src/os/linux/vm/os_linux.cpp

Thu, 24 May 2018 18:41:44 +0800

author
aoqi
date
Thu, 24 May 2018 18:41:44 +0800
changeset 8856
ac27a9c85bea
parent 8776
4a575a49e938
parent 8604
04d83ba48607
child 9041
95a08233f46c
permissions
-rw-r--r--

Merge

     1 /*
     2  * Copyright (c) 1999, 2016, 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  */
    25 // no precompiled headers
    26 #include "classfile/classLoader.hpp"
    27 #include "classfile/systemDictionary.hpp"
    28 #include "classfile/vmSymbols.hpp"
    29 #include "code/icBuffer.hpp"
    30 #include "code/vtableStubs.hpp"
    31 #include "compiler/compileBroker.hpp"
    32 #include "compiler/disassembler.hpp"
    33 #include "interpreter/interpreter.hpp"
    34 #include "jvm_linux.h"
    35 #include "memory/allocation.inline.hpp"
    36 #include "memory/filemap.hpp"
    37 #include "mutex_linux.inline.hpp"
    38 #include "oops/oop.inline.hpp"
    39 #include "os_share_linux.hpp"
    40 #include "prims/jniFastGetField.hpp"
    41 #include "prims/jvm.h"
    42 #include "prims/jvm_misc.hpp"
    43 #include "runtime/arguments.hpp"
    44 #include "runtime/extendedPC.hpp"
    45 #include "runtime/globals.hpp"
    46 #include "runtime/interfaceSupport.hpp"
    47 #include "runtime/init.hpp"
    48 #include "runtime/java.hpp"
    49 #include "runtime/javaCalls.hpp"
    50 #include "runtime/mutexLocker.hpp"
    51 #include "runtime/objectMonitor.hpp"
    52 #include "runtime/orderAccess.inline.hpp"
    53 #include "runtime/osThread.hpp"
    54 #include "runtime/perfMemory.hpp"
    55 #include "runtime/sharedRuntime.hpp"
    56 #include "runtime/statSampler.hpp"
    57 #include "runtime/stubRoutines.hpp"
    58 #include "runtime/thread.inline.hpp"
    59 #include "runtime/threadCritical.hpp"
    60 #include "runtime/timer.hpp"
    61 #include "services/attachListener.hpp"
    62 #include "services/memTracker.hpp"
    63 #include "services/runtimeService.hpp"
    64 #include "utilities/decoder.hpp"
    65 #include "utilities/defaultStream.hpp"
    66 #include "utilities/events.hpp"
    67 #include "utilities/elfFile.hpp"
    68 #include "utilities/growableArray.hpp"
    69 #include "utilities/vmError.hpp"
    71 // put OS-includes here
    72 # include <sys/types.h>
    73 # include <sys/mman.h>
    74 # include <sys/stat.h>
    75 # include <sys/select.h>
    76 # include <pthread.h>
    77 # include <signal.h>
    78 # include <errno.h>
    79 # include <dlfcn.h>
    80 # include <stdio.h>
    81 # include <unistd.h>
    82 # include <sys/resource.h>
    83 # include <pthread.h>
    84 # include <sys/stat.h>
    85 # include <sys/time.h>
    86 # include <sys/times.h>
    87 # include <sys/utsname.h>
    88 # include <sys/socket.h>
    89 # include <sys/wait.h>
    90 # include <pwd.h>
    91 # include <poll.h>
    92 # include <semaphore.h>
    93 # include <fcntl.h>
    94 # include <string.h>
    95 # include <syscall.h>
    96 # include <sys/sysinfo.h>
    97 # include <gnu/libc-version.h>
    98 # include <sys/ipc.h>
    99 # include <sys/shm.h>
   100 # include <link.h>
   101 # include <stdint.h>
   102 # include <inttypes.h>
   103 # include <sys/ioctl.h>
   105 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
   107 #ifndef _GNU_SOURCE
   108   #define _GNU_SOURCE
   109   #include <sched.h>
   110   #undef _GNU_SOURCE
   111 #else
   112   #include <sched.h>
   113 #endif
   115 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
   116 // getrusage() is prepared to handle the associated failure.
   117 #ifndef RUSAGE_THREAD
   118 #define RUSAGE_THREAD   (1)               /* only the calling thread */
   119 #endif
   121 #define MAX_PATH    (2 * K)
   123 #define MAX_SECS 100000000
   125 // for timer info max values which include all bits
   126 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
   128 #define LARGEPAGES_BIT (1 << 6)
   129 ////////////////////////////////////////////////////////////////////////////////
   130 // global variables
   131 julong os::Linux::_physical_memory = 0;
   133 address   os::Linux::_initial_thread_stack_bottom = NULL;
   134 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
   136 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
   137 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
   138 Mutex* os::Linux::_createThread_lock = NULL;
   139 pthread_t os::Linux::_main_thread;
   140 int os::Linux::_page_size = -1;
   141 const int os::Linux::_vm_default_page_size = (8 * K);
   142 bool os::Linux::_is_floating_stack = false;
   143 bool os::Linux::_is_NPTL = false;
   144 bool os::Linux::_supports_fast_thread_cpu_time = false;
   145 const char * os::Linux::_glibc_version = NULL;
   146 const char * os::Linux::_libpthread_version = NULL;
   147 pthread_condattr_t os::Linux::_condattr[1];
   149 static jlong initial_time_count=0;
   151 static int clock_tics_per_sec = 100;
   153 // For diagnostics to print a message once. see run_periodic_checks
   154 static sigset_t check_signal_done;
   155 static bool check_signals = true;
   157 static pid_t _initial_pid = 0;
   159 /* Signal number used to suspend/resume a thread */
   161 /* do not use any signal number less than SIGSEGV, see 4355769 */
   162 static int SR_signum = SIGUSR2;
   163 sigset_t SR_sigset;
   165 /* Used to protect dlsym() calls */
   166 static pthread_mutex_t dl_mutex;
   168 // Declarations
   169 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
   171 // utility functions
   173 static int SR_initialize();
   175 julong os::available_memory() {
   176   return Linux::available_memory();
   177 }
   179 julong os::Linux::available_memory() {
   180   // values in struct sysinfo are "unsigned long"
   181   struct sysinfo si;
   182   sysinfo(&si);
   184   return (julong)si.freeram * si.mem_unit;
   185 }
   187 julong os::physical_memory() {
   188   return Linux::physical_memory();
   189 }
   191 ////////////////////////////////////////////////////////////////////////////////
   192 // environment support
   194 bool os::getenv(const char* name, char* buf, int len) {
   195   const char* val = ::getenv(name);
   196   if (val != NULL && strlen(val) < (size_t)len) {
   197     strcpy(buf, val);
   198     return true;
   199   }
   200   if (len > 0) buf[0] = 0;  // return a null string
   201   return false;
   202 }
   205 // Return true if user is running as root.
   207 bool os::have_special_privileges() {
   208   static bool init = false;
   209   static bool privileges = false;
   210   if (!init) {
   211     privileges = (getuid() != geteuid()) || (getgid() != getegid());
   212     init = true;
   213   }
   214   return privileges;
   215 }
   218 #ifndef SYS_gettid
   219 // i386: 224, ia64: 1105, amd64: 186, sparc 143
   220   #ifdef __ia64__
   221     #define SYS_gettid 1105
   222   #else
   223     #ifdef __i386__
   224       #define SYS_gettid 224
   225     #else
   226       #ifdef __amd64__
   227         #define SYS_gettid 186
   228       #else
   229         #ifdef __sparc__
   230           #define SYS_gettid 143
   231         #else
   232           #error define gettid for the arch
   233         #endif
   234       #endif
   235     #endif
   236   #endif
   237 #endif
   239 // Cpu architecture string
   240 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
   242 // pid_t gettid()
   243 //
   244 // Returns the kernel thread id of the currently running thread. Kernel
   245 // thread id is used to access /proc.
   246 //
   247 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
   248 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
   249 //
   250 pid_t os::Linux::gettid() {
   251   int rslt = syscall(SYS_gettid);
   252   if (rslt == -1) {
   253      // old kernel, no NPTL support
   254      return getpid();
   255   } else {
   256      return (pid_t)rslt;
   257   }
   258 }
   260 // Most versions of linux have a bug where the number of processors are
   261 // determined by looking at the /proc file system.  In a chroot environment,
   262 // the system call returns 1.  This causes the VM to act as if it is
   263 // a single processor and elide locking (see is_MP() call).
   264 static bool unsafe_chroot_detected = false;
   265 static const char *unstable_chroot_error = "/proc file system not found.\n"
   266                      "Java may be unstable running multithreaded in a chroot "
   267                      "environment on Linux when /proc filesystem is not mounted.";
   269 void os::Linux::initialize_system_info() {
   270   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
   271   if (processor_count() == 1) {
   272     pid_t pid = os::Linux::gettid();
   273     char fname[32];
   274     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
   275     FILE *fp = fopen(fname, "r");
   276     if (fp == NULL) {
   277       unsafe_chroot_detected = true;
   278     } else {
   279       fclose(fp);
   280     }
   281   }
   282   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
   283   assert(processor_count() > 0, "linux error");
   284 }
   286 void os::init_system_properties_values() {
   287   // The next steps are taken in the product version:
   288   //
   289   // Obtain the JAVA_HOME value from the location of libjvm.so.
   290   // This library should be located at:
   291   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
   292   //
   293   // If "/jre/lib/" appears at the right place in the path, then we
   294   // assume libjvm.so is installed in a JDK and we use this path.
   295   //
   296   // Otherwise exit with message: "Could not create the Java virtual machine."
   297   //
   298   // The following extra steps are taken in the debugging version:
   299   //
   300   // If "/jre/lib/" does NOT appear at the right place in the path
   301   // instead of exit check for $JAVA_HOME environment variable.
   302   //
   303   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
   304   // then we append a fake suffix "hotspot/libjvm.so" to this path so
   305   // it looks like libjvm.so is installed there
   306   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
   307   //
   308   // Otherwise exit.
   309   //
   310   // Important note: if the location of libjvm.so changes this
   311   // code needs to be changed accordingly.
   313 // See ld(1):
   314 //      The linker uses the following search paths to locate required
   315 //      shared libraries:
   316 //        1: ...
   317 //        ...
   318 //        7: The default directories, normally /lib and /usr/lib.
   319 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
   320 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
   321 #else
   322 #define DEFAULT_LIBPATH "/lib:/usr/lib"
   323 #endif
   325 // Base path of extensions installed on the system.
   326 #define SYS_EXT_DIR     "/usr/java/packages"
   327 #define EXTENSIONS_DIR  "/lib/ext"
   328 #define ENDORSED_DIR    "/lib/endorsed"
   330   // Buffer that fits several sprintfs.
   331   // Note that the space for the colon and the trailing null are provided
   332   // by the nulls included by the sizeof operator.
   333   const size_t bufsize =
   334     MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
   335          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
   336          (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
   337   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
   339   // sysclasspath, java_home, dll_dir
   340   {
   341     char *pslash;
   342     os::jvm_path(buf, bufsize);
   344     // Found the full path to libjvm.so.
   345     // Now cut the path to <java_home>/jre if we can.
   346     *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
   347     pslash = strrchr(buf, '/');
   348     if (pslash != NULL) {
   349       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
   350     }
   351     Arguments::set_dll_dir(buf);
   353     if (pslash != NULL) {
   354       pslash = strrchr(buf, '/');
   355       if (pslash != NULL) {
   356         *pslash = '\0';          // Get rid of /<arch>.
   357         pslash = strrchr(buf, '/');
   358         if (pslash != NULL) {
   359           *pslash = '\0';        // Get rid of /lib.
   360         }
   361       }
   362     }
   363     Arguments::set_java_home(buf);
   364     set_boot_path('/', ':');
   365   }
   367   // Where to look for native libraries.
   368   //
   369   // Note: Due to a legacy implementation, most of the library path
   370   // is set in the launcher. This was to accomodate linking restrictions
   371   // on legacy Linux implementations (which are no longer supported).
   372   // Eventually, all the library path setting will be done here.
   373   //
   374   // However, to prevent the proliferation of improperly built native
   375   // libraries, the new path component /usr/java/packages is added here.
   376   // Eventually, all the library path setting will be done here.
   377   {
   378     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
   379     // should always exist (until the legacy problem cited above is
   380     // addressed).
   381     const char *v = ::getenv("LD_LIBRARY_PATH");
   382     const char *v_colon = ":";
   383     if (v == NULL) { v = ""; v_colon = ""; }
   384     // That's +1 for the colon and +1 for the trailing '\0'.
   385     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
   386                                                      strlen(v) + 1 +
   387                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
   388                                                      mtInternal);
   389     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
   390     Arguments::set_library_path(ld_library_path);
   391     FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
   392   }
   394   // Extensions directories.
   395   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
   396   Arguments::set_ext_dirs(buf);
   398   // Endorsed standards default directory.
   399   sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
   400   Arguments::set_endorsed_dirs(buf);
   402   FREE_C_HEAP_ARRAY(char, buf, mtInternal);
   404 #undef DEFAULT_LIBPATH
   405 #undef SYS_EXT_DIR
   406 #undef EXTENSIONS_DIR
   407 #undef ENDORSED_DIR
   408 }
   410 ////////////////////////////////////////////////////////////////////////////////
   411 // breakpoint support
   413 void os::breakpoint() {
   414   BREAKPOINT;
   415 }
   417 extern "C" void breakpoint() {
   418   // use debugger to set breakpoint here
   419 }
   421 ////////////////////////////////////////////////////////////////////////////////
   422 // signal support
   424 debug_only(static bool signal_sets_initialized = false);
   425 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
   427 bool os::Linux::is_sig_ignored(int sig) {
   428       struct sigaction oact;
   429       sigaction(sig, (struct sigaction*)NULL, &oact);
   430       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
   431                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
   432       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
   433            return true;
   434       else
   435            return false;
   436 }
   438 void os::Linux::signal_sets_init() {
   439   // Should also have an assertion stating we are still single-threaded.
   440   assert(!signal_sets_initialized, "Already initialized");
   441   // Fill in signals that are necessarily unblocked for all threads in
   442   // the VM. Currently, we unblock the following signals:
   443   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
   444   //                         by -Xrs (=ReduceSignalUsage));
   445   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
   446   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
   447   // the dispositions or masks wrt these signals.
   448   // Programs embedding the VM that want to use the above signals for their
   449   // own purposes must, at this time, use the "-Xrs" option to prevent
   450   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
   451   // (See bug 4345157, and other related bugs).
   452   // In reality, though, unblocking these signals is really a nop, since
   453   // these signals are not blocked by default.
   454   sigemptyset(&unblocked_sigs);
   455   sigemptyset(&allowdebug_blocked_sigs);
   456   sigaddset(&unblocked_sigs, SIGILL);
   457   sigaddset(&unblocked_sigs, SIGSEGV);
   458   sigaddset(&unblocked_sigs, SIGBUS);
   459   sigaddset(&unblocked_sigs, SIGFPE);
   460 #if defined(PPC64)
   461   sigaddset(&unblocked_sigs, SIGTRAP);
   462 #endif
   463   sigaddset(&unblocked_sigs, SR_signum);
   465   if (!ReduceSignalUsage) {
   466    if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
   467       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
   468       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
   469    }
   470    if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
   471       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
   472       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
   473    }
   474    if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
   475       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
   476       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
   477    }
   478   }
   479   // Fill in signals that are blocked by all but the VM thread.
   480   sigemptyset(&vm_sigs);
   481   if (!ReduceSignalUsage)
   482     sigaddset(&vm_sigs, BREAK_SIGNAL);
   483   debug_only(signal_sets_initialized = true);
   485 }
   487 // These are signals that are unblocked while a thread is running Java.
   488 // (For some reason, they get blocked by default.)
   489 sigset_t* os::Linux::unblocked_signals() {
   490   assert(signal_sets_initialized, "Not initialized");
   491   return &unblocked_sigs;
   492 }
   494 // These are the signals that are blocked while a (non-VM) thread is
   495 // running Java. Only the VM thread handles these signals.
   496 sigset_t* os::Linux::vm_signals() {
   497   assert(signal_sets_initialized, "Not initialized");
   498   return &vm_sigs;
   499 }
   501 // These are signals that are blocked during cond_wait to allow debugger in
   502 sigset_t* os::Linux::allowdebug_blocked_signals() {
   503   assert(signal_sets_initialized, "Not initialized");
   504   return &allowdebug_blocked_sigs;
   505 }
   507 void os::Linux::hotspot_sigmask(Thread* thread) {
   509   //Save caller's signal mask before setting VM signal mask
   510   sigset_t caller_sigmask;
   511   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
   513   OSThread* osthread = thread->osthread();
   514   osthread->set_caller_sigmask(caller_sigmask);
   516   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
   518   if (!ReduceSignalUsage) {
   519     if (thread->is_VM_thread()) {
   520       // Only the VM thread handles BREAK_SIGNAL ...
   521       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
   522     } else {
   523       // ... all other threads block BREAK_SIGNAL
   524       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
   525     }
   526   }
   527 }
   529 //////////////////////////////////////////////////////////////////////////////
   530 // detecting pthread library
   532 void os::Linux::libpthread_init() {
   533   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
   534   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
   535   // generic name for earlier versions.
   536   // Define macros here so we can build HotSpot on old systems.
   537 # ifndef _CS_GNU_LIBC_VERSION
   538 # define _CS_GNU_LIBC_VERSION 2
   539 # endif
   540 # ifndef _CS_GNU_LIBPTHREAD_VERSION
   541 # define _CS_GNU_LIBPTHREAD_VERSION 3
   542 # endif
   544   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
   545   if (n > 0) {
   546      char *str = (char *)malloc(n, mtInternal);
   547      confstr(_CS_GNU_LIBC_VERSION, str, n);
   548      os::Linux::set_glibc_version(str);
   549   } else {
   550      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
   551      static char _gnu_libc_version[32];
   552      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
   553               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
   554      os::Linux::set_glibc_version(_gnu_libc_version);
   555   }
   557   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
   558   if (n > 0) {
   559      char *str = (char *)malloc(n, mtInternal);
   560      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
   561      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
   562      // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
   563      // is the case. LinuxThreads has a hard limit on max number of threads.
   564      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
   565      // On the other hand, NPTL does not have such a limit, sysconf()
   566      // will return -1 and errno is not changed. Check if it is really NPTL.
   567      if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
   568          strstr(str, "NPTL") &&
   569          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
   570        free(str);
   571        os::Linux::set_libpthread_version("linuxthreads");
   572      } else {
   573        os::Linux::set_libpthread_version(str);
   574      }
   575   } else {
   576     // glibc before 2.3.2 only has LinuxThreads.
   577     os::Linux::set_libpthread_version("linuxthreads");
   578   }
   580   if (strstr(libpthread_version(), "NPTL")) {
   581      os::Linux::set_is_NPTL();
   582   } else {
   583      os::Linux::set_is_LinuxThreads();
   584   }
   586   // LinuxThreads have two flavors: floating-stack mode, which allows variable
   587   // stack size; and fixed-stack mode. NPTL is always floating-stack.
   588   if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
   589      os::Linux::set_is_floating_stack();
   590   }
   591 }
   593 /////////////////////////////////////////////////////////////////////////////
   594 // thread stack
   596 // Force Linux kernel to expand current thread stack. If "bottom" is close
   597 // to the stack guard, caller should block all signals.
   598 //
   599 // MAP_GROWSDOWN:
   600 //   A special mmap() flag that is used to implement thread stacks. It tells
   601 //   kernel that the memory region should extend downwards when needed. This
   602 //   allows early versions of LinuxThreads to only mmap the first few pages
   603 //   when creating a new thread. Linux kernel will automatically expand thread
   604 //   stack as needed (on page faults).
   605 //
   606 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
   607 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
   608 //   region, it's hard to tell if the fault is due to a legitimate stack
   609 //   access or because of reading/writing non-exist memory (e.g. buffer
   610 //   overrun). As a rule, if the fault happens below current stack pointer,
   611 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
   612 //   application (see Linux kernel fault.c).
   613 //
   614 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
   615 //   stack overflow detection.
   616 //
   617 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
   618 //   not use this flag. However, the stack of initial thread is not created
   619 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
   620 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
   621 //   and then attach the thread to JVM.
   622 //
   623 // To get around the problem and allow stack banging on Linux, we need to
   624 // manually expand thread stack after receiving the SIGSEGV.
   625 //
   626 // There are two ways to expand thread stack to address "bottom", we used
   627 // both of them in JVM before 1.5:
   628 //   1. adjust stack pointer first so that it is below "bottom", and then
   629 //      touch "bottom"
   630 //   2. mmap() the page in question
   631 //
   632 // Now alternate signal stack is gone, it's harder to use 2. For instance,
   633 // if current sp is already near the lower end of page 101, and we need to
   634 // call mmap() to map page 100, it is possible that part of the mmap() frame
   635 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
   636 // That will destroy the mmap() frame and cause VM to crash.
   637 //
   638 // The following code works by adjusting sp first, then accessing the "bottom"
   639 // page to force a page fault. Linux kernel will then automatically expand the
   640 // stack mapping.
   641 //
   642 // _expand_stack_to() assumes its frame size is less than page size, which
   643 // should always be true if the function is not inlined.
   645 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
   646 #define NOINLINE
   647 #else
   648 #define NOINLINE __attribute__ ((noinline))
   649 #endif
   651 static void _expand_stack_to(address bottom) NOINLINE;
   653 static void _expand_stack_to(address bottom) {
   654   address sp;
   655   size_t size;
   656   volatile char *p;
   658   // Adjust bottom to point to the largest address within the same page, it
   659   // gives us a one-page buffer if alloca() allocates slightly more memory.
   660   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
   661   bottom += os::Linux::page_size() - 1;
   663   // sp might be slightly above current stack pointer; if that's the case, we
   664   // will alloca() a little more space than necessary, which is OK. Don't use
   665   // os::current_stack_pointer(), as its result can be slightly below current
   666   // stack pointer, causing us to not alloca enough to reach "bottom".
   667   sp = (address)&sp;
   669   if (sp > bottom) {
   670     size = sp - bottom;
   671     p = (volatile char *)alloca(size);
   672     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
   673     p[0] = '\0';
   674   }
   675 }
   677 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
   678   assert(t!=NULL, "just checking");
   679   assert(t->osthread()->expanding_stack(), "expand should be set");
   680   assert(t->stack_base() != NULL, "stack_base was not initialized");
   682   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
   683     sigset_t mask_all, old_sigset;
   684     sigfillset(&mask_all);
   685     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
   686     _expand_stack_to(addr);
   687     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
   688     return true;
   689   }
   690   return false;
   691 }
   693 //////////////////////////////////////////////////////////////////////////////
   694 // create new thread
   696 static address highest_vm_reserved_address();
   698 // check if it's safe to start a new thread
   699 static bool _thread_safety_check(Thread* thread) {
   700   if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
   701     // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
   702     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
   703     //   allocated (MAP_FIXED) from high address space. Every thread stack
   704     //   occupies a fixed size slot (usually 2Mbytes, but user can change
   705     //   it to other values if they rebuild LinuxThreads).
   706     //
   707     // Problem with MAP_FIXED is that mmap() can still succeed even part of
   708     // the memory region has already been mmap'ed. That means if we have too
   709     // many threads and/or very large heap, eventually thread stack will
   710     // collide with heap.
   711     //
   712     // Here we try to prevent heap/stack collision by comparing current
   713     // stack bottom with the highest address that has been mmap'ed by JVM
   714     // plus a safety margin for memory maps created by native code.
   715     //
   716     // This feature can be disabled by setting ThreadSafetyMargin to 0
   717     //
   718     if (ThreadSafetyMargin > 0) {
   719       address stack_bottom = os::current_stack_base() - os::current_stack_size();
   721       // not safe if our stack extends below the safety margin
   722       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
   723     } else {
   724       return true;
   725     }
   726   } else {
   727     // Floating stack LinuxThreads or NPTL:
   728     //   Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
   729     //   there's not enough space left, pthread_create() will fail. If we come
   730     //   here, that means enough space has been reserved for stack.
   731     return true;
   732   }
   733 }
   735 // Thread start routine for all newly created threads
   736 static void *java_start(Thread *thread) {
   737   // Try to randomize the cache line index of hot stack frames.
   738   // This helps when threads of the same stack traces evict each other's
   739   // cache lines. The threads can be either from the same JVM instance, or
   740   // from different JVM instances. The benefit is especially true for
   741   // processors with hyperthreading technology.
