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 "oops/oop.inline.hpp"
  39 #include "os_linux.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "prims/jniFastGetField.hpp"
  42 #include "prims/jvm.h"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.hpp"
  49 #include "runtime/init.hpp"
  50 #include "runtime/java.hpp"
  51 #include "runtime/javaCalls.hpp"
  52 #include "runtime/mutexLocker.hpp"
  53 #include "runtime/objectMonitor.hpp"
  54 #include "runtime/orderAccess.inline.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/perfMemory.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/statSampler.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "runtime/thread.inline.hpp"
  61 #include "runtime/threadCritical.hpp"
  62 #include "runtime/timer.hpp"
  63 #include "semaphore_posix.hpp"
  64 #include "services/attachListener.hpp"
  65 #include "services/memTracker.hpp"
  66 #include "services/runtimeService.hpp"
  67 #include "utilities/decoder.hpp"
  68 #include "utilities/defaultStream.hpp"
  69 #include "utilities/events.hpp"
  70 #include "utilities/elfFile.hpp"
  71 #include "utilities/growableArray.hpp"
  72 #include "utilities/macros.hpp"
  73 #include "utilities/vmError.hpp"
  74 
  75 // put OS-includes here
  76 # include <sys/types.h>
  77 # include <sys/mman.h>
  78 # include <sys/stat.h>
  79 # include <sys/select.h>
  80 # include <pthread.h>
  81 # include <signal.h>
  82 # include <errno.h>
  83 # include <dlfcn.h>
  84 # include <stdio.h>
  85 # include <unistd.h>
  86 # include <sys/resource.h>
  87 # include <pthread.h>
  88 # include <sys/stat.h>
  89 # include <sys/time.h>
  90 # include <sys/times.h>
  91 # include <sys/utsname.h>
  92 # include <sys/socket.h>
  93 # include <sys/wait.h>
  94 # include <pwd.h>
  95 # include <poll.h>
  96 # include <semaphore.h>
  97 # include <fcntl.h>
  98 # include <string.h>
  99 # include <syscall.h>
 100 # include <sys/sysinfo.h>
 101 # include <gnu/libc-version.h>
 102 # include <sys/ipc.h>
 103 # include <sys/shm.h>
 104 # include <link.h>
 105 # include <stdint.h>
 106 # include <inttypes.h>
 107 # include <sys/ioctl.h>
 108 
 109 #ifndef _GNU_SOURCE
 110   #define _GNU_SOURCE
 111   #include <sched.h>
 112   #undef _GNU_SOURCE
 113 #else
 114   #include <sched.h>
 115 #endif
 116 
 117 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 118 // getrusage() is prepared to handle the associated failure.
 119 #ifndef RUSAGE_THREAD
 120   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 121 #endif
 122 
 123 #define MAX_PATH    (2 * K)
 124 
 125 #define MAX_SECS 100000000
 126 
 127 // for timer info max values which include all bits
 128 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 129 
 130 #define LARGEPAGES_BIT (1 << 6)
 131 ////////////////////////////////////////////////////////////////////////////////
 132 // global variables
 133 julong os::Linux::_physical_memory = 0;
 134 
 135 address   os::Linux::_initial_thread_stack_bottom = NULL;
 136 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 137 
 138 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 139 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 140 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 141 Mutex* os::Linux::_createThread_lock = NULL;
 142 pthread_t os::Linux::_main_thread;
 143 int os::Linux::_page_size = -1;
 144 bool os::Linux::_supports_fast_thread_cpu_time = false;
 145 uint32_t os::Linux::_os_version = 0;
 146 const char * os::Linux::_glibc_version = NULL;
 147 const char * os::Linux::_libpthread_version = NULL;
 148 pthread_condattr_t os::Linux::_condattr[1];
 149 
 150 static jlong initial_time_count=0;
 151 
 152 static int clock_tics_per_sec = 100;
 153 
 154 // For diagnostics to print a message once. see run_periodic_checks
 155 static sigset_t check_signal_done;
 156 static bool check_signals = true;
 157 
 158 // Signal number used to suspend/resume a thread
 159 
 160 // do not use any signal number less than SIGSEGV, see 4355769
 161 static int SR_signum = SIGUSR2;
 162 sigset_t SR_sigset;
 163 
 164 // Declarations
 165 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 166 
 167 // utility functions
 168 
 169 static int SR_initialize();
 170 
 171 julong os::available_memory() {
 172   return Linux::available_memory();
 173 }
 174 
 175 julong os::Linux::available_memory() {
 176   // values in struct sysinfo are "unsigned long"
 177   struct sysinfo si;
 178   sysinfo(&si);
 179 
 180   return (julong)si.freeram * si.mem_unit;
 181 }
 182 
 183 julong os::physical_memory() {
 184   return Linux::physical_memory();
 185 }
 186 
 187 // Return true if user is running as root.
 188 
 189 bool os::have_special_privileges() {
 190   static bool init = false;
 191   static bool privileges = false;
 192   if (!init) {
 193     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 194     init = true;
 195   }
 196   return privileges;
 197 }
 198 
 199 
 200 #ifndef SYS_gettid
 201 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 202   #ifdef __ia64__
 203     #define SYS_gettid 1105
 204   #else
 205     #ifdef __i386__
 206       #define SYS_gettid 224
 207     #else
 208       #ifdef __amd64__
 209         #define SYS_gettid 186
 210       #else
 211         #ifdef __sparc__
 212           #define SYS_gettid 143
 213         #else
 214           #error define gettid for the arch
 215         #endif
 216       #endif
 217     #endif
 218   #endif
 219 #endif
 220 
 221 // Cpu architecture string
 222 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 223 
 224 
 225 // pid_t gettid()
 226 //
 227 // Returns the kernel thread id of the currently running thread. Kernel
 228 // thread id is used to access /proc.
 229 pid_t os::Linux::gettid() {
 230   int rslt = syscall(SYS_gettid);
 231   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 232   return (pid_t)rslt;
 233 }
 234 
 235 // Most versions of linux have a bug where the number of processors are
 236 // determined by looking at the /proc file system.  In a chroot environment,
 237 // the system call returns 1.  This causes the VM to act as if it is
 238 // a single processor and elide locking (see is_MP() call).
 239 static bool unsafe_chroot_detected = false;
 240 static const char *unstable_chroot_error = "/proc file system not found.\n"
 241                      "Java may be unstable running multithreaded in a chroot "
 242                      "environment on Linux when /proc filesystem is not mounted.";
 243 
 244 void os::Linux::initialize_system_info() {
 245   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 246   if (processor_count() == 1) {
 247     pid_t pid = os::Linux::gettid();
 248     char fname[32];
 249     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 250     FILE *fp = fopen(fname, "r");
 251     if (fp == NULL) {
 252       unsafe_chroot_detected = true;
 253     } else {
 254       fclose(fp);
 255     }
 256   }
 257   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 258   assert(processor_count() > 0, "linux error");
 259 }
 260 
 261 void os::init_system_properties_values() {
 262   // The next steps are taken in the product version:
 263   //
 264   // Obtain the JAVA_HOME value from the location of libjvm.so.
 265   // This library should be located at:
 266   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 267   //
 268   // If "/jre/lib/" appears at the right place in the path, then we
 269   // assume libjvm.so is installed in a JDK and we use this path.
 270   //
 271   // Otherwise exit with message: "Could not create the Java virtual machine."
 272   //
 273   // The following extra steps are taken in the debugging version:
 274   //
 275   // If "/jre/lib/" does NOT appear at the right place in the path
 276   // instead of exit check for $JAVA_HOME environment variable.
 277   //
 278   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 279   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 280   // it looks like libjvm.so is installed there
 281   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 282   //
 283   // Otherwise exit.
 284   //
 285   // Important note: if the location of libjvm.so changes this
 286   // code needs to be changed accordingly.
 287 
 288   // See ld(1):
 289   //      The linker uses the following search paths to locate required
 290   //      shared libraries:
 291   //        1: ...
 292   //        ...
 293   //        7: The default directories, normally /lib and /usr/lib.
 294 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
 295   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 296 #else
 297   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 298 #endif
 299 
 300 // Base path of extensions installed on the system.
 301 #define SYS_EXT_DIR     "/usr/java/packages"
 302 #define EXTENSIONS_DIR  "/lib/ext"
 303 
 304   // Buffer that fits several sprintfs.
 305   // Note that the space for the colon and the trailing null are provided
 306   // by the nulls included by the sizeof operator.
 307   const size_t bufsize =
 308     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 309          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 310   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 311 
 312   // sysclasspath, java_home, dll_dir
 313   {
 314     char *pslash;
 315     os::jvm_path(buf, bufsize);
 316 
 317     // Found the full path to libjvm.so.
 318     // Now cut the path to <java_home>/jre if we can.
 319     pslash = strrchr(buf, '/');
 320     if (pslash != NULL) {
 321       *pslash = '\0';            // Get rid of /libjvm.so.
 322     }
 323     pslash = strrchr(buf, '/');
 324     if (pslash != NULL) {
 325       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 326     }
 327     Arguments::set_dll_dir(buf);
 328 
 329     if (pslash != NULL) {
 330       pslash = strrchr(buf, '/');
 331       if (pslash != NULL) {
 332         *pslash = '\0';          // Get rid of /<arch>.
 333         pslash = strrchr(buf, '/');
 334         if (pslash != NULL) {
 335           *pslash = '\0';        // Get rid of /lib.
 336         }
 337       }
 338     }
 339     Arguments::set_java_home(buf);
 340     set_boot_path('/', ':');
 341   }
 342 
 343   // Where to look for native libraries.
 344   //
 345   // Note: Due to a legacy implementation, most of the library path
 346   // is set in the launcher. This was to accomodate linking restrictions
 347   // on legacy Linux implementations (which are no longer supported).
 348   // Eventually, all the library path setting will be done here.
 349   //
 350   // However, to prevent the proliferation of improperly built native
 351   // libraries, the new path component /usr/java/packages is added here.
 352   // Eventually, all the library path setting will be done here.
 353   {
 354     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 355     // should always exist (until the legacy problem cited above is
 356     // addressed).
 357     const char *v = ::getenv("LD_LIBRARY_PATH");
 358     const char *v_colon = ":";
 359     if (v == NULL) { v = ""; v_colon = ""; }
 360     // That's +1 for the colon and +1 for the trailing '\0'.
 361     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 362                                                      strlen(v) + 1 +
 363                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 364                                                      mtInternal);
 365     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 366     Arguments::set_library_path(ld_library_path);
 367     FREE_C_HEAP_ARRAY(char, ld_library_path);
 368   }
 369 
 370   // Extensions directories.
 371   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 372   Arguments::set_ext_dirs(buf);
 373 
 374   FREE_C_HEAP_ARRAY(char, buf);
 375 
 376 #undef DEFAULT_LIBPATH
 377 #undef SYS_EXT_DIR
 378 #undef EXTENSIONS_DIR
 379 }
 380 
 381 ////////////////////////////////////////////////////////////////////////////////
 382 // breakpoint support
 383 
 384 void os::breakpoint() {
 385   BREAKPOINT;
 386 }
 387 
 388 extern "C" void breakpoint() {
 389   // use debugger to set breakpoint here
 390 }
 391 
 392 ////////////////////////////////////////////////////////////////////////////////
 393 // signal support
 394 
 395 debug_only(static bool signal_sets_initialized = false);
 396 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 397 
 398 bool os::Linux::is_sig_ignored(int sig) {
 399   struct sigaction oact;
 400   sigaction(sig, (struct sigaction*)NULL, &oact);
 401   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 402                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 403   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 404     return true;
 405   } else {
 406     return false;
 407   }
 408 }
 409 
 410 void os::Linux::signal_sets_init() {
 411   // Should also have an assertion stating we are still single-threaded.
 412   assert(!signal_sets_initialized, "Already initialized");
 413   // Fill in signals that are necessarily unblocked for all threads in
 414   // the VM. Currently, we unblock the following signals:
 415   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 416   //                         by -Xrs (=ReduceSignalUsage));
 417   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 418   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 419   // the dispositions or masks wrt these signals.
 420   // Programs embedding the VM that want to use the above signals for their
 421   // own purposes must, at this time, use the "-Xrs" option to prevent
 422   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 423   // (See bug 4345157, and other related bugs).
 424   // In reality, though, unblocking these signals is really a nop, since
 425   // these signals are not blocked by default.
 426   sigemptyset(&unblocked_sigs);
 427   sigemptyset(&allowdebug_blocked_sigs);
 428   sigaddset(&unblocked_sigs, SIGILL);
 429   sigaddset(&unblocked_sigs, SIGSEGV);
 430   sigaddset(&unblocked_sigs, SIGBUS);
 431   sigaddset(&unblocked_sigs, SIGFPE);
 432 #if defined(PPC64)
 433   sigaddset(&unblocked_sigs, SIGTRAP);
 434 #endif
 435   sigaddset(&unblocked_sigs, SR_signum);
 436 
 437   if (!ReduceSignalUsage) {
 438     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 439       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 440       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 441     }
 442     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 443       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 444       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 445     }
 446     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 447       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 448       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 449     }
 450   }
 451   // Fill in signals that are blocked by all but the VM thread.
 452   sigemptyset(&vm_sigs);
 453   if (!ReduceSignalUsage) {
 454     sigaddset(&vm_sigs, BREAK_SIGNAL);
 455   }
 456   debug_only(signal_sets_initialized = true);
 457 
 458 }
 459 
 460 // These are signals that are unblocked while a thread is running Java.
 461 // (For some reason, they get blocked by default.)
 462 sigset_t* os::Linux::unblocked_signals() {
 463   assert(signal_sets_initialized, "Not initialized");
 464   return &unblocked_sigs;
 465 }
 466 
 467 // These are the signals that are blocked while a (non-VM) thread is
 468 // running Java. Only the VM thread handles these signals.
 469 sigset_t* os::Linux::vm_signals() {
 470   assert(signal_sets_initialized, "Not initialized");
 471   return &vm_sigs;
 472 }
 473 
 474 // These are signals that are blocked during cond_wait to allow debugger in
 475 sigset_t* os::Linux::allowdebug_blocked_signals() {
 476   assert(signal_sets_initialized, "Not initialized");
 477   return &allowdebug_blocked_sigs;
 478 }
 479 
 480 void os::Linux::hotspot_sigmask(Thread* thread) {
 481 
 482   //Save caller's signal mask before setting VM signal mask
 483   sigset_t caller_sigmask;
 484   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 485 
 486   OSThread* osthread = thread->osthread();
 487   osthread->set_caller_sigmask(caller_sigmask);
 488 
 489   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 490 
 491   if (!ReduceSignalUsage) {
 492     if (thread->is_VM_thread()) {
 493       // Only the VM thread handles BREAK_SIGNAL ...
 494       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 495     } else {
 496       // ... all other threads block BREAK_SIGNAL
 497       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 498     }
 499   }
 500 }
 501 
 502 //////////////////////////////////////////////////////////////////////////////
 503 // detecting pthread library
 504 
 505 void os::Linux::libpthread_init() {
 506   // Save glibc and pthread version strings.
 507 #if !defined(_CS_GNU_LIBC_VERSION) || \
 508     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 509   #error "glibc too old (< 2.3.2)"
 510 #endif
 511 
 512   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 513   assert(n > 0, "cannot retrieve glibc version");
 514   char *str = (char *)malloc(n, mtInternal);
 515   confstr(_CS_GNU_LIBC_VERSION, str, n);
 516   os::Linux::set_glibc_version(str);
 517 
 518   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 519   assert(n > 0, "cannot retrieve pthread version");
 520   str = (char *)malloc(n, mtInternal);
 521   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 522   os::Linux::set_libpthread_version(str);
 523 }
 524 
 525 /////////////////////////////////////////////////////////////////////////////
 526 // thread stack expansion
 527 
 528 // os::Linux::manually_expand_stack() takes care of expanding the thread
 529 // stack. Note that this is normally not needed: pthread stacks allocate
 530 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 531 // committed. Therefore it is not necessary to expand the stack manually.
 532 //
 533 // Manually expanding the stack was historically needed on LinuxThreads
 534 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 535 // it is kept to deal with very rare corner cases:
 536 //
 537 // For one, user may run the VM on an own implementation of threads
 538 // whose stacks are - like the old LinuxThreads - implemented using
 539 // mmap(MAP_GROWSDOWN).
 540 //
 541 // Also, this coding may be needed if the VM is running on the primordial
 542 // thread. Normally we avoid running on the primordial thread; however,
 543 // user may still invoke the VM on the primordial thread.
 544 //
 545 // The following historical comment describes the details about running
 546 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 547 
 548 
 549 // Force Linux kernel to expand current thread stack. If "bottom" is close
 550 // to the stack guard, caller should block all signals.
 551 //
 552 // MAP_GROWSDOWN:
 553 //   A special mmap() flag that is used to implement thread stacks. It tells
 554 //   kernel that the memory region should extend downwards when needed. This
 555 //   allows early versions of LinuxThreads to only mmap the first few pages
 556 //   when creating a new thread. Linux kernel will automatically expand thread
 557 //   stack as needed (on page faults).
 558 //
 559 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 560 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 561 //   region, it's hard to tell if the fault is due to a legitimate stack
 562 //   access or because of reading/writing non-exist memory (e.g. buffer
 563 //   overrun). As a rule, if the fault happens below current stack pointer,
 564 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 565 //   application (see Linux kernel fault.c).
 566 //
 567 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 568 //   stack overflow detection.
 569 //
 570 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 571 //   not use MAP_GROWSDOWN.
 572 //
 573 // To get around the problem and allow stack banging on Linux, we need to
 574 // manually expand thread stack after receiving the SIGSEGV.
 575 //
 576 // There are two ways to expand thread stack to address "bottom", we used
 577 // both of them in JVM before 1.5:
 578 //   1. adjust stack pointer first so that it is below "bottom", and then
 579 //      touch "bottom"
 580 //   2. mmap() the page in question
 581 //
 582 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 583 // if current sp is already near the lower end of page 101, and we need to
 584 // call mmap() to map page 100, it is possible that part of the mmap() frame
 585 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 586 // That will destroy the mmap() frame and cause VM to crash.
 587 //
 588 // The following code works by adjusting sp first, then accessing the "bottom"
 589 // page to force a page fault. Linux kernel will then automatically expand the
 590 // stack mapping.
 591 //
 592 // _expand_stack_to() assumes its frame size is less than page size, which
 593 // should always be true if the function is not inlined.
 594 
 595 static void NOINLINE _expand_stack_to(address bottom) {
 596   address sp;
 597   size_t size;
 598   volatile char *p;
 599 
 600   // Adjust bottom to point to the largest address within the same page, it
 601   // gives us a one-page buffer if alloca() allocates slightly more memory.
 602   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 603   bottom += os::Linux::page_size() - 1;
 604 
 605   // sp might be slightly above current stack pointer; if that's the case, we
 606   // will alloca() a little more space than necessary, which is OK. Don't use
 607   // os::current_stack_pointer(), as its result can be slightly below current
 608   // stack pointer, causing us to not alloca enough to reach "bottom".
 609   sp = (address)&sp;
 610 
 611   if (sp > bottom) {
 612     size = sp - bottom;
 613     p = (volatile char *)alloca(size);
 614     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 615     p[0] = '\0';
 616   }
 617 }
 618 
 619 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 620   assert(t!=NULL, "just checking");
 621   assert(t->osthread()->expanding_stack(), "expand should be set");
 622   assert(t->stack_base() != NULL, "stack_base was not initialized");
 623 
 624   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 625     sigset_t mask_all, old_sigset;
 626     sigfillset(&mask_all);
 627     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 628     _expand_stack_to(addr);
 629     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 630     return true;
 631   }
 632   return false;
 633 }
 634 
 635 //////////////////////////////////////////////////////////////////////////////
 636 // create new thread
 637 
 638 // Thread start routine for all newly created threads
 639 static void *thread_native_entry(Thread *thread) {
 640   // Try to randomize the cache line index of hot stack frames.
 641   // This helps when threads of the same stack traces evict each other's
 642   // cache lines. The threads can be either from the same JVM instance, or
 643   // from different JVM instances. The benefit is especially true for
 644   // processors with hyperthreading technology.
 645   static int counter = 0;
 646   int pid = os::current_process_id();
 647   alloca(((pid ^ counter++) & 7) * 128);
 648 
 649   thread->initialize_thread_current();
 650 
 651   OSThread* osthread = thread->osthread();
 652   Monitor* sync = osthread->startThread_lock();
 653 
 654   osthread->set_thread_id(os::current_thread_id());
 655 
 656   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 657     os::current_thread_id(), (uintx) pthread_self());
 658 
 659   if (UseNUMA) {
 660     int lgrp_id = os::numa_get_group_id();
 661     if (lgrp_id != -1) {
 662       thread->set_lgrp_id(lgrp_id);
 663     }
 664   }
 665   // initialize signal mask for this thread
 666   os::Linux::hotspot_sigmask(thread);
 667 
 668   // initialize floating point control register
 669   os::Linux::init_thread_fpu_state();
 670 
 671   // handshaking with parent thread
 672   {
 673     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 674 
 675     // notify parent thread
 676     osthread->set_state(INITIALIZED);
 677     sync->notify_all();
 678 
 679     // wait until os::start_thread()
 680     while (osthread->get_state() == INITIALIZED) {
 681       sync->wait(Mutex::_no_safepoint_check_flag);
 682     }
 683   }
 684 
 685   // call one more level start routine
 686   thread->run();
 687 
 688   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 689     os::current_thread_id(), (uintx) pthread_self());
 690 
 691   // If a thread has not deleted itself ("delete this") as part of its
 692   // termination sequence, we have to ensure thread-local-storage is
 693   // cleared before we actually terminate. No threads should ever be
 694   // deleted asynchronously with respect to their termination.
 695   if (Thread::current_or_null_safe() != NULL) {
 696     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 697     thread->clear_thread_current();
 698   }
 699 
 700   return 0;
 701 }
 702 
 703 bool os::create_thread(Thread* thread, ThreadType thr_type,
 704                        size_t req_stack_size) {
 705   assert(thread->osthread() == NULL, "caller responsible");
 706 
 707   // Allocate the OSThread object
 708   OSThread* osthread = new OSThread(NULL, NULL);
 709   if (osthread == NULL) {
 710     return false;
 711   }
 712 
 713   // set the correct thread state
 714   osthread->set_thread_type(thr_type);
 715 
 716   // Initial state is ALLOCATED but not INITIALIZED
 717   osthread->set_state(ALLOCATED);
 718 
 719   thread->set_osthread(osthread);
 720 
 721   // init thread attributes
 722   pthread_attr_t attr;
 723   pthread_attr_init(&attr);
 724   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 725 
 726   // calculate stack size if it's not specified by caller
 727   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 728   pthread_attr_setstacksize(&attr, stack_size);
 729 
 730   // glibc guard page
 731   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 732 
 733   ThreadState state;
 734 
 735   {
 736     pthread_t tid;
 737     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 738 
 739     char buf[64];
 740     if (ret == 0) {
 741       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 742         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 743     } else {
 744       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 745         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 746     }
 747 
 748     pthread_attr_destroy(&attr);
 749 
 750     if (ret != 0) {
 751       // Need to clean up stuff we've allocated so far
 752       thread->set_osthread(NULL);
 753       delete osthread;
 754       return false;
 755     }
 756 
 757     // Store pthread info into the OSThread
 758     osthread->set_pthread_id(tid);
 759 
 760     // Wait until child thread is either initialized or aborted
 761     {
 762       Monitor* sync_with_child = osthread->startThread_lock();
 763       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 764       while ((state = osthread->get_state()) == ALLOCATED) {
 765         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 766       }
 767     }
 768   }
 769 
 770   // Aborted due to thread limit being reached
 771   if (state == ZOMBIE) {
 772     thread->set_osthread(NULL);
 773     delete osthread;
 774     return false;
 775   }
 776 
 777   // The thread is returned suspended (in state INITIALIZED),
 778   // and is started higher up in the call chain
 779   assert(state == INITIALIZED, "race condition");
 780   return true;
 781 }
 782 
 783 /////////////////////////////////////////////////////////////////////////////
 784 // attach existing thread
 785 
 786 // bootstrap the main thread
 787 bool os::create_main_thread(JavaThread* thread) {
 788   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 789   return create_attached_thread(thread);
 790 }
 791 
 792 bool os::create_attached_thread(JavaThread* thread) {
 793 #ifdef ASSERT
 794   thread->verify_not_published();
 795 #endif
 796 
 797   // Allocate the OSThread object
 798   OSThread* osthread = new OSThread(NULL, NULL);
 799 
 800   if (osthread == NULL) {
 801     return false;
 802   }
 803 
 804   // Store pthread info into the OSThread
 805   osthread->set_thread_id(os::Linux::gettid());
 806   osthread->set_pthread_id(::pthread_self());
 807 
 808   // initialize floating point control register
 809   os::Linux::init_thread_fpu_state();
 810 
 811   // Initial thread state is RUNNABLE
 812   osthread->set_state(RUNNABLE);
 813 
 814   thread->set_osthread(osthread);
 815 
 816   if (UseNUMA) {
 817     int lgrp_id = os::numa_get_group_id();
 818     if (lgrp_id != -1) {
 819       thread->set_lgrp_id(lgrp_id);
 820     }
 821   }
 822 
 823   if (os::Linux::is_initial_thread()) {
 824     // If current thread is initial thread, its stack is mapped on demand,
 825     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 826     // the entire stack region to avoid SEGV in stack banging.
