329     Arguments::set_dll_dir(buf);
 330 
 331     if (pslash != NULL) {
 332       pslash = strrchr(buf, '/');
 333       if (pslash != NULL) {
 334         *pslash = '\0';          // Get rid of /<arch>.
 335         pslash = strrchr(buf, '/');
 336         if (pslash != NULL) {
 337           *pslash = '\0';        // Get rid of /lib.
 338         }
 339       }
 340     }
 341     Arguments::set_java_home(buf);
 342     set_boot_path('/', ':');
 343   }
 344 
 345   // Where to look for native libraries.
 346   //
 347   // Note: Due to a legacy implementation, most of the library path
 348   // is set in the launcher. This was to accomodate linking restrictions
 349   // on legacy Linux implementations (which are no longer supported).
 350   // Eventually, all the library path setting will be done here.
 351   //
 352   // However, to prevent the proliferation of improperly built native
 353   // libraries, the new path component /usr/java/packages is added here.
 354   // Eventually, all the library path setting will be done here.
 355   {
 356     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 357     // should always exist (until the legacy problem cited above is
 358     // addressed).
 359     const char *v = ::getenv("LD_LIBRARY_PATH");
 360     const char *v_colon = ":";
 361     if (v == NULL) { v = ""; v_colon = ""; }
 362     // That's +1 for the colon and +1 for the trailing '\0'.
 363     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 364                                                      strlen(v) + 1 +
 365                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 366                                                      mtInternal);
 367     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 368     Arguments::set_library_path(ld_library_path);
 369     FREE_C_HEAP_ARRAY(char, ld_library_path);
 370   }
 371 
 372   // Extensions directories.
 373   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 374   Arguments::set_ext_dirs(buf);
 375 
 376   FREE_C_HEAP_ARRAY(char, buf);
 377 
 378 #undef DEFAULT_LIBPATH
 379 #undef SYS_EXT_DIR
 380 #undef EXTENSIONS_DIR
 381 }
 382 
 383 ////////////////////////////////////////////////////////////////////////////////
 384 // breakpoint support
 385 
 386 void os::breakpoint() {
 387   BREAKPOINT;
 388 }
 389 
 390 extern "C" void breakpoint() {
 391   // use debugger to set breakpoint here
 392 }
 393 
 394 ////////////////////////////////////////////////////////////////////////////////
 395 // signal support
 396 
 397 debug_only(static bool signal_sets_initialized = false);
 398 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 399 
 400 bool os::Linux::is_sig_ignored(int sig) {
 401   struct sigaction oact;
 402   sigaction(sig, (struct sigaction*)NULL, &oact);
 403   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 404                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 405   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 406     return true;
 407   } else {
 408     return false;
 409   }
 410 }
 411 
 412 void os::Linux::signal_sets_init() {
 413   // Should also have an assertion stating we are still single-threaded.
 414   assert(!signal_sets_initialized, "Already initialized");
 415   // Fill in signals that are necessarily unblocked for all threads in
 416   // the VM. Currently, we unblock the following signals:
 417   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 418   //                         by -Xrs (=ReduceSignalUsage));
 419   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 420   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 421   // the dispositions or masks wrt these signals.
 422   // Programs embedding the VM that want to use the above signals for their
 423   // own purposes must, at this time, use the "-Xrs" option to prevent
 424   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 425   // (See bug 4345157, and other related bugs).
 426   // In reality, though, unblocking these signals is really a nop, since
 427   // these signals are not blocked by default.
 428   sigemptyset(&unblocked_sigs);
 429   sigemptyset(&allowdebug_blocked_sigs);
 430   sigaddset(&unblocked_sigs, SIGILL);
 431   sigaddset(&unblocked_sigs, SIGSEGV);
 432   sigaddset(&unblocked_sigs, SIGBUS);
 433   sigaddset(&unblocked_sigs, SIGFPE);
 434 #if defined(PPC64)
 435   sigaddset(&unblocked_sigs, SIGTRAP);
 436 #endif
 437   sigaddset(&unblocked_sigs, SR_signum);
 438 
 439   if (!ReduceSignalUsage) {
 440     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 441       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 442       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 443     }
 444     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 445       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 446       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 447     }
 448     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 449       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 450       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 451     }
 452   }
 453   // Fill in signals that are blocked by all but the VM thread.
 454   sigemptyset(&vm_sigs);
 455   if (!ReduceSignalUsage) {
 456     sigaddset(&vm_sigs, BREAK_SIGNAL);
 457   }
 458   debug_only(signal_sets_initialized = true);
 459 
 460 }
 461 
 462 // These are signals that are unblocked while a thread is running Java.
 463 // (For some reason, they get blocked by default.)
 464 sigset_t* os::Linux::unblocked_signals() {
 465   assert(signal_sets_initialized, "Not initialized");
 466   return &unblocked_sigs;
 467 }
 468 
 469 // These are the signals that are blocked while a (non-VM) thread is
 470 // running Java. Only the VM thread handles these signals.
 471 sigset_t* os::Linux::vm_signals() {
 472   assert(signal_sets_initialized, "Not initialized");
 473   return &vm_sigs;
 474 }
 475 
 476 // These are signals that are blocked during cond_wait to allow debugger in
 477 sigset_t* os::Linux::allowdebug_blocked_signals() {
 478   assert(signal_sets_initialized, "Not initialized");
 479   return &allowdebug_blocked_sigs;
 480 }
 481 
 482 void os::Linux::hotspot_sigmask(Thread* thread) {
 483 
 484   //Save caller's signal mask before setting VM signal mask
 485   sigset_t caller_sigmask;
 486   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 487 
 488   OSThread* osthread = thread->osthread();
 489   osthread->set_caller_sigmask(caller_sigmask);
 490 
 491   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 492 
 493   if (!ReduceSignalUsage) {
 494     if (thread->is_VM_thread()) {
 495       // Only the VM thread handles BREAK_SIGNAL ...
 496       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 497     } else {
 498       // ... all other threads block BREAK_SIGNAL
 499       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 500     }
 501   }
 502 }
 503 
 504 //////////////////////////////////////////////////////////////////////////////
 505 // detecting pthread library
 506 
 507 void os::Linux::libpthread_init() {
 508   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 509   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 510   // generic name for earlier versions.
 511   // Define macros here so we can build HotSpot on old systems.
 512 #ifndef _CS_GNU_LIBC_VERSION
 513   #define _CS_GNU_LIBC_VERSION 2
 514 #endif
 515 #ifndef _CS_GNU_LIBPTHREAD_VERSION
 516   #define _CS_GNU_LIBPTHREAD_VERSION 3
 517 #endif
 518 
 519   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 520   if (n > 0) {
 521     char *str = (char *)malloc(n, mtInternal);
 522     confstr(_CS_GNU_LIBC_VERSION, str, n);
 523     os::Linux::set_glibc_version(str);
 524   } else {
 525     // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 526     static char _gnu_libc_version[32];
 527     jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 528                  "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 529     os::Linux::set_glibc_version(_gnu_libc_version);
 530   }
 531 
 532   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 533   if (n > 0) {
 534     char *str = (char *)malloc(n, mtInternal);
 535     confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 536     // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 537     // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
 538     // is the case. LinuxThreads has a hard limit on max number of threads.
 539     // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 540     // On the other hand, NPTL does not have such a limit, sysconf()
 541     // will return -1 and errno is not changed. Check if it is really NPTL.
 542     if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
 543         strstr(str, "NPTL") &&
 544         sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 545       free(str);
 546       os::Linux::set_libpthread_version("linuxthreads");
 547     } else {
 548       os::Linux::set_libpthread_version(str);
 549     }
 550   } else {
 551     // glibc before 2.3.2 only has LinuxThreads.
 552     os::Linux::set_libpthread_version("linuxthreads");
 553   }
 554 
 555   if (strstr(libpthread_version(), "NPTL")) {
 556     os::Linux::set_is_NPTL();
 557   } else {
 558     os::Linux::set_is_LinuxThreads();
 559   }
 560 
 561   // LinuxThreads have two flavors: floating-stack mode, which allows variable
 562   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 563   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
 564     os::Linux::set_is_floating_stack();
 565   }
 566 }
 567 
 568 /////////////////////////////////////////////////////////////////////////////
 569 // thread stack
 570 
 571 // Force Linux kernel to expand current thread stack. If "bottom" is close
 572 // to the stack guard, caller should block all signals.
 573 //
 574 // MAP_GROWSDOWN:
 575 //   A special mmap() flag that is used to implement thread stacks. It tells
 576 //   kernel that the memory region should extend downwards when needed. This
 577 //   allows early versions of LinuxThreads to only mmap the first few pages
 578 //   when creating a new thread. Linux kernel will automatically expand thread
 579 //   stack as needed (on page faults).
 580 //
 581 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 582 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 583 //   region, it's hard to tell if the fault is due to a legitimate stack
 584 //   access or because of reading/writing non-exist memory (e.g. buffer
 585 //   overrun). As a rule, if the fault happens below current stack pointer,
 586 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 587 //   application (see Linux kernel fault.c).
 588 //
 589 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 590 //   stack overflow detection.
 591 //
 592 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 593 //   not use this flag. However, the stack of initial thread is not created
 594 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 595 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 596 //   and then attach the thread to JVM.
 597 //
 598 // To get around the problem and allow stack banging on Linux, we need to
 599 // manually expand thread stack after receiving the SIGSEGV.
 600 //
 601 // There are two ways to expand thread stack to address "bottom", we used
 602 // both of them in JVM before 1.5:
 603 //   1. adjust stack pointer first so that it is below "bottom", and then
 604 //      touch "bottom"
 605 //   2. mmap() the page in question
 606 //
 607 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 608 // if current sp is already near the lower end of page 101, and we need to
 609 // call mmap() to map page 100, it is possible that part of the mmap() frame
 610 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 611 // That will destroy the mmap() frame and cause VM to crash.
 612 //
 613 // The following code works by adjusting sp first, then accessing the "bottom"
 614 // page to force a page fault. Linux kernel will then automatically expand the
 615 // stack mapping.
 616 //
 617 // _expand_stack_to() assumes its frame size is less than page size, which
 618 // should always be true if the function is not inlined.
 619 
 620 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 621   #define NOINLINE
 622 #else
 623   #define NOINLINE __attribute__ ((noinline))
 624 #endif
 625 
 626 static void _expand_stack_to(address bottom) NOINLINE;
 627 
 628 static void _expand_stack_to(address bottom) {
 629   address sp;
 630   size_t size;
 631   volatile char *p;
 632 
 633   // Adjust bottom to point to the largest address within the same page, it
 634   // gives us a one-page buffer if alloca() allocates slightly more memory.
 635   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 636   bottom += os::Linux::page_size() - 1;
 637 
 638   // sp might be slightly above current stack pointer; if that's the case, we
 639   // will alloca() a little more space than necessary, which is OK. Don't use
 640   // os::current_stack_pointer(), as its result can be slightly below current
 641   // stack pointer, causing us to not alloca enough to reach "bottom".
 642   sp = (address)&sp;
 643 
 644   if (sp > bottom) {
 645     size = sp - bottom;
 646     p = (volatile char *)alloca(size);
 647     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 648     p[0] = '\0';
 649   }
 650 }
 651 
 652 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 653   assert(t!=NULL, "just checking");
 654   assert(t->osthread()->expanding_stack(), "expand should be set");
 655   assert(t->stack_base() != NULL, "stack_base was not initialized");
 656 
 657   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 658     sigset_t mask_all, old_sigset;
 659     sigfillset(&mask_all);
 660     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 661     _expand_stack_to(addr);
 662     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 663     return true;
 664   }
 665   return false;
 666 }
 667 
 668 //////////////////////////////////////////////////////////////////////////////
 669 // create new thread
 670 
 671 static address highest_vm_reserved_address();
 672 
 673 // check if it's safe to start a new thread
 674 static bool _thread_safety_check(Thread* thread) {
 675   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
 676     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
 677     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 678     //   allocated (MAP_FIXED) from high address space. Every thread stack
 679     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 680     //   it to other values if they rebuild LinuxThreads).
 681     //
 682     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 683     // the memory region has already been mmap'ed. That means if we have too
 684     // many threads and/or very large heap, eventually thread stack will
 685     // collide with heap.
 686     //
 687     // Here we try to prevent heap/stack collision by comparing current
 688     // stack bottom with the highest address that has been mmap'ed by JVM
 689     // plus a safety margin for memory maps created by native code.
 690     //
 691     // This feature can be disabled by setting ThreadSafetyMargin to 0
 692     //
 693     if (ThreadSafetyMargin > 0) {
 694       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 695 
 696       // not safe if our stack extends below the safety margin
 697       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 698     } else {
 699       return true;
 700     }
 701   } else {
 702     // Floating stack LinuxThreads or NPTL:
 703     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
 704     //   there's not enough space left, pthread_create() will fail. If we come
 705     //   here, that means enough space has been reserved for stack.
 706     return true;
 707   }
 708 }
 709 
 710 // Thread start routine for all newly created threads
 711 static void *java_start(Thread *thread) {
 712   // Try to randomize the cache line index of hot stack frames.
 713   // This helps when threads of the same stack traces evict each other's
 714   // cache lines. The threads can be either from the same JVM instance, or
 715   // from different JVM instances. The benefit is especially true for
 716   // processors with hyperthreading technology.
 717   static int counter = 0;
 718   int pid = os::current_process_id();
 719   alloca(((pid ^ counter++) & 7) * 128);
 720 
 721   ThreadLocalStorage::set_thread(thread);
 722 
 723   OSThread* osthread = thread->osthread();
 724   Monitor* sync = osthread->startThread_lock();
 725 
 726   // non floating stack LinuxThreads needs extra check, see above
 727   if (!_thread_safety_check(thread)) {
 728     // notify parent thread
 729     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 730     osthread->set_state(ZOMBIE);
 731     sync->notify_all();
 732     return NULL;
 733   }
 734 
 735   // thread_id is kernel thread id (similar to Solaris LWP id)
 736   osthread->set_thread_id(os::Linux::gettid());
 737 
 738   if (UseNUMA) {
 739     int lgrp_id = os::numa_get_group_id();
 740     if (lgrp_id != -1) {
 741       thread->set_lgrp_id(lgrp_id);
 742     }
 743   }
 744   // initialize signal mask for this thread
 745   os::Linux::hotspot_sigmask(thread);
 746 
 747   // initialize floating point control register
 748   os::Linux::init_thread_fpu_state();
 749 
 750   // handshaking with parent thread
 751   {
 752     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 753 
 754     // notify parent thread
 755     osthread->set_state(INITIALIZED);
 756     sync->notify_all();
 757 
 758     // wait until os::start_thread()
 759     while (osthread->get_state() == INITIALIZED) {
 760       sync->wait(Mutex::_no_safepoint_check_flag);
 761     }
 762   }
 763 
 764   // call one more level start routine
 765   thread->run();
 766 
 767   return 0;
 768 }
 769 
 770 bool os::create_thread(Thread* thread, ThreadType thr_type,
 771                        size_t stack_size) {
 772   assert(thread->osthread() == NULL, "caller responsible");
 773 
 774   // Allocate the OSThread object
 775   OSThread* osthread = new OSThread(NULL, NULL);
 776   if (osthread == NULL) {
 777     return false;
 778   }
 779 
 780   // set the correct thread state
 781   osthread->set_thread_type(thr_type);
 782 
 783   // Initial state is ALLOCATED but not INITIALIZED
 784   osthread->set_state(ALLOCATED);
 785 
 786   thread->set_osthread(osthread);
 787 
 788   // init thread attributes
 789   pthread_attr_t attr;
 790   pthread_attr_init(&attr);
 791   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 792 
 793   // stack size
 794   if (os::Linux::supports_variable_stack_size()) {
 795     // calculate stack size if it's not specified by caller
 796     if (stack_size == 0) {
 797       stack_size = os::Linux::default_stack_size(thr_type);
 798 
 799       switch (thr_type) {
 800       case os::java_thread:
 801         // Java threads use ThreadStackSize which default value can be
 802         // changed with the flag -Xss
 803         assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 804         stack_size = JavaThread::stack_size_at_create();
 805         break;
 806       case os::compiler_thread:
 807         if (CompilerThreadStackSize > 0) {
 808           stack_size = (size_t)(CompilerThreadStackSize * K);
 809           break;
 810         } // else fall through:
 811           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 812       case os::vm_thread:
 813       case os::pgc_thread:
 814       case os::cgc_thread:
 815       case os::watcher_thread:
 816         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 817         break;
 818       }
 819     }
 820 
 821     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 822     pthread_attr_setstacksize(&attr, stack_size);
 823   } else {
 824     // let pthread_create() pick the default value.
 825   }
 826 
 827   // glibc guard page
 828   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 829 
 830   ThreadState state;
 831 
 832   {
 833     // Serialize thread creation if we are running with fixed stack LinuxThreads
 834     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
 835     if (lock) {
 836       os::Linux::createThread_lock()->lock_without_safepoint_check();
 837     }
 838 
 839     pthread_t tid;
 840     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 841 
 842     pthread_attr_destroy(&attr);
 843 
 844     if (ret != 0) {
 845       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 846         perror("pthread_create()");
 847       }
 848       // Need to clean up stuff we've allocated so far
 849       thread->set_osthread(NULL);
 850       delete osthread;
 851       if (lock) os::Linux::createThread_lock()->unlock();
 852       return false;
 853     }
 854 
 855     // Store pthread info into the OSThread
 856     osthread->set_pthread_id(tid);
 857 
 858     // Wait until child thread is either initialized or aborted
 859     {
 860       Monitor* sync_with_child = osthread->startThread_lock();
 861       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 862       while ((state = osthread->get_state()) == ALLOCATED) {
 863         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 864       }
 865     }
 866 
 867     if (lock) {
 868       os::Linux::createThread_lock()->unlock();
 869     }
 870   }
 871 
 872   // Aborted due to thread limit being reached
 873   if (state == ZOMBIE) {
 874     thread->set_osthread(NULL);
 875     delete osthread;
 876     return false;
 877   }
 878 
 879   // The thread is returned suspended (in state INITIALIZED),
 880   // and is started higher up in the call chain
 881   assert(state == INITIALIZED, "race condition");
 882   return true;
 883 }
 884 
 885 /////////////////////////////////////////////////////////////////////////////
 886 // attach existing thread
 887 
 888 // bootstrap the main thread
 889 bool os::create_main_thread(JavaThread* thread) {
 890   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 891   return create_attached_thread(thread);
 892 }
 893 
 894 bool os::create_attached_thread(JavaThread* thread) {
 895 #ifdef ASSERT
 896   thread->verify_not_published();
 897 #endif
 898 
 899   // Allocate the OSThread object
 900   OSThread* osthread = new OSThread(NULL, NULL);
 901 
 902   if (osthread == NULL) {
 903     return false;
 904   }
 905 
 906   // Store pthread info into the OSThread
 907   osthread->set_thread_id(os::Linux::gettid());
 908   osthread->set_pthread_id(::pthread_self());
 909 
 910   // initialize floating point control register
 911   os::Linux::init_thread_fpu_state();
 912 
 913   // Initial thread state is RUNNABLE
 914   osthread->set_state(RUNNABLE);
 915 
 916   thread->set_osthread(osthread);
 917 
 918   if (UseNUMA) {
 919     int lgrp_id = os::numa_get_group_id();
 920     if (lgrp_id != -1) {
 921       thread->set_lgrp_id(lgrp_id);
 922     }
 923   }
 924 
 925   if (os::Linux::is_initial_thread()) {
 926     // If current thread is initial thread, its stack is mapped on demand,
 927     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 928     // the entire stack region to avoid SEGV in stack banging.
 929     // It is also useful to get around the heap-stack-gap problem on SuSE
 930     // kernel (see 4821821 for details). We first expand stack to the top
 931     // of yellow zone, then enable stack yellow zone (order is significant,
 932     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 933     // is no gap between the last two virtual memory regions.
 934 
 935     JavaThread *jt = (JavaThread *)thread;
 936     address addr = jt->stack_yellow_zone_base();
 937     assert(addr != NULL, "initialization problem?");
 938     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 939 
 940     osthread->set_expanding_stack();
 941     os::Linux::manually_expand_stack(jt, addr);
 942     osthread->clear_expanding_stack();
 943   }
 944 
 945   // initialize signal mask for this thread
 946   // and save the caller's signal mask
 947   os::Linux::hotspot_sigmask(thread);
 948 
 949   return true;
 950 }
 951 
 952 void os::pd_start_thread(Thread* thread) {
 953   OSThread * osthread = thread->osthread();
 954   assert(osthread->get_state() != INITIALIZED, "just checking");
 955   Monitor* sync_with_child = osthread->startThread_lock();
 956   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 957   sync_with_child->notify();
 958 }
 959 
 960 // Free Linux resources related to the OSThread
 961 void os::free_thread(OSThread* osthread) {
 962   assert(osthread != NULL, "osthread not set");
 963 
 964   if (Thread::current()->osthread() == osthread) {
 965     // Restore caller's signal mask
 966     sigset_t sigmask = osthread->caller_sigmask();
 967     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 968   }
 969 
 970   delete osthread;
 971 }
 972 
 973 //////////////////////////////////////////////////////////////////////////////
 974 // thread local storage
 975 
 976 // Restore the thread pointer if the destructor is called. This is in case
 977 // someone from JNI code sets up a destructor with pthread_key_create to run
 978 // detachCurrentThread on thread death. Unless we restore the thread pointer we
 979 // will hang or crash. When detachCurrentThread is called the key will be set
 980 // to null and we will not be called again. If detachCurrentThread is never
 981 // called we could loop forever depending on the pthread implementation.
 982 static void restore_thread_pointer(void* p) {
 983   Thread* thread = (Thread*) p;
 984   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
 985 }
 986 
 987 int os::allocate_thread_local_storage() {
 988   pthread_key_t key;
 989   int rslt = pthread_key_create(&key, restore_thread_pointer);
 990   assert(rslt == 0, "cannot allocate thread local storage");
 991   return (int)key;
 992 }
 993 
 994 // Note: This is currently not used by VM, as we don't destroy TLS key
 995 // on VM exit.
 996 void os::free_thread_local_storage(int index) {
 997   int rslt = pthread_key_delete((pthread_key_t)index);
 998   assert(rslt == 0, "invalid index");
 999 }
1000 
1001 void os::thread_local_storage_at_put(int index, void* value) {
1002   int rslt = pthread_setspecific((pthread_key_t)index, value);
1003   assert(rslt == 0, "pthread_setspecific failed");
1004 }
1005 
1006 extern "C" Thread* get_thread() {
1007   return ThreadLocalStorage::thread();
1008 }
1009 
1010 //////////////////////////////////////////////////////////////////////////////
1011 // initial thread
1012 
1013 // Check if current thread is the initial thread, similar to Solaris thr_main.
