1 /*
   2  * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "mutex_linux.inline.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "os_linux.inline.hpp"
  41 #include "os_share_linux.hpp"
  42 #include "prims/jniFastGetField.hpp"
  43 #include "prims/jvm.h"
  44 #include "prims/jvm_misc.hpp"
  45 #include "runtime/arguments.hpp"
  46 #include "runtime/atomic.inline.hpp"
  47 #include "runtime/extendedPC.hpp"
  48 #include "runtime/globals.hpp"
  49 #include "runtime/interfaceSupport.hpp"
  50 #include "runtime/init.hpp"
  51 #include "runtime/java.hpp"
  52 #include "runtime/javaCalls.hpp"
  53 #include "runtime/mutexLocker.hpp"
  54 #include "runtime/objectMonitor.hpp"
  55 #include "runtime/orderAccess.inline.hpp"
  56 #include "runtime/osThread.hpp"
  57 #include "runtime/perfMemory.hpp"
  58 #include "runtime/sharedRuntime.hpp"
  59 #include "runtime/statSampler.hpp"
  60 #include "runtime/stubRoutines.hpp"
  61 #include "runtime/thread.inline.hpp"
  62 #include "runtime/threadCritical.hpp"
  63 #include "runtime/timer.hpp"
  64 #include "semaphore_posix.hpp"
  65 #include "services/attachListener.hpp"
  66 #include "services/memTracker.hpp"
  67 #include "services/runtimeService.hpp"
  68 #include "utilities/decoder.hpp"
  69 #include "utilities/defaultStream.hpp"
  70 #include "utilities/events.hpp"
  71 #include "utilities/elfFile.hpp"
  72 #include "utilities/growableArray.hpp"
  73 #include "utilities/macros.hpp"
  74 #include "utilities/vmError.hpp"
  75 
  76 // put OS-includes here
  77 # include <sys/types.h>
  78 # include <sys/mman.h>
  79 # include <sys/stat.h>
  80 # include <sys/select.h>
  81 # include <pthread.h>
  82 # include <signal.h>
  83 # include <errno.h>
  84 # include <dlfcn.h>
  85 # include <stdio.h>
  86 # include <unistd.h>
  87 # include <sys/resource.h>
  88 # include <pthread.h>
  89 # include <sys/stat.h>
  90 # include <sys/time.h>
  91 # include <sys/times.h>
  92 # include <sys/utsname.h>
  93 # include <sys/socket.h>
  94 # include <sys/wait.h>
  95 # include <pwd.h>
  96 # include <poll.h>
  97 # include <semaphore.h>
  98 # include <fcntl.h>
  99 # include <string.h>
 100 # include <syscall.h>
 101 # include <sys/sysinfo.h>
 102 # include <gnu/libc-version.h>
 103 # include <sys/ipc.h>
 104 # include <sys/shm.h>
 105 # include <link.h>
 106 # include <stdint.h>
 107 # include <inttypes.h>
 108 # include <sys/ioctl.h>
 109 
 110 #ifndef _GNU_SOURCE
 111   #define _GNU_SOURCE
 112   #include <sched.h>
 113   #undef _GNU_SOURCE
 114 #else
 115   #include <sched.h>
 116 #endif
 117 
 118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 119 // getrusage() is prepared to handle the associated failure.
 120 #ifndef RUSAGE_THREAD
 121   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 122 #endif
 123 
 124 #define MAX_PATH    (2 * K)
 125 
 126 #define MAX_SECS 100000000
 127 
 128 // for timer info max values which include all bits
 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 130 
 131 #define LARGEPAGES_BIT (1 << 6)
 132 ////////////////////////////////////////////////////////////////////////////////
 133 // global variables
 134 julong os::Linux::_physical_memory = 0;
 135 
 136 address   os::Linux::_initial_thread_stack_bottom = NULL;
 137 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 138 
 139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 142 Mutex* os::Linux::_createThread_lock = NULL;
 143 pthread_t os::Linux::_main_thread;
 144 int os::Linux::_page_size = -1;
 145 const int os::Linux::_vm_default_page_size = (8 * K);
 146 bool os::Linux::_supports_fast_thread_cpu_time = false;
 147 const char * os::Linux::_glibc_version = NULL;
 148 const char * os::Linux::_libpthread_version = NULL;
 149 pthread_condattr_t os::Linux::_condattr[1];
 150 
 151 static jlong initial_time_count=0;
 152 
 153 static int clock_tics_per_sec = 100;
 154 
 155 // For diagnostics to print a message once. see run_periodic_checks
 156 static sigset_t check_signal_done;
 157 static bool check_signals = true;
 158 
 159 // Signal number used to suspend/resume a thread
 160 
 161 // do not use any signal number less than SIGSEGV, see 4355769
 162 static int SR_signum = SIGUSR2;
 163 sigset_t SR_sigset;
 164 
 165 // Declarations
 166 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 167 
 168 // utility functions
 169 
 170 static int SR_initialize();
 171 
 172 julong os::available_memory() {
 173   return Linux::available_memory();
 174 }
 175 
 176 julong os::Linux::available_memory() {
 177   // values in struct sysinfo are "unsigned long"
 178   struct sysinfo si;
 179   sysinfo(&si);
 180 
 181   return (julong)si.freeram * si.mem_unit;
 182 }
 183 
 184 julong os::physical_memory() {
 185   return Linux::physical_memory();
 186 }
 187 
 188 // Return true if user is running as root.
 189 
 190 bool os::have_special_privileges() {
 191   static bool init = false;
 192   static bool privileges = false;
 193   if (!init) {
 194     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 195     init = true;
 196   }
 197   return privileges;
 198 }
 199 
 200 
 201 #ifndef SYS_gettid
 202 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 203   #ifdef __ia64__
 204     #define SYS_gettid 1105
 205   #else
 206     #ifdef __i386__
 207       #define SYS_gettid 224
 208     #else
 209       #ifdef __amd64__
 210         #define SYS_gettid 186
 211       #else
 212         #ifdef __sparc__
 213           #define SYS_gettid 143
 214         #else
 215           #error define gettid for the arch
 216         #endif
 217       #endif
 218     #endif
 219   #endif
 220 #endif
 221 
 222 // Cpu architecture string
 223 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 224 
 225 
 226 // pid_t gettid()
 227 //
 228 // Returns the kernel thread id of the currently running thread. Kernel
 229 // thread id is used to access /proc.
 230 pid_t os::Linux::gettid() {
 231   int rslt = syscall(SYS_gettid);
 232   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 233   return (pid_t)rslt;
 234 }
 235 
 236 // Most versions of linux have a bug where the number of processors are
 237 // determined by looking at the /proc file system.  In a chroot environment,
 238 // the system call returns 1.  This causes the VM to act as if it is
 239 // a single processor and elide locking (see is_MP() call).
 240 static bool unsafe_chroot_detected = false;
 241 static const char *unstable_chroot_error = "/proc file system not found.\n"
 242                      "Java may be unstable running multithreaded in a chroot "
 243                      "environment on Linux when /proc filesystem is not mounted.";
 244 
 245 void os::Linux::initialize_system_info() {
 246   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 247   if (processor_count() == 1) {
 248     pid_t pid = os::Linux::gettid();
 249     char fname[32];
 250     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 251     FILE *fp = fopen(fname, "r");
 252     if (fp == NULL) {
 253       unsafe_chroot_detected = true;
 254     } else {
 255       fclose(fp);
 256     }
 257   }
 258   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 259   assert(processor_count() > 0, "linux error");
 260 }
 261 
 262 void os::init_system_properties_values() {
 263   // The next steps are taken in the product version:
 264   //
 265   // Obtain the JAVA_HOME value from the location of libjvm.so.
 266   // This library should be located at:
 267   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 268   //
 269   // If "/jre/lib/" appears at the right place in the path, then we
 270   // assume libjvm.so is installed in a JDK and we use this path.
 271   //
 272   // Otherwise exit with message: "Could not create the Java virtual machine."
 273   //
 274   // The following extra steps are taken in the debugging version:
 275   //
 276   // If "/jre/lib/" does NOT appear at the right place in the path
 277   // instead of exit check for $JAVA_HOME environment variable.
 278   //
 279   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 280   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 281   // it looks like libjvm.so is installed there
 282   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 283   //
 284   // Otherwise exit.
 285   //
 286   // Important note: if the location of libjvm.so changes this
 287   // code needs to be changed accordingly.
 288 
 289   // See ld(1):
 290   //      The linker uses the following search paths to locate required
 291   //      shared libraries:
 292   //        1: ...
 293   //        ...
 294   //        7: The default directories, normally /lib and /usr/lib.
 295 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 296   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 297 #else
 298   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 299 #endif
 300 
 301 // Base path of extensions installed on the system.
 302 #define SYS_EXT_DIR     "/usr/java/packages"
 303 #define EXTENSIONS_DIR  "/lib/ext"
 304 
 305   // Buffer that fits several sprintfs.
 306   // Note that the space for the colon and the trailing null are provided
 307   // by the nulls included by the sizeof operator.
 308   const size_t bufsize =
 309     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 310          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 311   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 312 
 313   // sysclasspath, java_home, dll_dir
 314   {
 315     char *pslash;
 316     os::jvm_path(buf, bufsize);
 317 
 318     // Found the full path to libjvm.so.
 319     // Now cut the path to <java_home>/jre if we can.
 320     pslash = strrchr(buf, '/');
 321     if (pslash != NULL) {
 322       *pslash = '\0';            // Get rid of /libjvm.so.
 323     }
 324     pslash = strrchr(buf, '/');
 325     if (pslash != NULL) {
 326       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 327     }
 328     Arguments::set_dll_dir(buf);
 329 
 330     if (pslash != NULL) {
 331       pslash = strrchr(buf, '/');
 332       if (pslash != NULL) {
 333         *pslash = '\0';          // Get rid of /<arch>.
 334         pslash = strrchr(buf, '/');
 335         if (pslash != NULL) {
 336           *pslash = '\0';        // Get rid of /lib.
 337         }
 338       }
 339     }
 340     Arguments::set_java_home(buf);
 341     set_boot_path('/', ':');
 342   }
 343 
 344   // Where to look for native libraries.
 345   //
 346   // Note: Due to a legacy implementation, most of the library path
 347   // is set in the launcher. This was to accomodate linking restrictions
 348   // on legacy Linux implementations (which are no longer supported).
 349   // Eventually, all the library path setting will be done here.
 350   //
 351   // However, to prevent the proliferation of improperly built native
 352   // libraries, the new path component /usr/java/packages is added here.
 353   // Eventually, all the library path setting will be done here.
 354   {
 355     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 356     // should always exist (until the legacy problem cited above is
 357     // addressed).
 358     const char *v = ::getenv("LD_LIBRARY_PATH");
 359     const char *v_colon = ":";
 360     if (v == NULL) { v = ""; v_colon = ""; }
 361     // That's +1 for the colon and +1 for the trailing '\0'.
 362     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 363                                                      strlen(v) + 1 +
 364                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 365                                                      mtInternal);
 366     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 367     Arguments::set_library_path(ld_library_path);
 368     FREE_C_HEAP_ARRAY(char, ld_library_path);
 369   }
 370 
 371   // Extensions directories.
 372   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 373   Arguments::set_ext_dirs(buf);
 374 
 375   FREE_C_HEAP_ARRAY(char, buf);
 376 
 377 #undef DEFAULT_LIBPATH
 378 #undef SYS_EXT_DIR
 379 #undef EXTENSIONS_DIR
 380 }
 381 
 382 ////////////////////////////////////////////////////////////////////////////////
 383 // breakpoint support
 384 
 385 void os::breakpoint() {
 386   BREAKPOINT;
 387 }
 388 
 389 extern "C" void breakpoint() {
 390   // use debugger to set breakpoint here
 391 }
 392 
 393 ////////////////////////////////////////////////////////////////////////////////
 394 // signal support
 395 
 396 debug_only(static bool signal_sets_initialized = false);
 397 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 398 
 399 bool os::Linux::is_sig_ignored(int sig) {
 400   struct sigaction oact;
 401   sigaction(sig, (struct sigaction*)NULL, &oact);
 402   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 403                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 404   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 405     return true;
 406   } else {
 407     return false;
 408   }
 409 }
 410 
 411 void os::Linux::signal_sets_init() {
 412   // Should also have an assertion stating we are still single-threaded.
 413   assert(!signal_sets_initialized, "Already initialized");
 414   // Fill in signals that are necessarily unblocked for all threads in
 415   // the VM. Currently, we unblock the following signals:
 416   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 417   //                         by -Xrs (=ReduceSignalUsage));
 418   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 419   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 420   // the dispositions or masks wrt these signals.
 421   // Programs embedding the VM that want to use the above signals for their
 422   // own purposes must, at this time, use the "-Xrs" option to prevent
 423   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 424   // (See bug 4345157, and other related bugs).
 425   // In reality, though, unblocking these signals is really a nop, since
 426   // these signals are not blocked by default.
 427   sigemptyset(&unblocked_sigs);
 428   sigemptyset(&allowdebug_blocked_sigs);
 429   sigaddset(&unblocked_sigs, SIGILL);
 430   sigaddset(&unblocked_sigs, SIGSEGV);
 431   sigaddset(&unblocked_sigs, SIGBUS);
 432   sigaddset(&unblocked_sigs, SIGFPE);
 433 #if defined(PPC64)
 434   sigaddset(&unblocked_sigs, SIGTRAP);
 435 #endif
 436   sigaddset(&unblocked_sigs, SR_signum);
 437 
 438   if (!ReduceSignalUsage) {
 439     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 440       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 441       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 442     }
 443     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 444       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 445       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 446     }
 447     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 448       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 449       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 450     }
 451   }
 452   // Fill in signals that are blocked by all but the VM thread.
 453   sigemptyset(&vm_sigs);
 454   if (!ReduceSignalUsage) {
 455     sigaddset(&vm_sigs, BREAK_SIGNAL);
 456   }
 457   debug_only(signal_sets_initialized = true);
 458 
 459 }
 460 
 461 // These are signals that are unblocked while a thread is running Java.
 462 // (For some reason, they get blocked by default.)
 463 sigset_t* os::Linux::unblocked_signals() {
 464   assert(signal_sets_initialized, "Not initialized");
 465   return &unblocked_sigs;
 466 }
 467 
 468 // These are the signals that are blocked while a (non-VM) thread is
 469 // running Java. Only the VM thread handles these signals.
 470 sigset_t* os::Linux::vm_signals() {
 471   assert(signal_sets_initialized, "Not initialized");
 472   return &vm_sigs;
 473 }
 474 
 475 // These are signals that are blocked during cond_wait to allow debugger in
 476 sigset_t* os::Linux::allowdebug_blocked_signals() {
 477   assert(signal_sets_initialized, "Not initialized");
 478   return &allowdebug_blocked_sigs;
 479 }
 480 
 481 void os::Linux::hotspot_sigmask(Thread* thread) {
 482 
 483   //Save caller's signal mask before setting VM signal mask
 484   sigset_t caller_sigmask;
 485   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 486 
 487   OSThread* osthread = thread->osthread();
 488   osthread->set_caller_sigmask(caller_sigmask);
 489 
 490   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 491 
 492   if (!ReduceSignalUsage) {
 493     if (thread->is_VM_thread()) {
 494       // Only the VM thread handles BREAK_SIGNAL ...
 495       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 496     } else {
 497       // ... all other threads block BREAK_SIGNAL
 498       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 499     }
 500   }
 501 }
 502 
 503 //////////////////////////////////////////////////////////////////////////////
 504 // detecting pthread library
 505 
 506 void os::Linux::libpthread_init() {
 507   // Save glibc and pthread version strings.
 508 #if !defined(_CS_GNU_LIBC_VERSION) || \
 509     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 510   #error "glibc too old (< 2.3.2)"
 511 #endif
 512 
 513   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 514   assert(n > 0, "cannot retrieve glibc version");
 515   char *str = (char *)malloc(n, mtInternal);
 516   confstr(_CS_GNU_LIBC_VERSION, str, n);
 517   os::Linux::set_glibc_version(str);
 518 
 519   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 520   assert(n > 0, "cannot retrieve pthread version");
 521   str = (char *)malloc(n, mtInternal);
 522   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 523   os::Linux::set_libpthread_version(str);
 524 }
 525 
 526 /////////////////////////////////////////////////////////////////////////////
 527 // thread stack expansion
 528 
 529 // os::Linux::manually_expand_stack() takes care of expanding the thread
 530 // stack. Note that this is normally not needed: pthread stacks allocate
 531 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 532 // committed. Therefore it is not necessary to expand the stack manually.
 533 //
 534 // Manually expanding the stack was historically needed on LinuxThreads
 535 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 536 // it is kept to deal with very rare corner cases:
 537 //
 538 // For one, user may run the VM on an own implementation of threads
 539 // whose stacks are - like the old LinuxThreads - implemented using
 540 // mmap(MAP_GROWSDOWN).
 541 //
 542 // Also, this coding may be needed if the VM is running on the primordial
 543 // thread. Normally we avoid running on the primordial thread; however,
 544 // user may still invoke the VM on the primordial thread.
 545 //
 546 // The following historical comment describes the details about running
 547 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 548 
 549 
 550 // Force Linux kernel to expand current thread stack. If "bottom" is close
 551 // to the stack guard, caller should block all signals.
 552 //
 553 // MAP_GROWSDOWN:
 554 //   A special mmap() flag that is used to implement thread stacks. It tells
 555 //   kernel that the memory region should extend downwards when needed. This
 556 //   allows early versions of LinuxThreads to only mmap the first few pages
 557 //   when creating a new thread. Linux kernel will automatically expand thread
 558 //   stack as needed (on page faults).
 559 //
 560 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 561 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 562 //   region, it's hard to tell if the fault is due to a legitimate stack
 563 //   access or because of reading/writing non-exist memory (e.g. buffer
 564 //   overrun). As a rule, if the fault happens below current stack pointer,
 565 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 566 //   application (see Linux kernel fault.c).
 567 //
 568 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 569 //   stack overflow detection.
 570 //
 571 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 572 //   not use MAP_GROWSDOWN.
 573 //
 574 // To get around the problem and allow stack banging on Linux, we need to
 575 // manually expand thread stack after receiving the SIGSEGV.
 576 //
 577 // There are two ways to expand thread stack to address "bottom", we used
 578 // both of them in JVM before 1.5:
 579 //   1. adjust stack pointer first so that it is below "bottom", and then
 580 //      touch "bottom"
 581 //   2. mmap() the page in question
 582 //
 583 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 584 // if current sp is already near the lower end of page 101, and we need to
 585 // call mmap() to map page 100, it is possible that part of the mmap() frame
 586 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 587 // That will destroy the mmap() frame and cause VM to crash.
 588 //
 589 // The following code works by adjusting sp first, then accessing the "bottom"
 590 // page to force a page fault. Linux kernel will then automatically expand the
 591 // stack mapping.
 592 //
 593 // _expand_stack_to() assumes its frame size is less than page size, which
 594 // should always be true if the function is not inlined.
 595 
 596 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 597   #define NOINLINE
 598 #else
 599   #define NOINLINE __attribute__ ((noinline))
 600 #endif
 601 
 602 static void _expand_stack_to(address bottom) NOINLINE;
 603 
 604 static void _expand_stack_to(address bottom) {
 605   address sp;
 606   size_t size;
 607   volatile char *p;
 608 
 609   // Adjust bottom to point to the largest address within the same page, it
 610   // gives us a one-page buffer if alloca() allocates slightly more memory.
 611   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 612   bottom += os::Linux::page_size() - 1;
 613 
 614   // sp might be slightly above current stack pointer; if that's the case, we
 615   // will alloca() a little more space than necessary, which is OK. Don't use
 616   // os::current_stack_pointer(), as its result can be slightly below current
 617   // stack pointer, causing us to not alloca enough to reach "bottom".
 618   sp = (address)&sp;
 619 
 620   if (sp > bottom) {
 621     size = sp - bottom;
 622     p = (volatile char *)alloca(size);
 623     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 624     p[0] = '\0';
 625   }
 626 }
 627 
 628 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 629   assert(t!=NULL, "just checking");
 630   assert(t->osthread()->expanding_stack(), "expand should be set");
 631   assert(t->stack_base() != NULL, "stack_base was not initialized");
 632 
 633   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 634     sigset_t mask_all, old_sigset;
 635     sigfillset(&mask_all);
 636     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 637     _expand_stack_to(addr);
 638     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 639     return true;
 640   }
 641   return false;
 642 }
 643 
 644 //////////////////////////////////////////////////////////////////////////////
 645 // create new thread
 646 
 647 // Thread start routine for all newly created threads
 648 static void *java_start(Thread *thread) {
 649   // Try to randomize the cache line index of hot stack frames.
 650   // This helps when threads of the same stack traces evict each other's
 651   // cache lines. The threads can be either from the same JVM instance, or
 652   // from different JVM instances. The benefit is especially true for
 653   // processors with hyperthreading technology.