   742   static int counter = 0;
   743   int pid = os::current_process_id();
   744   alloca(((pid ^ counter++) & 7) * 128);
   746   ThreadLocalStorage::set_thread(thread);
   748   OSThread* osthread = thread->osthread();
   749   Monitor* sync = osthread->startThread_lock();
   751   // non floating stack LinuxThreads needs extra check, see above
   752   if (!_thread_safety_check(thread)) {
   753     // notify parent thread
   754     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
   755     osthread->set_state(ZOMBIE);
   756     sync->notify_all();
   757     return NULL;
   758   }
   760   // thread_id is kernel thread id (similar to Solaris LWP id)
   761   osthread->set_thread_id(os::Linux::gettid());
   763   if (UseNUMA) {
   764     int lgrp_id = os::numa_get_group_id();
   765     if (lgrp_id != -1) {
   766       thread->set_lgrp_id(lgrp_id);
   767     }
   768   }
   769   // initialize signal mask for this thread
   770   os::Linux::hotspot_sigmask(thread);
   772   // initialize floating point control register
   773   os::Linux::init_thread_fpu_state();
   775   // handshaking with parent thread
   776   {
   777     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
   779     // notify parent thread
   780     osthread->set_state(INITIALIZED);
   781     sync->notify_all();
   783     // wait until os::start_thread()
   784     while (osthread->get_state() == INITIALIZED) {
   785       sync->wait(Mutex::_no_safepoint_check_flag);
   786     }
   787   }
   789   // call one more level start routine
   790   thread->run();
   792   return 0;
   793 }
   795 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
   796   assert(thread->osthread() == NULL, "caller responsible");
   798   // Allocate the OSThread object
   799   OSThread* osthread = new OSThread(NULL, NULL);
   800   if (osthread == NULL) {
   801     return false;
   802   }
   804   // set the correct thread state
   805   osthread->set_thread_type(thr_type);
   807   // Initial state is ALLOCATED but not INITIALIZED
   808   osthread->set_state(ALLOCATED);
   810   thread->set_osthread(osthread);
   812   // init thread attributes
   813   pthread_attr_t attr;
   814   pthread_attr_init(&attr);
   815   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
   817   // stack size
   818   if (os::Linux::supports_variable_stack_size()) {
   819     // calculate stack size if it's not specified by caller
   820     if (stack_size == 0) {
   821       stack_size = os::Linux::default_stack_size(thr_type);
   823       switch (thr_type) {
   824       case os::java_thread:
   825         // Java threads use ThreadStackSize which default value can be
   826         // changed with the flag -Xss
   827         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
   828         stack_size = JavaThread::stack_size_at_create();
   829         break;
   830       case os::compiler_thread:
   831         if (CompilerThreadStackSize > 0) {
   832           stack_size = (size_t)(CompilerThreadStackSize * K);
   833           break;
   834         } // else fall through:
   835           // use VMThreadStackSize if CompilerThreadStackSize is not defined
   836       case os::vm_thread:
   837       case os::pgc_thread:
   838       case os::cgc_thread:
   839       case os::watcher_thread:
   840         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
   841         break;
   842       }
   843     }
   845     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
   846     pthread_attr_setstacksize(&attr, stack_size);
   847   } else {
   848     // let pthread_create() pick the default value.
   849   }
   851   // glibc guard page
   852   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
   854   ThreadState state;
   856   {
   857     // Serialize thread creation if we are running with fixed stack LinuxThreads
   858     bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
   859     if (lock) {
   860       os::Linux::createThread_lock()->lock_without_safepoint_check();
   861     }
   863     pthread_t tid;
   864     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
   866     pthread_attr_destroy(&attr);
   868     if (ret != 0) {
   869       if (PrintMiscellaneous && (Verbose || WizardMode)) {
   870         perror("pthread_create()");
   871       }
   872       // Need to clean up stuff we've allocated so far
   873       thread->set_osthread(NULL);
   874       delete osthread;
   875       if (lock) os::Linux::createThread_lock()->unlock();
   876       return false;
   877     }
   879     // Store pthread info into the OSThread
   880     osthread->set_pthread_id(tid);
   882     // Wait until child thread is either initialized or aborted
   883     {
   884       Monitor* sync_with_child = osthread->startThread_lock();
   885       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
   886       while ((state = osthread->get_state()) == ALLOCATED) {
   887         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
   888       }
   889     }
   891     if (lock) {
   892       os::Linux::createThread_lock()->unlock();
   893     }
   894   }
   896   // Aborted due to thread limit being reached
   897   if (state == ZOMBIE) {
   898       thread->set_osthread(NULL);
   899       delete osthread;
   900       return false;
   901   }
   903   // The thread is returned suspended (in state INITIALIZED),
   904   // and is started higher up in the call chain
   905   assert(state == INITIALIZED, "race condition");
   906   return true;
   907 }
   909 /////////////////////////////////////////////////////////////////////////////
   910 // attach existing thread
   912 // bootstrap the main thread
   913 bool os::create_main_thread(JavaThread* thread) {
   914   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
   915   return create_attached_thread(thread);
   916 }
   918 bool os::create_attached_thread(JavaThread* thread) {
   919 #ifdef ASSERT
   920     thread->verify_not_published();
   921 #endif
   923   // Allocate the OSThread object
   924   OSThread* osthread = new OSThread(NULL, NULL);
   926   if (osthread == NULL) {
   927     return false;
   928   }
   930   // Store pthread info into the OSThread
   931   osthread->set_thread_id(os::Linux::gettid());
   932   osthread->set_pthread_id(::pthread_self());
   934   // initialize floating point control register
   935   os::Linux::init_thread_fpu_state();
   937   // Initial thread state is RUNNABLE
   938   osthread->set_state(RUNNABLE);
   940   thread->set_osthread(osthread);
   942   if (UseNUMA) {
   943     int lgrp_id = os::numa_get_group_id();
   944     if (lgrp_id != -1) {
   945       thread->set_lgrp_id(lgrp_id);
   946     }
   947   }
   949   if (os::Linux::is_initial_thread()) {
   950     // If current thread is initial thread, its stack is mapped on demand,
   951     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
   952     // the entire stack region to avoid SEGV in stack banging.
   953     // It is also useful to get around the heap-stack-gap problem on SuSE
   954     // kernel (see 4821821 for details). We first expand stack to the top
   955     // of yellow zone, then enable stack yellow zone (order is significant,
   956     // enabling yellow zone first will crash JVM on SuSE Linux), so there
   957     // is no gap between the last two virtual memory regions.
   959     JavaThread *jt = (JavaThread *)thread;
   960     address addr = jt->stack_yellow_zone_base();
   961     assert(addr != NULL, "initialization problem?");
   962     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
   964     osthread->set_expanding_stack();
   965     os::Linux::manually_expand_stack(jt, addr);
   966     osthread->clear_expanding_stack();
   967   }
   969   // initialize signal mask for this thread
   970   // and save the caller's signal mask
   971   os::Linux::hotspot_sigmask(thread);
   973   return true;
   974 }
   976 void os::pd_start_thread(Thread* thread) {
   977   OSThread * osthread = thread->osthread();
   978   assert(osthread->get_state() != INITIALIZED, "just checking");
   979   Monitor* sync_with_child = osthread->startThread_lock();
   980   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
   981   sync_with_child->notify();
   983 #ifdef MIPS64
   984   /* 2013/11/5 Jin: To be accessed in NativeGeneralJump::patch_verified_entry() */
   985   if (thread->is_Java_thread())
   986   {
   987     ((JavaThread*)thread)->set_handle_wrong_method_stub(SharedRuntime::get_handle_wrong_method_stub());
   988   }
   989 #endif
   990 }
   992 // Free Linux resources related to the OSThread
   993 void os::free_thread(OSThread* osthread) {
   994   assert(osthread != NULL, "osthread not set");
   996   if (Thread::current()->osthread() == osthread) {
   997     // Restore caller's signal mask
   998     sigset_t sigmask = osthread->caller_sigmask();
   999     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
  1002   delete osthread;
  1005 //////////////////////////////////////////////////////////////////////////////
  1006 // thread local storage
  1008 // Restore the thread pointer if the destructor is called. This is in case
  1009 // someone from JNI code sets up a destructor with pthread_key_create to run
  1010 // detachCurrentThread on thread death. Unless we restore the thread pointer we
  1011 // will hang or crash. When detachCurrentThread is called the key will be set
  1012 // to null and we will not be called again. If detachCurrentThread is never
  1013 // called we could loop forever depending on the pthread implementation.
  1014 static void restore_thread_pointer(void* p) {
  1015   Thread* thread = (Thread*) p;
  1016   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
  1019 int os::allocate_thread_local_storage() {
  1020   pthread_key_t key;
  1021   int rslt = pthread_key_create(&key, restore_thread_pointer);
  1022   assert(rslt == 0, "cannot allocate thread local storage");
  1023   return (int)key;
  1026 // Note: This is currently not used by VM, as we don't destroy TLS key
  1027 // on VM exit.
  1028 void os::free_thread_local_storage(int index) {
  1029   int rslt = pthread_key_delete((pthread_key_t)index);
  1030   assert(rslt == 0, "invalid index");
  1033 void os::thread_local_storage_at_put(int index, void* value) {
  1034   int rslt = pthread_setspecific((pthread_key_t)index, value);
  1035   assert(rslt == 0, "pthread_setspecific failed");
  1038 extern "C" Thread* get_thread() {
  1039   return ThreadLocalStorage::thread();
  1042 //////////////////////////////////////////////////////////////////////////////
  1043 // initial thread
  1045 // Check if current thread is the initial thread, similar to Solaris thr_main.
  1046 bool os::Linux::is_initial_thread(void) {
  1047   char dummy;
  1048   // If called before init complete, thread stack bottom will be null.
  1049   // Can be called if fatal error occurs before initialization.
  1050   if (initial_thread_stack_bottom() == NULL) return false;
  1051   assert(initial_thread_stack_bottom() != NULL &&
  1052          initial_thread_stack_size()   != 0,
  1053          "os::init did not locate initial thread's stack region");
  1054   if ((address)&dummy >= initial_thread_stack_bottom() &&
  1055       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
  1056        return true;
  1057   else return false;
  1060 // Find the virtual memory area that contains addr
  1061 static bool find_vma(address addr, address* vma_low, address* vma_high) {
  1062   FILE *fp = fopen("/proc/self/maps", "r");
  1063   if (fp) {
  1064     address low, high;
  1065     while (!feof(fp)) {
  1066       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
  1067         if (low <= addr && addr < high) {
  1068            if (vma_low)  *vma_low  = low;
  1069            if (vma_high) *vma_high = high;
  1070            fclose (fp);
  1071            return true;
  1074       for (;;) {
  1075         int ch = fgetc(fp);
  1076         if (ch == EOF || ch == (int)'\n') break;
  1079     fclose(fp);
  1081   return false;
  1084 // Locate initial thread stack. This special handling of initial thread stack
  1085 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
  1086 // bogus value for the primordial process thread. While the launcher has created
  1087 // the VM in a new thread since JDK 6, we still have to allow for the use of the
  1088 // JNI invocation API from a primordial thread.
  1089 void os::Linux::capture_initial_stack(size_t max_size) {
  1091   // max_size is either 0 (which means accept OS default for thread stacks) or
  1092   // a user-specified value known to be at least the minimum needed. If we
  1093   // are actually on the primordial thread we can make it appear that we have a
  1094   // smaller max_size stack by inserting the guard pages at that location. But we
  1095   // cannot do anything to emulate a larger stack than what has been provided by
  1096   // the OS or threading library. In fact if we try to use a stack greater than
  1097   // what is set by rlimit then we will crash the hosting process.
  1099   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
  1100   // If this is "unlimited" then it will be a huge value.
  1101   struct rlimit rlim;
  1102   getrlimit(RLIMIT_STACK, &rlim);
  1103   size_t stack_size = rlim.rlim_cur;
  1105   // 6308388: a bug in ld.so will relocate its own .data section to the
  1106   //   lower end of primordial stack; reduce ulimit -s value a little bit
  1107   //   so we won't install guard page on ld.so's data section.
  1108   stack_size -= 2 * page_size();
  1110   // Try to figure out where the stack base (top) is. This is harder.
  1111   //
  1112   // When an application is started, glibc saves the initial stack pointer in
  1113   // a global variable "__libc_stack_end", which is then used by system
  1114   // libraries. __libc_stack_end should be pretty close to stack top. The
  1115   // variable is available since the very early days. However, because it is
  1116   // a private interface, it could disappear in the future.
  1117   //
  1118   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
  1119   // to __libc_stack_end, it is very close to stack top, but isn't the real
  1120   // stack top. Note that /proc may not exist if VM is running as a chroot
  1121   // program, so reading /proc/<pid>/stat could fail. Also the contents of
  1122   // /proc/<pid>/stat could change in the future (though unlikely).
  1123   //
  1124   // We try __libc_stack_end first. If that doesn't work, look for
  1125   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
  1126   // as a hint, which should work well in most cases.
  1128   uintptr_t stack_start;
  1130   // try __libc_stack_end first
  1131   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
  1132   if (p && *p) {
  1133     stack_start = *p;
  1134   } else {
  1135     // see if we can get the start_stack field from /proc/self/stat
  1136     FILE *fp;
  1137     int pid;
  1138     char state;
  1139     int ppid;
  1140     int pgrp;
  1141     int session;
  1142     int nr;
  1143     int tpgrp;
  1144     unsigned long flags;
  1145     unsigned long minflt;
  1146     unsigned long cminflt;
  1147     unsigned long majflt;
  1148     unsigned long cmajflt;
  1149     unsigned long utime;
  1150     unsigned long stime;
  1151     long cutime;
  1152     long cstime;
  1153     long prio;
  1154     long nice;
  1155     long junk;
  1156     long it_real;
  1157     uintptr_t start;
  1158     uintptr_t vsize;
  1159     intptr_t rss;
  1160     uintptr_t rsslim;
  1161     uintptr_t scodes;
  1162     uintptr_t ecode;
  1163     int i;
  1165     // Figure what the primordial thread stack base is. Code is inspired
  1166     // by email from Hans Boehm. /proc/self/stat begins with current pid,
  1167     // followed by command name surrounded by parentheses, state, etc.
  1168     char stat[2048];
  1169     int statlen;
  1171     fp = fopen("/proc/self/stat", "r");
  1172     if (fp) {
  1173       statlen = fread(stat, 1, 2047, fp);
  1174       stat[statlen] = '\0';
  1175       fclose(fp);
  1177       // Skip pid and the command string. Note that we could be dealing with
  1178       // weird command names, e.g. user could decide to rename java launcher
  1179       // to "java 1.4.2 :)", then the stat file would look like
  1180       //                1234 (java 1.4.2 :)) R ... ...
  1181       // We don't really need to know the command string, just find the last
  1182       // occurrence of ")" and then start parsing from there. See bug 4726580.
  1183       char * s = strrchr(stat, ')');
  1185       i = 0;
  1186       if (s) {
  1187         // Skip blank chars
  1188         do s++; while (isspace(*s));
  1190 #define _UFM UINTX_FORMAT
  1191 #define _DFM INTX_FORMAT
  1193         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
  1194         /*              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 */
  1195         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,
  1196              &state,          /* 3  %c  */
  1197              &ppid,           /* 4  %d  */
  1198              &pgrp,           /* 5  %d  */
  1199              &session,        /* 6  %d  */
  1200              &nr,             /* 7  %d  */
  1201              &tpgrp,          /* 8  %d  */
  1202              &flags,          /* 9  %lu  */
  1203              &minflt,         /* 10 %lu  */
  1204              &cminflt,        /* 11 %lu  */
  1205              &majflt,         /* 12 %lu  */
  1206              &cmajflt,        /* 13 %lu  */
  1207              &utime,          /* 14 %lu  */
  1208              &stime,          /* 15 %lu  */
  1209              &cutime,         /* 16 %ld  */
  1210              &cstime,         /* 17 %ld  */
  1211              &prio,           /* 18 %ld  */
  1212              &nice,           /* 19 %ld  */
  1213              &junk,           /* 20 %ld  */
  1214              &it_real,        /* 21 %ld  */
  1215              &start,          /* 22 UINTX_FORMAT */
  1216              &vsize,          /* 23 UINTX_FORMAT */
  1217              &rss,            /* 24 INTX_FORMAT  */
  1218              &rsslim,         /* 25 UINTX_FORMAT */
  1219              &scodes,         /* 26 UINTX_FORMAT */
  1220              &ecode,          /* 27 UINTX_FORMAT */
  1221              &stack_start);   /* 28 UINTX_FORMAT */
  1224 #undef _UFM
  1225 #undef _DFM
  1227       if (i != 28 - 2) {
  1228          assert(false, "Bad conversion from /proc/self/stat");
  1229          // product mode - assume we are the initial thread, good luck in the
  1230          // embedded case.
  1231          warning("Can't detect initial thread stack location - bad conversion");
  1232          stack_start = (uintptr_t) &rlim;
  1234     } else {
  1235       // For some reason we can't open /proc/self/stat (for example, running on
  1236       // FreeBSD with a Linux emulator, or inside chroot), this should work for
  1237       // most cases, so don't abort:
  1238       warning("Can't detect initial thread stack location - no /proc/self/stat");
  1239       stack_start = (uintptr_t) &rlim;
  1243   // Now we have a pointer (stack_start) very close to the stack top, the
  1244   // next thing to do is to figure out the exact location of stack top. We
  1245   // can find out the virtual memory area that contains stack_start by
  1246   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
  1247   // and its upper limit is the real stack top. (again, this would fail if
  1248   // running inside chroot, because /proc may not exist.)
  1250   uintptr_t stack_top;
  1251   address low, high;
  1252   if (find_vma((address)stack_start, &low, &high)) {
  1253     // success, "high" is the true stack top. (ignore "low", because initial
  1254     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
  1255     stack_top = (uintptr_t)high;
  1256   } else {
  1257     // failed, likely because /proc/self/maps does not exist
  1258     warning("Can't detect initial thread stack location - find_vma failed");
  1259     // best effort: stack_start is normally within a few pages below the real
  1260     // stack top, use it as stack top, and reduce stack size so we won't put
  1261     // guard page outside stack.
  1262     stack_top = stack_start;
  1263     stack_size -= 16 * page_size();
  1266   // stack_top could be partially down the page so align it
  1267   stack_top = align_size_up(stack_top, page_size());
  1269   // Allowed stack value is minimum of max_size and what we derived from rlimit
  1270   if (max_size > 0) {
  1271     _initial_thread_stack_size = MIN2(max_size, stack_size);
  1272   } else {
  1273     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
  1274     // clamp it at 8MB as we do on Solaris
  1275     _initial_thread_stack_size = MIN2(stack_size, 8*M);
  1278   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
  1279   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
  1280   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
  1283 ////////////////////////////////////////////////////////////////////////////////
  1284 // time support
  1286 // Time since start-up in seconds to a fine granularity.
  1287 // Used by VMSelfDestructTimer and the MemProfiler.
  1288 double os::elapsedTime() {
  1290   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
  1293 jlong os::elapsed_counter() {
  1294   return javaTimeNanos() - initial_time_count;
  1297 jlong os::elapsed_frequency() {
  1298   return NANOSECS_PER_SEC; // nanosecond resolution
  1301 bool os::supports_vtime() { return true; }
  1302 bool os::enable_vtime()   { return false; }
  1303 bool os::vtime_enabled()  { return false; }
  1305 double os::elapsedVTime() {
  1306   struct rusage usage;
  1307   int retval = getrusage(RUSAGE_THREAD, &usage);
  1308   if (retval == 0) {
  1309     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);
  1310   } else {
  1311     // better than nothing, but not much
  1312     return elapsedTime();
  1316 jlong os::javaTimeMillis() {
  1317   timeval time;
  1318   int status = gettimeofday(&time, NULL);
  1319   assert(status != -1, "linux error");
  1320   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
  1323 #ifndef CLOCK_MONOTONIC
  1324 #define CLOCK_MONOTONIC (1)
  1325 #endif
  1327 void os::Linux::clock_init() {
  1328   // we do dlopen's in this particular order due to bug in linux
  1329   // dynamical loader (see 6348968) leading to crash on exit
  1330   void* handle = dlopen("librt.so.1", RTLD_LAZY);
  1331   if (handle == NULL) {
  1332     handle = dlopen("librt.so", RTLD_LAZY);
  1335   if (handle) {
  1336     int (*clock_getres_func)(clockid_t, struct timespec*) =
  1337            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
  1338     int (*clock_gettime_func)(clockid_t, struct timespec*) =
  1339            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
  1340     if (clock_getres_func && clock_gettime_func) {
  1341       // See if monotonic clock is supported by the kernel. Note that some
  1342       // early implementations simply return kernel jiffies (updated every
  1343       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
  1344       // for nano time (though the monotonic property is still nice to have).
  1345       // It's fixed in newer kernels, however clock_getres() still returns
  1346       // 1/HZ. We check if clock_getres() works, but will ignore its reported
  1347       // resolution for now. Hopefully as people move to new kernels, this
  1348       // won't be a problem.
  1349       struct timespec res;
  1350       struct timespec tp;
  1351       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
  1352           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
  1353         // yes, monotonic clock is supported
  1354         _clock_gettime = clock_gettime_func;
  1355         return;
  1356       } else {
  1357         // close librt if there is no monotonic clock
  1358         dlclose(handle);
  1362   warning("No monotonic clock was available - timed services may " \
  1363           "be adversely affected if the time-of-day clock changes");
  1366 #ifndef SYS_clock_getres
  1368 #if defined(IA32) || defined(AMD64)
  1369 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
  1370 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
  1371 #else
  1372 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
  1373 #define sys_clock_getres(x,y)  -1
  1374 #endif
  1376 #else
  1377 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
  1378 #endif
  1380 void os::Linux::fast_thread_clock_init() {
  1381   if (!UseLinuxPosixThreadCPUClocks) {
  1382     return;
  1384   clockid_t clockid;
  1385   struct timespec tp;
  1386   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
  1387       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
  1389   // Switch to using fast clocks for thread cpu time if
  1390   // the sys_clock_getres() returns 0 error code.
  1391   // Note, that some kernels may support the current thread
  1392   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
  1393   // returned by the pthread_getcpuclockid().
  1394   // If the fast Posix clocks are supported then the sys_clock_getres()
  1395   // must return at least tp.tv_sec == 0 which means a resolution
  1396   // better than 1 sec. This is extra check for reliability.
  1398   if(pthread_getcpuclockid_func &&
  1399      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
  1400      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
  1402     _supports_fast_thread_cpu_time = true;
  1403     _pthread_getcpuclockid = pthread_getcpuclockid_func;
  1407 jlong os::javaTimeNanos() {
  1408   if (Linux::supports_monotonic_clock()) {
  1409     struct timespec tp;
  1410     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
  1411     assert(status == 0, "gettime error");
  1412     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
  1413     return result;
  1414   } else {
  1415     timeval time;
  1416     int status = gettimeofday(&time, NULL);
  1417     assert(status != -1, "linux error");
  1418     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
  1419     return 1000 * usecs;
  1423 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
  1424   if (Linux::supports_monotonic_clock()) {
  1425     info_ptr->max_value = ALL_64_BITS;
  1427     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
  1428     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
  1429     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
  1430   } else {
  1431     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
  1432     info_ptr->max_value = ALL_64_BITS;
  1434     // gettimeofday is a real time clock so it skips
  1435     info_ptr->may_skip_backward = true;
  1436     info_ptr->may_skip_forward = true;
  1439   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
  1442 // Return the real, user, and system times in seconds from an
  1443 // arbitrary fixed point in the past.
  1444 bool os::getTimesSecs(double* process_real_time,
  1445                       double* process_user_time,
  1446                       double* process_system_time) {
  1447   struct tms ticks;
  1448   clock_t real_ticks = times(&ticks);
  1450   if (real_ticks == (clock_t) (-1)) {
  1451     return false;
  1452   } else {
  1453     double ticks_per_second = (double) clock_tics_per_sec;
  1454     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
  1455     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
  1456     *process_real_time = ((double) real_ticks) / ticks_per_second;
  1458     return true;
  1463 char * os::local_time_string(char *buf, size_t buflen) {
  1464   struct tm t;
  1465   time_t long_time;
  1466   time(&long_time);
  1467   localtime_r(&long_time, &t);
  1468   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
  1469                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
  1470                t.tm_hour, t.tm_min, t.tm_sec);
  1471   return buf;
  1474 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
  1475   return localtime_r(clock, res);
  1478 ////////////////////////////////////////////////////////////////////////////////
  1479 // runtime exit support
  1481 // Note: os::shutdown() might be called very early during initialization, or
  1482 // called from signal handler. Before adding something to os::shutdown(), make
  1483 // sure it is async-safe and can handle partially initialized VM.
  1484 void os::shutdown() {
  1486   // allow PerfMemory to attempt cleanup of any persistent resources
  1487   perfMemory_exit();
  1489   // needs to remove object in file system
  1490   AttachListener::abort();
  1492   // flush buffered output, finish log files
  1493   ostream_abort();
  1495   // Check for abort hook
  1496   abort_hook_t abort_hook = Arguments::abort_hook();
  1497   if (abort_hook != NULL) {
  1498     abort_hook();
  1503 // Note: os::abort() might be called very early during initialization, or
  1504 // called from signal handler. Before adding something to os::abort(), make
  1505 // sure it is async-safe and can handle partially initialized VM.
  1506 void os::abort(bool dump_core) {
  1507   os::shutdown();
  1508   if (dump_core) {
  1509 #ifndef PRODUCT
  1510     fdStream out(defaultStream::output_fd());
  1511     out.print_raw("Current thread is ");
  1512     char buf[16];
  1513     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
  1514     out.print_raw_cr(buf);
  1515     out.print_raw_cr("Dumping core ...");
  1516 #endif
  1517     ::abort(); // dump core
  1520   ::exit(1);
  1523 // Die immediately, no exit hook, no abort hook, no cleanup.
