1 /*
   2  * Copyright (c) 1999, 2015, 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 "memory/allocation.inline.hpp"
  36 #include "memory/filemap.hpp"
  37 #include "mutex_linux.inline.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.inline.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 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
 110 
 111 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 112 // getrusage() is prepared to handle the associated failure.
 113 #ifndef RUSAGE_THREAD
 114   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 115 #endif
 116 
 117 #define MAX_PATH    (2 * K)
 118 
 119 #define MAX_SECS 100000000
 120 
 121 // for timer info max values which include all bits
 122 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 123 
 124 #define LARGEPAGES_BIT (1 << 6)
 125 ////////////////////////////////////////////////////////////////////////////////
 126 // global variables
 127 julong os::Linux::_physical_memory = 0;
 128 
 129 address   os::Linux::_initial_thread_stack_bottom = NULL;
 130 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 131 
 132 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 133 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 134 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 135 Mutex* os::Linux::_createThread_lock = NULL;
 136 pthread_t os::Linux::_main_thread;
 137 int os::Linux::_page_size = -1;
 138 const int os::Linux::_vm_default_page_size = (8 * K);
 139 bool os::Linux::_supports_fast_thread_cpu_time = false;
 140 const char * os::Linux::_glibc_version = NULL;
 141 const char * os::Linux::_libpthread_version = NULL;
 142 pthread_condattr_t os::Linux::_condattr[1];
 143 
 144 static jlong initial_time_count=0;
 145 
 146 static int clock_tics_per_sec = 100;
 147 
 148 // For diagnostics to print a message once. see run_periodic_checks
 149 static sigset_t check_signal_done;
 150 static bool check_signals = true;
 151 
 152 // Signal number used to suspend/resume a thread
 153 
 154 // do not use any signal number less than SIGSEGV, see 4355769
 155 static int SR_signum = SIGUSR2;
 156 sigset_t SR_sigset;
 157 
 158 // Declarations
 159 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 160 
 161 // utility functions
 162 
 163 static int SR_initialize();
 164 
 165 julong os::available_memory() {
 166   return Linux::available_memory();
 167 }
 168 
 169 julong os::Linux::available_memory() {
 170   // values in struct sysinfo are "unsigned long"
 171   struct sysinfo si;
 172   sysinfo(&si);
 173 
 174   return (julong)si.freeram * si.mem_unit;
 175 }
 176 
 177 julong os::physical_memory() {
 178   return Linux::physical_memory();
 179 }
 180 
 181 // Return true if user is running as root.
 182 
 183 bool os::have_special_privileges() {
 184   static bool init = false;
 185   static bool privileges = false;
 186   if (!init) {
 187     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 188     init = true;
 189   }
 190   return privileges;
 191 }
 192 
 193 
 194 #ifndef SYS_gettid
 195 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 196   #ifdef __ia64__
 197     #define SYS_gettid 1105
 198   #else
 199     #ifdef __i386__
 200       #define SYS_gettid 224
 201     #else
 202       #ifdef __amd64__
 203         #define SYS_gettid 186
 204       #else
 205         #ifdef __sparc__
 206           #define SYS_gettid 143
 207         #else
 208           #error define gettid for the arch
 209         #endif
 210       #endif
 211     #endif
 212   #endif
 213 #endif
 214 
 215 // Cpu architecture string
 216 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 217 
 218 
 219 // pid_t gettid()
 220 //
 221 // Returns the kernel thread id of the currently running thread. Kernel
 222 // thread id is used to access /proc.
 223 pid_t os::Linux::gettid() {
 224   int rslt = syscall(SYS_gettid);
 225   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 226   return (pid_t)rslt;
 227 }
 228 
 229 // Most versions of linux have a bug where the number of processors are
 230 // determined by looking at the /proc file system.  In a chroot environment,
 231 // the system call returns 1.  This causes the VM to act as if it is
 232 // a single processor and elide locking (see is_MP() call).
 233 static bool unsafe_chroot_detected = false;
 234 static const char *unstable_chroot_error = "/proc file system not found.\n"
 235                      "Java may be unstable running multithreaded in a chroot "
 236                      "environment on Linux when /proc filesystem is not mounted.";
 237 
 238 void os::Linux::initialize_system_info() {
 239   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 240   if (processor_count() == 1) {
 241     pid_t pid = os::Linux::gettid();
 242     char fname[32];
 243     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 244     FILE *fp = fopen(fname, "r");
 245     if (fp == NULL) {
 246       unsafe_chroot_detected = true;
 247     } else {
 248       fclose(fp);
 249     }
 250   }
 251   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 252   assert(processor_count() > 0, "linux error");
 253 }
 254 
 255 void os::init_system_properties_values() {
 256   // The next steps are taken in the product version:
 257   //
 258   // Obtain the JAVA_HOME value from the location of libjvm.so.
 259   // This library should be located at:
 260   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 261   //
 262   // If "/jre/lib/" appears at the right place in the path, then we
 263   // assume libjvm.so is installed in a JDK and we use this path.
 264   //
 265   // Otherwise exit with message: "Could not create the Java virtual machine."
 266   //
 267   // The following extra steps are taken in the debugging version:
 268   //
 269   // If "/jre/lib/" does NOT appear at the right place in the path
 270   // instead of exit check for $JAVA_HOME environment variable.
 271   //
 272   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 273   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 274   // it looks like libjvm.so is installed there
 275   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 276   //
 277   // Otherwise exit.
 278   //
 279   // Important note: if the location of libjvm.so changes this
 280   // code needs to be changed accordingly.
 281 
 282   // See ld(1):
 283   //      The linker uses the following search paths to locate required
 284   //      shared libraries:
 285   //        1: ...
 286   //        ...
 287   //        7: The default directories, normally /lib and /usr/lib.
 288 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 289   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 290 #else
 291   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 292 #endif
 293 
 294 // Base path of extensions installed on the system.
 295 #define SYS_EXT_DIR     "/usr/java/packages"
 296 #define EXTENSIONS_DIR  "/lib/ext"
 297 
 298   // Buffer that fits several sprintfs.
 299   // Note that the space for the colon and the trailing null are provided
 300   // by the nulls included by the sizeof operator.
 301   const size_t bufsize =
 302     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 303          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 304   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 305 
 306   // sysclasspath, java_home, dll_dir
 307   {
 308     char *pslash;
 309     os::jvm_path(buf, bufsize);
 310 
 311     // Found the full path to libjvm.so.
 312     // Now cut the path to <java_home>/jre if we can.
 313     pslash = strrchr(buf, '/');
 314     if (pslash != NULL) {
 315       *pslash = '\0';            // Get rid of /libjvm.so.
 316     }
 317     pslash = strrchr(buf, '/');
 318     if (pslash != NULL) {
 319       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 320     }
 321     Arguments::set_dll_dir(buf);
 322 
 323     if (pslash != NULL) {
 324       pslash = strrchr(buf, '/');
 325       if (pslash != NULL) {
 326         *pslash = '\0';          // Get rid of /<arch>.
 327         pslash = strrchr(buf, '/');
 328         if (pslash != NULL) {
 329           *pslash = '\0';        // Get rid of /lib.
 330         }
 331       }
 332     }
 333     Arguments::set_java_home(buf);
 334     set_boot_path('/', ':');
 335   }
 336 
 337   // Where to look for native libraries.
 338   //
 339   // Note: Due to a legacy implementation, most of the library path
 340   // is set in the launcher. This was to accomodate linking restrictions
 341   // on legacy Linux implementations (which are no longer supported).
 342   // Eventually, all the library path setting will be done here.
 343   //
 344   // However, to prevent the proliferation of improperly built native
 345   // libraries, the new path component /usr/java/packages is added here.
 346   // Eventually, all the library path setting will be done here.
 347   {
 348     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 349     // should always exist (until the legacy problem cited above is
 350     // addressed).
 351     const char *v = ::getenv("LD_LIBRARY_PATH");
 352     const char *v_colon = ":";
 353     if (v == NULL) { v = ""; v_colon = ""; }
 354     // That's +1 for the colon and +1 for the trailing '\0'.
 355     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 356                                                      strlen(v) + 1 +
 357                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 358                                                      mtInternal);
 359     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 360     Arguments::set_library_path(ld_library_path);
 361     FREE_C_HEAP_ARRAY(char, ld_library_path);
 362   }
 363 
 364   // Extensions directories.
 365   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 366   Arguments::set_ext_dirs(buf);
 367 
 368   FREE_C_HEAP_ARRAY(char, buf);
 369 
 370 #undef DEFAULT_LIBPATH
 371 #undef SYS_EXT_DIR
 372 #undef EXTENSIONS_DIR
 373 }
 374 
 375 ////////////////////////////////////////////////////////////////////////////////
 376 // breakpoint support
 377 
 378 void os::breakpoint() {
 379   BREAKPOINT;
 380 }
 381 
 382 extern "C" void breakpoint() {
 383   // use debugger to set breakpoint here
 384 }
 385 
 386 ////////////////////////////////////////////////////////////////////////////////
 387 // signal support
 388 
 389 debug_only(static bool signal_sets_initialized = false);
 390 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 391 
 392 bool os::Linux::is_sig_ignored(int sig) {
 393   struct sigaction oact;
 394   sigaction(sig, (struct sigaction*)NULL, &oact);
 395   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 396                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 397   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 398     return true;
 399   } else {
 400     return false;
 401   }
 402 }
 403 
 404 void os::Linux::signal_sets_init() {
 405   // Should also have an assertion stating we are still single-threaded.
 406   assert(!signal_sets_initialized, "Already initialized");
 407   // Fill in signals that are necessarily unblocked for all threads in
 408   // the VM. Currently, we unblock the following signals:
 409   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 410   //                         by -Xrs (=ReduceSignalUsage));
 411   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 412   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 413   // the dispositions or masks wrt these signals.
 414   // Programs embedding the VM that want to use the above signals for their
 415   // own purposes must, at this time, use the "-Xrs" option to prevent
 416   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 417   // (See bug 4345157, and other related bugs).
 418   // In reality, though, unblocking these signals is really a nop, since
 419   // these signals are not blocked by default.
 420   sigemptyset(&unblocked_sigs);
 421   sigemptyset(&allowdebug_blocked_sigs);
 422   sigaddset(&unblocked_sigs, SIGILL);
 423   sigaddset(&unblocked_sigs, SIGSEGV);
 424   sigaddset(&unblocked_sigs, SIGBUS);
 425   sigaddset(&unblocked_sigs, SIGFPE);
 426 #if defined(PPC64)
 427   sigaddset(&unblocked_sigs, SIGTRAP);
 428 #endif
 429   sigaddset(&unblocked_sigs, SR_signum);
 430 
 431   if (!ReduceSignalUsage) {
 432     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 433       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 434       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 435     }
 436     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 437       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 438       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 439     }
 440     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 441       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 442       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 443     }
 444   }
 445   // Fill in signals that are blocked by all but the VM thread.
 446   sigemptyset(&vm_sigs);
 447   if (!ReduceSignalUsage) {
 448     sigaddset(&vm_sigs, BREAK_SIGNAL);
 449   }
 450   debug_only(signal_sets_initialized = true);
 451 
 452 }
 453 
 454 // These are signals that are unblocked while a thread is running Java.
 455 // (For some reason, they get blocked by default.)
 456 sigset_t* os::Linux::unblocked_signals() {
 457   assert(signal_sets_initialized, "Not initialized");
 458   return &unblocked_sigs;
 459 }
 460 
 461 // These are the signals that are blocked while a (non-VM) thread is
 462 // running Java. Only the VM thread handles these signals.
 463 sigset_t* os::Linux::vm_signals() {
 464   assert(signal_sets_initialized, "Not initialized");
 465   return &vm_sigs;
 466 }
 467 
 468 // These are signals that are blocked during cond_wait to allow debugger in
 469 sigset_t* os::Linux::allowdebug_blocked_signals() {
 470   assert(signal_sets_initialized, "Not initialized");
 471   return &allowdebug_blocked_sigs;
 472 }
 473 
 474 void os::Linux::hotspot_sigmask(Thread* thread) {
 475 
 476   //Save caller's signal mask before setting VM signal mask
 477   sigset_t caller_sigmask;
 478   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 479 
 480   OSThread* osthread = thread->osthread();
 481   osthread->set_caller_sigmask(caller_sigmask);
 482 
 483   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 484 
 485   if (!ReduceSignalUsage) {
 486     if (thread->is_VM_thread()) {
 487       // Only the VM thread handles BREAK_SIGNAL ...
 488       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 489     } else {
 490       // ... all other threads block BREAK_SIGNAL
 491       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 492     }
 493   }
 494 }
 495 
 496 //////////////////////////////////////////////////////////////////////////////
 497 // detecting pthread library
 498 
 499 void os::Linux::libpthread_init() {
 500   // Save glibc and pthread version strings.
 501 #if !defined(_CS_GNU_LIBC_VERSION) || \
 502     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 503   #error "glibc too old (< 2.3.2)"
 504 #endif
 505 
 506   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 507   assert(n > 0, "cannot retrieve glibc version");
 508   char *str = (char *)malloc(n, mtInternal);
 509   confstr(_CS_GNU_LIBC_VERSION, str, n);
 510   os::Linux::set_glibc_version(str);
 511 
 512   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 513   assert(n > 0, "cannot retrieve pthread version");
 514   str = (char *)malloc(n, mtInternal);
 515   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 516   os::Linux::set_libpthread_version(str);
 517 }
 518 
 519 /////////////////////////////////////////////////////////////////////////////
 520 // thread stack expansion
 521 
 522 // os::Linux::manually_expand_stack() takes care of expanding the thread
 523 // stack. Note that this is normally not needed: pthread stacks allocate
 524 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 525 // committed. Therefore it is not necessary to expand the stack manually.
 526 //
 527 // Manually expanding the stack was historically needed on LinuxThreads
 528 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 529 // it is kept to deal with very rare corner cases:
 530 //
 531 // For one, user may run the VM on an own implementation of threads
 532 // whose stacks are - like the old LinuxThreads - implemented using
 533 // mmap(MAP_GROWSDOWN).
 534 //
 535 // Also, this coding may be needed if the VM is running on the primordial
 536 // thread. Normally we avoid running on the primordial thread; however,
 537 // user may still invoke the VM on the primordial thread.
 538 //
 539 // The following historical comment describes the details about running
 540 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 541 
 542 
 543 // Force Linux kernel to expand current thread stack. If "bottom" is close
 544 // to the stack guard, caller should block all signals.
 545 //
 546 // MAP_GROWSDOWN:
 547 //   A special mmap() flag that is used to implement thread stacks. It tells
 548 //   kernel that the memory region should extend downwards when needed. This
 549 //   allows early versions of LinuxThreads to only mmap the first few pages
 550 //   when creating a new thread. Linux kernel will automatically expand thread
 551 //   stack as needed (on page faults).
 552 //
 553 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 554 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 555 //   region, it's hard to tell if the fault is due to a legitimate stack
 556 //   access or because of reading/writing non-exist memory (e.g. buffer
 557 //   overrun). As a rule, if the fault happens below current stack pointer,
 558 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 559 //   application (see Linux kernel fault.c).
 560 //
 561 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 562 //   stack overflow detection.
 563 //
 564 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 565 //   not use MAP_GROWSDOWN.
 566 //
 567 // To get around the problem and allow stack banging on Linux, we need to
 568 // manually expand thread stack after receiving the SIGSEGV.
 569 //
 570 // There are two ways to expand thread stack to address "bottom", we used
 571 // both of them in JVM before 1.5:
 572 //   1. adjust stack pointer first so that it is below "bottom", and then
 573 //      touch "bottom"
 574 //   2. mmap() the page in question
 575 //
 576 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 577 // if current sp is already near the lower end of page 101, and we need to
 578 // call mmap() to map page 100, it is possible that part of the mmap() frame
 579 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 580 // That will destroy the mmap() frame and cause VM to crash.
 581 //
 582 // The following code works by adjusting sp first, then accessing the "bottom"
 583 // page to force a page fault. Linux kernel will then automatically expand the
 584 // stack mapping.
 585 //
 586 // _expand_stack_to() assumes its frame size is less than page size, which
 587 // should always be true if the function is not inlined.
 588 
 589 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 590   #define NOINLINE
 591 #else
 592   #define NOINLINE __attribute__ ((noinline))
 593 #endif
 594 
 595 static void _expand_stack_to(address bottom) NOINLINE;
 596 
 597 static void _expand_stack_to(address bottom) {
 598   address sp;
 599   size_t size;
 600   volatile char *p;
 601 
 602   // Adjust bottom to point to the largest address within the same page, it
 603   // gives us a one-page buffer if alloca() allocates slightly more memory.
 604   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 605   bottom += os::Linux::page_size() - 1;
 606 
 607   // sp might be slightly above current stack pointer; if that's the case, we
 608   // will alloca() a little more space than necessary, which is OK. Don't use
 609   // os::current_stack_pointer(), as its result can be slightly below current
 610   // stack pointer, causing us to not alloca enough to reach "bottom".
 611   sp = (address)&sp;
 612 
 613   if (sp > bottom) {
 614     size = sp - bottom;
 615     p = (volatile char *)alloca(size);
 616     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 617     p[0] = '\0';
 618   }
 619 }
 620 
 621 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 622   assert(t!=NULL, "just checking");
 623   assert(t->osthread()->expanding_stack(), "expand should be set");
 624   assert(t->stack_base() != NULL, "stack_base was not initialized");
 625 
 626   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 627     sigset_t mask_all, old_sigset;
 628     sigfillset(&mask_all);
 629     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 630     _expand_stack_to(addr);
 631     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 632     return true;
 633   }
 634   return false;
 635 }
 636 
 637 //////////////////////////////////////////////////////////////////////////////
 638 // create new thread
 639 
 640 // Thread start routine for all newly created threads
 641 static void *java_start(Thread *thread) {
 642   // Try to randomize the cache line index of hot stack frames.
 643   // This helps when threads of the same stack traces evict each other's
 644   // cache lines. The threads can be either from the same JVM instance, or
 645   // from different JVM instances. The benefit is especially true for
 646   // processors with hyperthreading technology.
 647   static int counter = 0;
 648   int pid = os::current_process_id();
 649   alloca(((pid ^ counter++) & 7) * 128);
 650 
 651   ThreadLocalStorage::set_thread(thread);
 652 
 653   OSThread* osthread = thread->osthread();
 654   Monitor* sync = osthread->startThread_lock();
 655 
 656   // thread_id is kernel thread id (similar to Solaris LWP id)
 657   osthread->set_thread_id(os::Linux::gettid());
 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   return 0;
 689 }
 690 
 691 bool os::create_thread(Thread* thread, ThreadType thr_type,
 692                        size_t stack_size) {
 693   assert(thread->osthread() == NULL, "caller responsible");
 694 
 695   // Allocate the OSThread object
 696   OSThread* osthread = new OSThread(NULL, NULL);
 697   if (osthread == NULL) {
 698     return false;
 699   }
 700 
 701   // set the correct thread state
 702   osthread->set_thread_type(thr_type);
 703 
 704   // Initial state is ALLOCATED but not INITIALIZED
 705   osthread->set_state(ALLOCATED);
 706 
 707   thread->set_osthread(osthread);
 708 
 709   // init thread attributes
 710   pthread_attr_t attr;
 711   pthread_attr_init(&attr);
 712   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 713 
 714   // stack size
 715   if (os::Linux::supports_variable_stack_size()) {
 716     // calculate stack size if it's not specified by caller
 717     if (stack_size == 0) {
 718       stack_size = os::Linux::default_stack_size(thr_type);
 719 
 720       switch (thr_type) {
 721       case os::java_thread:
 722         // Java threads use ThreadStackSize which default value can be
 723         // changed with the flag -Xss
 724         assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 725         stack_size = JavaThread::stack_size_at_create();
 726         break;
 727       case os::compiler_thread:
 728         if (CompilerThreadStackSize > 0) {
 729           stack_size = (size_t)(CompilerThreadStackSize * K);
 730           break;
 731         } // else fall through:
 732           // use VMThreadStackSize if CompilerThreadStackSize is not defined
 733       case os::vm_thread:
 734       case os::pgc_thread:
 735       case os::cgc_thread:
 736       case os::watcher_thread:
 737         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 738         break;
 739       }
 740     }
 741 
 742     stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 743     pthread_attr_setstacksize(&attr, stack_size);
 744   } else {
 745     // let pthread_create() pick the default value.
 746   }
 747 
 748   // glibc guard page
 749   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 750 
 751   ThreadState state;
 752 
 753   {
 754     pthread_t tid;
 755     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 756 
 757     pthread_attr_destroy(&attr);
 758 
 759     if (ret != 0) {
 760       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 761         perror("pthread_create()");
 762       }
 763       // Need to clean up stuff we've allocated so far
 764       thread->set_osthread(NULL);
 765       delete osthread;
 766       return false;
 767     }
 768 
 769     // Store pthread info into the OSThread
 770     osthread->set_pthread_id(tid);
 771 
 772     // Wait until child thread is either initialized or aborted
 773     {
 774       Monitor* sync_with_child = osthread->startThread_lock();
 775       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 776       while ((state = osthread->get_state()) == ALLOCATED) {
 777         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 778       }
 779     }
 780   }
 781 
 782   // Aborted due to thread limit being reached
 783   if (state == ZOMBIE) {
 784     thread->set_osthread(NULL);
 785     delete osthread;
 786     return false;
 787   }
 788 
 789   // The thread is returned suspended (in state INITIALIZED),
 790   // and is started higher up in the call chain
 791   assert(state == INITIALIZED, "race condition");
 792   return true;
 793 }
 794 
 795 /////////////////////////////////////////////////////////////////////////////
 796 // attach existing thread
 797 
 798 // bootstrap the main thread
 799 bool os::create_main_thread(JavaThread* thread) {
 800   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 801   return create_attached_thread(thread);
 802 }
 803 
 804 bool os::create_attached_thread(JavaThread* thread) {
 805 #ifdef ASSERT
 806   thread->verify_not_published();
 807 #endif
 808 
 809   // Allocate the OSThread object
 810   OSThread* osthread = new OSThread(NULL, NULL);
 811 
 812   if (osthread == NULL) {
 813     return false;
 814   }
 815 
 816   // Store pthread info into the OSThread
 817   osthread->set_thread_id(os::Linux::gettid());
 818   osthread->set_pthread_id(::pthread_self());
 819 
 820   // initialize floating point control register
 821   os::Linux::init_thread_fpu_state();
 822 
 823   // Initial thread state is RUNNABLE
 824   osthread->set_state(RUNNABLE);
 825 
 826   thread->set_osthread(osthread);
 827 
 828   if (UseNUMA) {
 829     int lgrp_id = os::numa_get_group_id();
 830     if (lgrp_id != -1) {
 831       thread->set_lgrp_id(lgrp_id);
 832     }
 833   }
 834 
 835   if (os::Linux::is_initial_thread()) {
 836     // If current thread is initial thread, its stack is mapped on demand,
 837     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 838     // the entire stack region to avoid SEGV in stack banging.
 839     // It is also useful to get around the heap-stack-gap problem on SuSE
 840     // kernel (see 4821821 for details). We first expand stack to the top
 841     // of yellow zone, then enable stack yellow zone (order is significant,
 842     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 843     // is no gap between the last two virtual memory regions.