 654   static int counter = 0;
 655   int pid = os::current_process_id();
 656   alloca(((pid ^ counter++) & 7) * 128);
 657 
 658   thread->initialize_thread_current();
 659 
 660   OSThread* osthread = thread->osthread();
 661   Monitor* sync = osthread->startThread_lock();
 662 
 663   osthread->set_thread_id(os::current_thread_id());
 664 
 665   log_debug(os)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 666     os::current_thread_id(), (uintx) pthread_self());
 667 
 668   if (UseNUMA) {
 669     int lgrp_id = os::numa_get_group_id();
 670     if (lgrp_id != -1) {
 671       thread->set_lgrp_id(lgrp_id);
 672     }
 673   }
 674   // initialize signal mask for this thread
 675   os::Linux::hotspot_sigmask(thread);
 676 
 677   // initialize floating point control register
 678   os::Linux::init_thread_fpu_state();
 679 
 680   // handshaking with parent thread
 681   {
 682     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 683 
 684     // notify parent thread
 685     osthread->set_state(INITIALIZED);
 686     sync->notify_all();
 687 
 688     // wait until os::start_thread()
 689     while (osthread->get_state() == INITIALIZED) {
 690       sync->wait(Mutex::_no_safepoint_check_flag);
 691     }
 692   }
 693 
 694   // call one more level start routine
 695   thread->run();
 696 
 697   log_debug(os)("Thread finished (tid " UINTX_FORMAT ", pthread id " UINTX_FORMAT ").",
 698     os::current_thread_id(), (uintx) pthread_self());
 699 
 700   return 0;
 701 }
 702 
 703 bool os::create_thread(Thread* thread, ThreadType thr_type,
 704                        size_t 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   // stack size
 727   // calculate stack size if it's not specified by caller
 728   if (stack_size == 0) {
 729     stack_size = os::Linux::default_stack_size(thr_type);
 730 
 731     switch (thr_type) {
 732     case os::java_thread:
 733       // Java threads use ThreadStackSize which default value can be
 734       // changed with the flag -Xss
 735       assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 736       stack_size = JavaThread::stack_size_at_create();
 737       break;
 738     case os::compiler_thread:
 739       if (CompilerThreadStackSize > 0) {
 740         stack_size = (size_t)(CompilerThreadStackSize * K);
 741         break;
 742       } // else fall through:
 743         // use VMThreadStackSize if CompilerThreadStackSize is not defined
 744     case os::vm_thread:
 745     case os::pgc_thread:
 746     case os::cgc_thread:
 747     case os::watcher_thread:
 748       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 749       break;
 750     }
 751   }
 752 
 753   stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 754   pthread_attr_setstacksize(&attr, stack_size);
 755 
 756   // glibc guard page
 757   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 758 
 759   ThreadState state;
 760 
 761   {
 762     pthread_t tid;
 763     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 764 
 765     LogHandle(os) log;
 766     if (log.is_debug()) {
 767       char buf[64];
 768       if (ret == 0) {
 769         log.debug("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 770           (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 771       } else {
 772         log.warning("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 773           strerror(errno), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 774       }
 775     }
 776 
 777     pthread_attr_destroy(&attr);
 778 
 779     if (ret != 0) {
 780       // Need to clean up stuff we've allocated so far
 781       thread->set_osthread(NULL);
 782       delete osthread;
 783       return false;
 784     }
 785 
 786     // Store pthread info into the OSThread
 787     osthread->set_pthread_id(tid);
 788 
 789     // Wait until child thread is either initialized or aborted
 790     {
 791       Monitor* sync_with_child = osthread->startThread_lock();
 792       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 793       while ((state = osthread->get_state()) == ALLOCATED) {
 794         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 795       }
 796     }
 797   }
 798 
 799   // Aborted due to thread limit being reached
 800   if (state == ZOMBIE) {
 801     thread->set_osthread(NULL);
 802     delete osthread;
 803     return false;
 804   }
 805 
 806   // The thread is returned suspended (in state INITIALIZED),
 807   // and is started higher up in the call chain
 808   assert(state == INITIALIZED, "race condition");
 809   return true;
 810 }
 811 
 812 /////////////////////////////////////////////////////////////////////////////
 813 // attach existing thread
 814 
 815 // bootstrap the main thread
 816 bool os::create_main_thread(JavaThread* thread) {
 817   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 818   return create_attached_thread(thread);
 819 }
 820 
 821 bool os::create_attached_thread(JavaThread* thread) {
 822 #ifdef ASSERT
 823   thread->verify_not_published();
 824 #endif
 825 
 826   // Allocate the OSThread object
 827   OSThread* osthread = new OSThread(NULL, NULL);
 828 
 829   if (osthread == NULL) {
 830     return false;
 831   }
 832 
 833   // Store pthread info into the OSThread
 834   osthread->set_thread_id(os::Linux::gettid());
 835   osthread->set_pthread_id(::pthread_self());
 836 
 837   // initialize floating point control register
 838   os::Linux::init_thread_fpu_state();
 839 
 840   // Initial thread state is RUNNABLE
 841   osthread->set_state(RUNNABLE);
 842 
 843   thread->set_osthread(osthread);
 844 
 845   if (UseNUMA) {
 846     int lgrp_id = os::numa_get_group_id();
 847     if (lgrp_id != -1) {
 848       thread->set_lgrp_id(lgrp_id);
 849     }
 850   }
 851 
 852   if (os::Linux::is_initial_thread()) {
 853     // If current thread is initial thread, its stack is mapped on demand,
 854     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 855     // the entire stack region to avoid SEGV in stack banging.
 856     // It is also useful to get around the heap-stack-gap problem on SuSE
 857     // kernel (see 4821821 for details). We first expand stack to the top
 858     // of yellow zone, then enable stack yellow zone (order is significant,
 859     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 860     // is no gap between the last two virtual memory regions.
 861 
 862     JavaThread *jt = (JavaThread *)thread;
 863     address addr = jt->stack_reserved_zone_base();
 864     assert(addr != NULL, "initialization problem?");
 865     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 866 
 867     osthread->set_expanding_stack();
 868     os::Linux::manually_expand_stack(jt, addr);
 869     osthread->clear_expanding_stack();
 870   }
 871 
 872   // initialize signal mask for this thread
 873   // and save the caller's signal mask
 874   os::Linux::hotspot_sigmask(thread);
 875 
 876   log_debug(os)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 877     os::current_thread_id(), (uintx) pthread_self());
 878 
 879   return true;
 880 }
 881 
 882 void os::pd_start_thread(Thread* thread) {
 883   OSThread * osthread = thread->osthread();
 884   assert(osthread->get_state() != INITIALIZED, "just checking");
 885   Monitor* sync_with_child = osthread->startThread_lock();
 886   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 887   sync_with_child->notify();
 888 }
 889 
 890 // Free Linux resources related to the OSThread
 891 void os::free_thread(OSThread* osthread) {
 892   assert(osthread != NULL, "osthread not set");
 893 
 894   if (Thread::current()->osthread() == osthread) {
 895     // Restore caller's signal mask
 896     sigset_t sigmask = osthread->caller_sigmask();
 897     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 898   }
 899 
 900   delete osthread;
 901 }
 902 
 903 //////////////////////////////////////////////////////////////////////////////
 904 // initial thread
 905 
 906 // Check if current thread is the initial thread, similar to Solaris thr_main.
 907 bool os::Linux::is_initial_thread(void) {
 908   char dummy;
 909   // If called before init complete, thread stack bottom will be null.
 910   // Can be called if fatal error occurs before initialization.
 911   if (initial_thread_stack_bottom() == NULL) return false;
 912   assert(initial_thread_stack_bottom() != NULL &&
 913          initial_thread_stack_size()   != 0,
 914          "os::init did not locate initial thread's stack region");
 915   if ((address)&dummy >= initial_thread_stack_bottom() &&
 916       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 917     return true;
 918   } else {
 919     return false;
 920   }
 921 }
 922 
 923 // Find the virtual memory area that contains addr
 924 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 925   FILE *fp = fopen("/proc/self/maps", "r");
 926   if (fp) {
 927     address low, high;
 928     while (!feof(fp)) {
 929       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 930         if (low <= addr && addr < high) {
 931           if (vma_low)  *vma_low  = low;
 932           if (vma_high) *vma_high = high;
 933           fclose(fp);
 934           return true;
 935         }
 936       }
 937       for (;;) {
 938         int ch = fgetc(fp);
 939         if (ch == EOF || ch == (int)'\n') break;
 940       }
 941     }
 942     fclose(fp);
 943   }
 944   return false;
 945 }
 946 
 947 // Locate initial thread stack. This special handling of initial thread stack
 948 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 949 // bogus value for initial thread.
 950 void os::Linux::capture_initial_stack(size_t max_size) {
 951   // stack size is the easy part, get it from RLIMIT_STACK
 952   size_t stack_size;
 953   struct rlimit rlim;
 954   getrlimit(RLIMIT_STACK, &rlim);
 955   stack_size = rlim.rlim_cur;
 956 
 957   // 6308388: a bug in ld.so will relocate its own .data section to the
 958   //   lower end of primordial stack; reduce ulimit -s value a little bit
 959   //   so we won't install guard page on ld.so's data section.
 960   stack_size -= 2 * page_size();
 961 
 962   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
 963   //   7.1, in both cases we will get 2G in return value.
 964   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
 965   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
 966   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
 967   //   in case other parts in glibc still assumes 2M max stack size.
 968   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
 969   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
 970   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
 971     stack_size = 2 * K * K IA64_ONLY(*2);
 972   }
 973   // Try to figure out where the stack base (top) is. This is harder.
 974   //
 975   // When an application is started, glibc saves the initial stack pointer in
 976   // a global variable "__libc_stack_end", which is then used by system
 977   // libraries. __libc_stack_end should be pretty close to stack top. The
 978   // variable is available since the very early days. However, because it is
 979   // a private interface, it could disappear in the future.
 980   //
 981   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 982   // to __libc_stack_end, it is very close to stack top, but isn't the real
 983   // stack top. Note that /proc may not exist if VM is running as a chroot
 984   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 985   // /proc/<pid>/stat could change in the future (though unlikely).
 986   //
 987   // We try __libc_stack_end first. If that doesn't work, look for
 988   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
 989   // as a hint, which should work well in most cases.
 990 
 991   uintptr_t stack_start;
 992 
 993   // try __libc_stack_end first
 994   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
 995   if (p && *p) {
 996     stack_start = *p;
 997   } else {
 998     // see if we can get the start_stack field from /proc/self/stat
 999     FILE *fp;
1000     int pid;
1001     char state;
1002     int ppid;
1003     int pgrp;
1004     int session;
1005     int nr;
1006     int tpgrp;
1007     unsigned long flags;
1008     unsigned long minflt;
1009     unsigned long cminflt;
1010     unsigned long majflt;
1011     unsigned long cmajflt;
1012     unsigned long utime;
1013     unsigned long stime;
1014     long cutime;
1015     long cstime;
1016     long prio;
1017     long nice;
1018     long junk;
1019     long it_real;
1020     uintptr_t start;
1021     uintptr_t vsize;
1022     intptr_t rss;
1023     uintptr_t rsslim;
1024     uintptr_t scodes;
1025     uintptr_t ecode;
1026     int i;
1027 
1028     // Figure what the primordial thread stack base is. Code is inspired
1029     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1030     // followed by command name surrounded by parentheses, state, etc.
1031     char stat[2048];
1032     int statlen;
1033 
1034     fp = fopen("/proc/self/stat", "r");
1035     if (fp) {
1036       statlen = fread(stat, 1, 2047, fp);
1037       stat[statlen] = '\0';
1038       fclose(fp);
1039 
1040       // Skip pid and the command string. Note that we could be dealing with
1041       // weird command names, e.g. user could decide to rename java launcher
1042       // to "java 1.4.2 :)", then the stat file would look like
1043       //                1234 (java 1.4.2 :)) R ... ...
1044       // We don't really need to know the command string, just find the last
1045       // occurrence of ")" and then start parsing from there. See bug 4726580.
1046       char * s = strrchr(stat, ')');
1047 
1048       i = 0;
1049       if (s) {
1050         // Skip blank chars
1051         do { s++; } while (s && isspace(*s));
1052 
1053 #define _UFM UINTX_FORMAT
1054 #define _DFM INTX_FORMAT
1055 
1056         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1057         //              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
1058         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,
1059                    &state,          // 3  %c
1060                    &ppid,           // 4  %d
1061                    &pgrp,           // 5  %d
1062                    &session,        // 6  %d
1063                    &nr,             // 7  %d
1064                    &tpgrp,          // 8  %d
1065                    &flags,          // 9  %lu
1066                    &minflt,         // 10 %lu
1067                    &cminflt,        // 11 %lu
1068                    &majflt,         // 12 %lu
1069                    &cmajflt,        // 13 %lu
1070                    &utime,          // 14 %lu
1071                    &stime,          // 15 %lu
1072                    &cutime,         // 16 %ld
1073                    &cstime,         // 17 %ld
1074                    &prio,           // 18 %ld
1075                    &nice,           // 19 %ld
1076                    &junk,           // 20 %ld
1077                    &it_real,        // 21 %ld
1078                    &start,          // 22 UINTX_FORMAT
1079                    &vsize,          // 23 UINTX_FORMAT
1080                    &rss,            // 24 INTX_FORMAT
1081                    &rsslim,         // 25 UINTX_FORMAT
1082                    &scodes,         // 26 UINTX_FORMAT
1083                    &ecode,          // 27 UINTX_FORMAT
1084                    &stack_start);   // 28 UINTX_FORMAT
1085       }
1086 
1087 #undef _UFM
1088 #undef _DFM
1089 
1090       if (i != 28 - 2) {
1091         assert(false, "Bad conversion from /proc/self/stat");
1092         // product mode - assume we are the initial thread, good luck in the
1093         // embedded case.
1094         warning("Can't detect initial thread stack location - bad conversion");
1095         stack_start = (uintptr_t) &rlim;
1096       }
1097     } else {
1098       // For some reason we can't open /proc/self/stat (for example, running on
1099       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1100       // most cases, so don't abort:
1101       warning("Can't detect initial thread stack location - no /proc/self/stat");
1102       stack_start = (uintptr_t) &rlim;
1103     }
1104   }
1105 
1106   // Now we have a pointer (stack_start) very close to the stack top, the
1107   // next thing to do is to figure out the exact location of stack top. We
1108   // can find out the virtual memory area that contains stack_start by
1109   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1110   // and its upper limit is the real stack top. (again, this would fail if
1111   // running inside chroot, because /proc may not exist.)
1112 
1113   uintptr_t stack_top;
1114   address low, high;
1115   if (find_vma((address)stack_start, &low, &high)) {
1116     // success, "high" is the true stack top. (ignore "low", because initial
1117     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1118     stack_top = (uintptr_t)high;
1119   } else {
1120     // failed, likely because /proc/self/maps does not exist
1121     warning("Can't detect initial thread stack location - find_vma failed");
1122     // best effort: stack_start is normally within a few pages below the real
1123     // stack top, use it as stack top, and reduce stack size so we won't put
1124     // guard page outside stack.
1125     stack_top = stack_start;
1126     stack_size -= 16 * page_size();
1127   }
1128 
1129   // stack_top could be partially down the page so align it
1130   stack_top = align_size_up(stack_top, page_size());
1131 
1132   if (max_size && stack_size > max_size) {
1133     _initial_thread_stack_size = max_size;
1134   } else {
1135     _initial_thread_stack_size = stack_size;
1136   }
1137 
1138   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1139   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1140 }
1141 
1142 ////////////////////////////////////////////////////////////////////////////////
1143 // time support
1144 
1145 // Time since start-up in seconds to a fine granularity.
1146 // Used by VMSelfDestructTimer and the MemProfiler.
1147 double os::elapsedTime() {
1148 
1149   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1150 }
1151 
1152 jlong os::elapsed_counter() {
1153   return javaTimeNanos() - initial_time_count;
1154 }
1155 
1156 jlong os::elapsed_frequency() {
1157   return NANOSECS_PER_SEC; // nanosecond resolution
1158 }
1159 
1160 bool os::supports_vtime() { return true; }
1161 bool os::enable_vtime()   { return false; }
1162 bool os::vtime_enabled()  { return false; }
1163 
1164 double os::elapsedVTime() {
1165   struct rusage usage;
1166   int retval = getrusage(RUSAGE_THREAD, &usage);
1167   if (retval == 0) {
1168     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);
1169   } else {
1170     // better than nothing, but not much
1171     return elapsedTime();
1172   }
1173 }
1174 
1175 jlong os::javaTimeMillis() {
1176   timeval time;
1177   int status = gettimeofday(&time, NULL);
1178   assert(status != -1, "linux error");
1179   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1180 }
1181 
1182 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1183   timeval time;
1184   int status = gettimeofday(&time, NULL);
1185   assert(status != -1, "linux error");
1186   seconds = jlong(time.tv_sec);
1187   nanos = jlong(time.tv_usec) * 1000;
1188 }
1189 
1190 
1191 #ifndef CLOCK_MONOTONIC
1192   #define CLOCK_MONOTONIC (1)
1193 #endif
1194 
1195 void os::Linux::clock_init() {
1196   // we do dlopen's in this particular order due to bug in linux
1197   // dynamical loader (see 6348968) leading to crash on exit
1198   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1199   if (handle == NULL) {
1200     handle = dlopen("librt.so", RTLD_LAZY);
1201   }
1202 
1203   if (handle) {
1204     int (*clock_getres_func)(clockid_t, struct timespec*) =
1205            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1206     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1207            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1208     if (clock_getres_func && clock_gettime_func) {
1209       // See if monotonic clock is supported by the kernel. Note that some
1210       // early implementations simply return kernel jiffies (updated every
1211       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1212       // for nano time (though the monotonic property is still nice to have).
1213       // It's fixed in newer kernels, however clock_getres() still returns
1214       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1215       // resolution for now. Hopefully as people move to new kernels, this
1216       // won't be a problem.
1217       struct timespec res;
1218       struct timespec tp;
1219       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1220           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1221         // yes, monotonic clock is supported
1222         _clock_gettime = clock_gettime_func;
1223         return;
1224       } else {
1225         // close librt if there is no monotonic clock
1226         dlclose(handle);
1227       }
1228     }
1229   }
1230   warning("No monotonic clock was available - timed services may " \
1231           "be adversely affected if the time-of-day clock changes");
1232 }
1233 
1234 #ifndef SYS_clock_getres
1235   #if defined(IA32) || defined(AMD64)
1236     #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1237     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1238   #else
1239     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1240     #define sys_clock_getres(x,y)  -1
1241   #endif
1242 #else
1243   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1244 #endif
1245 
1246 void os::Linux::fast_thread_clock_init() {
1247   if (!UseLinuxPosixThreadCPUClocks) {
1248     return;
1249   }
1250   clockid_t clockid;
1251   struct timespec tp;
1252   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1253       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1254 
1255   // Switch to using fast clocks for thread cpu time if
1256   // the sys_clock_getres() returns 0 error code.
1257   // Note, that some kernels may support the current thread
1258   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1259   // returned by the pthread_getcpuclockid().
1260   // If the fast Posix clocks are supported then the sys_clock_getres()
1261   // must return at least tp.tv_sec == 0 which means a resolution
1262   // better than 1 sec. This is extra check for reliability.
1263 
1264   if (pthread_getcpuclockid_func &&
1265       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1266       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1267     _supports_fast_thread_cpu_time = true;
1268     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1269   }
1270 }
1271 
1272 jlong os::javaTimeNanos() {
1273   if (os::supports_monotonic_clock()) {
1274     struct timespec tp;
1275     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1276     assert(status == 0, "gettime error");
1277     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1278     return result;
1279   } else {
1280     timeval time;
1281     int status = gettimeofday(&time, NULL);
1282     assert(status != -1, "linux error");
1283     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1284     return 1000 * usecs;
1285   }
1286 }
1287 
1288 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1289   if (os::supports_monotonic_clock()) {
1290     info_ptr->max_value = ALL_64_BITS;
1291 
1292     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1293     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1294     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1295   } else {
1296     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1297     info_ptr->max_value = ALL_64_BITS;
1298 
1299     // gettimeofday is a real time clock so it skips
1300     info_ptr->may_skip_backward = true;
1301     info_ptr->may_skip_forward = true;
1302   }
1303 
1304   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1305 }
1306 
1307 // Return the real, user, and system times in seconds from an
1308 // arbitrary fixed point in the past.
1309 bool os::getTimesSecs(double* process_real_time,
1310                       double* process_user_time,
1311                       double* process_system_time) {
1312   struct tms ticks;
1313   clock_t real_ticks = times(&ticks);
1314 
1315   if (real_ticks == (clock_t) (-1)) {
1316     return false;
1317   } else {
1318     double ticks_per_second = (double) clock_tics_per_sec;
1319     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1320     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1321     *process_real_time = ((double) real_ticks) / ticks_per_second;
1322 
1323     return true;
1324   }
1325 }
1326 
1327 
1328 char * os::local_time_string(char *buf, size_t buflen) {
1329   struct tm t;
1330   time_t long_time;
1331   time(&long_time);
1332   localtime_r(&long_time, &t);
1333   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1334                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1335                t.tm_hour, t.tm_min, t.tm_sec);
1336   return buf;
1337 }
1338 
1339 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1340   return localtime_r(clock, res);
1341 }
1342 
1343 ////////////////////////////////////////////////////////////////////////////////
1344 // runtime exit support
1345 
1346 // Note: os::shutdown() might be called very early during initialization, or
1347 // called from signal handler. Before adding something to os::shutdown(), make
1348 // sure it is async-safe and can handle partially initialized VM.
1349 void os::shutdown() {
1350 
1351   // allow PerfMemory to attempt cleanup of any persistent resources
1352   perfMemory_exit();
1353 
1354   // needs to remove object in file system
1355   AttachListener::abort();
1356 
1357   // flush buffered output, finish log files
1358   ostream_abort();
1359 
1360   // Check for abort hook
1361   abort_hook_t abort_hook = Arguments::abort_hook();
1362   if (abort_hook != NULL) {
1363     abort_hook();
1364   }
1365 
1366 }
1367 
1368 // Note: os::abort() might be called very early during initialization, or
1369 // called from signal handler. Before adding something to os::abort(), make
1370 // sure it is async-safe and can handle partially initialized VM.
1371 void os::abort(bool dump_core, void* siginfo, const void* context) {
1372   os::shutdown();
1373   if (dump_core) {
1374 #ifndef PRODUCT
1375     fdStream out(defaultStream::output_fd());
1376     out.print_raw("Current thread is ");
1377     char buf[16];
1378     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1379     out.print_raw_cr(buf);
1380     out.print_raw_cr("Dumping core ...");
1381 #endif
1382     ::abort(); // dump core
1383   }
1384 
1385   ::exit(1);
1386 }
1387 
1388 // Die immediately, no exit hook, no abort hook, no cleanup.
1389 void os::die() {
1390   ::abort();
1391 }
1392 
1393 
1394 // This method is a copy of JDK's sysGetLastErrorString
1395 // from src/solaris/hpi/src/system_md.c
1396 
1397 size_t os::lasterror(char *buf, size_t len) {
1398   if (errno == 0)  return 0;
1399 
1400   const char *s = ::strerror(errno);
1401   size_t n = ::strlen(s);
1402   if (n >= len) {
1403     n = len - 1;
1404   }
1405   ::strncpy(buf, s, n);
1406   buf[n] = '\0';
1407   return n;
1408 }
1409 
1410 // thread_id is kernel thread id (similar to Solaris LWP id)
1411 intx os::current_thread_id() { return os::Linux::gettid(); }
1412 int os::current_process_id() {
1413   return ::getpid();
1414 }
1415 
1416 // DLL functions
1417 
1418 const char* os::dll_file_extension() { return ".so"; }
1419 
1420 // This must be hard coded because it's the system's temporary
1421 // directory not the java application's temp directory, ala java.io.tmpdir.
1422 const char* os::get_temp_directory() { return "/tmp"; }
1423 
1424 static bool file_exists(const char* filename) {
1425   struct stat statbuf;
1426   if (filename == NULL || strlen(filename) == 0) {
1427     return false;
1428   }
1429   return os::stat(filename, &statbuf) == 0;
1430 }
1431 
1432 bool os::dll_build_name(char* buffer, size_t buflen,
1433                         const char* pname, const char* fname) {
1434   bool retval = false;
1435   // Copied from libhpi
1436   const size_t pnamelen = pname ? strlen(pname) : 0;
1437 
1438   // Return error on buffer overflow.
1439   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1440     return retval;
1441   }
1442 
1443   if (pnamelen == 0) {
1444     snprintf(buffer, buflen, "lib%s.so", fname);
1445     retval = true;
1446   } else if (strchr(pname, *os::path_separator()) != NULL) {
1447     int n;
1448     char** pelements = split_path(pname, &n);
1449     if (pelements == NULL) {
1450       return false;
1451     }
1452     for (int i = 0; i < n; i++) {
1453       // Really shouldn't be NULL, but check can't hurt
1454       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1455         continue; // skip the empty path values
1456       }
1457       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1458       if (file_exists(buffer)) {
1459         retval = true;
1460         break;
1461       }
1462     }
1463     // release the storage
1464     for (int i = 0; i < n; i++) {
1465       if (pelements[i] != NULL) {
1466         FREE_C_HEAP_ARRAY(char, pelements[i]);
1467       }
1468     }
1469     if (pelements != NULL) {
1470       FREE_C_HEAP_ARRAY(char*, pelements);
1471     }
1472   } else {
1473     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1474     retval = true;
1475   }
1476   return retval;
1477 }
1478 
1479 // check if addr is inside libjvm.so
1480 bool os::address_is_in_vm(address addr) {
1481   static address libjvm_base_addr;
1482   Dl_info dlinfo;
1483 
1484   if (libjvm_base_addr == NULL) {
1485     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1486       libjvm_base_addr = (address)dlinfo.dli_fbase;
1487     }
1488     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1489   }
1490 
1491   if (dladdr((void *)addr, &dlinfo) != 0) {
1492     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1493   }
1494 
1495   return false;
1496 }
1497 
1498 bool os::dll_address_to_function_name(address addr, char *buf,
1499                                       int buflen, int *offset,
1500                                       bool demangle) {
1501   // buf is not optional, but offset is optional
1502   assert(buf != NULL, "sanity check");
1503 
1504   Dl_info dlinfo;
1505 
1506   if (dladdr((void*)addr, &dlinfo) != 0) {
1507     // see if we have a matching symbol
1508     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1509       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1510         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1511       }
1512       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1513       return true;
1514     }
1515     // no matching symbol so try for just file info
1516     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1517       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1518                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1519         return true;
1520       }
1521     }
1522   }
1523 
1524   buf[0] = '\0';
1525   if (offset != NULL) *offset = -1;
1526   return false;
1527 }
1528 
1529 struct _address_to_library_name {
1530   address addr;          // input : memory address
1531   size_t  buflen;        //         size of fname
1532   char*   fname;         // output: library name
1533   address base;          //         library base addr
1534 };
1535 
1536 static int address_to_library_name_callback(struct dl_phdr_info *info,
1537                                             size_t size, void *data) {
1538   int i;
1539   bool found = false;
1540   address libbase = NULL;
1541   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1542 
1543   // iterate through all loadable segments
1544   for (i = 0; i < info->dlpi_phnum; i++) {
1545     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1546     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1547       // base address of a library is the lowest address of its loaded
1548       // segments.