  1524 void os::die() {
  1525   // _exit() on LinuxThreads only kills current thread
  1526   ::abort();
  1530 // This method is a copy of JDK's sysGetLastErrorString
  1531 // from src/solaris/hpi/src/system_md.c
  1533 size_t os::lasterror(char *buf, size_t len) {
  1535   if (errno == 0)  return 0;
  1537   const char *s = ::strerror(errno);
  1538   size_t n = ::strlen(s);
  1539   if (n >= len) {
  1540     n = len - 1;
  1542   ::strncpy(buf, s, n);
  1543   buf[n] = '\0';
  1544   return n;
  1547 intx os::current_thread_id() { return (intx)pthread_self(); }
  1548 int os::current_process_id() {
  1550   // Under the old linux thread library, linux gives each thread
  1551   // its own process id. Because of this each thread will return
  1552   // a different pid if this method were to return the result
  1553   // of getpid(2). Linux provides no api that returns the pid
  1554   // of the launcher thread for the vm. This implementation
  1555   // returns a unique pid, the pid of the launcher thread
  1556   // that starts the vm 'process'.
  1558   // Under the NPTL, getpid() returns the same pid as the
  1559   // launcher thread rather than a unique pid per thread.
  1560   // Use gettid() if you want the old pre NPTL behaviour.
  1562   // if you are looking for the result of a call to getpid() that
  1563   // returns a unique pid for the calling thread, then look at the
  1564   // OSThread::thread_id() method in osThread_linux.hpp file
  1566   return (int)(_initial_pid ? _initial_pid : getpid());
  1569 // DLL functions
  1571 const char* os::dll_file_extension() { return ".so"; }
  1573 // This must be hard coded because it's the system's temporary
  1574 // directory not the java application's temp directory, ala java.io.tmpdir.
  1575 const char* os::get_temp_directory() { return "/tmp"; }
  1577 static bool file_exists(const char* filename) {
  1578   struct stat statbuf;
  1579   if (filename == NULL || strlen(filename) == 0) {
  1580     return false;
  1582   return os::stat(filename, &statbuf) == 0;
  1585 bool os::dll_build_name(char* buffer, size_t buflen,
  1586                         const char* pname, const char* fname) {
  1587   bool retval = false;
  1588   // Copied from libhpi
  1589   const size_t pnamelen = pname ? strlen(pname) : 0;
  1591   // Return error on buffer overflow.
  1592   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
  1593     return retval;
  1596   if (pnamelen == 0) {
  1597     snprintf(buffer, buflen, "lib%s.so", fname);
  1598     retval = true;
  1599   } else if (strchr(pname, *os::path_separator()) != NULL) {
  1600     int n;
  1601     char** pelements = split_path(pname, &n);
  1602     if (pelements == NULL) {
  1603       return false;
  1605     for (int i = 0 ; i < n ; i++) {
  1606       // Really shouldn't be NULL, but check can't hurt
  1607       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
  1608         continue; // skip the empty path values
  1610       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
  1611       if (file_exists(buffer)) {
  1612         retval = true;
  1613         break;
  1616     // release the storage
  1617     for (int i = 0 ; i < n ; i++) {
  1618       if (pelements[i] != NULL) {
  1619         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
  1622     if (pelements != NULL) {
  1623       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
  1625   } else {
  1626     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
  1627     retval = true;
  1629   return retval;
  1632 // check if addr is inside libjvm.so
  1633 bool os::address_is_in_vm(address addr) {
  1634   static address libjvm_base_addr;
  1635   Dl_info dlinfo;
  1637   if (libjvm_base_addr == NULL) {
  1638     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
  1639       libjvm_base_addr = (address)dlinfo.dli_fbase;
  1641     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
  1644   if (dladdr((void *)addr, &dlinfo) != 0) {
  1645     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
  1648   return false;
  1651 bool os::dll_address_to_function_name(address addr, char *buf,
  1652                                       int buflen, int *offset) {
  1653   // buf is not optional, but offset is optional
  1654   assert(buf != NULL, "sanity check");
  1656   Dl_info dlinfo;
  1658   if (dladdr((void*)addr, &dlinfo) != 0) {
  1659     // see if we have a matching symbol
  1660     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
  1661       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
  1662         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
  1664       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
  1665       return true;
  1667     // no matching symbol so try for just file info
  1668     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
  1669       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
  1670                           buf, buflen, offset, dlinfo.dli_fname)) {
  1671         return true;
  1676   buf[0] = '\0';
  1677   if (offset != NULL) *offset = -1;
  1678   return false;
  1681 struct _address_to_library_name {
  1682   address addr;          // input : memory address
  1683   size_t  buflen;        //         size of fname
  1684   char*   fname;         // output: library name
  1685   address base;          //         library base addr
  1686 };
  1688 static int address_to_library_name_callback(struct dl_phdr_info *info,
  1689                                             size_t size, void *data) {
  1690   int i;
  1691   bool found = false;
  1692   address libbase = NULL;
  1693   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
  1695   // iterate through all loadable segments
  1696   for (i = 0; i < info->dlpi_phnum; i++) {
  1697     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
  1698     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
  1699       // base address of a library is the lowest address of its loaded
  1700       // segments.
  1701       if (libbase == NULL || libbase > segbase) {
  1702         libbase = segbase;
  1704       // see if 'addr' is within current segment
  1705       if (segbase <= d->addr &&
  1706           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
  1707         found = true;
  1712   // dlpi_name is NULL or empty if the ELF file is executable, return 0
  1713   // so dll_address_to_library_name() can fall through to use dladdr() which
  1714   // can figure out executable name from argv[0].
  1715   if (found && info->dlpi_name && info->dlpi_name[0]) {
  1716     d->base = libbase;
  1717     if (d->fname) {
  1718       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
  1720     return 1;
  1722   return 0;
  1725 bool os::dll_address_to_library_name(address addr, char* buf,
  1726                                      int buflen, int* offset) {
  1727   // buf is not optional, but offset is optional
  1728   assert(buf != NULL, "sanity check");
  1730   Dl_info dlinfo;
  1731   struct _address_to_library_name data;
  1733   // There is a bug in old glibc dladdr() implementation that it could resolve
  1734   // to wrong library name if the .so file has a base address != NULL. Here
  1735   // we iterate through the program headers of all loaded libraries to find
  1736   // out which library 'addr' really belongs to. This workaround can be
  1737   // removed once the minimum requirement for glibc is moved to 2.3.x.
  1738   data.addr = addr;
  1739   data.fname = buf;
  1740   data.buflen = buflen;
  1741   data.base = NULL;
  1742   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
  1744   if (rslt) {
  1745      // buf already contains library name
  1746      if (offset) *offset = addr - data.base;
  1747      return true;
  1749   if (dladdr((void*)addr, &dlinfo) != 0) {
  1750     if (dlinfo.dli_fname != NULL) {
  1751       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
  1753     if (dlinfo.dli_fbase != NULL && offset != NULL) {
  1754       *offset = addr - (address)dlinfo.dli_fbase;
  1756     return true;
  1759   buf[0] = '\0';
  1760   if (offset) *offset = -1;
  1761   return false;
  1764   // Loads .dll/.so and
  1765   // in case of error it checks if .dll/.so was built for the
  1766   // same architecture as Hotspot is running on
  1769 // Remember the stack's state. The Linux dynamic linker will change
  1770 // the stack to 'executable' at most once, so we must safepoint only once.
  1771 bool os::Linux::_stack_is_executable = false;
  1773 // VM operation that loads a library.  This is necessary if stack protection
  1774 // of the Java stacks can be lost during loading the library.  If we
  1775 // do not stop the Java threads, they can stack overflow before the stacks
  1776 // are protected again.
  1777 class VM_LinuxDllLoad: public VM_Operation {
  1778  private:
  1779   const char *_filename;
  1780   char *_ebuf;
  1781   int _ebuflen;
  1782   void *_lib;
  1783  public:
  1784   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
  1785     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
  1786   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
  1787   void doit() {
  1788     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
  1789     os::Linux::_stack_is_executable = true;
  1791   void* loaded_library() { return _lib; }
  1792 };
  1794 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
  1796   void * result = NULL;
  1797   bool load_attempted = false;
  1799   // Check whether the library to load might change execution rights
  1800   // of the stack. If they are changed, the protection of the stack
  1801   // guard pages will be lost. We need a safepoint to fix this.
  1802   //
  1803   // See Linux man page execstack(8) for more info.
  1804   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
  1805     ElfFile ef(filename);
  1806     if (!ef.specifies_noexecstack()) {
  1807       if (!is_init_completed()) {
  1808         os::Linux::_stack_is_executable = true;
  1809         // This is OK - No Java threads have been created yet, and hence no
  1810         // stack guard pages to fix.
  1811         //
  1812         // This should happen only when you are building JDK7 using a very
  1813         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
  1814         //
  1815         // Dynamic loader will make all stacks executable after
  1816         // this function returns, and will not do that again.
  1817         assert(Threads::first() == NULL, "no Java threads should exist yet.");
  1818       } else {
  1819         warning("You have loaded library %s which might have disabled stack guard. "
  1820                 "The VM will try to fix the stack guard now.\n"
  1821                 "It's highly recommended that you fix the library with "
  1822                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
  1823                 filename);
  1825         assert(Thread::current()->is_Java_thread(), "must be Java thread");
  1826         JavaThread *jt = JavaThread::current();
  1827         if (jt->thread_state() != _thread_in_native) {
  1828           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
  1829           // that requires ExecStack. Cannot enter safe point. Let's give up.
  1830           warning("Unable to fix stack guard. Giving up.");
  1831         } else {
  1832           if (!LoadExecStackDllInVMThread) {
  1833             // This is for the case where the DLL has an static
  1834             // constructor function that executes JNI code. We cannot
  1835             // load such DLLs in the VMThread.
  1836             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
  1839           ThreadInVMfromNative tiv(jt);
  1840           debug_only(VMNativeEntryWrapper vew;)
  1842           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
  1843           VMThread::execute(&op);
  1844           if (LoadExecStackDllInVMThread) {
  1845             result = op.loaded_library();
  1847           load_attempted = true;
  1853   if (!load_attempted) {
  1854     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
  1857   if (result != NULL) {
  1858     // Successful loading
  1859     return result;
  1862   Elf32_Ehdr elf_head;
  1863   int diag_msg_max_length=ebuflen-strlen(ebuf);
  1864   char* diag_msg_buf=ebuf+strlen(ebuf);
  1866   if (diag_msg_max_length==0) {
  1867     // No more space in ebuf for additional diagnostics message
  1868     return NULL;
  1872   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
  1874   if (file_descriptor < 0) {
  1875     // Can't open library, report dlerror() message
  1876     return NULL;
  1879   bool failed_to_read_elf_head=
  1880     (sizeof(elf_head)!=
  1881         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
  1883   ::close(file_descriptor);
  1884   if (failed_to_read_elf_head) {
  1885     // file i/o error - report dlerror() msg
  1886     return NULL;
  1889   typedef struct {
  1890     Elf32_Half  code;         // Actual value as defined in elf.h
  1891     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
  1892     char        elf_class;    // 32 or 64 bit
  1893     char        endianess;    // MSB or LSB
  1894     char*       name;         // String representation
  1895   } arch_t;
  1897   #ifndef EM_486
  1898   #define EM_486          6               /* Intel 80486 */
  1899   #endif
  1901   static const arch_t arch_array[]={
  1902     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
  1903     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
  1904     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
  1905     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
  1906     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
  1907     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
  1908     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
  1909     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
  1910 #if defined(VM_LITTLE_ENDIAN)
  1911     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
  1912 #else
  1913     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
  1914 #endif
  1915     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
  1916     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
  1917     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
  1918     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
  1919     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
  1920     {EM_MIPS,        EM_MIPS,    ELFCLASS64, ELFDATA2LSB, (char*)"MIPS64 LE"},
  1921     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
  1922     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
  1923   };
  1925   #if  (defined IA32)
  1926     static  Elf32_Half running_arch_code=EM_386;
  1927   #elif   (defined AMD64)
  1928     static  Elf32_Half running_arch_code=EM_X86_64;
  1929   #elif  (defined IA64)
  1930     static  Elf32_Half running_arch_code=EM_IA_64;
  1931   #elif  (defined __sparc) && (defined _LP64)
  1932     static  Elf32_Half running_arch_code=EM_SPARCV9;
  1933   #elif  (defined __sparc) && (!defined _LP64)
  1934     static  Elf32_Half running_arch_code=EM_SPARC;
  1935   #elif  (defined MIPS64)
  1936     static  Elf32_Half running_arch_code=EM_MIPS;
  1937   #elif  (defined __powerpc64__)
  1938     static  Elf32_Half running_arch_code=EM_PPC64;
  1939   #elif  (defined __powerpc__)
  1940     static  Elf32_Half running_arch_code=EM_PPC;
  1941   #elif  (defined ARM)
  1942     static  Elf32_Half running_arch_code=EM_ARM;
  1943   #elif  (defined S390)
  1944     static  Elf32_Half running_arch_code=EM_S390;
  1945   #elif  (defined ALPHA)
  1946     static  Elf32_Half running_arch_code=EM_ALPHA;
  1947   #elif  (defined MIPSEL)
  1948     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
  1949   #elif  (defined PARISC)
  1950     static  Elf32_Half running_arch_code=EM_PARISC;
  1951   #elif  (defined MIPS)
  1952     static  Elf32_Half running_arch_code=EM_MIPS;
  1953   #elif  (defined M68K)
  1954     static  Elf32_Half running_arch_code=EM_68K;
  1955   #else
  1956     #error Method os::dll_load requires that one of following is defined:\
  1957          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, __mips64, PARISC, M68K
  1958   #endif
  1960   // Identify compatability class for VM's architecture and library's architecture
  1961   // Obtain string descriptions for architectures
  1963   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
  1964   int running_arch_index=-1;
  1966   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
  1967     if (running_arch_code == arch_array[i].code) {
  1968       running_arch_index    = i;
  1970     if (lib_arch.code == arch_array[i].code) {
  1971       lib_arch.compat_class = arch_array[i].compat_class;
  1972       lib_arch.name         = arch_array[i].name;
  1976   assert(running_arch_index != -1,
  1977     "Didn't find running architecture code (running_arch_code) in arch_array");
  1978   if (running_arch_index == -1) {
  1979     // Even though running architecture detection failed
  1980     // we may still continue with reporting dlerror() message
  1981     return NULL;
  1984   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
  1985     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
  1986     return NULL;
  1989 #ifndef S390
  1990   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
  1991     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
  1992     return NULL;
  1994 #endif // !S390
  1996   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
  1997     if ( lib_arch.name!=NULL ) {
  1998       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
  1999         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
  2000         lib_arch.name, arch_array[running_arch_index].name);
  2001     } else {
  2002       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
  2003       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
  2004         lib_arch.code,
  2005         arch_array[running_arch_index].name);
  2009   return NULL;
  2012 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
  2013   void * result = ::dlopen(filename, RTLD_LAZY);
  2014   if (result == NULL) {
  2015     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
  2016     ebuf[ebuflen-1] = '\0';
  2018   return result;
  2021 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
  2022   void * result = NULL;
  2023   if (LoadExecStackDllInVMThread) {
  2024     result = dlopen_helper(filename, ebuf, ebuflen);
  2027   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
  2028   // library that requires an executable stack, or which does not have this
  2029   // stack attribute set, dlopen changes the stack attribute to executable. The
  2030   // read protection of the guard pages gets lost.
  2031   //
  2032   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
  2033   // may have been queued at the same time.
  2035   if (!_stack_is_executable) {
  2036     JavaThread *jt = Threads::first();
  2038     while (jt) {
  2039       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
  2040           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
  2041         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
  2042                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
  2043           warning("Attempt to reguard stack yellow zone failed.");
  2046       jt = jt->next();
  2050   return result;
  2053 /*
  2054  * glibc-2.0 libdl is not MT safe.  If you are building with any glibc,
  2055  * chances are you might want to run the generated bits against glibc-2.0
  2056  * libdl.so, so always use locking for any version of glibc.
  2057  */
  2058 void* os::dll_lookup(void* handle, const char* name) {
  2059   pthread_mutex_lock(&dl_mutex);
  2060   void* res = dlsym(handle, name);
  2061   pthread_mutex_unlock(&dl_mutex);
  2062   return res;
  2065 void* os::get_default_process_handle() {
  2066   return (void*)::dlopen(NULL, RTLD_LAZY);
  2069 static bool _print_ascii_file(const char* filename, outputStream* st) {
  2070   int fd = ::open(filename, O_RDONLY);
  2071   if (fd == -1) {
  2072      return false;
  2075   char buf[32];
  2076   int bytes;
  2077   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
  2078     st->print_raw(buf, bytes);
  2081   ::close(fd);
  2083   return true;
  2086 void os::print_dll_info(outputStream *st) {
  2087    st->print_cr("Dynamic libraries:");
  2089    char fname[32];
  2090    pid_t pid = os::Linux::gettid();
  2092    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
  2094    if (!_print_ascii_file(fname, st)) {
  2095      st->print("Can not get library information for pid = %d\n", pid);
  2099 void os::print_os_info_brief(outputStream* st) {
  2100   os::Linux::print_distro_info(st);
  2102   os::Posix::print_uname_info(st);
  2104   os::Linux::print_libversion_info(st);
  2108 void os::print_os_info(outputStream* st) {
  2109   st->print("OS:");
  2111   os::Linux::print_distro_info(st);
  2113   os::Posix::print_uname_info(st);
  2115   // Print warning if unsafe chroot environment detected
  2116   if (unsafe_chroot_detected) {
  2117     st->print("WARNING!! ");
  2118     st->print_cr("%s", unstable_chroot_error);
  2121   os::Linux::print_libversion_info(st);
  2123   os::Posix::print_rlimit_info(st);
  2125   os::Posix::print_load_average(st);
  2127   os::Linux::print_full_memory_info(st);
  2130 // Try to identify popular distros.
  2131 // Most Linux distributions have a /etc/XXX-release file, which contains
  2132 // the OS version string. Newer Linux distributions have a /etc/lsb-release
  2133 // file that also contains the OS version string. Some have more than one
  2134 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
  2135 // /etc/redhat-release.), so the order is important.
  2136 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
  2137 // their own specific XXX-release file as well as a redhat-release file.
  2138 // Because of this the XXX-release file needs to be searched for before the
  2139 // redhat-release file.
  2140 // Since Red Hat has a lsb-release file that is not very descriptive the
  2141 // search for redhat-release needs to be before lsb-release.
  2142 // Since the lsb-release file is the new standard it needs to be searched
  2143 // before the older style release files.
  2144 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
  2145 // next to last resort.  The os-release file is a new standard that contains
  2146 // distribution information and the system-release file seems to be an old
  2147 // standard that has been replaced by the lsb-release and os-release files.
  2148 // Searching for the debian_version file is the last resort.  It contains
  2149 // an informative string like "6.0.6" or "wheezy/sid". Because of this
  2150 // "Debian " is printed before the contents of the debian_version file.
  2151 void os::Linux::print_distro_info(outputStream* st) {
  2152    if (!_print_ascii_file("/etc/oracle-release", st) &&
  2153        !_print_ascii_file("/etc/mandriva-release", st) &&
  2154        !_print_ascii_file("/etc/mandrake-release", st) &&
  2155        !_print_ascii_file("/etc/sun-release", st) &&
  2156        !_print_ascii_file("/etc/redhat-release", st) &&
  2157        !_print_ascii_file("/etc/lsb-release", st) &&
  2158        !_print_ascii_file("/etc/SuSE-release", st) &&
  2159        !_print_ascii_file("/etc/turbolinux-release", st) &&
  2160        !_print_ascii_file("/etc/gentoo-release", st) &&
  2161        !_print_ascii_file("/etc/ltib-release", st) &&
  2162        !_print_ascii_file("/etc/angstrom-version", st) &&
  2163        !_print_ascii_file("/etc/system-release", st) &&
  2164        !_print_ascii_file("/etc/os-release", st)) {
  2166        if (file_exists("/etc/debian_version")) {
  2167          st->print("Debian ");
  2168          _print_ascii_file("/etc/debian_version", st);
  2169        } else {
  2170          st->print("Linux");
  2173    st->cr();
  2176 void os::Linux::print_libversion_info(outputStream* st) {
  2177   // libc, pthread
  2178   st->print("libc:");
  2179   st->print("%s ", os::Linux::glibc_version());
  2180   st->print("%s ", os::Linux::libpthread_version());
  2181   if (os::Linux::is_LinuxThreads()) {
  2182      st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
  2184   st->cr();
  2187 void os::Linux::print_full_memory_info(outputStream* st) {
  2188    st->print("\n/proc/meminfo:\n");
  2189    _print_ascii_file("/proc/meminfo", st);
  2190    st->cr();
  2193 void os::print_memory_info(outputStream* st) {
  2195   st->print("Memory:");
  2196   st->print(" %dk page", os::vm_page_size()>>10);
  2198   // values in struct sysinfo are "unsigned long"
  2199   struct sysinfo si;
  2200   sysinfo(&si);
  2202   st->print(", physical " UINT64_FORMAT "k",
  2203             os::physical_memory() >> 10);
  2204   st->print("(" UINT64_FORMAT "k free)",
  2205             os::available_memory() >> 10);
  2206   st->print(", swap " UINT64_FORMAT "k",
  2207             ((jlong)si.totalswap * si.mem_unit) >> 10);
  2208   st->print("(" UINT64_FORMAT "k free)",
  2209             ((jlong)si.freeswap * si.mem_unit) >> 10);
  2210   st->cr();
  2213 void os::pd_print_cpu_info(outputStream* st) {
  2214   st->print("\n/proc/cpuinfo:\n");
  2215   if (!_print_ascii_file("/proc/cpuinfo", st)) {
  2216     st->print("  <Not Available>");
  2218   st->cr();
  2221 void os::print_siginfo(outputStream* st, void* siginfo) {
  2222   const siginfo_t* si = (const siginfo_t*)siginfo;
  2224   os::Posix::print_siginfo_brief(st, si);
  2225 #if INCLUDE_CDS
  2226   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
  2227       UseSharedSpaces) {
  2228     FileMapInfo* mapinfo = FileMapInfo::current_info();
  2229     if (mapinfo->is_in_shared_space(si->si_addr)) {
  2230       st->print("\n\nError accessing class data sharing archive."   \
  2231                 " Mapped file inaccessible during execution, "      \
  2232                 " possible disk/network problem.");
  2235 #endif
  2236   st->cr();
  2240 static void print_signal_handler(outputStream* st, int sig,
  2241                                  char* buf, size_t buflen);
  2243 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
  2244   st->print_cr("Signal Handlers:");
  2245   print_signal_handler(st, SIGSEGV, buf, buflen);
  2246   print_signal_handler(st, SIGBUS , buf, buflen);
  2247   print_signal_handler(st, SIGFPE , buf, buflen);
  2248   print_signal_handler(st, SIGPIPE, buf, buflen);
  2249   print_signal_handler(st, SIGXFSZ, buf, buflen);
  2250   print_signal_handler(st, SIGILL , buf, buflen);
  2251   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
  2252   print_signal_handler(st, SR_signum, buf, buflen);
  2253   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
  2254   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
  2255   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
  2256   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
  2257 #if defined(PPC64)
  2258   print_signal_handler(st, SIGTRAP, buf, buflen);
  2259 #endif
  2262 static char saved_jvm_path[MAXPATHLEN] = {0};
  2264 // Find the full path to the current module, libjvm.so
  2265 void os::jvm_path(char *buf, jint buflen) {
  2266   // Error checking.
  2267   if (buflen < MAXPATHLEN) {
  2268     assert(false, "must use a large-enough buffer");
  2269     buf[0] = '\0';
  2270     return;
  2272   // Lazy resolve the path to current module.
  2273   if (saved_jvm_path[0] != 0) {
  2274     strcpy(buf, saved_jvm_path);
  2275     return;
  2278   char dli_fname[MAXPATHLEN];
  2279   bool ret = dll_address_to_library_name(
  2280                 CAST_FROM_FN_PTR(address, os::jvm_path),
  2281                 dli_fname, sizeof(dli_fname), NULL);
  2282   assert(ret, "cannot locate libjvm");
  2283   char *rp = NULL;
  2284   if (ret && dli_fname[0] != '\0') {
  2285     rp = realpath(dli_fname, buf);
  2287   if (rp == NULL)
  2288     return;
  2290   if (Arguments::created_by_gamma_launcher()) {
  2291     // Support for the gamma launcher.  Typical value for buf is
  2292     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
  2293     // the right place in the string, then assume we are installed in a JDK and
  2294     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
  2295     // up the path so it looks like libjvm.so is installed there (append a
  2296     // fake suffix hotspot/libjvm.so).
  2297     const char *p = buf + strlen(buf) - 1;
  2298     for (int count = 0; p > buf && count < 5; ++count) {
  2299       for (--p; p > buf && *p != '/'; --p)
  2300         /* empty */ ;
  2303     if (strncmp(p, "/jre/lib/", 9) != 0) {
  2304       // Look for JAVA_HOME in the environment.
  2305       char* java_home_var = ::getenv("JAVA_HOME");
  2306       if (java_home_var != NULL && java_home_var[0] != 0) {
  2307         char* jrelib_p;
  2308         int len;
  2310         // Check the current module name "libjvm.so".
