1 /*
   2  * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "mutex_linux.inline.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "os_linux.inline.hpp"
  41 #include "os_share_linux.hpp"
  42 #include "prims/jniFastGetField.hpp"
  43 #include "prims/jvm.h"
  44 #include "prims/jvm_misc.hpp"
  45 #include "runtime/arguments.hpp"
  46 #include "runtime/atomic.inline.hpp"
  47 #include "runtime/extendedPC.hpp"
  48 #include "runtime/globals.hpp"
  49 #include "runtime/interfaceSupport.hpp"
  50 #include "runtime/init.hpp"
  51 #include "runtime/java.hpp"
  52 #include "runtime/javaCalls.hpp"
  53 #include "runtime/mutexLocker.hpp"
  54 #include "runtime/objectMonitor.hpp"
  55 #include "runtime/orderAccess.inline.hpp"
  56 #include "runtime/osThread.hpp"
  57 #include "runtime/perfMemory.hpp"
  58 #include "runtime/sharedRuntime.hpp"
  59 #include "runtime/statSampler.hpp"
  60 #include "runtime/stubRoutines.hpp"
  61 #include "runtime/thread.inline.hpp"
  62 #include "runtime/threadCritical.hpp"
  63 #include "runtime/timer.hpp"
  64 #include "semaphore_posix.hpp"
  65 #include "services/attachListener.hpp"
  66 #include "services/memTracker.hpp"
  67 #include "services/runtimeService.hpp"
  68 #include "utilities/decoder.hpp"
  69 #include "utilities/defaultStream.hpp"
  70 #include "utilities/events.hpp"
  71 #include "utilities/elfFile.hpp"
  72 #include "utilities/growableArray.hpp"
  73 #include "utilities/macros.hpp"
  74 #include "utilities/vmError.hpp"
  75 
  76 // put OS-includes here
  77 # include <sys/types.h>
  78 # include <sys/mman.h>
  79 # include <sys/stat.h>
  80 # include <sys/select.h>
  81 # include <pthread.h>
  82 # include <signal.h>
  83 # include <errno.h>
  84 # include <dlfcn.h>
  85 # include <stdio.h>
  86 # include <unistd.h>
  87 # include <sys/resource.h>
  88 # include <pthread.h>
  89 # include <sys/stat.h>
  90 # include <sys/time.h>
  91 # include <sys/times.h>
  92 # include <sys/utsname.h>
  93 # include <sys/socket.h>
  94 # include <sys/wait.h>
  95 # include <pwd.h>
  96 # include <poll.h>
  97 # include <semaphore.h>
  98 # include <fcntl.h>
  99 # include <string.h>
 100 # include <syscall.h>
 101 # include <sys/sysinfo.h>
 102 # include <gnu/libc-version.h>
 103 # include <sys/ipc.h>
 104 # include <sys/shm.h>
 105 # include <link.h>
 106 # include <stdint.h>
 107 # include <inttypes.h>
 108 # include <sys/ioctl.h>
 109 
 110 #ifndef _GNU_SOURCE
 111   #define _GNU_SOURCE
 112   #include <sched.h>
 113   #undef _GNU_SOURCE
 114 #else
 115   #include <sched.h>
 116 #endif
 117 
 118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 119 // getrusage() is prepared to handle the associated failure.
 120 #ifndef RUSAGE_THREAD
 121   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 122 #endif
 123 
 124 #define MAX_PATH    (2 * K)
 125 
 126 #define MAX_SECS 100000000
 127 
 128 // for timer info max values which include all bits
 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 130 
 131 #define LARGEPAGES_BIT (1 << 6)
 132 ////////////////////////////////////////////////////////////////////////////////
 133 // global variables
 134 julong os::Linux::_physical_memory = 0;
 135 
 136 address   os::Linux::_initial_thread_stack_bottom = NULL;
 137 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 138 
 139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 142 Mutex* os::Linux::_createThread_lock = NULL;
 143 pthread_t os::Linux::_main_thread;
 144 int os::Linux::_page_size = -1;
 145 const int os::Linux::_vm_default_page_size = (8 * K);
 146 bool os::Linux::_supports_fast_thread_cpu_time = false;
 147 const char * os::Linux::_glibc_version = NULL;
 148 const char * os::Linux::_libpthread_version = NULL;
 149 pthread_condattr_t os::Linux::_condattr[1];
 150 
 151 static jlong initial_time_count=0;
 152 
 153 static int clock_tics_per_sec = 100;
 154 
 155 // For diagnostics to print a message once. see run_periodic_checks
 156 static sigset_t check_signal_done;
 157 static bool check_signals = true;
 158 
 159 // Signal number used to suspend/resume a thread
 160 
 161 // do not use any signal number less than SIGSEGV, see 4355769
 162 static int SR_signum = SIGUSR2;
 163 sigset_t SR_sigset;
 164 
 165 // Declarations
 166 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 167 
 168 // utility functions
 169 
 170 static int SR_initialize();
 171 
 172 julong os::available_memory() {
 173   return Linux::available_memory();
 174 }
 175 
 176 julong os::Linux::available_memory() {
 177   // values in struct sysinfo are "unsigned long"
 178   struct sysinfo si;
 179   sysinfo(&si);
 180 
 181   return (julong)si.freeram * si.mem_unit;
 182 }
 183 
 184 julong os::physical_memory() {
 185   return Linux::physical_memory();
 186 }
 187 
 188 // Return true if user is running as root.
