1 /*
   2  * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_linux.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "prims/jniFastGetField.hpp"
  42 #include "prims/jvm.h"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.hpp"
  49 #include "runtime/init.hpp"
  50 #include "runtime/java.hpp"
  51 #include "runtime/javaCalls.hpp"
  52 #include "runtime/mutexLocker.hpp"
  53 #include "runtime/objectMonitor.hpp"
  54 #include "runtime/orderAccess.inline.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/perfMemory.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/statSampler.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "runtime/thread.inline.hpp"
  61 #include "runtime/threadCritical.hpp"
  62 #include "runtime/timer.hpp"
  63 #include "semaphore_posix.hpp"
  64 #include "services/attachListener.hpp"
  65 #include "services/memTracker.hpp"
  66 #include "services/runtimeService.hpp"
  67 #include "utilities/decoder.hpp"
  68 #include "utilities/defaultStream.hpp"
  69 #include "utilities/events.hpp"
  70 #include "utilities/elfFile.hpp"
  71 #include "utilities/growableArray.hpp"
  72 #include "utilities/macros.hpp"
  73 #include "utilities/vmError.hpp"
  74 
  75 // put OS-includes here
  76 # include <sys/types.h>
  77 # include <sys/mman.h>
  78 # include <sys/stat.h>
  79 # include <sys/select.h>
  80 # include <pthread.h>
  81 # include <signal.h>
  82 # include <errno.h>
  83 # include <dlfcn.h>
  84 # include <stdio.h>
  85 # include <unistd.h>
  86 # include <sys/resource.h>
  87 # include <pthread.h>
  88 # include <sys/stat.h>
  89 # include <sys/time.h>
  90 # include <sys/times.h>
  91 # include <sys/utsname.h>
  92 # include <sys/socket.h>
  93 # include <sys/wait.h>
  94 # include <pwd.h>
  95 # include <poll.h>
  96 # include <semaphore.h>
  97 # include <fcntl.h>
  98 # include <string.h>
  99 # include <syscall.h>
 100 # include <sys/sysinfo.h>
 101 # include <gnu/libc-version.h>
 102 # include <sys/ipc.h>
 103 # include <sys/shm.h>
 104 # include <link.h>
 105 # include <stdint.h>
 106 # include <inttypes.h>
 107 # include <sys/ioctl.h>
 108 
 109 #ifndef _GNU_SOURCE
 110   #define _GNU_SOURCE
 111   #include <sched.h>
 112   #undef _GNU_SOURCE
 113 #else
 114   #include <sched.h>
 115 #endif
 116 
 117 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 118 // getrusage() is prepared to handle the associated failure.
 119 #ifndef RUSAGE_THREAD
 120   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 121 #endif
 122 
 123 #define MAX_PATH    (2 * K)
 124 
 125 #define MAX_SECS 100000000
 126 
 127 // for timer info max values which include all bits
 128 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 129 
 130 #define LARGEPAGES_BIT (1 << 6)
 131 ////////////////////////////////////////////////////////////////////////////////
 132 // global variables
 133 julong os::Linux::_physical_memory = 0;
 134 
 135 address   os::Linux::_initial_thread_stack_bottom = NULL;
 136 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 137 
 138 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 139 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 140 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 141 Mutex* os::Linux::_createThread_lock = NULL;
 142 pthread_t os::Linux::_main_thread;
 143 int os::Linux::_page_size = -1;
 144 bool os::Linux::_supports_fast_thread_cpu_time = false;
 145 uint32_t os::Linux::_os_version = 0;
 146 const char * os::Linux::_glibc_version = NULL;
 147 const char * os::Linux::_libpthread_version = NULL;
 148 pthread_condattr_t os::Linux::_condattr[1];
 149 
 150 static jlong initial_time_count=0;
 151 
 152 static int clock_tics_per_sec = 100;
 153 
 154 // For diagnostics to print a message once. see run_periodic_checks
 155 static sigset_t check_signal_done;
 156 static bool check_signals = true;
 157 
 158 // Signal number used to suspend/resume a thread
 159 
 160 // do not use any signal number less than SIGSEGV, see 4355769
 161 static int SR_signum = SIGUSR2;
 162 sigset_t SR_sigset;
 163 
 164 // Declarations
 165 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 166 
 167 // utility functions
 168 
 169 static int SR_initialize();
 170 
 171 julong os::available_memory() {
 172   return Linux::available_memory();
 173 }
 174 
 175 julong os::Linux::available_memory() {
 176   // values in struct sysinfo are "unsigned long"
 177   struct sysinfo si;
 178   sysinfo(&si);
 179 
 180   return (julong)si.freeram * si.mem_unit;
 181 }
 182 
 183 julong os::physical_memory() {
 184   return Linux::physical_memory();
 185 }
 186 
 187 // Return true if user is running as root.
 188 
 189 bool os::have_special_privileges() {
 190   static bool init = false;
 191   static bool privileges = false;
 192   if (!init) {
 193     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 194     init = true;
 195   }
 196   return privileges;
 197 }
 198 
 199 
 200 #ifndef SYS_gettid
 201 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 202   #ifdef __ia64__
 203     #define SYS_gettid 1105
 204   #else
 205     #ifdef __i386__
 206       #define SYS_gettid 224
 207     #else
 208       #ifdef __amd64__
 209         #define SYS_gettid 186
 210       #else
 211         #ifdef __sparc__
 212           #define SYS_gettid 143
 213         #else
 214           #error define gettid for the arch
 215         #endif
 216       #endif
 217     #endif
 218   #endif
 219 #endif
 220 
 221 // Cpu architecture string
 222 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 223 
 224 
 225 // pid_t gettid()
 226 //
 227 // Returns the kernel thread id of the currently running thread. Kernel
 228 // thread id is used to access /proc.
 229 pid_t os::Linux::gettid() {
 230   int rslt = syscall(SYS_gettid);
 231   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 232   return (pid_t)rslt;
 233 }
 234 
 235 // Most versions of linux have a bug where the number of processors are
 236 // determined by looking at the /proc file system.  In a chroot environment,
 237 // the system call returns 1.  This causes the VM to act as if it is
 238 // a single processor and elide locking (see is_MP() call).
 239 static bool unsafe_chroot_detected = false;
 240 static const char *unstable_chroot_error = "/proc file system not found.\n"
 241                      "Java may be unstable running multithreaded in a chroot "
 242                      "environment on Linux when /proc filesystem is not mounted.";
 243 
 244 void os::Linux::initialize_system_info() {
 245   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 246   if (processor_count() == 1) {
 247     pid_t pid = os::Linux::gettid();
 248     char fname[32];
 249     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 250     FILE *fp = fopen(fname, "r");
 251     if (fp == NULL) {
 252       unsafe_chroot_detected = true;
 253     } else {
 254       fclose(fp);
 255     }
 256   }
 257   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 258   assert(processor_count() > 0, "linux error");
 259 }
 260 
 261 void os::init_system_properties_values() {
 262   // The next steps are taken in the product version:
 263   //
 264   // Obtain the JAVA_HOME value from the location of libjvm.so.
 265   // This library should be located at:
 266   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 267   //
 268   // If "/jre/lib/" appears at the right place in the path, then we
 269   // assume libjvm.so is installed in a JDK and we use this path.
 270   //
 271   // Otherwise exit with message: "Could not create the Java virtual machine."
 272   //
 273   // The following extra steps are taken in the debugging version:
 274   //
 275   // If "/jre/lib/" does NOT appear at the right place in the path
 276   // instead of exit check for $JAVA_HOME environment variable.
 277   //
 278   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 279   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 280   // it looks like libjvm.so is installed there
 281   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 282   //
 283   // Otherwise exit.
 284   //
 285   // Important note: if the location of libjvm.so changes this
 286   // code needs to be changed accordingly.
 287 
 288   // See ld(1):
 289   //      The linker uses the following search paths to locate required
 290   //      shared libraries:
 291   //        1: ...
 292   //        ...
 293   //        7: The default directories, normally /lib and /usr/lib.
 294 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
 295   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 296 #else
 297   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 298 #endif
 299 
 300 // Base path of extensions installed on the system.
 301 #define SYS_EXT_DIR     "/usr/java/packages"
 302 #define EXTENSIONS_DIR  "/lib/ext"
 303 
 304   // Buffer that fits several sprintfs.
 305   // Note that the space for the colon and the trailing null are provided
 306   // by the nulls included by the sizeof operator.
 307   const size_t bufsize =
 308     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 309          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 310   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 311 
 312   // sysclasspath, java_home, dll_dir
 313   {
 314     char *pslash;
 315     os::jvm_path(buf, bufsize);
 316 
 317     // Found the full path to libjvm.so.
 318     // Now cut the path to <java_home>/jre if we can.
 319     pslash = strrchr(buf, '/');
 320     if (pslash != NULL) {
 321       *pslash = '\0';            // Get rid of /libjvm.so.
 322     }
 323     pslash = strrchr(buf, '/');
 324     if (pslash != NULL) {
 325       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 326     }
 327     Arguments::set_dll_dir(buf);
 328 
 329     if (pslash != NULL) {
 330       pslash = strrchr(buf, '/');
 331       if (pslash != NULL) {
 332         *pslash = '\0';          // Get rid of /<arch>.
 333         pslash = strrchr(buf, '/');
 334         if (pslash != NULL) {
 335           *pslash = '\0';        // Get rid of /lib.
 336         }
 337       }
 338     }
 339     Arguments::set_java_home(buf);
 340     set_boot_path('/', ':');
 341   }
 342 
 343   // Where to look for native libraries.
 344   //
 345   // Note: Due to a legacy implementation, most of the library path
 346   // is set in the launcher. This was to accomodate linking restrictions
 347   // on legacy Linux implementations (which are no longer supported).
 348   // Eventually, all the library path setting will be done here.
 349   //
 350   // However, to prevent the proliferation of improperly built native
 351   // libraries, the new path component /usr/java/packages is added here.
 352   // Eventually, all the library path setting will be done here.
 353   {
 354     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 355     // should always exist (until the legacy problem cited above is
 356     // addressed).
 357     const char *v = ::getenv("LD_LIBRARY_PATH");
 358     const char *v_colon = ":";
 359     if (v == NULL) { v = ""; v_colon = ""; }
 360     // That's +1 for the colon and +1 for the trailing '\0'.
 361     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 362                                                      strlen(v) + 1 +
 363                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 364                                                      mtInternal);
 365     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 366     Arguments::set_library_path(ld_library_path);
 367     FREE_C_HEAP_ARRAY(char, ld_library_path);
 368   }
 369 
 370   // Extensions directories.
 371   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 372   Arguments::set_ext_dirs(buf);
 373 
 374   FREE_C_HEAP_ARRAY(char, buf);
 375 
 376 #undef DEFAULT_LIBPATH
 377 #undef SYS_EXT_DIR
 378 #undef EXTENSIONS_DIR
 379 }
 380 
 381 ////////////////////////////////////////////////////////////////////////////////
 382 // breakpoint support
 383 
 384 void os::breakpoint() {
 385   BREAKPOINT;
 386 }
 387 
 388 extern "C" void breakpoint() {
 389   // use debugger to set breakpoint here
 390 }
 391 
 392 ////////////////////////////////////////////////////////////////////////////////
 393 // signal support
 394 
 395 debug_only(static bool signal_sets_initialized = false);
 396 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 397 
 398 bool os::Linux::is_sig_ignored(int sig) {
 399   struct sigaction oact;
 400   sigaction(sig, (struct sigaction*)NULL, &oact);
 401   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 402                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 403   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 404     return true;
 405   } else {
 406     return false;
 407   }
 408 }
 409 
 410 void os::Linux::signal_sets_init() {
 411   // Should also have an assertion stating we are still single-threaded.
 412   assert(!signal_sets_initialized, "Already initialized");
 413   // Fill in signals that are necessarily unblocked for all threads in
 414   // the VM. Currently, we unblock the following signals:
 415   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 416   //                         by -Xrs (=ReduceSignalUsage));
 417   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 418   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 419   // the dispositions or masks wrt these signals.
 420   // Programs embedding the VM that want to use the above signals for their
 421   // own purposes must, at this time, use the "-Xrs" option to prevent
 422   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 423   // (See bug 4345157, and other related bugs).
 424   // In reality, though, unblocking these signals is really a nop, since
 425   // these signals are not blocked by default.
 426   sigemptyset(&unblocked_sigs);
 427   sigemptyset(&allowdebug_blocked_sigs);
 428   sigaddset(&unblocked_sigs, SIGILL);
 429   sigaddset(&unblocked_sigs, SIGSEGV);
 430   sigaddset(&unblocked_sigs, SIGBUS);
 431   sigaddset(&unblocked_sigs, SIGFPE);
 432 #if defined(PPC64)
 433   sigaddset(&unblocked_sigs, SIGTRAP);
 434 #endif
 435   sigaddset(&unblocked_sigs, SR_signum);
 436 
 437   if (!ReduceSignalUsage) {
 438     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 439       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 440       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 441     }
 442     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 443       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 444       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 445     }
 446     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 447       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 448       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 449     }
 450   }
 451   // Fill in signals that are blocked by all but the VM thread.
 452   sigemptyset(&vm_sigs);
 453   if (!ReduceSignalUsage) {
 454     sigaddset(&vm_sigs, BREAK_SIGNAL);
 455   }
 456   debug_only(signal_sets_initialized = true);
 457 
 458 }
 459 
 460 // These are signals that are unblocked while a thread is running Java.
 461 // (For some reason, they get blocked by default.)
 462 sigset_t* os::Linux::unblocked_signals() {
 463   assert(signal_sets_initialized, "Not initialized");
 464   return &unblocked_sigs;
 465 }
 466 
 467 // These are the signals that are blocked while a (non-VM) thread is
 468 // running Java. Only the VM thread handles these signals.
 469 sigset_t* os::Linux::vm_signals() {
 470   assert(signal_sets_initialized, "Not initialized");
 471   return &vm_sigs;
 472 }
 473 
 474 // These are signals that are blocked during cond_wait to allow debugger in
 475 sigset_t* os::Linux::allowdebug_blocked_signals() {
 476   assert(signal_sets_initialized, "Not initialized");
 477   return &allowdebug_blocked_sigs;
 478 }
 479 
 480 void os::Linux::hotspot_sigmask(Thread* thread) {
 481 
 482   //Save caller's signal mask before setting VM signal mask
 483   sigset_t caller_sigmask;
 484   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 485 
 486   OSThread* osthread = thread->osthread();
 487   osthread->set_caller_sigmask(caller_sigmask);
 488 
 489   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 490 
 491   if (!ReduceSignalUsage) {
 492     if (thread->is_VM_thread()) {
 493       // Only the VM thread handles BREAK_SIGNAL ...
 494       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 495     } else {
 496       // ... all other threads block BREAK_SIGNAL
 497       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 498     }
 499   }
 500 }
 501 
 502 //////////////////////////////////////////////////////////////////////////////
 503 // detecting pthread library
 504 
 505 void os::Linux::libpthread_init() {
 506   // Save glibc and pthread version strings.
 507 #if !defined(_CS_GNU_LIBC_VERSION) || \
 508     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 509   #error "glibc too old (< 2.3.2)"
 510 #endif
 511 
 512   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 513   assert(n > 0, "cannot retrieve glibc version");
 514   char *str = (char *)malloc(n, mtInternal);
 515   confstr(_CS_GNU_LIBC_VERSION, str, n);
 516   os::Linux::set_glibc_version(str);
 517 
 518   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 519   assert(n > 0, "cannot retrieve pthread version");
 520   str = (char *)malloc(n, mtInternal);
 521   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 522   os::Linux::set_libpthread_version(str);
 523 }
 524 
 525 /////////////////////////////////////////////////////////////////////////////
 526 // thread stack expansion
 527 
 528 // os::Linux::manually_expand_stack() takes care of expanding the thread
 529 // stack. Note that this is normally not needed: pthread stacks allocate
 530 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 531 // committed. Therefore it is not necessary to expand the stack manually.
 532 //
 533 // Manually expanding the stack was historically needed on LinuxThreads
 534 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 535 // it is kept to deal with very rare corner cases:
 536 //
 537 // For one, user may run the VM on an own implementation of threads
 538 // whose stacks are - like the old LinuxThreads - implemented using
 539 // mmap(MAP_GROWSDOWN).
 540 //
 541 // Also, this coding may be needed if the VM is running on the primordial
 542 // thread. Normally we avoid running on the primordial thread; however,
 543 // user may still invoke the VM on the primordial thread.
 544 //
 545 // The following historical comment describes the details about running
 546 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 547 
 548 
 549 // Force Linux kernel to expand current thread stack. If "bottom" is close
 550 // to the stack guard, caller should block all signals.
 551 //
 552 // MAP_GROWSDOWN:
 553 //   A special mmap() flag that is used to implement thread stacks. It tells
 554 //   kernel that the memory region should extend downwards when needed. This
 555 //   allows early versions of LinuxThreads to only mmap the first few pages
 556 //   when creating a new thread. Linux kernel will automatically expand thread
 557 //   stack as needed (on page faults).
 558 //
 559 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 560 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 561 //   region, it's hard to tell if the fault is due to a legitimate stack
 562 //   access or because of reading/writing non-exist memory (e.g. buffer
 563 //   overrun). As a rule, if the fault happens below current stack pointer,
 564 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 565 //   application (see Linux kernel fault.c).
 566 //
 567 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 568 //   stack overflow detection.
 569 //
 570 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 571 //   not use MAP_GROWSDOWN.
 572 //
 573 // To get around the problem and allow stack banging on Linux, we need to
 574 // manually expand thread stack after receiving the SIGSEGV.
 575 //
 576 // There are two ways to expand thread stack to address "bottom", we used
 577 // both of them in JVM before 1.5:
 578 //   1. adjust stack pointer first so that it is below "bottom", and then
 579 //      touch "bottom"
 580 //   2. mmap() the page in question
 581 //
 582 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 583 // if current sp is already near the lower end of page 101, and we need to
 584 // call mmap() to map page 100, it is possible that part of the mmap() frame
 585 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 586 // That will destroy the mmap() frame and cause VM to crash.
 587 //
 588 // The following code works by adjusting sp first, then accessing the "bottom"
 589 // page to force a page fault. Linux kernel will then automatically expand the
 590 // stack mapping.
 591 //
 592 // _expand_stack_to() assumes its frame size is less than page size, which
 593 // should always be true if the function is not inlined.
 594 
 595 static void NOINLINE _expand_stack_to(address bottom) {
 596   address sp;
 597   size_t size;
 598   volatile char *p;
 599 
 600   // Adjust bottom to point to the largest address within the same page, it
 601   // gives us a one-page buffer if alloca() allocates slightly more memory.
 602   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 603   bottom += os::Linux::page_size() - 1;
 604 
 605   // sp might be slightly above current stack pointer; if that's the case, we
 606   // will alloca() a little more space than necessary, which is OK. Don't use
 607   // os::current_stack_pointer(), as its result can be slightly below current
 608   // stack pointer, causing us to not alloca enough to reach "bottom".
 609   sp = (address)&sp;
 610 
 611   if (sp > bottom) {
 612     size = sp - bottom;
 613     p = (volatile char *)alloca(size);
 614     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 615     p[0] = '\0';
 616   }
 617 }
 618 
 619 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 620   assert(t!=NULL, "just checking");
 621   assert(t->osthread()->expanding_stack(), "expand should be set");
 622   assert(t->stack_base() != NULL, "stack_base was not initialized");
 623 
 624   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 625     sigset_t mask_all, old_sigset;
 626     sigfillset(&mask_all);
 627     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 628     _expand_stack_to(addr);
 629     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 630     return true;
 631   }
 632   return false;
 633 }
 634 
 635 //////////////////////////////////////////////////////////////////////////////
 636 // create new thread
 637 
 638 // Thread start routine for all newly created threads
 639 static void *thread_native_entry(Thread *thread) {
 640   // Try to randomize the cache line index of hot stack frames.
 641   // This helps when threads of the same stack traces evict each other's
 642   // cache lines. The threads can be either from the same JVM instance, or
 643   // from different JVM instances. The benefit is especially true for
 644   // processors with hyperthreading technology.
 645   static int counter = 0;
 646   int pid = os::current_process_id();
 647   alloca(((pid ^ counter++) & 7) * 128);
 648 
 649   thread->initialize_thread_current();
 650 
 651   OSThread* osthread = thread->osthread();
 652   Monitor* sync = osthread->startThread_lock();
 653 
 654   osthread->set_thread_id(os::current_thread_id());
 655 
 656   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 657     os::current_thread_id(), (uintx) pthread_self());
 658 
 659   if (UseNUMA) {
 660     int lgrp_id = os::numa_get_group_id();
 661     if (lgrp_id != -1) {
 662       thread->set_lgrp_id(lgrp_id);
 663     }
 664   }
 665   // initialize signal mask for this thread
 666   os::Linux::hotspot_sigmask(thread);
 667 
 668   // initialize floating point control register
 669   os::Linux::init_thread_fpu_state();
 670 
 671   // handshaking with parent thread
 672   {
 673     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 674 
 675     // notify parent thread
 676     osthread->set_state(INITIALIZED);
 677     sync->notify_all();
 678 
 679     // wait until os::start_thread()
 680     while (osthread->get_state() == INITIALIZED) {
 681       sync->wait(Mutex::_no_safepoint_check_flag);
 682     }
 683   }
 684 
 685   // call one more level start routine
 686   thread->run();
 687 
 688   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 689     os::current_thread_id(), (uintx) pthread_self());
 690 
 691   // If a thread has not deleted itself ("delete this") as part of its
 692   // termination sequence, we have to ensure thread-local-storage is
 693   // cleared before we actually terminate. No threads should ever be
 694   // deleted asynchronously with respect to their termination.
 695   if (Thread::current_or_null_safe() != NULL) {
 696     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 697     thread->clear_thread_current();
 698   }
 699 
 700   return 0;
 701 }
 702 
 703 bool os::create_thread(Thread* thread, ThreadType thr_type,
 704                        size_t req_stack_size) {
 705   assert(thread->osthread() == NULL, "caller responsible");
 706 
 707   // Allocate the OSThread object
 708   OSThread* osthread = new OSThread(NULL, NULL);
 709   if (osthread == NULL) {
 710     return false;
 711   }
 712 
 713   // set the correct thread state
 714   osthread->set_thread_type(thr_type);
 715 
 716   // Initial state is ALLOCATED but not INITIALIZED
 717   osthread->set_state(ALLOCATED);
 718 
 719   thread->set_osthread(osthread);
 720 
 721   // init thread attributes
 722   pthread_attr_t attr;
 723   pthread_attr_init(&attr);
 724   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 725 
 726   // calculate stack size if it's not specified by caller
 727   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 728   pthread_attr_setstacksize(&attr, stack_size);
 729 
 730   // glibc guard page
 731   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 732 
 733   ThreadState state;
 734 
 735   {
 736     pthread_t tid;
 737     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 738 
 739     char buf[64];
 740     if (ret == 0) {
 741       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 742         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 743     } else {
 744       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 745         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 746     }
 747 
 748     pthread_attr_destroy(&attr);
 749 
 750     if (ret != 0) {
 751       // Need to clean up stuff we've allocated so far
 752       thread->set_osthread(NULL);
 753       delete osthread;
 754       return false;
 755     }
 756 
 757     // Store pthread info into the OSThread
 758     osthread->set_pthread_id(tid);
 759 
 760     // Wait until child thread is either initialized or aborted
 761     {
 762       Monitor* sync_with_child = osthread->startThread_lock();
 763       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 764       while ((state = osthread->get_state()) == ALLOCATED) {
 765         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 766       }
 767     }
 768   }
 769 
 770   // Aborted due to thread limit being reached
 771   if (state == ZOMBIE) {
 772     thread->set_osthread(NULL);
 773     delete osthread;
 774     return false;
 775   }
 776 
 777   // The thread is returned suspended (in state INITIALIZED),
 778   // and is started higher up in the call chain
 779   assert(state == INITIALIZED, "race condition");
 780   return true;
 781 }
 782 
 783 /////////////////////////////////////////////////////////////////////////////
 784 // attach existing thread
 785 
 786 // bootstrap the main thread
 787 bool os::create_main_thread(JavaThread* thread) {
 788   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 789   return create_attached_thread(thread);
 790 }
 791 
 792 bool os::create_attached_thread(JavaThread* thread) {
 793 #ifdef ASSERT
 794   thread->verify_not_published();
 795 #endif
 796 
 797   // Allocate the OSThread object
 798   OSThread* osthread = new OSThread(NULL, NULL);
 799 
 800   if (osthread == NULL) {
 801     return false;
 802   }
 803 
 804   // Store pthread info into the OSThread
 805   osthread->set_thread_id(os::Linux::gettid());
 806   osthread->set_pthread_id(::pthread_self());
 807 
 808   // initialize floating point control register
 809   os::Linux::init_thread_fpu_state();
 810 
 811   // Initial thread state is RUNNABLE
 812   osthread->set_state(RUNNABLE);
 813 
 814   thread->set_osthread(osthread);
 815 
 816   if (UseNUMA) {
 817     int lgrp_id = os::numa_get_group_id();
 818     if (lgrp_id != -1) {
 819       thread->set_lgrp_id(lgrp_id);
 820     }
 821   }
 822 
 823   if (os::Linux::is_initial_thread()) {
 824     // If current thread is initial thread, its stack is mapped on demand,
 825     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 826     // the entire stack region to avoid SEGV in stack banging.
