rev 12283 : 8169373: Work around linux NPTL stack guard error.
Summary: Also streamline OS guard page handling on linuxs390, linuxppc, aixppc.

   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   // In the Linux NPTL pthread implementation the guard size mechanism
 729   // is not implemented properly. The posix standard requires to add
 730   // the size of the guard pages to the stack size, instead Linux
 731   // takes the space out of 'stacksize'. Thus we adapt the requested
 732   // stack_size by the size of the guard pages to mimick proper
 733   // behaviour.
 734   stack_size = align_size_up(stack_size + os::Linux::default_guard_size(thr_type), vm_page_size());
 735   pthread_attr_setstacksize(&attr, stack_size);
 736 
 737   // glibc guard page
 738   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 739 
 740   ThreadState state;
 741 
 742   {
 743     pthread_t tid;
 744     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 745 
 746     char buf[64];
 747     if (ret == 0) {
 748       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 749         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 750     } else {
 751       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 752         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 753     }
 754 
 755     pthread_attr_destroy(&attr);
 756 
 757     if (ret != 0) {
 758       // Need to clean up stuff we've allocated so far
 759       thread->set_osthread(NULL);
 760       delete osthread;
 761       return false;
 762     }
 763 
 764     // Store pthread info into the OSThread
 765     osthread->set_pthread_id(tid);
 766 
 767     // Wait until child thread is either initialized or aborted
 768     {
 769       Monitor* sync_with_child = osthread->startThread_lock();
 770       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 771       while ((state = osthread->get_state()) == ALLOCATED) {
 772         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 773       }
 774     }
 775   }
 776 
 777   // Aborted due to thread limit being reached
 778   if (state == ZOMBIE) {
 779     thread->set_osthread(NULL);
 780     delete osthread;
 781     return false;
 782   }
 783 
 784   // The thread is returned suspended (in state INITIALIZED),
 785   // and is started higher up in the call chain
 786   assert(state == INITIALIZED, "race condition");
 787   return true;
 788 }
 789 
 790 /////////////////////////////////////////////////////////////////////////////
 791 // attach existing thread
 792 
 793 // bootstrap the main thread
 794 bool os::create_main_thread(JavaThread* thread) {
 795   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 796   return create_attached_thread(thread);
 797 }
 798 
 799 bool os::create_attached_thread(JavaThread* thread) {
 800 #ifdef ASSERT
 801   thread->verify_not_published();
 802 #endif
 803 
 804   // Allocate the OSThread object
 805   OSThread* osthread = new OSThread(NULL, NULL);
 806 
 807   if (osthread == NULL) {
 808     return false;
 809   }
 810 
 811   // Store pthread info into the OSThread
 812   osthread->set_thread_id(os::Linux::gettid());
 813   osthread->set_pthread_id(::pthread_self());
 814 
 815   // initialize floating point control register
 816   os::Linux::init_thread_fpu_state();
 817 
 818   // Initial thread state is RUNNABLE
 819   osthread->set_state(RUNNABLE);
 820 
 821   thread->set_osthread(osthread);
 822 
 823   if (UseNUMA) {
 824     int lgrp_id = os::numa_get_group_id();
 825     if (lgrp_id != -1) {
 826       thread->set_lgrp_id(lgrp_id);
 827     }
 828   }
 829 
 830   if (os::Linux::is_initial_thread()) {
 831     // If current thread is initial thread, its stack is mapped on demand,
 832     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 833     // the entire stack region to avoid SEGV in stack banging.
 834     // It is also useful to get around the heap-stack-gap problem on SuSE
 835     // kernel (see 4821821 for details). We first expand stack to the top
 836     // of yellow zone, then enable stack yellow zone (order is significant,
 837     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 838     // is no gap between the last two virtual memory regions.
 839 
 840     JavaThread *jt = (JavaThread *)thread;
 841     address addr = jt->stack_reserved_zone_base();
 842     assert(addr != NULL, "initialization problem?");
 843     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 844 
 845     osthread->set_expanding_stack();
 846     os::Linux::manually_expand_stack(jt, addr);
 847     osthread->clear_expanding_stack();
 848   }
 849 
 850   // initialize signal mask for this thread
 851   // and save the caller's signal mask
 852   os::Linux::hotspot_sigmask(thread);
 853 
 854   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 855     os::current_thread_id(), (uintx) pthread_self());
 856 
 857   return true;
 858 }
 859 
 860 void os::pd_start_thread(Thread* thread) {
 861   OSThread * osthread = thread->osthread();
 862   assert(osthread->get_state() != INITIALIZED, "just checking");
 863   Monitor* sync_with_child = osthread->startThread_lock();
 864   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 865   sync_with_child->notify();
 866 }
 867 
 868 // Free Linux resources related to the OSThread
 869 void os::free_thread(OSThread* osthread) {
 870   assert(osthread != NULL, "osthread not set");
 871 
 872   // We are told to free resources of the argument thread,
 873   // but we can only really operate on the current thread.
 874   assert(Thread::current()->osthread() == osthread,
 875          "os::free_thread but not current thread");
 876 
 877 #ifdef ASSERT
 878   sigset_t current;
 879   sigemptyset(&current);
 880   pthread_sigmask(SIG_SETMASK, NULL, &current);
 881   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 882 #endif
 883 
 884   // Restore caller's signal mask
 885   sigset_t sigmask = osthread->caller_sigmask();
 886   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 887 
 888   delete osthread;
 889 }
 890 
 891 //////////////////////////////////////////////////////////////////////////////
 892 // initial thread
 893 
 894 // Check if current thread is the initial thread, similar to Solaris thr_main.
 895 bool os::Linux::is_initial_thread(void) {
 896   char dummy;
 897   // If called before init complete, thread stack bottom will be null.
 898   // Can be called if fatal error occurs before initialization.
 899   if (initial_thread_stack_bottom() == NULL) return false;
 900   assert(initial_thread_stack_bottom() != NULL &&
 901          initial_thread_stack_size()   != 0,
 902          "os::init did not locate initial thread's stack region");
 903   if ((address)&dummy >= initial_thread_stack_bottom() &&
 904       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 905     return true;
 906   } else {
 907     return false;
 908   }
 909 }
 910 
 911 // Find the virtual memory area that contains addr
 912 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 913   FILE *fp = fopen("/proc/self/maps", "r");
 914   if (fp) {
 915     address low, high;
 916     while (!feof(fp)) {
 917       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 918         if (low <= addr && addr < high) {
 919           if (vma_low)  *vma_low  = low;
 920           if (vma_high) *vma_high = high;
 921           fclose(fp);
 922           return true;
 923         }
 924       }
 925       for (;;) {
 926         int ch = fgetc(fp);
 927         if (ch == EOF || ch == (int)'\n') break;
 928       }
 929     }
 930     fclose(fp);
 931   }
 932   return false;
 933 }
 934 
 935 // Locate initial thread stack. This special handling of initial thread stack
 936 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 937 // bogus value for initial thread.
 938 void os::Linux::capture_initial_stack(size_t max_size) {
 939   // stack size is the easy part, get it from RLIMIT_STACK
 940   struct rlimit rlim;
 941   getrlimit(RLIMIT_STACK, &rlim);
 942   size_t stack_size = rlim.rlim_cur;
 943 
 944   // 6308388: a bug in ld.so will relocate its own .data section to the
 945   //   lower end of primordial stack; reduce ulimit -s value a little bit
 946   //   so we won't install guard page on ld.so's data section.
 947   stack_size -= 2 * page_size();
 948 
 949   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
 950   //   7.1, in both cases we will get 2G in return value.
 951   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
 952   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
 953   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
 954   //   in case other parts in glibc still assumes 2M max stack size.
 955   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
 956   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
 957   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
 958     stack_size = 2 * K * K IA64_ONLY(*2);
 959   }
 960   // Try to figure out where the stack base (top) is. This is harder.
 961   //
 962   // When an application is started, glibc saves the initial stack pointer in
 963   // a global variable "__libc_stack_end", which is then used by system
 964   // libraries. __libc_stack_end should be pretty close to stack top. The
 965   // variable is available since the very early days. However, because it is
 966   // a private interface, it could disappear in the future.
 967   //
 968   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 969   // to __libc_stack_end, it is very close to stack top, but isn't the real
 970   // stack top. Note that /proc may not exist if VM is running as a chroot
 971   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 972   // /proc/<pid>/stat could change in the future (though unlikely).
 973   //
 974   // We try __libc_stack_end first. If that doesn't work, look for
 975   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
 976   // as a hint, which should work well in most cases.
 977 
 978   uintptr_t stack_start;
 979 
 980   // try __libc_stack_end first
 981   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
 982   if (p && *p) {
 983     stack_start = *p;
 984   } else {
 985     // see if we can get the start_stack field from /proc/self/stat
 986     FILE *fp;
 987     int pid;
 988     char state;
 989     int ppid;
 990     int pgrp;
 991     int session;
 992     int nr;
 993     int tpgrp;
 994     unsigned long flags;
 995     unsigned long minflt;
 996     unsigned long cminflt;
 997     unsigned long majflt;
 998     unsigned long cmajflt;
 999     unsigned long utime;
1000     unsigned long stime;
1001     long cutime;
1002     long cstime;
1003     long prio;
1004     long nice;
1005     long junk;
1006     long it_real;
1007     uintptr_t start;
1008     uintptr_t vsize;
1009     intptr_t rss;
1010     uintptr_t rsslim;
1011     uintptr_t scodes;
1012     uintptr_t ecode;
1013     int i;
1014 
1015     // Figure what the primordial thread stack base is. Code is inspired
1016     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1017     // followed by command name surrounded by parentheses, state, etc.
1018     char stat[2048];
1019     int statlen;
1020 
1021     fp = fopen("/proc/self/stat", "r");
1022     if (fp) {
1023       statlen = fread(stat, 1, 2047, fp);
1024       stat[statlen] = '\0';
1025       fclose(fp);
1026 
1027       // Skip pid and the command string. Note that we could be dealing with
1028       // weird command names, e.g. user could decide to rename java launcher
1029       // to "java 1.4.2 :)", then the stat file would look like
1030       //                1234 (java 1.4.2 :)) R ... ...
1031       // We don't really need to know the command string, just find the last
1032       // occurrence of ")" and then start parsing from there. See bug 4726580.
1033       char * s = strrchr(stat, ')');
1034 
1035       i = 0;
1036       if (s) {
1037         // Skip blank chars
1038         do { s++; } while (s && isspace(*s));
1039 
1040 #define _UFM UINTX_FORMAT
1041 #define _DFM INTX_FORMAT
1042 
1043         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1044         //              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
1045         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,
1046                    &state,          // 3  %c
1047                    &ppid,           // 4  %d
1048                    &pgrp,           // 5  %d
1049                    &session,        // 6  %d
1050                    &nr,             // 7  %d
1051                    &tpgrp,          // 8  %d
1052                    &flags,          // 9  %lu
1053                    &minflt,         // 10 %lu
1054                    &cminflt,        // 11 %lu
1055                    &majflt,         // 12 %lu
1056                    &cmajflt,        // 13 %lu
1057                    &utime,          // 14 %lu
1058                    &stime,          // 15 %lu
1059                    &cutime,         // 16 %ld
1060                    &cstime,         // 17 %ld
1061                    &prio,           // 18 %ld
1062                    &nice,           // 19 %ld
1063                    &junk,           // 20 %ld
1064                    &it_real,        // 21 %ld
1065                    &start,          // 22 UINTX_FORMAT
1066                    &vsize,          // 23 UINTX_FORMAT
1067                    &rss,            // 24 INTX_FORMAT
1068                    &rsslim,         // 25 UINTX_FORMAT
1069                    &scodes,         // 26 UINTX_FORMAT
1070                    &ecode,          // 27 UINTX_FORMAT
1071                    &stack_start);   // 28 UINTX_FORMAT
1072       }
1073 
1074 #undef _UFM
1075 #undef _DFM
1076 
1077       if (i != 28 - 2) {
1078         assert(false, "Bad conversion from /proc/self/stat");
1079         // product mode - assume we are the initial thread, good luck in the
1080         // embedded case.
1081         warning("Can't detect initial thread stack location - bad conversion");
1082         stack_start = (uintptr_t) &rlim;
1083       }
1084     } else {
1085       // For some reason we can't open /proc/self/stat (for example, running on
1086       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1087       // most cases, so don't abort:
1088       warning("Can't detect initial thread stack location - no /proc/self/stat");
1089       stack_start = (uintptr_t) &rlim;
1090     }
1091   }
1092 
1093   // Now we have a pointer (stack_start) very close to the stack top, the
1094   // next thing to do is to figure out the exact location of stack top. We
1095   // can find out the virtual memory area that contains stack_start by
1096   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1097   // and its upper limit is the real stack top. (again, this would fail if
1098   // running inside chroot, because /proc may not exist.)
1099 
1100   uintptr_t stack_top;
1101   address low, high;
1102   if (find_vma((address)stack_start, &low, &high)) {
1103     // success, "high" is the true stack top. (ignore "low", because initial
1104     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1105     stack_top = (uintptr_t)high;
1106   } else {
1107     // failed, likely because /proc/self/maps does not exist
1108     warning("Can't detect initial thread stack location - find_vma failed");
1109     // best effort: stack_start is normally within a few pages below the real
1110     // stack top, use it as stack top, and reduce stack size so we won't put
1111     // guard page outside stack.
1112     stack_top = stack_start;
1113     stack_size -= 16 * page_size();
1114   }
1115 
1116   // stack_top could be partially down the page so align it
1117   stack_top = align_size_up(stack_top, page_size());
1118 
1119   if (max_size && stack_size > max_size) {
1120     _initial_thread_stack_size = max_size;
1121   } else {
1122     _initial_thread_stack_size = stack_size;
1123   }
1124 
1125   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1126   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1127 }
1128 
1129 ////////////////////////////////////////////////////////////////////////////////
1130 // time support
1131 
1132 // Time since start-up in seconds to a fine granularity.
1133 // Used by VMSelfDestructTimer and the MemProfiler.
1134 double os::elapsedTime() {
1135 
1136   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1137 }
1138 
1139 jlong os::elapsed_counter() {
1140   return javaTimeNanos() - initial_time_count;
1141 }
1142 
1143 jlong os::elapsed_frequency() {
1144   return NANOSECS_PER_SEC; // nanosecond resolution
1145 }
1146 
1147 bool os::supports_vtime() { return true; }
1148 bool os::enable_vtime()   { return false; }
1149 bool os::vtime_enabled()  { return false; }
1150 
1151 double os::elapsedVTime() {
1152   struct rusage usage;
1153   int retval = getrusage(RUSAGE_THREAD, &usage);
1154   if (retval == 0) {
1155     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);
1156   } else {
1157     // better than nothing, but not much
1158     return elapsedTime();
1159   }
1160 }
1161 
1162 jlong os::javaTimeMillis() {
1163   timeval time;
1164   int status = gettimeofday(&time, NULL);
1165   assert(status != -1, "linux error");
1166   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1167 }
1168 
1169 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1170   timeval time;
1171   int status = gettimeofday(&time, NULL);
1172   assert(status != -1, "linux error");
1173   seconds = jlong(time.tv_sec);
1174   nanos = jlong(time.tv_usec) * 1000;
1175 }
1176 
1177 
1178 #ifndef CLOCK_MONOTONIC
1179   #define CLOCK_MONOTONIC (1)
1180 #endif
1181 
1182 void os::Linux::clock_init() {
1183   // we do dlopen's in this particular order due to bug in linux
1184   // dynamical loader (see 6348968) leading to crash on exit
1185   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1186   if (handle == NULL) {
1187     handle = dlopen("librt.so", RTLD_LAZY);
1188   }
1189 
1190   if (handle) {
1191     int (*clock_getres_func)(clockid_t, struct timespec*) =
1192            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1193     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1194            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1195     if (clock_getres_func && clock_gettime_func) {
1196       // See if monotonic clock is supported by the kernel. Note that some
1197       // early implementations simply return kernel jiffies (updated every
1198       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1199       // for nano time (though the monotonic property is still nice to have).
1200       // It's fixed in newer kernels, however clock_getres() still returns
1201       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1202       // resolution for now. Hopefully as people move to new kernels, this
1203       // won't be a problem.
1204       struct timespec res;
1205       struct timespec tp;
1206       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1207           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1208         // yes, monotonic clock is supported
1209         _clock_gettime = clock_gettime_func;
1210         return;
1211       } else {
1212         // close librt if there is no monotonic clock
1213         dlclose(handle);
1214       }
1215     }
1216   }
1217   warning("No monotonic clock was available - timed services may " \
1218           "be adversely affected if the time-of-day clock changes");
1219 }
1220 
1221 #ifndef SYS_clock_getres
1222   #if defined(X86) || defined(PPC64) || defined(S390)
1223     #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1224     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1225   #else
1226     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1227     #define sys_clock_getres(x,y)  -1
1228   #endif
1229 #else
1230   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1231 #endif
1232 
1233 void os::Linux::fast_thread_clock_init() {
1234   if (!UseLinuxPosixThreadCPUClocks) {
1235     return;
1236   }
1237   clockid_t clockid;
1238   struct timespec tp;
1239   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1240       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1241 
1242   // Switch to using fast clocks for thread cpu time if
1243   // the sys_clock_getres() returns 0 error code.
1244   // Note, that some kernels may support the current thread
1245   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1246   // returned by the pthread_getcpuclockid().
1247   // If the fast Posix clocks are supported then the sys_clock_getres()
1248   // must return at least tp.tv_sec == 0 which means a resolution
1249   // better than 1 sec. This is extra check for reliability.
1250 
1251   if (pthread_getcpuclockid_func &&
1252       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1253       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1254     _supports_fast_thread_cpu_time = true;
1255     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1256   }
1257 }
1258 
1259 jlong os::javaTimeNanos() {
1260   if (os::supports_monotonic_clock()) {
1261     struct timespec tp;
1262     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1263     assert(status == 0, "gettime error");
1264     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1265     return result;
1266   } else {
1267     timeval time;
1268     int status = gettimeofday(&time, NULL);
1269     assert(status != -1, "linux error");
1270     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1271     return 1000 * usecs;
1272   }
1273 }
1274 
1275 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1276   if (os::supports_monotonic_clock()) {
1277     info_ptr->max_value = ALL_64_BITS;
1278 
1279     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1280     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1281     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1282   } else {
1283     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1284     info_ptr->max_value = ALL_64_BITS;
1285 
1286     // gettimeofday is a real time clock so it skips
1287     info_ptr->may_skip_backward = true;
1288     info_ptr->may_skip_forward = true;
1289   }
1290 
1291   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1292 }
1293 
1294 // Return the real, user, and system times in seconds from an
1295 // arbitrary fixed point in the past.
1296 bool os::getTimesSecs(double* process_real_time,
1297                       double* process_user_time,
1298                       double* process_system_time) {
1299   struct tms ticks;
1300   clock_t real_ticks = times(&ticks);
1301 
1302   if (real_ticks == (clock_t) (-1)) {
1303     return false;
1304   } else {
1305     double ticks_per_second = (double) clock_tics_per_sec;
1306     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1307     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1308     *process_real_time = ((double) real_ticks) / ticks_per_second;
1309 
1310     return true;
1311   }
1312 }
1313 
1314 
1315 char * os::local_time_string(char *buf, size_t buflen) {
1316   struct tm t;
1317   time_t long_time;
1318   time(&long_time);
1319   localtime_r(&long_time, &t);
1320   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1321                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1322                t.tm_hour, t.tm_min, t.tm_sec);
1323   return buf;
1324 }
1325 
1326 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1327   return localtime_r(clock, res);
1328 }
1329 
1330 ////////////////////////////////////////////////////////////////////////////////
1331 // runtime exit support
1332 
1333 // Note: os::shutdown() might be called very early during initialization, or
1334 // called from signal handler. Before adding something to os::shutdown(), make
1335 // sure it is async-safe and can handle partially initialized VM.
1336 void os::shutdown() {
1337 
1338   // allow PerfMemory to attempt cleanup of any persistent resources
1339   perfMemory_exit();
1340 
1341   // needs to remove object in file system
1342   AttachListener::abort();
1343 
1344   // flush buffered output, finish log files
1345   ostream_abort();
1346 
1347   // Check for abort hook
1348   abort_hook_t abort_hook = Arguments::abort_hook();
1349   if (abort_hook != NULL) {
1350     abort_hook();
1351   }
1352 
1353 }
1354 
1355 // Note: os::abort() might be called very early during initialization, or
1356 // called from signal handler. Before adding something to os::abort(), make
1357 // sure it is async-safe and can handle partially initialized VM.
1358 void os::abort(bool dump_core, void* siginfo, const void* context) {
1359   os::shutdown();
1360   if (dump_core) {
1361 #ifndef PRODUCT
1362     fdStream out(defaultStream::output_fd());
1363     out.print_raw("Current thread is ");
1364     char buf[16];
1365     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1366     out.print_raw_cr(buf);
1367     out.print_raw_cr("Dumping core ...");
1368 #endif
1369     ::abort(); // dump core
1370   }
1371 
1372   ::exit(1);
1373 }
1374 
1375 // Die immediately, no exit hook, no abort hook, no cleanup.