  2311         p = strrchr(buf, '/');
  2312         assert(strstr(p, "/libjvm") == p, "invalid library name");
  2314         rp = realpath(java_home_var, buf);
  2315         if (rp == NULL)
  2316           return;
  2318         // determine if this is a legacy image or modules image
  2319         // modules image doesn't have "jre" subdirectory
  2320         len = strlen(buf);
  2321         assert(len < buflen, "Ran out of buffer room");
  2322         jrelib_p = buf + len;
  2323         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
  2324         if (0 != access(buf, F_OK)) {
  2325           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
  2328         if (0 == access(buf, F_OK)) {
  2329           // Use current module name "libjvm.so"
  2330           len = strlen(buf);
  2331           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
  2332         } else {
  2333           // Go back to path of .so
  2334           rp = realpath(dli_fname, buf);
  2335           if (rp == NULL)
  2336             return;
  2342   strncpy(saved_jvm_path, buf, MAXPATHLEN);
  2345 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
  2346   // no prefix required, not even "_"
  2349 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
  2350   // no suffix required
  2353 ////////////////////////////////////////////////////////////////////////////////
  2354 // sun.misc.Signal support
  2356 static volatile jint sigint_count = 0;
  2358 static void
  2359 UserHandler(int sig, void *siginfo, void *context) {
  2360   // 4511530 - sem_post is serialized and handled by the manager thread. When
  2361   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
  2362   // don't want to flood the manager thread with sem_post requests.
  2363   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
  2364       return;
  2366   // Ctrl-C is pressed during error reporting, likely because the error
  2367   // handler fails to abort. Let VM die immediately.
  2368   if (sig == SIGINT && is_error_reported()) {
  2369      os::die();
  2372   os::signal_notify(sig);
  2375 void* os::user_handler() {
  2376   return CAST_FROM_FN_PTR(void*, UserHandler);
  2379 class Semaphore : public StackObj {
  2380   public:
  2381     Semaphore();
  2382     ~Semaphore();
  2383     void signal();
  2384     void wait();
  2385     bool trywait();
  2386     bool timedwait(unsigned int sec, int nsec);
  2387   private:
  2388     sem_t _semaphore;
  2389 };
  2391 Semaphore::Semaphore() {
  2392   sem_init(&_semaphore, 0, 0);
  2395 Semaphore::~Semaphore() {
  2396   sem_destroy(&_semaphore);
  2399 void Semaphore::signal() {
  2400   sem_post(&_semaphore);
  2403 void Semaphore::wait() {
  2404   sem_wait(&_semaphore);
  2407 bool Semaphore::trywait() {
  2408   return sem_trywait(&_semaphore) == 0;
  2411 bool Semaphore::timedwait(unsigned int sec, int nsec) {
  2413   struct timespec ts;
  2414   // Semaphore's are always associated with CLOCK_REALTIME
  2415   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
  2416   // see unpackTime for discussion on overflow checking
  2417   if (sec >= MAX_SECS) {
  2418     ts.tv_sec += MAX_SECS;
  2419     ts.tv_nsec = 0;
  2420   } else {
  2421     ts.tv_sec += sec;
  2422     ts.tv_nsec += nsec;
  2423     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
  2424       ts.tv_nsec -= NANOSECS_PER_SEC;
  2425       ++ts.tv_sec; // note: this must be <= max_secs
  2429   while (1) {
  2430     int result = sem_timedwait(&_semaphore, &ts);
  2431     if (result == 0) {
  2432       return true;
  2433     } else if (errno == EINTR) {
  2434       continue;
  2435     } else if (errno == ETIMEDOUT) {
  2436       return false;
  2437     } else {
  2438       return false;
  2443 extern "C" {
  2444   typedef void (*sa_handler_t)(int);
  2445   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
  2448 void* os::signal(int signal_number, void* handler) {
  2449   struct sigaction sigAct, oldSigAct;
  2451   sigfillset(&(sigAct.sa_mask));
  2452   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
  2453   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
  2455   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
  2456     // -1 means registration failed
  2457     return (void *)-1;
  2460   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
  2463 void os::signal_raise(int signal_number) {
  2464   ::raise(signal_number);
  2467 /*
  2468  * The following code is moved from os.cpp for making this
  2469  * code platform specific, which it is by its very nature.
  2470  */
  2472 // Will be modified when max signal is changed to be dynamic
  2473 int os::sigexitnum_pd() {
  2474   return NSIG;
  2477 // a counter for each possible signal value
  2478 static volatile jint pending_signals[NSIG+1] = { 0 };
  2480 // Linux(POSIX) specific hand shaking semaphore.
  2481 static sem_t sig_sem;
  2482 static Semaphore sr_semaphore;
  2484 void os::signal_init_pd() {
  2485   // Initialize signal structures
  2486   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
  2488   // Initialize signal semaphore
  2489   ::sem_init(&sig_sem, 0, 0);
  2492 void os::signal_notify(int sig) {
  2493   Atomic::inc(&pending_signals[sig]);
  2494   ::sem_post(&sig_sem);
  2497 static int check_pending_signals(bool wait) {
  2498   Atomic::store(0, &sigint_count);
  2499   for (;;) {
  2500     for (int i = 0; i < NSIG + 1; i++) {
  2501       jint n = pending_signals[i];
  2502       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
  2503         return i;
  2506     if (!wait) {
  2507       return -1;
  2509     JavaThread *thread = JavaThread::current();
  2510     ThreadBlockInVM tbivm(thread);
  2512     bool threadIsSuspended;
  2513     do {
  2514       thread->set_suspend_equivalent();
  2515       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
  2516       ::sem_wait(&sig_sem);
  2518       // were we externally suspended while we were waiting?
  2519       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
  2520       if (threadIsSuspended) {
  2521         //
  2522         // The semaphore has been incremented, but while we were waiting
  2523         // another thread suspended us. We don't want to continue running
  2524         // while suspended because that would surprise the thread that
  2525         // suspended us.
  2526         //
  2527         ::sem_post(&sig_sem);
  2529         thread->java_suspend_self();
  2531     } while (threadIsSuspended);
  2535 int os::signal_lookup() {
  2536   return check_pending_signals(false);
  2539 int os::signal_wait() {
  2540   return check_pending_signals(true);
  2543 ////////////////////////////////////////////////////////////////////////////////
  2544 // Virtual Memory
  2546 int os::vm_page_size() {
  2547   // Seems redundant as all get out
  2548   assert(os::Linux::page_size() != -1, "must call os::init");
  2549   return os::Linux::page_size();
  2552 // Solaris allocates memory by pages.
  2553 int os::vm_allocation_granularity() {
  2554   assert(os::Linux::page_size() != -1, "must call os::init");
  2555   return os::Linux::page_size();
  2558 // Rationale behind this function:
  2559 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
  2560 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
  2561 //  samples for JITted code. Here we create private executable mapping over the code cache
  2562 //  and then we can use standard (well, almost, as mapping can change) way to provide
  2563 //  info for the reporting script by storing timestamp and location of symbol
  2564 void linux_wrap_code(char* base, size_t size) {
  2565   static volatile jint cnt = 0;
  2567   if (!UseOprofile) {
  2568     return;
  2571   char buf[PATH_MAX+1];
  2572   int num = Atomic::add(1, &cnt);
  2574   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
  2575            os::get_temp_directory(), os::current_process_id(), num);
  2576   unlink(buf);
  2578   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
  2580   if (fd != -1) {
  2581     off_t rv = ::lseek(fd, size-2, SEEK_SET);
  2582     if (rv != (off_t)-1) {
  2583       if (::write(fd, "", 1) == 1) {
  2584         mmap(base, size,
  2585              PROT_READ|PROT_WRITE|PROT_EXEC,
  2586              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
  2589     ::close(fd);
  2590     unlink(buf);
  2594 static bool recoverable_mmap_error(int err) {
  2595   // See if the error is one we can let the caller handle. This
  2596   // list of errno values comes from JBS-6843484. I can't find a
  2597   // Linux man page that documents this specific set of errno
  2598   // values so while this list currently matches Solaris, it may
  2599   // change as we gain experience with this failure mode.
  2600   switch (err) {
  2601   case EBADF:
  2602   case EINVAL:
  2603   case ENOTSUP:
  2604     // let the caller deal with these errors
  2605     return true;
  2607   default:
  2608     // Any remaining errors on this OS can cause our reserved mapping
  2609     // to be lost. That can cause confusion where different data
  2610     // structures think they have the same memory mapped. The worst
  2611     // scenario is if both the VM and a library think they have the
  2612     // same memory mapped.
  2613     return false;
  2617 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
  2618                                     int err) {
  2619   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
  2620           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
  2621           strerror(err), err);
  2624 static void warn_fail_commit_memory(char* addr, size_t size,
  2625                                     size_t alignment_hint, bool exec,
  2626                                     int err) {
  2627   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
  2628           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
  2629           alignment_hint, exec, strerror(err), err);
  2632 // NOTE: Linux kernel does not really reserve the pages for us.
  2633 //       All it does is to check if there are enough free pages
  2634 //       left at the time of mmap(). This could be a potential
  2635 //       problem.
  2636 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
  2637   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
  2638   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
  2639                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
  2640   if (res != (uintptr_t) MAP_FAILED) {
  2641     if (UseNUMAInterleaving) {
  2642       numa_make_global(addr, size);
  2644     return 0;
  2647   int err = errno;  // save errno from mmap() call above
  2649   if (!recoverable_mmap_error(err)) {
  2650     warn_fail_commit_memory(addr, size, exec, err);
  2651     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
  2654   return err;
  2657 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
  2658   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
  2661 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
  2662                                   const char* mesg) {
  2663   assert(mesg != NULL, "mesg must be specified");
  2664   int err = os::Linux::commit_memory_impl(addr, size, exec);
  2665   if (err != 0) {
  2666     // the caller wants all commit errors to exit with the specified mesg:
  2667     warn_fail_commit_memory(addr, size, exec, err);
  2668     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
  2672 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
  2673 #ifndef MAP_HUGETLB
  2674 #define MAP_HUGETLB 0x40000
  2675 #endif
  2677 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
  2678 #ifndef MADV_HUGEPAGE
  2679 #define MADV_HUGEPAGE 14
  2680 #endif
  2682 int os::Linux::commit_memory_impl(char* addr, size_t size,
  2683                                   size_t alignment_hint, bool exec) {
  2684   int err = os::Linux::commit_memory_impl(addr, size, exec);
  2685   if (err == 0) {
  2686     realign_memory(addr, size, alignment_hint);
  2688   return err;
  2691 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
  2692                           bool exec) {
  2693   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
  2696 void os::pd_commit_memory_or_exit(char* addr, size_t size,
  2697                                   size_t alignment_hint, bool exec,
  2698                                   const char* mesg) {
  2699   assert(mesg != NULL, "mesg must be specified");
  2700   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
  2701   if (err != 0) {
  2702     // the caller wants all commit errors to exit with the specified mesg:
  2703     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
  2704     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
  2708 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
  2709   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
  2710     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
  2711     // be supported or the memory may already be backed by huge pages.
  2712     ::madvise(addr, bytes, MADV_HUGEPAGE);
  2716 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
  2717   // This method works by doing an mmap over an existing mmaping and effectively discarding
  2718   // the existing pages. However it won't work for SHM-based large pages that cannot be
  2719   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
  2720   // small pages on top of the SHM segment. This method always works for small pages, so we
  2721   // allow that in any case.
  2722   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
  2723     commit_memory(addr, bytes, alignment_hint, !ExecMem);
  2727 void os::numa_make_global(char *addr, size_t bytes) {
  2728   Linux::numa_interleave_memory(addr, bytes);
  2731 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
  2732 // bind policy to MPOL_PREFERRED for the current thread.
  2733 #define USE_MPOL_PREFERRED 0
  2735 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
  2736   // To make NUMA and large pages more robust when both enabled, we need to ease
  2737   // the requirements on where the memory should be allocated. MPOL_BIND is the
  2738   // default policy and it will force memory to be allocated on the specified
  2739   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
  2740   // the specified node, but will not force it. Using this policy will prevent
  2741   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
  2742   // free large pages.
  2743   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
  2744   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
  2747 bool os::numa_topology_changed()   { return false; }
  2749 size_t os::numa_get_groups_num() {
  2750   // Return just the number of nodes in which it's possible to allocate memory
  2751   // (in numa terminology, configured nodes).
  2752   return Linux::numa_num_configured_nodes();
  2755 int os::numa_get_group_id() {
  2756   int cpu_id = Linux::sched_getcpu();
  2757   if (cpu_id != -1) {
  2758     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
  2759     if (lgrp_id != -1) {
  2760       return lgrp_id;
  2763   return 0;
  2766 int os::Linux::get_existing_num_nodes() {
  2767   size_t node;
  2768   size_t highest_node_number = Linux::numa_max_node();
  2769   int num_nodes = 0;
  2771   // Get the total number of nodes in the system including nodes without memory.
  2772   for (node = 0; node <= highest_node_number; node++) {
  2773     if (isnode_in_existing_nodes(node)) {
  2774       num_nodes++;
  2777   return num_nodes;
  2780 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
  2781   size_t highest_node_number = Linux::numa_max_node();
  2782   size_t i = 0;
  2784   // Map all node ids in which is possible to allocate memory. Also nodes are
  2785   // not always consecutively available, i.e. available from 0 to the highest
  2786   // node number.
  2787   for (size_t node = 0; node <= highest_node_number; node++) {
  2788     if (Linux::isnode_in_configured_nodes(node)) {
  2789       ids[i++] = node;
  2792   return i;
  2795 bool os::get_page_info(char *start, page_info* info) {
  2796   return false;
  2799 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
  2800   return end;
  2804 int os::Linux::sched_getcpu_syscall(void) {
  2805   unsigned int cpu = 0;
  2806   int retval = -1;
  2808 #if defined(IA32)
  2809 # ifndef SYS_getcpu
  2810 # define SYS_getcpu 318
  2811 # endif
  2812   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
  2813 #elif defined(AMD64)
  2814 // Unfortunately we have to bring all these macros here from vsyscall.h
  2815 // to be able to compile on old linuxes.
  2816 # define __NR_vgetcpu 2
  2817 # define VSYSCALL_START (-10UL << 20)
  2818 # define VSYSCALL_SIZE 1024
  2819 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
  2820   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
  2821   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
  2822   retval = vgetcpu(&cpu, NULL, NULL);
  2823 #endif
  2825   return (retval == -1) ? retval : cpu;
  2828 // Something to do with the numa-aware allocator needs these symbols
  2829 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
  2830 extern "C" JNIEXPORT void numa_error(char *where) { }
  2831 extern "C" JNIEXPORT int fork1() { return fork(); }
  2834 // If we are running with libnuma version > 2, then we should
  2835 // be trying to use symbols with versions 1.1
  2836 // If we are running with earlier version, which did not have symbol versions,
  2837 // we should use the base version.
  2838 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
  2839   void *f = dlvsym(handle, name, "libnuma_1.1");
  2840   if (f == NULL) {
  2841     f = dlsym(handle, name);
  2843   return f;
  2846 bool os::Linux::libnuma_init() {
  2847   // sched_getcpu() should be in libc.
  2848   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
  2849                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
  2851   // If it's not, try a direct syscall.
  2852   if (sched_getcpu() == -1)
  2853     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
  2855   if (sched_getcpu() != -1) { // Does it work?
  2856     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
  2857     if (handle != NULL) {
  2858       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
  2859                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
  2860       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
  2861                                        libnuma_dlsym(handle, "numa_max_node")));
  2862       set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
  2863                                                    libnuma_dlsym(handle, "numa_num_configured_nodes")));
  2864       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
  2865                                         libnuma_dlsym(handle, "numa_available")));
  2866       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
  2867                                             libnuma_dlsym(handle, "numa_tonode_memory")));
  2868       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
  2869                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
  2870       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
  2871                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
  2872       set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
  2873                                                libnuma_dlsym(handle, "numa_bitmask_isbitset")));
  2874       set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
  2875                                        libnuma_dlsym(handle, "numa_distance")));
  2877       if (numa_available() != -1) {
  2878         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
  2879         set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
  2880         set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
  2881         // Create an index -> node mapping, since nodes are not always consecutive
  2882         _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
  2883         rebuild_nindex_to_node_map();
  2884         // Create a cpu -> node mapping
  2885         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
  2886         rebuild_cpu_to_node_map();
  2887         return true;
  2891   return false;
  2894 void os::Linux::rebuild_nindex_to_node_map() {
  2895   int highest_node_number = Linux::numa_max_node();
  2897   nindex_to_node()->clear();
  2898   for (int node = 0; node <= highest_node_number; node++) {
  2899     if (Linux::isnode_in_existing_nodes(node)) {
  2900       nindex_to_node()->append(node);
  2905 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
  2906 // The table is later used in get_node_by_cpu().
  2907 void os::Linux::rebuild_cpu_to_node_map() {
  2908   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
  2909                               // in libnuma (possible values are starting from 16,
  2910                               // and continuing up with every other power of 2, but less
  2911                               // than the maximum number of CPUs supported by kernel), and
  2912                               // is a subject to change (in libnuma version 2 the requirements
  2913                               // are more reasonable) we'll just hardcode the number they use
  2914                               // in the library.
  2915   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
  2917   size_t cpu_num = processor_count();
  2918   size_t cpu_map_size = NCPUS / BitsPerCLong;
  2919   size_t cpu_map_valid_size =
  2920     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
  2922   cpu_to_node()->clear();
  2923   cpu_to_node()->at_grow(cpu_num - 1);
  2925   size_t node_num = get_existing_num_nodes();
  2927   int distance = 0;
  2928   int closest_distance = INT_MAX;
  2929   int closest_node = 0;
  2930   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
  2931   for (size_t i = 0; i < node_num; i++) {
  2932     // Check if node is configured (not a memory-less node). If it is not, find
  2933     // the closest configured node.
  2934     if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
  2935       closest_distance = INT_MAX;
  2936       // Check distance from all remaining nodes in the system. Ignore distance
  2937       // from itself and from another non-configured node.
  2938       for (size_t m = 0; m < node_num; m++) {
  2939         if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
  2940           distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
  2941           // If a closest node is found, update. There is always at least one
  2942           // configured node in the system so there is always at least one node
  2943           // close.
  2944           if (distance != 0 && distance < closest_distance) {
  2945             closest_distance = distance;
  2946             closest_node = nindex_to_node()->at(m);
  2950      } else {
  2951        // Current node is already a configured node.
  2952        closest_node = nindex_to_node()->at(i);
  2955     // Get cpus from the original node and map them to the closest node. If node
  2956     // is a configured node (not a memory-less node), then original node and
  2957     // closest node are the same.
  2958     if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
  2959       for (size_t j = 0; j < cpu_map_valid_size; j++) {
  2960         if (cpu_map[j] != 0) {
  2961           for (size_t k = 0; k < BitsPerCLong; k++) {
  2962             if (cpu_map[j] & (1UL << k)) {
  2963               cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
  2970   FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
  2973 int os::Linux::get_node_by_cpu(int cpu_id) {
  2974   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
  2975     return cpu_to_node()->at(cpu_id);
  2977   return -1;
  2980 GrowableArray<int>* os::Linux::_cpu_to_node;
  2981 GrowableArray<int>* os::Linux::_nindex_to_node;
  2982 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
  2983 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
  2984 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
  2985 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
  2986 os::Linux::numa_available_func_t os::Linux::_numa_available;
  2987 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
  2988 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
  2989 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
  2990 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
  2991 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
  2992 unsigned long* os::Linux::_numa_all_nodes;
  2993 struct bitmask* os::Linux::_numa_all_nodes_ptr;
  2994 struct bitmask* os::Linux::_numa_nodes_ptr;
  2996 bool os::pd_uncommit_memory(char* addr, size_t size) {
  2997   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
  2998                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
  2999   return res  != (uintptr_t) MAP_FAILED;
  3002 static
  3003 address get_stack_commited_bottom(address bottom, size_t size) {
  3004   address nbot = bottom;
  3005   address ntop = bottom + size;
  3007   size_t page_sz = os::vm_page_size();
  3008   unsigned pages = size / page_sz;
  3010   unsigned char vec[1];
  3011   unsigned imin = 1, imax = pages + 1, imid;
  3012   int mincore_return_value = 0;
  3014   assert(imin <= imax, "Unexpected page size");
  3016   while (imin < imax) {
  3017     imid = (imax + imin) / 2;
  3018     nbot = ntop - (imid * page_sz);
  3020     // Use a trick with mincore to check whether the page is mapped or not.
  3021     // mincore sets vec to 1 if page resides in memory and to 0 if page
  3022     // is swapped output but if page we are asking for is unmapped
  3023     // it returns -1,ENOMEM
  3024     mincore_return_value = mincore(nbot, page_sz, vec);
  3026     if (mincore_return_value == -1) {
  3027       // Page is not mapped go up
  3028       // to find first mapped page
  3029       if (errno != EAGAIN) {
  3030         assert(errno == ENOMEM, "Unexpected mincore errno");
  3031         imax = imid;
  3033     } else {
  3034       // Page is mapped go down
  3035       // to find first not mapped page
  3036       imin = imid + 1;
  3040   nbot = nbot + page_sz;
  3042   // Adjust stack bottom one page up if last checked page is not mapped
  3043   if (mincore_return_value == -1) {
  3044     nbot = nbot + page_sz;
  3047   return nbot;
  3051 // Linux uses a growable mapping for the stack, and if the mapping for
  3052 // the stack guard pages is not removed when we detach a thread the
  3053 // stack cannot grow beyond the pages where the stack guard was
  3054 // mapped.  If at some point later in the process the stack expands to
  3055 // that point, the Linux kernel cannot expand the stack any further
  3056 // because the guard pages are in the way, and a segfault occurs.
  3057 //
  3058 // However, it's essential not to split the stack region by unmapping
  3059 // a region (leaving a hole) that's already part of the stack mapping,
  3060 // so if the stack mapping has already grown beyond the guard pages at
  3061 // the time we create them, we have to truncate the stack mapping.
  3062 // So, we need to know the extent of the stack mapping when
  3063 // create_stack_guard_pages() is called.
  3065 // We only need this for stacks that are growable: at the time of
  3066 // writing thread stacks don't use growable mappings (i.e. those
  3067 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
  3068 // only applies to the main thread.
  3070 // If the (growable) stack mapping already extends beyond the point
  3071 // where we're going to put our guard pages, truncate the mapping at
  3072 // that point by munmap()ping it.  This ensures that when we later
  3073 // munmap() the guard pages we don't leave a hole in the stack
  3074 // mapping. This only affects the main/initial thread
  3076 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
  3078   if (os::Linux::is_initial_thread()) {
  3079     // As we manually grow stack up to bottom inside create_attached_thread(),
  3080     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
  3081     // we don't need to do anything special.
  3082     // Check it first, before calling heavy function.
  3083     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
  3084     unsigned char vec[1];
  3086     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
  3087       // Fallback to slow path on all errors, including EAGAIN
  3088       stack_extent = (uintptr_t) get_stack_commited_bottom(
  3089                                     os::Linux::initial_thread_stack_bottom(),
  3090                                     (size_t)addr - stack_extent);
  3093     if (stack_extent < (uintptr_t)addr) {
  3094       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
  3098   return os::commit_memory(addr, size, !ExecMem);
  3101 // If this is a growable mapping, remove the guard pages entirely by
  3102 // munmap()ping them.  If not, just call uncommit_memory(). This only
  3103 // affects the main/initial thread, but guard against future OS changes
  3104 // It's safe to always unmap guard pages for initial thread because we
  3105 // always place it right after end of the mapped region
  3107 bool os::remove_stack_guard_pages(char* addr, size_t size) {
  3108   uintptr_t stack_extent, stack_base;
  3110   if (os::Linux::is_initial_thread()) {
  3111     return ::munmap(addr, size) == 0;
  3114   return os::uncommit_memory(addr, size);
  3117 static address _highest_vm_reserved_address = NULL;
  3119 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
  3120 // at 'requested_addr'. If there are existing memory mappings at the same
  3121 // location, however, they will be overwritten. If 'fixed' is false,
  3122 // 'requested_addr' is only treated as a hint, the return value may or
  3123 // may not start from the requested address. Unlike Linux mmap(), this
  3124 // function returns NULL to indicate failure.
  3125 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
  3126   char * addr;
  3127   int flags;
  3129   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
  3130   if (fixed) {
  3131     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
  3132     flags |= MAP_FIXED;
  3135   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
  3136   // touch an uncommitted page. Otherwise, the read/write might
  3137   // succeed if we have enough swap space to back the physical page.
  3138   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
  3139                        flags, -1, 0);
  3141   if (addr != MAP_FAILED) {
  3142     // anon_mmap() should only get called during VM initialization,
  3143     // don't need lock (actually we can skip locking even it can be called
  3144     // from multiple threads, because _highest_vm_reserved_address is just a
  3145     // hint about the upper limit of non-stack memory regions.)
  3146     if ((address)addr + bytes > _highest_vm_reserved_address) {
  3147       _highest_vm_reserved_address = (address)addr + bytes;
  3151   return addr == MAP_FAILED ? NULL : addr;
  3154 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
  3155 //   (req_addr != NULL) or with a given alignment.
  3156 //  - bytes shall be a multiple of alignment.
  3157 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
  3158 //  - alignment sets the alignment at which memory shall be allocated.
  3159 //     It must be a multiple of allocation granularity.
  3160 // Returns address of memory or NULL. If req_addr was not NULL, will only return
  3161 //  req_addr or NULL.