1549       if (libbase == NULL || libbase > segbase) {
1550         libbase = segbase;
1551       }
1552       // see if 'addr' is within current segment
1553       if (segbase <= d->addr &&
1554           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1555         found = true;
1556       }
1557     }
1558   }
1559 
1560   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1561   // so dll_address_to_library_name() can fall through to use dladdr() which
1562   // can figure out executable name from argv[0].
1563   if (found && info->dlpi_name && info->dlpi_name[0]) {
1564     d->base = libbase;
1565     if (d->fname) {
1566       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1567     }
1568     return 1;
1569   }
1570   return 0;
1571 }
1572 
1573 bool os::dll_address_to_library_name(address addr, char* buf,
1574                                      int buflen, int* offset) {
1575   // buf is not optional, but offset is optional
1576   assert(buf != NULL, "sanity check");
1577 
1578   Dl_info dlinfo;
1579   struct _address_to_library_name data;
1580 
1581   // There is a bug in old glibc dladdr() implementation that it could resolve
1582   // to wrong library name if the .so file has a base address != NULL. Here
1583   // we iterate through the program headers of all loaded libraries to find
1584   // out which library 'addr' really belongs to. This workaround can be
1585   // removed once the minimum requirement for glibc is moved to 2.3.x.
1586   data.addr = addr;
1587   data.fname = buf;
1588   data.buflen = buflen;
1589   data.base = NULL;
1590   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1591 
1592   if (rslt) {
1593     // buf already contains library name
1594     if (offset) *offset = addr - data.base;
1595     return true;
1596   }
1597   if (dladdr((void*)addr, &dlinfo) != 0) {
1598     if (dlinfo.dli_fname != NULL) {
1599       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1600     }
1601     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1602       *offset = addr - (address)dlinfo.dli_fbase;
1603     }
1604     return true;
1605   }
1606 
1607   buf[0] = '\0';
1608   if (offset) *offset = -1;
1609   return false;
1610 }
1611 
1612 // Loads .dll/.so and
1613 // in case of error it checks if .dll/.so was built for the
1614 // same architecture as Hotspot is running on
1615 
1616 
1617 // Remember the stack's state. The Linux dynamic linker will change
1618 // the stack to 'executable' at most once, so we must safepoint only once.
1619 bool os::Linux::_stack_is_executable = false;
1620 
1621 // VM operation that loads a library.  This is necessary if stack protection
1622 // of the Java stacks can be lost during loading the library.  If we
1623 // do not stop the Java threads, they can stack overflow before the stacks
1624 // are protected again.
1625 class VM_LinuxDllLoad: public VM_Operation {
1626  private:
1627   const char *_filename;
1628   char *_ebuf;
1629   int _ebuflen;
1630   void *_lib;
1631  public:
1632   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1633     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1634   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1635   void doit() {
1636     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1637     os::Linux::_stack_is_executable = true;
1638   }
1639   void* loaded_library() { return _lib; }
1640 };
1641 
1642 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1643   void * result = NULL;
1644   bool load_attempted = false;
1645 
1646   // Check whether the library to load might change execution rights
1647   // of the stack. If they are changed, the protection of the stack
1648   // guard pages will be lost. We need a safepoint to fix this.
1649   //
1650   // See Linux man page execstack(8) for more info.
1651   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1652     ElfFile ef(filename);
1653     if (!ef.specifies_noexecstack()) {
1654       if (!is_init_completed()) {
1655         os::Linux::_stack_is_executable = true;
1656         // This is OK - No Java threads have been created yet, and hence no
1657         // stack guard pages to fix.
1658         //
1659         // This should happen only when you are building JDK7 using a very
1660         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1661         //
1662         // Dynamic loader will make all stacks executable after
1663         // this function returns, and will not do that again.
1664         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1665       } else {
1666         warning("You have loaded library %s which might have disabled stack guard. "
1667                 "The VM will try to fix the stack guard now.\n"
1668                 "It's highly recommended that you fix the library with "
1669                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1670                 filename);
1671 
1672         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1673         JavaThread *jt = JavaThread::current();
1674         if (jt->thread_state() != _thread_in_native) {
1675           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1676           // that requires ExecStack. Cannot enter safe point. Let's give up.
1677           warning("Unable to fix stack guard. Giving up.");
1678         } else {
1679           if (!LoadExecStackDllInVMThread) {
1680             // This is for the case where the DLL has an static
1681             // constructor function that executes JNI code. We cannot
1682             // load such DLLs in the VMThread.
1683             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1684           }
1685 
1686           ThreadInVMfromNative tiv(jt);
1687           debug_only(VMNativeEntryWrapper vew;)
1688 
1689           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1690           VMThread::execute(&op);
1691           if (LoadExecStackDllInVMThread) {
1692             result = op.loaded_library();
1693           }
1694           load_attempted = true;
1695         }
1696       }
1697     }
1698   }
1699 
1700   if (!load_attempted) {
1701     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1702   }
1703 
1704   if (result != NULL) {
1705     // Successful loading
1706     return result;
1707   }
1708 
1709   Elf32_Ehdr elf_head;
1710   int diag_msg_max_length=ebuflen-strlen(ebuf);
1711   char* diag_msg_buf=ebuf+strlen(ebuf);
1712 
1713   if (diag_msg_max_length==0) {
1714     // No more space in ebuf for additional diagnostics message
1715     return NULL;
1716   }
1717 
1718 
1719   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1720 
1721   if (file_descriptor < 0) {
1722     // Can't open library, report dlerror() message
1723     return NULL;
1724   }
1725 
1726   bool failed_to_read_elf_head=
1727     (sizeof(elf_head)!=
1728      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1729 
1730   ::close(file_descriptor);
1731   if (failed_to_read_elf_head) {
1732     // file i/o error - report dlerror() msg
1733     return NULL;
1734   }
1735 
1736   typedef struct {
1737     Elf32_Half  code;         // Actual value as defined in elf.h
1738     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1739     char        elf_class;    // 32 or 64 bit
1740     char        endianess;    // MSB or LSB
1741     char*       name;         // String representation
1742   } arch_t;
1743 
1744 #ifndef EM_486
1745   #define EM_486          6               /* Intel 80486 */
1746 #endif
1747 #ifndef EM_AARCH64
1748   #define EM_AARCH64    183               /* ARM AARCH64 */
1749 #endif
1750 
1751   static const arch_t arch_array[]={
1752     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1753     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1754     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1755     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1756     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1757     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1758     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1759     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1760 #if defined(VM_LITTLE_ENDIAN)
1761     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1762 #else
1763     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1764 #endif
1765     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1766     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1767     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1768     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1769     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1770     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1771     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1772     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1773   };
1774 
1775 #if  (defined IA32)
1776   static  Elf32_Half running_arch_code=EM_386;
1777 #elif   (defined AMD64)
1778   static  Elf32_Half running_arch_code=EM_X86_64;
1779 #elif  (defined IA64)
1780   static  Elf32_Half running_arch_code=EM_IA_64;
1781 #elif  (defined __sparc) && (defined _LP64)
1782   static  Elf32_Half running_arch_code=EM_SPARCV9;
1783 #elif  (defined __sparc) && (!defined _LP64)
1784   static  Elf32_Half running_arch_code=EM_SPARC;
1785 #elif  (defined __powerpc64__)
1786   static  Elf32_Half running_arch_code=EM_PPC64;
1787 #elif  (defined __powerpc__)
1788   static  Elf32_Half running_arch_code=EM_PPC;
1789 #elif  (defined ARM)
1790   static  Elf32_Half running_arch_code=EM_ARM;
1791 #elif  (defined S390)
1792   static  Elf32_Half running_arch_code=EM_S390;
1793 #elif  (defined ALPHA)
1794   static  Elf32_Half running_arch_code=EM_ALPHA;
1795 #elif  (defined MIPSEL)
1796   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1797 #elif  (defined PARISC)
1798   static  Elf32_Half running_arch_code=EM_PARISC;
1799 #elif  (defined MIPS)
1800   static  Elf32_Half running_arch_code=EM_MIPS;
1801 #elif  (defined M68K)
1802   static  Elf32_Half running_arch_code=EM_68K;
1803 #elif  (defined AARCH64)
1804   static  Elf32_Half running_arch_code=EM_AARCH64;
1805 #else
1806     #error Method os::dll_load requires that one of following is defined:\
1807          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1808 #endif
1809 
1810   // Identify compatability class for VM's architecture and library's architecture
1811   // Obtain string descriptions for architectures
1812 
1813   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1814   int running_arch_index=-1;
1815 
1816   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1817     if (running_arch_code == arch_array[i].code) {
1818       running_arch_index    = i;
1819     }
1820     if (lib_arch.code == arch_array[i].code) {
1821       lib_arch.compat_class = arch_array[i].compat_class;
1822       lib_arch.name         = arch_array[i].name;
1823     }
1824   }
1825 
1826   assert(running_arch_index != -1,
1827          "Didn't find running architecture code (running_arch_code) in arch_array");
1828   if (running_arch_index == -1) {
1829     // Even though running architecture detection failed
1830     // we may still continue with reporting dlerror() message
1831     return NULL;
1832   }
1833 
1834   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1835     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1836     return NULL;
1837   }
1838 
1839 #ifndef S390
1840   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1841     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1842     return NULL;
1843   }
1844 #endif // !S390
1845 
1846   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1847     if (lib_arch.name!=NULL) {
1848       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1849                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1850                  lib_arch.name, arch_array[running_arch_index].name);
1851     } else {
1852       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1853                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1854                  lib_arch.code,
1855                  arch_array[running_arch_index].name);
1856     }
1857   }
1858 
1859   return NULL;
1860 }
1861 
1862 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1863                                 int ebuflen) {
1864   void * result = ::dlopen(filename, RTLD_LAZY);
1865   if (result == NULL) {
1866     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1867     ebuf[ebuflen-1] = '\0';
1868   }
1869   return result;
1870 }
1871 
1872 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1873                                        int ebuflen) {
1874   void * result = NULL;
1875   if (LoadExecStackDllInVMThread) {
1876     result = dlopen_helper(filename, ebuf, ebuflen);
1877   }
1878 
1879   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1880   // library that requires an executable stack, or which does not have this
1881   // stack attribute set, dlopen changes the stack attribute to executable. The
1882   // read protection of the guard pages gets lost.
1883   //
1884   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1885   // may have been queued at the same time.
1886 
1887   if (!_stack_is_executable) {
1888     JavaThread *jt = Threads::first();
1889 
1890     while (jt) {
1891       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1892           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1893         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1894           warning("Attempt to reguard stack yellow zone failed.");
1895         }
1896       }
1897       jt = jt->next();
1898     }
1899   }
1900 
1901   return result;
1902 }
1903 
1904 void* os::dll_lookup(void* handle, const char* name) {
1905   void* res = dlsym(handle, name);
1906   return res;
1907 }
1908 
1909 void* os::get_default_process_handle() {
1910   return (void*)::dlopen(NULL, RTLD_LAZY);
1911 }
1912 
1913 static bool _print_ascii_file(const char* filename, outputStream* st) {
1914   int fd = ::open(filename, O_RDONLY);
1915   if (fd == -1) {
1916     return false;
1917   }
1918 
1919   char buf[32];
1920   int bytes;
1921   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1922     st->print_raw(buf, bytes);
1923   }
1924 
1925   ::close(fd);
1926 
1927   return true;
1928 }
1929 
1930 void os::print_dll_info(outputStream *st) {
1931   st->print_cr("Dynamic libraries:");
1932 
1933   char fname[32];
1934   pid_t pid = os::Linux::gettid();
1935 
1936   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1937 
1938   if (!_print_ascii_file(fname, st)) {
1939     st->print("Can not get library information for pid = %d\n", pid);
1940   }
1941 }
1942 
1943 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1944   FILE *procmapsFile = NULL;
1945 
1946   // Open the procfs maps file for the current process
1947   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1948     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1949     char line[PATH_MAX + 100];
1950 
1951     // Read line by line from 'file'
1952     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1953       u8 base, top, offset, inode;
1954       char permissions[5];
1955       char device[6];
1956       char name[PATH_MAX + 1];
1957 
1958       // Parse fields from line
1959       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1960              &base, &top, permissions, &offset, device, &inode, name);
1961 
1962       // Filter by device id '00:00' so that we only get file system mapped files.
1963       if (strcmp(device, "00:00") != 0) {
1964 
1965         // Call callback with the fields of interest
1966         if(callback(name, (address)base, (address)top, param)) {
1967           // Oops abort, callback aborted
1968           fclose(procmapsFile);
1969           return 1;
1970         }
1971       }
1972     }
1973     fclose(procmapsFile);
1974   }
1975   return 0;
1976 }
1977 
1978 void os::print_os_info_brief(outputStream* st) {
1979   os::Linux::print_distro_info(st);
1980 
1981   os::Posix::print_uname_info(st);
1982 
1983   os::Linux::print_libversion_info(st);
1984 
1985 }
1986 
1987 void os::print_os_info(outputStream* st) {
1988   st->print("OS:");
1989 
1990   os::Linux::print_distro_info(st);
1991 
1992   os::Posix::print_uname_info(st);
1993 
1994   // Print warning if unsafe chroot environment detected
1995   if (unsafe_chroot_detected) {
1996     st->print("WARNING!! ");
1997     st->print_cr("%s", unstable_chroot_error);
1998   }
1999 
2000   os::Linux::print_libversion_info(st);
2001 
2002   os::Posix::print_rlimit_info(st);
2003 
2004   os::Posix::print_load_average(st);
2005 
2006   os::Linux::print_full_memory_info(st);
2007 }
2008 
2009 // Try to identify popular distros.
2010 // Most Linux distributions have a /etc/XXX-release file, which contains
2011 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2012 // file that also contains the OS version string. Some have more than one
2013 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2014 // /etc/redhat-release.), so the order is important.
2015 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2016 // their own specific XXX-release file as well as a redhat-release file.
2017 // Because of this the XXX-release file needs to be searched for before the
2018 // redhat-release file.
2019 // Since Red Hat has a lsb-release file that is not very descriptive the
2020 // search for redhat-release needs to be before lsb-release.
2021 // Since the lsb-release file is the new standard it needs to be searched
2022 // before the older style release files.
2023 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2024 // next to last resort.  The os-release file is a new standard that contains
2025 // distribution information and the system-release file seems to be an old
2026 // standard that has been replaced by the lsb-release and os-release files.
2027 // Searching for the debian_version file is the last resort.  It contains
2028 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2029 // "Debian " is printed before the contents of the debian_version file.
2030 
2031 const char* distro_files[] = {
2032   "/etc/oracle-release",
2033   "/etc/mandriva-release",
2034   "/etc/mandrake-release",
2035   "/etc/sun-release",
2036   "/etc/redhat-release",
2037   "/etc/lsb-release",
2038   "/etc/SuSE-release",
2039   "/etc/turbolinux-release",
2040   "/etc/gentoo-release",
2041   "/etc/ltib-release",
2042   "/etc/angstrom-version",
2043   "/etc/system-release",
2044   "/etc/os-release",
2045   NULL };
2046 
2047 void os::Linux::print_distro_info(outputStream* st) {
2048   for (int i = 0;; i++) {
2049     const char* file = distro_files[i];
2050     if (file == NULL) {
2051       break;  // done
2052     }
2053     // If file prints, we found it.
2054     if (_print_ascii_file(file, st)) {
2055       return;
2056     }
2057   }
2058 
2059   if (file_exists("/etc/debian_version")) {
2060     st->print("Debian ");
2061     _print_ascii_file("/etc/debian_version", st);
2062   } else {
2063     st->print("Linux");
2064   }
2065   st->cr();
2066 }
2067 
2068 static void parse_os_info(char* distro, size_t length, const char* file) {
2069   FILE* fp = fopen(file, "r");
2070   if (fp != NULL) {
2071     char buf[256];
2072     // get last line of the file.
2073     while (fgets(buf, sizeof(buf), fp)) { }
2074     // Edit out extra stuff in expected ubuntu format
2075     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2076       char* ptr = strstr(buf, "\"");  // the name is in quotes
2077       if (ptr != NULL) {
2078         ptr++; // go beyond first quote
2079         char* nl = strchr(ptr, '\"');
2080         if (nl != NULL) *nl = '\0';
2081         strncpy(distro, ptr, length);
2082       } else {
2083         ptr = strstr(buf, "=");
2084         ptr++; // go beyond equals then
2085         char* nl = strchr(ptr, '\n');
2086         if (nl != NULL) *nl = '\0';
2087         strncpy(distro, ptr, length);
2088       }
2089     } else {
2090       // if not in expected Ubuntu format, print out whole line minus \n
2091       char* nl = strchr(buf, '\n');
2092       if (nl != NULL) *nl = '\0';
2093       strncpy(distro, buf, length);
2094     }
2095     // close distro file
2096     fclose(fp);
2097   }
2098 }
2099 
2100 void os::get_summary_os_info(char* buf, size_t buflen) {
2101   for (int i = 0;; i++) {
2102     const char* file = distro_files[i];
2103     if (file == NULL) {
2104       break; // ran out of distro_files
2105     }
2106     if (file_exists(file)) {
2107       parse_os_info(buf, buflen, file);
2108       return;
2109     }
2110   }
2111   // special case for debian
2112   if (file_exists("/etc/debian_version")) {
2113     strncpy(buf, "Debian ", buflen);
2114     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2115   } else {
2116     strncpy(buf, "Linux", buflen);
2117   }
2118 }
2119 
2120 void os::Linux::print_libversion_info(outputStream* st) {
2121   // libc, pthread
2122   st->print("libc:");
2123   st->print("%s ", os::Linux::glibc_version());
2124   st->print("%s ", os::Linux::libpthread_version());
2125   st->cr();
2126 }
2127 
2128 void os::Linux::print_full_memory_info(outputStream* st) {
2129   st->print("\n/proc/meminfo:\n");
2130   _print_ascii_file("/proc/meminfo", st);
2131   st->cr();
2132 }
2133 
2134 void os::print_memory_info(outputStream* st) {
2135 
2136   st->print("Memory:");
2137   st->print(" %dk page", os::vm_page_size()>>10);
2138 
2139   // values in struct sysinfo are "unsigned long"
2140   struct sysinfo si;
2141   sysinfo(&si);
2142 
2143   st->print(", physical " UINT64_FORMAT "k",
2144             os::physical_memory() >> 10);
2145   st->print("(" UINT64_FORMAT "k free)",
2146             os::available_memory() >> 10);
2147   st->print(", swap " UINT64_FORMAT "k",
2148             ((jlong)si.totalswap * si.mem_unit) >> 10);
2149   st->print("(" UINT64_FORMAT "k free)",
2150             ((jlong)si.freeswap * si.mem_unit) >> 10);
2151   st->cr();
2152 }
2153 
2154 // Print the first "model name" line and the first "flags" line
2155 // that we find and nothing more. We assume "model name" comes
2156 // before "flags" so if we find a second "model name", then the
2157 // "flags" field is considered missing.
2158 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2159 #if defined(IA32) || defined(AMD64)
2160   // Other platforms have less repetitive cpuinfo files
2161   FILE *fp = fopen("/proc/cpuinfo", "r");
2162   if (fp) {
2163     while (!feof(fp)) {
2164       if (fgets(buf, buflen, fp)) {
2165         // Assume model name comes before flags
2166         bool model_name_printed = false;
2167         if (strstr(buf, "model name") != NULL) {
2168           if (!model_name_printed) {
2169             st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2170             st->print_raw(buf);
2171             model_name_printed = true;
2172           } else {
2173             // model name printed but not flags?  Odd, just return
2174             fclose(fp);
2175             return true;
2176           }
2177         }
2178         // print the flags line too
2179         if (strstr(buf, "flags") != NULL) {
2180           st->print_raw(buf);
2181           fclose(fp);
2182           return true;
2183         }
2184       }
2185     }
2186     fclose(fp);
2187   }
2188 #endif // x86 platforms
2189   return false;
2190 }
2191 
2192 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2193   // Only print the model name if the platform provides this as a summary
2194   if (!print_model_name_and_flags(st, buf, buflen)) {
2195     st->print("\n/proc/cpuinfo:\n");
2196     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2197       st->print_cr("  <Not Available>");
2198     }
2199   }
2200 }
2201 
2202 #if defined(AMD64) || defined(IA32) || defined(X32)
2203 const char* search_string = "model name";
2204 #elif defined(SPARC)
2205 const char* search_string = "cpu";
2206 #elif defined(PPC64)
2207 const char* search_string = "cpu";
2208 #else
2209 const char* search_string = "Processor";
2210 #endif
2211 
2212 // Parses the cpuinfo file for string representing the model name.
2213 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2214   FILE* fp = fopen("/proc/cpuinfo", "r");
2215   if (fp != NULL) {
2216     while (!feof(fp)) {
2217       char buf[256];
2218       if (fgets(buf, sizeof(buf), fp)) {
2219         char* start = strstr(buf, search_string);
2220         if (start != NULL) {
2221           char *ptr = start + strlen(search_string);
2222           char *end = buf + strlen(buf);
2223           while (ptr != end) {
2224              // skip whitespace and colon for the rest of the name.
2225              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2226                break;
2227              }
2228              ptr++;
2229           }
2230           if (ptr != end) {
2231             // reasonable string, get rid of newline and keep the rest
2232             char* nl = strchr(buf, '\n');
2233             if (nl != NULL) *nl = '\0';
2234             strncpy(cpuinfo, ptr, length);
2235             fclose(fp);
2236             return;
2237           }
2238         }
2239       }
2240     }
2241     fclose(fp);
2242   }
2243   // cpuinfo not found or parsing failed, just print generic string.  The entire
2244   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2245 #if defined(AMD64)
2246   strncpy(cpuinfo, "x86_64", length);
2247 #elif defined(IA32)
2248   strncpy(cpuinfo, "x86_32", length);
2249 #elif defined(IA64)
2250   strncpy(cpuinfo, "IA64", length);
2251 #elif defined(SPARC)
2252   strncpy(cpuinfo, "sparcv9", length);
2253 #elif defined(AARCH64)
2254   strncpy(cpuinfo, "AArch64", length);
2255 #elif defined(ARM)
2256   strncpy(cpuinfo, "ARM", length);
2257 #elif defined(PPC)
2258   strncpy(cpuinfo, "PPC64", length);
2259 #elif defined(ZERO_LIBARCH)
2260   strncpy(cpuinfo, ZERO_LIBARCH, length);
2261 #else
2262   strncpy(cpuinfo, "unknown", length);
2263 #endif
2264 }
2265 
2266 static void print_signal_handler(outputStream* st, int sig,
2267                                  char* buf, size_t buflen);
2268 
2269 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2270   st->print_cr("Signal Handlers:");
2271   print_signal_handler(st, SIGSEGV, buf, buflen);
2272   print_signal_handler(st, SIGBUS , buf, buflen);
2273   print_signal_handler(st, SIGFPE , buf, buflen);
2274   print_signal_handler(st, SIGPIPE, buf, buflen);
2275   print_signal_handler(st, SIGXFSZ, buf, buflen);
2276   print_signal_handler(st, SIGILL , buf, buflen);
2277   print_signal_handler(st, SR_signum, buf, buflen);
2278   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2279   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2280   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2281   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2282 #if defined(PPC64)
2283   print_signal_handler(st, SIGTRAP, buf, buflen);
2284 #endif
2285 }
2286 
2287 static char saved_jvm_path[MAXPATHLEN] = {0};
2288 
2289 // Find the full path to the current module, libjvm.so
2290 void os::jvm_path(char *buf, jint buflen) {
2291   // Error checking.