 189 
 190 bool os::have_special_privileges() {
 191   static bool init = false;
 192   static bool privileges = false;
 193   if (!init) {
 194     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 195     init = true;
 196   }
 197   return privileges;
 198 }
 199 
 200 
 201 #ifndef SYS_gettid
 202 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 203   #ifdef __ia64__
 204     #define SYS_gettid 1105
 205   #else
 206     #ifdef __i386__
 207       #define SYS_gettid 224
 208     #else
 209       #ifdef __amd64__
 210         #define SYS_gettid 186
 211       #else
 212         #ifdef __sparc__
 213           #define SYS_gettid 143
 214         #else
 215           #error define gettid for the arch
 216         #endif
 217       #endif
 218     #endif
 219   #endif
 220 #endif
 221 
 222 // Cpu architecture string
 223 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 224 
 225 
 226 // pid_t gettid()
 227 //
 228 // Returns the kernel thread id of the currently running thread. Kernel
 229 // thread id is used to access /proc.
 230 pid_t os::Linux::gettid() {
 231   int rslt = syscall(SYS_gettid);
 232   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 233   return (pid_t)rslt;
 234 }
 235 
 236 // Most versions of linux have a bug where the number of processors are
 237 // determined by looking at the /proc file system.  In a chroot environment,
 238 // the system call returns 1.  This causes the VM to act as if it is
 239 // a single processor and elide locking (see is_MP() call).
 240 static bool unsafe_chroot_detected = false;
 241 static const char *unstable_chroot_error = "/proc file system not found.\n"
 242                      "Java may be unstable running multithreaded in a chroot "
 243                      "environment on Linux when /proc filesystem is not mounted.";
 244 
 245 void os::Linux::initialize_system_info() {
 246   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 247   if (processor_count() == 1) {
 248     pid_t pid = os::Linux::gettid();
 249     char fname[32];
 250     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 251     FILE *fp = fopen(fname, "r");
 252     if (fp == NULL) {
 253       unsafe_chroot_detected = true;
 254     } else {
 255       fclose(fp);
 256     }
 257   }
 258   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 259   assert(processor_count() > 0, "linux error");
 260 }
 261 
 262 void os::init_system_properties_values() {
 263   // The next steps are taken in the product version:
 264   //
 265   // Obtain the JAVA_HOME value from the location of libjvm.so.
 266   // This library should be located at:
 267   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 268   //
 269   // If "/jre/lib/" appears at the right place in the path, then we
 270   // assume libjvm.so is installed in a JDK and we use this path.
 271   //
 272   // Otherwise exit with message: "Could not create the Java virtual machine."
 273   //
 274   // The following extra steps are taken in the debugging version:
 275   //
 276   // If "/jre/lib/" does NOT appear at the right place in the path
 277   // instead of exit check for $JAVA_HOME environment variable.
 278   //
 279   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 280   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 281   // it looks like libjvm.so is installed there
 282   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 283   //
 284   // Otherwise exit.
 285   //
 286   // Important note: if the location of libjvm.so changes this
 287   // code needs to be changed accordingly.
 288 
 289   // See ld(1):
 290   //      The linker uses the following search paths to locate required
 291   //      shared libraries:
 292   //        1: ...
 293   //        ...
 294   //        7: The default directories, normally /lib and /usr/lib.
 295 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 296   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 297 #else
 298   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 299 #endif
 300 
 301 // Base path of extensions installed on the system.
 302 #define SYS_EXT_DIR     "/usr/java/packages"
 303 #define EXTENSIONS_DIR  "/lib/ext"
 304 
 305   // Buffer that fits several sprintfs.
 306   // Note that the space for the colon and the trailing null are provided
 307   // by the nulls included by the sizeof operator.
 308   const size_t bufsize =
 309     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 310          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 311   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 312 
 313   // sysclasspath, java_home, dll_dir
 314   {
 315     char *pslash;
 316     os::jvm_path(buf, bufsize);
 317 
 318     // Found the full path to libjvm.so.
 319     // Now cut the path to <java_home>/jre if we can.
 320     pslash = strrchr(buf, '/');
 321     if (pslash != NULL) {
 322       *pslash = '\0';            // Get rid of /libjvm.so.
 323     }
 324     pslash = strrchr(buf, '/');
 325     if (pslash != NULL) {
 326       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 327     }
 328     Arguments::set_dll_dir(buf);
 329 
 330     if (pslash != NULL) {
 331       pslash = strrchr(buf, '/');
 332       if (pslash != NULL) {
 333         *pslash = '\0';          // Get rid of /<arch>.
 334         pslash = strrchr(buf, '/');
 335         if (pslash != NULL) {
 336           *pslash = '\0';        // Get rid of /lib.
 337         }
 338       }
 339     }
 340     Arguments::set_java_home(buf);
 341     set_boot_path('/', ':');
 342   }
 343 
 344   // Where to look for native libraries.
 345   //
 346   // Note: Due to a legacy implementation, most of the library path
 347   // is set in the launcher. This was to accomodate linking restrictions
 348   // on legacy Linux implementations (which are no longer supported).
 349   // Eventually, all the library path setting will be done here.
 350   //
 351   // However, to prevent the proliferation of improperly built native
 352   // libraries, the new path component /usr/java/packages is added here.
 353   // Eventually, all the library path setting will be done here.
 354   {
 355     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 356     // should always exist (until the legacy problem cited above is
 357     // addressed).
 358     const char *v = ::getenv("LD_LIBRARY_PATH");
 359     const char *v_colon = ":";
 360     if (v == NULL) { v = ""; v_colon = ""; }
 361     // That's +1 for the colon and +1 for the trailing '\0'.