1376 void os::die() {
1377   ::abort();
1378 }
1379 
1380 
1381 // This method is a copy of JDK's sysGetLastErrorString
1382 // from src/solaris/hpi/src/system_md.c
1383 
1384 size_t os::lasterror(char *buf, size_t len) {
1385   if (errno == 0)  return 0;
1386 
1387   const char *s = os::strerror(errno);
1388   size_t n = ::strlen(s);
1389   if (n >= len) {
1390     n = len - 1;
1391   }
1392   ::strncpy(buf, s, n);
1393   buf[n] = '\0';
1394   return n;
1395 }
1396 
1397 // thread_id is kernel thread id (similar to Solaris LWP id)
1398 intx os::current_thread_id() { return os::Linux::gettid(); }
1399 int os::current_process_id() {
1400   return ::getpid();
1401 }
1402 
1403 // DLL functions
1404 
1405 const char* os::dll_file_extension() { return ".so"; }
1406 
1407 // This must be hard coded because it's the system's temporary
1408 // directory not the java application's temp directory, ala java.io.tmpdir.
1409 const char* os::get_temp_directory() { return "/tmp"; }
1410 
1411 static bool file_exists(const char* filename) {
1412   struct stat statbuf;
1413   if (filename == NULL || strlen(filename) == 0) {
1414     return false;
1415   }
1416   return os::stat(filename, &statbuf) == 0;
1417 }
1418 
1419 bool os::dll_build_name(char* buffer, size_t buflen,
1420                         const char* pname, const char* fname) {
1421   bool retval = false;
1422   // Copied from libhpi
1423   const size_t pnamelen = pname ? strlen(pname) : 0;
1424 
1425   // Return error on buffer overflow.
1426   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1427     return retval;
1428   }
1429 
1430   if (pnamelen == 0) {
1431     snprintf(buffer, buflen, "lib%s.so", fname);
1432     retval = true;
1433   } else if (strchr(pname, *os::path_separator()) != NULL) {
1434     int n;
1435     char** pelements = split_path(pname, &n);
1436     if (pelements == NULL) {
1437       return false;
1438     }
1439     for (int i = 0; i < n; i++) {
1440       // Really shouldn't be NULL, but check can't hurt
1441       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1442         continue; // skip the empty path values
1443       }
1444       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1445       if (file_exists(buffer)) {
1446         retval = true;
1447         break;
1448       }
1449     }
1450     // release the storage
1451     for (int i = 0; i < n; i++) {
1452       if (pelements[i] != NULL) {
1453         FREE_C_HEAP_ARRAY(char, pelements[i]);
1454       }
1455     }
1456     if (pelements != NULL) {
1457       FREE_C_HEAP_ARRAY(char*, pelements);
1458     }
1459   } else {
1460     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1461     retval = true;
1462   }
1463   return retval;
1464 }
1465 
1466 // check if addr is inside libjvm.so
1467 bool os::address_is_in_vm(address addr) {
1468   static address libjvm_base_addr;
1469   Dl_info dlinfo;
1470 
1471   if (libjvm_base_addr == NULL) {
1472     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1473       libjvm_base_addr = (address)dlinfo.dli_fbase;
1474     }
1475     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1476   }
1477 
1478   if (dladdr((void *)addr, &dlinfo) != 0) {
1479     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1480   }
1481 
1482   return false;
1483 }
1484 
1485 bool os::dll_address_to_function_name(address addr, char *buf,
1486                                       int buflen, int *offset,
1487                                       bool demangle) {
1488   // buf is not optional, but offset is optional
1489   assert(buf != NULL, "sanity check");
1490 
1491   Dl_info dlinfo;
1492 
1493   if (dladdr((void*)addr, &dlinfo) != 0) {
1494     // see if we have a matching symbol
1495     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1496       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1497         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1498       }
1499       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1500       return true;
1501     }
1502     // no matching symbol so try for just file info
1503     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1504       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1505                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1506         return true;
1507       }
1508     }
1509   }
1510 
1511   buf[0] = '\0';
1512   if (offset != NULL) *offset = -1;
1513   return false;
1514 }
1515 
1516 struct _address_to_library_name {
1517   address addr;          // input : memory address
1518   size_t  buflen;        //         size of fname
1519   char*   fname;         // output: library name
1520   address base;          //         library base addr
1521 };
1522 
1523 static int address_to_library_name_callback(struct dl_phdr_info *info,
1524                                             size_t size, void *data) {
1525   int i;
1526   bool found = false;
1527   address libbase = NULL;
1528   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1529 
1530   // iterate through all loadable segments
1531   for (i = 0; i < info->dlpi_phnum; i++) {
1532     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1533     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1534       // base address of a library is the lowest address of its loaded
1535       // segments.
1536       if (libbase == NULL || libbase > segbase) {
1537         libbase = segbase;
1538       }
1539       // see if 'addr' is within current segment
1540       if (segbase <= d->addr &&
1541           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1542         found = true;
1543       }
1544     }
1545   }
1546 
1547   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1548   // so dll_address_to_library_name() can fall through to use dladdr() which
1549   // can figure out executable name from argv[0].
1550   if (found && info->dlpi_name && info->dlpi_name[0]) {
1551     d->base = libbase;
1552     if (d->fname) {
1553       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1554     }
1555     return 1;
1556   }
1557   return 0;
1558 }
1559 
1560 bool os::dll_address_to_library_name(address addr, char* buf,
1561                                      int buflen, int* offset) {
1562   // buf is not optional, but offset is optional
1563   assert(buf != NULL, "sanity check");
1564 
1565   Dl_info dlinfo;
1566   struct _address_to_library_name data;
1567 
1568   // There is a bug in old glibc dladdr() implementation that it could resolve
1569   // to wrong library name if the .so file has a base address != NULL. Here
1570   // we iterate through the program headers of all loaded libraries to find
1571   // out which library 'addr' really belongs to. This workaround can be
1572   // removed once the minimum requirement for glibc is moved to 2.3.x.
1573   data.addr = addr;
1574   data.fname = buf;
1575   data.buflen = buflen;
1576   data.base = NULL;
1577   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1578 
1579   if (rslt) {
1580     // buf already contains library name
1581     if (offset) *offset = addr - data.base;
1582     return true;
1583   }
1584   if (dladdr((void*)addr, &dlinfo) != 0) {
1585     if (dlinfo.dli_fname != NULL) {
1586       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1587     }
1588     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1589       *offset = addr - (address)dlinfo.dli_fbase;
1590     }
1591     return true;
1592   }
1593 
1594   buf[0] = '\0';
1595   if (offset) *offset = -1;
1596   return false;
1597 }
1598 
1599 // Loads .dll/.so and
1600 // in case of error it checks if .dll/.so was built for the
1601 // same architecture as Hotspot is running on
1602 
1603 
1604 // Remember the stack's state. The Linux dynamic linker will change
1605 // the stack to 'executable' at most once, so we must safepoint only once.
1606 bool os::Linux::_stack_is_executable = false;
1607 
1608 // VM operation that loads a library.  This is necessary if stack protection
1609 // of the Java stacks can be lost during loading the library.  If we
1610 // do not stop the Java threads, they can stack overflow before the stacks
1611 // are protected again.
1612 class VM_LinuxDllLoad: public VM_Operation {
1613  private:
1614   const char *_filename;
1615   char *_ebuf;
1616   int _ebuflen;
1617   void *_lib;
1618  public:
1619   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1620     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1621   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1622   void doit() {
1623     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1624     os::Linux::_stack_is_executable = true;
1625   }
1626   void* loaded_library() { return _lib; }
1627 };
1628 
1629 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1630   void * result = NULL;
1631   bool load_attempted = false;
1632 
1633   // Check whether the library to load might change execution rights
1634   // of the stack. If they are changed, the protection of the stack
1635   // guard pages will be lost. We need a safepoint to fix this.
1636   //
1637   // See Linux man page execstack(8) for more info.
1638   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1639     ElfFile ef(filename);
1640     if (!ef.specifies_noexecstack()) {
1641       if (!is_init_completed()) {
1642         os::Linux::_stack_is_executable = true;
1643         // This is OK - No Java threads have been created yet, and hence no
1644         // stack guard pages to fix.
1645         //
1646         // This should happen only when you are building JDK7 using a very
1647         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1648         //
1649         // Dynamic loader will make all stacks executable after
1650         // this function returns, and will not do that again.
1651         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1652       } else {
1653         warning("You have loaded library %s which might have disabled stack guard. "
1654                 "The VM will try to fix the stack guard now.\n"
1655                 "It's highly recommended that you fix the library with "
1656                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1657                 filename);
1658 
1659         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1660         JavaThread *jt = JavaThread::current();
1661         if (jt->thread_state() != _thread_in_native) {
1662           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1663           // that requires ExecStack. Cannot enter safe point. Let's give up.
1664           warning("Unable to fix stack guard. Giving up.");
1665         } else {
1666           if (!LoadExecStackDllInVMThread) {
1667             // This is for the case where the DLL has an static
1668             // constructor function that executes JNI code. We cannot
1669             // load such DLLs in the VMThread.
1670             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1671           }
1672 
1673           ThreadInVMfromNative tiv(jt);
1674           debug_only(VMNativeEntryWrapper vew;)
1675 
1676           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1677           VMThread::execute(&op);
1678           if (LoadExecStackDllInVMThread) {
1679             result = op.loaded_library();
1680           }
1681           load_attempted = true;
1682         }
1683       }
1684     }
1685   }
1686 
1687   if (!load_attempted) {
1688     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1689   }
1690 
1691   if (result != NULL) {
1692     // Successful loading
1693     return result;
1694   }
1695 
1696   Elf32_Ehdr elf_head;
1697   int diag_msg_max_length=ebuflen-strlen(ebuf);
1698   char* diag_msg_buf=ebuf+strlen(ebuf);
1699 
1700   if (diag_msg_max_length==0) {
1701     // No more space in ebuf for additional diagnostics message
1702     return NULL;
1703   }
1704 
1705 
1706   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1707 
1708   if (file_descriptor < 0) {
1709     // Can't open library, report dlerror() message
1710     return NULL;
1711   }
1712 
1713   bool failed_to_read_elf_head=
1714     (sizeof(elf_head)!=
1715      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1716 
1717   ::close(file_descriptor);
1718   if (failed_to_read_elf_head) {
1719     // file i/o error - report dlerror() msg
1720     return NULL;
1721   }
1722 
1723   typedef struct {
1724     Elf32_Half    code;         // Actual value as defined in elf.h
1725     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1726     unsigned char elf_class;    // 32 or 64 bit
1727     unsigned char endianess;    // MSB or LSB
1728     char*         name;         // String representation
1729   } arch_t;
1730 
1731 #ifndef EM_486
1732   #define EM_486          6               /* Intel 80486 */
1733 #endif
1734 #ifndef EM_AARCH64
1735   #define EM_AARCH64    183               /* ARM AARCH64 */
1736 #endif
1737 
1738   static const arch_t arch_array[]={
1739     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1740     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1741     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1742     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1743     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1744     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1745     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1746     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1747 #if defined(VM_LITTLE_ENDIAN)
1748     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1749 #else
1750     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1751 #endif
1752     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1753     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1754     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1755     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1756     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1757     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1758     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1759     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1760   };
1761 
1762 #if  (defined IA32)
1763   static  Elf32_Half running_arch_code=EM_386;
1764 #elif   (defined AMD64)
1765   static  Elf32_Half running_arch_code=EM_X86_64;
1766 #elif  (defined IA64)
1767   static  Elf32_Half running_arch_code=EM_IA_64;
1768 #elif  (defined __sparc) && (defined _LP64)
1769   static  Elf32_Half running_arch_code=EM_SPARCV9;
1770 #elif  (defined __sparc) && (!defined _LP64)
1771   static  Elf32_Half running_arch_code=EM_SPARC;
1772 #elif  (defined __powerpc64__)
1773   static  Elf32_Half running_arch_code=EM_PPC64;
1774 #elif  (defined __powerpc__)
1775   static  Elf32_Half running_arch_code=EM_PPC;
1776 #elif  (defined AARCH64)
1777   static  Elf32_Half running_arch_code=EM_AARCH64;
1778 #elif  (defined ARM)
1779   static  Elf32_Half running_arch_code=EM_ARM;
1780 #elif  (defined S390)
1781   static  Elf32_Half running_arch_code=EM_S390;
1782 #elif  (defined ALPHA)
1783   static  Elf32_Half running_arch_code=EM_ALPHA;
1784 #elif  (defined MIPSEL)
1785   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1786 #elif  (defined PARISC)
1787   static  Elf32_Half running_arch_code=EM_PARISC;
1788 #elif  (defined MIPS)
1789   static  Elf32_Half running_arch_code=EM_MIPS;
1790 #elif  (defined M68K)
1791   static  Elf32_Half running_arch_code=EM_68K;
1792 #else
1793     #error Method os::dll_load requires that one of following is defined:\
1794         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, __sparc
1795 #endif
1796 
1797   // Identify compatability class for VM's architecture and library's architecture
1798   // Obtain string descriptions for architectures
1799 
1800   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1801   int running_arch_index=-1;
1802 
1803   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1804     if (running_arch_code == arch_array[i].code) {
1805       running_arch_index    = i;
1806     }
1807     if (lib_arch.code == arch_array[i].code) {
1808       lib_arch.compat_class = arch_array[i].compat_class;
1809       lib_arch.name         = arch_array[i].name;
1810     }
1811   }
1812 
1813   assert(running_arch_index != -1,
1814          "Didn't find running architecture code (running_arch_code) in arch_array");
1815   if (running_arch_index == -1) {
1816     // Even though running architecture detection failed
1817     // we may still continue with reporting dlerror() message
1818     return NULL;
1819   }
1820 
1821   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1822     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1823     return NULL;
1824   }
1825 
1826 #ifndef S390
1827   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1828     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1829     return NULL;
1830   }
1831 #endif // !S390
1832 
1833   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1834     if (lib_arch.name!=NULL) {
1835       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1836                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1837                  lib_arch.name, arch_array[running_arch_index].name);
1838     } else {
1839       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1840                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1841                  lib_arch.code,
1842                  arch_array[running_arch_index].name);
1843     }
1844   }
1845 
1846   return NULL;
1847 }
1848 
1849 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1850                                 int ebuflen) {
1851   void * result = ::dlopen(filename, RTLD_LAZY);
1852   if (result == NULL) {
1853     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1854     ebuf[ebuflen-1] = '\0';
1855   }
1856   return result;
1857 }
1858 
1859 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1860                                        int ebuflen) {
1861   void * result = NULL;
1862   if (LoadExecStackDllInVMThread) {
1863     result = dlopen_helper(filename, ebuf, ebuflen);
1864   }
1865 
1866   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1867   // library that requires an executable stack, or which does not have this
1868   // stack attribute set, dlopen changes the stack attribute to executable. The
1869   // read protection of the guard pages gets lost.
1870   //
1871   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1872   // may have been queued at the same time.
1873 
1874   if (!_stack_is_executable) {
1875     JavaThread *jt = Threads::first();
1876 
1877     while (jt) {
1878       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1879           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1880         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1881           warning("Attempt to reguard stack yellow zone failed.");
1882         }
1883       }
1884       jt = jt->next();
1885     }
1886   }
1887 
1888   return result;
1889 }
1890 
1891 void* os::dll_lookup(void* handle, const char* name) {
1892   void* res = dlsym(handle, name);
1893   return res;
1894 }
1895 
1896 void* os::get_default_process_handle() {
1897   return (void*)::dlopen(NULL, RTLD_LAZY);
1898 }
1899 
1900 static bool _print_ascii_file(const char* filename, outputStream* st) {
1901   int fd = ::open(filename, O_RDONLY);
1902   if (fd == -1) {
1903     return false;
1904   }
1905 
1906   char buf[33];
1907   int bytes;
1908   buf[32] = '\0';
1909   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1910     st->print_raw(buf, bytes);
1911   }
1912 
1913   ::close(fd);
1914 
1915   return true;
1916 }
1917 
1918 void os::print_dll_info(outputStream *st) {
1919   st->print_cr("Dynamic libraries:");
1920 
1921   char fname[32];
1922   pid_t pid = os::Linux::gettid();
1923 
1924   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1925 
1926   if (!_print_ascii_file(fname, st)) {
1927     st->print("Can not get library information for pid = %d\n", pid);
1928   }
1929 }
1930 
1931 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1932   FILE *procmapsFile = NULL;
1933 
1934   // Open the procfs maps file for the current process
1935   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1936     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1937     char line[PATH_MAX + 100];
1938 
1939     // Read line by line from 'file'
1940     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1941       u8 base, top, offset, inode;
1942       char permissions[5];
1943       char device[6];
1944       char name[PATH_MAX + 1];
1945 
1946       // Parse fields from line
1947       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1948              &base, &top, permissions, &offset, device, &inode, name);
1949 
1950       // Filter by device id '00:00' so that we only get file system mapped files.
1951       if (strcmp(device, "00:00") != 0) {
1952 
1953         // Call callback with the fields of interest
1954         if(callback(name, (address)base, (address)top, param)) {
1955           // Oops abort, callback aborted
1956           fclose(procmapsFile);
1957           return 1;
1958         }
1959       }
1960     }
1961     fclose(procmapsFile);
1962   }
1963   return 0;
1964 }
1965 
1966 void os::print_os_info_brief(outputStream* st) {
1967   os::Linux::print_distro_info(st);
1968 
1969   os::Posix::print_uname_info(st);
1970 
1971   os::Linux::print_libversion_info(st);
1972 
1973 }
1974 
1975 void os::print_os_info(outputStream* st) {
1976   st->print("OS:");
1977 
1978   os::Linux::print_distro_info(st);
1979 
1980   os::Posix::print_uname_info(st);
1981 
1982   // Print warning if unsafe chroot environment detected
1983   if (unsafe_chroot_detected) {
1984     st->print("WARNING!! ");
1985     st->print_cr("%s", unstable_chroot_error);
1986   }
1987 
1988   os::Linux::print_libversion_info(st);
1989 
1990   os::Posix::print_rlimit_info(st);
1991 
1992   os::Posix::print_load_average(st);
1993 
1994   os::Linux::print_full_memory_info(st);
1995 }
1996 
1997 // Try to identify popular distros.
1998 // Most Linux distributions have a /etc/XXX-release file, which contains
1999 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2000 // file that also contains the OS version string. Some have more than one
2001 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2002 // /etc/redhat-release.), so the order is important.
2003 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2004 // their own specific XXX-release file as well as a redhat-release file.
2005 // Because of this the XXX-release file needs to be searched for before the
2006 // redhat-release file.
2007 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2008 // search for redhat-release / SuSE-release needs to be before lsb-release.
2009 // Since the lsb-release file is the new standard it needs to be searched
2010 // before the older style release files.
2011 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2012 // next to last resort.  The os-release file is a new standard that contains
2013 // distribution information and the system-release file seems to be an old
2014 // standard that has been replaced by the lsb-release and os-release files.
2015 // Searching for the debian_version file is the last resort.  It contains
2016 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2017 // "Debian " is printed before the contents of the debian_version file.
2018 
2019 const char* distro_files[] = {
2020   "/etc/oracle-release",
2021   "/etc/mandriva-release",
2022   "/etc/mandrake-release",
2023   "/etc/sun-release",
2024   "/etc/redhat-release",
2025   "/etc/SuSE-release",
2026   "/etc/lsb-release",
2027   "/etc/turbolinux-release",
2028   "/etc/gentoo-release",
2029   "/etc/ltib-release",
2030   "/etc/angstrom-version",
2031   "/etc/system-release",
2032   "/etc/os-release",
2033   NULL };
2034 
2035 void os::Linux::print_distro_info(outputStream* st) {
2036   for (int i = 0;; i++) {
2037     const char* file = distro_files[i];
2038     if (file == NULL) {
2039       break;  // done
2040     }
2041     // If file prints, we found it.