 827     // It is also useful to get around the heap-stack-gap problem on SuSE
 828     // kernel (see 4821821 for details). We first expand stack to the top
 829     // of yellow zone, then enable stack yellow zone (order is significant,
 830     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 831     // is no gap between the last two virtual memory regions.
 832 
 833     JavaThread *jt = (JavaThread *)thread;
 834     address addr = jt->stack_reserved_zone_base();
 835     assert(addr != NULL, "initialization problem?");
 836     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 837 
 838     osthread->set_expanding_stack();
 839     os::Linux::manually_expand_stack(jt, addr);
 840     osthread->clear_expanding_stack();
 841   }
 842 
 843   // initialize signal mask for this thread
 844   // and save the caller's signal mask
 845   os::Linux::hotspot_sigmask(thread);
 846 
 847   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 848     os::current_thread_id(), (uintx) pthread_self());
 849 
 850   return true;
 851 }
 852 
 853 void os::pd_start_thread(Thread* thread) {
 854   OSThread * osthread = thread->osthread();
 855   assert(osthread->get_state() != INITIALIZED, "just checking");
 856   Monitor* sync_with_child = osthread->startThread_lock();
 857   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 858   sync_with_child->notify();
 859 }
 860 
 861 // Free Linux resources related to the OSThread
 862 void os::free_thread(OSThread* osthread) {
 863   assert(osthread != NULL, "osthread not set");
 864 
 865   // We are told to free resources of the argument thread,
 866   // but we can only really operate on the current thread.
 867   assert(Thread::current()->osthread() == osthread,
 868          "os::free_thread but not current thread");
 869 
 870 #ifdef ASSERT
 871   sigset_t current;
 872   sigemptyset(&current);
 873   pthread_sigmask(SIG_SETMASK, NULL, &current);
 874   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 875 #endif
 876 
 877   // Restore caller's signal mask
 878   sigset_t sigmask = osthread->caller_sigmask();
 879   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 880 
 881   delete osthread;
 882 }
 883 
 884 //////////////////////////////////////////////////////////////////////////////
 885 // initial thread
 886 
 887 // Check if current thread is the initial thread, similar to Solaris thr_main.
 888 bool os::Linux::is_initial_thread(void) {
 889   char dummy;
 890   // If called before init complete, thread stack bottom will be null.
 891   // Can be called if fatal error occurs before initialization.
 892   if (initial_thread_stack_bottom() == NULL) return false;
 893   assert(initial_thread_stack_bottom() != NULL &&
 894          initial_thread_stack_size()   != 0,
 895          "os::init did not locate initial thread's stack region");
 896   if ((address)&dummy >= initial_thread_stack_bottom() &&
 897       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 898     return true;
 899   } else {
 900     return false;
 901   }
 902 }
 903 
 904 // Find the virtual memory area that contains addr
 905 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 906   FILE *fp = fopen("/proc/self/maps", "r");
 907   if (fp) {
 908     address low, high;
 909     while (!feof(fp)) {
 910       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 911         if (low <= addr && addr < high) {
 912           if (vma_low)  *vma_low  = low;
 913           if (vma_high) *vma_high = high;
 914           fclose(fp);
 915           return true;
 916         }
 917       }
 918       for (;;) {
 919         int ch = fgetc(fp);
 920         if (ch == EOF || ch == (int)'\n') break;
 921       }
 922     }
 923     fclose(fp);
 924   }
 925   return false;
 926 }
 927 
 928 // Locate initial thread stack. This special handling of initial thread stack
 929 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 930 // bogus value for initial thread.
 931 void os::Linux::capture_initial_stack(size_t max_size) {
 932   // stack size is the easy part, get it from RLIMIT_STACK
 933   struct rlimit rlim;
 934   getrlimit(RLIMIT_STACK, &rlim);
 935   size_t stack_size = rlim.rlim_cur;
 936 
 937   // 6308388: a bug in ld.so will relocate its own .data section to the
 938   //   lower end of primordial stack; reduce ulimit -s value a little bit
 939   //   so we won't install guard page on ld.so's data section.
 940   stack_size -= 2 * page_size();
 941 
 942   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
 943   //   7.1, in both cases we will get 2G in return value.
 944   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
 945   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
 946   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
 947   //   in case other parts in glibc still assumes 2M max stack size.
 948   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
 949   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
 950   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
 951     stack_size = 2 * K * K IA64_ONLY(*2);
 952   }
 953   // Try to figure out where the stack base (top) is. This is harder.
 954   //
 955   // When an application is started, glibc saves the initial stack pointer in
 956   // a global variable "__libc_stack_end", which is then used by system
 957   // libraries. __libc_stack_end should be pretty close to stack top. The
 958   // variable is available since the very early days. However, because it is
 959   // a private interface, it could disappear in the future.
 960   //
 961   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 962   // to __libc_stack_end, it is very close to stack top, but isn't the real
 963   // stack top. Note that /proc may not exist if VM is running as a chroot
 964   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 965   // /proc/<pid>/stat could change in the future (though unlikely).
 966   //
 967   // We try __libc_stack_end first. If that doesn't work, look for
 968   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
 969   // as a hint, which should work well in most cases.
 970 
 971   uintptr_t stack_start;
 972 
 973   // try __libc_stack_end first
 974   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
 975   if (p && *p) {
 976     stack_start = *p;
 977   } else {
 978     // see if we can get the start_stack field from /proc/self/stat
 979     FILE *fp;
 980     int pid;
 981     char state;
 982     int ppid;
 983     int pgrp;
 984     int session;
 985     int nr;
 986     int tpgrp;
 987     unsigned long flags;
 988     unsigned long minflt;
 989     unsigned long cminflt;
 990     unsigned long majflt;
 991     unsigned long cmajflt;
 992     unsigned long utime;
 993     unsigned long stime;
 994     long cutime;
 995     long cstime;
 996     long prio;
 997     long nice;
 998     long junk;
 999     long it_real;
1000     uintptr_t start;
1001     uintptr_t vsize;
1002     intptr_t rss;
1003     uintptr_t rsslim;
1004     uintptr_t scodes;
1005     uintptr_t ecode;
1006     int i;
1007 
1008     // Figure what the primordial thread stack base is. Code is inspired
1009     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1010     // followed by command name surrounded by parentheses, state, etc.
1011     char stat[2048];
1012     int statlen;
1013 
1014     fp = fopen("/proc/self/stat", "r");
1015     if (fp) {
1016       statlen = fread(stat, 1, 2047, fp);
1017       stat[statlen] = '\0';
1018       fclose(fp);
1019 
1020       // Skip pid and the command string. Note that we could be dealing with
1021       // weird command names, e.g. user could decide to rename java launcher
1022       // to "java 1.4.2 :)", then the stat file would look like
1023       //                1234 (java 1.4.2 :)) R ... ...
1024       // We don't really need to know the command string, just find the last
1025       // occurrence of ")" and then start parsing from there. See bug 4726580.
1026       char * s = strrchr(stat, ')');
1027 
1028       i = 0;
1029       if (s) {
1030         // Skip blank chars
1031         do { s++; } while (s && isspace(*s));
1032 
1033 #define _UFM UINTX_FORMAT
1034 #define _DFM INTX_FORMAT
1035 
1036         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1037         //              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
1038         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,
1039                    &state,          // 3  %c
1040                    &ppid,           // 4  %d
1041                    &pgrp,           // 5  %d
1042                    &session,        // 6  %d
1043                    &nr,             // 7  %d
1044                    &tpgrp,          // 8  %d
1045                    &flags,          // 9  %lu
1046                    &minflt,         // 10 %lu
1047                    &cminflt,        // 11 %lu
1048                    &majflt,         // 12 %lu
1049                    &cmajflt,        // 13 %lu
1050                    &utime,          // 14 %lu
1051                    &stime,          // 15 %lu
1052                    &cutime,         // 16 %ld
1053                    &cstime,         // 17 %ld
1054                    &prio,           // 18 %ld
1055                    &nice,           // 19 %ld
1056                    &junk,           // 20 %ld
1057                    &it_real,        // 21 %ld
1058                    &start,          // 22 UINTX_FORMAT
1059                    &vsize,          // 23 UINTX_FORMAT
1060                    &rss,            // 24 INTX_FORMAT
1061                    &rsslim,         // 25 UINTX_FORMAT
1062                    &scodes,         // 26 UINTX_FORMAT
1063                    &ecode,          // 27 UINTX_FORMAT
1064                    &stack_start);   // 28 UINTX_FORMAT
1065       }
1066 
1067 #undef _UFM
1068 #undef _DFM
1069 
1070       if (i != 28 - 2) {
1071         assert(false, "Bad conversion from /proc/self/stat");
1072         // product mode - assume we are the initial thread, good luck in the
1073         // embedded case.
1074         warning("Can't detect initial thread stack location - bad conversion");
1075         stack_start = (uintptr_t) &rlim;
1076       }
1077     } else {
1078       // For some reason we can't open /proc/self/stat (for example, running on
1079       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1080       // most cases, so don't abort:
1081       warning("Can't detect initial thread stack location - no /proc/self/stat");
1082       stack_start = (uintptr_t) &rlim;
1083     }
1084   }
1085 
1086   // Now we have a pointer (stack_start) very close to the stack top, the
1087   // next thing to do is to figure out the exact location of stack top. We
1088   // can find out the virtual memory area that contains stack_start by
1089   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1090   // and its upper limit is the real stack top. (again, this would fail if
1091   // running inside chroot, because /proc may not exist.)
1092 
1093   uintptr_t stack_top;
1094   address low, high;
1095   if (find_vma((address)stack_start, &low, &high)) {
1096     // success, "high" is the true stack top. (ignore "low", because initial
1097     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1098     stack_top = (uintptr_t)high;
1099   } else {
1100     // failed, likely because /proc/self/maps does not exist
1101     warning("Can't detect initial thread stack location - find_vma failed");
1102     // best effort: stack_start is normally within a few pages below the real
1103     // stack top, use it as stack top, and reduce stack size so we won't put
1104     // guard page outside stack.
1105     stack_top = stack_start;
1106     stack_size -= 16 * page_size();
1107   }
1108 
1109   // stack_top could be partially down the page so align it
1110   stack_top = align_size_up(stack_top, page_size());
1111 
1112   if (max_size && stack_size > max_size) {
1113     _initial_thread_stack_size = max_size;
1114   } else {
1115     _initial_thread_stack_size = stack_size;
1116   }
1117 
1118   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1119   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1120 }
1121 
1122 ////////////////////////////////////////////////////////////////////////////////
1123 // time support
1124 
1125 // Time since start-up in seconds to a fine granularity.
1126 // Used by VMSelfDestructTimer and the MemProfiler.
1127 double os::elapsedTime() {
1128 
1129   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1130 }
1131 
1132 jlong os::elapsed_counter() {
1133   return javaTimeNanos() - initial_time_count;
1134 }
1135 
1136 jlong os::elapsed_frequency() {
1137   return NANOSECS_PER_SEC; // nanosecond resolution
1138 }
1139 
1140 bool os::supports_vtime() { return true; }
1141 bool os::enable_vtime()   { return false; }
1142 bool os::vtime_enabled()  { return false; }
1143 
1144 double os::elapsedVTime() {
1145   struct rusage usage;
1146   int retval = getrusage(RUSAGE_THREAD, &usage);
1147   if (retval == 0) {
1148     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);
1149   } else {
1150     // better than nothing, but not much
1151     return elapsedTime();
1152   }
1153 }
1154 
1155 jlong os::javaTimeMillis() {
1156   timeval time;
1157   int status = gettimeofday(&time, NULL);
1158   assert(status != -1, "linux error");
1159   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1160 }
1161 
1162 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1163   timeval time;
1164   int status = gettimeofday(&time, NULL);
1165   assert(status != -1, "linux error");
1166   seconds = jlong(time.tv_sec);
1167   nanos = jlong(time.tv_usec) * 1000;
1168 }
1169 
1170 
1171 #ifndef CLOCK_MONOTONIC
1172   #define CLOCK_MONOTONIC (1)
1173 #endif
1174 
1175 void os::Linux::clock_init() {
1176   // we do dlopen's in this particular order due to bug in linux
1177   // dynamical loader (see 6348968) leading to crash on exit
1178   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1179   if (handle == NULL) {
1180     handle = dlopen("librt.so", RTLD_LAZY);
1181   }
1182 
1183   if (handle) {
1184     int (*clock_getres_func)(clockid_t, struct timespec*) =
1185            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1186     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1187            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1188     if (clock_getres_func && clock_gettime_func) {
1189       // See if monotonic clock is supported by the kernel. Note that some
1190       // early implementations simply return kernel jiffies (updated every
1191       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1192       // for nano time (though the monotonic property is still nice to have).
1193       // It's fixed in newer kernels, however clock_getres() still returns
1194       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1195       // resolution for now. Hopefully as people move to new kernels, this
1196       // won't be a problem.
1197       struct timespec res;
1198       struct timespec tp;
1199       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1200           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1201         // yes, monotonic clock is supported
1202         _clock_gettime = clock_gettime_func;
1203         return;
1204       } else {
1205         // close librt if there is no monotonic clock
1206         dlclose(handle);
1207       }
1208     }
1209   }
1210   warning("No monotonic clock was available - timed services may " \
1211           "be adversely affected if the time-of-day clock changes");
1212 }
1213 
1214 #ifndef SYS_clock_getres
1215   #if defined(X86) || defined(PPC64) || defined(S390)
1216     #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1217     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1218   #else
1219     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1220     #define sys_clock_getres(x,y)  -1
1221   #endif
1222 #else
1223   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1224 #endif
1225 
1226 void os::Linux::fast_thread_clock_init() {
1227   if (!UseLinuxPosixThreadCPUClocks) {
1228     return;
1229   }
1230   clockid_t clockid;
1231   struct timespec tp;
1232   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1233       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1234 
1235   // Switch to using fast clocks for thread cpu time if
1236   // the sys_clock_getres() returns 0 error code.
1237   // Note, that some kernels may support the current thread
1238   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1239   // returned by the pthread_getcpuclockid().
1240   // If the fast Posix clocks are supported then the sys_clock_getres()
1241   // must return at least tp.tv_sec == 0 which means a resolution
1242   // better than 1 sec. This is extra check for reliability.
1243 
1244   if (pthread_getcpuclockid_func &&
1245       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1246       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1247     _supports_fast_thread_cpu_time = true;
1248     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1249   }
1250 }
1251 
1252 jlong os::javaTimeNanos() {
1253   if (os::supports_monotonic_clock()) {
1254     struct timespec tp;
1255     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1256     assert(status == 0, "gettime error");
1257     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1258     return result;
1259   } else {
1260     timeval time;
1261     int status = gettimeofday(&time, NULL);
1262     assert(status != -1, "linux error");
1263     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1264     return 1000 * usecs;
1265   }
1266 }
1267 
1268 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1269   if (os::supports_monotonic_clock()) {
1270     info_ptr->max_value = ALL_64_BITS;
1271 
1272     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1273     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1274     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1275   } else {
1276     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1277     info_ptr->max_value = ALL_64_BITS;
1278 
1279     // gettimeofday is a real time clock so it skips
1280     info_ptr->may_skip_backward = true;
1281     info_ptr->may_skip_forward = true;
1282   }
1283 
1284   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1285 }
1286 
1287 // Return the real, user, and system times in seconds from an
1288 // arbitrary fixed point in the past.
1289 bool os::getTimesSecs(double* process_real_time,
1290                       double* process_user_time,
1291                       double* process_system_time) {
1292   struct tms ticks;
1293   clock_t real_ticks = times(&ticks);
1294 
1295   if (real_ticks == (clock_t) (-1)) {
1296     return false;
1297   } else {
1298     double ticks_per_second = (double) clock_tics_per_sec;
1299     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1300     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1301     *process_real_time = ((double) real_ticks) / ticks_per_second;
1302 
1303     return true;
1304   }
1305 }
1306 
1307 
1308 char * os::local_time_string(char *buf, size_t buflen) {
1309   struct tm t;
1310   time_t long_time;
1311   time(&long_time);
1312   localtime_r(&long_time, &t);
1313   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1314                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1315                t.tm_hour, t.tm_min, t.tm_sec);
1316   return buf;
1317 }
1318 
1319 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1320   return localtime_r(clock, res);
1321 }
1322 
1323 ////////////////////////////////////////////////////////////////////////////////
1324 // runtime exit support
1325 
1326 // Note: os::shutdown() might be called very early during initialization, or
1327 // called from signal handler. Before adding something to os::shutdown(), make
1328 // sure it is async-safe and can handle partially initialized VM.
1329 void os::shutdown() {
1330 
1331   // allow PerfMemory to attempt cleanup of any persistent resources
1332   perfMemory_exit();
1333 
1334   // needs to remove object in file system
1335   AttachListener::abort();
1336 
1337   // flush buffered output, finish log files
1338   ostream_abort();
1339 
1340   // Check for abort hook
1341   abort_hook_t abort_hook = Arguments::abort_hook();
1342   if (abort_hook != NULL) {
1343     abort_hook();
1344   }
1345 
1346 }
1347 
1348 // Note: os::abort() might be called very early during initialization, or
1349 // called from signal handler. Before adding something to os::abort(), make
1350 // sure it is async-safe and can handle partially initialized VM.
1351 void os::abort(bool dump_core, void* siginfo, const void* context) {
1352   os::shutdown();
1353   if (dump_core) {
1354 #ifndef PRODUCT
1355     fdStream out(defaultStream::output_fd());
1356     out.print_raw("Current thread is ");
1357     char buf[16];
1358     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1359     out.print_raw_cr(buf);
1360     out.print_raw_cr("Dumping core ...");
1361 #endif
1362     ::abort(); // dump core
1363   }
1364 
1365   ::exit(1);
1366 }
1367 
1368 // Die immediately, no exit hook, no abort hook, no cleanup.
1369 void os::die() {
1370   ::abort();
1371 }
1372 
1373 
1374 // This method is a copy of JDK's sysGetLastErrorString
1375 // from src/solaris/hpi/src/system_md.c
1376 
1377 size_t os::lasterror(char *buf, size_t len) {
1378   if (errno == 0)  return 0;
1379 
1380   const char *s = os::strerror(errno);
1381   size_t n = ::strlen(s);
1382   if (n >= len) {
1383     n = len - 1;
1384   }
1385   ::strncpy(buf, s, n);
1386   buf[n] = '\0';
1387   return n;
1388 }
1389 
1390 // thread_id is kernel thread id (similar to Solaris LWP id)
1391 intx os::current_thread_id() { return os::Linux::gettid(); }
1392 int os::current_process_id() {
1393   return ::getpid();
1394 }
1395 
1396 // DLL functions
1397 
1398 const char* os::dll_file_extension() { return ".so"; }
1399 
1400 // This must be hard coded because it's the system's temporary
1401 // directory not the java application's temp directory, ala java.io.tmpdir.
1402 const char* os::get_temp_directory() { return "/tmp"; }
1403 
1404 static bool file_exists(const char* filename) {
1405   struct stat statbuf;
1406   if (filename == NULL || strlen(filename) == 0) {
1407     return false;
1408   }
1409   return os::stat(filename, &statbuf) == 0;
1410 }
1411 
1412 bool os::dll_build_name(char* buffer, size_t buflen,
1413                         const char* pname, const char* fname) {
1414   bool retval = false;
1415   // Copied from libhpi
1416   const size_t pnamelen = pname ? strlen(pname) : 0;
1417 
1418   // Return error on buffer overflow.
1419   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1420     return retval;
1421   }
1422 
1423   if (pnamelen == 0) {
1424     snprintf(buffer, buflen, "lib%s.so", fname);
1425     retval = true;
1426   } else if (strchr(pname, *os::path_separator()) != NULL) {
1427     int n;
1428     char** pelements = split_path(pname, &n);
1429     if (pelements == NULL) {
1430       return false;
1431     }
1432     for (int i = 0; i < n; i++) {
1433       // Really shouldn't be NULL, but check can't hurt
1434       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1435         continue; // skip the empty path values
1436       }
1437       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1438       if (file_exists(buffer)) {
1439         retval = true;
1440         break;
1441       }
1442     }
1443     // release the storage
1444     for (int i = 0; i < n; i++) {
1445       if (pelements[i] != NULL) {
1446         FREE_C_HEAP_ARRAY(char, pelements[i]);
1447       }
1448     }
1449     if (pelements != NULL) {
1450       FREE_C_HEAP_ARRAY(char*, pelements);
1451     }
1452   } else {
1453     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1454     retval = true;
1455   }
1456   return retval;
1457 }
1458 
1459 // check if addr is inside libjvm.so
1460 bool os::address_is_in_vm(address addr) {
1461   static address libjvm_base_addr;
1462   Dl_info dlinfo;
1463 
1464   if (libjvm_base_addr == NULL) {
1465     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1466       libjvm_base_addr = (address)dlinfo.dli_fbase;
1467     }
1468     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1469   }
1470 
1471   if (dladdr((void *)addr, &dlinfo) != 0) {
1472     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1473   }
1474 
1475   return false;
1476 }
1477 
1478 bool os::dll_address_to_function_name(address addr, char *buf,
1479                                       int buflen, int *offset,
1480                                       bool demangle) {
1481   // buf is not optional, but offset is optional
1482   assert(buf != NULL, "sanity check");
1483 
1484   Dl_info dlinfo;
1485 
1486   if (dladdr((void*)addr, &dlinfo) != 0) {
1487     // see if we have a matching symbol
1488     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1489       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1490         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1491       }
1492       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1493       return true;
1494     }
1495     // no matching symbol so try for just file info
1496     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1497       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1498                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1499         return true;
1500       }
1501     }
1502   }
1503 
1504   buf[0] = '\0';
1505   if (offset != NULL) *offset = -1;
1506   return false;
1507 }
1508 
1509 struct _address_to_library_name {
1510   address addr;          // input : memory address
1511   size_t  buflen;        //         size of fname
1512   char*   fname;         // output: library name
1513   address base;          //         library base addr
1514 };
1515 
1516 static int address_to_library_name_callback(struct dl_phdr_info *info,
1517                                             size_t size, void *data) {
1518   int i;
1519   bool found = false;
1520   address libbase = NULL;
1521   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1522 
1523   // iterate through all loadable segments
1524   for (i = 0; i < info->dlpi_phnum; i++) {
1525     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1526     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1527       // base address of a library is the lowest address of its loaded
1528       // segments.
1529       if (libbase == NULL || libbase > segbase) {
1530         libbase = segbase;
1531       }
1532       // see if 'addr' is within current segment
1533       if (segbase <= d->addr &&
1534           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1535         found = true;
1536       }
1537     }
1538   }
1539 
1540   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1541   // so dll_address_to_library_name() can fall through to use dladdr() which
1542   // can figure out executable name from argv[0].
