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  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "mutex_linux.inline.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "os_linux.inline.hpp"
  41 #include "os_share_linux.hpp"
  42 #include "prims/jniFastGetField.hpp"
  43 #include "prims/jvm.h"
  44 #include "prims/jvm_misc.hpp"
  45 #include "runtime/arguments.hpp"
  46 #include "runtime/atomic.inline.hpp"
  47 #include "runtime/extendedPC.hpp"
  48 #include "runtime/globals.hpp"
  49 #include "runtime/interfaceSupport.hpp"
  50 #include "runtime/init.hpp"
  51 #include "runtime/java.hpp"
  52 #include "runtime/javaCalls.hpp"
  53 #include "runtime/mutexLocker.hpp"
  54 #include "runtime/objectMonitor.hpp"
  55 #include "runtime/orderAccess.inline.hpp"
  56 #include "runtime/osThread.hpp"
  57 #include "runtime/perfMemory.hpp"
  58 #include "runtime/sharedRuntime.hpp"
  59 #include "runtime/statSampler.hpp"
  60 #include "runtime/stubRoutines.hpp"
  61 #include "runtime/thread.inline.hpp"
  62 #include "runtime/threadCritical.hpp"
  63 #include "runtime/timer.hpp"
  64 #include "semaphore_posix.hpp"
  65 #include "services/attachListener.hpp"
  66 #include "services/memTracker.hpp"
  67 #include "services/runtimeService.hpp"
  68 #include "utilities/decoder.hpp"
  69 #include "utilities/defaultStream.hpp"
  70 #include "utilities/events.hpp"
  71 #include "utilities/elfFile.hpp"
  72 #include "utilities/growableArray.hpp"
  73 #include "utilities/macros.hpp"
  74 #include "utilities/vmError.hpp"
  75 
  76 // put OS-includes here
  77 # include <sys/types.h>
  78 # include <sys/mman.h>
  79 # include <sys/stat.h>
  80 # include <sys/select.h>
  81 # include <pthread.h>
  82 # include <signal.h>
  83 # include <errno.h>
  84 # include <dlfcn.h>
  85 # include <stdio.h>
  86 # include <unistd.h>
  87 # include <sys/resource.h>
  88 # include <pthread.h>
  89 # include <sys/stat.h>
  90 # include <sys/time.h>
  91 # include <sys/times.h>
  92 # include <sys/utsname.h>
  93 # include <sys/socket.h>
  94 # include <sys/wait.h>
  95 # include <pwd.h>
  96 # include <poll.h>
  97 # include <semaphore.h>
  98 # include <fcntl.h>
  99 # include <string.h>
 100 # include <syscall.h>
 101 # include <sys/sysinfo.h>
 102 # include <gnu/libc-version.h>
 103 # include <sys/ipc.h>
 104 # include <sys/shm.h>
 105 # include <link.h>
 106 # include <stdint.h>
 107 # include <inttypes.h>
 108 # include <sys/ioctl.h>
 109 
 110 #ifndef _GNU_SOURCE
 111   #define _GNU_SOURCE
 112   #include <sched.h>
 113   #undef _GNU_SOURCE
 114 #else
 115   #include <sched.h>
 116 #endif
 117 
 118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 119 // getrusage() is prepared to handle the associated failure.
 120 #ifndef RUSAGE_THREAD
 121   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 122 #endif
 123 
 124 #define MAX_PATH    (2 * K)
 125 
 126 #define MAX_SECS 100000000
 127 
 128 // for timer info max values which include all bits
 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 130 
 131 #define LARGEPAGES_BIT (1 << 6)
 132 ////////////////////////////////////////////////////////////////////////////////
 133 // global variables
 134 julong os::Linux::_physical_memory = 0;
 135 
 136 address   os::Linux::_initial_thread_stack_bottom = NULL;
 137 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 138 
 139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 142 Mutex* os::Linux::_createThread_lock = NULL;
 143 pthread_t os::Linux::_main_thread;
 144 int os::Linux::_page_size = -1;
 145 const int os::Linux::_vm_default_page_size = (8 * K);
 146 bool os::Linux::_supports_fast_thread_cpu_time = false;
 147 uint32_t os::Linux::_os_version = 0;
 148 const char * os::Linux::_glibc_version = NULL;
 149 const char * os::Linux::_libpthread_version = NULL;
 150 pthread_condattr_t os::Linux::_condattr[1];
 151 
 152 static jlong initial_time_count=0;
 153 
 154 static int clock_tics_per_sec = 100;
 155 
 156 // For diagnostics to print a message once. see run_periodic_checks
 157 static sigset_t check_signal_done;
 158 static bool check_signals = true;
 159 
 160 // Signal number used to suspend/resume a thread
 161 
 162 // do not use any signal number less than SIGSEGV, see 4355769
 163 static int SR_signum = SIGUSR2;
 164 sigset_t SR_sigset;
 165 
 166 // Declarations
 167 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 168 
 169 // utility functions
 170 
 171 static int SR_initialize();
 172 
 173 julong os::available_memory() {
 174   return Linux::available_memory();
 175 }
 176 
 177 julong os::Linux::available_memory() {
 178   // values in struct sysinfo are "unsigned long"
 179   struct sysinfo si;
 180   sysinfo(&si);
 181 
 182   return (julong)si.freeram * si.mem_unit;
 183 }
 184 
 185 julong os::physical_memory() {
 186   return Linux::physical_memory();
 187 }
 188 
 189 // Return true if user is running as root.
 190 
 191 bool os::have_special_privileges() {
 192   static bool init = false;
 193   static bool privileges = false;
 194   if (!init) {
 195     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 196     init = true;
 197   }
 198   return privileges;
 199 }
 200 
 201 
 202 #ifndef SYS_gettid
 203 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 204   #ifdef __ia64__
 205     #define SYS_gettid 1105
 206   #else
 207     #ifdef __i386__
 208       #define SYS_gettid 224
 209     #else
 210       #ifdef __amd64__
 211         #define SYS_gettid 186
 212       #else
 213         #ifdef __sparc__
 214           #define SYS_gettid 143
 215         #else
 216           #error define gettid for the arch
 217         #endif
 218       #endif
 219     #endif
 220   #endif
 221 #endif
 222 
 223 // Cpu architecture string
 224 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 225 
 226 
 227 // pid_t gettid()
 228 //
 229 // Returns the kernel thread id of the currently running thread. Kernel
 230 // thread id is used to access /proc.
 231 pid_t os::Linux::gettid() {
 232   int rslt = syscall(SYS_gettid);
 233   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 234   return (pid_t)rslt;
 235 }
 236 
 237 // Most versions of linux have a bug where the number of processors are
 238 // determined by looking at the /proc file system.  In a chroot environment,
 239 // the system call returns 1.  This causes the VM to act as if it is
 240 // a single processor and elide locking (see is_MP() call).
 241 static bool unsafe_chroot_detected = false;
 242 static const char *unstable_chroot_error = "/proc file system not found.\n"
 243                      "Java may be unstable running multithreaded in a chroot "
 244                      "environment on Linux when /proc filesystem is not mounted.";
 245 
 246 void os::Linux::initialize_system_info() {
 247   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 248   if (processor_count() == 1) {
 249     pid_t pid = os::Linux::gettid();
 250     char fname[32];
 251     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 252     FILE *fp = fopen(fname, "r");
 253     if (fp == NULL) {
 254       unsafe_chroot_detected = true;
 255     } else {
 256       fclose(fp);
 257     }
 258   }
 259   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 260   assert(processor_count() > 0, "linux error");
 261 }
 262 
 263 void os::init_system_properties_values() {
 264   // The next steps are taken in the product version:
 265   //
 266   // Obtain the JAVA_HOME value from the location of libjvm.so.
 267   // This library should be located at:
 268   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 269   //
 270   // If "/jre/lib/" appears at the right place in the path, then we
 271   // assume libjvm.so is installed in a JDK and we use this path.
 272   //
 273   // Otherwise exit with message: "Could not create the Java virtual machine."
 274   //
 275   // The following extra steps are taken in the debugging version:
 276   //
 277   // If "/jre/lib/" does NOT appear at the right place in the path
 278   // instead of exit check for $JAVA_HOME environment variable.
 279   //
 280   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 281   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 282   // it looks like libjvm.so is installed there
 283   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 284   //
 285   // Otherwise exit.
 286   //
 287   // Important note: if the location of libjvm.so changes this
 288   // code needs to be changed accordingly.
 289 
 290   // See ld(1):
 291   //      The linker uses the following search paths to locate required
 292   //      shared libraries:
 293   //        1: ...
 294   //        ...
 295   //        7: The default directories, normally /lib and /usr/lib.
 296 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 297   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 298 #else
 299   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 300 #endif
 301 
 302 // Base path of extensions installed on the system.
 303 #define SYS_EXT_DIR     "/usr/java/packages"
 304 #define EXTENSIONS_DIR  "/lib/ext"
 305 
 306   // Buffer that fits several sprintfs.
 307   // Note that the space for the colon and the trailing null are provided
 308   // by the nulls included by the sizeof operator.
 309   const size_t bufsize =
 310     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 311          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 312   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 313 
 314   // sysclasspath, java_home, dll_dir
 315   {
 316     char *pslash;
 317     os::jvm_path(buf, bufsize);
 318 
 319     // Found the full path to libjvm.so.
 320     // Now cut the path to <java_home>/jre if we can.
 321     pslash = strrchr(buf, '/');
 322     if (pslash != NULL) {
 323       *pslash = '\0';            // Get rid of /libjvm.so.
 324     }
 325     pslash = strrchr(buf, '/');
 326     if (pslash != NULL) {
 327       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 328     }
 329     Arguments::set_dll_dir(buf);
 330 
 331     if (pslash != NULL) {
 332       pslash = strrchr(buf, '/');
 333       if (pslash != NULL) {
 334         *pslash = '\0';          // Get rid of /<arch>.
 335         pslash = strrchr(buf, '/');
 336         if (pslash != NULL) {
 337           *pslash = '\0';        // Get rid of /lib.
 338         }
 339       }
 340     }
 341     Arguments::set_java_home(buf);
 342     set_boot_path('/', ':');
 343   }
 344 
 345   // Where to look for native libraries.
 346   //
 347   // Note: Due to a legacy implementation, most of the library path
 348   // is set in the launcher. This was to accomodate linking restrictions
 349   // on legacy Linux implementations (which are no longer supported).
 350   // Eventually, all the library path setting will be done here.
 351   //
 352   // However, to prevent the proliferation of improperly built native
 353   // libraries, the new path component /usr/java/packages is added here.
 354   // Eventually, all the library path setting will be done here.
 355   {
 356     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 357     // should always exist (until the legacy problem cited above is
 358     // addressed).
 359     const char *v = ::getenv("LD_LIBRARY_PATH");
 360     const char *v_colon = ":";
 361     if (v == NULL) { v = ""; v_colon = ""; }
 362     // That's +1 for the colon and +1 for the trailing '\0'.
 363     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 364                                                      strlen(v) + 1 +
 365                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 366                                                      mtInternal);
 367     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 368     Arguments::set_library_path(ld_library_path);
 369     FREE_C_HEAP_ARRAY(char, ld_library_path);
 370   }
 371 
 372   // Extensions directories.
 373   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 374   Arguments::set_ext_dirs(buf);
 375 
 376   FREE_C_HEAP_ARRAY(char, buf);
 377 
 378 #undef DEFAULT_LIBPATH
 379 #undef SYS_EXT_DIR
 380 #undef EXTENSIONS_DIR
 381 }
 382 
 383 ////////////////////////////////////////////////////////////////////////////////
 384 // breakpoint support
 385 
 386 void os::breakpoint() {
 387   BREAKPOINT;
 388 }
 389 
 390 extern "C" void breakpoint() {
 391   // use debugger to set breakpoint here
 392 }
 393 
 394 ////////////////////////////////////////////////////////////////////////////////
 395 // signal support
 396 
 397 debug_only(static bool signal_sets_initialized = false);
 398 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 399 
 400 bool os::Linux::is_sig_ignored(int sig) {
 401   struct sigaction oact;
 402   sigaction(sig, (struct sigaction*)NULL, &oact);
 403   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 404                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 405   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 406     return true;
 407   } else {
 408     return false;
 409   }
 410 }
 411 
 412 void os::Linux::signal_sets_init() {
 413   // Should also have an assertion stating we are still single-threaded.
 414   assert(!signal_sets_initialized, "Already initialized");
 415   // Fill in signals that are necessarily unblocked for all threads in
 416   // the VM. Currently, we unblock the following signals:
 417   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 418   //                         by -Xrs (=ReduceSignalUsage));
 419   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 420   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 421   // the dispositions or masks wrt these signals.
 422   // Programs embedding the VM that want to use the above signals for their
 423   // own purposes must, at this time, use the "-Xrs" option to prevent
 424   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 425   // (See bug 4345157, and other related bugs).
 426   // In reality, though, unblocking these signals is really a nop, since
 427   // these signals are not blocked by default.
 428   sigemptyset(&unblocked_sigs);
 429   sigemptyset(&allowdebug_blocked_sigs);
 430   sigaddset(&unblocked_sigs, SIGILL);
 431   sigaddset(&unblocked_sigs, SIGSEGV);
 432   sigaddset(&unblocked_sigs, SIGBUS);
 433   sigaddset(&unblocked_sigs, SIGFPE);
 434 #if defined(PPC64)
 435   sigaddset(&unblocked_sigs, SIGTRAP);
 436 #endif
 437   sigaddset(&unblocked_sigs, SR_signum);
 438 
 439   if (!ReduceSignalUsage) {
 440     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 441       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 442       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 443     }
 444     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 445       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 446       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 447     }
 448     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 449       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 450       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 451     }
 452   }
 453   // Fill in signals that are blocked by all but the VM thread.
 454   sigemptyset(&vm_sigs);
 455   if (!ReduceSignalUsage) {
 456     sigaddset(&vm_sigs, BREAK_SIGNAL);
 457   }
 458   debug_only(signal_sets_initialized = true);
 459 
 460 }
 461 
 462 // These are signals that are unblocked while a thread is running Java.
 463 // (For some reason, they get blocked by default.)
 464 sigset_t* os::Linux::unblocked_signals() {
 465   assert(signal_sets_initialized, "Not initialized");
 466   return &unblocked_sigs;
 467 }
 468 
 469 // These are the signals that are blocked while a (non-VM) thread is
 470 // running Java. Only the VM thread handles these signals.
 471 sigset_t* os::Linux::vm_signals() {
 472   assert(signal_sets_initialized, "Not initialized");
 473   return &vm_sigs;
 474 }
 475 
 476 // These are signals that are blocked during cond_wait to allow debugger in
 477 sigset_t* os::Linux::allowdebug_blocked_signals() {
 478   assert(signal_sets_initialized, "Not initialized");
 479   return &allowdebug_blocked_sigs;
 480 }
 481 
 482 void os::Linux::hotspot_sigmask(Thread* thread) {
 483 
 484   //Save caller's signal mask before setting VM signal mask
 485   sigset_t caller_sigmask;
 486   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 487 
 488   OSThread* osthread = thread->osthread();
 489   osthread->set_caller_sigmask(caller_sigmask);
 490 
 491   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 492 
 493   if (!ReduceSignalUsage) {
 494     if (thread->is_VM_thread()) {
 495       // Only the VM thread handles BREAK_SIGNAL ...
 496       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 497     } else {
 498       // ... all other threads block BREAK_SIGNAL
 499       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 500     }
 501   }
 502 }
 503 
 504 //////////////////////////////////////////////////////////////////////////////
 505 // detecting pthread library
 506 
 507 void os::Linux::libpthread_init() {
 508   // Save glibc and pthread version strings.
 509 #if !defined(_CS_GNU_LIBC_VERSION) || \
 510     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 511   #error "glibc too old (< 2.3.2)"
 512 #endif
 513 
 514   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 515   assert(n > 0, "cannot retrieve glibc version");
 516   char *str = (char *)malloc(n, mtInternal);
 517   confstr(_CS_GNU_LIBC_VERSION, str, n);
 518   os::Linux::set_glibc_version(str);
 519 
 520   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 521   assert(n > 0, "cannot retrieve pthread version");
 522   str = (char *)malloc(n, mtInternal);
 523   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 524   os::Linux::set_libpthread_version(str);
 525 }
 526 
 527 /////////////////////////////////////////////////////////////////////////////
 528 // thread stack expansion
 529 
 530 // os::Linux::manually_expand_stack() takes care of expanding the thread
 531 // stack. Note that this is normally not needed: pthread stacks allocate
 532 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 533 // committed. Therefore it is not necessary to expand the stack manually.
 534 //
 535 // Manually expanding the stack was historically needed on LinuxThreads
 536 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 537 // it is kept to deal with very rare corner cases:
 538 //
 539 // For one, user may run the VM on an own implementation of threads
 540 // whose stacks are - like the old LinuxThreads - implemented using
 541 // mmap(MAP_GROWSDOWN).
 542 //
 543 // Also, this coding may be needed if the VM is running on the primordial
 544 // thread. Normally we avoid running on the primordial thread; however,
 545 // user may still invoke the VM on the primordial thread.
 546 //
 547 // The following historical comment describes the details about running
 548 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 549 
 550 
 551 // Force Linux kernel to expand current thread stack. If "bottom" is close
 552 // to the stack guard, caller should block all signals.
 553 //
 554 // MAP_GROWSDOWN:
 555 //   A special mmap() flag that is used to implement thread stacks. It tells
 556 //   kernel that the memory region should extend downwards when needed. This
 557 //   allows early versions of LinuxThreads to only mmap the first few pages
 558 //   when creating a new thread. Linux kernel will automatically expand thread
 559 //   stack as needed (on page faults).
 560 //
 561 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 562 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 563 //   region, it's hard to tell if the fault is due to a legitimate stack
 564 //   access or because of reading/writing non-exist memory (e.g. buffer
 565 //   overrun). As a rule, if the fault happens below current stack pointer,
 566 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 567 //   application (see Linux kernel fault.c).
 568 //
 569 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 570 //   stack overflow detection.
 571 //
 572 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 573 //   not use MAP_GROWSDOWN.
 574 //
 575 // To get around the problem and allow stack banging on Linux, we need to
 576 // manually expand thread stack after receiving the SIGSEGV.
 577 //
 578 // There are two ways to expand thread stack to address "bottom", we used
 579 // both of them in JVM before 1.5:
 580 //   1. adjust stack pointer first so that it is below "bottom", and then
 581 //      touch "bottom"
 582 //   2. mmap() the page in question
 583 //
 584 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 585 // if current sp is already near the lower end of page 101, and we need to
 586 // call mmap() to map page 100, it is possible that part of the mmap() frame
 587 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 588 // That will destroy the mmap() frame and cause VM to crash.
 589 //
 590 // The following code works by adjusting sp first, then accessing the "bottom"
 591 // page to force a page fault. Linux kernel will then automatically expand the
 592 // stack mapping.
 593 //
 594 // _expand_stack_to() assumes its frame size is less than page size, which
 595 // should always be true if the function is not inlined.
 596 
 597 static void NOINLINE _expand_stack_to(address bottom) {
 598   address sp;
 599   size_t size;
 600   volatile char *p;
 601 
 602   // Adjust bottom to point to the largest address within the same page, it
 603   // gives us a one-page buffer if alloca() allocates slightly more memory.
 604   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 605   bottom += os::Linux::page_size() - 1;
 606 
 607   // sp might be slightly above current stack pointer; if that's the case, we
 608   // will alloca() a little more space than necessary, which is OK. Don't use
 609   // os::current_stack_pointer(), as its result can be slightly below current
 610   // stack pointer, causing us to not alloca enough to reach "bottom".
 611   sp = (address)&sp;
 612 
 613   if (sp > bottom) {
 614     size = sp - bottom;
 615     p = (volatile char *)alloca(size);
 616     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 617     p[0] = '\0';
 618   }
 619 }
 620 
 621 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 622   assert(t!=NULL, "just checking");
 623   assert(t->osthread()->expanding_stack(), "expand should be set");
 624   assert(t->stack_base() != NULL, "stack_base was not initialized");
 625 
 626   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 627     sigset_t mask_all, old_sigset;
 628     sigfillset(&mask_all);
 629     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 630     _expand_stack_to(addr);
 631     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 632     return true;
 633   }
 634   return false;
 635 }
 636 
 637 //////////////////////////////////////////////////////////////////////////////
 638 // create new thread
 639 
 640 // Thread start routine for all newly created threads
 641 static void *thread_native_entry(Thread *thread) {
 642   // Try to randomize the cache line index of hot stack frames.
 643   // This helps when threads of the same stack traces evict each other's
 644   // cache lines. The threads can be either from the same JVM instance, or
 645   // from different JVM instances. The benefit is especially true for
 646   // processors with hyperthreading technology.
 647   static int counter = 0;
 648   int pid = os::current_process_id();
 649   alloca(((pid ^ counter++) & 7) * 128);
 650 
 651   thread->initialize_thread_current();
 652 
 653   OSThread* osthread = thread->osthread();
 654   Monitor* sync = osthread->startThread_lock();
 655 
 656   osthread->set_thread_id(os::current_thread_id());
 657 
 658   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 659     os::current_thread_id(), (uintx) pthread_self());
 660 
 661   if (UseNUMA) {
 662     int lgrp_id = os::numa_get_group_id();
 663     if (lgrp_id != -1) {
 664       thread->set_lgrp_id(lgrp_id);
 665     }
 666   }
 667   // initialize signal mask for this thread
 668   os::Linux::hotspot_sigmask(thread);
 669 
 670   // initialize floating point control register
 671   os::Linux::init_thread_fpu_state();
 672 
 673   // handshaking with parent thread
 674   {
 675     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 676 
 677     // notify parent thread
 678     osthread->set_state(INITIALIZED);
 679     sync->notify_all();
 680 
 681     // wait until os::start_thread()
 682     while (osthread->get_state() == INITIALIZED) {
 683       sync->wait(Mutex::_no_safepoint_check_flag);
 684     }
 685   }
 686 
 687   // call one more level start routine
 688   thread->run();
 689 
 690   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 691     os::current_thread_id(), (uintx) pthread_self());
 692 
 693   // If a thread has not deleted itself ("delete this") as part of its
 694   // termination sequence, we have to ensure thread-local-storage is
 695   // cleared before we actually terminate. No threads should ever be
 696   // deleted asynchronously with respect to their termination.