2292   if (buflen < MAXPATHLEN) {
2293     assert(false, "must use a large-enough buffer");
2294     buf[0] = '\0';
2295     return;
2296   }
2297   // Lazy resolve the path to current module.
2298   if (saved_jvm_path[0] != 0) {
2299     strcpy(buf, saved_jvm_path);
2300     return;
2301   }
2302 
2303   char dli_fname[MAXPATHLEN];
2304   bool ret = dll_address_to_library_name(
2305                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2306                                          dli_fname, sizeof(dli_fname), NULL);
2307   assert(ret, "cannot locate libjvm");
2308   char *rp = NULL;
2309   if (ret && dli_fname[0] != '\0') {
2310     rp = realpath(dli_fname, buf);
2311   }
2312   if (rp == NULL) {
2313     return;
2314   }
2315 
2316   if (Arguments::sun_java_launcher_is_altjvm()) {
2317     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2318     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2319     // If "/jre/lib/" appears at the right place in the string, then
2320     // assume we are installed in a JDK and we're done. Otherwise, check
2321     // for a JAVA_HOME environment variable and fix up the path so it
2322     // looks like libjvm.so is installed there (append a fake suffix
2323     // hotspot/libjvm.so).
2324     const char *p = buf + strlen(buf) - 1;
2325     for (int count = 0; p > buf && count < 5; ++count) {
2326       for (--p; p > buf && *p != '/'; --p)
2327         /* empty */ ;
2328     }
2329 
2330     if (strncmp(p, "/jre/lib/", 9) != 0) {
2331       // Look for JAVA_HOME in the environment.
2332       char* java_home_var = ::getenv("JAVA_HOME");
2333       if (java_home_var != NULL && java_home_var[0] != 0) {
2334         char* jrelib_p;
2335         int len;
2336 
2337         // Check the current module name "libjvm.so".
2338         p = strrchr(buf, '/');
2339         if (p == NULL) {
2340           return;
2341         }
2342         assert(strstr(p, "/libjvm") == p, "invalid library name");
2343 
2344         rp = realpath(java_home_var, buf);
2345         if (rp == NULL) {
2346           return;
2347         }
2348 
2349         // determine if this is a legacy image or modules image
2350         // modules image doesn't have "jre" subdirectory
2351         len = strlen(buf);
2352         assert(len < buflen, "Ran out of buffer room");
2353         jrelib_p = buf + len;
2354         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2355         if (0 != access(buf, F_OK)) {
2356           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2357         }
2358 
2359         if (0 == access(buf, F_OK)) {
2360           // Use current module name "libjvm.so"
2361           len = strlen(buf);
2362           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2363         } else {
2364           // Go back to path of .so
2365           rp = realpath(dli_fname, buf);
2366           if (rp == NULL) {
2367             return;
2368           }
2369         }
2370       }
2371     }
2372   }
2373 
2374   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2375   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2376 }
2377 
2378 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2379   // no prefix required, not even "_"
2380 }
2381 
2382 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2383   // no suffix required
2384 }
2385 
2386 ////////////////////////////////////////////////////////////////////////////////
2387 // sun.misc.Signal support
2388 
2389 static volatile jint sigint_count = 0;
2390 
2391 static void UserHandler(int sig, void *siginfo, void *context) {
2392   // 4511530 - sem_post is serialized and handled by the manager thread. When
2393   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2394   // don't want to flood the manager thread with sem_post requests.
2395   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2396     return;
2397   }
2398 
2399   // Ctrl-C is pressed during error reporting, likely because the error
2400   // handler fails to abort. Let VM die immediately.
2401   if (sig == SIGINT && is_error_reported()) {
2402     os::die();
2403   }
2404 
2405   os::signal_notify(sig);
2406 }
2407 
2408 void* os::user_handler() {
2409   return CAST_FROM_FN_PTR(void*, UserHandler);
2410 }
2411 
2412 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2413   struct timespec ts;
2414   // Semaphore's are always associated with CLOCK_REALTIME
2415   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2416   // see unpackTime for discussion on overflow checking
2417   if (sec >= MAX_SECS) {
2418     ts.tv_sec += MAX_SECS;
2419     ts.tv_nsec = 0;
2420   } else {
2421     ts.tv_sec += sec;
2422     ts.tv_nsec += nsec;
2423     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2424       ts.tv_nsec -= NANOSECS_PER_SEC;
2425       ++ts.tv_sec; // note: this must be <= max_secs
2426     }
2427   }
2428 
2429   return ts;
2430 }
2431 
2432 extern "C" {
2433   typedef void (*sa_handler_t)(int);
2434   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2435 }
2436 
2437 void* os::signal(int signal_number, void* handler) {
2438   struct sigaction sigAct, oldSigAct;
2439 
2440   sigfillset(&(sigAct.sa_mask));
2441   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2442   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2443 
2444   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2445     // -1 means registration failed
2446     return (void *)-1;
2447   }
2448 
2449   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2450 }
2451 
2452 void os::signal_raise(int signal_number) {
2453   ::raise(signal_number);
2454 }
2455 
2456 // The following code is moved from os.cpp for making this
2457 // code platform specific, which it is by its very nature.
2458 
2459 // Will be modified when max signal is changed to be dynamic
2460 int os::sigexitnum_pd() {
2461   return NSIG;
2462 }
2463 
2464 // a counter for each possible signal value
2465 static volatile jint pending_signals[NSIG+1] = { 0 };
2466 
2467 // Linux(POSIX) specific hand shaking semaphore.
2468 static sem_t sig_sem;
2469 static PosixSemaphore sr_semaphore;
2470 
2471 void os::signal_init_pd() {
2472   // Initialize signal structures
2473   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2474 
2475   // Initialize signal semaphore
2476   ::sem_init(&sig_sem, 0, 0);
2477 }
2478 
2479 void os::signal_notify(int sig) {
2480   Atomic::inc(&pending_signals[sig]);
2481   ::sem_post(&sig_sem);
2482 }
2483 
2484 static int check_pending_signals(bool wait) {
2485   Atomic::store(0, &sigint_count);
2486   for (;;) {
2487     for (int i = 0; i < NSIG + 1; i++) {
2488       jint n = pending_signals[i];
2489       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2490         return i;
2491       }
2492     }
2493     if (!wait) {
2494       return -1;
2495     }
2496     JavaThread *thread = JavaThread::current();
2497     ThreadBlockInVM tbivm(thread);
2498 
2499     bool threadIsSuspended;
2500     do {
2501       thread->set_suspend_equivalent();
2502       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2503       ::sem_wait(&sig_sem);
2504 
2505       // were we externally suspended while we were waiting?
2506       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2507       if (threadIsSuspended) {
2508         // The semaphore has been incremented, but while we were waiting
2509         // another thread suspended us. We don't want to continue running
2510         // while suspended because that would surprise the thread that
2511         // suspended us.
2512         ::sem_post(&sig_sem);
2513 
2514         thread->java_suspend_self();
2515       }
2516     } while (threadIsSuspended);
2517   }
2518 }
2519 
2520 int os::signal_lookup() {
2521   return check_pending_signals(false);
2522 }
2523 
2524 int os::signal_wait() {
2525   return check_pending_signals(true);
2526 }
2527 
2528 ////////////////////////////////////////////////////////////////////////////////
2529 // Virtual Memory
2530 
2531 int os::vm_page_size() {
2532   // Seems redundant as all get out
2533   assert(os::Linux::page_size() != -1, "must call os::init");
2534   return os::Linux::page_size();
2535 }
2536 
2537 // Solaris allocates memory by pages.
2538 int os::vm_allocation_granularity() {
2539   assert(os::Linux::page_size() != -1, "must call os::init");
2540   return os::Linux::page_size();
2541 }
2542 
2543 // Rationale behind this function:
2544 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2545 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2546 //  samples for JITted code. Here we create private executable mapping over the code cache
2547 //  and then we can use standard (well, almost, as mapping can change) way to provide
2548 //  info for the reporting script by storing timestamp and location of symbol
2549 void linux_wrap_code(char* base, size_t size) {
2550   static volatile jint cnt = 0;
2551 
2552   if (!UseOprofile) {
2553     return;
2554   }
2555 
2556   char buf[PATH_MAX+1];
2557   int num = Atomic::add(1, &cnt);
2558 
2559   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2560            os::get_temp_directory(), os::current_process_id(), num);
2561   unlink(buf);
2562 
2563   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2564 
2565   if (fd != -1) {
2566     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2567     if (rv != (off_t)-1) {
2568       if (::write(fd, "", 1) == 1) {
2569         mmap(base, size,
2570              PROT_READ|PROT_WRITE|PROT_EXEC,
2571              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2572       }
2573     }
2574     ::close(fd);
2575     unlink(buf);
2576   }
2577 }
2578 
2579 static bool recoverable_mmap_error(int err) {
2580   // See if the error is one we can let the caller handle. This
2581   // list of errno values comes from JBS-6843484. I can't find a
2582   // Linux man page that documents this specific set of errno
2583   // values so while this list currently matches Solaris, it may
2584   // change as we gain experience with this failure mode.
2585   switch (err) {
2586   case EBADF:
2587   case EINVAL:
2588   case ENOTSUP:
2589     // let the caller deal with these errors
2590     return true;
2591 
2592   default:
2593     // Any remaining errors on this OS can cause our reserved mapping
2594     // to be lost. That can cause confusion where different data
2595     // structures think they have the same memory mapped. The worst
2596     // scenario is if both the VM and a library think they have the
2597     // same memory mapped.
2598     return false;
2599   }
2600 }
2601 
2602 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2603                                     int err) {
2604   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2605           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2606           strerror(err), err);
2607 }
2608 
2609 static void warn_fail_commit_memory(char* addr, size_t size,
2610                                     size_t alignment_hint, bool exec,
2611                                     int err) {
2612   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2613           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2614           alignment_hint, exec, strerror(err), err);
2615 }
2616 
2617 // NOTE: Linux kernel does not really reserve the pages for us.
2618 //       All it does is to check if there are enough free pages
2619 //       left at the time of mmap(). This could be a potential
2620 //       problem.
2621 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2622   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2623   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2624                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2625   if (res != (uintptr_t) MAP_FAILED) {
2626     if (UseNUMAInterleaving) {
2627       numa_make_global(addr, size);
2628     }
2629     return 0;
2630   }
2631 
2632   int err = errno;  // save errno from mmap() call above
2633 
2634   if (!recoverable_mmap_error(err)) {
2635     warn_fail_commit_memory(addr, size, exec, err);
2636     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2637   }
2638 
2639   return err;
2640 }
2641 
2642 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2643   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2644 }
2645 
2646 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2647                                   const char* mesg) {
2648   assert(mesg != NULL, "mesg must be specified");
2649   int err = os::Linux::commit_memory_impl(addr, size, exec);
2650   if (err != 0) {
2651     // the caller wants all commit errors to exit with the specified mesg:
2652     warn_fail_commit_memory(addr, size, exec, err);
2653     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2654   }
2655 }
2656 
2657 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2658 #ifndef MAP_HUGETLB
2659   #define MAP_HUGETLB 0x40000
2660 #endif
2661 
2662 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2663 #ifndef MADV_HUGEPAGE
2664   #define MADV_HUGEPAGE 14
2665 #endif
2666 
2667 int os::Linux::commit_memory_impl(char* addr, size_t size,
2668                                   size_t alignment_hint, bool exec) {
2669   int err = os::Linux::commit_memory_impl(addr, size, exec);
2670   if (err == 0) {
2671     realign_memory(addr, size, alignment_hint);
2672   }
2673   return err;
2674 }
2675 
2676 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2677                           bool exec) {
2678   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2679 }
2680 
2681 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2682                                   size_t alignment_hint, bool exec,
2683                                   const char* mesg) {
2684   assert(mesg != NULL, "mesg must be specified");
2685   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2686   if (err != 0) {
2687     // the caller wants all commit errors to exit with the specified mesg:
2688     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2689     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2690   }
2691 }
2692 
2693 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2694   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2695     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2696     // be supported or the memory may already be backed by huge pages.
2697     ::madvise(addr, bytes, MADV_HUGEPAGE);
2698   }
2699 }
2700 
2701 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2702   // This method works by doing an mmap over an existing mmaping and effectively discarding
2703   // the existing pages. However it won't work for SHM-based large pages that cannot be
2704   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2705   // small pages on top of the SHM segment. This method always works for small pages, so we
2706   // allow that in any case.
2707   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2708     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2709   }
2710 }
2711 
2712 void os::numa_make_global(char *addr, size_t bytes) {
2713   Linux::numa_interleave_memory(addr, bytes);
2714 }
2715 
2716 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2717 // bind policy to MPOL_PREFERRED for the current thread.
2718 #define USE_MPOL_PREFERRED 0
2719 
2720 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2721   // To make NUMA and large pages more robust when both enabled, we need to ease
2722   // the requirements on where the memory should be allocated. MPOL_BIND is the
2723   // default policy and it will force memory to be allocated on the specified
2724   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2725   // the specified node, but will not force it. Using this policy will prevent
2726   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2727   // free large pages.
2728   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2729   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2730 }
2731 
2732 bool os::numa_topology_changed() { return false; }
2733 
2734 size_t os::numa_get_groups_num() {
2735   int max_node = Linux::numa_max_node();
2736   return max_node > 0 ? max_node + 1 : 1;
2737 }
2738 
2739 int os::numa_get_group_id() {
2740   int cpu_id = Linux::sched_getcpu();
2741   if (cpu_id != -1) {
2742     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2743     if (lgrp_id != -1) {
2744       return lgrp_id;
2745     }
2746   }
2747   return 0;
2748 }
2749 
2750 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2751   for (size_t i = 0; i < size; i++) {
2752     ids[i] = i;
2753   }
2754   return size;
2755 }
2756 
2757 bool os::get_page_info(char *start, page_info* info) {
2758   return false;
2759 }
2760 
2761 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2762                      page_info* page_found) {
2763   return end;
2764 }
2765 
2766 
2767 int os::Linux::sched_getcpu_syscall(void) {
2768   unsigned int cpu = 0;
2769   int retval = -1;
2770 
2771 #if defined(IA32)
2772   #ifndef SYS_getcpu
2773     #define SYS_getcpu 318
2774   #endif
2775   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2776 #elif defined(AMD64)
2777 // Unfortunately we have to bring all these macros here from vsyscall.h
2778 // to be able to compile on old linuxes.
2779   #define __NR_vgetcpu 2
2780   #define VSYSCALL_START (-10UL << 20)
2781   #define VSYSCALL_SIZE 1024
2782   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2783   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2784   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2785   retval = vgetcpu(&cpu, NULL, NULL);
2786 #endif
2787 
2788   return (retval == -1) ? retval : cpu;
2789 }
2790 
2791 // Something to do with the numa-aware allocator needs these symbols
2792 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2793 extern "C" JNIEXPORT void numa_error(char *where) { }
2794 
2795 
2796 // If we are running with libnuma version > 2, then we should
2797 // be trying to use symbols with versions 1.1
2798 // If we are running with earlier version, which did not have symbol versions,
2799 // we should use the base version.
2800 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2801   void *f = dlvsym(handle, name, "libnuma_1.1");
2802   if (f == NULL) {
2803     f = dlsym(handle, name);
2804   }
2805   return f;
2806 }
2807 
2808 bool os::Linux::libnuma_init() {
2809   // sched_getcpu() should be in libc.
2810   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2811                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2812 
2813   // If it's not, try a direct syscall.
2814   if (sched_getcpu() == -1) {
2815     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2816                                     (void*)&sched_getcpu_syscall));
2817   }
2818 
2819   if (sched_getcpu() != -1) { // Does it work?
2820     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2821     if (handle != NULL) {
2822       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2823                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2824       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2825                                        libnuma_dlsym(handle, "numa_max_node")));
2826       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2827                                         libnuma_dlsym(handle, "numa_available")));
2828       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2829                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2830       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2831                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2832       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2833                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2834 
2835 
2836       if (numa_available() != -1) {
2837         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2838         // Create a cpu -> node mapping
2839         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2840         rebuild_cpu_to_node_map();
2841         return true;
2842       }
2843     }
2844   }
2845   return false;
2846 }
2847 
2848 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2849 // The table is later used in get_node_by_cpu().
2850 void os::Linux::rebuild_cpu_to_node_map() {
2851   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2852                               // in libnuma (possible values are starting from 16,
2853                               // and continuing up with every other power of 2, but less
2854                               // than the maximum number of CPUs supported by kernel), and
2855                               // is a subject to change (in libnuma version 2 the requirements
2856                               // are more reasonable) we'll just hardcode the number they use
2857                               // in the library.
2858   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2859 
2860   size_t cpu_num = os::active_processor_count();
2861   size_t cpu_map_size = NCPUS / BitsPerCLong;
2862   size_t cpu_map_valid_size =
2863     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2864 
2865   cpu_to_node()->clear();
2866   cpu_to_node()->at_grow(cpu_num - 1);
2867   size_t node_num = numa_get_groups_num();
2868 
2869   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2870   for (size_t i = 0; i < node_num; i++) {
2871     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2872       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2873         if (cpu_map[j] != 0) {
2874           for (size_t k = 0; k < BitsPerCLong; k++) {
2875             if (cpu_map[j] & (1UL << k)) {
2876               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2877             }
2878           }
2879         }
2880       }
2881     }
2882   }
2883   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2884 }
2885 
2886 int os::Linux::get_node_by_cpu(int cpu_id) {
2887   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2888     return cpu_to_node()->at(cpu_id);
2889   }
2890   return -1;
2891 }
2892 
2893 GrowableArray<int>* os::Linux::_cpu_to_node;
2894 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2895 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2896 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2897 os::Linux::numa_available_func_t os::Linux::_numa_available;
2898 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2899 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2900 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2901 unsigned long* os::Linux::_numa_all_nodes;
2902 
2903 bool os::pd_uncommit_memory(char* addr, size_t size) {
2904   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2905                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2906   return res  != (uintptr_t) MAP_FAILED;
2907 }
2908 
2909 static address get_stack_commited_bottom(address bottom, size_t size) {
2910   address nbot = bottom;
2911   address ntop = bottom + size;
2912 
2913   size_t page_sz = os::vm_page_size();
2914   unsigned pages = size / page_sz;
2915 
2916   unsigned char vec[1];
2917   unsigned imin = 1, imax = pages + 1, imid;
2918   int mincore_return_value = 0;
2919 
2920   assert(imin <= imax, "Unexpected page size");
2921 
2922   while (imin < imax) {
2923     imid = (imax + imin) / 2;
2924     nbot = ntop - (imid * page_sz);
2925 
2926     // Use a trick with mincore to check whether the page is mapped or not.
2927     // mincore sets vec to 1 if page resides in memory and to 0 if page
2928     // is swapped output but if page we are asking for is unmapped
2929     // it returns -1,ENOMEM
2930     mincore_return_value = mincore(nbot, page_sz, vec);
2931 
2932     if (mincore_return_value == -1) {
2933       // Page is not mapped go up
2934       // to find first mapped page
2935       if (errno != EAGAIN) {
2936         assert(errno == ENOMEM, "Unexpected mincore errno");
2937         imax = imid;
2938       }
2939     } else {
2940       // Page is mapped go down
2941       // to find first not mapped page
2942       imin = imid + 1;
2943     }
2944   }
2945 
2946   nbot = nbot + page_sz;
2947 
2948   // Adjust stack bottom one page up if last checked page is not mapped
2949   if (mincore_return_value == -1) {
2950     nbot = nbot + page_sz;
2951   }
2952 
2953   return nbot;
2954 }
2955 
2956 
2957 // Linux uses a growable mapping for the stack, and if the mapping for
2958 // the stack guard pages is not removed when we detach a thread the
2959 // stack cannot grow beyond the pages where the stack guard was
2960 // mapped.  If at some point later in the process the stack expands to
2961 // that point, the Linux kernel cannot expand the stack any further
2962 // because the guard pages are in the way, and a segfault occurs.
2963 //
2964 // However, it's essential not to split the stack region by unmapping
2965 // a region (leaving a hole) that's already part of the stack mapping,
2966 // so if the stack mapping has already grown beyond the guard pages at
2967 // the time we create them, we have to truncate the stack mapping.
2968 // So, we need to know the extent of the stack mapping when
2969 // create_stack_guard_pages() is called.
2970 
2971 // We only need this for stacks that are growable: at the time of
2972 // writing thread stacks don't use growable mappings (i.e. those
2973 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2974 // only applies to the main thread.
2975 
2976 // If the (growable) stack mapping already extends beyond the point
2977 // where we're going to put our guard pages, truncate the mapping at
2978 // that point by munmap()ping it.  This ensures that when we later
2979 // munmap() the guard pages we don't leave a hole in the stack
2980 // mapping. This only affects the main/initial thread
2981 
2982 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2983   if (os::Linux::is_initial_thread()) {
2984     // As we manually grow stack up to bottom inside create_attached_thread(),
2985     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2986     // we don't need to do anything special.
2987     // Check it first, before calling heavy function.
2988     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2989     unsigned char vec[1];
2990 
2991     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2992       // Fallback to slow path on all errors, including EAGAIN
2993       stack_extent = (uintptr_t) get_stack_commited_bottom(
2994                                                            os::Linux::initial_thread_stack_bottom(),
2995                                                            (size_t)addr - stack_extent);
2996     }
2997 
2998     if (stack_extent < (uintptr_t)addr) {
2999       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3000     }
3001   }
3002 
3003   return os::commit_memory(addr, size, !ExecMem);
3004 }
3005 
3006 // If this is a growable mapping, remove the guard pages entirely by
3007 // munmap()ping them.  If not, just call uncommit_memory(). This only
3008 // affects the main/initial thread, but guard against future OS changes
3009 // It's safe to always unmap guard pages for initial thread because we
3010 // always place it right after end of the mapped region
3011 
3012 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3013   uintptr_t stack_extent, stack_base;
3014 
3015   if (os::Linux::is_initial_thread()) {
3016     return ::munmap(addr, size) == 0;
3017   }
3018 
3019   return os::uncommit_memory(addr, size);
3020 }
3021 
3022 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3023 // at 'requested_addr'. If there are existing memory mappings at the same
3024 // location, however, they will be overwritten. If 'fixed' is false,
3025 // 'requested_addr' is only treated as a hint, the return value may or
3026 // may not start from the requested address. Unlike Linux mmap(), this
3027 // function returns NULL to indicate failure.