1543   if (found && info->dlpi_name && info->dlpi_name[0]) {
1544     d->base = libbase;
1545     if (d->fname) {
1546       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1547     }
1548     return 1;
1549   }
1550   return 0;
1551 }
1552 
1553 bool os::dll_address_to_library_name(address addr, char* buf,
1554                                      int buflen, int* offset) {
1555   // buf is not optional, but offset is optional
1556   assert(buf != NULL, "sanity check");
1557 
1558   Dl_info dlinfo;
1559   struct _address_to_library_name data;
1560 
1561   // There is a bug in old glibc dladdr() implementation that it could resolve
1562   // to wrong library name if the .so file has a base address != NULL. Here
1563   // we iterate through the program headers of all loaded libraries to find
1564   // out which library 'addr' really belongs to. This workaround can be
1565   // removed once the minimum requirement for glibc is moved to 2.3.x.
1566   data.addr = addr;
1567   data.fname = buf;
1568   data.buflen = buflen;
1569   data.base = NULL;
1570   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1571 
1572   if (rslt) {
1573     // buf already contains library name
1574     if (offset) *offset = addr - data.base;
1575     return true;
1576   }
1577   if (dladdr((void*)addr, &dlinfo) != 0) {
1578     if (dlinfo.dli_fname != NULL) {
1579       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1580     }
1581     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1582       *offset = addr - (address)dlinfo.dli_fbase;
1583     }
1584     return true;
1585   }
1586 
1587   buf[0] = '\0';
1588   if (offset) *offset = -1;
1589   return false;
1590 }
1591 
1592 // Loads .dll/.so and
1593 // in case of error it checks if .dll/.so was built for the
1594 // same architecture as Hotspot is running on
1595 
1596 
1597 // Remember the stack's state. The Linux dynamic linker will change
1598 // the stack to 'executable' at most once, so we must safepoint only once.
1599 bool os::Linux::_stack_is_executable = false;
1600 
1601 // VM operation that loads a library.  This is necessary if stack protection
1602 // of the Java stacks can be lost during loading the library.  If we
1603 // do not stop the Java threads, they can stack overflow before the stacks
1604 // are protected again.
1605 class VM_LinuxDllLoad: public VM_Operation {
1606  private:
1607   const char *_filename;
1608   char *_ebuf;
1609   int _ebuflen;
1610   void *_lib;
1611  public:
1612   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1613     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1614   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1615   void doit() {
1616     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1617     os::Linux::_stack_is_executable = true;
1618   }
1619   void* loaded_library() { return _lib; }
1620 };
1621 
1622 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1623   void * result = NULL;
1624   bool load_attempted = false;
1625 
1626   // Check whether the library to load might change execution rights
1627   // of the stack. If they are changed, the protection of the stack
1628   // guard pages will be lost. We need a safepoint to fix this.
1629   //
1630   // See Linux man page execstack(8) for more info.
1631   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1632     ElfFile ef(filename);
1633     if (!ef.specifies_noexecstack()) {
1634       if (!is_init_completed()) {
1635         os::Linux::_stack_is_executable = true;
1636         // This is OK - No Java threads have been created yet, and hence no
1637         // stack guard pages to fix.
1638         //
1639         // This should happen only when you are building JDK7 using a very
1640         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1641         //
1642         // Dynamic loader will make all stacks executable after
1643         // this function returns, and will not do that again.
1644         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1645       } else {
1646         warning("You have loaded library %s which might have disabled stack guard. "
1647                 "The VM will try to fix the stack guard now.\n"
1648                 "It's highly recommended that you fix the library with "
1649                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1650                 filename);
1651 
1652         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1653         JavaThread *jt = JavaThread::current();
1654         if (jt->thread_state() != _thread_in_native) {
1655           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1656           // that requires ExecStack. Cannot enter safe point. Let's give up.
1657           warning("Unable to fix stack guard. Giving up.");
1658         } else {
1659           if (!LoadExecStackDllInVMThread) {
1660             // This is for the case where the DLL has an static
1661             // constructor function that executes JNI code. We cannot
1662             // load such DLLs in the VMThread.
1663             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1664           }
1665 
1666           ThreadInVMfromNative tiv(jt);
1667           debug_only(VMNativeEntryWrapper vew;)
1668 
1669           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1670           VMThread::execute(&op);
1671           if (LoadExecStackDllInVMThread) {
1672             result = op.loaded_library();
1673           }
1674           load_attempted = true;
1675         }
1676       }
1677     }
1678   }
1679 
1680   if (!load_attempted) {
1681     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1682   }
1683 
1684   if (result != NULL) {
1685     // Successful loading
1686     return result;
1687   }
1688 
1689   Elf32_Ehdr elf_head;
1690   int diag_msg_max_length=ebuflen-strlen(ebuf);
1691   char* diag_msg_buf=ebuf+strlen(ebuf);
1692 
1693   if (diag_msg_max_length==0) {
1694     // No more space in ebuf for additional diagnostics message
1695     return NULL;
1696   }
1697 
1698 
1699   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1700 
1701   if (file_descriptor < 0) {
1702     // Can't open library, report dlerror() message
1703     return NULL;
1704   }
1705 
1706   bool failed_to_read_elf_head=
1707     (sizeof(elf_head)!=
1708      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1709 
1710   ::close(file_descriptor);
1711   if (failed_to_read_elf_head) {
1712     // file i/o error - report dlerror() msg
1713     return NULL;
1714   }
1715 
1716   typedef struct {
1717     Elf32_Half    code;         // Actual value as defined in elf.h
1718     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1719     unsigned char elf_class;    // 32 or 64 bit
1720     unsigned char endianess;    // MSB or LSB
1721     char*         name;         // String representation
1722   } arch_t;
1723 
1724 #ifndef EM_486
1725   #define EM_486          6               /* Intel 80486 */
1726 #endif
1727 #ifndef EM_AARCH64
1728   #define EM_AARCH64    183               /* ARM AARCH64 */
1729 #endif
1730 
1731   static const arch_t arch_array[]={
1732     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1733     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1734     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1735     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1736     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1737     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1738     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1739     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1740 #if defined(VM_LITTLE_ENDIAN)
1741     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1742 #else
1743     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1744 #endif
1745     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1746     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1747     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1748     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1749     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1750     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1751     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1752     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1753   };
1754 
1755 #if  (defined IA32)
1756   static  Elf32_Half running_arch_code=EM_386;
1757 #elif   (defined AMD64)
1758   static  Elf32_Half running_arch_code=EM_X86_64;
1759 #elif  (defined IA64)
1760   static  Elf32_Half running_arch_code=EM_IA_64;
1761 #elif  (defined __sparc) && (defined _LP64)
1762   static  Elf32_Half running_arch_code=EM_SPARCV9;
1763 #elif  (defined __sparc) && (!defined _LP64)
1764   static  Elf32_Half running_arch_code=EM_SPARC;
1765 #elif  (defined __powerpc64__)
1766   static  Elf32_Half running_arch_code=EM_PPC64;
1767 #elif  (defined __powerpc__)
1768   static  Elf32_Half running_arch_code=EM_PPC;
1769 #elif  (defined AARCH64)
1770   static  Elf32_Half running_arch_code=EM_AARCH64;
1771 #elif  (defined ARM)
1772   static  Elf32_Half running_arch_code=EM_ARM;
1773 #elif  (defined S390)
1774   static  Elf32_Half running_arch_code=EM_S390;
1775 #elif  (defined ALPHA)
1776   static  Elf32_Half running_arch_code=EM_ALPHA;
1777 #elif  (defined MIPSEL)
1778   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1779 #elif  (defined PARISC)
1780   static  Elf32_Half running_arch_code=EM_PARISC;
1781 #elif  (defined MIPS)
1782   static  Elf32_Half running_arch_code=EM_MIPS;
1783 #elif  (defined M68K)
1784   static  Elf32_Half running_arch_code=EM_68K;
1785 #else
1786     #error Method os::dll_load requires that one of following is defined:\
1787         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, __sparc
1788 #endif
1789 
1790   // Identify compatability class for VM's architecture and library's architecture
1791   // Obtain string descriptions for architectures
1792 
1793   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1794   int running_arch_index=-1;
1795 
1796   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1797     if (running_arch_code == arch_array[i].code) {
1798       running_arch_index    = i;
1799     }
1800     if (lib_arch.code == arch_array[i].code) {
1801       lib_arch.compat_class = arch_array[i].compat_class;
1802       lib_arch.name         = arch_array[i].name;
1803     }
1804   }
1805 
1806   assert(running_arch_index != -1,
1807          "Didn't find running architecture code (running_arch_code) in arch_array");
1808   if (running_arch_index == -1) {
1809     // Even though running architecture detection failed
1810     // we may still continue with reporting dlerror() message
1811     return NULL;
1812   }
1813 
1814   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1815     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1816     return NULL;
1817   }
1818 
1819 #ifndef S390
1820   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1821     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1822     return NULL;
1823   }
1824 #endif // !S390
1825 
1826   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1827     if (lib_arch.name!=NULL) {
1828       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1829                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1830                  lib_arch.name, arch_array[running_arch_index].name);
1831     } else {
1832       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1833                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1834                  lib_arch.code,
1835                  arch_array[running_arch_index].name);
1836     }
1837   }
1838 
1839   return NULL;
1840 }
1841 
1842 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1843                                 int ebuflen) {
1844   void * result = ::dlopen(filename, RTLD_LAZY);
1845   if (result == NULL) {
1846     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1847     ebuf[ebuflen-1] = '\0';
1848   }
1849   return result;
1850 }
1851 
1852 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1853                                        int ebuflen) {
1854   void * result = NULL;
1855   if (LoadExecStackDllInVMThread) {
1856     result = dlopen_helper(filename, ebuf, ebuflen);
1857   }
1858 
1859   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1860   // library that requires an executable stack, or which does not have this
1861   // stack attribute set, dlopen changes the stack attribute to executable. The
1862   // read protection of the guard pages gets lost.
1863   //
1864   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1865   // may have been queued at the same time.
1866 
1867   if (!_stack_is_executable) {
1868     JavaThread *jt = Threads::first();
1869 
1870     while (jt) {
1871       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1872           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1873         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1874           warning("Attempt to reguard stack yellow zone failed.");
1875         }
1876       }
1877       jt = jt->next();
1878     }
1879   }
1880 
1881   return result;
1882 }
1883 
1884 void* os::dll_lookup(void* handle, const char* name) {
1885   void* res = dlsym(handle, name);
1886   return res;
1887 }
1888 
1889 void* os::get_default_process_handle() {
1890   return (void*)::dlopen(NULL, RTLD_LAZY);
1891 }
1892 
1893 static bool _print_ascii_file(const char* filename, outputStream* st) {
1894   int fd = ::open(filename, O_RDONLY);
1895   if (fd == -1) {
1896     return false;
1897   }
1898 
1899   char buf[33];
1900   int bytes;
1901   buf[32] = '\0';
1902   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1903     st->print_raw(buf, bytes);
1904   }
1905 
1906   ::close(fd);
1907 
1908   return true;
1909 }
1910 
1911 void os::print_dll_info(outputStream *st) {
1912   st->print_cr("Dynamic libraries:");
1913 
1914   char fname[32];
1915   pid_t pid = os::Linux::gettid();
1916 
1917   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1918 
1919   if (!_print_ascii_file(fname, st)) {
1920     st->print("Can not get library information for pid = %d\n", pid);
1921   }
1922 }
1923 
1924 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1925   FILE *procmapsFile = NULL;
1926 
1927   // Open the procfs maps file for the current process
1928   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1929     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1930     char line[PATH_MAX + 100];
1931 
1932     // Read line by line from 'file'
1933     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1934       u8 base, top, offset, inode;
1935       char permissions[5];
1936       char device[6];
1937       char name[PATH_MAX + 1];
1938 
1939       // Parse fields from line
1940       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1941              &base, &top, permissions, &offset, device, &inode, name);
1942 
1943       // Filter by device id '00:00' so that we only get file system mapped files.
1944       if (strcmp(device, "00:00") != 0) {
1945 
1946         // Call callback with the fields of interest
1947         if(callback(name, (address)base, (address)top, param)) {
1948           // Oops abort, callback aborted
1949           fclose(procmapsFile);
1950           return 1;
1951         }
1952       }
1953     }
1954     fclose(procmapsFile);
1955   }
1956   return 0;
1957 }
1958 
1959 void os::print_os_info_brief(outputStream* st) {
1960   os::Linux::print_distro_info(st);
1961 
1962   os::Posix::print_uname_info(st);
1963 
1964   os::Linux::print_libversion_info(st);
1965 
1966 }
1967 
1968 void os::print_os_info(outputStream* st) {
1969   st->print("OS:");
1970 
1971   os::Linux::print_distro_info(st);
1972 
1973   os::Posix::print_uname_info(st);
1974 
1975   // Print warning if unsafe chroot environment detected
1976   if (unsafe_chroot_detected) {
1977     st->print("WARNING!! ");
1978     st->print_cr("%s", unstable_chroot_error);
1979   }
1980 
1981   os::Linux::print_libversion_info(st);
1982 
1983   os::Posix::print_rlimit_info(st);
1984 
1985   os::Posix::print_load_average(st);
1986 
1987   os::Linux::print_full_memory_info(st);
1988 }
1989 
1990 // Try to identify popular distros.
1991 // Most Linux distributions have a /etc/XXX-release file, which contains
1992 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1993 // file that also contains the OS version string. Some have more than one
1994 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1995 // /etc/redhat-release.), so the order is important.
1996 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1997 // their own specific XXX-release file as well as a redhat-release file.
1998 // Because of this the XXX-release file needs to be searched for before the
1999 // redhat-release file.
2000 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2001 // search for redhat-release / SuSE-release needs to be before lsb-release.
2002 // Since the lsb-release file is the new standard it needs to be searched
2003 // before the older style release files.
2004 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2005 // next to last resort.  The os-release file is a new standard that contains
2006 // distribution information and the system-release file seems to be an old
2007 // standard that has been replaced by the lsb-release and os-release files.
2008 // Searching for the debian_version file is the last resort.  It contains
2009 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2010 // "Debian " is printed before the contents of the debian_version file.
2011 
2012 const char* distro_files[] = {
2013   "/etc/oracle-release",
2014   "/etc/mandriva-release",
2015   "/etc/mandrake-release",
2016   "/etc/sun-release",
2017   "/etc/redhat-release",
2018   "/etc/SuSE-release",
2019   "/etc/lsb-release",
2020   "/etc/turbolinux-release",
2021   "/etc/gentoo-release",
2022   "/etc/ltib-release",
2023   "/etc/angstrom-version",
2024   "/etc/system-release",
2025   "/etc/os-release",
2026   NULL };
2027 
2028 void os::Linux::print_distro_info(outputStream* st) {
2029   for (int i = 0;; i++) {
2030     const char* file = distro_files[i];
2031     if (file == NULL) {
2032       break;  // done
2033     }
2034     // If file prints, we found it.
2035     if (_print_ascii_file(file, st)) {
2036       return;
2037     }
2038   }
2039 
2040   if (file_exists("/etc/debian_version")) {
2041     st->print("Debian ");
2042     _print_ascii_file("/etc/debian_version", st);
2043   } else {
2044     st->print("Linux");
2045   }
2046   st->cr();
2047 }
2048 
2049 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2050   char buf[256];
2051   while (fgets(buf, sizeof(buf), fp)) {
2052     // Edit out extra stuff in expected format
2053     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2054       char* ptr = strstr(buf, "\"");  // the name is in quotes
2055       if (ptr != NULL) {
2056         ptr++; // go beyond first quote
2057         char* nl = strchr(ptr, '\"');
2058         if (nl != NULL) *nl = '\0';
2059         strncpy(distro, ptr, length);
2060       } else {
2061         ptr = strstr(buf, "=");
2062         ptr++; // go beyond equals then
2063         char* nl = strchr(ptr, '\n');
2064         if (nl != NULL) *nl = '\0';
2065         strncpy(distro, ptr, length);
2066       }
2067       return;
2068     } else if (get_first_line) {
2069       char* nl = strchr(buf, '\n');
2070       if (nl != NULL) *nl = '\0';
2071       strncpy(distro, buf, length);
2072       return;
2073     }
2074   }
2075   // print last line and close
2076   char* nl = strchr(buf, '\n');
2077   if (nl != NULL) *nl = '\0';
2078   strncpy(distro, buf, length);
2079 }
2080 
2081 static void parse_os_info(char* distro, size_t length, const char* file) {
2082   FILE* fp = fopen(file, "r");
2083   if (fp != NULL) {
2084     // if suse format, print out first line
2085     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2086     parse_os_info_helper(fp, distro, length, get_first_line);
2087     fclose(fp);
2088   }
2089 }
2090 
2091 void os::get_summary_os_info(char* buf, size_t buflen) {
2092   for (int i = 0;; i++) {
2093     const char* file = distro_files[i];
2094     if (file == NULL) {
2095       break; // ran out of distro_files
2096     }
2097     if (file_exists(file)) {
2098       parse_os_info(buf, buflen, file);
2099       return;
2100     }
2101   }
2102   // special case for debian
2103   if (file_exists("/etc/debian_version")) {
2104     strncpy(buf, "Debian ", buflen);
2105     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2106   } else {
2107     strncpy(buf, "Linux", buflen);
2108   }
2109 }
2110 
2111 void os::Linux::print_libversion_info(outputStream* st) {
2112   // libc, pthread
2113   st->print("libc:");
2114   st->print("%s ", os::Linux::glibc_version());
2115   st->print("%s ", os::Linux::libpthread_version());
2116   st->cr();
2117 }
2118 
2119 void os::Linux::print_full_memory_info(outputStream* st) {
2120   st->print("\n/proc/meminfo:\n");
2121   _print_ascii_file("/proc/meminfo", st);
2122   st->cr();
2123 }
2124 
2125 void os::print_memory_info(outputStream* st) {
2126 
2127   st->print("Memory:");
2128   st->print(" %dk page", os::vm_page_size()>>10);
2129 
2130   // values in struct sysinfo are "unsigned long"
2131   struct sysinfo si;
2132   sysinfo(&si);
2133 
2134   st->print(", physical " UINT64_FORMAT "k",
2135             os::physical_memory() >> 10);
2136   st->print("(" UINT64_FORMAT "k free)",
2137             os::available_memory() >> 10);
2138   st->print(", swap " UINT64_FORMAT "k",
2139             ((jlong)si.totalswap * si.mem_unit) >> 10);
2140   st->print("(" UINT64_FORMAT "k free)",
2141             ((jlong)si.freeswap * si.mem_unit) >> 10);
2142   st->cr();
2143 }
2144 
2145 // Print the first "model name" line and the first "flags" line
2146 // that we find and nothing more. We assume "model name" comes
2147 // before "flags" so if we find a second "model name", then the
2148 // "flags" field is considered missing.
2149 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2150 #if defined(IA32) || defined(AMD64)
2151   // Other platforms have less repetitive cpuinfo files
2152   FILE *fp = fopen("/proc/cpuinfo", "r");
2153   if (fp) {
2154     while (!feof(fp)) {
2155       if (fgets(buf, buflen, fp)) {
2156         // Assume model name comes before flags
2157         bool model_name_printed = false;
2158         if (strstr(buf, "model name") != NULL) {
2159           if (!model_name_printed) {
2160             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2161             st->print_raw(buf);
2162             model_name_printed = true;
2163           } else {
2164             // model name printed but not flags?  Odd, just return
2165             fclose(fp);
2166             return true;
2167           }
2168         }
2169         // print the flags line too
2170         if (strstr(buf, "flags") != NULL) {
2171           st->print_raw(buf);
2172           fclose(fp);
2173           return true;
2174         }
2175       }
2176     }
2177     fclose(fp);
2178   }
2179 #endif // x86 platforms
2180   return false;
2181 }
2182 
2183 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2184   // Only print the model name if the platform provides this as a summary
2185   if (!print_model_name_and_flags(st, buf, buflen)) {
2186     st->print("\n/proc/cpuinfo:\n");
2187     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2188       st->print_cr("  <Not Available>");
2189     }
2190   }
2191 }
2192 
2193 #if defined(AMD64) || defined(IA32) || defined(X32)
2194 const char* search_string = "model name";
2195 #elif defined(PPC64)
2196 const char* search_string = "cpu";
2197 #elif defined(S390)
2198 const char* search_string = "processor";
2199 #elif defined(SPARC)
2200 const char* search_string = "cpu";
2201 #else
2202 const char* search_string = "Processor";
2203 #endif
2204 
2205 // Parses the cpuinfo file for string representing the model name.
2206 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2207   FILE* fp = fopen("/proc/cpuinfo", "r");
2208   if (fp != NULL) {
2209     while (!feof(fp)) {
2210       char buf[256];
2211       if (fgets(buf, sizeof(buf), fp)) {
2212         char* start = strstr(buf, search_string);
2213         if (start != NULL) {
2214           char *ptr = start + strlen(search_string);
2215           char *end = buf + strlen(buf);
2216           while (ptr != end) {
2217              // skip whitespace and colon for the rest of the name.
2218              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2219                break;
2220              }
2221              ptr++;
2222           }
2223           if (ptr != end) {
2224             // reasonable string, get rid of newline and keep the rest
2225             char* nl = strchr(buf, '\n');
2226             if (nl != NULL) *nl = '\0';
2227             strncpy(cpuinfo, ptr, length);
2228             fclose(fp);
2229             return;
2230           }
2231         }
2232       }
2233     }
2234     fclose(fp);
2235   }
2236   // cpuinfo not found or parsing failed, just print generic string.  The entire
2237   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2238 #if   defined(AARCH64)
2239   strncpy(cpuinfo, "AArch64", length);
2240 #elif defined(AMD64)
2241   strncpy(cpuinfo, "x86_64", length);
2242 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2243   strncpy(cpuinfo, "ARM", length);
2244 #elif defined(IA32)
2245   strncpy(cpuinfo, "x86_32", length);
2246 #elif defined(IA64)
2247   strncpy(cpuinfo, "IA64", length);
2248 #elif defined(PPC)
2249   strncpy(cpuinfo, "PPC64", length);
2250 #elif defined(S390)
2251   strncpy(cpuinfo, "S390", length);
2252 #elif defined(SPARC)
2253   strncpy(cpuinfo, "sparcv9", length);
2254 #elif defined(ZERO_LIBARCH)
2255   strncpy(cpuinfo, ZERO_LIBARCH, length);
2256 #else
2257   strncpy(cpuinfo, "unknown", length);
2258 #endif
2259 }
2260 
2261 static void print_signal_handler(outputStream* st, int sig,
2262                                  char* buf, size_t buflen);
2263 
2264 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2265   st->print_cr("Signal Handlers:");
2266   print_signal_handler(st, SIGSEGV, buf, buflen);
2267   print_signal_handler(st, SIGBUS , buf, buflen);
2268   print_signal_handler(st, SIGFPE , buf, buflen);
2269   print_signal_handler(st, SIGPIPE, buf, buflen);
2270   print_signal_handler(st, SIGXFSZ, buf, buflen);
2271   print_signal_handler(st, SIGILL , buf, buflen);
2272   print_signal_handler(st, SR_signum, buf, buflen);
2273   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2274   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2275   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2276   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2277 #if defined(PPC64)
2278   print_signal_handler(st, SIGTRAP, buf, buflen);
2279 #endif
2280 }
2281 
2282 static char saved_jvm_path[MAXPATHLEN] = {0};
2283 
2284 // Find the full path to the current module, libjvm.so
2285 void os::jvm_path(char *buf, jint buflen) {
2286   // Error checking.
2287   if (buflen < MAXPATHLEN) {
2288     assert(false, "must use a large-enough buffer");
2289     buf[0] = '\0';
2290     return;
2291   }
2292   // Lazy resolve the path to current module.