  3162 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
  3164   size_t extra_size = bytes;
  3165   if (req_addr == NULL && alignment > 0) {
  3166     extra_size += alignment;
  3169   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
  3170     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
  3171     -1, 0);
  3172   if (start == MAP_FAILED) {
  3173     start = NULL;
  3174   } else {
  3175     if (req_addr != NULL) {
  3176       if (start != req_addr) {
  3177         ::munmap(start, extra_size);
  3178         start = NULL;
  3180     } else {
  3181       char* const start_aligned = (char*) align_ptr_up(start, alignment);
  3182       char* const end_aligned = start_aligned + bytes;
  3183       char* const end = start + extra_size;
  3184       if (start_aligned > start) {
  3185         ::munmap(start, start_aligned - start);
  3187       if (end_aligned < end) {
  3188         ::munmap(end_aligned, end - end_aligned);
  3190       start = start_aligned;
  3193   return start;
  3196 // Don't update _highest_vm_reserved_address, because there might be memory
  3197 // regions above addr + size. If so, releasing a memory region only creates
  3198 // a hole in the address space, it doesn't help prevent heap-stack collision.
  3199 //
  3200 static int anon_munmap(char * addr, size_t size) {
  3201   return ::munmap(addr, size) == 0;
  3204 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
  3205                          size_t alignment_hint) {
  3206   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
  3209 bool os::pd_release_memory(char* addr, size_t size) {
  3210   return anon_munmap(addr, size);
  3213 static address highest_vm_reserved_address() {
  3214   return _highest_vm_reserved_address;
  3217 static bool linux_mprotect(char* addr, size_t size, int prot) {
  3218   // Linux wants the mprotect address argument to be page aligned.
  3219   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
  3221   // According to SUSv3, mprotect() should only be used with mappings
  3222   // established by mmap(), and mmap() always maps whole pages. Unaligned
  3223   // 'addr' likely indicates problem in the VM (e.g. trying to change
  3224   // protection of malloc'ed or statically allocated memory). Check the
  3225   // caller if you hit this assert.
  3226   assert(addr == bottom, "sanity check");
  3228   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
  3229   return ::mprotect(bottom, size, prot) == 0;
  3232 // Set protections specified
  3233 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
  3234                         bool is_committed) {
  3235   unsigned int p = 0;
  3236   switch (prot) {
  3237   case MEM_PROT_NONE: p = PROT_NONE; break;
  3238   case MEM_PROT_READ: p = PROT_READ; break;
  3239   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
  3240   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
  3241   default:
  3242     ShouldNotReachHere();
  3244   // is_committed is unused.
  3245   return linux_mprotect(addr, bytes, p);
  3248 bool os::guard_memory(char* addr, size_t size) {
  3249   return linux_mprotect(addr, size, PROT_NONE);
  3252 bool os::unguard_memory(char* addr, size_t size) {
  3253   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
  3256 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
  3257   bool result = false;
  3258   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
  3259                  MAP_ANONYMOUS|MAP_PRIVATE,
  3260                  -1, 0);
  3261   if (p != MAP_FAILED) {
  3262     void *aligned_p = align_ptr_up(p, page_size);
  3264     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
  3266     munmap(p, page_size * 2);
  3269   if (warn && !result) {
  3270     warning("TransparentHugePages is not supported by the operating system.");
  3273   return result;
  3276 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
  3277   bool result = false;
  3278   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
  3279                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
  3280                  -1, 0);
  3282   if (p != MAP_FAILED) {
  3283     // We don't know if this really is a huge page or not.
  3284     FILE *fp = fopen("/proc/self/maps", "r");
  3285     if (fp) {
  3286       while (!feof(fp)) {
  3287         char chars[257];
  3288         long x = 0;
  3289         if (fgets(chars, sizeof(chars), fp)) {
  3290           if (sscanf(chars, "%lx-%*x", &x) == 1
  3291               && x == (long)p) {
  3292             if (strstr (chars, "hugepage")) {
  3293               result = true;
  3294               break;
  3299       fclose(fp);
  3301     munmap(p, page_size);
  3304   if (warn && !result) {
  3305     warning("HugeTLBFS is not supported by the operating system.");
  3308   return result;
  3311 /*
  3312 * Set the coredump_filter bits to include largepages in core dump (bit 6)
  3314 * From the coredump_filter documentation:
  3316 * - (bit 0) anonymous private memory
  3317 * - (bit 1) anonymous shared memory
  3318 * - (bit 2) file-backed private memory
  3319 * - (bit 3) file-backed shared memory
  3320 * - (bit 4) ELF header pages in file-backed private memory areas (it is
  3321 *           effective only if the bit 2 is cleared)
  3322 * - (bit 5) hugetlb private memory
  3323 * - (bit 6) hugetlb shared memory
  3324 */
  3325 static void set_coredump_filter(void) {
  3326   FILE *f;
  3327   long cdm;
  3329   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
  3330     return;
  3333   if (fscanf(f, "%lx", &cdm) != 1) {
  3334     fclose(f);
  3335     return;
  3338   rewind(f);
  3340   if ((cdm & LARGEPAGES_BIT) == 0) {
  3341     cdm |= LARGEPAGES_BIT;
  3342     fprintf(f, "%#lx", cdm);
  3345   fclose(f);
  3348 // Large page support
  3350 static size_t _large_page_size = 0;
  3352 size_t os::Linux::find_large_page_size() {
  3353   size_t large_page_size = 0;
  3355   // large_page_size on Linux is used to round up heap size. x86 uses either
  3356   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
  3357   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
  3358   // page as large as 256M.
  3359   //
  3360   // Here we try to figure out page size by parsing /proc/meminfo and looking
  3361   // for a line with the following format:
  3362   //    Hugepagesize:     2048 kB
  3363   //
  3364   // If we can't determine the value (e.g. /proc is not mounted, or the text
  3365   // format has been changed), we'll use the largest page size supported by
  3366   // the processor.
  3368 #ifndef ZERO
  3369   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
  3370                      ARM_ONLY(2 * M) PPC_ONLY(4 * M) MIPS64_ONLY(4 * M); //In MIPS _large_page_size is seted 4*M.
  3371 #endif // ZERO
  3373   FILE *fp = fopen("/proc/meminfo", "r");
  3374   if (fp) {
  3375     while (!feof(fp)) {
  3376       int x = 0;
  3377       char buf[16];
  3378       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
  3379         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
  3380           large_page_size = x * K;
  3381           break;
  3383       } else {
  3384         // skip to next line
  3385         for (;;) {
  3386           int ch = fgetc(fp);
  3387           if (ch == EOF || ch == (int)'\n') break;
  3391     fclose(fp);
  3394   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
  3395     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
  3396         SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
  3397         proper_unit_for_byte_size(large_page_size));
  3400   return large_page_size;
  3403 size_t os::Linux::setup_large_page_size() {
  3404   _large_page_size = Linux::find_large_page_size();
  3405   const size_t default_page_size = (size_t)Linux::page_size();
  3406   if (_large_page_size > default_page_size) {
  3407     _page_sizes[0] = _large_page_size;
  3408     _page_sizes[1] = default_page_size;
  3409     _page_sizes[2] = 0;
  3412   return _large_page_size;
  3415 bool os::Linux::setup_large_page_type(size_t page_size) {
  3416   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
  3417       FLAG_IS_DEFAULT(UseSHM) &&
  3418       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
  3420     // The type of large pages has not been specified by the user.
  3422     // Try UseHugeTLBFS and then UseSHM.
  3423     UseHugeTLBFS = UseSHM = true;
  3425     // Don't try UseTransparentHugePages since there are known
  3426     // performance issues with it turned on. This might change in the future.
  3427     UseTransparentHugePages = false;
  3430   if (UseTransparentHugePages) {
  3431     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
  3432     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
  3433       UseHugeTLBFS = false;
  3434       UseSHM = false;
  3435       return true;
  3437     UseTransparentHugePages = false;
  3440   if (UseHugeTLBFS) {
  3441     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
  3442     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
  3443       UseSHM = false;
  3444       return true;
  3446     UseHugeTLBFS = false;
  3449   return UseSHM;
  3452 void os::large_page_init() {
  3453   if (!UseLargePages &&
  3454       !UseTransparentHugePages &&
  3455       !UseHugeTLBFS &&
  3456       !UseSHM) {
  3457     // Not using large pages.
  3458     return;
  3461   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
  3462     // The user explicitly turned off large pages.
  3463     // Ignore the rest of the large pages flags.
  3464     UseTransparentHugePages = false;
  3465     UseHugeTLBFS = false;
  3466     UseSHM = false;
  3467     return;
  3470   size_t large_page_size = Linux::setup_large_page_size();
  3471   UseLargePages          = Linux::setup_large_page_type(large_page_size);
  3473   set_coredump_filter();
  3476 #ifndef SHM_HUGETLB
  3477 #define SHM_HUGETLB 04000
  3478 #endif
  3480 #define shm_warning_format(format, ...)              \
  3481   do {                                               \
  3482     if (UseLargePages &&                             \
  3483         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
  3484          !FLAG_IS_DEFAULT(UseSHM) ||                 \
  3485          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
  3486       warning(format, __VA_ARGS__);                  \
  3487     }                                                \
  3488   } while (0)
  3490 #define shm_warning(str) shm_warning_format("%s", str)
  3492 #define shm_warning_with_errno(str)                \
  3493   do {                                             \
  3494     int err = errno;                               \
  3495     shm_warning_format(str " (error = %d)", err);  \
  3496   } while (0)
  3498 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
  3499   assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
  3501   if (!is_size_aligned(alignment, SHMLBA)) {
  3502     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
  3503     return NULL;
  3506   // To ensure that we get 'alignment' aligned memory from shmat,
  3507   // we pre-reserve aligned virtual memory and then attach to that.
  3509   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
  3510   if (pre_reserved_addr == NULL) {
  3511     // Couldn't pre-reserve aligned memory.
  3512     shm_warning("Failed to pre-reserve aligned memory for shmat.");
  3513     return NULL;
  3516   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
  3517   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
  3519   if ((intptr_t)addr == -1) {
  3520     int err = errno;
  3521     shm_warning_with_errno("Failed to attach shared memory.");
  3523     assert(err != EACCES, "Unexpected error");
  3524     assert(err != EIDRM,  "Unexpected error");
  3525     assert(err != EINVAL, "Unexpected error");
  3527     // Since we don't know if the kernel unmapped the pre-reserved memory area
  3528     // we can't unmap it, since that would potentially unmap memory that was
  3529     // mapped from other threads.
  3530     return NULL;
  3533   return addr;
  3536 static char* shmat_at_address(int shmid, char* req_addr) {
  3537   if (!is_ptr_aligned(req_addr, SHMLBA)) {
  3538     assert(false, "Requested address needs to be SHMLBA aligned");
  3539     return NULL;
  3542   char* addr = (char*)shmat(shmid, req_addr, 0);
  3544   if ((intptr_t)addr == -1) {
  3545     shm_warning_with_errno("Failed to attach shared memory.");
  3546     return NULL;
  3549   return addr;
  3552 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
  3553   // If a req_addr has been provided, we assume that the caller has already aligned the address.
  3554   if (req_addr != NULL) {
  3555     assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
  3556     assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
  3557     return shmat_at_address(shmid, req_addr);
  3560   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
  3561   // return large page size aligned memory addresses when req_addr == NULL.
  3562   // However, if the alignment is larger than the large page size, we have
  3563   // to manually ensure that the memory returned is 'alignment' aligned.
  3564   if (alignment > os::large_page_size()) {
  3565     assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
  3566     return shmat_with_alignment(shmid, bytes, alignment);
  3567   } else {
  3568     return shmat_at_address(shmid, NULL);
  3572 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
  3573   // "exec" is passed in but not used.  Creating the shared image for
  3574   // the code cache doesn't have an SHM_X executable permission to check.
  3575   assert(UseLargePages && UseSHM, "only for SHM large pages");
  3576   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
  3577   assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
  3579   if (!is_size_aligned(bytes, os::large_page_size())) {
  3580     return NULL; // Fallback to small pages.
  3583   // Create a large shared memory region to attach to based on size.
  3584   // Currently, size is the total size of the heap.
  3585   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
  3586   if (shmid == -1) {
  3587     // Possible reasons for shmget failure:
  3588     // 1. shmmax is too small for Java heap.
  3589     //    > check shmmax value: cat /proc/sys/kernel/shmmax
  3590     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
  3591     // 2. not enough large page memory.
  3592     //    > check available large pages: cat /proc/meminfo
  3593     //    > increase amount of large pages:
  3594     //          echo new_value > /proc/sys/vm/nr_hugepages
  3595     //      Note 1: different Linux may use different name for this property,
  3596     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
  3597     //      Note 2: it's possible there's enough physical memory available but
  3598     //            they are so fragmented after a long run that they can't
  3599     //            coalesce into large pages. Try to reserve large pages when
  3600     //            the system is still "fresh".
  3601     shm_warning_with_errno("Failed to reserve shared memory.");
  3602     return NULL;
  3605   // Attach to the region.
  3606   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
  3608   // Remove shmid. If shmat() is successful, the actual shared memory segment
  3609   // will be deleted when it's detached by shmdt() or when the process
  3610   // terminates. If shmat() is not successful this will remove the shared
  3611   // segment immediately.
  3612   shmctl(shmid, IPC_RMID, NULL);
  3614   return addr;
  3617 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
  3618   assert(error == ENOMEM, "Only expect to fail if no memory is available");
  3620   bool warn_on_failure = UseLargePages &&
  3621       (!FLAG_IS_DEFAULT(UseLargePages) ||
  3622        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
  3623        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
  3625   if (warn_on_failure) {
  3626     char msg[128];
  3627     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
  3628         PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
  3629     warning("%s", msg);
  3633 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
  3634   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
  3635   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
  3636   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
  3638   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
  3639   char* addr = (char*)::mmap(req_addr, bytes, prot,
  3640                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
  3641                              -1, 0);
  3643   if (addr == MAP_FAILED) {
  3644     warn_on_large_pages_failure(req_addr, bytes, errno);
  3645     return NULL;
  3648   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
  3650   return addr;
  3653 // Reserve memory using mmap(MAP_HUGETLB).
  3654 //  - bytes shall be a multiple of alignment.
  3655 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
  3656 //  - alignment sets the alignment at which memory shall be allocated.
  3657 //     It must be a multiple of allocation granularity.
  3658 // Returns address of memory or NULL. If req_addr was not NULL, will only return
  3659 //  req_addr or NULL.
  3660 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
  3661   size_t large_page_size = os::large_page_size();
  3662   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
  3664   assert(is_ptr_aligned(req_addr, alignment), "Must be");
  3665   assert(is_size_aligned(bytes, alignment), "Must be");
  3667   // First reserve - but not commit - the address range in small pages.
  3668   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
  3670   if (start == NULL) {
  3671     return NULL;
  3674   assert(is_ptr_aligned(start, alignment), "Must be");
  3676   char* end = start + bytes;
  3678   // Find the regions of the allocated chunk that can be promoted to large pages.
  3679   char* lp_start = (char*)align_ptr_up(start, large_page_size);
  3680   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
  3682   size_t lp_bytes = lp_end - lp_start;
  3684   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
  3686   if (lp_bytes == 0) {
  3687     // The mapped region doesn't even span the start and the end of a large page.
  3688     // Fall back to allocate a non-special area.
  3689     ::munmap(start, end - start);
  3690     return NULL;
  3693   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
  3695   void* result;
  3697   // Commit small-paged leading area.
  3698   if (start != lp_start) {
  3699     result = ::mmap(start, lp_start - start, prot,
  3700                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
  3701                     -1, 0);
  3702     if (result == MAP_FAILED) {
  3703       ::munmap(lp_start, end - lp_start);
  3704       return NULL;
  3708   // Commit large-paged area.
  3709   result = ::mmap(lp_start, lp_bytes, prot,
  3710                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
  3711                   -1, 0);
  3712   if (result == MAP_FAILED) {
  3713     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
  3714     // If the mmap above fails, the large pages region will be unmapped and we
  3715     // have regions before and after with small pages. Release these regions.
  3716     //
  3717     // |  mapped  |  unmapped  |  mapped  |
  3718     // ^          ^            ^          ^
  3719     // start      lp_start     lp_end     end
  3720     //
  3721     ::munmap(start, lp_start - start);
  3722     ::munmap(lp_end, end - lp_end);
  3723     return NULL;
  3726   // Commit small-paged trailing area.
  3727   if (lp_end != end) {
  3728       result = ::mmap(lp_end, end - lp_end, prot,
  3729                       MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
  3730                       -1, 0);
  3731     if (result == MAP_FAILED) {
  3732       ::munmap(start, lp_end - start);
  3733       return NULL;
  3737   return start;
  3740 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
  3741   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
  3742   assert(is_ptr_aligned(req_addr, alignment), "Must be");
  3743   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
  3744   assert(is_power_of_2(os::large_page_size()), "Must be");
  3745   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
  3747   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
  3748     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
  3749   } else {
  3750     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
  3754 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
  3755   assert(UseLargePages, "only for large pages");
  3757   char* addr;
  3758   if (UseSHM) {
  3759     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
  3760   } else {
  3761     assert(UseHugeTLBFS, "must be");
  3762     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
  3765   if (addr != NULL) {
  3766     if (UseNUMAInterleaving) {
  3767       numa_make_global(addr, bytes);
  3770     // The memory is committed
  3771     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
  3774   return addr;
  3777 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
  3778   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
  3779   return shmdt(base) == 0;
  3782 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
  3783   return pd_release_memory(base, bytes);
  3786 bool os::release_memory_special(char* base, size_t bytes) {
  3787   bool res;
  3788   if (MemTracker::tracking_level() > NMT_minimal) {
  3789     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
  3790     res = os::Linux::release_memory_special_impl(base, bytes);
  3791     if (res) {
  3792       tkr.record((address)base, bytes);
  3795   } else {
  3796     res = os::Linux::release_memory_special_impl(base, bytes);
  3798   return res;
  3801 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
  3802   assert(UseLargePages, "only for large pages");
  3803   bool res;
  3805   if (UseSHM) {
  3806     res = os::Linux::release_memory_special_shm(base, bytes);
  3807   } else {
  3808     assert(UseHugeTLBFS, "must be");
  3809     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
  3811   return res;
  3814 size_t os::large_page_size() {
  3815   return _large_page_size;
  3818 // With SysV SHM the entire memory region must be allocated as shared
  3819 // memory.
  3820 // HugeTLBFS allows application to commit large page memory on demand.
  3821 // However, when committing memory with HugeTLBFS fails, the region
  3822 // that was supposed to be committed will lose the old reservation
  3823 // and allow other threads to steal that memory region. Because of this
  3824 // behavior we can't commit HugeTLBFS memory.
  3825 bool os::can_commit_large_page_memory() {
  3826   return UseTransparentHugePages;
  3829 bool os::can_execute_large_page_memory() {
  3830   return UseTransparentHugePages || UseHugeTLBFS;
  3833 // Reserve memory at an arbitrary address, only if that area is
  3834 // available (and not reserved for something else).
  3836 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
  3837   const int max_tries = 10;
  3838   char* base[max_tries];
  3839   size_t size[max_tries];
  3840   const size_t gap = 0x000000;
  3842   // Assert only that the size is a multiple of the page size, since
  3843   // that's all that mmap requires, and since that's all we really know
  3844   // about at this low abstraction level.  If we need higher alignment,
  3845   // we can either pass an alignment to this method or verify alignment
  3846   // in one of the methods further up the call chain.  See bug 5044738.
  3847   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
  3849   // Repeatedly allocate blocks until the block is allocated at the
  3850   // right spot. Give up after max_tries. Note that reserve_memory() will
  3851   // automatically update _highest_vm_reserved_address if the call is
  3852   // successful. The variable tracks the highest memory address every reserved
  3853   // by JVM. It is used to detect heap-stack collision if running with
  3854   // fixed-stack LinuxThreads. Because here we may attempt to reserve more
  3855   // space than needed, it could confuse the collision detecting code. To
  3856   // solve the problem, save current _highest_vm_reserved_address and
  3857   // calculate the correct value before return.
  3858   address old_highest = _highest_vm_reserved_address;
  3860   // Linux mmap allows caller to pass an address as hint; give it a try first,
  3861   // if kernel honors the hint then we can return immediately.
  3862   char * addr = anon_mmap(requested_addr, bytes, false);
  3863   if (addr == requested_addr) {
  3864      return requested_addr;
  3867   if (addr != NULL) {
  3868      // mmap() is successful but it fails to reserve at the requested address
  3869      anon_munmap(addr, bytes);
  3872   int i;
  3873   for (i = 0; i < max_tries; ++i) {
  3874     base[i] = reserve_memory(bytes);
  3876     if (base[i] != NULL) {
  3877       // Is this the block we wanted?
  3878       if (base[i] == requested_addr) {
  3879         size[i] = bytes;
  3880         break;
  3883       // Does this overlap the block we wanted? Give back the overlapped
  3884       // parts and try again.
  3886       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
  3887       if (top_overlap >= 0 && top_overlap < bytes) {
  3888         unmap_memory(base[i], top_overlap);
  3889         base[i] += top_overlap;
  3890         size[i] = bytes - top_overlap;
  3891       } else {
  3892         size_t bottom_overlap = base[i] + bytes - requested_addr;
  3893         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
  3894           unmap_memory(requested_addr, bottom_overlap);
  3895           size[i] = bytes - bottom_overlap;
  3896         } else {
  3897           size[i] = bytes;
  3903   // Give back the unused reserved pieces.
  3905   for (int j = 0; j < i; ++j) {
  3906     if (base[j] != NULL) {
  3907       unmap_memory(base[j], size[j]);
  3911   if (i < max_tries) {
  3912     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
  3913     return requested_addr;
  3914   } else {
  3915     _highest_vm_reserved_address = old_highest;
  3916     return NULL;
  3920 size_t os::read(int fd, void *buf, unsigned int nBytes) {
  3921   return ::read(fd, buf, nBytes);
  3924 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
  3925 // Solaris uses poll(), linux uses park().
  3926 // Poll() is likely a better choice, assuming that Thread.interrupt()
  3927 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
  3928 // SIGSEGV, see 4355769.
  3930 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
  3931   assert(thread == Thread::current(),  "thread consistency check");
  3933   ParkEvent * const slp = thread->_SleepEvent ;
  3934   slp->reset() ;
  3935   OrderAccess::fence() ;
  3937   if (interruptible) {
  3938     jlong prevtime = javaTimeNanos();
  3940     for (;;) {
  3941       if (os::is_interrupted(thread, true)) {
  3942         return OS_INTRPT;
  3945       jlong newtime = javaTimeNanos();
  3947       if (newtime - prevtime < 0) {
  3948         // time moving backwards, should only happen if no monotonic clock
  3949         // not a guarantee() because JVM should not abort on kernel/glibc bugs
  3950         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
  3951       } else {
  3952         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
  3955       if(millis <= 0) {
  3956         return OS_OK;
  3959       prevtime = newtime;
  3962         assert(thread->is_Java_thread(), "sanity check");
  3963         JavaThread *jt = (JavaThread *) thread;
  3964         ThreadBlockInVM tbivm(jt);
  3965         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
  3967         jt->set_suspend_equivalent();
  3968         // cleared by handle_special_suspend_equivalent_condition() or
  3969         // java_suspend_self() via check_and_wait_while_suspended()
  3971         slp->park(millis);
  3973         // were we externally suspended while we were waiting?
  3974         jt->check_and_wait_while_suspended();
  3977   } else {
  3978     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
  3979     jlong prevtime = javaTimeNanos();
  3981     for (;;) {
  3982       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
  3983       // the 1st iteration ...
  3984       jlong newtime = javaTimeNanos();
  3986       if (newtime - prevtime < 0) {
  3987         // time moving backwards, should only happen if no monotonic clock
  3988         // not a guarantee() because JVM should not abort on kernel/glibc bugs
  3989         assert(!Linux::supports_monotonic_clock(), "time moving backwards");
  3990       } else {
  3991         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
  3994       if(millis <= 0) break ;
  3996       prevtime = newtime;
  3997       slp->park(millis);
  3999     return OS_OK ;
  4003 //
  4004 // Short sleep, direct OS call.
  4005 //
  4006 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
  4007 // sched_yield(2) will actually give up the CPU:
  4008 //
  4009 //   * Alone on this pariticular CPU, keeps running.
  4010 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
  4011 //     (pre 2.6.39).
  4012 //
  4013 // So calling this with 0 is an alternative.
  4014 //
  4015 void os::naked_short_sleep(jlong ms) {
  4016   struct timespec req;
  4018   assert(ms < 1000, "Un-interruptable sleep, short time use only");
  4019   req.tv_sec = 0;
  4020   if (ms > 0) {
  4021     req.tv_nsec = (ms % 1000) * 1000000;
  4023   else {
  4024     req.tv_nsec = 1;
  4027   nanosleep(&req, NULL);
  4029   return;
  4032 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
  4033 void os::infinite_sleep() {
  4034   while (true) {    // sleep forever ...
  4035     ::sleep(100);   // ... 100 seconds at a time
  4039 // Used to convert frequent JVM_Yield() to nops
  4040 bool os::dont_yield() {
  4041   return DontYieldALot;
  4044 void os::yield() {
  4045   sched_yield();
  4048 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
  4050 void os::yield_all(int attempts) {
  4051   // Yields to all threads, including threads with lower priorities
  4052   // Threads on Linux are all with same priority. The Solaris style
  4053   // os::yield_all() with nanosleep(1ms) is not necessary.