 844 
 845     JavaThread *jt = (JavaThread *)thread;
 846     address addr = jt->stack_yellow_zone_base();
 847     assert(addr != NULL, "initialization problem?");
 848     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 849 
 850     osthread->set_expanding_stack();
 851     os::Linux::manually_expand_stack(jt, addr);
 852     osthread->clear_expanding_stack();
 853   }
 854 
 855   // initialize signal mask for this thread
 856   // and save the caller's signal mask
 857   os::Linux::hotspot_sigmask(thread);
 858 
 859   return true;
 860 }
 861 
 862 void os::pd_start_thread(Thread* thread) {
 863   OSThread * osthread = thread->osthread();
 864   assert(osthread->get_state() != INITIALIZED, "just checking");
 865   Monitor* sync_with_child = osthread->startThread_lock();
 866   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 867   sync_with_child->notify();
 868 }
 869 
 870 // Free Linux resources related to the OSThread
 871 void os::free_thread(OSThread* osthread) {
 872   assert(osthread != NULL, "osthread not set");
 873 
 874   if (Thread::current()->osthread() == osthread) {
 875     // Restore caller's signal mask
 876     sigset_t sigmask = osthread->caller_sigmask();
 877     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 878   }
 879 
 880   delete osthread;
 881 }
 882 
 883 //////////////////////////////////////////////////////////////////////////////
 884 // thread local storage
 885 
 886 // Restore the thread pointer if the destructor is called. This is in case
 887 // someone from JNI code sets up a destructor with pthread_key_create to run
 888 // detachCurrentThread on thread death. Unless we restore the thread pointer we
 889 // will hang or crash. When detachCurrentThread is called the key will be set
 890 // to null and we will not be called again. If detachCurrentThread is never
 891 // called we could loop forever depending on the pthread implementation.
 892 static void restore_thread_pointer(void* p) {
 893   Thread* thread = (Thread*) p;
 894   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
 895 }
 896 
 897 int os::allocate_thread_local_storage() {
 898   pthread_key_t key;
 899   int rslt = pthread_key_create(&key, restore_thread_pointer);
 900   assert(rslt == 0, "cannot allocate thread local storage");
 901   return (int)key;
 902 }
 903 
 904 // Note: This is currently not used by VM, as we don't destroy TLS key
 905 // on VM exit.
 906 void os::free_thread_local_storage(int index) {
 907   int rslt = pthread_key_delete((pthread_key_t)index);
 908   assert(rslt == 0, "invalid index");
 909 }
 910 
 911 void os::thread_local_storage_at_put(int index, void* value) {
 912   int rslt = pthread_setspecific((pthread_key_t)index, value);
 913   assert(rslt == 0, "pthread_setspecific failed");
 914 }
 915 
 916 extern "C" Thread* get_thread() {
 917   return ThreadLocalStorage::thread();
 918 }
 919 
 920 //////////////////////////////////////////////////////////////////////////////
 921 // initial thread
 922 
 923 // Check if current thread is the initial thread, similar to Solaris thr_main.
 924 bool os::Linux::is_initial_thread(void) {
 925   char dummy;
 926   // If called before init complete, thread stack bottom will be null.
 927   // Can be called if fatal error occurs before initialization.
 928   if (initial_thread_stack_bottom() == NULL) return false;
 929   assert(initial_thread_stack_bottom() != NULL &&
 930          initial_thread_stack_size()   != 0,
 931          "os::init did not locate initial thread's stack region");
 932   if ((address)&dummy >= initial_thread_stack_bottom() &&
 933       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 934     return true;
 935   } else {
 936     return false;
 937   }
 938 }
 939 
 940 // Find the virtual memory area that contains addr
 941 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 942   FILE *fp = fopen("/proc/self/maps", "r");
 943   if (fp) {
 944     address low, high;
 945     while (!feof(fp)) {
 946       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 947         if (low <= addr && addr < high) {
 948           if (vma_low)  *vma_low  = low;
 949           if (vma_high) *vma_high = high;
 950           fclose(fp);
 951           return true;
 952         }
 953       }
 954       for (;;) {
 955         int ch = fgetc(fp);
 956         if (ch == EOF || ch == (int)'\n') break;
 957       }
 958     }
 959     fclose(fp);
 960   }
 961   return false;
 962 }
 963 
 964 // Locate initial thread stack. This special handling of initial thread stack
 965 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 966 // bogus value for initial thread.
 967 void os::Linux::capture_initial_stack(size_t max_size) {
 968   // stack size is the easy part, get it from RLIMIT_STACK
 969   size_t stack_size;
 970   struct rlimit rlim;
 971   getrlimit(RLIMIT_STACK, &rlim);
 972   stack_size = rlim.rlim_cur;
 973 
 974   // 6308388: a bug in ld.so will relocate its own .data section to the
 975   //   lower end of primordial stack; reduce ulimit -s value a little bit
 976   //   so we won't install guard page on ld.so's data section.
 977   stack_size -= 2 * page_size();
 978 
 979   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
 980   //   7.1, in both cases we will get 2G in return value.
 981   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
 982   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
 983   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
 984   //   in case other parts in glibc still assumes 2M max stack size.
 985   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
 986   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
 987   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
 988     stack_size = 2 * K * K IA64_ONLY(*2);
 989   }
 990   // Try to figure out where the stack base (top) is. This is harder.
 991   //
 992   // When an application is started, glibc saves the initial stack pointer in
 993   // a global variable "__libc_stack_end", which is then used by system
 994   // libraries. __libc_stack_end should be pretty close to stack top. The
 995   // variable is available since the very early days. However, because it is
 996   // a private interface, it could disappear in the future.
 997   //
 998   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 999   // to __libc_stack_end, it is very close to stack top, but isn't the real
1000   // stack top. Note that /proc may not exist if VM is running as a chroot
1001   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1002   // /proc/<pid>/stat could change in the future (though unlikely).
1003   //
1004   // We try __libc_stack_end first. If that doesn't work, look for
1005   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1006   // as a hint, which should work well in most cases.
1007 
1008   uintptr_t stack_start;
1009 
1010   // try __libc_stack_end first
1011   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1012   if (p && *p) {
1013     stack_start = *p;
1014   } else {
1015     // see if we can get the start_stack field from /proc/self/stat
1016     FILE *fp;
1017     int pid;
1018     char state;
1019     int ppid;
1020     int pgrp;
1021     int session;
1022     int nr;
1023     int tpgrp;
1024     unsigned long flags;
1025     unsigned long minflt;
1026     unsigned long cminflt;
1027     unsigned long majflt;
1028     unsigned long cmajflt;
1029     unsigned long utime;
1030     unsigned long stime;
1031     long cutime;
1032     long cstime;
1033     long prio;
1034     long nice;
1035     long junk;
1036     long it_real;
1037     uintptr_t start;
1038     uintptr_t vsize;
1039     intptr_t rss;
1040     uintptr_t rsslim;
1041     uintptr_t scodes;
1042     uintptr_t ecode;
1043     int i;
1044 
1045     // Figure what the primordial thread stack base is. Code is inspired
1046     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1047     // followed by command name surrounded by parentheses, state, etc.
1048     char stat[2048];
1049     int statlen;
1050 
1051     fp = fopen("/proc/self/stat", "r");
1052     if (fp) {
1053       statlen = fread(stat, 1, 2047, fp);
1054       stat[statlen] = '\0';
1055       fclose(fp);
1056 
1057       // Skip pid and the command string. Note that we could be dealing with
1058       // weird command names, e.g. user could decide to rename java launcher
1059       // to "java 1.4.2 :)", then the stat file would look like
1060       //                1234 (java 1.4.2 :)) R ... ...
1061       // We don't really need to know the command string, just find the last
1062       // occurrence of ")" and then start parsing from there. See bug 4726580.
1063       char * s = strrchr(stat, ')');
1064 
1065       i = 0;
1066       if (s) {
1067         // Skip blank chars
1068         do { s++; } while (s && isspace(*s));
1069 
1070 #define _UFM UINTX_FORMAT
1071 #define _DFM INTX_FORMAT
1072 
1073         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1074         //              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
1075         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,
1076                    &state,          // 3  %c
1077                    &ppid,           // 4  %d
1078                    &pgrp,           // 5  %d
1079                    &session,        // 6  %d
1080                    &nr,             // 7  %d
1081                    &tpgrp,          // 8  %d
1082                    &flags,          // 9  %lu
1083                    &minflt,         // 10 %lu
1084                    &cminflt,        // 11 %lu
1085                    &majflt,         // 12 %lu
1086                    &cmajflt,        // 13 %lu
1087                    &utime,          // 14 %lu
1088                    &stime,          // 15 %lu
1089                    &cutime,         // 16 %ld
1090                    &cstime,         // 17 %ld
1091                    &prio,           // 18 %ld
1092                    &nice,           // 19 %ld
1093                    &junk,           // 20 %ld
1094                    &it_real,        // 21 %ld
1095                    &start,          // 22 UINTX_FORMAT
1096                    &vsize,          // 23 UINTX_FORMAT
1097                    &rss,            // 24 INTX_FORMAT
1098                    &rsslim,         // 25 UINTX_FORMAT
1099                    &scodes,         // 26 UINTX_FORMAT
1100                    &ecode,          // 27 UINTX_FORMAT
1101                    &stack_start);   // 28 UINTX_FORMAT
1102       }
1103 
1104 #undef _UFM
1105 #undef _DFM
1106 
1107       if (i != 28 - 2) {
1108         assert(false, "Bad conversion from /proc/self/stat");
1109         // product mode - assume we are the initial thread, good luck in the
1110         // embedded case.
1111         warning("Can't detect initial thread stack location - bad conversion");
1112         stack_start = (uintptr_t) &rlim;
1113       }
1114     } else {
1115       // For some reason we can't open /proc/self/stat (for example, running on
1116       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1117       // most cases, so don't abort:
1118       warning("Can't detect initial thread stack location - no /proc/self/stat");
1119       stack_start = (uintptr_t) &rlim;
1120     }
1121   }
1122 
1123   // Now we have a pointer (stack_start) very close to the stack top, the
1124   // next thing to do is to figure out the exact location of stack top. We
1125   // can find out the virtual memory area that contains stack_start by
1126   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1127   // and its upper limit is the real stack top. (again, this would fail if
1128   // running inside chroot, because /proc may not exist.)
1129 
1130   uintptr_t stack_top;
1131   address low, high;
1132   if (find_vma((address)stack_start, &low, &high)) {
1133     // success, "high" is the true stack top. (ignore "low", because initial
1134     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1135     stack_top = (uintptr_t)high;
1136   } else {
1137     // failed, likely because /proc/self/maps does not exist
1138     warning("Can't detect initial thread stack location - find_vma failed");
1139     // best effort: stack_start is normally within a few pages below the real
1140     // stack top, use it as stack top, and reduce stack size so we won't put
1141     // guard page outside stack.
1142     stack_top = stack_start;
1143     stack_size -= 16 * page_size();
1144   }
1145 
1146   // stack_top could be partially down the page so align it
1147   stack_top = align_size_up(stack_top, page_size());
1148 
1149   if (max_size && stack_size > max_size) {
1150     _initial_thread_stack_size = max_size;
1151   } else {
1152     _initial_thread_stack_size = stack_size;
1153   }
1154 
1155   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1156   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1157 }
1158 
1159 ////////////////////////////////////////////////////////////////////////////////
1160 // time support
1161 
1162 // Time since start-up in seconds to a fine granularity.
1163 // Used by VMSelfDestructTimer and the MemProfiler.
1164 double os::elapsedTime() {
1165 
1166   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1167 }
1168 
1169 jlong os::elapsed_counter() {
1170   return javaTimeNanos() - initial_time_count;
1171 }
1172 
1173 jlong os::elapsed_frequency() {
1174   return NANOSECS_PER_SEC; // nanosecond resolution
1175 }
1176 
1177 bool os::supports_vtime() { return true; }
1178 bool os::enable_vtime()   { return false; }
1179 bool os::vtime_enabled()  { return false; }
1180 
1181 double os::elapsedVTime() {
1182   struct rusage usage;
1183   int retval = getrusage(RUSAGE_THREAD, &usage);
1184   if (retval == 0) {
1185     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);
1186   } else {
1187     // better than nothing, but not much
1188     return elapsedTime();
1189   }
1190 }
1191 
1192 jlong os::javaTimeMillis() {
1193   timeval time;
1194   int status = gettimeofday(&time, NULL);
1195   assert(status != -1, "linux error");
1196   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1197 }
1198 
1199 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1200   timeval time;
1201   int status = gettimeofday(&time, NULL);
1202   assert(status != -1, "linux error");
1203   seconds = jlong(time.tv_sec);
1204   nanos = jlong(time.tv_usec) * 1000;
1205 }
1206 
1207 
1208 #ifndef CLOCK_MONOTONIC
1209   #define CLOCK_MONOTONIC (1)
1210 #endif
1211 
1212 void os::Linux::clock_init() {
1213   // we do dlopen's in this particular order due to bug in linux
1214   // dynamical loader (see 6348968) leading to crash on exit
1215   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1216   if (handle == NULL) {
1217     handle = dlopen("librt.so", RTLD_LAZY);
1218   }
1219 
1220   if (handle) {
1221     int (*clock_getres_func)(clockid_t, struct timespec*) =
1222            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1223     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1224            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1225     if (clock_getres_func && clock_gettime_func) {
1226       // See if monotonic clock is supported by the kernel. Note that some
1227       // early implementations simply return kernel jiffies (updated every
1228       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1229       // for nano time (though the monotonic property is still nice to have).
1230       // It's fixed in newer kernels, however clock_getres() still returns
1231       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1232       // resolution for now. Hopefully as people move to new kernels, this
1233       // won't be a problem.
1234       struct timespec res;
1235       struct timespec tp;
1236       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1237           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1238         // yes, monotonic clock is supported
1239         _clock_gettime = clock_gettime_func;
1240         return;
1241       } else {
1242         // close librt if there is no monotonic clock
1243         dlclose(handle);
1244       }
1245     }
1246   }
1247   warning("No monotonic clock was available - timed services may " \
1248           "be adversely affected if the time-of-day clock changes");
1249 }
1250 
1251 #ifndef SYS_clock_getres
1252   #if defined(IA32) || defined(AMD64)
1253     #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1254     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1255   #else
1256     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1257     #define sys_clock_getres(x,y)  -1
1258   #endif
1259 #else
1260   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1261 #endif
1262 
1263 void os::Linux::fast_thread_clock_init() {
1264   if (!UseLinuxPosixThreadCPUClocks) {
1265     return;
1266   }
1267   clockid_t clockid;
1268   struct timespec tp;
1269   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1270       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1271 
1272   // Switch to using fast clocks for thread cpu time if
1273   // the sys_clock_getres() returns 0 error code.
1274   // Note, that some kernels may support the current thread
1275   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1276   // returned by the pthread_getcpuclockid().
1277   // If the fast Posix clocks are supported then the sys_clock_getres()
1278   // must return at least tp.tv_sec == 0 which means a resolution
1279   // better than 1 sec. This is extra check for reliability.
1280 
1281   if (pthread_getcpuclockid_func &&
1282       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1283       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1284     _supports_fast_thread_cpu_time = true;
1285     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1286   }
1287 }
1288 
1289 jlong os::javaTimeNanos() {
1290   if (os::supports_monotonic_clock()) {
1291     struct timespec tp;
1292     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1293     assert(status == 0, "gettime error");
1294     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1295     return result;
1296   } else {
1297     timeval time;
1298     int status = gettimeofday(&time, NULL);
1299     assert(status != -1, "linux error");
1300     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1301     return 1000 * usecs;
1302   }
1303 }
1304 
1305 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1306   if (os::supports_monotonic_clock()) {
1307     info_ptr->max_value = ALL_64_BITS;
1308 
1309     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1310     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1311     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1312   } else {
1313     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1314     info_ptr->max_value = ALL_64_BITS;
1315 
1316     // gettimeofday is a real time clock so it skips
1317     info_ptr->may_skip_backward = true;
1318     info_ptr->may_skip_forward = true;
1319   }
1320 
1321   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1322 }
1323 
1324 // Return the real, user, and system times in seconds from an
1325 // arbitrary fixed point in the past.
1326 bool os::getTimesSecs(double* process_real_time,
1327                       double* process_user_time,
1328                       double* process_system_time) {
1329   struct tms ticks;
1330   clock_t real_ticks = times(&ticks);
1331 
1332   if (real_ticks == (clock_t) (-1)) {
1333     return false;
1334   } else {
1335     double ticks_per_second = (double) clock_tics_per_sec;
1336     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1337     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1338     *process_real_time = ((double) real_ticks) / ticks_per_second;
1339 
1340     return true;
1341   }
1342 }
1343 
1344 
1345 char * os::local_time_string(char *buf, size_t buflen) {
1346   struct tm t;
1347   time_t long_time;
1348   time(&long_time);
1349   localtime_r(&long_time, &t);
1350   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1351                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1352                t.tm_hour, t.tm_min, t.tm_sec);
1353   return buf;
1354 }
1355 
1356 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1357   return localtime_r(clock, res);
1358 }
1359 
1360 ////////////////////////////////////////////////////////////////////////////////
1361 // runtime exit support
1362 
1363 // Note: os::shutdown() might be called very early during initialization, or
1364 // called from signal handler. Before adding something to os::shutdown(), make
1365 // sure it is async-safe and can handle partially initialized VM.
1366 void os::shutdown() {
1367 
1368   // allow PerfMemory to attempt cleanup of any persistent resources
1369   perfMemory_exit();
1370 
1371   // needs to remove object in file system
1372   AttachListener::abort();
1373 
1374   // flush buffered output, finish log files
1375   ostream_abort();
1376 
1377   // Check for abort hook
1378   abort_hook_t abort_hook = Arguments::abort_hook();
1379   if (abort_hook != NULL) {
1380     abort_hook();
1381   }
1382 
1383 }
1384 
1385 // Note: os::abort() might be called very early during initialization, or
1386 // called from signal handler. Before adding something to os::abort(), make
1387 // sure it is async-safe and can handle partially initialized VM.
1388 void os::abort(bool dump_core, void* siginfo, void* context) {
1389   os::shutdown();
1390   if (dump_core) {
1391 #ifndef PRODUCT
1392     fdStream out(defaultStream::output_fd());
1393     out.print_raw("Current thread is ");
1394     char buf[16];
1395     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1396     out.print_raw_cr(buf);
1397     out.print_raw_cr("Dumping core ...");
1398 #endif
1399     ::abort(); // dump core
1400   }
1401 
1402   ::exit(1);
1403 }
1404 
1405 // Die immediately, no exit hook, no abort hook, no cleanup.
1406 void os::die() {
1407   ::abort();
1408 }
1409 
1410 
1411 // This method is a copy of JDK's sysGetLastErrorString
1412 // from src/solaris/hpi/src/system_md.c
1413 
1414 size_t os::lasterror(char *buf, size_t len) {
1415   if (errno == 0)  return 0;
1416 
1417   const char *s = ::strerror(errno);
1418   size_t n = ::strlen(s);
1419   if (n >= len) {
1420     n = len - 1;
1421   }
1422   ::strncpy(buf, s, n);
1423   buf[n] = '\0';
1424   return n;
1425 }
1426 
1427 intx os::current_thread_id() { return (intx)pthread_self(); }
1428 int os::current_process_id() {
1429   return ::getpid();
1430 }
1431 
1432 // DLL functions
1433 
1434 const char* os::dll_file_extension() { return ".so"; }
1435 
1436 // This must be hard coded because it's the system's temporary
1437 // directory not the java application's temp directory, ala java.io.tmpdir.
1438 const char* os::get_temp_directory() { return "/tmp"; }
1439 
1440 static bool file_exists(const char* filename) {
1441   struct stat statbuf;
1442   if (filename == NULL || strlen(filename) == 0) {
1443     return false;
1444   }
1445   return os::stat(filename, &statbuf) == 0;
1446 }
1447 
1448 bool os::dll_build_name(char* buffer, size_t buflen,
1449                         const char* pname, const char* fname) {
1450   bool retval = false;
1451   // Copied from libhpi
1452   const size_t pnamelen = pname ? strlen(pname) : 0;
1453 
1454   // Return error on buffer overflow.
1455   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1456     return retval;
1457   }
1458 
1459   if (pnamelen == 0) {
1460     snprintf(buffer, buflen, "lib%s.so", fname);
1461     retval = true;
1462   } else if (strchr(pname, *os::path_separator()) != NULL) {
1463     int n;
1464     char** pelements = split_path(pname, &n);
1465     if (pelements == NULL) {
1466       return false;
1467     }
1468     for (int i = 0; i < n; i++) {
1469       // Really shouldn't be NULL, but check can't hurt
1470       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1471         continue; // skip the empty path values
1472       }
1473       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1474       if (file_exists(buffer)) {
1475         retval = true;
1476         break;
1477       }
1478     }
1479     // release the storage
1480     for (int i = 0; i < n; i++) {
1481       if (pelements[i] != NULL) {
1482         FREE_C_HEAP_ARRAY(char, pelements[i]);
1483       }
1484     }
1485     if (pelements != NULL) {
1486       FREE_C_HEAP_ARRAY(char*, pelements);
1487     }
1488   } else {
1489     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1490     retval = true;
1491   }
1492   return retval;
1493 }
1494 
1495 // check if addr is inside libjvm.so
1496 bool os::address_is_in_vm(address addr) {
1497   static address libjvm_base_addr;
1498   Dl_info dlinfo;
1499 
1500   if (libjvm_base_addr == NULL) {
1501     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1502       libjvm_base_addr = (address)dlinfo.dli_fbase;
1503     }
1504     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1505   }
1506 
1507   if (dladdr((void *)addr, &dlinfo) != 0) {
1508     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1509   }
1510 
1511   return false;
1512 }
1513 
1514 bool os::dll_address_to_function_name(address addr, char *buf,
1515                                       int buflen, int *offset,
1516                                       bool demangle) {
1517   // buf is not optional, but offset is optional
1518   assert(buf != NULL, "sanity check");
1519 
1520   Dl_info dlinfo;
1521 
1522   if (dladdr((void*)addr, &dlinfo) != 0) {
1523     // see if we have a matching symbol
1524     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1525       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1526         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1527       }
1528       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1529       return true;
1530     }
1531     // no matching symbol so try for just file info
1532     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1533       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1534                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1535         return true;
1536       }
1537     }
1538   }
1539 
1540   buf[0] = '\0';
1541   if (offset != NULL) *offset = -1;
1542   return false;
1543 }
1544 
1545 struct _address_to_library_name {
1546   address addr;          // input : memory address
1547   size_t  buflen;        //         size of fname
1548   char*   fname;         // output: library name
1549   address base;          //         library base addr
1550 };
1551 
1552 static int address_to_library_name_callback(struct dl_phdr_info *info,
1553                                             size_t size, void *data) {
1554   int i;
1555   bool found = false;
1556   address libbase = NULL;
1557   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1558 
1559   // iterate through all loadable segments
1560   for (i = 0; i < info->dlpi_phnum; i++) {
1561     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1562     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1563       // base address of a library is the lowest address of its loaded
1564       // segments.