 362     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 363                                                      strlen(v) + 1 +
 364                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 365                                                      mtInternal);
 366     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 367     Arguments::set_library_path(ld_library_path);
 368     FREE_C_HEAP_ARRAY(char, ld_library_path);
 369   }
 370 
 371   // Extensions directories.
 372   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 373   Arguments::set_ext_dirs(buf);
 374 
 375   FREE_C_HEAP_ARRAY(char, buf);
 376 
 377 #undef DEFAULT_LIBPATH
 378 #undef SYS_EXT_DIR
 379 #undef EXTENSIONS_DIR
 380 }
 381 
 382 ////////////////////////////////////////////////////////////////////////////////
 383 // breakpoint support
 384 
 385 void os::breakpoint() {
 386   BREAKPOINT;
 387 }
 388 
 389 extern "C" void breakpoint() {
 390   // use debugger to set breakpoint here
 391 }
 392 
 393 ////////////////////////////////////////////////////////////////////////////////
 394 // signal support
 395 
 396 debug_only(static bool signal_sets_initialized = false);
 397 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 398 
 399 bool os::Linux::is_sig_ignored(int sig) {
 400   struct sigaction oact;
 401   sigaction(sig, (struct sigaction*)NULL, &oact);
 402   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 403                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 404   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 405     return true;
 406   } else {
 407     return false;
 408   }
 409 }
 410 
 411 void os::Linux::signal_sets_init() {
 412   // Should also have an assertion stating we are still single-threaded.
 413   assert(!signal_sets_initialized, "Already initialized");
 414   // Fill in signals that are necessarily unblocked for all threads in
 415   // the VM. Currently, we unblock the following signals:
 416   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 417   //                         by -Xrs (=ReduceSignalUsage));
 418   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 419   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 420   // the dispositions or masks wrt these signals.
 421   // Programs embedding the VM that want to use the above signals for their
 422   // own purposes must, at this time, use the "-Xrs" option to prevent
 423   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 424   // (See bug 4345157, and other related bugs).
 425   // In reality, though, unblocking these signals is really a nop, since
 426   // these signals are not blocked by default.
 427   sigemptyset(&unblocked_sigs);
 428   sigemptyset(&allowdebug_blocked_sigs);
 429   sigaddset(&unblocked_sigs, SIGILL);
 430   sigaddset(&unblocked_sigs, SIGSEGV);
 431   sigaddset(&unblocked_sigs, SIGBUS);
 432   sigaddset(&unblocked_sigs, SIGFPE);
 433 #if defined(PPC64)
 434   sigaddset(&unblocked_sigs, SIGTRAP);
 435 #endif
 436   sigaddset(&unblocked_sigs, SR_signum);
 437 
 438   if (!ReduceSignalUsage) {
 439     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 440       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 441       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 442     }
 443     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 444       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 445       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 446     }
 447     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 448       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 449       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 450     }
 451   }
 452   // Fill in signals that are blocked by all but the VM thread.
 453   sigemptyset(&vm_sigs);
 454   if (!ReduceSignalUsage) {
 455     sigaddset(&vm_sigs, BREAK_SIGNAL);
 456   }
 457   debug_only(signal_sets_initialized = true);
 458 
 459 }
 460 
 461 // These are signals that are unblocked while a thread is running Java.
 462 // (For some reason, they get blocked by default.)
 463 sigset_t* os::Linux::unblocked_signals() {
 464   assert(signal_sets_initialized, "Not initialized");
 465   return &unblocked_sigs;
 466 }
 467 
 468 // These are the signals that are blocked while a (non-VM) thread is
 469 // running Java. Only the VM thread handles these signals.
 470 sigset_t* os::Linux::vm_signals() {
 471   assert(signal_sets_initialized, "Not initialized");
 472   return &vm_sigs;
 473 }
 474 
 475 // These are signals that are blocked during cond_wait to allow debugger in
 476 sigset_t* os::Linux::allowdebug_blocked_signals() {
 477   assert(signal_sets_initialized, "Not initialized");
 478   return &allowdebug_blocked_sigs;
 479 }
 480 
 481 void os::Linux::hotspot_sigmask(Thread* thread) {
 482 
 483   //Save caller's signal mask before setting VM signal mask
 484   sigset_t caller_sigmask;
 485   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 486 
 487   OSThread* osthread = thread->osthread();
 488   osthread->set_caller_sigmask(caller_sigmask);
 489 
 490   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 491 
 492   if (!ReduceSignalUsage) {
 493     if (thread->is_VM_thread()) {
 494       // Only the VM thread handles BREAK_SIGNAL ...
 495       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 496     } else {
 497       // ... all other threads block BREAK_SIGNAL
 498       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 499     }
 500   }
 501 }
 502 
 503 //////////////////////////////////////////////////////////////////////////////
 504 // detecting pthread library
 505 
 506 void os::Linux::libpthread_init() {
 507   // Save glibc and pthread version strings.
 508 #if !defined(_CS_GNU_LIBC_VERSION) || \
 509     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 510   #error "glibc too old (< 2.3.2)"
 511 #endif
 512 
 513   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 514   assert(n > 0, "cannot retrieve glibc version");
 515   char *str = (char *)malloc(n, mtInternal);
 516   confstr(_CS_GNU_LIBC_VERSION, str, n);
 517   os::Linux::set_glibc_version(str);
 518 
 519   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 520   assert(n > 0, "cannot retrieve pthread version");
 521   str = (char *)malloc(n, mtInternal);
 522   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 523   os::Linux::set_libpthread_version(str);
 524 }
 525 
 526 /////////////////////////////////////////////////////////////////////////////
 527 // thread stack expansion
 528 
 529 // os::Linux::manually_expand_stack() takes care of expanding the thread
 530 // stack. Note that this is normally not needed: pthread stacks allocate
 531 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 532 // committed. Therefore it is not necessary to expand the stack manually.