3028 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3029   char * addr;
3030   int flags;
3031 
3032   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3033   if (fixed) {
3034     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3035     flags |= MAP_FIXED;
3036   }
3037 
3038   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3039   // touch an uncommitted page. Otherwise, the read/write might
3040   // succeed if we have enough swap space to back the physical page.
3041   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3042                        flags, -1, 0);
3043 
3044   return addr == MAP_FAILED ? NULL : addr;
3045 }
3046 
3047 static int anon_munmap(char * addr, size_t size) {
3048   return ::munmap(addr, size) == 0;
3049 }
3050 
3051 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3052                             size_t alignment_hint) {
3053   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3054 }
3055 
3056 bool os::pd_release_memory(char* addr, size_t size) {
3057   return anon_munmap(addr, size);
3058 }
3059 
3060 static bool linux_mprotect(char* addr, size_t size, int prot) {
3061   // Linux wants the mprotect address argument to be page aligned.
3062   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3063 
3064   // According to SUSv3, mprotect() should only be used with mappings
3065   // established by mmap(), and mmap() always maps whole pages. Unaligned
3066   // 'addr' likely indicates problem in the VM (e.g. trying to change
3067   // protection of malloc'ed or statically allocated memory). Check the
3068   // caller if you hit this assert.
3069   assert(addr == bottom, "sanity check");
3070 
3071   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3072   return ::mprotect(bottom, size, prot) == 0;
3073 }
3074 
3075 // Set protections specified
3076 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3077                         bool is_committed) {
3078   unsigned int p = 0;
3079   switch (prot) {
3080   case MEM_PROT_NONE: p = PROT_NONE; break;
3081   case MEM_PROT_READ: p = PROT_READ; break;
3082   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3083   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3084   default:
3085     ShouldNotReachHere();
3086   }
3087   // is_committed is unused.
3088   return linux_mprotect(addr, bytes, p);
3089 }
3090 
3091 bool os::guard_memory(char* addr, size_t size) {
3092   return linux_mprotect(addr, size, PROT_NONE);
3093 }
3094 
3095 bool os::unguard_memory(char* addr, size_t size) {
3096   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3097 }
3098 
3099 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3100                                                     size_t page_size) {
3101   bool result = false;
3102   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3103                  MAP_ANONYMOUS|MAP_PRIVATE,
3104                  -1, 0);
3105   if (p != MAP_FAILED) {
3106     void *aligned_p = align_ptr_up(p, page_size);
3107 
3108     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3109 
3110     munmap(p, page_size * 2);
3111   }
3112 
3113   if (warn && !result) {
3114     warning("TransparentHugePages is not supported by the operating system.");
3115   }
3116 
3117   return result;
3118 }
3119 
3120 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3121   bool result = false;
3122   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3123                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3124                  -1, 0);
3125 
3126   if (p != MAP_FAILED) {
3127     // We don't know if this really is a huge page or not.
3128     FILE *fp = fopen("/proc/self/maps", "r");
3129     if (fp) {
3130       while (!feof(fp)) {
3131         char chars[257];
3132         long x = 0;
3133         if (fgets(chars, sizeof(chars), fp)) {
3134           if (sscanf(chars, "%lx-%*x", &x) == 1
3135               && x == (long)p) {
3136             if (strstr (chars, "hugepage")) {
3137               result = true;
3138               break;
3139             }
3140           }
3141         }
3142       }
3143       fclose(fp);
3144     }
3145     munmap(p, page_size);
3146   }
3147 
3148   if (warn && !result) {
3149     warning("HugeTLBFS is not supported by the operating system.");
3150   }
3151 
3152   return result;
3153 }
3154 
3155 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3156 //
3157 // From the coredump_filter documentation:
3158 //
3159 // - (bit 0) anonymous private memory
3160 // - (bit 1) anonymous shared memory
3161 // - (bit 2) file-backed private memory
3162 // - (bit 3) file-backed shared memory
3163 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3164 //           effective only if the bit 2 is cleared)
3165 // - (bit 5) hugetlb private memory
3166 // - (bit 6) hugetlb shared memory
3167 //
3168 static void set_coredump_filter(void) {
3169   FILE *f;
3170   long cdm;
3171 
3172   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3173     return;
3174   }
3175 
3176   if (fscanf(f, "%lx", &cdm) != 1) {
3177     fclose(f);
3178     return;
3179   }
3180 
3181   rewind(f);
3182 
3183   if ((cdm & LARGEPAGES_BIT) == 0) {
3184     cdm |= LARGEPAGES_BIT;
3185     fprintf(f, "%#lx", cdm);
3186   }
3187 
3188   fclose(f);
3189 }
3190 
3191 // Large page support
3192 
3193 static size_t _large_page_size = 0;
3194 
3195 size_t os::Linux::find_large_page_size() {
3196   size_t large_page_size = 0;
3197 
3198   // large_page_size on Linux is used to round up heap size. x86 uses either
3199   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3200   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3201   // page as large as 256M.
3202   //
3203   // Here we try to figure out page size by parsing /proc/meminfo and looking
3204   // for a line with the following format:
3205   //    Hugepagesize:     2048 kB
3206   //
3207   // If we can't determine the value (e.g. /proc is not mounted, or the text
3208   // format has been changed), we'll use the largest page size supported by
3209   // the processor.
3210 
3211 #ifndef ZERO
3212   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3213                      ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3214 #endif // ZERO
3215 
3216   FILE *fp = fopen("/proc/meminfo", "r");
3217   if (fp) {
3218     while (!feof(fp)) {
3219       int x = 0;
3220       char buf[16];
3221       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3222         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3223           large_page_size = x * K;
3224           break;
3225         }
3226       } else {
3227         // skip to next line
3228         for (;;) {
3229           int ch = fgetc(fp);
3230           if (ch == EOF || ch == (int)'\n') break;
3231         }
3232       }
3233     }
3234     fclose(fp);
3235   }
3236 
3237   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3238     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3239             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3240             proper_unit_for_byte_size(large_page_size));
3241   }
3242 
3243   return large_page_size;
3244 }
3245 
3246 size_t os::Linux::setup_large_page_size() {
3247   _large_page_size = Linux::find_large_page_size();
3248   const size_t default_page_size = (size_t)Linux::page_size();
3249   if (_large_page_size > default_page_size) {
3250     _page_sizes[0] = _large_page_size;
3251     _page_sizes[1] = default_page_size;
3252     _page_sizes[2] = 0;
3253   }
3254 
3255   return _large_page_size;
3256 }
3257 
3258 bool os::Linux::setup_large_page_type(size_t page_size) {
3259   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3260       FLAG_IS_DEFAULT(UseSHM) &&
3261       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3262 
3263     // The type of large pages has not been specified by the user.
3264 
3265     // Try UseHugeTLBFS and then UseSHM.
3266     UseHugeTLBFS = UseSHM = true;
3267 
3268     // Don't try UseTransparentHugePages since there are known
3269     // performance issues with it turned on. This might change in the future.
3270     UseTransparentHugePages = false;
3271   }
3272 
3273   if (UseTransparentHugePages) {
3274     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3275     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3276       UseHugeTLBFS = false;
3277       UseSHM = false;
3278       return true;
3279     }
3280     UseTransparentHugePages = false;
3281   }
3282 
3283   if (UseHugeTLBFS) {
3284     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3285     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3286       UseSHM = false;
3287       return true;
3288     }
3289     UseHugeTLBFS = false;
3290   }
3291 
3292   return UseSHM;
3293 }
3294 
3295 void os::large_page_init() {
3296   if (!UseLargePages &&
3297       !UseTransparentHugePages &&
3298       !UseHugeTLBFS &&
3299       !UseSHM) {
3300     // Not using large pages.
3301     return;
3302   }
3303 
3304   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3305     // The user explicitly turned off large pages.
3306     // Ignore the rest of the large pages flags.
3307     UseTransparentHugePages = false;
3308     UseHugeTLBFS = false;
3309     UseSHM = false;
3310     return;
3311   }
3312 
3313   size_t large_page_size = Linux::setup_large_page_size();
3314   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3315 
3316   set_coredump_filter();
3317 }
3318 
3319 #ifndef SHM_HUGETLB
3320   #define SHM_HUGETLB 04000
3321 #endif
3322 
3323 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3324                                             char* req_addr, bool exec) {
3325   // "exec" is passed in but not used.  Creating the shared image for
3326   // the code cache doesn't have an SHM_X executable permission to check.
3327   assert(UseLargePages && UseSHM, "only for SHM large pages");
3328   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3329 
3330   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3331     return NULL; // Fallback to small pages.
3332   }
3333 
3334   key_t key = IPC_PRIVATE;
3335   char *addr;
3336 
3337   bool warn_on_failure = UseLargePages &&
3338                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3339                          !FLAG_IS_DEFAULT(UseSHM) ||
3340                          !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3341   char msg[128];
3342 
3343   // Create a large shared memory region to attach to based on size.
3344   // Currently, size is the total size of the heap
3345   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3346   if (shmid == -1) {
3347     // Possible reasons for shmget failure:
3348     // 1. shmmax is too small for Java heap.
3349     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3350     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3351     // 2. not enough large page memory.
3352     //    > check available large pages: cat /proc/meminfo
3353     //    > increase amount of large pages:
3354     //          echo new_value > /proc/sys/vm/nr_hugepages
3355     //      Note 1: different Linux may use different name for this property,
3356     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3357     //      Note 2: it's possible there's enough physical memory available but
3358     //            they are so fragmented after a long run that they can't
3359     //            coalesce into large pages. Try to reserve large pages when
3360     //            the system is still "fresh".
3361     if (warn_on_failure) {
3362       jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3363       warning("%s", msg);
3364     }
3365     return NULL;
3366   }
3367 
3368   // attach to the region
3369   addr = (char*)shmat(shmid, req_addr, 0);
3370   int err = errno;
3371 
3372   // Remove shmid. If shmat() is successful, the actual shared memory segment
3373   // will be deleted when it's detached by shmdt() or when the process
3374   // terminates. If shmat() is not successful this will remove the shared
3375   // segment immediately.
3376   shmctl(shmid, IPC_RMID, NULL);
3377 
3378   if ((intptr_t)addr == -1) {
3379     if (warn_on_failure) {
3380       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3381       warning("%s", msg);
3382     }
3383     return NULL;
3384   }
3385 
3386   return addr;
3387 }
3388 
3389 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3390                                         int error) {
3391   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3392 
3393   bool warn_on_failure = UseLargePages &&
3394       (!FLAG_IS_DEFAULT(UseLargePages) ||
3395        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3396        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3397 
3398   if (warn_on_failure) {
3399     char msg[128];
3400     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3401                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3402     warning("%s", msg);
3403   }
3404 }
3405 
3406 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3407                                                         char* req_addr,
3408                                                         bool exec) {
3409   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3410   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3411   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3412 
3413   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3414   char* addr = (char*)::mmap(req_addr, bytes, prot,
3415                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3416                              -1, 0);
3417 
3418   if (addr == MAP_FAILED) {
3419     warn_on_large_pages_failure(req_addr, bytes, errno);
3420     return NULL;
3421   }
3422 
3423   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3424 
3425   return addr;
3426 }
3427 
3428 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3429 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3430 //   (req_addr != NULL) or with a given alignment.
3431 //  - bytes shall be a multiple of alignment.
3432 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3433 //  - alignment sets the alignment at which memory shall be allocated.
3434 //     It must be a multiple of allocation granularity.
3435 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3436 //  req_addr or NULL.
3437 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3438 
3439   size_t extra_size = bytes;
3440   if (req_addr == NULL && alignment > 0) {
3441     extra_size += alignment;
3442   }
3443 
3444   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3445     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3446     -1, 0);
3447   if (start == MAP_FAILED) {
3448     start = NULL;
3449   } else {
3450     if (req_addr != NULL) {
3451       if (start != req_addr) {
3452         ::munmap(start, extra_size);
3453         start = NULL;
3454       }
3455     } else {
3456       char* const start_aligned = (char*) align_ptr_up(start, alignment);
3457       char* const end_aligned = start_aligned + bytes;
3458       char* const end = start + extra_size;
3459       if (start_aligned > start) {
3460         ::munmap(start, start_aligned - start);
3461       }
3462       if (end_aligned < end) {
3463         ::munmap(end_aligned, end - end_aligned);
3464       }
3465       start = start_aligned;
3466     }
3467   }
3468   return start;
3469 
3470 }
3471 
3472 // Reserve memory using mmap(MAP_HUGETLB).
3473 //  - bytes shall be a multiple of alignment.
3474 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3475 //  - alignment sets the alignment at which memory shall be allocated.
3476 //     It must be a multiple of allocation granularity.
3477 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3478 //  req_addr or NULL.
3479 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3480                                                          size_t alignment,
3481                                                          char* req_addr,
3482                                                          bool exec) {
3483   size_t large_page_size = os::large_page_size();
3484   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3485 
3486   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3487   assert(is_size_aligned(bytes, alignment), "Must be");
3488 
3489   // First reserve - but not commit - the address range in small pages.
3490   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3491 
3492   if (start == NULL) {
3493     return NULL;
3494   }
3495 
3496   assert(is_ptr_aligned(start, alignment), "Must be");
3497 
3498   char* end = start + bytes;
3499 
3500   // Find the regions of the allocated chunk that can be promoted to large pages.
3501   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3502   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3503 
3504   size_t lp_bytes = lp_end - lp_start;
3505 
3506   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3507 
3508   if (lp_bytes == 0) {
3509     // The mapped region doesn't even span the start and the end of a large page.
3510     // Fall back to allocate a non-special area.
3511     ::munmap(start, end - start);
3512     return NULL;
3513   }
3514 
3515   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3516 
3517   void* result;
3518 
3519   // Commit small-paged leading area.
3520   if (start != lp_start) {
3521     result = ::mmap(start, lp_start - start, prot,
3522                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3523                     -1, 0);
3524     if (result == MAP_FAILED) {
3525       ::munmap(lp_start, end - lp_start);
3526       return NULL;
3527     }
3528   }
3529 
3530   // Commit large-paged area.
3531   result = ::mmap(lp_start, lp_bytes, prot,
3532                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3533                   -1, 0);
3534   if (result == MAP_FAILED) {
3535     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3536     // If the mmap above fails, the large pages region will be unmapped and we
3537     // have regions before and after with small pages. Release these regions.
3538     //
3539     // |  mapped  |  unmapped  |  mapped  |
3540     // ^          ^            ^          ^
3541     // start      lp_start     lp_end     end
3542     //
3543     ::munmap(start, lp_start - start);
3544     ::munmap(lp_end, end - lp_end);
3545     return NULL;
3546   }
3547 
3548   // Commit small-paged trailing area.
3549   if (lp_end != end) {
3550     result = ::mmap(lp_end, end - lp_end, prot,
3551                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3552                     -1, 0);
3553     if (result == MAP_FAILED) {
3554       ::munmap(start, lp_end - start);
3555       return NULL;
3556     }
3557   }
3558 
3559   return start;
3560 }
3561 
3562 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3563                                                    size_t alignment,
3564                                                    char* req_addr,
3565                                                    bool exec) {
3566   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3567   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3568   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3569   assert(is_power_of_2(os::large_page_size()), "Must be");
3570   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3571 
3572   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3573     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3574   } else {
3575     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3576   }
3577 }
3578 
3579 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3580                                  char* req_addr, bool exec) {
3581   assert(UseLargePages, "only for large pages");
3582 
3583   char* addr;
3584   if (UseSHM) {
3585     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3586   } else {
3587     assert(UseHugeTLBFS, "must be");
3588     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3589   }
3590 
3591   if (addr != NULL) {
3592     if (UseNUMAInterleaving) {
3593       numa_make_global(addr, bytes);
3594     }
3595 
3596     // The memory is committed
3597     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3598   }
3599 
3600   return addr;
3601 }
3602 
3603 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3604   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3605   return shmdt(base) == 0;
3606 }
3607 
3608 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3609   return pd_release_memory(base, bytes);
3610 }
3611 
3612 bool os::release_memory_special(char* base, size_t bytes) {
3613   bool res;
3614   if (MemTracker::tracking_level() > NMT_minimal) {
3615     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3616     res = os::Linux::release_memory_special_impl(base, bytes);
3617     if (res) {
3618       tkr.record((address)base, bytes);
3619     }
3620 
3621   } else {
3622     res = os::Linux::release_memory_special_impl(base, bytes);
3623   }
3624   return res;
3625 }
3626 
3627 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3628   assert(UseLargePages, "only for large pages");
3629   bool res;
3630 
3631   if (UseSHM) {
3632     res = os::Linux::release_memory_special_shm(base, bytes);
3633   } else {
3634     assert(UseHugeTLBFS, "must be");
3635     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3636   }
3637   return res;
3638 }
3639 
3640 size_t os::large_page_size() {
3641   return _large_page_size;
3642 }
3643 
3644 // With SysV SHM the entire memory region must be allocated as shared
3645 // memory.
3646 // HugeTLBFS allows application to commit large page memory on demand.
3647 // However, when committing memory with HugeTLBFS fails, the region
3648 // that was supposed to be committed will lose the old reservation
3649 // and allow other threads to steal that memory region. Because of this
3650 // behavior we can't commit HugeTLBFS memory.
3651 bool os::can_commit_large_page_memory() {
3652   return UseTransparentHugePages;
3653 }
3654 
3655 bool os::can_execute_large_page_memory() {
3656   return UseTransparentHugePages || UseHugeTLBFS;
3657 }
3658 
3659 // Reserve memory at an arbitrary address, only if that area is
3660 // available (and not reserved for something else).
3661 
3662 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3663   const int max_tries = 10;
3664   char* base[max_tries];
3665   size_t size[max_tries];
3666   const size_t gap = 0x000000;
3667 
3668   // Assert only that the size is a multiple of the page size, since
3669   // that's all that mmap requires, and since that's all we really know
3670   // about at this low abstraction level.  If we need higher alignment,
3671   // we can either pass an alignment to this method or verify alignment
3672   // in one of the methods further up the call chain.  See bug 5044738.
3673   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3674 
3675   // Repeatedly allocate blocks until the block is allocated at the
3676   // right spot.
3677 
3678   // Linux mmap allows caller to pass an address as hint; give it a try first,
3679   // if kernel honors the hint then we can return immediately.
3680   char * addr = anon_mmap(requested_addr, bytes, false);
3681   if (addr == requested_addr) {
3682     return requested_addr;
3683   }
3684 
3685   if (addr != NULL) {
3686     // mmap() is successful but it fails to reserve at the requested address
3687     anon_munmap(addr, bytes);
3688   }
3689 
3690   int i;
3691   for (i = 0; i < max_tries; ++i) {
3692     base[i] = reserve_memory(bytes);
3693 
3694     if (base[i] != NULL) {
3695       // Is this the block we wanted?
3696       if (base[i] == requested_addr) {
3697         size[i] = bytes;
3698         break;
3699       }
3700 
3701       // Does this overlap the block we wanted? Give back the overlapped
3702       // parts and try again.
3703 
3704       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3705       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3706         unmap_memory(base[i], top_overlap);
3707         base[i] += top_overlap;
3708         size[i] = bytes - top_overlap;
3709       } else {
3710         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3711         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3712           unmap_memory(requested_addr, bottom_overlap);
3713           size[i] = bytes - bottom_overlap;
3714         } else {
3715           size[i] = bytes;
3716         }
3717       }
3718     }
3719   }
3720 
3721   // Give back the unused reserved pieces.
3722 
3723   for (int j = 0; j < i; ++j) {
3724     if (base[j] != NULL) {
3725       unmap_memory(base[j], size[j]);
3726     }
3727   }
3728 
3729   if (i < max_tries) {
3730     return requested_addr;
3731   } else {
3732     return NULL;
3733   }
3734 }
3735 
3736 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3737   return ::read(fd, buf, nBytes);
3738 }
3739 
3740 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3741   return ::pread(fd, buf, nBytes, offset);
3742 }
3743 
3744 // Short sleep, direct OS call.
3745 //
3746 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3747 // sched_yield(2) will actually give up the CPU:
3748 //
3749 //   * Alone on this pariticular CPU, keeps running.
3750 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3751 //     (pre 2.6.39).
3752 //
3753 // So calling this with 0 is an alternative.
3754 //
3755 void os::naked_short_sleep(jlong ms) {
3756   struct timespec req;
3757 
3758   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3759   req.tv_sec = 0;
3760   if (ms > 0) {
3761     req.tv_nsec = (ms % 1000) * 1000000;
3762   } else {
3763     req.tv_nsec = 1;
3764   }
3765 
3766   nanosleep(&req, NULL);
3767 
3768   return;
3769 }
3770 
3771 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3772 void os::infinite_sleep() {
3773   while (true) {    // sleep forever ...
3774     ::sleep(100);   // ... 100 seconds at a time
3775   }
3776 }
3777 
3778 // Used to convert frequent JVM_Yield() to nops
3779 bool os::dont_yield() {
3780   return DontYieldALot;
3781 }
3782 
3783 void os::naked_yield() {
3784   sched_yield();
3785 }
3786 
3787 ////////////////////////////////////////////////////////////////////////////////
3788 // thread priority support
3789 
3790 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3791 // only supports dynamic priority, static priority must be zero. For real-time
3792 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3793 // However, for large multi-threaded applications, SCHED_RR is not only slower
3794 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3795 // of 5 runs - Sep 2005).
3796 //
3797 // The following code actually changes the niceness of kernel-thread/LWP. It
3798 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3799 // not the entire user process, and user level threads are 1:1 mapped to kernel
3800 // threads. It has always been the case, but could change in the future. For
3801 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3802 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3803 
3804 int os::java_to_os_priority[CriticalPriority + 1] = {
3805   19,              // 0 Entry should never be used
3806 
3807    4,              // 1 MinPriority
3808    3,              // 2
3809    2,              // 3
3810 
3811    1,              // 4
3812    0,              // 5 NormPriority
3813   -1,              // 6
3814 
3815   -2,              // 7
3816   -3,              // 8
3817   -4,              // 9 NearMaxPriority
3818 
3819   -5,              // 10 MaxPriority
3820 
3821   -5               // 11 CriticalPriority
3822 };
3823 
3824 static int prio_init() {
3825   if (ThreadPriorityPolicy == 1) {
3826     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3827     // if effective uid is not root. Perhaps, a more elegant way of doing
3828     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3829     if (geteuid() != 0) {
3830       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3831         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3832       }
3833       ThreadPriorityPolicy = 0;
3834     }
3835   }
3836   if (UseCriticalJavaThreadPriority) {
3837     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3838   }
3839   return 0;
3840 }
3841 
3842 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3843   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3844 
3845   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3846   return (ret == 0) ? OS_OK : OS_ERR;
3847 }
3848 
3849 OSReturn os::get_native_priority(const Thread* const thread,
3850                                  int *priority_ptr) {
3851   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3852     *priority_ptr = java_to_os_priority[NormPriority];
3853     return OS_OK;
3854   }
3855 
3856   errno = 0;
3857   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3858   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3859 }
3860 
3861 // Hint to the underlying OS that a task switch would not be good.
3862 // Void return because it's a hint and can fail.
3863 void os::hint_no_preempt() {}
3864 
3865 ////////////////////////////////////////////////////////////////////////////////
3866 // suspend/resume support
3867 
3868 //  the low-level signal-based suspend/resume support is a remnant from the
3869 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3870 //  within hotspot. Now there is a single use-case for this:
3871 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3872 //      that runs in the watcher thread.
3873 //  The remaining code is greatly simplified from the more general suspension
3874 //  code that used to be used.