1565       if (libbase == NULL || libbase > segbase) {
1566         libbase = segbase;
1567       }
1568       // see if 'addr' is within current segment
1569       if (segbase <= d->addr &&
1570           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1571         found = true;
1572       }
1573     }
1574   }
1575 
1576   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1577   // so dll_address_to_library_name() can fall through to use dladdr() which
1578   // can figure out executable name from argv[0].
1579   if (found && info->dlpi_name && info->dlpi_name[0]) {
1580     d->base = libbase;
1581     if (d->fname) {
1582       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1583     }
1584     return 1;
1585   }
1586   return 0;
1587 }
1588 
1589 bool os::dll_address_to_library_name(address addr, char* buf,
1590                                      int buflen, int* offset) {
1591   // buf is not optional, but offset is optional
1592   assert(buf != NULL, "sanity check");
1593 
1594   Dl_info dlinfo;
1595   struct _address_to_library_name data;
1596 
1597   // There is a bug in old glibc dladdr() implementation that it could resolve
1598   // to wrong library name if the .so file has a base address != NULL. Here
1599   // we iterate through the program headers of all loaded libraries to find
1600   // out which library 'addr' really belongs to. This workaround can be
1601   // removed once the minimum requirement for glibc is moved to 2.3.x.
1602   data.addr = addr;
1603   data.fname = buf;
1604   data.buflen = buflen;
1605   data.base = NULL;
1606   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1607 
1608   if (rslt) {
1609     // buf already contains library name
1610     if (offset) *offset = addr - data.base;
1611     return true;
1612   }
1613   if (dladdr((void*)addr, &dlinfo) != 0) {
1614     if (dlinfo.dli_fname != NULL) {
1615       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1616     }
1617     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1618       *offset = addr - (address)dlinfo.dli_fbase;
1619     }
1620     return true;
1621   }
1622 
1623   buf[0] = '\0';
1624   if (offset) *offset = -1;
1625   return false;
1626 }
1627 
1628 // Loads .dll/.so and
1629 // in case of error it checks if .dll/.so was built for the
1630 // same architecture as Hotspot is running on
1631 
1632 
1633 // Remember the stack's state. The Linux dynamic linker will change
1634 // the stack to 'executable' at most once, so we must safepoint only once.
1635 bool os::Linux::_stack_is_executable = false;
1636 
1637 // VM operation that loads a library.  This is necessary if stack protection
1638 // of the Java stacks can be lost during loading the library.  If we
1639 // do not stop the Java threads, they can stack overflow before the stacks
1640 // are protected again.
1641 class VM_LinuxDllLoad: public VM_Operation {
1642  private:
1643   const char *_filename;
1644   char *_ebuf;
1645   int _ebuflen;
1646   void *_lib;
1647  public:
1648   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1649     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1650   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1651   void doit() {
1652     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1653     os::Linux::_stack_is_executable = true;
1654   }
1655   void* loaded_library() { return _lib; }
1656 };
1657 
1658 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1659   void * result = NULL;
1660   bool load_attempted = false;
1661 
1662   // Check whether the library to load might change execution rights
1663   // of the stack. If they are changed, the protection of the stack
1664   // guard pages will be lost. We need a safepoint to fix this.
1665   //
1666   // See Linux man page execstack(8) for more info.
1667   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1668     ElfFile ef(filename);
1669     if (!ef.specifies_noexecstack()) {
1670       if (!is_init_completed()) {
1671         os::Linux::_stack_is_executable = true;
1672         // This is OK - No Java threads have been created yet, and hence no
1673         // stack guard pages to fix.
1674         //
1675         // This should happen only when you are building JDK7 using a very
1676         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1677         //
1678         // Dynamic loader will make all stacks executable after
1679         // this function returns, and will not do that again.
1680         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1681       } else {
1682         warning("You have loaded library %s which might have disabled stack guard. "
1683                 "The VM will try to fix the stack guard now.\n"
1684                 "It's highly recommended that you fix the library with "
1685                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1686                 filename);
1687 
1688         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1689         JavaThread *jt = JavaThread::current();
1690         if (jt->thread_state() != _thread_in_native) {
1691           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1692           // that requires ExecStack. Cannot enter safe point. Let's give up.
1693           warning("Unable to fix stack guard. Giving up.");
1694         } else {
1695           if (!LoadExecStackDllInVMThread) {
1696             // This is for the case where the DLL has an static
1697             // constructor function that executes JNI code. We cannot
1698             // load such DLLs in the VMThread.
1699             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1700           }
1701 
1702           ThreadInVMfromNative tiv(jt);
1703           debug_only(VMNativeEntryWrapper vew;)
1704 
1705           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1706           VMThread::execute(&op);
1707           if (LoadExecStackDllInVMThread) {
1708             result = op.loaded_library();
1709           }
1710           load_attempted = true;
1711         }
1712       }
1713     }
1714   }
1715 
1716   if (!load_attempted) {
1717     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1718   }
1719 
1720   if (result != NULL) {
1721     // Successful loading
1722     return result;
1723   }
1724 
1725   Elf32_Ehdr elf_head;
1726   int diag_msg_max_length=ebuflen-strlen(ebuf);
1727   char* diag_msg_buf=ebuf+strlen(ebuf);
1728 
1729   if (diag_msg_max_length==0) {
1730     // No more space in ebuf for additional diagnostics message
1731     return NULL;
1732   }
1733 
1734 
1735   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1736 
1737   if (file_descriptor < 0) {
1738     // Can't open library, report dlerror() message
1739     return NULL;
1740   }
1741 
1742   bool failed_to_read_elf_head=
1743     (sizeof(elf_head)!=
1744      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1745 
1746   ::close(file_descriptor);
1747   if (failed_to_read_elf_head) {
1748     // file i/o error - report dlerror() msg
1749     return NULL;
1750   }
1751 
1752   typedef struct {
1753     Elf32_Half  code;         // Actual value as defined in elf.h
1754     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1755     char        elf_class;    // 32 or 64 bit
1756     char        endianess;    // MSB or LSB
1757     char*       name;         // String representation
1758   } arch_t;
1759 
1760 #ifndef EM_486
1761   #define EM_486          6               /* Intel 80486 */
1762 #endif
1763 #ifndef EM_AARCH64
1764   #define EM_AARCH64    183               /* ARM AARCH64 */
1765 #endif
1766 
1767   static const arch_t arch_array[]={
1768     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1769     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1770     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1771     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1772     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1773     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1774     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1775     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1776 #if defined(VM_LITTLE_ENDIAN)
1777     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1778 #else
1779     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1780 #endif
1781     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1782     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1783     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1784     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1785     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1786     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1787     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1788     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1789   };
1790 
1791 #if  (defined IA32)
1792   static  Elf32_Half running_arch_code=EM_386;
1793 #elif   (defined AMD64)
1794   static  Elf32_Half running_arch_code=EM_X86_64;
1795 #elif  (defined IA64)
1796   static  Elf32_Half running_arch_code=EM_IA_64;
1797 #elif  (defined __sparc) && (defined _LP64)
1798   static  Elf32_Half running_arch_code=EM_SPARCV9;
1799 #elif  (defined __sparc) && (!defined _LP64)
1800   static  Elf32_Half running_arch_code=EM_SPARC;
1801 #elif  (defined __powerpc64__)
1802   static  Elf32_Half running_arch_code=EM_PPC64;
1803 #elif  (defined __powerpc__)
1804   static  Elf32_Half running_arch_code=EM_PPC;
1805 #elif  (defined ARM)
1806   static  Elf32_Half running_arch_code=EM_ARM;
1807 #elif  (defined S390)
1808   static  Elf32_Half running_arch_code=EM_S390;
1809 #elif  (defined ALPHA)
1810   static  Elf32_Half running_arch_code=EM_ALPHA;
1811 #elif  (defined MIPSEL)
1812   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1813 #elif  (defined PARISC)
1814   static  Elf32_Half running_arch_code=EM_PARISC;
1815 #elif  (defined MIPS)
1816   static  Elf32_Half running_arch_code=EM_MIPS;
1817 #elif  (defined M68K)
1818   static  Elf32_Half running_arch_code=EM_68K;
1819 #elif  (defined AARCH64)
1820   static  Elf32_Half running_arch_code=EM_AARCH64;
1821 #else
1822     #error Method os::dll_load requires that one of following is defined:\
1823          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1824 #endif
1825 
1826   // Identify compatability class for VM's architecture and library's architecture
1827   // Obtain string descriptions for architectures
1828 
1829   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1830   int running_arch_index=-1;
1831 
1832   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1833     if (running_arch_code == arch_array[i].code) {
1834       running_arch_index    = i;
1835     }
1836     if (lib_arch.code == arch_array[i].code) {
1837       lib_arch.compat_class = arch_array[i].compat_class;
1838       lib_arch.name         = arch_array[i].name;
1839     }
1840   }
1841 
1842   assert(running_arch_index != -1,
1843          "Didn't find running architecture code (running_arch_code) in arch_array");
1844   if (running_arch_index == -1) {
1845     // Even though running architecture detection failed
1846     // we may still continue with reporting dlerror() message
1847     return NULL;
1848   }
1849 
1850   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1851     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1852     return NULL;
1853   }
1854 
1855 #ifndef S390
1856   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1857     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1858     return NULL;
1859   }
1860 #endif // !S390
1861 
1862   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1863     if (lib_arch.name!=NULL) {
1864       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1865                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1866                  lib_arch.name, arch_array[running_arch_index].name);
1867     } else {
1868       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1869                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1870                  lib_arch.code,
1871                  arch_array[running_arch_index].name);
1872     }
1873   }
1874 
1875   return NULL;
1876 }
1877 
1878 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1879                                 int ebuflen) {
1880   void * result = ::dlopen(filename, RTLD_LAZY);
1881   if (result == NULL) {
1882     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1883     ebuf[ebuflen-1] = '\0';
1884   }
1885   return result;
1886 }
1887 
1888 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1889                                        int ebuflen) {
1890   void * result = NULL;
1891   if (LoadExecStackDllInVMThread) {
1892     result = dlopen_helper(filename, ebuf, ebuflen);
1893   }
1894 
1895   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1896   // library that requires an executable stack, or which does not have this
1897   // stack attribute set, dlopen changes the stack attribute to executable. The
1898   // read protection of the guard pages gets lost.
1899   //
1900   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1901   // may have been queued at the same time.
1902 
1903   if (!_stack_is_executable) {
1904     JavaThread *jt = Threads::first();
1905 
1906     while (jt) {
1907       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
1908           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
1909         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
1910                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
1911           warning("Attempt to reguard stack yellow zone failed.");
1912         }
1913       }
1914       jt = jt->next();
1915     }
1916   }
1917 
1918   return result;
1919 }
1920 
1921 void* os::dll_lookup(void* handle, const char* name) {
1922   void* res = dlsym(handle, name);
1923   return res;
1924 }
1925 
1926 void* os::get_default_process_handle() {
1927   return (void*)::dlopen(NULL, RTLD_LAZY);
1928 }
1929 
1930 static bool _print_ascii_file(const char* filename, outputStream* st) {
1931   int fd = ::open(filename, O_RDONLY);
1932   if (fd == -1) {
1933     return false;
1934   }
1935 
1936   char buf[32];
1937   int bytes;
1938   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1939     st->print_raw(buf, bytes);
1940   }
1941 
1942   ::close(fd);
1943 
1944   return true;
1945 }
1946 
1947 void os::print_dll_info(outputStream *st) {
1948   st->print_cr("Dynamic libraries:");
1949 
1950   char fname[32];
1951   pid_t pid = os::Linux::gettid();
1952 
1953   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1954 
1955   if (!_print_ascii_file(fname, st)) {
1956     st->print("Can not get library information for pid = %d\n", pid);
1957   }
1958 }
1959 
1960 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1961   FILE *procmapsFile = NULL;
1962 
1963   // Open the procfs maps file for the current process
1964   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1965     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1966     char line[PATH_MAX + 100];
1967 
1968     // Read line by line from 'file'
1969     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1970       u8 base, top, offset, inode;
1971       char permissions[5];
1972       char device[6];
1973       char name[PATH_MAX + 1];
1974 
1975       // Parse fields from line
1976       sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
1977 
1978       // Filter by device id '00:00' so that we only get file system mapped files.
1979       if (strcmp(device, "00:00") != 0) {
1980 
1981         // Call callback with the fields of interest
1982         if(callback(name, (address)base, (address)top, param)) {
1983           // Oops abort, callback aborted
1984           fclose(procmapsFile);
1985           return 1;
1986         }
1987       }
1988     }
1989     fclose(procmapsFile);
1990   }
1991   return 0;
1992 }
1993 
1994 void os::print_os_info_brief(outputStream* st) {
1995   os::Linux::print_distro_info(st);
1996 
1997   os::Posix::print_uname_info(st);
1998 
1999   os::Linux::print_libversion_info(st);
2000 
2001 }
2002 
2003 void os::print_os_info(outputStream* st) {
2004   st->print("OS:");
2005 
2006   os::Linux::print_distro_info(st);
2007 
2008   os::Posix::print_uname_info(st);
2009 
2010   // Print warning if unsafe chroot environment detected
2011   if (unsafe_chroot_detected) {
2012     st->print("WARNING!! ");
2013     st->print_cr("%s", unstable_chroot_error);
2014   }
2015 
2016   os::Linux::print_libversion_info(st);
2017 
2018   os::Posix::print_rlimit_info(st);
2019 
2020   os::Posix::print_load_average(st);
2021 
2022   os::Linux::print_full_memory_info(st);
2023 }
2024 
2025 // Try to identify popular distros.
2026 // Most Linux distributions have a /etc/XXX-release file, which contains
2027 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2028 // file that also contains the OS version string. Some have more than one
2029 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2030 // /etc/redhat-release.), so the order is important.
2031 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2032 // their own specific XXX-release file as well as a redhat-release file.
2033 // Because of this the XXX-release file needs to be searched for before the
2034 // redhat-release file.
2035 // Since Red Hat has a lsb-release file that is not very descriptive the
2036 // search for redhat-release needs to be before lsb-release.
2037 // Since the lsb-release file is the new standard it needs to be searched
2038 // before the older style release files.
2039 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2040 // next to last resort.  The os-release file is a new standard that contains
2041 // distribution information and the system-release file seems to be an old
2042 // standard that has been replaced by the lsb-release and os-release files.
2043 // Searching for the debian_version file is the last resort.  It contains
2044 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2045 // "Debian " is printed before the contents of the debian_version file.
2046 
2047 const char* distro_files[] = {
2048   "/etc/oracle-release",
2049   "/etc/mandriva-release",
2050   "/etc/mandrake-release",
2051   "/etc/sun-release",
2052   "/etc/redhat-release",
2053   "/etc/lsb-release",
2054   "/etc/SuSE-release",
2055   "/etc/turbolinux-release",
2056   "/etc/gentoo-release",
2057   "/etc/ltib-release",
2058   "/etc/angstrom-version",
2059   "/etc/system-release",
2060   "/etc/os-release",
2061   NULL };
2062 
2063 void os::Linux::print_distro_info(outputStream* st) {
2064   for (int i = 0;; i++) {
2065     const char* file = distro_files[i];
2066     if (file == NULL) {
2067       break;  // done
2068     }
2069     // If file prints, we found it.
2070     if (_print_ascii_file(file, st)) {
2071       return;
2072     }
2073   }
2074 
2075   if (file_exists("/etc/debian_version")) {
2076     st->print("Debian ");
2077     _print_ascii_file("/etc/debian_version", st);
2078   } else {
2079     st->print("Linux");
2080   }
2081   st->cr();
2082 }
2083 
2084 static void parse_os_info(char* distro, size_t length, const char* file) {
2085   FILE* fp = fopen(file, "r");
2086   if (fp != NULL) {
2087     char buf[256];
2088     // get last line of the file.
2089     while (!feof(fp)) {
2090       fgets(buf, sizeof(buf), fp);
2091     }
2092     // Edit out ubuntu things
2093     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2094       char* ptr = strstr(buf, "\"");
2095       if (ptr != NULL) {
2096         ptr++; // go beyond first "
2097         char* nl = strchr(ptr, '\"');
2098         if (nl != NULL) *nl = '\0';
2099         strncpy(distro, ptr, length);
2100         fclose(fp);
2101         return;
2102       }
2103     } else {
2104       // if not Ubuntu, print out whole line minus \n
2105       char* nl = strchr(buf, '\n');
2106       if (nl != NULL) *nl = '\0';
2107       strncpy(distro, buf, length);
2108       fclose(fp);
2109       return;
2110     }
2111   }
2112 }
2113 
2114 void os::get_summary_os_info(char* buf, size_t buflen) {
2115   for (int i = 0;; i++) {
2116     const char* file = distro_files[i];
2117     if (file == NULL) {
2118       break; // ran out of distro_files
2119     }
2120     if (file_exists(file)) {
2121       parse_os_info(buf, buflen, file);
2122       return;
2123     }
2124   }
2125   // special case for debian
2126   if (file_exists("/etc/debian_version")) {
2127     strncpy(buf, "Debian ", buflen);
2128     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2129   } else {
2130     strncpy(buf, "Linux", buflen);
2131   }
2132 }
2133 
2134 void os::Linux::print_libversion_info(outputStream* st) {
2135   // libc, pthread
2136   st->print("libc:");
2137   st->print("%s ", os::Linux::glibc_version());
2138   st->print("%s ", os::Linux::libpthread_version());
2139   st->cr();
2140 }
2141 
2142 void os::Linux::print_full_memory_info(outputStream* st) {
2143   st->print("\n/proc/meminfo:\n");
2144   _print_ascii_file("/proc/meminfo", st);
2145   st->cr();
2146 }
2147 
2148 void os::print_memory_info(outputStream* st) {
2149 
2150   st->print("Memory:");
2151   st->print(" %dk page", os::vm_page_size()>>10);
2152 
2153   // values in struct sysinfo are "unsigned long"
2154   struct sysinfo si;
2155   sysinfo(&si);
2156 
2157   st->print(", physical " UINT64_FORMAT "k",
2158             os::physical_memory() >> 10);
2159   st->print("(" UINT64_FORMAT "k free)",
2160             os::available_memory() >> 10);
2161   st->print(", swap " UINT64_FORMAT "k",
2162             ((jlong)si.totalswap * si.mem_unit) >> 10);
2163   st->print("(" UINT64_FORMAT "k free)",
2164             ((jlong)si.freeswap * si.mem_unit) >> 10);
2165   st->cr();
2166 }
2167 
2168 // Print the first "model name" line and the first "flags" line
2169 // that we find and nothing more. We assume "model name" comes
2170 // before "flags" so if we find a second "model name", then the
2171 // "flags" field is considered missing.
2172 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2173 #if defined(IA32) || defined(AMD64)
2174   // Other platforms have less repetitive cpuinfo files
2175   FILE *fp = fopen("/proc/cpuinfo", "r");
2176   if (fp) {
2177     while (!feof(fp)) {
2178       if (fgets(buf, buflen, fp)) {
2179         // Assume model name comes before flags
2180         bool model_name_printed = false;
2181         if (strstr(buf, "model name") != NULL) {
2182           if (!model_name_printed) {
2183             st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2184             st->print_raw(buf);
2185             model_name_printed = true;
2186           } else {
2187             // model name printed but not flags?  Odd, just return
2188             fclose(fp);
2189             return true;
2190           }
2191         }
2192         // print the flags line too
2193         if (strstr(buf, "flags") != NULL) {
2194           st->print_raw(buf);
2195           fclose(fp);
2196           return true;
2197         }
2198       }
2199     }
2200     fclose(fp);
2201   }
2202 #endif // x86 platforms
2203   return false;
2204 }
2205 
2206 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2207   // Only print the model name if the platform provides this as a summary
2208   if (!print_model_name_and_flags(st, buf, buflen)) {
2209     st->print("\n/proc/cpuinfo:\n");
2210     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2211       st->print_cr("  <Not Available>");
2212     }
2213   }
2214 }
2215 
2216 const char* search_string = IA32_ONLY("model name") AMD64_ONLY("model name") 
2217                             IA64_ONLY("") SPARC_ONLY("cpu")
2218                             ARM32_ONLY("Processor") PPC_ONLY("Processor") AARCH64_ONLY("Processor");
2219 
2220 // Parses the cpuinfo file for string representing the model name.
2221 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2222   FILE* fp = fopen("/proc/cpuinfo", "r");
2223   if (fp != NULL) {
2224     while (!feof(fp)) {
2225       char buf[256];
2226       if (fgets(buf, sizeof(buf), fp)) {
2227         char* start = strstr(buf, search_string);
2228         if (start != NULL) {
2229           char *ptr = start + strlen(search_string);
2230           char *end = buf + strlen(buf);
2231           while (ptr != end) {
2232              // skip whitespace and colon for the rest of the name.
2233              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2234                break;
2235              }
2236              ptr++;
2237           }
2238           if (ptr != end) {
2239             // reasonable string, get rid of newline and keep the rest
2240             char* nl = strchr(buf, '\n');
2241             if (nl != NULL) *nl = '\0';
2242             strncpy(cpuinfo, ptr, length);
2243             fclose(fp);
2244             return;
2245           }
2246         }
2247       }
2248     }
2249   }
2250   // cpuinfo not found or parsing failed, just print generic string.  The entire
2251   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2252   strncpy(cpuinfo, IA32_ONLY("x86 IA32") AMD64_ONLY("x86 IA64") 
2253                    IA64_ONLY("IA64") SPARC_ONLY("sparcv9")
2254                    ARM32_ONLY("ARM32") PPC_ONLY("PPC64") AARCH64_ONLY("AArch64"), length);
2255 }
2256 
2257 void os::print_siginfo(outputStream* st, void* siginfo) {
2258   const siginfo_t* si = (const siginfo_t*)siginfo;
2259 
2260   os::Posix::print_siginfo_brief(st, si);
2261 #if INCLUDE_CDS
2262   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2263       UseSharedSpaces) {
2264     FileMapInfo* mapinfo = FileMapInfo::current_info();
2265     if (mapinfo->is_in_shared_space(si->si_addr)) {
2266       st->print("\n\nError accessing class data sharing archive."   \
2267                 " Mapped file inaccessible during execution, "      \
2268                 " possible disk/network problem.");
2269     }
2270   }
2271 #endif
2272   st->cr();
2273 }
2274 
2275 
2276 static void print_signal_handler(outputStream* st, int sig,
2277                                  char* buf, size_t buflen);
2278 
2279 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2280   st->print_cr("Signal Handlers:");
2281   print_signal_handler(st, SIGSEGV, buf, buflen);
2282   print_signal_handler(st, SIGBUS , buf, buflen);
2283   print_signal_handler(st, SIGFPE , buf, buflen);
2284   print_signal_handler(st, SIGPIPE, buf, buflen);
2285   print_signal_handler(st, SIGXFSZ, buf, buflen);
2286   print_signal_handler(st, SIGILL , buf, buflen);
2287   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2288   print_signal_handler(st, SR_signum, buf, buflen);
2289   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2290   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2291   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2292   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2293 #if defined(PPC64)
2294   print_signal_handler(st, SIGTRAP, buf, buflen);
2295 #endif
2296 }
2297 
2298 static char saved_jvm_path[MAXPATHLEN] = {0};
2299 
2300 // Find the full path to the current module, libjvm.so
2301 void os::jvm_path(char *buf, jint buflen) {
2302   // Error checking.