 697   if (Thread::current_or_null_safe() != NULL) {
 698     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 699     thread->clear_thread_current();
 700   }
 701 
 702   return 0;
 703 }
 704 
 705 bool os::create_thread(Thread* thread, ThreadType thr_type,
 706                        size_t stack_size) {
 707   assert(thread->osthread() == NULL, "caller responsible");
 708 
 709   // Allocate the OSThread object
 710   OSThread* osthread = new OSThread(NULL, NULL);
 711   if (osthread == NULL) {
 712     return false;
 713   }
 714 
 715   // set the correct thread state
 716   osthread->set_thread_type(thr_type);
 717 
 718   // Initial state is ALLOCATED but not INITIALIZED
 719   osthread->set_state(ALLOCATED);
 720 
 721   thread->set_osthread(osthread);
 722 
 723   // init thread attributes
 724   pthread_attr_t attr;
 725   pthread_attr_init(&attr);
 726   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 727 
 728   // stack size
 729   // calculate stack size if it's not specified by caller
 730   if (stack_size == 0) {
 731     stack_size = os::Linux::default_stack_size(thr_type);
 732 
 733     switch (thr_type) {
 734     case os::java_thread:
 735       // Java threads use ThreadStackSize which default value can be
 736       // changed with the flag -Xss
 737       assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 738       stack_size = JavaThread::stack_size_at_create();
 739       break;
 740     case os::compiler_thread:
 741       if (CompilerThreadStackSize > 0) {
 742         stack_size = (size_t)(CompilerThreadStackSize * K);
 743         break;
 744       } // else fall through:
 745         // use VMThreadStackSize if CompilerThreadStackSize is not defined
 746     case os::vm_thread:
 747     case os::pgc_thread:
 748     case os::cgc_thread:
 749     case os::watcher_thread:
 750       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 751       break;
 752     }
 753   }
 754 
 755   stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 756   pthread_attr_setstacksize(&attr, stack_size);
 757 
 758   // glibc guard page
 759   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 760 
 761   ThreadState state;
 762 
 763   {
 764     pthread_t tid;
 765     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 766 
 767     char buf[64];
 768     if (ret == 0) {
 769       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 770         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 771     } else {
 772       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 773         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 774     }
 775 
 776     pthread_attr_destroy(&attr);
 777 
 778     if (ret != 0) {
 779       // Need to clean up stuff we've allocated so far
 780       thread->set_osthread(NULL);
 781       delete osthread;
 782       return false;
 783     }
 784 
 785     // Store pthread info into the OSThread
 786     osthread->set_pthread_id(tid);
 787 
 788     // Wait until child thread is either initialized or aborted
 789     {
 790       Monitor* sync_with_child = osthread->startThread_lock();
 791       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 792       while ((state = osthread->get_state()) == ALLOCATED) {
 793         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 794       }
 795     }
 796   }
 797 
 798   // Aborted due to thread limit being reached
 799   if (state == ZOMBIE) {
 800     thread->set_osthread(NULL);
 801     delete osthread;
 802     return false;
 803   }
 804 
 805   // The thread is returned suspended (in state INITIALIZED),
 806   // and is started higher up in the call chain
 807   assert(state == INITIALIZED, "race condition");
 808   return true;
 809 }
 810 
 811 /////////////////////////////////////////////////////////////////////////////
 812 // attach existing thread
 813 
 814 // bootstrap the main thread
 815 bool os::create_main_thread(JavaThread* thread) {
 816   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 817   return create_attached_thread(thread);
 818 }
 819 
 820 bool os::create_attached_thread(JavaThread* thread) {
 821 #ifdef ASSERT
 822   thread->verify_not_published();
 823 #endif
 824 
 825   // Allocate the OSThread object
 826   OSThread* osthread = new OSThread(NULL, NULL);
 827 
 828   if (osthread == NULL) {
 829     return false;
 830   }
 831 
 832   // Store pthread info into the OSThread
 833   osthread->set_thread_id(os::Linux::gettid());
 834   osthread->set_pthread_id(::pthread_self());
 835 
 836   // initialize floating point control register
 837   os::Linux::init_thread_fpu_state();
 838 
 839   // Initial thread state is RUNNABLE
 840   osthread->set_state(RUNNABLE);
 841 
 842   thread->set_osthread(osthread);
 843 
 844   if (UseNUMA) {
 845     int lgrp_id = os::numa_get_group_id();
 846     if (lgrp_id != -1) {
 847       thread->set_lgrp_id(lgrp_id);
 848     }
 849   }
 850 
 851   if (os::Linux::is_initial_thread()) {
 852     // If current thread is initial thread, its stack is mapped on demand,
 853     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 854     // the entire stack region to avoid SEGV in stack banging.
 855     // It is also useful to get around the heap-stack-gap problem on SuSE
 856     // kernel (see 4821821 for details). We first expand stack to the top
 857     // of yellow zone, then enable stack yellow zone (order is significant,
 858     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 859     // is no gap between the last two virtual memory regions.
 860 
 861     JavaThread *jt = (JavaThread *)thread;
 862     address addr = jt->stack_reserved_zone_base();
 863     assert(addr != NULL, "initialization problem?");
 864     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 865 
 866     osthread->set_expanding_stack();
 867     os::Linux::manually_expand_stack(jt, addr);
 868     osthread->clear_expanding_stack();
 869   }
 870 
 871   // initialize signal mask for this thread
 872   // and save the caller's signal mask
 873   os::Linux::hotspot_sigmask(thread);
 874 
 875   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 876     os::current_thread_id(), (uintx) pthread_self());
 877 
 878   return true;
 879 }
 880 
 881 void os::pd_start_thread(Thread* thread) {
 882   OSThread * osthread = thread->osthread();
 883   assert(osthread->get_state() != INITIALIZED, "just checking");
 884   Monitor* sync_with_child = osthread->startThread_lock();
 885   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 886   sync_with_child->notify();
 887 }
 888 
 889 // Free Linux resources related to the OSThread
 890 void os::free_thread(OSThread* osthread) {
 891   assert(osthread != NULL, "osthread not set");
 892 
 893   // We are told to free resources of the argument thread,
 894   // but we can only really operate on the current thread.
 895   assert(Thread::current()->osthread() == osthread,
 896          "os::free_thread but not current thread");
 897 
 898 #ifdef ASSERT
 899   sigset_t current;
 900   sigemptyset(&current);
 901   pthread_sigmask(SIG_SETMASK, NULL, &current);
 902   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 903 #endif
 904 
 905   // Restore caller's signal mask
 906   sigset_t sigmask = osthread->caller_sigmask();
 907   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 908 
 909   delete osthread;
 910 }
 911 
 912 //////////////////////////////////////////////////////////////////////////////
 913 // initial thread
 914 
 915 // Check if current thread is the initial thread, similar to Solaris thr_main.
 916 bool os::Linux::is_initial_thread(void) {
 917   char dummy;
 918   // If called before init complete, thread stack bottom will be null.
 919   // Can be called if fatal error occurs before initialization.
 920   if (initial_thread_stack_bottom() == NULL) return false;
 921   assert(initial_thread_stack_bottom() != NULL &&
 922          initial_thread_stack_size()   != 0,
 923          "os::init did not locate initial thread's stack region");
 924   if ((address)&dummy >= initial_thread_stack_bottom() &&
 925       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 926     return true;
 927   } else {
 928     return false;
 929   }
 930 }
 931 
 932 // Find the virtual memory area that contains addr
 933 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 934   FILE *fp = fopen("/proc/self/maps", "r");
 935   if (fp) {
 936     address low, high;
 937     while (!feof(fp)) {
 938       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 939         if (low <= addr && addr < high) {
 940           if (vma_low)  *vma_low  = low;
 941           if (vma_high) *vma_high = high;
 942           fclose(fp);
 943           return true;
 944         }
 945       }
 946       for (;;) {
 947         int ch = fgetc(fp);
 948         if (ch == EOF || ch == (int)'\n') break;
 949       }
 950     }
 951     fclose(fp);
 952   }
 953   return false;
 954 }
 955 
 956 // Locate initial thread stack. This special handling of initial thread stack
 957 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 958 // bogus value for initial thread.
 959 void os::Linux::capture_initial_stack(size_t max_size) {
 960   // stack size is the easy part, get it from RLIMIT_STACK
 961   size_t stack_size;
 962   struct rlimit rlim;
 963   getrlimit(RLIMIT_STACK, &rlim);
 964   stack_size = rlim.rlim_cur;
 965 
 966   // 6308388: a bug in ld.so will relocate its own .data section to the
 967   //   lower end of primordial stack; reduce ulimit -s value a little bit
 968   //   so we won't install guard page on ld.so's data section.
 969   stack_size -= 2 * page_size();
 970 
 971   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
 972   //   7.1, in both cases we will get 2G in return value.
 973   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
 974   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
 975   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
 976   //   in case other parts in glibc still assumes 2M max stack size.
 977   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
 978   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
 979   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
 980     stack_size = 2 * K * K IA64_ONLY(*2);
 981   }
 982   // Try to figure out where the stack base (top) is. This is harder.
 983   //
 984   // When an application is started, glibc saves the initial stack pointer in
 985   // a global variable "__libc_stack_end", which is then used by system
 986   // libraries. __libc_stack_end should be pretty close to stack top. The
 987   // variable is available since the very early days. However, because it is
 988   // a private interface, it could disappear in the future.
 989   //
 990   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 991   // to __libc_stack_end, it is very close to stack top, but isn't the real
 992   // stack top. Note that /proc may not exist if VM is running as a chroot
 993   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 994   // /proc/<pid>/stat could change in the future (though unlikely).
 995   //
 996   // We try __libc_stack_end first. If that doesn't work, look for
 997   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
 998   // as a hint, which should work well in most cases.
 999 
1000   uintptr_t stack_start;
1001 
1002   // try __libc_stack_end first
1003   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1004   if (p && *p) {
1005     stack_start = *p;
1006   } else {
1007     // see if we can get the start_stack field from /proc/self/stat
1008     FILE *fp;
1009     int pid;
1010     char state;
1011     int ppid;
1012     int pgrp;
1013     int session;
1014     int nr;
1015     int tpgrp;
1016     unsigned long flags;
1017     unsigned long minflt;
1018     unsigned long cminflt;
1019     unsigned long majflt;
1020     unsigned long cmajflt;
1021     unsigned long utime;
1022     unsigned long stime;
1023     long cutime;
1024     long cstime;
1025     long prio;
1026     long nice;
1027     long junk;
1028     long it_real;
1029     uintptr_t start;
1030     uintptr_t vsize;
1031     intptr_t rss;
1032     uintptr_t rsslim;
1033     uintptr_t scodes;
1034     uintptr_t ecode;
1035     int i;
1036 
1037     // Figure what the primordial thread stack base is. Code is inspired
1038     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1039     // followed by command name surrounded by parentheses, state, etc.
1040     char stat[2048];
1041     int statlen;
1042 
1043     fp = fopen("/proc/self/stat", "r");
1044     if (fp) {
1045       statlen = fread(stat, 1, 2047, fp);
1046       stat[statlen] = '\0';
1047       fclose(fp);
1048 
1049       // Skip pid and the command string. Note that we could be dealing with
1050       // weird command names, e.g. user could decide to rename java launcher
1051       // to "java 1.4.2 :)", then the stat file would look like
1052       //                1234 (java 1.4.2 :)) R ... ...
1053       // We don't really need to know the command string, just find the last
1054       // occurrence of ")" and then start parsing from there. See bug 4726580.
1055       char * s = strrchr(stat, ')');
1056 
1057       i = 0;
1058       if (s) {
1059         // Skip blank chars
1060         do { s++; } while (s && isspace(*s));
1061 
1062 #define _UFM UINTX_FORMAT
1063 #define _DFM INTX_FORMAT
1064 
1065         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1066         //              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
1067         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,
1068                    &state,          // 3  %c
1069                    &ppid,           // 4  %d
1070                    &pgrp,           // 5  %d
1071                    &session,        // 6  %d
1072                    &nr,             // 7  %d
1073                    &tpgrp,          // 8  %d
1074                    &flags,          // 9  %lu
1075                    &minflt,         // 10 %lu
1076                    &cminflt,        // 11 %lu
1077                    &majflt,         // 12 %lu
1078                    &cmajflt,        // 13 %lu
1079                    &utime,          // 14 %lu
1080                    &stime,          // 15 %lu
1081                    &cutime,         // 16 %ld
1082                    &cstime,         // 17 %ld
1083                    &prio,           // 18 %ld
1084                    &nice,           // 19 %ld
1085                    &junk,           // 20 %ld
1086                    &it_real,        // 21 %ld
1087                    &start,          // 22 UINTX_FORMAT
1088                    &vsize,          // 23 UINTX_FORMAT
1089                    &rss,            // 24 INTX_FORMAT
1090                    &rsslim,         // 25 UINTX_FORMAT
1091                    &scodes,         // 26 UINTX_FORMAT
1092                    &ecode,          // 27 UINTX_FORMAT
1093                    &stack_start);   // 28 UINTX_FORMAT
1094       }
1095 
1096 #undef _UFM
1097 #undef _DFM
1098 
1099       if (i != 28 - 2) {
1100         assert(false, "Bad conversion from /proc/self/stat");
1101         // product mode - assume we are the initial thread, good luck in the
1102         // embedded case.
1103         warning("Can't detect initial thread stack location - bad conversion");
1104         stack_start = (uintptr_t) &rlim;
1105       }
1106     } else {
1107       // For some reason we can't open /proc/self/stat (for example, running on
1108       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1109       // most cases, so don't abort:
1110       warning("Can't detect initial thread stack location - no /proc/self/stat");
1111       stack_start = (uintptr_t) &rlim;
1112     }
1113   }
1114 
1115   // Now we have a pointer (stack_start) very close to the stack top, the
1116   // next thing to do is to figure out the exact location of stack top. We
1117   // can find out the virtual memory area that contains stack_start by
1118   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1119   // and its upper limit is the real stack top. (again, this would fail if
1120   // running inside chroot, because /proc may not exist.)
1121 
1122   uintptr_t stack_top;
1123   address low, high;
1124   if (find_vma((address)stack_start, &low, &high)) {
1125     // success, "high" is the true stack top. (ignore "low", because initial
1126     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1127     stack_top = (uintptr_t)high;
1128   } else {
1129     // failed, likely because /proc/self/maps does not exist
1130     warning("Can't detect initial thread stack location - find_vma failed");
1131     // best effort: stack_start is normally within a few pages below the real
1132     // stack top, use it as stack top, and reduce stack size so we won't put
1133     // guard page outside stack.
1134     stack_top = stack_start;
1135     stack_size -= 16 * page_size();
1136   }
1137 
1138   // stack_top could be partially down the page so align it
1139   stack_top = align_size_up(stack_top, page_size());
1140 
1141   if (max_size && stack_size > max_size) {
1142     _initial_thread_stack_size = max_size;
1143   } else {
1144     _initial_thread_stack_size = stack_size;
1145   }
1146 
1147   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1148   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1149 }
1150 
1151 ////////////////////////////////////////////////////////////////////////////////
1152 // time support
1153 
1154 // Time since start-up in seconds to a fine granularity.
1155 // Used by VMSelfDestructTimer and the MemProfiler.
1156 double os::elapsedTime() {
1157 
1158   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1159 }
1160 
1161 jlong os::elapsed_counter() {
1162   return javaTimeNanos() - initial_time_count;
1163 }
1164 
1165 jlong os::elapsed_frequency() {
1166   return NANOSECS_PER_SEC; // nanosecond resolution
1167 }
1168 
1169 bool os::supports_vtime() { return true; }
1170 bool os::enable_vtime()   { return false; }
1171 bool os::vtime_enabled()  { return false; }
1172 
1173 double os::elapsedVTime() {
1174   struct rusage usage;
1175   int retval = getrusage(RUSAGE_THREAD, &usage);
1176   if (retval == 0) {
1177     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);
1178   } else {
1179     // better than nothing, but not much
1180     return elapsedTime();
1181   }
1182 }
1183 
1184 jlong os::javaTimeMillis() {
1185   timeval time;
1186   int status = gettimeofday(&time, NULL);
1187   assert(status != -1, "linux error");
1188   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1189 }
1190 
1191 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1192   timeval time;
1193   int status = gettimeofday(&time, NULL);
1194   assert(status != -1, "linux error");
1195   seconds = jlong(time.tv_sec);
1196   nanos = jlong(time.tv_usec) * 1000;
1197 }
1198 
1199 
1200 #ifndef CLOCK_MONOTONIC
1201   #define CLOCK_MONOTONIC (1)
1202 #endif
1203 
1204 void os::Linux::clock_init() {
1205   // we do dlopen's in this particular order due to bug in linux
1206   // dynamical loader (see 6348968) leading to crash on exit
1207   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1208   if (handle == NULL) {
1209     handle = dlopen("librt.so", RTLD_LAZY);
1210   }
1211 
1212   if (handle) {
1213     int (*clock_getres_func)(clockid_t, struct timespec*) =
1214            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1215     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1216            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1217     if (clock_getres_func && clock_gettime_func) {
1218       // See if monotonic clock is supported by the kernel. Note that some
1219       // early implementations simply return kernel jiffies (updated every
1220       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1221       // for nano time (though the monotonic property is still nice to have).
1222       // It's fixed in newer kernels, however clock_getres() still returns
1223       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1224       // resolution for now. Hopefully as people move to new kernels, this
1225       // won't be a problem.
1226       struct timespec res;
1227       struct timespec tp;
1228       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1229           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1230         // yes, monotonic clock is supported
1231         _clock_gettime = clock_gettime_func;
1232         return;
1233       } else {
1234         // close librt if there is no monotonic clock
1235         dlclose(handle);
1236       }
1237     }
1238   }
1239   warning("No monotonic clock was available - timed services may " \
1240           "be adversely affected if the time-of-day clock changes");
1241 }
1242 
1243 #ifndef SYS_clock_getres
1244   #if defined(IA32) || defined(AMD64)
1245     #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1246     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1247   #else
1248     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1249     #define sys_clock_getres(x,y)  -1
1250   #endif
1251 #else
1252   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1253 #endif
1254 
1255 void os::Linux::fast_thread_clock_init() {
1256   if (!UseLinuxPosixThreadCPUClocks) {
1257     return;
1258   }
1259   clockid_t clockid;
1260   struct timespec tp;
1261   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1262       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1263 
1264   // Switch to using fast clocks for thread cpu time if
1265   // the sys_clock_getres() returns 0 error code.
1266   // Note, that some kernels may support the current thread
1267   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1268   // returned by the pthread_getcpuclockid().
1269   // If the fast Posix clocks are supported then the sys_clock_getres()
1270   // must return at least tp.tv_sec == 0 which means a resolution
1271   // better than 1 sec. This is extra check for reliability.
1272 
1273   if (pthread_getcpuclockid_func &&
1274       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1275       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1276     _supports_fast_thread_cpu_time = true;
1277     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1278   }
1279 }
1280 
1281 jlong os::javaTimeNanos() {
1282   if (os::supports_monotonic_clock()) {
1283     struct timespec tp;
1284     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1285     assert(status == 0, "gettime error");
1286     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1287     return result;
1288   } else {
1289     timeval time;
1290     int status = gettimeofday(&time, NULL);
1291     assert(status != -1, "linux error");
1292     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1293     return 1000 * usecs;
1294   }
1295 }
1296 
1297 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1298   if (os::supports_monotonic_clock()) {
1299     info_ptr->max_value = ALL_64_BITS;
1300 
1301     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1302     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1303     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1304   } else {
1305     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1306     info_ptr->max_value = ALL_64_BITS;
1307 
1308     // gettimeofday is a real time clock so it skips
1309     info_ptr->may_skip_backward = true;
1310     info_ptr->may_skip_forward = true;
1311   }
1312 
1313   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1314 }
1315 
1316 // Return the real, user, and system times in seconds from an
1317 // arbitrary fixed point in the past.
1318 bool os::getTimesSecs(double* process_real_time,
1319                       double* process_user_time,
1320                       double* process_system_time) {
1321   struct tms ticks;
1322   clock_t real_ticks = times(&ticks);
1323 
1324   if (real_ticks == (clock_t) (-1)) {
1325     return false;
1326   } else {
1327     double ticks_per_second = (double) clock_tics_per_sec;
1328     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1329     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1330     *process_real_time = ((double) real_ticks) / ticks_per_second;
1331 
1332     return true;
1333   }
1334 }
1335 
1336 
1337 char * os::local_time_string(char *buf, size_t buflen) {
1338   struct tm t;
1339   time_t long_time;
1340   time(&long_time);
1341   localtime_r(&long_time, &t);
1342   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1343                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1344                t.tm_hour, t.tm_min, t.tm_sec);
1345   return buf;
1346 }
1347 
1348 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1349   return localtime_r(clock, res);
1350 }
1351 
1352 ////////////////////////////////////////////////////////////////////////////////
1353 // runtime exit support
1354 
1355 // Note: os::shutdown() might be called very early during initialization, or
1356 // called from signal handler. Before adding something to os::shutdown(), make
1357 // sure it is async-safe and can handle partially initialized VM.
1358 void os::shutdown() {
1359 
1360   // allow PerfMemory to attempt cleanup of any persistent resources
1361   perfMemory_exit();
1362 
1363   // needs to remove object in file system
1364   AttachListener::abort();
1365 
1366   // flush buffered output, finish log files
1367   ostream_abort();
1368 
1369   // Check for abort hook
1370   abort_hook_t abort_hook = Arguments::abort_hook();
1371   if (abort_hook != NULL) {
1372     abort_hook();
1373   }
1374 
1375 }
1376 
1377 // Note: os::abort() might be called very early during initialization, or
1378 // called from signal handler. Before adding something to os::abort(), make
1379 // sure it is async-safe and can handle partially initialized VM.
1380 void os::abort(bool dump_core, void* siginfo, const void* context) {
1381   os::shutdown();
1382   if (dump_core) {
1383 #ifndef PRODUCT
1384     fdStream out(defaultStream::output_fd());
1385     out.print_raw("Current thread is ");
1386     char buf[16];
1387     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1388     out.print_raw_cr(buf);
1389     out.print_raw_cr("Dumping core ...");
1390 #endif
1391     ::abort(); // dump core
1392   }
1393 
1394   ::exit(1);
1395 }
1396 
1397 // Die immediately, no exit hook, no abort hook, no cleanup.
1398 void os::die() {
1399   ::abort();
1400 }
1401 
1402 
1403 // This method is a copy of JDK's sysGetLastErrorString
1404 // from src/solaris/hpi/src/system_md.c
1405 
1406 size_t os::lasterror(char *buf, size_t len) {
1407   if (errno == 0)  return 0;
1408 
1409   const char *s = os::strerror(errno);
1410   size_t n = ::strlen(s);
1411   if (n >= len) {
1412     n = len - 1;
1413   }
1414   ::strncpy(buf, s, n);
1415   buf[n] = '\0';
1416   return n;
1417 }
1418 
1419 // thread_id is kernel thread id (similar to Solaris LWP id)
1420 intx os::current_thread_id() { return os::Linux::gettid(); }
1421 int os::current_process_id() {
1422   return ::getpid();
1423 }
1424 
1425 // DLL functions
1426 
1427 const char* os::dll_file_extension() { return ".so"; }
1428 
1429 // This must be hard coded because it's the system's temporary
1430 // directory not the java application's temp directory, ala java.io.tmpdir.
1431 const char* os::get_temp_directory() { return "/tmp"; }
1432 
1433 static bool file_exists(const char* filename) {
1434   struct stat statbuf;
1435   if (filename == NULL || strlen(filename) == 0) {
1436     return false;
1437   }
1438   return os::stat(filename, &statbuf) == 0;
1439 }
1440 
1441 bool os::dll_build_name(char* buffer, size_t buflen,
1442                         const char* pname, const char* fname) {
1443   bool retval = false;
1444   // Copied from libhpi
1445   const size_t pnamelen = pname ? strlen(pname) : 0;
1446 
1447   // Return error on buffer overflow.
1448   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1449     return retval;
1450   }
1451 
1452   if (pnamelen == 0) {
1453     snprintf(buffer, buflen, "lib%s.so", fname);
1454     retval = true;
1455   } else if (strchr(pname, *os::path_separator()) != NULL) {
1456     int n;
1457     char** pelements = split_path(pname, &n);
1458     if (pelements == NULL) {
1459       return false;
1460     }
1461     for (int i = 0; i < n; i++) {
1462       // Really shouldn't be NULL, but check can't hurt
1463       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1464         continue; // skip the empty path values
1465       }
1466       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1467       if (file_exists(buffer)) {
1468         retval = true;
1469         break;
1470       }
1471     }
1472     // release the storage
1473     for (int i = 0; i < n; i++) {
1474       if (pelements[i] != NULL) {
1475         FREE_C_HEAP_ARRAY(char, pelements[i]);
1476       }
1477     }
1478     if (pelements != NULL) {
1479       FREE_C_HEAP_ARRAY(char*, pelements);
1480     }
1481   } else {
1482     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1483     retval = true;
1484   }
1485   return retval;
1486 }
1487 
1488 // check if addr is inside libjvm.so
1489 bool os::address_is_in_vm(address addr) {
1490   static address libjvm_base_addr;
1491   Dl_info dlinfo;
1492 
1493   if (libjvm_base_addr == NULL) {
1494     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1495       libjvm_base_addr = (address)dlinfo.dli_fbase;
1496     }
1497     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1498   }
1499 
1500   if (dladdr((void *)addr, &dlinfo) != 0) {
1501     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1502   }
1503 
1504   return false;
1505 }
1506 
1507 bool os::dll_address_to_function_name(address addr, char *buf,
1508                                       int buflen, int *offset,
1509                                       bool demangle) {
1510   // buf is not optional, but offset is optional
1511   assert(buf != NULL, "sanity check");
1512 
1513   Dl_info dlinfo;
1514 
1515   if (dladdr((void*)addr, &dlinfo) != 0) {
1516     // see if we have a matching symbol
1517     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1518       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1519         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1520       }
1521       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1522       return true;
1523     }
1524     // no matching symbol so try for just file info
1525     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1526       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1527                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1528         return true;
1529       }
1530     }
1531   }
1532 
1533   buf[0] = '\0';
1534   if (offset != NULL) *offset = -1;
1535   return false;
1536 }
1537 
1538 struct _address_to_library_name {
1539   address addr;          // input : memory address
1540   size_t  buflen;        //         size of fname
1541   char*   fname;         // output: library name
1542   address base;          //         library base addr
1543 };
1544 
1545 static int address_to_library_name_callback(struct dl_phdr_info *info,
1546                                             size_t size, void *data) {
1547   int i;
1548   bool found = false;
1549   address libbase = NULL;
1550   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1551 
1552   // iterate through all loadable segments
1553   for (i = 0; i < info->dlpi_phnum; i++) {
1554     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1555     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1556       // base address of a library is the lowest address of its loaded
1557       // segments.