  4054   sched_yield();
  4057 // Called from the tight loops to possibly influence time-sharing heuristics
  4058 void os::loop_breaker(int attempts) {
  4059   os::yield_all(attempts);
  4062 ////////////////////////////////////////////////////////////////////////////////
  4063 // thread priority support
  4065 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
  4066 // only supports dynamic priority, static priority must be zero. For real-time
  4067 // applications, Linux supports SCHED_RR which allows static priority (1-99).
  4068 // However, for large multi-threaded applications, SCHED_RR is not only slower
  4069 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
  4070 // of 5 runs - Sep 2005).
  4071 //
  4072 // The following code actually changes the niceness of kernel-thread/LWP. It
  4073 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
  4074 // not the entire user process, and user level threads are 1:1 mapped to kernel
  4075 // threads. It has always been the case, but could change in the future. For
  4076 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
  4077 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
  4079 int os::java_to_os_priority[CriticalPriority + 1] = {
  4080   19,              // 0 Entry should never be used
  4082    4,              // 1 MinPriority
  4083    3,              // 2
  4084    2,              // 3
  4086    1,              // 4
  4087    0,              // 5 NormPriority
  4088   -1,              // 6
  4090   -2,              // 7
  4091   -3,              // 8
  4092   -4,              // 9 NearMaxPriority
  4094   -5,              // 10 MaxPriority
  4096   -5               // 11 CriticalPriority
  4097 };
  4099 static int prio_init() {
  4100   if (ThreadPriorityPolicy == 1) {
  4101     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
  4102     // if effective uid is not root. Perhaps, a more elegant way of doing
  4103     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
  4104     if (geteuid() != 0) {
  4105       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
  4106         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
  4108       ThreadPriorityPolicy = 0;
  4111   if (UseCriticalJavaThreadPriority) {
  4112     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
  4114   return 0;
  4117 OSReturn os::set_native_priority(Thread* thread, int newpri) {
  4118   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
  4120   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
  4121   return (ret == 0) ? OS_OK : OS_ERR;
  4124 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
  4125   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
  4126     *priority_ptr = java_to_os_priority[NormPriority];
  4127     return OS_OK;
  4130   errno = 0;
  4131   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
  4132   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
  4135 // Hint to the underlying OS that a task switch would not be good.
  4136 // Void return because it's a hint and can fail.
  4137 void os::hint_no_preempt() {}
  4139 ////////////////////////////////////////////////////////////////////////////////
  4140 // suspend/resume support
  4142 //  the low-level signal-based suspend/resume support is a remnant from the
  4143 //  old VM-suspension that used to be for java-suspension, safepoints etc,
  4144 //  within hotspot. Now there is a single use-case for this:
  4145 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
  4146 //      that runs in the watcher thread.
  4147 //  The remaining code is greatly simplified from the more general suspension
  4148 //  code that used to be used.
  4149 //
  4150 //  The protocol is quite simple:
  4151 //  - suspend:
  4152 //      - sends a signal to the target thread
  4153 //      - polls the suspend state of the osthread using a yield loop
  4154 //      - target thread signal handler (SR_handler) sets suspend state
  4155 //        and blocks in sigsuspend until continued
  4156 //  - resume:
  4157 //      - sets target osthread state to continue
  4158 //      - sends signal to end the sigsuspend loop in the SR_handler
  4159 //
  4160 //  Note that the SR_lock plays no role in this suspend/resume protocol.
  4161 //
  4163 static void resume_clear_context(OSThread *osthread) {
  4164   osthread->set_ucontext(NULL);
  4165   osthread->set_siginfo(NULL);
  4168 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
  4169   osthread->set_ucontext(context);
  4170   osthread->set_siginfo(siginfo);
  4173 //
  4174 // Handler function invoked when a thread's execution is suspended or
  4175 // resumed. We have to be careful that only async-safe functions are
  4176 // called here (Note: most pthread functions are not async safe and
  4177 // should be avoided.)
  4178 //
  4179 // Note: sigwait() is a more natural fit than sigsuspend() from an
  4180 // interface point of view, but sigwait() prevents the signal hander
  4181 // from being run. libpthread would get very confused by not having
  4182 // its signal handlers run and prevents sigwait()'s use with the
  4183 // mutex granting granting signal.
  4184 //
  4185 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
  4186 //
  4187 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
  4188   // Save and restore errno to avoid confusing native code with EINTR
  4189   // after sigsuspend.
  4190   int old_errno = errno;
  4192   Thread* thread = Thread::current();
  4193   OSThread* osthread = thread->osthread();
  4194   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
  4196   os::SuspendResume::State current = osthread->sr.state();
  4197   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
  4198     suspend_save_context(osthread, siginfo, context);
  4200     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
  4201     os::SuspendResume::State state = osthread->sr.suspended();
  4202     if (state == os::SuspendResume::SR_SUSPENDED) {
  4203       sigset_t suspend_set;  // signals for sigsuspend()
  4205       // get current set of blocked signals and unblock resume signal
  4206       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
  4207       sigdelset(&suspend_set, SR_signum);
  4209       sr_semaphore.signal();
  4210       // wait here until we are resumed
  4211       while (1) {
  4212         sigsuspend(&suspend_set);
  4214         os::SuspendResume::State result = osthread->sr.running();
  4215         if (result == os::SuspendResume::SR_RUNNING) {
  4216           sr_semaphore.signal();
  4217           break;
  4221     } else if (state == os::SuspendResume::SR_RUNNING) {
  4222       // request was cancelled, continue
  4223     } else {
  4224       ShouldNotReachHere();
  4227     resume_clear_context(osthread);
  4228   } else if (current == os::SuspendResume::SR_RUNNING) {
  4229     // request was cancelled, continue
  4230   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
  4231     // ignore
  4232   } else {
  4233     // ignore
  4236   errno = old_errno;
  4240 static int SR_initialize() {
  4241   struct sigaction act;
  4242   char *s;
  4243   /* Get signal number to use for suspend/resume */
  4244   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
  4245     int sig = ::strtol(s, 0, 10);
  4246     if (sig > 0 || sig < _NSIG) {
  4247         SR_signum = sig;
  4251   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
  4252         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
  4254   sigemptyset(&SR_sigset);
  4255   sigaddset(&SR_sigset, SR_signum);
  4257   /* Set up signal handler for suspend/resume */
  4258   act.sa_flags = SA_RESTART|SA_SIGINFO;
  4259   act.sa_handler = (void (*)(int)) SR_handler;
  4261   // SR_signum is blocked by default.
  4262   // 4528190 - We also need to block pthread restart signal (32 on all
  4263   // supported Linux platforms). Note that LinuxThreads need to block
  4264   // this signal for all threads to work properly. So we don't have
  4265   // to use hard-coded signal number when setting up the mask.
  4266   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
  4268   if (sigaction(SR_signum, &act, 0) == -1) {
  4269     return -1;
  4272   // Save signal flag
  4273   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
  4274   return 0;
  4277 static int sr_notify(OSThread* osthread) {
  4278   int status = pthread_kill(osthread->pthread_id(), SR_signum);
  4279   assert_status(status == 0, status, "pthread_kill");
  4280   return status;
  4283 // "Randomly" selected value for how long we want to spin
  4284 // before bailing out on suspending a thread, also how often
  4285 // we send a signal to a thread we want to resume
  4286 static const int RANDOMLY_LARGE_INTEGER = 1000000;
  4287 static const int RANDOMLY_LARGE_INTEGER2 = 100;
  4289 // returns true on success and false on error - really an error is fatal
  4290 // but this seems the normal response to library errors
  4291 static bool do_suspend(OSThread* osthread) {
  4292   assert(osthread->sr.is_running(), "thread should be running");
  4293   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
  4295   // mark as suspended and send signal
  4296   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
  4297     // failed to switch, state wasn't running?
  4298     ShouldNotReachHere();
  4299     return false;
  4302   if (sr_notify(osthread) != 0) {
  4303     ShouldNotReachHere();
  4306   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
  4307   while (true) {
  4308     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
  4309       break;
  4310     } else {
  4311       // timeout
  4312       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
  4313       if (cancelled == os::SuspendResume::SR_RUNNING) {
  4314         return false;
  4315       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
  4316         // make sure that we consume the signal on the semaphore as well
  4317         sr_semaphore.wait();
  4318         break;
  4319       } else {
  4320         ShouldNotReachHere();
  4321         return false;
  4326   guarantee(osthread->sr.is_suspended(), "Must be suspended");
  4327   return true;
  4330 static void do_resume(OSThread* osthread) {
  4331   assert(osthread->sr.is_suspended(), "thread should be suspended");
  4332   assert(!sr_semaphore.trywait(), "invalid semaphore state");
  4334   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
  4335     // failed to switch to WAKEUP_REQUEST
  4336     ShouldNotReachHere();
  4337     return;
  4340   while (true) {
  4341     if (sr_notify(osthread) == 0) {
  4342       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
  4343         if (osthread->sr.is_running()) {
  4344           return;
  4347     } else {
  4348       ShouldNotReachHere();
  4352   guarantee(osthread->sr.is_running(), "Must be running!");
  4355 ////////////////////////////////////////////////////////////////////////////////
  4356 // interrupt support
  4358 void os::interrupt(Thread* thread) {
  4359   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
  4360     "possibility of dangling Thread pointer");
  4362   OSThread* osthread = thread->osthread();
  4364   if (!osthread->interrupted()) {
  4365     osthread->set_interrupted(true);
  4366     // More than one thread can get here with the same value of osthread,
  4367     // resulting in multiple notifications.  We do, however, want the store
  4368     // to interrupted() to be visible to other threads before we execute unpark().
  4369     OrderAccess::fence();
  4370     ParkEvent * const slp = thread->_SleepEvent ;
  4371     if (slp != NULL) slp->unpark() ;
  4374   // For JSR166. Unpark even if interrupt status already was set
  4375   if (thread->is_Java_thread())
  4376     ((JavaThread*)thread)->parker()->unpark();
  4378   ParkEvent * ev = thread->_ParkEvent ;
  4379   if (ev != NULL) ev->unpark() ;
  4383 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
  4384   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
  4385     "possibility of dangling Thread pointer");
  4387   OSThread* osthread = thread->osthread();
  4389   bool interrupted = osthread->interrupted();
  4391   if (interrupted && clear_interrupted) {
  4392     osthread->set_interrupted(false);
  4393     // consider thread->_SleepEvent->reset() ... optional optimization
  4396   return interrupted;
  4399 ///////////////////////////////////////////////////////////////////////////////////
  4400 // signal handling (except suspend/resume)
  4402 // This routine may be used by user applications as a "hook" to catch signals.
  4403 // The user-defined signal handler must pass unrecognized signals to this
  4404 // routine, and if it returns true (non-zero), then the signal handler must
  4405 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
  4406 // routine will never retun false (zero), but instead will execute a VM panic
  4407 // routine kill the process.
  4408 //
  4409 // If this routine returns false, it is OK to call it again.  This allows
  4410 // the user-defined signal handler to perform checks either before or after
  4411 // the VM performs its own checks.  Naturally, the user code would be making
  4412 // a serious error if it tried to handle an exception (such as a null check
  4413 // or breakpoint) that the VM was generating for its own correct operation.
  4414 //
  4415 // This routine may recognize any of the following kinds of signals:
  4416 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
  4417 // It should be consulted by handlers for any of those signals.
  4418 //
  4419 // The caller of this routine must pass in the three arguments supplied
  4420 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
  4421 // field of the structure passed to sigaction().  This routine assumes that
  4422 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
  4423 //
  4424 // Note that the VM will print warnings if it detects conflicting signal
  4425 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
  4426 //
  4427 extern "C" JNIEXPORT int
  4428 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
  4429                         void* ucontext, int abort_if_unrecognized);
  4431 void signalHandler(int sig, siginfo_t* info, void* uc) {
  4432   assert(info != NULL && uc != NULL, "it must be old kernel");
  4433   int orig_errno = errno;  // Preserve errno value over signal handler.
  4434   JVM_handle_linux_signal(sig, info, uc, true);
  4435   errno = orig_errno;
  4439 // This boolean allows users to forward their own non-matching signals
  4440 // to JVM_handle_linux_signal, harmlessly.
  4441 bool os::Linux::signal_handlers_are_installed = false;
  4443 // For signal-chaining
  4444 struct sigaction os::Linux::sigact[MAXSIGNUM];
  4445 unsigned int os::Linux::sigs = 0;
  4446 bool os::Linux::libjsig_is_loaded = false;
  4447 typedef struct sigaction *(*get_signal_t)(int);
  4448 get_signal_t os::Linux::get_signal_action = NULL;
  4450 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
  4451   struct sigaction *actp = NULL;
  4453   if (libjsig_is_loaded) {
  4454     // Retrieve the old signal handler from libjsig
  4455     actp = (*get_signal_action)(sig);
  4457   if (actp == NULL) {
  4458     // Retrieve the preinstalled signal handler from jvm
  4459     actp = get_preinstalled_handler(sig);
  4462   return actp;
  4465 static bool call_chained_handler(struct sigaction *actp, int sig,
  4466                                  siginfo_t *siginfo, void *context) {
  4467   // Call the old signal handler
  4468   if (actp->sa_handler == SIG_DFL) {
  4469     // It's more reasonable to let jvm treat it as an unexpected exception
  4470     // instead of taking the default action.
  4471     return false;
  4472   } else if (actp->sa_handler != SIG_IGN) {
  4473     if ((actp->sa_flags & SA_NODEFER) == 0) {
  4474       // automaticlly block the signal
  4475       sigaddset(&(actp->sa_mask), sig);
  4478     sa_handler_t hand = NULL;
  4479     sa_sigaction_t sa = NULL;
  4480     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
  4481     // retrieve the chained handler
  4482     if (siginfo_flag_set) {
  4483       sa = actp->sa_sigaction;
  4484     } else {
  4485       hand = actp->sa_handler;
  4488     if ((actp->sa_flags & SA_RESETHAND) != 0) {
  4489       actp->sa_handler = SIG_DFL;
  4492     // try to honor the signal mask
  4493     sigset_t oset;
  4494     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
  4496     // call into the chained handler
  4497     if (siginfo_flag_set) {
  4498       (*sa)(sig, siginfo, context);
  4499     } else {
  4500       (*hand)(sig);
  4503     // restore the signal mask
  4504     pthread_sigmask(SIG_SETMASK, &oset, 0);
  4506   // Tell jvm's signal handler the signal is taken care of.
  4507   return true;
  4510 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
  4511   bool chained = false;
  4512   // signal-chaining
  4513   if (UseSignalChaining) {
  4514     struct sigaction *actp = get_chained_signal_action(sig);
  4515     if (actp != NULL) {
  4516       chained = call_chained_handler(actp, sig, siginfo, context);
  4519   return chained;
  4522 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
  4523   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
  4524     return &sigact[sig];
  4526   return NULL;
  4529 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
  4530   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
  4531   sigact[sig] = oldAct;
  4532   sigs |= (unsigned int)1 << sig;
  4535 // for diagnostic
  4536 int os::Linux::sigflags[MAXSIGNUM];
  4538 int os::Linux::get_our_sigflags(int sig) {
  4539   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
  4540   return sigflags[sig];
  4543 void os::Linux::set_our_sigflags(int sig, int flags) {
  4544   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
  4545   sigflags[sig] = flags;
  4548 void os::Linux::set_signal_handler(int sig, bool set_installed) {
  4549   // Check for overwrite.
  4550   struct sigaction oldAct;
  4551   sigaction(sig, (struct sigaction*)NULL, &oldAct);
  4553   void* oldhand = oldAct.sa_sigaction
  4554                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
  4555                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
  4556   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
  4557       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
  4558       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
  4559     if (AllowUserSignalHandlers || !set_installed) {
  4560       // Do not overwrite; user takes responsibility to forward to us.
  4561       return;
  4562     } else if (UseSignalChaining) {
  4563       // save the old handler in jvm
  4564       save_preinstalled_handler(sig, oldAct);
  4565       // libjsig also interposes the sigaction() call below and saves the
  4566       // old sigaction on it own.
  4567     } else {
  4568       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
  4569                     "%#lx for signal %d.", (long)oldhand, sig));
  4573   struct sigaction sigAct;
  4574   sigfillset(&(sigAct.sa_mask));
  4575   sigAct.sa_handler = SIG_DFL;
  4576   if (!set_installed) {
  4577     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
  4578   } else {
  4579     sigAct.sa_sigaction = signalHandler;
  4580     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
  4582   // Save flags, which are set by ours
  4583   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
  4584   sigflags[sig] = sigAct.sa_flags;
  4586   int ret = sigaction(sig, &sigAct, &oldAct);
  4587   assert(ret == 0, "check");
  4589   void* oldhand2  = oldAct.sa_sigaction
  4590                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
  4591                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
  4592   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
  4595 // install signal handlers for signals that HotSpot needs to
  4596 // handle in order to support Java-level exception handling.
  4598 void os::Linux::install_signal_handlers() {
  4599   if (!signal_handlers_are_installed) {
  4600     signal_handlers_are_installed = true;
  4602     // signal-chaining
  4603     typedef void (*signal_setting_t)();
  4604     signal_setting_t begin_signal_setting = NULL;
  4605     signal_setting_t end_signal_setting = NULL;
  4606     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
  4607                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
  4608     if (begin_signal_setting != NULL) {
  4609       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
  4610                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
  4611       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
  4612                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
  4613       libjsig_is_loaded = true;
  4614       assert(UseSignalChaining, "should enable signal-chaining");
  4616     if (libjsig_is_loaded) {
  4617       // Tell libjsig jvm is setting signal handlers
  4618       (*begin_signal_setting)();
  4621     set_signal_handler(SIGSEGV, true);
  4622     set_signal_handler(SIGPIPE, true);
  4623     set_signal_handler(SIGBUS, true);
  4624     set_signal_handler(SIGILL, true);
  4625     set_signal_handler(SIGFPE, true);
  4626 #if defined(PPC64)
  4627     set_signal_handler(SIGTRAP, true);
  4628 #endif
  4629     set_signal_handler(SIGXFSZ, true);
  4631     if (libjsig_is_loaded) {
  4632       // Tell libjsig jvm finishes setting signal handlers
  4633       (*end_signal_setting)();
  4636     // We don't activate signal checker if libjsig is in place, we trust ourselves
  4637     // and if UserSignalHandler is installed all bets are off.
  4638     // Log that signal checking is off only if -verbose:jni is specified.
  4639     if (CheckJNICalls) {
  4640       if (libjsig_is_loaded) {
  4641         if (PrintJNIResolving) {
  4642           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
  4644         check_signals = false;
  4646       if (AllowUserSignalHandlers) {
  4647         if (PrintJNIResolving) {
  4648           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
  4650         check_signals = false;
  4656 // This is the fastest way to get thread cpu time on Linux.
  4657 // Returns cpu time (user+sys) for any thread, not only for current.
  4658 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
  4659 // It might work on 2.6.10+ with a special kernel/glibc patch.
  4660 // For reference, please, see IEEE Std 1003.1-2004:
  4661 //   http://www.unix.org/single_unix_specification
  4663 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
  4664   struct timespec tp;
  4665   int rc = os::Linux::clock_gettime(clockid, &tp);
  4666   assert(rc == 0, "clock_gettime is expected to return 0 code");
  4668   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
  4671 /////
  4672 // glibc on Linux platform uses non-documented flag
  4673 // to indicate, that some special sort of signal
  4674 // trampoline is used.
  4675 // We will never set this flag, and we should
  4676 // ignore this flag in our diagnostic
  4677 #ifdef SIGNIFICANT_SIGNAL_MASK
  4678 #undef SIGNIFICANT_SIGNAL_MASK
  4679 #endif
  4680 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
  4682 static const char* get_signal_handler_name(address handler,
  4683                                            char* buf, int buflen) {
  4684   int offset = 0;
  4685   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
  4686   if (found) {
  4687     // skip directory names
  4688     const char *p1, *p2;
  4689     p1 = buf;
  4690     size_t len = strlen(os::file_separator());
  4691     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
  4692     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
  4693   } else {
  4694     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
  4696   return buf;
  4699 static void print_signal_handler(outputStream* st, int sig,
  4700                                  char* buf, size_t buflen) {
  4701   struct sigaction sa;
  4703   sigaction(sig, NULL, &sa);
  4705   // See comment for SIGNIFICANT_SIGNAL_MASK define
  4706   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
  4708   st->print("%s: ", os::exception_name(sig, buf, buflen));
  4710   address handler = (sa.sa_flags & SA_SIGINFO)
  4711     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
  4712     : CAST_FROM_FN_PTR(address, sa.sa_handler);
  4714   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
  4715     st->print("SIG_DFL");
  4716   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
  4717     st->print("SIG_IGN");
  4718   } else {
  4719     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
  4722   st->print(", sa_mask[0]=");
  4723   os::Posix::print_signal_set_short(st, &sa.sa_mask);
  4725   address rh = VMError::get_resetted_sighandler(sig);
  4726   // May be, handler was resetted by VMError?
  4727   if(rh != NULL) {
  4728     handler = rh;
  4729     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
  4732   st->print(", sa_flags=");
  4733   os::Posix::print_sa_flags(st, sa.sa_flags);
  4735   // Check: is it our handler?
  4736   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
  4737      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
  4738     // It is our signal handler
  4739     // check for flags, reset system-used one!
  4740     if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
  4741       st->print(
  4742                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
  4743                 os::Linux::get_our_sigflags(sig));
  4746   st->cr();
  4750 #define DO_SIGNAL_CHECK(sig) \
  4751   if (!sigismember(&check_signal_done, sig)) \
  4752     os::Linux::check_signal_handler(sig)
  4754 // This method is a periodic task to check for misbehaving JNI applications
  4755 // under CheckJNI, we can add any periodic checks here
  4757 void os::run_periodic_checks() {
  4759   if (check_signals == false) return;
  4761   // SEGV and BUS if overridden could potentially prevent
  4762   // generation of hs*.log in the event of a crash, debugging
  4763   // such a case can be very challenging, so we absolutely
  4764   // check the following for a good measure:
  4765   DO_SIGNAL_CHECK(SIGSEGV);
  4766   DO_SIGNAL_CHECK(SIGILL);
  4767   DO_SIGNAL_CHECK(SIGFPE);
  4768   DO_SIGNAL_CHECK(SIGBUS);
  4769   DO_SIGNAL_CHECK(SIGPIPE);
  4770   DO_SIGNAL_CHECK(SIGXFSZ);
  4771 #if defined(PPC64)
  4772   DO_SIGNAL_CHECK(SIGTRAP);
  4773 #endif
  4775   // ReduceSignalUsage allows the user to override these handlers
  4776   // see comments at the very top and jvm_solaris.h
  4777   if (!ReduceSignalUsage) {
  4778     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
  4779     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
  4780     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
  4781     DO_SIGNAL_CHECK(BREAK_SIGNAL);
  4784   DO_SIGNAL_CHECK(SR_signum);
  4785   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
  4788 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
  4790 static os_sigaction_t os_sigaction = NULL;
  4792 void os::Linux::check_signal_handler(int sig) {
  4793   char buf[O_BUFLEN];
  4794   address jvmHandler = NULL;
  4797   struct sigaction act;
  4798   if (os_sigaction == NULL) {
  4799     // only trust the default sigaction, in case it has been interposed
  4800     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
  4801     if (os_sigaction == NULL) return;
  4804   os_sigaction(sig, (struct sigaction*)NULL, &act);
  4807   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
  4809   address thisHandler = (act.sa_flags & SA_SIGINFO)
  4810     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
  4811     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
  4814   switch(sig) {
  4815   case SIGSEGV:
  4816   case SIGBUS:
  4817   case SIGFPE:
  4818   case SIGPIPE:
  4819   case SIGILL:
  4820   case SIGXFSZ:
  4821     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
  4822     break;
  4824   case SHUTDOWN1_SIGNAL:
  4825   case SHUTDOWN2_SIGNAL:
  4826   case SHUTDOWN3_SIGNAL:
  4827   case BREAK_SIGNAL:
  4828     jvmHandler = (address)user_handler();
  4829     break;
  4831   case INTERRUPT_SIGNAL:
  4832     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
  4833     break;
  4835   default:
  4836     if (sig == SR_signum) {
  4837       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
  4838     } else {
  4839       return;
  4841     break;
  4844   if (thisHandler != jvmHandler) {
  4845     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
  4846     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
  4847     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
  4848     // No need to check this sig any longer
  4849     sigaddset(&check_signal_done, sig);
  4850     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
  4851     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
  4852       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
  4853                     exception_name(sig, buf, O_BUFLEN));
  4855   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
  4856     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
  4857     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
  4858     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
  4859     // No need to check this sig any longer
  4860     sigaddset(&check_signal_done, sig);
  4863   // Dump all the signal
  4864   if (sigismember(&check_signal_done, sig)) {
  4865     print_signal_handlers(tty, buf, O_BUFLEN);
  4869 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
  4871 extern bool signal_name(int signo, char* buf, size_t len);
  4873 const char* os::exception_name(int exception_code, char* buf, size_t size) {
  4874   if (0 < exception_code && exception_code <= SIGRTMAX) {
  4875     // signal
  4876     if (!signal_name(exception_code, buf, size)) {
  4877       jio_snprintf(buf, size, "SIG%d", exception_code);
  4879     return buf;
  4880   } else {
  4881     return NULL;
  4885 // this is called _before_ the most of global arguments have been parsed
  4886 void os::init(void) {
  4887   char dummy;   /* used to get a guess on initial stack address */
  4888 //  first_hrtime = gethrtime();
  4890   // With LinuxThreads the JavaMain thread pid (primordial thread)
  4891   // is different than the pid of the java launcher thread.