 533 //
 534 // Manually expanding the stack was historically needed on LinuxThreads
 535 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 536 // it is kept to deal with very rare corner cases:
 537 //
 538 // For one, user may run the VM on an own implementation of threads
 539 // whose stacks are - like the old LinuxThreads - implemented using
 540 // mmap(MAP_GROWSDOWN).
 541 //
 542 // Also, this coding may be needed if the VM is running on the primordial
 543 // thread. Normally we avoid running on the primordial thread; however,
 544 // user may still invoke the VM on the primordial thread.
 545 //
 546 // The following historical comment describes the details about running
 547 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 548 
 549 
 550 // Force Linux kernel to expand current thread stack. If "bottom" is close
 551 // to the stack guard, caller should block all signals.
 552 //
 553 // MAP_GROWSDOWN:
 554 //   A special mmap() flag that is used to implement thread stacks. It tells
 555 //   kernel that the memory region should extend downwards when needed. This
 556 //   allows early versions of LinuxThreads to only mmap the first few pages
 557 //   when creating a new thread. Linux kernel will automatically expand thread
 558 //   stack as needed (on page faults).
 559 //
 560 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 561 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 562 //   region, it's hard to tell if the fault is due to a legitimate stack
 563 //   access or because of reading/writing non-exist memory (e.g. buffer
 564 //   overrun). As a rule, if the fault happens below current stack pointer,
 565 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 566 //   application (see Linux kernel fault.c).
 567 //
 568 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 569 //   stack overflow detection.
 570 //
 571 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 572 //   not use MAP_GROWSDOWN.
 573 //
 574 // To get around the problem and allow stack banging on Linux, we need to
 575 // manually expand thread stack after receiving the SIGSEGV.
 576 //
 577 // There are two ways to expand thread stack to address "bottom", we used
 578 // both of them in JVM before 1.5:
 579 //   1. adjust stack pointer first so that it is below "bottom", and then
 580 //      touch "bottom"
 581 //   2. mmap() the page in question
 582 //
 583 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 584 // if current sp is already near the lower end of page 101, and we need to
 585 // call mmap() to map page 100, it is possible that part of the mmap() frame
 586 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 587 // That will destroy the mmap() frame and cause VM to crash.
 588 //
 589 // The following code works by adjusting sp first, then accessing the "bottom"
 590 // page to force a page fault. Linux kernel will then automatically expand the
 591 // stack mapping.
 592 //
 593 // _expand_stack_to() assumes its frame size is less than page size, which
 594 // should always be true if the function is not inlined.
 595 
 596 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 597   #define NOINLINE
 598 #else
 599   #define NOINLINE __attribute__ ((noinline))
 600 #endif
 601 
 602 static void _expand_stack_to(address bottom) NOINLINE;
 603 
 604 static void _expand_stack_to(address bottom) {
 605   address sp;
 606   size_t size;
 607   volatile char *p;
 608 
 609   // Adjust bottom to point to the largest address within the same page, it
 610   // gives us a one-page buffer if alloca() allocates slightly more memory.
 611   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 612   bottom += os::Linux::page_size() - 1;
 613 
 614   // sp might be slightly above current stack pointer; if that's the case, we
 615   // will alloca() a little more space than necessary, which is OK. Don't use
 616   // os::current_stack_pointer(), as its result can be slightly below current
 617   // stack pointer, causing us to not alloca enough to reach "bottom".
 618   sp = (address)&sp;
 619 
 620   if (sp > bottom) {
 621     size = sp - bottom;
 622     p = (volatile char *)alloca(size);
 623     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 624     p[0] = '\0';
 625   }
 626 }
 627 
 628 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 629   assert(t!=NULL, "just checking");
 630   assert(t->osthread()->expanding_stack(), "expand should be set");
 631   assert(t->stack_base() != NULL, "stack_base was not initialized");
 632 
 633   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 634     sigset_t mask_all, old_sigset;
 635     sigfillset(&mask_all);
 636     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 637     _expand_stack_to(addr);
 638     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 639     return true;
 640   }
 641   return false;
 642 }
 643 
 644 //////////////////////////////////////////////////////////////////////////////
 645 // create new thread
 646 
 647 // Thread start routine for all newly created threads
 648 static void *java_start(Thread *thread) {
 649   // Try to randomize the cache line index of hot stack frames.
 650   // This helps when threads of the same stack traces evict each other's
 651   // cache lines. The threads can be either from the same JVM instance, or
 652   // from different JVM instances. The benefit is especially true for
 653   // processors with hyperthreading technology.