1558       if (libbase == NULL || libbase > segbase) {
1559         libbase = segbase;
1560       }
1561       // see if 'addr' is within current segment
1562       if (segbase <= d->addr &&
1563           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1564         found = true;
1565       }
1566     }
1567   }
1568 
1569   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1570   // so dll_address_to_library_name() can fall through to use dladdr() which
1571   // can figure out executable name from argv[0].
1572   if (found && info->dlpi_name && info->dlpi_name[0]) {
1573     d->base = libbase;
1574     if (d->fname) {
1575       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1576     }
1577     return 1;
1578   }
1579   return 0;
1580 }
1581 
1582 bool os::dll_address_to_library_name(address addr, char* buf,
1583                                      int buflen, int* offset) {
1584   // buf is not optional, but offset is optional
1585   assert(buf != NULL, "sanity check");
1586 
1587   Dl_info dlinfo;
1588   struct _address_to_library_name data;
1589 
1590   // There is a bug in old glibc dladdr() implementation that it could resolve
1591   // to wrong library name if the .so file has a base address != NULL. Here
1592   // we iterate through the program headers of all loaded libraries to find
1593   // out which library 'addr' really belongs to. This workaround can be
1594   // removed once the minimum requirement for glibc is moved to 2.3.x.
1595   data.addr = addr;
1596   data.fname = buf;
1597   data.buflen = buflen;
1598   data.base = NULL;
1599   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1600 
1601   if (rslt) {
1602     // buf already contains library name
1603     if (offset) *offset = addr - data.base;
1604     return true;
1605   }
1606   if (dladdr((void*)addr, &dlinfo) != 0) {
1607     if (dlinfo.dli_fname != NULL) {
1608       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1609     }
1610     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1611       *offset = addr - (address)dlinfo.dli_fbase;
1612     }
1613     return true;
1614   }
1615 
1616   buf[0] = '\0';
1617   if (offset) *offset = -1;
1618   return false;
1619 }
1620 
1621 // Loads .dll/.so and
1622 // in case of error it checks if .dll/.so was built for the
1623 // same architecture as Hotspot is running on
1624 
1625 
1626 // Remember the stack's state. The Linux dynamic linker will change
1627 // the stack to 'executable' at most once, so we must safepoint only once.
1628 bool os::Linux::_stack_is_executable = false;
1629 
1630 // VM operation that loads a library.  This is necessary if stack protection
1631 // of the Java stacks can be lost during loading the library.  If we
1632 // do not stop the Java threads, they can stack overflow before the stacks
1633 // are protected again.
1634 class VM_LinuxDllLoad: public VM_Operation {
1635  private:
1636   const char *_filename;
1637   char *_ebuf;
1638   int _ebuflen;
1639   void *_lib;
1640  public:
1641   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1642     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1643   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1644   void doit() {
1645     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1646     os::Linux::_stack_is_executable = true;
1647   }
1648   void* loaded_library() { return _lib; }
1649 };
1650 
1651 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1652   void * result = NULL;
1653   bool load_attempted = false;
1654 
1655   // Check whether the library to load might change execution rights
1656   // of the stack. If they are changed, the protection of the stack
1657   // guard pages will be lost. We need a safepoint to fix this.
1658   //
1659   // See Linux man page execstack(8) for more info.
1660   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1661     ElfFile ef(filename);
1662     if (!ef.specifies_noexecstack()) {
1663       if (!is_init_completed()) {
1664         os::Linux::_stack_is_executable = true;
1665         // This is OK - No Java threads have been created yet, and hence no
1666         // stack guard pages to fix.
1667         //
1668         // This should happen only when you are building JDK7 using a very
1669         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1670         //
1671         // Dynamic loader will make all stacks executable after
1672         // this function returns, and will not do that again.
1673         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1674       } else {
1675         warning("You have loaded library %s which might have disabled stack guard. "
1676                 "The VM will try to fix the stack guard now.\n"
1677                 "It's highly recommended that you fix the library with "
1678                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1679                 filename);
1680 
1681         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1682         JavaThread *jt = JavaThread::current();
1683         if (jt->thread_state() != _thread_in_native) {
1684           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1685           // that requires ExecStack. Cannot enter safe point. Let's give up.
1686           warning("Unable to fix stack guard. Giving up.");
1687         } else {
1688           if (!LoadExecStackDllInVMThread) {
1689             // This is for the case where the DLL has an static
1690             // constructor function that executes JNI code. We cannot
1691             // load such DLLs in the VMThread.
1692             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1693           }
1694 
1695           ThreadInVMfromNative tiv(jt);
1696           debug_only(VMNativeEntryWrapper vew;)
1697 
1698           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1699           VMThread::execute(&op);
1700           if (LoadExecStackDllInVMThread) {
1701             result = op.loaded_library();
1702           }
1703           load_attempted = true;
1704         }
1705       }
1706     }
1707   }
1708 
1709   if (!load_attempted) {
1710     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1711   }
1712 
1713   if (result != NULL) {
1714     // Successful loading
1715     return result;
1716   }
1717 
1718   Elf32_Ehdr elf_head;
1719   int diag_msg_max_length=ebuflen-strlen(ebuf);
1720   char* diag_msg_buf=ebuf+strlen(ebuf);
1721 
1722   if (diag_msg_max_length==0) {
1723     // No more space in ebuf for additional diagnostics message
1724     return NULL;
1725   }
1726 
1727 
1728   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1729 
1730   if (file_descriptor < 0) {
1731     // Can't open library, report dlerror() message
1732     return NULL;
1733   }
1734 
1735   bool failed_to_read_elf_head=
1736     (sizeof(elf_head)!=
1737      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1738 
1739   ::close(file_descriptor);
1740   if (failed_to_read_elf_head) {
1741     // file i/o error - report dlerror() msg
1742     return NULL;
1743   }
1744 
1745   typedef struct {
1746     Elf32_Half  code;         // Actual value as defined in elf.h
1747     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1748     char        elf_class;    // 32 or 64 bit
1749     char        endianess;    // MSB or LSB
1750     char*       name;         // String representation
1751   } arch_t;
1752 
1753 #ifndef EM_486
1754   #define EM_486          6               /* Intel 80486 */
1755 #endif
1756 #ifndef EM_AARCH64
1757   #define EM_AARCH64    183               /* ARM AARCH64 */
1758 #endif
1759 
1760   static const arch_t arch_array[]={
1761     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1762     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1763     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1764     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1765     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1766     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1767     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1768     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1769 #if defined(VM_LITTLE_ENDIAN)
1770     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1771 #else
1772     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1773 #endif
1774     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1775     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1776     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1777     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1778     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1779     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1780     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1781     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1782   };
1783 
1784 #if  (defined IA32)
1785   static  Elf32_Half running_arch_code=EM_386;
1786 #elif   (defined AMD64)
1787   static  Elf32_Half running_arch_code=EM_X86_64;
1788 #elif  (defined IA64)
1789   static  Elf32_Half running_arch_code=EM_IA_64;
1790 #elif  (defined __sparc) && (defined _LP64)
1791   static  Elf32_Half running_arch_code=EM_SPARCV9;
1792 #elif  (defined __sparc) && (!defined _LP64)
1793   static  Elf32_Half running_arch_code=EM_SPARC;
1794 #elif  (defined __powerpc64__)
1795   static  Elf32_Half running_arch_code=EM_PPC64;
1796 #elif  (defined __powerpc__)
1797   static  Elf32_Half running_arch_code=EM_PPC;
1798 #elif  (defined ARM)
1799   static  Elf32_Half running_arch_code=EM_ARM;
1800 #elif  (defined S390)
1801   static  Elf32_Half running_arch_code=EM_S390;
1802 #elif  (defined ALPHA)
1803   static  Elf32_Half running_arch_code=EM_ALPHA;
1804 #elif  (defined MIPSEL)
1805   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1806 #elif  (defined PARISC)
1807   static  Elf32_Half running_arch_code=EM_PARISC;
1808 #elif  (defined MIPS)
1809   static  Elf32_Half running_arch_code=EM_MIPS;
1810 #elif  (defined M68K)
1811   static  Elf32_Half running_arch_code=EM_68K;
1812 #elif  (defined AARCH64)
1813   static  Elf32_Half running_arch_code=EM_AARCH64;
1814 #else
1815     #error Method os::dll_load requires that one of following is defined:\
1816          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1817 #endif
1818 
1819   // Identify compatability class for VM's architecture and library's architecture
1820   // Obtain string descriptions for architectures
1821 
1822   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1823   int running_arch_index=-1;
1824 
1825   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1826     if (running_arch_code == arch_array[i].code) {
1827       running_arch_index    = i;
1828     }
1829     if (lib_arch.code == arch_array[i].code) {
1830       lib_arch.compat_class = arch_array[i].compat_class;
1831       lib_arch.name         = arch_array[i].name;
1832     }
1833   }
1834 
1835   assert(running_arch_index != -1,
1836          "Didn't find running architecture code (running_arch_code) in arch_array");
1837   if (running_arch_index == -1) {
1838     // Even though running architecture detection failed
1839     // we may still continue with reporting dlerror() message
1840     return NULL;
1841   }
1842 
1843   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1844     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1845     return NULL;
1846   }
1847 
1848 #ifndef S390
1849   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1850     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1851     return NULL;
1852   }
1853 #endif // !S390
1854 
1855   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1856     if (lib_arch.name!=NULL) {
1857       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1858                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1859                  lib_arch.name, arch_array[running_arch_index].name);
1860     } else {
1861       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1862                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1863                  lib_arch.code,
1864                  arch_array[running_arch_index].name);
1865     }
1866   }
1867 
1868   return NULL;
1869 }
1870 
1871 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1872                                 int ebuflen) {
1873   void * result = ::dlopen(filename, RTLD_LAZY);
1874   if (result == NULL) {
1875     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1876     ebuf[ebuflen-1] = '\0';
1877   }
1878   return result;
1879 }
1880 
1881 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1882                                        int ebuflen) {
1883   void * result = NULL;
1884   if (LoadExecStackDllInVMThread) {
1885     result = dlopen_helper(filename, ebuf, ebuflen);
1886   }
1887 
1888   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1889   // library that requires an executable stack, or which does not have this
1890   // stack attribute set, dlopen changes the stack attribute to executable. The
1891   // read protection of the guard pages gets lost.
1892   //
1893   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1894   // may have been queued at the same time.
1895 
1896   if (!_stack_is_executable) {
1897     JavaThread *jt = Threads::first();
1898 
1899     while (jt) {
1900       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1901           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1902         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1903           warning("Attempt to reguard stack yellow zone failed.");
1904         }
1905       }
1906       jt = jt->next();
1907     }
1908   }
1909 
1910   return result;
1911 }
1912 
1913 void* os::dll_lookup(void* handle, const char* name) {
1914   void* res = dlsym(handle, name);
1915   return res;
1916 }
1917 
1918 void* os::get_default_process_handle() {
1919   return (void*)::dlopen(NULL, RTLD_LAZY);
1920 }
1921 
1922 static bool _print_ascii_file(const char* filename, outputStream* st) {
1923   int fd = ::open(filename, O_RDONLY);
1924   if (fd == -1) {
1925     return false;
1926   }
1927 
1928   char buf[32];
1929   int bytes;
1930   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1931     st->print_raw(buf, bytes);
1932   }
1933 
1934   ::close(fd);
1935 
1936   return true;
1937 }
1938 
1939 void os::print_dll_info(outputStream *st) {
1940   st->print_cr("Dynamic libraries:");
1941 
1942   char fname[32];
1943   pid_t pid = os::Linux::gettid();
1944 
1945   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1946 
1947   if (!_print_ascii_file(fname, st)) {
1948     st->print("Can not get library information for pid = %d\n", pid);
1949   }
1950 }
1951 
1952 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1953   FILE *procmapsFile = NULL;
1954 
1955   // Open the procfs maps file for the current process
1956   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1957     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1958     char line[PATH_MAX + 100];
1959 
1960     // Read line by line from 'file'
1961     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1962       u8 base, top, offset, inode;
1963       char permissions[5];
1964       char device[6];
1965       char name[PATH_MAX + 1];
1966 
1967       // Parse fields from line
1968       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1969              &base, &top, permissions, &offset, device, &inode, name);
1970 
1971       // Filter by device id '00:00' so that we only get file system mapped files.
1972       if (strcmp(device, "00:00") != 0) {
1973 
1974         // Call callback with the fields of interest
1975         if(callback(name, (address)base, (address)top, param)) {
1976           // Oops abort, callback aborted
1977           fclose(procmapsFile);
1978           return 1;
1979         }
1980       }
1981     }
1982     fclose(procmapsFile);
1983   }
1984   return 0;
1985 }
1986 
1987 void os::print_os_info_brief(outputStream* st) {
1988   os::Linux::print_distro_info(st);
1989 
1990   os::Posix::print_uname_info(st);
1991 
1992   os::Linux::print_libversion_info(st);
1993 
1994 }
1995 
1996 void os::print_os_info(outputStream* st) {
1997   st->print("OS:");
1998 
1999   os::Linux::print_distro_info(st);
2000 
2001   os::Posix::print_uname_info(st);
2002 
2003   // Print warning if unsafe chroot environment detected
2004   if (unsafe_chroot_detected) {
2005     st->print("WARNING!! ");
2006     st->print_cr("%s", unstable_chroot_error);
2007   }
2008 
2009   os::Linux::print_libversion_info(st);
2010 
2011   os::Posix::print_rlimit_info(st);
2012 
2013   os::Posix::print_load_average(st);
2014 
2015   os::Linux::print_full_memory_info(st);
2016 }
2017 
2018 // Try to identify popular distros.
2019 // Most Linux distributions have a /etc/XXX-release file, which contains
2020 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2021 // file that also contains the OS version string. Some have more than one
2022 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2023 // /etc/redhat-release.), so the order is important.
2024 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2025 // their own specific XXX-release file as well as a redhat-release file.
2026 // Because of this the XXX-release file needs to be searched for before the
2027 // redhat-release file.
2028 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2029 // search for redhat-release / SuSE-release needs to be before lsb-release.
2030 // Since the lsb-release file is the new standard it needs to be searched
2031 // before the older style release files.
2032 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2033 // next to last resort.  The os-release file is a new standard that contains
2034 // distribution information and the system-release file seems to be an old
2035 // standard that has been replaced by the lsb-release and os-release files.
2036 // Searching for the debian_version file is the last resort.  It contains
2037 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2038 // "Debian " is printed before the contents of the debian_version file.
2039 
2040 const char* distro_files[] = {
2041   "/etc/oracle-release",
2042   "/etc/mandriva-release",
2043   "/etc/mandrake-release",
2044   "/etc/sun-release",
2045   "/etc/redhat-release",
2046   "/etc/SuSE-release",
2047   "/etc/lsb-release",
2048   "/etc/turbolinux-release",
2049   "/etc/gentoo-release",
2050   "/etc/ltib-release",
2051   "/etc/angstrom-version",
2052   "/etc/system-release",
2053   "/etc/os-release",
2054   NULL };
2055 
2056 void os::Linux::print_distro_info(outputStream* st) {
2057   for (int i = 0;; i++) {
2058     const char* file = distro_files[i];
2059     if (file == NULL) {
2060       break;  // done
2061     }
2062     // If file prints, we found it.
2063     if (_print_ascii_file(file, st)) {
2064       return;
2065     }
2066   }
2067 
2068   if (file_exists("/etc/debian_version")) {
2069     st->print("Debian ");
2070     _print_ascii_file("/etc/debian_version", st);
2071   } else {
2072     st->print("Linux");
2073   }
2074   st->cr();
2075 }
2076 
2077 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2078   char buf[256];
2079   while (fgets(buf, sizeof(buf), fp)) {
2080     // Edit out extra stuff in expected format
2081     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2082       char* ptr = strstr(buf, "\"");  // the name is in quotes
2083       if (ptr != NULL) {
2084         ptr++; // go beyond first quote
2085         char* nl = strchr(ptr, '\"');
2086         if (nl != NULL) *nl = '\0';
2087         strncpy(distro, ptr, length);
2088       } else {
2089         ptr = strstr(buf, "=");
2090         ptr++; // go beyond equals then
2091         char* nl = strchr(ptr, '\n');
2092         if (nl != NULL) *nl = '\0';
2093         strncpy(distro, ptr, length);
2094       }
2095       return;
2096     } else if (get_first_line) {
2097       char* nl = strchr(buf, '\n');
2098       if (nl != NULL) *nl = '\0';
2099       strncpy(distro, buf, length);
2100       return;
2101     }
2102   }
2103   // print last line and close
2104   char* nl = strchr(buf, '\n');
2105   if (nl != NULL) *nl = '\0';
2106   strncpy(distro, buf, length);
2107 }
2108 
2109 static void parse_os_info(char* distro, size_t length, const char* file) {
2110   FILE* fp = fopen(file, "r");
2111   if (fp != NULL) {
2112     // if suse format, print out first line
2113     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2114     parse_os_info_helper(fp, distro, length, get_first_line);
2115     fclose(fp);
2116   }
2117 }
2118 
2119 void os::get_summary_os_info(char* buf, size_t buflen) {
2120   for (int i = 0;; i++) {
2121     const char* file = distro_files[i];
2122     if (file == NULL) {
2123       break; // ran out of distro_files
2124     }
2125     if (file_exists(file)) {
2126       parse_os_info(buf, buflen, file);
2127       return;
2128     }
2129   }
2130   // special case for debian
2131   if (file_exists("/etc/debian_version")) {
2132     strncpy(buf, "Debian ", buflen);
2133     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2134   } else {
2135     strncpy(buf, "Linux", buflen);
2136   }
2137 }
2138 
2139 void os::Linux::print_libversion_info(outputStream* st) {
2140   // libc, pthread
2141   st->print("libc:");
2142   st->print("%s ", os::Linux::glibc_version());
2143   st->print("%s ", os::Linux::libpthread_version());
2144   st->cr();
2145 }
2146 
2147 void os::Linux::print_full_memory_info(outputStream* st) {
2148   st->print("\n/proc/meminfo:\n");
2149   _print_ascii_file("/proc/meminfo", st);
2150   st->cr();
2151 }
2152 
2153 void os::print_memory_info(outputStream* st) {
2154 
2155   st->print("Memory:");
2156   st->print(" %dk page", os::vm_page_size()>>10);
2157 
2158   // values in struct sysinfo are "unsigned long"
2159   struct sysinfo si;
2160   sysinfo(&si);
2161 
2162   st->print(", physical " UINT64_FORMAT "k",
2163             os::physical_memory() >> 10);
2164   st->print("(" UINT64_FORMAT "k free)",
2165             os::available_memory() >> 10);
2166   st->print(", swap " UINT64_FORMAT "k",
2167             ((jlong)si.totalswap * si.mem_unit) >> 10);
2168   st->print("(" UINT64_FORMAT "k free)",
2169             ((jlong)si.freeswap * si.mem_unit) >> 10);
2170   st->cr();
2171 }
2172 
2173 // Print the first "model name" line and the first "flags" line
2174 // that we find and nothing more. We assume "model name" comes
2175 // before "flags" so if we find a second "model name", then the
2176 // "flags" field is considered missing.
2177 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2178 #if defined(IA32) || defined(AMD64)
2179   // Other platforms have less repetitive cpuinfo files
2180   FILE *fp = fopen("/proc/cpuinfo", "r");
2181   if (fp) {
2182     while (!feof(fp)) {
2183       if (fgets(buf, buflen, fp)) {
2184         // Assume model name comes before flags
2185         bool model_name_printed = false;
2186         if (strstr(buf, "model name") != NULL) {
2187           if (!model_name_printed) {
2188             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2189             st->print_raw(buf);
2190             model_name_printed = true;
2191           } else {
2192             // model name printed but not flags?  Odd, just return
2193             fclose(fp);
2194             return true;
2195           }
2196         }
2197         // print the flags line too
2198         if (strstr(buf, "flags") != NULL) {
2199           st->print_raw(buf);
2200           fclose(fp);
2201           return true;
2202         }
2203       }
2204     }
2205     fclose(fp);
2206   }
2207 #endif // x86 platforms
2208   return false;
2209 }
2210 
2211 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2212   // Only print the model name if the platform provides this as a summary
2213   if (!print_model_name_and_flags(st, buf, buflen)) {
2214     st->print("\n/proc/cpuinfo:\n");
2215     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2216       st->print_cr("  <Not Available>");
2217     }
2218   }
2219 }
2220 
2221 #if defined(AMD64) || defined(IA32) || defined(X32)
2222 const char* search_string = "model name";
2223 #elif defined(SPARC)
2224 const char* search_string = "cpu";
2225 #elif defined(PPC64)
2226 const char* search_string = "cpu";
2227 #else
2228 const char* search_string = "Processor";
2229 #endif
2230 
2231 // Parses the cpuinfo file for string representing the model name.
2232 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2233   FILE* fp = fopen("/proc/cpuinfo", "r");
2234   if (fp != NULL) {
2235     while (!feof(fp)) {
2236       char buf[256];
2237       if (fgets(buf, sizeof(buf), fp)) {
2238         char* start = strstr(buf, search_string);
2239         if (start != NULL) {
2240           char *ptr = start + strlen(search_string);
2241           char *end = buf + strlen(buf);
2242           while (ptr != end) {
2243              // skip whitespace and colon for the rest of the name.
2244              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2245                break;
2246              }
2247              ptr++;
2248           }
2249           if (ptr != end) {
2250             // reasonable string, get rid of newline and keep the rest
2251             char* nl = strchr(buf, '\n');
2252             if (nl != NULL) *nl = '\0';
2253             strncpy(cpuinfo, ptr, length);
2254             fclose(fp);
2255             return;
2256           }
2257         }
2258       }
2259     }
2260     fclose(fp);
2261   }
2262   // cpuinfo not found or parsing failed, just print generic string.  The entire
2263   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2264 #if defined(AMD64)
2265   strncpy(cpuinfo, "x86_64", length);
2266 #elif defined(IA32)
2267   strncpy(cpuinfo, "x86_32", length);
2268 #elif defined(IA64)
2269   strncpy(cpuinfo, "IA64", length);
2270 #elif defined(SPARC)
2271   strncpy(cpuinfo, "sparcv9", length);
2272 #elif defined(AARCH64)
2273   strncpy(cpuinfo, "AArch64", length);
2274 #elif defined(ARM)
2275   strncpy(cpuinfo, "ARM", length);
2276 #elif defined(PPC)
2277   strncpy(cpuinfo, "PPC64", length);
2278 #elif defined(ZERO_LIBARCH)
2279   strncpy(cpuinfo, ZERO_LIBARCH, length);
2280 #else
2281   strncpy(cpuinfo, "unknown", length);
2282 #endif
2283 }
2284 
2285 static void print_signal_handler(outputStream* st, int sig,
2286                                  char* buf, size_t buflen);
2287 
2288 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2289   st->print_cr("Signal Handlers:");
2290   print_signal_handler(st, SIGSEGV, buf, buflen);
2291   print_signal_handler(st, SIGBUS , buf, buflen);
2292   print_signal_handler(st, SIGFPE , buf, buflen);
2293   print_signal_handler(st, SIGPIPE, buf, buflen);
2294   print_signal_handler(st, SIGXFSZ, buf, buflen);
2295   print_signal_handler(st, SIGILL , buf, buflen);
2296   print_signal_handler(st, SR_signum, buf, buflen);
2297   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2298   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2299   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2300   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2301 #if defined(PPC64)
2302   print_signal_handler(st, SIGTRAP, buf, buflen);
2303 #endif
2304 }
2305 
2306 static char saved_jvm_path[MAXPATHLEN] = {0};
2307 
2308 // Find the full path to the current module, libjvm.so
2309 void os::jvm_path(char *buf, jint buflen) {
2310   // Error checking.