 827     // It is also useful to get around the heap-stack-gap problem on SuSE
 828     // kernel (see 4821821 for details). We first expand stack to the top
 829     // of yellow zone, then enable stack yellow zone (order is significant,
 830     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 831     // is no gap between the last two virtual memory regions.
 832 
 833     JavaThread *jt = (JavaThread *)thread;
 834     address addr = jt->stack_reserved_zone_base();
 835     assert(addr != NULL, "initialization problem?");
 836     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 837 
 838     osthread->set_expanding_stack();
 839     os::Linux::manually_expand_stack(jt, addr);
 840     osthread->clear_expanding_stack();
 841   }
 842 
 843   // initialize signal mask for this thread
 844   // and save the caller's signal mask
 845   os::Linux::hotspot_sigmask(thread);
 846 
 847   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 848     os::current_thread_id(), (uintx) pthread_self());
 849 
 850   return true;
 851 }
 852 
 853 void os::pd_start_thread(Thread* thread) {
 854   OSThread * osthread = thread->osthread();
 855   assert(osthread->get_state() != INITIALIZED, "just checking");
 856   Monitor* sync_with_child = osthread->startThread_lock();
 857   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 858   sync_with_child->notify();
 859 }
 860 
 861 // Free Linux resources related to the OSThread
 862 void os::free_thread(OSThread* osthread) {
 863   assert(osthread != NULL, "osthread not set");
 864 
 865   // We are told to free resources of the argument thread,
 866   // but we can only really operate on the current thread.
 867   assert(Thread::current()->osthread() == osthread,
 868          "os::free_thread but not current thread");
 869 
 870 #ifdef ASSERT
 871   sigset_t current;
 872   sigemptyset(&current);
 873   pthread_sigmask(SIG_SETMASK, NULL, &current);
 874   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 875 #endif
 876 
 877   // Restore caller's signal mask
 878   sigset_t sigmask = osthread->caller_sigmask();
 879   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 880 
 881   delete osthread;
 882 }
 883 
 884 //////////////////////////////////////////////////////////////////////////////
 885 // initial thread
 886 
 887 // Check if current thread is the initial thread, similar to Solaris thr_main.
 888 bool os::Linux::is_initial_thread(void) {
 889   char dummy;
 890   // If called before init complete, thread stack bottom will be null.
 891   // Can be called if fatal error occurs before initialization.
 892   if (initial_thread_stack_bottom() == NULL) return false;
 893   assert(initial_thread_stack_bottom() != NULL &&
 894          initial_thread_stack_size()   != 0,
 895          "os::init did not locate initial thread's stack region");
 896   if ((address)&dummy >= initial_thread_stack_bottom() &&
 897       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 898     return true;
 899   } else {
 900     return false;
 901   }
 902 }
 903 
 904 // Find the virtual memory area that contains addr
 905 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 906   FILE *fp = fopen("/proc/self/maps", "r");
 907   if (fp) {
 908     address low, high;
 909     while (!feof(fp)) {
 910       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 911         if (low <= addr && addr < high) {
 912           if (vma_low)  *vma_low  = low;
 913           if (vma_high) *vma_high = high;
 914           fclose(fp);
 915           return true;
 916         }
 917       }
 918       for (;;) {
 919         int ch = fgetc(fp);
 920         if (ch == EOF || ch == (int)'\n') break;
 921       }
 922     }
 923     fclose(fp);
 924   }
 925   return false;
 926 }
 927 
 928 // Locate initial thread stack. This special handling of initial thread stack
 929 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 930 // bogus value for the primordial process thread. While the launcher has created
 931 // the VM in a new thread since JDK 6, we still have to allow for the use of the
 932 // JNI invocation API from a primordial thread.
 933 void os::Linux::capture_initial_stack(size_t max_size) {
 934 
 935   // max_size is either 0 (which means accept OS default for thread stacks) or
 936   // a user-specified value known to be greater than the minimum needed. If we
 937   // are actually on the primordial thread we can make it appear that we have a
 938   // smaller max_size stack by inserting the guard pages at that location. But we
 939   // can not do anything to emulate a larger stack than what has been provided by
 940   // the OS or threading library. In fact if we try to use a stack greater than
 941   // what is set by rlimit then we will crash the hosting process.
 942 
 943   // Mamimum stack size is the easy part, get it from RLIMIT_STACK
 944   // If this is "unlimited" then it will be a huge value.
 945   struct rlimit rlim;
 946   getrlimit(RLIMIT_STACK, &rlim);
 947   size_t stack_size = rlim.rlim_cur;
 948 
 949   // 6308388: a bug in ld.so will relocate its own .data section to the
 950   //   lower end of primordial stack; reduce ulimit -s value a little bit
 951   //   so we won't install guard page on ld.so's data section.
 952   stack_size -= 2 * page_size();
 953 











 954   // Try to figure out where the stack base (top) is. This is harder.
 955   //
 956   // When an application is started, glibc saves the initial stack pointer in
 957   // a global variable "__libc_stack_end", which is then used by system
 958   // libraries. __libc_stack_end should be pretty close to stack top. The
 959   // variable is available since the very early days. However, because it is
 960   // a private interface, it could disappear in the future.
 961   //
 962   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 963   // to __libc_stack_end, it is very close to stack top, but isn't the real
 964   // stack top. Note that /proc may not exist if VM is running as a chroot
 965   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 966   // /proc/<pid>/stat could change in the future (though unlikely).
 967   //
 968   // We try __libc_stack_end first. If that doesn't work, look for
 969   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
 970   // as a hint, which should work well in most cases.
 971 
 972   uintptr_t stack_start;
 973 
 974   // try __libc_stack_end first
 975   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
 976   if (p && *p) {
 977     stack_start = *p;
 978   } else {
 979     // see if we can get the start_stack field from /proc/self/stat
 980     FILE *fp;
 981     int pid;
 982     char state;
 983     int ppid;
 984     int pgrp;
 985     int session;
 986     int nr;
 987     int tpgrp;
 988     unsigned long flags;
 989     unsigned long minflt;
 990     unsigned long cminflt;
 991     unsigned long majflt;
 992     unsigned long cmajflt;
 993     unsigned long utime;
 994     unsigned long stime;
 995     long cutime;
 996     long cstime;
 997     long prio;
 998     long nice;
 999     long junk;
1000     long it_real;
1001     uintptr_t start;
1002     uintptr_t vsize;
1003     intptr_t rss;
1004     uintptr_t rsslim;
1005     uintptr_t scodes;
1006     uintptr_t ecode;
1007     int i;
1008 
1009     // Figure what the primordial thread stack base is. Code is inspired
1010     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1011     // followed by command name surrounded by parentheses, state, etc.
1012     char stat[2048];
1013     int statlen;
1014 
1015     fp = fopen("/proc/self/stat", "r");
1016     if (fp) {
1017       statlen = fread(stat, 1, 2047, fp);
1018       stat[statlen] = '\0';
1019       fclose(fp);
1020 
1021       // Skip pid and the command string. Note that we could be dealing with
1022       // weird command names, e.g. user could decide to rename java launcher
1023       // to "java 1.4.2 :)", then the stat file would look like
1024       //                1234 (java 1.4.2 :)) R ... ...
1025       // We don't really need to know the command string, just find the last
1026       // occurrence of ")" and then start parsing from there. See bug 4726580.
1027       char * s = strrchr(stat, ')');
1028 
1029       i = 0;
1030       if (s) {
1031         // Skip blank chars
1032         do { s++; } while (s && isspace(*s));
1033 
1034 #define _UFM UINTX_FORMAT
1035 #define _DFM INTX_FORMAT
1036 
1037         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1038         //              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
1039         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,
1040                    &state,          // 3  %c
1041                    &ppid,           // 4  %d
1042                    &pgrp,           // 5  %d
1043                    &session,        // 6  %d
1044                    &nr,             // 7  %d
1045                    &tpgrp,          // 8  %d
1046                    &flags,          // 9  %lu
1047                    &minflt,         // 10 %lu
1048                    &cminflt,        // 11 %lu
1049                    &majflt,         // 12 %lu
1050                    &cmajflt,        // 13 %lu
1051                    &utime,          // 14 %lu
1052                    &stime,          // 15 %lu
1053                    &cutime,         // 16 %ld
1054                    &cstime,         // 17 %ld
1055                    &prio,           // 18 %ld
1056                    &nice,           // 19 %ld
1057                    &junk,           // 20 %ld
1058                    &it_real,        // 21 %ld
1059                    &start,          // 22 UINTX_FORMAT
1060                    &vsize,          // 23 UINTX_FORMAT
1061                    &rss,            // 24 INTX_FORMAT
1062                    &rsslim,         // 25 UINTX_FORMAT
1063                    &scodes,         // 26 UINTX_FORMAT
1064                    &ecode,          // 27 UINTX_FORMAT
1065                    &stack_start);   // 28 UINTX_FORMAT
1066       }
1067 
1068 #undef _UFM
1069 #undef _DFM
1070 
1071       if (i != 28 - 2) {
1072         assert(false, "Bad conversion from /proc/self/stat");
1073         // product mode - assume we are the initial thread, good luck in the
1074         // embedded case.
1075         warning("Can't detect initial thread stack location - bad conversion");
1076         stack_start = (uintptr_t) &rlim;
1077       }
1078     } else {
1079       // For some reason we can't open /proc/self/stat (for example, running on
1080       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1081       // most cases, so don't abort:
1082       warning("Can't detect initial thread stack location - no /proc/self/stat");
1083       stack_start = (uintptr_t) &rlim;
1084     }
1085   }
1086 
1087   // Now we have a pointer (stack_start) very close to the stack top, the
1088   // next thing to do is to figure out the exact location of stack top. We
1089   // can find out the virtual memory area that contains stack_start by
1090   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1091   // and its upper limit is the real stack top. (again, this would fail if
1092   // running inside chroot, because /proc may not exist.)
1093 
1094   uintptr_t stack_top;
1095   address low, high;
1096   if (find_vma((address)stack_start, &low, &high)) {
1097     // success, "high" is the true stack top. (ignore "low", because initial
1098     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1099     stack_top = (uintptr_t)high;
1100   } else {
1101     // failed, likely because /proc/self/maps does not exist
1102     warning("Can't detect initial thread stack location - find_vma failed");
1103     // best effort: stack_start is normally within a few pages below the real
1104     // stack top, use it as stack top, and reduce stack size so we won't put
1105     // guard page outside stack.
1106     stack_top = stack_start;
1107     stack_size -= 16 * page_size();
1108   }
1109 
1110   // stack_top could be partially down the page so align it
1111   stack_top = align_size_up(stack_top, page_size());
1112 
1113   // Allowed stack value is minimum of max_size and what we derived from rlimit
1114   if (max_size > 0 && stack_size > max_size) {
1115     _initial_thread_stack_size = max_size;
1116   } else {
1117     _initial_thread_stack_size = stack_size;
1118   }
1119 
1120   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1121   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1122 }
1123 
1124 ////////////////////////////////////////////////////////////////////////////////
1125 // time support
1126 
1127 // Time since start-up in seconds to a fine granularity.
1128 // Used by VMSelfDestructTimer and the MemProfiler.
1129 double os::elapsedTime() {
1130 
1131   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1132 }
1133 
1134 jlong os::elapsed_counter() {
1135   return javaTimeNanos() - initial_time_count;
1136 }
1137 
1138 jlong os::elapsed_frequency() {
1139   return NANOSECS_PER_SEC; // nanosecond resolution
1140 }
1141 
1142 bool os::supports_vtime() { return true; }
1143 bool os::enable_vtime()   { return false; }
1144 bool os::vtime_enabled()  { return false; }
1145 
1146 double os::elapsedVTime() {
1147   struct rusage usage;
1148   int retval = getrusage(RUSAGE_THREAD, &usage);
1149   if (retval == 0) {
1150     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);
1151   } else {
1152     // better than nothing, but not much
1153     return elapsedTime();
1154   }
1155 }
1156 
1157 jlong os::javaTimeMillis() {
1158   timeval time;
1159   int status = gettimeofday(&time, NULL);
1160   assert(status != -1, "linux error");
1161   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1162 }
1163 
1164 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1165   timeval time;
1166   int status = gettimeofday(&time, NULL);
1167   assert(status != -1, "linux error");
1168   seconds = jlong(time.tv_sec);
1169   nanos = jlong(time.tv_usec) * 1000;
1170 }
1171 
1172 
1173 #ifndef CLOCK_MONOTONIC
1174   #define CLOCK_MONOTONIC (1)
1175 #endif
1176 
1177 void os::Linux::clock_init() {
1178   // we do dlopen's in this particular order due to bug in linux
1179   // dynamical loader (see 6348968) leading to crash on exit
1180   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1181   if (handle == NULL) {
1182     handle = dlopen("librt.so", RTLD_LAZY);
1183   }
1184 
1185   if (handle) {
1186     int (*clock_getres_func)(clockid_t, struct timespec*) =
1187            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1188     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1189            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1190     if (clock_getres_func && clock_gettime_func) {
1191       // See if monotonic clock is supported by the kernel. Note that some
1192       // early implementations simply return kernel jiffies (updated every
1193       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1194       // for nano time (though the monotonic property is still nice to have).
1195       // It's fixed in newer kernels, however clock_getres() still returns
1196       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1197       // resolution for now. Hopefully as people move to new kernels, this
1198       // won't be a problem.
1199       struct timespec res;
1200       struct timespec tp;
1201       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1202           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1203         // yes, monotonic clock is supported
1204         _clock_gettime = clock_gettime_func;
1205         return;
1206       } else {
1207         // close librt if there is no monotonic clock
1208         dlclose(handle);
1209       }
1210     }
1211   }
1212   warning("No monotonic clock was available - timed services may " \
1213           "be adversely affected if the time-of-day clock changes");
1214 }
1215 
1216 #ifndef SYS_clock_getres
1217   #if defined(X86) || defined(PPC64) || defined(S390)
1218     #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1219     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1220   #else
1221     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1222     #define sys_clock_getres(x,y)  -1
1223   #endif
1224 #else
1225   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1226 #endif
1227 
1228 void os::Linux::fast_thread_clock_init() {
1229   if (!UseLinuxPosixThreadCPUClocks) {
1230     return;
1231   }
1232   clockid_t clockid;
1233   struct timespec tp;
1234   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1235       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1236 
1237   // Switch to using fast clocks for thread cpu time if
1238   // the sys_clock_getres() returns 0 error code.
1239   // Note, that some kernels may support the current thread
1240   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1241   // returned by the pthread_getcpuclockid().
1242   // If the fast Posix clocks are supported then the sys_clock_getres()
1243   // must return at least tp.tv_sec == 0 which means a resolution
1244   // better than 1 sec. This is extra check for reliability.
1245 
1246   if (pthread_getcpuclockid_func &&
1247       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1248       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1249     _supports_fast_thread_cpu_time = true;
1250     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1251   }
1252 }
1253 
1254 jlong os::javaTimeNanos() {
1255   if (os::supports_monotonic_clock()) {
1256     struct timespec tp;
1257     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1258     assert(status == 0, "gettime error");
1259     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1260     return result;
1261   } else {
1262     timeval time;
1263     int status = gettimeofday(&time, NULL);
1264     assert(status != -1, "linux error");
1265     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1266     return 1000 * usecs;
1267   }
1268 }
1269 
1270 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1271   if (os::supports_monotonic_clock()) {
1272     info_ptr->max_value = ALL_64_BITS;
1273 
1274     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1275     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1276     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1277   } else {
1278     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1279     info_ptr->max_value = ALL_64_BITS;
1280 
1281     // gettimeofday is a real time clock so it skips
1282     info_ptr->may_skip_backward = true;
1283     info_ptr->may_skip_forward = true;
1284   }
1285 
1286   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1287 }
1288 
1289 // Return the real, user, and system times in seconds from an
1290 // arbitrary fixed point in the past.
1291 bool os::getTimesSecs(double* process_real_time,
1292                       double* process_user_time,
1293                       double* process_system_time) {
1294   struct tms ticks;
1295   clock_t real_ticks = times(&ticks);
1296 
1297   if (real_ticks == (clock_t) (-1)) {
1298     return false;
1299   } else {
1300     double ticks_per_second = (double) clock_tics_per_sec;
1301     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1302     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1303     *process_real_time = ((double) real_ticks) / ticks_per_second;
1304 
1305     return true;
1306   }
1307 }
1308 
1309 
1310 char * os::local_time_string(char *buf, size_t buflen) {
1311   struct tm t;
1312   time_t long_time;
1313   time(&long_time);
1314   localtime_r(&long_time, &t);
1315   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1316                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1317                t.tm_hour, t.tm_min, t.tm_sec);
1318   return buf;
1319 }
1320 
1321 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1322   return localtime_r(clock, res);
1323 }
1324 
1325 ////////////////////////////////////////////////////////////////////////////////
1326 // runtime exit support
1327 
1328 // Note: os::shutdown() might be called very early during initialization, or
1329 // called from signal handler. Before adding something to os::shutdown(), make
1330 // sure it is async-safe and can handle partially initialized VM.
1331 void os::shutdown() {
1332 
1333   // allow PerfMemory to attempt cleanup of any persistent resources
1334   perfMemory_exit();
1335 
1336   // needs to remove object in file system
1337   AttachListener::abort();
1338 
1339   // flush buffered output, finish log files
1340   ostream_abort();
1341 
1342   // Check for abort hook
1343   abort_hook_t abort_hook = Arguments::abort_hook();
1344   if (abort_hook != NULL) {
1345     abort_hook();
1346   }
1347 
1348 }
1349 
1350 // Note: os::abort() might be called very early during initialization, or
1351 // called from signal handler. Before adding something to os::abort(), make
1352 // sure it is async-safe and can handle partially initialized VM.
1353 void os::abort(bool dump_core, void* siginfo, const void* context) {
1354   os::shutdown();
1355   if (dump_core) {
1356 #ifndef PRODUCT
1357     fdStream out(defaultStream::output_fd());
1358     out.print_raw("Current thread is ");
1359     char buf[16];
1360     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1361     out.print_raw_cr(buf);
1362     out.print_raw_cr("Dumping core ...");
1363 #endif
1364     ::abort(); // dump core
1365   }
1366 
1367   ::exit(1);
1368 }
1369 
1370 // Die immediately, no exit hook, no abort hook, no cleanup.
1371 void os::die() {
1372   ::abort();
1373 }
1374 
1375 
1376 // This method is a copy of JDK's sysGetLastErrorString
1377 // from src/solaris/hpi/src/system_md.c
1378 
1379 size_t os::lasterror(char *buf, size_t len) {
1380   if (errno == 0)  return 0;
1381 
1382   const char *s = os::strerror(errno);
1383   size_t n = ::strlen(s);
1384   if (n >= len) {
1385     n = len - 1;
1386   }
1387   ::strncpy(buf, s, n);
1388   buf[n] = '\0';
1389   return n;
1390 }
1391 
1392 // thread_id is kernel thread id (similar to Solaris LWP id)
1393 intx os::current_thread_id() { return os::Linux::gettid(); }
1394 int os::current_process_id() {
1395   return ::getpid();
1396 }
1397 
1398 // DLL functions
1399 
1400 const char* os::dll_file_extension() { return ".so"; }
1401 
1402 // This must be hard coded because it's the system's temporary
1403 // directory not the java application's temp directory, ala java.io.tmpdir.
1404 const char* os::get_temp_directory() { return "/tmp"; }
1405 
1406 static bool file_exists(const char* filename) {
1407   struct stat statbuf;
1408   if (filename == NULL || strlen(filename) == 0) {
1409     return false;
1410   }
1411   return os::stat(filename, &statbuf) == 0;
1412 }
1413 
1414 bool os::dll_build_name(char* buffer, size_t buflen,
1415                         const char* pname, const char* fname) {
1416   bool retval = false;
1417   // Copied from libhpi
1418   const size_t pnamelen = pname ? strlen(pname) : 0;
1419 
1420   // Return error on buffer overflow.
1421   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1422     return retval;
1423   }
1424 
1425   if (pnamelen == 0) {
1426     snprintf(buffer, buflen, "lib%s.so", fname);
1427     retval = true;
1428   } else if (strchr(pname, *os::path_separator()) != NULL) {
1429     int n;
1430     char** pelements = split_path(pname, &n);
1431     if (pelements == NULL) {
1432       return false;
1433     }
1434     for (int i = 0; i < n; i++) {
1435       // Really shouldn't be NULL, but check can't hurt
1436       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1437         continue; // skip the empty path values
1438       }
1439       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1440       if (file_exists(buffer)) {
1441         retval = true;
1442         break;
1443       }
1444     }
1445     // release the storage
1446     for (int i = 0; i < n; i++) {
1447       if (pelements[i] != NULL) {
1448         FREE_C_HEAP_ARRAY(char, pelements[i]);
1449       }
1450     }
1451     if (pelements != NULL) {
1452       FREE_C_HEAP_ARRAY(char*, pelements);
1453     }
1454   } else {
1455     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1456     retval = true;
1457   }
1458   return retval;
1459 }
1460 
1461 // check if addr is inside libjvm.so
1462 bool os::address_is_in_vm(address addr) {
1463   static address libjvm_base_addr;
1464   Dl_info dlinfo;
1465 
1466   if (libjvm_base_addr == NULL) {
1467     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1468       libjvm_base_addr = (address)dlinfo.dli_fbase;
1469     }
1470     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1471   }
1472 
1473   if (dladdr((void *)addr, &dlinfo) != 0) {
1474     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1475   }
1476 
1477   return false;
1478 }
1479 
1480 bool os::dll_address_to_function_name(address addr, char *buf,
1481                                       int buflen, int *offset,
1482                                       bool demangle) {
1483   // buf is not optional, but offset is optional
1484   assert(buf != NULL, "sanity check");
1485 
1486   Dl_info dlinfo;
1487 
1488   if (dladdr((void*)addr, &dlinfo) != 0) {
1489     // see if we have a matching symbol
1490     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1491       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1492         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1493       }
1494       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1495       return true;
1496     }
1497     // no matching symbol so try for just file info
1498     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1499       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1500                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1501         return true;
1502       }
1503     }
1504   }
1505 
1506   buf[0] = '\0';
1507   if (offset != NULL) *offset = -1;
1508   return false;
1509 }
1510 
1511 struct _address_to_library_name {
1512   address addr;          // input : memory address
1513   size_t  buflen;        //         size of fname
1514   char*   fname;         // output: library name
1515   address base;          //         library base addr
1516 };
1517 
1518 static int address_to_library_name_callback(struct dl_phdr_info *info,
1519                                             size_t size, void *data) {
1520   int i;
1521   bool found = false;
1522   address libbase = NULL;
1523   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1524 
1525   // iterate through all loadable segments
1526   for (i = 0; i < info->dlpi_phnum; i++) {
1527     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1528     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1529       // base address of a library is the lowest address of its loaded
1530       // segments.