2303   if (buflen < MAXPATHLEN) {
2304     assert(false, "must use a large-enough buffer");
2305     buf[0] = '\0';
2306     return;
2307   }
2308   // Lazy resolve the path to current module.
2309   if (saved_jvm_path[0] != 0) {
2310     strcpy(buf, saved_jvm_path);
2311     return;
2312   }
2313 
2314   char dli_fname[MAXPATHLEN];
2315   bool ret = dll_address_to_library_name(
2316                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2317                                          dli_fname, sizeof(dli_fname), NULL);
2318   assert(ret, "cannot locate libjvm");
2319   char *rp = NULL;
2320   if (ret && dli_fname[0] != '\0') {
2321     rp = realpath(dli_fname, buf);
2322   }
2323   if (rp == NULL) {
2324     return;
2325   }
2326 
2327   if (Arguments::sun_java_launcher_is_altjvm()) {
2328     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2329     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2330     // If "/jre/lib/" appears at the right place in the string, then
2331     // assume we are installed in a JDK and we're done. Otherwise, check
2332     // for a JAVA_HOME environment variable and fix up the path so it
2333     // looks like libjvm.so is installed there (append a fake suffix
2334     // hotspot/libjvm.so).
2335     const char *p = buf + strlen(buf) - 1;
2336     for (int count = 0; p > buf && count < 5; ++count) {
2337       for (--p; p > buf && *p != '/'; --p)
2338         /* empty */ ;
2339     }
2340 
2341     if (strncmp(p, "/jre/lib/", 9) != 0) {
2342       // Look for JAVA_HOME in the environment.
2343       char* java_home_var = ::getenv("JAVA_HOME");
2344       if (java_home_var != NULL && java_home_var[0] != 0) {
2345         char* jrelib_p;
2346         int len;
2347 
2348         // Check the current module name "libjvm.so".
2349         p = strrchr(buf, '/');
2350         if (p == NULL) {
2351           return;
2352         }
2353         assert(strstr(p, "/libjvm") == p, "invalid library name");
2354 
2355         rp = realpath(java_home_var, buf);
2356         if (rp == NULL) {
2357           return;
2358         }
2359 
2360         // determine if this is a legacy image or modules image
2361         // modules image doesn't have "jre" subdirectory
2362         len = strlen(buf);
2363         assert(len < buflen, "Ran out of buffer room");
2364         jrelib_p = buf + len;
2365         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2366         if (0 != access(buf, F_OK)) {
2367           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2368         }
2369 
2370         if (0 == access(buf, F_OK)) {
2371           // Use current module name "libjvm.so"
2372           len = strlen(buf);
2373           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2374         } else {
2375           // Go back to path of .so
2376           rp = realpath(dli_fname, buf);
2377           if (rp == NULL) {
2378             return;
2379           }
2380         }
2381       }
2382     }
2383   }
2384 
2385   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2386   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2387 }
2388 
2389 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2390   // no prefix required, not even "_"
2391 }
2392 
2393 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2394   // no suffix required
2395 }
2396 
2397 ////////////////////////////////////////////////////////////////////////////////
2398 // sun.misc.Signal support
2399 
2400 static volatile jint sigint_count = 0;
2401 
2402 static void UserHandler(int sig, void *siginfo, void *context) {
2403   // 4511530 - sem_post is serialized and handled by the manager thread. When
2404   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2405   // don't want to flood the manager thread with sem_post requests.
2406   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2407     return;
2408   }
2409 
2410   // Ctrl-C is pressed during error reporting, likely because the error
2411   // handler fails to abort. Let VM die immediately.
2412   if (sig == SIGINT && is_error_reported()) {
2413     os::die();
2414   }
2415 
2416   os::signal_notify(sig);
2417 }
2418 
2419 void* os::user_handler() {
2420   return CAST_FROM_FN_PTR(void*, UserHandler);
2421 }
2422 
2423 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2424   struct timespec ts;
2425   // Semaphore's are always associated with CLOCK_REALTIME
2426   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2427   // see unpackTime for discussion on overflow checking
2428   if (sec >= MAX_SECS) {
2429     ts.tv_sec += MAX_SECS;
2430     ts.tv_nsec = 0;
2431   } else {
2432     ts.tv_sec += sec;
2433     ts.tv_nsec += nsec;
2434     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2435       ts.tv_nsec -= NANOSECS_PER_SEC;
2436       ++ts.tv_sec; // note: this must be <= max_secs
2437     }
2438   }
2439 
2440   return ts;
2441 }
2442 
2443 extern "C" {
2444   typedef void (*sa_handler_t)(int);
2445   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2446 }
2447 
2448 void* os::signal(int signal_number, void* handler) {
2449   struct sigaction sigAct, oldSigAct;
2450 
2451   sigfillset(&(sigAct.sa_mask));
2452   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2453   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2454 
2455   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2456     // -1 means registration failed
2457     return (void *)-1;
2458   }
2459 
2460   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2461 }
2462 
2463 void os::signal_raise(int signal_number) {
2464   ::raise(signal_number);
2465 }
2466 
2467 // The following code is moved from os.cpp for making this
2468 // code platform specific, which it is by its very nature.
2469 
2470 // Will be modified when max signal is changed to be dynamic
2471 int os::sigexitnum_pd() {
2472   return NSIG;
2473 }
2474 
2475 // a counter for each possible signal value
2476 static volatile jint pending_signals[NSIG+1] = { 0 };
2477 
2478 // Linux(POSIX) specific hand shaking semaphore.
2479 static sem_t sig_sem;
2480 static PosixSemaphore sr_semaphore;
2481 
2482 void os::signal_init_pd() {
2483   // Initialize signal structures
2484   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2485 
2486   // Initialize signal semaphore
2487   ::sem_init(&sig_sem, 0, 0);
2488 }
2489 
2490 void os::signal_notify(int sig) {
2491   Atomic::inc(&pending_signals[sig]);
2492   ::sem_post(&sig_sem);
2493 }
2494 
2495 static int check_pending_signals(bool wait) {
2496   Atomic::store(0, &sigint_count);
2497   for (;;) {
2498     for (int i = 0; i < NSIG + 1; i++) {
2499       jint n = pending_signals[i];
2500       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2501         return i;
2502       }
2503     }
2504     if (!wait) {
2505       return -1;
2506     }
2507     JavaThread *thread = JavaThread::current();
2508     ThreadBlockInVM tbivm(thread);
2509 
2510     bool threadIsSuspended;
2511     do {
2512       thread->set_suspend_equivalent();
2513       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2514       ::sem_wait(&sig_sem);
2515 
2516       // were we externally suspended while we were waiting?
2517       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2518       if (threadIsSuspended) {
2519         // The semaphore has been incremented, but while we were waiting
2520         // another thread suspended us. We don't want to continue running
2521         // while suspended because that would surprise the thread that
2522         // suspended us.
2523         ::sem_post(&sig_sem);
2524 
2525         thread->java_suspend_self();
2526       }
2527     } while (threadIsSuspended);
2528   }
2529 }
2530 
2531 int os::signal_lookup() {
2532   return check_pending_signals(false);
2533 }
2534 
2535 int os::signal_wait() {
2536   return check_pending_signals(true);
2537 }
2538 
2539 ////////////////////////////////////////////////////////////////////////////////
2540 // Virtual Memory
2541 
2542 int os::vm_page_size() {
2543   // Seems redundant as all get out
2544   assert(os::Linux::page_size() != -1, "must call os::init");
2545   return os::Linux::page_size();
2546 }
2547 
2548 // Solaris allocates memory by pages.
2549 int os::vm_allocation_granularity() {
2550   assert(os::Linux::page_size() != -1, "must call os::init");
2551   return os::Linux::page_size();
2552 }
2553 
2554 // Rationale behind this function:
2555 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2556 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2557 //  samples for JITted code. Here we create private executable mapping over the code cache
2558 //  and then we can use standard (well, almost, as mapping can change) way to provide
2559 //  info for the reporting script by storing timestamp and location of symbol
2560 void linux_wrap_code(char* base, size_t size) {
2561   static volatile jint cnt = 0;
2562 
2563   if (!UseOprofile) {
2564     return;
2565   }
2566 
2567   char buf[PATH_MAX+1];
2568   int num = Atomic::add(1, &cnt);
2569 
2570   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2571            os::get_temp_directory(), os::current_process_id(), num);
2572   unlink(buf);
2573 
2574   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2575 
2576   if (fd != -1) {
2577     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2578     if (rv != (off_t)-1) {
2579       if (::write(fd, "", 1) == 1) {
2580         mmap(base, size,
2581              PROT_READ|PROT_WRITE|PROT_EXEC,
2582              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2583       }
2584     }
2585     ::close(fd);
2586     unlink(buf);
2587   }
2588 }
2589 
2590 static bool recoverable_mmap_error(int err) {
2591   // See if the error is one we can let the caller handle. This
2592   // list of errno values comes from JBS-6843484. I can't find a
2593   // Linux man page that documents this specific set of errno
2594   // values so while this list currently matches Solaris, it may
2595   // change as we gain experience with this failure mode.
2596   switch (err) {
2597   case EBADF:
2598   case EINVAL:
2599   case ENOTSUP:
2600     // let the caller deal with these errors
2601     return true;
2602 
2603   default:
2604     // Any remaining errors on this OS can cause our reserved mapping
2605     // to be lost. That can cause confusion where different data
2606     // structures think they have the same memory mapped. The worst
2607     // scenario is if both the VM and a library think they have the
2608     // same memory mapped.
2609     return false;
2610   }
2611 }
2612 
2613 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2614                                     int err) {
2615   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2616           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2617           strerror(err), err);
2618 }
2619 
2620 static void warn_fail_commit_memory(char* addr, size_t size,
2621                                     size_t alignment_hint, bool exec,
2622                                     int err) {
2623   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2624           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2625           alignment_hint, exec, strerror(err), err);
2626 }
2627 
2628 // NOTE: Linux kernel does not really reserve the pages for us.
2629 //       All it does is to check if there are enough free pages
2630 //       left at the time of mmap(). This could be a potential
2631 //       problem.
2632 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2633   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2634   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2635                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2636   if (res != (uintptr_t) MAP_FAILED) {
2637     if (UseNUMAInterleaving) {
2638       numa_make_global(addr, size);
2639     }
2640     return 0;
2641   }
2642 
2643   int err = errno;  // save errno from mmap() call above
2644 
2645   if (!recoverable_mmap_error(err)) {
2646     warn_fail_commit_memory(addr, size, exec, err);
2647     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2648   }
2649 
2650   return err;
2651 }
2652 
2653 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2654   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2655 }
2656 
2657 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2658                                   const char* mesg) {
2659   assert(mesg != NULL, "mesg must be specified");
2660   int err = os::Linux::commit_memory_impl(addr, size, exec);
2661   if (err != 0) {
2662     // the caller wants all commit errors to exit with the specified mesg:
2663     warn_fail_commit_memory(addr, size, exec, err);
2664     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2665   }
2666 }
2667 
2668 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2669 #ifndef MAP_HUGETLB
2670   #define MAP_HUGETLB 0x40000
2671 #endif
2672 
2673 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2674 #ifndef MADV_HUGEPAGE
2675   #define MADV_HUGEPAGE 14
2676 #endif
2677 
2678 int os::Linux::commit_memory_impl(char* addr, size_t size,
2679                                   size_t alignment_hint, bool exec) {
2680   int err = os::Linux::commit_memory_impl(addr, size, exec);
2681   if (err == 0) {
2682     realign_memory(addr, size, alignment_hint);
2683   }
2684   return err;
2685 }
2686 
2687 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2688                           bool exec) {
2689   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2690 }
2691 
2692 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2693                                   size_t alignment_hint, bool exec,
2694                                   const char* mesg) {
2695   assert(mesg != NULL, "mesg must be specified");
2696   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2697   if (err != 0) {
2698     // the caller wants all commit errors to exit with the specified mesg:
2699     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2700     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2701   }
2702 }
2703 
2704 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2705   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2706     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2707     // be supported or the memory may already be backed by huge pages.
2708     ::madvise(addr, bytes, MADV_HUGEPAGE);
2709   }
2710 }
2711 
2712 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2713   // This method works by doing an mmap over an existing mmaping and effectively discarding
2714   // the existing pages. However it won't work for SHM-based large pages that cannot be
2715   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2716   // small pages on top of the SHM segment. This method always works for small pages, so we
2717   // allow that in any case.
2718   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2719     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2720   }
2721 }
2722 
2723 void os::numa_make_global(char *addr, size_t bytes) {
2724   Linux::numa_interleave_memory(addr, bytes);
2725 }
2726 
2727 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2728 // bind policy to MPOL_PREFERRED for the current thread.
2729 #define USE_MPOL_PREFERRED 0
2730 
2731 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2732   // To make NUMA and large pages more robust when both enabled, we need to ease
2733   // the requirements on where the memory should be allocated. MPOL_BIND is the
2734   // default policy and it will force memory to be allocated on the specified
2735   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2736   // the specified node, but will not force it. Using this policy will prevent
2737   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2738   // free large pages.
2739   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2740   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2741 }
2742 
2743 bool os::numa_topology_changed() { return false; }
2744 
2745 size_t os::numa_get_groups_num() {
2746   int max_node = Linux::numa_max_node();
2747   return max_node > 0 ? max_node + 1 : 1;
2748 }
2749 
2750 int os::numa_get_group_id() {
2751   int cpu_id = Linux::sched_getcpu();
2752   if (cpu_id != -1) {
2753     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2754     if (lgrp_id != -1) {
2755       return lgrp_id;
2756     }
2757   }
2758   return 0;
2759 }
2760 
2761 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2762   for (size_t i = 0; i < size; i++) {
2763     ids[i] = i;
2764   }
2765   return size;
2766 }
2767 
2768 bool os::get_page_info(char *start, page_info* info) {
2769   return false;
2770 }
2771 
2772 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2773                      page_info* page_found) {
2774   return end;
2775 }
2776 
2777 
2778 int os::Linux::sched_getcpu_syscall(void) {
2779   unsigned int cpu;
2780   int retval = -1;
2781 
2782 #if defined(IA32)
2783   #ifndef SYS_getcpu
2784     #define SYS_getcpu 318
2785   #endif
2786   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2787 #elif defined(AMD64)
2788 // Unfortunately we have to bring all these macros here from vsyscall.h
2789 // to be able to compile on old linuxes.
2790   #define __NR_vgetcpu 2
2791   #define VSYSCALL_START (-10UL << 20)
2792   #define VSYSCALL_SIZE 1024
2793   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2794   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2795   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2796   retval = vgetcpu(&cpu, NULL, NULL);
2797 #endif
2798 
2799   return (retval == -1) ? retval : cpu;
2800 }
2801 
2802 // Something to do with the numa-aware allocator needs these symbols
2803 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2804 extern "C" JNIEXPORT void numa_error(char *where) { }
2805 extern "C" JNIEXPORT int fork1() { return fork(); }
2806 
2807 
2808 // If we are running with libnuma version > 2, then we should
2809 // be trying to use symbols with versions 1.1
2810 // If we are running with earlier version, which did not have symbol versions,
2811 // we should use the base version.
2812 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2813   void *f = dlvsym(handle, name, "libnuma_1.1");
2814   if (f == NULL) {
2815     f = dlsym(handle, name);
2816   }
2817   return f;
2818 }
2819 
2820 bool os::Linux::libnuma_init() {
2821   // sched_getcpu() should be in libc.
2822   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2823                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2824 
2825   // If it's not, try a direct syscall.
2826   if (sched_getcpu() == -1) {
2827     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2828                                     (void*)&sched_getcpu_syscall));
2829   }
2830 
2831   if (sched_getcpu() != -1) { // Does it work?
2832     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2833     if (handle != NULL) {
2834       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2835                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2836       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2837                                        libnuma_dlsym(handle, "numa_max_node")));
2838       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2839                                         libnuma_dlsym(handle, "numa_available")));
2840       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2841                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2842       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2843                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2844       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2845                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2846 
2847 
2848       if (numa_available() != -1) {
2849         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2850         // Create a cpu -> node mapping
2851         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2852         rebuild_cpu_to_node_map();
2853         return true;
2854       }
2855     }
2856   }
2857   return false;
2858 }
2859 
2860 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2861 // The table is later used in get_node_by_cpu().
2862 void os::Linux::rebuild_cpu_to_node_map() {
2863   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2864                               // in libnuma (possible values are starting from 16,
2865                               // and continuing up with every other power of 2, but less
2866                               // than the maximum number of CPUs supported by kernel), and
2867                               // is a subject to change (in libnuma version 2 the requirements
2868                               // are more reasonable) we'll just hardcode the number they use
2869                               // in the library.
2870   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2871 
2872   size_t cpu_num = os::active_processor_count();
2873   size_t cpu_map_size = NCPUS / BitsPerCLong;
2874   size_t cpu_map_valid_size =
2875     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2876 
2877   cpu_to_node()->clear();
2878   cpu_to_node()->at_grow(cpu_num - 1);
2879   size_t node_num = numa_get_groups_num();
2880 
2881   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2882   for (size_t i = 0; i < node_num; i++) {
2883     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2884       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2885         if (cpu_map[j] != 0) {
2886           for (size_t k = 0; k < BitsPerCLong; k++) {
2887             if (cpu_map[j] & (1UL << k)) {
2888               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2889             }
2890           }
2891         }
2892       }
2893     }
2894   }
2895   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2896 }
2897 
2898 int os::Linux::get_node_by_cpu(int cpu_id) {
2899   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2900     return cpu_to_node()->at(cpu_id);
2901   }
2902   return -1;
2903 }
2904 
2905 GrowableArray<int>* os::Linux::_cpu_to_node;
2906 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2907 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2908 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2909 os::Linux::numa_available_func_t os::Linux::_numa_available;
2910 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2911 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2912 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2913 unsigned long* os::Linux::_numa_all_nodes;
2914 
2915 bool os::pd_uncommit_memory(char* addr, size_t size) {
2916   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2917                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2918   return res  != (uintptr_t) MAP_FAILED;
2919 }
2920 
2921 static address get_stack_commited_bottom(address bottom, size_t size) {
2922   address nbot = bottom;
2923   address ntop = bottom + size;
2924 
2925   size_t page_sz = os::vm_page_size();
2926   unsigned pages = size / page_sz;
2927 
2928   unsigned char vec[1];
2929   unsigned imin = 1, imax = pages + 1, imid;
2930   int mincore_return_value = 0;
2931 
2932   assert(imin <= imax, "Unexpected page size");
2933 
2934   while (imin < imax) {
2935     imid = (imax + imin) / 2;
2936     nbot = ntop - (imid * page_sz);
2937 
2938     // Use a trick with mincore to check whether the page is mapped or not.
2939     // mincore sets vec to 1 if page resides in memory and to 0 if page
2940     // is swapped output but if page we are asking for is unmapped
2941     // it returns -1,ENOMEM
2942     mincore_return_value = mincore(nbot, page_sz, vec);
2943 
2944     if (mincore_return_value == -1) {
2945       // Page is not mapped go up
2946       // to find first mapped page
2947       if (errno != EAGAIN) {
2948         assert(errno == ENOMEM, "Unexpected mincore errno");
2949         imax = imid;
2950       }
2951     } else {
2952       // Page is mapped go down
2953       // to find first not mapped page
2954       imin = imid + 1;
2955     }
2956   }
2957 
2958   nbot = nbot + page_sz;
2959 
2960   // Adjust stack bottom one page up if last checked page is not mapped
2961   if (mincore_return_value == -1) {
2962     nbot = nbot + page_sz;
2963   }
2964 
2965   return nbot;
2966 }
2967 
2968 
2969 // Linux uses a growable mapping for the stack, and if the mapping for
2970 // the stack guard pages is not removed when we detach a thread the
2971 // stack cannot grow beyond the pages where the stack guard was
2972 // mapped.  If at some point later in the process the stack expands to
2973 // that point, the Linux kernel cannot expand the stack any further
2974 // because the guard pages are in the way, and a segfault occurs.
2975 //
2976 // However, it's essential not to split the stack region by unmapping
2977 // a region (leaving a hole) that's already part of the stack mapping,
2978 // so if the stack mapping has already grown beyond the guard pages at
2979 // the time we create them, we have to truncate the stack mapping.
2980 // So, we need to know the extent of the stack mapping when
2981 // create_stack_guard_pages() is called.
2982 
2983 // We only need this for stacks that are growable: at the time of
2984 // writing thread stacks don't use growable mappings (i.e. those
2985 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2986 // only applies to the main thread.
2987 
2988 // If the (growable) stack mapping already extends beyond the point
2989 // where we're going to put our guard pages, truncate the mapping at
2990 // that point by munmap()ping it.  This ensures that when we later
2991 // munmap() the guard pages we don't leave a hole in the stack
2992 // mapping. This only affects the main/initial thread
2993 
2994 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2995   if (os::Linux::is_initial_thread()) {
2996     // As we manually grow stack up to bottom inside create_attached_thread(),
2997     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2998     // we don't need to do anything special.
2999     // Check it first, before calling heavy function.
3000     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3001     unsigned char vec[1];
3002 
3003     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3004       // Fallback to slow path on all errors, including EAGAIN
3005       stack_extent = (uintptr_t) get_stack_commited_bottom(
3006                                                            os::Linux::initial_thread_stack_bottom(),
3007                                                            (size_t)addr - stack_extent);
3008     }
3009 
3010     if (stack_extent < (uintptr_t)addr) {
3011       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3012     }
3013   }
3014 
3015   return os::commit_memory(addr, size, !ExecMem);
3016 }
3017 
3018 // If this is a growable mapping, remove the guard pages entirely by
3019 // munmap()ping them.  If not, just call uncommit_memory(). This only
3020 // affects the main/initial thread, but guard against future OS changes
3021 // It's safe to always unmap guard pages for initial thread because we
3022 // always place it right after end of the mapped region
3023 
3024 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3025   uintptr_t stack_extent, stack_base;
3026 
3027   if (os::Linux::is_initial_thread()) {
3028     return ::munmap(addr, size) == 0;
3029   }
3030 
3031   return os::uncommit_memory(addr, size);
3032 }
3033 
3034 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3035 // at 'requested_addr'. If there are existing memory mappings at the same
3036 // location, however, they will be overwritten. If 'fixed' is false,
3037 // 'requested_addr' is only treated as a hint, the return value may or
3038 // may not start from the requested address. Unlike Linux mmap(), this
3039 // function returns NULL to indicate failure.
3040 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3041   char * addr;
3042   int flags;
3043 
3044   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3045   if (fixed) {
3046     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3047     flags |= MAP_FIXED;
3048   }
3049 
3050   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3051   // touch an uncommitted page. Otherwise, the read/write might
3052   // succeed if we have enough swap space to back the physical page.
3053   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3054                        flags, -1, 0);
3055 
3056   return addr == MAP_FAILED ? NULL : addr;
3057 }
3058 
3059 static int anon_munmap(char * addr, size_t size) {
3060   return ::munmap(addr, size) == 0;
3061 }
3062 
3063 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3064                             size_t alignment_hint) {
3065   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3066 }
3067 
3068 bool os::pd_release_memory(char* addr, size_t size) {
3069   return anon_munmap(addr, size);
3070 }
3071 
3072 static bool linux_mprotect(char* addr, size_t size, int prot) {
3073   // Linux wants the mprotect address argument to be page aligned.