2042     if (_print_ascii_file(file, st)) {
2043       return;
2044     }
2045   }
2046 
2047   if (file_exists("/etc/debian_version")) {
2048     st->print("Debian ");
2049     _print_ascii_file("/etc/debian_version", st);
2050   } else {
2051     st->print("Linux");
2052   }
2053   st->cr();
2054 }
2055 
2056 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2057   char buf[256];
2058   while (fgets(buf, sizeof(buf), fp)) {
2059     // Edit out extra stuff in expected format
2060     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2061       char* ptr = strstr(buf, "\"");  // the name is in quotes
2062       if (ptr != NULL) {
2063         ptr++; // go beyond first quote
2064         char* nl = strchr(ptr, '\"');
2065         if (nl != NULL) *nl = '\0';
2066         strncpy(distro, ptr, length);
2067       } else {
2068         ptr = strstr(buf, "=");
2069         ptr++; // go beyond equals then
2070         char* nl = strchr(ptr, '\n');
2071         if (nl != NULL) *nl = '\0';
2072         strncpy(distro, ptr, length);
2073       }
2074       return;
2075     } else if (get_first_line) {
2076       char* nl = strchr(buf, '\n');
2077       if (nl != NULL) *nl = '\0';
2078       strncpy(distro, buf, length);
2079       return;
2080     }
2081   }
2082   // print last line and close
2083   char* nl = strchr(buf, '\n');
2084   if (nl != NULL) *nl = '\0';
2085   strncpy(distro, buf, length);
2086 }
2087 
2088 static void parse_os_info(char* distro, size_t length, const char* file) {
2089   FILE* fp = fopen(file, "r");
2090   if (fp != NULL) {
2091     // if suse format, print out first line
2092     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2093     parse_os_info_helper(fp, distro, length, get_first_line);
2094     fclose(fp);
2095   }
2096 }
2097 
2098 void os::get_summary_os_info(char* buf, size_t buflen) {
2099   for (int i = 0;; i++) {
2100     const char* file = distro_files[i];
2101     if (file == NULL) {
2102       break; // ran out of distro_files
2103     }
2104     if (file_exists(file)) {
2105       parse_os_info(buf, buflen, file);
2106       return;
2107     }
2108   }
2109   // special case for debian
2110   if (file_exists("/etc/debian_version")) {
2111     strncpy(buf, "Debian ", buflen);
2112     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2113   } else {
2114     strncpy(buf, "Linux", buflen);
2115   }
2116 }
2117 
2118 void os::Linux::print_libversion_info(outputStream* st) {
2119   // libc, pthread
2120   st->print("libc:");
2121   st->print("%s ", os::Linux::glibc_version());
2122   st->print("%s ", os::Linux::libpthread_version());
2123   st->cr();
2124 }
2125 
2126 void os::Linux::print_full_memory_info(outputStream* st) {
2127   st->print("\n/proc/meminfo:\n");
2128   _print_ascii_file("/proc/meminfo", st);
2129   st->cr();
2130 }
2131 
2132 void os::print_memory_info(outputStream* st) {
2133 
2134   st->print("Memory:");
2135   st->print(" %dk page", os::vm_page_size()>>10);
2136 
2137   // values in struct sysinfo are "unsigned long"
2138   struct sysinfo si;
2139   sysinfo(&si);
2140 
2141   st->print(", physical " UINT64_FORMAT "k",
2142             os::physical_memory() >> 10);
2143   st->print("(" UINT64_FORMAT "k free)",
2144             os::available_memory() >> 10);
2145   st->print(", swap " UINT64_FORMAT "k",
2146             ((jlong)si.totalswap * si.mem_unit) >> 10);
2147   st->print("(" UINT64_FORMAT "k free)",
2148             ((jlong)si.freeswap * si.mem_unit) >> 10);
2149   st->cr();
2150 }
2151 
2152 // Print the first "model name" line and the first "flags" line
2153 // that we find and nothing more. We assume "model name" comes
2154 // before "flags" so if we find a second "model name", then the
2155 // "flags" field is considered missing.
2156 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2157 #if defined(IA32) || defined(AMD64)
2158   // Other platforms have less repetitive cpuinfo files
2159   FILE *fp = fopen("/proc/cpuinfo", "r");
2160   if (fp) {
2161     while (!feof(fp)) {
2162       if (fgets(buf, buflen, fp)) {
2163         // Assume model name comes before flags
2164         bool model_name_printed = false;
2165         if (strstr(buf, "model name") != NULL) {
2166           if (!model_name_printed) {
2167             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2168             st->print_raw(buf);
2169             model_name_printed = true;
2170           } else {
2171             // model name printed but not flags?  Odd, just return
2172             fclose(fp);
2173             return true;
2174           }
2175         }
2176         // print the flags line too
2177         if (strstr(buf, "flags") != NULL) {
2178           st->print_raw(buf);
2179           fclose(fp);
2180           return true;
2181         }
2182       }
2183     }
2184     fclose(fp);
2185   }
2186 #endif // x86 platforms
2187   return false;
2188 }
2189 
2190 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2191   // Only print the model name if the platform provides this as a summary
2192   if (!print_model_name_and_flags(st, buf, buflen)) {
2193     st->print("\n/proc/cpuinfo:\n");
2194     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2195       st->print_cr("  <Not Available>");
2196     }
2197   }
2198 }
2199 
2200 #if defined(AMD64) || defined(IA32) || defined(X32)
2201 const char* search_string = "model name";
2202 #elif defined(PPC64)
2203 const char* search_string = "cpu";
2204 #elif defined(S390)
2205 const char* search_string = "processor";
2206 #elif defined(SPARC)
2207 const char* search_string = "cpu";
2208 #else
2209 const char* search_string = "Processor";
2210 #endif
2211 
2212 // Parses the cpuinfo file for string representing the model name.
2213 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2214   FILE* fp = fopen("/proc/cpuinfo", "r");
2215   if (fp != NULL) {
2216     while (!feof(fp)) {
2217       char buf[256];
2218       if (fgets(buf, sizeof(buf), fp)) {
2219         char* start = strstr(buf, search_string);
2220         if (start != NULL) {
2221           char *ptr = start + strlen(search_string);
2222           char *end = buf + strlen(buf);
2223           while (ptr != end) {
2224              // skip whitespace and colon for the rest of the name.
2225              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2226                break;
2227              }
2228              ptr++;
2229           }
2230           if (ptr != end) {
2231             // reasonable string, get rid of newline and keep the rest
2232             char* nl = strchr(buf, '\n');
2233             if (nl != NULL) *nl = '\0';
2234             strncpy(cpuinfo, ptr, length);
2235             fclose(fp);
2236             return;
2237           }
2238         }
2239       }
2240     }
2241     fclose(fp);
2242   }
2243   // cpuinfo not found or parsing failed, just print generic string.  The entire
2244   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2245 #if   defined(AARCH64)
2246   strncpy(cpuinfo, "AArch64", length);
2247 #elif defined(AMD64)
2248   strncpy(cpuinfo, "x86_64", length);
2249 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2250   strncpy(cpuinfo, "ARM", length);
2251 #elif defined(IA32)
2252   strncpy(cpuinfo, "x86_32", length);
2253 #elif defined(IA64)
2254   strncpy(cpuinfo, "IA64", length);
2255 #elif defined(PPC)
2256   strncpy(cpuinfo, "PPC64", length);
2257 #elif defined(S390)
2258   strncpy(cpuinfo, "S390", length);
2259 #elif defined(SPARC)
2260   strncpy(cpuinfo, "sparcv9", length);
2261 #elif defined(ZERO_LIBARCH)
2262   strncpy(cpuinfo, ZERO_LIBARCH, length);
2263 #else
2264   strncpy(cpuinfo, "unknown", length);
2265 #endif
2266 }
2267 
2268 static void print_signal_handler(outputStream* st, int sig,
2269                                  char* buf, size_t buflen);
2270 
2271 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2272   st->print_cr("Signal Handlers:");
2273   print_signal_handler(st, SIGSEGV, buf, buflen);
2274   print_signal_handler(st, SIGBUS , buf, buflen);
2275   print_signal_handler(st, SIGFPE , buf, buflen);
2276   print_signal_handler(st, SIGPIPE, buf, buflen);
2277   print_signal_handler(st, SIGXFSZ, buf, buflen);
2278   print_signal_handler(st, SIGILL , buf, buflen);
2279   print_signal_handler(st, SR_signum, buf, buflen);
2280   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2281   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2282   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2283   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2284 #if defined(PPC64)
2285   print_signal_handler(st, SIGTRAP, buf, buflen);
2286 #endif
2287 }
2288 
2289 static char saved_jvm_path[MAXPATHLEN] = {0};
2290 
2291 // Find the full path to the current module, libjvm.so
2292 void os::jvm_path(char *buf, jint buflen) {
2293   // Error checking.
2294   if (buflen < MAXPATHLEN) {
2295     assert(false, "must use a large-enough buffer");
2296     buf[0] = '\0';
2297     return;
2298   }
2299   // Lazy resolve the path to current module.
2300   if (saved_jvm_path[0] != 0) {
2301     strcpy(buf, saved_jvm_path);
2302     return;
2303   }
2304 
2305   char dli_fname[MAXPATHLEN];
2306   bool ret = dll_address_to_library_name(
2307                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2308                                          dli_fname, sizeof(dli_fname), NULL);
2309   assert(ret, "cannot locate libjvm");
2310   char *rp = NULL;
2311   if (ret && dli_fname[0] != '\0') {
2312     rp = realpath(dli_fname, buf);
2313   }
2314   if (rp == NULL) {
2315     return;
2316   }
2317 
2318   if (Arguments::sun_java_launcher_is_altjvm()) {
2319     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2320     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2321     // If "/jre/lib/" appears at the right place in the string, then
2322     // assume we are installed in a JDK and we're done. Otherwise, check
2323     // for a JAVA_HOME environment variable and fix up the path so it
2324     // looks like libjvm.so is installed there (append a fake suffix
2325     // hotspot/libjvm.so).
2326     const char *p = buf + strlen(buf) - 1;
2327     for (int count = 0; p > buf && count < 5; ++count) {
2328       for (--p; p > buf && *p != '/'; --p)
2329         /* empty */ ;
2330     }
2331 
2332     if (strncmp(p, "/jre/lib/", 9) != 0) {
2333       // Look for JAVA_HOME in the environment.
2334       char* java_home_var = ::getenv("JAVA_HOME");
2335       if (java_home_var != NULL && java_home_var[0] != 0) {
2336         char* jrelib_p;
2337         int len;
2338 
2339         // Check the current module name "libjvm.so".
2340         p = strrchr(buf, '/');
2341         if (p == NULL) {
2342           return;
2343         }
2344         assert(strstr(p, "/libjvm") == p, "invalid library name");
2345 
2346         rp = realpath(java_home_var, buf);
2347         if (rp == NULL) {
2348           return;
2349         }
2350 
2351         // determine if this is a legacy image or modules image
2352         // modules image doesn't have "jre" subdirectory
2353         len = strlen(buf);
2354         assert(len < buflen, "Ran out of buffer room");
2355         jrelib_p = buf + len;
2356         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2357         if (0 != access(buf, F_OK)) {
2358           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2359         }
2360 
2361         if (0 == access(buf, F_OK)) {
2362           // Use current module name "libjvm.so"
2363           len = strlen(buf);
2364           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2365         } else {
2366           // Go back to path of .so
2367           rp = realpath(dli_fname, buf);
2368           if (rp == NULL) {
2369             return;
2370           }
2371         }
2372       }
2373     }
2374   }
2375 
2376   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2377   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2378 }
2379 
2380 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2381   // no prefix required, not even "_"
2382 }
2383 
2384 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2385   // no suffix required
2386 }
2387 
2388 ////////////////////////////////////////////////////////////////////////////////
2389 // sun.misc.Signal support
2390 
2391 static volatile jint sigint_count = 0;
2392 
2393 static void UserHandler(int sig, void *siginfo, void *context) {
2394   // 4511530 - sem_post is serialized and handled by the manager thread. When
2395   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2396   // don't want to flood the manager thread with sem_post requests.
2397   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2398     return;
2399   }
2400 
2401   // Ctrl-C is pressed during error reporting, likely because the error
2402   // handler fails to abort. Let VM die immediately.
2403   if (sig == SIGINT && is_error_reported()) {
2404     os::die();
2405   }
2406 
2407   os::signal_notify(sig);
2408 }
2409 
2410 void* os::user_handler() {
2411   return CAST_FROM_FN_PTR(void*, UserHandler);
2412 }
2413 
2414 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2415   struct timespec ts;
2416   // Semaphore's are always associated with CLOCK_REALTIME
2417   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2418   // see unpackTime for discussion on overflow checking
2419   if (sec >= MAX_SECS) {
2420     ts.tv_sec += MAX_SECS;
2421     ts.tv_nsec = 0;
2422   } else {
2423     ts.tv_sec += sec;
2424     ts.tv_nsec += nsec;
2425     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2426       ts.tv_nsec -= NANOSECS_PER_SEC;
2427       ++ts.tv_sec; // note: this must be <= max_secs
2428     }
2429   }
2430 
2431   return ts;
2432 }
2433 
2434 extern "C" {
2435   typedef void (*sa_handler_t)(int);
2436   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2437 }
2438 
2439 void* os::signal(int signal_number, void* handler) {
2440   struct sigaction sigAct, oldSigAct;
2441 
2442   sigfillset(&(sigAct.sa_mask));
2443   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2444   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2445 
2446   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2447     // -1 means registration failed
2448     return (void *)-1;
2449   }
2450 
2451   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2452 }
2453 
2454 void os::signal_raise(int signal_number) {
2455   ::raise(signal_number);
2456 }
2457 
2458 // The following code is moved from os.cpp for making this
2459 // code platform specific, which it is by its very nature.
2460 
2461 // Will be modified when max signal is changed to be dynamic
2462 int os::sigexitnum_pd() {
2463   return NSIG;
2464 }
2465 
2466 // a counter for each possible signal value
2467 static volatile jint pending_signals[NSIG+1] = { 0 };
2468 
2469 // Linux(POSIX) specific hand shaking semaphore.
2470 static sem_t sig_sem;
2471 static PosixSemaphore sr_semaphore;
2472 
2473 void os::signal_init_pd() {
2474   // Initialize signal structures
2475   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2476 
2477   // Initialize signal semaphore
2478   ::sem_init(&sig_sem, 0, 0);
2479 }
2480 
2481 void os::signal_notify(int sig) {
2482   Atomic::inc(&pending_signals[sig]);
2483   ::sem_post(&sig_sem);
2484 }
2485 
2486 static int check_pending_signals(bool wait) {
2487   Atomic::store(0, &sigint_count);
2488   for (;;) {
2489     for (int i = 0; i < NSIG + 1; i++) {
2490       jint n = pending_signals[i];
2491       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2492         return i;
2493       }
2494     }
2495     if (!wait) {
2496       return -1;
2497     }
2498     JavaThread *thread = JavaThread::current();
2499     ThreadBlockInVM tbivm(thread);
2500 
2501     bool threadIsSuspended;
2502     do {
2503       thread->set_suspend_equivalent();
2504       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2505       ::sem_wait(&sig_sem);
2506 
2507       // were we externally suspended while we were waiting?
2508       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2509       if (threadIsSuspended) {
2510         // The semaphore has been incremented, but while we were waiting
2511         // another thread suspended us. We don't want to continue running
2512         // while suspended because that would surprise the thread that
2513         // suspended us.
2514         ::sem_post(&sig_sem);
2515 
2516         thread->java_suspend_self();
2517       }
2518     } while (threadIsSuspended);
2519   }
2520 }
2521 
2522 int os::signal_lookup() {
2523   return check_pending_signals(false);
2524 }
2525 
2526 int os::signal_wait() {
2527   return check_pending_signals(true);
2528 }
2529 
2530 ////////////////////////////////////////////////////////////////////////////////
2531 // Virtual Memory
2532 
2533 int os::vm_page_size() {
2534   // Seems redundant as all get out
2535   assert(os::Linux::page_size() != -1, "must call os::init");
2536   return os::Linux::page_size();
2537 }
2538 
2539 // Solaris allocates memory by pages.
2540 int os::vm_allocation_granularity() {
2541   assert(os::Linux::page_size() != -1, "must call os::init");
2542   return os::Linux::page_size();
2543 }
2544 
2545 // Rationale behind this function:
2546 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2547 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2548 //  samples for JITted code. Here we create private executable mapping over the code cache
2549 //  and then we can use standard (well, almost, as mapping can change) way to provide
2550 //  info for the reporting script by storing timestamp and location of symbol
2551 void linux_wrap_code(char* base, size_t size) {
2552   static volatile jint cnt = 0;
2553 
2554   if (!UseOprofile) {
2555     return;
2556   }
2557 
2558   char buf[PATH_MAX+1];
2559   int num = Atomic::add(1, &cnt);
2560 
2561   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2562            os::get_temp_directory(), os::current_process_id(), num);
2563   unlink(buf);
2564 
2565   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2566 
2567   if (fd != -1) {
2568     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2569     if (rv != (off_t)-1) {
2570       if (::write(fd, "", 1) == 1) {
2571         mmap(base, size,
2572              PROT_READ|PROT_WRITE|PROT_EXEC,
2573              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2574       }
2575     }
2576     ::close(fd);
2577     unlink(buf);
2578   }
2579 }
2580 
2581 static bool recoverable_mmap_error(int err) {
2582   // See if the error is one we can let the caller handle. This
2583   // list of errno values comes from JBS-6843484. I can't find a
2584   // Linux man page that documents this specific set of errno
2585   // values so while this list currently matches Solaris, it may
2586   // change as we gain experience with this failure mode.
2587   switch (err) {
2588   case EBADF:
2589   case EINVAL:
2590   case ENOTSUP:
2591     // let the caller deal with these errors
2592     return true;
2593 
2594   default:
2595     // Any remaining errors on this OS can cause our reserved mapping
2596     // to be lost. That can cause confusion where different data
2597     // structures think they have the same memory mapped. The worst
2598     // scenario is if both the VM and a library think they have the
2599     // same memory mapped.
2600     return false;
2601   }
2602 }
2603 
2604 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2605                                     int err) {
2606   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2607           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2608           os::strerror(err), err);
2609 }
2610 
2611 static void warn_fail_commit_memory(char* addr, size_t size,
2612                                     size_t alignment_hint, bool exec,
2613                                     int err) {
2614   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2615           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2616           alignment_hint, exec, os::strerror(err), err);
2617 }
2618 
2619 // NOTE: Linux kernel does not really reserve the pages for us.
2620 //       All it does is to check if there are enough free pages
2621 //       left at the time of mmap(). This could be a potential
2622 //       problem.
2623 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2624   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2625   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2626                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2627   if (res != (uintptr_t) MAP_FAILED) {
2628     if (UseNUMAInterleaving) {
2629       numa_make_global(addr, size);
2630     }
2631     return 0;
2632   }
2633 
2634   int err = errno;  // save errno from mmap() call above
2635 
2636   if (!recoverable_mmap_error(err)) {
2637     warn_fail_commit_memory(addr, size, exec, err);
2638     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2639   }
2640 
2641   return err;
2642 }
2643 
2644 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2645   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2646 }
2647 
2648 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2649                                   const char* mesg) {
2650   assert(mesg != NULL, "mesg must be specified");
2651   int err = os::Linux::commit_memory_impl(addr, size, exec);
2652   if (err != 0) {
2653     // the caller wants all commit errors to exit with the specified mesg:
2654     warn_fail_commit_memory(addr, size, exec, err);
2655     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2656   }
2657 }
2658 
2659 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2660 #ifndef MAP_HUGETLB
2661   #define MAP_HUGETLB 0x40000
2662 #endif
2663 
2664 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2665 #ifndef MADV_HUGEPAGE
2666   #define MADV_HUGEPAGE 14
2667 #endif
2668 
2669 int os::Linux::commit_memory_impl(char* addr, size_t size,
2670                                   size_t alignment_hint, bool exec) {
2671   int err = os::Linux::commit_memory_impl(addr, size, exec);
2672   if (err == 0) {
2673     realign_memory(addr, size, alignment_hint);
2674   }
2675   return err;
2676 }
2677 
2678 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2679                           bool exec) {
2680   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2681 }
2682 
2683 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2684                                   size_t alignment_hint, bool exec,
2685                                   const char* mesg) {
2686   assert(mesg != NULL, "mesg must be specified");
2687   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2688   if (err != 0) {
2689     // the caller wants all commit errors to exit with the specified mesg:
2690     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2691     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2692   }
2693 }
2694 
2695 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2696   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2697     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2698     // be supported or the memory may already be backed by huge pages.
2699     ::madvise(addr, bytes, MADV_HUGEPAGE);
2700   }
2701 }
2702 
2703 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2704   // This method works by doing an mmap over an existing mmaping and effectively discarding
2705   // the existing pages. However it won't work for SHM-based large pages that cannot be
2706   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2707   // small pages on top of the SHM segment. This method always works for small pages, so we
2708   // allow that in any case.
2709   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2710     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2711   }
2712 }
2713 
2714 void os::numa_make_global(char *addr, size_t bytes) {
2715   Linux::numa_interleave_memory(addr, bytes);
2716 }
2717 
2718 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2719 // bind policy to MPOL_PREFERRED for the current thread.
2720 #define USE_MPOL_PREFERRED 0
2721 
2722 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2723   // To make NUMA and large pages more robust when both enabled, we need to ease
2724   // the requirements on where the memory should be allocated. MPOL_BIND is the
2725   // default policy and it will force memory to be allocated on the specified
2726   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2727   // the specified node, but will not force it. Using this policy will prevent
2728   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2729   // free large pages.
2730   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2731   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2732 }
2733 
2734 bool os::numa_topology_changed() { return false; }
2735 
2736 size_t os::numa_get_groups_num() {
2737   int max_node = Linux::numa_max_node();
2738   return max_node > 0 ? max_node + 1 : 1;
2739 }
2740 
2741 int os::numa_get_group_id() {
2742   int cpu_id = Linux::sched_getcpu();
2743   if (cpu_id != -1) {
2744     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2745     if (lgrp_id != -1) {
2746       return lgrp_id;
2747     }
2748   }
2749   return 0;
2750 }
2751 
2752 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2753   for (size_t i = 0; i < size; i++) {
2754     ids[i] = i;
2755   }
2756   return size;
2757 }
2758 
2759 bool os::get_page_info(char *start, page_info* info) {
2760   return false;
2761 }
2762 
2763 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2764                      page_info* page_found) {
2765   return end;
2766 }
2767 
2768 
2769 int os::Linux::sched_getcpu_syscall(void) {
2770   unsigned int cpu = 0;
2771   int retval = -1;
2772 
2773 #if defined(IA32)
2774   #ifndef SYS_getcpu
2775     #define SYS_getcpu 318
2776   #endif
2777   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2778 #elif defined(AMD64)
2779 // Unfortunately we have to bring all these macros here from vsyscall.h
2780 // to be able to compile on old linuxes.