2311   if (buflen < MAXPATHLEN) {
2312     assert(false, "must use a large-enough buffer");
2313     buf[0] = '\0';
2314     return;
2315   }
2316   // Lazy resolve the path to current module.
2317   if (saved_jvm_path[0] != 0) {
2318     strcpy(buf, saved_jvm_path);
2319     return;
2320   }
2321 
2322   char dli_fname[MAXPATHLEN];
2323   bool ret = dll_address_to_library_name(
2324                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2325                                          dli_fname, sizeof(dli_fname), NULL);
2326   assert(ret, "cannot locate libjvm");
2327   char *rp = NULL;
2328   if (ret && dli_fname[0] != '\0') {
2329     rp = realpath(dli_fname, buf);
2330   }
2331   if (rp == NULL) {
2332     return;
2333   }
2334 
2335   if (Arguments::sun_java_launcher_is_altjvm()) {
2336     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2337     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2338     // If "/jre/lib/" appears at the right place in the string, then
2339     // assume we are installed in a JDK and we're done. Otherwise, check
2340     // for a JAVA_HOME environment variable and fix up the path so it
2341     // looks like libjvm.so is installed there (append a fake suffix
2342     // hotspot/libjvm.so).
2343     const char *p = buf + strlen(buf) - 1;
2344     for (int count = 0; p > buf && count < 5; ++count) {
2345       for (--p; p > buf && *p != '/'; --p)
2346         /* empty */ ;
2347     }
2348 
2349     if (strncmp(p, "/jre/lib/", 9) != 0) {
2350       // Look for JAVA_HOME in the environment.
2351       char* java_home_var = ::getenv("JAVA_HOME");
2352       if (java_home_var != NULL && java_home_var[0] != 0) {
2353         char* jrelib_p;
2354         int len;
2355 
2356         // Check the current module name "libjvm.so".
2357         p = strrchr(buf, '/');
2358         if (p == NULL) {
2359           return;
2360         }
2361         assert(strstr(p, "/libjvm") == p, "invalid library name");
2362 
2363         rp = realpath(java_home_var, buf);
2364         if (rp == NULL) {
2365           return;
2366         }
2367 
2368         // determine if this is a legacy image or modules image
2369         // modules image doesn't have "jre" subdirectory
2370         len = strlen(buf);
2371         assert(len < buflen, "Ran out of buffer room");
2372         jrelib_p = buf + len;
2373         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2374         if (0 != access(buf, F_OK)) {
2375           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2376         }
2377 
2378         if (0 == access(buf, F_OK)) {
2379           // Use current module name "libjvm.so"
2380           len = strlen(buf);
2381           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2382         } else {
2383           // Go back to path of .so
2384           rp = realpath(dli_fname, buf);
2385           if (rp == NULL) {
2386             return;
2387           }
2388         }
2389       }
2390     }
2391   }
2392 
2393   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2394   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2395 }
2396 
2397 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2398   // no prefix required, not even "_"
2399 }
2400 
2401 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2402   // no suffix required
2403 }
2404 
2405 ////////////////////////////////////////////////////////////////////////////////
2406 // sun.misc.Signal support
2407 
2408 static volatile jint sigint_count = 0;
2409 
2410 static void UserHandler(int sig, void *siginfo, void *context) {
2411   // 4511530 - sem_post is serialized and handled by the manager thread. When
2412   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2413   // don't want to flood the manager thread with sem_post requests.
2414   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2415     return;
2416   }
2417 
2418   // Ctrl-C is pressed during error reporting, likely because the error
2419   // handler fails to abort. Let VM die immediately.
2420   if (sig == SIGINT && is_error_reported()) {
2421     os::die();
2422   }
2423 
2424   os::signal_notify(sig);
2425 }
2426 
2427 void* os::user_handler() {
2428   return CAST_FROM_FN_PTR(void*, UserHandler);
2429 }
2430 
2431 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2432   struct timespec ts;
2433   // Semaphore's are always associated with CLOCK_REALTIME
2434   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2435   // see unpackTime for discussion on overflow checking
2436   if (sec >= MAX_SECS) {
2437     ts.tv_sec += MAX_SECS;
2438     ts.tv_nsec = 0;
2439   } else {
2440     ts.tv_sec += sec;
2441     ts.tv_nsec += nsec;
2442     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2443       ts.tv_nsec -= NANOSECS_PER_SEC;
2444       ++ts.tv_sec; // note: this must be <= max_secs
2445     }
2446   }
2447 
2448   return ts;
2449 }
2450 
2451 extern "C" {
2452   typedef void (*sa_handler_t)(int);
2453   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2454 }
2455 
2456 void* os::signal(int signal_number, void* handler) {
2457   struct sigaction sigAct, oldSigAct;
2458 
2459   sigfillset(&(sigAct.sa_mask));
2460   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2461   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2462 
2463   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2464     // -1 means registration failed
2465     return (void *)-1;
2466   }
2467 
2468   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2469 }
2470 
2471 void os::signal_raise(int signal_number) {
2472   ::raise(signal_number);
2473 }
2474 
2475 // The following code is moved from os.cpp for making this
2476 // code platform specific, which it is by its very nature.
2477 
2478 // Will be modified when max signal is changed to be dynamic
2479 int os::sigexitnum_pd() {
2480   return NSIG;
2481 }
2482 
2483 // a counter for each possible signal value
2484 static volatile jint pending_signals[NSIG+1] = { 0 };
2485 
2486 // Linux(POSIX) specific hand shaking semaphore.
2487 static sem_t sig_sem;
2488 static PosixSemaphore sr_semaphore;
2489 
2490 void os::signal_init_pd() {
2491   // Initialize signal structures
2492   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2493 
2494   // Initialize signal semaphore
2495   ::sem_init(&sig_sem, 0, 0);
2496 }
2497 
2498 void os::signal_notify(int sig) {
2499   Atomic::inc(&pending_signals[sig]);
2500   ::sem_post(&sig_sem);
2501 }
2502 
2503 static int check_pending_signals(bool wait) {
2504   Atomic::store(0, &sigint_count);
2505   for (;;) {
2506     for (int i = 0; i < NSIG + 1; i++) {
2507       jint n = pending_signals[i];
2508       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2509         return i;
2510       }
2511     }
2512     if (!wait) {
2513       return -1;
2514     }
2515     JavaThread *thread = JavaThread::current();
2516     ThreadBlockInVM tbivm(thread);
2517 
2518     bool threadIsSuspended;
2519     do {
2520       thread->set_suspend_equivalent();
2521       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2522       ::sem_wait(&sig_sem);
2523 
2524       // were we externally suspended while we were waiting?
2525       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2526       if (threadIsSuspended) {
2527         // The semaphore has been incremented, but while we were waiting
2528         // another thread suspended us. We don't want to continue running
2529         // while suspended because that would surprise the thread that
2530         // suspended us.
2531         ::sem_post(&sig_sem);
2532 
2533         thread->java_suspend_self();
2534       }
2535     } while (threadIsSuspended);
2536   }
2537 }
2538 
2539 int os::signal_lookup() {
2540   return check_pending_signals(false);
2541 }
2542 
2543 int os::signal_wait() {
2544   return check_pending_signals(true);
2545 }
2546 
2547 ////////////////////////////////////////////////////////////////////////////////
2548 // Virtual Memory
2549 
2550 int os::vm_page_size() {
2551   // Seems redundant as all get out
2552   assert(os::Linux::page_size() != -1, "must call os::init");
2553   return os::Linux::page_size();
2554 }
2555 
2556 // Solaris allocates memory by pages.
2557 int os::vm_allocation_granularity() {
2558   assert(os::Linux::page_size() != -1, "must call os::init");
2559   return os::Linux::page_size();
2560 }
2561 
2562 // Rationale behind this function:
2563 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2564 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2565 //  samples for JITted code. Here we create private executable mapping over the code cache
2566 //  and then we can use standard (well, almost, as mapping can change) way to provide
2567 //  info for the reporting script by storing timestamp and location of symbol
2568 void linux_wrap_code(char* base, size_t size) {
2569   static volatile jint cnt = 0;
2570 
2571   if (!UseOprofile) {
2572     return;
2573   }
2574 
2575   char buf[PATH_MAX+1];
2576   int num = Atomic::add(1, &cnt);
2577 
2578   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2579            os::get_temp_directory(), os::current_process_id(), num);
2580   unlink(buf);
2581 
2582   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2583 
2584   if (fd != -1) {
2585     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2586     if (rv != (off_t)-1) {
2587       if (::write(fd, "", 1) == 1) {
2588         mmap(base, size,
2589              PROT_READ|PROT_WRITE|PROT_EXEC,
2590              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2591       }
2592     }
2593     ::close(fd);
2594     unlink(buf);
2595   }
2596 }
2597 
2598 static bool recoverable_mmap_error(int err) {
2599   // See if the error is one we can let the caller handle. This
2600   // list of errno values comes from JBS-6843484. I can't find a
2601   // Linux man page that documents this specific set of errno
2602   // values so while this list currently matches Solaris, it may
2603   // change as we gain experience with this failure mode.
2604   switch (err) {
2605   case EBADF:
2606   case EINVAL:
2607   case ENOTSUP:
2608     // let the caller deal with these errors
2609     return true;
2610 
2611   default:
2612     // Any remaining errors on this OS can cause our reserved mapping
2613     // to be lost. That can cause confusion where different data
2614     // structures think they have the same memory mapped. The worst
2615     // scenario is if both the VM and a library think they have the
2616     // same memory mapped.
2617     return false;
2618   }
2619 }
2620 
2621 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2622                                     int err) {
2623   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2624           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2625           os::strerror(err), err);
2626 }
2627 
2628 static void warn_fail_commit_memory(char* addr, size_t size,
2629                                     size_t alignment_hint, bool exec,
2630                                     int err) {
2631   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2632           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2633           alignment_hint, exec, os::strerror(err), err);
2634 }
2635 
2636 // NOTE: Linux kernel does not really reserve the pages for us.
2637 //       All it does is to check if there are enough free pages
2638 //       left at the time of mmap(). This could be a potential
2639 //       problem.
2640 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2641   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2642   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2643                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2644   if (res != (uintptr_t) MAP_FAILED) {
2645     if (UseNUMAInterleaving) {
2646       numa_make_global(addr, size);
2647     }
2648     return 0;
2649   }
2650 
2651   int err = errno;  // save errno from mmap() call above
2652 
2653   if (!recoverable_mmap_error(err)) {
2654     warn_fail_commit_memory(addr, size, exec, err);
2655     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2656   }
2657 
2658   return err;
2659 }
2660 
2661 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2662   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2663 }
2664 
2665 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2666                                   const char* mesg) {
2667   assert(mesg != NULL, "mesg must be specified");
2668   int err = os::Linux::commit_memory_impl(addr, size, exec);
2669   if (err != 0) {
2670     // the caller wants all commit errors to exit with the specified mesg:
2671     warn_fail_commit_memory(addr, size, exec, err);
2672     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2673   }
2674 }
2675 
2676 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2677 #ifndef MAP_HUGETLB
2678   #define MAP_HUGETLB 0x40000
2679 #endif
2680 
2681 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2682 #ifndef MADV_HUGEPAGE
2683   #define MADV_HUGEPAGE 14
2684 #endif
2685 
2686 int os::Linux::commit_memory_impl(char* addr, size_t size,
2687                                   size_t alignment_hint, bool exec) {
2688   int err = os::Linux::commit_memory_impl(addr, size, exec);
2689   if (err == 0) {
2690     realign_memory(addr, size, alignment_hint);
2691   }
2692   return err;
2693 }
2694 
2695 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2696                           bool exec) {
2697   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2698 }
2699 
2700 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2701                                   size_t alignment_hint, bool exec,
2702                                   const char* mesg) {
2703   assert(mesg != NULL, "mesg must be specified");
2704   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2705   if (err != 0) {
2706     // the caller wants all commit errors to exit with the specified mesg:
2707     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2708     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2709   }
2710 }
2711 
2712 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2713   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2714     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2715     // be supported or the memory may already be backed by huge pages.
2716     ::madvise(addr, bytes, MADV_HUGEPAGE);
2717   }
2718 }
2719 
2720 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2721   // This method works by doing an mmap over an existing mmaping and effectively discarding
2722   // the existing pages. However it won't work for SHM-based large pages that cannot be
2723   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2724   // small pages on top of the SHM segment. This method always works for small pages, so we
2725   // allow that in any case.
2726   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2727     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2728   }
2729 }
2730 
2731 void os::numa_make_global(char *addr, size_t bytes) {
2732   Linux::numa_interleave_memory(addr, bytes);
2733 }
2734 
2735 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2736 // bind policy to MPOL_PREFERRED for the current thread.
2737 #define USE_MPOL_PREFERRED 0
2738 
2739 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2740   // To make NUMA and large pages more robust when both enabled, we need to ease
2741   // the requirements on where the memory should be allocated. MPOL_BIND is the
2742   // default policy and it will force memory to be allocated on the specified
2743   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2744   // the specified node, but will not force it. Using this policy will prevent
2745   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2746   // free large pages.
2747   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2748   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2749 }
2750 
2751 bool os::numa_topology_changed() { return false; }
2752 
2753 size_t os::numa_get_groups_num() {
2754   int max_node = Linux::numa_max_node();
2755   return max_node > 0 ? max_node + 1 : 1;
2756 }
2757 
2758 int os::numa_get_group_id() {
2759   int cpu_id = Linux::sched_getcpu();
2760   if (cpu_id != -1) {
2761     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2762     if (lgrp_id != -1) {
2763       return lgrp_id;
2764     }
2765   }
2766   return 0;
2767 }
2768 
2769 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2770   for (size_t i = 0; i < size; i++) {
2771     ids[i] = i;
2772   }
2773   return size;
2774 }
2775 
2776 bool os::get_page_info(char *start, page_info* info) {
2777   return false;
2778 }
2779 
2780 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2781                      page_info* page_found) {
2782   return end;
2783 }
2784 
2785 
2786 int os::Linux::sched_getcpu_syscall(void) {
2787   unsigned int cpu = 0;
2788   int retval = -1;
2789 
2790 #if defined(IA32)
2791   #ifndef SYS_getcpu
2792     #define SYS_getcpu 318
2793   #endif
2794   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2795 #elif defined(AMD64)
2796 // Unfortunately we have to bring all these macros here from vsyscall.h
2797 // to be able to compile on old linuxes.
2798   #define __NR_vgetcpu 2
2799   #define VSYSCALL_START (-10UL << 20)
2800   #define VSYSCALL_SIZE 1024
2801   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2802   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2803   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2804   retval = vgetcpu(&cpu, NULL, NULL);
2805 #endif
2806 
2807   return (retval == -1) ? retval : cpu;
2808 }
2809 
2810 // Something to do with the numa-aware allocator needs these symbols
2811 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2812 extern "C" JNIEXPORT void numa_error(char *where) { }
2813 
2814 
2815 // If we are running with libnuma version > 2, then we should
2816 // be trying to use symbols with versions 1.1
2817 // If we are running with earlier version, which did not have symbol versions,
2818 // we should use the base version.
2819 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2820   void *f = dlvsym(handle, name, "libnuma_1.1");
2821   if (f == NULL) {
2822     f = dlsym(handle, name);
2823   }
2824   return f;
2825 }
2826 
2827 bool os::Linux::libnuma_init() {
2828   // sched_getcpu() should be in libc.
2829   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2830                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2831 
2832   // If it's not, try a direct syscall.
2833   if (sched_getcpu() == -1) {
2834     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2835                                     (void*)&sched_getcpu_syscall));
2836   }
2837 
2838   if (sched_getcpu() != -1) { // Does it work?
2839     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2840     if (handle != NULL) {
2841       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2842                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2843       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2844                                        libnuma_dlsym(handle, "numa_max_node")));
2845       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2846                                         libnuma_dlsym(handle, "numa_available")));
2847       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2848                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2849       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2850                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2851       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2852                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2853 
2854 
2855       if (numa_available() != -1) {
2856         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2857         // Create a cpu -> node mapping
2858         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2859         rebuild_cpu_to_node_map();
2860         return true;
2861       }
2862     }
2863   }
2864   return false;
2865 }
2866 
2867 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2868 // The table is later used in get_node_by_cpu().
2869 void os::Linux::rebuild_cpu_to_node_map() {
2870   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2871                               // in libnuma (possible values are starting from 16,
2872                               // and continuing up with every other power of 2, but less
2873                               // than the maximum number of CPUs supported by kernel), and
2874                               // is a subject to change (in libnuma version 2 the requirements
2875                               // are more reasonable) we'll just hardcode the number they use
2876                               // in the library.
2877   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2878 
2879   size_t cpu_num = os::active_processor_count();
2880   size_t cpu_map_size = NCPUS / BitsPerCLong;
2881   size_t cpu_map_valid_size =
2882     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2883 
2884   cpu_to_node()->clear();
2885   cpu_to_node()->at_grow(cpu_num - 1);
2886   size_t node_num = numa_get_groups_num();
2887 
2888   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2889   for (size_t i = 0; i < node_num; i++) {
2890     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2891       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2892         if (cpu_map[j] != 0) {
2893           for (size_t k = 0; k < BitsPerCLong; k++) {
2894             if (cpu_map[j] & (1UL << k)) {
2895               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2896             }
2897           }
2898         }
2899       }
2900     }
2901   }
2902   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2903 }
2904 
2905 int os::Linux::get_node_by_cpu(int cpu_id) {
2906   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2907     return cpu_to_node()->at(cpu_id);
2908   }
2909   return -1;
2910 }
2911 
2912 GrowableArray<int>* os::Linux::_cpu_to_node;
2913 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2914 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2915 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2916 os::Linux::numa_available_func_t os::Linux::_numa_available;
2917 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2918 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2919 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2920 unsigned long* os::Linux::_numa_all_nodes;
2921 
2922 bool os::pd_uncommit_memory(char* addr, size_t size) {
2923   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2924                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2925   return res  != (uintptr_t) MAP_FAILED;
2926 }
2927 
2928 static address get_stack_commited_bottom(address bottom, size_t size) {
2929   address nbot = bottom;
2930   address ntop = bottom + size;
2931 
2932   size_t page_sz = os::vm_page_size();
2933   unsigned pages = size / page_sz;
2934 
2935   unsigned char vec[1];
2936   unsigned imin = 1, imax = pages + 1, imid;
2937   int mincore_return_value = 0;
2938 
2939   assert(imin <= imax, "Unexpected page size");
2940 
2941   while (imin < imax) {
2942     imid = (imax + imin) / 2;
2943     nbot = ntop - (imid * page_sz);
2944 
2945     // Use a trick with mincore to check whether the page is mapped or not.
2946     // mincore sets vec to 1 if page resides in memory and to 0 if page
2947     // is swapped output but if page we are asking for is unmapped
2948     // it returns -1,ENOMEM
2949     mincore_return_value = mincore(nbot, page_sz, vec);
2950 
2951     if (mincore_return_value == -1) {
2952       // Page is not mapped go up
2953       // to find first mapped page
2954       if (errno != EAGAIN) {
2955         assert(errno == ENOMEM, "Unexpected mincore errno");
2956         imax = imid;
2957       }
2958     } else {
2959       // Page is mapped go down
2960       // to find first not mapped page
2961       imin = imid + 1;
2962     }
2963   }
2964 
2965   nbot = nbot + page_sz;
2966 
2967   // Adjust stack bottom one page up if last checked page is not mapped
2968   if (mincore_return_value == -1) {
2969     nbot = nbot + page_sz;
2970   }
2971 
2972   return nbot;
2973 }
2974 
2975 
2976 // Linux uses a growable mapping for the stack, and if the mapping for
2977 // the stack guard pages is not removed when we detach a thread the
2978 // stack cannot grow beyond the pages where the stack guard was
2979 // mapped.  If at some point later in the process the stack expands to
2980 // that point, the Linux kernel cannot expand the stack any further
2981 // because the guard pages are in the way, and a segfault occurs.
2982 //
2983 // However, it's essential not to split the stack region by unmapping
2984 // a region (leaving a hole) that's already part of the stack mapping,
2985 // so if the stack mapping has already grown beyond the guard pages at
2986 // the time we create them, we have to truncate the stack mapping.
2987 // So, we need to know the extent of the stack mapping when
2988 // create_stack_guard_pages() is called.
2989 
2990 // We only need this for stacks that are growable: at the time of
2991 // writing thread stacks don't use growable mappings (i.e. those
2992 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2993 // only applies to the main thread.
2994 
2995 // If the (growable) stack mapping already extends beyond the point
2996 // where we're going to put our guard pages, truncate the mapping at
2997 // that point by munmap()ping it.  This ensures that when we later
2998 // munmap() the guard pages we don't leave a hole in the stack
2999 // mapping. This only affects the main/initial thread
3000 
3001 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3002   if (os::Linux::is_initial_thread()) {
3003     // As we manually grow stack up to bottom inside create_attached_thread(),
3004     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3005     // we don't need to do anything special.
3006     // Check it first, before calling heavy function.
3007     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3008     unsigned char vec[1];
3009 
3010     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3011       // Fallback to slow path on all errors, including EAGAIN
3012       stack_extent = (uintptr_t) get_stack_commited_bottom(
3013                                                            os::Linux::initial_thread_stack_bottom(),
3014                                                            (size_t)addr - stack_extent);
3015     }
3016 
3017     if (stack_extent < (uintptr_t)addr) {
3018       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3019     }
3020   }
3021 
3022   return os::commit_memory(addr, size, !ExecMem);
3023 }
3024 
3025 // If this is a growable mapping, remove the guard pages entirely by
3026 // munmap()ping them.  If not, just call uncommit_memory(). This only
3027 // affects the main/initial thread, but guard against future OS changes
3028 // It's safe to always unmap guard pages for initial thread because we
3029 // always place it right after end of the mapped region
3030 
3031 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3032   uintptr_t stack_extent, stack_base;
3033 
3034   if (os::Linux::is_initial_thread()) {
3035     return ::munmap(addr, size) == 0;
3036   }
3037 
3038   return os::uncommit_memory(addr, size);
3039 }
3040 
3041 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3042 // at 'requested_addr'. If there are existing memory mappings at the same
3043 // location, however, they will be overwritten. If 'fixed' is false,
3044 // 'requested_addr' is only treated as a hint, the return value may or
3045 // may not start from the requested address. Unlike Linux mmap(), this
3046 // function returns NULL to indicate failure.
3047 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3048   char * addr;
3049   int flags;
3050 
3051   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3052   if (fixed) {
3053     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3054     flags |= MAP_FIXED;
3055   }
3056 
3057   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3058   // touch an uncommitted page. Otherwise, the read/write might
3059   // succeed if we have enough swap space to back the physical page.
3060   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3061                        flags, -1, 0);
3062 
3063   return addr == MAP_FAILED ? NULL : addr;
3064 }
3065 
3066 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3067 //   (req_addr != NULL) or with a given alignment.
3068 //  - bytes shall be a multiple of alignment.
3069 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3070 //  - alignment sets the alignment at which memory shall be allocated.
3071 //     It must be a multiple of allocation granularity.
3072 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3073 //  req_addr or NULL.