1531       if (libbase == NULL || libbase > segbase) {
1532         libbase = segbase;
1533       }
1534       // see if 'addr' is within current segment
1535       if (segbase <= d->addr &&
1536           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1537         found = true;
1538       }
1539     }
1540   }
1541 
1542   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1543   // so dll_address_to_library_name() can fall through to use dladdr() which
1544   // can figure out executable name from argv[0].
1545   if (found && info->dlpi_name && info->dlpi_name[0]) {
1546     d->base = libbase;
1547     if (d->fname) {
1548       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1549     }
1550     return 1;
1551   }
1552   return 0;
1553 }
1554 
1555 bool os::dll_address_to_library_name(address addr, char* buf,
1556                                      int buflen, int* offset) {
1557   // buf is not optional, but offset is optional
1558   assert(buf != NULL, "sanity check");
1559 
1560   Dl_info dlinfo;
1561   struct _address_to_library_name data;
1562 
1563   // There is a bug in old glibc dladdr() implementation that it could resolve
1564   // to wrong library name if the .so file has a base address != NULL. Here
1565   // we iterate through the program headers of all loaded libraries to find
1566   // out which library 'addr' really belongs to. This workaround can be
1567   // removed once the minimum requirement for glibc is moved to 2.3.x.
1568   data.addr = addr;
1569   data.fname = buf;
1570   data.buflen = buflen;
1571   data.base = NULL;
1572   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1573 
1574   if (rslt) {
1575     // buf already contains library name
1576     if (offset) *offset = addr - data.base;
1577     return true;
1578   }
1579   if (dladdr((void*)addr, &dlinfo) != 0) {
1580     if (dlinfo.dli_fname != NULL) {
1581       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1582     }
1583     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1584       *offset = addr - (address)dlinfo.dli_fbase;
1585     }
1586     return true;
1587   }
1588 
1589   buf[0] = '\0';
1590   if (offset) *offset = -1;
1591   return false;
1592 }
1593 
1594 // Loads .dll/.so and
1595 // in case of error it checks if .dll/.so was built for the
1596 // same architecture as Hotspot is running on
1597 
1598 
1599 // Remember the stack's state. The Linux dynamic linker will change
1600 // the stack to 'executable' at most once, so we must safepoint only once.
1601 bool os::Linux::_stack_is_executable = false;
1602 
1603 // VM operation that loads a library.  This is necessary if stack protection
1604 // of the Java stacks can be lost during loading the library.  If we
1605 // do not stop the Java threads, they can stack overflow before the stacks
1606 // are protected again.
1607 class VM_LinuxDllLoad: public VM_Operation {
1608  private:
1609   const char *_filename;
1610   char *_ebuf;
1611   int _ebuflen;
1612   void *_lib;
1613  public:
1614   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1615     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1616   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1617   void doit() {
1618     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1619     os::Linux::_stack_is_executable = true;
1620   }
1621   void* loaded_library() { return _lib; }
1622 };
1623 
1624 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1625   void * result = NULL;
1626   bool load_attempted = false;
1627 
1628   // Check whether the library to load might change execution rights
1629   // of the stack. If they are changed, the protection of the stack
1630   // guard pages will be lost. We need a safepoint to fix this.
1631   //
1632   // See Linux man page execstack(8) for more info.
1633   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1634     ElfFile ef(filename);
1635     if (!ef.specifies_noexecstack()) {
1636       if (!is_init_completed()) {
1637         os::Linux::_stack_is_executable = true;
1638         // This is OK - No Java threads have been created yet, and hence no
1639         // stack guard pages to fix.
1640         //
1641         // This should happen only when you are building JDK7 using a very
1642         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1643         //
1644         // Dynamic loader will make all stacks executable after
1645         // this function returns, and will not do that again.
1646         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1647       } else {
1648         warning("You have loaded library %s which might have disabled stack guard. "
1649                 "The VM will try to fix the stack guard now.\n"
1650                 "It's highly recommended that you fix the library with "
1651                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1652                 filename);
1653 
1654         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1655         JavaThread *jt = JavaThread::current();
1656         if (jt->thread_state() != _thread_in_native) {
1657           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1658           // that requires ExecStack. Cannot enter safe point. Let's give up.
1659           warning("Unable to fix stack guard. Giving up.");
1660         } else {
1661           if (!LoadExecStackDllInVMThread) {
1662             // This is for the case where the DLL has an static
1663             // constructor function that executes JNI code. We cannot
1664             // load such DLLs in the VMThread.
1665             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1666           }
1667 
1668           ThreadInVMfromNative tiv(jt);
1669           debug_only(VMNativeEntryWrapper vew;)
1670 
1671           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1672           VMThread::execute(&op);
1673           if (LoadExecStackDllInVMThread) {
1674             result = op.loaded_library();
1675           }
1676           load_attempted = true;
1677         }
1678       }
1679     }
1680   }
1681 
1682   if (!load_attempted) {
1683     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1684   }
1685 
1686   if (result != NULL) {
1687     // Successful loading
1688     return result;
1689   }
1690 
1691   Elf32_Ehdr elf_head;
1692   int diag_msg_max_length=ebuflen-strlen(ebuf);
1693   char* diag_msg_buf=ebuf+strlen(ebuf);
1694 
1695   if (diag_msg_max_length==0) {
1696     // No more space in ebuf for additional diagnostics message
1697     return NULL;
1698   }
1699 
1700 
1701   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1702 
1703   if (file_descriptor < 0) {
1704     // Can't open library, report dlerror() message
1705     return NULL;
1706   }
1707 
1708   bool failed_to_read_elf_head=
1709     (sizeof(elf_head)!=
1710      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1711 
1712   ::close(file_descriptor);
1713   if (failed_to_read_elf_head) {
1714     // file i/o error - report dlerror() msg
1715     return NULL;
1716   }
1717 
1718   typedef struct {
1719     Elf32_Half    code;         // Actual value as defined in elf.h
1720     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1721     unsigned char elf_class;    // 32 or 64 bit
1722     unsigned char endianess;    // MSB or LSB
1723     char*         name;         // String representation
1724   } arch_t;
1725 
1726 #ifndef EM_486
1727   #define EM_486          6               /* Intel 80486 */
1728 #endif
1729 #ifndef EM_AARCH64
1730   #define EM_AARCH64    183               /* ARM AARCH64 */
1731 #endif
1732 
1733   static const arch_t arch_array[]={
1734     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1735     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1736     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1737     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1738     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1739     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1740     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1741     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1742 #if defined(VM_LITTLE_ENDIAN)
1743     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1744 #else
1745     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1746 #endif
1747     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1748     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1749     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1750     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1751     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1752     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1753     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1754     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1755   };
1756 
1757 #if  (defined IA32)
1758   static  Elf32_Half running_arch_code=EM_386;
1759 #elif   (defined AMD64)
1760   static  Elf32_Half running_arch_code=EM_X86_64;
1761 #elif  (defined IA64)
1762   static  Elf32_Half running_arch_code=EM_IA_64;
1763 #elif  (defined __sparc) && (defined _LP64)
1764   static  Elf32_Half running_arch_code=EM_SPARCV9;
1765 #elif  (defined __sparc) && (!defined _LP64)
1766   static  Elf32_Half running_arch_code=EM_SPARC;
1767 #elif  (defined __powerpc64__)
1768   static  Elf32_Half running_arch_code=EM_PPC64;
1769 #elif  (defined __powerpc__)
1770   static  Elf32_Half running_arch_code=EM_PPC;
1771 #elif  (defined AARCH64)
1772   static  Elf32_Half running_arch_code=EM_AARCH64;
1773 #elif  (defined ARM)
1774   static  Elf32_Half running_arch_code=EM_ARM;
1775 #elif  (defined S390)
1776   static  Elf32_Half running_arch_code=EM_S390;
1777 #elif  (defined ALPHA)
1778   static  Elf32_Half running_arch_code=EM_ALPHA;
1779 #elif  (defined MIPSEL)
1780   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1781 #elif  (defined PARISC)
1782   static  Elf32_Half running_arch_code=EM_PARISC;
1783 #elif  (defined MIPS)
1784   static  Elf32_Half running_arch_code=EM_MIPS;
1785 #elif  (defined M68K)
1786   static  Elf32_Half running_arch_code=EM_68K;
1787 #else
1788     #error Method os::dll_load requires that one of following is defined:\
1789         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, __sparc
1790 #endif
1791 
1792   // Identify compatability class for VM's architecture and library's architecture
1793   // Obtain string descriptions for architectures
1794 
1795   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1796   int running_arch_index=-1;
1797 
1798   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1799     if (running_arch_code == arch_array[i].code) {
1800       running_arch_index    = i;
1801     }
1802     if (lib_arch.code == arch_array[i].code) {
1803       lib_arch.compat_class = arch_array[i].compat_class;
1804       lib_arch.name         = arch_array[i].name;
1805     }
1806   }
1807 
1808   assert(running_arch_index != -1,
1809          "Didn't find running architecture code (running_arch_code) in arch_array");
1810   if (running_arch_index == -1) {
1811     // Even though running architecture detection failed
1812     // we may still continue with reporting dlerror() message
1813     return NULL;
1814   }
1815 
1816   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1817     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1818     return NULL;
1819   }
1820 
1821 #ifndef S390
1822   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1823     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1824     return NULL;
1825   }
1826 #endif // !S390
1827 
1828   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1829     if (lib_arch.name!=NULL) {
1830       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1831                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1832                  lib_arch.name, arch_array[running_arch_index].name);
1833     } else {
1834       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1835                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1836                  lib_arch.code,
1837                  arch_array[running_arch_index].name);
1838     }
1839   }
1840 
1841   return NULL;
1842 }
1843 
1844 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1845                                 int ebuflen) {
1846   void * result = ::dlopen(filename, RTLD_LAZY);
1847   if (result == NULL) {
1848     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1849     ebuf[ebuflen-1] = '\0';
1850   }
1851   return result;
1852 }
1853 
1854 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1855                                        int ebuflen) {
1856   void * result = NULL;
1857   if (LoadExecStackDllInVMThread) {
1858     result = dlopen_helper(filename, ebuf, ebuflen);
1859   }
1860 
1861   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1862   // library that requires an executable stack, or which does not have this
1863   // stack attribute set, dlopen changes the stack attribute to executable. The
1864   // read protection of the guard pages gets lost.
1865   //
1866   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1867   // may have been queued at the same time.
1868 
1869   if (!_stack_is_executable) {
1870     JavaThread *jt = Threads::first();
1871 
1872     while (jt) {
1873       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1874           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1875         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1876           warning("Attempt to reguard stack yellow zone failed.");
1877         }
1878       }
1879       jt = jt->next();
1880     }
1881   }
1882 
1883   return result;
1884 }
1885 
1886 void* os::dll_lookup(void* handle, const char* name) {
1887   void* res = dlsym(handle, name);
1888   return res;
1889 }
1890 
1891 void* os::get_default_process_handle() {
1892   return (void*)::dlopen(NULL, RTLD_LAZY);
1893 }
1894 
1895 static bool _print_ascii_file(const char* filename, outputStream* st) {
1896   int fd = ::open(filename, O_RDONLY);
1897   if (fd == -1) {
1898     return false;
1899   }
1900 
1901   char buf[33];
1902   int bytes;
1903   buf[32] = '\0';
1904   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1905     st->print_raw(buf, bytes);
1906   }
1907 
1908   ::close(fd);
1909 
1910   return true;
1911 }
1912 
1913 void os::print_dll_info(outputStream *st) {
1914   st->print_cr("Dynamic libraries:");
1915 
1916   char fname[32];
1917   pid_t pid = os::Linux::gettid();
1918 
1919   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1920 
1921   if (!_print_ascii_file(fname, st)) {
1922     st->print("Can not get library information for pid = %d\n", pid);
1923   }
1924 }
1925 
1926 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1927   FILE *procmapsFile = NULL;
1928 
1929   // Open the procfs maps file for the current process
1930   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1931     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1932     char line[PATH_MAX + 100];
1933 
1934     // Read line by line from 'file'
1935     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1936       u8 base, top, offset, inode;
1937       char permissions[5];
1938       char device[6];
1939       char name[PATH_MAX + 1];
1940 
1941       // Parse fields from line
1942       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1943              &base, &top, permissions, &offset, device, &inode, name);
1944 
1945       // Filter by device id '00:00' so that we only get file system mapped files.
1946       if (strcmp(device, "00:00") != 0) {
1947 
1948         // Call callback with the fields of interest
1949         if(callback(name, (address)base, (address)top, param)) {
1950           // Oops abort, callback aborted
1951           fclose(procmapsFile);
1952           return 1;
1953         }
1954       }
1955     }
1956     fclose(procmapsFile);
1957   }
1958   return 0;
1959 }
1960 
1961 void os::print_os_info_brief(outputStream* st) {
1962   os::Linux::print_distro_info(st);
1963 
1964   os::Posix::print_uname_info(st);
1965 
1966   os::Linux::print_libversion_info(st);
1967 
1968 }
1969 
1970 void os::print_os_info(outputStream* st) {
1971   st->print("OS:");
1972 
1973   os::Linux::print_distro_info(st);
1974 
1975   os::Posix::print_uname_info(st);
1976 
1977   // Print warning if unsafe chroot environment detected
1978   if (unsafe_chroot_detected) {
1979     st->print("WARNING!! ");
1980     st->print_cr("%s", unstable_chroot_error);
1981   }
1982 
1983   os::Linux::print_libversion_info(st);
1984 
1985   os::Posix::print_rlimit_info(st);
1986 
1987   os::Posix::print_load_average(st);
1988 
1989   os::Linux::print_full_memory_info(st);
1990 }
1991 
1992 // Try to identify popular distros.
1993 // Most Linux distributions have a /etc/XXX-release file, which contains
1994 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1995 // file that also contains the OS version string. Some have more than one
1996 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1997 // /etc/redhat-release.), so the order is important.
1998 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1999 // their own specific XXX-release file as well as a redhat-release file.
2000 // Because of this the XXX-release file needs to be searched for before the
2001 // redhat-release file.
2002 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2003 // search for redhat-release / SuSE-release needs to be before lsb-release.
2004 // Since the lsb-release file is the new standard it needs to be searched
2005 // before the older style release files.
2006 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2007 // next to last resort.  The os-release file is a new standard that contains
2008 // distribution information and the system-release file seems to be an old
2009 // standard that has been replaced by the lsb-release and os-release files.
2010 // Searching for the debian_version file is the last resort.  It contains
2011 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2012 // "Debian " is printed before the contents of the debian_version file.
2013 
2014 const char* distro_files[] = {
2015   "/etc/oracle-release",
2016   "/etc/mandriva-release",
2017   "/etc/mandrake-release",
2018   "/etc/sun-release",
2019   "/etc/redhat-release",
2020   "/etc/SuSE-release",
2021   "/etc/lsb-release",
2022   "/etc/turbolinux-release",
2023   "/etc/gentoo-release",
2024   "/etc/ltib-release",
2025   "/etc/angstrom-version",
2026   "/etc/system-release",
2027   "/etc/os-release",
2028   NULL };
2029 
2030 void os::Linux::print_distro_info(outputStream* st) {
2031   for (int i = 0;; i++) {
2032     const char* file = distro_files[i];
2033     if (file == NULL) {
2034       break;  // done
2035     }
2036     // If file prints, we found it.
2037     if (_print_ascii_file(file, st)) {
2038       return;
2039     }
2040   }
2041 
2042   if (file_exists("/etc/debian_version")) {
2043     st->print("Debian ");
2044     _print_ascii_file("/etc/debian_version", st);
2045   } else {
2046     st->print("Linux");
2047   }
2048   st->cr();
2049 }
2050 
2051 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2052   char buf[256];
2053   while (fgets(buf, sizeof(buf), fp)) {
2054     // Edit out extra stuff in expected format
2055     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2056       char* ptr = strstr(buf, "\"");  // the name is in quotes
2057       if (ptr != NULL) {
2058         ptr++; // go beyond first quote
2059         char* nl = strchr(ptr, '\"');
2060         if (nl != NULL) *nl = '\0';
2061         strncpy(distro, ptr, length);
2062       } else {
2063         ptr = strstr(buf, "=");
2064         ptr++; // go beyond equals then
2065         char* nl = strchr(ptr, '\n');
2066         if (nl != NULL) *nl = '\0';
2067         strncpy(distro, ptr, length);
2068       }
2069       return;
2070     } else if (get_first_line) {
2071       char* nl = strchr(buf, '\n');
2072       if (nl != NULL) *nl = '\0';
2073       strncpy(distro, buf, length);
2074       return;
2075     }
2076   }
2077   // print last line and close
2078   char* nl = strchr(buf, '\n');
2079   if (nl != NULL) *nl = '\0';
2080   strncpy(distro, buf, length);
2081 }
2082 
2083 static void parse_os_info(char* distro, size_t length, const char* file) {
2084   FILE* fp = fopen(file, "r");
2085   if (fp != NULL) {
2086     // if suse format, print out first line
2087     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2088     parse_os_info_helper(fp, distro, length, get_first_line);
2089     fclose(fp);
2090   }
2091 }
2092 
2093 void os::get_summary_os_info(char* buf, size_t buflen) {
2094   for (int i = 0;; i++) {
2095     const char* file = distro_files[i];
2096     if (file == NULL) {
2097       break; // ran out of distro_files
2098     }
2099     if (file_exists(file)) {
2100       parse_os_info(buf, buflen, file);
2101       return;
2102     }
2103   }
2104   // special case for debian
2105   if (file_exists("/etc/debian_version")) {
2106     strncpy(buf, "Debian ", buflen);
2107     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2108   } else {
2109     strncpy(buf, "Linux", buflen);
2110   }
2111 }
2112 
2113 void os::Linux::print_libversion_info(outputStream* st) {
2114   // libc, pthread
2115   st->print("libc:");
2116   st->print("%s ", os::Linux::glibc_version());
2117   st->print("%s ", os::Linux::libpthread_version());
2118   st->cr();
2119 }
2120 
2121 void os::Linux::print_full_memory_info(outputStream* st) {
2122   st->print("\n/proc/meminfo:\n");
2123   _print_ascii_file("/proc/meminfo", st);
2124   st->cr();
2125 }
2126 
2127 void os::print_memory_info(outputStream* st) {
2128 
2129   st->print("Memory:");
2130   st->print(" %dk page", os::vm_page_size()>>10);
2131 
2132   // values in struct sysinfo are "unsigned long"
2133   struct sysinfo si;
2134   sysinfo(&si);
2135 
2136   st->print(", physical " UINT64_FORMAT "k",
2137             os::physical_memory() >> 10);
2138   st->print("(" UINT64_FORMAT "k free)",
2139             os::available_memory() >> 10);
2140   st->print(", swap " UINT64_FORMAT "k",
2141             ((jlong)si.totalswap * si.mem_unit) >> 10);
2142   st->print("(" UINT64_FORMAT "k free)",
2143             ((jlong)si.freeswap * si.mem_unit) >> 10);
2144   st->cr();
2145 }
2146 
2147 // Print the first "model name" line and the first "flags" line
2148 // that we find and nothing more. We assume "model name" comes
2149 // before "flags" so if we find a second "model name", then the
2150 // "flags" field is considered missing.
2151 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2152 #if defined(IA32) || defined(AMD64)
2153   // Other platforms have less repetitive cpuinfo files
2154   FILE *fp = fopen("/proc/cpuinfo", "r");
2155   if (fp) {
2156     while (!feof(fp)) {
2157       if (fgets(buf, buflen, fp)) {
2158         // Assume model name comes before flags
2159         bool model_name_printed = false;
2160         if (strstr(buf, "model name") != NULL) {
2161           if (!model_name_printed) {
2162             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2163             st->print_raw(buf);
2164             model_name_printed = true;
2165           } else {
2166             // model name printed but not flags?  Odd, just return
2167             fclose(fp);
2168             return true;
2169           }
2170         }
2171         // print the flags line too
2172         if (strstr(buf, "flags") != NULL) {
2173           st->print_raw(buf);
2174           fclose(fp);
2175           return true;
2176         }
2177       }
2178     }
2179     fclose(fp);
2180   }
2181 #endif // x86 platforms
2182   return false;
2183 }
2184 
2185 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2186   // Only print the model name if the platform provides this as a summary
2187   if (!print_model_name_and_flags(st, buf, buflen)) {
2188     st->print("\n/proc/cpuinfo:\n");
2189     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2190       st->print_cr("  <Not Available>");
2191     }
2192   }
2193 }
2194 
2195 #if defined(AMD64) || defined(IA32) || defined(X32)
2196 const char* search_string = "model name";
2197 #elif defined(PPC64)
2198 const char* search_string = "cpu";
2199 #elif defined(S390)
2200 const char* search_string = "processor";
2201 #elif defined(SPARC)
2202 const char* search_string = "cpu";
2203 #else
2204 const char* search_string = "Processor";
2205 #endif
2206 
2207 // Parses the cpuinfo file for string representing the model name.
2208 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2209   FILE* fp = fopen("/proc/cpuinfo", "r");
2210   if (fp != NULL) {
2211     while (!feof(fp)) {
2212       char buf[256];
2213       if (fgets(buf, sizeof(buf), fp)) {
2214         char* start = strstr(buf, search_string);
2215         if (start != NULL) {
2216           char *ptr = start + strlen(search_string);
2217           char *end = buf + strlen(buf);
2218           while (ptr != end) {
2219              // skip whitespace and colon for the rest of the name.
2220              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2221                break;
2222              }
2223              ptr++;
2224           }
2225           if (ptr != end) {
2226             // reasonable string, get rid of newline and keep the rest
2227             char* nl = strchr(buf, '\n');
2228             if (nl != NULL) *nl = '\0';
2229             strncpy(cpuinfo, ptr, length);
2230             fclose(fp);
2231             return;
2232           }
2233         }
2234       }
2235     }
2236     fclose(fp);
2237   }
2238   // cpuinfo not found or parsing failed, just print generic string.  The entire
2239   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2240 #if   defined(AARCH64)
2241   strncpy(cpuinfo, "AArch64", length);
2242 #elif defined(AMD64)
2243   strncpy(cpuinfo, "x86_64", length);
2244 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2245   strncpy(cpuinfo, "ARM", length);
2246 #elif defined(IA32)
2247   strncpy(cpuinfo, "x86_32", length);
2248 #elif defined(IA64)
2249   strncpy(cpuinfo, "IA64", length);
2250 #elif defined(PPC)
2251   strncpy(cpuinfo, "PPC64", length);
2252 #elif defined(S390)
2253   strncpy(cpuinfo, "S390", length);
2254 #elif defined(SPARC)
2255   strncpy(cpuinfo, "sparcv9", length);
2256 #elif defined(ZERO_LIBARCH)
2257   strncpy(cpuinfo, ZERO_LIBARCH, length);
2258 #else
2259   strncpy(cpuinfo, "unknown", length);
2260 #endif
2261 }
2262 
2263 static void print_signal_handler(outputStream* st, int sig,
2264                                  char* buf, size_t buflen);
2265 
2266 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2267   st->print_cr("Signal Handlers:");
2268   print_signal_handler(st, SIGSEGV, buf, buflen);
2269   print_signal_handler(st, SIGBUS , buf, buflen);
2270   print_signal_handler(st, SIGFPE , buf, buflen);
2271   print_signal_handler(st, SIGPIPE, buf, buflen);
2272   print_signal_handler(st, SIGXFSZ, buf, buflen);
2273   print_signal_handler(st, SIGILL , buf, buflen);
2274   print_signal_handler(st, SR_signum, buf, buflen);
2275   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2276   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2277   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2278   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2279 #if defined(PPC64)
2280   print_signal_handler(st, SIGTRAP, buf, buflen);
2281 #endif
2282 }
2283 
2284 static char saved_jvm_path[MAXPATHLEN] = {0};
2285 
2286 // Find the full path to the current module, libjvm.so
2287 void os::jvm_path(char *buf, jint buflen) {
2288   // Error checking.