2293   if (saved_jvm_path[0] != 0) {
2294     strcpy(buf, saved_jvm_path);
2295     return;
2296   }
2297 
2298   char dli_fname[MAXPATHLEN];
2299   bool ret = dll_address_to_library_name(
2300                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2301                                          dli_fname, sizeof(dli_fname), NULL);
2302   assert(ret, "cannot locate libjvm");
2303   char *rp = NULL;
2304   if (ret && dli_fname[0] != '\0') {
2305     rp = realpath(dli_fname, buf);
2306   }
2307   if (rp == NULL) {
2308     return;
2309   }
2310 
2311   if (Arguments::sun_java_launcher_is_altjvm()) {
2312     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2313     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2314     // If "/jre/lib/" appears at the right place in the string, then
2315     // assume we are installed in a JDK and we're done. Otherwise, check
2316     // for a JAVA_HOME environment variable and fix up the path so it
2317     // looks like libjvm.so is installed there (append a fake suffix
2318     // hotspot/libjvm.so).
2319     const char *p = buf + strlen(buf) - 1;
2320     for (int count = 0; p > buf && count < 5; ++count) {
2321       for (--p; p > buf && *p != '/'; --p)
2322         /* empty */ ;
2323     }
2324 
2325     if (strncmp(p, "/jre/lib/", 9) != 0) {
2326       // Look for JAVA_HOME in the environment.
2327       char* java_home_var = ::getenv("JAVA_HOME");
2328       if (java_home_var != NULL && java_home_var[0] != 0) {
2329         char* jrelib_p;
2330         int len;
2331 
2332         // Check the current module name "libjvm.so".
2333         p = strrchr(buf, '/');
2334         if (p == NULL) {
2335           return;
2336         }
2337         assert(strstr(p, "/libjvm") == p, "invalid library name");
2338 
2339         rp = realpath(java_home_var, buf);
2340         if (rp == NULL) {
2341           return;
2342         }
2343 
2344         // determine if this is a legacy image or modules image
2345         // modules image doesn't have "jre" subdirectory
2346         len = strlen(buf);
2347         assert(len < buflen, "Ran out of buffer room");
2348         jrelib_p = buf + len;
2349         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2350         if (0 != access(buf, F_OK)) {
2351           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2352         }
2353 
2354         if (0 == access(buf, F_OK)) {
2355           // Use current module name "libjvm.so"
2356           len = strlen(buf);
2357           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2358         } else {
2359           // Go back to path of .so
2360           rp = realpath(dli_fname, buf);
2361           if (rp == NULL) {
2362             return;
2363           }
2364         }
2365       }
2366     }
2367   }
2368 
2369   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2370   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2371 }
2372 
2373 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2374   // no prefix required, not even "_"
2375 }
2376 
2377 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2378   // no suffix required
2379 }
2380 
2381 ////////////////////////////////////////////////////////////////////////////////
2382 // sun.misc.Signal support
2383 
2384 static volatile jint sigint_count = 0;
2385 
2386 static void UserHandler(int sig, void *siginfo, void *context) {
2387   // 4511530 - sem_post is serialized and handled by the manager thread. When
2388   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2389   // don't want to flood the manager thread with sem_post requests.
2390   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2391     return;
2392   }
2393 
2394   // Ctrl-C is pressed during error reporting, likely because the error
2395   // handler fails to abort. Let VM die immediately.
2396   if (sig == SIGINT && is_error_reported()) {
2397     os::die();
2398   }
2399 
2400   os::signal_notify(sig);
2401 }
2402 
2403 void* os::user_handler() {
2404   return CAST_FROM_FN_PTR(void*, UserHandler);
2405 }
2406 
2407 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2408   struct timespec ts;
2409   // Semaphore's are always associated with CLOCK_REALTIME
2410   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2411   // see unpackTime for discussion on overflow checking
2412   if (sec >= MAX_SECS) {
2413     ts.tv_sec += MAX_SECS;
2414     ts.tv_nsec = 0;
2415   } else {
2416     ts.tv_sec += sec;
2417     ts.tv_nsec += nsec;
2418     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2419       ts.tv_nsec -= NANOSECS_PER_SEC;
2420       ++ts.tv_sec; // note: this must be <= max_secs
2421     }
2422   }
2423 
2424   return ts;
2425 }
2426 
2427 extern "C" {
2428   typedef void (*sa_handler_t)(int);
2429   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2430 }
2431 
2432 void* os::signal(int signal_number, void* handler) {
2433   struct sigaction sigAct, oldSigAct;
2434 
2435   sigfillset(&(sigAct.sa_mask));
2436   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2437   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2438 
2439   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2440     // -1 means registration failed
2441     return (void *)-1;
2442   }
2443 
2444   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2445 }
2446 
2447 void os::signal_raise(int signal_number) {
2448   ::raise(signal_number);
2449 }
2450 
2451 // The following code is moved from os.cpp for making this
2452 // code platform specific, which it is by its very nature.
2453 
2454 // Will be modified when max signal is changed to be dynamic
2455 int os::sigexitnum_pd() {
2456   return NSIG;
2457 }
2458 
2459 // a counter for each possible signal value
2460 static volatile jint pending_signals[NSIG+1] = { 0 };
2461 
2462 // Linux(POSIX) specific hand shaking semaphore.
2463 static sem_t sig_sem;
2464 static PosixSemaphore sr_semaphore;
2465 
2466 void os::signal_init_pd() {
2467   // Initialize signal structures
2468   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2469 
2470   // Initialize signal semaphore
2471   ::sem_init(&sig_sem, 0, 0);
2472 }
2473 
2474 void os::signal_notify(int sig) {
2475   Atomic::inc(&pending_signals[sig]);
2476   ::sem_post(&sig_sem);
2477 }
2478 
2479 static int check_pending_signals(bool wait) {
2480   Atomic::store(0, &sigint_count);
2481   for (;;) {
2482     for (int i = 0; i < NSIG + 1; i++) {
2483       jint n = pending_signals[i];
2484       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2485         return i;
2486       }
2487     }
2488     if (!wait) {
2489       return -1;
2490     }
2491     JavaThread *thread = JavaThread::current();
2492     ThreadBlockInVM tbivm(thread);
2493 
2494     bool threadIsSuspended;
2495     do {
2496       thread->set_suspend_equivalent();
2497       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2498       ::sem_wait(&sig_sem);
2499 
2500       // were we externally suspended while we were waiting?
2501       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2502       if (threadIsSuspended) {
2503         // The semaphore has been incremented, but while we were waiting
2504         // another thread suspended us. We don't want to continue running
2505         // while suspended because that would surprise the thread that
2506         // suspended us.
2507         ::sem_post(&sig_sem);
2508 
2509         thread->java_suspend_self();
2510       }
2511     } while (threadIsSuspended);
2512   }
2513 }
2514 
2515 int os::signal_lookup() {
2516   return check_pending_signals(false);
2517 }
2518 
2519 int os::signal_wait() {
2520   return check_pending_signals(true);
2521 }
2522 
2523 ////////////////////////////////////////////////////////////////////////////////
2524 // Virtual Memory
2525 
2526 int os::vm_page_size() {
2527   // Seems redundant as all get out
2528   assert(os::Linux::page_size() != -1, "must call os::init");
2529   return os::Linux::page_size();
2530 }
2531 
2532 // Solaris allocates memory by pages.
2533 int os::vm_allocation_granularity() {
2534   assert(os::Linux::page_size() != -1, "must call os::init");
2535   return os::Linux::page_size();
2536 }
2537 
2538 // Rationale behind this function:
2539 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2540 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2541 //  samples for JITted code. Here we create private executable mapping over the code cache
2542 //  and then we can use standard (well, almost, as mapping can change) way to provide
2543 //  info for the reporting script by storing timestamp and location of symbol
2544 void linux_wrap_code(char* base, size_t size) {
2545   static volatile jint cnt = 0;
2546 
2547   if (!UseOprofile) {
2548     return;
2549   }
2550 
2551   char buf[PATH_MAX+1];
2552   int num = Atomic::add(1, &cnt);
2553 
2554   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2555            os::get_temp_directory(), os::current_process_id(), num);
2556   unlink(buf);
2557 
2558   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2559 
2560   if (fd != -1) {
2561     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2562     if (rv != (off_t)-1) {
2563       if (::write(fd, "", 1) == 1) {
2564         mmap(base, size,
2565              PROT_READ|PROT_WRITE|PROT_EXEC,
2566              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2567       }
2568     }
2569     ::close(fd);
2570     unlink(buf);
2571   }
2572 }
2573 
2574 static bool recoverable_mmap_error(int err) {
2575   // See if the error is one we can let the caller handle. This
2576   // list of errno values comes from JBS-6843484. I can't find a
2577   // Linux man page that documents this specific set of errno
2578   // values so while this list currently matches Solaris, it may
2579   // change as we gain experience with this failure mode.
2580   switch (err) {
2581   case EBADF:
2582   case EINVAL:
2583   case ENOTSUP:
2584     // let the caller deal with these errors
2585     return true;
2586 
2587   default:
2588     // Any remaining errors on this OS can cause our reserved mapping
2589     // to be lost. That can cause confusion where different data
2590     // structures think they have the same memory mapped. The worst
2591     // scenario is if both the VM and a library think they have the
2592     // same memory mapped.
2593     return false;
2594   }
2595 }
2596 
2597 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2598                                     int err) {
2599   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2600           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2601           os::strerror(err), err);
2602 }
2603 
2604 static void warn_fail_commit_memory(char* addr, size_t size,
2605                                     size_t alignment_hint, bool exec,
2606                                     int err) {
2607   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2608           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2609           alignment_hint, exec, os::strerror(err), err);
2610 }
2611 
2612 // NOTE: Linux kernel does not really reserve the pages for us.
2613 //       All it does is to check if there are enough free pages
2614 //       left at the time of mmap(). This could be a potential
2615 //       problem.
2616 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2617   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2618   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2619                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2620   if (res != (uintptr_t) MAP_FAILED) {
2621     if (UseNUMAInterleaving) {
2622       numa_make_global(addr, size);
2623     }
2624     return 0;
2625   }
2626 
2627   int err = errno;  // save errno from mmap() call above
2628 
2629   if (!recoverable_mmap_error(err)) {
2630     warn_fail_commit_memory(addr, size, exec, err);
2631     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2632   }
2633 
2634   return err;
2635 }
2636 
2637 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2638   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2639 }
2640 
2641 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2642                                   const char* mesg) {
2643   assert(mesg != NULL, "mesg must be specified");
2644   int err = os::Linux::commit_memory_impl(addr, size, exec);
2645   if (err != 0) {
2646     // the caller wants all commit errors to exit with the specified mesg:
2647     warn_fail_commit_memory(addr, size, exec, err);
2648     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2649   }
2650 }
2651 
2652 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2653 #ifndef MAP_HUGETLB
2654   #define MAP_HUGETLB 0x40000
2655 #endif
2656 
2657 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2658 #ifndef MADV_HUGEPAGE
2659   #define MADV_HUGEPAGE 14
2660 #endif
2661 
2662 int os::Linux::commit_memory_impl(char* addr, size_t size,
2663                                   size_t alignment_hint, bool exec) {
2664   int err = os::Linux::commit_memory_impl(addr, size, exec);
2665   if (err == 0) {
2666     realign_memory(addr, size, alignment_hint);
2667   }
2668   return err;
2669 }
2670 
2671 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2672                           bool exec) {
2673   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2674 }
2675 
2676 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2677                                   size_t alignment_hint, bool exec,
2678                                   const char* mesg) {
2679   assert(mesg != NULL, "mesg must be specified");
2680   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2681   if (err != 0) {
2682     // the caller wants all commit errors to exit with the specified mesg:
2683     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2684     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2685   }
2686 }
2687 
2688 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2689   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2690     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2691     // be supported or the memory may already be backed by huge pages.
2692     ::madvise(addr, bytes, MADV_HUGEPAGE);
2693   }
2694 }
2695 
2696 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2697   // This method works by doing an mmap over an existing mmaping and effectively discarding
2698   // the existing pages. However it won't work for SHM-based large pages that cannot be
2699   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2700   // small pages on top of the SHM segment. This method always works for small pages, so we
2701   // allow that in any case.
2702   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2703     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2704   }
2705 }
2706 
2707 void os::numa_make_global(char *addr, size_t bytes) {
2708   Linux::numa_interleave_memory(addr, bytes);
2709 }
2710 
2711 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2712 // bind policy to MPOL_PREFERRED for the current thread.
2713 #define USE_MPOL_PREFERRED 0
2714 
2715 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2716   // To make NUMA and large pages more robust when both enabled, we need to ease
2717   // the requirements on where the memory should be allocated. MPOL_BIND is the
2718   // default policy and it will force memory to be allocated on the specified
2719   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2720   // the specified node, but will not force it. Using this policy will prevent
2721   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2722   // free large pages.
2723   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2724   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2725 }
2726 
2727 bool os::numa_topology_changed() { return false; }
2728 
2729 size_t os::numa_get_groups_num() {
2730   int max_node = Linux::numa_max_node();
2731   return max_node > 0 ? max_node + 1 : 1;
2732 }
2733 
2734 int os::numa_get_group_id() {
2735   int cpu_id = Linux::sched_getcpu();
2736   if (cpu_id != -1) {
2737     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2738     if (lgrp_id != -1) {
2739       return lgrp_id;
2740     }
2741   }
2742   return 0;
2743 }
2744 
2745 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2746   for (size_t i = 0; i < size; i++) {
2747     ids[i] = i;
2748   }
2749   return size;
2750 }
2751 
2752 bool os::get_page_info(char *start, page_info* info) {
2753   return false;
2754 }
2755 
2756 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2757                      page_info* page_found) {
2758   return end;
2759 }
2760 
2761 
2762 int os::Linux::sched_getcpu_syscall(void) {
2763   unsigned int cpu = 0;
2764   int retval = -1;
2765 
2766 #if defined(IA32)
2767   #ifndef SYS_getcpu
2768     #define SYS_getcpu 318
2769   #endif
2770   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2771 #elif defined(AMD64)
2772 // Unfortunately we have to bring all these macros here from vsyscall.h
2773 // to be able to compile on old linuxes.
2774   #define __NR_vgetcpu 2
2775   #define VSYSCALL_START (-10UL << 20)
2776   #define VSYSCALL_SIZE 1024
2777   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2778   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2779   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2780   retval = vgetcpu(&cpu, NULL, NULL);
2781 #endif
2782 
2783   return (retval == -1) ? retval : cpu;
2784 }
2785 
2786 // Something to do with the numa-aware allocator needs these symbols
2787 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2788 extern "C" JNIEXPORT void numa_error(char *where) { }
2789 
2790 
2791 // If we are running with libnuma version > 2, then we should
2792 // be trying to use symbols with versions 1.1
2793 // If we are running with earlier version, which did not have symbol versions,
2794 // we should use the base version.
2795 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2796   void *f = dlvsym(handle, name, "libnuma_1.1");
2797   if (f == NULL) {
2798     f = dlsym(handle, name);
2799   }
2800   return f;
2801 }
2802 
2803 bool os::Linux::libnuma_init() {
2804   // sched_getcpu() should be in libc.
2805   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2806                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2807 
2808   // If it's not, try a direct syscall.
2809   if (sched_getcpu() == -1) {
2810     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2811                                     (void*)&sched_getcpu_syscall));
2812   }
2813 
2814   if (sched_getcpu() != -1) { // Does it work?
2815     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2816     if (handle != NULL) {
2817       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2818                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2819       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2820                                        libnuma_dlsym(handle, "numa_max_node")));
2821       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2822                                         libnuma_dlsym(handle, "numa_available")));
2823       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2824                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2825       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2826                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2827       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2828                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2829 
2830 
2831       if (numa_available() != -1) {
2832         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2833         // Create a cpu -> node mapping
2834         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2835         rebuild_cpu_to_node_map();
2836         return true;
2837       }
2838     }
2839   }
2840   return false;
2841 }
2842 
2843 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2844 // The table is later used in get_node_by_cpu().
2845 void os::Linux::rebuild_cpu_to_node_map() {
2846   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2847                               // in libnuma (possible values are starting from 16,
2848                               // and continuing up with every other power of 2, but less
2849                               // than the maximum number of CPUs supported by kernel), and
2850                               // is a subject to change (in libnuma version 2 the requirements
2851                               // are more reasonable) we'll just hardcode the number they use
2852                               // in the library.
2853   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2854 
2855   size_t cpu_num = processor_count();
2856   size_t cpu_map_size = NCPUS / BitsPerCLong;
2857   size_t cpu_map_valid_size =
2858     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2859 
2860   cpu_to_node()->clear();
2861   cpu_to_node()->at_grow(cpu_num - 1);
2862   size_t node_num = numa_get_groups_num();
2863 
2864   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2865   for (size_t i = 0; i < node_num; i++) {
2866     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2867       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2868         if (cpu_map[j] != 0) {
2869           for (size_t k = 0; k < BitsPerCLong; k++) {
2870             if (cpu_map[j] & (1UL << k)) {
2871               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2872             }
2873           }
2874         }
2875       }
2876     }
2877   }
2878   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2879 }
2880 
2881 int os::Linux::get_node_by_cpu(int cpu_id) {
2882   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2883     return cpu_to_node()->at(cpu_id);
2884   }
2885   return -1;
2886 }
2887 
2888 GrowableArray<int>* os::Linux::_cpu_to_node;
2889 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2890 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2891 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2892 os::Linux::numa_available_func_t os::Linux::_numa_available;
2893 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2894 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2895 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2896 unsigned long* os::Linux::_numa_all_nodes;
2897 
2898 bool os::pd_uncommit_memory(char* addr, size_t size) {
2899   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2900                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2901   return res  != (uintptr_t) MAP_FAILED;
2902 }
2903 
2904 static address get_stack_commited_bottom(address bottom, size_t size) {
2905   address nbot = bottom;
2906   address ntop = bottom + size;
2907 
2908   size_t page_sz = os::vm_page_size();
2909   unsigned pages = size / page_sz;
2910 
2911   unsigned char vec[1];
2912   unsigned imin = 1, imax = pages + 1, imid;
2913   int mincore_return_value = 0;
2914 
2915   assert(imin <= imax, "Unexpected page size");
2916 
2917   while (imin < imax) {
2918     imid = (imax + imin) / 2;
2919     nbot = ntop - (imid * page_sz);
2920 
2921     // Use a trick with mincore to check whether the page is mapped or not.
2922     // mincore sets vec to 1 if page resides in memory and to 0 if page
2923     // is swapped output but if page we are asking for is unmapped
2924     // it returns -1,ENOMEM
2925     mincore_return_value = mincore(nbot, page_sz, vec);
2926 
2927     if (mincore_return_value == -1) {
2928       // Page is not mapped go up
2929       // to find first mapped page
2930       if (errno != EAGAIN) {
2931         assert(errno == ENOMEM, "Unexpected mincore errno");
2932         imax = imid;
2933       }
2934     } else {
2935       // Page is mapped go down
2936       // to find first not mapped page
2937       imin = imid + 1;
2938     }
2939   }
2940 
2941   nbot = nbot + page_sz;
2942 
2943   // Adjust stack bottom one page up if last checked page is not mapped
2944   if (mincore_return_value == -1) {
2945     nbot = nbot + page_sz;
2946   }
2947 
2948   return nbot;
2949 }
2950 
2951 
2952 // Linux uses a growable mapping for the stack, and if the mapping for
2953 // the stack guard pages is not removed when we detach a thread the
2954 // stack cannot grow beyond the pages where the stack guard was
2955 // mapped.  If at some point later in the process the stack expands to
2956 // that point, the Linux kernel cannot expand the stack any further
2957 // because the guard pages are in the way, and a segfault occurs.
2958 //
2959 // However, it's essential not to split the stack region by unmapping
2960 // a region (leaving a hole) that's already part of the stack mapping,
2961 // so if the stack mapping has already grown beyond the guard pages at
2962 // the time we create them, we have to truncate the stack mapping.
2963 // So, we need to know the extent of the stack mapping when
2964 // create_stack_guard_pages() is called.
2965 
2966 // We only need this for stacks that are growable: at the time of
2967 // writing thread stacks don't use growable mappings (i.e. those
2968 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2969 // only applies to the main thread.
2970 
2971 // If the (growable) stack mapping already extends beyond the point
2972 // where we're going to put our guard pages, truncate the mapping at
2973 // that point by munmap()ping it.  This ensures that when we later
2974 // munmap() the guard pages we don't leave a hole in the stack
2975 // mapping. This only affects the main/initial thread
2976 
2977 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2978   if (os::Linux::is_initial_thread()) {
2979     // As we manually grow stack up to bottom inside create_attached_thread(),
2980     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2981     // we don't need to do anything special.
2982     // Check it first, before calling heavy function.
2983     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2984     unsigned char vec[1];
2985 
2986     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2987       // Fallback to slow path on all errors, including EAGAIN
2988       stack_extent = (uintptr_t) get_stack_commited_bottom(
2989                                                            os::Linux::initial_thread_stack_bottom(),
2990                                                            (size_t)addr - stack_extent);
2991     }
2992 
2993     if (stack_extent < (uintptr_t)addr) {
2994       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2995     }
2996   }
2997 
2998   return os::commit_memory(addr, size, !ExecMem);
2999 }
3000 
3001 // If this is a growable mapping, remove the guard pages entirely by
3002 // munmap()ping them.  If not, just call uncommit_memory(). This only
3003 // affects the main/initial thread, but guard against future OS changes
3004 // It's safe to always unmap guard pages for initial thread because we
3005 // always place it right after end of the mapped region
3006 
3007 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3008   uintptr_t stack_extent, stack_base;
3009 
3010   if (os::Linux::is_initial_thread()) {
3011     return ::munmap(addr, size) == 0;
3012   }
3013 
3014   return os::uncommit_memory(addr, size);
3015 }
3016 
3017 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3018 // at 'requested_addr'. If there are existing memory mappings at the same
3019 // location, however, they will be overwritten. If 'fixed' is false,
3020 // 'requested_addr' is only treated as a hint, the return value may or
3021 // may not start from the requested address. Unlike Linux mmap(), this
3022 // function returns NULL to indicate failure.
3023 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3024   char * addr;
3025   int flags;
3026 
3027   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3028   if (fixed) {
3029     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3030     flags |= MAP_FIXED;
3031   }
3032 
3033   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3034   // touch an uncommitted page. Otherwise, the read/write might
3035   // succeed if we have enough swap space to back the physical page.
3036   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3037                        flags, -1, 0);
3038 
3039   return addr == MAP_FAILED ? NULL : addr;
3040 }
3041 
3042 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3043 //   (req_addr != NULL) or with a given alignment.
3044 //  - bytes shall be a multiple of alignment.
3045 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3046 //  - alignment sets the alignment at which memory shall be allocated.
3047 //     It must be a multiple of allocation granularity.
3048 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3049 //  req_addr or NULL.
3050 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3051 
3052   size_t extra_size = bytes;
3053   if (req_addr == NULL && alignment > 0) {
3054     extra_size += alignment;
3055   }
3056 
3057   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3058     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3059     -1, 0);
3060   if (start == MAP_FAILED) {
3061     start = NULL;
3062   } else {
3063     if (req_addr != NULL) {
3064       if (start != req_addr) {
3065         ::munmap(start, extra_size);
3066         start = NULL;
3067       }
3068     } else {
3069       char* const start_aligned = (char*) align_ptr_up(start, alignment);
3070       char* const end_aligned = start_aligned + bytes;
3071       char* const end = start + extra_size;
3072       if (start_aligned > start) {
3073         ::munmap(start, start_aligned - start);
3074       }
3075       if (end_aligned < end) {
3076         ::munmap(end_aligned, end - end_aligned);
3077       }
3078       start = start_aligned;
3079     }
3080   }
3081   return start;
3082 }
3083 
3084 static int anon_munmap(char * addr, size_t size) {
3085   return ::munmap(addr, size) == 0;
3086 }
3087 
3088 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3089                             size_t alignment_hint) {
3090   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3091 }
3092 
3093 bool os::pd_release_memory(char* addr, size_t size) {
3094   return anon_munmap(addr, size);
3095 }
3096 
3097 static bool linux_mprotect(char* addr, size_t size, int prot) {
3098   // Linux wants the mprotect address argument to be page aligned.