3875 //
3876 //  The protocol is quite simple:
3877 //  - suspend:
3878 //      - sends a signal to the target thread
3879 //      - polls the suspend state of the osthread using a yield loop
3880 //      - target thread signal handler (SR_handler) sets suspend state
3881 //        and blocks in sigsuspend until continued
3882 //  - resume:
3883 //      - sets target osthread state to continue
3884 //      - sends signal to end the sigsuspend loop in the SR_handler
3885 //
3886 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3887 
3888 static void resume_clear_context(OSThread *osthread) {
3889   osthread->set_ucontext(NULL);
3890   osthread->set_siginfo(NULL);
3891 }
3892 
3893 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3894                                  ucontext_t* context) {
3895   osthread->set_ucontext(context);
3896   osthread->set_siginfo(siginfo);
3897 }
3898 
3899 // Handler function invoked when a thread's execution is suspended or
3900 // resumed. We have to be careful that only async-safe functions are
3901 // called here (Note: most pthread functions are not async safe and
3902 // should be avoided.)
3903 //
3904 // Note: sigwait() is a more natural fit than sigsuspend() from an
3905 // interface point of view, but sigwait() prevents the signal hander
3906 // from being run. libpthread would get very confused by not having
3907 // its signal handlers run and prevents sigwait()'s use with the
3908 // mutex granting granting signal.
3909 //
3910 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3911 //
3912 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3913   // Save and restore errno to avoid confusing native code with EINTR
3914   // after sigsuspend.
3915   int old_errno = errno;
3916 
3917   Thread* thread = Thread::current();
3918   OSThread* osthread = thread->osthread();
3919   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3920 
3921   os::SuspendResume::State current = osthread->sr.state();
3922   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3923     suspend_save_context(osthread, siginfo, context);
3924 
3925     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3926     os::SuspendResume::State state = osthread->sr.suspended();
3927     if (state == os::SuspendResume::SR_SUSPENDED) {
3928       sigset_t suspend_set;  // signals for sigsuspend()
3929 
3930       // get current set of blocked signals and unblock resume signal
3931       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3932       sigdelset(&suspend_set, SR_signum);
3933 
3934       sr_semaphore.signal();
3935       // wait here until we are resumed
3936       while (1) {
3937         sigsuspend(&suspend_set);
3938 
3939         os::SuspendResume::State result = osthread->sr.running();
3940         if (result == os::SuspendResume::SR_RUNNING) {
3941           sr_semaphore.signal();
3942           break;
3943         }
3944       }
3945 
3946     } else if (state == os::SuspendResume::SR_RUNNING) {
3947       // request was cancelled, continue
3948     } else {
3949       ShouldNotReachHere();
3950     }
3951 
3952     resume_clear_context(osthread);
3953   } else if (current == os::SuspendResume::SR_RUNNING) {
3954     // request was cancelled, continue
3955   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3956     // ignore
3957   } else {
3958     // ignore
3959   }
3960 
3961   errno = old_errno;
3962 }
3963 
3964 static int SR_initialize() {
3965   struct sigaction act;
3966   char *s;
3967 
3968   // Get signal number to use for suspend/resume
3969   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3970     int sig = ::strtol(s, 0, 10);
3971     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
3972         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
3973       SR_signum = sig;
3974     } else {
3975       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
3976               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
3977     }
3978   }
3979 
3980   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3981          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3982 
3983   sigemptyset(&SR_sigset);
3984   sigaddset(&SR_sigset, SR_signum);
3985 
3986   // Set up signal handler for suspend/resume
3987   act.sa_flags = SA_RESTART|SA_SIGINFO;
3988   act.sa_handler = (void (*)(int)) SR_handler;
3989 
3990   // SR_signum is blocked by default.
3991   // 4528190 - We also need to block pthread restart signal (32 on all
3992   // supported Linux platforms). Note that LinuxThreads need to block
3993   // this signal for all threads to work properly. So we don't have
3994   // to use hard-coded signal number when setting up the mask.
3995   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3996 
3997   if (sigaction(SR_signum, &act, 0) == -1) {
3998     return -1;
3999   }
4000 
4001   // Save signal flag
4002   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4003   return 0;
4004 }
4005 
4006 static int sr_notify(OSThread* osthread) {
4007   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4008   assert_status(status == 0, status, "pthread_kill");
4009   return status;
4010 }
4011 
4012 // "Randomly" selected value for how long we want to spin
4013 // before bailing out on suspending a thread, also how often
4014 // we send a signal to a thread we want to resume
4015 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4016 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4017 
4018 // returns true on success and false on error - really an error is fatal
4019 // but this seems the normal response to library errors
4020 static bool do_suspend(OSThread* osthread) {
4021   assert(osthread->sr.is_running(), "thread should be running");
4022   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4023 
4024   // mark as suspended and send signal
4025   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4026     // failed to switch, state wasn't running?
4027     ShouldNotReachHere();
4028     return false;
4029   }
4030 
4031   if (sr_notify(osthread) != 0) {
4032     ShouldNotReachHere();
4033   }
4034 
4035   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4036   while (true) {
4037     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4038       break;
4039     } else {
4040       // timeout
4041       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4042       if (cancelled == os::SuspendResume::SR_RUNNING) {
4043         return false;
4044       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4045         // make sure that we consume the signal on the semaphore as well
4046         sr_semaphore.wait();
4047         break;
4048       } else {
4049         ShouldNotReachHere();
4050         return false;
4051       }
4052     }
4053   }
4054 
4055   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4056   return true;
4057 }
4058 
4059 static void do_resume(OSThread* osthread) {
4060   assert(osthread->sr.is_suspended(), "thread should be suspended");
4061   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4062 
4063   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4064     // failed to switch to WAKEUP_REQUEST
4065     ShouldNotReachHere();
4066     return;
4067   }
4068 
4069   while (true) {
4070     if (sr_notify(osthread) == 0) {
4071       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4072         if (osthread->sr.is_running()) {
4073           return;
4074         }
4075       }
4076     } else {
4077       ShouldNotReachHere();
4078     }
4079   }
4080 
4081   guarantee(osthread->sr.is_running(), "Must be running!");
4082 }
4083 
4084 ///////////////////////////////////////////////////////////////////////////////////
4085 // signal handling (except suspend/resume)
4086 
4087 // This routine may be used by user applications as a "hook" to catch signals.
4088 // The user-defined signal handler must pass unrecognized signals to this
4089 // routine, and if it returns true (non-zero), then the signal handler must
4090 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4091 // routine will never retun false (zero), but instead will execute a VM panic
4092 // routine kill the process.
4093 //
4094 // If this routine returns false, it is OK to call it again.  This allows
4095 // the user-defined signal handler to perform checks either before or after
4096 // the VM performs its own checks.  Naturally, the user code would be making
4097 // a serious error if it tried to handle an exception (such as a null check
4098 // or breakpoint) that the VM was generating for its own correct operation.
4099 //
4100 // This routine may recognize any of the following kinds of signals:
4101 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4102 // It should be consulted by handlers for any of those signals.
4103 //
4104 // The caller of this routine must pass in the three arguments supplied
4105 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4106 // field of the structure passed to sigaction().  This routine assumes that
4107 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4108 //
4109 // Note that the VM will print warnings if it detects conflicting signal
4110 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4111 //
4112 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4113                                                  siginfo_t* siginfo,
4114                                                  void* ucontext,
4115                                                  int abort_if_unrecognized);
4116 
4117 void signalHandler(int sig, siginfo_t* info, void* uc) {
4118   assert(info != NULL && uc != NULL, "it must be old kernel");
4119   int orig_errno = errno;  // Preserve errno value over signal handler.
4120   JVM_handle_linux_signal(sig, info, uc, true);
4121   errno = orig_errno;
4122 }
4123 
4124 
4125 // This boolean allows users to forward their own non-matching signals
4126 // to JVM_handle_linux_signal, harmlessly.
4127 bool os::Linux::signal_handlers_are_installed = false;
4128 
4129 // For signal-chaining
4130 struct sigaction sigact[NSIG];
4131 uint64_t sigs = 0;
4132 #if (64 < NSIG-1)
4133 #error "Not all signals can be encoded in sigs. Adapt its type!"
4134 #endif
4135 bool os::Linux::libjsig_is_loaded = false;
4136 typedef struct sigaction *(*get_signal_t)(int);
4137 get_signal_t os::Linux::get_signal_action = NULL;
4138 
4139 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4140   struct sigaction *actp = NULL;
4141 
4142   if (libjsig_is_loaded) {
4143     // Retrieve the old signal handler from libjsig
4144     actp = (*get_signal_action)(sig);
4145   }
4146   if (actp == NULL) {
4147     // Retrieve the preinstalled signal handler from jvm
4148     actp = get_preinstalled_handler(sig);
4149   }
4150 
4151   return actp;
4152 }
4153 
4154 static bool call_chained_handler(struct sigaction *actp, int sig,
4155                                  siginfo_t *siginfo, void *context) {
4156   // Call the old signal handler
4157   if (actp->sa_handler == SIG_DFL) {
4158     // It's more reasonable to let jvm treat it as an unexpected exception
4159     // instead of taking the default action.
4160     return false;
4161   } else if (actp->sa_handler != SIG_IGN) {
4162     if ((actp->sa_flags & SA_NODEFER) == 0) {
4163       // automaticlly block the signal
4164       sigaddset(&(actp->sa_mask), sig);
4165     }
4166 
4167     sa_handler_t hand = NULL;
4168     sa_sigaction_t sa = NULL;
4169     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4170     // retrieve the chained handler
4171     if (siginfo_flag_set) {
4172       sa = actp->sa_sigaction;
4173     } else {
4174       hand = actp->sa_handler;
4175     }
4176 
4177     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4178       actp->sa_handler = SIG_DFL;
4179     }
4180 
4181     // try to honor the signal mask
4182     sigset_t oset;
4183     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4184 
4185     // call into the chained handler
4186     if (siginfo_flag_set) {
4187       (*sa)(sig, siginfo, context);
4188     } else {
4189       (*hand)(sig);
4190     }
4191 
4192     // restore the signal mask
4193     pthread_sigmask(SIG_SETMASK, &oset, 0);
4194   }
4195   // Tell jvm's signal handler the signal is taken care of.
4196   return true;
4197 }
4198 
4199 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4200   bool chained = false;
4201   // signal-chaining
4202   if (UseSignalChaining) {
4203     struct sigaction *actp = get_chained_signal_action(sig);
4204     if (actp != NULL) {
4205       chained = call_chained_handler(actp, sig, siginfo, context);
4206     }
4207   }
4208   return chained;
4209 }
4210 
4211 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4212   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4213     return &sigact[sig];
4214   }
4215   return NULL;
4216 }
4217 
4218 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4219   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4220   sigact[sig] = oldAct;
4221   sigs |= (uint64_t)1 << (sig-1);
4222 }
4223 
4224 // for diagnostic
4225 int sigflags[NSIG];
4226 
4227 int os::Linux::get_our_sigflags(int sig) {
4228   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4229   return sigflags[sig];
4230 }
4231 
4232 void os::Linux::set_our_sigflags(int sig, int flags) {
4233   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4234   if (sig > 0 && sig < NSIG) {
4235     sigflags[sig] = flags;
4236   }
4237 }
4238 
4239 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4240   // Check for overwrite.
4241   struct sigaction oldAct;
4242   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4243 
4244   void* oldhand = oldAct.sa_sigaction
4245                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4246                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4247   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4248       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4249       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4250     if (AllowUserSignalHandlers || !set_installed) {
4251       // Do not overwrite; user takes responsibility to forward to us.
4252       return;
4253     } else if (UseSignalChaining) {
4254       // save the old handler in jvm
4255       save_preinstalled_handler(sig, oldAct);
4256       // libjsig also interposes the sigaction() call below and saves the
4257       // old sigaction on it own.
4258     } else {
4259       fatal("Encountered unexpected pre-existing sigaction handler "
4260             "%#lx for signal %d.", (long)oldhand, sig);
4261     }
4262   }
4263 
4264   struct sigaction sigAct;
4265   sigfillset(&(sigAct.sa_mask));
4266   sigAct.sa_handler = SIG_DFL;
4267   if (!set_installed) {
4268     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4269   } else {
4270     sigAct.sa_sigaction = signalHandler;
4271     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4272   }
4273   // Save flags, which are set by ours
4274   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4275   sigflags[sig] = sigAct.sa_flags;
4276 
4277   int ret = sigaction(sig, &sigAct, &oldAct);
4278   assert(ret == 0, "check");
4279 
4280   void* oldhand2  = oldAct.sa_sigaction
4281                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4282                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4283   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4284 }
4285 
4286 // install signal handlers for signals that HotSpot needs to
4287 // handle in order to support Java-level exception handling.
4288 
4289 void os::Linux::install_signal_handlers() {
4290   if (!signal_handlers_are_installed) {
4291     signal_handlers_are_installed = true;
4292 
4293     // signal-chaining
4294     typedef void (*signal_setting_t)();
4295     signal_setting_t begin_signal_setting = NULL;
4296     signal_setting_t end_signal_setting = NULL;
4297     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4298                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4299     if (begin_signal_setting != NULL) {
4300       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4301                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4302       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4303                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4304       libjsig_is_loaded = true;
4305       assert(UseSignalChaining, "should enable signal-chaining");
4306     }
4307     if (libjsig_is_loaded) {
4308       // Tell libjsig jvm is setting signal handlers
4309       (*begin_signal_setting)();
4310     }
4311 
4312     set_signal_handler(SIGSEGV, true);
4313     set_signal_handler(SIGPIPE, true);
4314     set_signal_handler(SIGBUS, true);
4315     set_signal_handler(SIGILL, true);
4316     set_signal_handler(SIGFPE, true);
4317 #if defined(PPC64)
4318     set_signal_handler(SIGTRAP, true);
4319 #endif
4320     set_signal_handler(SIGXFSZ, true);
4321 
4322     if (libjsig_is_loaded) {
4323       // Tell libjsig jvm finishes setting signal handlers
4324       (*end_signal_setting)();
4325     }
4326 
4327     // We don't activate signal checker if libjsig is in place, we trust ourselves
4328     // and if UserSignalHandler is installed all bets are off.
4329     // Log that signal checking is off only if -verbose:jni is specified.
4330     if (CheckJNICalls) {
4331       if (libjsig_is_loaded) {
4332         if (PrintJNIResolving) {
4333           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4334         }
4335         check_signals = false;
4336       }
4337       if (AllowUserSignalHandlers) {
4338         if (PrintJNIResolving) {
4339           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4340         }
4341         check_signals = false;
4342       }
4343     }
4344   }
4345 }
4346 
4347 // This is the fastest way to get thread cpu time on Linux.
4348 // Returns cpu time (user+sys) for any thread, not only for current.
4349 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4350 // It might work on 2.6.10+ with a special kernel/glibc patch.
4351 // For reference, please, see IEEE Std 1003.1-2004:
4352 //   http://www.unix.org/single_unix_specification
4353 
4354 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4355   struct timespec tp;
4356   int rc = os::Linux::clock_gettime(clockid, &tp);
4357   assert(rc == 0, "clock_gettime is expected to return 0 code");
4358 
4359   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4360 }
4361 
4362 /////
4363 // glibc on Linux platform uses non-documented flag
4364 // to indicate, that some special sort of signal
4365 // trampoline is used.
4366 // We will never set this flag, and we should
4367 // ignore this flag in our diagnostic
4368 #ifdef SIGNIFICANT_SIGNAL_MASK
4369   #undef SIGNIFICANT_SIGNAL_MASK
4370 #endif
4371 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4372 
4373 static const char* get_signal_handler_name(address handler,
4374                                            char* buf, int buflen) {
4375   int offset = 0;
4376   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4377   if (found) {
4378     // skip directory names
4379     const char *p1, *p2;
4380     p1 = buf;
4381     size_t len = strlen(os::file_separator());
4382     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4383     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4384   } else {
4385     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4386   }
4387   return buf;
4388 }
4389 
4390 static void print_signal_handler(outputStream* st, int sig,
4391                                  char* buf, size_t buflen) {
4392   struct sigaction sa;
4393 
4394   sigaction(sig, NULL, &sa);
4395 
4396   // See comment for SIGNIFICANT_SIGNAL_MASK define
4397   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4398 
4399   st->print("%s: ", os::exception_name(sig, buf, buflen));
4400 
4401   address handler = (sa.sa_flags & SA_SIGINFO)
4402     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4403     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4404 
4405   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4406     st->print("SIG_DFL");
4407   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4408     st->print("SIG_IGN");
4409   } else {
4410     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4411   }
4412 
4413   st->print(", sa_mask[0]=");
4414   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4415 
4416   address rh = VMError::get_resetted_sighandler(sig);
4417   // May be, handler was resetted by VMError?
4418   if (rh != NULL) {
4419     handler = rh;
4420     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4421   }
4422 
4423   st->print(", sa_flags=");
4424   os::Posix::print_sa_flags(st, sa.sa_flags);
4425 
4426   // Check: is it our handler?
4427   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4428       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4429     // It is our signal handler
4430     // check for flags, reset system-used one!
4431     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4432       st->print(
4433                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4434                 os::Linux::get_our_sigflags(sig));
4435     }
4436   }
4437   st->cr();
4438 }
4439 
4440 
4441 #define DO_SIGNAL_CHECK(sig)                      \
4442   do {                                            \
4443     if (!sigismember(&check_signal_done, sig)) {  \
4444       os::Linux::check_signal_handler(sig);       \
4445     }                                             \
4446   } while (0)
4447 
4448 // This method is a periodic task to check for misbehaving JNI applications
4449 // under CheckJNI, we can add any periodic checks here
4450 
4451 void os::run_periodic_checks() {
4452   if (check_signals == false) return;
4453 
4454   // SEGV and BUS if overridden could potentially prevent
4455   // generation of hs*.log in the event of a crash, debugging
4456   // such a case can be very challenging, so we absolutely
4457   // check the following for a good measure:
4458   DO_SIGNAL_CHECK(SIGSEGV);
4459   DO_SIGNAL_CHECK(SIGILL);
4460   DO_SIGNAL_CHECK(SIGFPE);
4461   DO_SIGNAL_CHECK(SIGBUS);
4462   DO_SIGNAL_CHECK(SIGPIPE);
4463   DO_SIGNAL_CHECK(SIGXFSZ);
4464 #if defined(PPC64)
4465   DO_SIGNAL_CHECK(SIGTRAP);
4466 #endif
4467 
4468   // ReduceSignalUsage allows the user to override these handlers
4469   // see comments at the very top and jvm_solaris.h
4470   if (!ReduceSignalUsage) {
4471     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4472     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4473     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4474     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4475   }
4476 
4477   DO_SIGNAL_CHECK(SR_signum);
4478 }
4479 
4480 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4481 
4482 static os_sigaction_t os_sigaction = NULL;
4483 
4484 void os::Linux::check_signal_handler(int sig) {
4485   char buf[O_BUFLEN];
4486   address jvmHandler = NULL;
4487 
4488 
4489   struct sigaction act;
4490   if (os_sigaction == NULL) {
4491     // only trust the default sigaction, in case it has been interposed
4492     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4493     if (os_sigaction == NULL) return;
4494   }
4495 
4496   os_sigaction(sig, (struct sigaction*)NULL, &act);
4497 
4498 
4499   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4500 
4501   address thisHandler = (act.sa_flags & SA_SIGINFO)
4502     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4503     : CAST_FROM_FN_PTR(address, act.sa_handler);
4504 
4505 
4506   switch (sig) {
4507   case SIGSEGV:
4508   case SIGBUS:
4509   case SIGFPE:
4510   case SIGPIPE:
4511   case SIGILL:
4512   case SIGXFSZ:
4513     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4514     break;
4515 
4516   case SHUTDOWN1_SIGNAL:
4517   case SHUTDOWN2_SIGNAL:
4518   case SHUTDOWN3_SIGNAL:
4519   case BREAK_SIGNAL:
4520     jvmHandler = (address)user_handler();
4521     break;
4522 
4523   default:
4524     if (sig == SR_signum) {
4525       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4526     } else {
4527       return;
4528     }
4529     break;
4530   }
4531 
4532   if (thisHandler != jvmHandler) {
4533     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4534     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4535     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4536     // No need to check this sig any longer
4537     sigaddset(&check_signal_done, sig);
4538     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4539     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4540       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4541                     exception_name(sig, buf, O_BUFLEN));
4542     }
4543   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4544     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4545     tty->print("expected:");
4546     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4547     tty->cr();
4548     tty->print("  found:");
4549     os::Posix::print_sa_flags(tty, act.sa_flags);
4550     tty->cr();
4551     // No need to check this sig any longer
4552     sigaddset(&check_signal_done, sig);
4553   }
4554 
4555   // Dump all the signal
4556   if (sigismember(&check_signal_done, sig)) {
4557     print_signal_handlers(tty, buf, O_BUFLEN);
4558   }
4559 }
4560 
4561 extern void report_error(char* file_name, int line_no, char* title,
4562                          char* format, ...);
4563 
4564 // this is called _before_ the most of global arguments have been parsed
4565 void os::init(void) {
4566   char dummy;   // used to get a guess on initial stack address
4567 //  first_hrtime = gethrtime();
4568 
4569   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4570 
4571   init_random(1234567);
4572 
4573   ThreadCritical::initialize();
4574 
4575   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4576   if (Linux::page_size() == -1) {
4577     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4578           strerror(errno));
4579   }
4580   init_page_sizes((size_t) Linux::page_size());
4581 
4582   Linux::initialize_system_info();
4583 
4584   // main_thread points to the aboriginal thread
4585   Linux::_main_thread = pthread_self();
4586 
4587   Linux::clock_init();
4588   initial_time_count = javaTimeNanos();
4589 
4590   // pthread_condattr initialization for monotonic clock
4591   int status;
4592   pthread_condattr_t* _condattr = os::Linux::condAttr();
4593   if ((status = pthread_condattr_init(_condattr)) != 0) {
4594     fatal("pthread_condattr_init: %s", strerror(status));
4595   }
4596   // Only set the clock if CLOCK_MONOTONIC is available
4597   if (os::supports_monotonic_clock()) {
4598     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4599       if (status == EINVAL) {
4600         warning("Unable to use monotonic clock with relative timed-waits" \
4601                 " - changes to the time-of-day clock may have adverse affects");
4602       } else {
4603         fatal("pthread_condattr_setclock: %s", strerror(status));
4604       }
4605     }
4606   }
4607   // else it defaults to CLOCK_REALTIME
4608 
4609   // retrieve entry point for pthread_setname_np
4610   Linux::_pthread_setname_np =
4611     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4612 
4613 }
4614 
4615 // To install functions for atexit system call
4616 extern "C" {
4617   static void perfMemory_exit_helper() {
4618     perfMemory_exit();
4619   }
4620 }
4621 
4622 // this is called _after_ the global arguments have been parsed
4623 jint os::init_2(void) {
4624   Linux::fast_thread_clock_init();
4625 
4626   // Allocate a single page and mark it as readable for safepoint polling
4627   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4628   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4629 
4630   os::set_polling_page(polling_page);
4631 
4632 #ifndef PRODUCT
4633   if (Verbose && PrintMiscellaneous) {
4634     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4635                (intptr_t)polling_page);
4636   }
4637 #endif
4638 
4639   if (!UseMembar) {
4640     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4641     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4642     os::set_memory_serialize_page(mem_serialize_page);
4643 
4644 #ifndef PRODUCT
4645     if (Verbose && PrintMiscellaneous) {
4646       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n",
4647                  (intptr_t)mem_serialize_page);
4648     }
4649 #endif
4650   }
4651 
4652   // initialize suspend/resume support - must do this before signal_sets_init()
4653   if (SR_initialize() != 0) {
4654     perror("SR_initialize failed");
4655     return JNI_ERR;
4656   }
4657 
4658   Linux::signal_sets_init();
4659   Linux::install_signal_handlers();
4660 
4661   // Check minimum allowable stack size for thread creation and to initialize
4662   // the java system classes, including StackOverflowError - depends on page
4663   // size.  Add a page for compiler2 recursion in main thread.