2289   if (buflen < MAXPATHLEN) {
2290     assert(false, "must use a large-enough buffer");
2291     buf[0] = '\0';
2292     return;
2293   }
2294   // Lazy resolve the path to current module.
2295   if (saved_jvm_path[0] != 0) {
2296     strcpy(buf, saved_jvm_path);
2297     return;
2298   }
2299 
2300   char dli_fname[MAXPATHLEN];
2301   bool ret = dll_address_to_library_name(
2302                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2303                                          dli_fname, sizeof(dli_fname), NULL);
2304   assert(ret, "cannot locate libjvm");
2305   char *rp = NULL;
2306   if (ret && dli_fname[0] != '\0') {
2307     rp = realpath(dli_fname, buf);
2308   }
2309   if (rp == NULL) {
2310     return;
2311   }
2312 
2313   if (Arguments::sun_java_launcher_is_altjvm()) {
2314     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2315     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2316     // If "/jre/lib/" appears at the right place in the string, then
2317     // assume we are installed in a JDK and we're done. Otherwise, check
2318     // for a JAVA_HOME environment variable and fix up the path so it
2319     // looks like libjvm.so is installed there (append a fake suffix
2320     // hotspot/libjvm.so).
2321     const char *p = buf + strlen(buf) - 1;
2322     for (int count = 0; p > buf && count < 5; ++count) {
2323       for (--p; p > buf && *p != '/'; --p)
2324         /* empty */ ;
2325     }
2326 
2327     if (strncmp(p, "/jre/lib/", 9) != 0) {
2328       // Look for JAVA_HOME in the environment.
2329       char* java_home_var = ::getenv("JAVA_HOME");
2330       if (java_home_var != NULL && java_home_var[0] != 0) {
2331         char* jrelib_p;
2332         int len;
2333 
2334         // Check the current module name "libjvm.so".
2335         p = strrchr(buf, '/');
2336         if (p == NULL) {
2337           return;
2338         }
2339         assert(strstr(p, "/libjvm") == p, "invalid library name");
2340 
2341         rp = realpath(java_home_var, buf);
2342         if (rp == NULL) {
2343           return;
2344         }
2345 
2346         // determine if this is a legacy image or modules image
2347         // modules image doesn't have "jre" subdirectory
2348         len = strlen(buf);
2349         assert(len < buflen, "Ran out of buffer room");
2350         jrelib_p = buf + len;
2351         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2352         if (0 != access(buf, F_OK)) {
2353           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2354         }
2355 
2356         if (0 == access(buf, F_OK)) {
2357           // Use current module name "libjvm.so"
2358           len = strlen(buf);
2359           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2360         } else {
2361           // Go back to path of .so
2362           rp = realpath(dli_fname, buf);
2363           if (rp == NULL) {
2364             return;
2365           }
2366         }
2367       }
2368     }
2369   }
2370 
2371   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2372   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2373 }
2374 
2375 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2376   // no prefix required, not even "_"
2377 }
2378 
2379 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2380   // no suffix required
2381 }
2382 
2383 ////////////////////////////////////////////////////////////////////////////////
2384 // sun.misc.Signal support
2385 
2386 static volatile jint sigint_count = 0;
2387 
2388 static void UserHandler(int sig, void *siginfo, void *context) {
2389   // 4511530 - sem_post is serialized and handled by the manager thread. When
2390   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2391   // don't want to flood the manager thread with sem_post requests.
2392   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2393     return;
2394   }
2395 
2396   // Ctrl-C is pressed during error reporting, likely because the error
2397   // handler fails to abort. Let VM die immediately.
2398   if (sig == SIGINT && is_error_reported()) {
2399     os::die();
2400   }
2401 
2402   os::signal_notify(sig);
2403 }
2404 
2405 void* os::user_handler() {
2406   return CAST_FROM_FN_PTR(void*, UserHandler);
2407 }
2408 
2409 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2410   struct timespec ts;
2411   // Semaphore's are always associated with CLOCK_REALTIME
2412   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2413   // see unpackTime for discussion on overflow checking
2414   if (sec >= MAX_SECS) {
2415     ts.tv_sec += MAX_SECS;
2416     ts.tv_nsec = 0;
2417   } else {
2418     ts.tv_sec += sec;
2419     ts.tv_nsec += nsec;
2420     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2421       ts.tv_nsec -= NANOSECS_PER_SEC;
2422       ++ts.tv_sec; // note: this must be <= max_secs
2423     }
2424   }
2425 
2426   return ts;
2427 }
2428 
2429 extern "C" {
2430   typedef void (*sa_handler_t)(int);
2431   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2432 }
2433 
2434 void* os::signal(int signal_number, void* handler) {
2435   struct sigaction sigAct, oldSigAct;
2436 
2437   sigfillset(&(sigAct.sa_mask));
2438   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2439   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2440 
2441   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2442     // -1 means registration failed
2443     return (void *)-1;
2444   }
2445 
2446   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2447 }
2448 
2449 void os::signal_raise(int signal_number) {
2450   ::raise(signal_number);
2451 }
2452 
2453 // The following code is moved from os.cpp for making this
2454 // code platform specific, which it is by its very nature.
2455 
2456 // Will be modified when max signal is changed to be dynamic
2457 int os::sigexitnum_pd() {
2458   return NSIG;
2459 }
2460 
2461 // a counter for each possible signal value
2462 static volatile jint pending_signals[NSIG+1] = { 0 };
2463 
2464 // Linux(POSIX) specific hand shaking semaphore.
2465 static sem_t sig_sem;
2466 static PosixSemaphore sr_semaphore;
2467 
2468 void os::signal_init_pd() {
2469   // Initialize signal structures
2470   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2471 
2472   // Initialize signal semaphore
2473   ::sem_init(&sig_sem, 0, 0);
2474 }
2475 
2476 void os::signal_notify(int sig) {
2477   Atomic::inc(&pending_signals[sig]);
2478   ::sem_post(&sig_sem);
2479 }
2480 
2481 static int check_pending_signals(bool wait) {
2482   Atomic::store(0, &sigint_count);
2483   for (;;) {
2484     for (int i = 0; i < NSIG + 1; i++) {
2485       jint n = pending_signals[i];
2486       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2487         return i;
2488       }
2489     }
2490     if (!wait) {
2491       return -1;
2492     }
2493     JavaThread *thread = JavaThread::current();
2494     ThreadBlockInVM tbivm(thread);
2495 
2496     bool threadIsSuspended;
2497     do {
2498       thread->set_suspend_equivalent();
2499       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2500       ::sem_wait(&sig_sem);
2501 
2502       // were we externally suspended while we were waiting?
2503       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2504       if (threadIsSuspended) {
2505         // The semaphore has been incremented, but while we were waiting
2506         // another thread suspended us. We don't want to continue running
2507         // while suspended because that would surprise the thread that
2508         // suspended us.
2509         ::sem_post(&sig_sem);
2510 
2511         thread->java_suspend_self();
2512       }
2513     } while (threadIsSuspended);
2514   }
2515 }
2516 
2517 int os::signal_lookup() {
2518   return check_pending_signals(false);
2519 }
2520 
2521 int os::signal_wait() {
2522   return check_pending_signals(true);
2523 }
2524 
2525 ////////////////////////////////////////////////////////////////////////////////
2526 // Virtual Memory
2527 
2528 int os::vm_page_size() {
2529   // Seems redundant as all get out
2530   assert(os::Linux::page_size() != -1, "must call os::init");
2531   return os::Linux::page_size();
2532 }
2533 
2534 // Solaris allocates memory by pages.
2535 int os::vm_allocation_granularity() {
2536   assert(os::Linux::page_size() != -1, "must call os::init");
2537   return os::Linux::page_size();
2538 }
2539 
2540 // Rationale behind this function:
2541 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2542 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2543 //  samples for JITted code. Here we create private executable mapping over the code cache
2544 //  and then we can use standard (well, almost, as mapping can change) way to provide
2545 //  info for the reporting script by storing timestamp and location of symbol
2546 void linux_wrap_code(char* base, size_t size) {
2547   static volatile jint cnt = 0;
2548 
2549   if (!UseOprofile) {
2550     return;
2551   }
2552 
2553   char buf[PATH_MAX+1];
2554   int num = Atomic::add(1, &cnt);
2555 
2556   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2557            os::get_temp_directory(), os::current_process_id(), num);
2558   unlink(buf);
2559 
2560   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2561 
2562   if (fd != -1) {
2563     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2564     if (rv != (off_t)-1) {
2565       if (::write(fd, "", 1) == 1) {
2566         mmap(base, size,
2567              PROT_READ|PROT_WRITE|PROT_EXEC,
2568              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2569       }
2570     }
2571     ::close(fd);
2572     unlink(buf);
2573   }
2574 }
2575 
2576 static bool recoverable_mmap_error(int err) {
2577   // See if the error is one we can let the caller handle. This
2578   // list of errno values comes from JBS-6843484. I can't find a
2579   // Linux man page that documents this specific set of errno
2580   // values so while this list currently matches Solaris, it may
2581   // change as we gain experience with this failure mode.
2582   switch (err) {
2583   case EBADF:
2584   case EINVAL:
2585   case ENOTSUP:
2586     // let the caller deal with these errors
2587     return true;
2588 
2589   default:
2590     // Any remaining errors on this OS can cause our reserved mapping
2591     // to be lost. That can cause confusion where different data
2592     // structures think they have the same memory mapped. The worst
2593     // scenario is if both the VM and a library think they have the
2594     // same memory mapped.
2595     return false;
2596   }
2597 }
2598 
2599 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2600                                     int err) {
2601   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2602           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2603           os::strerror(err), err);
2604 }
2605 
2606 static void warn_fail_commit_memory(char* addr, size_t size,
2607                                     size_t alignment_hint, bool exec,
2608                                     int err) {
2609   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2610           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2611           alignment_hint, exec, os::strerror(err), err);
2612 }
2613 
2614 // NOTE: Linux kernel does not really reserve the pages for us.
2615 //       All it does is to check if there are enough free pages
2616 //       left at the time of mmap(). This could be a potential
2617 //       problem.
2618 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2619   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2620   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2621                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2622   if (res != (uintptr_t) MAP_FAILED) {
2623     if (UseNUMAInterleaving) {
2624       numa_make_global(addr, size);
2625     }
2626     return 0;
2627   }
2628 
2629   int err = errno;  // save errno from mmap() call above
2630 
2631   if (!recoverable_mmap_error(err)) {
2632     warn_fail_commit_memory(addr, size, exec, err);
2633     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2634   }
2635 
2636   return err;
2637 }
2638 
2639 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2640   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2641 }
2642 
2643 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2644                                   const char* mesg) {
2645   assert(mesg != NULL, "mesg must be specified");
2646   int err = os::Linux::commit_memory_impl(addr, size, exec);
2647   if (err != 0) {
2648     // the caller wants all commit errors to exit with the specified mesg:
2649     warn_fail_commit_memory(addr, size, exec, err);
2650     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2651   }
2652 }
2653 
2654 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2655 #ifndef MAP_HUGETLB
2656   #define MAP_HUGETLB 0x40000
2657 #endif
2658 
2659 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2660 #ifndef MADV_HUGEPAGE
2661   #define MADV_HUGEPAGE 14
2662 #endif
2663 
2664 int os::Linux::commit_memory_impl(char* addr, size_t size,
2665                                   size_t alignment_hint, bool exec) {
2666   int err = os::Linux::commit_memory_impl(addr, size, exec);
2667   if (err == 0) {
2668     realign_memory(addr, size, alignment_hint);
2669   }
2670   return err;
2671 }
2672 
2673 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2674                           bool exec) {
2675   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2676 }
2677 
2678 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2679                                   size_t alignment_hint, bool exec,
2680                                   const char* mesg) {
2681   assert(mesg != NULL, "mesg must be specified");
2682   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2683   if (err != 0) {
2684     // the caller wants all commit errors to exit with the specified mesg:
2685     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2686     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2687   }
2688 }
2689 
2690 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2691   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2692     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2693     // be supported or the memory may already be backed by huge pages.
2694     ::madvise(addr, bytes, MADV_HUGEPAGE);
2695   }
2696 }
2697 
2698 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2699   // This method works by doing an mmap over an existing mmaping and effectively discarding
2700   // the existing pages. However it won't work for SHM-based large pages that cannot be
2701   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2702   // small pages on top of the SHM segment. This method always works for small pages, so we
2703   // allow that in any case.
2704   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2705     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2706   }
2707 }
2708 
2709 void os::numa_make_global(char *addr, size_t bytes) {
2710   Linux::numa_interleave_memory(addr, bytes);
2711 }
2712 
2713 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2714 // bind policy to MPOL_PREFERRED for the current thread.
2715 #define USE_MPOL_PREFERRED 0
2716 
2717 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2718   // To make NUMA and large pages more robust when both enabled, we need to ease
2719   // the requirements on where the memory should be allocated. MPOL_BIND is the
2720   // default policy and it will force memory to be allocated on the specified
2721   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2722   // the specified node, but will not force it. Using this policy will prevent
2723   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2724   // free large pages.
2725   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2726   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2727 }
2728 
2729 bool os::numa_topology_changed() { return false; }
2730 
2731 size_t os::numa_get_groups_num() {
2732   int max_node = Linux::numa_max_node();
2733   return max_node > 0 ? max_node + 1 : 1;
2734 }
2735 
2736 int os::numa_get_group_id() {
2737   int cpu_id = Linux::sched_getcpu();
2738   if (cpu_id != -1) {
2739     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2740     if (lgrp_id != -1) {
2741       return lgrp_id;
2742     }
2743   }
2744   return 0;
2745 }
2746 
2747 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2748   for (size_t i = 0; i < size; i++) {
2749     ids[i] = i;
2750   }
2751   return size;
2752 }
2753 
2754 bool os::get_page_info(char *start, page_info* info) {
2755   return false;
2756 }
2757 
2758 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2759                      page_info* page_found) {
2760   return end;
2761 }
2762 
2763 
2764 int os::Linux::sched_getcpu_syscall(void) {
2765   unsigned int cpu = 0;
2766   int retval = -1;
2767 
2768 #if defined(IA32)
2769   #ifndef SYS_getcpu
2770     #define SYS_getcpu 318
2771   #endif
2772   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2773 #elif defined(AMD64)
2774 // Unfortunately we have to bring all these macros here from vsyscall.h
2775 // to be able to compile on old linuxes.
2776   #define __NR_vgetcpu 2
2777   #define VSYSCALL_START (-10UL << 20)
2778   #define VSYSCALL_SIZE 1024
2779   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2780   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2781   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2782   retval = vgetcpu(&cpu, NULL, NULL);
2783 #endif
2784 
2785   return (retval == -1) ? retval : cpu;
2786 }
2787 
2788 // Something to do with the numa-aware allocator needs these symbols
2789 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2790 extern "C" JNIEXPORT void numa_error(char *where) { }
2791 
2792 
2793 // If we are running with libnuma version > 2, then we should
2794 // be trying to use symbols with versions 1.1
2795 // If we are running with earlier version, which did not have symbol versions,
2796 // we should use the base version.
2797 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2798   void *f = dlvsym(handle, name, "libnuma_1.1");
2799   if (f == NULL) {
2800     f = dlsym(handle, name);
2801   }
2802   return f;
2803 }
2804 
2805 bool os::Linux::libnuma_init() {
2806   // sched_getcpu() should be in libc.
2807   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2808                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2809 
2810   // If it's not, try a direct syscall.
2811   if (sched_getcpu() == -1) {
2812     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2813                                     (void*)&sched_getcpu_syscall));
2814   }
2815 
2816   if (sched_getcpu() != -1) { // Does it work?
2817     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2818     if (handle != NULL) {
2819       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2820                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2821       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2822                                        libnuma_dlsym(handle, "numa_max_node")));
2823       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2824                                         libnuma_dlsym(handle, "numa_available")));
2825       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2826                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2827       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2828                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2829       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2830                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2831 
2832 
2833       if (numa_available() != -1) {
2834         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2835         // Create a cpu -> node mapping
2836         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2837         rebuild_cpu_to_node_map();
2838         return true;
2839       }
2840     }
2841   }
2842   return false;
2843 }
2844 
2845 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2846 // The table is later used in get_node_by_cpu().
2847 void os::Linux::rebuild_cpu_to_node_map() {
2848   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2849                               // in libnuma (possible values are starting from 16,
2850                               // and continuing up with every other power of 2, but less
2851                               // than the maximum number of CPUs supported by kernel), and
2852                               // is a subject to change (in libnuma version 2 the requirements
2853                               // are more reasonable) we'll just hardcode the number they use
2854                               // in the library.
2855   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2856 
2857   size_t cpu_num = processor_count();
2858   size_t cpu_map_size = NCPUS / BitsPerCLong;
2859   size_t cpu_map_valid_size =
2860     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2861 
2862   cpu_to_node()->clear();
2863   cpu_to_node()->at_grow(cpu_num - 1);
2864   size_t node_num = numa_get_groups_num();
2865 
2866   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2867   for (size_t i = 0; i < node_num; i++) {
2868     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2869       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2870         if (cpu_map[j] != 0) {
2871           for (size_t k = 0; k < BitsPerCLong; k++) {
2872             if (cpu_map[j] & (1UL << k)) {
2873               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2874             }
2875           }
2876         }
2877       }
2878     }
2879   }
2880   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2881 }
2882 
2883 int os::Linux::get_node_by_cpu(int cpu_id) {
2884   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2885     return cpu_to_node()->at(cpu_id);
2886   }
2887   return -1;
2888 }
2889 
2890 GrowableArray<int>* os::Linux::_cpu_to_node;
2891 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2892 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2893 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2894 os::Linux::numa_available_func_t os::Linux::_numa_available;
2895 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2896 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2897 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2898 unsigned long* os::Linux::_numa_all_nodes;
2899 
2900 bool os::pd_uncommit_memory(char* addr, size_t size) {
2901   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2902                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2903   return res  != (uintptr_t) MAP_FAILED;
2904 }
2905 
2906 static address get_stack_commited_bottom(address bottom, size_t size) {
2907   address nbot = bottom;
2908   address ntop = bottom + size;
2909 
2910   size_t page_sz = os::vm_page_size();
2911   unsigned pages = size / page_sz;
2912 
2913   unsigned char vec[1];
2914   unsigned imin = 1, imax = pages + 1, imid;
2915   int mincore_return_value = 0;
2916 
2917   assert(imin <= imax, "Unexpected page size");
2918 
2919   while (imin < imax) {
2920     imid = (imax + imin) / 2;
2921     nbot = ntop - (imid * page_sz);
2922 
2923     // Use a trick with mincore to check whether the page is mapped or not.
2924     // mincore sets vec to 1 if page resides in memory and to 0 if page
2925     // is swapped output but if page we are asking for is unmapped
2926     // it returns -1,ENOMEM
2927     mincore_return_value = mincore(nbot, page_sz, vec);
2928 
2929     if (mincore_return_value == -1) {
2930       // Page is not mapped go up
2931       // to find first mapped page
2932       if (errno != EAGAIN) {
2933         assert(errno == ENOMEM, "Unexpected mincore errno");
2934         imax = imid;
2935       }
2936     } else {
2937       // Page is mapped go down
2938       // to find first not mapped page
2939       imin = imid + 1;
2940     }
2941   }
2942 
2943   nbot = nbot + page_sz;
2944 
2945   // Adjust stack bottom one page up if last checked page is not mapped
2946   if (mincore_return_value == -1) {
2947     nbot = nbot + page_sz;
2948   }
2949 
2950   return nbot;
2951 }
2952 
2953 
2954 // Linux uses a growable mapping for the stack, and if the mapping for
2955 // the stack guard pages is not removed when we detach a thread the
2956 // stack cannot grow beyond the pages where the stack guard was
2957 // mapped.  If at some point later in the process the stack expands to
2958 // that point, the Linux kernel cannot expand the stack any further
2959 // because the guard pages are in the way, and a segfault occurs.
2960 //
2961 // However, it's essential not to split the stack region by unmapping
2962 // a region (leaving a hole) that's already part of the stack mapping,
2963 // so if the stack mapping has already grown beyond the guard pages at
2964 // the time we create them, we have to truncate the stack mapping.
2965 // So, we need to know the extent of the stack mapping when
2966 // create_stack_guard_pages() is called.
2967 
2968 // We only need this for stacks that are growable: at the time of
2969 // writing thread stacks don't use growable mappings (i.e. those
2970 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2971 // only applies to the main thread.
2972 
2973 // If the (growable) stack mapping already extends beyond the point
2974 // where we're going to put our guard pages, truncate the mapping at
2975 // that point by munmap()ping it.  This ensures that when we later
2976 // munmap() the guard pages we don't leave a hole in the stack
2977 // mapping. This only affects the main/initial thread
2978 
2979 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2980   if (os::Linux::is_initial_thread()) {
2981     // As we manually grow stack up to bottom inside create_attached_thread(),
2982     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2983     // we don't need to do anything special.
2984     // Check it first, before calling heavy function.
2985     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2986     unsigned char vec[1];
2987 
2988     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2989       // Fallback to slow path on all errors, including EAGAIN
2990       stack_extent = (uintptr_t) get_stack_commited_bottom(
2991                                                            os::Linux::initial_thread_stack_bottom(),
2992                                                            (size_t)addr - stack_extent);
2993     }
2994 
2995     if (stack_extent < (uintptr_t)addr) {
2996       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2997     }
2998   }
2999 
3000   return os::commit_memory(addr, size, !ExecMem);
3001 }
3002 
3003 // If this is a growable mapping, remove the guard pages entirely by
3004 // munmap()ping them.  If not, just call uncommit_memory(). This only
3005 // affects the main/initial thread, but guard against future OS changes
3006 // It's safe to always unmap guard pages for initial thread because we
3007 // always place it right after end of the mapped region
3008 
3009 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3010   uintptr_t stack_extent, stack_base;
3011 
3012   if (os::Linux::is_initial_thread()) {
3013     return ::munmap(addr, size) == 0;
3014   }
3015 
3016   return os::uncommit_memory(addr, size);
3017 }
3018 
3019 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3020 // at 'requested_addr'. If there are existing memory mappings at the same
3021 // location, however, they will be overwritten. If 'fixed' is false,
3022 // 'requested_addr' is only treated as a hint, the return value may or
3023 // may not start from the requested address. Unlike Linux mmap(), this
3024 // function returns NULL to indicate failure.
3025 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3026   char * addr;
3027   int flags;
3028 
3029   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3030   if (fixed) {
3031     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3032     flags |= MAP_FIXED;
3033   }
3034 
3035   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3036   // touch an uncommitted page. Otherwise, the read/write might
3037   // succeed if we have enough swap space to back the physical page.
3038   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3039                        flags, -1, 0);
3040 
3041   return addr == MAP_FAILED ? NULL : addr;
3042 }
3043 
3044 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3045 //   (req_addr != NULL) or with a given alignment.
3046 //  - bytes shall be a multiple of alignment.
3047 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3048 //  - alignment sets the alignment at which memory shall be allocated.
3049 //     It must be a multiple of allocation granularity.
3050 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3051 //  req_addr or NULL.
3052 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3053 
3054   size_t extra_size = bytes;
3055   if (req_addr == NULL && alignment > 0) {
3056     extra_size += alignment;
3057   }
3058 
3059   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3060     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3061     -1, 0);
3062   if (start == MAP_FAILED) {
3063     start = NULL;
3064   } else {
3065     if (req_addr != NULL) {
3066       if (start != req_addr) {
3067         ::munmap(start, extra_size);
3068         start = NULL;
3069       }
3070     } else {
3071       char* const start_aligned = (char*) align_ptr_up(start, alignment);
3072       char* const end_aligned = start_aligned + bytes;
3073       char* const end = start + extra_size;
3074       if (start_aligned > start) {
3075         ::munmap(start, start_aligned - start);
3076       }
3077       if (end_aligned < end) {
3078         ::munmap(end_aligned, end - end_aligned);
3079       }
3080       start = start_aligned;
3081     }
3082   }
3083   return start;
3084 }
3085 
3086 static int anon_munmap(char * addr, size_t size) {
3087   return ::munmap(addr, size) == 0;
3088 }
3089 
3090 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3091                             size_t alignment_hint) {
3092   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3093 }
3094 
3095 bool os::pd_release_memory(char* addr, size_t size) {
3096   return anon_munmap(addr, size);
3097 }
3098 
3099 static bool linux_mprotect(char* addr, size_t size, int prot) {
3100   // Linux wants the mprotect address argument to be page aligned.