 654   static int counter = 0;
 655   int pid = os::current_process_id();
 656   alloca(((pid ^ counter++) & 7) * 128);
 657 
 658   thread->initialize_thread_current();
 659 
 660   OSThread* osthread = thread->osthread();
 661   Monitor* sync = osthread->startThread_lock();
 662 
 663   osthread->set_thread_id(os::current_thread_id());
 664 
 665   if (UseNUMA) {
 666     int lgrp_id = os::numa_get_group_id();
 667     if (lgrp_id != -1) {
 668       thread->set_lgrp_id(lgrp_id);
 669     }
 670   }
 671   // initialize signal mask for this thread
 672   os::Linux::hotspot_sigmask(thread);
 673 
 674   // initialize floating point control register
 675   os::Linux::init_thread_fpu_state();
 676 
 677   // handshaking with parent thread
 678   {
 679     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 680 
 681     // notify parent thread
 682     osthread->set_state(INITIALIZED);
 683     sync->notify_all();
 684 
 685     // wait until os::start_thread()
 686     while (osthread->get_state() == INITIALIZED) {
 687       sync->wait(Mutex::_no_safepoint_check_flag);
 688     }
 689   }
 690 
 691   // call one more level start routine
 692   thread->run();
 693 
 694   return 0;
 695 }
 696 
 697 bool os::create_thread(Thread* thread, ThreadType thr_type,
 698                        size_t stack_size) {
 699   assert(thread->osthread() == NULL, "caller responsible");
 700 
 701   // Allocate the OSThread object
 702   OSThread* osthread = new OSThread(NULL, NULL);
 703   if (osthread == NULL) {
 704     return false;
 705   }
 706 
 707   // set the correct thread state
 708   osthread->set_thread_type(thr_type);
 709 
 710   // Initial state is ALLOCATED but not INITIALIZED
 711   osthread->set_state(ALLOCATED);
 712 
 713   thread->set_osthread(osthread);
 714 
 715   // init thread attributes
 716   pthread_attr_t attr;
 717   pthread_attr_init(&attr);
 718   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 719 
 720   // stack size
 721   // calculate stack size if it's not specified by caller
 722   if (stack_size == 0) {
 723     stack_size = os::Linux::default_stack_size(thr_type);
 724 
 725     switch (thr_type) {
 726     case os::java_thread:
 727       // Java threads use ThreadStackSize which default value can be
 728       // changed with the flag -Xss
 729       assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 730       stack_size = JavaThread::stack_size_at_create();
 731       break;
 732     case os::compiler_thread:
 733       if (CompilerThreadStackSize > 0) {
 734         stack_size = (size_t)(CompilerThreadStackSize * K);
 735         break;
 736       } // else fall through:
 737         // use VMThreadStackSize if CompilerThreadStackSize is not defined
 738     case os::vm_thread:
 739     case os::pgc_thread:
 740     case os::cgc_thread:
 741     case os::watcher_thread:
 742       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 743       break;
 744     }
 745   }
 746 
 747   stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 748   pthread_attr_setstacksize(&attr, stack_size);
 749 
 750   // glibc guard page
 751   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 752 
 753   ThreadState state;
 754 
 755   {
 756     pthread_t tid;
 757     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 758 
 759     pthread_attr_destroy(&attr);
 760 
 761     if (ret != 0) {
 762       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 763         perror("pthread_create()");
 764       }
 765       // Need to clean up stuff we've allocated so far
 766       thread->set_osthread(NULL);
 767       delete osthread;
 768       return false;
 769     }
 770 
 771     // Store pthread info into the OSThread
 772     osthread->set_pthread_id(tid);
 773 
 774     // Wait until child thread is either initialized or aborted
 775     {
 776       Monitor* sync_with_child = osthread->startThread_lock();
 777       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 778       while ((state = osthread->get_state()) == ALLOCATED) {
 779         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 780       }
 781     }
 782   }
 783 
 784   // Aborted due to thread limit being reached
 785   if (state == ZOMBIE) {
 786     thread->set_osthread(NULL);
 787     delete osthread;
 788     return false;
 789   }
 790 
 791   // The thread is returned suspended (in state INITIALIZED),
 792   // and is started higher up in the call chain
 793   assert(state == INITIALIZED, "race condition");
 794   return true;
 795 }
 796 
 797 /////////////////////////////////////////////////////////////////////////////
 798 // attach existing thread
 799 
 800 // bootstrap the main thread
 801 bool os::create_main_thread(JavaThread* thread) {
 802   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 803   return create_attached_thread(thread);
 804 }
 805 
 806 bool os::create_attached_thread(JavaThread* thread) {
 807 #ifdef ASSERT
 808   thread->verify_not_published();
 809 #endif
 810 
 811   // Allocate the OSThread object
 812   OSThread* osthread = new OSThread(NULL, NULL);
 813 
 814   if (osthread == NULL) {
 815     return false;
 816   }
 817 
 818   // Store pthread info into the OSThread
 819   osthread->set_thread_id(os::Linux::gettid());
 820   osthread->set_pthread_id(::pthread_self());
 821 
 822   // initialize floating point control register
 823   os::Linux::init_thread_fpu_state();
 824 
 825   // Initial thread state is RUNNABLE
 826   osthread->set_state(RUNNABLE);
 827 
 828   thread->set_osthread(osthread);
 829 
 830   if (UseNUMA) {
 831     int lgrp_id = os::numa_get_group_id();
 832     if (lgrp_id != -1) {
 833       thread->set_lgrp_id(lgrp_id);
 834     }
 835   }
 836 
 837   if (os::Linux::is_initial_thread()) {
 838     // If current thread is initial thread, its stack is mapped on demand,
 839     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 840     // the entire stack region to avoid SEGV in stack banging.
 841     // It is also useful to get around the heap-stack-gap problem on SuSE
 842     // kernel (see 4821821 for details). We first expand stack to the top
 843     // of yellow zone, then enable stack yellow zone (order is significant,
 844     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 845     // is no gap between the last two virtual memory regions.