1014 bool os::Linux::is_initial_thread(void) {
1015   char dummy;
1016   // If called before init complete, thread stack bottom will be null.
1017   // Can be called if fatal error occurs before initialization.
1018   if (initial_thread_stack_bottom() == NULL) return false;
1019   assert(initial_thread_stack_bottom() != NULL &&
1020          initial_thread_stack_size()   != 0,
1021          "os::init did not locate initial thread's stack region");
1022   if ((address)&dummy >= initial_thread_stack_bottom() &&
1023       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1024     return true;
1025   } else {
1026     return false;
1027   }
1028 }
1029 
1030 // Find the virtual memory area that contains addr
1031 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1032   FILE *fp = fopen("/proc/self/maps", "r");
1033   if (fp) {
1034     address low, high;
1035     while (!feof(fp)) {
1036       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1037         if (low <= addr && addr < high) {
1038           if (vma_low)  *vma_low  = low;
1039           if (vma_high) *vma_high = high;
1040           fclose(fp);
1041           return true;
1042         }
1043       }
1044       for (;;) {
1045         int ch = fgetc(fp);
1046         if (ch == EOF || ch == (int)'\n') break;
1047       }
1048     }
1049     fclose(fp);
1050   }
1051   return false;
1052 }
1053 
1054 // Locate initial thread stack. This special handling of initial thread stack
1055 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1056 // bogus value for initial thread.
1057 void os::Linux::capture_initial_stack(size_t max_size) {
1058   // stack size is the easy part, get it from RLIMIT_STACK
1059   size_t stack_size;
1060   struct rlimit rlim;
1061   getrlimit(RLIMIT_STACK, &rlim);
1062   stack_size = rlim.rlim_cur;
1063 
1064   // 6308388: a bug in ld.so will relocate its own .data section to the
1065   //   lower end of primordial stack; reduce ulimit -s value a little bit
1066   //   so we won't install guard page on ld.so's data section.
1067   stack_size -= 2 * page_size();
1068 
1069   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1070   //   7.1, in both cases we will get 2G in return value.
1071   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1072   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1073   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1074   //   in case other parts in glibc still assumes 2M max stack size.
1075   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1076   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1077   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1078     stack_size = 2 * K * K IA64_ONLY(*2);
1079   }
1080   // Try to figure out where the stack base (top) is. This is harder.
1081   //
1082   // When an application is started, glibc saves the initial stack pointer in
1083   // a global variable "__libc_stack_end", which is then used by system
1084   // libraries. __libc_stack_end should be pretty close to stack top. The
1085   // variable is available since the very early days. However, because it is
1086   // a private interface, it could disappear in the future.
1087   //
1088   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1089   // to __libc_stack_end, it is very close to stack top, but isn't the real
1090   // stack top. Note that /proc may not exist if VM is running as a chroot
1091   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1092   // /proc/<pid>/stat could change in the future (though unlikely).
1093   //
1094   // We try __libc_stack_end first. If that doesn't work, look for
1095   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1096   // as a hint, which should work well in most cases.
1097 
1098   uintptr_t stack_start;
1099 
1100   // try __libc_stack_end first
1101   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1102   if (p && *p) {
1103     stack_start = *p;
1104   } else {
1105     // see if we can get the start_stack field from /proc/self/stat
1106     FILE *fp;
1107     int pid;
1108     char state;
1109     int ppid;
1110     int pgrp;
1111     int session;
1112     int nr;
1113     int tpgrp;
1114     unsigned long flags;
1115     unsigned long minflt;
1116     unsigned long cminflt;
1117     unsigned long majflt;
1118     unsigned long cmajflt;
1119     unsigned long utime;
1120     unsigned long stime;
1121     long cutime;
1122     long cstime;
1123     long prio;
1124     long nice;
1125     long junk;
1126     long it_real;
1127     uintptr_t start;
1128     uintptr_t vsize;
1129     intptr_t rss;
1130     uintptr_t rsslim;
1131     uintptr_t scodes;
1132     uintptr_t ecode;
1133     int i;
1134 
1135     // Figure what the primordial thread stack base is. Code is inspired
1136     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1137     // followed by command name surrounded by parentheses, state, etc.
1138     char stat[2048];
1139     int statlen;
1140 
1141     fp = fopen("/proc/self/stat", "r");
1142     if (fp) {
1143       statlen = fread(stat, 1, 2047, fp);
1144       stat[statlen] = '\0';
1145       fclose(fp);
1146 
1147       // Skip pid and the command string. Note that we could be dealing with
1148       // weird command names, e.g. user could decide to rename java launcher
1149       // to "java 1.4.2 :)", then the stat file would look like
1150       //                1234 (java 1.4.2 :)) R ... ...
1151       // We don't really need to know the command string, just find the last
1152       // occurrence of ")" and then start parsing from there. See bug 4726580.
1153       char * s = strrchr(stat, ')');
1154 
1155       i = 0;
1156       if (s) {
1157         // Skip blank chars
1158         do { s++; } while (s && isspace(*s));
1159 
1160 #define _UFM UINTX_FORMAT
1161 #define _DFM INTX_FORMAT
1162 
1163         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1164         //              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
1165         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,
1166                    &state,          // 3  %c
1167                    &ppid,           // 4  %d
1168                    &pgrp,           // 5  %d
1169                    &session,        // 6  %d
1170                    &nr,             // 7  %d
1171                    &tpgrp,          // 8  %d
1172                    &flags,          // 9  %lu
1173                    &minflt,         // 10 %lu
1174                    &cminflt,        // 11 %lu
1175                    &majflt,         // 12 %lu
1176                    &cmajflt,        // 13 %lu
1177                    &utime,          // 14 %lu
1178                    &stime,          // 15 %lu
1179                    &cutime,         // 16 %ld
1180                    &cstime,         // 17 %ld
1181                    &prio,           // 18 %ld
1182                    &nice,           // 19 %ld
1183                    &junk,           // 20 %ld
1184                    &it_real,        // 21 %ld
1185                    &start,          // 22 UINTX_FORMAT
1186                    &vsize,          // 23 UINTX_FORMAT
1187                    &rss,            // 24 INTX_FORMAT
1188                    &rsslim,         // 25 UINTX_FORMAT
1189                    &scodes,         // 26 UINTX_FORMAT
1190                    &ecode,          // 27 UINTX_FORMAT
1191                    &stack_start);   // 28 UINTX_FORMAT
1192       }
1193 
1194 #undef _UFM
1195 #undef _DFM
1196 
1197       if (i != 28 - 2) {
1198         assert(false, "Bad conversion from /proc/self/stat");
1199         // product mode - assume we are the initial thread, good luck in the
1200         // embedded case.
1201         warning("Can't detect initial thread stack location - bad conversion");
1202         stack_start = (uintptr_t) &rlim;
1203       }
1204     } else {
1205       // For some reason we can't open /proc/self/stat (for example, running on
1206       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1207       // most cases, so don't abort:
1208       warning("Can't detect initial thread stack location - no /proc/self/stat");
1209       stack_start = (uintptr_t) &rlim;
1210     }
1211   }
1212 
1213   // Now we have a pointer (stack_start) very close to the stack top, the
1214   // next thing to do is to figure out the exact location of stack top. We
1215   // can find out the virtual memory area that contains stack_start by
1216   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1217   // and its upper limit is the real stack top. (again, this would fail if
1218   // running inside chroot, because /proc may not exist.)
1219 
1220   uintptr_t stack_top;
1221   address low, high;
1222   if (find_vma((address)stack_start, &low, &high)) {
1223     // success, "high" is the true stack top. (ignore "low", because initial
1224     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1225     stack_top = (uintptr_t)high;
1226   } else {
1227     // failed, likely because /proc/self/maps does not exist
1228     warning("Can't detect initial thread stack location - find_vma failed");
1229     // best effort: stack_start is normally within a few pages below the real
1230     // stack top, use it as stack top, and reduce stack size so we won't put
1231     // guard page outside stack.
1232     stack_top = stack_start;
1233     stack_size -= 16 * page_size();
1234   }
1235 
1236   // stack_top could be partially down the page so align it
1237   stack_top = align_size_up(stack_top, page_size());
1238 
1239   if (max_size && stack_size > max_size) {
1240     _initial_thread_stack_size = max_size;
1241   } else {
1242     _initial_thread_stack_size = stack_size;
1243   }
1244 
1245   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1246   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1247 }
1248 
1249 ////////////////////////////////////////////////////////////////////////////////
1250 // time support
1251 
1252 // Time since start-up in seconds to a fine granularity.
1253 // Used by VMSelfDestructTimer and the MemProfiler.
1254 double os::elapsedTime() {
1255 
1256   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1257 }
1258 
1259 jlong os::elapsed_counter() {
1260   return javaTimeNanos() - initial_time_count;
1261 }
1262 
1263 jlong os::elapsed_frequency() {
1264   return NANOSECS_PER_SEC; // nanosecond resolution
1265 }
1266 
1267 bool os::supports_vtime() { return true; }
1268 bool os::enable_vtime()   { return false; }
1269 bool os::vtime_enabled()  { return false; }
1270 
1271 double os::elapsedVTime() {
1272   struct rusage usage;
1273   int retval = getrusage(RUSAGE_THREAD, &usage);
1274   if (retval == 0) {
1275     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);
1276   } else {
1277     // better than nothing, but not much
1278     return elapsedTime();
1279   }
1280 }
1281 
1282 jlong os::javaTimeMillis() {
1283   timeval time;
1284   int status = gettimeofday(&time, NULL);
1285   assert(status != -1, "linux error");
1286   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1287 }
1288 
1289 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1290   timeval time;
1291   int status = gettimeofday(&time, NULL);
1292   assert(status != -1, "linux error");
1293   seconds = jlong(time.tv_sec);
1294   nanos = jlong(time.tv_usec) * 1000;
1295 }
1296 
1297 
1298 #ifndef CLOCK_MONOTONIC
1299   #define CLOCK_MONOTONIC (1)
1300 #endif
1301 
1302 void os::Linux::clock_init() {
1303   // we do dlopen's in this particular order due to bug in linux
1304   // dynamical loader (see 6348968) leading to crash on exit
1305   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1306   if (handle == NULL) {
1307     handle = dlopen("librt.so", RTLD_LAZY);
1308   }
1309 
1310   if (handle) {
1311     int (*clock_getres_func)(clockid_t, struct timespec*) =
1312            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1313     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1314            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1315     if (clock_getres_func && clock_gettime_func) {
1316       // See if monotonic clock is supported by the kernel. Note that some
1317       // early implementations simply return kernel jiffies (updated every
1318       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1319       // for nano time (though the monotonic property is still nice to have).
1320       // It's fixed in newer kernels, however clock_getres() still returns
1321       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1322       // resolution for now. Hopefully as people move to new kernels, this
1323       // won't be a problem.
1324       struct timespec res;
1325       struct timespec tp;
1326       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1327           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1328         // yes, monotonic clock is supported
1329         _clock_gettime = clock_gettime_func;
1330         return;
1331       } else {
1332         // close librt if there is no monotonic clock
1333         dlclose(handle);
1334       }
1335     }
1336   }
1337   warning("No monotonic clock was available - timed services may " \
1338           "be adversely affected if the time-of-day clock changes");
1339 }
1340 
1341 #ifndef SYS_clock_getres
1342   #if defined(IA32) || defined(AMD64)
1343     #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1344     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1345   #else
1346     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1347     #define sys_clock_getres(x,y)  -1
1348   #endif
1349 #else
1350   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1351 #endif
1352 
1353 void os::Linux::fast_thread_clock_init() {
1354   if (!UseLinuxPosixThreadCPUClocks) {
1355     return;
1356   }
1357   clockid_t clockid;
1358   struct timespec tp;
1359   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1360       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1361 
1362   // Switch to using fast clocks for thread cpu time if
1363   // the sys_clock_getres() returns 0 error code.
1364   // Note, that some kernels may support the current thread
1365   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1366   // returned by the pthread_getcpuclockid().
1367   // If the fast Posix clocks are supported then the sys_clock_getres()
1368   // must return at least tp.tv_sec == 0 which means a resolution
1369   // better than 1 sec. This is extra check for reliability.
1370 
1371   if (pthread_getcpuclockid_func &&
1372       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1373       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1374     _supports_fast_thread_cpu_time = true;
1375     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1376   }
1377 }
1378 
1379 jlong os::javaTimeNanos() {
1380   if (os::supports_monotonic_clock()) {
1381     struct timespec tp;
1382     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1383     assert(status == 0, "gettime error");
1384     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1385     return result;
1386   } else {
1387     timeval time;
1388     int status = gettimeofday(&time, NULL);
1389     assert(status != -1, "linux error");
1390     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1391     return 1000 * usecs;
1392   }
1393 }
1394 
1395 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1396   if (os::supports_monotonic_clock()) {
1397     info_ptr->max_value = ALL_64_BITS;
1398 
1399     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1400     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1401     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1402   } else {
1403     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1404     info_ptr->max_value = ALL_64_BITS;
1405 
1406     // gettimeofday is a real time clock so it skips
1407     info_ptr->may_skip_backward = true;
1408     info_ptr->may_skip_forward = true;
1409   }
1410 
1411   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1412 }
1413 
1414 // Return the real, user, and system times in seconds from an
1415 // arbitrary fixed point in the past.
1416 bool os::getTimesSecs(double* process_real_time,
1417                       double* process_user_time,
1418                       double* process_system_time) {
1419   struct tms ticks;
1420   clock_t real_ticks = times(&ticks);
1421 
1422   if (real_ticks == (clock_t) (-1)) {
1423     return false;
1424   } else {
1425     double ticks_per_second = (double) clock_tics_per_sec;
1426     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1427     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1428     *process_real_time = ((double) real_ticks) / ticks_per_second;
1429 
1430     return true;
1431   }
1432 }
1433 
1434 
1435 char * os::local_time_string(char *buf, size_t buflen) {
1436   struct tm t;
1437   time_t long_time;
1438   time(&long_time);
1439   localtime_r(&long_time, &t);
1440   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1441                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1442                t.tm_hour, t.tm_min, t.tm_sec);
1443   return buf;
1444 }
1445 
1446 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1447   return localtime_r(clock, res);
1448 }
1449 
1450 ////////////////////////////////////////////////////////////////////////////////
1451 // runtime exit support
1452 
1453 // Note: os::shutdown() might be called very early during initialization, or
1454 // called from signal handler. Before adding something to os::shutdown(), make
1455 // sure it is async-safe and can handle partially initialized VM.
1456 void os::shutdown() {
1457 
1458   // allow PerfMemory to attempt cleanup of any persistent resources
1459   perfMemory_exit();
1460 
1461   // needs to remove object in file system
1462   AttachListener::abort();
1463 
1464   // flush buffered output, finish log files
1465   ostream_abort();
1466 
1467   // Check for abort hook
1468   abort_hook_t abort_hook = Arguments::abort_hook();
1469   if (abort_hook != NULL) {
1470     abort_hook();
1471   }
1472 
1473 }
1474 
1475 // Note: os::abort() might be called very early during initialization, or
1476 // called from signal handler. Before adding something to os::abort(), make
1477 // sure it is async-safe and can handle partially initialized VM.
1478 void os::abort(bool dump_core) {
1479   os::shutdown();
1480   if (dump_core) {
1481 #ifndef PRODUCT
1482     fdStream out(defaultStream::output_fd());
1483     out.print_raw("Current thread is ");
1484     char buf[16];
1485     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1486     out.print_raw_cr(buf);
1487     out.print_raw_cr("Dumping core ...");
1488 #endif
1489     ::abort(); // dump core
1490   }
1491 
1492   ::exit(1);
1493 }
1494 
1495 // Die immediately, no exit hook, no abort hook, no cleanup.
1496 void os::die() {
1497   // _exit() on LinuxThreads only kills current thread
1498   ::abort();
1499 }
1500 
1501 
1502 // This method is a copy of JDK's sysGetLastErrorString
1503 // from src/solaris/hpi/src/system_md.c
1504 
1505 size_t os::lasterror(char *buf, size_t len) {
1506   if (errno == 0)  return 0;
1507 
1508   const char *s = ::strerror(errno);
1509   size_t n = ::strlen(s);
1510   if (n >= len) {
1511     n = len - 1;
1512   }
1513   ::strncpy(buf, s, n);
1514   buf[n] = '\0';
1515   return n;
1516 }
1517 
1518 intx os::current_thread_id() { return (intx)pthread_self(); }
1519 int os::current_process_id() {
1520 
1521   // Under the old linux thread library, linux gives each thread
1522   // its own process id. Because of this each thread will return
1523   // a different pid if this method were to return the result
1524   // of getpid(2). Linux provides no api that returns the pid
1525   // of the launcher thread for the vm. This implementation
1526   // returns a unique pid, the pid of the launcher thread
1527   // that starts the vm 'process'.
1528 
1529   // Under the NPTL, getpid() returns the same pid as the
1530   // launcher thread rather than a unique pid per thread.
1531   // Use gettid() if you want the old pre NPTL behaviour.
1532 
1533   // if you are looking for the result of a call to getpid() that
1534   // returns a unique pid for the calling thread, then look at the
1535   // OSThread::thread_id() method in osThread_linux.hpp file
1536 
1537   return (int)(_initial_pid ? _initial_pid : getpid());
1538 }
1539 
1540 // DLL functions
1541 
1542 const char* os::dll_file_extension() { return ".so"; }
1543 
1544 // This must be hard coded because it's the system's temporary
1545 // directory not the java application's temp directory, ala java.io.tmpdir.
1546 const char* os::get_temp_directory() { return "/tmp"; }
1547 
1548 static bool file_exists(const char* filename) {
1549   struct stat statbuf;
1550   if (filename == NULL || strlen(filename) == 0) {
1551     return false;
1552   }
1553   return os::stat(filename, &statbuf) == 0;
1554 }
1555 
1556 bool os::dll_build_name(char* buffer, size_t buflen,
1557                         const char* pname, const char* fname) {
1558   bool retval = false;
1559   // Copied from libhpi
1560   const size_t pnamelen = pname ? strlen(pname) : 0;
1561 
1562   // Return error on buffer overflow.
1563   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1564     return retval;
1565   }
1566 
1567   if (pnamelen == 0) {
1568     snprintf(buffer, buflen, "lib%s.so", fname);
1569     retval = true;
1570   } else if (strchr(pname, *os::path_separator()) != NULL) {
1571     int n;
1572     char** pelements = split_path(pname, &n);
1573     if (pelements == NULL) {
1574       return false;
1575     }
1576     for (int i = 0; i < n; i++) {
1577       // Really shouldn't be NULL, but check can't hurt
1578       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1579         continue; // skip the empty path values
1580       }
1581       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1582       if (file_exists(buffer)) {
1583         retval = true;
1584         break;
1585       }
1586     }
1587     // release the storage
1588     for (int i = 0; i < n; i++) {
1589       if (pelements[i] != NULL) {
1590         FREE_C_HEAP_ARRAY(char, pelements[i]);
1591       }
1592     }
1593     if (pelements != NULL) {
1594       FREE_C_HEAP_ARRAY(char*, pelements);
1595     }
1596   } else {
1597     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1598     retval = true;
1599   }
1600   return retval;
1601 }
1602 
1603 // check if addr is inside libjvm.so
1604 bool os::address_is_in_vm(address addr) {
1605   static address libjvm_base_addr;
1606   Dl_info dlinfo;
1607 
1608   if (libjvm_base_addr == NULL) {
1609     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1610       libjvm_base_addr = (address)dlinfo.dli_fbase;
1611     }
1612     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1613   }
1614 
1615   if (dladdr((void *)addr, &dlinfo) != 0) {
1616     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1617   }
1618 
1619   return false;
1620 }
1621 
1622 bool os::dll_address_to_function_name(address addr, char *buf,
1623                                       int buflen, int *offset) {
1624   // buf is not optional, but offset is optional
1625   assert(buf != NULL, "sanity check");
1626 
1627   Dl_info dlinfo;
1628 
1629   if (dladdr((void*)addr, &dlinfo) != 0) {
1630     // see if we have a matching symbol
1631     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1632       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1633         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1634       }
1635       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1636       return true;
1637     }
1638     // no matching symbol so try for just file info
1639     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1640       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1641                           buf, buflen, offset, dlinfo.dli_fname)) {
1642         return true;
1643       }
1644     }
1645   }
1646 
1647   buf[0] = '\0';
1648   if (offset != NULL) *offset = -1;
1649   return false;
1650 }
1651 
1652 struct _address_to_library_name {
1653   address addr;          // input : memory address
1654   size_t  buflen;        //         size of fname
1655   char*   fname;         // output: library name
1656   address base;          //         library base addr
1657 };
1658 
1659 static int address_to_library_name_callback(struct dl_phdr_info *info,
1660                                             size_t size, void *data) {
1661   int i;
1662   bool found = false;
1663   address libbase = NULL;
1664   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1665 
1666   // iterate through all loadable segments
1667   for (i = 0; i < info->dlpi_phnum; i++) {
1668     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1669     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1670       // base address of a library is the lowest address of its loaded
1671       // segments.
1672       if (libbase == NULL || libbase > segbase) {
1673         libbase = segbase;
1674       }
1675       // see if 'addr' is within current segment
1676       if (segbase <= d->addr &&
1677           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1678         found = true;
1679       }
1680     }
1681   }
1682 
1683   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1684   // so dll_address_to_library_name() can fall through to use dladdr() which
1685   // can figure out executable name from argv[0].
1686   if (found && info->dlpi_name && info->dlpi_name[0]) {
1687     d->base = libbase;
1688     if (d->fname) {
1689       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1690     }
1691     return 1;
1692   }
1693   return 0;
1694 }
1695 
1696 bool os::dll_address_to_library_name(address addr, char* buf,
1697                                      int buflen, int* offset) {
1698   // buf is not optional, but offset is optional
1699   assert(buf != NULL, "sanity check");
1700 
1701   Dl_info dlinfo;
1702   struct _address_to_library_name data;
1703 
1704   // There is a bug in old glibc dladdr() implementation that it could resolve
1705   // to wrong library name if the .so file has a base address != NULL. Here
1706   // we iterate through the program headers of all loaded libraries to find
1707   // out which library 'addr' really belongs to. This workaround can be
1708   // removed once the minimum requirement for glibc is moved to 2.3.x.