3099   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3100 
3101   // According to SUSv3, mprotect() should only be used with mappings
3102   // established by mmap(), and mmap() always maps whole pages. Unaligned
3103   // 'addr' likely indicates problem in the VM (e.g. trying to change
3104   // protection of malloc'ed or statically allocated memory). Check the
3105   // caller if you hit this assert.
3106   assert(addr == bottom, "sanity check");
3107 
3108   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3109   return ::mprotect(bottom, size, prot) == 0;
3110 }
3111 
3112 // Set protections specified
3113 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3114                         bool is_committed) {
3115   unsigned int p = 0;
3116   switch (prot) {
3117   case MEM_PROT_NONE: p = PROT_NONE; break;
3118   case MEM_PROT_READ: p = PROT_READ; break;
3119   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3120   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3121   default:
3122     ShouldNotReachHere();
3123   }
3124   // is_committed is unused.
3125   return linux_mprotect(addr, bytes, p);
3126 }
3127 
3128 bool os::guard_memory(char* addr, size_t size) {
3129   return linux_mprotect(addr, size, PROT_NONE);
3130 }
3131 
3132 bool os::unguard_memory(char* addr, size_t size) {
3133   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3134 }
3135 
3136 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3137                                                     size_t page_size) {
3138   bool result = false;
3139   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3140                  MAP_ANONYMOUS|MAP_PRIVATE,
3141                  -1, 0);
3142   if (p != MAP_FAILED) {
3143     void *aligned_p = align_ptr_up(p, page_size);
3144 
3145     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3146 
3147     munmap(p, page_size * 2);
3148   }
3149 
3150   if (warn && !result) {
3151     warning("TransparentHugePages is not supported by the operating system.");
3152   }
3153 
3154   return result;
3155 }
3156 
3157 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3158   bool result = false;
3159   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3160                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3161                  -1, 0);
3162 
3163   if (p != MAP_FAILED) {
3164     // We don't know if this really is a huge page or not.
3165     FILE *fp = fopen("/proc/self/maps", "r");
3166     if (fp) {
3167       while (!feof(fp)) {
3168         char chars[257];
3169         long x = 0;
3170         if (fgets(chars, sizeof(chars), fp)) {
3171           if (sscanf(chars, "%lx-%*x", &x) == 1
3172               && x == (long)p) {
3173             if (strstr (chars, "hugepage")) {
3174               result = true;
3175               break;
3176             }
3177           }
3178         }
3179       }
3180       fclose(fp);
3181     }
3182     munmap(p, page_size);
3183   }
3184 
3185   if (warn && !result) {
3186     warning("HugeTLBFS is not supported by the operating system.");
3187   }
3188 
3189   return result;
3190 }
3191 
3192 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3193 //
3194 // From the coredump_filter documentation:
3195 //
3196 // - (bit 0) anonymous private memory
3197 // - (bit 1) anonymous shared memory
3198 // - (bit 2) file-backed private memory
3199 // - (bit 3) file-backed shared memory
3200 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3201 //           effective only if the bit 2 is cleared)
3202 // - (bit 5) hugetlb private memory
3203 // - (bit 6) hugetlb shared memory
3204 //
3205 static void set_coredump_filter(void) {
3206   FILE *f;
3207   long cdm;
3208 
3209   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3210     return;
3211   }
3212 
3213   if (fscanf(f, "%lx", &cdm) != 1) {
3214     fclose(f);
3215     return;
3216   }
3217 
3218   rewind(f);
3219 
3220   if ((cdm & LARGEPAGES_BIT) == 0) {
3221     cdm |= LARGEPAGES_BIT;
3222     fprintf(f, "%#lx", cdm);
3223   }
3224 
3225   fclose(f);
3226 }
3227 
3228 // Large page support
3229 
3230 static size_t _large_page_size = 0;
3231 
3232 size_t os::Linux::find_large_page_size() {
3233   size_t large_page_size = 0;
3234 
3235   // large_page_size on Linux is used to round up heap size. x86 uses either
3236   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3237   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3238   // page as large as 256M.
3239   //
3240   // Here we try to figure out page size by parsing /proc/meminfo and looking
3241   // for a line with the following format:
3242   //    Hugepagesize:     2048 kB
3243   //
3244   // If we can't determine the value (e.g. /proc is not mounted, or the text
3245   // format has been changed), we'll use the largest page size supported by
3246   // the processor.
3247 
3248 #ifndef ZERO
3249   large_page_size =
3250     AARCH64_ONLY(2 * M)
3251     AMD64_ONLY(2 * M)
3252     ARM32_ONLY(2 * M)
3253     IA32_ONLY(4 * M)
3254     IA64_ONLY(256 * M)
3255     PPC_ONLY(4 * M)
3256     S390_ONLY(1 * M)
3257     SPARC_ONLY(4 * M);
3258 #endif // ZERO
3259 
3260   FILE *fp = fopen("/proc/meminfo", "r");
3261   if (fp) {
3262     while (!feof(fp)) {
3263       int x = 0;
3264       char buf[16];
3265       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3266         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3267           large_page_size = x * K;
3268           break;
3269         }
3270       } else {
3271         // skip to next line
3272         for (;;) {
3273           int ch = fgetc(fp);
3274           if (ch == EOF || ch == (int)'\n') break;
3275         }
3276       }
3277     }
3278     fclose(fp);
3279   }
3280 
3281   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3282     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3283             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3284             proper_unit_for_byte_size(large_page_size));
3285   }
3286 
3287   return large_page_size;
3288 }
3289 
3290 size_t os::Linux::setup_large_page_size() {
3291   _large_page_size = Linux::find_large_page_size();
3292   const size_t default_page_size = (size_t)Linux::page_size();
3293   if (_large_page_size > default_page_size) {
3294     _page_sizes[0] = _large_page_size;
3295     _page_sizes[1] = default_page_size;
3296     _page_sizes[2] = 0;
3297   }
3298 
3299   return _large_page_size;
3300 }
3301 
3302 bool os::Linux::setup_large_page_type(size_t page_size) {
3303   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3304       FLAG_IS_DEFAULT(UseSHM) &&
3305       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3306 
3307     // The type of large pages has not been specified by the user.
3308 
3309     // Try UseHugeTLBFS and then UseSHM.
3310     UseHugeTLBFS = UseSHM = true;
3311 
3312     // Don't try UseTransparentHugePages since there are known
3313     // performance issues with it turned on. This might change in the future.
3314     UseTransparentHugePages = false;
3315   }
3316 
3317   if (UseTransparentHugePages) {
3318     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3319     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3320       UseHugeTLBFS = false;
3321       UseSHM = false;
3322       return true;
3323     }
3324     UseTransparentHugePages = false;
3325   }
3326 
3327   if (UseHugeTLBFS) {
3328     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3329     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3330       UseSHM = false;
3331       return true;
3332     }
3333     UseHugeTLBFS = false;
3334   }
3335 
3336   return UseSHM;
3337 }
3338 
3339 void os::large_page_init() {
3340   if (!UseLargePages &&
3341       !UseTransparentHugePages &&
3342       !UseHugeTLBFS &&
3343       !UseSHM) {
3344     // Not using large pages.
3345     return;
3346   }
3347 
3348   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3349     // The user explicitly turned off large pages.
3350     // Ignore the rest of the large pages flags.
3351     UseTransparentHugePages = false;
3352     UseHugeTLBFS = false;
3353     UseSHM = false;
3354     return;
3355   }
3356 
3357   size_t large_page_size = Linux::setup_large_page_size();
3358   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3359 
3360   set_coredump_filter();
3361 }
3362 
3363 #ifndef SHM_HUGETLB
3364   #define SHM_HUGETLB 04000
3365 #endif
3366 
3367 #define shm_warning_format(format, ...)              \
3368   do {                                               \
3369     if (UseLargePages &&                             \
3370         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3371          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3372          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3373       warning(format, __VA_ARGS__);                  \
3374     }                                                \
3375   } while (0)
3376 
3377 #define shm_warning(str) shm_warning_format("%s", str)
3378 
3379 #define shm_warning_with_errno(str)                \
3380   do {                                             \
3381     int err = errno;                               \
3382     shm_warning_format(str " (error = %d)", err);  \
3383   } while (0)
3384 
3385 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3386   assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3387 
3388   if (!is_size_aligned(alignment, SHMLBA)) {
3389     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3390     return NULL;
3391   }
3392 
3393   // To ensure that we get 'alignment' aligned memory from shmat,
3394   // we pre-reserve aligned virtual memory and then attach to that.
3395 
3396   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3397   if (pre_reserved_addr == NULL) {
3398     // Couldn't pre-reserve aligned memory.
3399     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3400     return NULL;
3401   }
3402 
3403   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3404   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3405 
3406   if ((intptr_t)addr == -1) {
3407     int err = errno;
3408     shm_warning_with_errno("Failed to attach shared memory.");
3409 
3410     assert(err != EACCES, "Unexpected error");
3411     assert(err != EIDRM,  "Unexpected error");
3412     assert(err != EINVAL, "Unexpected error");
3413 
3414     // Since we don't know if the kernel unmapped the pre-reserved memory area
3415     // we can't unmap it, since that would potentially unmap memory that was
3416     // mapped from other threads.
3417     return NULL;
3418   }
3419 
3420   return addr;
3421 }
3422 
3423 static char* shmat_at_address(int shmid, char* req_addr) {
3424   if (!is_ptr_aligned(req_addr, SHMLBA)) {
3425     assert(false, "Requested address needs to be SHMLBA aligned");
3426     return NULL;
3427   }
3428 
3429   char* addr = (char*)shmat(shmid, req_addr, 0);
3430 
3431   if ((intptr_t)addr == -1) {
3432     shm_warning_with_errno("Failed to attach shared memory.");
3433     return NULL;
3434   }
3435 
3436   return addr;
3437 }
3438 
3439 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3440   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3441   if (req_addr != NULL) {
3442     assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3443     assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3444     return shmat_at_address(shmid, req_addr);
3445   }
3446 
3447   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3448   // return large page size aligned memory addresses when req_addr == NULL.
3449   // However, if the alignment is larger than the large page size, we have
3450   // to manually ensure that the memory returned is 'alignment' aligned.
3451   if (alignment > os::large_page_size()) {
3452     assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3453     return shmat_with_alignment(shmid, bytes, alignment);
3454   } else {
3455     return shmat_at_address(shmid, NULL);
3456   }
3457 }
3458 
3459 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3460                                             char* req_addr, bool exec) {
3461   // "exec" is passed in but not used.  Creating the shared image for
3462   // the code cache doesn't have an SHM_X executable permission to check.
3463   assert(UseLargePages && UseSHM, "only for SHM large pages");
3464   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3465   assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3466 
3467   if (!is_size_aligned(bytes, os::large_page_size())) {
3468     return NULL; // Fallback to small pages.
3469   }
3470 
3471   // Create a large shared memory region to attach to based on size.
3472   // Currently, size is the total size of the heap.
3473   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3474   if (shmid == -1) {
3475     // Possible reasons for shmget failure:
3476     // 1. shmmax is too small for Java heap.
3477     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3478     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3479     // 2. not enough large page memory.
3480     //    > check available large pages: cat /proc/meminfo
3481     //    > increase amount of large pages:
3482     //          echo new_value > /proc/sys/vm/nr_hugepages
3483     //      Note 1: different Linux may use different name for this property,
3484     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3485     //      Note 2: it's possible there's enough physical memory available but
3486     //            they are so fragmented after a long run that they can't
3487     //            coalesce into large pages. Try to reserve large pages when
3488     //            the system is still "fresh".
3489     shm_warning_with_errno("Failed to reserve shared memory.");
3490     return NULL;
3491   }
3492 
3493   // Attach to the region.
3494   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3495 
3496   // Remove shmid. If shmat() is successful, the actual shared memory segment
3497   // will be deleted when it's detached by shmdt() or when the process
3498   // terminates. If shmat() is not successful this will remove the shared
3499   // segment immediately.
3500   shmctl(shmid, IPC_RMID, NULL);
3501 
3502   return addr;
3503 }
3504 
3505 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3506                                         int error) {
3507   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3508 
3509   bool warn_on_failure = UseLargePages &&
3510       (!FLAG_IS_DEFAULT(UseLargePages) ||
3511        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3512        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3513 
3514   if (warn_on_failure) {
3515     char msg[128];
3516     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3517                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3518     warning("%s", msg);
3519   }
3520 }
3521 
3522 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3523                                                         char* req_addr,
3524                                                         bool exec) {
3525   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3526   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3527   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3528 
3529   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3530   char* addr = (char*)::mmap(req_addr, bytes, prot,
3531                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3532                              -1, 0);
3533 
3534   if (addr == MAP_FAILED) {
3535     warn_on_large_pages_failure(req_addr, bytes, errno);
3536     return NULL;
3537   }
3538 
3539   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3540 
3541   return addr;
3542 }
3543 
3544 // Reserve memory using mmap(MAP_HUGETLB).
3545 //  - bytes shall be a multiple of alignment.
3546 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3547 //  - alignment sets the alignment at which memory shall be allocated.
3548 //     It must be a multiple of allocation granularity.
3549 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3550 //  req_addr or NULL.
3551 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3552                                                          size_t alignment,
3553                                                          char* req_addr,
3554                                                          bool exec) {
3555   size_t large_page_size = os::large_page_size();
3556   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3557 
3558   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3559   assert(is_size_aligned(bytes, alignment), "Must be");
3560 
3561   // First reserve - but not commit - the address range in small pages.
3562   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3563 
3564   if (start == NULL) {
3565     return NULL;
3566   }
3567 
3568   assert(is_ptr_aligned(start, alignment), "Must be");
3569 
3570   char* end = start + bytes;
3571 
3572   // Find the regions of the allocated chunk that can be promoted to large pages.
3573   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3574   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3575 
3576   size_t lp_bytes = lp_end - lp_start;
3577 
3578   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3579 
3580   if (lp_bytes == 0) {
3581     // The mapped region doesn't even span the start and the end of a large page.
3582     // Fall back to allocate a non-special area.
3583     ::munmap(start, end - start);
3584     return NULL;
3585   }
3586 
3587   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3588 
3589   void* result;
3590 
3591   // Commit small-paged leading area.
3592   if (start != lp_start) {
3593     result = ::mmap(start, lp_start - start, prot,
3594                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3595                     -1, 0);
3596     if (result == MAP_FAILED) {
3597       ::munmap(lp_start, end - lp_start);
3598       return NULL;
3599     }
3600   }
3601 
3602   // Commit large-paged area.
3603   result = ::mmap(lp_start, lp_bytes, prot,
3604                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3605                   -1, 0);
3606   if (result == MAP_FAILED) {
3607     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3608     // If the mmap above fails, the large pages region will be unmapped and we
3609     // have regions before and after with small pages. Release these regions.
3610     //
3611     // |  mapped  |  unmapped  |  mapped  |
3612     // ^          ^            ^          ^
3613     // start      lp_start     lp_end     end
3614     //
3615     ::munmap(start, lp_start - start);
3616     ::munmap(lp_end, end - lp_end);
3617     return NULL;
3618   }
3619 
3620   // Commit small-paged trailing area.
3621   if (lp_end != end) {
3622     result = ::mmap(lp_end, end - lp_end, prot,
3623                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3624                     -1, 0);
3625     if (result == MAP_FAILED) {
3626       ::munmap(start, lp_end - start);
3627       return NULL;
3628     }
3629   }
3630 
3631   return start;
3632 }
3633 
3634 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3635                                                    size_t alignment,
3636                                                    char* req_addr,
3637                                                    bool exec) {
3638   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3639   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3640   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3641   assert(is_power_of_2(os::large_page_size()), "Must be");
3642   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3643 
3644   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3645     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3646   } else {
3647     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3648   }
3649 }
3650 
3651 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3652                                  char* req_addr, bool exec) {
3653   assert(UseLargePages, "only for large pages");
3654 
3655   char* addr;
3656   if (UseSHM) {
3657     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3658   } else {
3659     assert(UseHugeTLBFS, "must be");
3660     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3661   }
3662 
3663   if (addr != NULL) {
3664     if (UseNUMAInterleaving) {
3665       numa_make_global(addr, bytes);
3666     }
3667 
3668     // The memory is committed
3669     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3670   }
3671 
3672   return addr;
3673 }
3674 
3675 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3676   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3677   return shmdt(base) == 0;
3678 }
3679 
3680 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3681   return pd_release_memory(base, bytes);
3682 }
3683 
3684 bool os::release_memory_special(char* base, size_t bytes) {
3685   bool res;
3686   if (MemTracker::tracking_level() > NMT_minimal) {
3687     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3688     res = os::Linux::release_memory_special_impl(base, bytes);
3689     if (res) {
3690       tkr.record((address)base, bytes);
3691     }
3692 
3693   } else {
3694     res = os::Linux::release_memory_special_impl(base, bytes);
3695   }
3696   return res;
3697 }
3698 
3699 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3700   assert(UseLargePages, "only for large pages");
3701   bool res;
3702 
3703   if (UseSHM) {
3704     res = os::Linux::release_memory_special_shm(base, bytes);
3705   } else {
3706     assert(UseHugeTLBFS, "must be");
3707     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3708   }
3709   return res;
3710 }
3711 
3712 size_t os::large_page_size() {
3713   return _large_page_size;
3714 }
3715 
3716 // With SysV SHM the entire memory region must be allocated as shared
3717 // memory.
3718 // HugeTLBFS allows application to commit large page memory on demand.
3719 // However, when committing memory with HugeTLBFS fails, the region
3720 // that was supposed to be committed will lose the old reservation
3721 // and allow other threads to steal that memory region. Because of this
3722 // behavior we can't commit HugeTLBFS memory.
3723 bool os::can_commit_large_page_memory() {
3724   return UseTransparentHugePages;
3725 }
3726 
3727 bool os::can_execute_large_page_memory() {
3728   return UseTransparentHugePages || UseHugeTLBFS;
3729 }
3730 
3731 // Reserve memory at an arbitrary address, only if that area is
3732 // available (and not reserved for something else).
3733 
3734 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3735   const int max_tries = 10;
3736   char* base[max_tries];
3737   size_t size[max_tries];
3738   const size_t gap = 0x000000;
3739 
3740   // Assert only that the size is a multiple of the page size, since
3741   // that's all that mmap requires, and since that's all we really know
3742   // about at this low abstraction level.  If we need higher alignment,
3743   // we can either pass an alignment to this method or verify alignment
3744   // in one of the methods further up the call chain.  See bug 5044738.
3745   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3746 
3747   // Repeatedly allocate blocks until the block is allocated at the
3748   // right spot.
3749 
3750   // Linux mmap allows caller to pass an address as hint; give it a try first,
3751   // if kernel honors the hint then we can return immediately.
3752   char * addr = anon_mmap(requested_addr, bytes, false);
3753   if (addr == requested_addr) {
3754     return requested_addr;
3755   }
3756 
3757   if (addr != NULL) {
3758     // mmap() is successful but it fails to reserve at the requested address
3759     anon_munmap(addr, bytes);
3760   }
3761 
3762   int i;
3763   for (i = 0; i < max_tries; ++i) {
3764     base[i] = reserve_memory(bytes);
3765 
3766     if (base[i] != NULL) {
3767       // Is this the block we wanted?
3768       if (base[i] == requested_addr) {
3769         size[i] = bytes;
3770         break;
3771       }
3772 
3773       // Does this overlap the block we wanted? Give back the overlapped
3774       // parts and try again.
3775 
3776       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3777       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3778         unmap_memory(base[i], top_overlap);
3779         base[i] += top_overlap;
3780         size[i] = bytes - top_overlap;
3781       } else {
3782         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3783         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3784           unmap_memory(requested_addr, bottom_overlap);
3785           size[i] = bytes - bottom_overlap;
3786         } else {
3787           size[i] = bytes;
3788         }
3789       }
3790     }
3791   }
3792 
3793   // Give back the unused reserved pieces.
3794 
3795   for (int j = 0; j < i; ++j) {
3796     if (base[j] != NULL) {
3797       unmap_memory(base[j], size[j]);
3798     }
3799   }
3800 
3801   if (i < max_tries) {
3802     return requested_addr;
3803   } else {
3804     return NULL;
3805   }
3806 }
3807 
3808 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3809   return ::read(fd, buf, nBytes);
3810 }
3811 
3812 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3813   return ::pread(fd, buf, nBytes, offset);
3814 }
3815 
3816 // Short sleep, direct OS call.
3817 //
3818 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3819 // sched_yield(2) will actually give up the CPU:
3820 //
3821 //   * Alone on this pariticular CPU, keeps running.
3822 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3823 //     (pre 2.6.39).
3824 //
3825 // So calling this with 0 is an alternative.
3826 //
3827 void os::naked_short_sleep(jlong ms) {
3828   struct timespec req;
3829 
3830   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3831   req.tv_sec = 0;
3832   if (ms > 0) {
3833     req.tv_nsec = (ms % 1000) * 1000000;
3834   } else {
3835     req.tv_nsec = 1;
3836   }
3837 
3838   nanosleep(&req, NULL);
3839 
3840   return;
3841 }
3842 
3843 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3844 void os::infinite_sleep() {
3845   while (true) {    // sleep forever ...
3846     ::sleep(100);   // ... 100 seconds at a time
3847   }
3848 }
3849 
3850 // Used to convert frequent JVM_Yield() to nops
3851 bool os::dont_yield() {
3852   return DontYieldALot;
3853 }
3854 
3855 void os::naked_yield() {
3856   sched_yield();
3857 }
3858 
3859 ////////////////////////////////////////////////////////////////////////////////
3860 // thread priority support
3861 
3862 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3863 // only supports dynamic priority, static priority must be zero. For real-time
3864 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3865 // However, for large multi-threaded applications, SCHED_RR is not only slower
3866 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3867 // of 5 runs - Sep 2005).
3868 //
3869 // The following code actually changes the niceness of kernel-thread/LWP. It
3870 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3871 // not the entire user process, and user level threads are 1:1 mapped to kernel
3872 // threads. It has always been the case, but could change in the future. For
3873 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3874 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3875 
3876 int os::java_to_os_priority[CriticalPriority + 1] = {
3877   19,              // 0 Entry should never be used
3878 
3879    4,              // 1 MinPriority
3880    3,              // 2
3881    2,              // 3
3882 
3883    1,              // 4
3884    0,              // 5 NormPriority
3885   -1,              // 6
3886 
3887   -2,              // 7
3888   -3,              // 8
3889   -4,              // 9 NearMaxPriority
3890 
3891   -5,              // 10 MaxPriority
3892 
3893   -5               // 11 CriticalPriority
3894 };
3895 
3896 static int prio_init() {
3897   if (ThreadPriorityPolicy == 1) {
3898     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3899     // if effective uid is not root. Perhaps, a more elegant way of doing
3900     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3901     if (geteuid() != 0) {
3902       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3903         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3904       }
3905       ThreadPriorityPolicy = 0;
3906     }
3907   }
3908   if (UseCriticalJavaThreadPriority) {
3909     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3910   }
3911   return 0;
3912 }
3913 
3914 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3915   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3916 
3917   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3918   return (ret == 0) ? OS_OK : OS_ERR;
3919 }
3920 
3921 OSReturn os::get_native_priority(const Thread* const thread,
3922                                  int *priority_ptr) {
3923   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3924     *priority_ptr = java_to_os_priority[NormPriority];
3925     return OS_OK;
3926   }
3927 
3928   errno = 0;
3929   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3930   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3931 }
3932 
3933 // Hint to the underlying OS that a task switch would not be good.
3934 // Void return because it's a hint and can fail.
3935 void os::hint_no_preempt() {}
3936 
3937 ////////////////////////////////////////////////////////////////////////////////
3938 // suspend/resume support
3939 
3940 //  the low-level signal-based suspend/resume support is a remnant from the
3941 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3942 //  within hotspot. Now there is a single use-case for this:
3943 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3944 //      that runs in the watcher thread.