 846 
 847     JavaThread *jt = (JavaThread *)thread;
 848     address addr = jt->stack_reserved_zone_base();
 849     assert(addr != NULL, "initialization problem?");
 850     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 851 
 852     osthread->set_expanding_stack();
 853     os::Linux::manually_expand_stack(jt, addr);
 854     osthread->clear_expanding_stack();
 855   }
 856 
 857   // initialize signal mask for this thread
 858   // and save the caller's signal mask
 859   os::Linux::hotspot_sigmask(thread);
 860 
 861   return true;
 862 }
 863 
 864 void os::pd_start_thread(Thread* thread) {
 865   OSThread * osthread = thread->osthread();
 866   assert(osthread->get_state() != INITIALIZED, "just checking");
 867   Monitor* sync_with_child = osthread->startThread_lock();
 868   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 869   sync_with_child->notify();
 870 }
 871 
 872 // Free Linux resources related to the OSThread
 873 void os::free_thread(OSThread* osthread) {
 874   assert(osthread != NULL, "osthread not set");
 875 
 876   if (Thread::current()->osthread() == osthread) {
 877     // Restore caller's signal mask
 878     sigset_t sigmask = osthread->caller_sigmask();
 879     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 880   }
 881 
 882   delete osthread;
 883 }
 884 
 885 //////////////////////////////////////////////////////////////////////////////
 886 // initial thread
 887 
 888 // Check if current thread is the initial thread, similar to Solaris thr_main.
 889 bool os::Linux::is_initial_thread(void) {
 890   char dummy;
 891   // If called before init complete, thread stack bottom will be null.
 892   // Can be called if fatal error occurs before initialization.
 893   if (initial_thread_stack_bottom() == NULL) return false;
 894   assert(initial_thread_stack_bottom() != NULL &&
 895          initial_thread_stack_size()   != 0,
 896          "os::init did not locate initial thread's stack region");
 897   if ((address)&dummy >= initial_thread_stack_bottom() &&
 898       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 899     return true;
 900   } else {
 901     return false;
 902   }
 903 }
 904 
 905 // Find the virtual memory area that contains addr
 906 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 907   FILE *fp = fopen("/proc/self/maps", "r");
 908   if (fp) {
 909     address low, high;
 910     while (!feof(fp)) {
 911       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 912         if (low <= addr && addr < high) {
 913           if (vma_low)  *vma_low  = low;
 914           if (vma_high) *vma_high = high;
 915           fclose(fp);
 916           return true;
 917         }
 918       }
 919       for (;;) {
 920         int ch = fgetc(fp);
 921         if (ch == EOF || ch == (int)'\n') break;
 922       }
 923     }
 924     fclose(fp);
 925   }
 926   return false;
 927 }
 928 
 929 // Locate initial thread stack. This special handling of initial thread stack
 930 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 931 // bogus value for initial thread.
 932 void os::Linux::capture_initial_stack(size_t max_size) {
 933   // stack size is the easy part, get it from RLIMIT_STACK
 934   size_t stack_size;
 935   struct rlimit rlim;
 936   getrlimit(RLIMIT_STACK, &rlim);
 937   stack_size = rlim.rlim_cur;
 938 
 939   // 6308388: a bug in ld.so will relocate its own .data section to the
 940   //   lower end of primordial stack; reduce ulimit -s value a little bit
 941   //   so we won't install guard page on ld.so's data section.
 942   stack_size -= 2 * page_size();
 943 
 944   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
 945   //   7.1, in both cases we will get 2G in return value.
 946   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
 947   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
 948   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
 949   //   in case other parts in glibc still assumes 2M max stack size.
 950   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
 951   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
 952   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
 953     stack_size = 2 * K * K IA64_ONLY(*2);
 954   }
 955   // Try to figure out where the stack base (top) is. This is harder.
 956   //
 957   // When an application is started, glibc saves the initial stack pointer in
 958   // a global variable "__libc_stack_end", which is then used by system
 959   // libraries. __libc_stack_end should be pretty close to stack top. The
 960   // variable is available since the very early days. However, because it is
 961   // a private interface, it could disappear in the future.
 962   //
 963   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 964   // to __libc_stack_end, it is very close to stack top, but isn't the real
 965   // stack top. Note that /proc may not exist if VM is running as a chroot
 966   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 967   // /proc/<pid>/stat could change in the future (though unlikely).
 968   //
 969   // We try __libc_stack_end first. If that doesn't work, look for
 970   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
 971   // as a hint, which should work well in most cases.
 972 
 973   uintptr_t stack_start;
 974 
 975   // try __libc_stack_end first
 976   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
 977   if (p && *p) {
 978     stack_start = *p;
 979   } else {
 980     // see if we can get the start_stack field from /proc/self/stat
 981     FILE *fp;
 982     int pid;
 983     char state;
 984     int ppid;
 985     int pgrp;
 986     int session;
 987     int nr;
 988     int tpgrp;
 989     unsigned long flags;
 990     unsigned long minflt;
 991     unsigned long cminflt;
 992     unsigned long majflt;
 993     unsigned long cmajflt;
 994     unsigned long utime;
 995     unsigned long stime;
 996     long cutime;
 997     long cstime;
 998     long prio;
 999     long nice;
1000     long junk;
1001     long it_real;
1002     uintptr_t start;
1003     uintptr_t vsize;
1004     intptr_t rss;
1005     uintptr_t rsslim;
1006     uintptr_t scodes;
1007     uintptr_t ecode;
1008     int i;
1009 
1010     // Figure what the primordial thread stack base is. Code is inspired
1011     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1012     // followed by command name surrounded by parentheses, state, etc.
1013     char stat[2048];
1014     int statlen;
1015 
1016     fp = fopen("/proc/self/stat", "r");
1017     if (fp) {
1018       statlen = fread(stat, 1, 2047, fp);
1019       stat[statlen] = '\0';
1020       fclose(fp);
1021 
1022       // Skip pid and the command string. Note that we could be dealing with
1023       // weird command names, e.g. user could decide to rename java launcher
1024       // to "java 1.4.2 :)", then the stat file would look like
1025       //                1234 (java 1.4.2 :)) R ... ...