4664   // Add in 2*BytesPerWord times page size to account for VM stack during
4665   // class initialization depending on 32 or 64 bit VM.
4666   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4667                                       JavaThread::stack_guard_zone_size() +
4668                                       JavaThread::stack_shadow_zone_size() +
4669                                       (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4670 
4671   size_t threadStackSizeInBytes = ThreadStackSize * K;
4672   if (threadStackSizeInBytes != 0 &&
4673       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4674     tty->print_cr("\nThe stack size specified is too small, "
4675                   "Specify at least " SIZE_FORMAT "k",
4676                   os::Linux::min_stack_allowed/ K);
4677     return JNI_ERR;
4678   }
4679 
4680   // Make the stack size a multiple of the page size so that
4681   // the yellow/red zones can be guarded.
4682   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4683                                                 vm_page_size()));
4684 
4685   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4686 
4687 #if defined(IA32)
4688   workaround_expand_exec_shield_cs_limit();
4689 #endif
4690 
4691   Linux::libpthread_init();
4692   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4693     tty->print_cr("[HotSpot is running with %s, %s]\n",
4694                   Linux::glibc_version(), Linux::libpthread_version());
4695   }
4696 
4697   if (UseNUMA) {
4698     if (!Linux::libnuma_init()) {
4699       UseNUMA = false;
4700     } else {
4701       if ((Linux::numa_max_node() < 1)) {
4702         // There's only one node(they start from 0), disable NUMA.
4703         UseNUMA = false;
4704       }
4705     }
4706     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4707     // we can make the adaptive lgrp chunk resizing work. If the user specified
4708     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4709     // disable adaptive resizing.
4710     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4711       if (FLAG_IS_DEFAULT(UseNUMA)) {
4712         UseNUMA = false;
4713       } else {
4714         if (FLAG_IS_DEFAULT(UseLargePages) &&
4715             FLAG_IS_DEFAULT(UseSHM) &&
4716             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4717           UseLargePages = false;
4718         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4719           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4720           UseAdaptiveSizePolicy = false;
4721           UseAdaptiveNUMAChunkSizing = false;
4722         }
4723       }
4724     }
4725     if (!UseNUMA && ForceNUMA) {
4726       UseNUMA = true;
4727     }
4728   }
4729 
4730   if (MaxFDLimit) {
4731     // set the number of file descriptors to max. print out error
4732     // if getrlimit/setrlimit fails but continue regardless.
4733     struct rlimit nbr_files;
4734     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4735     if (status != 0) {
4736       if (PrintMiscellaneous && (Verbose || WizardMode)) {
4737         perror("os::init_2 getrlimit failed");
4738       }
4739     } else {
4740       nbr_files.rlim_cur = nbr_files.rlim_max;
4741       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4742       if (status != 0) {
4743         if (PrintMiscellaneous && (Verbose || WizardMode)) {
4744           perror("os::init_2 setrlimit failed");
4745         }
4746       }
4747     }
4748   }
4749 
4750   // Initialize lock used to serialize thread creation (see os::create_thread)
4751   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4752 
4753   // at-exit methods are called in the reverse order of their registration.
4754   // atexit functions are called on return from main or as a result of a
4755   // call to exit(3C). There can be only 32 of these functions registered
4756   // and atexit() does not set errno.
4757 
4758   if (PerfAllowAtExitRegistration) {
4759     // only register atexit functions if PerfAllowAtExitRegistration is set.
4760     // atexit functions can be delayed until process exit time, which
4761     // can be problematic for embedded VM situations. Embedded VMs should
4762     // call DestroyJavaVM() to assure that VM resources are released.
4763 
4764     // note: perfMemory_exit_helper atexit function may be removed in
4765     // the future if the appropriate cleanup code can be added to the
4766     // VM_Exit VMOperation's doit method.
4767     if (atexit(perfMemory_exit_helper) != 0) {
4768       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4769     }
4770   }
4771 
4772   // initialize thread priority policy
4773   prio_init();
4774 
4775   return JNI_OK;
4776 }
4777 
4778 // Mark the polling page as unreadable
4779 void os::make_polling_page_unreadable(void) {
4780   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4781     fatal("Could not disable polling page");
4782   }
4783 }
4784 
4785 // Mark the polling page as readable
4786 void os::make_polling_page_readable(void) {
4787   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4788     fatal("Could not enable polling page");
4789   }
4790 }
4791 
4792 // older glibc versions don't have this macro (which expands to
4793 // an optimized bit-counting function) so we have to roll our own
4794 #ifndef CPU_COUNT
4795 
4796 static int _cpu_count(const cpu_set_t* cpus) {
4797   int count = 0;
4798   // only look up to the number of configured processors
4799   for (int i = 0; i < os::processor_count(); i++) {
4800     if (CPU_ISSET(i, cpus)) {
4801       count++;
4802     }
4803   }
4804   return count;
4805 }
4806 
4807 #define CPU_COUNT(cpus) _cpu_count(cpus)
4808 
4809 #endif // CPU_COUNT
4810 
4811 // Get the current number of available processors for this process.
4812 // This value can change at any time during a process's lifetime.
4813 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4814 // If it appears there may be more than 1024 processors then we do a
4815 // dynamic check - see 6515172 for details.
4816 // If anything goes wrong we fallback to returning the number of online
4817 // processors - which can be greater than the number available to the process.
4818 int os::active_processor_count() {
4819   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
4820   cpu_set_t* cpus_p = &cpus;
4821   int cpus_size = sizeof(cpu_set_t);
4822 
4823   int configured_cpus = processor_count();  // upper bound on available cpus
4824   int cpu_count = 0;
4825 
4826 // old build platforms may not support dynamic cpu sets
4827 #ifdef CPU_ALLOC
4828 
4829   // To enable easy testing of the dynamic path on different platforms we
4830   // introduce a diagnostic flag: UseCpuAllocPath
4831   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4832     // kernel may use a mask bigger than cpu_set_t
4833     log_trace(os)("active_processor_count: using dynamic path %s"
4834                   "- configured processors: %d",
4835                   UseCpuAllocPath ? "(forced) " : "",
4836                   configured_cpus);
4837     cpus_p = CPU_ALLOC(configured_cpus);
4838     if (cpus_p != NULL) {
4839       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4840       // zero it just to be safe
4841       CPU_ZERO_S(cpus_size, cpus_p);
4842     }
4843     else {
4844        // failed to allocate so fallback to online cpus
4845        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4846        log_trace(os)("active_processor_count: "
4847                      "CPU_ALLOC failed (%s) - using "
4848                      "online processor count: %d",
4849                      strerror(errno), online_cpus);
4850        return online_cpus;
4851     }
4852   }
4853   else {
4854     log_trace(os)("active_processor_count: using static path - configured processors: %d",
4855                   configured_cpus);
4856   }
4857 #else // CPU_ALLOC
4858 // these stubs won't be executed
4859 #define CPU_COUNT_S(size, cpus) -1
4860 #define CPU_FREE(cpus)
4861 
4862   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4863                 configured_cpus);
4864 #endif // CPU_ALLOC
4865 
4866   // pid 0 means the current thread - which we have to assume represents the process
4867   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4868     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4869       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4870     }
4871     else {
4872       cpu_count = CPU_COUNT(cpus_p);
4873     }
4874     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4875   }
4876   else {
4877     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4878     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4879             "which may exceed available processors", strerror(errno), cpu_count);
4880   }
4881 
4882   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4883     CPU_FREE(cpus_p);
4884   }
4885 
4886   assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
4887   return cpu_count;
4888 }
4889 
4890 void os::set_native_thread_name(const char *name) {
4891   if (Linux::_pthread_setname_np) {
4892     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4893     snprintf(buf, sizeof(buf), "%s", name);
4894     buf[sizeof(buf) - 1] = '\0';
4895     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4896     // ERANGE should not happen; all other errors should just be ignored.
4897     assert(rc != ERANGE, "pthread_setname_np failed");
4898   }
4899 }
4900 
4901 bool os::distribute_processes(uint length, uint* distribution) {
4902   // Not yet implemented.
4903   return false;
4904 }
4905 
4906 bool os::bind_to_processor(uint processor_id) {
4907   // Not yet implemented.
4908   return false;
4909 }
4910 
4911 ///
4912 
4913 void os::SuspendedThreadTask::internal_do_task() {
4914   if (do_suspend(_thread->osthread())) {
4915     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4916     do_task(context);
4917     do_resume(_thread->osthread());
4918   }
4919 }
4920 
4921 class PcFetcher : public os::SuspendedThreadTask {
4922  public:
4923   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4924   ExtendedPC result();
4925  protected:
4926   void do_task(const os::SuspendedThreadTaskContext& context);
4927  private:
4928   ExtendedPC _epc;
4929 };
4930 
4931 ExtendedPC PcFetcher::result() {
4932   guarantee(is_done(), "task is not done yet.");
4933   return _epc;
4934 }
4935 
4936 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4937   Thread* thread = context.thread();
4938   OSThread* osthread = thread->osthread();
4939   if (osthread->ucontext() != NULL) {
4940     _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
4941   } else {
4942     // NULL context is unexpected, double-check this is the VMThread
4943     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4944   }
4945 }
4946 
4947 // Suspends the target using the signal mechanism and then grabs the PC before
4948 // resuming the target. Used by the flat-profiler only
4949 ExtendedPC os::get_thread_pc(Thread* thread) {
4950   // Make sure that it is called by the watcher for the VMThread
4951   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4952   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4953 
4954   PcFetcher fetcher(thread);
4955   fetcher.run();
4956   return fetcher.result();
4957 }
4958 
4959 ////////////////////////////////////////////////////////////////////////////////
4960 // debug support
4961 
4962 bool os::find(address addr, outputStream* st) {
4963   Dl_info dlinfo;
4964   memset(&dlinfo, 0, sizeof(dlinfo));
4965   if (dladdr(addr, &dlinfo) != 0) {
4966     st->print(PTR_FORMAT ": ", p2i(addr));
4967     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4968       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
4969                 p2i(addr) - p2i(dlinfo.dli_saddr));
4970     } else if (dlinfo.dli_fbase != NULL) {
4971       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
4972     } else {
4973       st->print("<absolute address>");
4974     }
4975     if (dlinfo.dli_fname != NULL) {
4976       st->print(" in %s", dlinfo.dli_fname);
4977     }
4978     if (dlinfo.dli_fbase != NULL) {
4979       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
4980     }
4981     st->cr();
4982 
4983     if (Verbose) {
4984       // decode some bytes around the PC
4985       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4986       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4987       address       lowest = (address) dlinfo.dli_sname;
4988       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4989       if (begin < lowest)  begin = lowest;
4990       Dl_info dlinfo2;
4991       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4992           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4993         end = (address) dlinfo2.dli_saddr;
4994       }
4995       Disassembler::decode(begin, end, st);
4996     }
4997     return true;
4998   }
4999   return false;
5000 }
5001 
5002 ////////////////////////////////////////////////////////////////////////////////
5003 // misc
5004 
5005 // This does not do anything on Linux. This is basically a hook for being
5006 // able to use structured exception handling (thread-local exception filters)
5007 // on, e.g., Win32.
5008 void
5009 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5010                          JavaCallArguments* args, Thread* thread) {
5011   f(value, method, args, thread);
5012 }
5013 
5014 void os::print_statistics() {
5015 }
5016 
5017 bool os::message_box(const char* title, const char* message) {
5018   int i;
5019   fdStream err(defaultStream::error_fd());
5020   for (i = 0; i < 78; i++) err.print_raw("=");
5021   err.cr();
5022   err.print_raw_cr(title);
5023   for (i = 0; i < 78; i++) err.print_raw("-");
5024   err.cr();
5025   err.print_raw_cr(message);
5026   for (i = 0; i < 78; i++) err.print_raw("=");
5027   err.cr();
5028 
5029   char buf[16];
5030   // Prevent process from exiting upon "read error" without consuming all CPU
5031   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5032 
5033   return buf[0] == 'y' || buf[0] == 'Y';
5034 }
5035 
5036 int os::stat(const char *path, struct stat *sbuf) {
5037   char pathbuf[MAX_PATH];
5038   if (strlen(path) > MAX_PATH - 1) {
5039     errno = ENAMETOOLONG;
5040     return -1;
5041   }
5042   os::native_path(strcpy(pathbuf, path));
5043   return ::stat(pathbuf, sbuf);
5044 }
5045 
5046 bool os::check_heap(bool force) {
5047   return true;
5048 }
5049 
5050 // Is a (classpath) directory empty?
5051 bool os::dir_is_empty(const char* path) {
5052   DIR *dir = NULL;
5053   struct dirent *ptr;
5054 
5055   dir = opendir(path);
5056   if (dir == NULL) return true;
5057 
5058   // Scan the directory
5059   bool result = true;
5060   char buf[sizeof(struct dirent) + MAX_PATH];
5061   while (result && (ptr = ::readdir(dir)) != NULL) {
5062     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5063       result = false;
5064     }
5065   }
5066   closedir(dir);
5067   return result;
5068 }
5069 
5070 // This code originates from JDK's sysOpen and open64_w
5071 // from src/solaris/hpi/src/system_md.c
5072 
5073 int os::open(const char *path, int oflag, int mode) {
5074   if (strlen(path) > MAX_PATH - 1) {
5075     errno = ENAMETOOLONG;
5076     return -1;
5077   }
5078 
5079   // All file descriptors that are opened in the Java process and not
5080   // specifically destined for a subprocess should have the close-on-exec
5081   // flag set.  If we don't set it, then careless 3rd party native code
5082   // might fork and exec without closing all appropriate file descriptors
5083   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5084   // turn might:
5085   //
5086   // - cause end-of-file to fail to be detected on some file
5087   //   descriptors, resulting in mysterious hangs, or
5088   //
5089   // - might cause an fopen in the subprocess to fail on a system
5090   //   suffering from bug 1085341.
5091   //
5092   // (Yes, the default setting of the close-on-exec flag is a Unix
5093   // design flaw)
5094   //
5095   // See:
5096   // 1085341: 32-bit stdio routines should support file descriptors >255
5097   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5098   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5099   //
5100   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5101   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5102   // because it saves a system call and removes a small window where the flag
5103   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5104   // and we fall back to using FD_CLOEXEC (see below).
5105 #ifdef O_CLOEXEC
5106   oflag |= O_CLOEXEC;
5107 #endif
5108 
5109   int fd = ::open64(path, oflag, mode);
5110   if (fd == -1) return -1;
5111 
5112   //If the open succeeded, the file might still be a directory
5113   {
5114     struct stat64 buf64;
5115     int ret = ::fstat64(fd, &buf64);
5116     int st_mode = buf64.st_mode;
5117 
5118     if (ret != -1) {
5119       if ((st_mode & S_IFMT) == S_IFDIR) {
5120         errno = EISDIR;
5121         ::close(fd);
5122         return -1;
5123       }
5124     } else {
5125       ::close(fd);
5126       return -1;
5127     }
5128   }
5129 
5130 #ifdef FD_CLOEXEC
5131   // Validate that the use of the O_CLOEXEC flag on open above worked.
5132   // With recent kernels, we will perform this check exactly once.
5133   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5134   if (!O_CLOEXEC_is_known_to_work) {
5135     int flags = ::fcntl(fd, F_GETFD);
5136     if (flags != -1) {
5137       if ((flags & FD_CLOEXEC) != 0)
5138         O_CLOEXEC_is_known_to_work = 1;
5139       else
5140         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5141     }
5142   }
5143 #endif
5144 
5145   return fd;
5146 }
5147 
5148 
5149 // create binary file, rewriting existing file if required
5150 int os::create_binary_file(const char* path, bool rewrite_existing) {
5151   int oflags = O_WRONLY | O_CREAT;
5152   if (!rewrite_existing) {
5153     oflags |= O_EXCL;
5154   }
5155   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5156 }
5157 
5158 // return current position of file pointer
5159 jlong os::current_file_offset(int fd) {
5160   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5161 }
5162 
5163 // move file pointer to the specified offset
5164 jlong os::seek_to_file_offset(int fd, jlong offset) {
5165   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5166 }
5167 
5168 // This code originates from JDK's sysAvailable
5169 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5170 
5171 int os::available(int fd, jlong *bytes) {
5172   jlong cur, end;
5173   int mode;
5174   struct stat64 buf64;
5175 
5176   if (::fstat64(fd, &buf64) >= 0) {
5177     mode = buf64.st_mode;
5178     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5179       int n;
5180       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5181         *bytes = n;
5182         return 1;
5183       }
5184     }
5185   }
5186   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5187     return 0;
5188   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5189     return 0;
5190   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5191     return 0;
5192   }
5193   *bytes = end - cur;
5194   return 1;
5195 }
5196 
5197 // Map a block of memory.
5198 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5199                         char *addr, size_t bytes, bool read_only,
5200                         bool allow_exec) {
5201   int prot;
5202   int flags = MAP_PRIVATE;
5203 
5204   if (read_only) {
5205     prot = PROT_READ;
5206   } else {
5207     prot = PROT_READ | PROT_WRITE;
5208   }
5209 
5210   if (allow_exec) {
5211     prot |= PROT_EXEC;
5212   }
5213 
5214   if (addr != NULL) {
5215     flags |= MAP_FIXED;
5216   }
5217 
5218   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5219                                      fd, file_offset);
5220   if (mapped_address == MAP_FAILED) {
5221     return NULL;
5222   }
5223   return mapped_address;
5224 }
5225 
5226 
5227 // Remap a block of memory.
5228 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5229                           char *addr, size_t bytes, bool read_only,
5230                           bool allow_exec) {
5231   // same as map_memory() on this OS
5232   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5233                         allow_exec);
5234 }
5235 
5236 
5237 // Unmap a block of memory.
5238 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5239   return munmap(addr, bytes) == 0;
5240 }
5241 
5242 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5243 
5244 static clockid_t thread_cpu_clockid(Thread* thread) {
5245   pthread_t tid = thread->osthread()->pthread_id();
5246   clockid_t clockid;
5247 
5248   // Get thread clockid
5249   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5250   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5251   return clockid;
5252 }
5253 
5254 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5255 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5256 // of a thread.
5257 //
5258 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5259 // the fast estimate available on the platform.
5260 
5261 jlong os::current_thread_cpu_time() {
5262   if (os::Linux::supports_fast_thread_cpu_time()) {
5263     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5264   } else {
5265     // return user + sys since the cost is the same
5266     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5267   }
5268 }
5269 
5270 jlong os::thread_cpu_time(Thread* thread) {
5271   // consistent with what current_thread_cpu_time() returns
5272   if (os::Linux::supports_fast_thread_cpu_time()) {
5273     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5274   } else {
5275     return slow_thread_cpu_time(thread, true /* user + sys */);
5276   }
5277 }
5278 
5279 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5280   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5281     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5282   } else {
5283     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5284   }
5285 }
5286 
5287 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5288   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5289     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5290   } else {
5291     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5292   }
5293 }
5294 
5295 //  -1 on error.
5296 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5297   pid_t  tid = thread->osthread()->thread_id();
5298   char *s;
5299   char stat[2048];
5300   int statlen;
5301   char proc_name[64];
5302   int count;
5303   long sys_time, user_time;
5304   char cdummy;
5305   int idummy;
5306   long ldummy;
5307   FILE *fp;
5308 
5309   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5310   fp = fopen(proc_name, "r");
5311   if (fp == NULL) return -1;
5312   statlen = fread(stat, 1, 2047, fp);
5313   stat[statlen] = '\0';
5314   fclose(fp);
5315 
5316   // Skip pid and the command string. Note that we could be dealing with
5317   // weird command names, e.g. user could decide to rename java launcher
5318   // to "java 1.4.2 :)", then the stat file would look like
5319   //                1234 (java 1.4.2 :)) R ... ...
5320   // We don't really need to know the command string, just find the last
5321   // occurrence of ")" and then start parsing from there. See bug 4726580.
5322   s = strrchr(stat, ')');
5323   if (s == NULL) return -1;
5324 
5325   // Skip blank chars
5326   do { s++; } while (s && isspace(*s));
5327 
5328   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5329                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5330                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5331                  &user_time, &sys_time);
5332   if (count != 13) return -1;
5333   if (user_sys_cpu_time) {
5334     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5335   } else {
5336     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5337   }
5338 }
5339 
5340 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5341   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5342   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5343   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5344   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5345 }
5346 
5347 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5348   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5349   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5350   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5351   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5352 }
5353 
5354 bool os::is_thread_cpu_time_supported() {
5355   return true;
5356 }
5357 
5358 // System loadavg support.  Returns -1 if load average cannot be obtained.
5359 // Linux doesn't yet have a (official) notion of processor sets,
5360 // so just return the system wide load average.
5361 int os::loadavg(double loadavg[], int nelem) {
5362   return ::getloadavg(loadavg, nelem);
5363 }
5364 
5365 void os::pause() {
5366   char filename[MAX_PATH];
5367   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5368     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5369   } else {
5370     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5371   }
5372 
5373   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5374   if (fd != -1) {
5375     struct stat buf;
5376     ::close(fd);
5377     while (::stat(filename, &buf) == 0) {
5378       (void)::poll(NULL, 0, 100);
5379     }
5380   } else {
5381     jio_fprintf(stderr,
5382                 "Could not open pause file '%s', continuing immediately.\n", filename);
5383   }
5384 }
5385 
5386 
5387 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5388 // comment paragraphs are worth repeating here:
5389 //
5390 // Assumption:
5391 //    Only one parker can exist on an event, which is why we allocate
5392 //    them per-thread. Multiple unparkers can coexist.
5393 //
5394 // _Event serves as a restricted-range semaphore.
5395 //   -1 : thread is blocked, i.e. there is a waiter
5396 //    0 : neutral: thread is running or ready,
5397 //        could have been signaled after a wait started
5398 //    1 : signaled - thread is running or ready
5399 //
5400 
5401 // utility to compute the abstime argument to timedwait:
5402 // millis is the relative timeout time
5403 // abstime will be the absolute timeout time
5404 // TODO: replace compute_abstime() with unpackTime()
5405 
5406 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5407   if (millis < 0)  millis = 0;
5408 
5409   jlong seconds = millis / 1000;
5410   millis %= 1000;
5411   if (seconds > 50000000) { // see man cond_timedwait(3T)
5412     seconds = 50000000;
5413   }
5414 
5415   if (os::supports_monotonic_clock()) {
5416     struct timespec now;
5417     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5418     assert_status(status == 0, status, "clock_gettime");
5419     abstime->tv_sec = now.tv_sec  + seconds;
5420     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5421     if (nanos >= NANOSECS_PER_SEC) {
5422       abstime->tv_sec += 1;
5423       nanos -= NANOSECS_PER_SEC;
5424     }
5425     abstime->tv_nsec = nanos;
5426   } else {
5427     struct timeval now;
5428     int status = gettimeofday(&now, NULL);
5429     assert(status == 0, "gettimeofday");
5430     abstime->tv_sec = now.tv_sec  + seconds;
5431     long usec = now.tv_usec + millis * 1000;
5432     if (usec >= 1000000) {
5433       abstime->tv_sec += 1;
5434       usec -= 1000000;
5435     }
5436     abstime->tv_nsec = usec * 1000;
5437   }
5438   return abstime;
5439 }
5440 
5441 void os::PlatformEvent::park() {       // AKA "down()"
5442   // Transitions for _Event:
5443   //   -1 => -1 : illegal
5444   //    1 =>  0 : pass - return immediately
5445   //    0 => -1 : block; then set _Event to 0 before returning
5446 
5447   // Invariant: Only the thread associated with the Event/PlatformEvent
5448   // may call park().