3074 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3075 
3076   size_t extra_size = bytes;
3077   if (req_addr == NULL && alignment > 0) {
3078     extra_size += alignment;
3079   }
3080 
3081   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3082     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3083     -1, 0);
3084   if (start == MAP_FAILED) {
3085     start = NULL;
3086   } else {
3087     if (req_addr != NULL) {
3088       if (start != req_addr) {
3089         ::munmap(start, extra_size);
3090         start = NULL;
3091       }
3092     } else {
3093       char* const start_aligned = (char*) align_ptr_up(start, alignment);
3094       char* const end_aligned = start_aligned + bytes;
3095       char* const end = start + extra_size;
3096       if (start_aligned > start) {
3097         ::munmap(start, start_aligned - start);
3098       }
3099       if (end_aligned < end) {
3100         ::munmap(end_aligned, end - end_aligned);
3101       }
3102       start = start_aligned;
3103     }
3104   }
3105   return start;
3106 }
3107 
3108 static int anon_munmap(char * addr, size_t size) {
3109   return ::munmap(addr, size) == 0;
3110 }
3111 
3112 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3113                             size_t alignment_hint) {
3114   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3115 }
3116 
3117 bool os::pd_release_memory(char* addr, size_t size) {
3118   return anon_munmap(addr, size);
3119 }
3120 
3121 static bool linux_mprotect(char* addr, size_t size, int prot) {
3122   // Linux wants the mprotect address argument to be page aligned.
3123   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3124 
3125   // According to SUSv3, mprotect() should only be used with mappings
3126   // established by mmap(), and mmap() always maps whole pages. Unaligned
3127   // 'addr' likely indicates problem in the VM (e.g. trying to change
3128   // protection of malloc'ed or statically allocated memory). Check the
3129   // caller if you hit this assert.
3130   assert(addr == bottom, "sanity check");
3131 
3132   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3133   return ::mprotect(bottom, size, prot) == 0;
3134 }
3135 
3136 // Set protections specified
3137 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3138                         bool is_committed) {
3139   unsigned int p = 0;
3140   switch (prot) {
3141   case MEM_PROT_NONE: p = PROT_NONE; break;
3142   case MEM_PROT_READ: p = PROT_READ; break;
3143   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3144   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3145   default:
3146     ShouldNotReachHere();
3147   }
3148   // is_committed is unused.
3149   return linux_mprotect(addr, bytes, p);
3150 }
3151 
3152 bool os::guard_memory(char* addr, size_t size) {
3153   return linux_mprotect(addr, size, PROT_NONE);
3154 }
3155 
3156 bool os::unguard_memory(char* addr, size_t size) {
3157   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3158 }
3159 
3160 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3161                                                     size_t page_size) {
3162   bool result = false;
3163   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3164                  MAP_ANONYMOUS|MAP_PRIVATE,
3165                  -1, 0);
3166   if (p != MAP_FAILED) {
3167     void *aligned_p = align_ptr_up(p, page_size);
3168 
3169     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3170 
3171     munmap(p, page_size * 2);
3172   }
3173 
3174   if (warn && !result) {
3175     warning("TransparentHugePages is not supported by the operating system.");
3176   }
3177 
3178   return result;
3179 }
3180 
3181 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3182   bool result = false;
3183   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3184                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3185                  -1, 0);
3186 
3187   if (p != MAP_FAILED) {
3188     // We don't know if this really is a huge page or not.
3189     FILE *fp = fopen("/proc/self/maps", "r");
3190     if (fp) {
3191       while (!feof(fp)) {
3192         char chars[257];
3193         long x = 0;
3194         if (fgets(chars, sizeof(chars), fp)) {
3195           if (sscanf(chars, "%lx-%*x", &x) == 1
3196               && x == (long)p) {
3197             if (strstr (chars, "hugepage")) {
3198               result = true;
3199               break;
3200             }
3201           }
3202         }
3203       }
3204       fclose(fp);
3205     }
3206     munmap(p, page_size);
3207   }
3208 
3209   if (warn && !result) {
3210     warning("HugeTLBFS is not supported by the operating system.");
3211   }
3212 
3213   return result;
3214 }
3215 
3216 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3217 //
3218 // From the coredump_filter documentation:
3219 //
3220 // - (bit 0) anonymous private memory
3221 // - (bit 1) anonymous shared memory
3222 // - (bit 2) file-backed private memory
3223 // - (bit 3) file-backed shared memory
3224 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3225 //           effective only if the bit 2 is cleared)
3226 // - (bit 5) hugetlb private memory
3227 // - (bit 6) hugetlb shared memory
3228 //
3229 static void set_coredump_filter(void) {
3230   FILE *f;
3231   long cdm;
3232 
3233   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3234     return;
3235   }
3236 
3237   if (fscanf(f, "%lx", &cdm) != 1) {
3238     fclose(f);
3239     return;
3240   }
3241 
3242   rewind(f);
3243 
3244   if ((cdm & LARGEPAGES_BIT) == 0) {
3245     cdm |= LARGEPAGES_BIT;
3246     fprintf(f, "%#lx", cdm);
3247   }
3248 
3249   fclose(f);
3250 }
3251 
3252 // Large page support
3253 
3254 static size_t _large_page_size = 0;
3255 
3256 size_t os::Linux::find_large_page_size() {
3257   size_t large_page_size = 0;
3258 
3259   // large_page_size on Linux is used to round up heap size. x86 uses either
3260   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3261   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3262   // page as large as 256M.
3263   //
3264   // Here we try to figure out page size by parsing /proc/meminfo and looking
3265   // for a line with the following format:
3266   //    Hugepagesize:     2048 kB
3267   //
3268   // If we can't determine the value (e.g. /proc is not mounted, or the text
3269   // format has been changed), we'll use the largest page size supported by
3270   // the processor.
3271 
3272 #ifndef ZERO
3273   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3274                      ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3275 #endif // ZERO
3276 
3277   FILE *fp = fopen("/proc/meminfo", "r");
3278   if (fp) {
3279     while (!feof(fp)) {
3280       int x = 0;
3281       char buf[16];
3282       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3283         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3284           large_page_size = x * K;
3285           break;
3286         }
3287       } else {
3288         // skip to next line
3289         for (;;) {
3290           int ch = fgetc(fp);
3291           if (ch == EOF || ch == (int)'\n') break;
3292         }
3293       }
3294     }
3295     fclose(fp);
3296   }
3297 
3298   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3299     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3300             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3301             proper_unit_for_byte_size(large_page_size));
3302   }
3303 
3304   return large_page_size;
3305 }
3306 
3307 size_t os::Linux::setup_large_page_size() {
3308   _large_page_size = Linux::find_large_page_size();
3309   const size_t default_page_size = (size_t)Linux::page_size();
3310   if (_large_page_size > default_page_size) {
3311     _page_sizes[0] = _large_page_size;
3312     _page_sizes[1] = default_page_size;
3313     _page_sizes[2] = 0;
3314   }
3315 
3316   return _large_page_size;
3317 }
3318 
3319 bool os::Linux::setup_large_page_type(size_t page_size) {
3320   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3321       FLAG_IS_DEFAULT(UseSHM) &&
3322       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3323 
3324     // The type of large pages has not been specified by the user.
3325 
3326     // Try UseHugeTLBFS and then UseSHM.
3327     UseHugeTLBFS = UseSHM = true;
3328 
3329     // Don't try UseTransparentHugePages since there are known
3330     // performance issues with it turned on. This might change in the future.
3331     UseTransparentHugePages = false;
3332   }
3333 
3334   if (UseTransparentHugePages) {
3335     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3336     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3337       UseHugeTLBFS = false;
3338       UseSHM = false;
3339       return true;
3340     }
3341     UseTransparentHugePages = false;
3342   }
3343 
3344   if (UseHugeTLBFS) {
3345     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3346     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3347       UseSHM = false;
3348       return true;
3349     }
3350     UseHugeTLBFS = false;
3351   }
3352 
3353   return UseSHM;
3354 }
3355 
3356 void os::large_page_init() {
3357   if (!UseLargePages &&
3358       !UseTransparentHugePages &&
3359       !UseHugeTLBFS &&
3360       !UseSHM) {
3361     // Not using large pages.
3362     return;
3363   }
3364 
3365   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3366     // The user explicitly turned off large pages.
3367     // Ignore the rest of the large pages flags.
3368     UseTransparentHugePages = false;
3369     UseHugeTLBFS = false;
3370     UseSHM = false;
3371     return;
3372   }
3373 
3374   size_t large_page_size = Linux::setup_large_page_size();
3375   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3376 
3377   set_coredump_filter();
3378 }
3379 
3380 #ifndef SHM_HUGETLB
3381   #define SHM_HUGETLB 04000
3382 #endif
3383 
3384 #define shm_warning_format(format, ...)              \
3385   do {                                               \
3386     if (UseLargePages &&                             \
3387         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3388          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3389          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3390       warning(format, __VA_ARGS__);                  \
3391     }                                                \
3392   } while (0)
3393 
3394 #define shm_warning(str) shm_warning_format("%s", str)
3395 
3396 #define shm_warning_with_errno(str)                \
3397   do {                                             \
3398     int err = errno;                               \
3399     shm_warning_format(str " (error = %d)", err);  \
3400   } while (0)
3401 
3402 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3403   assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3404 
3405   if (!is_size_aligned(alignment, SHMLBA)) {
3406     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3407     return NULL;
3408   }
3409 
3410   // To ensure that we get 'alignment' aligned memory from shmat,
3411   // we pre-reserve aligned virtual memory and then attach to that.
3412 
3413   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3414   if (pre_reserved_addr == NULL) {
3415     // Couldn't pre-reserve aligned memory.
3416     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3417     return NULL;
3418   }
3419 
3420   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3421   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3422 
3423   if ((intptr_t)addr == -1) {
3424     int err = errno;
3425     shm_warning_with_errno("Failed to attach shared memory.");
3426 
3427     assert(err != EACCES, "Unexpected error");
3428     assert(err != EIDRM,  "Unexpected error");
3429     assert(err != EINVAL, "Unexpected error");
3430 
3431     // Since we don't know if the kernel unmapped the pre-reserved memory area
3432     // we can't unmap it, since that would potentially unmap memory that was
3433     // mapped from other threads.
3434     return NULL;
3435   }
3436 
3437   return addr;
3438 }
3439 
3440 static char* shmat_at_address(int shmid, char* req_addr) {
3441   if (!is_ptr_aligned(req_addr, SHMLBA)) {
3442     assert(false, "Requested address needs to be SHMLBA aligned");
3443     return NULL;
3444   }
3445 
3446   char* addr = (char*)shmat(shmid, req_addr, 0);
3447 
3448   if ((intptr_t)addr == -1) {
3449     shm_warning_with_errno("Failed to attach shared memory.");
3450     return NULL;
3451   }
3452 
3453   return addr;
3454 }
3455 
3456 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3457   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3458   if (req_addr != NULL) {
3459     assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3460     assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3461     return shmat_at_address(shmid, req_addr);
3462   }
3463 
3464   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3465   // return large page size aligned memory addresses when req_addr == NULL.
3466   // However, if the alignment is larger than the large page size, we have
3467   // to manually ensure that the memory returned is 'alignment' aligned.
3468   if (alignment > os::large_page_size()) {
3469     assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3470     return shmat_with_alignment(shmid, bytes, alignment);
3471   } else {
3472     return shmat_at_address(shmid, NULL);
3473   }
3474 }
3475 
3476 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3477                                             char* req_addr, bool exec) {
3478   // "exec" is passed in but not used.  Creating the shared image for
3479   // the code cache doesn't have an SHM_X executable permission to check.
3480   assert(UseLargePages && UseSHM, "only for SHM large pages");
3481   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3482   assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3483 
3484   if (!is_size_aligned(bytes, os::large_page_size())) {
3485     return NULL; // Fallback to small pages.
3486   }
3487 
3488   // Create a large shared memory region to attach to based on size.
3489   // Currently, size is the total size of the heap.
3490   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3491   if (shmid == -1) {
3492     // Possible reasons for shmget failure:
3493     // 1. shmmax is too small for Java heap.
3494     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3495     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3496     // 2. not enough large page memory.
3497     //    > check available large pages: cat /proc/meminfo
3498     //    > increase amount of large pages:
3499     //          echo new_value > /proc/sys/vm/nr_hugepages
3500     //      Note 1: different Linux may use different name for this property,
3501     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3502     //      Note 2: it's possible there's enough physical memory available but
3503     //            they are so fragmented after a long run that they can't
3504     //            coalesce into large pages. Try to reserve large pages when
3505     //            the system is still "fresh".
3506     shm_warning_with_errno("Failed to reserve shared memory.");
3507     return NULL;
3508   }
3509 
3510   // Attach to the region.
3511   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3512 
3513   // Remove shmid. If shmat() is successful, the actual shared memory segment
3514   // will be deleted when it's detached by shmdt() or when the process
3515   // terminates. If shmat() is not successful this will remove the shared
3516   // segment immediately.
3517   shmctl(shmid, IPC_RMID, NULL);
3518 
3519   return addr;
3520 }
3521 
3522 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3523                                         int error) {
3524   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3525 
3526   bool warn_on_failure = UseLargePages &&
3527       (!FLAG_IS_DEFAULT(UseLargePages) ||
3528        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3529        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3530 
3531   if (warn_on_failure) {
3532     char msg[128];
3533     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3534                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3535     warning("%s", msg);
3536   }
3537 }
3538 
3539 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3540                                                         char* req_addr,
3541                                                         bool exec) {
3542   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3543   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3544   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3545 
3546   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3547   char* addr = (char*)::mmap(req_addr, bytes, prot,
3548                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3549                              -1, 0);
3550 
3551   if (addr == MAP_FAILED) {
3552     warn_on_large_pages_failure(req_addr, bytes, errno);
3553     return NULL;
3554   }
3555 
3556   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3557 
3558   return addr;
3559 }
3560 
3561 // Reserve memory using mmap(MAP_HUGETLB).
3562 //  - bytes shall be a multiple of alignment.
3563 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3564 //  - alignment sets the alignment at which memory shall be allocated.
3565 //     It must be a multiple of allocation granularity.
3566 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3567 //  req_addr or NULL.
3568 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3569                                                          size_t alignment,
3570                                                          char* req_addr,
3571                                                          bool exec) {
3572   size_t large_page_size = os::large_page_size();
3573   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3574 
3575   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3576   assert(is_size_aligned(bytes, alignment), "Must be");
3577 
3578   // First reserve - but not commit - the address range in small pages.
3579   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3580 
3581   if (start == NULL) {
3582     return NULL;
3583   }
3584 
3585   assert(is_ptr_aligned(start, alignment), "Must be");
3586 
3587   char* end = start + bytes;
3588 
3589   // Find the regions of the allocated chunk that can be promoted to large pages.
3590   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3591   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3592 
3593   size_t lp_bytes = lp_end - lp_start;
3594 
3595   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3596 
3597   if (lp_bytes == 0) {
3598     // The mapped region doesn't even span the start and the end of a large page.
3599     // Fall back to allocate a non-special area.
3600     ::munmap(start, end - start);
3601     return NULL;
3602   }
3603 
3604   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3605 
3606   void* result;
3607 
3608   // Commit small-paged leading area.
3609   if (start != lp_start) {
3610     result = ::mmap(start, lp_start - start, prot,
3611                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3612                     -1, 0);
3613     if (result == MAP_FAILED) {
3614       ::munmap(lp_start, end - lp_start);
3615       return NULL;
3616     }
3617   }
3618 
3619   // Commit large-paged area.
3620   result = ::mmap(lp_start, lp_bytes, prot,
3621                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3622                   -1, 0);
3623   if (result == MAP_FAILED) {
3624     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3625     // If the mmap above fails, the large pages region will be unmapped and we
3626     // have regions before and after with small pages. Release these regions.
3627     //
3628     // |  mapped  |  unmapped  |  mapped  |
3629     // ^          ^            ^          ^
3630     // start      lp_start     lp_end     end
3631     //
3632     ::munmap(start, lp_start - start);
3633     ::munmap(lp_end, end - lp_end);
3634     return NULL;
3635   }
3636 
3637   // Commit small-paged trailing area.
3638   if (lp_end != end) {
3639     result = ::mmap(lp_end, end - lp_end, prot,
3640                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3641                     -1, 0);
3642     if (result == MAP_FAILED) {
3643       ::munmap(start, lp_end - start);
3644       return NULL;
3645     }
3646   }
3647 
3648   return start;
3649 }
3650 
3651 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3652                                                    size_t alignment,
3653                                                    char* req_addr,
3654                                                    bool exec) {
3655   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3656   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3657   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3658   assert(is_power_of_2(os::large_page_size()), "Must be");
3659   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3660 
3661   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3662     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3663   } else {
3664     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3665   }
3666 }
3667 
3668 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3669                                  char* req_addr, bool exec) {
3670   assert(UseLargePages, "only for large pages");
3671 
3672   char* addr;
3673   if (UseSHM) {
3674     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3675   } else {
3676     assert(UseHugeTLBFS, "must be");
3677     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3678   }
3679 
3680   if (addr != NULL) {
3681     if (UseNUMAInterleaving) {
3682       numa_make_global(addr, bytes);
3683     }
3684 
3685     // The memory is committed
3686     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3687   }
3688 
3689   return addr;
3690 }
3691 
3692 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3693   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3694   return shmdt(base) == 0;
3695 }
3696 
3697 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3698   return pd_release_memory(base, bytes);
3699 }
3700 
3701 bool os::release_memory_special(char* base, size_t bytes) {
3702   bool res;
3703   if (MemTracker::tracking_level() > NMT_minimal) {
3704     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3705     res = os::Linux::release_memory_special_impl(base, bytes);
3706     if (res) {
3707       tkr.record((address)base, bytes);
3708     }
3709 
3710   } else {
3711     res = os::Linux::release_memory_special_impl(base, bytes);
3712   }
3713   return res;
3714 }
3715 
3716 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3717   assert(UseLargePages, "only for large pages");
3718   bool res;
3719 
3720   if (UseSHM) {
3721     res = os::Linux::release_memory_special_shm(base, bytes);
3722   } else {
3723     assert(UseHugeTLBFS, "must be");
3724     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3725   }
3726   return res;
3727 }
3728 
3729 size_t os::large_page_size() {
3730   return _large_page_size;
3731 }
3732 
3733 // With SysV SHM the entire memory region must be allocated as shared
3734 // memory.
3735 // HugeTLBFS allows application to commit large page memory on demand.
3736 // However, when committing memory with HugeTLBFS fails, the region
3737 // that was supposed to be committed will lose the old reservation
3738 // and allow other threads to steal that memory region. Because of this
3739 // behavior we can't commit HugeTLBFS memory.
3740 bool os::can_commit_large_page_memory() {
3741   return UseTransparentHugePages;
3742 }
3743 
3744 bool os::can_execute_large_page_memory() {
3745   return UseTransparentHugePages || UseHugeTLBFS;
3746 }
3747 
3748 // Reserve memory at an arbitrary address, only if that area is
3749 // available (and not reserved for something else).
3750 
3751 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3752   const int max_tries = 10;
3753   char* base[max_tries];
3754   size_t size[max_tries];
3755   const size_t gap = 0x000000;
3756 
3757   // Assert only that the size is a multiple of the page size, since
3758   // that's all that mmap requires, and since that's all we really know
3759   // about at this low abstraction level.  If we need higher alignment,
3760   // we can either pass an alignment to this method or verify alignment
3761   // in one of the methods further up the call chain.  See bug 5044738.
3762   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3763 
3764   // Repeatedly allocate blocks until the block is allocated at the
3765   // right spot.
3766 
3767   // Linux mmap allows caller to pass an address as hint; give it a try first,
3768   // if kernel honors the hint then we can return immediately.
3769   char * addr = anon_mmap(requested_addr, bytes, false);
3770   if (addr == requested_addr) {
3771     return requested_addr;
3772   }
3773 
3774   if (addr != NULL) {
3775     // mmap() is successful but it fails to reserve at the requested address
3776     anon_munmap(addr, bytes);
3777   }
3778 
3779   int i;
3780   for (i = 0; i < max_tries; ++i) {
3781     base[i] = reserve_memory(bytes);
3782 
3783     if (base[i] != NULL) {
3784       // Is this the block we wanted?
3785       if (base[i] == requested_addr) {
3786         size[i] = bytes;
3787         break;
3788       }
3789 
3790       // Does this overlap the block we wanted? Give back the overlapped
3791       // parts and try again.
3792 
3793       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3794       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3795         unmap_memory(base[i], top_overlap);
3796         base[i] += top_overlap;
3797         size[i] = bytes - top_overlap;
3798       } else {
3799         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3800         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3801           unmap_memory(requested_addr, bottom_overlap);
3802           size[i] = bytes - bottom_overlap;
3803         } else {
3804           size[i] = bytes;
3805         }
3806       }
3807     }
3808   }
3809 
3810   // Give back the unused reserved pieces.
3811 
3812   for (int j = 0; j < i; ++j) {
3813     if (base[j] != NULL) {
3814       unmap_memory(base[j], size[j]);
3815     }
3816   }
3817 
3818   if (i < max_tries) {
3819     return requested_addr;
3820   } else {
3821     return NULL;
3822   }
3823 }
3824 
3825 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3826   return ::read(fd, buf, nBytes);
3827 }
3828 
3829 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3830   return ::pread(fd, buf, nBytes, offset);
3831 }
3832 
3833 // Short sleep, direct OS call.
3834 //
3835 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3836 // sched_yield(2) will actually give up the CPU:
3837 //
3838 //   * Alone on this pariticular CPU, keeps running.
3839 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3840 //     (pre 2.6.39).
3841 //
3842 // So calling this with 0 is an alternative.
3843 //
3844 void os::naked_short_sleep(jlong ms) {
3845   struct timespec req;
3846 
3847   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3848   req.tv_sec = 0;
3849   if (ms > 0) {
3850     req.tv_nsec = (ms % 1000) * 1000000;
3851   } else {
3852     req.tv_nsec = 1;
3853   }
3854 
3855   nanosleep(&req, NULL);
3856 
3857   return;
3858 }
3859 
3860 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3861 void os::infinite_sleep() {
3862   while (true) {    // sleep forever ...
3863     ::sleep(100);   // ... 100 seconds at a time
3864   }
3865 }
3866 
3867 // Used to convert frequent JVM_Yield() to nops
3868 bool os::dont_yield() {
3869   return DontYieldALot;
3870 }
3871 
3872 void os::naked_yield() {
3873   sched_yield();
3874 }
3875 
3876 ////////////////////////////////////////////////////////////////////////////////
3877 // thread priority support
3878 
3879 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3880 // only supports dynamic priority, static priority must be zero. For real-time
3881 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3882 // However, for large multi-threaded applications, SCHED_RR is not only slower
3883 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3884 // of 5 runs - Sep 2005).
3885 //
3886 // The following code actually changes the niceness of kernel-thread/LWP. It
3887 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3888 // not the entire user process, and user level threads are 1:1 mapped to kernel
3889 // threads. It has always been the case, but could change in the future. For
3890 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3891 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3892 
3893 int os::java_to_os_priority[CriticalPriority + 1] = {
3894   19,              // 0 Entry should never be used
3895 
3896    4,              // 1 MinPriority
3897    3,              // 2
3898    2,              // 3
3899 
3900    1,              // 4
3901    0,              // 5 NormPriority
3902   -1,              // 6
3903 
3904   -2,              // 7
3905   -3,              // 8
3906   -4,              // 9 NearMaxPriority
3907 
3908   -5,              // 10 MaxPriority
3909 
3910   -5               // 11 CriticalPriority
3911 };
3912 
3913 static int prio_init() {
3914   if (ThreadPriorityPolicy == 1) {
3915     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3916     // if effective uid is not root. Perhaps, a more elegant way of doing
3917     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3918     if (geteuid() != 0) {
3919       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3920         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3921       }
3922       ThreadPriorityPolicy = 0;
3923     }
3924   }
3925   if (UseCriticalJavaThreadPriority) {
3926     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3927   }
3928   return 0;
3929 }
3930 
3931 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3932   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3933 
3934   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3935   return (ret == 0) ? OS_OK : OS_ERR;
3936 }
3937 
3938 OSReturn os::get_native_priority(const Thread* const thread,
3939                                  int *priority_ptr) {
3940   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3941     *priority_ptr = java_to_os_priority[NormPriority];
3942     return OS_OK;
3943   }
3944 
3945   errno = 0;
3946   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3947   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3948 }
3949 
3950 // Hint to the underlying OS that a task switch would not be good.
3951 // Void return because it's a hint and can fail.
3952 void os::hint_no_preempt() {}
3953 
3954 ////////////////////////////////////////////////////////////////////////////////
3955 // suspend/resume support
3956 
3957 //  the low-level signal-based suspend/resume support is a remnant from the
3958 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3959 //  within hotspot. Now there is a single use-case for this:
3960 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3961 //      that runs in the watcher thread.