3945 //  The remaining code is greatly simplified from the more general suspension
3946 //  code that used to be used.
3947 //
3948 //  The protocol is quite simple:
3949 //  - suspend:
3950 //      - sends a signal to the target thread
3951 //      - polls the suspend state of the osthread using a yield loop
3952 //      - target thread signal handler (SR_handler) sets suspend state
3953 //        and blocks in sigsuspend until continued
3954 //  - resume:
3955 //      - sets target osthread state to continue
3956 //      - sends signal to end the sigsuspend loop in the SR_handler
3957 //
3958 //  Note that the SR_lock plays no role in this suspend/resume protocol,
3959 //  but is checked for NULL in SR_handler as a thread termination indicator.
3960 
3961 static void resume_clear_context(OSThread *osthread) {
3962   osthread->set_ucontext(NULL);
3963   osthread->set_siginfo(NULL);
3964 }
3965 
3966 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3967                                  ucontext_t* context) {
3968   osthread->set_ucontext(context);
3969   osthread->set_siginfo(siginfo);
3970 }
3971 
3972 // Handler function invoked when a thread's execution is suspended or
3973 // resumed. We have to be careful that only async-safe functions are
3974 // called here (Note: most pthread functions are not async safe and
3975 // should be avoided.)
3976 //
3977 // Note: sigwait() is a more natural fit than sigsuspend() from an
3978 // interface point of view, but sigwait() prevents the signal hander
3979 // from being run. libpthread would get very confused by not having
3980 // its signal handlers run and prevents sigwait()'s use with the
3981 // mutex granting granting signal.
3982 //
3983 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3984 //
3985 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3986   // Save and restore errno to avoid confusing native code with EINTR
3987   // after sigsuspend.
3988   int old_errno = errno;
3989 
3990   Thread* thread = Thread::current_or_null_safe();
3991   assert(thread != NULL, "Missing current thread in SR_handler");
3992 
3993   // On some systems we have seen signal delivery get "stuck" until the signal
3994   // mask is changed as part of thread termination. Check that the current thread
3995   // has not already terminated (via SR_lock()) - else the following assertion
3996   // will fail because the thread is no longer a JavaThread as the ~JavaThread
3997   // destructor has completed.
3998 
3999   if (thread->SR_lock() == NULL) {
4000     return;
4001   }
4002 
4003   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4004 
4005   OSThread* osthread = thread->osthread();
4006 
4007   os::SuspendResume::State current = osthread->sr.state();
4008   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4009     suspend_save_context(osthread, siginfo, context);
4010 
4011     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4012     os::SuspendResume::State state = osthread->sr.suspended();
4013     if (state == os::SuspendResume::SR_SUSPENDED) {
4014       sigset_t suspend_set;  // signals for sigsuspend()
4015       sigemptyset(&suspend_set);
4016       // get current set of blocked signals and unblock resume signal
4017       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4018       sigdelset(&suspend_set, SR_signum);
4019 
4020       sr_semaphore.signal();
4021       // wait here until we are resumed
4022       while (1) {
4023         sigsuspend(&suspend_set);
4024 
4025         os::SuspendResume::State result = osthread->sr.running();
4026         if (result == os::SuspendResume::SR_RUNNING) {
4027           sr_semaphore.signal();
4028           break;
4029         }
4030       }
4031 
4032     } else if (state == os::SuspendResume::SR_RUNNING) {
4033       // request was cancelled, continue
4034     } else {
4035       ShouldNotReachHere();
4036     }
4037 
4038     resume_clear_context(osthread);
4039   } else if (current == os::SuspendResume::SR_RUNNING) {
4040     // request was cancelled, continue
4041   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4042     // ignore
4043   } else {
4044     // ignore
4045   }
4046 
4047   errno = old_errno;
4048 }
4049 
4050 static int SR_initialize() {
4051   struct sigaction act;
4052   char *s;
4053 
4054   // Get signal number to use for suspend/resume
4055   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4056     int sig = ::strtol(s, 0, 10);
4057     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4058         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4059       SR_signum = sig;
4060     } else {
4061       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4062               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4063     }
4064   }
4065 
4066   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4067          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4068 
4069   sigemptyset(&SR_sigset);
4070   sigaddset(&SR_sigset, SR_signum);
4071 
4072   // Set up signal handler for suspend/resume
4073   act.sa_flags = SA_RESTART|SA_SIGINFO;
4074   act.sa_handler = (void (*)(int)) SR_handler;
4075 
4076   // SR_signum is blocked by default.
4077   // 4528190 - We also need to block pthread restart signal (32 on all
4078   // supported Linux platforms). Note that LinuxThreads need to block
4079   // this signal for all threads to work properly. So we don't have
4080   // to use hard-coded signal number when setting up the mask.
4081   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4082 
4083   if (sigaction(SR_signum, &act, 0) == -1) {
4084     return -1;
4085   }
4086 
4087   // Save signal flag
4088   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4089   return 0;
4090 }
4091 
4092 static int sr_notify(OSThread* osthread) {
4093   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4094   assert_status(status == 0, status, "pthread_kill");
4095   return status;
4096 }
4097 
4098 // "Randomly" selected value for how long we want to spin
4099 // before bailing out on suspending a thread, also how often
4100 // we send a signal to a thread we want to resume
4101 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4102 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4103 
4104 // returns true on success and false on error - really an error is fatal
4105 // but this seems the normal response to library errors
4106 static bool do_suspend(OSThread* osthread) {
4107   assert(osthread->sr.is_running(), "thread should be running");
4108   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4109 
4110   // mark as suspended and send signal
4111   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4112     // failed to switch, state wasn't running?
4113     ShouldNotReachHere();
4114     return false;
4115   }
4116 
4117   if (sr_notify(osthread) != 0) {
4118     ShouldNotReachHere();
4119   }
4120 
4121   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4122   while (true) {
4123     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4124       break;
4125     } else {
4126       // timeout
4127       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4128       if (cancelled == os::SuspendResume::SR_RUNNING) {
4129         return false;
4130       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4131         // make sure that we consume the signal on the semaphore as well
4132         sr_semaphore.wait();
4133         break;
4134       } else {
4135         ShouldNotReachHere();
4136         return false;
4137       }
4138     }
4139   }
4140 
4141   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4142   return true;
4143 }
4144 
4145 static void do_resume(OSThread* osthread) {
4146   assert(osthread->sr.is_suspended(), "thread should be suspended");
4147   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4148 
4149   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4150     // failed to switch to WAKEUP_REQUEST
4151     ShouldNotReachHere();
4152     return;
4153   }
4154 
4155   while (true) {
4156     if (sr_notify(osthread) == 0) {
4157       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4158         if (osthread->sr.is_running()) {
4159           return;
4160         }
4161       }
4162     } else {
4163       ShouldNotReachHere();
4164     }
4165   }
4166 
4167   guarantee(osthread->sr.is_running(), "Must be running!");
4168 }
4169 
4170 ///////////////////////////////////////////////////////////////////////////////////
4171 // signal handling (except suspend/resume)
4172 
4173 // This routine may be used by user applications as a "hook" to catch signals.
4174 // The user-defined signal handler must pass unrecognized signals to this
4175 // routine, and if it returns true (non-zero), then the signal handler must
4176 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4177 // routine will never retun false (zero), but instead will execute a VM panic
4178 // routine kill the process.
4179 //
4180 // If this routine returns false, it is OK to call it again.  This allows
4181 // the user-defined signal handler to perform checks either before or after
4182 // the VM performs its own checks.  Naturally, the user code would be making
4183 // a serious error if it tried to handle an exception (such as a null check
4184 // or breakpoint) that the VM was generating for its own correct operation.
4185 //
4186 // This routine may recognize any of the following kinds of signals:
4187 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4188 // It should be consulted by handlers for any of those signals.
4189 //
4190 // The caller of this routine must pass in the three arguments supplied
4191 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4192 // field of the structure passed to sigaction().  This routine assumes that
4193 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4194 //
4195 // Note that the VM will print warnings if it detects conflicting signal
4196 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4197 //
4198 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4199                                                  siginfo_t* siginfo,
4200                                                  void* ucontext,
4201                                                  int abort_if_unrecognized);
4202 
4203 void signalHandler(int sig, siginfo_t* info, void* uc) {
4204   assert(info != NULL && uc != NULL, "it must be old kernel");
4205   int orig_errno = errno;  // Preserve errno value over signal handler.
4206   JVM_handle_linux_signal(sig, info, uc, true);
4207   errno = orig_errno;
4208 }
4209 
4210 
4211 // This boolean allows users to forward their own non-matching signals
4212 // to JVM_handle_linux_signal, harmlessly.
4213 bool os::Linux::signal_handlers_are_installed = false;
4214 
4215 // For signal-chaining
4216 struct sigaction sigact[NSIG];
4217 uint64_t sigs = 0;
4218 #if (64 < NSIG-1)
4219 #error "Not all signals can be encoded in sigs. Adapt its type!"
4220 #endif
4221 bool os::Linux::libjsig_is_loaded = false;
4222 typedef struct sigaction *(*get_signal_t)(int);
4223 get_signal_t os::Linux::get_signal_action = NULL;
4224 
4225 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4226   struct sigaction *actp = NULL;
4227 
4228   if (libjsig_is_loaded) {
4229     // Retrieve the old signal handler from libjsig
4230     actp = (*get_signal_action)(sig);
4231   }
4232   if (actp == NULL) {
4233     // Retrieve the preinstalled signal handler from jvm
4234     actp = get_preinstalled_handler(sig);
4235   }
4236 
4237   return actp;
4238 }
4239 
4240 static bool call_chained_handler(struct sigaction *actp, int sig,
4241                                  siginfo_t *siginfo, void *context) {
4242   // Call the old signal handler
4243   if (actp->sa_handler == SIG_DFL) {
4244     // It's more reasonable to let jvm treat it as an unexpected exception
4245     // instead of taking the default action.
4246     return false;
4247   } else if (actp->sa_handler != SIG_IGN) {
4248     if ((actp->sa_flags & SA_NODEFER) == 0) {
4249       // automaticlly block the signal
4250       sigaddset(&(actp->sa_mask), sig);
4251     }
4252 
4253     sa_handler_t hand = NULL;
4254     sa_sigaction_t sa = NULL;
4255     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4256     // retrieve the chained handler
4257     if (siginfo_flag_set) {
4258       sa = actp->sa_sigaction;
4259     } else {
4260       hand = actp->sa_handler;
4261     }
4262 
4263     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4264       actp->sa_handler = SIG_DFL;
4265     }
4266 
4267     // try to honor the signal mask
4268     sigset_t oset;
4269     sigemptyset(&oset);
4270     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4271 
4272     // call into the chained handler
4273     if (siginfo_flag_set) {
4274       (*sa)(sig, siginfo, context);
4275     } else {
4276       (*hand)(sig);
4277     }
4278 
4279     // restore the signal mask
4280     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4281   }
4282   // Tell jvm's signal handler the signal is taken care of.
4283   return true;
4284 }
4285 
4286 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4287   bool chained = false;
4288   // signal-chaining
4289   if (UseSignalChaining) {
4290     struct sigaction *actp = get_chained_signal_action(sig);
4291     if (actp != NULL) {
4292       chained = call_chained_handler(actp, sig, siginfo, context);
4293     }
4294   }
4295   return chained;
4296 }
4297 
4298 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4299   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4300     return &sigact[sig];
4301   }
4302   return NULL;
4303 }
4304 
4305 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4306   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4307   sigact[sig] = oldAct;
4308   sigs |= (uint64_t)1 << (sig-1);
4309 }
4310 
4311 // for diagnostic
4312 int sigflags[NSIG];
4313 
4314 int os::Linux::get_our_sigflags(int sig) {
4315   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4316   return sigflags[sig];
4317 }
4318 
4319 void os::Linux::set_our_sigflags(int sig, int flags) {
4320   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4321   if (sig > 0 && sig < NSIG) {
4322     sigflags[sig] = flags;
4323   }
4324 }
4325 
4326 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4327   // Check for overwrite.
4328   struct sigaction oldAct;
4329   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4330 
4331   void* oldhand = oldAct.sa_sigaction
4332                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4333                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4334   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4335       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4336       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4337     if (AllowUserSignalHandlers || !set_installed) {
4338       // Do not overwrite; user takes responsibility to forward to us.
4339       return;
4340     } else if (UseSignalChaining) {
4341       // save the old handler in jvm
4342       save_preinstalled_handler(sig, oldAct);
4343       // libjsig also interposes the sigaction() call below and saves the
4344       // old sigaction on it own.
4345     } else {
4346       fatal("Encountered unexpected pre-existing sigaction handler "
4347             "%#lx for signal %d.", (long)oldhand, sig);
4348     }
4349   }
4350 
4351   struct sigaction sigAct;
4352   sigfillset(&(sigAct.sa_mask));
4353   sigAct.sa_handler = SIG_DFL;
4354   if (!set_installed) {
4355     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4356   } else {
4357     sigAct.sa_sigaction = signalHandler;
4358     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4359   }
4360   // Save flags, which are set by ours
4361   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4362   sigflags[sig] = sigAct.sa_flags;
4363 
4364   int ret = sigaction(sig, &sigAct, &oldAct);
4365   assert(ret == 0, "check");
4366 
4367   void* oldhand2  = oldAct.sa_sigaction
4368                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4369                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4370   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4371 }
4372 
4373 // install signal handlers for signals that HotSpot needs to
4374 // handle in order to support Java-level exception handling.
4375 
4376 void os::Linux::install_signal_handlers() {
4377   if (!signal_handlers_are_installed) {
4378     signal_handlers_are_installed = true;
4379 
4380     // signal-chaining
4381     typedef void (*signal_setting_t)();
4382     signal_setting_t begin_signal_setting = NULL;
4383     signal_setting_t end_signal_setting = NULL;
4384     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4385                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4386     if (begin_signal_setting != NULL) {
4387       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4388                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4389       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4390                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4391       libjsig_is_loaded = true;
4392       assert(UseSignalChaining, "should enable signal-chaining");
4393     }
4394     if (libjsig_is_loaded) {
4395       // Tell libjsig jvm is setting signal handlers
4396       (*begin_signal_setting)();
4397     }
4398 
4399     set_signal_handler(SIGSEGV, true);
4400     set_signal_handler(SIGPIPE, true);
4401     set_signal_handler(SIGBUS, true);
4402     set_signal_handler(SIGILL, true);
4403     set_signal_handler(SIGFPE, true);
4404 #if defined(PPC64)
4405     set_signal_handler(SIGTRAP, true);
4406 #endif
4407     set_signal_handler(SIGXFSZ, true);
4408 
4409     if (libjsig_is_loaded) {
4410       // Tell libjsig jvm finishes setting signal handlers
4411       (*end_signal_setting)();
4412     }
4413 
4414     // We don't activate signal checker if libjsig is in place, we trust ourselves
4415     // and if UserSignalHandler is installed all bets are off.
4416     // Log that signal checking is off only if -verbose:jni is specified.
4417     if (CheckJNICalls) {
4418       if (libjsig_is_loaded) {
4419         if (PrintJNIResolving) {
4420           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4421         }
4422         check_signals = false;
4423       }
4424       if (AllowUserSignalHandlers) {
4425         if (PrintJNIResolving) {
4426           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4427         }
4428         check_signals = false;
4429       }
4430     }
4431   }
4432 }
4433 
4434 // This is the fastest way to get thread cpu time on Linux.
4435 // Returns cpu time (user+sys) for any thread, not only for current.
4436 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4437 // It might work on 2.6.10+ with a special kernel/glibc patch.
4438 // For reference, please, see IEEE Std 1003.1-2004:
4439 //   http://www.unix.org/single_unix_specification
4440 
4441 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4442   struct timespec tp;
4443   int rc = os::Linux::clock_gettime(clockid, &tp);
4444   assert(rc == 0, "clock_gettime is expected to return 0 code");
4445 
4446   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4447 }
4448 
4449 void os::Linux::initialize_os_info() {
4450   assert(_os_version == 0, "OS info already initialized");
4451 
4452   struct utsname _uname;
4453 
4454   uint32_t major;
4455   uint32_t minor;
4456   uint32_t fix;
4457 
4458   int rc;
4459 
4460   // Kernel version is unknown if
4461   // verification below fails.
4462   _os_version = 0x01000000;
4463 
4464   rc = uname(&_uname);
4465   if (rc != -1) {
4466 
4467     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4468     if (rc == 3) {
4469 
4470       if (major < 256 && minor < 256 && fix < 256) {
4471         // Kernel version format is as expected,
4472         // set it overriding unknown state.
4473         _os_version = (major << 16) |
4474                       (minor << 8 ) |
4475                       (fix   << 0 ) ;
4476       }
4477     }
4478   }
4479 }
4480 
4481 uint32_t os::Linux::os_version() {
4482   assert(_os_version != 0, "not initialized");
4483   return _os_version & 0x00FFFFFF;
4484 }
4485 
4486 bool os::Linux::os_version_is_known() {
4487   assert(_os_version != 0, "not initialized");
4488   return _os_version & 0x01000000 ? false : true;
4489 }
4490 
4491 /////
4492 // glibc on Linux platform uses non-documented flag
4493 // to indicate, that some special sort of signal
4494 // trampoline is used.
4495 // We will never set this flag, and we should
4496 // ignore this flag in our diagnostic
4497 #ifdef SIGNIFICANT_SIGNAL_MASK
4498   #undef SIGNIFICANT_SIGNAL_MASK
4499 #endif
4500 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4501 
4502 static const char* get_signal_handler_name(address handler,
4503                                            char* buf, int buflen) {
4504   int offset = 0;
4505   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4506   if (found) {
4507     // skip directory names
4508     const char *p1, *p2;
4509     p1 = buf;
4510     size_t len = strlen(os::file_separator());
4511     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4512     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4513   } else {
4514     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4515   }
4516   return buf;
4517 }
4518 
4519 static void print_signal_handler(outputStream* st, int sig,
4520                                  char* buf, size_t buflen) {
4521   struct sigaction sa;
4522 
4523   sigaction(sig, NULL, &sa);
4524 
4525   // See comment for SIGNIFICANT_SIGNAL_MASK define
4526   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4527 
4528   st->print("%s: ", os::exception_name(sig, buf, buflen));
4529 
4530   address handler = (sa.sa_flags & SA_SIGINFO)
4531     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4532     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4533 
4534   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4535     st->print("SIG_DFL");
4536   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4537     st->print("SIG_IGN");
4538   } else {
4539     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4540   }
4541 
4542   st->print(", sa_mask[0]=");
4543   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4544 
4545   address rh = VMError::get_resetted_sighandler(sig);
4546   // May be, handler was resetted by VMError?
4547   if (rh != NULL) {
4548     handler = rh;
4549     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4550   }
4551 
4552   st->print(", sa_flags=");
4553   os::Posix::print_sa_flags(st, sa.sa_flags);
4554 
4555   // Check: is it our handler?
4556   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4557       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4558     // It is our signal handler
4559     // check for flags, reset system-used one!
4560     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4561       st->print(
4562                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4563                 os::Linux::get_our_sigflags(sig));
4564     }
4565   }
4566   st->cr();
4567 }
4568 
4569 
4570 #define DO_SIGNAL_CHECK(sig)                      \
4571   do {                                            \
4572     if (!sigismember(&check_signal_done, sig)) {  \
4573       os::Linux::check_signal_handler(sig);       \
4574     }                                             \
4575   } while (0)
4576 
4577 // This method is a periodic task to check for misbehaving JNI applications
4578 // under CheckJNI, we can add any periodic checks here
4579 
4580 void os::run_periodic_checks() {
4581   if (check_signals == false) return;
4582 
4583   // SEGV and BUS if overridden could potentially prevent
4584   // generation of hs*.log in the event of a crash, debugging
4585   // such a case can be very challenging, so we absolutely
4586   // check the following for a good measure:
4587   DO_SIGNAL_CHECK(SIGSEGV);
4588   DO_SIGNAL_CHECK(SIGILL);
4589   DO_SIGNAL_CHECK(SIGFPE);
4590   DO_SIGNAL_CHECK(SIGBUS);
4591   DO_SIGNAL_CHECK(SIGPIPE);
4592   DO_SIGNAL_CHECK(SIGXFSZ);
4593 #if defined(PPC64)
4594   DO_SIGNAL_CHECK(SIGTRAP);
4595 #endif
4596 
4597   // ReduceSignalUsage allows the user to override these handlers
4598   // see comments at the very top and jvm_solaris.h
4599   if (!ReduceSignalUsage) {
4600     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4601     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4602     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4603     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4604   }
4605 
4606   DO_SIGNAL_CHECK(SR_signum);
4607 }
4608 
4609 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4610 
4611 static os_sigaction_t os_sigaction = NULL;
4612 
4613 void os::Linux::check_signal_handler(int sig) {
4614   char buf[O_BUFLEN];
4615   address jvmHandler = NULL;
4616 
4617 
4618   struct sigaction act;
4619   if (os_sigaction == NULL) {
4620     // only trust the default sigaction, in case it has been interposed
4621     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4622     if (os_sigaction == NULL) return;
4623   }
4624 
4625   os_sigaction(sig, (struct sigaction*)NULL, &act);
4626 
4627 
4628   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4629 
4630   address thisHandler = (act.sa_flags & SA_SIGINFO)
4631     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4632     : CAST_FROM_FN_PTR(address, act.sa_handler);
4633 
4634 
4635   switch (sig) {
4636   case SIGSEGV:
4637   case SIGBUS:
4638   case SIGFPE:
4639   case SIGPIPE:
4640   case SIGILL:
4641   case SIGXFSZ:
4642     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4643     break;
4644 
4645   case SHUTDOWN1_SIGNAL:
4646   case SHUTDOWN2_SIGNAL:
4647   case SHUTDOWN3_SIGNAL:
4648   case BREAK_SIGNAL:
4649     jvmHandler = (address)user_handler();
4650     break;
4651 
4652   default:
4653     if (sig == SR_signum) {
4654       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4655     } else {
4656       return;
4657     }
4658     break;
4659   }
4660 
4661   if (thisHandler != jvmHandler) {
4662     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4663     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4664     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4665     // No need to check this sig any longer
4666     sigaddset(&check_signal_done, sig);
4667     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4668     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4669       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4670                     exception_name(sig, buf, O_BUFLEN));
4671     }
4672   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4673     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4674     tty->print("expected:");
4675     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4676     tty->cr();
4677     tty->print("  found:");
4678     os::Posix::print_sa_flags(tty, act.sa_flags);
4679     tty->cr();
4680     // No need to check this sig any longer
4681     sigaddset(&check_signal_done, sig);
4682   }
4683 
4684   // Dump all the signal
4685   if (sigismember(&check_signal_done, sig)) {
4686     print_signal_handlers(tty, buf, O_BUFLEN);
4687   }
4688 }
4689 
4690 extern void report_error(char* file_name, int line_no, char* title,
4691                          char* format, ...);
4692 
4693 // this is called _before_ the most of global arguments have been parsed
4694 void os::init(void) {
4695   char dummy;   // used to get a guess on initial stack address
4696 //  first_hrtime = gethrtime();
4697 
4698   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4699 
4700   init_random(1234567);
4701 
4702   ThreadCritical::initialize();
4703 
4704   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4705   if (Linux::page_size() == -1) {
4706     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4707           os::strerror(errno));
4708   }
4709   init_page_sizes((size_t) Linux::page_size());
4710 
4711   Linux::initialize_system_info();
4712 
4713   Linux::initialize_os_info();
4714 
4715   // main_thread points to the aboriginal thread
4716   Linux::_main_thread = pthread_self();
4717 
4718   Linux::clock_init();
4719   initial_time_count = javaTimeNanos();
4720 
4721   // pthread_condattr initialization for monotonic clock
4722   int status;
4723   pthread_condattr_t* _condattr = os::Linux::condAttr();
4724   if ((status = pthread_condattr_init(_condattr)) != 0) {
4725     fatal("pthread_condattr_init: %s", os::strerror(status));
4726   }
4727   // Only set the clock if CLOCK_MONOTONIC is available
4728   if (os::supports_monotonic_clock()) {
4729     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4730       if (status == EINVAL) {
4731         warning("Unable to use monotonic clock with relative timed-waits" \
4732                 " - changes to the time-of-day clock may have adverse affects");
4733       } else {
4734         fatal("pthread_condattr_setclock: %s", os::strerror(status));
4735       }
4736     }
4737   }
4738   // else it defaults to CLOCK_REALTIME
4739 
4740   // retrieve entry point for pthread_setname_np
4741   Linux::_pthread_setname_np =
4742     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4743 
4744 }
4745 
4746 // To install functions for atexit system call
4747 extern "C" {
4748   static void perfMemory_exit_helper() {
4749     perfMemory_exit();
4750   }
4751 }
4752 
4753 // this is called _after_ the global arguments have been parsed
4754 jint os::init_2(void) {
4755   Linux::fast_thread_clock_init();
4756 
4757   // Allocate a single page and mark it as readable for safepoint polling
4758   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4759   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4760 
4761   os::set_polling_page(polling_page);
4762   log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4763 
4764   if (!UseMembar) {
4765     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4766     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4767     os::set_memory_serialize_page(mem_serialize_page);
4768     log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4769   }
4770 
4771   // initialize suspend/resume support - must do this before signal_sets_init()
4772   if (SR_initialize() != 0) {
4773     perror("SR_initialize failed");
4774     return JNI_ERR;
4775   }
4776 
4777   Linux::signal_sets_init();
4778   Linux::install_signal_handlers();
4779 
4780   // Check and sets minimum stack sizes against command line options
4781   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4782     return JNI_ERR;
4783   }
4784   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4785 
4786 #if defined(IA32)
4787   workaround_expand_exec_shield_cs_limit();
4788 #endif
4789 
4790   Linux::libpthread_init();
4791   log_info(os)("HotSpot is running with %s, %s",
4792                Linux::glibc_version(), Linux::libpthread_version());
4793 
4794   if (UseNUMA) {
4795     if (!Linux::libnuma_init()) {
4796       UseNUMA = false;
4797     } else {
4798       if ((Linux::numa_max_node() < 1)) {
4799         // There's only one node(they start from 0), disable NUMA.