3101   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3102 
3103   // According to SUSv3, mprotect() should only be used with mappings
3104   // established by mmap(), and mmap() always maps whole pages. Unaligned
3105   // 'addr' likely indicates problem in the VM (e.g. trying to change
3106   // protection of malloc'ed or statically allocated memory). Check the
3107   // caller if you hit this assert.
3108   assert(addr == bottom, "sanity check");
3109 
3110   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3111   return ::mprotect(bottom, size, prot) == 0;
3112 }
3113 
3114 // Set protections specified
3115 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3116                         bool is_committed) {
3117   unsigned int p = 0;
3118   switch (prot) {
3119   case MEM_PROT_NONE: p = PROT_NONE; break;
3120   case MEM_PROT_READ: p = PROT_READ; break;
3121   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3122   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3123   default:
3124     ShouldNotReachHere();
3125   }
3126   // is_committed is unused.
3127   return linux_mprotect(addr, bytes, p);
3128 }
3129 
3130 bool os::guard_memory(char* addr, size_t size) {
3131   return linux_mprotect(addr, size, PROT_NONE);
3132 }
3133 
3134 bool os::unguard_memory(char* addr, size_t size) {
3135   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3136 }
3137 
3138 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3139                                                     size_t page_size) {
3140   bool result = false;
3141   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3142                  MAP_ANONYMOUS|MAP_PRIVATE,
3143                  -1, 0);
3144   if (p != MAP_FAILED) {
3145     void *aligned_p = align_ptr_up(p, page_size);
3146 
3147     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3148 
3149     munmap(p, page_size * 2);
3150   }
3151 
3152   if (warn && !result) {
3153     warning("TransparentHugePages is not supported by the operating system.");
3154   }
3155 
3156   return result;
3157 }
3158 
3159 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3160   bool result = false;
3161   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3162                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3163                  -1, 0);
3164 
3165   if (p != MAP_FAILED) {
3166     // We don't know if this really is a huge page or not.
3167     FILE *fp = fopen("/proc/self/maps", "r");
3168     if (fp) {
3169       while (!feof(fp)) {
3170         char chars[257];
3171         long x = 0;
3172         if (fgets(chars, sizeof(chars), fp)) {
3173           if (sscanf(chars, "%lx-%*x", &x) == 1
3174               && x == (long)p) {
3175             if (strstr (chars, "hugepage")) {
3176               result = true;
3177               break;
3178             }
3179           }
3180         }
3181       }
3182       fclose(fp);
3183     }
3184     munmap(p, page_size);
3185   }
3186 
3187   if (warn && !result) {
3188     warning("HugeTLBFS is not supported by the operating system.");
3189   }
3190 
3191   return result;
3192 }
3193 
3194 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3195 //
3196 // From the coredump_filter documentation:
3197 //
3198 // - (bit 0) anonymous private memory
3199 // - (bit 1) anonymous shared memory
3200 // - (bit 2) file-backed private memory
3201 // - (bit 3) file-backed shared memory
3202 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3203 //           effective only if the bit 2 is cleared)
3204 // - (bit 5) hugetlb private memory
3205 // - (bit 6) hugetlb shared memory
3206 //
3207 static void set_coredump_filter(void) {
3208   FILE *f;
3209   long cdm;
3210 
3211   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3212     return;
3213   }
3214 
3215   if (fscanf(f, "%lx", &cdm) != 1) {
3216     fclose(f);
3217     return;
3218   }
3219 
3220   rewind(f);
3221 
3222   if ((cdm & LARGEPAGES_BIT) == 0) {
3223     cdm |= LARGEPAGES_BIT;
3224     fprintf(f, "%#lx", cdm);
3225   }
3226 
3227   fclose(f);
3228 }
3229 
3230 // Large page support
3231 
3232 static size_t _large_page_size = 0;
3233 
3234 size_t os::Linux::find_large_page_size() {
3235   size_t large_page_size = 0;
3236 
3237   // large_page_size on Linux is used to round up heap size. x86 uses either
3238   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3239   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3240   // page as large as 256M.
3241   //
3242   // Here we try to figure out page size by parsing /proc/meminfo and looking
3243   // for a line with the following format:
3244   //    Hugepagesize:     2048 kB
3245   //
3246   // If we can't determine the value (e.g. /proc is not mounted, or the text
3247   // format has been changed), we'll use the largest page size supported by
3248   // the processor.
3249 
3250 #ifndef ZERO
3251   large_page_size =
3252     AARCH64_ONLY(2 * M)
3253     AMD64_ONLY(2 * M)
3254     ARM32_ONLY(2 * M)
3255     IA32_ONLY(4 * M)
3256     IA64_ONLY(256 * M)
3257     PPC_ONLY(4 * M)
3258     S390_ONLY(1 * M)
3259     SPARC_ONLY(4 * M);
3260 #endif // ZERO
3261 
3262   FILE *fp = fopen("/proc/meminfo", "r");
3263   if (fp) {
3264     while (!feof(fp)) {
3265       int x = 0;
3266       char buf[16];
3267       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3268         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3269           large_page_size = x * K;
3270           break;
3271         }
3272       } else {
3273         // skip to next line
3274         for (;;) {
3275           int ch = fgetc(fp);
3276           if (ch == EOF || ch == (int)'\n') break;
3277         }
3278       }
3279     }
3280     fclose(fp);
3281   }
3282 
3283   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3284     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3285             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3286             proper_unit_for_byte_size(large_page_size));
3287   }
3288 
3289   return large_page_size;
3290 }
3291 
3292 size_t os::Linux::setup_large_page_size() {
3293   _large_page_size = Linux::find_large_page_size();
3294   const size_t default_page_size = (size_t)Linux::page_size();
3295   if (_large_page_size > default_page_size) {
3296     _page_sizes[0] = _large_page_size;
3297     _page_sizes[1] = default_page_size;
3298     _page_sizes[2] = 0;
3299   }
3300 
3301   return _large_page_size;
3302 }
3303 
3304 bool os::Linux::setup_large_page_type(size_t page_size) {
3305   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3306       FLAG_IS_DEFAULT(UseSHM) &&
3307       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3308 
3309     // The type of large pages has not been specified by the user.
3310 
3311     // Try UseHugeTLBFS and then UseSHM.
3312     UseHugeTLBFS = UseSHM = true;
3313 
3314     // Don't try UseTransparentHugePages since there are known
3315     // performance issues with it turned on. This might change in the future.
3316     UseTransparentHugePages = false;
3317   }
3318 
3319   if (UseTransparentHugePages) {
3320     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3321     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3322       UseHugeTLBFS = false;
3323       UseSHM = false;
3324       return true;
3325     }
3326     UseTransparentHugePages = false;
3327   }
3328 
3329   if (UseHugeTLBFS) {
3330     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3331     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3332       UseSHM = false;
3333       return true;
3334     }
3335     UseHugeTLBFS = false;
3336   }
3337 
3338   return UseSHM;
3339 }
3340 
3341 void os::large_page_init() {
3342   if (!UseLargePages &&
3343       !UseTransparentHugePages &&
3344       !UseHugeTLBFS &&
3345       !UseSHM) {
3346     // Not using large pages.
3347     return;
3348   }
3349 
3350   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3351     // The user explicitly turned off large pages.
3352     // Ignore the rest of the large pages flags.
3353     UseTransparentHugePages = false;
3354     UseHugeTLBFS = false;
3355     UseSHM = false;
3356     return;
3357   }
3358 
3359   size_t large_page_size = Linux::setup_large_page_size();
3360   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3361 
3362   set_coredump_filter();
3363 }
3364 
3365 #ifndef SHM_HUGETLB
3366   #define SHM_HUGETLB 04000
3367 #endif
3368 
3369 #define shm_warning_format(format, ...)              \
3370   do {                                               \
3371     if (UseLargePages &&                             \
3372         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3373          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3374          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3375       warning(format, __VA_ARGS__);                  \
3376     }                                                \
3377   } while (0)
3378 
3379 #define shm_warning(str) shm_warning_format("%s", str)
3380 
3381 #define shm_warning_with_errno(str)                \
3382   do {                                             \
3383     int err = errno;                               \
3384     shm_warning_format(str " (error = %d)", err);  \
3385   } while (0)
3386 
3387 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3388   assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3389 
3390   if (!is_size_aligned(alignment, SHMLBA)) {
3391     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3392     return NULL;
3393   }
3394 
3395   // To ensure that we get 'alignment' aligned memory from shmat,
3396   // we pre-reserve aligned virtual memory and then attach to that.
3397 
3398   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3399   if (pre_reserved_addr == NULL) {
3400     // Couldn't pre-reserve aligned memory.
3401     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3402     return NULL;
3403   }
3404 
3405   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3406   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3407 
3408   if ((intptr_t)addr == -1) {
3409     int err = errno;
3410     shm_warning_with_errno("Failed to attach shared memory.");
3411 
3412     assert(err != EACCES, "Unexpected error");
3413     assert(err != EIDRM,  "Unexpected error");
3414     assert(err != EINVAL, "Unexpected error");
3415 
3416     // Since we don't know if the kernel unmapped the pre-reserved memory area
3417     // we can't unmap it, since that would potentially unmap memory that was
3418     // mapped from other threads.
3419     return NULL;
3420   }
3421 
3422   return addr;
3423 }
3424 
3425 static char* shmat_at_address(int shmid, char* req_addr) {
3426   if (!is_ptr_aligned(req_addr, SHMLBA)) {
3427     assert(false, "Requested address needs to be SHMLBA aligned");
3428     return NULL;
3429   }
3430 
3431   char* addr = (char*)shmat(shmid, req_addr, 0);
3432 
3433   if ((intptr_t)addr == -1) {
3434     shm_warning_with_errno("Failed to attach shared memory.");
3435     return NULL;
3436   }
3437 
3438   return addr;
3439 }
3440 
3441 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3442   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3443   if (req_addr != NULL) {
3444     assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3445     assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3446     return shmat_at_address(shmid, req_addr);
3447   }
3448 
3449   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3450   // return large page size aligned memory addresses when req_addr == NULL.
3451   // However, if the alignment is larger than the large page size, we have
3452   // to manually ensure that the memory returned is 'alignment' aligned.
3453   if (alignment > os::large_page_size()) {
3454     assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3455     return shmat_with_alignment(shmid, bytes, alignment);
3456   } else {
3457     return shmat_at_address(shmid, NULL);
3458   }
3459 }
3460 
3461 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3462                                             char* req_addr, bool exec) {
3463   // "exec" is passed in but not used.  Creating the shared image for
3464   // the code cache doesn't have an SHM_X executable permission to check.
3465   assert(UseLargePages && UseSHM, "only for SHM large pages");
3466   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3467   assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3468 
3469   if (!is_size_aligned(bytes, os::large_page_size())) {
3470     return NULL; // Fallback to small pages.
3471   }
3472 
3473   // Create a large shared memory region to attach to based on size.
3474   // Currently, size is the total size of the heap.
3475   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3476   if (shmid == -1) {
3477     // Possible reasons for shmget failure:
3478     // 1. shmmax is too small for Java heap.
3479     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3480     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3481     // 2. not enough large page memory.
3482     //    > check available large pages: cat /proc/meminfo
3483     //    > increase amount of large pages:
3484     //          echo new_value > /proc/sys/vm/nr_hugepages
3485     //      Note 1: different Linux may use different name for this property,
3486     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3487     //      Note 2: it's possible there's enough physical memory available but
3488     //            they are so fragmented after a long run that they can't
3489     //            coalesce into large pages. Try to reserve large pages when
3490     //            the system is still "fresh".
3491     shm_warning_with_errno("Failed to reserve shared memory.");
3492     return NULL;
3493   }
3494 
3495   // Attach to the region.
3496   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3497 
3498   // Remove shmid. If shmat() is successful, the actual shared memory segment
3499   // will be deleted when it's detached by shmdt() or when the process
3500   // terminates. If shmat() is not successful this will remove the shared
3501   // segment immediately.
3502   shmctl(shmid, IPC_RMID, NULL);
3503 
3504   return addr;
3505 }
3506 
3507 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3508                                         int error) {
3509   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3510 
3511   bool warn_on_failure = UseLargePages &&
3512       (!FLAG_IS_DEFAULT(UseLargePages) ||
3513        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3514        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3515 
3516   if (warn_on_failure) {
3517     char msg[128];
3518     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3519                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3520     warning("%s", msg);
3521   }
3522 }
3523 
3524 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3525                                                         char* req_addr,
3526                                                         bool exec) {
3527   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3528   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3529   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3530 
3531   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3532   char* addr = (char*)::mmap(req_addr, bytes, prot,
3533                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3534                              -1, 0);
3535 
3536   if (addr == MAP_FAILED) {
3537     warn_on_large_pages_failure(req_addr, bytes, errno);
3538     return NULL;
3539   }
3540 
3541   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3542 
3543   return addr;
3544 }
3545 
3546 // Reserve memory using mmap(MAP_HUGETLB).
3547 //  - bytes shall be a multiple of alignment.
3548 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3549 //  - alignment sets the alignment at which memory shall be allocated.
3550 //     It must be a multiple of allocation granularity.
3551 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3552 //  req_addr or NULL.
3553 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3554                                                          size_t alignment,
3555                                                          char* req_addr,
3556                                                          bool exec) {
3557   size_t large_page_size = os::large_page_size();
3558   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3559 
3560   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3561   assert(is_size_aligned(bytes, alignment), "Must be");
3562 
3563   // First reserve - but not commit - the address range in small pages.
3564   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3565 
3566   if (start == NULL) {
3567     return NULL;
3568   }
3569 
3570   assert(is_ptr_aligned(start, alignment), "Must be");
3571 
3572   char* end = start + bytes;
3573 
3574   // Find the regions of the allocated chunk that can be promoted to large pages.
3575   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3576   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3577 
3578   size_t lp_bytes = lp_end - lp_start;
3579 
3580   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3581 
3582   if (lp_bytes == 0) {
3583     // The mapped region doesn't even span the start and the end of a large page.
3584     // Fall back to allocate a non-special area.
3585     ::munmap(start, end - start);
3586     return NULL;
3587   }
3588 
3589   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3590 
3591   void* result;
3592 
3593   // Commit small-paged leading area.
3594   if (start != lp_start) {
3595     result = ::mmap(start, lp_start - start, prot,
3596                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3597                     -1, 0);
3598     if (result == MAP_FAILED) {
3599       ::munmap(lp_start, end - lp_start);
3600       return NULL;
3601     }
3602   }
3603 
3604   // Commit large-paged area.
3605   result = ::mmap(lp_start, lp_bytes, prot,
3606                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3607                   -1, 0);
3608   if (result == MAP_FAILED) {
3609     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3610     // If the mmap above fails, the large pages region will be unmapped and we
3611     // have regions before and after with small pages. Release these regions.
3612     //
3613     // |  mapped  |  unmapped  |  mapped  |
3614     // ^          ^            ^          ^
3615     // start      lp_start     lp_end     end
3616     //
3617     ::munmap(start, lp_start - start);
3618     ::munmap(lp_end, end - lp_end);
3619     return NULL;
3620   }
3621 
3622   // Commit small-paged trailing area.
3623   if (lp_end != end) {
3624     result = ::mmap(lp_end, end - lp_end, prot,
3625                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3626                     -1, 0);
3627     if (result == MAP_FAILED) {
3628       ::munmap(start, lp_end - start);
3629       return NULL;
3630     }
3631   }
3632 
3633   return start;
3634 }
3635 
3636 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3637                                                    size_t alignment,
3638                                                    char* req_addr,
3639                                                    bool exec) {
3640   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3641   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3642   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3643   assert(is_power_of_2(os::large_page_size()), "Must be");
3644   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3645 
3646   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3647     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3648   } else {
3649     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3650   }
3651 }
3652 
3653 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3654                                  char* req_addr, bool exec) {
3655   assert(UseLargePages, "only for large pages");
3656 
3657   char* addr;
3658   if (UseSHM) {
3659     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3660   } else {
3661     assert(UseHugeTLBFS, "must be");
3662     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3663   }
3664 
3665   if (addr != NULL) {
3666     if (UseNUMAInterleaving) {
3667       numa_make_global(addr, bytes);
3668     }
3669 
3670     // The memory is committed
3671     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3672   }
3673 
3674   return addr;
3675 }
3676 
3677 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3678   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3679   return shmdt(base) == 0;
3680 }
3681 
3682 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3683   return pd_release_memory(base, bytes);
3684 }
3685 
3686 bool os::release_memory_special(char* base, size_t bytes) {
3687   bool res;
3688   if (MemTracker::tracking_level() > NMT_minimal) {
3689     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3690     res = os::Linux::release_memory_special_impl(base, bytes);
3691     if (res) {
3692       tkr.record((address)base, bytes);
3693     }
3694 
3695   } else {
3696     res = os::Linux::release_memory_special_impl(base, bytes);
3697   }
3698   return res;
3699 }
3700 
3701 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3702   assert(UseLargePages, "only for large pages");
3703   bool res;
3704 
3705   if (UseSHM) {
3706     res = os::Linux::release_memory_special_shm(base, bytes);
3707   } else {
3708     assert(UseHugeTLBFS, "must be");
3709     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3710   }
3711   return res;
3712 }
3713 
3714 size_t os::large_page_size() {
3715   return _large_page_size;
3716 }
3717 
3718 // With SysV SHM the entire memory region must be allocated as shared
3719 // memory.
3720 // HugeTLBFS allows application to commit large page memory on demand.
3721 // However, when committing memory with HugeTLBFS fails, the region
3722 // that was supposed to be committed will lose the old reservation
3723 // and allow other threads to steal that memory region. Because of this
3724 // behavior we can't commit HugeTLBFS memory.
3725 bool os::can_commit_large_page_memory() {
3726   return UseTransparentHugePages;
3727 }
3728 
3729 bool os::can_execute_large_page_memory() {
3730   return UseTransparentHugePages || UseHugeTLBFS;
3731 }
3732 
3733 // Reserve memory at an arbitrary address, only if that area is
3734 // available (and not reserved for something else).
3735 
3736 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3737   const int max_tries = 10;
3738   char* base[max_tries];
3739   size_t size[max_tries];
3740   const size_t gap = 0x000000;
3741 
3742   // Assert only that the size is a multiple of the page size, since
3743   // that's all that mmap requires, and since that's all we really know
3744   // about at this low abstraction level.  If we need higher alignment,
3745   // we can either pass an alignment to this method or verify alignment
3746   // in one of the methods further up the call chain.  See bug 5044738.
3747   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3748 
3749   // Repeatedly allocate blocks until the block is allocated at the
3750   // right spot.
3751 
3752   // Linux mmap allows caller to pass an address as hint; give it a try first,
3753   // if kernel honors the hint then we can return immediately.
3754   char * addr = anon_mmap(requested_addr, bytes, false);
3755   if (addr == requested_addr) {
3756     return requested_addr;
3757   }
3758 
3759   if (addr != NULL) {
3760     // mmap() is successful but it fails to reserve at the requested address
3761     anon_munmap(addr, bytes);
3762   }
3763 
3764   int i;
3765   for (i = 0; i < max_tries; ++i) {
3766     base[i] = reserve_memory(bytes);
3767 
3768     if (base[i] != NULL) {
3769       // Is this the block we wanted?
3770       if (base[i] == requested_addr) {
3771         size[i] = bytes;
3772         break;
3773       }
3774 
3775       // Does this overlap the block we wanted? Give back the overlapped
3776       // parts and try again.
3777 
3778       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3779       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3780         unmap_memory(base[i], top_overlap);
3781         base[i] += top_overlap;
3782         size[i] = bytes - top_overlap;
3783       } else {
3784         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3785         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3786           unmap_memory(requested_addr, bottom_overlap);
3787           size[i] = bytes - bottom_overlap;
3788         } else {
3789           size[i] = bytes;
3790         }
3791       }
3792     }
3793   }
3794 
3795   // Give back the unused reserved pieces.
3796 
3797   for (int j = 0; j < i; ++j) {
3798     if (base[j] != NULL) {
3799       unmap_memory(base[j], size[j]);
3800     }
3801   }
3802 
3803   if (i < max_tries) {
3804     return requested_addr;
3805   } else {
3806     return NULL;
3807   }
3808 }
3809 
3810 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3811   return ::read(fd, buf, nBytes);
3812 }
3813 
3814 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3815   return ::pread(fd, buf, nBytes, offset);
3816 }
3817 
3818 // Short sleep, direct OS call.
3819 //
3820 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3821 // sched_yield(2) will actually give up the CPU:
3822 //
3823 //   * Alone on this pariticular CPU, keeps running.
3824 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3825 //     (pre 2.6.39).
3826 //
3827 // So calling this with 0 is an alternative.
3828 //
3829 void os::naked_short_sleep(jlong ms) {
3830   struct timespec req;
3831 
3832   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3833   req.tv_sec = 0;
3834   if (ms > 0) {
3835     req.tv_nsec = (ms % 1000) * 1000000;
3836   } else {
3837     req.tv_nsec = 1;
3838   }
3839 
3840   nanosleep(&req, NULL);
3841 
3842   return;
3843 }
3844 
3845 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3846 void os::infinite_sleep() {
3847   while (true) {    // sleep forever ...
3848     ::sleep(100);   // ... 100 seconds at a time
3849   }
3850 }
3851 
3852 // Used to convert frequent JVM_Yield() to nops
3853 bool os::dont_yield() {
3854   return DontYieldALot;
3855 }
3856 
3857 void os::naked_yield() {
3858   sched_yield();
3859 }
3860 
3861 ////////////////////////////////////////////////////////////////////////////////
3862 // thread priority support
3863 
3864 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3865 // only supports dynamic priority, static priority must be zero. For real-time
3866 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3867 // However, for large multi-threaded applications, SCHED_RR is not only slower
3868 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3869 // of 5 runs - Sep 2005).
3870 //
3871 // The following code actually changes the niceness of kernel-thread/LWP. It
3872 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3873 // not the entire user process, and user level threads are 1:1 mapped to kernel
3874 // threads. It has always been the case, but could change in the future. For
3875 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3876 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3877 
3878 int os::java_to_os_priority[CriticalPriority + 1] = {
3879   19,              // 0 Entry should never be used
3880 
3881    4,              // 1 MinPriority
3882    3,              // 2
3883    2,              // 3
3884 
3885    1,              // 4
3886    0,              // 5 NormPriority
3887   -1,              // 6
3888 
3889   -2,              // 7
3890   -3,              // 8
3891   -4,              // 9 NearMaxPriority
3892 
3893   -5,              // 10 MaxPriority
3894 
3895   -5               // 11 CriticalPriority
3896 };
3897 
3898 static int prio_init() {
3899   if (ThreadPriorityPolicy == 1) {
3900     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3901     // if effective uid is not root. Perhaps, a more elegant way of doing
3902     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3903     if (geteuid() != 0) {
3904       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3905         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3906       }
3907       ThreadPriorityPolicy = 0;
3908     }
3909   }
3910   if (UseCriticalJavaThreadPriority) {
3911     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3912   }
3913   return 0;
3914 }
3915 
3916 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3917   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3918 
3919   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3920   return (ret == 0) ? OS_OK : OS_ERR;
3921 }
3922 
3923 OSReturn os::get_native_priority(const Thread* const thread,
3924                                  int *priority_ptr) {
3925   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3926     *priority_ptr = java_to_os_priority[NormPriority];
3927     return OS_OK;
3928   }
3929 
3930   errno = 0;
3931   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3932   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3933 }
3934 
3935 // Hint to the underlying OS that a task switch would not be good.
3936 // Void return because it's a hint and can fail.