3962 //  The remaining code is greatly simplified from the more general suspension
3963 //  code that used to be used.
3964 //
3965 //  The protocol is quite simple:
3966 //  - suspend:
3967 //      - sends a signal to the target thread
3968 //      - polls the suspend state of the osthread using a yield loop
3969 //      - target thread signal handler (SR_handler) sets suspend state
3970 //        and blocks in sigsuspend until continued
3971 //  - resume:
3972 //      - sets target osthread state to continue
3973 //      - sends signal to end the sigsuspend loop in the SR_handler
3974 //
3975 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3976 
3977 static void resume_clear_context(OSThread *osthread) {
3978   osthread->set_ucontext(NULL);
3979   osthread->set_siginfo(NULL);
3980 }
3981 
3982 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3983                                  ucontext_t* context) {
3984   osthread->set_ucontext(context);
3985   osthread->set_siginfo(siginfo);
3986 }
3987 
3988 // Handler function invoked when a thread's execution is suspended or
3989 // resumed. We have to be careful that only async-safe functions are
3990 // called here (Note: most pthread functions are not async safe and
3991 // should be avoided.)
3992 //
3993 // Note: sigwait() is a more natural fit than sigsuspend() from an
3994 // interface point of view, but sigwait() prevents the signal hander
3995 // from being run. libpthread would get very confused by not having
3996 // its signal handlers run and prevents sigwait()'s use with the
3997 // mutex granting granting signal.
3998 //
3999 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4000 //
4001 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4002   // Save and restore errno to avoid confusing native code with EINTR
4003   // after sigsuspend.
4004   int old_errno = errno;
4005 
4006   Thread* thread = Thread::current_or_null_safe();
4007   assert(thread != NULL, "Missing current thread in SR_handler");
4008   OSThread* osthread = thread->osthread();
4009   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4010 
4011   os::SuspendResume::State current = osthread->sr.state();
4012   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4013     suspend_save_context(osthread, siginfo, context);
4014 
4015     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4016     os::SuspendResume::State state = osthread->sr.suspended();
4017     if (state == os::SuspendResume::SR_SUSPENDED) {
4018       sigset_t suspend_set;  // signals for sigsuspend()
4019       sigemptyset(&suspend_set);
4020       // get current set of blocked signals and unblock resume signal
4021       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4022       sigdelset(&suspend_set, SR_signum);
4023 
4024       sr_semaphore.signal();
4025       // wait here until we are resumed
4026       while (1) {
4027         sigsuspend(&suspend_set);
4028 
4029         os::SuspendResume::State result = osthread->sr.running();
4030         if (result == os::SuspendResume::SR_RUNNING) {
4031           sr_semaphore.signal();
4032           break;
4033         }
4034       }
4035 
4036     } else if (state == os::SuspendResume::SR_RUNNING) {
4037       // request was cancelled, continue
4038     } else {
4039       ShouldNotReachHere();
4040     }
4041 
4042     resume_clear_context(osthread);
4043   } else if (current == os::SuspendResume::SR_RUNNING) {
4044     // request was cancelled, continue
4045   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4046     // ignore
4047   } else {
4048     // ignore
4049   }
4050 
4051   errno = old_errno;
4052 }
4053 
4054 static int SR_initialize() {
4055   struct sigaction act;
4056   char *s;
4057 
4058   // Get signal number to use for suspend/resume
4059   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4060     int sig = ::strtol(s, 0, 10);
4061     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4062         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4063       SR_signum = sig;
4064     } else {
4065       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4066               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4067     }
4068   }
4069 
4070   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4071          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4072 
4073   sigemptyset(&SR_sigset);
4074   sigaddset(&SR_sigset, SR_signum);
4075 
4076   // Set up signal handler for suspend/resume
4077   act.sa_flags = SA_RESTART|SA_SIGINFO;
4078   act.sa_handler = (void (*)(int)) SR_handler;
4079 
4080   // SR_signum is blocked by default.
4081   // 4528190 - We also need to block pthread restart signal (32 on all
4082   // supported Linux platforms). Note that LinuxThreads need to block
4083   // this signal for all threads to work properly. So we don't have
4084   // to use hard-coded signal number when setting up the mask.
4085   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4086 
4087   if (sigaction(SR_signum, &act, 0) == -1) {
4088     return -1;
4089   }
4090 
4091   // Save signal flag
4092   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4093   return 0;
4094 }
4095 
4096 static int sr_notify(OSThread* osthread) {
4097   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4098   assert_status(status == 0, status, "pthread_kill");
4099   return status;
4100 }
4101 
4102 // "Randomly" selected value for how long we want to spin
4103 // before bailing out on suspending a thread, also how often
4104 // we send a signal to a thread we want to resume
4105 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4106 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4107 
4108 // returns true on success and false on error - really an error is fatal
4109 // but this seems the normal response to library errors
4110 static bool do_suspend(OSThread* osthread) {
4111   assert(osthread->sr.is_running(), "thread should be running");
4112   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4113 
4114   // mark as suspended and send signal
4115   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4116     // failed to switch, state wasn't running?
4117     ShouldNotReachHere();
4118     return false;
4119   }
4120 
4121   if (sr_notify(osthread) != 0) {
4122     ShouldNotReachHere();
4123   }
4124 
4125   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4126   while (true) {
4127     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4128       break;
4129     } else {
4130       // timeout
4131       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4132       if (cancelled == os::SuspendResume::SR_RUNNING) {
4133         return false;
4134       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4135         // make sure that we consume the signal on the semaphore as well
4136         sr_semaphore.wait();
4137         break;
4138       } else {
4139         ShouldNotReachHere();
4140         return false;
4141       }
4142     }
4143   }
4144 
4145   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4146   return true;
4147 }
4148 
4149 static void do_resume(OSThread* osthread) {
4150   assert(osthread->sr.is_suspended(), "thread should be suspended");
4151   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4152 
4153   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4154     // failed to switch to WAKEUP_REQUEST
4155     ShouldNotReachHere();
4156     return;
4157   }
4158 
4159   while (true) {
4160     if (sr_notify(osthread) == 0) {
4161       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4162         if (osthread->sr.is_running()) {
4163           return;
4164         }
4165       }
4166     } else {
4167       ShouldNotReachHere();
4168     }
4169   }
4170 
4171   guarantee(osthread->sr.is_running(), "Must be running!");
4172 }
4173 
4174 ///////////////////////////////////////////////////////////////////////////////////
4175 // signal handling (except suspend/resume)
4176 
4177 // This routine may be used by user applications as a "hook" to catch signals.
4178 // The user-defined signal handler must pass unrecognized signals to this
4179 // routine, and if it returns true (non-zero), then the signal handler must
4180 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4181 // routine will never retun false (zero), but instead will execute a VM panic
4182 // routine kill the process.
4183 //
4184 // If this routine returns false, it is OK to call it again.  This allows
4185 // the user-defined signal handler to perform checks either before or after
4186 // the VM performs its own checks.  Naturally, the user code would be making
4187 // a serious error if it tried to handle an exception (such as a null check
4188 // or breakpoint) that the VM was generating for its own correct operation.
4189 //
4190 // This routine may recognize any of the following kinds of signals:
4191 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4192 // It should be consulted by handlers for any of those signals.
4193 //
4194 // The caller of this routine must pass in the three arguments supplied
4195 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4196 // field of the structure passed to sigaction().  This routine assumes that
4197 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4198 //
4199 // Note that the VM will print warnings if it detects conflicting signal
4200 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4201 //
4202 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4203                                                  siginfo_t* siginfo,
4204                                                  void* ucontext,
4205                                                  int abort_if_unrecognized);
4206 
4207 void signalHandler(int sig, siginfo_t* info, void* uc) {
4208   assert(info != NULL && uc != NULL, "it must be old kernel");
4209   int orig_errno = errno;  // Preserve errno value over signal handler.
4210   JVM_handle_linux_signal(sig, info, uc, true);
4211   errno = orig_errno;
4212 }
4213 
4214 
4215 // This boolean allows users to forward their own non-matching signals
4216 // to JVM_handle_linux_signal, harmlessly.
4217 bool os::Linux::signal_handlers_are_installed = false;
4218 
4219 // For signal-chaining
4220 struct sigaction sigact[NSIG];
4221 uint64_t sigs = 0;
4222 #if (64 < NSIG-1)
4223 #error "Not all signals can be encoded in sigs. Adapt its type!"
4224 #endif
4225 bool os::Linux::libjsig_is_loaded = false;
4226 typedef struct sigaction *(*get_signal_t)(int);
4227 get_signal_t os::Linux::get_signal_action = NULL;
4228 
4229 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4230   struct sigaction *actp = NULL;
4231 
4232   if (libjsig_is_loaded) {
4233     // Retrieve the old signal handler from libjsig
4234     actp = (*get_signal_action)(sig);
4235   }
4236   if (actp == NULL) {
4237     // Retrieve the preinstalled signal handler from jvm
4238     actp = get_preinstalled_handler(sig);
4239   }
4240 
4241   return actp;
4242 }
4243 
4244 static bool call_chained_handler(struct sigaction *actp, int sig,
4245                                  siginfo_t *siginfo, void *context) {
4246   // Call the old signal handler
4247   if (actp->sa_handler == SIG_DFL) {
4248     // It's more reasonable to let jvm treat it as an unexpected exception
4249     // instead of taking the default action.
4250     return false;
4251   } else if (actp->sa_handler != SIG_IGN) {
4252     if ((actp->sa_flags & SA_NODEFER) == 0) {
4253       // automaticlly block the signal
4254       sigaddset(&(actp->sa_mask), sig);
4255     }
4256 
4257     sa_handler_t hand = NULL;
4258     sa_sigaction_t sa = NULL;
4259     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4260     // retrieve the chained handler
4261     if (siginfo_flag_set) {
4262       sa = actp->sa_sigaction;
4263     } else {
4264       hand = actp->sa_handler;
4265     }
4266 
4267     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4268       actp->sa_handler = SIG_DFL;
4269     }
4270 
4271     // try to honor the signal mask
4272     sigset_t oset;
4273     sigemptyset(&oset);
4274     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4275 
4276     // call into the chained handler
4277     if (siginfo_flag_set) {
4278       (*sa)(sig, siginfo, context);
4279     } else {
4280       (*hand)(sig);
4281     }
4282 
4283     // restore the signal mask
4284     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4285   }
4286   // Tell jvm's signal handler the signal is taken care of.
4287   return true;
4288 }
4289 
4290 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4291   bool chained = false;
4292   // signal-chaining
4293   if (UseSignalChaining) {
4294     struct sigaction *actp = get_chained_signal_action(sig);
4295     if (actp != NULL) {
4296       chained = call_chained_handler(actp, sig, siginfo, context);
4297     }
4298   }
4299   return chained;
4300 }
4301 
4302 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4303   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4304     return &sigact[sig];
4305   }
4306   return NULL;
4307 }
4308 
4309 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4310   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4311   sigact[sig] = oldAct;
4312   sigs |= (uint64_t)1 << (sig-1);
4313 }
4314 
4315 // for diagnostic
4316 int sigflags[NSIG];
4317 
4318 int os::Linux::get_our_sigflags(int sig) {
4319   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4320   return sigflags[sig];
4321 }
4322 
4323 void os::Linux::set_our_sigflags(int sig, int flags) {
4324   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4325   if (sig > 0 && sig < NSIG) {
4326     sigflags[sig] = flags;
4327   }
4328 }
4329 
4330 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4331   // Check for overwrite.
4332   struct sigaction oldAct;
4333   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4334 
4335   void* oldhand = oldAct.sa_sigaction
4336                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4337                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4338   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4339       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4340       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4341     if (AllowUserSignalHandlers || !set_installed) {
4342       // Do not overwrite; user takes responsibility to forward to us.
4343       return;
4344     } else if (UseSignalChaining) {
4345       // save the old handler in jvm
4346       save_preinstalled_handler(sig, oldAct);
4347       // libjsig also interposes the sigaction() call below and saves the
4348       // old sigaction on it own.
4349     } else {
4350       fatal("Encountered unexpected pre-existing sigaction handler "
4351             "%#lx for signal %d.", (long)oldhand, sig);
4352     }
4353   }
4354 
4355   struct sigaction sigAct;
4356   sigfillset(&(sigAct.sa_mask));
4357   sigAct.sa_handler = SIG_DFL;
4358   if (!set_installed) {
4359     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4360   } else {
4361     sigAct.sa_sigaction = signalHandler;
4362     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4363   }
4364   // Save flags, which are set by ours
4365   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4366   sigflags[sig] = sigAct.sa_flags;
4367 
4368   int ret = sigaction(sig, &sigAct, &oldAct);
4369   assert(ret == 0, "check");
4370 
4371   void* oldhand2  = oldAct.sa_sigaction
4372                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4373                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4374   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4375 }
4376 
4377 // install signal handlers for signals that HotSpot needs to
4378 // handle in order to support Java-level exception handling.
4379 
4380 void os::Linux::install_signal_handlers() {
4381   if (!signal_handlers_are_installed) {
4382     signal_handlers_are_installed = true;
4383 
4384     // signal-chaining
4385     typedef void (*signal_setting_t)();
4386     signal_setting_t begin_signal_setting = NULL;
4387     signal_setting_t end_signal_setting = NULL;
4388     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4389                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4390     if (begin_signal_setting != NULL) {
4391       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4392                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4393       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4394                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4395       libjsig_is_loaded = true;
4396       assert(UseSignalChaining, "should enable signal-chaining");
4397     }
4398     if (libjsig_is_loaded) {
4399       // Tell libjsig jvm is setting signal handlers
4400       (*begin_signal_setting)();
4401     }
4402 
4403     set_signal_handler(SIGSEGV, true);
4404     set_signal_handler(SIGPIPE, true);
4405     set_signal_handler(SIGBUS, true);
4406     set_signal_handler(SIGILL, true);
4407     set_signal_handler(SIGFPE, true);
4408 #if defined(PPC64)
4409     set_signal_handler(SIGTRAP, true);
4410 #endif
4411     set_signal_handler(SIGXFSZ, true);
4412 
4413     if (libjsig_is_loaded) {
4414       // Tell libjsig jvm finishes setting signal handlers
4415       (*end_signal_setting)();
4416     }
4417 
4418     // We don't activate signal checker if libjsig is in place, we trust ourselves
4419     // and if UserSignalHandler is installed all bets are off.
4420     // Log that signal checking is off only if -verbose:jni is specified.
4421     if (CheckJNICalls) {
4422       if (libjsig_is_loaded) {
4423         if (PrintJNIResolving) {
4424           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4425         }
4426         check_signals = false;
4427       }
4428       if (AllowUserSignalHandlers) {
4429         if (PrintJNIResolving) {
4430           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4431         }
4432         check_signals = false;
4433       }
4434     }
4435   }
4436 }
4437 
4438 // This is the fastest way to get thread cpu time on Linux.
4439 // Returns cpu time (user+sys) for any thread, not only for current.
4440 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4441 // It might work on 2.6.10+ with a special kernel/glibc patch.
4442 // For reference, please, see IEEE Std 1003.1-2004:
4443 //   http://www.unix.org/single_unix_specification
4444 
4445 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4446   struct timespec tp;
4447   int rc = os::Linux::clock_gettime(clockid, &tp);
4448   assert(rc == 0, "clock_gettime is expected to return 0 code");
4449 
4450   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4451 }
4452 
4453 void os::Linux::initialize_os_info() {
4454   assert(_os_version == 0, "OS info already initialized");
4455 
4456   struct utsname _uname;
4457 
4458   uint32_t major;
4459   uint32_t minor;
4460   uint32_t fix;
4461 
4462   int rc;
4463 
4464   // Kernel version is unknown if
4465   // verification below fails.
4466   _os_version = 0x01000000;
4467 
4468   rc = uname(&_uname);
4469   if (rc != -1) {
4470 
4471     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4472     if (rc == 3) {
4473 
4474       if (major < 256 && minor < 256 && fix < 256) {
4475         // Kernel version format is as expected,
4476         // set it overriding unknown state.
4477         _os_version = (major << 16) |
4478                       (minor << 8 ) |
4479                       (fix   << 0 ) ;
4480       }
4481     }
4482   }
4483 }
4484 
4485 uint32_t os::Linux::os_version() {
4486   assert(_os_version != 0, "not initialized");
4487   return _os_version & 0x00FFFFFF;
4488 }
4489 
4490 bool os::Linux::os_version_is_known() {
4491   assert(_os_version != 0, "not initialized");
4492   return _os_version & 0x01000000 ? false : true;
4493 }
4494 
4495 /////
4496 // glibc on Linux platform uses non-documented flag
4497 // to indicate, that some special sort of signal
4498 // trampoline is used.
4499 // We will never set this flag, and we should
4500 // ignore this flag in our diagnostic
4501 #ifdef SIGNIFICANT_SIGNAL_MASK
4502   #undef SIGNIFICANT_SIGNAL_MASK
4503 #endif
4504 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4505 
4506 static const char* get_signal_handler_name(address handler,
4507                                            char* buf, int buflen) {
4508   int offset = 0;
4509   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4510   if (found) {
4511     // skip directory names
4512     const char *p1, *p2;
4513     p1 = buf;
4514     size_t len = strlen(os::file_separator());
4515     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4516     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4517   } else {
4518     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4519   }
4520   return buf;
4521 }
4522 
4523 static void print_signal_handler(outputStream* st, int sig,
4524                                  char* buf, size_t buflen) {
4525   struct sigaction sa;
4526 
4527   sigaction(sig, NULL, &sa);
4528 
4529   // See comment for SIGNIFICANT_SIGNAL_MASK define
4530   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4531 
4532   st->print("%s: ", os::exception_name(sig, buf, buflen));
4533 
4534   address handler = (sa.sa_flags & SA_SIGINFO)
4535     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4536     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4537 
4538   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4539     st->print("SIG_DFL");
4540   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4541     st->print("SIG_IGN");
4542   } else {
4543     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4544   }
4545 
4546   st->print(", sa_mask[0]=");
4547   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4548 
4549   address rh = VMError::get_resetted_sighandler(sig);
4550   // May be, handler was resetted by VMError?
4551   if (rh != NULL) {
4552     handler = rh;
4553     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4554   }
4555 
4556   st->print(", sa_flags=");
4557   os::Posix::print_sa_flags(st, sa.sa_flags);
4558 
4559   // Check: is it our handler?
4560   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4561       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4562     // It is our signal handler
4563     // check for flags, reset system-used one!
4564     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4565       st->print(
4566                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4567                 os::Linux::get_our_sigflags(sig));
4568     }
4569   }
4570   st->cr();
4571 }
4572 
4573 
4574 #define DO_SIGNAL_CHECK(sig)                      \
4575   do {                                            \
4576     if (!sigismember(&check_signal_done, sig)) {  \
4577       os::Linux::check_signal_handler(sig);       \
4578     }                                             \
4579   } while (0)
4580 
4581 // This method is a periodic task to check for misbehaving JNI applications
4582 // under CheckJNI, we can add any periodic checks here
4583 
4584 void os::run_periodic_checks() {
4585   if (check_signals == false) return;
4586 
4587   // SEGV and BUS if overridden could potentially prevent
4588   // generation of hs*.log in the event of a crash, debugging
4589   // such a case can be very challenging, so we absolutely
4590   // check the following for a good measure:
4591   DO_SIGNAL_CHECK(SIGSEGV);
4592   DO_SIGNAL_CHECK(SIGILL);
4593   DO_SIGNAL_CHECK(SIGFPE);
4594   DO_SIGNAL_CHECK(SIGBUS);
4595   DO_SIGNAL_CHECK(SIGPIPE);
4596   DO_SIGNAL_CHECK(SIGXFSZ);
4597 #if defined(PPC64)
4598   DO_SIGNAL_CHECK(SIGTRAP);
4599 #endif
4600 
4601   // ReduceSignalUsage allows the user to override these handlers
4602   // see comments at the very top and jvm_solaris.h
4603   if (!ReduceSignalUsage) {
4604     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4605     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4606     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4607     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4608   }
4609 
4610   DO_SIGNAL_CHECK(SR_signum);
4611 }
4612 
4613 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4614 
4615 static os_sigaction_t os_sigaction = NULL;
4616 
4617 void os::Linux::check_signal_handler(int sig) {
4618   char buf[O_BUFLEN];
4619   address jvmHandler = NULL;
4620 
4621 
4622   struct sigaction act;
4623   if (os_sigaction == NULL) {
4624     // only trust the default sigaction, in case it has been interposed
4625     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4626     if (os_sigaction == NULL) return;
4627   }
4628 
4629   os_sigaction(sig, (struct sigaction*)NULL, &act);
4630 
4631 
4632   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4633 
4634   address thisHandler = (act.sa_flags & SA_SIGINFO)
4635     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4636     : CAST_FROM_FN_PTR(address, act.sa_handler);
4637 
4638 
4639   switch (sig) {
4640   case SIGSEGV:
4641   case SIGBUS:
4642   case SIGFPE:
4643   case SIGPIPE:
4644   case SIGILL:
4645   case SIGXFSZ:
4646     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4647     break;
4648 
4649   case SHUTDOWN1_SIGNAL:
4650   case SHUTDOWN2_SIGNAL:
4651   case SHUTDOWN3_SIGNAL:
4652   case BREAK_SIGNAL:
4653     jvmHandler = (address)user_handler();
4654     break;
4655 
4656   default:
4657     if (sig == SR_signum) {
4658       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4659     } else {
4660       return;
4661     }
4662     break;
4663   }
4664 
4665   if (thisHandler != jvmHandler) {
4666     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4667     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4668     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4669     // No need to check this sig any longer
4670     sigaddset(&check_signal_done, sig);
4671     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4672     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4673       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4674                     exception_name(sig, buf, O_BUFLEN));
4675     }
4676   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4677     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4678     tty->print("expected:");
4679     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4680     tty->cr();
4681     tty->print("  found:");
4682     os::Posix::print_sa_flags(tty, act.sa_flags);
4683     tty->cr();
4684     // No need to check this sig any longer
4685     sigaddset(&check_signal_done, sig);
4686   }
4687 
4688   // Dump all the signal
4689   if (sigismember(&check_signal_done, sig)) {
4690     print_signal_handlers(tty, buf, O_BUFLEN);
4691   }
4692 }
4693 
4694 extern void report_error(char* file_name, int line_no, char* title,
4695                          char* format, ...);
4696 
4697 // this is called _before_ the most of global arguments have been parsed
4698 void os::init(void) {
4699   char dummy;   // used to get a guess on initial stack address
4700 //  first_hrtime = gethrtime();
4701 
4702   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4703 
4704   init_random(1234567);
4705 
4706   ThreadCritical::initialize();
4707 
4708   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4709   if (Linux::page_size() == -1) {
4710     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4711           os::strerror(errno));
4712   }
4713   init_page_sizes((size_t) Linux::page_size());
4714 
4715   Linux::initialize_system_info();
4716 
4717   Linux::initialize_os_info();
4718 
4719   // main_thread points to the aboriginal thread
4720   Linux::_main_thread = pthread_self();
4721 
4722   Linux::clock_init();
4723   initial_time_count = javaTimeNanos();
4724 
4725   // pthread_condattr initialization for monotonic clock
4726   int status;
4727   pthread_condattr_t* _condattr = os::Linux::condAttr();
4728   if ((status = pthread_condattr_init(_condattr)) != 0) {
4729     fatal("pthread_condattr_init: %s", os::strerror(status));
4730   }
4731   // Only set the clock if CLOCK_MONOTONIC is available
4732   if (os::supports_monotonic_clock()) {
4733     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4734       if (status == EINVAL) {
4735         warning("Unable to use monotonic clock with relative timed-waits" \
4736                 " - changes to the time-of-day clock may have adverse affects");
4737       } else {
4738         fatal("pthread_condattr_setclock: %s", os::strerror(status));
4739       }
4740     }
4741   }
4742   // else it defaults to CLOCK_REALTIME
4743 
4744   // retrieve entry point for pthread_setname_np
4745   Linux::_pthread_setname_np =
4746     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4747 
4748 }
4749 
4750 // To install functions for atexit system call
4751 extern "C" {
4752   static void perfMemory_exit_helper() {
4753     perfMemory_exit();
4754   }
4755 }
4756 
4757 // this is called _after_ the global arguments have been parsed
4758 jint os::init_2(void) {
4759   Linux::fast_thread_clock_init();
4760 
4761   // Allocate a single page and mark it as readable for safepoint polling
4762   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4763   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4764 
4765   os::set_polling_page(polling_page);
4766   log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4767 
4768   if (!UseMembar) {
4769     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4770     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4771     os::set_memory_serialize_page(mem_serialize_page);
4772     log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4773   }
4774 
4775   // initialize suspend/resume support - must do this before signal_sets_init()
4776   if (SR_initialize() != 0) {
4777     perror("SR_initialize failed");
4778     return JNI_ERR;
4779   }
4780 
4781   Linux::signal_sets_init();
4782   Linux::install_signal_handlers();
4783 
4784   // Check minimum allowable stack size for thread creation and to initialize
4785   // the java system classes, including StackOverflowError - depends on page
4786   // size.  Add a page for compiler2 recursion in main thread.