  4892   // So, on Linux, the launcher thread pid is passed to the VM
  4893   // via the sun.java.launcher.pid property.
  4894   // Use this property instead of getpid() if it was correctly passed.
  4895   // See bug 6351349.
  4896   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
  4898   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
  4900   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
  4902   init_random(1234567);
  4904   ThreadCritical::initialize();
  4906   Linux::set_page_size(sysconf(_SC_PAGESIZE));
  4907   if (Linux::page_size() == -1) {
  4908     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
  4909                   strerror(errno)));
  4911   init_page_sizes((size_t) Linux::page_size());
  4913   Linux::initialize_system_info();
  4915   // main_thread points to the aboriginal thread
  4916   Linux::_main_thread = pthread_self();
  4918   Linux::clock_init();
  4919   initial_time_count = javaTimeNanos();
  4921   // pthread_condattr initialization for monotonic clock
  4922   int status;
  4923   pthread_condattr_t* _condattr = os::Linux::condAttr();
  4924   if ((status = pthread_condattr_init(_condattr)) != 0) {
  4925     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
  4927   // Only set the clock if CLOCK_MONOTONIC is available
  4928   if (Linux::supports_monotonic_clock()) {
  4929     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
  4930       if (status == EINVAL) {
  4931         warning("Unable to use monotonic clock with relative timed-waits" \
  4932                 " - changes to the time-of-day clock may have adverse affects");
  4933       } else {
  4934         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
  4938   // else it defaults to CLOCK_REALTIME
  4940   pthread_mutex_init(&dl_mutex, NULL);
  4942   // If the pagesize of the VM is greater than 8K determine the appropriate
  4943   // number of initial guard pages.  The user can change this with the
  4944   // command line arguments, if needed.
  4945   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
  4946     StackYellowPages = 1;
  4947     StackRedPages = 1;
  4948     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
  4952 // To install functions for atexit system call
  4953 extern "C" {
  4954   static void perfMemory_exit_helper() {
  4955     perfMemory_exit();
  4959 // this is called _after_ the global arguments have been parsed
  4960 jint os::init_2(void)
  4962   Linux::fast_thread_clock_init();
  4964   // Allocate a single page and mark it as readable for safepoint polling
  4965 #ifdef OPT_SAFEPOINT
  4966   void * p = (void *)(0x10000);
  4967   address polling_page = (address) ::mmap(p, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
  4968 #else
  4969   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
  4970 #endif
  4971   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
  4973   os::set_polling_page( polling_page );
  4975 #ifndef PRODUCT
  4976   if(Verbose && PrintMiscellaneous)
  4977     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
  4978 #endif
  4980   if (!UseMembar) {
  4981     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
  4982     guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
  4983     os::set_memory_serialize_page( mem_serialize_page );
  4985 #ifndef PRODUCT
  4986     if(Verbose && PrintMiscellaneous)
  4987       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
  4988 #endif
  4991   // initialize suspend/resume support - must do this before signal_sets_init()
  4992   if (SR_initialize() != 0) {
  4993     perror("SR_initialize failed");
  4994     return JNI_ERR;
  4997   Linux::signal_sets_init();
  4998   Linux::install_signal_handlers();
  5000   // Check minimum allowable stack size for thread creation and to initialize
  5001   // the java system classes, including StackOverflowError - depends on page
  5002   // size.  Add a page for compiler2 recursion in main thread.
  5003   // Add in 2*BytesPerWord times page size to account for VM stack during
  5004   // class initialization depending on 32 or 64 bit VM.
  5006   /* 2014/1/2 Liao: JDK8 requires larger -Xss option.
  5007    *   TongWeb cannot run with -Xss192K.
  5008    *   We are not sure whether this causes errors, so simply print a warning. */
  5009   size_t min_stack_allowed_jdk6 = os::Linux::min_stack_allowed;
  5010   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
  5011             (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
  5012                     (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
  5014   size_t threadStackSizeInBytes = ThreadStackSize * K;
  5015   if (threadStackSizeInBytes != 0 &&
  5016       threadStackSizeInBytes < min_stack_allowed_jdk6) {
  5017         tty->print_cr("\nThe stack size specified is too small, "
  5018                       "Specify at least %dk",
  5019                       os::Linux::min_stack_allowed/ K);
  5020         return JNI_ERR;
  5023   // Make the stack size a multiple of the page size so that
  5024   // the yellow/red zones can be guarded.
  5025   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
  5026         vm_page_size()));
  5028   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
  5030 #if defined(IA32)
  5031   workaround_expand_exec_shield_cs_limit();
  5032 #endif
  5034   Linux::libpthread_init();
  5035   if (PrintMiscellaneous && (Verbose || WizardMode)) {
  5036      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
  5037           Linux::glibc_version(), Linux::libpthread_version(),
  5038           Linux::is_floating_stack() ? "floating stack" : "fixed stack");
  5041   if (UseNUMA) {
  5042     if (!Linux::libnuma_init()) {
  5043       UseNUMA = false;
  5044     } else {
  5045       if ((Linux::numa_max_node() < 1)) {
  5046         // There's only one node(they start from 0), disable NUMA.
  5047         UseNUMA = false;
  5050     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
  5051     // we can make the adaptive lgrp chunk resizing work. If the user specified
  5052     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
  5053     // disable adaptive resizing.
  5054     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
  5055       if (FLAG_IS_DEFAULT(UseNUMA)) {
  5056         UseNUMA = false;
  5057       } else {
  5058         if (FLAG_IS_DEFAULT(UseLargePages) &&
  5059             FLAG_IS_DEFAULT(UseSHM) &&
  5060             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
  5061           UseLargePages = false;
  5062         } else {
  5063           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
  5064           UseAdaptiveSizePolicy = false;
  5065           UseAdaptiveNUMAChunkSizing = false;
  5069     if (!UseNUMA && ForceNUMA) {
  5070       UseNUMA = true;
  5074   if (MaxFDLimit) {
  5075     // set the number of file descriptors to max. print out error
  5076     // if getrlimit/setrlimit fails but continue regardless.
  5077     struct rlimit nbr_files;
  5078     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
  5079     if (status != 0) {
  5080       if (PrintMiscellaneous && (Verbose || WizardMode))
  5081         perror("os::init_2 getrlimit failed");
  5082     } else {
  5083       nbr_files.rlim_cur = nbr_files.rlim_max;
  5084       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
  5085       if (status != 0) {
  5086         if (PrintMiscellaneous && (Verbose || WizardMode))
  5087           perror("os::init_2 setrlimit failed");
  5092   // Initialize lock used to serialize thread creation (see os::create_thread)
  5093   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
  5095   // at-exit methods are called in the reverse order of their registration.
  5096   // atexit functions are called on return from main or as a result of a
  5097   // call to exit(3C). There can be only 32 of these functions registered
  5098   // and atexit() does not set errno.
  5100   if (PerfAllowAtExitRegistration) {
  5101     // only register atexit functions if PerfAllowAtExitRegistration is set.
  5102     // atexit functions can be delayed until process exit time, which
  5103     // can be problematic for embedded VM situations. Embedded VMs should
  5104     // call DestroyJavaVM() to assure that VM resources are released.
  5106     // note: perfMemory_exit_helper atexit function may be removed in
  5107     // the future if the appropriate cleanup code can be added to the
  5108     // VM_Exit VMOperation's doit method.
  5109     if (atexit(perfMemory_exit_helper) != 0) {
  5110       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
  5114   // initialize thread priority policy
  5115   prio_init();
  5117   return JNI_OK;
  5120 // Mark the polling page as unreadable
  5121 void os::make_polling_page_unreadable(void) {
  5122   if( !guard_memory((char*)_polling_page, Linux::page_size()) )
  5123     fatal("Could not disable polling page");
  5124 };
  5126 // Mark the polling page as readable
  5127 void os::make_polling_page_readable(void) {
  5128   if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
  5129     fatal("Could not enable polling page");
  5131 };
  5133 static int os_cpu_count(const cpu_set_t* cpus) {
  5134   int count = 0;
  5135   // only look up to the number of configured processors
  5136   for (int i = 0; i < os::processor_count(); i++) {
  5137     if (CPU_ISSET(i, cpus)) {
  5138       count++;
  5141   return count;
  5144 // Get the current number of available processors for this process.
  5145 // This value can change at any time during a process's lifetime.
  5146 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
  5147 // If anything goes wrong we fallback to returning the number of online
  5148 // processors - which can be greater than the number available to the process.
  5149 int os::active_processor_count() {
  5150   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
  5151   int cpus_size = sizeof(cpu_set_t);
  5152   int cpu_count = 0;
  5154   // pid 0 means the current thread - which we have to assume represents the process
  5155   if (sched_getaffinity(0, cpus_size, &cpus) == 0) {
  5156     cpu_count = os_cpu_count(&cpus);
  5157     if (PrintActiveCpus) {
  5158       tty->print_cr("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
  5161   else {
  5162     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
  5163     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
  5164             "which may exceed available processors", strerror(errno), cpu_count);
  5167   assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
  5168   return cpu_count;
  5171 void os::set_native_thread_name(const char *name) {
  5172   // Not yet implemented.
  5173   return;
  5176 bool os::distribute_processes(uint length, uint* distribution) {
  5177   // Not yet implemented.
  5178   return false;
  5181 bool os::bind_to_processor(uint processor_id) {
  5182   // Not yet implemented.
  5183   return false;
  5186 ///
  5188 void os::SuspendedThreadTask::internal_do_task() {
  5189   if (do_suspend(_thread->osthread())) {
  5190     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
  5191     do_task(context);
  5192     do_resume(_thread->osthread());
  5196 class PcFetcher : public os::SuspendedThreadTask {
  5197 public:
  5198   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
  5199   ExtendedPC result();
  5200 protected:
  5201   void do_task(const os::SuspendedThreadTaskContext& context);
  5202 private:
  5203   ExtendedPC _epc;
  5204 };
  5206 ExtendedPC PcFetcher::result() {
  5207   guarantee(is_done(), "task is not done yet.");
  5208   return _epc;
  5211 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
  5212   Thread* thread = context.thread();
  5213   OSThread* osthread = thread->osthread();
  5214   if (osthread->ucontext() != NULL) {
  5215     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
  5216   } else {
  5217     // NULL context is unexpected, double-check this is the VMThread
  5218     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
  5222 // Suspends the target using the signal mechanism and then grabs the PC before
  5223 // resuming the target. Used by the flat-profiler only
  5224 ExtendedPC os::get_thread_pc(Thread* thread) {
  5225   // Make sure that it is called by the watcher for the VMThread
  5226   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
  5227   assert(thread->is_VM_thread(), "Can only be called for VMThread");
  5229   PcFetcher fetcher(thread);
  5230   fetcher.run();
  5231   return fetcher.result();
  5234 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
  5236    if (is_NPTL()) {
  5237       return pthread_cond_timedwait(_cond, _mutex, _abstime);
  5238    } else {
  5239       // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
  5240       // word back to default 64bit precision if condvar is signaled. Java
  5241       // wants 53bit precision.  Save and restore current value.
  5242       int fpu = get_fpu_control_word();
  5243       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
  5244       set_fpu_control_word(fpu);
  5245       return status;
  5249 ////////////////////////////////////////////////////////////////////////////////
  5250 // debug support
  5252 bool os::find(address addr, outputStream* st) {
  5253   Dl_info dlinfo;
  5254   memset(&dlinfo, 0, sizeof(dlinfo));
  5255   if (dladdr(addr, &dlinfo) != 0) {
  5256     st->print(PTR_FORMAT ": ", addr);
  5257     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
  5258       st->print("%s+%#x", dlinfo.dli_sname,
  5259                  addr - (intptr_t)dlinfo.dli_saddr);
  5260     } else if (dlinfo.dli_fbase != NULL) {
  5261       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
  5262     } else {
  5263       st->print("<absolute address>");
  5265     if (dlinfo.dli_fname != NULL) {
  5266       st->print(" in %s", dlinfo.dli_fname);
  5268     if (dlinfo.dli_fbase != NULL) {
  5269       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
  5271     st->cr();
  5273     if (Verbose) {
  5274       // decode some bytes around the PC
  5275       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
  5276       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
  5277       address       lowest = (address) dlinfo.dli_sname;
  5278       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
  5279       if (begin < lowest)  begin = lowest;
  5280       Dl_info dlinfo2;
  5281       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
  5282           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
  5283         end = (address) dlinfo2.dli_saddr;
  5284       Disassembler::decode(begin, end, st);
  5286     return true;
  5288   return false;
  5291 ////////////////////////////////////////////////////////////////////////////////
  5292 // misc
  5294 // This does not do anything on Linux. This is basically a hook for being
  5295 // able to use structured exception handling (thread-local exception filters)
  5296 // on, e.g., Win32.
  5297 void
  5298 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
  5299                          JavaCallArguments* args, Thread* thread) {
  5300   f(value, method, args, thread);
  5303 void os::print_statistics() {
  5306 int os::message_box(const char* title, const char* message) {
  5307   int i;
  5308   fdStream err(defaultStream::error_fd());
  5309   for (i = 0; i < 78; i++) err.print_raw("=");
  5310   err.cr();
  5311   err.print_raw_cr(title);
  5312   for (i = 0; i < 78; i++) err.print_raw("-");
  5313   err.cr();
  5314   err.print_raw_cr(message);
  5315   for (i = 0; i < 78; i++) err.print_raw("=");
  5316   err.cr();
  5318   char buf[16];
  5319   // Prevent process from exiting upon "read error" without consuming all CPU
  5320   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
  5322   return buf[0] == 'y' || buf[0] == 'Y';
  5325 int os::stat(const char *path, struct stat *sbuf) {
  5326   char pathbuf[MAX_PATH];
  5327   if (strlen(path) > MAX_PATH - 1) {
  5328     errno = ENAMETOOLONG;
  5329     return -1;
  5331   os::native_path(strcpy(pathbuf, path));
  5332   return ::stat(pathbuf, sbuf);
  5335 bool os::check_heap(bool force) {
  5336   return true;
  5339 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
  5340   return ::vsnprintf(buf, count, format, args);
  5343 // Is a (classpath) directory empty?
  5344 bool os::dir_is_empty(const char* path) {
  5345   DIR *dir = NULL;
  5346   struct dirent *ptr;
  5348   dir = opendir(path);
  5349   if (dir == NULL) return true;
  5351   /* Scan the directory */
  5352   bool result = true;
  5353   char buf[sizeof(struct dirent) + MAX_PATH];
  5354   while (result && (ptr = ::readdir(dir)) != NULL) {
  5355     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
  5356       result = false;
  5359   closedir(dir);
  5360   return result;
  5363 // This code originates from JDK's sysOpen and open64_w
  5364 // from src/solaris/hpi/src/system_md.c
  5366 #ifndef O_DELETE
  5367 #define O_DELETE 0x10000
  5368 #endif
  5370 // Open a file. Unlink the file immediately after open returns
  5371 // if the specified oflag has the O_DELETE flag set.
  5372 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
  5374 int os::open(const char *path, int oflag, int mode) {
  5376   if (strlen(path) > MAX_PATH - 1) {
  5377     errno = ENAMETOOLONG;
  5378     return -1;
  5380   int fd;
  5381   int o_delete = (oflag & O_DELETE);
  5382   oflag = oflag & ~O_DELETE;
  5384   fd = ::open64(path, oflag, mode);
  5385   if (fd == -1) return -1;
  5387   //If the open succeeded, the file might still be a directory
  5389     struct stat64 buf64;
  5390     int ret = ::fstat64(fd, &buf64);
  5391     int st_mode = buf64.st_mode;
  5393     if (ret != -1) {
  5394       if ((st_mode & S_IFMT) == S_IFDIR) {
  5395         errno = EISDIR;
  5396         ::close(fd);
  5397         return -1;
  5399     } else {
  5400       ::close(fd);
  5401       return -1;
  5405     /*
  5406      * All file descriptors that are opened in the JVM and not
  5407      * specifically destined for a subprocess should have the
  5408      * close-on-exec flag set.  If we don't set it, then careless 3rd
  5409      * party native code might fork and exec without closing all
  5410      * appropriate file descriptors (e.g. as we do in closeDescriptors in
  5411      * UNIXProcess.c), and this in turn might:
  5413      * - cause end-of-file to fail to be detected on some file
  5414      *   descriptors, resulting in mysterious hangs, or
  5416      * - might cause an fopen in the subprocess to fail on a system
  5417      *   suffering from bug 1085341.
  5419      * (Yes, the default setting of the close-on-exec flag is a Unix
  5420      * design flaw)
  5422      * See:
  5423      * 1085341: 32-bit stdio routines should support file descriptors >255
  5424      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
  5425      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
  5426      */
  5427 #ifdef FD_CLOEXEC
  5429         int flags = ::fcntl(fd, F_GETFD);
  5430         if (flags != -1)
  5431             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
  5433 #endif
  5435   if (o_delete != 0) {
  5436     ::unlink(path);
  5438   return fd;
  5442 // create binary file, rewriting existing file if required
  5443 int os::create_binary_file(const char* path, bool rewrite_existing) {
  5444   int oflags = O_WRONLY | O_CREAT;
  5445   if (!rewrite_existing) {
  5446     oflags |= O_EXCL;
  5448   return ::open64(path, oflags, S_IREAD | S_IWRITE);
  5451 // return current position of file pointer
  5452 jlong os::current_file_offset(int fd) {
  5453   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
  5456 // move file pointer to the specified offset
  5457 jlong os::seek_to_file_offset(int fd, jlong offset) {
  5458   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
  5461 // This code originates from JDK's sysAvailable
  5462 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
  5464 int os::available(int fd, jlong *bytes) {
  5465   jlong cur, end;
  5466   int mode;
  5467   struct stat64 buf64;
  5469   if (::fstat64(fd, &buf64) >= 0) {
  5470     mode = buf64.st_mode;
  5471     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
  5472       /*
  5473       * XXX: is the following call interruptible? If so, this might
  5474       * need to go through the INTERRUPT_IO() wrapper as for other
  5475       * blocking, interruptible calls in this file.
  5476       */
  5477       int n;
  5478       if (::ioctl(fd, FIONREAD, &n) >= 0) {
  5479         *bytes = n;
  5480         return 1;
  5484   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
  5485     return 0;
  5486   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
  5487     return 0;
  5488   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
  5489     return 0;
  5491   *bytes = end - cur;
  5492   return 1;
  5495 int os::socket_available(int fd, jint *pbytes) {
  5496   // Linux doc says EINTR not returned, unlike Solaris
  5497   int ret = ::ioctl(fd, FIONREAD, pbytes);
  5499   //%% note ioctl can return 0 when successful, JVM_SocketAvailable
  5500   // is expected to return 0 on failure and 1 on success to the jdk.
  5501   return (ret < 0) ? 0 : 1;
  5504 // Map a block of memory.
  5505 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
  5506                      char *addr, size_t bytes, bool read_only,
  5507                      bool allow_exec) {
  5508   int prot;
  5509   int flags = MAP_PRIVATE;
  5511   if (read_only) {
  5512     prot = PROT_READ;
  5513   } else {
  5514     prot = PROT_READ | PROT_WRITE;
  5517   if (allow_exec) {
  5518     prot |= PROT_EXEC;
  5521   if (addr != NULL) {
  5522     flags |= MAP_FIXED;
  5525   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
  5526                                      fd, file_offset);
  5527   if (mapped_address == MAP_FAILED) {
  5528     return NULL;
  5530   return mapped_address;
  5534 // Remap a block of memory.
  5535 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
  5536                        char *addr, size_t bytes, bool read_only,
  5537                        bool allow_exec) {
  5538   // same as map_memory() on this OS
  5539   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
  5540                         allow_exec);
  5544 // Unmap a block of memory.
  5545 bool os::pd_unmap_memory(char* addr, size_t bytes) {
  5546   return munmap(addr, bytes) == 0;
  5549 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
  5551 static clockid_t thread_cpu_clockid(Thread* thread) {
  5552   pthread_t tid = thread->osthread()->pthread_id();
  5553   clockid_t clockid;
  5555   // Get thread clockid
  5556   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
  5557   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
  5558   return clockid;
  5561 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
  5562 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
  5563 // of a thread.
  5564 //
  5565 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
  5566 // the fast estimate available on the platform.
  5568 jlong os::current_thread_cpu_time() {
  5569   if (os::Linux::supports_fast_thread_cpu_time()) {
  5570     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
  5571   } else {
  5572     // return user + sys since the cost is the same
  5573     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
  5577 jlong os::thread_cpu_time(Thread* thread) {
  5578   // consistent with what current_thread_cpu_time() returns
  5579   if (os::Linux::supports_fast_thread_cpu_time()) {
  5580     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
  5581   } else {
  5582     return slow_thread_cpu_time(thread, true /* user + sys */);
  5586 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
  5587   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
  5588     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
  5589   } else {
  5590     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
  5594 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
  5595   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
  5596     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
  5597   } else {
  5598     return slow_thread_cpu_time(thread, user_sys_cpu_time);
  5602 //
  5603 //  -1 on error.
  5604 //
  5606 PRAGMA_DIAG_PUSH
  5607 PRAGMA_FORMAT_NONLITERAL_IGNORED
  5608 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
  5609   static bool proc_task_unchecked = true;
  5610   static const char *proc_stat_path = "/proc/%d/stat";
  5611   pid_t  tid = thread->osthread()->thread_id();
  5612   char *s;
  5613   char stat[2048];
  5614   int statlen;
  5615   char proc_name[64];
  5616   int count;
  5617   long sys_time, user_time;
  5618   char cdummy;
  5619   int idummy;
  5620   long ldummy;
  5621   FILE *fp;
  5623   // The /proc/<tid>/stat aggregates per-process usage on
  5624   // new Linux kernels 2.6+ where NPTL is supported.
  5625   // The /proc/self/task/<tid>/stat still has the per-thread usage.
  5626   // See bug 6328462.
  5627   // There possibly can be cases where there is no directory
  5628   // /proc/self/task, so we check its availability.
  5629   if (proc_task_unchecked && os::Linux::is_NPTL()) {
  5630     // This is executed only once
  5631     proc_task_unchecked = false;
  5632     fp = fopen("/proc/self/task", "r");
  5633     if (fp != NULL) {
  5634       proc_stat_path = "/proc/self/task/%d/stat";
  5635       fclose(fp);
  5639   sprintf(proc_name, proc_stat_path, tid);
  5640   fp = fopen(proc_name, "r");
  5641   if ( fp == NULL ) return -1;
  5642   statlen = fread(stat, 1, 2047, fp);
  5643   stat[statlen] = '\0';
  5644   fclose(fp);
  5646   // Skip pid and the command string. Note that we could be dealing with
  5647   // weird command names, e.g. user could decide to rename java launcher
  5648   // to "java 1.4.2 :)", then the stat file would look like
  5649   //                1234 (java 1.4.2 :)) R ... ...
  5650   // We don't really need to know the command string, just find the last
  5651   // occurrence of ")" and then start parsing from there. See bug 4726580.
  5652   s = strrchr(stat, ')');
  5653   if (s == NULL ) return -1;
  5655   // Skip blank chars
  5656   do s++; while (isspace(*s));
  5658   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
  5659                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
  5660                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
  5661                  &user_time, &sys_time);
  5662   if ( count != 13 ) return -1;
  5663   if (user_sys_cpu_time) {
  5664     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
  5665   } else {
  5666     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
  5669 PRAGMA_DIAG_POP
  5671 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
  5672   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
  5673   info_ptr->may_skip_backward = false;     // elapsed time not wall time
  5674   info_ptr->may_skip_forward = false;      // elapsed time not wall time
  5675   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
  5678 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
  5679   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
  5680   info_ptr->may_skip_backward = false;     // elapsed time not wall time
  5681   info_ptr->may_skip_forward = false;      // elapsed time not wall time
  5682   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
  5685 bool os::is_thread_cpu_time_supported() {
  5686   return true;
  5689 // System loadavg support.  Returns -1 if load average cannot be obtained.
  5690 // Linux doesn't yet have a (official) notion of processor sets,
  5691 // so just return the system wide load average.
  5692 int os::loadavg(double loadavg[], int nelem) {
  5693   return ::getloadavg(loadavg, nelem);
  5696 void os::pause() {
  5697   char filename[MAX_PATH];
  5698   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
  5699     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
  5700   } else {
  5701     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
  5704   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
  5705   if (fd != -1) {
  5706     struct stat buf;
  5707     ::close(fd);
  5708     while (::stat(filename, &buf) == 0) {
  5709       (void)::poll(NULL, 0, 100);
  5711   } else {
  5712     jio_fprintf(stderr,
  5713       "Could not open pause file '%s', continuing immediately.\n", filename);
  5718 // Refer to the comments in os_solaris.cpp park-unpark.
  5719 //
  5720 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
  5721 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
  5722 // For specifics regarding the bug see GLIBC BUGID 261237 :
  5723 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
  5724 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
  5725 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
  5726 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
  5727 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
  5728 // and monitorenter when we're using 1-0 locking.  All those operations may result in
  5729 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
  5730 // of libpthread avoids the problem, but isn't practical.
  5731 //
  5732 // Possible remedies:
  5733 //
  5734 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
  5735 //      This is palliative and probabilistic, however.  If the thread is preempted
  5736 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
  5737 //      than the minimum period may have passed, and the abstime may be stale (in the
  5738 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
  5739 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
  5740 //
  5741 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
  5742 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
  5743 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
  5744 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
  5745 //      thread.