1026       // We don't really need to know the command string, just find the last
1027       // occurrence of ")" and then start parsing from there. See bug 4726580.
1028       char * s = strrchr(stat, ')');
1029 
1030       i = 0;
1031       if (s) {
1032         // Skip blank chars
1033         do { s++; } while (s && isspace(*s));
1034 
1035 #define _UFM UINTX_FORMAT
1036 #define _DFM INTX_FORMAT
1037 
1038         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1039         //              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
1040         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,
1041                    &state,          // 3  %c
1042                    &ppid,           // 4  %d
1043                    &pgrp,           // 5  %d
1044                    &session,        // 6  %d
1045                    &nr,             // 7  %d
1046                    &tpgrp,          // 8  %d
1047                    &flags,          // 9  %lu
1048                    &minflt,         // 10 %lu
1049                    &cminflt,        // 11 %lu
1050                    &majflt,         // 12 %lu
1051                    &cmajflt,        // 13 %lu
1052                    &utime,          // 14 %lu
1053                    &stime,          // 15 %lu
1054                    &cutime,         // 16 %ld
1055                    &cstime,         // 17 %ld
1056                    &prio,           // 18 %ld
1057                    &nice,           // 19 %ld
1058                    &junk,           // 20 %ld
1059                    &it_real,        // 21 %ld
1060                    &start,          // 22 UINTX_FORMAT
1061                    &vsize,          // 23 UINTX_FORMAT
1062                    &rss,            // 24 INTX_FORMAT
1063                    &rsslim,         // 25 UINTX_FORMAT
1064                    &scodes,         // 26 UINTX_FORMAT
1065                    &ecode,          // 27 UINTX_FORMAT
1066                    &stack_start);   // 28 UINTX_FORMAT
1067       }
1068 
1069 #undef _UFM
1070 #undef _DFM
1071 
1072       if (i != 28 - 2) {
1073         assert(false, "Bad conversion from /proc/self/stat");
1074         // product mode - assume we are the initial thread, good luck in the
1075         // embedded case.
1076         warning("Can't detect initial thread stack location - bad conversion");
1077         stack_start = (uintptr_t) &rlim;
1078       }
1079     } else {
1080       // For some reason we can't open /proc/self/stat (for example, running on
1081       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1082       // most cases, so don't abort:
1083       warning("Can't detect initial thread stack location - no /proc/self/stat");
1084       stack_start = (uintptr_t) &rlim;
1085     }
1086   }
1087 
1088   // Now we have a pointer (stack_start) very close to the stack top, the
1089   // next thing to do is to figure out the exact location of stack top. We
1090   // can find out the virtual memory area that contains stack_start by
1091   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1092   // and its upper limit is the real stack top. (again, this would fail if
1093   // running inside chroot, because /proc may not exist.)
1094 
1095   uintptr_t stack_top;
1096   address low, high;
1097   if (find_vma((address)stack_start, &low, &high)) {
1098     // success, "high" is the true stack top. (ignore "low", because initial
1099     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1100     stack_top = (uintptr_t)high;
1101   } else {
1102     // failed, likely because /proc/self/maps does not exist
1103     warning("Can't detect initial thread stack location - find_vma failed");
1104     // best effort: stack_start is normally within a few pages below the real
1105     // stack top, use it as stack top, and reduce stack size so we won't put
1106     // guard page outside stack.
1107     stack_top = stack_start;
1108     stack_size -= 16 * page_size();
1109   }
1110 
1111   // stack_top could be partially down the page so align it
1112   stack_top = align_size_up(stack_top, page_size());
1113 
1114   if (max_size && stack_size > max_size) {
1115     _initial_thread_stack_size = max_size;
1116   } else {
1117     _initial_thread_stack_size = stack_size;
1118   }
1119 
1120   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1121   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1122 }
1123 
1124 ////////////////////////////////////////////////////////////////////////////////
1125 // time support
1126 
1127 // Time since start-up in seconds to a fine granularity.
1128 // Used by VMSelfDestructTimer and the MemProfiler.
1129 double os::elapsedTime() {
1130 
1131   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1132 }
1133 
1134 jlong os::elapsed_counter() {
1135   return javaTimeNanos() - initial_time_count;
1136 }
1137 
1138 jlong os::elapsed_frequency() {
1139   return NANOSECS_PER_SEC; // nanosecond resolution
1140 }
1141 
1142 bool os::supports_vtime() { return true; }
1143 bool os::enable_vtime()   { return false; }
1144 bool os::vtime_enabled()  { return false; }
1145 
1146 double os::elapsedVTime() {
1147   struct rusage usage;
1148   int retval = getrusage(RUSAGE_THREAD, &usage);
1149   if (retval == 0) {
1150     return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1151   } else {
1152     // better than nothing, but not much
1153     return elapsedTime();
1154   }
1155 }
1156 
1157 jlong os::javaTimeMillis() {
1158   timeval time;
1159   int status = gettimeofday(&time, NULL);
1160   assert(status != -1, "linux error");
1161   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1162 }
1163 
1164 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1165   timeval time;
1166   int status = gettimeofday(&time, NULL);
1167   assert(status != -1, "linux error");
1168   seconds = jlong(time.tv_sec);
1169   nanos = jlong(time.tv_usec) * 1000;
1170 }
1171 
1172 
1173 #ifndef CLOCK_MONOTONIC
1174   #define CLOCK_MONOTONIC (1)
1175 #endif
1176 
1177 void os::Linux::clock_init() {
1178   // we do dlopen's in this particular order due to bug in linux
1179   // dynamical loader (see 6348968) leading to crash on exit
1180   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1181   if (handle == NULL) {
1182     handle = dlopen("librt.so", RTLD_LAZY);
1183   }
1184 
1185   if (handle) {
1186     int (*clock_getres_func)(clockid_t, struct timespec*) =
1187            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1188     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1189            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1190     if (clock_getres_func && clock_gettime_func) {
1191       // See if monotonic clock is supported by the kernel. Note that some
1192       // early implementations simply return kernel jiffies (updated every
1193       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1194       // for nano time (though the monotonic property is still nice to have).