2781   #define __NR_vgetcpu 2
2782   #define VSYSCALL_START (-10UL << 20)
2783   #define VSYSCALL_SIZE 1024
2784   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2785   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2786   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2787   retval = vgetcpu(&cpu, NULL, NULL);
2788 #endif
2789 
2790   return (retval == -1) ? retval : cpu;
2791 }
2792 
2793 // Something to do with the numa-aware allocator needs these symbols
2794 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2795 extern "C" JNIEXPORT void numa_error(char *where) { }
2796 
2797 
2798 // If we are running with libnuma version > 2, then we should
2799 // be trying to use symbols with versions 1.1
2800 // If we are running with earlier version, which did not have symbol versions,
2801 // we should use the base version.
2802 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2803   void *f = dlvsym(handle, name, "libnuma_1.1");
2804   if (f == NULL) {
2805     f = dlsym(handle, name);
2806   }
2807   return f;
2808 }
2809 
2810 bool os::Linux::libnuma_init() {
2811   // sched_getcpu() should be in libc.
2812   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2813                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2814 
2815   // If it's not, try a direct syscall.
2816   if (sched_getcpu() == -1) {
2817     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2818                                     (void*)&sched_getcpu_syscall));
2819   }
2820 
2821   if (sched_getcpu() != -1) { // Does it work?
2822     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2823     if (handle != NULL) {
2824       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2825                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2826       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2827                                        libnuma_dlsym(handle, "numa_max_node")));
2828       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2829                                         libnuma_dlsym(handle, "numa_available")));
2830       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2831                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2832       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2833                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2834       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2835                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2836 
2837 
2838       if (numa_available() != -1) {
2839         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2840         // Create a cpu -> node mapping
2841         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2842         rebuild_cpu_to_node_map();
2843         return true;
2844       }
2845     }
2846   }
2847   return false;
2848 }
2849 
2850 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2851 // The table is later used in get_node_by_cpu().
2852 void os::Linux::rebuild_cpu_to_node_map() {
2853   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2854                               // in libnuma (possible values are starting from 16,
2855                               // and continuing up with every other power of 2, but less
2856                               // than the maximum number of CPUs supported by kernel), and
2857                               // is a subject to change (in libnuma version 2 the requirements
2858                               // are more reasonable) we'll just hardcode the number they use
2859                               // in the library.
2860   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2861 
2862   size_t cpu_num = processor_count();
2863   size_t cpu_map_size = NCPUS / BitsPerCLong;
2864   size_t cpu_map_valid_size =
2865     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2866 
2867   cpu_to_node()->clear();
2868   cpu_to_node()->at_grow(cpu_num - 1);
2869   size_t node_num = numa_get_groups_num();
2870 
2871   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2872   for (size_t i = 0; i < node_num; i++) {
2873     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2874       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2875         if (cpu_map[j] != 0) {
2876           for (size_t k = 0; k < BitsPerCLong; k++) {
2877             if (cpu_map[j] & (1UL << k)) {
2878               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2879             }
2880           }
2881         }
2882       }
2883     }
2884   }
2885   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2886 }
2887 
2888 int os::Linux::get_node_by_cpu(int cpu_id) {
2889   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2890     return cpu_to_node()->at(cpu_id);
2891   }
2892   return -1;
2893 }
2894 
2895 GrowableArray<int>* os::Linux::_cpu_to_node;
2896 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2897 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2898 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2899 os::Linux::numa_available_func_t os::Linux::_numa_available;
2900 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2901 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2902 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2903 unsigned long* os::Linux::_numa_all_nodes;
2904 
2905 bool os::pd_uncommit_memory(char* addr, size_t size) {
2906   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2907                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2908   return res  != (uintptr_t) MAP_FAILED;
2909 }
2910 
2911 static address get_stack_commited_bottom(address bottom, size_t size) {
2912   address nbot = bottom;
2913   address ntop = bottom + size;
2914 
2915   size_t page_sz = os::vm_page_size();
2916   unsigned pages = size / page_sz;
2917 
2918   unsigned char vec[1];
2919   unsigned imin = 1, imax = pages + 1, imid;
2920   int mincore_return_value = 0;
2921 
2922   assert(imin <= imax, "Unexpected page size");
2923 
2924   while (imin < imax) {
2925     imid = (imax + imin) / 2;
2926     nbot = ntop - (imid * page_sz);
2927 
2928     // Use a trick with mincore to check whether the page is mapped or not.
2929     // mincore sets vec to 1 if page resides in memory and to 0 if page
2930     // is swapped output but if page we are asking for is unmapped
2931     // it returns -1,ENOMEM
2932     mincore_return_value = mincore(nbot, page_sz, vec);
2933 
2934     if (mincore_return_value == -1) {
2935       // Page is not mapped go up
2936       // to find first mapped page
2937       if (errno != EAGAIN) {
2938         assert(errno == ENOMEM, "Unexpected mincore errno");
2939         imax = imid;
2940       }
2941     } else {
2942       // Page is mapped go down
2943       // to find first not mapped page
2944       imin = imid + 1;
2945     }
2946   }
2947 
2948   nbot = nbot + page_sz;
2949 
2950   // Adjust stack bottom one page up if last checked page is not mapped
2951   if (mincore_return_value == -1) {
2952     nbot = nbot + page_sz;
2953   }
2954 
2955   return nbot;
2956 }
2957 
2958 
2959 // Linux uses a growable mapping for the stack, and if the mapping for
2960 // the stack guard pages is not removed when we detach a thread the
2961 // stack cannot grow beyond the pages where the stack guard was
2962 // mapped.  If at some point later in the process the stack expands to
2963 // that point, the Linux kernel cannot expand the stack any further
2964 // because the guard pages are in the way, and a segfault occurs.
2965 //
2966 // However, it's essential not to split the stack region by unmapping
2967 // a region (leaving a hole) that's already part of the stack mapping,
2968 // so if the stack mapping has already grown beyond the guard pages at
2969 // the time we create them, we have to truncate the stack mapping.
2970 // So, we need to know the extent of the stack mapping when
2971 // create_stack_guard_pages() is called.
2972 
2973 // We only need this for stacks that are growable: at the time of
2974 // writing thread stacks don't use growable mappings (i.e. those
2975 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2976 // only applies to the main thread.
2977 
2978 // If the (growable) stack mapping already extends beyond the point
2979 // where we're going to put our guard pages, truncate the mapping at
2980 // that point by munmap()ping it.  This ensures that when we later
2981 // munmap() the guard pages we don't leave a hole in the stack
2982 // mapping. This only affects the main/initial thread
2983 
2984 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2985   if (os::Linux::is_initial_thread()) {
2986     // As we manually grow stack up to bottom inside create_attached_thread(),
2987     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2988     // we don't need to do anything special.
2989     // Check it first, before calling heavy function.
2990     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2991     unsigned char vec[1];
2992 
2993     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2994       // Fallback to slow path on all errors, including EAGAIN
2995       stack_extent = (uintptr_t) get_stack_commited_bottom(
2996                                                            os::Linux::initial_thread_stack_bottom(),
2997                                                            (size_t)addr - stack_extent);
2998     }
2999 
3000     if (stack_extent < (uintptr_t)addr) {
3001       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3002     }
3003   }
3004 
3005   return os::commit_memory(addr, size, !ExecMem);
3006 }
3007 
3008 // If this is a growable mapping, remove the guard pages entirely by
3009 // munmap()ping them.  If not, just call uncommit_memory(). This only
3010 // affects the main/initial thread, but guard against future OS changes
3011 // It's safe to always unmap guard pages for initial thread because we
3012 // always place it right after end of the mapped region
3013 
3014 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3015   uintptr_t stack_extent, stack_base;
3016 
3017   if (os::Linux::is_initial_thread()) {
3018     return ::munmap(addr, size) == 0;
3019   }
3020 
3021   return os::uncommit_memory(addr, size);
3022 }
3023 
3024 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3025 // at 'requested_addr'. If there are existing memory mappings at the same
3026 // location, however, they will be overwritten. If 'fixed' is false,
3027 // 'requested_addr' is only treated as a hint, the return value may or
3028 // may not start from the requested address. Unlike Linux mmap(), this
3029 // function returns NULL to indicate failure.
3030 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3031   char * addr;
3032   int flags;
3033 
3034   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3035   if (fixed) {
3036     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3037     flags |= MAP_FIXED;
3038   }
3039 
3040   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3041   // touch an uncommitted page. Otherwise, the read/write might
3042   // succeed if we have enough swap space to back the physical page.
3043   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3044                        flags, -1, 0);
3045 
3046   return addr == MAP_FAILED ? NULL : addr;
3047 }
3048 
3049 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3050 //   (req_addr != NULL) or with a given alignment.
3051 //  - bytes shall be a multiple of alignment.
3052 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3053 //  - alignment sets the alignment at which memory shall be allocated.
3054 //     It must be a multiple of allocation granularity.
3055 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3056 //  req_addr or NULL.
3057 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3058 
3059   size_t extra_size = bytes;
3060   if (req_addr == NULL && alignment > 0) {
3061     extra_size += alignment;
3062   }
3063 
3064   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3065     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3066     -1, 0);
3067   if (start == MAP_FAILED) {
3068     start = NULL;
3069   } else {
3070     if (req_addr != NULL) {
3071       if (start != req_addr) {
3072         ::munmap(start, extra_size);
3073         start = NULL;
3074       }
3075     } else {
3076       char* const start_aligned = (char*) align_ptr_up(start, alignment);
3077       char* const end_aligned = start_aligned + bytes;
3078       char* const end = start + extra_size;
3079       if (start_aligned > start) {
3080         ::munmap(start, start_aligned - start);
3081       }
3082       if (end_aligned < end) {
3083         ::munmap(end_aligned, end - end_aligned);
3084       }
3085       start = start_aligned;
3086     }
3087   }
3088   return start;
3089 }
3090 
3091 static int anon_munmap(char * addr, size_t size) {
3092   return ::munmap(addr, size) == 0;
3093 }
3094 
3095 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3096                             size_t alignment_hint) {
3097   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3098 }
3099 
3100 bool os::pd_release_memory(char* addr, size_t size) {
3101   return anon_munmap(addr, size);
3102 }
3103 
3104 static bool linux_mprotect(char* addr, size_t size, int prot) {
3105   // Linux wants the mprotect address argument to be page aligned.
3106   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3107 
3108   // According to SUSv3, mprotect() should only be used with mappings
3109   // established by mmap(), and mmap() always maps whole pages. Unaligned
3110   // 'addr' likely indicates problem in the VM (e.g. trying to change
3111   // protection of malloc'ed or statically allocated memory). Check the
3112   // caller if you hit this assert.
3113   assert(addr == bottom, "sanity check");
3114 
3115   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3116   return ::mprotect(bottom, size, prot) == 0;
3117 }
3118 
3119 // Set protections specified
3120 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3121                         bool is_committed) {
3122   unsigned int p = 0;
3123   switch (prot) {
3124   case MEM_PROT_NONE: p = PROT_NONE; break;
3125   case MEM_PROT_READ: p = PROT_READ; break;
3126   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3127   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3128   default:
3129     ShouldNotReachHere();
3130   }
3131   // is_committed is unused.
3132   return linux_mprotect(addr, bytes, p);
3133 }
3134 
3135 bool os::guard_memory(char* addr, size_t size) {
3136   return linux_mprotect(addr, size, PROT_NONE);
3137 }
3138 
3139 bool os::unguard_memory(char* addr, size_t size) {
3140   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3141 }
3142 
3143 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3144                                                     size_t page_size) {
3145   bool result = false;
3146   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3147                  MAP_ANONYMOUS|MAP_PRIVATE,
3148                  -1, 0);
3149   if (p != MAP_FAILED) {
3150     void *aligned_p = align_ptr_up(p, page_size);
3151 
3152     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3153 
3154     munmap(p, page_size * 2);
3155   }
3156 
3157   if (warn && !result) {
3158     warning("TransparentHugePages is not supported by the operating system.");
3159   }
3160 
3161   return result;
3162 }
3163 
3164 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3165   bool result = false;
3166   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3167                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3168                  -1, 0);
3169 
3170   if (p != MAP_FAILED) {
3171     // We don't know if this really is a huge page or not.
3172     FILE *fp = fopen("/proc/self/maps", "r");
3173     if (fp) {
3174       while (!feof(fp)) {
3175         char chars[257];
3176         long x = 0;
3177         if (fgets(chars, sizeof(chars), fp)) {
3178           if (sscanf(chars, "%lx-%*x", &x) == 1
3179               && x == (long)p) {
3180             if (strstr (chars, "hugepage")) {
3181               result = true;
3182               break;
3183             }
3184           }
3185         }
3186       }
3187       fclose(fp);
3188     }
3189     munmap(p, page_size);
3190   }
3191 
3192   if (warn && !result) {
3193     warning("HugeTLBFS is not supported by the operating system.");
3194   }
3195 
3196   return result;
3197 }
3198 
3199 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3200 //
3201 // From the coredump_filter documentation:
3202 //
3203 // - (bit 0) anonymous private memory
3204 // - (bit 1) anonymous shared memory
3205 // - (bit 2) file-backed private memory
3206 // - (bit 3) file-backed shared memory
3207 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3208 //           effective only if the bit 2 is cleared)
3209 // - (bit 5) hugetlb private memory
3210 // - (bit 6) hugetlb shared memory
3211 //
3212 static void set_coredump_filter(void) {
3213   FILE *f;
3214   long cdm;
3215 
3216   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3217     return;
3218   }
3219 
3220   if (fscanf(f, "%lx", &cdm) != 1) {
3221     fclose(f);
3222     return;
3223   }
3224 
3225   rewind(f);
3226 
3227   if ((cdm & LARGEPAGES_BIT) == 0) {
3228     cdm |= LARGEPAGES_BIT;
3229     fprintf(f, "%#lx", cdm);
3230   }
3231 
3232   fclose(f);
3233 }
3234 
3235 // Large page support
3236 
3237 static size_t _large_page_size = 0;
3238 
3239 size_t os::Linux::find_large_page_size() {
3240   size_t large_page_size = 0;
3241 
3242   // large_page_size on Linux is used to round up heap size. x86 uses either
3243   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3244   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3245   // page as large as 256M.
3246   //
3247   // Here we try to figure out page size by parsing /proc/meminfo and looking
3248   // for a line with the following format:
3249   //    Hugepagesize:     2048 kB
3250   //
3251   // If we can't determine the value (e.g. /proc is not mounted, or the text
3252   // format has been changed), we'll use the largest page size supported by
3253   // the processor.
3254 
3255 #ifndef ZERO
3256   large_page_size =
3257     AARCH64_ONLY(2 * M)
3258     AMD64_ONLY(2 * M)
3259     ARM32_ONLY(2 * M)
3260     IA32_ONLY(4 * M)
3261     IA64_ONLY(256 * M)
3262     PPC_ONLY(4 * M)
3263     S390_ONLY(1 * M)
3264     SPARC_ONLY(4 * M);
3265 #endif // ZERO
3266 
3267   FILE *fp = fopen("/proc/meminfo", "r");
3268   if (fp) {
3269     while (!feof(fp)) {
3270       int x = 0;
3271       char buf[16];
3272       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3273         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3274           large_page_size = x * K;
3275           break;
3276         }
3277       } else {
3278         // skip to next line
3279         for (;;) {
3280           int ch = fgetc(fp);
3281           if (ch == EOF || ch == (int)'\n') break;
3282         }
3283       }
3284     }
3285     fclose(fp);
3286   }
3287 
3288   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3289     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3290             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3291             proper_unit_for_byte_size(large_page_size));
3292   }
3293 
3294   return large_page_size;
3295 }
3296 
3297 size_t os::Linux::setup_large_page_size() {
3298   _large_page_size = Linux::find_large_page_size();
3299   const size_t default_page_size = (size_t)Linux::page_size();
3300   if (_large_page_size > default_page_size) {
3301     _page_sizes[0] = _large_page_size;
3302     _page_sizes[1] = default_page_size;
3303     _page_sizes[2] = 0;
3304   }
3305 
3306   return _large_page_size;
3307 }
3308 
3309 bool os::Linux::setup_large_page_type(size_t page_size) {
3310   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3311       FLAG_IS_DEFAULT(UseSHM) &&
3312       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3313 
3314     // The type of large pages has not been specified by the user.
3315 
3316     // Try UseHugeTLBFS and then UseSHM.
3317     UseHugeTLBFS = UseSHM = true;
3318 
3319     // Don't try UseTransparentHugePages since there are known
3320     // performance issues with it turned on. This might change in the future.
3321     UseTransparentHugePages = false;
3322   }
3323 
3324   if (UseTransparentHugePages) {
3325     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3326     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3327       UseHugeTLBFS = false;
3328       UseSHM = false;
3329       return true;
3330     }
3331     UseTransparentHugePages = false;
3332   }
3333 
3334   if (UseHugeTLBFS) {
3335     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3336     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3337       UseSHM = false;
3338       return true;
3339     }
3340     UseHugeTLBFS = false;
3341   }
3342 
3343   return UseSHM;
3344 }
3345 
3346 void os::large_page_init() {
3347   if (!UseLargePages &&
3348       !UseTransparentHugePages &&
3349       !UseHugeTLBFS &&
3350       !UseSHM) {
3351     // Not using large pages.
3352     return;
3353   }
3354 
3355   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3356     // The user explicitly turned off large pages.
3357     // Ignore the rest of the large pages flags.
3358     UseTransparentHugePages = false;
3359     UseHugeTLBFS = false;
3360     UseSHM = false;
3361     return;
3362   }
3363 
3364   size_t large_page_size = Linux::setup_large_page_size();
3365   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3366 
3367   set_coredump_filter();
3368 }
3369 
3370 #ifndef SHM_HUGETLB
3371   #define SHM_HUGETLB 04000
3372 #endif
3373 
3374 #define shm_warning_format(format, ...)              \
3375   do {                                               \
3376     if (UseLargePages &&                             \
3377         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3378          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3379          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3380       warning(format, __VA_ARGS__);                  \
3381     }                                                \
3382   } while (0)
3383 
3384 #define shm_warning(str) shm_warning_format("%s", str)
3385 
3386 #define shm_warning_with_errno(str)                \
3387   do {                                             \
3388     int err = errno;                               \
3389     shm_warning_format(str " (error = %d)", err);  \
3390   } while (0)
3391 
3392 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3393   assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3394 
3395   if (!is_size_aligned(alignment, SHMLBA)) {
3396     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3397     return NULL;
3398   }
3399 
3400   // To ensure that we get 'alignment' aligned memory from shmat,
3401   // we pre-reserve aligned virtual memory and then attach to that.
3402 
3403   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3404   if (pre_reserved_addr == NULL) {
3405     // Couldn't pre-reserve aligned memory.
3406     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3407     return NULL;
3408   }
3409 
3410   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3411   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3412 
3413   if ((intptr_t)addr == -1) {
3414     int err = errno;
3415     shm_warning_with_errno("Failed to attach shared memory.");
3416 
3417     assert(err != EACCES, "Unexpected error");
3418     assert(err != EIDRM,  "Unexpected error");
3419     assert(err != EINVAL, "Unexpected error");
3420 
3421     // Since we don't know if the kernel unmapped the pre-reserved memory area
3422     // we can't unmap it, since that would potentially unmap memory that was
3423     // mapped from other threads.
3424     return NULL;
3425   }
3426 
3427   return addr;
3428 }
3429 
3430 static char* shmat_at_address(int shmid, char* req_addr) {
3431   if (!is_ptr_aligned(req_addr, SHMLBA)) {
3432     assert(false, "Requested address needs to be SHMLBA aligned");
3433     return NULL;
3434   }
3435 
3436   char* addr = (char*)shmat(shmid, req_addr, 0);
3437 
3438   if ((intptr_t)addr == -1) {
3439     shm_warning_with_errno("Failed to attach shared memory.");
3440     return NULL;
3441   }
3442 
3443   return addr;
3444 }
3445 
3446 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3447   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3448   if (req_addr != NULL) {
3449     assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3450     assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3451     return shmat_at_address(shmid, req_addr);
3452   }
3453 
3454   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3455   // return large page size aligned memory addresses when req_addr == NULL.
3456   // However, if the alignment is larger than the large page size, we have
3457   // to manually ensure that the memory returned is 'alignment' aligned.
3458   if (alignment > os::large_page_size()) {
3459     assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3460     return shmat_with_alignment(shmid, bytes, alignment);
3461   } else {
3462     return shmat_at_address(shmid, NULL);
3463   }
3464 }
3465 
3466 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3467                                             char* req_addr, bool exec) {
3468   // "exec" is passed in but not used.  Creating the shared image for
3469   // the code cache doesn't have an SHM_X executable permission to check.
3470   assert(UseLargePages && UseSHM, "only for SHM large pages");
3471   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3472   assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3473 
3474   if (!is_size_aligned(bytes, os::large_page_size())) {
3475     return NULL; // Fallback to small pages.
3476   }
3477 
3478   // Create a large shared memory region to attach to based on size.