3074   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3075 
3076   // According to SUSv3, mprotect() should only be used with mappings
3077   // established by mmap(), and mmap() always maps whole pages. Unaligned
3078   // 'addr' likely indicates problem in the VM (e.g. trying to change
3079   // protection of malloc'ed or statically allocated memory). Check the
3080   // caller if you hit this assert.
3081   assert(addr == bottom, "sanity check");
3082 
3083   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3084   return ::mprotect(bottom, size, prot) == 0;
3085 }
3086 
3087 // Set protections specified
3088 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3089                         bool is_committed) {
3090   unsigned int p = 0;
3091   switch (prot) {
3092   case MEM_PROT_NONE: p = PROT_NONE; break;
3093   case MEM_PROT_READ: p = PROT_READ; break;
3094   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3095   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3096   default:
3097     ShouldNotReachHere();
3098   }
3099   // is_committed is unused.
3100   return linux_mprotect(addr, bytes, p);
3101 }
3102 
3103 bool os::guard_memory(char* addr, size_t size) {
3104   return linux_mprotect(addr, size, PROT_NONE);
3105 }
3106 
3107 bool os::unguard_memory(char* addr, size_t size) {
3108   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3109 }
3110 
3111 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3112                                                     size_t page_size) {
3113   bool result = false;
3114   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3115                  MAP_ANONYMOUS|MAP_PRIVATE,
3116                  -1, 0);
3117   if (p != MAP_FAILED) {
3118     void *aligned_p = align_ptr_up(p, page_size);
3119 
3120     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3121 
3122     munmap(p, page_size * 2);
3123   }
3124 
3125   if (warn && !result) {
3126     warning("TransparentHugePages is not supported by the operating system.");
3127   }
3128 
3129   return result;
3130 }
3131 
3132 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3133   bool result = false;
3134   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3135                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3136                  -1, 0);
3137 
3138   if (p != MAP_FAILED) {
3139     // We don't know if this really is a huge page or not.
3140     FILE *fp = fopen("/proc/self/maps", "r");
3141     if (fp) {
3142       while (!feof(fp)) {
3143         char chars[257];
3144         long x = 0;
3145         if (fgets(chars, sizeof(chars), fp)) {
3146           if (sscanf(chars, "%lx-%*x", &x) == 1
3147               && x == (long)p) {
3148             if (strstr (chars, "hugepage")) {
3149               result = true;
3150               break;
3151             }
3152           }
3153         }
3154       }
3155       fclose(fp);
3156     }
3157     munmap(p, page_size);
3158   }
3159 
3160   if (warn && !result) {
3161     warning("HugeTLBFS is not supported by the operating system.");
3162   }
3163 
3164   return result;
3165 }
3166 
3167 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3168 //
3169 // From the coredump_filter documentation:
3170 //
3171 // - (bit 0) anonymous private memory
3172 // - (bit 1) anonymous shared memory
3173 // - (bit 2) file-backed private memory
3174 // - (bit 3) file-backed shared memory
3175 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3176 //           effective only if the bit 2 is cleared)
3177 // - (bit 5) hugetlb private memory
3178 // - (bit 6) hugetlb shared memory
3179 //
3180 static void set_coredump_filter(void) {
3181   FILE *f;
3182   long cdm;
3183 
3184   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3185     return;
3186   }
3187 
3188   if (fscanf(f, "%lx", &cdm) != 1) {
3189     fclose(f);
3190     return;
3191   }
3192 
3193   rewind(f);
3194 
3195   if ((cdm & LARGEPAGES_BIT) == 0) {
3196     cdm |= LARGEPAGES_BIT;
3197     fprintf(f, "%#lx", cdm);
3198   }
3199 
3200   fclose(f);
3201 }
3202 
3203 // Large page support
3204 
3205 static size_t _large_page_size = 0;
3206 
3207 size_t os::Linux::find_large_page_size() {
3208   size_t large_page_size = 0;
3209 
3210   // large_page_size on Linux is used to round up heap size. x86 uses either
3211   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3212   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3213   // page as large as 256M.
3214   //
3215   // Here we try to figure out page size by parsing /proc/meminfo and looking
3216   // for a line with the following format:
3217   //    Hugepagesize:     2048 kB
3218   //
3219   // If we can't determine the value (e.g. /proc is not mounted, or the text
3220   // format has been changed), we'll use the largest page size supported by
3221   // the processor.
3222 
3223 #ifndef ZERO
3224   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3225                      ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3226 #endif // ZERO
3227 
3228   FILE *fp = fopen("/proc/meminfo", "r");
3229   if (fp) {
3230     while (!feof(fp)) {
3231       int x = 0;
3232       char buf[16];
3233       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3234         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3235           large_page_size = x * K;
3236           break;
3237         }
3238       } else {
3239         // skip to next line
3240         for (;;) {
3241           int ch = fgetc(fp);
3242           if (ch == EOF || ch == (int)'\n') break;
3243         }
3244       }
3245     }
3246     fclose(fp);
3247   }
3248 
3249   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3250     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3251             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3252             proper_unit_for_byte_size(large_page_size));
3253   }
3254 
3255   return large_page_size;
3256 }
3257 
3258 size_t os::Linux::setup_large_page_size() {
3259   _large_page_size = Linux::find_large_page_size();
3260   const size_t default_page_size = (size_t)Linux::page_size();
3261   if (_large_page_size > default_page_size) {
3262     _page_sizes[0] = _large_page_size;
3263     _page_sizes[1] = default_page_size;
3264     _page_sizes[2] = 0;
3265   }
3266 
3267   return _large_page_size;
3268 }
3269 
3270 bool os::Linux::setup_large_page_type(size_t page_size) {
3271   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3272       FLAG_IS_DEFAULT(UseSHM) &&
3273       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3274 
3275     // The type of large pages has not been specified by the user.
3276 
3277     // Try UseHugeTLBFS and then UseSHM.
3278     UseHugeTLBFS = UseSHM = true;
3279 
3280     // Don't try UseTransparentHugePages since there are known
3281     // performance issues with it turned on. This might change in the future.
3282     UseTransparentHugePages = false;
3283   }
3284 
3285   if (UseTransparentHugePages) {
3286     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3287     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3288       UseHugeTLBFS = false;
3289       UseSHM = false;
3290       return true;
3291     }
3292     UseTransparentHugePages = false;
3293   }
3294 
3295   if (UseHugeTLBFS) {
3296     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3297     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3298       UseSHM = false;
3299       return true;
3300     }
3301     UseHugeTLBFS = false;
3302   }
3303 
3304   return UseSHM;
3305 }
3306 
3307 void os::large_page_init() {
3308   if (!UseLargePages &&
3309       !UseTransparentHugePages &&
3310       !UseHugeTLBFS &&
3311       !UseSHM) {
3312     // Not using large pages.
3313     return;
3314   }
3315 
3316   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3317     // The user explicitly turned off large pages.
3318     // Ignore the rest of the large pages flags.
3319     UseTransparentHugePages = false;
3320     UseHugeTLBFS = false;
3321     UseSHM = false;
3322     return;
3323   }
3324 
3325   size_t large_page_size = Linux::setup_large_page_size();
3326   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3327 
3328   set_coredump_filter();
3329 }
3330 
3331 #ifndef SHM_HUGETLB
3332   #define SHM_HUGETLB 04000
3333 #endif
3334 
3335 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3336                                             char* req_addr, bool exec) {
3337   // "exec" is passed in but not used.  Creating the shared image for
3338   // the code cache doesn't have an SHM_X executable permission to check.
3339   assert(UseLargePages && UseSHM, "only for SHM large pages");
3340   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3341 
3342   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3343     return NULL; // Fallback to small pages.
3344   }
3345 
3346   key_t key = IPC_PRIVATE;
3347   char *addr;
3348 
3349   bool warn_on_failure = UseLargePages &&
3350                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3351                          !FLAG_IS_DEFAULT(UseSHM) ||
3352                          !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3353   char msg[128];
3354 
3355   // Create a large shared memory region to attach to based on size.
3356   // Currently, size is the total size of the heap
3357   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3358   if (shmid == -1) {
3359     // Possible reasons for shmget failure:
3360     // 1. shmmax is too small for Java heap.
3361     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3362     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3363     // 2. not enough large page memory.
3364     //    > check available large pages: cat /proc/meminfo
3365     //    > increase amount of large pages:
3366     //          echo new_value > /proc/sys/vm/nr_hugepages
3367     //      Note 1: different Linux may use different name for this property,
3368     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3369     //      Note 2: it's possible there's enough physical memory available but
3370     //            they are so fragmented after a long run that they can't
3371     //            coalesce into large pages. Try to reserve large pages when
3372     //            the system is still "fresh".
3373     if (warn_on_failure) {
3374       jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3375       warning("%s", msg);
3376     }
3377     return NULL;
3378   }
3379 
3380   // attach to the region
3381   addr = (char*)shmat(shmid, req_addr, 0);
3382   int err = errno;
3383 
3384   // Remove shmid. If shmat() is successful, the actual shared memory segment
3385   // will be deleted when it's detached by shmdt() or when the process
3386   // terminates. If shmat() is not successful this will remove the shared
3387   // segment immediately.
3388   shmctl(shmid, IPC_RMID, NULL);
3389 
3390   if ((intptr_t)addr == -1) {
3391     if (warn_on_failure) {
3392       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3393       warning("%s", msg);
3394     }
3395     return NULL;
3396   }
3397 
3398   return addr;
3399 }
3400 
3401 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3402                                         int error) {
3403   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3404 
3405   bool warn_on_failure = UseLargePages &&
3406       (!FLAG_IS_DEFAULT(UseLargePages) ||
3407        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3408        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3409 
3410   if (warn_on_failure) {
3411     char msg[128];
3412     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3413                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3414     warning("%s", msg);
3415   }
3416 }
3417 
3418 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3419                                                         char* req_addr,
3420                                                         bool exec) {
3421   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3422   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3423   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3424 
3425   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3426   char* addr = (char*)::mmap(req_addr, bytes, prot,
3427                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3428                              -1, 0);
3429 
3430   if (addr == MAP_FAILED) {
3431     warn_on_large_pages_failure(req_addr, bytes, errno);
3432     return NULL;
3433   }
3434 
3435   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3436 
3437   return addr;
3438 }
3439 
3440 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3441 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3442 //   (req_addr != NULL) or with a given alignment.
3443 //  - bytes shall be a multiple of alignment.
3444 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3445 //  - alignment sets the alignment at which memory shall be allocated.
3446 //     It must be a multiple of allocation granularity.
3447 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3448 //  req_addr or NULL.
3449 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3450 
3451   size_t extra_size = bytes;
3452   if (req_addr == NULL && alignment > 0) {
3453     extra_size += alignment;
3454   }
3455 
3456   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3457     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3458     -1, 0);
3459   if (start == MAP_FAILED) {
3460     start = NULL;
3461   } else {
3462     if (req_addr != NULL) {
3463       if (start != req_addr) {
3464         ::munmap(start, extra_size);
3465         start = NULL;
3466       }
3467     } else {
3468       char* const start_aligned = (char*) align_ptr_up(start, alignment);
3469       char* const end_aligned = start_aligned + bytes;
3470       char* const end = start + extra_size;
3471       if (start_aligned > start) {
3472         ::munmap(start, start_aligned - start);
3473       }
3474       if (end_aligned < end) {
3475         ::munmap(end_aligned, end - end_aligned);
3476       }
3477       start = start_aligned;
3478     }
3479   }
3480   return start;
3481 
3482 }
3483 
3484 // Reserve memory using mmap(MAP_HUGETLB).
3485 //  - bytes shall be a multiple of alignment.
3486 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3487 //  - alignment sets the alignment at which memory shall be allocated.
3488 //     It must be a multiple of allocation granularity.
3489 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3490 //  req_addr or NULL.
3491 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3492                                                          size_t alignment,
3493                                                          char* req_addr,
3494                                                          bool exec) {
3495   size_t large_page_size = os::large_page_size();
3496   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3497 
3498   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3499   assert(is_size_aligned(bytes, alignment), "Must be");
3500 
3501   // First reserve - but not commit - the address range in small pages.
3502   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3503 
3504   if (start == NULL) {
3505     return NULL;
3506   }
3507 
3508   assert(is_ptr_aligned(start, alignment), "Must be");
3509 
3510   char* end = start + bytes;
3511 
3512   // Find the regions of the allocated chunk that can be promoted to large pages.
3513   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3514   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3515 
3516   size_t lp_bytes = lp_end - lp_start;
3517 
3518   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3519 
3520   if (lp_bytes == 0) {
3521     // The mapped region doesn't even span the start and the end of a large page.
3522     // Fall back to allocate a non-special area.
3523     ::munmap(start, end - start);
3524     return NULL;
3525   }
3526 
3527   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3528 
3529   void* result;
3530 
3531   // Commit small-paged leading area.
3532   if (start != lp_start) {
3533     result = ::mmap(start, lp_start - start, prot,
3534                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3535                     -1, 0);
3536     if (result == MAP_FAILED) {
3537       ::munmap(lp_start, end - lp_start);
3538       return NULL;
3539     }
3540   }
3541 
3542   // Commit large-paged area.
3543   result = ::mmap(lp_start, lp_bytes, prot,
3544                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3545                   -1, 0);
3546   if (result == MAP_FAILED) {
3547     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3548     // If the mmap above fails, the large pages region will be unmapped and we
3549     // have regions before and after with small pages. Release these regions.
3550     //
3551     // |  mapped  |  unmapped  |  mapped  |
3552     // ^          ^            ^          ^
3553     // start      lp_start     lp_end     end
3554     //
3555     ::munmap(start, lp_start - start);
3556     ::munmap(lp_end, end - lp_end);
3557     return NULL;
3558   }
3559 
3560   // Commit small-paged trailing area.
3561   if (lp_end != end) {
3562     result = ::mmap(lp_end, end - lp_end, prot,
3563                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3564                     -1, 0);
3565     if (result == MAP_FAILED) {
3566       ::munmap(start, lp_end - start);
3567       return NULL;
3568     }
3569   }
3570 
3571   return start;
3572 }
3573 
3574 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3575                                                    size_t alignment,
3576                                                    char* req_addr,
3577                                                    bool exec) {
3578   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3579   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3580   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3581   assert(is_power_of_2(os::large_page_size()), "Must be");
3582   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3583 
3584   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3585     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3586   } else {
3587     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3588   }
3589 }
3590 
3591 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3592                                  char* req_addr, bool exec) {
3593   assert(UseLargePages, "only for large pages");
3594 
3595   char* addr;
3596   if (UseSHM) {
3597     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3598   } else {
3599     assert(UseHugeTLBFS, "must be");
3600     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3601   }
3602 
3603   if (addr != NULL) {
3604     if (UseNUMAInterleaving) {
3605       numa_make_global(addr, bytes);
3606     }
3607 
3608     // The memory is committed
3609     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3610   }
3611 
3612   return addr;
3613 }
3614 
3615 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3616   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3617   return shmdt(base) == 0;
3618 }
3619 
3620 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3621   return pd_release_memory(base, bytes);
3622 }
3623 
3624 bool os::release_memory_special(char* base, size_t bytes) {
3625   bool res;
3626   if (MemTracker::tracking_level() > NMT_minimal) {
3627     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3628     res = os::Linux::release_memory_special_impl(base, bytes);
3629     if (res) {
3630       tkr.record((address)base, bytes);
3631     }
3632 
3633   } else {
3634     res = os::Linux::release_memory_special_impl(base, bytes);
3635   }
3636   return res;
3637 }
3638 
3639 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3640   assert(UseLargePages, "only for large pages");
3641   bool res;
3642 
3643   if (UseSHM) {
3644     res = os::Linux::release_memory_special_shm(base, bytes);
3645   } else {
3646     assert(UseHugeTLBFS, "must be");
3647     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3648   }
3649   return res;
3650 }
3651 
3652 size_t os::large_page_size() {
3653   return _large_page_size;
3654 }
3655 
3656 // With SysV SHM the entire memory region must be allocated as shared
3657 // memory.
3658 // HugeTLBFS allows application to commit large page memory on demand.
3659 // However, when committing memory with HugeTLBFS fails, the region
3660 // that was supposed to be committed will lose the old reservation
3661 // and allow other threads to steal that memory region. Because of this
3662 // behavior we can't commit HugeTLBFS memory.
3663 bool os::can_commit_large_page_memory() {
3664   return UseTransparentHugePages;
3665 }
3666 
3667 bool os::can_execute_large_page_memory() {
3668   return UseTransparentHugePages || UseHugeTLBFS;
3669 }
3670 
3671 // Reserve memory at an arbitrary address, only if that area is
3672 // available (and not reserved for something else).
3673 
3674 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3675   const int max_tries = 10;
3676   char* base[max_tries];
3677   size_t size[max_tries];
3678   const size_t gap = 0x000000;
3679 
3680   // Assert only that the size is a multiple of the page size, since
3681   // that's all that mmap requires, and since that's all we really know
3682   // about at this low abstraction level.  If we need higher alignment,
3683   // we can either pass an alignment to this method or verify alignment
3684   // in one of the methods further up the call chain.  See bug 5044738.
3685   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3686 
3687   // Repeatedly allocate blocks until the block is allocated at the
3688   // right spot.
3689 
3690   // Linux mmap allows caller to pass an address as hint; give it a try first,
3691   // if kernel honors the hint then we can return immediately.
3692   char * addr = anon_mmap(requested_addr, bytes, false);
3693   if (addr == requested_addr) {
3694     return requested_addr;
3695   }
3696 
3697   if (addr != NULL) {
3698     // mmap() is successful but it fails to reserve at the requested address
3699     anon_munmap(addr, bytes);
3700   }
3701 
3702   int i;
3703   for (i = 0; i < max_tries; ++i) {
3704     base[i] = reserve_memory(bytes);
3705 
3706     if (base[i] != NULL) {
3707       // Is this the block we wanted?
3708       if (base[i] == requested_addr) {
3709         size[i] = bytes;
3710         break;
3711       }
3712 
3713       // Does this overlap the block we wanted? Give back the overlapped
3714       // parts and try again.
3715 
3716       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3717       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3718         unmap_memory(base[i], top_overlap);
3719         base[i] += top_overlap;
3720         size[i] = bytes - top_overlap;
3721       } else {
3722         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3723         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3724           unmap_memory(requested_addr, bottom_overlap);
3725           size[i] = bytes - bottom_overlap;
3726         } else {
3727           size[i] = bytes;
3728         }
3729       }
3730     }
3731   }
3732 
3733   // Give back the unused reserved pieces.
3734 
3735   for (int j = 0; j < i; ++j) {
3736     if (base[j] != NULL) {
3737       unmap_memory(base[j], size[j]);
3738     }
3739   }
3740 
3741   if (i < max_tries) {
3742     return requested_addr;
3743   } else {
3744     return NULL;
3745   }
3746 }
3747 
3748 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3749   return ::read(fd, buf, nBytes);
3750 }
3751 
3752 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3753   return ::pread(fd, buf, nBytes, offset);
3754 }
3755 
3756 // Short sleep, direct OS call.
3757 //
3758 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3759 // sched_yield(2) will actually give up the CPU:
3760 //
3761 //   * Alone on this pariticular CPU, keeps running.
3762 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3763 //     (pre 2.6.39).
3764 //
3765 // So calling this with 0 is an alternative.
3766 //
3767 void os::naked_short_sleep(jlong ms) {
3768   struct timespec req;
3769 
3770   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3771   req.tv_sec = 0;
3772   if (ms > 0) {
3773     req.tv_nsec = (ms % 1000) * 1000000;
3774   } else {
3775     req.tv_nsec = 1;
3776   }
3777 
3778   nanosleep(&req, NULL);
3779 
3780   return;
3781 }
3782 
3783 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3784 void os::infinite_sleep() {
3785   while (true) {    // sleep forever ...
3786     ::sleep(100);   // ... 100 seconds at a time
3787   }
3788 }
3789 
3790 // Used to convert frequent JVM_Yield() to nops
3791 bool os::dont_yield() {
3792   return DontYieldALot;
3793 }
3794 
3795 void os::naked_yield() {
3796   sched_yield();
3797 }
3798 
3799 ////////////////////////////////////////////////////////////////////////////////
3800 // thread priority support
3801 
3802 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3803 // only supports dynamic priority, static priority must be zero. For real-time
3804 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3805 // However, for large multi-threaded applications, SCHED_RR is not only slower
3806 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3807 // of 5 runs - Sep 2005).
3808 //
3809 // The following code actually changes the niceness of kernel-thread/LWP. It
3810 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3811 // not the entire user process, and user level threads are 1:1 mapped to kernel
3812 // threads. It has always been the case, but could change in the future. For
3813 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3814 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3815 
3816 int os::java_to_os_priority[CriticalPriority + 1] = {
3817   19,              // 0 Entry should never be used
3818 
3819    4,              // 1 MinPriority
3820    3,              // 2
3821    2,              // 3
3822 
3823    1,              // 4
3824    0,              // 5 NormPriority
3825   -1,              // 6
3826 
3827   -2,              // 7
3828   -3,              // 8
3829   -4,              // 9 NearMaxPriority
3830 
3831   -5,              // 10 MaxPriority
3832 
3833   -5               // 11 CriticalPriority
3834 };
3835 
3836 static int prio_init() {
3837   if (ThreadPriorityPolicy == 1) {
3838     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3839     // if effective uid is not root. Perhaps, a more elegant way of doing
3840     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3841     if (geteuid() != 0) {
3842       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3843         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3844       }
3845       ThreadPriorityPolicy = 0;
3846     }
3847   }
3848   if (UseCriticalJavaThreadPriority) {
3849     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3850   }
3851   return 0;
3852 }
3853 
3854 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3855   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3856 
3857   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3858   return (ret == 0) ? OS_OK : OS_ERR;
3859 }
3860 
3861 OSReturn os::get_native_priority(const Thread* const thread,
3862                                  int *priority_ptr) {
3863   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3864     *priority_ptr = java_to_os_priority[NormPriority];
3865     return OS_OK;
3866   }
3867 
3868   errno = 0;
3869   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3870   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3871 }
3872 
3873 // Hint to the underlying OS that a task switch would not be good.
3874 // Void return because it's a hint and can fail.
3875 void os::hint_no_preempt() {}
3876 
3877 ////////////////////////////////////////////////////////////////////////////////
3878 // suspend/resume support
3879 
3880 //  the low-level signal-based suspend/resume support is a remnant from the
3881 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3882 //  within hotspot. Now there is a single use-case for this:
3883 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3884 //      that runs in the watcher thread.
3885 //  The remaining code is greatly simplified from the more general suspension
3886 //  code that used to be used.