1709   data.addr = addr;
1710   data.fname = buf;
1711   data.buflen = buflen;
1712   data.base = NULL;
1713   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1714 
1715   if (rslt) {
1716     // buf already contains library name
1717     if (offset) *offset = addr - data.base;
1718     return true;
1719   }
1720   if (dladdr((void*)addr, &dlinfo) != 0) {
1721     if (dlinfo.dli_fname != NULL) {
1722       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1723     }
1724     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1725       *offset = addr - (address)dlinfo.dli_fbase;
1726     }
1727     return true;
1728   }
1729 
1730   buf[0] = '\0';
1731   if (offset) *offset = -1;
1732   return false;
1733 }
1734 
1735 // Loads .dll/.so and
1736 // in case of error it checks if .dll/.so was built for the
1737 // same architecture as Hotspot is running on
1738 
1739 
1740 // Remember the stack's state. The Linux dynamic linker will change
1741 // the stack to 'executable' at most once, so we must safepoint only once.
1742 bool os::Linux::_stack_is_executable = false;
1743 
1744 // VM operation that loads a library.  This is necessary if stack protection
1745 // of the Java stacks can be lost during loading the library.  If we
1746 // do not stop the Java threads, they can stack overflow before the stacks
1747 // are protected again.
1748 class VM_LinuxDllLoad: public VM_Operation {
1749  private:
1750   const char *_filename;
1751   char *_ebuf;
1752   int _ebuflen;
1753   void *_lib;
1754  public:
1755   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1756     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1757   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1758   void doit() {
1759     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1760     os::Linux::_stack_is_executable = true;
1761   }
1762   void* loaded_library() { return _lib; }
1763 };
1764 
1765 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1766   void * result = NULL;
1767   bool load_attempted = false;
1768 
1769   // Check whether the library to load might change execution rights
1770   // of the stack. If they are changed, the protection of the stack
1771   // guard pages will be lost. We need a safepoint to fix this.
1772   //
1773   // See Linux man page execstack(8) for more info.
1774   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1775     ElfFile ef(filename);
1776     if (!ef.specifies_noexecstack()) {
1777       if (!is_init_completed()) {
1778         os::Linux::_stack_is_executable = true;
1779         // This is OK - No Java threads have been created yet, and hence no
1780         // stack guard pages to fix.
1781         //
1782         // This should happen only when you are building JDK7 using a very
1783         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1784         //
1785         // Dynamic loader will make all stacks executable after
1786         // this function returns, and will not do that again.
1787         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1788       } else {
1789         warning("You have loaded library %s which might have disabled stack guard. "
1790                 "The VM will try to fix the stack guard now.\n"
1791                 "It's highly recommended that you fix the library with "
1792                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1793                 filename);
1794 
1795         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1796         JavaThread *jt = JavaThread::current();
1797         if (jt->thread_state() != _thread_in_native) {
1798           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1799           // that requires ExecStack. Cannot enter safe point. Let's give up.
1800           warning("Unable to fix stack guard. Giving up.");
1801         } else {
1802           if (!LoadExecStackDllInVMThread) {
1803             // This is for the case where the DLL has an static
1804             // constructor function that executes JNI code. We cannot
1805             // load such DLLs in the VMThread.
1806             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1807           }
1808 
1809           ThreadInVMfromNative tiv(jt);
1810           debug_only(VMNativeEntryWrapper vew;)
1811 
1812           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1813           VMThread::execute(&op);
1814           if (LoadExecStackDllInVMThread) {
1815             result = op.loaded_library();
1816           }
1817           load_attempted = true;
1818         }
1819       }
1820     }
1821   }
1822 
1823   if (!load_attempted) {
1824     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1825   }
1826 
1827   if (result != NULL) {
1828     // Successful loading
1829     return result;
1830   }
1831 
1832   Elf32_Ehdr elf_head;
1833   int diag_msg_max_length=ebuflen-strlen(ebuf);
1834   char* diag_msg_buf=ebuf+strlen(ebuf);
1835 
1836   if (diag_msg_max_length==0) {
1837     // No more space in ebuf for additional diagnostics message
1838     return NULL;
1839   }
1840 
1841 
1842   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1843 
1844   if (file_descriptor < 0) {
1845     // Can't open library, report dlerror() message
1846     return NULL;
1847   }
1848 
1849   bool failed_to_read_elf_head=
1850     (sizeof(elf_head)!=
1851      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1852 
1853   ::close(file_descriptor);
1854   if (failed_to_read_elf_head) {
1855     // file i/o error - report dlerror() msg
1856     return NULL;
1857   }
1858 
1859   typedef struct {
1860     Elf32_Half  code;         // Actual value as defined in elf.h
1861     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1862     char        elf_class;    // 32 or 64 bit
1863     char        endianess;    // MSB or LSB
1864     char*       name;         // String representation
1865   } arch_t;
1866 
1867 #ifndef EM_486
1868   #define EM_486          6               /* Intel 80486 */
1869 #endif
1870 #ifndef EM_AARCH64
1871   #define EM_AARCH64    183               /* ARM AARCH64 */
1872 #endif
1873 
1874   static const arch_t arch_array[]={
1875     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1876     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1877     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1878     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1879     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1880     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1881     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1882     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1883 #if defined(VM_LITTLE_ENDIAN)
1884     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1885 #else
1886     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1887 #endif
1888     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1889     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1890     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1891     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1892     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1893     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1894     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1895     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1896   };
1897 
1898 #if  (defined IA32)
1899   static  Elf32_Half running_arch_code=EM_386;
1900 #elif   (defined AMD64)
1901   static  Elf32_Half running_arch_code=EM_X86_64;
1902 #elif  (defined IA64)
1903   static  Elf32_Half running_arch_code=EM_IA_64;
1904 #elif  (defined __sparc) && (defined _LP64)
1905   static  Elf32_Half running_arch_code=EM_SPARCV9;
1906 #elif  (defined __sparc) && (!defined _LP64)
1907   static  Elf32_Half running_arch_code=EM_SPARC;
1908 #elif  (defined __powerpc64__)
1909   static  Elf32_Half running_arch_code=EM_PPC64;
1910 #elif  (defined __powerpc__)
1911   static  Elf32_Half running_arch_code=EM_PPC;
1912 #elif  (defined ARM)
1913   static  Elf32_Half running_arch_code=EM_ARM;
1914 #elif  (defined S390)
1915   static  Elf32_Half running_arch_code=EM_S390;
1916 #elif  (defined ALPHA)
1917   static  Elf32_Half running_arch_code=EM_ALPHA;
1918 #elif  (defined MIPSEL)
1919   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1920 #elif  (defined PARISC)
1921   static  Elf32_Half running_arch_code=EM_PARISC;
1922 #elif  (defined MIPS)
1923   static  Elf32_Half running_arch_code=EM_MIPS;
1924 #elif  (defined M68K)
1925   static  Elf32_Half running_arch_code=EM_68K;
1926 #elif  (defined AARCH64)
1927   static  Elf32_Half running_arch_code=EM_AARCH64;
1928 #else
1929     #error Method os::dll_load requires that one of following is defined:\
1930          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1931 #endif
1932 
1933   // Identify compatability class for VM's architecture and library's architecture
1934   // Obtain string descriptions for architectures
1935 
1936   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1937   int running_arch_index=-1;
1938 
1939   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1940     if (running_arch_code == arch_array[i].code) {
1941       running_arch_index    = i;
1942     }
1943     if (lib_arch.code == arch_array[i].code) {
1944       lib_arch.compat_class = arch_array[i].compat_class;
1945       lib_arch.name         = arch_array[i].name;
1946     }
1947   }
1948 
1949   assert(running_arch_index != -1,
1950          "Didn't find running architecture code (running_arch_code) in arch_array");
1951   if (running_arch_index == -1) {
1952     // Even though running architecture detection failed
1953     // we may still continue with reporting dlerror() message
1954     return NULL;
1955   }
1956 
1957   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1958     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1959     return NULL;
1960   }
1961 
1962 #ifndef S390
1963   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1964     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1965     return NULL;
1966   }
1967 #endif // !S390
1968 
1969   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1970     if (lib_arch.name!=NULL) {
1971       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1972                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1973                  lib_arch.name, arch_array[running_arch_index].name);
1974     } else {
1975       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1976                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1977                  lib_arch.code,
1978                  arch_array[running_arch_index].name);
1979     }
1980   }
1981 
1982   return NULL;
1983 }
1984 
1985 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1986                                 int ebuflen) {
1987   void * result = ::dlopen(filename, RTLD_LAZY);
1988   if (result == NULL) {
1989     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1990     ebuf[ebuflen-1] = '\0';
1991   }
1992   return result;
1993 }
1994 
1995 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1996                                        int ebuflen) {
1997   void * result = NULL;
1998   if (LoadExecStackDllInVMThread) {
1999     result = dlopen_helper(filename, ebuf, ebuflen);
2000   }
2001 
2002   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2003   // library that requires an executable stack, or which does not have this
2004   // stack attribute set, dlopen changes the stack attribute to executable. The
2005   // read protection of the guard pages gets lost.
2006   //
2007   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2008   // may have been queued at the same time.
2009 
2010   if (!_stack_is_executable) {
2011     JavaThread *jt = Threads::first();
2012 
2013     while (jt) {
2014       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
2015           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
2016         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2017                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2018           warning("Attempt to reguard stack yellow zone failed.");
2019         }
2020       }
2021       jt = jt->next();
2022     }
2023   }
2024 
2025   return result;
2026 }
2027 
2028 // glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
2029 // chances are you might want to run the generated bits against glibc-2.0
2030 // libdl.so, so always use locking for any version of glibc.
2031 //
2032 void* os::dll_lookup(void* handle, const char* name) {
2033   pthread_mutex_lock(&dl_mutex);
2034   void* res = dlsym(handle, name);
2035   pthread_mutex_unlock(&dl_mutex);
2036   return res;
2037 }
2038 
2039 void* os::get_default_process_handle() {
2040   return (void*)::dlopen(NULL, RTLD_LAZY);
2041 }
2042 
2043 static bool _print_ascii_file(const char* filename, outputStream* st) {
2044   int fd = ::open(filename, O_RDONLY);
2045   if (fd == -1) {
2046     return false;
2047   }
2048 
2049   char buf[32];
2050   int bytes;
2051   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2052     st->print_raw(buf, bytes);
2053   }
2054 
2055   ::close(fd);
2056 
2057   return true;
2058 }
2059 
2060 void os::print_dll_info(outputStream *st) {
2061   st->print_cr("Dynamic libraries:");
2062 
2063   char fname[32];
2064   pid_t pid = os::Linux::gettid();
2065 
2066   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2067 
2068   if (!_print_ascii_file(fname, st)) {
2069     st->print("Can not get library information for pid = %d\n", pid);
2070   }
2071 }
2072 
2073 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2074   FILE *procmapsFile = NULL;
2075 
2076   // Open the procfs maps file for the current process
2077   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2078     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2079     char line[PATH_MAX + 100];
2080 
2081     // Read line by line from 'file'
2082     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2083       u8 base, top, offset, inode;
2084       char permissions[5];
2085       char device[6];
2086       char name[PATH_MAX + 1];
2087 
2088       // Parse fields from line
2089       sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2090 
2091       // Filter by device id '00:00' so that we only get file system mapped files.
2092       if (strcmp(device, "00:00") != 0) {
2093 
2094         // Call callback with the fields of interest
2095         if(callback(name, (address)base, (address)top, param)) {
2096           // Oops abort, callback aborted
2097           fclose(procmapsFile);
2098           return 1;
2099         }
2100       }
2101     }
2102     fclose(procmapsFile);
2103   }
2104   return 0;
2105 }
2106 
2107 void os::print_os_info_brief(outputStream* st) {
2108   os::Linux::print_distro_info(st);
2109 
2110   os::Posix::print_uname_info(st);
2111 
2112   os::Linux::print_libversion_info(st);
2113 
2114 }
2115 
2116 void os::print_os_info(outputStream* st) {
2117   st->print("OS:");
2118 
2119   os::Linux::print_distro_info(st);
2120 
2121   os::Posix::print_uname_info(st);
2122 
2123   // Print warning if unsafe chroot environment detected
2124   if (unsafe_chroot_detected) {
2125     st->print("WARNING!! ");
2126     st->print_cr("%s", unstable_chroot_error);
2127   }
2128 
2129   os::Linux::print_libversion_info(st);
2130 
2131   os::Posix::print_rlimit_info(st);
2132 
2133   os::Posix::print_load_average(st);
2134 
2135   os::Linux::print_full_memory_info(st);
2136 }
2137 
2138 // Try to identify popular distros.
2139 // Most Linux distributions have a /etc/XXX-release file, which contains
2140 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2141 // file that also contains the OS version string. Some have more than one
2142 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2143 // /etc/redhat-release.), so the order is important.
2144 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2145 // their own specific XXX-release file as well as a redhat-release file.
2146 // Because of this the XXX-release file needs to be searched for before the
2147 // redhat-release file.
2148 // Since Red Hat has a lsb-release file that is not very descriptive the
2149 // search for redhat-release needs to be before lsb-release.
2150 // Since the lsb-release file is the new standard it needs to be searched
2151 // before the older style release files.
2152 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2153 // next to last resort.  The os-release file is a new standard that contains
2154 // distribution information and the system-release file seems to be an old
2155 // standard that has been replaced by the lsb-release and os-release files.
2156 // Searching for the debian_version file is the last resort.  It contains
2157 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2158 // "Debian " is printed before the contents of the debian_version file.
2159 void os::Linux::print_distro_info(outputStream* st) {
2160   if (!_print_ascii_file("/etc/oracle-release", st) &&
2161       !_print_ascii_file("/etc/mandriva-release", st) &&
2162       !_print_ascii_file("/etc/mandrake-release", st) &&
2163       !_print_ascii_file("/etc/sun-release", st) &&
2164       !_print_ascii_file("/etc/redhat-release", st) &&
2165       !_print_ascii_file("/etc/lsb-release", st) &&
2166       !_print_ascii_file("/etc/SuSE-release", st) &&
2167       !_print_ascii_file("/etc/turbolinux-release", st) &&
2168       !_print_ascii_file("/etc/gentoo-release", st) &&
2169       !_print_ascii_file("/etc/ltib-release", st) &&
2170       !_print_ascii_file("/etc/angstrom-version", st) &&
2171       !_print_ascii_file("/etc/system-release", st) &&
2172       !_print_ascii_file("/etc/os-release", st)) {
2173 
2174     if (file_exists("/etc/debian_version")) {
2175       st->print("Debian ");
2176       _print_ascii_file("/etc/debian_version", st);
2177     } else {
2178       st->print("Linux");
2179     }
2180   }
2181   st->cr();
2182 }
2183 
2184 void os::Linux::print_libversion_info(outputStream* st) {
2185   // libc, pthread
2186   st->print("libc:");
2187   st->print("%s ", os::Linux::glibc_version());
2188   st->print("%s ", os::Linux::libpthread_version());
2189   if (os::Linux::is_LinuxThreads()) {
2190     st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2191   }
2192   st->cr();
2193 }
2194 
2195 void os::Linux::print_full_memory_info(outputStream* st) {
2196   st->print("\n/proc/meminfo:\n");
2197   _print_ascii_file("/proc/meminfo", st);
2198   st->cr();
2199 }
2200 
2201 void os::print_memory_info(outputStream* st) {
2202 
2203   st->print("Memory:");
2204   st->print(" %dk page", os::vm_page_size()>>10);
2205 
2206   // values in struct sysinfo are "unsigned long"
2207   struct sysinfo si;
2208   sysinfo(&si);
2209 
2210   st->print(", physical " UINT64_FORMAT "k",
2211             os::physical_memory() >> 10);
2212   st->print("(" UINT64_FORMAT "k free)",
2213             os::available_memory() >> 10);
2214   st->print(", swap " UINT64_FORMAT "k",
2215             ((jlong)si.totalswap * si.mem_unit) >> 10);
2216   st->print("(" UINT64_FORMAT "k free)",
2217             ((jlong)si.freeswap * si.mem_unit) >> 10);
2218   st->cr();
2219 }
2220 
2221 void os::pd_print_cpu_info(outputStream* st) {
2222   st->print("\n/proc/cpuinfo:\n");
2223   if (!_print_ascii_file("/proc/cpuinfo", st)) {
2224     st->print("  <Not Available>");
2225   }
2226   st->cr();
2227 }
2228 
2229 void os::print_siginfo(outputStream* st, void* siginfo) {
2230   const siginfo_t* si = (const siginfo_t*)siginfo;
2231 
2232   os::Posix::print_siginfo_brief(st, si);
2233 #if INCLUDE_CDS
2234   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2235       UseSharedSpaces) {
2236     FileMapInfo* mapinfo = FileMapInfo::current_info();
2237     if (mapinfo->is_in_shared_space(si->si_addr)) {
2238       st->print("\n\nError accessing class data sharing archive."   \
2239                 " Mapped file inaccessible during execution, "      \
2240                 " possible disk/network problem.");
2241     }
2242   }
2243 #endif
2244   st->cr();
2245 }
2246 
2247 
2248 static void print_signal_handler(outputStream* st, int sig,
2249                                  char* buf, size_t buflen);
2250 
2251 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2252   st->print_cr("Signal Handlers:");
2253   print_signal_handler(st, SIGSEGV, buf, buflen);
2254   print_signal_handler(st, SIGBUS , buf, buflen);
2255   print_signal_handler(st, SIGFPE , buf, buflen);
2256   print_signal_handler(st, SIGPIPE, buf, buflen);
2257   print_signal_handler(st, SIGXFSZ, buf, buflen);
2258   print_signal_handler(st, SIGILL , buf, buflen);
2259   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2260   print_signal_handler(st, SR_signum, buf, buflen);
2261   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2262   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2263   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2264   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2265 #if defined(PPC64)
2266   print_signal_handler(st, SIGTRAP, buf, buflen);
2267 #endif
2268 }
2269 
2270 static char saved_jvm_path[MAXPATHLEN] = {0};
2271 
2272 // Find the full path to the current module, libjvm.so
2273 void os::jvm_path(char *buf, jint buflen) {
2274   // Error checking.
2275   if (buflen < MAXPATHLEN) {
2276     assert(false, "must use a large-enough buffer");
2277     buf[0] = '\0';
2278     return;
2279   }
2280   // Lazy resolve the path to current module.
2281   if (saved_jvm_path[0] != 0) {
2282     strcpy(buf, saved_jvm_path);
2283     return;
2284   }
2285 
2286   char dli_fname[MAXPATHLEN];
2287   bool ret = dll_address_to_library_name(
2288                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2289                                          dli_fname, sizeof(dli_fname), NULL);
2290   assert(ret, "cannot locate libjvm");
2291   char *rp = NULL;
2292   if (ret && dli_fname[0] != '\0') {
2293     rp = realpath(dli_fname, buf);
2294   }
2295   if (rp == NULL) {
2296     return;
2297   }
2298 
2299   if (Arguments::sun_java_launcher_is_altjvm()) {
2300     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2301     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2302     // If "/jre/lib/" appears at the right place in the string, then
2303     // assume we are installed in a JDK and we're done. Otherwise, check
2304     // for a JAVA_HOME environment variable and fix up the path so it
2305     // looks like libjvm.so is installed there (append a fake suffix
2306     // hotspot/libjvm.so).
2307     const char *p = buf + strlen(buf) - 1;
2308     for (int count = 0; p > buf && count < 5; ++count) {
2309       for (--p; p > buf && *p != '/'; --p)
2310         /* empty */ ;
2311     }
2312 
2313     if (strncmp(p, "/jre/lib/", 9) != 0) {
2314       // Look for JAVA_HOME in the environment.
2315       char* java_home_var = ::getenv("JAVA_HOME");
2316       if (java_home_var != NULL && java_home_var[0] != 0) {
2317         char* jrelib_p;
2318         int len;
2319 
2320         // Check the current module name "libjvm.so".
2321         p = strrchr(buf, '/');
2322         if (p == NULL) {
2323           return;
2324         }
2325         assert(strstr(p, "/libjvm") == p, "invalid library name");
2326 
2327         rp = realpath(java_home_var, buf);
2328         if (rp == NULL) {
2329           return;
2330         }
2331 
2332         // determine if this is a legacy image or modules image
2333         // modules image doesn't have "jre" subdirectory
2334         len = strlen(buf);
2335         assert(len < buflen, "Ran out of buffer room");
2336         jrelib_p = buf + len;
2337         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2338         if (0 != access(buf, F_OK)) {
2339           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2340         }
2341 
2342         if (0 == access(buf, F_OK)) {
2343           // Use current module name "libjvm.so"
2344           len = strlen(buf);
2345           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2346         } else {
2347           // Go back to path of .so
2348           rp = realpath(dli_fname, buf);
2349           if (rp == NULL) {
2350             return;
2351           }
2352         }
2353       }
2354     }
2355   }
2356 
2357   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2358   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2359 }
2360 
2361 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2362   // no prefix required, not even "_"
2363 }
2364 
2365 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2366   // no suffix required
2367 }
2368 
2369 ////////////////////////////////////////////////////////////////////////////////
2370 // sun.misc.Signal support
2371 
2372 static volatile jint sigint_count = 0;
2373 
2374 static void UserHandler(int sig, void *siginfo, void *context) {
2375   // 4511530 - sem_post is serialized and handled by the manager thread. When
2376   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2377   // don't want to flood the manager thread with sem_post requests.
2378   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2379     return;
2380   }
2381 
2382   // Ctrl-C is pressed during error reporting, likely because the error
2383   // handler fails to abort. Let VM die immediately.