3937 void os::hint_no_preempt() {}
3938 
3939 ////////////////////////////////////////////////////////////////////////////////
3940 // suspend/resume support
3941 
3942 //  the low-level signal-based suspend/resume support is a remnant from the
3943 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3944 //  within hotspot. Now there is a single use-case for this:
3945 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3946 //      that runs in the watcher thread.
3947 //  The remaining code is greatly simplified from the more general suspension
3948 //  code that used to be used.
3949 //
3950 //  The protocol is quite simple:
3951 //  - suspend:
3952 //      - sends a signal to the target thread
3953 //      - polls the suspend state of the osthread using a yield loop
3954 //      - target thread signal handler (SR_handler) sets suspend state
3955 //        and blocks in sigsuspend until continued
3956 //  - resume:
3957 //      - sets target osthread state to continue
3958 //      - sends signal to end the sigsuspend loop in the SR_handler
3959 //
3960 //  Note that the SR_lock plays no role in this suspend/resume protocol,
3961 //  but is checked for NULL in SR_handler as a thread termination indicator.
3962 
3963 static void resume_clear_context(OSThread *osthread) {
3964   osthread->set_ucontext(NULL);
3965   osthread->set_siginfo(NULL);
3966 }
3967 
3968 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3969                                  ucontext_t* context) {
3970   osthread->set_ucontext(context);
3971   osthread->set_siginfo(siginfo);
3972 }
3973 
3974 // Handler function invoked when a thread's execution is suspended or
3975 // resumed. We have to be careful that only async-safe functions are
3976 // called here (Note: most pthread functions are not async safe and
3977 // should be avoided.)
3978 //
3979 // Note: sigwait() is a more natural fit than sigsuspend() from an
3980 // interface point of view, but sigwait() prevents the signal hander
3981 // from being run. libpthread would get very confused by not having
3982 // its signal handlers run and prevents sigwait()'s use with the
3983 // mutex granting granting signal.
3984 //
3985 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3986 //
3987 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3988   // Save and restore errno to avoid confusing native code with EINTR
3989   // after sigsuspend.
3990   int old_errno = errno;
3991 
3992   Thread* thread = Thread::current_or_null_safe();
3993   assert(thread != NULL, "Missing current thread in SR_handler");
3994 
3995   // On some systems we have seen signal delivery get "stuck" until the signal
3996   // mask is changed as part of thread termination. Check that the current thread
3997   // has not already terminated (via SR_lock()) - else the following assertion
3998   // will fail because the thread is no longer a JavaThread as the ~JavaThread
3999   // destructor has completed.
4000 
4001   if (thread->SR_lock() == NULL) {
4002     return;
4003   }
4004 
4005   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4006 
4007   OSThread* osthread = thread->osthread();
4008 
4009   os::SuspendResume::State current = osthread->sr.state();
4010   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4011     suspend_save_context(osthread, siginfo, context);
4012 
4013     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4014     os::SuspendResume::State state = osthread->sr.suspended();
4015     if (state == os::SuspendResume::SR_SUSPENDED) {
4016       sigset_t suspend_set;  // signals for sigsuspend()
4017       sigemptyset(&suspend_set);
4018       // get current set of blocked signals and unblock resume signal
4019       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4020       sigdelset(&suspend_set, SR_signum);
4021 
4022       sr_semaphore.signal();
4023       // wait here until we are resumed
4024       while (1) {
4025         sigsuspend(&suspend_set);
4026 
4027         os::SuspendResume::State result = osthread->sr.running();
4028         if (result == os::SuspendResume::SR_RUNNING) {
4029           sr_semaphore.signal();
4030           break;
4031         }
4032       }
4033 
4034     } else if (state == os::SuspendResume::SR_RUNNING) {
4035       // request was cancelled, continue
4036     } else {
4037       ShouldNotReachHere();
4038     }
4039 
4040     resume_clear_context(osthread);
4041   } else if (current == os::SuspendResume::SR_RUNNING) {
4042     // request was cancelled, continue
4043   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4044     // ignore
4045   } else {
4046     // ignore
4047   }
4048 
4049   errno = old_errno;
4050 }
4051 
4052 static int SR_initialize() {
4053   struct sigaction act;
4054   char *s;
4055 
4056   // Get signal number to use for suspend/resume
4057   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4058     int sig = ::strtol(s, 0, 10);
4059     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4060         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4061       SR_signum = sig;
4062     } else {
4063       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4064               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4065     }
4066   }
4067 
4068   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4069          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4070 
4071   sigemptyset(&SR_sigset);
4072   sigaddset(&SR_sigset, SR_signum);
4073 
4074   // Set up signal handler for suspend/resume
4075   act.sa_flags = SA_RESTART|SA_SIGINFO;
4076   act.sa_handler = (void (*)(int)) SR_handler;
4077 
4078   // SR_signum is blocked by default.
4079   // 4528190 - We also need to block pthread restart signal (32 on all
4080   // supported Linux platforms). Note that LinuxThreads need to block
4081   // this signal for all threads to work properly. So we don't have
4082   // to use hard-coded signal number when setting up the mask.
4083   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4084 
4085   if (sigaction(SR_signum, &act, 0) == -1) {
4086     return -1;
4087   }
4088 
4089   // Save signal flag
4090   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4091   return 0;
4092 }
4093 
4094 static int sr_notify(OSThread* osthread) {
4095   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4096   assert_status(status == 0, status, "pthread_kill");
4097   return status;
4098 }
4099 
4100 // "Randomly" selected value for how long we want to spin
4101 // before bailing out on suspending a thread, also how often
4102 // we send a signal to a thread we want to resume
4103 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4104 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4105 
4106 // returns true on success and false on error - really an error is fatal
4107 // but this seems the normal response to library errors
4108 static bool do_suspend(OSThread* osthread) {
4109   assert(osthread->sr.is_running(), "thread should be running");
4110   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4111 
4112   // mark as suspended and send signal
4113   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4114     // failed to switch, state wasn't running?
4115     ShouldNotReachHere();
4116     return false;
4117   }
4118 
4119   if (sr_notify(osthread) != 0) {
4120     ShouldNotReachHere();
4121   }
4122 
4123   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4124   while (true) {
4125     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4126       break;
4127     } else {
4128       // timeout
4129       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4130       if (cancelled == os::SuspendResume::SR_RUNNING) {
4131         return false;
4132       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4133         // make sure that we consume the signal on the semaphore as well
4134         sr_semaphore.wait();
4135         break;
4136       } else {
4137         ShouldNotReachHere();
4138         return false;
4139       }
4140     }
4141   }
4142 
4143   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4144   return true;
4145 }
4146 
4147 static void do_resume(OSThread* osthread) {
4148   assert(osthread->sr.is_suspended(), "thread should be suspended");
4149   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4150 
4151   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4152     // failed to switch to WAKEUP_REQUEST
4153     ShouldNotReachHere();
4154     return;
4155   }
4156 
4157   while (true) {
4158     if (sr_notify(osthread) == 0) {
4159       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4160         if (osthread->sr.is_running()) {
4161           return;
4162         }
4163       }
4164     } else {
4165       ShouldNotReachHere();
4166     }
4167   }
4168 
4169   guarantee(osthread->sr.is_running(), "Must be running!");
4170 }
4171 
4172 ///////////////////////////////////////////////////////////////////////////////////
4173 // signal handling (except suspend/resume)
4174 
4175 // This routine may be used by user applications as a "hook" to catch signals.
4176 // The user-defined signal handler must pass unrecognized signals to this
4177 // routine, and if it returns true (non-zero), then the signal handler must
4178 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4179 // routine will never retun false (zero), but instead will execute a VM panic
4180 // routine kill the process.
4181 //
4182 // If this routine returns false, it is OK to call it again.  This allows
4183 // the user-defined signal handler to perform checks either before or after
4184 // the VM performs its own checks.  Naturally, the user code would be making
4185 // a serious error if it tried to handle an exception (such as a null check
4186 // or breakpoint) that the VM was generating for its own correct operation.
4187 //
4188 // This routine may recognize any of the following kinds of signals:
4189 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4190 // It should be consulted by handlers for any of those signals.
4191 //
4192 // The caller of this routine must pass in the three arguments supplied
4193 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4194 // field of the structure passed to sigaction().  This routine assumes that
4195 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4196 //
4197 // Note that the VM will print warnings if it detects conflicting signal
4198 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4199 //
4200 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4201                                                  siginfo_t* siginfo,
4202                                                  void* ucontext,
4203                                                  int abort_if_unrecognized);
4204 
4205 void signalHandler(int sig, siginfo_t* info, void* uc) {
4206   assert(info != NULL && uc != NULL, "it must be old kernel");
4207   int orig_errno = errno;  // Preserve errno value over signal handler.
4208   JVM_handle_linux_signal(sig, info, uc, true);
4209   errno = orig_errno;
4210 }
4211 
4212 
4213 // This boolean allows users to forward their own non-matching signals
4214 // to JVM_handle_linux_signal, harmlessly.
4215 bool os::Linux::signal_handlers_are_installed = false;
4216 
4217 // For signal-chaining
4218 struct sigaction sigact[NSIG];
4219 uint64_t sigs = 0;
4220 #if (64 < NSIG-1)
4221 #error "Not all signals can be encoded in sigs. Adapt its type!"
4222 #endif
4223 bool os::Linux::libjsig_is_loaded = false;
4224 typedef struct sigaction *(*get_signal_t)(int);
4225 get_signal_t os::Linux::get_signal_action = NULL;
4226 
4227 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4228   struct sigaction *actp = NULL;
4229 
4230   if (libjsig_is_loaded) {
4231     // Retrieve the old signal handler from libjsig
4232     actp = (*get_signal_action)(sig);
4233   }
4234   if (actp == NULL) {
4235     // Retrieve the preinstalled signal handler from jvm
4236     actp = get_preinstalled_handler(sig);
4237   }
4238 
4239   return actp;
4240 }
4241 
4242 static bool call_chained_handler(struct sigaction *actp, int sig,
4243                                  siginfo_t *siginfo, void *context) {
4244   // Call the old signal handler
4245   if (actp->sa_handler == SIG_DFL) {
4246     // It's more reasonable to let jvm treat it as an unexpected exception
4247     // instead of taking the default action.
4248     return false;
4249   } else if (actp->sa_handler != SIG_IGN) {
4250     if ((actp->sa_flags & SA_NODEFER) == 0) {
4251       // automaticlly block the signal
4252       sigaddset(&(actp->sa_mask), sig);
4253     }
4254 
4255     sa_handler_t hand = NULL;
4256     sa_sigaction_t sa = NULL;
4257     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4258     // retrieve the chained handler
4259     if (siginfo_flag_set) {
4260       sa = actp->sa_sigaction;
4261     } else {
4262       hand = actp->sa_handler;
4263     }
4264 
4265     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4266       actp->sa_handler = SIG_DFL;
4267     }
4268 
4269     // try to honor the signal mask
4270     sigset_t oset;
4271     sigemptyset(&oset);
4272     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4273 
4274     // call into the chained handler
4275     if (siginfo_flag_set) {
4276       (*sa)(sig, siginfo, context);
4277     } else {
4278       (*hand)(sig);
4279     }
4280 
4281     // restore the signal mask
4282     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4283   }
4284   // Tell jvm's signal handler the signal is taken care of.
4285   return true;
4286 }
4287 
4288 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4289   bool chained = false;
4290   // signal-chaining
4291   if (UseSignalChaining) {
4292     struct sigaction *actp = get_chained_signal_action(sig);
4293     if (actp != NULL) {
4294       chained = call_chained_handler(actp, sig, siginfo, context);
4295     }
4296   }
4297   return chained;
4298 }
4299 
4300 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4301   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4302     return &sigact[sig];
4303   }
4304   return NULL;
4305 }
4306 
4307 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4308   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4309   sigact[sig] = oldAct;
4310   sigs |= (uint64_t)1 << (sig-1);
4311 }
4312 
4313 // for diagnostic
4314 int sigflags[NSIG];
4315 
4316 int os::Linux::get_our_sigflags(int sig) {
4317   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4318   return sigflags[sig];
4319 }
4320 
4321 void os::Linux::set_our_sigflags(int sig, int flags) {
4322   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4323   if (sig > 0 && sig < NSIG) {
4324     sigflags[sig] = flags;
4325   }
4326 }
4327 
4328 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4329   // Check for overwrite.
4330   struct sigaction oldAct;
4331   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4332 
4333   void* oldhand = oldAct.sa_sigaction
4334                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4335                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4336   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4337       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4338       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4339     if (AllowUserSignalHandlers || !set_installed) {
4340       // Do not overwrite; user takes responsibility to forward to us.
4341       return;
4342     } else if (UseSignalChaining) {
4343       // save the old handler in jvm
4344       save_preinstalled_handler(sig, oldAct);
4345       // libjsig also interposes the sigaction() call below and saves the
4346       // old sigaction on it own.
4347     } else {
4348       fatal("Encountered unexpected pre-existing sigaction handler "
4349             "%#lx for signal %d.", (long)oldhand, sig);
4350     }
4351   }
4352 
4353   struct sigaction sigAct;
4354   sigfillset(&(sigAct.sa_mask));
4355   sigAct.sa_handler = SIG_DFL;
4356   if (!set_installed) {
4357     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4358   } else {
4359     sigAct.sa_sigaction = signalHandler;
4360     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4361   }
4362   // Save flags, which are set by ours
4363   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4364   sigflags[sig] = sigAct.sa_flags;
4365 
4366   int ret = sigaction(sig, &sigAct, &oldAct);
4367   assert(ret == 0, "check");
4368 
4369   void* oldhand2  = oldAct.sa_sigaction
4370                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4371                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4372   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4373 }
4374 
4375 // install signal handlers for signals that HotSpot needs to
4376 // handle in order to support Java-level exception handling.
4377 
4378 void os::Linux::install_signal_handlers() {
4379   if (!signal_handlers_are_installed) {
4380     signal_handlers_are_installed = true;
4381 
4382     // signal-chaining
4383     typedef void (*signal_setting_t)();
4384     signal_setting_t begin_signal_setting = NULL;
4385     signal_setting_t end_signal_setting = NULL;
4386     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4387                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4388     if (begin_signal_setting != NULL) {
4389       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4390                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4391       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4392                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4393       libjsig_is_loaded = true;
4394       assert(UseSignalChaining, "should enable signal-chaining");
4395     }
4396     if (libjsig_is_loaded) {
4397       // Tell libjsig jvm is setting signal handlers
4398       (*begin_signal_setting)();
4399     }
4400 
4401     set_signal_handler(SIGSEGV, true);
4402     set_signal_handler(SIGPIPE, true);
4403     set_signal_handler(SIGBUS, true);
4404     set_signal_handler(SIGILL, true);
4405     set_signal_handler(SIGFPE, true);
4406 #if defined(PPC64)
4407     set_signal_handler(SIGTRAP, true);
4408 #endif
4409     set_signal_handler(SIGXFSZ, true);
4410 
4411     if (libjsig_is_loaded) {
4412       // Tell libjsig jvm finishes setting signal handlers
4413       (*end_signal_setting)();
4414     }
4415 
4416     // We don't activate signal checker if libjsig is in place, we trust ourselves
4417     // and if UserSignalHandler is installed all bets are off.
4418     // Log that signal checking is off only if -verbose:jni is specified.
4419     if (CheckJNICalls) {
4420       if (libjsig_is_loaded) {
4421         if (PrintJNIResolving) {
4422           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4423         }
4424         check_signals = false;
4425       }
4426       if (AllowUserSignalHandlers) {
4427         if (PrintJNIResolving) {
4428           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4429         }
4430         check_signals = false;
4431       }
4432     }
4433   }
4434 }
4435 
4436 // This is the fastest way to get thread cpu time on Linux.
4437 // Returns cpu time (user+sys) for any thread, not only for current.
4438 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4439 // It might work on 2.6.10+ with a special kernel/glibc patch.
4440 // For reference, please, see IEEE Std 1003.1-2004:
4441 //   http://www.unix.org/single_unix_specification
4442 
4443 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4444   struct timespec tp;
4445   int rc = os::Linux::clock_gettime(clockid, &tp);
4446   assert(rc == 0, "clock_gettime is expected to return 0 code");
4447 
4448   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4449 }
4450 
4451 void os::Linux::initialize_os_info() {
4452   assert(_os_version == 0, "OS info already initialized");
4453 
4454   struct utsname _uname;
4455 
4456   uint32_t major;
4457   uint32_t minor;
4458   uint32_t fix;
4459 
4460   int rc;
4461 
4462   // Kernel version is unknown if
4463   // verification below fails.
4464   _os_version = 0x01000000;
4465 
4466   rc = uname(&_uname);
4467   if (rc != -1) {
4468 
4469     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4470     if (rc == 3) {
4471 
4472       if (major < 256 && minor < 256 && fix < 256) {
4473         // Kernel version format is as expected,
4474         // set it overriding unknown state.
4475         _os_version = (major << 16) |
4476                       (minor << 8 ) |
4477                       (fix   << 0 ) ;
4478       }
4479     }
4480   }
4481 }
4482 
4483 uint32_t os::Linux::os_version() {
4484   assert(_os_version != 0, "not initialized");
4485   return _os_version & 0x00FFFFFF;
4486 }
4487 
4488 bool os::Linux::os_version_is_known() {
4489   assert(_os_version != 0, "not initialized");
4490   return _os_version & 0x01000000 ? false : true;
4491 }
4492 
4493 /////
4494 // glibc on Linux platform uses non-documented flag
4495 // to indicate, that some special sort of signal
4496 // trampoline is used.
4497 // We will never set this flag, and we should
4498 // ignore this flag in our diagnostic
4499 #ifdef SIGNIFICANT_SIGNAL_MASK
4500   #undef SIGNIFICANT_SIGNAL_MASK
4501 #endif
4502 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4503 
4504 static const char* get_signal_handler_name(address handler,
4505                                            char* buf, int buflen) {
4506   int offset = 0;
4507   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4508   if (found) {
4509     // skip directory names
4510     const char *p1, *p2;
4511     p1 = buf;
4512     size_t len = strlen(os::file_separator());
4513     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4514     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4515   } else {
4516     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4517   }
4518   return buf;
4519 }
4520 
4521 static void print_signal_handler(outputStream* st, int sig,
4522                                  char* buf, size_t buflen) {
4523   struct sigaction sa;
4524 
4525   sigaction(sig, NULL, &sa);
4526 
4527   // See comment for SIGNIFICANT_SIGNAL_MASK define
4528   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4529 
4530   st->print("%s: ", os::exception_name(sig, buf, buflen));
4531 
4532   address handler = (sa.sa_flags & SA_SIGINFO)
4533     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4534     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4535 
4536   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4537     st->print("SIG_DFL");
4538   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4539     st->print("SIG_IGN");
4540   } else {
4541     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4542   }
4543 
4544   st->print(", sa_mask[0]=");
4545   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4546 
4547   address rh = VMError::get_resetted_sighandler(sig);
4548   // May be, handler was resetted by VMError?
4549   if (rh != NULL) {
4550     handler = rh;
4551     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4552   }
4553 
4554   st->print(", sa_flags=");
4555   os::Posix::print_sa_flags(st, sa.sa_flags);
4556 
4557   // Check: is it our handler?
4558   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4559       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4560     // It is our signal handler
4561     // check for flags, reset system-used one!
4562     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4563       st->print(
4564                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4565                 os::Linux::get_our_sigflags(sig));
4566     }
4567   }
4568   st->cr();
4569 }
4570 
4571 
4572 #define DO_SIGNAL_CHECK(sig)                      \
4573   do {                                            \
4574     if (!sigismember(&check_signal_done, sig)) {  \
4575       os::Linux::check_signal_handler(sig);       \
4576     }                                             \
4577   } while (0)
4578 
4579 // This method is a periodic task to check for misbehaving JNI applications
4580 // under CheckJNI, we can add any periodic checks here
4581 
4582 void os::run_periodic_checks() {
4583   if (check_signals == false) return;
4584 
4585   // SEGV and BUS if overridden could potentially prevent
4586   // generation of hs*.log in the event of a crash, debugging
4587   // such a case can be very challenging, so we absolutely
4588   // check the following for a good measure:
4589   DO_SIGNAL_CHECK(SIGSEGV);
4590   DO_SIGNAL_CHECK(SIGILL);
4591   DO_SIGNAL_CHECK(SIGFPE);
4592   DO_SIGNAL_CHECK(SIGBUS);
4593   DO_SIGNAL_CHECK(SIGPIPE);
4594   DO_SIGNAL_CHECK(SIGXFSZ);
4595 #if defined(PPC64)
4596   DO_SIGNAL_CHECK(SIGTRAP);
4597 #endif
4598 
4599   // ReduceSignalUsage allows the user to override these handlers
4600   // see comments at the very top and jvm_solaris.h
4601   if (!ReduceSignalUsage) {
4602     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4603     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4604     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4605     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4606   }
4607 
4608   DO_SIGNAL_CHECK(SR_signum);
4609 }
4610 
4611 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4612 
4613 static os_sigaction_t os_sigaction = NULL;
4614 
4615 void os::Linux::check_signal_handler(int sig) {
4616   char buf[O_BUFLEN];
4617   address jvmHandler = NULL;
4618 
4619 
4620   struct sigaction act;
4621   if (os_sigaction == NULL) {
4622     // only trust the default sigaction, in case it has been interposed
4623     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4624     if (os_sigaction == NULL) return;
4625   }
4626 
4627   os_sigaction(sig, (struct sigaction*)NULL, &act);
4628 
4629 
4630   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4631 
4632   address thisHandler = (act.sa_flags & SA_SIGINFO)
4633     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4634     : CAST_FROM_FN_PTR(address, act.sa_handler);
4635 
4636 
4637   switch (sig) {
4638   case SIGSEGV:
4639   case SIGBUS:
4640   case SIGFPE:
4641   case SIGPIPE:
4642   case SIGILL:
4643   case SIGXFSZ:
4644     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4645     break;
4646 
4647   case SHUTDOWN1_SIGNAL:
4648   case SHUTDOWN2_SIGNAL:
4649   case SHUTDOWN3_SIGNAL:
4650   case BREAK_SIGNAL:
4651     jvmHandler = (address)user_handler();
4652     break;
4653 
4654   default:
4655     if (sig == SR_signum) {
4656       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4657     } else {
4658       return;
4659     }
4660     break;
4661   }
4662 
4663   if (thisHandler != jvmHandler) {
4664     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4665     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4666     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4667     // No need to check this sig any longer
4668     sigaddset(&check_signal_done, sig);
4669     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4670     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4671       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4672                     exception_name(sig, buf, O_BUFLEN));
4673     }
4674   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4675     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4676     tty->print("expected:");
4677     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4678     tty->cr();
4679     tty->print("  found:");
4680     os::Posix::print_sa_flags(tty, act.sa_flags);
4681     tty->cr();
4682     // No need to check this sig any longer
4683     sigaddset(&check_signal_done, sig);
4684   }
4685 
4686   // Dump all the signal
4687   if (sigismember(&check_signal_done, sig)) {
4688     print_signal_handlers(tty, buf, O_BUFLEN);
4689   }
4690 }
4691 
4692 extern void report_error(char* file_name, int line_no, char* title,
4693                          char* format, ...);
4694 
4695 // this is called _before_ the most of global arguments have been parsed
4696 void os::init(void) {
4697   char dummy;   // used to get a guess on initial stack address
4698 //  first_hrtime = gethrtime();
4699 
4700   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4701 
4702   init_random(1234567);
4703 
4704   ThreadCritical::initialize();
4705 
4706   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4707   if (Linux::page_size() == -1) {
4708     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4709           os::strerror(errno));
4710   }
4711   init_page_sizes((size_t) Linux::page_size());
4712 
4713   Linux::initialize_system_info();
4714 
4715   Linux::initialize_os_info();
4716 
4717   // main_thread points to the aboriginal thread
4718   Linux::_main_thread = pthread_self();
4719 
4720   Linux::clock_init();
4721   initial_time_count = javaTimeNanos();
4722 
4723   // pthread_condattr initialization for monotonic clock
4724   int status;
4725   pthread_condattr_t* _condattr = os::Linux::condAttr();
4726   if ((status = pthread_condattr_init(_condattr)) != 0) {
4727     fatal("pthread_condattr_init: %s", os::strerror(status));
4728   }
4729   // Only set the clock if CLOCK_MONOTONIC is available
4730   if (os::supports_monotonic_clock()) {
4731     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4732       if (status == EINVAL) {
4733         warning("Unable to use monotonic clock with relative timed-waits" \
4734                 " - changes to the time-of-day clock may have adverse affects");
4735       } else {
4736         fatal("pthread_condattr_setclock: %s", os::strerror(status));
4737       }
4738     }
4739   }
4740   // else it defaults to CLOCK_REALTIME
4741 
4742   // retrieve entry point for pthread_setname_np
4743   Linux::_pthread_setname_np =
4744     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4745 
4746 }
4747 
4748 // To install functions for atexit system call
4749 extern "C" {
4750   static void perfMemory_exit_helper() {
4751     perfMemory_exit();
4752   }
4753 }
4754 
4755 // this is called _after_ the global arguments have been parsed
4756 jint os::init_2(void) {
4757   Linux::fast_thread_clock_init();
4758 
4759   // Allocate a single page and mark it as readable for safepoint polling
4760   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4761   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4762 
4763   os::set_polling_page(polling_page);
4764   log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4765 
4766   if (!UseMembar) {
4767     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4768     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4769     os::set_memory_serialize_page(mem_serialize_page);
4770     log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4771   }
4772 
4773   // initialize suspend/resume support - must do this before signal_sets_init()
4774   if (SR_initialize() != 0) {
4775     perror("SR_initialize failed");
4776     return JNI_ERR;
4777   }
4778 
4779   Linux::signal_sets_init();
4780   Linux::install_signal_handlers();
4781 
4782   // Check and sets minimum stack sizes against command line options
4783   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4784     return JNI_ERR;
4785   }
4786   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4787 
4788 #if defined(IA32)
4789   workaround_expand_exec_shield_cs_limit();
4790 #endif
4791 
4792   Linux::libpthread_init();
4793   log_info(os)("HotSpot is running with %s, %s",
4794                Linux::glibc_version(), Linux::libpthread_version());
4795 
4796   if (UseNUMA) {
4797     if (!Linux::libnuma_init()) {
4798       UseNUMA = false;
4799     } else {
4800       if ((Linux::numa_max_node() < 1)) {
4801         // There's only one node(they start from 0), disable NUMA.