4800         UseNUMA = false;
4801       }
4802     }
4803     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4804     // we can make the adaptive lgrp chunk resizing work. If the user specified
4805     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4806     // disable adaptive resizing.
4807     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4808       if (FLAG_IS_DEFAULT(UseNUMA)) {
4809         UseNUMA = false;
4810       } else {
4811         if (FLAG_IS_DEFAULT(UseLargePages) &&
4812             FLAG_IS_DEFAULT(UseSHM) &&
4813             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4814           UseLargePages = false;
4815         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4816           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4817           UseAdaptiveSizePolicy = false;
4818           UseAdaptiveNUMAChunkSizing = false;
4819         }
4820       }
4821     }
4822     if (!UseNUMA && ForceNUMA) {
4823       UseNUMA = true;
4824     }
4825   }
4826 
4827   if (MaxFDLimit) {
4828     // set the number of file descriptors to max. print out error
4829     // if getrlimit/setrlimit fails but continue regardless.
4830     struct rlimit nbr_files;
4831     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4832     if (status != 0) {
4833       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4834     } else {
4835       nbr_files.rlim_cur = nbr_files.rlim_max;
4836       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4837       if (status != 0) {
4838         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4839       }
4840     }
4841   }
4842 
4843   // Initialize lock used to serialize thread creation (see os::create_thread)
4844   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4845 
4846   // at-exit methods are called in the reverse order of their registration.
4847   // atexit functions are called on return from main or as a result of a
4848   // call to exit(3C). There can be only 32 of these functions registered
4849   // and atexit() does not set errno.
4850 
4851   if (PerfAllowAtExitRegistration) {
4852     // only register atexit functions if PerfAllowAtExitRegistration is set.
4853     // atexit functions can be delayed until process exit time, which
4854     // can be problematic for embedded VM situations. Embedded VMs should
4855     // call DestroyJavaVM() to assure that VM resources are released.
4856 
4857     // note: perfMemory_exit_helper atexit function may be removed in
4858     // the future if the appropriate cleanup code can be added to the
4859     // VM_Exit VMOperation's doit method.
4860     if (atexit(perfMemory_exit_helper) != 0) {
4861       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4862     }
4863   }
4864 
4865   // initialize thread priority policy
4866   prio_init();
4867 
4868   return JNI_OK;
4869 }
4870 
4871 // Mark the polling page as unreadable
4872 void os::make_polling_page_unreadable(void) {
4873   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4874     fatal("Could not disable polling page");
4875   }
4876 }
4877 
4878 // Mark the polling page as readable
4879 void os::make_polling_page_readable(void) {
4880   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4881     fatal("Could not enable polling page");
4882   }
4883 }
4884 
4885 // older glibc versions don't have this macro (which expands to
4886 // an optimized bit-counting function) so we have to roll our own
4887 #ifndef CPU_COUNT
4888 
4889 static int _cpu_count(const cpu_set_t* cpus) {
4890   int count = 0;
4891   // only look up to the number of configured processors
4892   for (int i = 0; i < os::processor_count(); i++) {
4893     if (CPU_ISSET(i, cpus)) {
4894       count++;
4895     }
4896   }
4897   return count;
4898 }
4899 
4900 #define CPU_COUNT(cpus) _cpu_count(cpus)
4901 
4902 #endif // CPU_COUNT
4903 
4904 // Get the current number of available processors for this process.
4905 // This value can change at any time during a process's lifetime.
4906 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4907 // If it appears there may be more than 1024 processors then we do a
4908 // dynamic check - see 6515172 for details.
4909 // If anything goes wrong we fallback to returning the number of online
4910 // processors - which can be greater than the number available to the process.
4911 int os::active_processor_count() {
4912   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
4913   cpu_set_t* cpus_p = &cpus;
4914   int cpus_size = sizeof(cpu_set_t);
4915 
4916   int configured_cpus = processor_count();  // upper bound on available cpus
4917   int cpu_count = 0;
4918 
4919 // old build platforms may not support dynamic cpu sets
4920 #ifdef CPU_ALLOC
4921 
4922   // To enable easy testing of the dynamic path on different platforms we
4923   // introduce a diagnostic flag: UseCpuAllocPath
4924   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4925     // kernel may use a mask bigger than cpu_set_t
4926     log_trace(os)("active_processor_count: using dynamic path %s"
4927                   "- configured processors: %d",
4928                   UseCpuAllocPath ? "(forced) " : "",
4929                   configured_cpus);
4930     cpus_p = CPU_ALLOC(configured_cpus);
4931     if (cpus_p != NULL) {
4932       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4933       // zero it just to be safe
4934       CPU_ZERO_S(cpus_size, cpus_p);
4935     }
4936     else {
4937        // failed to allocate so fallback to online cpus
4938        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4939        log_trace(os)("active_processor_count: "
4940                      "CPU_ALLOC failed (%s) - using "
4941                      "online processor count: %d",
4942                      os::strerror(errno), online_cpus);
4943        return online_cpus;
4944     }
4945   }
4946   else {
4947     log_trace(os)("active_processor_count: using static path - configured processors: %d",
4948                   configured_cpus);
4949   }
4950 #else // CPU_ALLOC
4951 // these stubs won't be executed
4952 #define CPU_COUNT_S(size, cpus) -1
4953 #define CPU_FREE(cpus)
4954 
4955   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4956                 configured_cpus);
4957 #endif // CPU_ALLOC
4958 
4959   // pid 0 means the current thread - which we have to assume represents the process
4960   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4961     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4962       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4963     }
4964     else {
4965       cpu_count = CPU_COUNT(cpus_p);
4966     }
4967     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4968   }
4969   else {
4970     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4971     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4972             "which may exceed available processors", os::strerror(errno), cpu_count);
4973   }
4974 
4975   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4976     CPU_FREE(cpus_p);
4977   }
4978 
4979   assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
4980   return cpu_count;
4981 }
4982 
4983 void os::set_native_thread_name(const char *name) {
4984   if (Linux::_pthread_setname_np) {
4985     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4986     snprintf(buf, sizeof(buf), "%s", name);
4987     buf[sizeof(buf) - 1] = '\0';
4988     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4989     // ERANGE should not happen; all other errors should just be ignored.
4990     assert(rc != ERANGE, "pthread_setname_np failed");
4991   }
4992 }
4993 
4994 bool os::distribute_processes(uint length, uint* distribution) {
4995   // Not yet implemented.
4996   return false;
4997 }
4998 
4999 bool os::bind_to_processor(uint processor_id) {
5000   // Not yet implemented.
5001   return false;
5002 }
5003 
5004 ///
5005 
5006 void os::SuspendedThreadTask::internal_do_task() {
5007   if (do_suspend(_thread->osthread())) {
5008     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5009     do_task(context);
5010     do_resume(_thread->osthread());
5011   }
5012 }
5013 
5014 class PcFetcher : public os::SuspendedThreadTask {
5015  public:
5016   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5017   ExtendedPC result();
5018  protected:
5019   void do_task(const os::SuspendedThreadTaskContext& context);
5020  private:
5021   ExtendedPC _epc;
5022 };
5023 
5024 ExtendedPC PcFetcher::result() {
5025   guarantee(is_done(), "task is not done yet.");
5026   return _epc;
5027 }
5028 
5029 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5030   Thread* thread = context.thread();
5031   OSThread* osthread = thread->osthread();
5032   if (osthread->ucontext() != NULL) {
5033     _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5034   } else {
5035     // NULL context is unexpected, double-check this is the VMThread
5036     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5037   }
5038 }
5039 
5040 // Suspends the target using the signal mechanism and then grabs the PC before
5041 // resuming the target. Used by the flat-profiler only
5042 ExtendedPC os::get_thread_pc(Thread* thread) {
5043   // Make sure that it is called by the watcher for the VMThread
5044   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5045   assert(thread->is_VM_thread(), "Can only be called for VMThread");
5046 
5047   PcFetcher fetcher(thread);
5048   fetcher.run();
5049   return fetcher.result();
5050 }
5051 
5052 ////////////////////////////////////////////////////////////////////////////////
5053 // debug support
5054 
5055 bool os::find(address addr, outputStream* st) {
5056   Dl_info dlinfo;
5057   memset(&dlinfo, 0, sizeof(dlinfo));
5058   if (dladdr(addr, &dlinfo) != 0) {
5059     st->print(PTR_FORMAT ": ", p2i(addr));
5060     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5061       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5062                 p2i(addr) - p2i(dlinfo.dli_saddr));
5063     } else if (dlinfo.dli_fbase != NULL) {
5064       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5065     } else {
5066       st->print("<absolute address>");
5067     }
5068     if (dlinfo.dli_fname != NULL) {
5069       st->print(" in %s", dlinfo.dli_fname);
5070     }
5071     if (dlinfo.dli_fbase != NULL) {
5072       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5073     }
5074     st->cr();
5075 
5076     if (Verbose) {
5077       // decode some bytes around the PC
5078       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5079       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5080       address       lowest = (address) dlinfo.dli_sname;
5081       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5082       if (begin < lowest)  begin = lowest;
5083       Dl_info dlinfo2;
5084       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5085           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5086         end = (address) dlinfo2.dli_saddr;
5087       }
5088       Disassembler::decode(begin, end, st);
5089     }
5090     return true;
5091   }
5092   return false;
5093 }
5094 
5095 ////////////////////////////////////////////////////////////////////////////////
5096 // misc
5097 
5098 // This does not do anything on Linux. This is basically a hook for being
5099 // able to use structured exception handling (thread-local exception filters)
5100 // on, e.g., Win32.
5101 void
5102 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5103                          JavaCallArguments* args, Thread* thread) {
5104   f(value, method, args, thread);
5105 }
5106 
5107 void os::print_statistics() {
5108 }
5109 
5110 bool os::message_box(const char* title, const char* message) {
5111   int i;
5112   fdStream err(defaultStream::error_fd());
5113   for (i = 0; i < 78; i++) err.print_raw("=");
5114   err.cr();
5115   err.print_raw_cr(title);
5116   for (i = 0; i < 78; i++) err.print_raw("-");
5117   err.cr();
5118   err.print_raw_cr(message);
5119   for (i = 0; i < 78; i++) err.print_raw("=");
5120   err.cr();
5121 
5122   char buf[16];
5123   // Prevent process from exiting upon "read error" without consuming all CPU
5124   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5125 
5126   return buf[0] == 'y' || buf[0] == 'Y';
5127 }
5128 
5129 int os::stat(const char *path, struct stat *sbuf) {
5130   char pathbuf[MAX_PATH];
5131   if (strlen(path) > MAX_PATH - 1) {
5132     errno = ENAMETOOLONG;
5133     return -1;
5134   }
5135   os::native_path(strcpy(pathbuf, path));
5136   return ::stat(pathbuf, sbuf);
5137 }
5138 
5139 // Is a (classpath) directory empty?
5140 bool os::dir_is_empty(const char* path) {
5141   DIR *dir = NULL;
5142   struct dirent *ptr;
5143 
5144   dir = opendir(path);
5145   if (dir == NULL) return true;
5146 
5147   // Scan the directory
5148   bool result = true;
5149   char buf[sizeof(struct dirent) + MAX_PATH];
5150   while (result && (ptr = ::readdir(dir)) != NULL) {
5151     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5152       result = false;
5153     }
5154   }
5155   closedir(dir);
5156   return result;
5157 }
5158 
5159 // This code originates from JDK's sysOpen and open64_w
5160 // from src/solaris/hpi/src/system_md.c
5161 
5162 int os::open(const char *path, int oflag, int mode) {
5163   if (strlen(path) > MAX_PATH - 1) {
5164     errno = ENAMETOOLONG;
5165     return -1;
5166   }
5167 
5168   // All file descriptors that are opened in the Java process and not
5169   // specifically destined for a subprocess should have the close-on-exec
5170   // flag set.  If we don't set it, then careless 3rd party native code
5171   // might fork and exec without closing all appropriate file descriptors
5172   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5173   // turn might:
5174   //
5175   // - cause end-of-file to fail to be detected on some file
5176   //   descriptors, resulting in mysterious hangs, or
5177   //
5178   // - might cause an fopen in the subprocess to fail on a system
5179   //   suffering from bug 1085341.
5180   //
5181   // (Yes, the default setting of the close-on-exec flag is a Unix
5182   // design flaw)
5183   //
5184   // See:
5185   // 1085341: 32-bit stdio routines should support file descriptors >255
5186   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5187   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5188   //
5189   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5190   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5191   // because it saves a system call and removes a small window where the flag
5192   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5193   // and we fall back to using FD_CLOEXEC (see below).
5194 #ifdef O_CLOEXEC
5195   oflag |= O_CLOEXEC;
5196 #endif
5197 
5198   int fd = ::open64(path, oflag, mode);
5199   if (fd == -1) return -1;
5200 
5201   //If the open succeeded, the file might still be a directory
5202   {
5203     struct stat64 buf64;
5204     int ret = ::fstat64(fd, &buf64);
5205     int st_mode = buf64.st_mode;
5206 
5207     if (ret != -1) {
5208       if ((st_mode & S_IFMT) == S_IFDIR) {
5209         errno = EISDIR;
5210         ::close(fd);
5211         return -1;
5212       }
5213     } else {
5214       ::close(fd);
5215       return -1;
5216     }
5217   }
5218 
5219 #ifdef FD_CLOEXEC
5220   // Validate that the use of the O_CLOEXEC flag on open above worked.
5221   // With recent kernels, we will perform this check exactly once.
5222   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5223   if (!O_CLOEXEC_is_known_to_work) {
5224     int flags = ::fcntl(fd, F_GETFD);
5225     if (flags != -1) {
5226       if ((flags & FD_CLOEXEC) != 0)
5227         O_CLOEXEC_is_known_to_work = 1;
5228       else
5229         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5230     }
5231   }
5232 #endif
5233 
5234   return fd;
5235 }
5236 
5237 
5238 // create binary file, rewriting existing file if required
5239 int os::create_binary_file(const char* path, bool rewrite_existing) {
5240   int oflags = O_WRONLY | O_CREAT;
5241   if (!rewrite_existing) {
5242     oflags |= O_EXCL;
5243   }
5244   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5245 }
5246 
5247 // return current position of file pointer
5248 jlong os::current_file_offset(int fd) {
5249   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5250 }
5251 
5252 // move file pointer to the specified offset
5253 jlong os::seek_to_file_offset(int fd, jlong offset) {
5254   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5255 }
5256 
5257 // This code originates from JDK's sysAvailable
5258 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5259 
5260 int os::available(int fd, jlong *bytes) {
5261   jlong cur, end;
5262   int mode;
5263   struct stat64 buf64;
5264 
5265   if (::fstat64(fd, &buf64) >= 0) {
5266     mode = buf64.st_mode;
5267     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5268       int n;
5269       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5270         *bytes = n;
5271         return 1;
5272       }
5273     }
5274   }
5275   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5276     return 0;
5277   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5278     return 0;
5279   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5280     return 0;
5281   }
5282   *bytes = end - cur;
5283   return 1;
5284 }
5285 
5286 // Map a block of memory.
5287 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5288                         char *addr, size_t bytes, bool read_only,
5289                         bool allow_exec) {
5290   int prot;
5291   int flags = MAP_PRIVATE;
5292 
5293   if (read_only) {
5294     prot = PROT_READ;
5295   } else {
5296     prot = PROT_READ | PROT_WRITE;
5297   }
5298 
5299   if (allow_exec) {
5300     prot |= PROT_EXEC;
5301   }
5302 
5303   if (addr != NULL) {
5304     flags |= MAP_FIXED;
5305   }
5306 
5307   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5308                                      fd, file_offset);
5309   if (mapped_address == MAP_FAILED) {
5310     return NULL;
5311   }
5312   return mapped_address;
5313 }
5314 
5315 
5316 // Remap a block of memory.
5317 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5318                           char *addr, size_t bytes, bool read_only,
5319                           bool allow_exec) {
5320   // same as map_memory() on this OS
5321   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5322                         allow_exec);
5323 }
5324 
5325 
5326 // Unmap a block of memory.
5327 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5328   return munmap(addr, bytes) == 0;
5329 }
5330 
5331 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5332 
5333 static clockid_t thread_cpu_clockid(Thread* thread) {
5334   pthread_t tid = thread->osthread()->pthread_id();
5335   clockid_t clockid;
5336 
5337   // Get thread clockid
5338   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5339   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5340   return clockid;
5341 }
5342 
5343 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5344 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5345 // of a thread.
5346 //
5347 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5348 // the fast estimate available on the platform.
5349 
5350 jlong os::current_thread_cpu_time() {
5351   if (os::Linux::supports_fast_thread_cpu_time()) {
5352     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5353   } else {
5354     // return user + sys since the cost is the same
5355     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5356   }
5357 }
5358 
5359 jlong os::thread_cpu_time(Thread* thread) {
5360   // consistent with what current_thread_cpu_time() returns
5361   if (os::Linux::supports_fast_thread_cpu_time()) {
5362     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5363   } else {
5364     return slow_thread_cpu_time(thread, true /* user + sys */);
5365   }
5366 }
5367 
5368 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5369   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5370     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5371   } else {
5372     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5373   }
5374 }
5375 
5376 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5377   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5378     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5379   } else {
5380     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5381   }
5382 }
5383 
5384 //  -1 on error.
5385 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5386   pid_t  tid = thread->osthread()->thread_id();
5387   char *s;
5388   char stat[2048];
5389   int statlen;
5390   char proc_name[64];
5391   int count;
5392   long sys_time, user_time;
5393   char cdummy;
5394   int idummy;
5395   long ldummy;
5396   FILE *fp;
5397 
5398   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5399   fp = fopen(proc_name, "r");
5400   if (fp == NULL) return -1;
5401   statlen = fread(stat, 1, 2047, fp);
5402   stat[statlen] = '\0';
5403   fclose(fp);
5404 
5405   // Skip pid and the command string. Note that we could be dealing with
5406   // weird command names, e.g. user could decide to rename java launcher
5407   // to "java 1.4.2 :)", then the stat file would look like
5408   //                1234 (java 1.4.2 :)) R ... ...
5409   // We don't really need to know the command string, just find the last
5410   // occurrence of ")" and then start parsing from there. See bug 4726580.
5411   s = strrchr(stat, ')');
5412   if (s == NULL) return -1;
5413 
5414   // Skip blank chars
5415   do { s++; } while (s && isspace(*s));
5416 
5417   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5418                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5419                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5420                  &user_time, &sys_time);
5421   if (count != 13) return -1;
5422   if (user_sys_cpu_time) {
5423     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5424   } else {
5425     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5426   }
5427 }
5428 
5429 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5430   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5431   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5432   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5433   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5434 }
5435 
5436 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5437   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5438   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5439   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5440   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5441 }
5442 
5443 bool os::is_thread_cpu_time_supported() {
5444   return true;
5445 }
5446 
5447 // System loadavg support.  Returns -1 if load average cannot be obtained.
5448 // Linux doesn't yet have a (official) notion of processor sets,
5449 // so just return the system wide load average.
5450 int os::loadavg(double loadavg[], int nelem) {
5451   return ::getloadavg(loadavg, nelem);
5452 }
5453 
5454 void os::pause() {
5455   char filename[MAX_PATH];
5456   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5457     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5458   } else {
5459     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5460   }
5461 
5462   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5463   if (fd != -1) {
5464     struct stat buf;
5465     ::close(fd);
5466     while (::stat(filename, &buf) == 0) {
5467       (void)::poll(NULL, 0, 100);
5468     }
5469   } else {
5470     jio_fprintf(stderr,
5471                 "Could not open pause file '%s', continuing immediately.\n", filename);
5472   }
5473 }
5474 
5475 
5476 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5477 // comment paragraphs are worth repeating here:
5478 //
5479 // Assumption:
5480 //    Only one parker can exist on an event, which is why we allocate
5481 //    them per-thread. Multiple unparkers can coexist.
5482 //
5483 // _Event serves as a restricted-range semaphore.
5484 //   -1 : thread is blocked, i.e. there is a waiter
5485 //    0 : neutral: thread is running or ready,
5486 //        could have been signaled after a wait started
5487 //    1 : signaled - thread is running or ready
5488 //
5489 
5490 // utility to compute the abstime argument to timedwait:
5491 // millis is the relative timeout time
5492 // abstime will be the absolute timeout time
5493 // TODO: replace compute_abstime() with unpackTime()
5494 
5495 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5496   if (millis < 0)  millis = 0;
5497 
5498   jlong seconds = millis / 1000;
5499   millis %= 1000;
5500   if (seconds > 50000000) { // see man cond_timedwait(3T)
5501     seconds = 50000000;
5502   }
5503 
5504   if (os::supports_monotonic_clock()) {
5505     struct timespec now;
5506     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5507     assert_status(status == 0, status, "clock_gettime");
5508     abstime->tv_sec = now.tv_sec  + seconds;
5509     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5510     if (nanos >= NANOSECS_PER_SEC) {
5511       abstime->tv_sec += 1;
5512       nanos -= NANOSECS_PER_SEC;
5513     }
5514     abstime->tv_nsec = nanos;
5515   } else {
5516     struct timeval now;
5517     int status = gettimeofday(&now, NULL);
5518     assert(status == 0, "gettimeofday");
5519     abstime->tv_sec = now.tv_sec  + seconds;
5520     long usec = now.tv_usec + millis * 1000;
5521     if (usec >= 1000000) {
5522       abstime->tv_sec += 1;
5523       usec -= 1000000;
5524     }
5525     abstime->tv_nsec = usec * 1000;
5526   }
5527   return abstime;
5528 }
5529 
5530 void os::PlatformEvent::park() {       // AKA "down()"
5531   // Transitions for _Event:
5532   //   -1 => -1 : illegal
5533   //    1 =>  0 : pass - return immediately
5534   //    0 => -1 : block; then set _Event to 0 before returning
5535 
5536   // Invariant: Only the thread associated with the Event/PlatformEvent
5537   // may call park().