3479   // Currently, size is the total size of the heap.
3480   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3481   if (shmid == -1) {
3482     // Possible reasons for shmget failure:
3483     // 1. shmmax is too small for Java heap.
3484     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3485     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3486     // 2. not enough large page memory.
3487     //    > check available large pages: cat /proc/meminfo
3488     //    > increase amount of large pages:
3489     //          echo new_value > /proc/sys/vm/nr_hugepages
3490     //      Note 1: different Linux may use different name for this property,
3491     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3492     //      Note 2: it's possible there's enough physical memory available but
3493     //            they are so fragmented after a long run that they can't
3494     //            coalesce into large pages. Try to reserve large pages when
3495     //            the system is still "fresh".
3496     shm_warning_with_errno("Failed to reserve shared memory.");
3497     return NULL;
3498   }
3499 
3500   // Attach to the region.
3501   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3502 
3503   // Remove shmid. If shmat() is successful, the actual shared memory segment
3504   // will be deleted when it's detached by shmdt() or when the process
3505   // terminates. If shmat() is not successful this will remove the shared
3506   // segment immediately.
3507   shmctl(shmid, IPC_RMID, NULL);
3508 
3509   return addr;
3510 }
3511 
3512 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3513                                         int error) {
3514   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3515 
3516   bool warn_on_failure = UseLargePages &&
3517       (!FLAG_IS_DEFAULT(UseLargePages) ||
3518        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3519        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3520 
3521   if (warn_on_failure) {
3522     char msg[128];
3523     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3524                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3525     warning("%s", msg);
3526   }
3527 }
3528 
3529 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3530                                                         char* req_addr,
3531                                                         bool exec) {
3532   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3533   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3534   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3535 
3536   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3537   char* addr = (char*)::mmap(req_addr, bytes, prot,
3538                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3539                              -1, 0);
3540 
3541   if (addr == MAP_FAILED) {
3542     warn_on_large_pages_failure(req_addr, bytes, errno);
3543     return NULL;
3544   }
3545 
3546   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3547 
3548   return addr;
3549 }
3550 
3551 // Reserve memory using mmap(MAP_HUGETLB).
3552 //  - bytes shall be a multiple of alignment.
3553 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3554 //  - alignment sets the alignment at which memory shall be allocated.
3555 //     It must be a multiple of allocation granularity.
3556 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3557 //  req_addr or NULL.
3558 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3559                                                          size_t alignment,
3560                                                          char* req_addr,
3561                                                          bool exec) {
3562   size_t large_page_size = os::large_page_size();
3563   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3564 
3565   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3566   assert(is_size_aligned(bytes, alignment), "Must be");
3567 
3568   // First reserve - but not commit - the address range in small pages.
3569   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3570 
3571   if (start == NULL) {
3572     return NULL;
3573   }
3574 
3575   assert(is_ptr_aligned(start, alignment), "Must be");
3576 
3577   char* end = start + bytes;
3578 
3579   // Find the regions of the allocated chunk that can be promoted to large pages.
3580   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3581   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3582 
3583   size_t lp_bytes = lp_end - lp_start;
3584 
3585   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3586 
3587   if (lp_bytes == 0) {
3588     // The mapped region doesn't even span the start and the end of a large page.
3589     // Fall back to allocate a non-special area.
3590     ::munmap(start, end - start);
3591     return NULL;
3592   }
3593 
3594   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3595 
3596   void* result;
3597 
3598   // Commit small-paged leading area.
3599   if (start != lp_start) {
3600     result = ::mmap(start, lp_start - start, prot,
3601                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3602                     -1, 0);
3603     if (result == MAP_FAILED) {
3604       ::munmap(lp_start, end - lp_start);
3605       return NULL;
3606     }
3607   }
3608 
3609   // Commit large-paged area.
3610   result = ::mmap(lp_start, lp_bytes, prot,
3611                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3612                   -1, 0);
3613   if (result == MAP_FAILED) {
3614     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3615     // If the mmap above fails, the large pages region will be unmapped and we
3616     // have regions before and after with small pages. Release these regions.
3617     //
3618     // |  mapped  |  unmapped  |  mapped  |
3619     // ^          ^            ^          ^
3620     // start      lp_start     lp_end     end
3621     //
3622     ::munmap(start, lp_start - start);
3623     ::munmap(lp_end, end - lp_end);
3624     return NULL;
3625   }
3626 
3627   // Commit small-paged trailing area.
3628   if (lp_end != end) {
3629     result = ::mmap(lp_end, end - lp_end, prot,
3630                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3631                     -1, 0);
3632     if (result == MAP_FAILED) {
3633       ::munmap(start, lp_end - start);
3634       return NULL;
3635     }
3636   }
3637 
3638   return start;
3639 }
3640 
3641 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3642                                                    size_t alignment,
3643                                                    char* req_addr,
3644                                                    bool exec) {
3645   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3646   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3647   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3648   assert(is_power_of_2(os::large_page_size()), "Must be");
3649   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3650 
3651   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3652     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3653   } else {
3654     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3655   }
3656 }
3657 
3658 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3659                                  char* req_addr, bool exec) {
3660   assert(UseLargePages, "only for large pages");
3661 
3662   char* addr;
3663   if (UseSHM) {
3664     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3665   } else {
3666     assert(UseHugeTLBFS, "must be");
3667     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3668   }
3669 
3670   if (addr != NULL) {
3671     if (UseNUMAInterleaving) {
3672       numa_make_global(addr, bytes);
3673     }
3674 
3675     // The memory is committed
3676     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3677   }
3678 
3679   return addr;
3680 }
3681 
3682 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3683   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3684   return shmdt(base) == 0;
3685 }
3686 
3687 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3688   return pd_release_memory(base, bytes);
3689 }
3690 
3691 bool os::release_memory_special(char* base, size_t bytes) {
3692   bool res;
3693   if (MemTracker::tracking_level() > NMT_minimal) {
3694     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3695     res = os::Linux::release_memory_special_impl(base, bytes);
3696     if (res) {
3697       tkr.record((address)base, bytes);
3698     }
3699 
3700   } else {
3701     res = os::Linux::release_memory_special_impl(base, bytes);
3702   }
3703   return res;
3704 }
3705 
3706 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3707   assert(UseLargePages, "only for large pages");
3708   bool res;
3709 
3710   if (UseSHM) {
3711     res = os::Linux::release_memory_special_shm(base, bytes);
3712   } else {
3713     assert(UseHugeTLBFS, "must be");
3714     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3715   }
3716   return res;
3717 }
3718 
3719 size_t os::large_page_size() {
3720   return _large_page_size;
3721 }
3722 
3723 // With SysV SHM the entire memory region must be allocated as shared
3724 // memory.
3725 // HugeTLBFS allows application to commit large page memory on demand.
3726 // However, when committing memory with HugeTLBFS fails, the region
3727 // that was supposed to be committed will lose the old reservation
3728 // and allow other threads to steal that memory region. Because of this
3729 // behavior we can't commit HugeTLBFS memory.
3730 bool os::can_commit_large_page_memory() {
3731   return UseTransparentHugePages;
3732 }
3733 
3734 bool os::can_execute_large_page_memory() {
3735   return UseTransparentHugePages || UseHugeTLBFS;
3736 }
3737 
3738 // Reserve memory at an arbitrary address, only if that area is
3739 // available (and not reserved for something else).
3740 
3741 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3742   const int max_tries = 10;
3743   char* base[max_tries];
3744   size_t size[max_tries];
3745   const size_t gap = 0x000000;
3746 
3747   // Assert only that the size is a multiple of the page size, since
3748   // that's all that mmap requires, and since that's all we really know
3749   // about at this low abstraction level.  If we need higher alignment,
3750   // we can either pass an alignment to this method or verify alignment
3751   // in one of the methods further up the call chain.  See bug 5044738.
3752   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3753 
3754   // Repeatedly allocate blocks until the block is allocated at the
3755   // right spot.
3756 
3757   // Linux mmap allows caller to pass an address as hint; give it a try first,
3758   // if kernel honors the hint then we can return immediately.
3759   char * addr = anon_mmap(requested_addr, bytes, false);
3760   if (addr == requested_addr) {
3761     return requested_addr;
3762   }
3763 
3764   if (addr != NULL) {
3765     // mmap() is successful but it fails to reserve at the requested address
3766     anon_munmap(addr, bytes);
3767   }
3768 
3769   int i;
3770   for (i = 0; i < max_tries; ++i) {
3771     base[i] = reserve_memory(bytes);
3772 
3773     if (base[i] != NULL) {
3774       // Is this the block we wanted?
3775       if (base[i] == requested_addr) {
3776         size[i] = bytes;
3777         break;
3778       }
3779 
3780       // Does this overlap the block we wanted? Give back the overlapped
3781       // parts and try again.
3782 
3783       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3784       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3785         unmap_memory(base[i], top_overlap);
3786         base[i] += top_overlap;
3787         size[i] = bytes - top_overlap;
3788       } else {
3789         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3790         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3791           unmap_memory(requested_addr, bottom_overlap);
3792           size[i] = bytes - bottom_overlap;
3793         } else {
3794           size[i] = bytes;
3795         }
3796       }
3797     }
3798   }
3799 
3800   // Give back the unused reserved pieces.
3801 
3802   for (int j = 0; j < i; ++j) {
3803     if (base[j] != NULL) {
3804       unmap_memory(base[j], size[j]);
3805     }
3806   }
3807 
3808   if (i < max_tries) {
3809     return requested_addr;
3810   } else {
3811     return NULL;
3812   }
3813 }
3814 
3815 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3816   return ::read(fd, buf, nBytes);
3817 }
3818 
3819 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3820   return ::pread(fd, buf, nBytes, offset);
3821 }
3822 
3823 // Short sleep, direct OS call.
3824 //
3825 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3826 // sched_yield(2) will actually give up the CPU:
3827 //
3828 //   * Alone on this pariticular CPU, keeps running.
3829 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3830 //     (pre 2.6.39).
3831 //
3832 // So calling this with 0 is an alternative.
3833 //
3834 void os::naked_short_sleep(jlong ms) {
3835   struct timespec req;
3836 
3837   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3838   req.tv_sec = 0;
3839   if (ms > 0) {
3840     req.tv_nsec = (ms % 1000) * 1000000;
3841   } else {
3842     req.tv_nsec = 1;
3843   }
3844 
3845   nanosleep(&req, NULL);
3846 
3847   return;
3848 }
3849 
3850 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3851 void os::infinite_sleep() {
3852   while (true) {    // sleep forever ...
3853     ::sleep(100);   // ... 100 seconds at a time
3854   }
3855 }
3856 
3857 // Used to convert frequent JVM_Yield() to nops
3858 bool os::dont_yield() {
3859   return DontYieldALot;
3860 }
3861 
3862 void os::naked_yield() {
3863   sched_yield();
3864 }
3865 
3866 ////////////////////////////////////////////////////////////////////////////////
3867 // thread priority support
3868 
3869 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3870 // only supports dynamic priority, static priority must be zero. For real-time
3871 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3872 // However, for large multi-threaded applications, SCHED_RR is not only slower
3873 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3874 // of 5 runs - Sep 2005).
3875 //
3876 // The following code actually changes the niceness of kernel-thread/LWP. It
3877 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3878 // not the entire user process, and user level threads are 1:1 mapped to kernel
3879 // threads. It has always been the case, but could change in the future. For
3880 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3881 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3882 
3883 int os::java_to_os_priority[CriticalPriority + 1] = {
3884   19,              // 0 Entry should never be used
3885 
3886    4,              // 1 MinPriority
3887    3,              // 2
3888    2,              // 3
3889 
3890    1,              // 4
3891    0,              // 5 NormPriority
3892   -1,              // 6
3893 
3894   -2,              // 7
3895   -3,              // 8
3896   -4,              // 9 NearMaxPriority
3897 
3898   -5,              // 10 MaxPriority
3899 
3900   -5               // 11 CriticalPriority
3901 };
3902 
3903 static int prio_init() {
3904   if (ThreadPriorityPolicy == 1) {
3905     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3906     // if effective uid is not root. Perhaps, a more elegant way of doing
3907     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3908     if (geteuid() != 0) {
3909       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3910         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3911       }
3912       ThreadPriorityPolicy = 0;
3913     }
3914   }
3915   if (UseCriticalJavaThreadPriority) {
3916     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3917   }
3918   return 0;
3919 }
3920 
3921 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3922   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3923 
3924   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3925   return (ret == 0) ? OS_OK : OS_ERR;
3926 }
3927 
3928 OSReturn os::get_native_priority(const Thread* const thread,
3929                                  int *priority_ptr) {
3930   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3931     *priority_ptr = java_to_os_priority[NormPriority];
3932     return OS_OK;
3933   }
3934 
3935   errno = 0;
3936   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3937   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3938 }
3939 
3940 // Hint to the underlying OS that a task switch would not be good.
3941 // Void return because it's a hint and can fail.
3942 void os::hint_no_preempt() {}
3943 
3944 ////////////////////////////////////////////////////////////////////////////////
3945 // suspend/resume support
3946 
3947 //  the low-level signal-based suspend/resume support is a remnant from the
3948 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3949 //  within hotspot. Now there is a single use-case for this:
3950 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3951 //      that runs in the watcher thread.
3952 //  The remaining code is greatly simplified from the more general suspension
3953 //  code that used to be used.
3954 //
3955 //  The protocol is quite simple:
3956 //  - suspend:
3957 //      - sends a signal to the target thread
3958 //      - polls the suspend state of the osthread using a yield loop
3959 //      - target thread signal handler (SR_handler) sets suspend state
3960 //        and blocks in sigsuspend until continued
3961 //  - resume:
3962 //      - sets target osthread state to continue
3963 //      - sends signal to end the sigsuspend loop in the SR_handler
3964 //
3965 //  Note that the SR_lock plays no role in this suspend/resume protocol,
3966 //  but is checked for NULL in SR_handler as a thread termination indicator.
3967 
3968 static void resume_clear_context(OSThread *osthread) {
3969   osthread->set_ucontext(NULL);
3970   osthread->set_siginfo(NULL);
3971 }
3972 
3973 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3974                                  ucontext_t* context) {
3975   osthread->set_ucontext(context);
3976   osthread->set_siginfo(siginfo);
3977 }
3978 
3979 // Handler function invoked when a thread's execution is suspended or
3980 // resumed. We have to be careful that only async-safe functions are
3981 // called here (Note: most pthread functions are not async safe and
3982 // should be avoided.)
3983 //
3984 // Note: sigwait() is a more natural fit than sigsuspend() from an
3985 // interface point of view, but sigwait() prevents the signal hander
3986 // from being run. libpthread would get very confused by not having
3987 // its signal handlers run and prevents sigwait()'s use with the
3988 // mutex granting granting signal.
3989 //
3990 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3991 //
3992 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3993   // Save and restore errno to avoid confusing native code with EINTR
3994   // after sigsuspend.
3995   int old_errno = errno;
3996 
3997   Thread* thread = Thread::current_or_null_safe();
3998   assert(thread != NULL, "Missing current thread in SR_handler");
3999 
4000   // On some systems we have seen signal delivery get "stuck" until the signal
4001   // mask is changed as part of thread termination. Check that the current thread
4002   // has not already terminated (via SR_lock()) - else the following assertion
4003   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4004   // destructor has completed.
4005 
4006   if (thread->SR_lock() == NULL) {
4007     return;
4008   }
4009 
4010   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4011 
4012   OSThread* osthread = thread->osthread();
4013 
4014   os::SuspendResume::State current = osthread->sr.state();
4015   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4016     suspend_save_context(osthread, siginfo, context);
4017 
4018     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4019     os::SuspendResume::State state = osthread->sr.suspended();
4020     if (state == os::SuspendResume::SR_SUSPENDED) {
4021       sigset_t suspend_set;  // signals for sigsuspend()
4022       sigemptyset(&suspend_set);
4023       // get current set of blocked signals and unblock resume signal
4024       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4025       sigdelset(&suspend_set, SR_signum);
4026 
4027       sr_semaphore.signal();
4028       // wait here until we are resumed
4029       while (1) {
4030         sigsuspend(&suspend_set);
4031 
4032         os::SuspendResume::State result = osthread->sr.running();
4033         if (result == os::SuspendResume::SR_RUNNING) {
4034           sr_semaphore.signal();
4035           break;
4036         }
4037       }
4038 
4039     } else if (state == os::SuspendResume::SR_RUNNING) {
4040       // request was cancelled, continue
4041     } else {
4042       ShouldNotReachHere();
4043     }
4044 
4045     resume_clear_context(osthread);
4046   } else if (current == os::SuspendResume::SR_RUNNING) {
4047     // request was cancelled, continue
4048   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4049     // ignore
4050   } else {
4051     // ignore
4052   }
4053 
4054   errno = old_errno;
4055 }
4056 
4057 static int SR_initialize() {
4058   struct sigaction act;
4059   char *s;
4060 
4061   // Get signal number to use for suspend/resume
4062   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4063     int sig = ::strtol(s, 0, 10);
4064     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4065         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4066       SR_signum = sig;
4067     } else {
4068       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4069               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4070     }
4071   }
4072 
4073   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4074          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4075 
4076   sigemptyset(&SR_sigset);
4077   sigaddset(&SR_sigset, SR_signum);
4078 
4079   // Set up signal handler for suspend/resume
4080   act.sa_flags = SA_RESTART|SA_SIGINFO;
4081   act.sa_handler = (void (*)(int)) SR_handler;
4082 
4083   // SR_signum is blocked by default.
4084   // 4528190 - We also need to block pthread restart signal (32 on all
4085   // supported Linux platforms). Note that LinuxThreads need to block
4086   // this signal for all threads to work properly. So we don't have
4087   // to use hard-coded signal number when setting up the mask.
4088   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4089 
4090   if (sigaction(SR_signum, &act, 0) == -1) {
4091     return -1;
4092   }
4093 
4094   // Save signal flag
4095   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4096   return 0;
4097 }
4098 
4099 static int sr_notify(OSThread* osthread) {
4100   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4101   assert_status(status == 0, status, "pthread_kill");
4102   return status;
4103 }
4104 
4105 // "Randomly" selected value for how long we want to spin
4106 // before bailing out on suspending a thread, also how often
4107 // we send a signal to a thread we want to resume
4108 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4109 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4110 
4111 // returns true on success and false on error - really an error is fatal
4112 // but this seems the normal response to library errors
4113 static bool do_suspend(OSThread* osthread) {
4114   assert(osthread->sr.is_running(), "thread should be running");
4115   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4116 
4117   // mark as suspended and send signal
4118   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4119     // failed to switch, state wasn't running?
4120     ShouldNotReachHere();
4121     return false;
4122   }
4123 
4124   if (sr_notify(osthread) != 0) {
4125     ShouldNotReachHere();
4126   }
4127 
4128   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4129   while (true) {
4130     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4131       break;
4132     } else {
4133       // timeout
4134       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4135       if (cancelled == os::SuspendResume::SR_RUNNING) {
4136         return false;
4137       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4138         // make sure that we consume the signal on the semaphore as well
4139         sr_semaphore.wait();
4140         break;
4141       } else {
4142         ShouldNotReachHere();
4143         return false;
4144       }
4145     }
4146   }
4147 
4148   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4149   return true;
4150 }
4151 
4152 static void do_resume(OSThread* osthread) {
4153   assert(osthread->sr.is_suspended(), "thread should be suspended");
4154   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4155 
4156   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4157     // failed to switch to WAKEUP_REQUEST
4158     ShouldNotReachHere();
4159     return;
4160   }
4161 
4162   while (true) {
4163     if (sr_notify(osthread) == 0) {
4164       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4165         if (osthread->sr.is_running()) {
4166           return;
4167         }
4168       }
4169     } else {
4170       ShouldNotReachHere();
4171     }
4172   }
4173 
4174   guarantee(osthread->sr.is_running(), "Must be running!");
4175 }
4176 
4177 ///////////////////////////////////////////////////////////////////////////////////
4178 // signal handling (except suspend/resume)
4179 
4180 // This routine may be used by user applications as a "hook" to catch signals.
4181 // The user-defined signal handler must pass unrecognized signals to this
4182 // routine, and if it returns true (non-zero), then the signal handler must
4183 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4184 // routine will never retun false (zero), but instead will execute a VM panic
4185 // routine kill the process.
4186 //
4187 // If this routine returns false, it is OK to call it again.  This allows
4188 // the user-defined signal handler to perform checks either before or after
4189 // the VM performs its own checks.  Naturally, the user code would be making
4190 // a serious error if it tried to handle an exception (such as a null check
4191 // or breakpoint) that the VM was generating for its own correct operation.
4192 //
4193 // This routine may recognize any of the following kinds of signals:
4194 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4195 // It should be consulted by handlers for any of those signals.
4196 //
4197 // The caller of this routine must pass in the three arguments supplied
4198 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4199 // field of the structure passed to sigaction().  This routine assumes that
4200 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4201 //
4202 // Note that the VM will print warnings if it detects conflicting signal
4203 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4204 //
4205 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4206                                                  siginfo_t* siginfo,
4207                                                  void* ucontext,
4208                                                  int abort_if_unrecognized);
4209 
4210 void signalHandler(int sig, siginfo_t* info, void* uc) {
4211   assert(info != NULL && uc != NULL, "it must be old kernel");
4212   int orig_errno = errno;  // Preserve errno value over signal handler.