4787   // Add in 2*BytesPerWord times page size to account for VM stack during
4788   // class initialization depending on 32 or 64 bit VM.
4789   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4790                                       JavaThread::stack_guard_zone_size() +
4791                                       JavaThread::stack_shadow_zone_size() +
4792                                       (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4793 
4794   size_t threadStackSizeInBytes = ThreadStackSize * K;
4795   if (threadStackSizeInBytes != 0 &&
4796       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4797     tty->print_cr("\nThe stack size specified is too small, "
4798                   "Specify at least " SIZE_FORMAT "k",
4799                   os::Linux::min_stack_allowed/ K);
4800     return JNI_ERR;
4801   }
4802 
4803   // Make the stack size a multiple of the page size so that
4804   // the yellow/red zones can be guarded.
4805   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4806                                                 vm_page_size()));
4807 
4808   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4809 
4810 #if defined(IA32)
4811   workaround_expand_exec_shield_cs_limit();
4812 #endif
4813 
4814   Linux::libpthread_init();
4815   log_info(os)("HotSpot is running with %s, %s",
4816                Linux::glibc_version(), Linux::libpthread_version());
4817 
4818   if (UseNUMA) {
4819     if (!Linux::libnuma_init()) {
4820       UseNUMA = false;
4821     } else {
4822       if ((Linux::numa_max_node() < 1)) {
4823         // There's only one node(they start from 0), disable NUMA.
4824         UseNUMA = false;
4825       }
4826     }
4827     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4828     // we can make the adaptive lgrp chunk resizing work. If the user specified
4829     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4830     // disable adaptive resizing.
4831     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4832       if (FLAG_IS_DEFAULT(UseNUMA)) {
4833         UseNUMA = false;
4834       } else {
4835         if (FLAG_IS_DEFAULT(UseLargePages) &&
4836             FLAG_IS_DEFAULT(UseSHM) &&
4837             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4838           UseLargePages = false;
4839         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4840           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4841           UseAdaptiveSizePolicy = false;
4842           UseAdaptiveNUMAChunkSizing = false;
4843         }
4844       }
4845     }
4846     if (!UseNUMA && ForceNUMA) {
4847       UseNUMA = true;
4848     }
4849   }
4850 
4851   if (MaxFDLimit) {
4852     // set the number of file descriptors to max. print out error
4853     // if getrlimit/setrlimit fails but continue regardless.
4854     struct rlimit nbr_files;
4855     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4856     if (status != 0) {
4857       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4858     } else {
4859       nbr_files.rlim_cur = nbr_files.rlim_max;
4860       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4861       if (status != 0) {
4862         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4863       }
4864     }
4865   }
4866 
4867   // Initialize lock used to serialize thread creation (see os::create_thread)
4868   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4869 
4870   // at-exit methods are called in the reverse order of their registration.
4871   // atexit functions are called on return from main or as a result of a
4872   // call to exit(3C). There can be only 32 of these functions registered
4873   // and atexit() does not set errno.
4874 
4875   if (PerfAllowAtExitRegistration) {
4876     // only register atexit functions if PerfAllowAtExitRegistration is set.
4877     // atexit functions can be delayed until process exit time, which
4878     // can be problematic for embedded VM situations. Embedded VMs should
4879     // call DestroyJavaVM() to assure that VM resources are released.
4880 
4881     // note: perfMemory_exit_helper atexit function may be removed in
4882     // the future if the appropriate cleanup code can be added to the
4883     // VM_Exit VMOperation's doit method.
4884     if (atexit(perfMemory_exit_helper) != 0) {
4885       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4886     }
4887   }
4888 
4889   // initialize thread priority policy
4890   prio_init();
4891 
4892   return JNI_OK;
4893 }
4894 
4895 // Mark the polling page as unreadable
4896 void os::make_polling_page_unreadable(void) {
4897   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4898     fatal("Could not disable polling page");
4899   }
4900 }
4901 
4902 // Mark the polling page as readable
4903 void os::make_polling_page_readable(void) {
4904   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4905     fatal("Could not enable polling page");
4906   }
4907 }
4908 
4909 // older glibc versions don't have this macro (which expands to
4910 // an optimized bit-counting function) so we have to roll our own
4911 #ifndef CPU_COUNT
4912 
4913 static int _cpu_count(const cpu_set_t* cpus) {
4914   int count = 0;
4915   // only look up to the number of configured processors
4916   for (int i = 0; i < os::processor_count(); i++) {
4917     if (CPU_ISSET(i, cpus)) {
4918       count++;
4919     }
4920   }
4921   return count;
4922 }
4923 
4924 #define CPU_COUNT(cpus) _cpu_count(cpus)
4925 
4926 #endif // CPU_COUNT
4927 
4928 // Get the current number of available processors for this process.
4929 // This value can change at any time during a process's lifetime.
4930 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4931 // If it appears there may be more than 1024 processors then we do a
4932 // dynamic check - see 6515172 for details.
4933 // If anything goes wrong we fallback to returning the number of online
4934 // processors - which can be greater than the number available to the process.
4935 int os::active_processor_count() {
4936   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
4937   cpu_set_t* cpus_p = &cpus;
4938   int cpus_size = sizeof(cpu_set_t);
4939 
4940   int configured_cpus = processor_count();  // upper bound on available cpus
4941   int cpu_count = 0;
4942 
4943 // old build platforms may not support dynamic cpu sets
4944 #ifdef CPU_ALLOC
4945 
4946   // To enable easy testing of the dynamic path on different platforms we
4947   // introduce a diagnostic flag: UseCpuAllocPath
4948   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4949     // kernel may use a mask bigger than cpu_set_t
4950     log_trace(os)("active_processor_count: using dynamic path %s"
4951                   "- configured processors: %d",
4952                   UseCpuAllocPath ? "(forced) " : "",
4953                   configured_cpus);
4954     cpus_p = CPU_ALLOC(configured_cpus);
4955     if (cpus_p != NULL) {
4956       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4957       // zero it just to be safe
4958       CPU_ZERO_S(cpus_size, cpus_p);
4959     }
4960     else {
4961        // failed to allocate so fallback to online cpus
4962        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4963        log_trace(os)("active_processor_count: "
4964                      "CPU_ALLOC failed (%s) - using "
4965                      "online processor count: %d",
4966                      os::strerror(errno), online_cpus);
4967        return online_cpus;
4968     }
4969   }
4970   else {
4971     log_trace(os)("active_processor_count: using static path - configured processors: %d",
4972                   configured_cpus);
4973   }
4974 #else // CPU_ALLOC
4975 // these stubs won't be executed
4976 #define CPU_COUNT_S(size, cpus) -1
4977 #define CPU_FREE(cpus)
4978 
4979   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4980                 configured_cpus);
4981 #endif // CPU_ALLOC
4982 
4983   // pid 0 means the current thread - which we have to assume represents the process
4984   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4985     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4986       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4987     }
4988     else {
4989       cpu_count = CPU_COUNT(cpus_p);
4990     }
4991     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4992   }
4993   else {
4994     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4995     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4996             "which may exceed available processors", os::strerror(errno), cpu_count);
4997   }
4998 
4999   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5000     CPU_FREE(cpus_p);
5001   }
5002 
5003   assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
5004   return cpu_count;
5005 }
5006 
5007 void os::set_native_thread_name(const char *name) {
5008   if (Linux::_pthread_setname_np) {
5009     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5010     snprintf(buf, sizeof(buf), "%s", name);
5011     buf[sizeof(buf) - 1] = '\0';
5012     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5013     // ERANGE should not happen; all other errors should just be ignored.
5014     assert(rc != ERANGE, "pthread_setname_np failed");
5015   }
5016 }
5017 
5018 bool os::distribute_processes(uint length, uint* distribution) {
5019   // Not yet implemented.
5020   return false;
5021 }
5022 
5023 bool os::bind_to_processor(uint processor_id) {
5024   // Not yet implemented.
5025   return false;
5026 }
5027 
5028 ///
5029 
5030 void os::SuspendedThreadTask::internal_do_task() {
5031   if (do_suspend(_thread->osthread())) {
5032     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5033     do_task(context);
5034     do_resume(_thread->osthread());
5035   }
5036 }
5037 
5038 class PcFetcher : public os::SuspendedThreadTask {
5039  public:
5040   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5041   ExtendedPC result();
5042  protected:
5043   void do_task(const os::SuspendedThreadTaskContext& context);
5044  private:
5045   ExtendedPC _epc;
5046 };
5047 
5048 ExtendedPC PcFetcher::result() {
5049   guarantee(is_done(), "task is not done yet.");
5050   return _epc;
5051 }
5052 
5053 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5054   Thread* thread = context.thread();
5055   OSThread* osthread = thread->osthread();
5056   if (osthread->ucontext() != NULL) {
5057     _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5058   } else {
5059     // NULL context is unexpected, double-check this is the VMThread
5060     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5061   }
5062 }
5063 
5064 // Suspends the target using the signal mechanism and then grabs the PC before
5065 // resuming the target. Used by the flat-profiler only
5066 ExtendedPC os::get_thread_pc(Thread* thread) {
5067   // Make sure that it is called by the watcher for the VMThread
5068   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5069   assert(thread->is_VM_thread(), "Can only be called for VMThread");
5070 
5071   PcFetcher fetcher(thread);
5072   fetcher.run();
5073   return fetcher.result();
5074 }
5075 
5076 ////////////////////////////////////////////////////////////////////////////////
5077 // debug support
5078 
5079 bool os::find(address addr, outputStream* st) {
5080   Dl_info dlinfo;
5081   memset(&dlinfo, 0, sizeof(dlinfo));
5082   if (dladdr(addr, &dlinfo) != 0) {
5083     st->print(PTR_FORMAT ": ", p2i(addr));
5084     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5085       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5086                 p2i(addr) - p2i(dlinfo.dli_saddr));
5087     } else if (dlinfo.dli_fbase != NULL) {
5088       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5089     } else {
5090       st->print("<absolute address>");
5091     }
5092     if (dlinfo.dli_fname != NULL) {
5093       st->print(" in %s", dlinfo.dli_fname);
5094     }
5095     if (dlinfo.dli_fbase != NULL) {
5096       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5097     }
5098     st->cr();
5099 
5100     if (Verbose) {
5101       // decode some bytes around the PC
5102       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5103       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5104       address       lowest = (address) dlinfo.dli_sname;
5105       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5106       if (begin < lowest)  begin = lowest;
5107       Dl_info dlinfo2;
5108       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5109           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5110         end = (address) dlinfo2.dli_saddr;
5111       }
5112       Disassembler::decode(begin, end, st);
5113     }
5114     return true;
5115   }
5116   return false;
5117 }
5118 
5119 ////////////////////////////////////////////////////////////////////////////////
5120 // misc
5121 
5122 // This does not do anything on Linux. This is basically a hook for being
5123 // able to use structured exception handling (thread-local exception filters)
5124 // on, e.g., Win32.
5125 void
5126 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5127                          JavaCallArguments* args, Thread* thread) {
5128   f(value, method, args, thread);
5129 }
5130 
5131 void os::print_statistics() {
5132 }
5133 
5134 bool os::message_box(const char* title, const char* message) {
5135   int i;
5136   fdStream err(defaultStream::error_fd());
5137   for (i = 0; i < 78; i++) err.print_raw("=");
5138   err.cr();
5139   err.print_raw_cr(title);
5140   for (i = 0; i < 78; i++) err.print_raw("-");
5141   err.cr();
5142   err.print_raw_cr(message);
5143   for (i = 0; i < 78; i++) err.print_raw("=");
5144   err.cr();
5145 
5146   char buf[16];
5147   // Prevent process from exiting upon "read error" without consuming all CPU
5148   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5149 
5150   return buf[0] == 'y' || buf[0] == 'Y';
5151 }
5152 
5153 int os::stat(const char *path, struct stat *sbuf) {
5154   char pathbuf[MAX_PATH];
5155   if (strlen(path) > MAX_PATH - 1) {
5156     errno = ENAMETOOLONG;
5157     return -1;
5158   }
5159   os::native_path(strcpy(pathbuf, path));
5160   return ::stat(pathbuf, sbuf);
5161 }
5162 
5163 bool os::check_heap(bool force) {
5164   return true;
5165 }
5166 
5167 // Is a (classpath) directory empty?
5168 bool os::dir_is_empty(const char* path) {
5169   DIR *dir = NULL;
5170   struct dirent *ptr;
5171 
5172   dir = opendir(path);
5173   if (dir == NULL) return true;
5174 
5175   // Scan the directory
5176   bool result = true;
5177   char buf[sizeof(struct dirent) + MAX_PATH];
5178   while (result && (ptr = ::readdir(dir)) != NULL) {
5179     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5180       result = false;
5181     }
5182   }
5183   closedir(dir);
5184   return result;
5185 }
5186 
5187 // This code originates from JDK's sysOpen and open64_w
5188 // from src/solaris/hpi/src/system_md.c
5189 
5190 int os::open(const char *path, int oflag, int mode) {
5191   if (strlen(path) > MAX_PATH - 1) {
5192     errno = ENAMETOOLONG;
5193     return -1;
5194   }
5195 
5196   // All file descriptors that are opened in the Java process and not
5197   // specifically destined for a subprocess should have the close-on-exec
5198   // flag set.  If we don't set it, then careless 3rd party native code
5199   // might fork and exec without closing all appropriate file descriptors
5200   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5201   // turn might:
5202   //
5203   // - cause end-of-file to fail to be detected on some file
5204   //   descriptors, resulting in mysterious hangs, or
5205   //
5206   // - might cause an fopen in the subprocess to fail on a system
5207   //   suffering from bug 1085341.
5208   //
5209   // (Yes, the default setting of the close-on-exec flag is a Unix
5210   // design flaw)
5211   //
5212   // See:
5213   // 1085341: 32-bit stdio routines should support file descriptors >255
5214   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5215   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5216   //
5217   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5218   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5219   // because it saves a system call and removes a small window where the flag
5220   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5221   // and we fall back to using FD_CLOEXEC (see below).
5222 #ifdef O_CLOEXEC
5223   oflag |= O_CLOEXEC;
5224 #endif
5225 
5226   int fd = ::open64(path, oflag, mode);
5227   if (fd == -1) return -1;
5228 
5229   //If the open succeeded, the file might still be a directory
5230   {
5231     struct stat64 buf64;
5232     int ret = ::fstat64(fd, &buf64);
5233     int st_mode = buf64.st_mode;
5234 
5235     if (ret != -1) {
5236       if ((st_mode & S_IFMT) == S_IFDIR) {
5237         errno = EISDIR;
5238         ::close(fd);
5239         return -1;
5240       }
5241     } else {
5242       ::close(fd);
5243       return -1;
5244     }
5245   }
5246 
5247 #ifdef FD_CLOEXEC
5248   // Validate that the use of the O_CLOEXEC flag on open above worked.
5249   // With recent kernels, we will perform this check exactly once.
5250   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5251   if (!O_CLOEXEC_is_known_to_work) {
5252     int flags = ::fcntl(fd, F_GETFD);
5253     if (flags != -1) {
5254       if ((flags & FD_CLOEXEC) != 0)
5255         O_CLOEXEC_is_known_to_work = 1;
5256       else
5257         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5258     }
5259   }
5260 #endif
5261 
5262   return fd;
5263 }
5264 
5265 
5266 // create binary file, rewriting existing file if required
5267 int os::create_binary_file(const char* path, bool rewrite_existing) {
5268   int oflags = O_WRONLY | O_CREAT;
5269   if (!rewrite_existing) {
5270     oflags |= O_EXCL;
5271   }
5272   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5273 }
5274 
5275 // return current position of file pointer
5276 jlong os::current_file_offset(int fd) {
5277   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5278 }
5279 
5280 // move file pointer to the specified offset
5281 jlong os::seek_to_file_offset(int fd, jlong offset) {
5282   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5283 }
5284 
5285 // This code originates from JDK's sysAvailable
5286 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5287 
5288 int os::available(int fd, jlong *bytes) {
5289   jlong cur, end;
5290   int mode;
5291   struct stat64 buf64;
5292 
5293   if (::fstat64(fd, &buf64) >= 0) {
5294     mode = buf64.st_mode;
5295     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5296       int n;
5297       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5298         *bytes = n;
5299         return 1;
5300       }
5301     }
5302   }
5303   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5304     return 0;
5305   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5306     return 0;
5307   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5308     return 0;
5309   }
5310   *bytes = end - cur;
5311   return 1;
5312 }
5313 
5314 // Map a block of memory.
5315 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5316                         char *addr, size_t bytes, bool read_only,
5317                         bool allow_exec) {
5318   int prot;
5319   int flags = MAP_PRIVATE;
5320 
5321   if (read_only) {
5322     prot = PROT_READ;
5323   } else {
5324     prot = PROT_READ | PROT_WRITE;
5325   }
5326 
5327   if (allow_exec) {
5328     prot |= PROT_EXEC;
5329   }
5330 
5331   if (addr != NULL) {
5332     flags |= MAP_FIXED;
5333   }
5334 
5335   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5336                                      fd, file_offset);
5337   if (mapped_address == MAP_FAILED) {
5338     return NULL;
5339   }
5340   return mapped_address;
5341 }
5342 
5343 
5344 // Remap a block of memory.
5345 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5346                           char *addr, size_t bytes, bool read_only,
5347                           bool allow_exec) {
5348   // same as map_memory() on this OS
5349   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5350                         allow_exec);
5351 }
5352 
5353 
5354 // Unmap a block of memory.
5355 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5356   return munmap(addr, bytes) == 0;
5357 }
5358 
5359 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5360 
5361 static clockid_t thread_cpu_clockid(Thread* thread) {
5362   pthread_t tid = thread->osthread()->pthread_id();
5363   clockid_t clockid;
5364 
5365   // Get thread clockid
5366   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5367   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5368   return clockid;
5369 }
5370 
5371 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5372 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5373 // of a thread.
5374 //
5375 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5376 // the fast estimate available on the platform.
5377 
5378 jlong os::current_thread_cpu_time() {
5379   if (os::Linux::supports_fast_thread_cpu_time()) {
5380     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5381   } else {
5382     // return user + sys since the cost is the same
5383     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5384   }
5385 }
5386 
5387 jlong os::thread_cpu_time(Thread* thread) {
5388   // consistent with what current_thread_cpu_time() returns
5389   if (os::Linux::supports_fast_thread_cpu_time()) {
5390     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5391   } else {
5392     return slow_thread_cpu_time(thread, true /* user + sys */);
5393   }
5394 }
5395 
5396 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5397   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5398     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5399   } else {
5400     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5401   }
5402 }
5403 
5404 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5405   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5406     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5407   } else {
5408     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5409   }
5410 }
5411 
5412 //  -1 on error.
5413 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5414   pid_t  tid = thread->osthread()->thread_id();
5415   char *s;
5416   char stat[2048];
5417   int statlen;
5418   char proc_name[64];
5419   int count;
5420   long sys_time, user_time;
5421   char cdummy;
5422   int idummy;
5423   long ldummy;
5424   FILE *fp;
5425 
5426   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5427   fp = fopen(proc_name, "r");
5428   if (fp == NULL) return -1;
5429   statlen = fread(stat, 1, 2047, fp);
5430   stat[statlen] = '\0';
5431   fclose(fp);
5432 
5433   // Skip pid and the command string. Note that we could be dealing with
5434   // weird command names, e.g. user could decide to rename java launcher
5435   // to "java 1.4.2 :)", then the stat file would look like
5436   //                1234 (java 1.4.2 :)) R ... ...
5437   // We don't really need to know the command string, just find the last
5438   // occurrence of ")" and then start parsing from there. See bug 4726580.
5439   s = strrchr(stat, ')');
5440   if (s == NULL) return -1;
5441 
5442   // Skip blank chars
5443   do { s++; } while (s && isspace(*s));
5444 
5445   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5446                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5447                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5448                  &user_time, &sys_time);
5449   if (count != 13) return -1;
5450   if (user_sys_cpu_time) {
5451     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5452   } else {
5453     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5454   }
5455 }
5456 
5457 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5458   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5459   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5460   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5461   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5462 }
5463 
5464 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5465   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5466   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5467   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5468   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5469 }
5470 
5471 bool os::is_thread_cpu_time_supported() {
5472   return true;
5473 }
5474 
5475 // System loadavg support.  Returns -1 if load average cannot be obtained.
5476 // Linux doesn't yet have a (official) notion of processor sets,
5477 // so just return the system wide load average.
5478 int os::loadavg(double loadavg[], int nelem) {
5479   return ::getloadavg(loadavg, nelem);
5480 }
5481 
5482 void os::pause() {
5483   char filename[MAX_PATH];
5484   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5485     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5486   } else {
5487     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5488   }
5489 
5490   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5491   if (fd != -1) {
5492     struct stat buf;
5493     ::close(fd);
5494     while (::stat(filename, &buf) == 0) {
5495       (void)::poll(NULL, 0, 100);
5496     }
5497   } else {
5498     jio_fprintf(stderr,
5499                 "Could not open pause file '%s', continuing immediately.\n", filename);
5500   }
5501 }
5502 
5503 
5504 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5505 // comment paragraphs are worth repeating here:
5506 //
5507 // Assumption:
5508 //    Only one parker can exist on an event, which is why we allocate
5509 //    them per-thread. Multiple unparkers can coexist.
5510 //
5511 // _Event serves as a restricted-range semaphore.
5512 //   -1 : thread is blocked, i.e. there is a waiter
5513 //    0 : neutral: thread is running or ready,
5514 //        could have been signaled after a wait started
5515 //    1 : signaled - thread is running or ready
5516 //
5517 
5518 // utility to compute the abstime argument to timedwait:
5519 // millis is the relative timeout time
5520 // abstime will be the absolute timeout time
5521 // TODO: replace compute_abstime() with unpackTime()
5522 
5523 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5524   if (millis < 0)  millis = 0;
5525 
5526   jlong seconds = millis / 1000;
5527   millis %= 1000;
5528   if (seconds > 50000000) { // see man cond_timedwait(3T)
5529     seconds = 50000000;
5530   }
5531 
5532   if (os::supports_monotonic_clock()) {
5533     struct timespec now;
5534     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5535     assert_status(status == 0, status, "clock_gettime");
5536     abstime->tv_sec = now.tv_sec  + seconds;
5537     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5538     if (nanos >= NANOSECS_PER_SEC) {
5539       abstime->tv_sec += 1;
5540       nanos -= NANOSECS_PER_SEC;
5541     }
5542     abstime->tv_nsec = nanos;
5543   } else {
5544     struct timeval now;
5545     int status = gettimeofday(&now, NULL);
5546     assert(status == 0, "gettimeofday");
5547     abstime->tv_sec = now.tv_sec  + seconds;
5548     long usec = now.tv_usec + millis * 1000;
5549     if (usec >= 1000000) {
5550       abstime->tv_sec += 1;
5551       usec -= 1000000;
5552     }
5553     abstime->tv_nsec = usec * 1000;
5554   }
5555   return abstime;
5556 }
5557 
5558 void os::PlatformEvent::park() {       // AKA "down()"
5559   // Transitions for _Event:
5560   //   -1 => -1 : illegal
5561   //    1 =>  0 : pass - return immediately
5562   //    0 => -1 : block; then set _Event to 0 before returning
5563 
5564   // Invariant: Only the thread associated with the Event/PlatformEvent
5565   // may call park().