5538   // TODO: assert that _Assoc != NULL or _Assoc == Self
5539   assert(_nParked == 0, "invariant");
5540 
5541   int v;
5542   for (;;) {
5543     v = _Event;
5544     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5545   }
5546   guarantee(v >= 0, "invariant");
5547   if (v == 0) {
5548     // Do this the hard way by blocking ...
5549     int status = pthread_mutex_lock(_mutex);
5550     assert_status(status == 0, status, "mutex_lock");
5551     guarantee(_nParked == 0, "invariant");
5552     ++_nParked;
5553     while (_Event < 0) {
5554       status = pthread_cond_wait(_cond, _mutex);
5555       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5556       // Treat this the same as if the wait was interrupted
5557       if (status == ETIME) { status = EINTR; }
5558       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5559     }
5560     --_nParked;
5561 
5562     _Event = 0;
5563     status = pthread_mutex_unlock(_mutex);
5564     assert_status(status == 0, status, "mutex_unlock");
5565     // Paranoia to ensure our locked and lock-free paths interact
5566     // correctly with each other.
5567     OrderAccess::fence();
5568   }
5569   guarantee(_Event >= 0, "invariant");
5570 }
5571 
5572 int os::PlatformEvent::park(jlong millis) {
5573   // Transitions for _Event:
5574   //   -1 => -1 : illegal
5575   //    1 =>  0 : pass - return immediately
5576   //    0 => -1 : block; then set _Event to 0 before returning
5577 
5578   guarantee(_nParked == 0, "invariant");
5579 
5580   int v;
5581   for (;;) {
5582     v = _Event;
5583     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5584   }
5585   guarantee(v >= 0, "invariant");
5586   if (v != 0) return OS_OK;
5587 
5588   // We do this the hard way, by blocking the thread.
5589   // Consider enforcing a minimum timeout value.
5590   struct timespec abst;
5591   compute_abstime(&abst, millis);
5592 
5593   int ret = OS_TIMEOUT;
5594   int status = pthread_mutex_lock(_mutex);
5595   assert_status(status == 0, status, "mutex_lock");
5596   guarantee(_nParked == 0, "invariant");
5597   ++_nParked;
5598 
5599   // Object.wait(timo) will return because of
5600   // (a) notification
5601   // (b) timeout
5602   // (c) thread.interrupt
5603   //
5604   // Thread.interrupt and object.notify{All} both call Event::set.
5605   // That is, we treat thread.interrupt as a special case of notification.
5606   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5607   // We assume all ETIME returns are valid.
5608   //
5609   // TODO: properly differentiate simultaneous notify+interrupt.
5610   // In that case, we should propagate the notify to another waiter.
5611 
5612   while (_Event < 0) {
5613     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5614     assert_status(status == 0 || status == EINTR ||
5615                   status == ETIME || status == ETIMEDOUT,
5616                   status, "cond_timedwait");
5617     if (!FilterSpuriousWakeups) break;                 // previous semantics
5618     if (status == ETIME || status == ETIMEDOUT) break;
5619     // We consume and ignore EINTR and spurious wakeups.
5620   }
5621   --_nParked;
5622   if (_Event >= 0) {
5623     ret = OS_OK;
5624   }
5625   _Event = 0;
5626   status = pthread_mutex_unlock(_mutex);
5627   assert_status(status == 0, status, "mutex_unlock");
5628   assert(_nParked == 0, "invariant");
5629   // Paranoia to ensure our locked and lock-free paths interact
5630   // correctly with each other.
5631   OrderAccess::fence();
5632   return ret;
5633 }
5634 
5635 void os::PlatformEvent::unpark() {
5636   // Transitions for _Event:
5637   //    0 => 1 : just return
5638   //    1 => 1 : just return
5639   //   -1 => either 0 or 1; must signal target thread
5640   //         That is, we can safely transition _Event from -1 to either
5641   //         0 or 1.
5642   // See also: "Semaphores in Plan 9" by Mullender & Cox
5643   //
5644   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5645   // that it will take two back-to-back park() calls for the owning
5646   // thread to block. This has the benefit of forcing a spurious return
5647   // from the first park() call after an unpark() call which will help
5648   // shake out uses of park() and unpark() without condition variables.
5649 
5650   if (Atomic::xchg(1, &_Event) >= 0) return;
5651 
5652   // Wait for the thread associated with the event to vacate
5653   int status = pthread_mutex_lock(_mutex);
5654   assert_status(status == 0, status, "mutex_lock");
5655   int AnyWaiters = _nParked;
5656   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5657   status = pthread_mutex_unlock(_mutex);
5658   assert_status(status == 0, status, "mutex_unlock");
5659   if (AnyWaiters != 0) {
5660     // Note that we signal() *after* dropping the lock for "immortal" Events.
5661     // This is safe and avoids a common class of  futile wakeups.  In rare
5662     // circumstances this can cause a thread to return prematurely from
5663     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5664     // will simply re-test the condition and re-park itself.
5665     // This provides particular benefit if the underlying platform does not
5666     // provide wait morphing.
5667     status = pthread_cond_signal(_cond);
5668     assert_status(status == 0, status, "cond_signal");
5669   }
5670 }
5671 
5672 
5673 // JSR166
5674 // -------------------------------------------------------
5675 
5676 // The solaris and linux implementations of park/unpark are fairly
5677 // conservative for now, but can be improved. They currently use a
5678 // mutex/condvar pair, plus a a count.
5679 // Park decrements count if > 0, else does a condvar wait.  Unpark
5680 // sets count to 1 and signals condvar.  Only one thread ever waits
5681 // on the condvar. Contention seen when trying to park implies that someone
5682 // is unparking you, so don't wait. And spurious returns are fine, so there
5683 // is no need to track notifications.
5684 
5685 // This code is common to linux and solaris and will be moved to a
5686 // common place in dolphin.
5687 //
5688 // The passed in time value is either a relative time in nanoseconds
5689 // or an absolute time in milliseconds. Either way it has to be unpacked
5690 // into suitable seconds and nanoseconds components and stored in the
5691 // given timespec structure.
5692 // Given time is a 64-bit value and the time_t used in the timespec is only
5693 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5694 // overflow if times way in the future are given. Further on Solaris versions
5695 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5696 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5697 // As it will be 28 years before "now + 100000000" will overflow we can
5698 // ignore overflow and just impose a hard-limit on seconds using the value
5699 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5700 // years from "now".
5701 
5702 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5703   assert(time > 0, "convertTime");
5704   time_t max_secs = 0;
5705 
5706   if (!os::supports_monotonic_clock() || isAbsolute) {
5707     struct timeval now;
5708     int status = gettimeofday(&now, NULL);
5709     assert(status == 0, "gettimeofday");
5710 
5711     max_secs = now.tv_sec + MAX_SECS;
5712 
5713     if (isAbsolute) {
5714       jlong secs = time / 1000;
5715       if (secs > max_secs) {
5716         absTime->tv_sec = max_secs;
5717       } else {
5718         absTime->tv_sec = secs;
5719       }
5720       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5721     } else {
5722       jlong secs = time / NANOSECS_PER_SEC;
5723       if (secs >= MAX_SECS) {
5724         absTime->tv_sec = max_secs;
5725         absTime->tv_nsec = 0;
5726       } else {
5727         absTime->tv_sec = now.tv_sec + secs;
5728         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5729         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5730           absTime->tv_nsec -= NANOSECS_PER_SEC;
5731           ++absTime->tv_sec; // note: this must be <= max_secs
5732         }
5733       }
5734     }
5735   } else {
5736     // must be relative using monotonic clock
5737     struct timespec now;
5738     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5739     assert_status(status == 0, status, "clock_gettime");
5740     max_secs = now.tv_sec + MAX_SECS;
5741     jlong secs = time / NANOSECS_PER_SEC;
5742     if (secs >= MAX_SECS) {
5743       absTime->tv_sec = max_secs;
5744       absTime->tv_nsec = 0;
5745     } else {
5746       absTime->tv_sec = now.tv_sec + secs;
5747       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5748       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5749         absTime->tv_nsec -= NANOSECS_PER_SEC;
5750         ++absTime->tv_sec; // note: this must be <= max_secs
5751       }
5752     }
5753   }
5754   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5755   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5756   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5757   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5758 }
5759 
5760 void Parker::park(bool isAbsolute, jlong time) {
5761   // Ideally we'd do something useful while spinning, such
5762   // as calling unpackTime().
5763 
5764   // Optional fast-path check:
5765   // Return immediately if a permit is available.
5766   // We depend on Atomic::xchg() having full barrier semantics
5767   // since we are doing a lock-free update to _counter.
5768   if (Atomic::xchg(0, &_counter) > 0) return;
5769 
5770   Thread* thread = Thread::current();
5771   assert(thread->is_Java_thread(), "Must be JavaThread");
5772   JavaThread *jt = (JavaThread *)thread;
5773 
5774   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5775   // Check interrupt before trying to wait
5776   if (Thread::is_interrupted(thread, false)) {
5777     return;
5778   }
5779 
5780   // Next, demultiplex/decode time arguments
5781   timespec absTime;
5782   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5783     return;
5784   }
5785   if (time > 0) {
5786     unpackTime(&absTime, isAbsolute, time);
5787   }
5788 
5789 
5790   // Enter safepoint region
5791   // Beware of deadlocks such as 6317397.
5792   // The per-thread Parker:: mutex is a classic leaf-lock.
5793   // In particular a thread must never block on the Threads_lock while
5794   // holding the Parker:: mutex.  If safepoints are pending both the
5795   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5796   ThreadBlockInVM tbivm(jt);
5797 
5798   // Don't wait if cannot get lock since interference arises from
5799   // unblocking.  Also. check interrupt before trying wait
5800   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5801     return;
5802   }
5803 
5804   int status;
5805   if (_counter > 0)  { // no wait needed
5806     _counter = 0;
5807     status = pthread_mutex_unlock(_mutex);
5808     assert_status(status == 0, status, "invariant");
5809     // Paranoia to ensure our locked and lock-free paths interact
5810     // correctly with each other and Java-level accesses.
5811     OrderAccess::fence();
5812     return;
5813   }
5814 
5815 #ifdef ASSERT
5816   // Don't catch signals while blocked; let the running threads have the signals.
5817   // (This allows a debugger to break into the running thread.)
5818   sigset_t oldsigs;
5819   sigemptyset(&oldsigs);
5820   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5821   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5822 #endif
5823 
5824   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5825   jt->set_suspend_equivalent();
5826   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5827 
5828   assert(_cur_index == -1, "invariant");
5829   if (time == 0) {
5830     _cur_index = REL_INDEX; // arbitrary choice when not timed
5831     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5832   } else {
5833     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5834     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5835   }
5836   _cur_index = -1;
5837   assert_status(status == 0 || status == EINTR ||
5838                 status == ETIME || status == ETIMEDOUT,
5839                 status, "cond_timedwait");
5840 
5841 #ifdef ASSERT
5842   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5843 #endif
5844 
5845   _counter = 0;
5846   status = pthread_mutex_unlock(_mutex);
5847   assert_status(status == 0, status, "invariant");
5848   // Paranoia to ensure our locked and lock-free paths interact
5849   // correctly with each other and Java-level accesses.
5850   OrderAccess::fence();
5851 
5852   // If externally suspended while waiting, re-suspend
5853   if (jt->handle_special_suspend_equivalent_condition()) {
5854     jt->java_suspend_self();
5855   }
5856 }
5857 
5858 void Parker::unpark() {
5859   int status = pthread_mutex_lock(_mutex);
5860   assert_status(status == 0, status, "invariant");
5861   const int s = _counter;
5862   _counter = 1;
5863   // must capture correct index before unlocking
5864   int index = _cur_index;
5865   status = pthread_mutex_unlock(_mutex);
5866   assert_status(status == 0, status, "invariant");
5867   if (s < 1 && index != -1) {
5868     // thread is definitely parked
5869     status = pthread_cond_signal(&_cond[index]);
5870     assert_status(status == 0, status, "invariant");
5871   }
5872 }
5873 
5874 
5875 extern char** environ;
5876 
5877 // Run the specified command in a separate process. Return its exit value,
5878 // or -1 on failure (e.g. can't fork a new process).
5879 // Unlike system(), this function can be called from signal handler. It
5880 // doesn't block SIGINT et al.
5881 int os::fork_and_exec(char* cmd) {
5882   const char * argv[4] = {"sh", "-c", cmd, NULL};
5883 
5884   pid_t pid = fork();
5885 
5886   if (pid < 0) {
5887     // fork failed
5888     return -1;
5889 
5890   } else if (pid == 0) {
5891     // child process
5892 
5893     execve("/bin/sh", (char* const*)argv, environ);
5894 
5895     // execve failed
5896     _exit(-1);
5897 
5898   } else  {
5899     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5900     // care about the actual exit code, for now.
5901 
5902     int status;
5903 
5904     // Wait for the child process to exit.  This returns immediately if
5905     // the child has already exited. */
5906     while (waitpid(pid, &status, 0) < 0) {
5907       switch (errno) {
5908       case ECHILD: return 0;
5909       case EINTR: break;
5910       default: return -1;
5911       }
5912     }
5913 
5914     if (WIFEXITED(status)) {
5915       // The child exited normally; get its exit code.
5916       return WEXITSTATUS(status);
5917     } else if (WIFSIGNALED(status)) {
5918       // The child exited because of a signal
5919       // The best value to return is 0x80 + signal number,
5920       // because that is what all Unix shells do, and because
5921       // it allows callers to distinguish between process exit and
5922       // process death by signal.
5923       return 0x80 + WTERMSIG(status);
5924     } else {
5925       // Unknown exit code; pass it through
5926       return status;
5927     }
5928   }
5929 }
5930 
5931 // is_headless_jre()
5932 //
5933 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5934 // in order to report if we are running in a headless jre
5935 //
5936 // Since JDK8 xawt/libmawt.so was moved into the same directory
5937 // as libawt.so, and renamed libawt_xawt.so
5938 //
5939 bool os::is_headless_jre() {
5940   struct stat statbuf;
5941   char buf[MAXPATHLEN];
5942   char libmawtpath[MAXPATHLEN];
5943   const char *xawtstr  = "/xawt/libmawt.so";
5944   const char *new_xawtstr = "/libawt_xawt.so";
5945   char *p;
5946 
5947   // Get path to libjvm.so
5948   os::jvm_path(buf, sizeof(buf));
5949 
5950   // Get rid of libjvm.so
5951   p = strrchr(buf, '/');
5952   if (p == NULL) {
5953     return false;
5954   } else {
5955     *p = '\0';
5956   }
5957 
5958   // Get rid of client or server
5959   p = strrchr(buf, '/');
5960   if (p == NULL) {
5961     return false;
5962   } else {
5963     *p = '\0';
5964   }
5965 
5966   // check xawt/libmawt.so
5967   strcpy(libmawtpath, buf);
5968   strcat(libmawtpath, xawtstr);
5969   if (::stat(libmawtpath, &statbuf) == 0) return false;
5970 
5971   // check libawt_xawt.so
5972   strcpy(libmawtpath, buf);
5973   strcat(libmawtpath, new_xawtstr);
5974   if (::stat(libmawtpath, &statbuf) == 0) return false;
5975 
5976   return true;
5977 }
5978 
5979 // Get the default path to the core file
5980 // Returns the length of the string
5981 int os::get_core_path(char* buffer, size_t bufferSize) {
5982   /*
5983    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5984    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5985    */
5986   const int core_pattern_len = 129;
5987   char core_pattern[core_pattern_len] = {0};
5988 
5989   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5990   if (core_pattern_file == -1) {
5991     return -1;
5992   }
5993 
5994   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5995   ::close(core_pattern_file);
5996   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5997     return -1;
5998   }
5999   if (core_pattern[ret-1] == '\n') {
6000     core_pattern[ret-1] = '\0';
6001   } else {
6002     core_pattern[ret] = '\0';
6003   }
6004 
6005   char *pid_pos = strstr(core_pattern, "%p");
6006   int written;
6007 
6008   if (core_pattern[0] == '/') {
6009     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6010   } else {
6011     char cwd[PATH_MAX];
6012 
6013     const char* p = get_current_directory(cwd, PATH_MAX);
6014     if (p == NULL) {
6015       return -1;
6016     }
6017 
6018     if (core_pattern[0] == '|') {
6019       written = jio_snprintf(buffer, bufferSize,
6020                              "\"%s\" (or dumping to %s/core.%d)",
6021                              &core_pattern[1], p, current_process_id());
6022     } else {
6023       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6024     }
6025   }
6026 
6027   if (written < 0) {
6028     return -1;
6029   }
6030 
6031   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6032     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6033 
6034     if (core_uses_pid_file != -1) {
6035       char core_uses_pid = 0;
6036       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6037       ::close(core_uses_pid_file);
6038 
6039       if (core_uses_pid == '1') {
6040         jio_snprintf(buffer + written, bufferSize - written,
6041                                           ".%d", current_process_id());
6042       }
6043     }
6044   }
6045 
6046   return strlen(buffer);
6047 }
6048 
6049 bool os::start_debugging(char *buf, int buflen) {
6050   int len = (int)strlen(buf);
6051   char *p = &buf[len];
6052 
6053   jio_snprintf(p, buflen-len,
6054                "\n\n"
6055                "Do you want to debug the problem?\n\n"
6056                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6057                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6058                "Otherwise, press RETURN to abort...",
6059                os::current_process_id(), os::current_process_id(),
6060                os::current_thread_id(), os::current_thread_id());
6061 
6062   bool yes = os::message_box("Unexpected Error", buf);
6063 
6064   if (yes) {
6065     // yes, user asked VM to launch debugger
6066     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6067                  os::current_process_id(), os::current_process_id());
6068 
6069     os::fork_and_exec(buf);
6070     yes = false;
6071   }
6072   return yes;
6073 }
6074 
6075 static inline struct timespec get_mtime(const char* filename) {
6076   struct stat st;
6077   int ret = os::stat(filename, &st);
6078   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6079   return st.st_mtim;
6080 }
6081 
6082 int os::compare_file_modified_times(const char* file1, const char* file2) {
6083   struct timespec filetime1 = get_mtime(file1);
6084   struct timespec filetime2 = get_mtime(file2);
6085   int diff = filetime1.tv_sec - filetime2.tv_sec;
6086   if (diff == 0) {
6087     return filetime1.tv_nsec - filetime2.tv_nsec;
6088   }
6089   return diff;
6090 }
6091 
6092 /////////////// Unit tests ///////////////
6093 
6094 #ifndef PRODUCT
6095 
6096 #define test_log(...)              \
6097   do {                             \
6098     if (VerboseInternalVMTests) {  \
6099       tty->print_cr(__VA_ARGS__);  \
6100       tty->flush();                \
6101     }                              \
6102   } while (false)
6103 
6104 class TestReserveMemorySpecial : AllStatic {
6105  public:
6106   static void small_page_write(void* addr, size_t size) {
6107     size_t page_size = os::vm_page_size();
6108 
6109     char* end = (char*)addr + size;
6110     for (char* p = (char*)addr; p < end; p += page_size) {
6111       *p = 1;
6112     }
6113   }
6114 
6115   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6116     if (!UseHugeTLBFS) {
6117       return;
6118     }
6119 
6120     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6121 
6122     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6123 
6124     if (addr != NULL) {
6125       small_page_write(addr, size);
6126 
6127       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6128     }
6129   }
6130 
6131   static void test_reserve_memory_special_huge_tlbfs_only() {
6132     if (!UseHugeTLBFS) {
6133       return;
6134     }
6135 
6136     size_t lp = os::large_page_size();
6137 
6138     for (size_t size = lp; size <= lp * 10; size += lp) {
6139       test_reserve_memory_special_huge_tlbfs_only(size);
6140     }
6141   }
6142 
6143   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6144     size_t lp = os::large_page_size();
6145     size_t ag = os::vm_allocation_granularity();
6146 
6147     // sizes to test
6148     const size_t sizes[] = {
6149       lp, lp + ag, lp + lp / 2, lp * 2,
6150       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6151       lp * 10, lp * 10 + lp / 2
6152     };
6153     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6154 
6155     // For each size/alignment combination, we test three scenarios:
6156     // 1) with req_addr == NULL
6157     // 2) with a non-null req_addr at which we expect to successfully allocate
6158     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6159     //    expect the allocation to either fail or to ignore req_addr
6160 
6161     // Pre-allocate two areas; they shall be as large as the largest allocation
6162     //  and aligned to the largest alignment we will be testing.
6163     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6164     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6165       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6166       -1, 0);
6167     assert(mapping1 != MAP_FAILED, "should work");
6168 
6169     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6170       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6171       -1, 0);
6172     assert(mapping2 != MAP_FAILED, "should work");
6173 
6174     // Unmap the first mapping, but leave the second mapping intact: the first
6175     // mapping will serve as a value for a "good" req_addr (case 2). The second
6176     // mapping, still intact, as "bad" req_addr (case 3).
6177     ::munmap(mapping1, mapping_size);
6178 
6179     // Case 1
6180     test_log("%s, req_addr NULL:", __FUNCTION__);
6181     test_log("size            align           result");
6182 
6183     for (int i = 0; i < num_sizes; i++) {
6184       const size_t size = sizes[i];
6185       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6186         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6187         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6188                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6189         if (p != NULL) {
6190           assert(is_ptr_aligned(p, alignment), "must be");
6191           small_page_write(p, size);
6192           os::Linux::release_memory_special_huge_tlbfs(p, size);
6193         }
6194       }
6195     }
6196 
6197     // Case 2
6198     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6199     test_log("size            align           req_addr         result");
6200 
6201     for (int i = 0; i < num_sizes; i++) {
6202       const size_t size = sizes[i];
6203       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6204         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6205         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6206         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6207                  size, alignment, p2i(req_addr), p2i(p),
6208                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6209         if (p != NULL) {
6210           assert(p == req_addr, "must be");
6211           small_page_write(p, size);
6212           os::Linux::release_memory_special_huge_tlbfs(p, size);
6213         }
6214       }
6215     }
6216 
6217     // Case 3
6218     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6219     test_log("size            align           req_addr         result");
6220 
6221     for (int i = 0; i < num_sizes; i++) {
6222       const size_t size = sizes[i];
6223       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6224         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6225         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6226         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6227                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6228         // as the area around req_addr contains already existing mappings, the API should always
6229         // return NULL (as per contract, it cannot return another address)
6230         assert(p == NULL, "must be");
6231       }
6232     }
6233 
6234     ::munmap(mapping2, mapping_size);
6235 
6236   }
6237 
6238   static void test_reserve_memory_special_huge_tlbfs() {
6239     if (!UseHugeTLBFS) {
6240       return;
6241     }
6242 
6243     test_reserve_memory_special_huge_tlbfs_only();
6244     test_reserve_memory_special_huge_tlbfs_mixed();
6245   }
6246 
6247   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6248     if (!UseSHM) {
6249       return;
6250     }
6251 
6252     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6253 
6254     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6255 
6256     if (addr != NULL) {
6257       assert(is_ptr_aligned(addr, alignment), "Check");
6258       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6259 
6260       small_page_write(addr, size);
6261 
6262       os::Linux::release_memory_special_shm(addr, size);
6263     }
6264   }
6265 
6266   static void test_reserve_memory_special_shm() {
6267     size_t lp = os::large_page_size();
6268     size_t ag = os::vm_allocation_granularity();
6269 
6270     for (size_t size = ag; size < lp * 3; size += ag) {
6271       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6272         test_reserve_memory_special_shm(size, alignment);
6273       }
6274     }
6275   }
6276 
6277   static void test() {
6278     test_reserve_memory_special_huge_tlbfs();
6279     test_reserve_memory_special_shm();
6280   }
6281 };
6282 
6283 void TestReserveMemorySpecial_test() {
6284   TestReserveMemorySpecial::test();
6285 }
6286 
6287 #endif