5449   // TODO: assert that _Assoc != NULL or _Assoc == Self
5450   assert(_nParked == 0, "invariant");
5451 
5452   int v;
5453   for (;;) {
5454     v = _Event;
5455     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5456   }
5457   guarantee(v >= 0, "invariant");
5458   if (v == 0) {
5459     // Do this the hard way by blocking ...
5460     int status = pthread_mutex_lock(_mutex);
5461     assert_status(status == 0, status, "mutex_lock");
5462     guarantee(_nParked == 0, "invariant");
5463     ++_nParked;
5464     while (_Event < 0) {
5465       status = pthread_cond_wait(_cond, _mutex);
5466       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5467       // Treat this the same as if the wait was interrupted
5468       if (status == ETIME) { status = EINTR; }
5469       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5470     }
5471     --_nParked;
5472 
5473     _Event = 0;
5474     status = pthread_mutex_unlock(_mutex);
5475     assert_status(status == 0, status, "mutex_unlock");
5476     // Paranoia to ensure our locked and lock-free paths interact
5477     // correctly with each other.
5478     OrderAccess::fence();
5479   }
5480   guarantee(_Event >= 0, "invariant");
5481 }
5482 
5483 int os::PlatformEvent::park(jlong millis) {
5484   // Transitions for _Event:
5485   //   -1 => -1 : illegal
5486   //    1 =>  0 : pass - return immediately
5487   //    0 => -1 : block; then set _Event to 0 before returning
5488 
5489   guarantee(_nParked == 0, "invariant");
5490 
5491   int v;
5492   for (;;) {
5493     v = _Event;
5494     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5495   }
5496   guarantee(v >= 0, "invariant");
5497   if (v != 0) return OS_OK;
5498 
5499   // We do this the hard way, by blocking the thread.
5500   // Consider enforcing a minimum timeout value.
5501   struct timespec abst;
5502   compute_abstime(&abst, millis);
5503 
5504   int ret = OS_TIMEOUT;
5505   int status = pthread_mutex_lock(_mutex);
5506   assert_status(status == 0, status, "mutex_lock");
5507   guarantee(_nParked == 0, "invariant");
5508   ++_nParked;
5509 
5510   // Object.wait(timo) will return because of
5511   // (a) notification
5512   // (b) timeout
5513   // (c) thread.interrupt
5514   //
5515   // Thread.interrupt and object.notify{All} both call Event::set.
5516   // That is, we treat thread.interrupt as a special case of notification.
5517   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5518   // We assume all ETIME returns are valid.
5519   //
5520   // TODO: properly differentiate simultaneous notify+interrupt.
5521   // In that case, we should propagate the notify to another waiter.
5522 
5523   while (_Event < 0) {
5524     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5525     assert_status(status == 0 || status == EINTR ||
5526                   status == ETIME || status == ETIMEDOUT,
5527                   status, "cond_timedwait");
5528     if (!FilterSpuriousWakeups) break;                 // previous semantics
5529     if (status == ETIME || status == ETIMEDOUT) break;
5530     // We consume and ignore EINTR and spurious wakeups.
5531   }
5532   --_nParked;
5533   if (_Event >= 0) {
5534     ret = OS_OK;
5535   }
5536   _Event = 0;
5537   status = pthread_mutex_unlock(_mutex);
5538   assert_status(status == 0, status, "mutex_unlock");
5539   assert(_nParked == 0, "invariant");
5540   // Paranoia to ensure our locked and lock-free paths interact
5541   // correctly with each other.
5542   OrderAccess::fence();
5543   return ret;
5544 }
5545 
5546 void os::PlatformEvent::unpark() {
5547   // Transitions for _Event:
5548   //    0 => 1 : just return
5549   //    1 => 1 : just return
5550   //   -1 => either 0 or 1; must signal target thread
5551   //         That is, we can safely transition _Event from -1 to either
5552   //         0 or 1.
5553   // See also: "Semaphores in Plan 9" by Mullender & Cox
5554   //
5555   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5556   // that it will take two back-to-back park() calls for the owning
5557   // thread to block. This has the benefit of forcing a spurious return
5558   // from the first park() call after an unpark() call which will help
5559   // shake out uses of park() and unpark() without condition variables.
5560 
5561   if (Atomic::xchg(1, &_Event) >= 0) return;
5562 
5563   // Wait for the thread associated with the event to vacate
5564   int status = pthread_mutex_lock(_mutex);
5565   assert_status(status == 0, status, "mutex_lock");
5566   int AnyWaiters = _nParked;
5567   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5568   status = pthread_mutex_unlock(_mutex);
5569   assert_status(status == 0, status, "mutex_unlock");
5570   if (AnyWaiters != 0) {
5571     // Note that we signal() *after* dropping the lock for "immortal" Events.
5572     // This is safe and avoids a common class of  futile wakeups.  In rare
5573     // circumstances this can cause a thread to return prematurely from
5574     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5575     // will simply re-test the condition and re-park itself.
5576     // This provides particular benefit if the underlying platform does not
5577     // provide wait morphing.
5578     status = pthread_cond_signal(_cond);
5579     assert_status(status == 0, status, "cond_signal");
5580   }
5581 }
5582 
5583 
5584 // JSR166
5585 // -------------------------------------------------------
5586 
5587 // The solaris and linux implementations of park/unpark are fairly
5588 // conservative for now, but can be improved. They currently use a
5589 // mutex/condvar pair, plus a a count.
5590 // Park decrements count if > 0, else does a condvar wait.  Unpark
5591 // sets count to 1 and signals condvar.  Only one thread ever waits
5592 // on the condvar. Contention seen when trying to park implies that someone
5593 // is unparking you, so don't wait. And spurious returns are fine, so there
5594 // is no need to track notifications.
5595 
5596 // This code is common to linux and solaris and will be moved to a
5597 // common place in dolphin.
5598 //
5599 // The passed in time value is either a relative time in nanoseconds
5600 // or an absolute time in milliseconds. Either way it has to be unpacked
5601 // into suitable seconds and nanoseconds components and stored in the
5602 // given timespec structure.
5603 // Given time is a 64-bit value and the time_t used in the timespec is only
5604 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5605 // overflow if times way in the future are given. Further on Solaris versions
5606 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5607 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5608 // As it will be 28 years before "now + 100000000" will overflow we can
5609 // ignore overflow and just impose a hard-limit on seconds using the value
5610 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5611 // years from "now".
5612 
5613 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5614   assert(time > 0, "convertTime");
5615   time_t max_secs = 0;
5616 
5617   if (!os::supports_monotonic_clock() || isAbsolute) {
5618     struct timeval now;
5619     int status = gettimeofday(&now, NULL);
5620     assert(status == 0, "gettimeofday");
5621 
5622     max_secs = now.tv_sec + MAX_SECS;
5623 
5624     if (isAbsolute) {
5625       jlong secs = time / 1000;
5626       if (secs > max_secs) {
5627         absTime->tv_sec = max_secs;
5628       } else {
5629         absTime->tv_sec = secs;
5630       }
5631       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5632     } else {
5633       jlong secs = time / NANOSECS_PER_SEC;
5634       if (secs >= MAX_SECS) {
5635         absTime->tv_sec = max_secs;
5636         absTime->tv_nsec = 0;
5637       } else {
5638         absTime->tv_sec = now.tv_sec + secs;
5639         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5640         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5641           absTime->tv_nsec -= NANOSECS_PER_SEC;
5642           ++absTime->tv_sec; // note: this must be <= max_secs
5643         }
5644       }
5645     }
5646   } else {
5647     // must be relative using monotonic clock
5648     struct timespec now;
5649     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5650     assert_status(status == 0, status, "clock_gettime");
5651     max_secs = now.tv_sec + MAX_SECS;
5652     jlong secs = time / NANOSECS_PER_SEC;
5653     if (secs >= MAX_SECS) {
5654       absTime->tv_sec = max_secs;
5655       absTime->tv_nsec = 0;
5656     } else {
5657       absTime->tv_sec = now.tv_sec + secs;
5658       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5659       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5660         absTime->tv_nsec -= NANOSECS_PER_SEC;
5661         ++absTime->tv_sec; // note: this must be <= max_secs
5662       }
5663     }
5664   }
5665   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5666   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5667   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5668   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5669 }
5670 
5671 void Parker::park(bool isAbsolute, jlong time) {
5672   // Ideally we'd do something useful while spinning, such
5673   // as calling unpackTime().
5674 
5675   // Optional fast-path check:
5676   // Return immediately if a permit is available.
5677   // We depend on Atomic::xchg() having full barrier semantics
5678   // since we are doing a lock-free update to _counter.
5679   if (Atomic::xchg(0, &_counter) > 0) return;
5680 
5681   Thread* thread = Thread::current();
5682   assert(thread->is_Java_thread(), "Must be JavaThread");
5683   JavaThread *jt = (JavaThread *)thread;
5684 
5685   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5686   // Check interrupt before trying to wait
5687   if (Thread::is_interrupted(thread, false)) {
5688     return;
5689   }
5690 
5691   // Next, demultiplex/decode time arguments
5692   timespec absTime;
5693   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5694     return;
5695   }
5696   if (time > 0) {
5697     unpackTime(&absTime, isAbsolute, time);
5698   }
5699 
5700 
5701   // Enter safepoint region
5702   // Beware of deadlocks such as 6317397.
5703   // The per-thread Parker:: mutex is a classic leaf-lock.
5704   // In particular a thread must never block on the Threads_lock while
5705   // holding the Parker:: mutex.  If safepoints are pending both the
5706   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5707   ThreadBlockInVM tbivm(jt);
5708 
5709   // Don't wait if cannot get lock since interference arises from
5710   // unblocking.  Also. check interrupt before trying wait
5711   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5712     return;
5713   }
5714 
5715   int status;
5716   if (_counter > 0)  { // no wait needed
5717     _counter = 0;
5718     status = pthread_mutex_unlock(_mutex);
5719     assert_status(status == 0, status, "invariant");
5720     // Paranoia to ensure our locked and lock-free paths interact
5721     // correctly with each other and Java-level accesses.
5722     OrderAccess::fence();
5723     return;
5724   }
5725 
5726 #ifdef ASSERT
5727   // Don't catch signals while blocked; let the running threads have the signals.
5728   // (This allows a debugger to break into the running thread.)
5729   sigset_t oldsigs;
5730   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5731   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5732 #endif
5733 
5734   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5735   jt->set_suspend_equivalent();
5736   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5737 
5738   assert(_cur_index == -1, "invariant");
5739   if (time == 0) {
5740     _cur_index = REL_INDEX; // arbitrary choice when not timed
5741     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5742   } else {
5743     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5744     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5745   }
5746   _cur_index = -1;
5747   assert_status(status == 0 || status == EINTR ||
5748                 status == ETIME || status == ETIMEDOUT,
5749                 status, "cond_timedwait");
5750 
5751 #ifdef ASSERT
5752   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5753 #endif
5754 
5755   _counter = 0;
5756   status = pthread_mutex_unlock(_mutex);
5757   assert_status(status == 0, status, "invariant");
5758   // Paranoia to ensure our locked and lock-free paths interact
5759   // correctly with each other and Java-level accesses.
5760   OrderAccess::fence();
5761 
5762   // If externally suspended while waiting, re-suspend
5763   if (jt->handle_special_suspend_equivalent_condition()) {
5764     jt->java_suspend_self();
5765   }
5766 }
5767 
5768 void Parker::unpark() {
5769   int status = pthread_mutex_lock(_mutex);
5770   assert_status(status == 0, status, "invariant");
5771   const int s = _counter;
5772   _counter = 1;
5773   // must capture correct index before unlocking
5774   int index = _cur_index;
5775   status = pthread_mutex_unlock(_mutex);
5776   assert_status(status == 0, status, "invariant");
5777   if (s < 1 && index != -1) {
5778     // thread is definitely parked
5779     status = pthread_cond_signal(&_cond[index]);
5780     assert_status(status == 0, status, "invariant");
5781   }
5782 }
5783 
5784 
5785 extern char** environ;
5786 
5787 // Run the specified command in a separate process. Return its exit value,
5788 // or -1 on failure (e.g. can't fork a new process).
5789 // Unlike system(), this function can be called from signal handler. It
5790 // doesn't block SIGINT et al.
5791 int os::fork_and_exec(char* cmd) {
5792   const char * argv[4] = {"sh", "-c", cmd, NULL};
5793 
5794   pid_t pid = fork();
5795 
5796   if (pid < 0) {
5797     // fork failed
5798     return -1;
5799 
5800   } else if (pid == 0) {
5801     // child process
5802 
5803     execve("/bin/sh", (char* const*)argv, environ);
5804 
5805     // execve failed
5806     _exit(-1);
5807 
5808   } else  {
5809     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5810     // care about the actual exit code, for now.
5811 
5812     int status;
5813 
5814     // Wait for the child process to exit.  This returns immediately if
5815     // the child has already exited. */
5816     while (waitpid(pid, &status, 0) < 0) {
5817       switch (errno) {
5818       case ECHILD: return 0;
5819       case EINTR: break;
5820       default: return -1;
5821       }
5822     }
5823 
5824     if (WIFEXITED(status)) {
5825       // The child exited normally; get its exit code.
5826       return WEXITSTATUS(status);
5827     } else if (WIFSIGNALED(status)) {
5828       // The child exited because of a signal
5829       // The best value to return is 0x80 + signal number,
5830       // because that is what all Unix shells do, and because
5831       // it allows callers to distinguish between process exit and
5832       // process death by signal.
5833       return 0x80 + WTERMSIG(status);
5834     } else {
5835       // Unknown exit code; pass it through
5836       return status;
5837     }
5838   }
5839 }
5840 
5841 // is_headless_jre()
5842 //
5843 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5844 // in order to report if we are running in a headless jre
5845 //
5846 // Since JDK8 xawt/libmawt.so was moved into the same directory
5847 // as libawt.so, and renamed libawt_xawt.so
5848 //
5849 bool os::is_headless_jre() {
5850   struct stat statbuf;
5851   char buf[MAXPATHLEN];
5852   char libmawtpath[MAXPATHLEN];
5853   const char *xawtstr  = "/xawt/libmawt.so";
5854   const char *new_xawtstr = "/libawt_xawt.so";
5855   char *p;
5856 
5857   // Get path to libjvm.so
5858   os::jvm_path(buf, sizeof(buf));
5859 
5860   // Get rid of libjvm.so
5861   p = strrchr(buf, '/');
5862   if (p == NULL) {
5863     return false;
5864   } else {
5865     *p = '\0';
5866   }
5867 
5868   // Get rid of client or server
5869   p = strrchr(buf, '/');
5870   if (p == NULL) {
5871     return false;
5872   } else {
5873     *p = '\0';
5874   }
5875 
5876   // check xawt/libmawt.so
5877   strcpy(libmawtpath, buf);
5878   strcat(libmawtpath, xawtstr);
5879   if (::stat(libmawtpath, &statbuf) == 0) return false;
5880 
5881   // check libawt_xawt.so
5882   strcpy(libmawtpath, buf);
5883   strcat(libmawtpath, new_xawtstr);
5884   if (::stat(libmawtpath, &statbuf) == 0) return false;
5885 
5886   return true;
5887 }
5888 
5889 // Get the default path to the core file
5890 // Returns the length of the string
5891 int os::get_core_path(char* buffer, size_t bufferSize) {
5892   /*
5893    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5894    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5895    */
5896   const int core_pattern_len = 129;
5897   char core_pattern[core_pattern_len] = {0};
5898 
5899   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5900   if (core_pattern_file == -1) {
5901     return -1;
5902   }
5903 
5904   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5905   ::close(core_pattern_file);
5906   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5907     return -1;
5908   }
5909   if (core_pattern[ret-1] == '\n') {
5910     core_pattern[ret-1] = '\0';
5911   } else {
5912     core_pattern[ret] = '\0';
5913   }
5914 
5915   char *pid_pos = strstr(core_pattern, "%p");
5916   int written;
5917 
5918   if (core_pattern[0] == '/') {
5919     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5920   } else {
5921     char cwd[PATH_MAX];
5922 
5923     const char* p = get_current_directory(cwd, PATH_MAX);
5924     if (p == NULL) {
5925       return -1;
5926     }
5927 
5928     if (core_pattern[0] == '|') {
5929       written = jio_snprintf(buffer, bufferSize,
5930                         "\"%s\" (or dumping to %s/core.%d)",
5931                                      &core_pattern[1], p, current_process_id());
5932     } else {
5933       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5934     }
5935   }
5936 
5937   if (written < 0) {
5938     return -1;
5939   }
5940 
5941   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5942     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5943 
5944     if (core_uses_pid_file != -1) {
5945       char core_uses_pid = 0;
5946       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5947       ::close(core_uses_pid_file);
5948 
5949       if (core_uses_pid == '1') {
5950         jio_snprintf(buffer + written, bufferSize - written,
5951                                           ".%d", current_process_id());
5952       }
5953     }
5954   }
5955 
5956   return strlen(buffer);
5957 }
5958 
5959 bool os::start_debugging(char *buf, int buflen) {
5960   int len = (int)strlen(buf);
5961   char *p = &buf[len];
5962 
5963   jio_snprintf(p, buflen-len,
5964              "\n\n"
5965              "Do you want to debug the problem?\n\n"
5966              "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5967              "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5968              "Otherwise, press RETURN to abort...",
5969              os::current_process_id(), os::current_process_id(),
5970              os::current_thread_id(), os::current_thread_id());
5971 
5972   bool yes = os::message_box("Unexpected Error", buf);
5973 
5974   if (yes) {
5975     // yes, user asked VM to launch debugger
5976     jio_snprintf(buf, sizeof(buf), "gdb /proc/%d/exe %d",
5977                      os::current_process_id(), os::current_process_id());
5978 
5979     os::fork_and_exec(buf);
5980     yes = false;
5981   }
5982   return yes;
5983 }
5984 
5985 
5986 
5987 /////////////// Unit tests ///////////////
5988 
5989 #ifndef PRODUCT
5990 
5991 #define test_log(...)              \
5992   do {                             \
5993     if (VerboseInternalVMTests) {  \
5994       tty->print_cr(__VA_ARGS__);  \
5995       tty->flush();                \
5996     }                              \
5997   } while (false)
5998 
5999 class TestReserveMemorySpecial : AllStatic {
6000  public:
6001   static void small_page_write(void* addr, size_t size) {
6002     size_t page_size = os::vm_page_size();
6003 
6004     char* end = (char*)addr + size;
6005     for (char* p = (char*)addr; p < end; p += page_size) {
6006       *p = 1;
6007     }
6008   }
6009 
6010   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6011     if (!UseHugeTLBFS) {
6012       return;
6013     }
6014 
6015     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6016 
6017     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6018 
6019     if (addr != NULL) {
6020       small_page_write(addr, size);
6021 
6022       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6023     }
6024   }
6025 
6026   static void test_reserve_memory_special_huge_tlbfs_only() {
6027     if (!UseHugeTLBFS) {
6028       return;
6029     }
6030 
6031     size_t lp = os::large_page_size();
6032 
6033     for (size_t size = lp; size <= lp * 10; size += lp) {
6034       test_reserve_memory_special_huge_tlbfs_only(size);
6035     }
6036   }
6037 
6038   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6039     size_t lp = os::large_page_size();
6040     size_t ag = os::vm_allocation_granularity();
6041 
6042     // sizes to test
6043     const size_t sizes[] = {
6044       lp, lp + ag, lp + lp / 2, lp * 2,
6045       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6046       lp * 10, lp * 10 + lp / 2
6047     };
6048     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6049 
6050     // For each size/alignment combination, we test three scenarios:
6051     // 1) with req_addr == NULL
6052     // 2) with a non-null req_addr at which we expect to successfully allocate
6053     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6054     //    expect the allocation to either fail or to ignore req_addr
6055 
6056     // Pre-allocate two areas; they shall be as large as the largest allocation
6057     //  and aligned to the largest alignment we will be testing.
6058     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6059     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6060       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6061       -1, 0);
6062     assert(mapping1 != MAP_FAILED, "should work");
6063 
6064     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6065       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6066       -1, 0);
6067     assert(mapping2 != MAP_FAILED, "should work");
6068 
6069     // Unmap the first mapping, but leave the second mapping intact: the first
6070     // mapping will serve as a value for a "good" req_addr (case 2). The second
6071     // mapping, still intact, as "bad" req_addr (case 3).
6072     ::munmap(mapping1, mapping_size);
6073 
6074     // Case 1
6075     test_log("%s, req_addr NULL:", __FUNCTION__);
6076     test_log("size            align           result");
6077 
6078     for (int i = 0; i < num_sizes; i++) {
6079       const size_t size = sizes[i];
6080       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6081         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6082         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6083                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6084         if (p != NULL) {
6085           assert(is_ptr_aligned(p, alignment), "must be");
6086           small_page_write(p, size);
6087           os::Linux::release_memory_special_huge_tlbfs(p, size);
6088         }
6089       }
6090     }
6091 
6092     // Case 2
6093     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6094     test_log("size            align           req_addr         result");
6095 
6096     for (int i = 0; i < num_sizes; i++) {
6097       const size_t size = sizes[i];
6098       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6099         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6100         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6101         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6102                  size, alignment, p2i(req_addr), p2i(p),
6103                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6104         if (p != NULL) {
6105           assert(p == req_addr, "must be");
6106           small_page_write(p, size);
6107           os::Linux::release_memory_special_huge_tlbfs(p, size);
6108         }
6109       }
6110     }
6111 
6112     // Case 3
6113     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6114     test_log("size            align           req_addr         result");
6115 
6116     for (int i = 0; i < num_sizes; i++) {
6117       const size_t size = sizes[i];
6118       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6119         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6120         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6121         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6122                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6123         // as the area around req_addr contains already existing mappings, the API should always
6124         // return NULL (as per contract, it cannot return another address)
6125         assert(p == NULL, "must be");
6126       }
6127     }
6128 
6129     ::munmap(mapping2, mapping_size);
6130 
6131   }
6132 
6133   static void test_reserve_memory_special_huge_tlbfs() {
6134     if (!UseHugeTLBFS) {
6135       return;
6136     }
6137 
6138     test_reserve_memory_special_huge_tlbfs_only();
6139     test_reserve_memory_special_huge_tlbfs_mixed();
6140   }
6141 
6142   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6143     if (!UseSHM) {
6144       return;
6145     }
6146 
6147     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6148 
6149     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6150 
6151     if (addr != NULL) {
6152       assert(is_ptr_aligned(addr, alignment), "Check");
6153       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6154 
6155       small_page_write(addr, size);
6156 
6157       os::Linux::release_memory_special_shm(addr, size);
6158     }
6159   }
6160 
6161   static void test_reserve_memory_special_shm() {
6162     size_t lp = os::large_page_size();
6163     size_t ag = os::vm_allocation_granularity();
6164 
6165     for (size_t size = ag; size < lp * 3; size += ag) {
6166       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6167         test_reserve_memory_special_shm(size, alignment);
6168       }
6169     }
6170   }
6171 
6172   static void test() {
6173     test_reserve_memory_special_huge_tlbfs();
6174     test_reserve_memory_special_shm();
6175   }
6176 };
6177 
6178 void TestReserveMemorySpecial_test() {
6179   TestReserveMemorySpecial::test();
6180 }
6181 
6182 #endif