3887 //
3888 //  The protocol is quite simple:
3889 //  - suspend:
3890 //      - sends a signal to the target thread
3891 //      - polls the suspend state of the osthread using a yield loop
3892 //      - target thread signal handler (SR_handler) sets suspend state
3893 //        and blocks in sigsuspend until continued
3894 //  - resume:
3895 //      - sets target osthread state to continue
3896 //      - sends signal to end the sigsuspend loop in the SR_handler
3897 //
3898 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3899 
3900 static void resume_clear_context(OSThread *osthread) {
3901   osthread->set_ucontext(NULL);
3902   osthread->set_siginfo(NULL);
3903 }
3904 
3905 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3906                                  ucontext_t* context) {
3907   osthread->set_ucontext(context);
3908   osthread->set_siginfo(siginfo);
3909 }
3910 
3911 // Handler function invoked when a thread's execution is suspended or
3912 // resumed. We have to be careful that only async-safe functions are
3913 // called here (Note: most pthread functions are not async safe and
3914 // should be avoided.)
3915 //
3916 // Note: sigwait() is a more natural fit than sigsuspend() from an
3917 // interface point of view, but sigwait() prevents the signal hander
3918 // from being run. libpthread would get very confused by not having
3919 // its signal handlers run and prevents sigwait()'s use with the
3920 // mutex granting granting signal.
3921 //
3922 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3923 //
3924 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3925   // Save and restore errno to avoid confusing native code with EINTR
3926   // after sigsuspend.
3927   int old_errno = errno;
3928 
3929   Thread* thread = Thread::current();
3930   OSThread* osthread = thread->osthread();
3931   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3932 
3933   os::SuspendResume::State current = osthread->sr.state();
3934   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3935     suspend_save_context(osthread, siginfo, context);
3936 
3937     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3938     os::SuspendResume::State state = osthread->sr.suspended();
3939     if (state == os::SuspendResume::SR_SUSPENDED) {
3940       sigset_t suspend_set;  // signals for sigsuspend()
3941 
3942       // get current set of blocked signals and unblock resume signal
3943       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3944       sigdelset(&suspend_set, SR_signum);
3945 
3946       sr_semaphore.signal();
3947       // wait here until we are resumed
3948       while (1) {
3949         sigsuspend(&suspend_set);
3950 
3951         os::SuspendResume::State result = osthread->sr.running();
3952         if (result == os::SuspendResume::SR_RUNNING) {
3953           sr_semaphore.signal();
3954           break;
3955         }
3956       }
3957 
3958     } else if (state == os::SuspendResume::SR_RUNNING) {
3959       // request was cancelled, continue
3960     } else {
3961       ShouldNotReachHere();
3962     }
3963 
3964     resume_clear_context(osthread);
3965   } else if (current == os::SuspendResume::SR_RUNNING) {
3966     // request was cancelled, continue
3967   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3968     // ignore
3969   } else {
3970     // ignore
3971   }
3972 
3973   errno = old_errno;
3974 }
3975 
3976 
3977 static int SR_initialize() {
3978   struct sigaction act;
3979   char *s;
3980   // Get signal number to use for suspend/resume
3981   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3982     int sig = ::strtol(s, 0, 10);
3983     if (sig > 0 || sig < _NSIG) {
3984       SR_signum = sig;
3985     }
3986   }
3987 
3988   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3989          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3990 
3991   sigemptyset(&SR_sigset);
3992   sigaddset(&SR_sigset, SR_signum);
3993 
3994   // Set up signal handler for suspend/resume
3995   act.sa_flags = SA_RESTART|SA_SIGINFO;
3996   act.sa_handler = (void (*)(int)) SR_handler;
3997 
3998   // SR_signum is blocked by default.
3999   // 4528190 - We also need to block pthread restart signal (32 on all
4000   // supported Linux platforms). Note that LinuxThreads need to block
4001   // this signal for all threads to work properly. So we don't have
4002   // to use hard-coded signal number when setting up the mask.
4003   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4004 
4005   if (sigaction(SR_signum, &act, 0) == -1) {
4006     return -1;
4007   }
4008 
4009   // Save signal flag
4010   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4011   return 0;
4012 }
4013 
4014 static int sr_notify(OSThread* osthread) {
4015   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4016   assert_status(status == 0, status, "pthread_kill");
4017   return status;
4018 }
4019 
4020 // "Randomly" selected value for how long we want to spin
4021 // before bailing out on suspending a thread, also how often
4022 // we send a signal to a thread we want to resume
4023 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4024 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4025 
4026 // returns true on success and false on error - really an error is fatal
4027 // but this seems the normal response to library errors
4028 static bool do_suspend(OSThread* osthread) {
4029   assert(osthread->sr.is_running(), "thread should be running");
4030   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4031 
4032   // mark as suspended and send signal
4033   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4034     // failed to switch, state wasn't running?
4035     ShouldNotReachHere();
4036     return false;
4037   }
4038 
4039   if (sr_notify(osthread) != 0) {
4040     ShouldNotReachHere();
4041   }
4042 
4043   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4044   while (true) {
4045     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4046       break;
4047     } else {
4048       // timeout
4049       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4050       if (cancelled == os::SuspendResume::SR_RUNNING) {
4051         return false;
4052       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4053         // make sure that we consume the signal on the semaphore as well
4054         sr_semaphore.wait();
4055         break;
4056       } else {
4057         ShouldNotReachHere();
4058         return false;
4059       }
4060     }
4061   }
4062 
4063   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4064   return true;
4065 }
4066 
4067 static void do_resume(OSThread* osthread) {
4068   assert(osthread->sr.is_suspended(), "thread should be suspended");
4069   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4070 
4071   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4072     // failed to switch to WAKEUP_REQUEST
4073     ShouldNotReachHere();
4074     return;
4075   }
4076 
4077   while (true) {
4078     if (sr_notify(osthread) == 0) {
4079       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4080         if (osthread->sr.is_running()) {
4081           return;
4082         }
4083       }
4084     } else {
4085       ShouldNotReachHere();
4086     }
4087   }
4088 
4089   guarantee(osthread->sr.is_running(), "Must be running!");
4090 }
4091 
4092 ///////////////////////////////////////////////////////////////////////////////////
4093 // signal handling (except suspend/resume)
4094 
4095 // This routine may be used by user applications as a "hook" to catch signals.
4096 // The user-defined signal handler must pass unrecognized signals to this
4097 // routine, and if it returns true (non-zero), then the signal handler must
4098 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4099 // routine will never retun false (zero), but instead will execute a VM panic
4100 // routine kill the process.
4101 //
4102 // If this routine returns false, it is OK to call it again.  This allows
4103 // the user-defined signal handler to perform checks either before or after
4104 // the VM performs its own checks.  Naturally, the user code would be making
4105 // a serious error if it tried to handle an exception (such as a null check
4106 // or breakpoint) that the VM was generating for its own correct operation.
4107 //
4108 // This routine may recognize any of the following kinds of signals:
4109 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4110 // It should be consulted by handlers for any of those signals.
4111 //
4112 // The caller of this routine must pass in the three arguments supplied
4113 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4114 // field of the structure passed to sigaction().  This routine assumes that
4115 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4116 //
4117 // Note that the VM will print warnings if it detects conflicting signal
4118 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4119 //
4120 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4121                                                  siginfo_t* siginfo,
4122                                                  void* ucontext,
4123                                                  int abort_if_unrecognized);
4124 
4125 void signalHandler(int sig, siginfo_t* info, void* uc) {
4126   assert(info != NULL && uc != NULL, "it must be old kernel");
4127   int orig_errno = errno;  // Preserve errno value over signal handler.
4128   JVM_handle_linux_signal(sig, info, uc, true);
4129   errno = orig_errno;
4130 }
4131 
4132 
4133 // This boolean allows users to forward their own non-matching signals
4134 // to JVM_handle_linux_signal, harmlessly.
4135 bool os::Linux::signal_handlers_are_installed = false;
4136 
4137 // For signal-chaining
4138 struct sigaction os::Linux::sigact[MAXSIGNUM];
4139 unsigned int os::Linux::sigs = 0;
4140 bool os::Linux::libjsig_is_loaded = false;
4141 typedef struct sigaction *(*get_signal_t)(int);
4142 get_signal_t os::Linux::get_signal_action = NULL;
4143 
4144 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4145   struct sigaction *actp = NULL;
4146 
4147   if (libjsig_is_loaded) {
4148     // Retrieve the old signal handler from libjsig
4149     actp = (*get_signal_action)(sig);
4150   }
4151   if (actp == NULL) {
4152     // Retrieve the preinstalled signal handler from jvm
4153     actp = get_preinstalled_handler(sig);
4154   }
4155 
4156   return actp;
4157 }
4158 
4159 static bool call_chained_handler(struct sigaction *actp, int sig,
4160                                  siginfo_t *siginfo, void *context) {
4161   // Call the old signal handler
4162   if (actp->sa_handler == SIG_DFL) {
4163     // It's more reasonable to let jvm treat it as an unexpected exception
4164     // instead of taking the default action.
4165     return false;
4166   } else if (actp->sa_handler != SIG_IGN) {
4167     if ((actp->sa_flags & SA_NODEFER) == 0) {
4168       // automaticlly block the signal
4169       sigaddset(&(actp->sa_mask), sig);
4170     }
4171 
4172     sa_handler_t hand;
4173     sa_sigaction_t sa;
4174     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4175     // retrieve the chained handler
4176     if (siginfo_flag_set) {
4177       sa = actp->sa_sigaction;
4178     } else {
4179       hand = actp->sa_handler;
4180     }
4181 
4182     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4183       actp->sa_handler = SIG_DFL;
4184     }
4185 
4186     // try to honor the signal mask
4187     sigset_t oset;
4188     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4189 
4190     // call into the chained handler
4191     if (siginfo_flag_set) {
4192       (*sa)(sig, siginfo, context);
4193     } else {
4194       (*hand)(sig);
4195     }
4196 
4197     // restore the signal mask
4198     pthread_sigmask(SIG_SETMASK, &oset, 0);
4199   }
4200   // Tell jvm's signal handler the signal is taken care of.
4201   return true;
4202 }
4203 
4204 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4205   bool chained = false;
4206   // signal-chaining
4207   if (UseSignalChaining) {
4208     struct sigaction *actp = get_chained_signal_action(sig);
4209     if (actp != NULL) {
4210       chained = call_chained_handler(actp, sig, siginfo, context);
4211     }
4212   }
4213   return chained;
4214 }
4215 
4216 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4217   if ((((unsigned int)1 << sig) & sigs) != 0) {
4218     return &sigact[sig];
4219   }
4220   return NULL;
4221 }
4222 
4223 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4224   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4225   sigact[sig] = oldAct;
4226   sigs |= (unsigned int)1 << sig;
4227 }
4228 
4229 // for diagnostic
4230 int os::Linux::sigflags[MAXSIGNUM];
4231 
4232 int os::Linux::get_our_sigflags(int sig) {
4233   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4234   return sigflags[sig];
4235 }
4236 
4237 void os::Linux::set_our_sigflags(int sig, int flags) {
4238   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4239   sigflags[sig] = flags;
4240 }
4241 
4242 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4243   // Check for overwrite.
4244   struct sigaction oldAct;
4245   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4246 
4247   void* oldhand = oldAct.sa_sigaction
4248                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4249                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4250   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4251       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4252       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4253     if (AllowUserSignalHandlers || !set_installed) {
4254       // Do not overwrite; user takes responsibility to forward to us.
4255       return;
4256     } else if (UseSignalChaining) {
4257       // save the old handler in jvm
4258       save_preinstalled_handler(sig, oldAct);
4259       // libjsig also interposes the sigaction() call below and saves the
4260       // old sigaction on it own.
4261     } else {
4262       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4263                     "%#lx for signal %d.", (long)oldhand, sig));
4264     }
4265   }
4266 
4267   struct sigaction sigAct;
4268   sigfillset(&(sigAct.sa_mask));
4269   sigAct.sa_handler = SIG_DFL;
4270   if (!set_installed) {
4271     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4272   } else {
4273     sigAct.sa_sigaction = signalHandler;
4274     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4275   }
4276   // Save flags, which are set by ours
4277   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4278   sigflags[sig] = sigAct.sa_flags;
4279 
4280   int ret = sigaction(sig, &sigAct, &oldAct);
4281   assert(ret == 0, "check");
4282 
4283   void* oldhand2  = oldAct.sa_sigaction
4284                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4285                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4286   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4287 }
4288 
4289 // install signal handlers for signals that HotSpot needs to
4290 // handle in order to support Java-level exception handling.
4291 
4292 void os::Linux::install_signal_handlers() {
4293   if (!signal_handlers_are_installed) {
4294     signal_handlers_are_installed = true;
4295 
4296     // signal-chaining
4297     typedef void (*signal_setting_t)();
4298     signal_setting_t begin_signal_setting = NULL;
4299     signal_setting_t end_signal_setting = NULL;
4300     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4301                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4302     if (begin_signal_setting != NULL) {
4303       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4304                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4305       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4306                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4307       libjsig_is_loaded = true;
4308       assert(UseSignalChaining, "should enable signal-chaining");
4309     }
4310     if (libjsig_is_loaded) {
4311       // Tell libjsig jvm is setting signal handlers
4312       (*begin_signal_setting)();
4313     }
4314 
4315     set_signal_handler(SIGSEGV, true);
4316     set_signal_handler(SIGPIPE, true);
4317     set_signal_handler(SIGBUS, true);
4318     set_signal_handler(SIGILL, true);
4319     set_signal_handler(SIGFPE, true);
4320 #if defined(PPC64)
4321     set_signal_handler(SIGTRAP, true);
4322 #endif
4323     set_signal_handler(SIGXFSZ, true);
4324 
4325     if (libjsig_is_loaded) {
4326       // Tell libjsig jvm finishes setting signal handlers
4327       (*end_signal_setting)();
4328     }
4329 
4330     // We don't activate signal checker if libjsig is in place, we trust ourselves
4331     // and if UserSignalHandler is installed all bets are off.
4332     // Log that signal checking is off only if -verbose:jni is specified.
4333     if (CheckJNICalls) {
4334       if (libjsig_is_loaded) {
4335         if (PrintJNIResolving) {
4336           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4337         }
4338         check_signals = false;
4339       }
4340       if (AllowUserSignalHandlers) {
4341         if (PrintJNIResolving) {
4342           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4343         }
4344         check_signals = false;
4345       }
4346     }
4347   }
4348 }
4349 
4350 // This is the fastest way to get thread cpu time on Linux.
4351 // Returns cpu time (user+sys) for any thread, not only for current.
4352 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4353 // It might work on 2.6.10+ with a special kernel/glibc patch.
4354 // For reference, please, see IEEE Std 1003.1-2004:
4355 //   http://www.unix.org/single_unix_specification
4356 
4357 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4358   struct timespec tp;
4359   int rc = os::Linux::clock_gettime(clockid, &tp);
4360   assert(rc == 0, "clock_gettime is expected to return 0 code");
4361 
4362   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4363 }
4364 
4365 /////
4366 // glibc on Linux platform uses non-documented flag
4367 // to indicate, that some special sort of signal
4368 // trampoline is used.
4369 // We will never set this flag, and we should
4370 // ignore this flag in our diagnostic
4371 #ifdef SIGNIFICANT_SIGNAL_MASK
4372   #undef SIGNIFICANT_SIGNAL_MASK
4373 #endif
4374 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4375 
4376 static const char* get_signal_handler_name(address handler,
4377                                            char* buf, int buflen) {
4378   int offset;
4379   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4380   if (found) {
4381     // skip directory names
4382     const char *p1, *p2;
4383     p1 = buf;
4384     size_t len = strlen(os::file_separator());
4385     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4386     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4387   } else {
4388     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4389   }
4390   return buf;
4391 }
4392 
4393 static void print_signal_handler(outputStream* st, int sig,
4394                                  char* buf, size_t buflen) {
4395   struct sigaction sa;
4396 
4397   sigaction(sig, NULL, &sa);
4398 
4399   // See comment for SIGNIFICANT_SIGNAL_MASK define
4400   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4401 
4402   st->print("%s: ", os::exception_name(sig, buf, buflen));
4403 
4404   address handler = (sa.sa_flags & SA_SIGINFO)
4405     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4406     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4407 
4408   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4409     st->print("SIG_DFL");
4410   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4411     st->print("SIG_IGN");
4412   } else {
4413     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4414   }
4415 
4416   st->print(", sa_mask[0]=");
4417   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4418 
4419   address rh = VMError::get_resetted_sighandler(sig);
4420   // May be, handler was resetted by VMError?
4421   if (rh != NULL) {
4422     handler = rh;
4423     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4424   }
4425 
4426   st->print(", sa_flags=");
4427   os::Posix::print_sa_flags(st, sa.sa_flags);
4428 
4429   // Check: is it our handler?
4430   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4431       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4432     // It is our signal handler
4433     // check for flags, reset system-used one!
4434     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4435       st->print(
4436                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4437                 os::Linux::get_our_sigflags(sig));
4438     }
4439   }
4440   st->cr();
4441 }
4442 
4443 
4444 #define DO_SIGNAL_CHECK(sig)                      \
4445   do {                                            \
4446     if (!sigismember(&check_signal_done, sig)) {  \
4447       os::Linux::check_signal_handler(sig);       \
4448     }                                             \
4449   } while (0)
4450 
4451 // This method is a periodic task to check for misbehaving JNI applications
4452 // under CheckJNI, we can add any periodic checks here
4453 
4454 void os::run_periodic_checks() {
4455   if (check_signals == false) return;
4456 
4457   // SEGV and BUS if overridden could potentially prevent
4458   // generation of hs*.log in the event of a crash, debugging
4459   // such a case can be very challenging, so we absolutely
4460   // check the following for a good measure:
4461   DO_SIGNAL_CHECK(SIGSEGV);
4462   DO_SIGNAL_CHECK(SIGILL);
4463   DO_SIGNAL_CHECK(SIGFPE);
4464   DO_SIGNAL_CHECK(SIGBUS);
4465   DO_SIGNAL_CHECK(SIGPIPE);
4466   DO_SIGNAL_CHECK(SIGXFSZ);
4467 #if defined(PPC64)
4468   DO_SIGNAL_CHECK(SIGTRAP);
4469 #endif
4470 
4471   // ReduceSignalUsage allows the user to override these handlers
4472   // see comments at the very top and jvm_solaris.h
4473   if (!ReduceSignalUsage) {
4474     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4475     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4476     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4477     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4478   }
4479 
4480   DO_SIGNAL_CHECK(SR_signum);
4481   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4482 }
4483 
4484 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4485 
4486 static os_sigaction_t os_sigaction = NULL;
4487 
4488 void os::Linux::check_signal_handler(int sig) {
4489   char buf[O_BUFLEN];
4490   address jvmHandler = NULL;
4491 
4492 
4493   struct sigaction act;
4494   if (os_sigaction == NULL) {
4495     // only trust the default sigaction, in case it has been interposed
4496     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4497     if (os_sigaction == NULL) return;
4498   }
4499 
4500   os_sigaction(sig, (struct sigaction*)NULL, &act);
4501 
4502 
4503   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4504 
4505   address thisHandler = (act.sa_flags & SA_SIGINFO)
4506     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4507     : CAST_FROM_FN_PTR(address, act.sa_handler);
4508 
4509 
4510   switch (sig) {
4511   case SIGSEGV:
4512   case SIGBUS:
4513   case SIGFPE:
4514   case SIGPIPE:
4515   case SIGILL:
4516   case SIGXFSZ:
4517     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4518     break;
4519 
4520   case SHUTDOWN1_SIGNAL:
4521   case SHUTDOWN2_SIGNAL:
4522   case SHUTDOWN3_SIGNAL:
4523   case BREAK_SIGNAL:
4524     jvmHandler = (address)user_handler();
4525     break;
4526 
4527   case INTERRUPT_SIGNAL:
4528     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4529     break;
4530 
4531   default:
4532     if (sig == SR_signum) {
4533       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4534     } else {
4535       return;
4536     }
4537     break;
4538   }
4539 
4540   if (thisHandler != jvmHandler) {
4541     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4542     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4543     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4544     // No need to check this sig any longer
4545     sigaddset(&check_signal_done, sig);
4546     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4547     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4548       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4549                     exception_name(sig, buf, O_BUFLEN));
4550     }
4551   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4552     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4553     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4554     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4555     // No need to check this sig any longer
4556     sigaddset(&check_signal_done, sig);
4557   }
4558 
4559   // Dump all the signal
4560   if (sigismember(&check_signal_done, sig)) {
4561     print_signal_handlers(tty, buf, O_BUFLEN);
4562   }
4563 }
4564 
4565 extern void report_error(char* file_name, int line_no, char* title,
4566                          char* format, ...);
4567 
4568 extern bool signal_name(int signo, char* buf, size_t len);
4569 
4570 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4571   if (0 < exception_code && exception_code <= SIGRTMAX) {
4572     // signal
4573     if (!signal_name(exception_code, buf, size)) {
4574       jio_snprintf(buf, size, "SIG%d", exception_code);
4575     }
4576     return buf;
4577   } else {
4578     return NULL;
4579   }
4580 }
4581 
4582 // this is called _before_ the most of global arguments have been parsed
4583 void os::init(void) {
4584   char dummy;   // used to get a guess on initial stack address
4585 //  first_hrtime = gethrtime();
4586 
4587   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4588 
4589   init_random(1234567);
4590 
4591   ThreadCritical::initialize();
4592 
4593   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4594   if (Linux::page_size() == -1) {
4595     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4596                   strerror(errno)));
4597   }
4598   init_page_sizes((size_t) Linux::page_size());
4599 
4600   Linux::initialize_system_info();
4601 
4602   // main_thread points to the aboriginal thread
4603   Linux::_main_thread = pthread_self();
4604 
4605   Linux::clock_init();
4606   initial_time_count = javaTimeNanos();
4607 
4608   // pthread_condattr initialization for monotonic clock
4609   int status;
4610   pthread_condattr_t* _condattr = os::Linux::condAttr();
4611   if ((status = pthread_condattr_init(_condattr)) != 0) {
4612     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4613   }
4614   // Only set the clock if CLOCK_MONOTONIC is available
4615   if (os::supports_monotonic_clock()) {
4616     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4617       if (status == EINVAL) {
4618         warning("Unable to use monotonic clock with relative timed-waits" \
4619                 " - changes to the time-of-day clock may have adverse affects");
4620       } else {
4621         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4622       }
4623     }
4624   }
4625   // else it defaults to CLOCK_REALTIME
4626 
4627   // If the pagesize of the VM is greater than 8K determine the appropriate
4628   // number of initial guard pages.  The user can change this with the
4629   // command line arguments, if needed.
4630   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4631     StackYellowPages = 1;
4632     StackRedPages = 1;
4633     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4634   }
4635 
4636   // retrieve entry point for pthread_setname_np
4637   Linux::_pthread_setname_np =
4638     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4639 
4640 }
4641 
4642 // To install functions for atexit system call
4643 extern "C" {
4644   static void perfMemory_exit_helper() {
4645     perfMemory_exit();
4646   }
4647 }
4648 
4649 // this is called _after_ the global arguments have been parsed
4650 jint os::init_2(void) {
4651   Linux::fast_thread_clock_init();
4652 
4653   // Allocate a single page and mark it as readable for safepoint polling
4654   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4655   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4656 
4657   os::set_polling_page(polling_page);
4658 
4659 #ifndef PRODUCT
4660   if (Verbose && PrintMiscellaneous) {
4661     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4662                (intptr_t)polling_page);
4663   }
4664 #endif
4665 
4666   if (!UseMembar) {
4667     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4668     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4669     os::set_memory_serialize_page(mem_serialize_page);
4670 
4671 #ifndef PRODUCT
4672     if (Verbose && PrintMiscellaneous) {
4673       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n",
4674                  (intptr_t)mem_serialize_page);
4675     }
4676 #endif
4677   }
4678 
4679   // initialize suspend/resume support - must do this before signal_sets_init()
4680   if (SR_initialize() != 0) {
4681     perror("SR_initialize failed");
4682     return JNI_ERR;
4683   }
4684 
4685   Linux::signal_sets_init();
4686   Linux::install_signal_handlers();
4687 
4688   // Check minimum allowable stack size for thread creation and to initialize
4689   // the java system classes, including StackOverflowError - depends on page
4690   // size.  Add a page for compiler2 recursion in main thread.