5566   // TODO: assert that _Assoc != NULL or _Assoc == Self
5567   assert(_nParked == 0, "invariant");
5568 
5569   int v;
5570   for (;;) {
5571     v = _Event;
5572     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5573   }
5574   guarantee(v >= 0, "invariant");
5575   if (v == 0) {
5576     // Do this the hard way by blocking ...
5577     int status = pthread_mutex_lock(_mutex);
5578     assert_status(status == 0, status, "mutex_lock");
5579     guarantee(_nParked == 0, "invariant");
5580     ++_nParked;
5581     while (_Event < 0) {
5582       status = pthread_cond_wait(_cond, _mutex);
5583       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5584       // Treat this the same as if the wait was interrupted
5585       if (status == ETIME) { status = EINTR; }
5586       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5587     }
5588     --_nParked;
5589 
5590     _Event = 0;
5591     status = pthread_mutex_unlock(_mutex);
5592     assert_status(status == 0, status, "mutex_unlock");
5593     // Paranoia to ensure our locked and lock-free paths interact
5594     // correctly with each other.
5595     OrderAccess::fence();
5596   }
5597   guarantee(_Event >= 0, "invariant");
5598 }
5599 
5600 int os::PlatformEvent::park(jlong millis) {
5601   // Transitions for _Event:
5602   //   -1 => -1 : illegal
5603   //    1 =>  0 : pass - return immediately
5604   //    0 => -1 : block; then set _Event to 0 before returning
5605 
5606   guarantee(_nParked == 0, "invariant");
5607 
5608   int v;
5609   for (;;) {
5610     v = _Event;
5611     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5612   }
5613   guarantee(v >= 0, "invariant");
5614   if (v != 0) return OS_OK;
5615 
5616   // We do this the hard way, by blocking the thread.
5617   // Consider enforcing a minimum timeout value.
5618   struct timespec abst;
5619   compute_abstime(&abst, millis);
5620 
5621   int ret = OS_TIMEOUT;
5622   int status = pthread_mutex_lock(_mutex);
5623   assert_status(status == 0, status, "mutex_lock");
5624   guarantee(_nParked == 0, "invariant");
5625   ++_nParked;
5626 
5627   // Object.wait(timo) will return because of
5628   // (a) notification
5629   // (b) timeout
5630   // (c) thread.interrupt
5631   //
5632   // Thread.interrupt and object.notify{All} both call Event::set.
5633   // That is, we treat thread.interrupt as a special case of notification.
5634   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5635   // We assume all ETIME returns are valid.
5636   //
5637   // TODO: properly differentiate simultaneous notify+interrupt.
5638   // In that case, we should propagate the notify to another waiter.
5639 
5640   while (_Event < 0) {
5641     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5642     assert_status(status == 0 || status == EINTR ||
5643                   status == ETIME || status == ETIMEDOUT,
5644                   status, "cond_timedwait");
5645     if (!FilterSpuriousWakeups) break;                 // previous semantics
5646     if (status == ETIME || status == ETIMEDOUT) break;
5647     // We consume and ignore EINTR and spurious wakeups.
5648   }
5649   --_nParked;
5650   if (_Event >= 0) {
5651     ret = OS_OK;
5652   }
5653   _Event = 0;
5654   status = pthread_mutex_unlock(_mutex);
5655   assert_status(status == 0, status, "mutex_unlock");
5656   assert(_nParked == 0, "invariant");
5657   // Paranoia to ensure our locked and lock-free paths interact
5658   // correctly with each other.
5659   OrderAccess::fence();
5660   return ret;
5661 }
5662 
5663 void os::PlatformEvent::unpark() {
5664   // Transitions for _Event:
5665   //    0 => 1 : just return
5666   //    1 => 1 : just return
5667   //   -1 => either 0 or 1; must signal target thread
5668   //         That is, we can safely transition _Event from -1 to either
5669   //         0 or 1.
5670   // See also: "Semaphores in Plan 9" by Mullender & Cox
5671   //
5672   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5673   // that it will take two back-to-back park() calls for the owning
5674   // thread to block. This has the benefit of forcing a spurious return
5675   // from the first park() call after an unpark() call which will help
5676   // shake out uses of park() and unpark() without condition variables.
5677 
5678   if (Atomic::xchg(1, &_Event) >= 0) return;
5679 
5680   // Wait for the thread associated with the event to vacate
5681   int status = pthread_mutex_lock(_mutex);
5682   assert_status(status == 0, status, "mutex_lock");
5683   int AnyWaiters = _nParked;
5684   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5685   status = pthread_mutex_unlock(_mutex);
5686   assert_status(status == 0, status, "mutex_unlock");
5687   if (AnyWaiters != 0) {
5688     // Note that we signal() *after* dropping the lock for "immortal" Events.
5689     // This is safe and avoids a common class of  futile wakeups.  In rare
5690     // circumstances this can cause a thread to return prematurely from
5691     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5692     // will simply re-test the condition and re-park itself.
5693     // This provides particular benefit if the underlying platform does not
5694     // provide wait morphing.
5695     status = pthread_cond_signal(_cond);
5696     assert_status(status == 0, status, "cond_signal");
5697   }
5698 }
5699 
5700 
5701 // JSR166
5702 // -------------------------------------------------------
5703 
5704 // The solaris and linux implementations of park/unpark are fairly
5705 // conservative for now, but can be improved. They currently use a
5706 // mutex/condvar pair, plus a a count.
5707 // Park decrements count if > 0, else does a condvar wait.  Unpark
5708 // sets count to 1 and signals condvar.  Only one thread ever waits
5709 // on the condvar. Contention seen when trying to park implies that someone
5710 // is unparking you, so don't wait. And spurious returns are fine, so there
5711 // is no need to track notifications.
5712 
5713 // This code is common to linux and solaris and will be moved to a
5714 // common place in dolphin.
5715 //
5716 // The passed in time value is either a relative time in nanoseconds
5717 // or an absolute time in milliseconds. Either way it has to be unpacked
5718 // into suitable seconds and nanoseconds components and stored in the
5719 // given timespec structure.
5720 // Given time is a 64-bit value and the time_t used in the timespec is only
5721 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5722 // overflow if times way in the future are given. Further on Solaris versions
5723 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5724 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5725 // As it will be 28 years before "now + 100000000" will overflow we can
5726 // ignore overflow and just impose a hard-limit on seconds using the value
5727 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5728 // years from "now".
5729 
5730 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5731   assert(time > 0, "convertTime");
5732   time_t max_secs = 0;
5733 
5734   if (!os::supports_monotonic_clock() || isAbsolute) {
5735     struct timeval now;
5736     int status = gettimeofday(&now, NULL);
5737     assert(status == 0, "gettimeofday");
5738 
5739     max_secs = now.tv_sec + MAX_SECS;
5740 
5741     if (isAbsolute) {
5742       jlong secs = time / 1000;
5743       if (secs > max_secs) {
5744         absTime->tv_sec = max_secs;
5745       } else {
5746         absTime->tv_sec = secs;
5747       }
5748       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5749     } else {
5750       jlong secs = time / NANOSECS_PER_SEC;
5751       if (secs >= MAX_SECS) {
5752         absTime->tv_sec = max_secs;
5753         absTime->tv_nsec = 0;
5754       } else {
5755         absTime->tv_sec = now.tv_sec + secs;
5756         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5757         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5758           absTime->tv_nsec -= NANOSECS_PER_SEC;
5759           ++absTime->tv_sec; // note: this must be <= max_secs
5760         }
5761       }
5762     }
5763   } else {
5764     // must be relative using monotonic clock
5765     struct timespec now;
5766     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5767     assert_status(status == 0, status, "clock_gettime");
5768     max_secs = now.tv_sec + MAX_SECS;
5769     jlong secs = time / NANOSECS_PER_SEC;
5770     if (secs >= MAX_SECS) {
5771       absTime->tv_sec = max_secs;
5772       absTime->tv_nsec = 0;
5773     } else {
5774       absTime->tv_sec = now.tv_sec + secs;
5775       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5776       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5777         absTime->tv_nsec -= NANOSECS_PER_SEC;
5778         ++absTime->tv_sec; // note: this must be <= max_secs
5779       }
5780     }
5781   }
5782   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5783   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5784   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5785   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5786 }
5787 
5788 void Parker::park(bool isAbsolute, jlong time) {
5789   // Ideally we'd do something useful while spinning, such
5790   // as calling unpackTime().
5791 
5792   // Optional fast-path check:
5793   // Return immediately if a permit is available.
5794   // We depend on Atomic::xchg() having full barrier semantics
5795   // since we are doing a lock-free update to _counter.
5796   if (Atomic::xchg(0, &_counter) > 0) return;
5797 
5798   Thread* thread = Thread::current();
5799   assert(thread->is_Java_thread(), "Must be JavaThread");
5800   JavaThread *jt = (JavaThread *)thread;
5801 
5802   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5803   // Check interrupt before trying to wait
5804   if (Thread::is_interrupted(thread, false)) {
5805     return;
5806   }
5807 
5808   // Next, demultiplex/decode time arguments
5809   timespec absTime;
5810   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5811     return;
5812   }
5813   if (time > 0) {
5814     unpackTime(&absTime, isAbsolute, time);
5815   }
5816 
5817 
5818   // Enter safepoint region
5819   // Beware of deadlocks such as 6317397.
5820   // The per-thread Parker:: mutex is a classic leaf-lock.
5821   // In particular a thread must never block on the Threads_lock while
5822   // holding the Parker:: mutex.  If safepoints are pending both the
5823   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5824   ThreadBlockInVM tbivm(jt);
5825 
5826   // Don't wait if cannot get lock since interference arises from
5827   // unblocking.  Also. check interrupt before trying wait
5828   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5829     return;
5830   }
5831 
5832   int status;
5833   if (_counter > 0)  { // no wait needed
5834     _counter = 0;
5835     status = pthread_mutex_unlock(_mutex);
5836     assert_status(status == 0, status, "invariant");
5837     // Paranoia to ensure our locked and lock-free paths interact
5838     // correctly with each other and Java-level accesses.
5839     OrderAccess::fence();
5840     return;
5841   }
5842 
5843 #ifdef ASSERT
5844   // Don't catch signals while blocked; let the running threads have the signals.
5845   // (This allows a debugger to break into the running thread.)
5846   sigset_t oldsigs;
5847   sigemptyset(&oldsigs);
5848   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5849   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5850 #endif
5851 
5852   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5853   jt->set_suspend_equivalent();
5854   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5855 
5856   assert(_cur_index == -1, "invariant");
5857   if (time == 0) {
5858     _cur_index = REL_INDEX; // arbitrary choice when not timed
5859     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5860   } else {
5861     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5862     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5863   }
5864   _cur_index = -1;
5865   assert_status(status == 0 || status == EINTR ||
5866                 status == ETIME || status == ETIMEDOUT,
5867                 status, "cond_timedwait");
5868 
5869 #ifdef ASSERT
5870   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5871 #endif
5872 
5873   _counter = 0;
5874   status = pthread_mutex_unlock(_mutex);
5875   assert_status(status == 0, status, "invariant");
5876   // Paranoia to ensure our locked and lock-free paths interact
5877   // correctly with each other and Java-level accesses.
5878   OrderAccess::fence();
5879 
5880   // If externally suspended while waiting, re-suspend
5881   if (jt->handle_special_suspend_equivalent_condition()) {
5882     jt->java_suspend_self();
5883   }
5884 }
5885 
5886 void Parker::unpark() {
5887   int status = pthread_mutex_lock(_mutex);
5888   assert_status(status == 0, status, "invariant");
5889   const int s = _counter;
5890   _counter = 1;
5891   // must capture correct index before unlocking
5892   int index = _cur_index;
5893   status = pthread_mutex_unlock(_mutex);
5894   assert_status(status == 0, status, "invariant");
5895   if (s < 1 && index != -1) {
5896     // thread is definitely parked
5897     status = pthread_cond_signal(&_cond[index]);
5898     assert_status(status == 0, status, "invariant");
5899   }
5900 }
5901 
5902 
5903 extern char** environ;
5904 
5905 // Run the specified command in a separate process. Return its exit value,
5906 // or -1 on failure (e.g. can't fork a new process).
5907 // Unlike system(), this function can be called from signal handler. It
5908 // doesn't block SIGINT et al.
5909 int os::fork_and_exec(char* cmd) {
5910   const char * argv[4] = {"sh", "-c", cmd, NULL};
5911 
5912   pid_t pid = fork();
5913 
5914   if (pid < 0) {
5915     // fork failed
5916     return -1;
5917 
5918   } else if (pid == 0) {
5919     // child process
5920 
5921     execve("/bin/sh", (char* const*)argv, environ);
5922 
5923     // execve failed
5924     _exit(-1);
5925 
5926   } else  {
5927     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5928     // care about the actual exit code, for now.
5929 
5930     int status;
5931 
5932     // Wait for the child process to exit.  This returns immediately if
5933     // the child has already exited. */
5934     while (waitpid(pid, &status, 0) < 0) {
5935       switch (errno) {
5936       case ECHILD: return 0;
5937       case EINTR: break;
5938       default: return -1;
5939       }
5940     }
5941 
5942     if (WIFEXITED(status)) {
5943       // The child exited normally; get its exit code.
5944       return WEXITSTATUS(status);
5945     } else if (WIFSIGNALED(status)) {
5946       // The child exited because of a signal
5947       // The best value to return is 0x80 + signal number,
5948       // because that is what all Unix shells do, and because
5949       // it allows callers to distinguish between process exit and
5950       // process death by signal.
5951       return 0x80 + WTERMSIG(status);
5952     } else {
5953       // Unknown exit code; pass it through
5954       return status;
5955     }
5956   }
5957 }
5958 
5959 // is_headless_jre()
5960 //
5961 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5962 // in order to report if we are running in a headless jre
5963 //
5964 // Since JDK8 xawt/libmawt.so was moved into the same directory
5965 // as libawt.so, and renamed libawt_xawt.so
5966 //
5967 bool os::is_headless_jre() {
5968   struct stat statbuf;
5969   char buf[MAXPATHLEN];
5970   char libmawtpath[MAXPATHLEN];
5971   const char *xawtstr  = "/xawt/libmawt.so";
5972   const char *new_xawtstr = "/libawt_xawt.so";
5973   char *p;
5974 
5975   // Get path to libjvm.so
5976   os::jvm_path(buf, sizeof(buf));
5977 
5978   // Get rid of libjvm.so
5979   p = strrchr(buf, '/');
5980   if (p == NULL) {
5981     return false;
5982   } else {
5983     *p = '\0';
5984   }
5985 
5986   // Get rid of client or server
5987   p = strrchr(buf, '/');
5988   if (p == NULL) {
5989     return false;
5990   } else {
5991     *p = '\0';
5992   }
5993 
5994   // check xawt/libmawt.so
5995   strcpy(libmawtpath, buf);
5996   strcat(libmawtpath, xawtstr);
5997   if (::stat(libmawtpath, &statbuf) == 0) return false;
5998 
5999   // check libawt_xawt.so
6000   strcpy(libmawtpath, buf);
6001   strcat(libmawtpath, new_xawtstr);
6002   if (::stat(libmawtpath, &statbuf) == 0) return false;
6003 
6004   return true;
6005 }
6006 
6007 // Get the default path to the core file
6008 // Returns the length of the string
6009 int os::get_core_path(char* buffer, size_t bufferSize) {
6010   /*
6011    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
6012    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
6013    */
6014   const int core_pattern_len = 129;
6015   char core_pattern[core_pattern_len] = {0};
6016 
6017   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6018   if (core_pattern_file == -1) {
6019     return -1;
6020   }
6021 
6022   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6023   ::close(core_pattern_file);
6024   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
6025     return -1;
6026   }
6027   if (core_pattern[ret-1] == '\n') {
6028     core_pattern[ret-1] = '\0';
6029   } else {
6030     core_pattern[ret] = '\0';
6031   }
6032 
6033   char *pid_pos = strstr(core_pattern, "%p");
6034   int written;
6035 
6036   if (core_pattern[0] == '/') {
6037     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6038   } else {
6039     char cwd[PATH_MAX];
6040 
6041     const char* p = get_current_directory(cwd, PATH_MAX);
6042     if (p == NULL) {
6043       return -1;
6044     }
6045 
6046     if (core_pattern[0] == '|') {
6047       written = jio_snprintf(buffer, bufferSize,
6048                         "\"%s\" (or dumping to %s/core.%d)",
6049                                      &core_pattern[1], p, current_process_id());
6050     } else {
6051       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6052     }
6053   }
6054 
6055   if (written < 0) {
6056     return -1;
6057   }
6058 
6059   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6060     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6061 
6062     if (core_uses_pid_file != -1) {
6063       char core_uses_pid = 0;
6064       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6065       ::close(core_uses_pid_file);
6066 
6067       if (core_uses_pid == '1') {
6068         jio_snprintf(buffer + written, bufferSize - written,
6069                                           ".%d", current_process_id());
6070       }
6071     }
6072   }
6073 
6074   return strlen(buffer);
6075 }
6076 
6077 bool os::start_debugging(char *buf, int buflen) {
6078   int len = (int)strlen(buf);
6079   char *p = &buf[len];
6080 
6081   jio_snprintf(p, buflen-len,
6082              "\n\n"
6083              "Do you want to debug the problem?\n\n"
6084              "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6085              "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6086              "Otherwise, press RETURN to abort...",
6087              os::current_process_id(), os::current_process_id(),
6088              os::current_thread_id(), os::current_thread_id());
6089 
6090   bool yes = os::message_box("Unexpected Error", buf);
6091 
6092   if (yes) {
6093     // yes, user asked VM to launch debugger
6094     jio_snprintf(buf, sizeof(buf), "gdb /proc/%d/exe %d",
6095                      os::current_process_id(), os::current_process_id());
6096 
6097     os::fork_and_exec(buf);
6098     yes = false;
6099   }
6100   return yes;
6101 }
6102 
6103 static inline struct timespec get_mtime(const char* filename) {
6104   struct stat st;
6105   int ret = os::stat(filename, &st);
6106   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6107   return st.st_mtim;
6108 }
6109 
6110 int os::compare_file_modified_times(const char* file1, const char* file2) {
6111   struct timespec filetime1 = get_mtime(file1);
6112   struct timespec filetime2 = get_mtime(file2);
6113   int diff = filetime1.tv_sec - filetime2.tv_sec;
6114   if (diff == 0) {
6115     return filetime1.tv_nsec - filetime2.tv_nsec;
6116   }
6117   return diff;
6118 }
6119 
6120 /////////////// Unit tests ///////////////
6121 
6122 #ifndef PRODUCT
6123 
6124 #define test_log(...)              \
6125   do {                             \
6126     if (VerboseInternalVMTests) {  \
6127       tty->print_cr(__VA_ARGS__);  \
6128       tty->flush();                \
6129     }                              \
6130   } while (false)
6131 
6132 class TestReserveMemorySpecial : AllStatic {
6133  public:
6134   static void small_page_write(void* addr, size_t size) {
6135     size_t page_size = os::vm_page_size();
6136 
6137     char* end = (char*)addr + size;
6138     for (char* p = (char*)addr; p < end; p += page_size) {
6139       *p = 1;
6140     }
6141   }
6142 
6143   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6144     if (!UseHugeTLBFS) {
6145       return;
6146     }
6147 
6148     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6149 
6150     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6151 
6152     if (addr != NULL) {
6153       small_page_write(addr, size);
6154 
6155       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6156     }
6157   }
6158 
6159   static void test_reserve_memory_special_huge_tlbfs_only() {
6160     if (!UseHugeTLBFS) {
6161       return;
6162     }
6163 
6164     size_t lp = os::large_page_size();
6165 
6166     for (size_t size = lp; size <= lp * 10; size += lp) {
6167       test_reserve_memory_special_huge_tlbfs_only(size);
6168     }
6169   }
6170 
6171   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6172     size_t lp = os::large_page_size();
6173     size_t ag = os::vm_allocation_granularity();
6174 
6175     // sizes to test
6176     const size_t sizes[] = {
6177       lp, lp + ag, lp + lp / 2, lp * 2,
6178       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6179       lp * 10, lp * 10 + lp / 2
6180     };
6181     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6182 
6183     // For each size/alignment combination, we test three scenarios:
6184     // 1) with req_addr == NULL
6185     // 2) with a non-null req_addr at which we expect to successfully allocate
6186     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6187     //    expect the allocation to either fail or to ignore req_addr
6188 
6189     // Pre-allocate two areas; they shall be as large as the largest allocation
6190     //  and aligned to the largest alignment we will be testing.
6191     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6192     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6193       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6194       -1, 0);
6195     assert(mapping1 != MAP_FAILED, "should work");
6196 
6197     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6198       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6199       -1, 0);
6200     assert(mapping2 != MAP_FAILED, "should work");
6201 
6202     // Unmap the first mapping, but leave the second mapping intact: the first
6203     // mapping will serve as a value for a "good" req_addr (case 2). The second
6204     // mapping, still intact, as "bad" req_addr (case 3).
6205     ::munmap(mapping1, mapping_size);
6206 
6207     // Case 1
6208     test_log("%s, req_addr NULL:", __FUNCTION__);
6209     test_log("size            align           result");
6210 
6211     for (int i = 0; i < num_sizes; i++) {
6212       const size_t size = sizes[i];
6213       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6214         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6215         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6216                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6217         if (p != NULL) {
6218           assert(is_ptr_aligned(p, alignment), "must be");
6219           small_page_write(p, size);
6220           os::Linux::release_memory_special_huge_tlbfs(p, size);
6221         }
6222       }
6223     }
6224 
6225     // Case 2
6226     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6227     test_log("size            align           req_addr         result");
6228 
6229     for (int i = 0; i < num_sizes; i++) {
6230       const size_t size = sizes[i];
6231       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6232         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6233         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6234         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6235                  size, alignment, p2i(req_addr), p2i(p),
6236                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6237         if (p != NULL) {
6238           assert(p == req_addr, "must be");
6239           small_page_write(p, size);
6240           os::Linux::release_memory_special_huge_tlbfs(p, size);
6241         }
6242       }
6243     }
6244 
6245     // Case 3
6246     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6247     test_log("size            align           req_addr         result");
6248 
6249     for (int i = 0; i < num_sizes; i++) {
6250       const size_t size = sizes[i];
6251       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6252         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6253         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6254         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6255                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6256         // as the area around req_addr contains already existing mappings, the API should always
6257         // return NULL (as per contract, it cannot return another address)
6258         assert(p == NULL, "must be");
6259       }
6260     }
6261 
6262     ::munmap(mapping2, mapping_size);
6263 
6264   }
6265 
6266   static void test_reserve_memory_special_huge_tlbfs() {
6267     if (!UseHugeTLBFS) {
6268       return;
6269     }
6270 
6271     test_reserve_memory_special_huge_tlbfs_only();
6272     test_reserve_memory_special_huge_tlbfs_mixed();
6273   }
6274 
6275   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6276     if (!UseSHM) {
6277       return;
6278     }
6279 
6280     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6281 
6282     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6283 
6284     if (addr != NULL) {
6285       assert(is_ptr_aligned(addr, alignment), "Check");
6286       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6287 
6288       small_page_write(addr, size);
6289 
6290       os::Linux::release_memory_special_shm(addr, size);
6291     }
6292   }
6293 
6294   static void test_reserve_memory_special_shm() {
6295     size_t lp = os::large_page_size();
6296     size_t ag = os::vm_allocation_granularity();
6297 
6298     for (size_t size = ag; size < lp * 3; size += ag) {
6299       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6300         test_reserve_memory_special_shm(size, alignment);
6301       }
6302     }
6303   }
6304 
6305   static void test() {
6306     test_reserve_memory_special_huge_tlbfs();
6307     test_reserve_memory_special_shm();
6308   }
6309 };
6310 
6311 void TestReserveMemorySpecial_test() {
6312   TestReserveMemorySpecial::test();
6313 }
6314 
6315 #endif