2384   if (sig == SIGINT && is_error_reported()) {
2385     os::die();
2386   }
2387 
2388   os::signal_notify(sig);
2389 }
2390 
2391 void* os::user_handler() {
2392   return CAST_FROM_FN_PTR(void*, UserHandler);
2393 }
2394 
2395 class Semaphore : public StackObj {
2396  public:
2397   Semaphore();
2398   ~Semaphore();
2399   void signal();
2400   void wait();
2401   bool trywait();
2402   bool timedwait(unsigned int sec, int nsec);
2403  private:
2404   sem_t _semaphore;
2405 };
2406 
2407 Semaphore::Semaphore() {
2408   sem_init(&_semaphore, 0, 0);
2409 }
2410 
2411 Semaphore::~Semaphore() {
2412   sem_destroy(&_semaphore);
2413 }
2414 
2415 void Semaphore::signal() {
2416   sem_post(&_semaphore);
2417 }
2418 
2419 void Semaphore::wait() {
2420   sem_wait(&_semaphore);
2421 }
2422 
2423 bool Semaphore::trywait() {
2424   return sem_trywait(&_semaphore) == 0;
2425 }
2426 
2427 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2428 
2429   struct timespec ts;
2430   // Semaphore's are always associated with CLOCK_REALTIME
2431   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2432   // see unpackTime for discussion on overflow checking
2433   if (sec >= MAX_SECS) {
2434     ts.tv_sec += MAX_SECS;
2435     ts.tv_nsec = 0;
2436   } else {
2437     ts.tv_sec += sec;
2438     ts.tv_nsec += nsec;
2439     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2440       ts.tv_nsec -= NANOSECS_PER_SEC;
2441       ++ts.tv_sec; // note: this must be <= max_secs
2442     }
2443   }
2444 
2445   while (1) {
2446     int result = sem_timedwait(&_semaphore, &ts);
2447     if (result == 0) {
2448       return true;
2449     } else if (errno == EINTR) {
2450       continue;
2451     } else if (errno == ETIMEDOUT) {
2452       return false;
2453     } else {
2454       return false;
2455     }
2456   }
2457 }
2458 
2459 extern "C" {
2460   typedef void (*sa_handler_t)(int);
2461   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2462 }
2463 
2464 void* os::signal(int signal_number, void* handler) {
2465   struct sigaction sigAct, oldSigAct;
2466 
2467   sigfillset(&(sigAct.sa_mask));
2468   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2469   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2470 
2471   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2472     // -1 means registration failed
2473     return (void *)-1;
2474   }
2475 
2476   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2477 }
2478 
2479 void os::signal_raise(int signal_number) {
2480   ::raise(signal_number);
2481 }
2482 
2483 // The following code is moved from os.cpp for making this
2484 // code platform specific, which it is by its very nature.
2485 
2486 // Will be modified when max signal is changed to be dynamic
2487 int os::sigexitnum_pd() {
2488   return NSIG;
2489 }
2490 
2491 // a counter for each possible signal value
2492 static volatile jint pending_signals[NSIG+1] = { 0 };
2493 
2494 // Linux(POSIX) specific hand shaking semaphore.
2495 static sem_t sig_sem;
2496 static Semaphore sr_semaphore;
2497 
2498 void os::signal_init_pd() {
2499   // Initialize signal structures
2500   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2501 
2502   // Initialize signal semaphore
2503   ::sem_init(&sig_sem, 0, 0);
2504 }
2505 
2506 void os::signal_notify(int sig) {
2507   Atomic::inc(&pending_signals[sig]);
2508   ::sem_post(&sig_sem);
2509 }
2510 
2511 static int check_pending_signals(bool wait) {
2512   Atomic::store(0, &sigint_count);
2513   for (;;) {
2514     for (int i = 0; i < NSIG + 1; i++) {
2515       jint n = pending_signals[i];
2516       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2517         return i;
2518       }
2519     }
2520     if (!wait) {
2521       return -1;
2522     }
2523     JavaThread *thread = JavaThread::current();
2524     ThreadBlockInVM tbivm(thread);
2525 
2526     bool threadIsSuspended;
2527     do {
2528       thread->set_suspend_equivalent();
2529       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2530       ::sem_wait(&sig_sem);
2531 
2532       // were we externally suspended while we were waiting?
2533       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2534       if (threadIsSuspended) {
2535         // The semaphore has been incremented, but while we were waiting
2536         // another thread suspended us. We don't want to continue running
2537         // while suspended because that would surprise the thread that
2538         // suspended us.
2539         ::sem_post(&sig_sem);
2540 
2541         thread->java_suspend_self();
2542       }
2543     } while (threadIsSuspended);
2544   }
2545 }
2546 
2547 int os::signal_lookup() {
2548   return check_pending_signals(false);
2549 }
2550 
2551 int os::signal_wait() {
2552   return check_pending_signals(true);
2553 }
2554 
2555 ////////////////////////////////////////////////////////////////////////////////
2556 // Virtual Memory
2557 
2558 int os::vm_page_size() {
2559   // Seems redundant as all get out
2560   assert(os::Linux::page_size() != -1, "must call os::init");
2561   return os::Linux::page_size();
2562 }
2563 
2564 // Solaris allocates memory by pages.
2565 int os::vm_allocation_granularity() {
2566   assert(os::Linux::page_size() != -1, "must call os::init");
2567   return os::Linux::page_size();
2568 }
2569 
2570 // Rationale behind this function:
2571 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2572 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2573 //  samples for JITted code. Here we create private executable mapping over the code cache
2574 //  and then we can use standard (well, almost, as mapping can change) way to provide
2575 //  info for the reporting script by storing timestamp and location of symbol
2576 void linux_wrap_code(char* base, size_t size) {
2577   static volatile jint cnt = 0;
2578 
2579   if (!UseOprofile) {
2580     return;
2581   }
2582 
2583   char buf[PATH_MAX+1];
2584   int num = Atomic::add(1, &cnt);
2585 
2586   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2587            os::get_temp_directory(), os::current_process_id(), num);
2588   unlink(buf);
2589 
2590   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2591 
2592   if (fd != -1) {
2593     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2594     if (rv != (off_t)-1) {
2595       if (::write(fd, "", 1) == 1) {
2596         mmap(base, size,
2597              PROT_READ|PROT_WRITE|PROT_EXEC,
2598              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2599       }
2600     }
2601     ::close(fd);
2602     unlink(buf);
2603   }
2604 }
2605 
2606 static bool recoverable_mmap_error(int err) {
2607   // See if the error is one we can let the caller handle. This
2608   // list of errno values comes from JBS-6843484. I can't find a
2609   // Linux man page that documents this specific set of errno
2610   // values so while this list currently matches Solaris, it may
2611   // change as we gain experience with this failure mode.
2612   switch (err) {
2613   case EBADF:
2614   case EINVAL:
2615   case ENOTSUP:
2616     // let the caller deal with these errors
2617     return true;
2618 
2619   default:
2620     // Any remaining errors on this OS can cause our reserved mapping
2621     // to be lost. That can cause confusion where different data
2622     // structures think they have the same memory mapped. The worst
2623     // scenario is if both the VM and a library think they have the
2624     // same memory mapped.
2625     return false;
2626   }
2627 }
2628 
2629 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2630                                     int err) {
2631   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2632           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2633           strerror(err), err);
2634 }
2635 
2636 static void warn_fail_commit_memory(char* addr, size_t size,
2637                                     size_t alignment_hint, bool exec,
2638                                     int err) {
2639   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2640           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2641           alignment_hint, exec, strerror(err), err);
2642 }
2643 
2644 // NOTE: Linux kernel does not really reserve the pages for us.
2645 //       All it does is to check if there are enough free pages
2646 //       left at the time of mmap(). This could be a potential
2647 //       problem.
2648 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2649   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2650   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2651                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2652   if (res != (uintptr_t) MAP_FAILED) {
2653     if (UseNUMAInterleaving) {
2654       numa_make_global(addr, size);
2655     }
2656     return 0;
2657   }
2658 
2659   int err = errno;  // save errno from mmap() call above
2660 
2661   if (!recoverable_mmap_error(err)) {
2662     warn_fail_commit_memory(addr, size, exec, err);
2663     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2664   }
2665 
2666   return err;
2667 }
2668 
2669 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2670   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2671 }
2672 
2673 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2674                                   const char* mesg) {
2675   assert(mesg != NULL, "mesg must be specified");
2676   int err = os::Linux::commit_memory_impl(addr, size, exec);
2677   if (err != 0) {
2678     // the caller wants all commit errors to exit with the specified mesg:
2679     warn_fail_commit_memory(addr, size, exec, err);
2680     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2681   }
2682 }
2683 
2684 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2685 #ifndef MAP_HUGETLB
2686   #define MAP_HUGETLB 0x40000
2687 #endif
2688 
2689 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2690 #ifndef MADV_HUGEPAGE
2691   #define MADV_HUGEPAGE 14
2692 #endif
2693 
2694 int os::Linux::commit_memory_impl(char* addr, size_t size,
2695                                   size_t alignment_hint, bool exec) {
2696   int err = os::Linux::commit_memory_impl(addr, size, exec);
2697   if (err == 0) {
2698     realign_memory(addr, size, alignment_hint);
2699   }
2700   return err;
2701 }
2702 
2703 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2704                           bool exec) {
2705   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2706 }
2707 
2708 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2709                                   size_t alignment_hint, bool exec,
2710                                   const char* mesg) {
2711   assert(mesg != NULL, "mesg must be specified");
2712   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2713   if (err != 0) {
2714     // the caller wants all commit errors to exit with the specified mesg:
2715     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2716     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2717   }
2718 }
2719 
2720 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2721   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2722     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2723     // be supported or the memory may already be backed by huge pages.
2724     ::madvise(addr, bytes, MADV_HUGEPAGE);
2725   }
2726 }
2727 
2728 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2729   // This method works by doing an mmap over an existing mmaping and effectively discarding
2730   // the existing pages. However it won't work for SHM-based large pages that cannot be
2731   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2732   // small pages on top of the SHM segment. This method always works for small pages, so we
2733   // allow that in any case.
2734   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2735     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2736   }
2737 }
2738 
2739 void os::numa_make_global(char *addr, size_t bytes) {
2740   Linux::numa_interleave_memory(addr, bytes);
2741 }
2742 
2743 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2744 // bind policy to MPOL_PREFERRED for the current thread.
2745 #define USE_MPOL_PREFERRED 0
2746 
2747 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2748   // To make NUMA and large pages more robust when both enabled, we need to ease
2749   // the requirements on where the memory should be allocated. MPOL_BIND is the
2750   // default policy and it will force memory to be allocated on the specified
2751   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2752   // the specified node, but will not force it. Using this policy will prevent
2753   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2754   // free large pages.
2755   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2756   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2757 }
2758 
2759 bool os::numa_topology_changed() { return false; }
2760 
2761 size_t os::numa_get_groups_num() {
2762   int max_node = Linux::numa_max_node();
2763   return max_node > 0 ? max_node + 1 : 1;
2764 }
2765 
2766 int os::numa_get_group_id() {
2767   int cpu_id = Linux::sched_getcpu();
2768   if (cpu_id != -1) {
2769     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2770     if (lgrp_id != -1) {
2771       return lgrp_id;
2772     }
2773   }
2774   return 0;
2775 }
2776 
2777 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2778   for (size_t i = 0; i < size; i++) {
2779     ids[i] = i;
2780   }
2781   return size;
2782 }
2783 
2784 bool os::get_page_info(char *start, page_info* info) {
2785   return false;
2786 }
2787 
2788 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2789                      page_info* page_found) {
2790   return end;
2791 }
2792 
2793 
2794 int os::Linux::sched_getcpu_syscall(void) {
2795   unsigned int cpu;
2796   int retval = -1;
2797 
2798 #if defined(IA32)
2799   #ifndef SYS_getcpu
2800     #define SYS_getcpu 318
2801   #endif
2802   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2803 #elif defined(AMD64)
2804 // Unfortunately we have to bring all these macros here from vsyscall.h
2805 // to be able to compile on old linuxes.
2806   #define __NR_vgetcpu 2
2807   #define VSYSCALL_START (-10UL << 20)
2808   #define VSYSCALL_SIZE 1024
2809   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2810   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2811   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2812   retval = vgetcpu(&cpu, NULL, NULL);
2813 #endif
2814 
2815   return (retval == -1) ? retval : cpu;
2816 }
2817 
2818 // Something to do with the numa-aware allocator needs these symbols
2819 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2820 extern "C" JNIEXPORT void numa_error(char *where) { }
2821 extern "C" JNIEXPORT int fork1() { return fork(); }
2822 
2823 
2824 // If we are running with libnuma version > 2, then we should
2825 // be trying to use symbols with versions 1.1
2826 // If we are running with earlier version, which did not have symbol versions,
2827 // we should use the base version.
2828 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2829   void *f = dlvsym(handle, name, "libnuma_1.1");
2830   if (f == NULL) {
2831     f = dlsym(handle, name);
2832   }
2833   return f;
2834 }
2835 
2836 bool os::Linux::libnuma_init() {
2837   // sched_getcpu() should be in libc.
2838   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2839                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2840 
2841   // If it's not, try a direct syscall.
2842   if (sched_getcpu() == -1) {
2843     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2844                                     (void*)&sched_getcpu_syscall));
2845   }
2846 
2847   if (sched_getcpu() != -1) { // Does it work?
2848     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2849     if (handle != NULL) {
2850       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2851                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2852       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2853                                        libnuma_dlsym(handle, "numa_max_node")));
2854       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2855                                         libnuma_dlsym(handle, "numa_available")));
2856       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2857                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2858       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2859                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2860       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2861                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2862 
2863 
2864       if (numa_available() != -1) {
2865         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2866         // Create a cpu -> node mapping
2867         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2868         rebuild_cpu_to_node_map();
2869         return true;
2870       }
2871     }
2872   }
2873   return false;
2874 }
2875 
2876 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2877 // The table is later used in get_node_by_cpu().
2878 void os::Linux::rebuild_cpu_to_node_map() {
2879   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2880                               // in libnuma (possible values are starting from 16,
2881                               // and continuing up with every other power of 2, but less
2882                               // than the maximum number of CPUs supported by kernel), and
2883                               // is a subject to change (in libnuma version 2 the requirements
2884                               // are more reasonable) we'll just hardcode the number they use
2885                               // in the library.
2886   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2887 
2888   size_t cpu_num = os::active_processor_count();
2889   size_t cpu_map_size = NCPUS / BitsPerCLong;
2890   size_t cpu_map_valid_size =
2891     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2892 
2893   cpu_to_node()->clear();
2894   cpu_to_node()->at_grow(cpu_num - 1);
2895   size_t node_num = numa_get_groups_num();
2896 
2897   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2898   for (size_t i = 0; i < node_num; i++) {
2899     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2900       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2901         if (cpu_map[j] != 0) {
2902           for (size_t k = 0; k < BitsPerCLong; k++) {
2903             if (cpu_map[j] & (1UL << k)) {
2904               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2905             }
2906           }
2907         }
2908       }
2909     }
2910   }
2911   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2912 }
2913 
2914 int os::Linux::get_node_by_cpu(int cpu_id) {
2915   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2916     return cpu_to_node()->at(cpu_id);
2917   }
2918   return -1;
2919 }
2920 
2921 GrowableArray<int>* os::Linux::_cpu_to_node;
2922 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2923 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2924 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2925 os::Linux::numa_available_func_t os::Linux::_numa_available;
2926 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2927 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2928 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2929 unsigned long* os::Linux::_numa_all_nodes;
2930 
2931 bool os::pd_uncommit_memory(char* addr, size_t size) {
2932   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2933                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2934   return res  != (uintptr_t) MAP_FAILED;
2935 }
2936 
2937 static address get_stack_commited_bottom(address bottom, size_t size) {
2938   address nbot = bottom;
2939   address ntop = bottom + size;
2940 
2941   size_t page_sz = os::vm_page_size();
2942   unsigned pages = size / page_sz;
2943 
2944   unsigned char vec[1];
2945   unsigned imin = 1, imax = pages + 1, imid;
2946   int mincore_return_value = 0;
2947 
2948   assert(imin <= imax, "Unexpected page size");
2949 
2950   while (imin < imax) {
2951     imid = (imax + imin) / 2;
2952     nbot = ntop - (imid * page_sz);
2953 
2954     // Use a trick with mincore to check whether the page is mapped or not.
2955     // mincore sets vec to 1 if page resides in memory and to 0 if page
2956     // is swapped output but if page we are asking for is unmapped
2957     // it returns -1,ENOMEM
2958     mincore_return_value = mincore(nbot, page_sz, vec);
2959 
2960     if (mincore_return_value == -1) {
2961       // Page is not mapped go up
2962       // to find first mapped page
2963       if (errno != EAGAIN) {
2964         assert(errno == ENOMEM, "Unexpected mincore errno");
2965         imax = imid;
2966       }
2967     } else {
2968       // Page is mapped go down
2969       // to find first not mapped page
2970       imin = imid + 1;
2971     }
2972   }
2973 
2974   nbot = nbot + page_sz;
2975 
2976   // Adjust stack bottom one page up if last checked page is not mapped
2977   if (mincore_return_value == -1) {
2978     nbot = nbot + page_sz;
2979   }
2980 
2981   return nbot;
2982 }
2983 
2984 
2985 // Linux uses a growable mapping for the stack, and if the mapping for
2986 // the stack guard pages is not removed when we detach a thread the
2987 // stack cannot grow beyond the pages where the stack guard was
2988 // mapped.  If at some point later in the process the stack expands to
2989 // that point, the Linux kernel cannot expand the stack any further
2990 // because the guard pages are in the way, and a segfault occurs.
2991 //
2992 // However, it's essential not to split the stack region by unmapping
2993 // a region (leaving a hole) that's already part of the stack mapping,
2994 // so if the stack mapping has already grown beyond the guard pages at
2995 // the time we create them, we have to truncate the stack mapping.
2996 // So, we need to know the extent of the stack mapping when
2997 // create_stack_guard_pages() is called.
2998 
2999 // We only need this for stacks that are growable: at the time of
3000 // writing thread stacks don't use growable mappings (i.e. those
3001 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3002 // only applies to the main thread.
3003 
3004 // If the (growable) stack mapping already extends beyond the point
3005 // where we're going to put our guard pages, truncate the mapping at
3006 // that point by munmap()ping it.  This ensures that when we later
3007 // munmap() the guard pages we don't leave a hole in the stack
3008 // mapping. This only affects the main/initial thread
3009 
3010 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3011   if (os::Linux::is_initial_thread()) {
3012     // As we manually grow stack up to bottom inside create_attached_thread(),
3013     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3014     // we don't need to do anything special.
3015     // Check it first, before calling heavy function.
3016     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3017     unsigned char vec[1];
3018 
3019     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3020       // Fallback to slow path on all errors, including EAGAIN
3021       stack_extent = (uintptr_t) get_stack_commited_bottom(
3022                                                            os::Linux::initial_thread_stack_bottom(),
3023                                                            (size_t)addr - stack_extent);
3024     }
3025 
3026     if (stack_extent < (uintptr_t)addr) {
3027       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3028     }
3029   }
3030 
3031   return os::commit_memory(addr, size, !ExecMem);
3032 }
3033 
3034 // If this is a growable mapping, remove the guard pages entirely by
3035 // munmap()ping them.  If not, just call uncommit_memory(). This only
3036 // affects the main/initial thread, but guard against future OS changes
3037 // It's safe to always unmap guard pages for initial thread because we
3038 // always place it right after end of the mapped region
3039 
3040 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3041   uintptr_t stack_extent, stack_base;
3042 
3043   if (os::Linux::is_initial_thread()) {
3044     return ::munmap(addr, size) == 0;
3045   }
3046 
3047   return os::uncommit_memory(addr, size);
3048 }
3049 
3050 static address _highest_vm_reserved_address = NULL;
3051 
3052 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3053 // at 'requested_addr'. If there are existing memory mappings at the same
3054 // location, however, they will be overwritten. If 'fixed' is false,
3055 // 'requested_addr' is only treated as a hint, the return value may or
3056 // may not start from the requested address. Unlike Linux mmap(), this
3057 // function returns NULL to indicate failure.
3058 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3059   char * addr;
3060   int flags;
3061 
3062   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3063   if (fixed) {
3064     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3065     flags |= MAP_FIXED;
3066   }
3067 
3068   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3069   // touch an uncommitted page. Otherwise, the read/write might
3070   // succeed if we have enough swap space to back the physical page.
3071   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3072                        flags, -1, 0);
3073 
3074   if (addr != MAP_FAILED) {
3075     // anon_mmap() should only get called during VM initialization,
3076     // don't need lock (actually we can skip locking even it can be called
3077     // from multiple threads, because _highest_vm_reserved_address is just a
3078     // hint about the upper limit of non-stack memory regions.)
3079     if ((address)addr + bytes > _highest_vm_reserved_address) {
3080       _highest_vm_reserved_address = (address)addr + bytes;
3081     }
3082   }
3083 
3084   return addr == MAP_FAILED ? NULL : addr;
3085 }
3086 
3087 // Don't update _highest_vm_reserved_address, because there might be memory
3088 // regions above addr + size. If so, releasing a memory region only creates
3089 // a hole in the address space, it doesn't help prevent heap-stack collision.
3090 //
3091 static int anon_munmap(char * addr, size_t size) {
3092   return ::munmap(addr, size) == 0;
3093 }
3094 
3095 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3096                             size_t alignment_hint) {
3097   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3098 }
3099 
3100 bool os::pd_release_memory(char* addr, size_t size) {
3101   return anon_munmap(addr, size);
3102 }
3103 
3104 static address highest_vm_reserved_address() {
3105   return _highest_vm_reserved_address;
3106 }
3107 
3108 static bool linux_mprotect(char* addr, size_t size, int prot) {
3109   // Linux wants the mprotect address argument to be page aligned.
3110   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3111 
3112   // According to SUSv3, mprotect() should only be used with mappings
3113   // established by mmap(), and mmap() always maps whole pages. Unaligned
3114   // 'addr' likely indicates problem in the VM (e.g. trying to change
3115   // protection of malloc'ed or statically allocated memory). Check the
3116   // caller if you hit this assert.
3117   assert(addr == bottom, "sanity check");
3118 
3119   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3120   return ::mprotect(bottom, size, prot) == 0;
3121 }
3122 
3123 // Set protections specified
3124 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3125                         bool is_committed) {
3126   unsigned int p = 0;
3127   switch (prot) {
3128   case MEM_PROT_NONE: p = PROT_NONE; break;
3129   case MEM_PROT_READ: p = PROT_READ; break;
3130   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3131   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3132   default:
3133     ShouldNotReachHere();
3134   }
3135   // is_committed is unused.
3136   return linux_mprotect(addr, bytes, p);
3137 }
3138 
3139 bool os::guard_memory(char* addr, size_t size) {
3140   return linux_mprotect(addr, size, PROT_NONE);
3141 }
3142 
3143 bool os::unguard_memory(char* addr, size_t size) {
3144   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3145 }
3146 
3147 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3148                                                     size_t page_size) {
3149   bool result = false;
3150   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3151                  MAP_ANONYMOUS|MAP_PRIVATE,
3152                  -1, 0);
3153   if (p != MAP_FAILED) {
3154     void *aligned_p = align_ptr_up(p, page_size);
3155 
3156     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3157 
3158     munmap(p, page_size * 2);
3159   }
3160 
3161   if (warn && !result) {
3162     warning("TransparentHugePages is not supported by the operating system.");
3163   }
3164 
3165   return result;
3166 }
3167 
3168 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3169   bool result = false;
3170   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3171                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3172                  -1, 0);
3173 
3174   if (p != MAP_FAILED) {
3175     // We don't know if this really is a huge page or not.