4802         UseNUMA = false;
4803       }
4804     }
4805     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4806     // we can make the adaptive lgrp chunk resizing work. If the user specified
4807     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4808     // disable adaptive resizing.
4809     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4810       if (FLAG_IS_DEFAULT(UseNUMA)) {
4811         UseNUMA = false;
4812       } else {
4813         if (FLAG_IS_DEFAULT(UseLargePages) &&
4814             FLAG_IS_DEFAULT(UseSHM) &&
4815             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4816           UseLargePages = false;
4817         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4818           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4819           UseAdaptiveSizePolicy = false;
4820           UseAdaptiveNUMAChunkSizing = false;
4821         }
4822       }
4823     }
4824     if (!UseNUMA && ForceNUMA) {
4825       UseNUMA = true;
4826     }
4827   }
4828 
4829   if (MaxFDLimit) {
4830     // set the number of file descriptors to max. print out error
4831     // if getrlimit/setrlimit fails but continue regardless.
4832     struct rlimit nbr_files;
4833     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4834     if (status != 0) {
4835       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4836     } else {
4837       nbr_files.rlim_cur = nbr_files.rlim_max;
4838       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4839       if (status != 0) {
4840         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4841       }
4842     }
4843   }
4844 
4845   // Initialize lock used to serialize thread creation (see os::create_thread)
4846   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4847 
4848   // at-exit methods are called in the reverse order of their registration.
4849   // atexit functions are called on return from main or as a result of a
4850   // call to exit(3C). There can be only 32 of these functions registered
4851   // and atexit() does not set errno.
4852 
4853   if (PerfAllowAtExitRegistration) {
4854     // only register atexit functions if PerfAllowAtExitRegistration is set.
4855     // atexit functions can be delayed until process exit time, which
4856     // can be problematic for embedded VM situations. Embedded VMs should
4857     // call DestroyJavaVM() to assure that VM resources are released.
4858 
4859     // note: perfMemory_exit_helper atexit function may be removed in
4860     // the future if the appropriate cleanup code can be added to the
4861     // VM_Exit VMOperation's doit method.
4862     if (atexit(perfMemory_exit_helper) != 0) {
4863       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4864     }
4865   }
4866 
4867   // initialize thread priority policy
4868   prio_init();
4869 
4870   return JNI_OK;
4871 }
4872 
4873 // Mark the polling page as unreadable
4874 void os::make_polling_page_unreadable(void) {
4875   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4876     fatal("Could not disable polling page");
4877   }
4878 }
4879 
4880 // Mark the polling page as readable
4881 void os::make_polling_page_readable(void) {
4882   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4883     fatal("Could not enable polling page");
4884   }
4885 }
4886 
4887 // older glibc versions don't have this macro (which expands to
4888 // an optimized bit-counting function) so we have to roll our own
4889 #ifndef CPU_COUNT
4890 
4891 static int _cpu_count(const cpu_set_t* cpus) {
4892   int count = 0;
4893   // only look up to the number of configured processors
4894   for (int i = 0; i < os::processor_count(); i++) {
4895     if (CPU_ISSET(i, cpus)) {
4896       count++;
4897     }
4898   }
4899   return count;
4900 }
4901 
4902 #define CPU_COUNT(cpus) _cpu_count(cpus)
4903 
4904 #endif // CPU_COUNT
4905 
4906 // Get the current number of available processors for this process.
4907 // This value can change at any time during a process's lifetime.
4908 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4909 // If it appears there may be more than 1024 processors then we do a
4910 // dynamic check - see 6515172 for details.
4911 // If anything goes wrong we fallback to returning the number of online
4912 // processors - which can be greater than the number available to the process.
4913 int os::active_processor_count() {
4914   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
4915   cpu_set_t* cpus_p = &cpus;
4916   int cpus_size = sizeof(cpu_set_t);
4917 
4918   int configured_cpus = processor_count();  // upper bound on available cpus
4919   int cpu_count = 0;
4920 
4921 // old build platforms may not support dynamic cpu sets
4922 #ifdef CPU_ALLOC
4923 
4924   // To enable easy testing of the dynamic path on different platforms we
4925   // introduce a diagnostic flag: UseCpuAllocPath
4926   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4927     // kernel may use a mask bigger than cpu_set_t
4928     log_trace(os)("active_processor_count: using dynamic path %s"
4929                   "- configured processors: %d",
4930                   UseCpuAllocPath ? "(forced) " : "",
4931                   configured_cpus);
4932     cpus_p = CPU_ALLOC(configured_cpus);
4933     if (cpus_p != NULL) {
4934       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4935       // zero it just to be safe
4936       CPU_ZERO_S(cpus_size, cpus_p);
4937     }
4938     else {
4939        // failed to allocate so fallback to online cpus
4940        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4941        log_trace(os)("active_processor_count: "
4942                      "CPU_ALLOC failed (%s) - using "
4943                      "online processor count: %d",
4944                      os::strerror(errno), online_cpus);
4945        return online_cpus;
4946     }
4947   }
4948   else {
4949     log_trace(os)("active_processor_count: using static path - configured processors: %d",
4950                   configured_cpus);
4951   }
4952 #else // CPU_ALLOC
4953 // these stubs won't be executed
4954 #define CPU_COUNT_S(size, cpus) -1
4955 #define CPU_FREE(cpus)
4956 
4957   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4958                 configured_cpus);
4959 #endif // CPU_ALLOC
4960 
4961   // pid 0 means the current thread - which we have to assume represents the process
4962   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4963     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4964       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4965     }
4966     else {
4967       cpu_count = CPU_COUNT(cpus_p);
4968     }
4969     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4970   }
4971   else {
4972     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4973     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4974             "which may exceed available processors", os::strerror(errno), cpu_count);
4975   }
4976 
4977   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4978     CPU_FREE(cpus_p);
4979   }
4980 
4981   assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
4982   return cpu_count;
4983 }
4984 
4985 void os::set_native_thread_name(const char *name) {
4986   if (Linux::_pthread_setname_np) {
4987     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4988     snprintf(buf, sizeof(buf), "%s", name);
4989     buf[sizeof(buf) - 1] = '\0';
4990     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4991     // ERANGE should not happen; all other errors should just be ignored.
4992     assert(rc != ERANGE, "pthread_setname_np failed");
4993   }
4994 }
4995 
4996 bool os::distribute_processes(uint length, uint* distribution) {
4997   // Not yet implemented.
4998   return false;
4999 }
5000 
5001 bool os::bind_to_processor(uint processor_id) {
5002   // Not yet implemented.
5003   return false;
5004 }
5005 
5006 ///
5007 
5008 void os::SuspendedThreadTask::internal_do_task() {
5009   if (do_suspend(_thread->osthread())) {
5010     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5011     do_task(context);
5012     do_resume(_thread->osthread());
5013   }
5014 }
5015 
5016 class PcFetcher : public os::SuspendedThreadTask {
5017  public:
5018   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5019   ExtendedPC result();
5020  protected:
5021   void do_task(const os::SuspendedThreadTaskContext& context);
5022  private:
5023   ExtendedPC _epc;
5024 };
5025 
5026 ExtendedPC PcFetcher::result() {
5027   guarantee(is_done(), "task is not done yet.");
5028   return _epc;
5029 }
5030 
5031 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5032   Thread* thread = context.thread();
5033   OSThread* osthread = thread->osthread();
5034   if (osthread->ucontext() != NULL) {
5035     _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5036   } else {
5037     // NULL context is unexpected, double-check this is the VMThread
5038     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5039   }
5040 }
5041 
5042 // Suspends the target using the signal mechanism and then grabs the PC before
5043 // resuming the target. Used by the flat-profiler only
5044 ExtendedPC os::get_thread_pc(Thread* thread) {
5045   // Make sure that it is called by the watcher for the VMThread
5046   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5047   assert(thread->is_VM_thread(), "Can only be called for VMThread");
5048 
5049   PcFetcher fetcher(thread);
5050   fetcher.run();
5051   return fetcher.result();
5052 }
5053 
5054 ////////////////////////////////////////////////////////////////////////////////
5055 // debug support
5056 
5057 bool os::find(address addr, outputStream* st) {
5058   Dl_info dlinfo;
5059   memset(&dlinfo, 0, sizeof(dlinfo));
5060   if (dladdr(addr, &dlinfo) != 0) {
5061     st->print(PTR_FORMAT ": ", p2i(addr));
5062     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5063       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5064                 p2i(addr) - p2i(dlinfo.dli_saddr));
5065     } else if (dlinfo.dli_fbase != NULL) {
5066       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5067     } else {
5068       st->print("<absolute address>");
5069     }
5070     if (dlinfo.dli_fname != NULL) {
5071       st->print(" in %s", dlinfo.dli_fname);
5072     }
5073     if (dlinfo.dli_fbase != NULL) {
5074       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5075     }
5076     st->cr();
5077 
5078     if (Verbose) {
5079       // decode some bytes around the PC
5080       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5081       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5082       address       lowest = (address) dlinfo.dli_sname;
5083       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5084       if (begin < lowest)  begin = lowest;
5085       Dl_info dlinfo2;
5086       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5087           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5088         end = (address) dlinfo2.dli_saddr;
5089       }
5090       Disassembler::decode(begin, end, st);
5091     }
5092     return true;
5093   }
5094   return false;
5095 }
5096 
5097 ////////////////////////////////////////////////////////////////////////////////
5098 // misc
5099 
5100 // This does not do anything on Linux. This is basically a hook for being
5101 // able to use structured exception handling (thread-local exception filters)
5102 // on, e.g., Win32.
5103 void
5104 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5105                          JavaCallArguments* args, Thread* thread) {
5106   f(value, method, args, thread);
5107 }
5108 
5109 void os::print_statistics() {
5110 }
5111 
5112 bool os::message_box(const char* title, const char* message) {
5113   int i;
5114   fdStream err(defaultStream::error_fd());
5115   for (i = 0; i < 78; i++) err.print_raw("=");
5116   err.cr();
5117   err.print_raw_cr(title);
5118   for (i = 0; i < 78; i++) err.print_raw("-");
5119   err.cr();
5120   err.print_raw_cr(message);
5121   for (i = 0; i < 78; i++) err.print_raw("=");
5122   err.cr();
5123 
5124   char buf[16];
5125   // Prevent process from exiting upon "read error" without consuming all CPU
5126   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5127 
5128   return buf[0] == 'y' || buf[0] == 'Y';
5129 }
5130 
5131 int os::stat(const char *path, struct stat *sbuf) {
5132   char pathbuf[MAX_PATH];
5133   if (strlen(path) > MAX_PATH - 1) {
5134     errno = ENAMETOOLONG;
5135     return -1;
5136   }
5137   os::native_path(strcpy(pathbuf, path));
5138   return ::stat(pathbuf, sbuf);
5139 }
5140 
5141 // Is a (classpath) directory empty?
5142 bool os::dir_is_empty(const char* path) {
5143   DIR *dir = NULL;
5144   struct dirent *ptr;
5145 
5146   dir = opendir(path);
5147   if (dir == NULL) return true;
5148 
5149   // Scan the directory
5150   bool result = true;
5151   char buf[sizeof(struct dirent) + MAX_PATH];
5152   while (result && (ptr = ::readdir(dir)) != NULL) {
5153     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5154       result = false;
5155     }
5156   }
5157   closedir(dir);
5158   return result;
5159 }
5160 
5161 // This code originates from JDK's sysOpen and open64_w
5162 // from src/solaris/hpi/src/system_md.c
5163 
5164 int os::open(const char *path, int oflag, int mode) {
5165   if (strlen(path) > MAX_PATH - 1) {
5166     errno = ENAMETOOLONG;
5167     return -1;
5168   }
5169 
5170   // All file descriptors that are opened in the Java process and not
5171   // specifically destined for a subprocess should have the close-on-exec
5172   // flag set.  If we don't set it, then careless 3rd party native code
5173   // might fork and exec without closing all appropriate file descriptors
5174   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5175   // turn might:
5176   //
5177   // - cause end-of-file to fail to be detected on some file
5178   //   descriptors, resulting in mysterious hangs, or
5179   //
5180   // - might cause an fopen in the subprocess to fail on a system
5181   //   suffering from bug 1085341.
5182   //
5183   // (Yes, the default setting of the close-on-exec flag is a Unix
5184   // design flaw)
5185   //
5186   // See:
5187   // 1085341: 32-bit stdio routines should support file descriptors >255
5188   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5189   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5190   //
5191   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5192   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5193   // because it saves a system call and removes a small window where the flag
5194   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5195   // and we fall back to using FD_CLOEXEC (see below).
5196 #ifdef O_CLOEXEC
5197   oflag |= O_CLOEXEC;
5198 #endif
5199 
5200   int fd = ::open64(path, oflag, mode);
5201   if (fd == -1) return -1;
5202 
5203   //If the open succeeded, the file might still be a directory
5204   {
5205     struct stat64 buf64;
5206     int ret = ::fstat64(fd, &buf64);
5207     int st_mode = buf64.st_mode;
5208 
5209     if (ret != -1) {
5210       if ((st_mode & S_IFMT) == S_IFDIR) {
5211         errno = EISDIR;
5212         ::close(fd);
5213         return -1;
5214       }
5215     } else {
5216       ::close(fd);
5217       return -1;
5218     }
5219   }
5220 
5221 #ifdef FD_CLOEXEC
5222   // Validate that the use of the O_CLOEXEC flag on open above worked.
5223   // With recent kernels, we will perform this check exactly once.
5224   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5225   if (!O_CLOEXEC_is_known_to_work) {
5226     int flags = ::fcntl(fd, F_GETFD);
5227     if (flags != -1) {
5228       if ((flags & FD_CLOEXEC) != 0)
5229         O_CLOEXEC_is_known_to_work = 1;
5230       else
5231         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5232     }
5233   }
5234 #endif
5235 
5236   return fd;
5237 }
5238 
5239 
5240 // create binary file, rewriting existing file if required
5241 int os::create_binary_file(const char* path, bool rewrite_existing) {
5242   int oflags = O_WRONLY | O_CREAT;
5243   if (!rewrite_existing) {
5244     oflags |= O_EXCL;
5245   }
5246   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5247 }
5248 
5249 // return current position of file pointer
5250 jlong os::current_file_offset(int fd) {
5251   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5252 }
5253 
5254 // move file pointer to the specified offset
5255 jlong os::seek_to_file_offset(int fd, jlong offset) {
5256   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5257 }
5258 
5259 // This code originates from JDK's sysAvailable
5260 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5261 
5262 int os::available(int fd, jlong *bytes) {
5263   jlong cur, end;
5264   int mode;
5265   struct stat64 buf64;
5266 
5267   if (::fstat64(fd, &buf64) >= 0) {
5268     mode = buf64.st_mode;
5269     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5270       int n;
5271       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5272         *bytes = n;
5273         return 1;
5274       }
5275     }
5276   }
5277   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5278     return 0;
5279   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5280     return 0;
5281   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5282     return 0;
5283   }
5284   *bytes = end - cur;
5285   return 1;
5286 }
5287 
5288 // Map a block of memory.
5289 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5290                         char *addr, size_t bytes, bool read_only,
5291                         bool allow_exec) {
5292   int prot;
5293   int flags = MAP_PRIVATE;
5294 
5295   if (read_only) {
5296     prot = PROT_READ;
5297   } else {
5298     prot = PROT_READ | PROT_WRITE;
5299   }
5300 
5301   if (allow_exec) {
5302     prot |= PROT_EXEC;
5303   }
5304 
5305   if (addr != NULL) {
5306     flags |= MAP_FIXED;
5307   }
5308 
5309   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5310                                      fd, file_offset);
5311   if (mapped_address == MAP_FAILED) {
5312     return NULL;
5313   }
5314   return mapped_address;
5315 }
5316 
5317 
5318 // Remap a block of memory.
5319 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5320                           char *addr, size_t bytes, bool read_only,
5321                           bool allow_exec) {
5322   // same as map_memory() on this OS
5323   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5324                         allow_exec);
5325 }
5326 
5327 
5328 // Unmap a block of memory.
5329 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5330   return munmap(addr, bytes) == 0;
5331 }
5332 
5333 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5334 
5335 static clockid_t thread_cpu_clockid(Thread* thread) {
5336   pthread_t tid = thread->osthread()->pthread_id();
5337   clockid_t clockid;
5338 
5339   // Get thread clockid
5340   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5341   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5342   return clockid;
5343 }
5344 
5345 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5346 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5347 // of a thread.
5348 //
5349 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5350 // the fast estimate available on the platform.
5351 
5352 jlong os::current_thread_cpu_time() {
5353   if (os::Linux::supports_fast_thread_cpu_time()) {
5354     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5355   } else {
5356     // return user + sys since the cost is the same
5357     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5358   }
5359 }
5360 
5361 jlong os::thread_cpu_time(Thread* thread) {
5362   // consistent with what current_thread_cpu_time() returns
5363   if (os::Linux::supports_fast_thread_cpu_time()) {
5364     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5365   } else {
5366     return slow_thread_cpu_time(thread, true /* user + sys */);
5367   }
5368 }
5369 
5370 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5371   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5372     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5373   } else {
5374     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5375   }
5376 }
5377 
5378 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5379   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5380     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5381   } else {
5382     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5383   }
5384 }
5385 
5386 //  -1 on error.
5387 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5388   pid_t  tid = thread->osthread()->thread_id();
5389   char *s;
5390   char stat[2048];
5391   int statlen;
5392   char proc_name[64];
5393   int count;
5394   long sys_time, user_time;
5395   char cdummy;
5396   int idummy;
5397   long ldummy;
5398   FILE *fp;
5399 
5400   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5401   fp = fopen(proc_name, "r");
5402   if (fp == NULL) return -1;
5403   statlen = fread(stat, 1, 2047, fp);
5404   stat[statlen] = '\0';
5405   fclose(fp);
5406 
5407   // Skip pid and the command string. Note that we could be dealing with
5408   // weird command names, e.g. user could decide to rename java launcher
5409   // to "java 1.4.2 :)", then the stat file would look like
5410   //                1234 (java 1.4.2 :)) R ... ...
5411   // We don't really need to know the command string, just find the last
5412   // occurrence of ")" and then start parsing from there. See bug 4726580.
5413   s = strrchr(stat, ')');
5414   if (s == NULL) return -1;
5415 
5416   // Skip blank chars
5417   do { s++; } while (s && isspace(*s));
5418 
5419   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5420                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5421                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5422                  &user_time, &sys_time);
5423   if (count != 13) return -1;
5424   if (user_sys_cpu_time) {
5425     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5426   } else {
5427     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5428   }
5429 }
5430 
5431 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5432   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5433   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5434   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5435   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5436 }
5437 
5438 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5439   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5440   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5441   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5442   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5443 }
5444 
5445 bool os::is_thread_cpu_time_supported() {
5446   return true;
5447 }
5448 
5449 // System loadavg support.  Returns -1 if load average cannot be obtained.
5450 // Linux doesn't yet have a (official) notion of processor sets,
5451 // so just return the system wide load average.
5452 int os::loadavg(double loadavg[], int nelem) {
5453   return ::getloadavg(loadavg, nelem);
5454 }
5455 
5456 void os::pause() {
5457   char filename[MAX_PATH];
5458   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5459     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5460   } else {
5461     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5462   }
5463 
5464   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5465   if (fd != -1) {
5466     struct stat buf;
5467     ::close(fd);
5468     while (::stat(filename, &buf) == 0) {
5469       (void)::poll(NULL, 0, 100);
5470     }
5471   } else {
5472     jio_fprintf(stderr,
5473                 "Could not open pause file '%s', continuing immediately.\n", filename);
5474   }
5475 }
5476 
5477 
5478 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5479 // comment paragraphs are worth repeating here:
5480 //
5481 // Assumption:
5482 //    Only one parker can exist on an event, which is why we allocate
5483 //    them per-thread. Multiple unparkers can coexist.
5484 //
5485 // _Event serves as a restricted-range semaphore.
5486 //   -1 : thread is blocked, i.e. there is a waiter
5487 //    0 : neutral: thread is running or ready,
5488 //        could have been signaled after a wait started
5489 //    1 : signaled - thread is running or ready
5490 //
5491 
5492 // utility to compute the abstime argument to timedwait:
5493 // millis is the relative timeout time
5494 // abstime will be the absolute timeout time
5495 // TODO: replace compute_abstime() with unpackTime()
5496 
5497 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5498   if (millis < 0)  millis = 0;
5499 
5500   jlong seconds = millis / 1000;
5501   millis %= 1000;
5502   if (seconds > 50000000) { // see man cond_timedwait(3T)
5503     seconds = 50000000;
5504   }
5505 
5506   if (os::supports_monotonic_clock()) {
5507     struct timespec now;
5508     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5509     assert_status(status == 0, status, "clock_gettime");
5510     abstime->tv_sec = now.tv_sec  + seconds;
5511     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5512     if (nanos >= NANOSECS_PER_SEC) {
5513       abstime->tv_sec += 1;
5514       nanos -= NANOSECS_PER_SEC;
5515     }
5516     abstime->tv_nsec = nanos;
5517   } else {
5518     struct timeval now;
5519     int status = gettimeofday(&now, NULL);
5520     assert(status == 0, "gettimeofday");
5521     abstime->tv_sec = now.tv_sec  + seconds;
5522     long usec = now.tv_usec + millis * 1000;
5523     if (usec >= 1000000) {
5524       abstime->tv_sec += 1;
5525       usec -= 1000000;
5526     }
5527     abstime->tv_nsec = usec * 1000;
5528   }
5529   return abstime;
5530 }
5531 
5532 void os::PlatformEvent::park() {       // AKA "down()"
5533   // Transitions for _Event:
5534   //   -1 => -1 : illegal
5535   //    1 =>  0 : pass - return immediately
5536   //    0 => -1 : block; then set _Event to 0 before returning
5537 
5538   // Invariant: Only the thread associated with the Event/PlatformEvent
5539   // may call park().