4213   JVM_handle_linux_signal(sig, info, uc, true);
4214   errno = orig_errno;
4215 }
4216 
4217 
4218 // This boolean allows users to forward their own non-matching signals
4219 // to JVM_handle_linux_signal, harmlessly.
4220 bool os::Linux::signal_handlers_are_installed = false;
4221 
4222 // For signal-chaining
4223 struct sigaction sigact[NSIG];
4224 uint64_t sigs = 0;
4225 #if (64 < NSIG-1)
4226 #error "Not all signals can be encoded in sigs. Adapt its type!"
4227 #endif
4228 bool os::Linux::libjsig_is_loaded = false;
4229 typedef struct sigaction *(*get_signal_t)(int);
4230 get_signal_t os::Linux::get_signal_action = NULL;
4231 
4232 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4233   struct sigaction *actp = NULL;
4234 
4235   if (libjsig_is_loaded) {
4236     // Retrieve the old signal handler from libjsig
4237     actp = (*get_signal_action)(sig);
4238   }
4239   if (actp == NULL) {
4240     // Retrieve the preinstalled signal handler from jvm
4241     actp = get_preinstalled_handler(sig);
4242   }
4243 
4244   return actp;
4245 }
4246 
4247 static bool call_chained_handler(struct sigaction *actp, int sig,
4248                                  siginfo_t *siginfo, void *context) {
4249   // Call the old signal handler
4250   if (actp->sa_handler == SIG_DFL) {
4251     // It's more reasonable to let jvm treat it as an unexpected exception
4252     // instead of taking the default action.
4253     return false;
4254   } else if (actp->sa_handler != SIG_IGN) {
4255     if ((actp->sa_flags & SA_NODEFER) == 0) {
4256       // automaticlly block the signal
4257       sigaddset(&(actp->sa_mask), sig);
4258     }
4259 
4260     sa_handler_t hand = NULL;
4261     sa_sigaction_t sa = NULL;
4262     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4263     // retrieve the chained handler
4264     if (siginfo_flag_set) {
4265       sa = actp->sa_sigaction;
4266     } else {
4267       hand = actp->sa_handler;
4268     }
4269 
4270     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4271       actp->sa_handler = SIG_DFL;
4272     }
4273 
4274     // try to honor the signal mask
4275     sigset_t oset;
4276     sigemptyset(&oset);
4277     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4278 
4279     // call into the chained handler
4280     if (siginfo_flag_set) {
4281       (*sa)(sig, siginfo, context);
4282     } else {
4283       (*hand)(sig);
4284     }
4285 
4286     // restore the signal mask
4287     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4288   }
4289   // Tell jvm's signal handler the signal is taken care of.
4290   return true;
4291 }
4292 
4293 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4294   bool chained = false;
4295   // signal-chaining
4296   if (UseSignalChaining) {
4297     struct sigaction *actp = get_chained_signal_action(sig);
4298     if (actp != NULL) {
4299       chained = call_chained_handler(actp, sig, siginfo, context);
4300     }
4301   }
4302   return chained;
4303 }
4304 
4305 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4306   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4307     return &sigact[sig];
4308   }
4309   return NULL;
4310 }
4311 
4312 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4313   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4314   sigact[sig] = oldAct;
4315   sigs |= (uint64_t)1 << (sig-1);
4316 }
4317 
4318 // for diagnostic
4319 int sigflags[NSIG];
4320 
4321 int os::Linux::get_our_sigflags(int sig) {
4322   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4323   return sigflags[sig];
4324 }
4325 
4326 void os::Linux::set_our_sigflags(int sig, int flags) {
4327   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4328   if (sig > 0 && sig < NSIG) {
4329     sigflags[sig] = flags;
4330   }
4331 }
4332 
4333 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4334   // Check for overwrite.
4335   struct sigaction oldAct;
4336   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4337 
4338   void* oldhand = oldAct.sa_sigaction
4339                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4340                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4341   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4342       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4343       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4344     if (AllowUserSignalHandlers || !set_installed) {
4345       // Do not overwrite; user takes responsibility to forward to us.
4346       return;
4347     } else if (UseSignalChaining) {
4348       // save the old handler in jvm
4349       save_preinstalled_handler(sig, oldAct);
4350       // libjsig also interposes the sigaction() call below and saves the
4351       // old sigaction on it own.
4352     } else {
4353       fatal("Encountered unexpected pre-existing sigaction handler "
4354             "%#lx for signal %d.", (long)oldhand, sig);
4355     }
4356   }
4357 
4358   struct sigaction sigAct;
4359   sigfillset(&(sigAct.sa_mask));
4360   sigAct.sa_handler = SIG_DFL;
4361   if (!set_installed) {
4362     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4363   } else {
4364     sigAct.sa_sigaction = signalHandler;
4365     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4366   }
4367   // Save flags, which are set by ours
4368   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4369   sigflags[sig] = sigAct.sa_flags;
4370 
4371   int ret = sigaction(sig, &sigAct, &oldAct);
4372   assert(ret == 0, "check");
4373 
4374   void* oldhand2  = oldAct.sa_sigaction
4375                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4376                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4377   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4378 }
4379 
4380 // install signal handlers for signals that HotSpot needs to
4381 // handle in order to support Java-level exception handling.
4382 
4383 void os::Linux::install_signal_handlers() {
4384   if (!signal_handlers_are_installed) {
4385     signal_handlers_are_installed = true;
4386 
4387     // signal-chaining
4388     typedef void (*signal_setting_t)();
4389     signal_setting_t begin_signal_setting = NULL;
4390     signal_setting_t end_signal_setting = NULL;
4391     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4392                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4393     if (begin_signal_setting != NULL) {
4394       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4395                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4396       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4397                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4398       libjsig_is_loaded = true;
4399       assert(UseSignalChaining, "should enable signal-chaining");
4400     }
4401     if (libjsig_is_loaded) {
4402       // Tell libjsig jvm is setting signal handlers
4403       (*begin_signal_setting)();
4404     }
4405 
4406     set_signal_handler(SIGSEGV, true);
4407     set_signal_handler(SIGPIPE, true);
4408     set_signal_handler(SIGBUS, true);
4409     set_signal_handler(SIGILL, true);
4410     set_signal_handler(SIGFPE, true);
4411 #if defined(PPC64)
4412     set_signal_handler(SIGTRAP, true);
4413 #endif
4414     set_signal_handler(SIGXFSZ, true);
4415 
4416     if (libjsig_is_loaded) {
4417       // Tell libjsig jvm finishes setting signal handlers
4418       (*end_signal_setting)();
4419     }
4420 
4421     // We don't activate signal checker if libjsig is in place, we trust ourselves
4422     // and if UserSignalHandler is installed all bets are off.
4423     // Log that signal checking is off only if -verbose:jni is specified.
4424     if (CheckJNICalls) {
4425       if (libjsig_is_loaded) {
4426         if (PrintJNIResolving) {
4427           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4428         }
4429         check_signals = false;
4430       }
4431       if (AllowUserSignalHandlers) {
4432         if (PrintJNIResolving) {
4433           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4434         }
4435         check_signals = false;
4436       }
4437     }
4438   }
4439 }
4440 
4441 // This is the fastest way to get thread cpu time on Linux.
4442 // Returns cpu time (user+sys) for any thread, not only for current.
4443 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4444 // It might work on 2.6.10+ with a special kernel/glibc patch.
4445 // For reference, please, see IEEE Std 1003.1-2004:
4446 //   http://www.unix.org/single_unix_specification
4447 
4448 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4449   struct timespec tp;
4450   int rc = os::Linux::clock_gettime(clockid, &tp);
4451   assert(rc == 0, "clock_gettime is expected to return 0 code");
4452 
4453   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4454 }
4455 
4456 void os::Linux::initialize_os_info() {
4457   assert(_os_version == 0, "OS info already initialized");
4458 
4459   struct utsname _uname;
4460 
4461   uint32_t major;
4462   uint32_t minor;
4463   uint32_t fix;
4464 
4465   int rc;
4466 
4467   // Kernel version is unknown if
4468   // verification below fails.
4469   _os_version = 0x01000000;
4470 
4471   rc = uname(&_uname);
4472   if (rc != -1) {
4473 
4474     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4475     if (rc == 3) {
4476 
4477       if (major < 256 && minor < 256 && fix < 256) {
4478         // Kernel version format is as expected,
4479         // set it overriding unknown state.
4480         _os_version = (major << 16) |
4481                       (minor << 8 ) |
4482                       (fix   << 0 ) ;
4483       }
4484     }
4485   }
4486 }
4487 
4488 uint32_t os::Linux::os_version() {
4489   assert(_os_version != 0, "not initialized");
4490   return _os_version & 0x00FFFFFF;
4491 }
4492 
4493 bool os::Linux::os_version_is_known() {
4494   assert(_os_version != 0, "not initialized");
4495   return _os_version & 0x01000000 ? false : true;
4496 }
4497 
4498 /////
4499 // glibc on Linux platform uses non-documented flag
4500 // to indicate, that some special sort of signal
4501 // trampoline is used.
4502 // We will never set this flag, and we should
4503 // ignore this flag in our diagnostic
4504 #ifdef SIGNIFICANT_SIGNAL_MASK
4505   #undef SIGNIFICANT_SIGNAL_MASK
4506 #endif
4507 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4508 
4509 static const char* get_signal_handler_name(address handler,
4510                                            char* buf, int buflen) {
4511   int offset = 0;
4512   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4513   if (found) {
4514     // skip directory names
4515     const char *p1, *p2;
4516     p1 = buf;
4517     size_t len = strlen(os::file_separator());
4518     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4519     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4520   } else {
4521     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4522   }
4523   return buf;
4524 }
4525 
4526 static void print_signal_handler(outputStream* st, int sig,
4527                                  char* buf, size_t buflen) {
4528   struct sigaction sa;
4529 
4530   sigaction(sig, NULL, &sa);
4531 
4532   // See comment for SIGNIFICANT_SIGNAL_MASK define
4533   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4534 
4535   st->print("%s: ", os::exception_name(sig, buf, buflen));
4536 
4537   address handler = (sa.sa_flags & SA_SIGINFO)
4538     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4539     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4540 
4541   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4542     st->print("SIG_DFL");
4543   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4544     st->print("SIG_IGN");
4545   } else {
4546     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4547   }
4548 
4549   st->print(", sa_mask[0]=");
4550   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4551 
4552   address rh = VMError::get_resetted_sighandler(sig);
4553   // May be, handler was resetted by VMError?
4554   if (rh != NULL) {
4555     handler = rh;
4556     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4557   }
4558 
4559   st->print(", sa_flags=");
4560   os::Posix::print_sa_flags(st, sa.sa_flags);
4561 
4562   // Check: is it our handler?
4563   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4564       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4565     // It is our signal handler
4566     // check for flags, reset system-used one!
4567     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4568       st->print(
4569                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4570                 os::Linux::get_our_sigflags(sig));
4571     }
4572   }
4573   st->cr();
4574 }
4575 
4576 
4577 #define DO_SIGNAL_CHECK(sig)                      \
4578   do {                                            \
4579     if (!sigismember(&check_signal_done, sig)) {  \
4580       os::Linux::check_signal_handler(sig);       \
4581     }                                             \
4582   } while (0)
4583 
4584 // This method is a periodic task to check for misbehaving JNI applications
4585 // under CheckJNI, we can add any periodic checks here
4586 
4587 void os::run_periodic_checks() {
4588   if (check_signals == false) return;
4589 
4590   // SEGV and BUS if overridden could potentially prevent
4591   // generation of hs*.log in the event of a crash, debugging
4592   // such a case can be very challenging, so we absolutely
4593   // check the following for a good measure:
4594   DO_SIGNAL_CHECK(SIGSEGV);
4595   DO_SIGNAL_CHECK(SIGILL);
4596   DO_SIGNAL_CHECK(SIGFPE);
4597   DO_SIGNAL_CHECK(SIGBUS);
4598   DO_SIGNAL_CHECK(SIGPIPE);
4599   DO_SIGNAL_CHECK(SIGXFSZ);
4600 #if defined(PPC64)
4601   DO_SIGNAL_CHECK(SIGTRAP);
4602 #endif
4603 
4604   // ReduceSignalUsage allows the user to override these handlers
4605   // see comments at the very top and jvm_solaris.h
4606   if (!ReduceSignalUsage) {
4607     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4608     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4609     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4610     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4611   }
4612 
4613   DO_SIGNAL_CHECK(SR_signum);
4614 }
4615 
4616 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4617 
4618 static os_sigaction_t os_sigaction = NULL;
4619 
4620 void os::Linux::check_signal_handler(int sig) {
4621   char buf[O_BUFLEN];
4622   address jvmHandler = NULL;
4623 
4624 
4625   struct sigaction act;
4626   if (os_sigaction == NULL) {
4627     // only trust the default sigaction, in case it has been interposed
4628     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4629     if (os_sigaction == NULL) return;
4630   }
4631 
4632   os_sigaction(sig, (struct sigaction*)NULL, &act);
4633 
4634 
4635   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4636 
4637   address thisHandler = (act.sa_flags & SA_SIGINFO)
4638     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4639     : CAST_FROM_FN_PTR(address, act.sa_handler);
4640 
4641 
4642   switch (sig) {
4643   case SIGSEGV:
4644   case SIGBUS:
4645   case SIGFPE:
4646   case SIGPIPE:
4647   case SIGILL:
4648   case SIGXFSZ:
4649     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4650     break;
4651 
4652   case SHUTDOWN1_SIGNAL:
4653   case SHUTDOWN2_SIGNAL:
4654   case SHUTDOWN3_SIGNAL:
4655   case BREAK_SIGNAL:
4656     jvmHandler = (address)user_handler();
4657     break;
4658 
4659   default:
4660     if (sig == SR_signum) {
4661       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4662     } else {
4663       return;
4664     }
4665     break;
4666   }
4667 
4668   if (thisHandler != jvmHandler) {
4669     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4670     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4671     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4672     // No need to check this sig any longer
4673     sigaddset(&check_signal_done, sig);
4674     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4675     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4676       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4677                     exception_name(sig, buf, O_BUFLEN));
4678     }
4679   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4680     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4681     tty->print("expected:");
4682     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4683     tty->cr();
4684     tty->print("  found:");
4685     os::Posix::print_sa_flags(tty, act.sa_flags);
4686     tty->cr();
4687     // No need to check this sig any longer
4688     sigaddset(&check_signal_done, sig);
4689   }
4690 
4691   // Dump all the signal
4692   if (sigismember(&check_signal_done, sig)) {
4693     print_signal_handlers(tty, buf, O_BUFLEN);
4694   }
4695 }
4696 
4697 extern void report_error(char* file_name, int line_no, char* title,
4698                          char* format, ...);
4699 
4700 // this is called _before_ the most of global arguments have been parsed
4701 void os::init(void) {
4702   char dummy;   // used to get a guess on initial stack address
4703 //  first_hrtime = gethrtime();
4704 
4705   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4706 
4707   init_random(1234567);
4708 
4709   ThreadCritical::initialize();
4710 
4711   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4712   if (Linux::page_size() == -1) {
4713     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4714           os::strerror(errno));
4715   }
4716   init_page_sizes((size_t) Linux::page_size());
4717 
4718   Linux::initialize_system_info();
4719 
4720   Linux::initialize_os_info();
4721 
4722   // main_thread points to the aboriginal thread
4723   Linux::_main_thread = pthread_self();
4724 
4725   Linux::clock_init();
4726   initial_time_count = javaTimeNanos();
4727 
4728   // pthread_condattr initialization for monotonic clock
4729   int status;
4730   pthread_condattr_t* _condattr = os::Linux::condAttr();
4731   if ((status = pthread_condattr_init(_condattr)) != 0) {
4732     fatal("pthread_condattr_init: %s", os::strerror(status));
4733   }
4734   // Only set the clock if CLOCK_MONOTONIC is available
4735   if (os::supports_monotonic_clock()) {
4736     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4737       if (status == EINVAL) {
4738         warning("Unable to use monotonic clock with relative timed-waits" \
4739                 " - changes to the time-of-day clock may have adverse affects");
4740       } else {
4741         fatal("pthread_condattr_setclock: %s", os::strerror(status));
4742       }
4743     }
4744   }
4745   // else it defaults to CLOCK_REALTIME
4746 
4747   // retrieve entry point for pthread_setname_np
4748   Linux::_pthread_setname_np =
4749     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4750 
4751 }
4752 
4753 // To install functions for atexit system call
4754 extern "C" {
4755   static void perfMemory_exit_helper() {
4756     perfMemory_exit();
4757   }
4758 }
4759 
4760 // this is called _after_ the global arguments have been parsed
4761 jint os::init_2(void) {
4762   Linux::fast_thread_clock_init();
4763 
4764   // Allocate a single page and mark it as readable for safepoint polling
4765   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4766   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4767 
4768   os::set_polling_page(polling_page);
4769   log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4770 
4771   if (!UseMembar) {
4772     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4773     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4774     os::set_memory_serialize_page(mem_serialize_page);
4775     log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4776   }
4777 
4778   // initialize suspend/resume support - must do this before signal_sets_init()
4779   if (SR_initialize() != 0) {
4780     perror("SR_initialize failed");
4781     return JNI_ERR;
4782   }
4783 
4784   Linux::signal_sets_init();
4785   Linux::install_signal_handlers();
4786 
4787   // Check and sets minimum stack sizes against command line options
4788   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4789     return JNI_ERR;
4790   }
4791   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4792 
4793 #if defined(IA32)
4794   workaround_expand_exec_shield_cs_limit();
4795 #endif
4796 
4797   Linux::libpthread_init();
4798   log_info(os)("HotSpot is running with %s, %s",
4799                Linux::glibc_version(), Linux::libpthread_version());
4800 
4801   if (UseNUMA) {
4802     if (!Linux::libnuma_init()) {
4803       UseNUMA = false;
4804     } else {
4805       if ((Linux::numa_max_node() < 1)) {
4806         // There's only one node(they start from 0), disable NUMA.
4807         UseNUMA = false;
4808       }
4809     }
4810     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4811     // we can make the adaptive lgrp chunk resizing work. If the user specified
4812     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4813     // disable adaptive resizing.
4814     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4815       if (FLAG_IS_DEFAULT(UseNUMA)) {
4816         UseNUMA = false;
4817       } else {
4818         if (FLAG_IS_DEFAULT(UseLargePages) &&
4819             FLAG_IS_DEFAULT(UseSHM) &&
4820             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4821           UseLargePages = false;
4822         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4823           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4824           UseAdaptiveSizePolicy = false;
4825           UseAdaptiveNUMAChunkSizing = false;
4826         }
4827       }
4828     }
4829     if (!UseNUMA && ForceNUMA) {
4830       UseNUMA = true;
4831     }
4832   }
4833 
4834   if (MaxFDLimit) {
4835     // set the number of file descriptors to max. print out error
4836     // if getrlimit/setrlimit fails but continue regardless.
4837     struct rlimit nbr_files;
4838     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4839     if (status != 0) {
4840       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4841     } else {
4842       nbr_files.rlim_cur = nbr_files.rlim_max;
4843       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4844       if (status != 0) {
4845         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4846       }
4847     }
4848   }
4849 
4850   // Initialize lock used to serialize thread creation (see os::create_thread)
4851   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4852 
4853   // at-exit methods are called in the reverse order of their registration.
4854   // atexit functions are called on return from main or as a result of a
4855   // call to exit(3C). There can be only 32 of these functions registered
4856   // and atexit() does not set errno.
4857 
4858   if (PerfAllowAtExitRegistration) {
4859     // only register atexit functions if PerfAllowAtExitRegistration is set.
4860     // atexit functions can be delayed until process exit time, which
4861     // can be problematic for embedded VM situations. Embedded VMs should
4862     // call DestroyJavaVM() to assure that VM resources are released.
4863 
4864     // note: perfMemory_exit_helper atexit function may be removed in
4865     // the future if the appropriate cleanup code can be added to the
4866     // VM_Exit VMOperation's doit method.
4867     if (atexit(perfMemory_exit_helper) != 0) {
4868       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4869     }
4870   }
4871 
4872   // initialize thread priority policy
4873   prio_init();
4874 
4875   return JNI_OK;
4876 }
4877 
4878 // Mark the polling page as unreadable
4879 void os::make_polling_page_unreadable(void) {
4880   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4881     fatal("Could not disable polling page");
4882   }
4883 }
4884 
4885 // Mark the polling page as readable
4886 void os::make_polling_page_readable(void) {
4887   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4888     fatal("Could not enable polling page");
4889   }
4890 }
4891 
4892 // older glibc versions don't have this macro (which expands to
4893 // an optimized bit-counting function) so we have to roll our own
4894 #ifndef CPU_COUNT
4895 
4896 static int _cpu_count(const cpu_set_t* cpus) {
4897   int count = 0;
4898   // only look up to the number of configured processors
4899   for (int i = 0; i < os::processor_count(); i++) {
4900     if (CPU_ISSET(i, cpus)) {
4901       count++;
4902     }
4903   }
4904   return count;
4905 }
4906 
4907 #define CPU_COUNT(cpus) _cpu_count(cpus)
4908 
4909 #endif // CPU_COUNT
4910 
4911 // Get the current number of available processors for this process.