3176     FILE *fp = fopen("/proc/self/maps", "r");
3177     if (fp) {
3178       while (!feof(fp)) {
3179         char chars[257];
3180         long x = 0;
3181         if (fgets(chars, sizeof(chars), fp)) {
3182           if (sscanf(chars, "%lx-%*x", &x) == 1
3183               && x == (long)p) {
3184             if (strstr (chars, "hugepage")) {
3185               result = true;
3186               break;
3187             }
3188           }
3189         }
3190       }
3191       fclose(fp);
3192     }
3193     munmap(p, page_size);
3194   }
3195 
3196   if (warn && !result) {
3197     warning("HugeTLBFS is not supported by the operating system.");
3198   }
3199 
3200   return result;
3201 }
3202 
3203 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3204 //
3205 // From the coredump_filter documentation:
3206 //
3207 // - (bit 0) anonymous private memory
3208 // - (bit 1) anonymous shared memory
3209 // - (bit 2) file-backed private memory
3210 // - (bit 3) file-backed shared memory
3211 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3212 //           effective only if the bit 2 is cleared)
3213 // - (bit 5) hugetlb private memory
3214 // - (bit 6) hugetlb shared memory
3215 //
3216 static void set_coredump_filter(void) {
3217   FILE *f;
3218   long cdm;
3219 
3220   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3221     return;
3222   }
3223 
3224   if (fscanf(f, "%lx", &cdm) != 1) {
3225     fclose(f);
3226     return;
3227   }
3228 
3229   rewind(f);
3230 
3231   if ((cdm & LARGEPAGES_BIT) == 0) {
3232     cdm |= LARGEPAGES_BIT;
3233     fprintf(f, "%#lx", cdm);
3234   }
3235 
3236   fclose(f);
3237 }
3238 
3239 // Large page support
3240 
3241 static size_t _large_page_size = 0;
3242 
3243 size_t os::Linux::find_large_page_size() {
3244   size_t large_page_size = 0;
3245 
3246   // large_page_size on Linux is used to round up heap size. x86 uses either
3247   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3248   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3249   // page as large as 256M.
3250   //
3251   // Here we try to figure out page size by parsing /proc/meminfo and looking
3252   // for a line with the following format:
3253   //    Hugepagesize:     2048 kB
3254   //
3255   // If we can't determine the value (e.g. /proc is not mounted, or the text
3256   // format has been changed), we'll use the largest page size supported by
3257   // the processor.
3258 
3259 #ifndef ZERO
3260   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3261                      ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3262 #endif // ZERO
3263 
3264   FILE *fp = fopen("/proc/meminfo", "r");
3265   if (fp) {
3266     while (!feof(fp)) {
3267       int x = 0;
3268       char buf[16];
3269       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3270         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3271           large_page_size = x * K;
3272           break;
3273         }
3274       } else {
3275         // skip to next line
3276         for (;;) {
3277           int ch = fgetc(fp);
3278           if (ch == EOF || ch == (int)'\n') break;
3279         }
3280       }
3281     }
3282     fclose(fp);
3283   }
3284 
3285   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3286     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3287             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3288             proper_unit_for_byte_size(large_page_size));
3289   }
3290 
3291   return large_page_size;
3292 }
3293 
3294 size_t os::Linux::setup_large_page_size() {
3295   _large_page_size = Linux::find_large_page_size();
3296   const size_t default_page_size = (size_t)Linux::page_size();
3297   if (_large_page_size > default_page_size) {
3298     _page_sizes[0] = _large_page_size;
3299     _page_sizes[1] = default_page_size;
3300     _page_sizes[2] = 0;
3301   }
3302 
3303   return _large_page_size;
3304 }
3305 
3306 bool os::Linux::setup_large_page_type(size_t page_size) {
3307   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3308       FLAG_IS_DEFAULT(UseSHM) &&
3309       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3310 
3311     // The type of large pages has not been specified by the user.
3312 
3313     // Try UseHugeTLBFS and then UseSHM.
3314     UseHugeTLBFS = UseSHM = true;
3315 
3316     // Don't try UseTransparentHugePages since there are known
3317     // performance issues with it turned on. This might change in the future.
3318     UseTransparentHugePages = false;
3319   }
3320 
3321   if (UseTransparentHugePages) {
3322     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3323     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3324       UseHugeTLBFS = false;
3325       UseSHM = false;
3326       return true;
3327     }
3328     UseTransparentHugePages = false;
3329   }
3330 
3331   if (UseHugeTLBFS) {
3332     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3333     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3334       UseSHM = false;
3335       return true;
3336     }
3337     UseHugeTLBFS = false;
3338   }
3339 
3340   return UseSHM;
3341 }
3342 
3343 void os::large_page_init() {
3344   if (!UseLargePages &&
3345       !UseTransparentHugePages &&
3346       !UseHugeTLBFS &&
3347       !UseSHM) {
3348     // Not using large pages.
3349     return;
3350   }
3351 
3352   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3353     // The user explicitly turned off large pages.
3354     // Ignore the rest of the large pages flags.
3355     UseTransparentHugePages = false;
3356     UseHugeTLBFS = false;
3357     UseSHM = false;
3358     return;
3359   }
3360 
3361   size_t large_page_size = Linux::setup_large_page_size();
3362   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3363 
3364   set_coredump_filter();
3365 }
3366 
3367 #ifndef SHM_HUGETLB
3368   #define SHM_HUGETLB 04000
3369 #endif
3370 
3371 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3372                                             char* req_addr, bool exec) {
3373   // "exec" is passed in but not used.  Creating the shared image for
3374   // the code cache doesn't have an SHM_X executable permission to check.
3375   assert(UseLargePages && UseSHM, "only for SHM large pages");
3376   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3377 
3378   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3379     return NULL; // Fallback to small pages.
3380   }
3381 
3382   key_t key = IPC_PRIVATE;
3383   char *addr;
3384 
3385   bool warn_on_failure = UseLargePages &&
3386                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3387                          !FLAG_IS_DEFAULT(UseSHM) ||
3388                          !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3389   char msg[128];
3390 
3391   // Create a large shared memory region to attach to based on size.
3392   // Currently, size is the total size of the heap
3393   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3394   if (shmid == -1) {
3395     // Possible reasons for shmget failure:
3396     // 1. shmmax is too small for Java heap.
3397     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3398     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3399     // 2. not enough large page memory.
3400     //    > check available large pages: cat /proc/meminfo
3401     //    > increase amount of large pages:
3402     //          echo new_value > /proc/sys/vm/nr_hugepages
3403     //      Note 1: different Linux may use different name for this property,
3404     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3405     //      Note 2: it's possible there's enough physical memory available but
3406     //            they are so fragmented after a long run that they can't
3407     //            coalesce into large pages. Try to reserve large pages when
3408     //            the system is still "fresh".
3409     if (warn_on_failure) {
3410       jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3411       warning("%s", msg);
3412     }
3413     return NULL;
3414   }
3415 
3416   // attach to the region
3417   addr = (char*)shmat(shmid, req_addr, 0);
3418   int err = errno;
3419 
3420   // Remove shmid. If shmat() is successful, the actual shared memory segment
3421   // will be deleted when it's detached by shmdt() or when the process
3422   // terminates. If shmat() is not successful this will remove the shared
3423   // segment immediately.
3424   shmctl(shmid, IPC_RMID, NULL);
3425 
3426   if ((intptr_t)addr == -1) {
3427     if (warn_on_failure) {
3428       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3429       warning("%s", msg);
3430     }
3431     return NULL;
3432   }
3433 
3434   return addr;
3435 }
3436 
3437 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3438                                         int error) {
3439   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3440 
3441   bool warn_on_failure = UseLargePages &&
3442       (!FLAG_IS_DEFAULT(UseLargePages) ||
3443        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3444        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3445 
3446   if (warn_on_failure) {
3447     char msg[128];
3448     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3449                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3450     warning("%s", msg);
3451   }
3452 }
3453 
3454 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3455                                                         char* req_addr,
3456                                                         bool exec) {
3457   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3458   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3459   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3460 
3461   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3462   char* addr = (char*)::mmap(req_addr, bytes, prot,
3463                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3464                              -1, 0);
3465 
3466   if (addr == MAP_FAILED) {
3467     warn_on_large_pages_failure(req_addr, bytes, errno);
3468     return NULL;
3469   }
3470 
3471   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3472 
3473   return addr;
3474 }
3475 
3476 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3477                                                          size_t alignment,
3478                                                          char* req_addr,
3479                                                          bool exec) {
3480   size_t large_page_size = os::large_page_size();
3481 
3482   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3483 
3484   // Allocate small pages.
3485 
3486   char* start;
3487   if (req_addr != NULL) {
3488     assert(is_ptr_aligned(req_addr, alignment), "Must be");
3489     assert(is_size_aligned(bytes, alignment), "Must be");
3490     start = os::reserve_memory(bytes, req_addr);
3491     assert(start == NULL || start == req_addr, "Must be");
3492   } else {
3493     start = os::reserve_memory_aligned(bytes, alignment);
3494   }
3495 
3496   if (start == NULL) {
3497     return NULL;
3498   }
3499 
3500   assert(is_ptr_aligned(start, alignment), "Must be");
3501 
3502   if (MemTracker::tracking_level() > NMT_minimal) {
3503     // os::reserve_memory_special will record this memory area.
3504     // Need to release it here to prevent overlapping reservations.
3505     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3506     tkr.record((address)start, bytes);
3507   }
3508 
3509   char* end = start + bytes;
3510 
3511   // Find the regions of the allocated chunk that can be promoted to large pages.
3512   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3513   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3514 
3515   size_t lp_bytes = lp_end - lp_start;
3516 
3517   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3518 
3519   if (lp_bytes == 0) {
3520     // The mapped region doesn't even span the start and the end of a large page.
3521     // Fall back to allocate a non-special area.
3522     ::munmap(start, end - start);
3523     return NULL;
3524   }
3525 
3526   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3527 
3528 
3529   void* result;
3530 
3531   if (start != lp_start) {
3532     result = ::mmap(start, lp_start - start, prot,
3533                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3534                     -1, 0);
3535     if (result == MAP_FAILED) {
3536       ::munmap(lp_start, end - lp_start);
3537       return NULL;
3538     }
3539   }
3540 
3541   result = ::mmap(lp_start, lp_bytes, prot,
3542                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3543                   -1, 0);
3544   if (result == MAP_FAILED) {
3545     warn_on_large_pages_failure(req_addr, bytes, errno);
3546     // If the mmap above fails, the large pages region will be unmapped and we
3547     // have regions before and after with small pages. Release these regions.
3548     //
3549     // |  mapped  |  unmapped  |  mapped  |
3550     // ^          ^            ^          ^
3551     // start      lp_start     lp_end     end
3552     //
3553     ::munmap(start, lp_start - start);
3554     ::munmap(lp_end, end - lp_end);
3555     return NULL;
3556   }
3557 
3558   if (lp_end != end) {
3559     result = ::mmap(lp_end, end - lp_end, prot,
3560                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3561                     -1, 0);
3562     if (result == MAP_FAILED) {
3563       ::munmap(start, lp_end - start);
3564       return NULL;
3565     }
3566   }
3567 
3568   return start;
3569 }
3570 
3571 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3572                                                    size_t alignment,
3573                                                    char* req_addr,
3574                                                    bool exec) {
3575   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3576   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3577   assert(is_power_of_2(alignment), "Must be");
3578   assert(is_power_of_2(os::large_page_size()), "Must be");
3579   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3580 
3581   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3582     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3583   } else {
3584     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3585   }
3586 }
3587 
3588 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3589                                  char* req_addr, bool exec) {
3590   assert(UseLargePages, "only for large pages");
3591 
3592   char* addr;
3593   if (UseSHM) {
3594     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3595   } else {
3596     assert(UseHugeTLBFS, "must be");
3597     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3598   }
3599 
3600   if (addr != NULL) {
3601     if (UseNUMAInterleaving) {
3602       numa_make_global(addr, bytes);
3603     }
3604 
3605     // The memory is committed
3606     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3607   }
3608 
3609   return addr;
3610 }
3611 
3612 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3613   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3614   return shmdt(base) == 0;
3615 }
3616 
3617 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3618   return pd_release_memory(base, bytes);
3619 }
3620 
3621 bool os::release_memory_special(char* base, size_t bytes) {
3622   bool res;
3623   if (MemTracker::tracking_level() > NMT_minimal) {
3624     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3625     res = os::Linux::release_memory_special_impl(base, bytes);
3626     if (res) {
3627       tkr.record((address)base, bytes);
3628     }
3629 
3630   } else {
3631     res = os::Linux::release_memory_special_impl(base, bytes);
3632   }
3633   return res;
3634 }
3635 
3636 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3637   assert(UseLargePages, "only for large pages");
3638   bool res;
3639 
3640   if (UseSHM) {
3641     res = os::Linux::release_memory_special_shm(base, bytes);
3642   } else {
3643     assert(UseHugeTLBFS, "must be");
3644     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3645   }
3646   return res;
3647 }
3648 
3649 size_t os::large_page_size() {
3650   return _large_page_size;
3651 }
3652 
3653 // With SysV SHM the entire memory region must be allocated as shared
3654 // memory.
3655 // HugeTLBFS allows application to commit large page memory on demand.
3656 // However, when committing memory with HugeTLBFS fails, the region
3657 // that was supposed to be committed will lose the old reservation
3658 // and allow other threads to steal that memory region. Because of this
3659 // behavior we can't commit HugeTLBFS memory.
3660 bool os::can_commit_large_page_memory() {
3661   return UseTransparentHugePages;
3662 }
3663 
3664 bool os::can_execute_large_page_memory() {
3665   return UseTransparentHugePages || UseHugeTLBFS;
3666 }
3667 
3668 // Reserve memory at an arbitrary address, only if that area is
3669 // available (and not reserved for something else).
3670 
3671 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3672   const int max_tries = 10;
3673   char* base[max_tries];
3674   size_t size[max_tries];
3675   const size_t gap = 0x000000;
3676 
3677   // Assert only that the size is a multiple of the page size, since
3678   // that's all that mmap requires, and since that's all we really know
3679   // about at this low abstraction level.  If we need higher alignment,
3680   // we can either pass an alignment to this method or verify alignment
3681   // in one of the methods further up the call chain.  See bug 5044738.
3682   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3683 
3684   // Repeatedly allocate blocks until the block is allocated at the
3685   // right spot. Give up after max_tries. Note that reserve_memory() will
3686   // automatically update _highest_vm_reserved_address if the call is
3687   // successful. The variable tracks the highest memory address every reserved
3688   // by JVM. It is used to detect heap-stack collision if running with
3689   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3690   // space than needed, it could confuse the collision detecting code. To
3691   // solve the problem, save current _highest_vm_reserved_address and
3692   // calculate the correct value before return.
3693   address old_highest = _highest_vm_reserved_address;
3694 
3695   // Linux mmap allows caller to pass an address as hint; give it a try first,
3696   // if kernel honors the hint then we can return immediately.
3697   char * addr = anon_mmap(requested_addr, bytes, false);
3698   if (addr == requested_addr) {
3699     return requested_addr;
3700   }
3701 
3702   if (addr != NULL) {
3703     // mmap() is successful but it fails to reserve at the requested address
3704     anon_munmap(addr, bytes);
3705   }
3706 
3707   int i;
3708   for (i = 0; i < max_tries; ++i) {
3709     base[i] = reserve_memory(bytes);
3710 
3711     if (base[i] != NULL) {
3712       // Is this the block we wanted?
3713       if (base[i] == requested_addr) {
3714         size[i] = bytes;
3715         break;
3716       }
3717 
3718       // Does this overlap the block we wanted? Give back the overlapped
3719       // parts and try again.
3720 
3721       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3722       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3723         unmap_memory(base[i], top_overlap);
3724         base[i] += top_overlap;
3725         size[i] = bytes - top_overlap;
3726       } else {
3727         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3728         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3729           unmap_memory(requested_addr, bottom_overlap);
3730           size[i] = bytes - bottom_overlap;
3731         } else {
3732           size[i] = bytes;
3733         }
3734       }
3735     }
3736   }
3737 
3738   // Give back the unused reserved pieces.
3739 
3740   for (int j = 0; j < i; ++j) {
3741     if (base[j] != NULL) {
3742       unmap_memory(base[j], size[j]);
3743     }
3744   }
3745 
3746   if (i < max_tries) {
3747     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3748     return requested_addr;
3749   } else {
3750     _highest_vm_reserved_address = old_highest;
3751     return NULL;
3752   }
3753 }
3754 
3755 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3756   return ::read(fd, buf, nBytes);
3757 }
3758 
3759 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3760   return ::pread(fd, buf, nBytes, offset);
3761 }
3762 
3763 // Short sleep, direct OS call.
3764 //
3765 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3766 // sched_yield(2) will actually give up the CPU:
3767 //
3768 //   * Alone on this pariticular CPU, keeps running.
3769 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3770 //     (pre 2.6.39).
3771 //
3772 // So calling this with 0 is an alternative.
3773 //
3774 void os::naked_short_sleep(jlong ms) {
3775   struct timespec req;
3776 
3777   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3778   req.tv_sec = 0;
3779   if (ms > 0) {
3780     req.tv_nsec = (ms % 1000) * 1000000;
3781   } else {
3782     req.tv_nsec = 1;
3783   }
3784 
3785   nanosleep(&req, NULL);
3786 
3787   return;
3788 }
3789 
3790 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3791 void os::infinite_sleep() {
3792   while (true) {    // sleep forever ...
3793     ::sleep(100);   // ... 100 seconds at a time
3794   }
3795 }
3796 
3797 // Used to convert frequent JVM_Yield() to nops
3798 bool os::dont_yield() {
3799   return DontYieldALot;
3800 }
3801 
3802 void os::naked_yield() {
3803   sched_yield();
3804 }
3805 
3806 ////////////////////////////////////////////////////////////////////////////////
3807 // thread priority support
3808 
3809 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3810 // only supports dynamic priority, static priority must be zero. For real-time
3811 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3812 // However, for large multi-threaded applications, SCHED_RR is not only slower
3813 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3814 // of 5 runs - Sep 2005).
3815 //
3816 // The following code actually changes the niceness of kernel-thread/LWP. It
3817 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3818 // not the entire user process, and user level threads are 1:1 mapped to kernel
3819 // threads. It has always been the case, but could change in the future. For
3820 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3821 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3822 
3823 int os::java_to_os_priority[CriticalPriority + 1] = {
3824   19,              // 0 Entry should never be used
3825 
3826    4,              // 1 MinPriority
3827    3,              // 2
3828    2,              // 3
3829 
3830    1,              // 4
3831    0,              // 5 NormPriority
3832   -1,              // 6
3833 
3834   -2,              // 7
3835   -3,              // 8
3836   -4,              // 9 NearMaxPriority
3837 
3838   -5,              // 10 MaxPriority
3839 
3840   -5               // 11 CriticalPriority
3841 };
3842 
3843 static int prio_init() {
3844   if (ThreadPriorityPolicy == 1) {
3845     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3846     // if effective uid is not root. Perhaps, a more elegant way of doing
3847     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3848     if (geteuid() != 0) {
3849       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3850         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3851       }
3852       ThreadPriorityPolicy = 0;
3853     }
3854   }
3855   if (UseCriticalJavaThreadPriority) {
3856     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3857   }
3858   return 0;
3859 }
3860 
3861 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3862   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3863 
3864   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3865   return (ret == 0) ? OS_OK : OS_ERR;
3866 }
3867 
3868 OSReturn os::get_native_priority(const Thread* const thread,
3869                                  int *priority_ptr) {
3870   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3871     *priority_ptr = java_to_os_priority[NormPriority];
3872     return OS_OK;
3873   }
3874 
3875   errno = 0;
3876   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3877   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3878 }
3879 
3880 // Hint to the underlying OS that a task switch would not be good.
3881 // Void return because it's a hint and can fail.
3882 void os::hint_no_preempt() {}
3883 
3884 ////////////////////////////////////////////////////////////////////////////////
3885 // suspend/resume support
3886 
3887 //  the low-level signal-based suspend/resume support is a remnant from the
3888 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3889 //  within hotspot. Now there is a single use-case for this:
3890 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3891 //      that runs in the watcher thread.
3892 //  The remaining code is greatly simplified from the more general suspension
3893 //  code that used to be used.
3894 //
3895 //  The protocol is quite simple:
3896 //  - suspend:
3897 //      - sends a signal to the target thread
3898 //      - polls the suspend state of the osthread using a yield loop
3899 //      - target thread signal handler (SR_handler) sets suspend state
3900 //        and blocks in sigsuspend until continued
3901 //  - resume:
3902 //      - sets target osthread state to continue
3903 //      - sends signal to end the sigsuspend loop in the SR_handler
3904 //
3905 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3906 
3907 static void resume_clear_context(OSThread *osthread) {
3908   osthread->set_ucontext(NULL);
3909   osthread->set_siginfo(NULL);
3910 }
3911 
3912 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3913                                  ucontext_t* context) {
3914   osthread->set_ucontext(context);
3915   osthread->set_siginfo(siginfo);
3916 }
3917 
3918 // Handler function invoked when a thread's execution is suspended or
3919 // resumed. We have to be careful that only async-safe functions are
3920 // called here (Note: most pthread functions are not async safe and
3921 // should be avoided.)
3922 //
3923 // Note: sigwait() is a more natural fit than sigsuspend() from an
3924 // interface point of view, but sigwait() prevents the signal hander
3925 // from being run. libpthread would get very confused by not having
3926 // its signal handlers run and prevents sigwait()'s use with the
3927 // mutex granting granting signal.
3928 //
3929 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3930 //
3931 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3932   // Save and restore errno to avoid confusing native code with EINTR
3933   // after sigsuspend.