5540   // TODO: assert that _Assoc != NULL or _Assoc == Self
5541   assert(_nParked == 0, "invariant");
5542 
5543   int v;
5544   for (;;) {
5545     v = _Event;
5546     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5547   }
5548   guarantee(v >= 0, "invariant");
5549   if (v == 0) {
5550     // Do this the hard way by blocking ...
5551     int status = pthread_mutex_lock(_mutex);
5552     assert_status(status == 0, status, "mutex_lock");
5553     guarantee(_nParked == 0, "invariant");
5554     ++_nParked;
5555     while (_Event < 0) {
5556       status = pthread_cond_wait(_cond, _mutex);
5557       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5558       // Treat this the same as if the wait was interrupted
5559       if (status == ETIME) { status = EINTR; }
5560       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5561     }
5562     --_nParked;
5563 
5564     _Event = 0;
5565     status = pthread_mutex_unlock(_mutex);
5566     assert_status(status == 0, status, "mutex_unlock");
5567     // Paranoia to ensure our locked and lock-free paths interact
5568     // correctly with each other.
5569     OrderAccess::fence();
5570   }
5571   guarantee(_Event >= 0, "invariant");
5572 }
5573 
5574 int os::PlatformEvent::park(jlong millis) {
5575   // Transitions for _Event:
5576   //   -1 => -1 : illegal
5577   //    1 =>  0 : pass - return immediately
5578   //    0 => -1 : block; then set _Event to 0 before returning
5579 
5580   guarantee(_nParked == 0, "invariant");
5581 
5582   int v;
5583   for (;;) {
5584     v = _Event;
5585     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5586   }
5587   guarantee(v >= 0, "invariant");
5588   if (v != 0) return OS_OK;
5589 
5590   // We do this the hard way, by blocking the thread.
5591   // Consider enforcing a minimum timeout value.
5592   struct timespec abst;
5593   compute_abstime(&abst, millis);
5594 
5595   int ret = OS_TIMEOUT;
5596   int status = pthread_mutex_lock(_mutex);
5597   assert_status(status == 0, status, "mutex_lock");
5598   guarantee(_nParked == 0, "invariant");
5599   ++_nParked;
5600 
5601   // Object.wait(timo) will return because of
5602   // (a) notification
5603   // (b) timeout
5604   // (c) thread.interrupt
5605   //
5606   // Thread.interrupt and object.notify{All} both call Event::set.
5607   // That is, we treat thread.interrupt as a special case of notification.
5608   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5609   // We assume all ETIME returns are valid.
5610   //
5611   // TODO: properly differentiate simultaneous notify+interrupt.
5612   // In that case, we should propagate the notify to another waiter.
5613 
5614   while (_Event < 0) {
5615     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5616     assert_status(status == 0 || status == EINTR ||
5617                   status == ETIME || status == ETIMEDOUT,
5618                   status, "cond_timedwait");
5619     if (!FilterSpuriousWakeups) break;                 // previous semantics
5620     if (status == ETIME || status == ETIMEDOUT) break;
5621     // We consume and ignore EINTR and spurious wakeups.
5622   }
5623   --_nParked;
5624   if (_Event >= 0) {
5625     ret = OS_OK;
5626   }
5627   _Event = 0;
5628   status = pthread_mutex_unlock(_mutex);
5629   assert_status(status == 0, status, "mutex_unlock");
5630   assert(_nParked == 0, "invariant");
5631   // Paranoia to ensure our locked and lock-free paths interact
5632   // correctly with each other.
5633   OrderAccess::fence();
5634   return ret;
5635 }
5636 
5637 void os::PlatformEvent::unpark() {
5638   // Transitions for _Event:
5639   //    0 => 1 : just return
5640   //    1 => 1 : just return
5641   //   -1 => either 0 or 1; must signal target thread
5642   //         That is, we can safely transition _Event from -1 to either
5643   //         0 or 1.
5644   // See also: "Semaphores in Plan 9" by Mullender & Cox
5645   //
5646   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5647   // that it will take two back-to-back park() calls for the owning
5648   // thread to block. This has the benefit of forcing a spurious return
5649   // from the first park() call after an unpark() call which will help
5650   // shake out uses of park() and unpark() without condition variables.
5651 
5652   if (Atomic::xchg(1, &_Event) >= 0) return;
5653 
5654   // Wait for the thread associated with the event to vacate
5655   int status = pthread_mutex_lock(_mutex);
5656   assert_status(status == 0, status, "mutex_lock");
5657   int AnyWaiters = _nParked;
5658   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5659   status = pthread_mutex_unlock(_mutex);
5660   assert_status(status == 0, status, "mutex_unlock");
5661   if (AnyWaiters != 0) {
5662     // Note that we signal() *after* dropping the lock for "immortal" Events.
5663     // This is safe and avoids a common class of  futile wakeups.  In rare
5664     // circumstances this can cause a thread to return prematurely from
5665     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5666     // will simply re-test the condition and re-park itself.
5667     // This provides particular benefit if the underlying platform does not
5668     // provide wait morphing.
5669     status = pthread_cond_signal(_cond);
5670     assert_status(status == 0, status, "cond_signal");
5671   }
5672 }
5673 
5674 
5675 // JSR166
5676 // -------------------------------------------------------
5677 
5678 // The solaris and linux implementations of park/unpark are fairly
5679 // conservative for now, but can be improved. They currently use a
5680 // mutex/condvar pair, plus a a count.
5681 // Park decrements count if > 0, else does a condvar wait.  Unpark
5682 // sets count to 1 and signals condvar.  Only one thread ever waits
5683 // on the condvar. Contention seen when trying to park implies that someone
5684 // is unparking you, so don't wait. And spurious returns are fine, so there
5685 // is no need to track notifications.
5686 
5687 // This code is common to linux and solaris and will be moved to a
5688 // common place in dolphin.
5689 //
5690 // The passed in time value is either a relative time in nanoseconds
5691 // or an absolute time in milliseconds. Either way it has to be unpacked
5692 // into suitable seconds and nanoseconds components and stored in the
5693 // given timespec structure.
5694 // Given time is a 64-bit value and the time_t used in the timespec is only
5695 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5696 // overflow if times way in the future are given. Further on Solaris versions
5697 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5698 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5699 // As it will be 28 years before "now + 100000000" will overflow we can
5700 // ignore overflow and just impose a hard-limit on seconds using the value
5701 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5702 // years from "now".
5703 
5704 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5705   assert(time > 0, "convertTime");
5706   time_t max_secs = 0;
5707 
5708   if (!os::supports_monotonic_clock() || isAbsolute) {
5709     struct timeval now;
5710     int status = gettimeofday(&now, NULL);
5711     assert(status == 0, "gettimeofday");
5712 
5713     max_secs = now.tv_sec + MAX_SECS;
5714 
5715     if (isAbsolute) {
5716       jlong secs = time / 1000;
5717       if (secs > max_secs) {
5718         absTime->tv_sec = max_secs;
5719       } else {
5720         absTime->tv_sec = secs;
5721       }
5722       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5723     } else {
5724       jlong secs = time / NANOSECS_PER_SEC;
5725       if (secs >= MAX_SECS) {
5726         absTime->tv_sec = max_secs;
5727         absTime->tv_nsec = 0;
5728       } else {
5729         absTime->tv_sec = now.tv_sec + secs;
5730         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5731         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5732           absTime->tv_nsec -= NANOSECS_PER_SEC;
5733           ++absTime->tv_sec; // note: this must be <= max_secs
5734         }
5735       }
5736     }
5737   } else {
5738     // must be relative using monotonic clock
5739     struct timespec now;
5740     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5741     assert_status(status == 0, status, "clock_gettime");
5742     max_secs = now.tv_sec + MAX_SECS;
5743     jlong secs = time / NANOSECS_PER_SEC;
5744     if (secs >= MAX_SECS) {
5745       absTime->tv_sec = max_secs;
5746       absTime->tv_nsec = 0;
5747     } else {
5748       absTime->tv_sec = now.tv_sec + secs;
5749       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5750       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5751         absTime->tv_nsec -= NANOSECS_PER_SEC;
5752         ++absTime->tv_sec; // note: this must be <= max_secs
5753       }
5754     }
5755   }
5756   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5757   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5758   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5759   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5760 }
5761 
5762 void Parker::park(bool isAbsolute, jlong time) {
5763   // Ideally we'd do something useful while spinning, such
5764   // as calling unpackTime().
5765 
5766   // Optional fast-path check:
5767   // Return immediately if a permit is available.
5768   // We depend on Atomic::xchg() having full barrier semantics
5769   // since we are doing a lock-free update to _counter.
5770   if (Atomic::xchg(0, &_counter) > 0) return;
5771 
5772   Thread* thread = Thread::current();
5773   assert(thread->is_Java_thread(), "Must be JavaThread");
5774   JavaThread *jt = (JavaThread *)thread;
5775 
5776   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5777   // Check interrupt before trying to wait
5778   if (Thread::is_interrupted(thread, false)) {
5779     return;
5780   }
5781 
5782   // Next, demultiplex/decode time arguments
5783   timespec absTime;
5784   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5785     return;
5786   }
5787   if (time > 0) {
5788     unpackTime(&absTime, isAbsolute, time);
5789   }
5790 
5791 
5792   // Enter safepoint region
5793   // Beware of deadlocks such as 6317397.
5794   // The per-thread Parker:: mutex is a classic leaf-lock.
5795   // In particular a thread must never block on the Threads_lock while
5796   // holding the Parker:: mutex.  If safepoints are pending both the
5797   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5798   ThreadBlockInVM tbivm(jt);
5799 
5800   // Don't wait if cannot get lock since interference arises from
5801   // unblocking.  Also. check interrupt before trying wait
5802   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5803     return;
5804   }
5805 
5806   int status;
5807   if (_counter > 0)  { // no wait needed
5808     _counter = 0;
5809     status = pthread_mutex_unlock(_mutex);
5810     assert_status(status == 0, status, "invariant");
5811     // Paranoia to ensure our locked and lock-free paths interact
5812     // correctly with each other and Java-level accesses.
5813     OrderAccess::fence();
5814     return;
5815   }
5816 
5817 #ifdef ASSERT
5818   // Don't catch signals while blocked; let the running threads have the signals.
5819   // (This allows a debugger to break into the running thread.)
5820   sigset_t oldsigs;
5821   sigemptyset(&oldsigs);
5822   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5823   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5824 #endif
5825 
5826   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5827   jt->set_suspend_equivalent();
5828   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5829 
5830   assert(_cur_index == -1, "invariant");
5831   if (time == 0) {
5832     _cur_index = REL_INDEX; // arbitrary choice when not timed
5833     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5834   } else {
5835     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5836     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5837   }
5838   _cur_index = -1;
5839   assert_status(status == 0 || status == EINTR ||
5840                 status == ETIME || status == ETIMEDOUT,
5841                 status, "cond_timedwait");
5842 
5843 #ifdef ASSERT
5844   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5845 #endif
5846 
5847   _counter = 0;
5848   status = pthread_mutex_unlock(_mutex);
5849   assert_status(status == 0, status, "invariant");
5850   // Paranoia to ensure our locked and lock-free paths interact
5851   // correctly with each other and Java-level accesses.
5852   OrderAccess::fence();
5853 
5854   // If externally suspended while waiting, re-suspend
5855   if (jt->handle_special_suspend_equivalent_condition()) {
5856     jt->java_suspend_self();
5857   }
5858 }
5859 
5860 void Parker::unpark() {
5861   int status = pthread_mutex_lock(_mutex);
5862   assert_status(status == 0, status, "invariant");
5863   const int s = _counter;
5864   _counter = 1;
5865   // must capture correct index before unlocking
5866   int index = _cur_index;
5867   status = pthread_mutex_unlock(_mutex);
5868   assert_status(status == 0, status, "invariant");
5869   if (s < 1 && index != -1) {
5870     // thread is definitely parked
5871     status = pthread_cond_signal(&_cond[index]);
5872     assert_status(status == 0, status, "invariant");
5873   }
5874 }
5875 
5876 
5877 extern char** environ;
5878 
5879 // Run the specified command in a separate process. Return its exit value,
5880 // or -1 on failure (e.g. can't fork a new process).
5881 // Unlike system(), this function can be called from signal handler. It
5882 // doesn't block SIGINT et al.
5883 int os::fork_and_exec(char* cmd) {
5884   const char * argv[4] = {"sh", "-c", cmd, NULL};
5885 
5886   pid_t pid = fork();
5887 
5888   if (pid < 0) {
5889     // fork failed
5890     return -1;
5891 
5892   } else if (pid == 0) {
5893     // child process
5894 
5895     execve("/bin/sh", (char* const*)argv, environ);
5896 
5897     // execve failed
5898     _exit(-1);
5899 
5900   } else  {
5901     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5902     // care about the actual exit code, for now.
5903 
5904     int status;
5905 
5906     // Wait for the child process to exit.  This returns immediately if
5907     // the child has already exited. */
5908     while (waitpid(pid, &status, 0) < 0) {
5909       switch (errno) {
5910       case ECHILD: return 0;
5911       case EINTR: break;
5912       default: return -1;
5913       }
5914     }
5915 
5916     if (WIFEXITED(status)) {
5917       // The child exited normally; get its exit code.
5918       return WEXITSTATUS(status);
5919     } else if (WIFSIGNALED(status)) {
5920       // The child exited because of a signal
5921       // The best value to return is 0x80 + signal number,
5922       // because that is what all Unix shells do, and because
5923       // it allows callers to distinguish between process exit and
5924       // process death by signal.
5925       return 0x80 + WTERMSIG(status);
5926     } else {
5927       // Unknown exit code; pass it through
5928       return status;
5929     }
5930   }
5931 }
5932 
5933 // is_headless_jre()
5934 //
5935 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5936 // in order to report if we are running in a headless jre
5937 //
5938 // Since JDK8 xawt/libmawt.so was moved into the same directory
5939 // as libawt.so, and renamed libawt_xawt.so
5940 //
5941 bool os::is_headless_jre() {
5942   struct stat statbuf;
5943   char buf[MAXPATHLEN];
5944   char libmawtpath[MAXPATHLEN];
5945   const char *xawtstr  = "/xawt/libmawt.so";
5946   const char *new_xawtstr = "/libawt_xawt.so";
5947   char *p;
5948 
5949   // Get path to libjvm.so
5950   os::jvm_path(buf, sizeof(buf));
5951 
5952   // Get rid of libjvm.so
5953   p = strrchr(buf, '/');
5954   if (p == NULL) {
5955     return false;
5956   } else {
5957     *p = '\0';
5958   }
5959 
5960   // Get rid of client or server
5961   p = strrchr(buf, '/');
5962   if (p == NULL) {
5963     return false;
5964   } else {
5965     *p = '\0';
5966   }
5967 
5968   // check xawt/libmawt.so
5969   strcpy(libmawtpath, buf);
5970   strcat(libmawtpath, xawtstr);
5971   if (::stat(libmawtpath, &statbuf) == 0) return false;
5972 
5973   // check libawt_xawt.so
5974   strcpy(libmawtpath, buf);
5975   strcat(libmawtpath, new_xawtstr);
5976   if (::stat(libmawtpath, &statbuf) == 0) return false;
5977 
5978   return true;
5979 }
5980 
5981 // Get the default path to the core file
5982 // Returns the length of the string
5983 int os::get_core_path(char* buffer, size_t bufferSize) {
5984   /*
5985    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5986    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5987    */
5988   const int core_pattern_len = 129;
5989   char core_pattern[core_pattern_len] = {0};
5990 
5991   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5992   if (core_pattern_file == -1) {
5993     return -1;
5994   }
5995 
5996   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5997   ::close(core_pattern_file);
5998   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5999     return -1;
6000   }
6001   if (core_pattern[ret-1] == '\n') {
6002     core_pattern[ret-1] = '\0';
6003   } else {
6004     core_pattern[ret] = '\0';
6005   }
6006 
6007   char *pid_pos = strstr(core_pattern, "%p");
6008   int written;
6009 
6010   if (core_pattern[0] == '/') {
6011     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6012   } else {
6013     char cwd[PATH_MAX];
6014 
6015     const char* p = get_current_directory(cwd, PATH_MAX);
6016     if (p == NULL) {
6017       return -1;
6018     }
6019 
6020     if (core_pattern[0] == '|') {
6021       written = jio_snprintf(buffer, bufferSize,
6022                              "\"%s\" (or dumping to %s/core.%d)",
6023                              &core_pattern[1], p, current_process_id());
6024     } else {
6025       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6026     }
6027   }
6028 
6029   if (written < 0) {
6030     return -1;
6031   }
6032 
6033   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6034     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6035 
6036     if (core_uses_pid_file != -1) {
6037       char core_uses_pid = 0;
6038       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6039       ::close(core_uses_pid_file);
6040 
6041       if (core_uses_pid == '1') {
6042         jio_snprintf(buffer + written, bufferSize - written,
6043                                           ".%d", current_process_id());
6044       }
6045     }
6046   }
6047 
6048   return strlen(buffer);
6049 }
6050 
6051 bool os::start_debugging(char *buf, int buflen) {
6052   int len = (int)strlen(buf);
6053   char *p = &buf[len];
6054 
6055   jio_snprintf(p, buflen-len,
6056                "\n\n"
6057                "Do you want to debug the problem?\n\n"
6058                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6059                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6060                "Otherwise, press RETURN to abort...",
6061                os::current_process_id(), os::current_process_id(),
6062                os::current_thread_id(), os::current_thread_id());
6063 
6064   bool yes = os::message_box("Unexpected Error", buf);
6065 
6066   if (yes) {
6067     // yes, user asked VM to launch debugger
6068     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6069                  os::current_process_id(), os::current_process_id());
6070 
6071     os::fork_and_exec(buf);
6072     yes = false;
6073   }
6074   return yes;
6075 }
6076 
6077 static inline struct timespec get_mtime(const char* filename) {
6078   struct stat st;
6079   int ret = os::stat(filename, &st);
6080   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6081   return st.st_mtim;
6082 }
6083 
6084 int os::compare_file_modified_times(const char* file1, const char* file2) {
6085   struct timespec filetime1 = get_mtime(file1);
6086   struct timespec filetime2 = get_mtime(file2);
6087   int diff = filetime1.tv_sec - filetime2.tv_sec;
6088   if (diff == 0) {
6089     return filetime1.tv_nsec - filetime2.tv_nsec;
6090   }
6091   return diff;
6092 }
6093 
6094 /////////////// Unit tests ///////////////
6095 
6096 #ifndef PRODUCT
6097 
6098 #define test_log(...)              \
6099   do {                             \
6100     if (VerboseInternalVMTests) {  \
6101       tty->print_cr(__VA_ARGS__);  \
6102       tty->flush();                \
6103     }                              \
6104   } while (false)
6105 
6106 class TestReserveMemorySpecial : AllStatic {
6107  public:
6108   static void small_page_write(void* addr, size_t size) {
6109     size_t page_size = os::vm_page_size();
6110 
6111     char* end = (char*)addr + size;
6112     for (char* p = (char*)addr; p < end; p += page_size) {
6113       *p = 1;
6114     }
6115   }
6116 
6117   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6118     if (!UseHugeTLBFS) {
6119       return;
6120     }
6121 
6122     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6123 
6124     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6125 
6126     if (addr != NULL) {
6127       small_page_write(addr, size);
6128 
6129       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6130     }
6131   }
6132 
6133   static void test_reserve_memory_special_huge_tlbfs_only() {
6134     if (!UseHugeTLBFS) {
6135       return;
6136     }
6137 
6138     size_t lp = os::large_page_size();
6139 
6140     for (size_t size = lp; size <= lp * 10; size += lp) {
6141       test_reserve_memory_special_huge_tlbfs_only(size);
6142     }
6143   }
6144 
6145   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6146     size_t lp = os::large_page_size();
6147     size_t ag = os::vm_allocation_granularity();
6148 
6149     // sizes to test
6150     const size_t sizes[] = {
6151       lp, lp + ag, lp + lp / 2, lp * 2,
6152       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6153       lp * 10, lp * 10 + lp / 2
6154     };
6155     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6156 
6157     // For each size/alignment combination, we test three scenarios:
6158     // 1) with req_addr == NULL
6159     // 2) with a non-null req_addr at which we expect to successfully allocate
6160     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6161     //    expect the allocation to either fail or to ignore req_addr
6162 
6163     // Pre-allocate two areas; they shall be as large as the largest allocation
6164     //  and aligned to the largest alignment we will be testing.
6165     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6166     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6167       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6168       -1, 0);
6169     assert(mapping1 != MAP_FAILED, "should work");
6170 
6171     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6172       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6173       -1, 0);
6174     assert(mapping2 != MAP_FAILED, "should work");
6175 
6176     // Unmap the first mapping, but leave the second mapping intact: the first
6177     // mapping will serve as a value for a "good" req_addr (case 2). The second
6178     // mapping, still intact, as "bad" req_addr (case 3).
6179     ::munmap(mapping1, mapping_size);
6180 
6181     // Case 1
6182     test_log("%s, req_addr NULL:", __FUNCTION__);
6183     test_log("size            align           result");
6184 
6185     for (int i = 0; i < num_sizes; i++) {
6186       const size_t size = sizes[i];
6187       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6188         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6189         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6190                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6191         if (p != NULL) {
6192           assert(is_ptr_aligned(p, alignment), "must be");
6193           small_page_write(p, size);
6194           os::Linux::release_memory_special_huge_tlbfs(p, size);
6195         }
6196       }
6197     }
6198 
6199     // Case 2
6200     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6201     test_log("size            align           req_addr         result");
6202 
6203     for (int i = 0; i < num_sizes; i++) {
6204       const size_t size = sizes[i];
6205       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6206         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6207         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6208         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6209                  size, alignment, p2i(req_addr), p2i(p),
6210                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6211         if (p != NULL) {
6212           assert(p == req_addr, "must be");
6213           small_page_write(p, size);
6214           os::Linux::release_memory_special_huge_tlbfs(p, size);
6215         }
6216       }
6217     }
6218 
6219     // Case 3
6220     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6221     test_log("size            align           req_addr         result");
6222 
6223     for (int i = 0; i < num_sizes; i++) {
6224       const size_t size = sizes[i];
6225       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6226         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6227         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6228         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6229                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6230         // as the area around req_addr contains already existing mappings, the API should always
6231         // return NULL (as per contract, it cannot return another address)
6232         assert(p == NULL, "must be");
6233       }
6234     }
6235 
6236     ::munmap(mapping2, mapping_size);
6237 
6238   }
6239 
6240   static void test_reserve_memory_special_huge_tlbfs() {
6241     if (!UseHugeTLBFS) {
6242       return;
6243     }
6244 
6245     test_reserve_memory_special_huge_tlbfs_only();
6246     test_reserve_memory_special_huge_tlbfs_mixed();
6247   }
6248 
6249   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6250     if (!UseSHM) {
6251       return;
6252     }
6253 
6254     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6255 
6256     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6257 
6258     if (addr != NULL) {
6259       assert(is_ptr_aligned(addr, alignment), "Check");
6260       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6261 
6262       small_page_write(addr, size);
6263 
6264       os::Linux::release_memory_special_shm(addr, size);
6265     }
6266   }
6267 
6268   static void test_reserve_memory_special_shm() {
6269     size_t lp = os::large_page_size();
6270     size_t ag = os::vm_allocation_granularity();
6271 
6272     for (size_t size = ag; size < lp * 3; size += ag) {
6273       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6274         test_reserve_memory_special_shm(size, alignment);
6275       }
6276     }
6277   }
6278 
6279   static void test() {
6280     test_reserve_memory_special_huge_tlbfs();
6281     test_reserve_memory_special_shm();
6282   }
6283 };
6284 
6285 void TestReserveMemorySpecial_test() {
6286   TestReserveMemorySpecial::test();
6287 }
6288 
6289 #endif
--- EOF ---