4912 // This value can change at any time during a process's lifetime.
4913 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4914 // If it appears there may be more than 1024 processors then we do a
4915 // dynamic check - see 6515172 for details.
4916 // If anything goes wrong we fallback to returning the number of online
4917 // processors - which can be greater than the number available to the process.
4918 int os::active_processor_count() {
4919   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
4920   cpu_set_t* cpus_p = &cpus;
4921   int cpus_size = sizeof(cpu_set_t);
4922 
4923   int configured_cpus = processor_count();  // upper bound on available cpus
4924   int cpu_count = 0;
4925 
4926 // old build platforms may not support dynamic cpu sets
4927 #ifdef CPU_ALLOC
4928 
4929   // To enable easy testing of the dynamic path on different platforms we
4930   // introduce a diagnostic flag: UseCpuAllocPath
4931   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4932     // kernel may use a mask bigger than cpu_set_t
4933     log_trace(os)("active_processor_count: using dynamic path %s"
4934                   "- configured processors: %d",
4935                   UseCpuAllocPath ? "(forced) " : "",
4936                   configured_cpus);
4937     cpus_p = CPU_ALLOC(configured_cpus);
4938     if (cpus_p != NULL) {
4939       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4940       // zero it just to be safe
4941       CPU_ZERO_S(cpus_size, cpus_p);
4942     }
4943     else {
4944        // failed to allocate so fallback to online cpus
4945        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4946        log_trace(os)("active_processor_count: "
4947                      "CPU_ALLOC failed (%s) - using "
4948                      "online processor count: %d",
4949                      os::strerror(errno), online_cpus);
4950        return online_cpus;
4951     }
4952   }
4953   else {
4954     log_trace(os)("active_processor_count: using static path - configured processors: %d",
4955                   configured_cpus);
4956   }
4957 #else // CPU_ALLOC
4958 // these stubs won't be executed
4959 #define CPU_COUNT_S(size, cpus) -1
4960 #define CPU_FREE(cpus)
4961 
4962   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4963                 configured_cpus);
4964 #endif // CPU_ALLOC
4965 
4966   // pid 0 means the current thread - which we have to assume represents the process
4967   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4968     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4969       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4970     }
4971     else {
4972       cpu_count = CPU_COUNT(cpus_p);
4973     }
4974     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4975   }
4976   else {
4977     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4978     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4979             "which may exceed available processors", os::strerror(errno), cpu_count);
4980   }
4981 
4982   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4983     CPU_FREE(cpus_p);
4984   }
4985 
4986   assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
4987   return cpu_count;
4988 }
4989 
4990 void os::set_native_thread_name(const char *name) {
4991   if (Linux::_pthread_setname_np) {
4992     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4993     snprintf(buf, sizeof(buf), "%s", name);
4994     buf[sizeof(buf) - 1] = '\0';
4995     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4996     // ERANGE should not happen; all other errors should just be ignored.
4997     assert(rc != ERANGE, "pthread_setname_np failed");
4998   }
4999 }
5000 
5001 bool os::distribute_processes(uint length, uint* distribution) {
5002   // Not yet implemented.
5003   return false;
5004 }
5005 
5006 bool os::bind_to_processor(uint processor_id) {
5007   // Not yet implemented.
5008   return false;
5009 }
5010 
5011 ///
5012 
5013 void os::SuspendedThreadTask::internal_do_task() {
5014   if (do_suspend(_thread->osthread())) {
5015     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5016     do_task(context);
5017     do_resume(_thread->osthread());
5018   }
5019 }
5020 
5021 class PcFetcher : public os::SuspendedThreadTask {
5022  public:
5023   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5024   ExtendedPC result();
5025  protected:
5026   void do_task(const os::SuspendedThreadTaskContext& context);
5027  private:
5028   ExtendedPC _epc;
5029 };
5030 
5031 ExtendedPC PcFetcher::result() {
5032   guarantee(is_done(), "task is not done yet.");
5033   return _epc;
5034 }
5035 
5036 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5037   Thread* thread = context.thread();
5038   OSThread* osthread = thread->osthread();
5039   if (osthread->ucontext() != NULL) {
5040     _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5041   } else {
5042     // NULL context is unexpected, double-check this is the VMThread
5043     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5044   }
5045 }
5046 
5047 // Suspends the target using the signal mechanism and then grabs the PC before
5048 // resuming the target. Used by the flat-profiler only
5049 ExtendedPC os::get_thread_pc(Thread* thread) {
5050   // Make sure that it is called by the watcher for the VMThread
5051   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5052   assert(thread->is_VM_thread(), "Can only be called for VMThread");
5053 
5054   PcFetcher fetcher(thread);
5055   fetcher.run();
5056   return fetcher.result();
5057 }
5058 
5059 ////////////////////////////////////////////////////////////////////////////////
5060 // debug support
5061 
5062 bool os::find(address addr, outputStream* st) {
5063   Dl_info dlinfo;
5064   memset(&dlinfo, 0, sizeof(dlinfo));
5065   if (dladdr(addr, &dlinfo) != 0) {
5066     st->print(PTR_FORMAT ": ", p2i(addr));
5067     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5068       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5069                 p2i(addr) - p2i(dlinfo.dli_saddr));
5070     } else if (dlinfo.dli_fbase != NULL) {
5071       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5072     } else {
5073       st->print("<absolute address>");
5074     }
5075     if (dlinfo.dli_fname != NULL) {
5076       st->print(" in %s", dlinfo.dli_fname);
5077     }
5078     if (dlinfo.dli_fbase != NULL) {
5079       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5080     }
5081     st->cr();
5082 
5083     if (Verbose) {
5084       // decode some bytes around the PC
5085       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5086       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5087       address       lowest = (address) dlinfo.dli_sname;
5088       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5089       if (begin < lowest)  begin = lowest;
5090       Dl_info dlinfo2;
5091       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5092           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5093         end = (address) dlinfo2.dli_saddr;
5094       }
5095       Disassembler::decode(begin, end, st);
5096     }
5097     return true;
5098   }
5099   return false;
5100 }
5101 
5102 ////////////////////////////////////////////////////////////////////////////////
5103 // misc
5104 
5105 // This does not do anything on Linux. This is basically a hook for being
5106 // able to use structured exception handling (thread-local exception filters)
5107 // on, e.g., Win32.
5108 void
5109 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5110                          JavaCallArguments* args, Thread* thread) {
5111   f(value, method, args, thread);
5112 }
5113 
5114 void os::print_statistics() {
5115 }
5116 
5117 bool os::message_box(const char* title, const char* message) {
5118   int i;
5119   fdStream err(defaultStream::error_fd());
5120   for (i = 0; i < 78; i++) err.print_raw("=");
5121   err.cr();
5122   err.print_raw_cr(title);
5123   for (i = 0; i < 78; i++) err.print_raw("-");
5124   err.cr();
5125   err.print_raw_cr(message);
5126   for (i = 0; i < 78; i++) err.print_raw("=");
5127   err.cr();
5128 
5129   char buf[16];
5130   // Prevent process from exiting upon "read error" without consuming all CPU
5131   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5132 
5133   return buf[0] == 'y' || buf[0] == 'Y';
5134 }
5135 
5136 int os::stat(const char *path, struct stat *sbuf) {
5137   char pathbuf[MAX_PATH];
5138   if (strlen(path) > MAX_PATH - 1) {
5139     errno = ENAMETOOLONG;
5140     return -1;
5141   }
5142   os::native_path(strcpy(pathbuf, path));
5143   return ::stat(pathbuf, sbuf);
5144 }
5145 
5146 // Is a (classpath) directory empty?
5147 bool os::dir_is_empty(const char* path) {
5148   DIR *dir = NULL;
5149   struct dirent *ptr;
5150 
5151   dir = opendir(path);
5152   if (dir == NULL) return true;
5153 
5154   // Scan the directory
5155   bool result = true;
5156   char buf[sizeof(struct dirent) + MAX_PATH];
5157   while (result && (ptr = ::readdir(dir)) != NULL) {
5158     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5159       result = false;
5160     }
5161   }
5162   closedir(dir);
5163   return result;
5164 }
5165 
5166 // This code originates from JDK's sysOpen and open64_w
5167 // from src/solaris/hpi/src/system_md.c
5168 
5169 int os::open(const char *path, int oflag, int mode) {
5170   if (strlen(path) > MAX_PATH - 1) {
5171     errno = ENAMETOOLONG;
5172     return -1;
5173   }
5174 
5175   // All file descriptors that are opened in the Java process and not
5176   // specifically destined for a subprocess should have the close-on-exec
5177   // flag set.  If we don't set it, then careless 3rd party native code
5178   // might fork and exec without closing all appropriate file descriptors
5179   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5180   // turn might:
5181   //
5182   // - cause end-of-file to fail to be detected on some file
5183   //   descriptors, resulting in mysterious hangs, or
5184   //
5185   // - might cause an fopen in the subprocess to fail on a system
5186   //   suffering from bug 1085341.
5187   //
5188   // (Yes, the default setting of the close-on-exec flag is a Unix
5189   // design flaw)
5190   //
5191   // See:
5192   // 1085341: 32-bit stdio routines should support file descriptors >255
5193   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5194   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5195   //
5196   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5197   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5198   // because it saves a system call and removes a small window where the flag
5199   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5200   // and we fall back to using FD_CLOEXEC (see below).
5201 #ifdef O_CLOEXEC
5202   oflag |= O_CLOEXEC;
5203 #endif
5204 
5205   int fd = ::open64(path, oflag, mode);
5206   if (fd == -1) return -1;
5207 
5208   //If the open succeeded, the file might still be a directory
5209   {
5210     struct stat64 buf64;
5211     int ret = ::fstat64(fd, &buf64);
5212     int st_mode = buf64.st_mode;
5213 
5214     if (ret != -1) {
5215       if ((st_mode & S_IFMT) == S_IFDIR) {
5216         errno = EISDIR;
5217         ::close(fd);
5218         return -1;
5219       }
5220     } else {
5221       ::close(fd);
5222       return -1;
5223     }
5224   }
5225 
5226 #ifdef FD_CLOEXEC
5227   // Validate that the use of the O_CLOEXEC flag on open above worked.
5228   // With recent kernels, we will perform this check exactly once.
5229   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5230   if (!O_CLOEXEC_is_known_to_work) {
5231     int flags = ::fcntl(fd, F_GETFD);
5232     if (flags != -1) {
5233       if ((flags & FD_CLOEXEC) != 0)
5234         O_CLOEXEC_is_known_to_work = 1;
5235       else
5236         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5237     }
5238   }
5239 #endif
5240 
5241   return fd;
5242 }
5243 
5244 
5245 // create binary file, rewriting existing file if required
5246 int os::create_binary_file(const char* path, bool rewrite_existing) {
5247   int oflags = O_WRONLY | O_CREAT;
5248   if (!rewrite_existing) {
5249     oflags |= O_EXCL;
5250   }
5251   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5252 }
5253 
5254 // return current position of file pointer
5255 jlong os::current_file_offset(int fd) {
5256   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5257 }
5258 
5259 // move file pointer to the specified offset
5260 jlong os::seek_to_file_offset(int fd, jlong offset) {
5261   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5262 }
5263 
5264 // This code originates from JDK's sysAvailable
5265 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5266 
5267 int os::available(int fd, jlong *bytes) {
5268   jlong cur, end;
5269   int mode;
5270   struct stat64 buf64;
5271 
5272   if (::fstat64(fd, &buf64) >= 0) {
5273     mode = buf64.st_mode;
5274     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5275       int n;
5276       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5277         *bytes = n;
5278         return 1;
5279       }
5280     }
5281   }
5282   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5283     return 0;
5284   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5285     return 0;
5286   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5287     return 0;
5288   }
5289   *bytes = end - cur;
5290   return 1;
5291 }
5292 
5293 // Map a block of memory.
5294 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5295                         char *addr, size_t bytes, bool read_only,
5296                         bool allow_exec) {
5297   int prot;
5298   int flags = MAP_PRIVATE;
5299 
5300   if (read_only) {
5301     prot = PROT_READ;
5302   } else {
5303     prot = PROT_READ | PROT_WRITE;
5304   }
5305 
5306   if (allow_exec) {
5307     prot |= PROT_EXEC;
5308   }
5309 
5310   if (addr != NULL) {
5311     flags |= MAP_FIXED;
5312   }
5313 
5314   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5315                                      fd, file_offset);
5316   if (mapped_address == MAP_FAILED) {
5317     return NULL;
5318   }
5319   return mapped_address;
5320 }
5321 
5322 
5323 // Remap a block of memory.
5324 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5325                           char *addr, size_t bytes, bool read_only,
5326                           bool allow_exec) {
5327   // same as map_memory() on this OS
5328   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5329                         allow_exec);
5330 }
5331 
5332 
5333 // Unmap a block of memory.
5334 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5335   return munmap(addr, bytes) == 0;
5336 }
5337 
5338 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5339 
5340 static clockid_t thread_cpu_clockid(Thread* thread) {
5341   pthread_t tid = thread->osthread()->pthread_id();
5342   clockid_t clockid;
5343 
5344   // Get thread clockid
5345   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5346   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5347   return clockid;
5348 }
5349 
5350 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5351 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5352 // of a thread.
5353 //
5354 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5355 // the fast estimate available on the platform.
5356 
5357 jlong os::current_thread_cpu_time() {
5358   if (os::Linux::supports_fast_thread_cpu_time()) {
5359     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5360   } else {
5361     // return user + sys since the cost is the same
5362     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5363   }
5364 }
5365 
5366 jlong os::thread_cpu_time(Thread* thread) {
5367   // consistent with what current_thread_cpu_time() returns
5368   if (os::Linux::supports_fast_thread_cpu_time()) {
5369     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5370   } else {
5371     return slow_thread_cpu_time(thread, true /* user + sys */);
5372   }
5373 }
5374 
5375 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5376   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5377     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5378   } else {
5379     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5380   }
5381 }
5382 
5383 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5384   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5385     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5386   } else {
5387     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5388   }
5389 }
5390 
5391 //  -1 on error.
5392 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5393   pid_t  tid = thread->osthread()->thread_id();
5394   char *s;
5395   char stat[2048];
5396   int statlen;
5397   char proc_name[64];
5398   int count;
5399   long sys_time, user_time;
5400   char cdummy;
5401   int idummy;
5402   long ldummy;
5403   FILE *fp;
5404 
5405   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5406   fp = fopen(proc_name, "r");
5407   if (fp == NULL) return -1;
5408   statlen = fread(stat, 1, 2047, fp);
5409   stat[statlen] = '\0';
5410   fclose(fp);
5411 
5412   // Skip pid and the command string. Note that we could be dealing with
5413   // weird command names, e.g. user could decide to rename java launcher
5414   // to "java 1.4.2 :)", then the stat file would look like
5415   //                1234 (java 1.4.2 :)) R ... ...
5416   // We don't really need to know the command string, just find the last
5417   // occurrence of ")" and then start parsing from there. See bug 4726580.
5418   s = strrchr(stat, ')');
5419   if (s == NULL) return -1;
5420 
5421   // Skip blank chars
5422   do { s++; } while (s && isspace(*s));
5423 
5424   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5425                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5426                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5427                  &user_time, &sys_time);
5428   if (count != 13) return -1;
5429   if (user_sys_cpu_time) {
5430     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5431   } else {
5432     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5433   }
5434 }
5435 
5436 void os::current_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 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5444   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5445   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5446   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5447   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5448 }
5449 
5450 bool os::is_thread_cpu_time_supported() {
5451   return true;
5452 }
5453 
5454 // System loadavg support.  Returns -1 if load average cannot be obtained.
5455 // Linux doesn't yet have a (official) notion of processor sets,
5456 // so just return the system wide load average.
5457 int os::loadavg(double loadavg[], int nelem) {
5458   return ::getloadavg(loadavg, nelem);
5459 }
5460 
5461 void os::pause() {
5462   char filename[MAX_PATH];
5463   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5464     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5465   } else {
5466     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5467   }
5468 
5469   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5470   if (fd != -1) {
5471     struct stat buf;
5472     ::close(fd);
5473     while (::stat(filename, &buf) == 0) {
5474       (void)::poll(NULL, 0, 100);
5475     }
5476   } else {
5477     jio_fprintf(stderr,
5478                 "Could not open pause file '%s', continuing immediately.\n", filename);
5479   }
5480 }
5481 
5482 
5483 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5484 // comment paragraphs are worth repeating here:
5485 //
5486 // Assumption:
5487 //    Only one parker can exist on an event, which is why we allocate
5488 //    them per-thread. Multiple unparkers can coexist.
5489 //
5490 // _Event serves as a restricted-range semaphore.
5491 //   -1 : thread is blocked, i.e. there is a waiter
5492 //    0 : neutral: thread is running or ready,
5493 //        could have been signaled after a wait started
5494 //    1 : signaled - thread is running or ready
5495 //
5496 
5497 // utility to compute the abstime argument to timedwait:
5498 // millis is the relative timeout time
5499 // abstime will be the absolute timeout time
5500 // TODO: replace compute_abstime() with unpackTime()
5501 
5502 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5503   if (millis < 0)  millis = 0;
5504 
5505   jlong seconds = millis / 1000;
5506   millis %= 1000;
5507   if (seconds > 50000000) { // see man cond_timedwait(3T)
5508     seconds = 50000000;
5509   }
5510 
5511   if (os::supports_monotonic_clock()) {
5512     struct timespec now;
5513     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5514     assert_status(status == 0, status, "clock_gettime");
5515     abstime->tv_sec = now.tv_sec  + seconds;
5516     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5517     if (nanos >= NANOSECS_PER_SEC) {
5518       abstime->tv_sec += 1;
5519       nanos -= NANOSECS_PER_SEC;
5520     }
5521     abstime->tv_nsec = nanos;
5522   } else {
5523     struct timeval now;
5524     int status = gettimeofday(&now, NULL);
5525     assert(status == 0, "gettimeofday");
5526     abstime->tv_sec = now.tv_sec  + seconds;
5527     long usec = now.tv_usec + millis * 1000;
5528     if (usec >= 1000000) {
5529       abstime->tv_sec += 1;
5530       usec -= 1000000;
5531     }
5532     abstime->tv_nsec = usec * 1000;
5533   }
5534   return abstime;
5535 }
5536 
5537 void os::PlatformEvent::park() {       // AKA "down()"
5538   // Transitions for _Event:
5539   //   -1 => -1 : illegal
5540   //    1 =>  0 : pass - return immediately
5541   //    0 => -1 : block; then set _Event to 0 before returning
5542 
5543   // Invariant: Only the thread associated with the Event/PlatformEvent
5544   // may call park().
5545   // TODO: assert that _Assoc != NULL or _Assoc == Self
5546   assert(_nParked == 0, "invariant");
5547 
5548   int v;
5549   for (;;) {
5550     v = _Event;
5551     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5552   }
5553   guarantee(v >= 0, "invariant");
5554   if (v == 0) {
5555     // Do this the hard way by blocking ...
5556     int status = pthread_mutex_lock(_mutex);
5557     assert_status(status == 0, status, "mutex_lock");
5558     guarantee(_nParked == 0, "invariant");
5559     ++_nParked;
5560     while (_Event < 0) {
5561       status = pthread_cond_wait(_cond, _mutex);
5562       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5563       // Treat this the same as if the wait was interrupted
5564       if (status == ETIME) { status = EINTR; }
5565       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5566     }
5567     --_nParked;
5568 
5569     _Event = 0;
5570     status = pthread_mutex_unlock(_mutex);
5571     assert_status(status == 0, status, "mutex_unlock");
5572     // Paranoia to ensure our locked and lock-free paths interact
5573     // correctly with each other.
5574     OrderAccess::fence();
5575   }
5576   guarantee(_Event >= 0, "invariant");
5577 }
5578 
5579 int os::PlatformEvent::park(jlong millis) {
5580   // Transitions for _Event:
5581   //   -1 => -1 : illegal
5582   //    1 =>  0 : pass - return immediately
5583   //    0 => -1 : block; then set _Event to 0 before returning
5584 
5585   guarantee(_nParked == 0, "invariant");
5586 
5587   int v;
5588   for (;;) {
5589     v = _Event;
5590     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5591   }
5592   guarantee(v >= 0, "invariant");
5593   if (v != 0) return OS_OK;
5594 
5595   // We do this the hard way, by blocking the thread.
5596   // Consider enforcing a minimum timeout value.
5597   struct timespec abst;
5598   compute_abstime(&abst, millis);
5599 
5600   int ret = OS_TIMEOUT;
5601   int status = pthread_mutex_lock(_mutex);
5602   assert_status(status == 0, status, "mutex_lock");
5603   guarantee(_nParked == 0, "invariant");
5604   ++_nParked;
5605 
5606   // Object.wait(timo) will return because of
5607   // (a) notification
5608   // (b) timeout
5609   // (c) thread.interrupt
5610   //
5611   // Thread.interrupt and object.notify{All} both call Event::set.
5612   // That is, we treat thread.interrupt as a special case of notification.
5613   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5614   // We assume all ETIME returns are valid.
5615   //
5616   // TODO: properly differentiate simultaneous notify+interrupt.
5617   // In that case, we should propagate the notify to another waiter.