3934   int old_errno = errno;
3935 
3936   Thread* thread = Thread::current();
3937   OSThread* osthread = thread->osthread();
3938   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3939 
3940   os::SuspendResume::State current = osthread->sr.state();
3941   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3942     suspend_save_context(osthread, siginfo, context);
3943 
3944     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3945     os::SuspendResume::State state = osthread->sr.suspended();
3946     if (state == os::SuspendResume::SR_SUSPENDED) {
3947       sigset_t suspend_set;  // signals for sigsuspend()
3948 
3949       // get current set of blocked signals and unblock resume signal
3950       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3951       sigdelset(&suspend_set, SR_signum);
3952 
3953       sr_semaphore.signal();
3954       // wait here until we are resumed
3955       while (1) {
3956         sigsuspend(&suspend_set);
3957 
3958         os::SuspendResume::State result = osthread->sr.running();
3959         if (result == os::SuspendResume::SR_RUNNING) {
3960           sr_semaphore.signal();
3961           break;
3962         }
3963       }
3964 
3965     } else if (state == os::SuspendResume::SR_RUNNING) {
3966       // request was cancelled, continue
3967     } else {
3968       ShouldNotReachHere();
3969     }
3970 
3971     resume_clear_context(osthread);
3972   } else if (current == os::SuspendResume::SR_RUNNING) {
3973     // request was cancelled, continue
3974   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3975     // ignore
3976   } else {
3977     // ignore
3978   }
3979 
3980   errno = old_errno;
3981 }
3982 
3983 
3984 static int SR_initialize() {
3985   struct sigaction act;
3986   char *s;
3987   // Get signal number to use for suspend/resume
3988   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3989     int sig = ::strtol(s, 0, 10);
3990     if (sig > 0 || sig < _NSIG) {
3991       SR_signum = sig;
3992     }
3993   }
3994 
3995   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3996          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3997 
3998   sigemptyset(&SR_sigset);
3999   sigaddset(&SR_sigset, SR_signum);
4000 
4001   // Set up signal handler for suspend/resume
4002   act.sa_flags = SA_RESTART|SA_SIGINFO;
4003   act.sa_handler = (void (*)(int)) SR_handler;
4004 
4005   // SR_signum is blocked by default.
4006   // 4528190 - We also need to block pthread restart signal (32 on all
4007   // supported Linux platforms). Note that LinuxThreads need to block
4008   // this signal for all threads to work properly. So we don't have
4009   // to use hard-coded signal number when setting up the mask.
4010   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4011 
4012   if (sigaction(SR_signum, &act, 0) == -1) {
4013     return -1;
4014   }
4015 
4016   // Save signal flag
4017   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4018   return 0;
4019 }
4020 
4021 static int sr_notify(OSThread* osthread) {
4022   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4023   assert_status(status == 0, status, "pthread_kill");
4024   return status;
4025 }
4026 
4027 // "Randomly" selected value for how long we want to spin
4028 // before bailing out on suspending a thread, also how often
4029 // we send a signal to a thread we want to resume
4030 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4031 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4032 
4033 // returns true on success and false on error - really an error is fatal
4034 // but this seems the normal response to library errors
4035 static bool do_suspend(OSThread* osthread) {
4036   assert(osthread->sr.is_running(), "thread should be running");
4037   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4038 
4039   // mark as suspended and send signal
4040   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4041     // failed to switch, state wasn't running?
4042     ShouldNotReachHere();
4043     return false;
4044   }
4045 
4046   if (sr_notify(osthread) != 0) {
4047     ShouldNotReachHere();
4048   }
4049 
4050   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4051   while (true) {
4052     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4053       break;
4054     } else {
4055       // timeout
4056       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4057       if (cancelled == os::SuspendResume::SR_RUNNING) {
4058         return false;
4059       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4060         // make sure that we consume the signal on the semaphore as well
4061         sr_semaphore.wait();
4062         break;
4063       } else {
4064         ShouldNotReachHere();
4065         return false;
4066       }
4067     }
4068   }
4069 
4070   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4071   return true;
4072 }
4073 
4074 static void do_resume(OSThread* osthread) {
4075   assert(osthread->sr.is_suspended(), "thread should be suspended");
4076   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4077 
4078   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4079     // failed to switch to WAKEUP_REQUEST
4080     ShouldNotReachHere();
4081     return;
4082   }
4083 
4084   while (true) {
4085     if (sr_notify(osthread) == 0) {
4086       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4087         if (osthread->sr.is_running()) {
4088           return;
4089         }
4090       }
4091     } else {
4092       ShouldNotReachHere();
4093     }
4094   }
4095 
4096   guarantee(osthread->sr.is_running(), "Must be running!");
4097 }
4098 
4099 ///////////////////////////////////////////////////////////////////////////////////
4100 // signal handling (except suspend/resume)
4101 
4102 // This routine may be used by user applications as a "hook" to catch signals.
4103 // The user-defined signal handler must pass unrecognized signals to this
4104 // routine, and if it returns true (non-zero), then the signal handler must
4105 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4106 // routine will never retun false (zero), but instead will execute a VM panic
4107 // routine kill the process.
4108 //
4109 // If this routine returns false, it is OK to call it again.  This allows
4110 // the user-defined signal handler to perform checks either before or after
4111 // the VM performs its own checks.  Naturally, the user code would be making
4112 // a serious error if it tried to handle an exception (such as a null check
4113 // or breakpoint) that the VM was generating for its own correct operation.
4114 //
4115 // This routine may recognize any of the following kinds of signals:
4116 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4117 // It should be consulted by handlers for any of those signals.
4118 //
4119 // The caller of this routine must pass in the three arguments supplied
4120 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4121 // field of the structure passed to sigaction().  This routine assumes that
4122 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4123 //
4124 // Note that the VM will print warnings if it detects conflicting signal
4125 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4126 //
4127 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4128                                                  siginfo_t* siginfo,
4129                                                  void* ucontext,
4130                                                  int abort_if_unrecognized);
4131 
4132 void signalHandler(int sig, siginfo_t* info, void* uc) {
4133   assert(info != NULL && uc != NULL, "it must be old kernel");
4134   int orig_errno = errno;  // Preserve errno value over signal handler.
4135   JVM_handle_linux_signal(sig, info, uc, true);
4136   errno = orig_errno;
4137 }
4138 
4139 
4140 // This boolean allows users to forward their own non-matching signals
4141 // to JVM_handle_linux_signal, harmlessly.
4142 bool os::Linux::signal_handlers_are_installed = false;
4143 
4144 // For signal-chaining
4145 struct sigaction os::Linux::sigact[MAXSIGNUM];
4146 unsigned int os::Linux::sigs = 0;
4147 bool os::Linux::libjsig_is_loaded = false;
4148 typedef struct sigaction *(*get_signal_t)(int);
4149 get_signal_t os::Linux::get_signal_action = NULL;
4150 
4151 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4152   struct sigaction *actp = NULL;
4153 
4154   if (libjsig_is_loaded) {
4155     // Retrieve the old signal handler from libjsig
4156     actp = (*get_signal_action)(sig);
4157   }
4158   if (actp == NULL) {
4159     // Retrieve the preinstalled signal handler from jvm
4160     actp = get_preinstalled_handler(sig);
4161   }
4162 
4163   return actp;
4164 }
4165 
4166 static bool call_chained_handler(struct sigaction *actp, int sig,
4167                                  siginfo_t *siginfo, void *context) {
4168   // Call the old signal handler
4169   if (actp->sa_handler == SIG_DFL) {
4170     // It's more reasonable to let jvm treat it as an unexpected exception
4171     // instead of taking the default action.
4172     return false;
4173   } else if (actp->sa_handler != SIG_IGN) {
4174     if ((actp->sa_flags & SA_NODEFER) == 0) {
4175       // automaticlly block the signal
4176       sigaddset(&(actp->sa_mask), sig);
4177     }
4178 
4179     sa_handler_t hand;
4180     sa_sigaction_t sa;
4181     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4182     // retrieve the chained handler
4183     if (siginfo_flag_set) {
4184       sa = actp->sa_sigaction;
4185     } else {
4186       hand = actp->sa_handler;
4187     }
4188 
4189     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4190       actp->sa_handler = SIG_DFL;
4191     }
4192 
4193     // try to honor the signal mask
4194     sigset_t oset;
4195     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4196 
4197     // call into the chained handler
4198     if (siginfo_flag_set) {
4199       (*sa)(sig, siginfo, context);
4200     } else {
4201       (*hand)(sig);
4202     }
4203 
4204     // restore the signal mask
4205     pthread_sigmask(SIG_SETMASK, &oset, 0);
4206   }
4207   // Tell jvm's signal handler the signal is taken care of.
4208   return true;
4209 }
4210 
4211 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4212   bool chained = false;
4213   // signal-chaining
4214   if (UseSignalChaining) {
4215     struct sigaction *actp = get_chained_signal_action(sig);
4216     if (actp != NULL) {
4217       chained = call_chained_handler(actp, sig, siginfo, context);
4218     }
4219   }
4220   return chained;
4221 }
4222 
4223 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4224   if ((((unsigned int)1 << sig) & sigs) != 0) {
4225     return &sigact[sig];
4226   }
4227   return NULL;
4228 }
4229 
4230 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4231   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4232   sigact[sig] = oldAct;
4233   sigs |= (unsigned int)1 << sig;
4234 }
4235 
4236 // for diagnostic
4237 int os::Linux::sigflags[MAXSIGNUM];
4238 
4239 int os::Linux::get_our_sigflags(int sig) {
4240   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4241   return sigflags[sig];
4242 }
4243 
4244 void os::Linux::set_our_sigflags(int sig, int flags) {
4245   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4246   sigflags[sig] = flags;
4247 }
4248 
4249 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4250   // Check for overwrite.
4251   struct sigaction oldAct;
4252   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4253 
4254   void* oldhand = oldAct.sa_sigaction
4255                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4256                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4257   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4258       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4259       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4260     if (AllowUserSignalHandlers || !set_installed) {
4261       // Do not overwrite; user takes responsibility to forward to us.
4262       return;
4263     } else if (UseSignalChaining) {
4264       // save the old handler in jvm
4265       save_preinstalled_handler(sig, oldAct);
4266       // libjsig also interposes the sigaction() call below and saves the
4267       // old sigaction on it own.
4268     } else {
4269       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4270                     "%#lx for signal %d.", (long)oldhand, sig));
4271     }
4272   }
4273 
4274   struct sigaction sigAct;
4275   sigfillset(&(sigAct.sa_mask));
4276   sigAct.sa_handler = SIG_DFL;
4277   if (!set_installed) {
4278     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4279   } else {
4280     sigAct.sa_sigaction = signalHandler;
4281     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4282   }
4283   // Save flags, which are set by ours
4284   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4285   sigflags[sig] = sigAct.sa_flags;
4286 
4287   int ret = sigaction(sig, &sigAct, &oldAct);
4288   assert(ret == 0, "check");
4289 
4290   void* oldhand2  = oldAct.sa_sigaction
4291                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4292                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4293   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4294 }
4295 
4296 // install signal handlers for signals that HotSpot needs to
4297 // handle in order to support Java-level exception handling.
4298 
4299 void os::Linux::install_signal_handlers() {
4300   if (!signal_handlers_are_installed) {
4301     signal_handlers_are_installed = true;
4302 
4303     // signal-chaining
4304     typedef void (*signal_setting_t)();
4305     signal_setting_t begin_signal_setting = NULL;
4306     signal_setting_t end_signal_setting = NULL;
4307     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4308                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4309     if (begin_signal_setting != NULL) {
4310       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4311                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4312       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4313                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4314       libjsig_is_loaded = true;
4315       assert(UseSignalChaining, "should enable signal-chaining");
4316     }
4317     if (libjsig_is_loaded) {
4318       // Tell libjsig jvm is setting signal handlers
4319       (*begin_signal_setting)();
4320     }
4321 
4322     set_signal_handler(SIGSEGV, true);
4323     set_signal_handler(SIGPIPE, true);
4324     set_signal_handler(SIGBUS, true);
4325     set_signal_handler(SIGILL, true);
4326     set_signal_handler(SIGFPE, true);
4327 #if defined(PPC64)
4328     set_signal_handler(SIGTRAP, true);
4329 #endif
4330     set_signal_handler(SIGXFSZ, true);
4331 
4332     if (libjsig_is_loaded) {
4333       // Tell libjsig jvm finishes setting signal handlers
4334       (*end_signal_setting)();
4335     }
4336 
4337     // We don't activate signal checker if libjsig is in place, we trust ourselves
4338     // and if UserSignalHandler is installed all bets are off.
4339     // Log that signal checking is off only if -verbose:jni is specified.
4340     if (CheckJNICalls) {
4341       if (libjsig_is_loaded) {
4342         if (PrintJNIResolving) {
4343           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4344         }
4345         check_signals = false;
4346       }
4347       if (AllowUserSignalHandlers) {
4348         if (PrintJNIResolving) {
4349           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4350         }
4351         check_signals = false;
4352       }
4353     }
4354   }
4355 }
4356 
4357 // This is the fastest way to get thread cpu time on Linux.
4358 // Returns cpu time (user+sys) for any thread, not only for current.
4359 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4360 // It might work on 2.6.10+ with a special kernel/glibc patch.
4361 // For reference, please, see IEEE Std 1003.1-2004:
4362 //   http://www.unix.org/single_unix_specification
4363 
4364 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4365   struct timespec tp;
4366   int rc = os::Linux::clock_gettime(clockid, &tp);
4367   assert(rc == 0, "clock_gettime is expected to return 0 code");
4368 
4369   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4370 }
4371 
4372 /////
4373 // glibc on Linux platform uses non-documented flag
4374 // to indicate, that some special sort of signal
4375 // trampoline is used.
4376 // We will never set this flag, and we should
4377 // ignore this flag in our diagnostic
4378 #ifdef SIGNIFICANT_SIGNAL_MASK
4379   #undef SIGNIFICANT_SIGNAL_MASK
4380 #endif
4381 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4382 
4383 static const char* get_signal_handler_name(address handler,
4384                                            char* buf, int buflen) {
4385   int offset;
4386   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4387   if (found) {
4388     // skip directory names
4389     const char *p1, *p2;
4390     p1 = buf;
4391     size_t len = strlen(os::file_separator());
4392     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4393     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4394   } else {
4395     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4396   }
4397   return buf;
4398 }
4399 
4400 static void print_signal_handler(outputStream* st, int sig,
4401                                  char* buf, size_t buflen) {
4402   struct sigaction sa;
4403 
4404   sigaction(sig, NULL, &sa);
4405 
4406   // See comment for SIGNIFICANT_SIGNAL_MASK define
4407   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4408 
4409   st->print("%s: ", os::exception_name(sig, buf, buflen));
4410 
4411   address handler = (sa.sa_flags & SA_SIGINFO)
4412     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4413     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4414 
4415   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4416     st->print("SIG_DFL");
4417   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4418     st->print("SIG_IGN");
4419   } else {
4420     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4421   }
4422 
4423   st->print(", sa_mask[0]=");
4424   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4425 
4426   address rh = VMError::get_resetted_sighandler(sig);
4427   // May be, handler was resetted by VMError?
4428   if (rh != NULL) {
4429     handler = rh;
4430     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4431   }
4432 
4433   st->print(", sa_flags=");
4434   os::Posix::print_sa_flags(st, sa.sa_flags);
4435 
4436   // Check: is it our handler?
4437   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4438       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4439     // It is our signal handler
4440     // check for flags, reset system-used one!
4441     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4442       st->print(
4443                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4444                 os::Linux::get_our_sigflags(sig));
4445     }
4446   }
4447   st->cr();
4448 }
4449 
4450 
4451 #define DO_SIGNAL_CHECK(sig)                      \
4452   do {                                            \
4453     if (!sigismember(&check_signal_done, sig)) {  \
4454       os::Linux::check_signal_handler(sig);       \
4455     }                                             \
4456   } while (0)
4457 
4458 // This method is a periodic task to check for misbehaving JNI applications
4459 // under CheckJNI, we can add any periodic checks here
4460 
4461 void os::run_periodic_checks() {
4462   if (check_signals == false) return;
4463 
4464   // SEGV and BUS if overridden could potentially prevent
4465   // generation of hs*.log in the event of a crash, debugging
4466   // such a case can be very challenging, so we absolutely
4467   // check the following for a good measure:
4468   DO_SIGNAL_CHECK(SIGSEGV);
4469   DO_SIGNAL_CHECK(SIGILL);
4470   DO_SIGNAL_CHECK(SIGFPE);
4471   DO_SIGNAL_CHECK(SIGBUS);
4472   DO_SIGNAL_CHECK(SIGPIPE);
4473   DO_SIGNAL_CHECK(SIGXFSZ);
4474 #if defined(PPC64)
4475   DO_SIGNAL_CHECK(SIGTRAP);
4476 #endif
4477 
4478   // ReduceSignalUsage allows the user to override these handlers
4479   // see comments at the very top and jvm_solaris.h
4480   if (!ReduceSignalUsage) {
4481     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4482     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4483     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4484     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4485   }
4486 
4487   DO_SIGNAL_CHECK(SR_signum);
4488   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4489 }
4490 
4491 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4492 
4493 static os_sigaction_t os_sigaction = NULL;
4494 
4495 void os::Linux::check_signal_handler(int sig) {
4496   char buf[O_BUFLEN];
4497   address jvmHandler = NULL;
4498 
4499 
4500   struct sigaction act;
4501   if (os_sigaction == NULL) {
4502     // only trust the default sigaction, in case it has been interposed
4503     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4504     if (os_sigaction == NULL) return;
4505   }
4506 
4507   os_sigaction(sig, (struct sigaction*)NULL, &act);
4508 
4509 
4510   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4511 
4512   address thisHandler = (act.sa_flags & SA_SIGINFO)
4513     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4514     : CAST_FROM_FN_PTR(address, act.sa_handler);
4515 
4516 
4517   switch (sig) {
4518   case SIGSEGV:
4519   case SIGBUS:
4520   case SIGFPE:
4521   case SIGPIPE:
4522   case SIGILL:
4523   case SIGXFSZ:
4524     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4525     break;
4526 
4527   case SHUTDOWN1_SIGNAL:
4528   case SHUTDOWN2_SIGNAL:
4529   case SHUTDOWN3_SIGNAL:
4530   case BREAK_SIGNAL:
4531     jvmHandler = (address)user_handler();
4532     break;
4533 
4534   case INTERRUPT_SIGNAL:
4535     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4536     break;
4537 
4538   default:
4539     if (sig == SR_signum) {
4540       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4541     } else {
4542       return;
4543     }
4544     break;
4545   }
4546 
4547   if (thisHandler != jvmHandler) {
4548     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4549     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4550     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4551     // No need to check this sig any longer
4552     sigaddset(&check_signal_done, sig);
4553     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4554     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4555       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4556                     exception_name(sig, buf, O_BUFLEN));
4557     }
4558   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4559     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4560     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4561     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4562     // No need to check this sig any longer
4563     sigaddset(&check_signal_done, sig);
4564   }
4565 
4566   // Dump all the signal
4567   if (sigismember(&check_signal_done, sig)) {
4568     print_signal_handlers(tty, buf, O_BUFLEN);
4569   }
4570 }
4571 
4572 extern void report_error(char* file_name, int line_no, char* title,
4573                          char* format, ...);
4574 
4575 extern bool signal_name(int signo, char* buf, size_t len);
4576 
4577 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4578   if (0 < exception_code && exception_code <= SIGRTMAX) {
4579     // signal
4580     if (!signal_name(exception_code, buf, size)) {
4581       jio_snprintf(buf, size, "SIG%d", exception_code);
4582     }
4583     return buf;
4584   } else {
4585     return NULL;
4586   }
4587 }
4588 
4589 // this is called _before_ the most of global arguments have been parsed
4590 void os::init(void) {
4591   char dummy;   // used to get a guess on initial stack address
4592 //  first_hrtime = gethrtime();
4593 
4594   // With LinuxThreads the JavaMain thread pid (primordial thread)
4595   // is different than the pid of the java launcher thread.
4596   // So, on Linux, the launcher thread pid is passed to the VM
4597   // via the sun.java.launcher.pid property.
4598   // Use this property instead of getpid() if it was correctly passed.
4599   // See bug 6351349.
4600   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4601 
4602   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4603 
4604   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4605 
4606   init_random(1234567);
4607 
4608   ThreadCritical::initialize();
4609 
4610   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4611   if (Linux::page_size() == -1) {
4612     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4613                   strerror(errno)));
4614   }
4615   init_page_sizes((size_t) Linux::page_size());
4616 
4617   Linux::initialize_system_info();
4618 
4619   // main_thread points to the aboriginal thread
4620   Linux::_main_thread = pthread_self();
4621 
4622   Linux::clock_init();
4623   initial_time_count = javaTimeNanos();
4624 
4625   // pthread_condattr initialization for monotonic clock
4626   int status;
4627   pthread_condattr_t* _condattr = os::Linux::condAttr();
4628   if ((status = pthread_condattr_init(_condattr)) != 0) {
4629     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4630   }
4631   // Only set the clock if CLOCK_MONOTONIC is available
4632   if (os::supports_monotonic_clock()) {
4633     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4634       if (status == EINVAL) {
4635         warning("Unable to use monotonic clock with relative timed-waits" \
4636                 " - changes to the time-of-day clock may have adverse affects");
4637       } else {
4638         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4639       }
4640     }
4641   }
4642   // else it defaults to CLOCK_REALTIME
4643 
4644   pthread_mutex_init(&dl_mutex, NULL);
4645 
4646   // If the pagesize of the VM is greater than 8K determine the appropriate
4647   // number of initial guard pages.  The user can change this with the
4648   // command line arguments, if needed.
4649   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4650     StackYellowPages = 1;
4651     StackRedPages = 1;
4652     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4653   }
4654 
4655   // retrieve entry point for pthread_setname_np
4656   Linux::_pthread_setname_np =
4657     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4658 
4659 }
4660 
4661 // To install functions for atexit system call
4662 extern "C" {
4663   static void perfMemory_exit_helper() {
4664     perfMemory_exit();
4665   }
4666 }
4667 
4668 // this is called _after_ the global arguments have been parsed
4669 jint os::init_2(void) {
4670   Linux::fast_thread_clock_init();
4671 
4672   // Allocate a single page and mark it as readable for safepoint polling
4673   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4674   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4675 
4676   os::set_polling_page(polling_page);
4677 
4678 #ifndef PRODUCT
4679   if (Verbose && PrintMiscellaneous) {
4680     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4681                (intptr_t)polling_page);
4682   }
4683 #endif
4684 
4685   if (!UseMembar) {
4686     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4687     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4688     os::set_memory_serialize_page(mem_serialize_page);
4689 
4690 #ifndef PRODUCT
4691     if (Verbose && PrintMiscellaneous) {
4692       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n",
4693                  (intptr_t)mem_serialize_page);
4694     }
4695 #endif
4696   }
4697 
4698   // initialize suspend/resume support - must do this before signal_sets_init()
4699   if (SR_initialize() != 0) {
4700     perror("SR_initialize failed");
4701     return JNI_ERR;
4702   }
4703 
4704   Linux::signal_sets_init();
4705   Linux::install_signal_handlers();
4706 
4707   // Check minimum allowable stack size for thread creation and to initialize
4708   // the java system classes, including StackOverflowError - depends on page
4709   // size.  Add a page for compiler2 recursion in main thread.