4691   // Add in 2*BytesPerWord times page size to account for VM stack during
4692   // class initialization depending on 32 or 64 bit VM.
4693   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4694                                       (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4695                                       (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4696 
4697   size_t threadStackSizeInBytes = ThreadStackSize * K;
4698   if (threadStackSizeInBytes != 0 &&
4699       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4700     tty->print_cr("\nThe stack size specified is too small, "
4701                   "Specify at least %dk",
4702                   os::Linux::min_stack_allowed/ K);
4703     return JNI_ERR;
4704   }
4705 
4706   // Make the stack size a multiple of the page size so that
4707   // the yellow/red zones can be guarded.
4708   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4709                                                 vm_page_size()));
4710 
4711   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4712 
4713 #if defined(IA32)
4714   workaround_expand_exec_shield_cs_limit();
4715 #endif
4716 
4717   Linux::libpthread_init();
4718   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4719     tty->print_cr("[HotSpot is running with %s, %s]\n",
4720                   Linux::glibc_version(), Linux::libpthread_version());
4721   }
4722 
4723   if (UseNUMA) {
4724     if (!Linux::libnuma_init()) {
4725       UseNUMA = false;
4726     } else {
4727       if ((Linux::numa_max_node() < 1)) {
4728         // There's only one node(they start from 0), disable NUMA.
4729         UseNUMA = false;
4730       }
4731     }
4732     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4733     // we can make the adaptive lgrp chunk resizing work. If the user specified
4734     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4735     // disable adaptive resizing.
4736     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4737       if (FLAG_IS_DEFAULT(UseNUMA)) {
4738         UseNUMA = false;
4739       } else {
4740         if (FLAG_IS_DEFAULT(UseLargePages) &&
4741             FLAG_IS_DEFAULT(UseSHM) &&
4742             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4743           UseLargePages = false;
4744         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4745           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4746           UseAdaptiveSizePolicy = false;
4747           UseAdaptiveNUMAChunkSizing = false;
4748         }
4749       }
4750     }
4751     if (!UseNUMA && ForceNUMA) {
4752       UseNUMA = true;
4753     }
4754   }
4755 
4756   if (MaxFDLimit) {
4757     // set the number of file descriptors to max. print out error
4758     // if getrlimit/setrlimit fails but continue regardless.
4759     struct rlimit nbr_files;
4760     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4761     if (status != 0) {
4762       if (PrintMiscellaneous && (Verbose || WizardMode)) {
4763         perror("os::init_2 getrlimit failed");
4764       }
4765     } else {
4766       nbr_files.rlim_cur = nbr_files.rlim_max;
4767       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4768       if (status != 0) {
4769         if (PrintMiscellaneous && (Verbose || WizardMode)) {
4770           perror("os::init_2 setrlimit failed");
4771         }
4772       }
4773     }
4774   }
4775 
4776   // Initialize lock used to serialize thread creation (see os::create_thread)
4777   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4778 
4779   // at-exit methods are called in the reverse order of their registration.
4780   // atexit functions are called on return from main or as a result of a
4781   // call to exit(3C). There can be only 32 of these functions registered
4782   // and atexit() does not set errno.
4783 
4784   if (PerfAllowAtExitRegistration) {
4785     // only register atexit functions if PerfAllowAtExitRegistration is set.
4786     // atexit functions can be delayed until process exit time, which
4787     // can be problematic for embedded VM situations. Embedded VMs should
4788     // call DestroyJavaVM() to assure that VM resources are released.
4789 
4790     // note: perfMemory_exit_helper atexit function may be removed in
4791     // the future if the appropriate cleanup code can be added to the
4792     // VM_Exit VMOperation's doit method.
4793     if (atexit(perfMemory_exit_helper) != 0) {
4794       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4795     }
4796   }
4797 
4798   // initialize thread priority policy
4799   prio_init();
4800 
4801   return JNI_OK;
4802 }
4803 
4804 // Mark the polling page as unreadable
4805 void os::make_polling_page_unreadable(void) {
4806   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4807     fatal("Could not disable polling page");
4808   }
4809 }
4810 
4811 // Mark the polling page as readable
4812 void os::make_polling_page_readable(void) {
4813   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4814     fatal("Could not enable polling page");
4815   }
4816 }
4817 
4818 int os::active_processor_count() {
4819   // Linux doesn't yet have a (official) notion of processor sets,
4820   // so just return the number of online processors.
4821   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4822   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4823   return online_cpus;
4824 }
4825 
4826 void os::set_native_thread_name(const char *name) {
4827   if (Linux::_pthread_setname_np) {
4828     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4829     snprintf(buf, sizeof(buf), "%s", name);
4830     buf[sizeof(buf) - 1] = '\0';
4831     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4832     // ERANGE should not happen; all other errors should just be ignored.
4833     assert(rc != ERANGE, "pthread_setname_np failed");
4834   }
4835 }
4836 
4837 bool os::distribute_processes(uint length, uint* distribution) {
4838   // Not yet implemented.
4839   return false;
4840 }
4841 
4842 bool os::bind_to_processor(uint processor_id) {
4843   // Not yet implemented.
4844   return false;
4845 }
4846 
4847 ///
4848 
4849 void os::SuspendedThreadTask::internal_do_task() {
4850   if (do_suspend(_thread->osthread())) {
4851     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4852     do_task(context);
4853     do_resume(_thread->osthread());
4854   }
4855 }
4856 
4857 class PcFetcher : public os::SuspendedThreadTask {
4858  public:
4859   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4860   ExtendedPC result();
4861  protected:
4862   void do_task(const os::SuspendedThreadTaskContext& context);
4863  private:
4864   ExtendedPC _epc;
4865 };
4866 
4867 ExtendedPC PcFetcher::result() {
4868   guarantee(is_done(), "task is not done yet.");
4869   return _epc;
4870 }
4871 
4872 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4873   Thread* thread = context.thread();
4874   OSThread* osthread = thread->osthread();
4875   if (osthread->ucontext() != NULL) {
4876     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4877   } else {
4878     // NULL context is unexpected, double-check this is the VMThread
4879     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4880   }
4881 }
4882 
4883 // Suspends the target using the signal mechanism and then grabs the PC before
4884 // resuming the target. Used by the flat-profiler only
4885 ExtendedPC os::get_thread_pc(Thread* thread) {
4886   // Make sure that it is called by the watcher for the VMThread
4887   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4888   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4889 
4890   PcFetcher fetcher(thread);
4891   fetcher.run();
4892   return fetcher.result();
4893 }
4894 
4895 ////////////////////////////////////////////////////////////////////////////////
4896 // debug support
4897 
4898 bool os::find(address addr, outputStream* st) {
4899   Dl_info dlinfo;
4900   memset(&dlinfo, 0, sizeof(dlinfo));
4901   if (dladdr(addr, &dlinfo) != 0) {
4902     st->print(PTR_FORMAT ": ", addr);
4903     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4904       st->print("%s+%#x", dlinfo.dli_sname,
4905                 addr - (intptr_t)dlinfo.dli_saddr);
4906     } else if (dlinfo.dli_fbase != NULL) {
4907       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4908     } else {
4909       st->print("<absolute address>");
4910     }
4911     if (dlinfo.dli_fname != NULL) {
4912       st->print(" in %s", dlinfo.dli_fname);
4913     }
4914     if (dlinfo.dli_fbase != NULL) {
4915       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4916     }
4917     st->cr();
4918 
4919     if (Verbose) {
4920       // decode some bytes around the PC
4921       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4922       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4923       address       lowest = (address) dlinfo.dli_sname;
4924       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4925       if (begin < lowest)  begin = lowest;
4926       Dl_info dlinfo2;
4927       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4928           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4929         end = (address) dlinfo2.dli_saddr;
4930       }
4931       Disassembler::decode(begin, end, st);
4932     }
4933     return true;
4934   }
4935   return false;
4936 }
4937 
4938 ////////////////////////////////////////////////////////////////////////////////
4939 // misc
4940 
4941 // This does not do anything on Linux. This is basically a hook for being
4942 // able to use structured exception handling (thread-local exception filters)
4943 // on, e.g., Win32.
4944 void
4945 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4946                          JavaCallArguments* args, Thread* thread) {
4947   f(value, method, args, thread);
4948 }
4949 
4950 void os::print_statistics() {
4951 }
4952 
4953 int os::message_box(const char* title, const char* message) {
4954   int i;
4955   fdStream err(defaultStream::error_fd());
4956   for (i = 0; i < 78; i++) err.print_raw("=");
4957   err.cr();
4958   err.print_raw_cr(title);
4959   for (i = 0; i < 78; i++) err.print_raw("-");
4960   err.cr();
4961   err.print_raw_cr(message);
4962   for (i = 0; i < 78; i++) err.print_raw("=");
4963   err.cr();
4964 
4965   char buf[16];
4966   // Prevent process from exiting upon "read error" without consuming all CPU
4967   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4968 
4969   return buf[0] == 'y' || buf[0] == 'Y';
4970 }
4971 
4972 int os::stat(const char *path, struct stat *sbuf) {
4973   char pathbuf[MAX_PATH];
4974   if (strlen(path) > MAX_PATH - 1) {
4975     errno = ENAMETOOLONG;
4976     return -1;
4977   }
4978   os::native_path(strcpy(pathbuf, path));
4979   return ::stat(pathbuf, sbuf);
4980 }
4981 
4982 bool os::check_heap(bool force) {
4983   return true;
4984 }
4985 
4986 // Is a (classpath) directory empty?
4987 bool os::dir_is_empty(const char* path) {
4988   DIR *dir = NULL;
4989   struct dirent *ptr;
4990 
4991   dir = opendir(path);
4992   if (dir == NULL) return true;
4993 
4994   // Scan the directory
4995   bool result = true;
4996   char buf[sizeof(struct dirent) + MAX_PATH];
4997   while (result && (ptr = ::readdir(dir)) != NULL) {
4998     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4999       result = false;
5000     }
5001   }
5002   closedir(dir);
5003   return result;
5004 }
5005 
5006 // This code originates from JDK's sysOpen and open64_w
5007 // from src/solaris/hpi/src/system_md.c
5008 
5009 int os::open(const char *path, int oflag, int mode) {
5010   if (strlen(path) > MAX_PATH - 1) {
5011     errno = ENAMETOOLONG;
5012     return -1;
5013   }
5014 
5015   // All file descriptors that are opened in the Java process and not
5016   // specifically destined for a subprocess should have the close-on-exec
5017   // flag set.  If we don't set it, then careless 3rd party native code
5018   // might fork and exec without closing all appropriate file descriptors
5019   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5020   // turn might:
5021   //
5022   // - cause end-of-file to fail to be detected on some file
5023   //   descriptors, resulting in mysterious hangs, or
5024   //
5025   // - might cause an fopen in the subprocess to fail on a system
5026   //   suffering from bug 1085341.
5027   //
5028   // (Yes, the default setting of the close-on-exec flag is a Unix
5029   // design flaw)
5030   //
5031   // See:
5032   // 1085341: 32-bit stdio routines should support file descriptors >255
5033   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5034   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5035   //
5036   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5037   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5038   // because it saves a system call and removes a small window where the flag
5039   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5040   // and we fall back to using FD_CLOEXEC (see below).
5041 #ifdef O_CLOEXEC
5042   oflag |= O_CLOEXEC;
5043 #endif
5044 
5045   int fd = ::open64(path, oflag, mode);
5046   if (fd == -1) return -1;
5047 
5048   //If the open succeeded, the file might still be a directory
5049   {
5050     struct stat64 buf64;
5051     int ret = ::fstat64(fd, &buf64);
5052     int st_mode = buf64.st_mode;
5053 
5054     if (ret != -1) {
5055       if ((st_mode & S_IFMT) == S_IFDIR) {
5056         errno = EISDIR;
5057         ::close(fd);
5058         return -1;
5059       }
5060     } else {
5061       ::close(fd);
5062       return -1;
5063     }
5064   }
5065 
5066 #ifdef FD_CLOEXEC
5067   // Validate that the use of the O_CLOEXEC flag on open above worked.
5068   // With recent kernels, we will perform this check exactly once.
5069   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5070   if (!O_CLOEXEC_is_known_to_work) {
5071     int flags = ::fcntl(fd, F_GETFD);
5072     if (flags != -1) {
5073       if ((flags & FD_CLOEXEC) != 0)
5074         O_CLOEXEC_is_known_to_work = 1;
5075       else
5076         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5077     }
5078   }
5079 #endif
5080 
5081   return fd;
5082 }
5083 
5084 
5085 // create binary file, rewriting existing file if required
5086 int os::create_binary_file(const char* path, bool rewrite_existing) {
5087   int oflags = O_WRONLY | O_CREAT;
5088   if (!rewrite_existing) {
5089     oflags |= O_EXCL;
5090   }
5091   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5092 }
5093 
5094 // return current position of file pointer
5095 jlong os::current_file_offset(int fd) {
5096   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5097 }
5098 
5099 // move file pointer to the specified offset
5100 jlong os::seek_to_file_offset(int fd, jlong offset) {
5101   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5102 }
5103 
5104 // This code originates from JDK's sysAvailable
5105 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5106 
5107 int os::available(int fd, jlong *bytes) {
5108   jlong cur, end;
5109   int mode;
5110   struct stat64 buf64;
5111 
5112   if (::fstat64(fd, &buf64) >= 0) {
5113     mode = buf64.st_mode;
5114     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5115       // XXX: is the following call interruptible? If so, this might
5116       // need to go through the INTERRUPT_IO() wrapper as for other
5117       // blocking, interruptible calls in this file.
5118       int n;
5119       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5120         *bytes = n;
5121         return 1;
5122       }
5123     }
5124   }
5125   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5126     return 0;
5127   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5128     return 0;
5129   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5130     return 0;
5131   }
5132   *bytes = end - cur;
5133   return 1;
5134 }
5135 
5136 // Map a block of memory.
5137 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5138                         char *addr, size_t bytes, bool read_only,
5139                         bool allow_exec) {
5140   int prot;
5141   int flags = MAP_PRIVATE;
5142 
5143   if (read_only) {
5144     prot = PROT_READ;
5145   } else {
5146     prot = PROT_READ | PROT_WRITE;
5147   }
5148 
5149   if (allow_exec) {
5150     prot |= PROT_EXEC;
5151   }
5152 
5153   if (addr != NULL) {
5154     flags |= MAP_FIXED;
5155   }
5156 
5157   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5158                                      fd, file_offset);
5159   if (mapped_address == MAP_FAILED) {
5160     return NULL;
5161   }
5162   return mapped_address;
5163 }
5164 
5165 
5166 // Remap a block of memory.
5167 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5168                           char *addr, size_t bytes, bool read_only,
5169                           bool allow_exec) {
5170   // same as map_memory() on this OS
5171   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5172                         allow_exec);
5173 }
5174 
5175 
5176 // Unmap a block of memory.
5177 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5178   return munmap(addr, bytes) == 0;
5179 }
5180 
5181 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5182 
5183 static clockid_t thread_cpu_clockid(Thread* thread) {
5184   pthread_t tid = thread->osthread()->pthread_id();
5185   clockid_t clockid;
5186 
5187   // Get thread clockid
5188   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5189   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5190   return clockid;
5191 }
5192 
5193 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5194 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5195 // of a thread.
5196 //
5197 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5198 // the fast estimate available on the platform.
5199 
5200 jlong os::current_thread_cpu_time() {
5201   if (os::Linux::supports_fast_thread_cpu_time()) {
5202     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5203   } else {
5204     // return user + sys since the cost is the same
5205     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5206   }
5207 }
5208 
5209 jlong os::thread_cpu_time(Thread* thread) {
5210   // consistent with what current_thread_cpu_time() returns
5211   if (os::Linux::supports_fast_thread_cpu_time()) {
5212     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5213   } else {
5214     return slow_thread_cpu_time(thread, true /* user + sys */);
5215   }
5216 }
5217 
5218 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5219   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5220     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5221   } else {
5222     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5223   }
5224 }
5225 
5226 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5227   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5228     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5229   } else {
5230     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5231   }
5232 }
5233 
5234 //  -1 on error.
5235 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5236   pid_t  tid = thread->osthread()->thread_id();
5237   char *s;
5238   char stat[2048];
5239   int statlen;
5240   char proc_name[64];
5241   int count;
5242   long sys_time, user_time;
5243   char cdummy;
5244   int idummy;
5245   long ldummy;
5246   FILE *fp;
5247 
5248   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5249   fp = fopen(proc_name, "r");
5250   if (fp == NULL) return -1;
5251   statlen = fread(stat, 1, 2047, fp);
5252   stat[statlen] = '\0';
5253   fclose(fp);
5254 
5255   // Skip pid and the command string. Note that we could be dealing with
5256   // weird command names, e.g. user could decide to rename java launcher
5257   // to "java 1.4.2 :)", then the stat file would look like
5258   //                1234 (java 1.4.2 :)) R ... ...
5259   // We don't really need to know the command string, just find the last
5260   // occurrence of ")" and then start parsing from there. See bug 4726580.
5261   s = strrchr(stat, ')');
5262   if (s == NULL) return -1;
5263 
5264   // Skip blank chars
5265   do { s++; } while (s && isspace(*s));
5266 
5267   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5268                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5269                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5270                  &user_time, &sys_time);
5271   if (count != 13) return -1;
5272   if (user_sys_cpu_time) {
5273     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5274   } else {
5275     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5276   }
5277 }
5278 
5279 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5280   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5281   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5282   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5283   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5284 }
5285 
5286 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5287   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5288   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5289   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5290   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5291 }
5292 
5293 bool os::is_thread_cpu_time_supported() {
5294   return true;
5295 }
5296 
5297 // System loadavg support.  Returns -1 if load average cannot be obtained.
5298 // Linux doesn't yet have a (official) notion of processor sets,
5299 // so just return the system wide load average.
5300 int os::loadavg(double loadavg[], int nelem) {
5301   return ::getloadavg(loadavg, nelem);
5302 }
5303 
5304 void os::pause() {
5305   char filename[MAX_PATH];
5306   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5307     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5308   } else {
5309     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5310   }
5311 
5312   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5313   if (fd != -1) {
5314     struct stat buf;
5315     ::close(fd);
5316     while (::stat(filename, &buf) == 0) {
5317       (void)::poll(NULL, 0, 100);
5318     }
5319   } else {
5320     jio_fprintf(stderr,
5321                 "Could not open pause file '%s', continuing immediately.\n", filename);
5322   }
5323 }
5324 
5325 
5326 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5327 // comment paragraphs are worth repeating here:
5328 //
5329 // Assumption:
5330 //    Only one parker can exist on an event, which is why we allocate
5331 //    them per-thread. Multiple unparkers can coexist.
5332 //
5333 // _Event serves as a restricted-range semaphore.
5334 //   -1 : thread is blocked, i.e. there is a waiter
5335 //    0 : neutral: thread is running or ready,
5336 //        could have been signaled after a wait started
5337 //    1 : signaled - thread is running or ready
5338 //
5339 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5340 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5341 // For specifics regarding the bug see GLIBC BUGID 261237 :
5342 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5343 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5344 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5345 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5346 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5347 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5348 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5349 // of libpthread avoids the problem, but isn't practical.
5350 //
5351 // Possible remedies:
5352 //
5353 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5354 //      This is palliative and probabilistic, however.  If the thread is preempted
5355 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5356 //      than the minimum period may have passed, and the abstime may be stale (in the
5357 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5358 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5359 //
5360 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5361 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5362 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5363 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5364 //      thread.
5365 //
5366 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5367 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5368 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5369 //      This also works well.  In fact it avoids kernel-level scalability impediments
5370 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5371 //      timers in a graceful fashion.
5372 //
5373 // 4.   When the abstime value is in the past it appears that control returns
5374 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5375 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5376 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5377 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5378 //      It may be possible to avoid reinitialization by checking the return
5379 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5380 //      condvar we must establish the invariant that cond_signal() is only called
5381 //      within critical sections protected by the adjunct mutex.  This prevents
5382 //      cond_signal() from "seeing" a condvar that's in the midst of being
5383 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5384 //      desirable signal-after-unlock optimization that avoids futile context switching.
5385 //
5386 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5387 //      structure when a condvar is used or initialized.  cond_destroy()  would
5388 //      release the helper structure.  Our reinitialize-after-timedwait fix
5389 //      put excessive stress on malloc/free and locks protecting the c-heap.
5390 //
5391 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5392 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5393 // and only enabling the work-around for vulnerable environments.
5394 
5395 // utility to compute the abstime argument to timedwait:
5396 // millis is the relative timeout time
5397 // abstime will be the absolute timeout time
5398 // TODO: replace compute_abstime() with unpackTime()
5399 
5400 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5401   if (millis < 0)  millis = 0;
5402 
5403   jlong seconds = millis / 1000;
5404   millis %= 1000;
5405   if (seconds > 50000000) { // see man cond_timedwait(3T)
5406     seconds = 50000000;
5407   }
5408 
5409   if (os::supports_monotonic_clock()) {
5410     struct timespec now;
5411     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5412     assert_status(status == 0, status, "clock_gettime");
5413     abstime->tv_sec = now.tv_sec  + seconds;
5414     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5415     if (nanos >= NANOSECS_PER_SEC) {
5416       abstime->tv_sec += 1;
5417       nanos -= NANOSECS_PER_SEC;
5418     }
5419     abstime->tv_nsec = nanos;
5420   } else {
5421     struct timeval now;
5422     int status = gettimeofday(&now, NULL);
5423     assert(status == 0, "gettimeofday");
5424     abstime->tv_sec = now.tv_sec  + seconds;
5425     long usec = now.tv_usec + millis * 1000;
5426     if (usec >= 1000000) {
5427       abstime->tv_sec += 1;
5428       usec -= 1000000;
5429     }
5430     abstime->tv_nsec = usec * 1000;
5431   }
5432   return abstime;
5433 }
5434 
5435 void os::PlatformEvent::park() {       // AKA "down()"
5436   // Transitions for _Event:
5437   //   -1 => -1 : illegal
5438   //    1 =>  0 : pass - return immediately
5439   //    0 => -1 : block; then set _Event to 0 before returning
5440 
5441   // Invariant: Only the thread associated with the Event/PlatformEvent
5442   // may call park().
5443   // TODO: assert that _Assoc != NULL or _Assoc == Self
5444   assert(_nParked == 0, "invariant");
5445 
5446   int v;
5447   for (;;) {
5448     v = _Event;
5449     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5450   }
5451   guarantee(v >= 0, "invariant");
5452   if (v == 0) {
5453     // Do this the hard way by blocking ...