  5746 //
  5747 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
  5748 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
  5749 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
  5750 //      This also works well.  In fact it avoids kernel-level scalability impediments
  5751 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
  5752 //      timers in a graceful fashion.
  5753 //
  5754 // 4.   When the abstime value is in the past it appears that control returns
  5755 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
  5756 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
  5757 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
  5758 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
  5759 //      It may be possible to avoid reinitialization by checking the return
  5760 //      value from pthread_cond_timedwait().  In addition to reinitializing the
  5761 //      condvar we must establish the invariant that cond_signal() is only called
  5762 //      within critical sections protected by the adjunct mutex.  This prevents
  5763 //      cond_signal() from "seeing" a condvar that's in the midst of being
  5764 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
  5765 //      desirable signal-after-unlock optimization that avoids futile context switching.
  5766 //
  5767 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
  5768 //      structure when a condvar is used or initialized.  cond_destroy()  would
  5769 //      release the helper structure.  Our reinitialize-after-timedwait fix
  5770 //      put excessive stress on malloc/free and locks protecting the c-heap.
  5771 //
  5772 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
  5773 // It may be possible to refine (4) by checking the kernel and NTPL verisons
  5774 // and only enabling the work-around for vulnerable environments.
  5776 // utility to compute the abstime argument to timedwait:
  5777 // millis is the relative timeout time
  5778 // abstime will be the absolute timeout time
  5779 // TODO: replace compute_abstime() with unpackTime()
  5781 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
  5782   if (millis < 0)  millis = 0;
  5784   jlong seconds = millis / 1000;
  5785   millis %= 1000;
  5786   if (seconds > 50000000) { // see man cond_timedwait(3T)
  5787     seconds = 50000000;
  5790   if (os::Linux::supports_monotonic_clock()) {
  5791     struct timespec now;
  5792     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
  5793     assert_status(status == 0, status, "clock_gettime");
  5794     abstime->tv_sec = now.tv_sec  + seconds;
  5795     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
  5796     if (nanos >= NANOSECS_PER_SEC) {
  5797       abstime->tv_sec += 1;
  5798       nanos -= NANOSECS_PER_SEC;
  5800     abstime->tv_nsec = nanos;
  5801   } else {
  5802     struct timeval now;
  5803     int status = gettimeofday(&now, NULL);
  5804     assert(status == 0, "gettimeofday");
  5805     abstime->tv_sec = now.tv_sec  + seconds;
  5806     long usec = now.tv_usec + millis * 1000;
  5807     if (usec >= 1000000) {
  5808       abstime->tv_sec += 1;
  5809       usec -= 1000000;
  5811     abstime->tv_nsec = usec * 1000;
  5813   return abstime;
  5817 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
  5818 // Conceptually TryPark() should be equivalent to park(0).
  5820 int os::PlatformEvent::TryPark() {
  5821   for (;;) {
  5822     const int v = _Event ;
  5823     guarantee ((v == 0) || (v == 1), "invariant") ;
  5824     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
  5828 void os::PlatformEvent::park() {       // AKA "down()"
  5829   // Invariant: Only the thread associated with the Event/PlatformEvent
  5830   // may call park().
  5831   // TODO: assert that _Assoc != NULL or _Assoc == Self
  5832   int v ;
  5833   for (;;) {
  5834       v = _Event ;
  5835       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
  5837   guarantee (v >= 0, "invariant") ;
  5838   if (v == 0) {
  5839      // Do this the hard way by blocking ...
  5840      int status = pthread_mutex_lock(_mutex);
  5841      assert_status(status == 0, status, "mutex_lock");
  5842      guarantee (_nParked == 0, "invariant") ;
  5843      ++ _nParked ;
  5844      while (_Event < 0) {
  5845         status = pthread_cond_wait(_cond, _mutex);
  5846         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
  5847         // Treat this the same as if the wait was interrupted
  5848         if (status == ETIME) { status = EINTR; }
  5849         assert_status(status == 0 || status == EINTR, status, "cond_wait");
  5851      -- _nParked ;
  5853     _Event = 0 ;
  5854      status = pthread_mutex_unlock(_mutex);
  5855      assert_status(status == 0, status, "mutex_unlock");
  5856     // Paranoia to ensure our locked and lock-free paths interact
  5857     // correctly with each other.
  5858     OrderAccess::fence();
  5860   guarantee (_Event >= 0, "invariant") ;
  5863 int os::PlatformEvent::park(jlong millis) {
  5864   guarantee (_nParked == 0, "invariant") ;
  5866   int v ;
  5867   for (;;) {
  5868       v = _Event ;
  5869       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
  5871   guarantee (v >= 0, "invariant") ;
  5872   if (v != 0) return OS_OK ;
  5874   // We do this the hard way, by blocking the thread.
  5875   // Consider enforcing a minimum timeout value.
  5876   struct timespec abst;
  5877   compute_abstime(&abst, millis);
  5879   int ret = OS_TIMEOUT;
  5880   int status = pthread_mutex_lock(_mutex);
  5881   assert_status(status == 0, status, "mutex_lock");
  5882   guarantee (_nParked == 0, "invariant") ;
  5883   ++_nParked ;
  5885   // Object.wait(timo) will return because of
  5886   // (a) notification
  5887   // (b) timeout
  5888   // (c) thread.interrupt
  5889   //
  5890   // Thread.interrupt and object.notify{All} both call Event::set.
  5891   // That is, we treat thread.interrupt as a special case of notification.
  5892   // The underlying Solaris implementation, cond_timedwait, admits
  5893   // spurious/premature wakeups, but the JLS/JVM spec prevents the
  5894   // JVM from making those visible to Java code.  As such, we must
  5895   // filter out spurious wakeups.  We assume all ETIME returns are valid.
  5896   //
  5897   // TODO: properly differentiate simultaneous notify+interrupt.
  5898   // In that case, we should propagate the notify to another waiter.
  5900   while (_Event < 0) {
  5901     status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
  5902     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
  5903       pthread_cond_destroy (_cond);
  5904       pthread_cond_init (_cond, os::Linux::condAttr()) ;
  5906     assert_status(status == 0 || status == EINTR ||
  5907                   status == ETIME || status == ETIMEDOUT,
  5908                   status, "cond_timedwait");
  5909     if (!FilterSpuriousWakeups) break ;                 // previous semantics
  5910     if (status == ETIME || status == ETIMEDOUT) break ;
  5911     // We consume and ignore EINTR and spurious wakeups.
  5913   --_nParked ;
  5914   if (_Event >= 0) {
  5915      ret = OS_OK;
  5917   _Event = 0 ;
  5918   status = pthread_mutex_unlock(_mutex);
  5919   assert_status(status == 0, status, "mutex_unlock");
  5920   assert (_nParked == 0, "invariant") ;
  5921   // Paranoia to ensure our locked and lock-free paths interact
  5922   // correctly with each other.
  5923   OrderAccess::fence();
  5924   return ret;
  5927 void os::PlatformEvent::unpark() {
  5928   // Transitions for _Event:
  5929   //    0 :=> 1
  5930   //    1 :=> 1
  5931   //   -1 :=> either 0 or 1; must signal target thread
  5932   //          That is, we can safely transition _Event from -1 to either
  5933   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
  5934   //          unpark() calls.
  5935   // See also: "Semaphores in Plan 9" by Mullender & Cox
  5936   //
  5937   // Note: Forcing a transition from "-1" to "1" on an unpark() means
  5938   // that it will take two back-to-back park() calls for the owning
  5939   // thread to block. This has the benefit of forcing a spurious return
  5940   // from the first park() call after an unpark() call which will help
  5941   // shake out uses of park() and unpark() without condition variables.
  5943   if (Atomic::xchg(1, &_Event) >= 0) return;
  5945   // Wait for the thread associated with the event to vacate
  5946   int status = pthread_mutex_lock(_mutex);
  5947   assert_status(status == 0, status, "mutex_lock");
  5948   int AnyWaiters = _nParked;
  5949   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
  5950   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
  5951     AnyWaiters = 0;
  5952     pthread_cond_signal(_cond);
  5954   status = pthread_mutex_unlock(_mutex);
  5955   assert_status(status == 0, status, "mutex_unlock");
  5956   if (AnyWaiters != 0) {
  5957     status = pthread_cond_signal(_cond);
  5958     assert_status(status == 0, status, "cond_signal");
  5961   // Note that we signal() _after dropping the lock for "immortal" Events.
  5962   // This is safe and avoids a common class of  futile wakeups.  In rare
  5963   // circumstances this can cause a thread to return prematurely from
  5964   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
  5965   // simply re-test the condition and re-park itself.
  5969 // JSR166
  5970 // -------------------------------------------------------
  5972 /*
  5973  * The solaris and linux implementations of park/unpark are fairly
  5974  * conservative for now, but can be improved. They currently use a
  5975  * mutex/condvar pair, plus a a count.
  5976  * Park decrements count if > 0, else does a condvar wait.  Unpark
  5977  * sets count to 1 and signals condvar.  Only one thread ever waits
  5978  * on the condvar. Contention seen when trying to park implies that someone
  5979  * is unparking you, so don't wait. And spurious returns are fine, so there
  5980  * is no need to track notifications.
  5981  */
  5983 /*
  5984  * This code is common to linux and solaris and will be moved to a
  5985  * common place in dolphin.
  5987  * The passed in time value is either a relative time in nanoseconds
  5988  * or an absolute time in milliseconds. Either way it has to be unpacked
  5989  * into suitable seconds and nanoseconds components and stored in the
  5990  * given timespec structure.
  5991  * Given time is a 64-bit value and the time_t used in the timespec is only
  5992  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
  5993  * overflow if times way in the future are given. Further on Solaris versions
  5994  * prior to 10 there is a restriction (see cond_timedwait) that the specified
  5995  * number of seconds, in abstime, is less than current_time  + 100,000,000.
  5996  * As it will be 28 years before "now + 100000000" will overflow we can
  5997  * ignore overflow and just impose a hard-limit on seconds using the value
  5998  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
  5999  * years from "now".
  6000  */
  6002 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
  6003   assert (time > 0, "convertTime");
  6004   time_t max_secs = 0;
  6006   if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
  6007     struct timeval now;
  6008     int status = gettimeofday(&now, NULL);
  6009     assert(status == 0, "gettimeofday");
  6011     max_secs = now.tv_sec + MAX_SECS;
  6013     if (isAbsolute) {
  6014       jlong secs = time / 1000;
  6015       if (secs > max_secs) {
  6016         absTime->tv_sec = max_secs;
  6017       } else {
  6018         absTime->tv_sec = secs;
  6020       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
  6021     } else {
  6022       jlong secs = time / NANOSECS_PER_SEC;
  6023       if (secs >= MAX_SECS) {
  6024         absTime->tv_sec = max_secs;
  6025         absTime->tv_nsec = 0;
  6026       } else {
  6027         absTime->tv_sec = now.tv_sec + secs;
  6028         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
  6029         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
  6030           absTime->tv_nsec -= NANOSECS_PER_SEC;
  6031           ++absTime->tv_sec; // note: this must be <= max_secs
  6035   } else {
  6036     // must be relative using monotonic clock
  6037     struct timespec now;
  6038     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
  6039     assert_status(status == 0, status, "clock_gettime");
  6040     max_secs = now.tv_sec + MAX_SECS;
  6041     jlong secs = time / NANOSECS_PER_SEC;
  6042     if (secs >= MAX_SECS) {
  6043       absTime->tv_sec = max_secs;
  6044       absTime->tv_nsec = 0;
  6045     } else {
  6046       absTime->tv_sec = now.tv_sec + secs;
  6047       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
  6048       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
  6049         absTime->tv_nsec -= NANOSECS_PER_SEC;
  6050         ++absTime->tv_sec; // note: this must be <= max_secs
  6054   assert(absTime->tv_sec >= 0, "tv_sec < 0");
  6055   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
  6056   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
  6057   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
  6060 void Parker::park(bool isAbsolute, jlong time) {
  6061   // Ideally we'd do something useful while spinning, such
  6062   // as calling unpackTime().
  6064   // Optional fast-path check:
  6065   // Return immediately if a permit is available.
  6066   // We depend on Atomic::xchg() having full barrier semantics
  6067   // since we are doing a lock-free update to _counter.
  6068   if (Atomic::xchg(0, &_counter) > 0) return;
  6070   Thread* thread = Thread::current();
  6071   assert(thread->is_Java_thread(), "Must be JavaThread");
  6072   JavaThread *jt = (JavaThread *)thread;
  6074   // Optional optimization -- avoid state transitions if there's an interrupt pending.
  6075   // Check interrupt before trying to wait
  6076   if (Thread::is_interrupted(thread, false)) {
  6077     return;
  6080   // Next, demultiplex/decode time arguments
  6081   timespec absTime;
  6082   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
  6083     return;
  6085   if (time > 0) {
  6086     unpackTime(&absTime, isAbsolute, time);
  6090   // Enter safepoint region
  6091   // Beware of deadlocks such as 6317397.
  6092   // The per-thread Parker:: mutex is a classic leaf-lock.
  6093   // In particular a thread must never block on the Threads_lock while
  6094   // holding the Parker:: mutex.  If safepoints are pending both the
  6095   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
  6096   ThreadBlockInVM tbivm(jt);
  6098   // Don't wait if cannot get lock since interference arises from
  6099   // unblocking.  Also. check interrupt before trying wait
  6100   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
  6101     return;
  6104   int status ;
  6105   if (_counter > 0)  { // no wait needed
  6106     _counter = 0;
  6107     status = pthread_mutex_unlock(_mutex);
  6108     assert (status == 0, "invariant") ;
  6109     // Paranoia to ensure our locked and lock-free paths interact
  6110     // correctly with each other and Java-level accesses.
  6111     OrderAccess::fence();
  6112     return;
  6115 #ifdef ASSERT
  6116   // Don't catch signals while blocked; let the running threads have the signals.
  6117   // (This allows a debugger to break into the running thread.)
  6118   sigset_t oldsigs;
  6119   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
  6120   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
  6121 #endif
  6123   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
  6124   jt->set_suspend_equivalent();
  6125   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
  6127   assert(_cur_index == -1, "invariant");
  6128   if (time == 0) {
  6129     _cur_index = REL_INDEX; // arbitrary choice when not timed
  6130     status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
  6131   } else {
  6132     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
  6133     status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
  6134     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
  6135       pthread_cond_destroy (&_cond[_cur_index]) ;
  6136       pthread_cond_init    (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
  6139   _cur_index = -1;
  6140   assert_status(status == 0 || status == EINTR ||
  6141                 status == ETIME || status == ETIMEDOUT,
  6142                 status, "cond_timedwait");
  6144 #ifdef ASSERT
  6145   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
  6146 #endif
  6148   _counter = 0 ;
  6149   status = pthread_mutex_unlock(_mutex) ;
  6150   assert_status(status == 0, status, "invariant") ;
  6151   // Paranoia to ensure our locked and lock-free paths interact
  6152   // correctly with each other and Java-level accesses.
  6153   OrderAccess::fence();
  6155   // If externally suspended while waiting, re-suspend
  6156   if (jt->handle_special_suspend_equivalent_condition()) {
  6157     jt->java_suspend_self();
  6161 void Parker::unpark() {
  6162   int s, status ;
  6163   status = pthread_mutex_lock(_mutex);
  6164   assert (status == 0, "invariant") ;
  6165   s = _counter;
  6166   _counter = 1;
  6167   if (s < 1) {
  6168     // thread might be parked
  6169     if (_cur_index != -1) {
  6170       // thread is definitely parked
  6171       if (WorkAroundNPTLTimedWaitHang) {
  6172         status = pthread_cond_signal (&_cond[_cur_index]);
  6173         assert (status == 0, "invariant");
  6174         status = pthread_mutex_unlock(_mutex);
  6175         assert (status == 0, "invariant");
  6176       } else {
  6177         // must capture correct index before unlocking
  6178         int index = _cur_index;
  6179         status = pthread_mutex_unlock(_mutex);
  6180         assert (status == 0, "invariant");
  6181         status = pthread_cond_signal (&_cond[index]);
  6182         assert (status == 0, "invariant");
  6184     } else {
  6185       pthread_mutex_unlock(_mutex);
  6186       assert (status == 0, "invariant") ;
  6188   } else {
  6189     pthread_mutex_unlock(_mutex);
  6190     assert (status == 0, "invariant") ;
  6195 extern char** environ;
  6197 // Run the specified command in a separate process. Return its exit value,
  6198 // or -1 on failure (e.g. can't fork a new process).
  6199 // Unlike system(), this function can be called from signal handler. It
  6200 // doesn't block SIGINT et al.
  6201 int os::fork_and_exec(char* cmd) {
  6202   const char * argv[4] = {"sh", "-c", cmd, NULL};
  6204   pid_t pid = fork();
  6206   if (pid < 0) {
  6207     // fork failed
  6208     return -1;
  6210   } else if (pid == 0) {
  6211     // child process
  6213     execve("/bin/sh", (char* const*)argv, environ);
  6215     // execve failed
  6216     _exit(-1);
  6218   } else  {
  6219     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
  6220     // care about the actual exit code, for now.
  6222     int status;
  6224     // Wait for the child process to exit.  This returns immediately if
  6225     // the child has already exited. */
  6226     while (waitpid(pid, &status, 0) < 0) {
  6227         switch (errno) {
  6228         case ECHILD: return 0;
  6229         case EINTR: break;
  6230         default: return -1;
  6234     if (WIFEXITED(status)) {
  6235        // The child exited normally; get its exit code.
  6236        return WEXITSTATUS(status);
  6237     } else if (WIFSIGNALED(status)) {
  6238        // The child exited because of a signal
  6239        // The best value to return is 0x80 + signal number,
  6240        // because that is what all Unix shells do, and because
  6241        // it allows callers to distinguish between process exit and
  6242        // process death by signal.
  6243        return 0x80 + WTERMSIG(status);
  6244     } else {
  6245        // Unknown exit code; pass it through
  6246        return status;
  6251 // is_headless_jre()
  6252 //
  6253 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
  6254 // in order to report if we are running in a headless jre
  6255 //
  6256 // Since JDK8 xawt/libmawt.so was moved into the same directory
  6257 // as libawt.so, and renamed libawt_xawt.so
  6258 //
  6259 bool os::is_headless_jre() {
  6260     struct stat statbuf;
  6261     char buf[MAXPATHLEN];
  6262     char libmawtpath[MAXPATHLEN];
  6263     const char *xawtstr  = "/xawt/libmawt.so";
  6264     const char *new_xawtstr = "/libawt_xawt.so";
  6265     char *p;
  6267     // Get path to libjvm.so
  6268     os::jvm_path(buf, sizeof(buf));
  6270     // Get rid of libjvm.so
  6271     p = strrchr(buf, '/');
  6272     if (p == NULL) return false;
  6273     else *p = '\0';
  6275     // Get rid of client or server
  6276     p = strrchr(buf, '/');
  6277     if (p == NULL) return false;
  6278     else *p = '\0';
  6280     // check xawt/libmawt.so
  6281     strcpy(libmawtpath, buf);
  6282     strcat(libmawtpath, xawtstr);
  6283     if (::stat(libmawtpath, &statbuf) == 0) return false;
  6285     // check libawt_xawt.so
  6286     strcpy(libmawtpath, buf);
  6287     strcat(libmawtpath, new_xawtstr);
  6288     if (::stat(libmawtpath, &statbuf) == 0) return false;
  6290     return true;
  6293 // Get the default path to the core file
  6294 // Returns the length of the string
  6295 int os::get_core_path(char* buffer, size_t bufferSize) {
  6296   const char* p = get_current_directory(buffer, bufferSize);
  6298   if (p == NULL) {
  6299     assert(p != NULL, "failed to get current directory");
  6300     return 0;
  6303   return strlen(buffer);
  6306 /////////////// Unit tests ///////////////
  6308 #ifndef PRODUCT
  6310 #define test_log(...) \
  6311   do {\
  6312     if (VerboseInternalVMTests) { \
  6313       tty->print_cr(__VA_ARGS__); \
  6314       tty->flush(); \
  6315     }\
  6316   } while (false)
  6318 class TestReserveMemorySpecial : AllStatic {
  6319  public:
  6320   static void small_page_write(void* addr, size_t size) {
  6321     size_t page_size = os::vm_page_size();
  6323     char* end = (char*)addr + size;
  6324     for (char* p = (char*)addr; p < end; p += page_size) {
  6325       *p = 1;
  6329   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
  6330     if (!UseHugeTLBFS) {
  6331       return;
  6334     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
  6336     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
  6338     if (addr != NULL) {
  6339       small_page_write(addr, size);
  6341       os::Linux::release_memory_special_huge_tlbfs(addr, size);
  6345   static void test_reserve_memory_special_huge_tlbfs_only() {
  6346     if (!UseHugeTLBFS) {
  6347       return;
  6350     size_t lp = os::large_page_size();
  6352     for (size_t size = lp; size <= lp * 10; size += lp) {
  6353       test_reserve_memory_special_huge_tlbfs_only(size);
  6357   static void test_reserve_memory_special_huge_tlbfs_mixed() {
  6358     size_t lp = os::large_page_size();
  6359     size_t ag = os::vm_allocation_granularity();
  6361     // sizes to test
  6362     const size_t sizes[] = {
  6363       lp, lp + ag, lp + lp / 2, lp * 2,
  6364       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
  6365       lp * 10, lp * 10 + lp / 2
  6366     };
  6367     const int num_sizes = sizeof(sizes) / sizeof(size_t);
  6369     // For each size/alignment combination, we test three scenarios:
  6370     // 1) with req_addr == NULL
  6371     // 2) with a non-null req_addr at which we expect to successfully allocate
  6372     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
  6373     //    expect the allocation to either fail or to ignore req_addr
  6375     // Pre-allocate two areas; they shall be as large as the largest allocation
  6376     //  and aligned to the largest alignment we will be testing.
  6377     const size_t mapping_size = sizes[num_sizes - 1] * 2;
  6378     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
  6379       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
  6380       -1, 0);
  6381     assert(mapping1 != MAP_FAILED, "should work");
  6383     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
  6384       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
  6385       -1, 0);
  6386     assert(mapping2 != MAP_FAILED, "should work");
  6388     // Unmap the first mapping, but leave the second mapping intact: the first
  6389     // mapping will serve as a value for a "good" req_addr (case 2). The second
  6390     // mapping, still intact, as "bad" req_addr (case 3).
  6391     ::munmap(mapping1, mapping_size);
  6393     // Case 1
  6394     test_log("%s, req_addr NULL:", __FUNCTION__);
  6395     test_log("size            align           result");
  6397     for (int i = 0; i < num_sizes; i++) {
  6398       const size_t size = sizes[i];
  6399       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
  6400         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
  6401         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
  6402             size, alignment, p, (p != NULL ? "" : "(failed)"));
  6403         if (p != NULL) {
  6404           assert(is_ptr_aligned(p, alignment), "must be");
  6405           small_page_write(p, size);
  6406           os::Linux::release_memory_special_huge_tlbfs(p, size);
  6411     // Case 2
  6412     test_log("%s, req_addr non-NULL:", __FUNCTION__);
  6413     test_log("size            align           req_addr         result");
  6415     for (int i = 0; i < num_sizes; i++) {
  6416       const size_t size = sizes[i];
  6417       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
  6418         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
  6419         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
  6420         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
  6421             size, alignment, req_addr, p,
  6422             ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
  6423         if (p != NULL) {
  6424           assert(p == req_addr, "must be");
  6425           small_page_write(p, size);
  6426           os::Linux::release_memory_special_huge_tlbfs(p, size);
  6431     // Case 3
  6432     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
  6433     test_log("size            align           req_addr         result");
  6435     for (int i = 0; i < num_sizes; i++) {
  6436       const size_t size = sizes[i];
  6437       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
  6438         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
  6439         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
  6440         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
  6441             size, alignment, req_addr, p,
  6442             ((p != NULL ? "" : "(failed)")));
  6443         // as the area around req_addr contains already existing mappings, the API should always
  6444         // return NULL (as per contract, it cannot return another address)
  6445         assert(p == NULL, "must be");
  6449     ::munmap(mapping2, mapping_size);
  6453   static void test_reserve_memory_special_huge_tlbfs() {
  6454     if (!UseHugeTLBFS) {
  6455       return;
  6458     test_reserve_memory_special_huge_tlbfs_only();
  6459     test_reserve_memory_special_huge_tlbfs_mixed();
  6462   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
  6463     if (!UseSHM) {
  6464       return;
  6467     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
  6469     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
  6471     if (addr != NULL) {
  6472       assert(is_ptr_aligned(addr, alignment), "Check");
  6473       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
  6475       small_page_write(addr, size);
  6477       os::Linux::release_memory_special_shm(addr, size);
  6481   static void test_reserve_memory_special_shm() {
  6482     size_t lp = os::large_page_size();
  6483     size_t ag = os::vm_allocation_granularity();
  6485     for (size_t size = ag; size < lp * 3; size += ag) {
  6486       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
  6487         test_reserve_memory_special_shm(size, alignment);
  6492   static void test() {
  6493     test_reserve_memory_special_huge_tlbfs();
  6494     test_reserve_memory_special_shm();
  6496 };
  6498 void TestReserveMemorySpecial_test() {
  6499   TestReserveMemorySpecial::test();
  6502 #endif

mercurial