4710   // Add in 2*BytesPerWord times page size to account for VM stack during
4711   // class initialization depending on 32 or 64 bit VM.
4712   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4713                                       (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4714                                       (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4715 
4716   size_t threadStackSizeInBytes = ThreadStackSize * K;
4717   if (threadStackSizeInBytes != 0 &&
4718       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4719     tty->print_cr("\nThe stack size specified is too small, "
4720                   "Specify at least %dk",
4721                   os::Linux::min_stack_allowed/ K);
4722     return JNI_ERR;
4723   }
4724 
4725   // Make the stack size a multiple of the page size so that
4726   // the yellow/red zones can be guarded.
4727   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4728                                                 vm_page_size()));
4729 
4730   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4731 
4732 #if defined(IA32)
4733   workaround_expand_exec_shield_cs_limit();
4734 #endif
4735 
4736   Linux::libpthread_init();
4737   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4738     tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4739                   Linux::glibc_version(), Linux::libpthread_version(),
4740                   Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4741   }
4742 
4743   if (UseNUMA) {
4744     if (!Linux::libnuma_init()) {
4745       UseNUMA = false;
4746     } else {
4747       if ((Linux::numa_max_node() < 1)) {
4748         // There's only one node(they start from 0), disable NUMA.
4749         UseNUMA = false;
4750       }
4751     }
4752     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4753     // we can make the adaptive lgrp chunk resizing work. If the user specified
4754     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4755     // disable adaptive resizing.
4756     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4757       if (FLAG_IS_DEFAULT(UseNUMA)) {
4758         UseNUMA = false;
4759       } else {
4760         if (FLAG_IS_DEFAULT(UseLargePages) &&
4761             FLAG_IS_DEFAULT(UseSHM) &&
4762             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4763           UseLargePages = false;
4764         } else {
4765           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4766           UseAdaptiveSizePolicy = false;
4767           UseAdaptiveNUMAChunkSizing = false;
4768         }
4769       }
4770     }
4771     if (!UseNUMA && ForceNUMA) {
4772       UseNUMA = true;
4773     }
4774   }
4775 
4776   if (MaxFDLimit) {
4777     // set the number of file descriptors to max. print out error
4778     // if getrlimit/setrlimit fails but continue regardless.
4779     struct rlimit nbr_files;
4780     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4781     if (status != 0) {
4782       if (PrintMiscellaneous && (Verbose || WizardMode)) {
4783         perror("os::init_2 getrlimit failed");
4784       }
4785     } else {
4786       nbr_files.rlim_cur = nbr_files.rlim_max;
4787       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4788       if (status != 0) {
4789         if (PrintMiscellaneous && (Verbose || WizardMode)) {
4790           perror("os::init_2 setrlimit failed");
4791         }
4792       }
4793     }
4794   }
4795 
4796   // Initialize lock used to serialize thread creation (see os::create_thread)
4797   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4798 
4799   // at-exit methods are called in the reverse order of their registration.
4800   // atexit functions are called on return from main or as a result of a
4801   // call to exit(3C). There can be only 32 of these functions registered
4802   // and atexit() does not set errno.
4803 
4804   if (PerfAllowAtExitRegistration) {
4805     // only register atexit functions if PerfAllowAtExitRegistration is set.
4806     // atexit functions can be delayed until process exit time, which
4807     // can be problematic for embedded VM situations. Embedded VMs should
4808     // call DestroyJavaVM() to assure that VM resources are released.
4809 
4810     // note: perfMemory_exit_helper atexit function may be removed in
4811     // the future if the appropriate cleanup code can be added to the
4812     // VM_Exit VMOperation's doit method.
4813     if (atexit(perfMemory_exit_helper) != 0) {
4814       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4815     }
4816   }
4817 
4818   // initialize thread priority policy
4819   prio_init();
4820 
4821   return JNI_OK;
4822 }
4823 
4824 // Mark the polling page as unreadable
4825 void os::make_polling_page_unreadable(void) {
4826   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4827     fatal("Could not disable polling page");
4828   }
4829 }
4830 
4831 // Mark the polling page as readable
4832 void os::make_polling_page_readable(void) {
4833   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4834     fatal("Could not enable polling page");
4835   }
4836 }
4837 
4838 int os::active_processor_count() {
4839   // Linux doesn't yet have a (official) notion of processor sets,
4840   // so just return the number of online processors.
4841   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4842   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4843   return online_cpus;
4844 }
4845 
4846 void os::set_native_thread_name(const char *name) {
4847   if (Linux::_pthread_setname_np) {
4848     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4849     snprintf(buf, sizeof(buf), "%s", name);
4850     buf[sizeof(buf) - 1] = '\0';
4851     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4852     // ERANGE should not happen; all other errors should just be ignored.
4853     assert(rc != ERANGE, "pthread_setname_np failed");
4854   }
4855 }
4856 
4857 bool os::distribute_processes(uint length, uint* distribution) {
4858   // Not yet implemented.
4859   return false;
4860 }
4861 
4862 bool os::bind_to_processor(uint processor_id) {
4863   // Not yet implemented.
4864   return false;
4865 }
4866 
4867 ///
4868 
4869 void os::SuspendedThreadTask::internal_do_task() {
4870   if (do_suspend(_thread->osthread())) {
4871     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4872     do_task(context);
4873     do_resume(_thread->osthread());
4874   }
4875 }
4876 
4877 class PcFetcher : public os::SuspendedThreadTask {
4878  public:
4879   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4880   ExtendedPC result();
4881  protected:
4882   void do_task(const os::SuspendedThreadTaskContext& context);
4883  private:
4884   ExtendedPC _epc;
4885 };
4886 
4887 ExtendedPC PcFetcher::result() {
4888   guarantee(is_done(), "task is not done yet.");
4889   return _epc;
4890 }
4891 
4892 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4893   Thread* thread = context.thread();
4894   OSThread* osthread = thread->osthread();
4895   if (osthread->ucontext() != NULL) {
4896     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4897   } else {
4898     // NULL context is unexpected, double-check this is the VMThread
4899     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4900   }
4901 }
4902 
4903 // Suspends the target using the signal mechanism and then grabs the PC before
4904 // resuming the target. Used by the flat-profiler only
4905 ExtendedPC os::get_thread_pc(Thread* thread) {
4906   // Make sure that it is called by the watcher for the VMThread
4907   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4908   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4909 
4910   PcFetcher fetcher(thread);
4911   fetcher.run();
4912   return fetcher.result();
4913 }
4914 
4915 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4916                                    pthread_mutex_t *_mutex,
4917                                    const struct timespec *_abstime) {
4918   if (is_NPTL()) {
4919     return pthread_cond_timedwait(_cond, _mutex, _abstime);
4920   } else {
4921     // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4922     // word back to default 64bit precision if condvar is signaled. Java
4923     // wants 53bit precision.  Save and restore current value.
4924     int fpu = get_fpu_control_word();
4925     int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4926     set_fpu_control_word(fpu);
4927     return status;
4928   }
4929 }
4930 
4931 ////////////////////////////////////////////////////////////////////////////////
4932 // debug support
4933 
4934 bool os::find(address addr, outputStream* st) {
4935   Dl_info dlinfo;
4936   memset(&dlinfo, 0, sizeof(dlinfo));
4937   if (dladdr(addr, &dlinfo) != 0) {
4938     st->print(PTR_FORMAT ": ", addr);
4939     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4940       st->print("%s+%#x", dlinfo.dli_sname,
4941                 addr - (intptr_t)dlinfo.dli_saddr);
4942     } else if (dlinfo.dli_fbase != NULL) {
4943       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4944     } else {
4945       st->print("<absolute address>");
4946     }
4947     if (dlinfo.dli_fname != NULL) {
4948       st->print(" in %s", dlinfo.dli_fname);
4949     }
4950     if (dlinfo.dli_fbase != NULL) {
4951       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4952     }
4953     st->cr();
4954 
4955     if (Verbose) {
4956       // decode some bytes around the PC
4957       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4958       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4959       address       lowest = (address) dlinfo.dli_sname;
4960       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4961       if (begin < lowest)  begin = lowest;
4962       Dl_info dlinfo2;
4963       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4964           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4965         end = (address) dlinfo2.dli_saddr;
4966       }
4967       Disassembler::decode(begin, end, st);
4968     }
4969     return true;
4970   }
4971   return false;
4972 }
4973 
4974 ////////////////////////////////////////////////////////////////////////////////
4975 // misc
4976 
4977 // This does not do anything on Linux. This is basically a hook for being
4978 // able to use structured exception handling (thread-local exception filters)
4979 // on, e.g., Win32.
4980 void
4981 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4982                          JavaCallArguments* args, Thread* thread) {
4983   f(value, method, args, thread);
4984 }
4985 
4986 void os::print_statistics() {
4987 }
4988 
4989 int os::message_box(const char* title, const char* message) {
4990   int i;
4991   fdStream err(defaultStream::error_fd());
4992   for (i = 0; i < 78; i++) err.print_raw("=");
4993   err.cr();
4994   err.print_raw_cr(title);
4995   for (i = 0; i < 78; i++) err.print_raw("-");
4996   err.cr();
4997   err.print_raw_cr(message);
4998   for (i = 0; i < 78; i++) err.print_raw("=");
4999   err.cr();
5000 
5001   char buf[16];
5002   // Prevent process from exiting upon "read error" without consuming all CPU
5003   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5004 
5005   return buf[0] == 'y' || buf[0] == 'Y';
5006 }
5007 
5008 int os::stat(const char *path, struct stat *sbuf) {
5009   char pathbuf[MAX_PATH];
5010   if (strlen(path) > MAX_PATH - 1) {
5011     errno = ENAMETOOLONG;
5012     return -1;
5013   }
5014   os::native_path(strcpy(pathbuf, path));
5015   return ::stat(pathbuf, sbuf);
5016 }
5017 
5018 bool os::check_heap(bool force) {
5019   return true;
5020 }
5021 
5022 // Is a (classpath) directory empty?
5023 bool os::dir_is_empty(const char* path) {
5024   DIR *dir = NULL;
5025   struct dirent *ptr;
5026 
5027   dir = opendir(path);
5028   if (dir == NULL) return true;
5029 
5030   // Scan the directory
5031   bool result = true;
5032   char buf[sizeof(struct dirent) + MAX_PATH];
5033   while (result && (ptr = ::readdir(dir)) != NULL) {
5034     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5035       result = false;
5036     }
5037   }
5038   closedir(dir);
5039   return result;
5040 }
5041 
5042 // This code originates from JDK's sysOpen and open64_w
5043 // from src/solaris/hpi/src/system_md.c
5044 
5045 int os::open(const char *path, int oflag, int mode) {
5046   if (strlen(path) > MAX_PATH - 1) {
5047     errno = ENAMETOOLONG;
5048     return -1;
5049   }
5050 
5051   // All file descriptors that are opened in the Java process and not
5052   // specifically destined for a subprocess should have the close-on-exec
5053   // flag set.  If we don't set it, then careless 3rd party native code
5054   // might fork and exec without closing all appropriate file descriptors
5055   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5056   // turn might:
5057   //
5058   // - cause end-of-file to fail to be detected on some file
5059   //   descriptors, resulting in mysterious hangs, or
5060   //
5061   // - might cause an fopen in the subprocess to fail on a system
5062   //   suffering from bug 1085341.
5063   //
5064   // (Yes, the default setting of the close-on-exec flag is a Unix
5065   // design flaw)
5066   //
5067   // See:
5068   // 1085341: 32-bit stdio routines should support file descriptors >255
5069   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5070   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5071   //
5072   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5073   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5074   // because it saves a system call and removes a small window where the flag
5075   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5076   // and we fall back to using FD_CLOEXEC (see below).
5077 #ifdef O_CLOEXEC
5078   oflag |= O_CLOEXEC;
5079 #endif
5080 
5081   int fd = ::open64(path, oflag, mode);
5082   if (fd == -1) return -1;
5083 
5084   //If the open succeeded, the file might still be a directory
5085   {
5086     struct stat64 buf64;
5087     int ret = ::fstat64(fd, &buf64);
5088     int st_mode = buf64.st_mode;
5089 
5090     if (ret != -1) {
5091       if ((st_mode & S_IFMT) == S_IFDIR) {
5092         errno = EISDIR;
5093         ::close(fd);
5094         return -1;
5095       }
5096     } else {
5097       ::close(fd);
5098       return -1;
5099     }
5100   }
5101 
5102 #ifdef FD_CLOEXEC
5103   // Validate that the use of the O_CLOEXEC flag on open above worked.
5104   // With recent kernels, we will perform this check exactly once.
5105   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5106   if (!O_CLOEXEC_is_known_to_work) {
5107     int flags = ::fcntl(fd, F_GETFD);
5108     if (flags != -1) {
5109       if ((flags & FD_CLOEXEC) != 0)
5110         O_CLOEXEC_is_known_to_work = 1;
5111       else
5112         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5113     }
5114   }
5115 #endif
5116 
5117   return fd;
5118 }
5119 
5120 
5121 // create binary file, rewriting existing file if required
5122 int os::create_binary_file(const char* path, bool rewrite_existing) {
5123   int oflags = O_WRONLY | O_CREAT;
5124   if (!rewrite_existing) {
5125     oflags |= O_EXCL;
5126   }
5127   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5128 }
5129 
5130 // return current position of file pointer
5131 jlong os::current_file_offset(int fd) {
5132   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5133 }
5134 
5135 // move file pointer to the specified offset
5136 jlong os::seek_to_file_offset(int fd, jlong offset) {
5137   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5138 }
5139 
5140 // This code originates from JDK's sysAvailable
5141 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5142 
5143 int os::available(int fd, jlong *bytes) {
5144   jlong cur, end;
5145   int mode;
5146   struct stat64 buf64;
5147 
5148   if (::fstat64(fd, &buf64) >= 0) {
5149     mode = buf64.st_mode;
5150     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5151       // XXX: is the following call interruptible? If so, this might
5152       // need to go through the INTERRUPT_IO() wrapper as for other
5153       // blocking, interruptible calls in this file.
5154       int n;
5155       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5156         *bytes = n;
5157         return 1;
5158       }
5159     }
5160   }
5161   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5162     return 0;
5163   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5164     return 0;
5165   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5166     return 0;
5167   }
5168   *bytes = end - cur;
5169   return 1;
5170 }
5171 
5172 // Map a block of memory.
5173 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5174                         char *addr, size_t bytes, bool read_only,
5175                         bool allow_exec) {
5176   int prot;
5177   int flags = MAP_PRIVATE;
5178 
5179   if (read_only) {
5180     prot = PROT_READ;
5181   } else {
5182     prot = PROT_READ | PROT_WRITE;
5183   }
5184 
5185   if (allow_exec) {
5186     prot |= PROT_EXEC;
5187   }
5188 
5189   if (addr != NULL) {
5190     flags |= MAP_FIXED;
5191   }
5192 
5193   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5194                                      fd, file_offset);
5195   if (mapped_address == MAP_FAILED) {
5196     return NULL;
5197   }
5198   return mapped_address;
5199 }
5200 
5201 
5202 // Remap a block of memory.
5203 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5204                           char *addr, size_t bytes, bool read_only,
5205                           bool allow_exec) {
5206   // same as map_memory() on this OS
5207   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5208                         allow_exec);
5209 }
5210 
5211 
5212 // Unmap a block of memory.
5213 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5214   return munmap(addr, bytes) == 0;
5215 }
5216 
5217 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5218 
5219 static clockid_t thread_cpu_clockid(Thread* thread) {
5220   pthread_t tid = thread->osthread()->pthread_id();
5221   clockid_t clockid;
5222 
5223   // Get thread clockid
5224   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5225   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5226   return clockid;
5227 }
5228 
5229 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5230 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5231 // of a thread.
5232 //
5233 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5234 // the fast estimate available on the platform.
5235 
5236 jlong os::current_thread_cpu_time() {
5237   if (os::Linux::supports_fast_thread_cpu_time()) {
5238     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5239   } else {
5240     // return user + sys since the cost is the same
5241     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5242   }
5243 }
5244 
5245 jlong os::thread_cpu_time(Thread* thread) {
5246   // consistent with what current_thread_cpu_time() returns
5247   if (os::Linux::supports_fast_thread_cpu_time()) {
5248     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5249   } else {
5250     return slow_thread_cpu_time(thread, true /* user + sys */);
5251   }
5252 }
5253 
5254 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5255   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5256     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5257   } else {
5258     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5259   }
5260 }
5261 
5262 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5263   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5264     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5265   } else {
5266     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5267   }
5268 }
5269 
5270 //  -1 on error.
5271 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5272   pid_t  tid = thread->osthread()->thread_id();
5273   char *s;
5274   char stat[2048];
5275   int statlen;
5276   char proc_name[64];
5277   int count;
5278   long sys_time, user_time;
5279   char cdummy;
5280   int idummy;
5281   long ldummy;
5282   FILE *fp;
5283 
5284   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5285   fp = fopen(proc_name, "r");
5286   if (fp == NULL) return -1;
5287   statlen = fread(stat, 1, 2047, fp);
5288   stat[statlen] = '\0';
5289   fclose(fp);
5290 
5291   // Skip pid and the command string. Note that we could be dealing with
5292   // weird command names, e.g. user could decide to rename java launcher
5293   // to "java 1.4.2 :)", then the stat file would look like
5294   //                1234 (java 1.4.2 :)) R ... ...
5295   // We don't really need to know the command string, just find the last
5296   // occurrence of ")" and then start parsing from there. See bug 4726580.
5297   s = strrchr(stat, ')');
5298   if (s == NULL) return -1;
5299 
5300   // Skip blank chars
5301   do { s++; } while (s && isspace(*s));
5302 
5303   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5304                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5305                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5306                  &user_time, &sys_time);
5307   if (count != 13) return -1;
5308   if (user_sys_cpu_time) {
5309     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5310   } else {
5311     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5312   }
5313 }
5314 
5315 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5316   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5317   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5318   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5319   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5320 }
5321 
5322 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5323   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5324   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5325   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5326   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5327 }
5328 
5329 bool os::is_thread_cpu_time_supported() {
5330   return true;
5331 }
5332 
5333 // System loadavg support.  Returns -1 if load average cannot be obtained.
5334 // Linux doesn't yet have a (official) notion of processor sets,
5335 // so just return the system wide load average.
5336 int os::loadavg(double loadavg[], int nelem) {
5337   return ::getloadavg(loadavg, nelem);
5338 }
5339 
5340 void os::pause() {
5341   char filename[MAX_PATH];
5342   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5343     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5344   } else {
5345     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5346   }
5347 
5348   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5349   if (fd != -1) {
5350     struct stat buf;
5351     ::close(fd);
5352     while (::stat(filename, &buf) == 0) {
5353       (void)::poll(NULL, 0, 100);
5354     }
5355   } else {
5356     jio_fprintf(stderr,
5357                 "Could not open pause file '%s', continuing immediately.\n", filename);
5358   }
5359 }
5360 
5361 
5362 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5363 // comment paragraphs are worth repeating here:
5364 //
5365 // Assumption:
5366 //    Only one parker can exist on an event, which is why we allocate
5367 //    them per-thread. Multiple unparkers can coexist.
5368 //
5369 // _Event serves as a restricted-range semaphore.
5370 //   -1 : thread is blocked, i.e. there is a waiter
5371 //    0 : neutral: thread is running or ready,
5372 //        could have been signaled after a wait started
5373 //    1 : signaled - thread is running or ready
5374 //
5375 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5376 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5377 // For specifics regarding the bug see GLIBC BUGID 261237 :
5378 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5379 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5380 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5381 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5382 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5383 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5384 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5385 // of libpthread avoids the problem, but isn't practical.
5386 //
5387 // Possible remedies:
5388 //
5389 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5390 //      This is palliative and probabilistic, however.  If the thread is preempted
5391 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5392 //      than the minimum period may have passed, and the abstime may be stale (in the
5393 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5394 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5395 //
5396 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5397 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5398 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5399 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5400 //      thread.
5401 //
5402 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5403 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5404 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5405 //      This also works well.  In fact it avoids kernel-level scalability impediments
5406 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5407 //      timers in a graceful fashion.
5408 //
5409 // 4.   When the abstime value is in the past it appears that control returns
5410 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5411 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5412 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5413 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5414 //      It may be possible to avoid reinitialization by checking the return
5415 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5416 //      condvar we must establish the invariant that cond_signal() is only called
5417 //      within critical sections protected by the adjunct mutex.  This prevents
5418 //      cond_signal() from "seeing" a condvar that's in the midst of being
5419 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5420 //      desirable signal-after-unlock optimization that avoids futile context switching.
5421 //
5422 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5423 //      structure when a condvar is used or initialized.  cond_destroy()  would
5424 //      release the helper structure.  Our reinitialize-after-timedwait fix
5425 //      put excessive stress on malloc/free and locks protecting the c-heap.
5426 //
5427 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5428 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5429 // and only enabling the work-around for vulnerable environments.
5430 
5431 // utility to compute the abstime argument to timedwait:
5432 // millis is the relative timeout time
5433 // abstime will be the absolute timeout time
5434 // TODO: replace compute_abstime() with unpackTime()
5435 
5436 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5437   if (millis < 0)  millis = 0;
5438 
5439   jlong seconds = millis / 1000;
5440   millis %= 1000;
5441   if (seconds > 50000000) { // see man cond_timedwait(3T)
5442     seconds = 50000000;
5443   }
5444 
5445   if (os::supports_monotonic_clock()) {
5446     struct timespec now;
5447     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5448     assert_status(status == 0, status, "clock_gettime");
5449     abstime->tv_sec = now.tv_sec  + seconds;
5450     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5451     if (nanos >= NANOSECS_PER_SEC) {
5452       abstime->tv_sec += 1;
5453       nanos -= NANOSECS_PER_SEC;
5454     }
5455     abstime->tv_nsec = nanos;
5456   } else {
5457     struct timeval now;
5458     int status = gettimeofday(&now, NULL);
5459     assert(status == 0, "gettimeofday");
5460     abstime->tv_sec = now.tv_sec  + seconds;
5461     long usec = now.tv_usec + millis * 1000;
5462     if (usec >= 1000000) {
5463       abstime->tv_sec += 1;
5464       usec -= 1000000;
5465     }
5466     abstime->tv_nsec = usec * 1000;
5467   }
5468   return abstime;
5469 }
5470 
5471 void os::PlatformEvent::park() {       // AKA "down()"
5472   // Transitions for _Event:
5473   //   -1 => -1 : illegal
5474   //    1 =>  0 : pass - return immediately
5475   //    0 => -1 : block; then set _Event to 0 before returning
5476 
5477   // Invariant: Only the thread associated with the Event/PlatformEvent
5478   // may call park().