5454     int status = pthread_mutex_lock(_mutex);
5455     assert_status(status == 0, status, "mutex_lock");
5456     guarantee(_nParked == 0, "invariant");
5457     ++_nParked;
5458     while (_Event < 0) {
5459       status = pthread_cond_wait(_cond, _mutex);
5460       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5461       // Treat this the same as if the wait was interrupted
5462       if (status == ETIME) { status = EINTR; }
5463       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5464     }
5465     --_nParked;
5466 
5467     _Event = 0;
5468     status = pthread_mutex_unlock(_mutex);
5469     assert_status(status == 0, status, "mutex_unlock");
5470     // Paranoia to ensure our locked and lock-free paths interact
5471     // correctly with each other.
5472     OrderAccess::fence();
5473   }
5474   guarantee(_Event >= 0, "invariant");
5475 }
5476 
5477 int os::PlatformEvent::park(jlong millis) {
5478   // Transitions for _Event:
5479   //   -1 => -1 : illegal
5480   //    1 =>  0 : pass - return immediately
5481   //    0 => -1 : block; then set _Event to 0 before returning
5482 
5483   guarantee(_nParked == 0, "invariant");
5484 
5485   int v;
5486   for (;;) {
5487     v = _Event;
5488     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5489   }
5490   guarantee(v >= 0, "invariant");
5491   if (v != 0) return OS_OK;
5492 
5493   // We do this the hard way, by blocking the thread.
5494   // Consider enforcing a minimum timeout value.
5495   struct timespec abst;
5496   compute_abstime(&abst, millis);
5497 
5498   int ret = OS_TIMEOUT;
5499   int status = pthread_mutex_lock(_mutex);
5500   assert_status(status == 0, status, "mutex_lock");
5501   guarantee(_nParked == 0, "invariant");
5502   ++_nParked;
5503 
5504   // Object.wait(timo) will return because of
5505   // (a) notification
5506   // (b) timeout
5507   // (c) thread.interrupt
5508   //
5509   // Thread.interrupt and object.notify{All} both call Event::set.
5510   // That is, we treat thread.interrupt as a special case of notification.
5511   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5512   // We assume all ETIME returns are valid.
5513   //
5514   // TODO: properly differentiate simultaneous notify+interrupt.
5515   // In that case, we should propagate the notify to another waiter.
5516 
5517   while (_Event < 0) {
5518     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5519     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5520       pthread_cond_destroy(_cond);
5521       pthread_cond_init(_cond, os::Linux::condAttr());
5522     }
5523     assert_status(status == 0 || status == EINTR ||
5524                   status == ETIME || status == ETIMEDOUT,
5525                   status, "cond_timedwait");
5526     if (!FilterSpuriousWakeups) break;                 // previous semantics
5527     if (status == ETIME || status == ETIMEDOUT) break;
5528     // We consume and ignore EINTR and spurious wakeups.
5529   }
5530   --_nParked;
5531   if (_Event >= 0) {
5532     ret = OS_OK;
5533   }
5534   _Event = 0;
5535   status = pthread_mutex_unlock(_mutex);
5536   assert_status(status == 0, status, "mutex_unlock");
5537   assert(_nParked == 0, "invariant");
5538   // Paranoia to ensure our locked and lock-free paths interact
5539   // correctly with each other.
5540   OrderAccess::fence();
5541   return ret;
5542 }
5543 
5544 void os::PlatformEvent::unpark() {
5545   // Transitions for _Event:
5546   //    0 => 1 : just return
5547   //    1 => 1 : just return
5548   //   -1 => either 0 or 1; must signal target thread
5549   //         That is, we can safely transition _Event from -1 to either
5550   //         0 or 1.
5551   // See also: "Semaphores in Plan 9" by Mullender & Cox
5552   //
5553   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5554   // that it will take two back-to-back park() calls for the owning
5555   // thread to block. This has the benefit of forcing a spurious return
5556   // from the first park() call after an unpark() call which will help
5557   // shake out uses of park() and unpark() without condition variables.
5558 
5559   if (Atomic::xchg(1, &_Event) >= 0) return;
5560 
5561   // Wait for the thread associated with the event to vacate
5562   int status = pthread_mutex_lock(_mutex);
5563   assert_status(status == 0, status, "mutex_lock");
5564   int AnyWaiters = _nParked;
5565   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5566   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5567     AnyWaiters = 0;
5568     pthread_cond_signal(_cond);
5569   }
5570   status = pthread_mutex_unlock(_mutex);
5571   assert_status(status == 0, status, "mutex_unlock");
5572   if (AnyWaiters != 0) {
5573     // Note that we signal() *after* dropping the lock for "immortal" Events.
5574     // This is safe and avoids a common class of  futile wakeups.  In rare
5575     // circumstances this can cause a thread to return prematurely from
5576     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5577     // will simply re-test the condition and re-park itself.
5578     // This provides particular benefit if the underlying platform does not
5579     // provide wait morphing.
5580     status = pthread_cond_signal(_cond);
5581     assert_status(status == 0, status, "cond_signal");
5582   }
5583 }
5584 
5585 
5586 // JSR166
5587 // -------------------------------------------------------
5588 
5589 // The solaris and linux implementations of park/unpark are fairly
5590 // conservative for now, but can be improved. They currently use a
5591 // mutex/condvar pair, plus a a count.
5592 // Park decrements count if > 0, else does a condvar wait.  Unpark
5593 // sets count to 1 and signals condvar.  Only one thread ever waits
5594 // on the condvar. Contention seen when trying to park implies that someone
5595 // is unparking you, so don't wait. And spurious returns are fine, so there
5596 // is no need to track notifications.
5597 
5598 // This code is common to linux and solaris and will be moved to a
5599 // common place in dolphin.
5600 //
5601 // The passed in time value is either a relative time in nanoseconds
5602 // or an absolute time in milliseconds. Either way it has to be unpacked
5603 // into suitable seconds and nanoseconds components and stored in the
5604 // given timespec structure.
5605 // Given time is a 64-bit value and the time_t used in the timespec is only
5606 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5607 // overflow if times way in the future are given. Further on Solaris versions
5608 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5609 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5610 // As it will be 28 years before "now + 100000000" will overflow we can
5611 // ignore overflow and just impose a hard-limit on seconds using the value
5612 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5613 // years from "now".
5614 
5615 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5616   assert(time > 0, "convertTime");
5617   time_t max_secs = 0;
5618 
5619   if (!os::supports_monotonic_clock() || isAbsolute) {
5620     struct timeval now;
5621     int status = gettimeofday(&now, NULL);
5622     assert(status == 0, "gettimeofday");
5623 
5624     max_secs = now.tv_sec + MAX_SECS;
5625 
5626     if (isAbsolute) {
5627       jlong secs = time / 1000;
5628       if (secs > max_secs) {
5629         absTime->tv_sec = max_secs;
5630       } else {
5631         absTime->tv_sec = secs;
5632       }
5633       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5634     } else {
5635       jlong secs = time / NANOSECS_PER_SEC;
5636       if (secs >= MAX_SECS) {
5637         absTime->tv_sec = max_secs;
5638         absTime->tv_nsec = 0;
5639       } else {
5640         absTime->tv_sec = now.tv_sec + secs;
5641         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5642         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5643           absTime->tv_nsec -= NANOSECS_PER_SEC;
5644           ++absTime->tv_sec; // note: this must be <= max_secs
5645         }
5646       }
5647     }
5648   } else {
5649     // must be relative using monotonic clock
5650     struct timespec now;
5651     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5652     assert_status(status == 0, status, "clock_gettime");
5653     max_secs = now.tv_sec + MAX_SECS;
5654     jlong secs = time / NANOSECS_PER_SEC;
5655     if (secs >= MAX_SECS) {
5656       absTime->tv_sec = max_secs;
5657       absTime->tv_nsec = 0;
5658     } else {
5659       absTime->tv_sec = now.tv_sec + secs;
5660       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5661       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5662         absTime->tv_nsec -= NANOSECS_PER_SEC;
5663         ++absTime->tv_sec; // note: this must be <= max_secs
5664       }
5665     }
5666   }
5667   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5668   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5669   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5670   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5671 }
5672 
5673 void Parker::park(bool isAbsolute, jlong time) {
5674   // Ideally we'd do something useful while spinning, such
5675   // as calling unpackTime().
5676 
5677   // Optional fast-path check:
5678   // Return immediately if a permit is available.
5679   // We depend on Atomic::xchg() having full barrier semantics
5680   // since we are doing a lock-free update to _counter.
5681   if (Atomic::xchg(0, &_counter) > 0) return;
5682 
5683   Thread* thread = Thread::current();
5684   assert(thread->is_Java_thread(), "Must be JavaThread");
5685   JavaThread *jt = (JavaThread *)thread;
5686 
5687   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5688   // Check interrupt before trying to wait
5689   if (Thread::is_interrupted(thread, false)) {
5690     return;
5691   }
5692 
5693   // Next, demultiplex/decode time arguments
5694   timespec absTime;
5695   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5696     return;
5697   }
5698   if (time > 0) {
5699     unpackTime(&absTime, isAbsolute, time);
5700   }
5701 
5702 
5703   // Enter safepoint region
5704   // Beware of deadlocks such as 6317397.
5705   // The per-thread Parker:: mutex is a classic leaf-lock.
5706   // In particular a thread must never block on the Threads_lock while
5707   // holding the Parker:: mutex.  If safepoints are pending both the
5708   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5709   ThreadBlockInVM tbivm(jt);
5710 
5711   // Don't wait if cannot get lock since interference arises from
5712   // unblocking.  Also. check interrupt before trying wait
5713   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5714     return;
5715   }
5716 
5717   int status;
5718   if (_counter > 0)  { // no wait needed
5719     _counter = 0;
5720     status = pthread_mutex_unlock(_mutex);
5721     assert(status == 0, "invariant");
5722     // Paranoia to ensure our locked and lock-free paths interact
5723     // correctly with each other and Java-level accesses.
5724     OrderAccess::fence();
5725     return;
5726   }
5727 
5728 #ifdef ASSERT
5729   // Don't catch signals while blocked; let the running threads have the signals.
5730   // (This allows a debugger to break into the running thread.)
5731   sigset_t oldsigs;
5732   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5733   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5734 #endif
5735 
5736   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5737   jt->set_suspend_equivalent();
5738   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5739 
5740   assert(_cur_index == -1, "invariant");
5741   if (time == 0) {
5742     _cur_index = REL_INDEX; // arbitrary choice when not timed
5743     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5744   } else {
5745     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5746     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5747     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5748       pthread_cond_destroy(&_cond[_cur_index]);
5749       pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5750     }
5751   }
5752   _cur_index = -1;
5753   assert_status(status == 0 || status == EINTR ||
5754                 status == ETIME || status == ETIMEDOUT,
5755                 status, "cond_timedwait");
5756 
5757 #ifdef ASSERT
5758   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5759 #endif
5760 
5761   _counter = 0;
5762   status = pthread_mutex_unlock(_mutex);
5763   assert_status(status == 0, status, "invariant");
5764   // Paranoia to ensure our locked and lock-free paths interact
5765   // correctly with each other and Java-level accesses.
5766   OrderAccess::fence();
5767 
5768   // If externally suspended while waiting, re-suspend
5769   if (jt->handle_special_suspend_equivalent_condition()) {
5770     jt->java_suspend_self();
5771   }
5772 }
5773 
5774 void Parker::unpark() {
5775   int status = pthread_mutex_lock(_mutex);
5776   assert(status == 0, "invariant");
5777   const int s = _counter;
5778   _counter = 1;
5779   if (s < 1) {
5780     // thread might be parked
5781     if (_cur_index != -1) {
5782       // thread is definitely parked
5783       if (WorkAroundNPTLTimedWaitHang) {
5784         status = pthread_cond_signal(&_cond[_cur_index]);
5785         assert(status == 0, "invariant");
5786         status = pthread_mutex_unlock(_mutex);
5787         assert(status == 0, "invariant");
5788       } else {
5789         status = pthread_mutex_unlock(_mutex);
5790         assert(status == 0, "invariant");
5791         status = pthread_cond_signal(&_cond[_cur_index]);
5792         assert(status == 0, "invariant");
5793       }
5794     } else {
5795       pthread_mutex_unlock(_mutex);
5796       assert(status == 0, "invariant");
5797     }
5798   } else {
5799     pthread_mutex_unlock(_mutex);
5800     assert(status == 0, "invariant");
5801   }
5802 }
5803 
5804 
5805 extern char** environ;
5806 
5807 // Run the specified command in a separate process. Return its exit value,
5808 // or -1 on failure (e.g. can't fork a new process).
5809 // Unlike system(), this function can be called from signal handler. It
5810 // doesn't block SIGINT et al.
5811 int os::fork_and_exec(char* cmd) {
5812   const char * argv[4] = {"sh", "-c", cmd, NULL};
5813 
5814   pid_t pid = fork();
5815 
5816   if (pid < 0) {
5817     // fork failed
5818     return -1;
5819 
5820   } else if (pid == 0) {
5821     // child process
5822 
5823     execve("/bin/sh", (char* const*)argv, environ);
5824 
5825     // execve failed
5826     _exit(-1);
5827 
5828   } else  {
5829     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5830     // care about the actual exit code, for now.
5831 
5832     int status;
5833 
5834     // Wait for the child process to exit.  This returns immediately if
5835     // the child has already exited. */
5836     while (waitpid(pid, &status, 0) < 0) {
5837       switch (errno) {
5838       case ECHILD: return 0;
5839       case EINTR: break;
5840       default: return -1;
5841       }
5842     }
5843 
5844     if (WIFEXITED(status)) {
5845       // The child exited normally; get its exit code.
5846       return WEXITSTATUS(status);
5847     } else if (WIFSIGNALED(status)) {
5848       // The child exited because of a signal
5849       // The best value to return is 0x80 + signal number,
5850       // because that is what all Unix shells do, and because
5851       // it allows callers to distinguish between process exit and
5852       // process death by signal.
5853       return 0x80 + WTERMSIG(status);
5854     } else {
5855       // Unknown exit code; pass it through
5856       return status;
5857     }
5858   }
5859 }
5860 
5861 // is_headless_jre()
5862 //
5863 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5864 // in order to report if we are running in a headless jre
5865 //
5866 // Since JDK8 xawt/libmawt.so was moved into the same directory
5867 // as libawt.so, and renamed libawt_xawt.so
5868 //
5869 bool os::is_headless_jre() {
5870   struct stat statbuf;
5871   char buf[MAXPATHLEN];
5872   char libmawtpath[MAXPATHLEN];
5873   const char *xawtstr  = "/xawt/libmawt.so";
5874   const char *new_xawtstr = "/libawt_xawt.so";
5875   char *p;
5876 
5877   // Get path to libjvm.so
5878   os::jvm_path(buf, sizeof(buf));
5879 
5880   // Get rid of libjvm.so
5881   p = strrchr(buf, '/');
5882   if (p == NULL) {
5883     return false;
5884   } else {
5885     *p = '\0';
5886   }
5887 
5888   // Get rid of client or server
5889   p = strrchr(buf, '/');
5890   if (p == NULL) {
5891     return false;
5892   } else {
5893     *p = '\0';
5894   }
5895 
5896   // check xawt/libmawt.so
5897   strcpy(libmawtpath, buf);
5898   strcat(libmawtpath, xawtstr);
5899   if (::stat(libmawtpath, &statbuf) == 0) return false;
5900 
5901   // check libawt_xawt.so
5902   strcpy(libmawtpath, buf);
5903   strcat(libmawtpath, new_xawtstr);
5904   if (::stat(libmawtpath, &statbuf) == 0) return false;
5905 
5906   return true;
5907 }
5908 
5909 // Get the default path to the core file
5910 // Returns the length of the string
5911 int os::get_core_path(char* buffer, size_t bufferSize) {
5912   /*
5913    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5914    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5915    */
5916   const int core_pattern_len = 129;
5917   char core_pattern[core_pattern_len] = {0};
5918 
5919   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5920   if (core_pattern_file != -1) {
5921     ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5922     ::close(core_pattern_file);
5923 
5924     if (ret > 0) {
5925       char *last_char = core_pattern + strlen(core_pattern) - 1;
5926 
5927       if (*last_char == '\n') {
5928         *last_char = '\0';
5929       }
5930     }
5931   }
5932 
5933   if (strlen(core_pattern) == 0) {
5934     return -1;
5935   }
5936 
5937   char *pid_pos = strstr(core_pattern, "%p");
5938   int written;
5939 
5940   if (core_pattern[0] == '/') {
5941     written = jio_snprintf(buffer, bufferSize, core_pattern);
5942   } else {
5943     char cwd[PATH_MAX];
5944 
5945     const char* p = get_current_directory(cwd, PATH_MAX);
5946     if (p == NULL) {
5947       return -1;
5948     }
5949 
5950     if (core_pattern[0] == '|') {
5951       written = jio_snprintf(buffer, bufferSize,
5952                         "\"%s\" (or dumping to %s/core.%d)",
5953                                      &core_pattern[1], p, current_process_id());
5954     } else {
5955       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5956     }
5957   }
5958 
5959   if (written < 0) {
5960     return -1;
5961   }
5962 
5963   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5964     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5965 
5966     if (core_uses_pid_file != -1) {
5967       char core_uses_pid = 0;
5968       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5969       ::close(core_uses_pid_file);
5970 
5971       if (core_uses_pid == '1') {
5972         jio_snprintf(buffer + written, bufferSize - written,
5973                                           ".%d", current_process_id());
5974       }
5975     }
5976   }
5977 
5978   return strlen(buffer);
5979 }
5980 
5981 /////////////// Unit tests ///////////////
5982 
5983 #ifndef PRODUCT
5984 
5985 #define test_log(...)              \
5986   do {                             \
5987     if (VerboseInternalVMTests) {  \
5988       tty->print_cr(__VA_ARGS__);  \
5989       tty->flush();                \
5990     }                              \
5991   } while (false)
5992 
5993 class TestReserveMemorySpecial : AllStatic {
5994  public:
5995   static void small_page_write(void* addr, size_t size) {
5996     size_t page_size = os::vm_page_size();
5997 
5998     char* end = (char*)addr + size;
5999     for (char* p = (char*)addr; p < end; p += page_size) {
6000       *p = 1;
6001     }
6002   }
6003 
6004   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6005     if (!UseHugeTLBFS) {
6006       return;
6007     }
6008 
6009     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6010 
6011     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6012 
6013     if (addr != NULL) {
6014       small_page_write(addr, size);
6015 
6016       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6017     }
6018   }
6019 
6020   static void test_reserve_memory_special_huge_tlbfs_only() {
6021     if (!UseHugeTLBFS) {
6022       return;
6023     }
6024 
6025     size_t lp = os::large_page_size();
6026 
6027     for (size_t size = lp; size <= lp * 10; size += lp) {
6028       test_reserve_memory_special_huge_tlbfs_only(size);
6029     }
6030   }
6031 
6032   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6033     size_t lp = os::large_page_size();
6034     size_t ag = os::vm_allocation_granularity();
6035 
6036     // sizes to test
6037     const size_t sizes[] = {
6038       lp, lp + ag, lp + lp / 2, lp * 2,
6039       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6040       lp * 10, lp * 10 + lp / 2
6041     };
6042     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6043 
6044     // For each size/alignment combination, we test three scenarios:
6045     // 1) with req_addr == NULL
6046     // 2) with a non-null req_addr at which we expect to successfully allocate
6047     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6048     //    expect the allocation to either fail or to ignore req_addr
6049 
6050     // Pre-allocate two areas; they shall be as large as the largest allocation
6051     //  and aligned to the largest alignment we will be testing.
6052     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6053     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6054       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6055       -1, 0);
6056     assert(mapping1 != MAP_FAILED, "should work");
6057 
6058     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6059       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6060       -1, 0);
6061     assert(mapping2 != MAP_FAILED, "should work");
6062 
6063     // Unmap the first mapping, but leave the second mapping intact: the first
6064     // mapping will serve as a value for a "good" req_addr (case 2). The second
6065     // mapping, still intact, as "bad" req_addr (case 3).
6066     ::munmap(mapping1, mapping_size);
6067 
6068     // Case 1
6069     test_log("%s, req_addr NULL:", __FUNCTION__);
6070     test_log("size            align           result");
6071 
6072     for (int i = 0; i < num_sizes; i++) {
6073       const size_t size = sizes[i];
6074       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6075         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6076         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6077             size, alignment, p, (p != NULL ? "" : "(failed)"));
6078         if (p != NULL) {
6079           assert(is_ptr_aligned(p, alignment), "must be");
6080           small_page_write(p, size);
6081           os::Linux::release_memory_special_huge_tlbfs(p, size);
6082         }
6083       }
6084     }
6085 
6086     // Case 2
6087     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6088     test_log("size            align           req_addr         result");
6089 
6090     for (int i = 0; i < num_sizes; i++) {
6091       const size_t size = sizes[i];
6092       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6093         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6094         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6095         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6096             size, alignment, req_addr, p,
6097             ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6098         if (p != NULL) {
6099           assert(p == req_addr, "must be");
6100           small_page_write(p, size);
6101           os::Linux::release_memory_special_huge_tlbfs(p, size);
6102         }
6103       }
6104     }
6105 
6106     // Case 3
6107     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6108     test_log("size            align           req_addr         result");
6109 
6110     for (int i = 0; i < num_sizes; i++) {
6111       const size_t size = sizes[i];
6112       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6113         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6114         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6115         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6116             size, alignment, req_addr, p,
6117             ((p != NULL ? "" : "(failed)")));
6118         // as the area around req_addr contains already existing mappings, the API should always
6119         // return NULL (as per contract, it cannot return another address)
6120         assert(p == NULL, "must be");
6121       }
6122     }
6123 
6124     ::munmap(mapping2, mapping_size);
6125 
6126   }
6127 
6128   static void test_reserve_memory_special_huge_tlbfs() {
6129     if (!UseHugeTLBFS) {
6130       return;
6131     }
6132 
6133     test_reserve_memory_special_huge_tlbfs_only();
6134     test_reserve_memory_special_huge_tlbfs_mixed();
6135   }
6136 
6137   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6138     if (!UseSHM) {
6139       return;
6140     }
6141 
6142     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6143 
6144     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6145 
6146     if (addr != NULL) {
6147       assert(is_ptr_aligned(addr, alignment), "Check");
6148       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6149 
6150       small_page_write(addr, size);
6151 
6152       os::Linux::release_memory_special_shm(addr, size);
6153     }
6154   }
6155 
6156   static void test_reserve_memory_special_shm() {
6157     size_t lp = os::large_page_size();
6158     size_t ag = os::vm_allocation_granularity();
6159 
6160     for (size_t size = ag; size < lp * 3; size += ag) {
6161       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6162         test_reserve_memory_special_shm(size, alignment);
6163       }
6164     }
6165   }
6166 
6167   static void test() {
6168     test_reserve_memory_special_huge_tlbfs();
6169     test_reserve_memory_special_shm();
6170   }
6171 };
6172 
6173 void TestReserveMemorySpecial_test() {
6174   TestReserveMemorySpecial::test();
6175 }
6176 
6177 #endif