5618 
5619   while (_Event < 0) {
5620     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5621     assert_status(status == 0 || status == EINTR ||
5622                   status == ETIME || status == ETIMEDOUT,
5623                   status, "cond_timedwait");
5624     if (!FilterSpuriousWakeups) break;                 // previous semantics
5625     if (status == ETIME || status == ETIMEDOUT) break;
5626     // We consume and ignore EINTR and spurious wakeups.
5627   }
5628   --_nParked;
5629   if (_Event >= 0) {
5630     ret = OS_OK;
5631   }
5632   _Event = 0;
5633   status = pthread_mutex_unlock(_mutex);
5634   assert_status(status == 0, status, "mutex_unlock");
5635   assert(_nParked == 0, "invariant");
5636   // Paranoia to ensure our locked and lock-free paths interact
5637   // correctly with each other.
5638   OrderAccess::fence();
5639   return ret;
5640 }
5641 
5642 void os::PlatformEvent::unpark() {
5643   // Transitions for _Event:
5644   //    0 => 1 : just return
5645   //    1 => 1 : just return
5646   //   -1 => either 0 or 1; must signal target thread
5647   //         That is, we can safely transition _Event from -1 to either
5648   //         0 or 1.
5649   // See also: "Semaphores in Plan 9" by Mullender & Cox
5650   //
5651   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5652   // that it will take two back-to-back park() calls for the owning
5653   // thread to block. This has the benefit of forcing a spurious return
5654   // from the first park() call after an unpark() call which will help
5655   // shake out uses of park() and unpark() without condition variables.
5656 
5657   if (Atomic::xchg(1, &_Event) >= 0) return;
5658 
5659   // Wait for the thread associated with the event to vacate
5660   int status = pthread_mutex_lock(_mutex);
5661   assert_status(status == 0, status, "mutex_lock");
5662   int AnyWaiters = _nParked;
5663   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5664   status = pthread_mutex_unlock(_mutex);
5665   assert_status(status == 0, status, "mutex_unlock");
5666   if (AnyWaiters != 0) {
5667     // Note that we signal() *after* dropping the lock for "immortal" Events.
5668     // This is safe and avoids a common class of  futile wakeups.  In rare
5669     // circumstances this can cause a thread to return prematurely from
5670     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5671     // will simply re-test the condition and re-park itself.
5672     // This provides particular benefit if the underlying platform does not
5673     // provide wait morphing.
5674     status = pthread_cond_signal(_cond);
5675     assert_status(status == 0, status, "cond_signal");
5676   }
5677 }
5678 
5679 
5680 // JSR166
5681 // -------------------------------------------------------
5682 
5683 // The solaris and linux implementations of park/unpark are fairly
5684 // conservative for now, but can be improved. They currently use a
5685 // mutex/condvar pair, plus a a count.
5686 // Park decrements count if > 0, else does a condvar wait.  Unpark
5687 // sets count to 1 and signals condvar.  Only one thread ever waits
5688 // on the condvar. Contention seen when trying to park implies that someone
5689 // is unparking you, so don't wait. And spurious returns are fine, so there
5690 // is no need to track notifications.
5691 
5692 // This code is common to linux and solaris and will be moved to a
5693 // common place in dolphin.
5694 //
5695 // The passed in time value is either a relative time in nanoseconds
5696 // or an absolute time in milliseconds. Either way it has to be unpacked
5697 // into suitable seconds and nanoseconds components and stored in the
5698 // given timespec structure.
5699 // Given time is a 64-bit value and the time_t used in the timespec is only
5700 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5701 // overflow if times way in the future are given. Further on Solaris versions
5702 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5703 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5704 // As it will be 28 years before "now + 100000000" will overflow we can
5705 // ignore overflow and just impose a hard-limit on seconds using the value
5706 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5707 // years from "now".
5708 
5709 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5710   assert(time > 0, "convertTime");
5711   time_t max_secs = 0;
5712 
5713   if (!os::supports_monotonic_clock() || isAbsolute) {
5714     struct timeval now;
5715     int status = gettimeofday(&now, NULL);
5716     assert(status == 0, "gettimeofday");
5717 
5718     max_secs = now.tv_sec + MAX_SECS;
5719 
5720     if (isAbsolute) {
5721       jlong secs = time / 1000;
5722       if (secs > max_secs) {
5723         absTime->tv_sec = max_secs;
5724       } else {
5725         absTime->tv_sec = secs;
5726       }
5727       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5728     } else {
5729       jlong secs = time / NANOSECS_PER_SEC;
5730       if (secs >= MAX_SECS) {
5731         absTime->tv_sec = max_secs;
5732         absTime->tv_nsec = 0;
5733       } else {
5734         absTime->tv_sec = now.tv_sec + secs;
5735         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5736         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5737           absTime->tv_nsec -= NANOSECS_PER_SEC;
5738           ++absTime->tv_sec; // note: this must be <= max_secs
5739         }
5740       }
5741     }
5742   } else {
5743     // must be relative using monotonic clock
5744     struct timespec now;
5745     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5746     assert_status(status == 0, status, "clock_gettime");
5747     max_secs = now.tv_sec + MAX_SECS;
5748     jlong secs = time / NANOSECS_PER_SEC;
5749     if (secs >= MAX_SECS) {
5750       absTime->tv_sec = max_secs;
5751       absTime->tv_nsec = 0;
5752     } else {
5753       absTime->tv_sec = now.tv_sec + secs;
5754       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5755       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5756         absTime->tv_nsec -= NANOSECS_PER_SEC;
5757         ++absTime->tv_sec; // note: this must be <= max_secs
5758       }
5759     }
5760   }
5761   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5762   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5763   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5764   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5765 }
5766 
5767 void Parker::park(bool isAbsolute, jlong time) {
5768   // Ideally we'd do something useful while spinning, such
5769   // as calling unpackTime().
5770 
5771   // Optional fast-path check:
5772   // Return immediately if a permit is available.
5773   // We depend on Atomic::xchg() having full barrier semantics
5774   // since we are doing a lock-free update to _counter.
5775   if (Atomic::xchg(0, &_counter) > 0) return;
5776 
5777   Thread* thread = Thread::current();
5778   assert(thread->is_Java_thread(), "Must be JavaThread");
5779   JavaThread *jt = (JavaThread *)thread;
5780 
5781   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5782   // Check interrupt before trying to wait
5783   if (Thread::is_interrupted(thread, false)) {
5784     return;
5785   }
5786 
5787   // Next, demultiplex/decode time arguments
5788   timespec absTime;
5789   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5790     return;
5791   }
5792   if (time > 0) {
5793     unpackTime(&absTime, isAbsolute, time);
5794   }
5795 
5796 
5797   // Enter safepoint region
5798   // Beware of deadlocks such as 6317397.
5799   // The per-thread Parker:: mutex is a classic leaf-lock.
5800   // In particular a thread must never block on the Threads_lock while
5801   // holding the Parker:: mutex.  If safepoints are pending both the
5802   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5803   ThreadBlockInVM tbivm(jt);
5804 
5805   // Don't wait if cannot get lock since interference arises from
5806   // unblocking.  Also. check interrupt before trying wait
5807   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5808     return;
5809   }
5810 
5811   int status;
5812   if (_counter > 0)  { // no wait needed
5813     _counter = 0;
5814     status = pthread_mutex_unlock(_mutex);
5815     assert_status(status == 0, status, "invariant");
5816     // Paranoia to ensure our locked and lock-free paths interact
5817     // correctly with each other and Java-level accesses.
5818     OrderAccess::fence();
5819     return;
5820   }
5821 
5822 #ifdef ASSERT
5823   // Don't catch signals while blocked; let the running threads have the signals.
5824   // (This allows a debugger to break into the running thread.)
5825   sigset_t oldsigs;
5826   sigemptyset(&oldsigs);
5827   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5828   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5829 #endif
5830 
5831   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5832   jt->set_suspend_equivalent();
5833   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5834 
5835   assert(_cur_index == -1, "invariant");
5836   if (time == 0) {
5837     _cur_index = REL_INDEX; // arbitrary choice when not timed
5838     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5839   } else {
5840     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5841     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5842   }
5843   _cur_index = -1;
5844   assert_status(status == 0 || status == EINTR ||
5845                 status == ETIME || status == ETIMEDOUT,
5846                 status, "cond_timedwait");
5847 
5848 #ifdef ASSERT
5849   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5850 #endif
5851 
5852   _counter = 0;
5853   status = pthread_mutex_unlock(_mutex);
5854   assert_status(status == 0, status, "invariant");
5855   // Paranoia to ensure our locked and lock-free paths interact
5856   // correctly with each other and Java-level accesses.
5857   OrderAccess::fence();
5858 
5859   // If externally suspended while waiting, re-suspend
5860   if (jt->handle_special_suspend_equivalent_condition()) {
5861     jt->java_suspend_self();
5862   }
5863 }
5864 
5865 void Parker::unpark() {
5866   int status = pthread_mutex_lock(_mutex);
5867   assert_status(status == 0, status, "invariant");
5868   const int s = _counter;
5869   _counter = 1;
5870   // must capture correct index before unlocking
5871   int index = _cur_index;
5872   status = pthread_mutex_unlock(_mutex);
5873   assert_status(status == 0, status, "invariant");
5874   if (s < 1 && index != -1) {
5875     // thread is definitely parked
5876     status = pthread_cond_signal(&_cond[index]);
5877     assert_status(status == 0, status, "invariant");
5878   }
5879 }
5880 
5881 
5882 extern char** environ;
5883 
5884 // Run the specified command in a separate process. Return its exit value,
5885 // or -1 on failure (e.g. can't fork a new process).
5886 // Unlike system(), this function can be called from signal handler. It
5887 // doesn't block SIGINT et al.
5888 int os::fork_and_exec(char* cmd) {
5889   const char * argv[4] = {"sh", "-c", cmd, NULL};
5890 
5891   pid_t pid = fork();
5892 
5893   if (pid < 0) {
5894     // fork failed
5895     return -1;
5896 
5897   } else if (pid == 0) {
5898     // child process
5899 
5900     execve("/bin/sh", (char* const*)argv, environ);
5901 
5902     // execve failed
5903     _exit(-1);
5904 
5905   } else  {
5906     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5907     // care about the actual exit code, for now.
5908 
5909     int status;
5910 
5911     // Wait for the child process to exit.  This returns immediately if
5912     // the child has already exited. */
5913     while (waitpid(pid, &status, 0) < 0) {
5914       switch (errno) {
5915       case ECHILD: return 0;
5916       case EINTR: break;
5917       default: return -1;
5918       }
5919     }
5920 
5921     if (WIFEXITED(status)) {
5922       // The child exited normally; get its exit code.
5923       return WEXITSTATUS(status);
5924     } else if (WIFSIGNALED(status)) {
5925       // The child exited because of a signal
5926       // The best value to return is 0x80 + signal number,
5927       // because that is what all Unix shells do, and because
5928       // it allows callers to distinguish between process exit and
5929       // process death by signal.
5930       return 0x80 + WTERMSIG(status);
5931     } else {
5932       // Unknown exit code; pass it through
5933       return status;
5934     }
5935   }
5936 }
5937 
5938 // is_headless_jre()
5939 //
5940 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5941 // in order to report if we are running in a headless jre
5942 //
5943 // Since JDK8 xawt/libmawt.so was moved into the same directory
5944 // as libawt.so, and renamed libawt_xawt.so
5945 //
5946 bool os::is_headless_jre() {
5947   struct stat statbuf;
5948   char buf[MAXPATHLEN];
5949   char libmawtpath[MAXPATHLEN];
5950   const char *xawtstr  = "/xawt/libmawt.so";
5951   const char *new_xawtstr = "/libawt_xawt.so";
5952   char *p;
5953 
5954   // Get path to libjvm.so
5955   os::jvm_path(buf, sizeof(buf));
5956 
5957   // Get rid of libjvm.so
5958   p = strrchr(buf, '/');
5959   if (p == NULL) {
5960     return false;
5961   } else {
5962     *p = '\0';
5963   }
5964 
5965   // Get rid of client or server
5966   p = strrchr(buf, '/');
5967   if (p == NULL) {
5968     return false;
5969   } else {
5970     *p = '\0';
5971   }
5972 
5973   // check xawt/libmawt.so
5974   strcpy(libmawtpath, buf);
5975   strcat(libmawtpath, xawtstr);
5976   if (::stat(libmawtpath, &statbuf) == 0) return false;
5977 
5978   // check libawt_xawt.so
5979   strcpy(libmawtpath, buf);
5980   strcat(libmawtpath, new_xawtstr);
5981   if (::stat(libmawtpath, &statbuf) == 0) return false;
5982 
5983   return true;
5984 }
5985 
5986 // Get the default path to the core file
5987 // Returns the length of the string
5988 int os::get_core_path(char* buffer, size_t bufferSize) {
5989   /*
5990    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5991    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5992    */
5993   const int core_pattern_len = 129;
5994   char core_pattern[core_pattern_len] = {0};
5995 
5996   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5997   if (core_pattern_file == -1) {
5998     return -1;
5999   }
6000 
6001   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6002   ::close(core_pattern_file);
6003   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
6004     return -1;
6005   }
6006   if (core_pattern[ret-1] == '\n') {
6007     core_pattern[ret-1] = '\0';
6008   } else {
6009     core_pattern[ret] = '\0';
6010   }
6011 
6012   char *pid_pos = strstr(core_pattern, "%p");
6013   int written;
6014 
6015   if (core_pattern[0] == '/') {
6016     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6017   } else {
6018     char cwd[PATH_MAX];
6019 
6020     const char* p = get_current_directory(cwd, PATH_MAX);
6021     if (p == NULL) {
6022       return -1;
6023     }
6024 
6025     if (core_pattern[0] == '|') {
6026       written = jio_snprintf(buffer, bufferSize,
6027                              "\"%s\" (or dumping to %s/core.%d)",
6028                              &core_pattern[1], p, current_process_id());
6029     } else {
6030       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6031     }
6032   }
6033 
6034   if (written < 0) {
6035     return -1;
6036   }
6037 
6038   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6039     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6040 
6041     if (core_uses_pid_file != -1) {
6042       char core_uses_pid = 0;
6043       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6044       ::close(core_uses_pid_file);
6045 
6046       if (core_uses_pid == '1') {
6047         jio_snprintf(buffer + written, bufferSize - written,
6048                                           ".%d", current_process_id());
6049       }
6050     }
6051   }
6052 
6053   return strlen(buffer);
6054 }
6055 
6056 bool os::start_debugging(char *buf, int buflen) {
6057   int len = (int)strlen(buf);
6058   char *p = &buf[len];
6059 
6060   jio_snprintf(p, buflen-len,
6061                "\n\n"
6062                "Do you want to debug the problem?\n\n"
6063                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6064                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6065                "Otherwise, press RETURN to abort...",
6066                os::current_process_id(), os::current_process_id(),
6067                os::current_thread_id(), os::current_thread_id());
6068 
6069   bool yes = os::message_box("Unexpected Error", buf);
6070 
6071   if (yes) {
6072     // yes, user asked VM to launch debugger
6073     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6074                  os::current_process_id(), os::current_process_id());
6075 
6076     os::fork_and_exec(buf);
6077     yes = false;
6078   }
6079   return yes;
6080 }
6081 
6082 static inline struct timespec get_mtime(const char* filename) {
6083   struct stat st;
6084   int ret = os::stat(filename, &st);
6085   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6086   return st.st_mtim;
6087 }
6088 
6089 int os::compare_file_modified_times(const char* file1, const char* file2) {
6090   struct timespec filetime1 = get_mtime(file1);
6091   struct timespec filetime2 = get_mtime(file2);
6092   int diff = filetime1.tv_sec - filetime2.tv_sec;
6093   if (diff == 0) {
6094     return filetime1.tv_nsec - filetime2.tv_nsec;
6095   }
6096   return diff;
6097 }
6098 
6099 /////////////// Unit tests ///////////////
6100 
6101 #ifndef PRODUCT
6102 
6103 #define test_log(...)              \
6104   do {                             \
6105     if (VerboseInternalVMTests) {  \
6106       tty->print_cr(__VA_ARGS__);  \
6107       tty->flush();                \
6108     }                              \
6109   } while (false)
6110 
6111 class TestReserveMemorySpecial : AllStatic {
6112  public:
6113   static void small_page_write(void* addr, size_t size) {
6114     size_t page_size = os::vm_page_size();
6115 
6116     char* end = (char*)addr + size;
6117     for (char* p = (char*)addr; p < end; p += page_size) {
6118       *p = 1;
6119     }
6120   }
6121 
6122   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6123     if (!UseHugeTLBFS) {
6124       return;
6125     }
6126 
6127     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6128 
6129     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6130 
6131     if (addr != NULL) {
6132       small_page_write(addr, size);
6133 
6134       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6135     }
6136   }
6137 
6138   static void test_reserve_memory_special_huge_tlbfs_only() {
6139     if (!UseHugeTLBFS) {
6140       return;
6141     }
6142 
6143     size_t lp = os::large_page_size();
6144 
6145     for (size_t size = lp; size <= lp * 10; size += lp) {
6146       test_reserve_memory_special_huge_tlbfs_only(size);
6147     }
6148   }
6149 
6150   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6151     size_t lp = os::large_page_size();
6152     size_t ag = os::vm_allocation_granularity();
6153 
6154     // sizes to test
6155     const size_t sizes[] = {
6156       lp, lp + ag, lp + lp / 2, lp * 2,
6157       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6158       lp * 10, lp * 10 + lp / 2
6159     };
6160     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6161 
6162     // For each size/alignment combination, we test three scenarios:
6163     // 1) with req_addr == NULL
6164     // 2) with a non-null req_addr at which we expect to successfully allocate
6165     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6166     //    expect the allocation to either fail or to ignore req_addr
6167 
6168     // Pre-allocate two areas; they shall be as large as the largest allocation
6169     //  and aligned to the largest alignment we will be testing.
6170     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6171     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6172       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6173       -1, 0);
6174     assert(mapping1 != MAP_FAILED, "should work");
6175 
6176     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6177       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6178       -1, 0);
6179     assert(mapping2 != MAP_FAILED, "should work");
6180 
6181     // Unmap the first mapping, but leave the second mapping intact: the first
6182     // mapping will serve as a value for a "good" req_addr (case 2). The second
6183     // mapping, still intact, as "bad" req_addr (case 3).
6184     ::munmap(mapping1, mapping_size);
6185 
6186     // Case 1
6187     test_log("%s, req_addr NULL:", __FUNCTION__);
6188     test_log("size            align           result");
6189 
6190     for (int i = 0; i < num_sizes; i++) {
6191       const size_t size = sizes[i];
6192       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6193         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6194         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6195                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6196         if (p != NULL) {
6197           assert(is_ptr_aligned(p, alignment), "must be");
6198           small_page_write(p, size);
6199           os::Linux::release_memory_special_huge_tlbfs(p, size);
6200         }
6201       }
6202     }
6203 
6204     // Case 2
6205     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6206     test_log("size            align           req_addr         result");
6207 
6208     for (int i = 0; i < num_sizes; i++) {
6209       const size_t size = sizes[i];
6210       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6211         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6212         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6213         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6214                  size, alignment, p2i(req_addr), p2i(p),
6215                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6216         if (p != NULL) {
6217           assert(p == req_addr, "must be");
6218           small_page_write(p, size);
6219           os::Linux::release_memory_special_huge_tlbfs(p, size);
6220         }
6221       }
6222     }
6223 
6224     // Case 3
6225     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6226     test_log("size            align           req_addr         result");
6227 
6228     for (int i = 0; i < num_sizes; i++) {
6229       const size_t size = sizes[i];
6230       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6231         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6232         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6233         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6234                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6235         // as the area around req_addr contains already existing mappings, the API should always
6236         // return NULL (as per contract, it cannot return another address)
6237         assert(p == NULL, "must be");
6238       }
6239     }
6240 
6241     ::munmap(mapping2, mapping_size);
6242 
6243   }
6244 
6245   static void test_reserve_memory_special_huge_tlbfs() {
6246     if (!UseHugeTLBFS) {
6247       return;
6248     }
6249 
6250     test_reserve_memory_special_huge_tlbfs_only();
6251     test_reserve_memory_special_huge_tlbfs_mixed();
6252   }
6253 
6254   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6255     if (!UseSHM) {
6256       return;
6257     }
6258 
6259     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6260 
6261     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6262 
6263     if (addr != NULL) {
6264       assert(is_ptr_aligned(addr, alignment), "Check");
6265       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6266 
6267       small_page_write(addr, size);
6268 
6269       os::Linux::release_memory_special_shm(addr, size);
6270     }
6271   }
6272 
6273   static void test_reserve_memory_special_shm() {
6274     size_t lp = os::large_page_size();
6275     size_t ag = os::vm_allocation_granularity();
6276 
6277     for (size_t size = ag; size < lp * 3; size += ag) {
6278       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6279         test_reserve_memory_special_shm(size, alignment);
6280       }
6281     }
6282   }
6283 
6284   static void test() {
6285     test_reserve_memory_special_huge_tlbfs();
6286     test_reserve_memory_special_shm();
6287   }
6288 };
6289 
6290 void TestReserveMemorySpecial_test() {
6291   TestReserveMemorySpecial::test();
6292 }
6293 
6294 #endif
--- EOF ---