5479   // TODO: assert that _Assoc != NULL or _Assoc == Self
5480   assert(_nParked == 0, "invariant");
5481 
5482   int v;
5483   for (;;) {
5484     v = _Event;
5485     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5486   }
5487   guarantee(v >= 0, "invariant");
5488   if (v == 0) {
5489     // Do this the hard way by blocking ...
5490     int status = pthread_mutex_lock(_mutex);
5491     assert_status(status == 0, status, "mutex_lock");
5492     guarantee(_nParked == 0, "invariant");
5493     ++_nParked;
5494     while (_Event < 0) {
5495       status = pthread_cond_wait(_cond, _mutex);
5496       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5497       // Treat this the same as if the wait was interrupted
5498       if (status == ETIME) { status = EINTR; }
5499       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5500     }
5501     --_nParked;
5502 
5503     _Event = 0;
5504     status = pthread_mutex_unlock(_mutex);
5505     assert_status(status == 0, status, "mutex_unlock");
5506     // Paranoia to ensure our locked and lock-free paths interact
5507     // correctly with each other.
5508     OrderAccess::fence();
5509   }
5510   guarantee(_Event >= 0, "invariant");
5511 }
5512 
5513 int os::PlatformEvent::park(jlong millis) {
5514   // Transitions for _Event:
5515   //   -1 => -1 : illegal
5516   //    1 =>  0 : pass - return immediately
5517   //    0 => -1 : block; then set _Event to 0 before returning
5518 
5519   guarantee(_nParked == 0, "invariant");
5520 
5521   int v;
5522   for (;;) {
5523     v = _Event;
5524     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5525   }
5526   guarantee(v >= 0, "invariant");
5527   if (v != 0) return OS_OK;
5528 
5529   // We do this the hard way, by blocking the thread.
5530   // Consider enforcing a minimum timeout value.
5531   struct timespec abst;
5532   compute_abstime(&abst, millis);
5533 
5534   int ret = OS_TIMEOUT;
5535   int status = pthread_mutex_lock(_mutex);
5536   assert_status(status == 0, status, "mutex_lock");
5537   guarantee(_nParked == 0, "invariant");
5538   ++_nParked;
5539 
5540   // Object.wait(timo) will return because of
5541   // (a) notification
5542   // (b) timeout
5543   // (c) thread.interrupt
5544   //
5545   // Thread.interrupt and object.notify{All} both call Event::set.
5546   // That is, we treat thread.interrupt as a special case of notification.
5547   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5548   // We assume all ETIME returns are valid.
5549   //
5550   // TODO: properly differentiate simultaneous notify+interrupt.
5551   // In that case, we should propagate the notify to another waiter.
5552 
5553   while (_Event < 0) {
5554     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5555     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5556       pthread_cond_destroy(_cond);
5557       pthread_cond_init(_cond, os::Linux::condAttr());
5558     }
5559     assert_status(status == 0 || status == EINTR ||
5560                   status == ETIME || status == ETIMEDOUT,
5561                   status, "cond_timedwait");
5562     if (!FilterSpuriousWakeups) break;                 // previous semantics
5563     if (status == ETIME || status == ETIMEDOUT) break;
5564     // We consume and ignore EINTR and spurious wakeups.
5565   }
5566   --_nParked;
5567   if (_Event >= 0) {
5568     ret = OS_OK;
5569   }
5570   _Event = 0;
5571   status = pthread_mutex_unlock(_mutex);
5572   assert_status(status == 0, status, "mutex_unlock");
5573   assert(_nParked == 0, "invariant");
5574   // Paranoia to ensure our locked and lock-free paths interact
5575   // correctly with each other.
5576   OrderAccess::fence();
5577   return ret;
5578 }
5579 
5580 void os::PlatformEvent::unpark() {
5581   // Transitions for _Event:
5582   //    0 => 1 : just return
5583   //    1 => 1 : just return
5584   //   -1 => either 0 or 1; must signal target thread
5585   //         That is, we can safely transition _Event from -1 to either
5586   //         0 or 1.
5587   // See also: "Semaphores in Plan 9" by Mullender & Cox
5588   //
5589   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5590   // that it will take two back-to-back park() calls for the owning
5591   // thread to block. This has the benefit of forcing a spurious return
5592   // from the first park() call after an unpark() call which will help
5593   // shake out uses of park() and unpark() without condition variables.
5594 
5595   if (Atomic::xchg(1, &_Event) >= 0) return;
5596 
5597   // Wait for the thread associated with the event to vacate
5598   int status = pthread_mutex_lock(_mutex);
5599   assert_status(status == 0, status, "mutex_lock");
5600   int AnyWaiters = _nParked;
5601   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5602   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5603     AnyWaiters = 0;
5604     pthread_cond_signal(_cond);
5605   }
5606   status = pthread_mutex_unlock(_mutex);
5607   assert_status(status == 0, status, "mutex_unlock");
5608   if (AnyWaiters != 0) {
5609     // Note that we signal() *after* dropping the lock for "immortal" Events.
5610     // This is safe and avoids a common class of  futile wakeups.  In rare
5611     // circumstances this can cause a thread to return prematurely from
5612     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5613     // will simply re-test the condition and re-park itself.
5614     // This provides particular benefit if the underlying platform does not
5615     // provide wait morphing.
5616     status = pthread_cond_signal(_cond);
5617     assert_status(status == 0, status, "cond_signal");
5618   }
5619 }
5620 
5621 
5622 // JSR166
5623 // -------------------------------------------------------
5624 
5625 // The solaris and linux implementations of park/unpark are fairly
5626 // conservative for now, but can be improved. They currently use a
5627 // mutex/condvar pair, plus a a count.
5628 // Park decrements count if > 0, else does a condvar wait.  Unpark
5629 // sets count to 1 and signals condvar.  Only one thread ever waits
5630 // on the condvar. Contention seen when trying to park implies that someone
5631 // is unparking you, so don't wait. And spurious returns are fine, so there
5632 // is no need to track notifications.
5633 
5634 // This code is common to linux and solaris and will be moved to a
5635 // common place in dolphin.
5636 //
5637 // The passed in time value is either a relative time in nanoseconds
5638 // or an absolute time in milliseconds. Either way it has to be unpacked
5639 // into suitable seconds and nanoseconds components and stored in the
5640 // given timespec structure.
5641 // Given time is a 64-bit value and the time_t used in the timespec is only
5642 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5643 // overflow if times way in the future are given. Further on Solaris versions
5644 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5645 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5646 // As it will be 28 years before "now + 100000000" will overflow we can
5647 // ignore overflow and just impose a hard-limit on seconds using the value
5648 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5649 // years from "now".
5650 
5651 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5652   assert(time > 0, "convertTime");
5653   time_t max_secs = 0;
5654 
5655   if (!os::supports_monotonic_clock() || isAbsolute) {
5656     struct timeval now;
5657     int status = gettimeofday(&now, NULL);
5658     assert(status == 0, "gettimeofday");
5659 
5660     max_secs = now.tv_sec + MAX_SECS;
5661 
5662     if (isAbsolute) {
5663       jlong secs = time / 1000;
5664       if (secs > max_secs) {
5665         absTime->tv_sec = max_secs;
5666       } else {
5667         absTime->tv_sec = secs;
5668       }
5669       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5670     } else {
5671       jlong secs = time / NANOSECS_PER_SEC;
5672       if (secs >= MAX_SECS) {
5673         absTime->tv_sec = max_secs;
5674         absTime->tv_nsec = 0;
5675       } else {
5676         absTime->tv_sec = now.tv_sec + secs;
5677         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5678         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5679           absTime->tv_nsec -= NANOSECS_PER_SEC;
5680           ++absTime->tv_sec; // note: this must be <= max_secs
5681         }
5682       }
5683     }
5684   } else {
5685     // must be relative using monotonic clock
5686     struct timespec now;
5687     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5688     assert_status(status == 0, status, "clock_gettime");
5689     max_secs = now.tv_sec + MAX_SECS;
5690     jlong secs = time / NANOSECS_PER_SEC;
5691     if (secs >= MAX_SECS) {
5692       absTime->tv_sec = max_secs;
5693       absTime->tv_nsec = 0;
5694     } else {
5695       absTime->tv_sec = now.tv_sec + secs;
5696       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5697       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5698         absTime->tv_nsec -= NANOSECS_PER_SEC;
5699         ++absTime->tv_sec; // note: this must be <= max_secs
5700       }
5701     }
5702   }
5703   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5704   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5705   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5706   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5707 }
5708 
5709 void Parker::park(bool isAbsolute, jlong time) {
5710   // Ideally we'd do something useful while spinning, such
5711   // as calling unpackTime().
5712 
5713   // Optional fast-path check:
5714   // Return immediately if a permit is available.
5715   // We depend on Atomic::xchg() having full barrier semantics
5716   // since we are doing a lock-free update to _counter.
5717   if (Atomic::xchg(0, &_counter) > 0) return;
5718 
5719   Thread* thread = Thread::current();
5720   assert(thread->is_Java_thread(), "Must be JavaThread");
5721   JavaThread *jt = (JavaThread *)thread;
5722 
5723   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5724   // Check interrupt before trying to wait
5725   if (Thread::is_interrupted(thread, false)) {
5726     return;
5727   }
5728 
5729   // Next, demultiplex/decode time arguments
5730   timespec absTime;
5731   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5732     return;
5733   }
5734   if (time > 0) {
5735     unpackTime(&absTime, isAbsolute, time);
5736   }
5737 
5738 
5739   // Enter safepoint region
5740   // Beware of deadlocks such as 6317397.
5741   // The per-thread Parker:: mutex is a classic leaf-lock.
5742   // In particular a thread must never block on the Threads_lock while
5743   // holding the Parker:: mutex.  If safepoints are pending both the
5744   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5745   ThreadBlockInVM tbivm(jt);
5746 
5747   // Don't wait if cannot get lock since interference arises from
5748   // unblocking.  Also. check interrupt before trying wait
5749   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5750     return;
5751   }
5752 
5753   int status;
5754   if (_counter > 0)  { // no wait needed
5755     _counter = 0;
5756     status = pthread_mutex_unlock(_mutex);
5757     assert(status == 0, "invariant");
5758     // Paranoia to ensure our locked and lock-free paths interact
5759     // correctly with each other and Java-level accesses.
5760     OrderAccess::fence();
5761     return;
5762   }
5763 
5764 #ifdef ASSERT
5765   // Don't catch signals while blocked; let the running threads have the signals.
5766   // (This allows a debugger to break into the running thread.)
5767   sigset_t oldsigs;
5768   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5769   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5770 #endif
5771 
5772   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5773   jt->set_suspend_equivalent();
5774   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5775 
5776   assert(_cur_index == -1, "invariant");
5777   if (time == 0) {
5778     _cur_index = REL_INDEX; // arbitrary choice when not timed
5779     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5780   } else {
5781     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5782     status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5783     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5784       pthread_cond_destroy(&_cond[_cur_index]);
5785       pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5786     }
5787   }
5788   _cur_index = -1;
5789   assert_status(status == 0 || status == EINTR ||
5790                 status == ETIME || status == ETIMEDOUT,
5791                 status, "cond_timedwait");
5792 
5793 #ifdef ASSERT
5794   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5795 #endif
5796 
5797   _counter = 0;
5798   status = pthread_mutex_unlock(_mutex);
5799   assert_status(status == 0, status, "invariant");
5800   // Paranoia to ensure our locked and lock-free paths interact
5801   // correctly with each other and Java-level accesses.
5802   OrderAccess::fence();
5803 
5804   // If externally suspended while waiting, re-suspend
5805   if (jt->handle_special_suspend_equivalent_condition()) {
5806     jt->java_suspend_self();
5807   }
5808 }
5809 
5810 void Parker::unpark() {
5811   int status = pthread_mutex_lock(_mutex);
5812   assert(status == 0, "invariant");
5813   const int s = _counter;
5814   _counter = 1;
5815   if (s < 1) {
5816     // thread might be parked
5817     if (_cur_index != -1) {
5818       // thread is definitely parked
5819       if (WorkAroundNPTLTimedWaitHang) {
5820         status = pthread_cond_signal(&_cond[_cur_index]);
5821         assert(status == 0, "invariant");
5822         status = pthread_mutex_unlock(_mutex);
5823         assert(status == 0, "invariant");
5824       } else {
5825         status = pthread_mutex_unlock(_mutex);
5826         assert(status == 0, "invariant");
5827         status = pthread_cond_signal(&_cond[_cur_index]);
5828         assert(status == 0, "invariant");
5829       }
5830     } else {
5831       pthread_mutex_unlock(_mutex);
5832       assert(status == 0, "invariant");
5833     }
5834   } else {
5835     pthread_mutex_unlock(_mutex);
5836     assert(status == 0, "invariant");
5837   }
5838 }
5839 
5840 
5841 extern char** environ;
5842 
5843 #ifndef __NR_fork
5844   #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) AARCH64_ONLY(1079)
5845 #endif
5846 
5847 #ifndef __NR_execve
5848   #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) AARCH64_ONLY(221)
5849 #endif
5850 
5851 // Run the specified command in a separate process. Return its exit value,
5852 // or -1 on failure (e.g. can't fork a new process).
5853 // Unlike system(), this function can be called from signal handler. It
5854 // doesn't block SIGINT et al.
5855 int os::fork_and_exec(char* cmd) {
5856   const char * argv[4] = {"sh", "-c", cmd, NULL};
5857 
5858   // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5859   // pthread_atfork handlers and reset pthread library. All we need is a
5860   // separate process to execve. Make a direct syscall to fork process.
5861   // On IA64 there's no fork syscall, we have to use fork() and hope for
5862   // the best...
5863   pid_t pid = NOT_IA64(syscall(__NR_fork);)
5864   IA64_ONLY(fork();)
5865 
5866   if (pid < 0) {
5867     // fork failed
5868     return -1;
5869 
5870   } else if (pid == 0) {
5871     // child process
5872 
5873     // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5874     // first to kill every thread on the thread list. Because this list is
5875     // not reset by fork() (see notes above), execve() will instead kill
5876     // every thread in the parent process. We know this is the only thread
5877     // in the new process, so make a system call directly.
5878     // IA64 should use normal execve() from glibc to match the glibc fork()
5879     // above.
5880     NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5881     IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5882 
5883     // execve failed
5884     _exit(-1);
5885 
5886   } else  {
5887     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5888     // care about the actual exit code, for now.
5889 
5890     int status;
5891 
5892     // Wait for the child process to exit.  This returns immediately if
5893     // the child has already exited. */
5894     while (waitpid(pid, &status, 0) < 0) {
5895       switch (errno) {
5896       case ECHILD: return 0;
5897       case EINTR: break;
5898       default: return -1;
5899       }
5900     }
5901 
5902     if (WIFEXITED(status)) {
5903       // The child exited normally; get its exit code.
5904       return WEXITSTATUS(status);
5905     } else if (WIFSIGNALED(status)) {
5906       // The child exited because of a signal
5907       // The best value to return is 0x80 + signal number,
5908       // because that is what all Unix shells do, and because
5909       // it allows callers to distinguish between process exit and
5910       // process death by signal.
5911       return 0x80 + WTERMSIG(status);
5912     } else {
5913       // Unknown exit code; pass it through
5914       return status;
5915     }
5916   }
5917 }
5918 
5919 // is_headless_jre()
5920 //
5921 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5922 // in order to report if we are running in a headless jre
5923 //
5924 // Since JDK8 xawt/libmawt.so was moved into the same directory
5925 // as libawt.so, and renamed libawt_xawt.so
5926 //
5927 bool os::is_headless_jre() {
5928   struct stat statbuf;
5929   char buf[MAXPATHLEN];
5930   char libmawtpath[MAXPATHLEN];
5931   const char *xawtstr  = "/xawt/libmawt.so";
5932   const char *new_xawtstr = "/libawt_xawt.so";
5933   char *p;
5934 
5935   // Get path to libjvm.so
5936   os::jvm_path(buf, sizeof(buf));
5937 
5938   // Get rid of libjvm.so
5939   p = strrchr(buf, '/');
5940   if (p == NULL) {
5941     return false;
5942   } else {
5943     *p = '\0';
5944   }
5945 
5946   // Get rid of client or server
5947   p = strrchr(buf, '/');
5948   if (p == NULL) {
5949     return false;
5950   } else {
5951     *p = '\0';
5952   }
5953 
5954   // check xawt/libmawt.so
5955   strcpy(libmawtpath, buf);
5956   strcat(libmawtpath, xawtstr);
5957   if (::stat(libmawtpath, &statbuf) == 0) return false;
5958 
5959   // check libawt_xawt.so
5960   strcpy(libmawtpath, buf);
5961   strcat(libmawtpath, new_xawtstr);
5962   if (::stat(libmawtpath, &statbuf) == 0) return false;
5963 
5964   return true;
5965 }
5966 
5967 // Get the default path to the core file
5968 // Returns the length of the string
5969 int os::get_core_path(char* buffer, size_t bufferSize) {
5970   /*
5971    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5972    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5973    */
5974   const int core_pattern_len = 129;
5975   char core_pattern[core_pattern_len] = {0};
5976 
5977   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5978   if (core_pattern_file != -1) {
5979     ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5980     ::close(core_pattern_file);
5981 
5982     if (ret > 0) {
5983       char *last_char = core_pattern + strlen(core_pattern) - 1;
5984 
5985       if (*last_char == '\n') {
5986         *last_char = '\0';
5987       }
5988     }
5989   }
5990 
5991   if (strlen(core_pattern) == 0) {
5992     return -1;
5993   }
5994 
5995   char *pid_pos = strstr(core_pattern, "%p");
5996   int written;
5997 
5998   if (core_pattern[0] == '/') {
5999     written = jio_snprintf(buffer, bufferSize, core_pattern);
6000   } else {
6001     char cwd[PATH_MAX];
6002 
6003     const char* p = get_current_directory(cwd, PATH_MAX);
6004     if (p == NULL) {
6005       return -1;
6006     }
6007 
6008     if (core_pattern[0] == '|') {
6009       written = jio_snprintf(buffer, bufferSize,
6010                         "\"%s\" (or dumping to %s/core.%d)",
6011                                      &core_pattern[1], p, current_process_id());
6012     } else {
6013       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6014     }
6015   }
6016 
6017   if (written < 0) {
6018     return -1;
6019   }
6020 
6021   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6022     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6023 
6024     if (core_uses_pid_file != -1) {
6025       char core_uses_pid = 0;
6026       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6027       ::close(core_uses_pid_file);
6028 
6029       if (core_uses_pid == '1') {
6030         jio_snprintf(buffer + written, bufferSize - written,
6031                                           ".%d", current_process_id());
6032       }
6033     }
6034   }
6035 
6036   return strlen(buffer);
6037 }
6038 
6039 /////////////// Unit tests ///////////////
6040 
6041 #ifndef PRODUCT
6042 
6043 #define test_log(...)              \
6044   do {                             \
6045     if (VerboseInternalVMTests) {  \
6046       tty->print_cr(__VA_ARGS__);  \
6047       tty->flush();                \
6048     }                              \
6049   } while (false)
6050 
6051 class TestReserveMemorySpecial : AllStatic {
6052  public:
6053   static void small_page_write(void* addr, size_t size) {
6054     size_t page_size = os::vm_page_size();
6055 
6056     char* end = (char*)addr + size;
6057     for (char* p = (char*)addr; p < end; p += page_size) {
6058       *p = 1;
6059     }
6060   }
6061 
6062   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6063     if (!UseHugeTLBFS) {
6064       return;
6065     }
6066 
6067     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6068 
6069     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6070 
6071     if (addr != NULL) {
6072       small_page_write(addr, size);
6073 
6074       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6075     }
6076   }
6077 
6078   static void test_reserve_memory_special_huge_tlbfs_only() {
6079     if (!UseHugeTLBFS) {
6080       return;
6081     }
6082 
6083     size_t lp = os::large_page_size();
6084 
6085     for (size_t size = lp; size <= lp * 10; size += lp) {
6086       test_reserve_memory_special_huge_tlbfs_only(size);
6087     }
6088   }
6089 
6090   static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6091     if (!UseHugeTLBFS) {
6092       return;
6093     }
6094 
6095     test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6096              size, alignment);
6097 
6098     assert(size >= os::large_page_size(), "Incorrect input to test");
6099 
6100     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6101 
6102     if (addr != NULL) {
6103       small_page_write(addr, size);
6104 
6105       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6106     }
6107   }
6108 
6109   static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6110     size_t lp = os::large_page_size();
6111     size_t ag = os::vm_allocation_granularity();
6112 
6113     for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6114       test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6115     }
6116   }
6117 
6118   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6119     size_t lp = os::large_page_size();
6120     size_t ag = os::vm_allocation_granularity();
6121 
6122     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6123     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6124     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6125     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6126     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6127     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6128     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6129     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6130     test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6131   }
6132 
6133   static void test_reserve_memory_special_huge_tlbfs() {
6134     if (!UseHugeTLBFS) {
6135       return;
6136     }
6137 
6138     test_reserve_memory_special_huge_tlbfs_only();
6139     test_reserve_memory_special_huge_tlbfs_mixed();
6140   }
6141 
6142   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6143     if (!UseSHM) {
6144       return;
6145     }
6146 
6147     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6148 
6149     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6150 
6151     if (addr != NULL) {
6152       assert(is_ptr_aligned(addr, alignment), "Check");
6153       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6154 
6155       small_page_write(addr, size);
6156 
6157       os::Linux::release_memory_special_shm(addr, size);
6158     }
6159   }
6160 
6161   static void test_reserve_memory_special_shm() {
6162     size_t lp = os::large_page_size();
6163     size_t ag = os::vm_allocation_granularity();
6164 
6165     for (size_t size = ag; size < lp * 3; size += ag) {
6166       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6167         test_reserve_memory_special_shm(size, alignment);
6168       }
6169     }
6170   }
6171 
6172   static void test() {
6173     test_reserve_memory_special_huge_tlbfs();
6174     test_reserve_memory_special_shm();
6175   }
6176 };
6177 
6178 void TestReserveMemorySpecial_test() {
6179   TestReserveMemorySpecial::test();
6180 }
6181 
6182 #endif