1195       // It's fixed in newer kernels, however clock_getres() still returns
1196       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1197       // resolution for now. Hopefully as people move to new kernels, this
1198       // won't be a problem.
1199       struct timespec res;
1200       struct timespec tp;
1201       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1202           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1203         // yes, monotonic clock is supported
1204         _clock_gettime = clock_gettime_func;
1205         return;
1206       } else {
1207         // close librt if there is no monotonic clock
1208         dlclose(handle);
1209       }
1210     }
1211   }
1212   warning("No monotonic clock was available - timed services may " \
1213           "be adversely affected if the time-of-day clock changes");
1214 }
1215 
1216 #ifndef SYS_clock_getres
1217   #if defined(IA32) || defined(AMD64)
1218     #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1219     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1220   #else
1221     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1222     #define sys_clock_getres(x,y)  -1
1223   #endif
1224 #else
1225   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1226 #endif
1227 
1228 void os::Linux::fast_thread_clock_init() {
1229   if (!UseLinuxPosixThreadCPUClocks) {
1230     return;
1231   }
1232   clockid_t clockid;
1233   struct timespec tp;
1234   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1235       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1236 
1237   // Switch to using fast clocks for thread cpu time if
1238   // the sys_clock_getres() returns 0 error code.
1239   // Note, that some kernels may support the current thread
1240   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1241   // returned by the pthread_getcpuclockid().
1242   // If the fast Posix clocks are supported then the sys_clock_getres()
1243   // must return at least tp.tv_sec == 0 which means a resolution
1244   // better than 1 sec. This is extra check for reliability.
1245 
1246   if (pthread_getcpuclockid_func &&
1247       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1248       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1249     _supports_fast_thread_cpu_time = true;
1250     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1251   }
1252 }
1253 
1254 jlong os::javaTimeNanos() {
1255   if (os::supports_monotonic_clock()) {
1256     struct timespec tp;
1257     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1258     assert(status == 0, "gettime error");
1259     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1260     return result;
1261   } else {
1262     timeval time;
1263     int status = gettimeofday(&time, NULL);
1264     assert(status != -1, "linux error");
1265     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1266     return 1000 * usecs;
1267   }
1268 }
1269 
1270 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1271   if (os::supports_monotonic_clock()) {
1272     info_ptr->max_value = ALL_64_BITS;
1273 
1274     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1275     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1276     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1277   } else {
1278     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1279     info_ptr->max_value = ALL_64_BITS;
1280 
1281     // gettimeofday is a real time clock so it skips
1282     info_ptr->may_skip_backward = true;
1283     info_ptr->may_skip_forward = true;
1284   }
1285 
1286   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1287 }
1288 
1289 // Return the real, user, and system times in seconds from an
1290 // arbitrary fixed point in the past.
1291 bool os::getTimesSecs(double* process_real_time,
1292                       double* process_user_time,
1293                       double* process_system_time) {
1294   struct tms ticks;
1295   clock_t real_ticks = times(&ticks);
1296 
1297   if (real_ticks == (clock_t) (-1)) {
1298     return false;
1299   } else {
1300     double ticks_per_second = (double) clock_tics_per_sec;
1301     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1302     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1303     *process_real_time = ((double) real_ticks) / ticks_per_second;
1304 
1305     return true;
1306   }
1307 }
1308 
1309 
1310 char * os::local_time_string(char *buf, size_t buflen) {
1311   struct tm t;
1312   time_t long_time;
1313   time(&long_time);
1314   localtime_r(&long_time, &t);
1315   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1316                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1317                t.tm_hour, t.tm_min, t.tm_sec);
1318   return buf;
1319 }
1320 
1321 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1322   return localtime_r(clock, res);
1323 }
1324 
1325 ////////////////////////////////////////////////////////////////////////////////
1326 // runtime exit support
1327 
1328 // Note: os::shutdown() might be called very early during initialization, or
1329 // called from signal handler. Before adding something to os::shutdown(), make
1330 // sure it is async-safe and can handle partially initialized VM.
1331 void os::shutdown() {
1332 
1333   // allow PerfMemory to attempt cleanup of any persistent resources
1334   perfMemory_exit();
1335 
1336   // needs to remove object in file system
1337   AttachListener::abort();
1338 
1339   // flush buffered output, finish log files
1340   ostream_abort();
1341 
1342   // Check for abort hook
1343   abort_hook_t abort_hook = Arguments::abort_hook();
1344   if (abort_hook != NULL) {
1345     abort_hook();
1346   }
1347 
1348 }
1349 
1350 // Note: os::abort() might be called very early during initialization, or
1351 // called from signal handler. Before adding something to os::abort(), make
1352 // sure it is async-safe and can handle partially initialized VM.
1353 void os::abort(bool dump_core, void* siginfo, const void* context) {
1354   os::shutdown();
1355   if (dump_core) {
1356 #ifndef PRODUCT
1357     fdStream out(defaultStream::output_fd());
1358     out.print_raw("Current thread is ");
1359     char buf[16];
1360     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1361     out.print_raw_cr(buf);
1362     out.print_raw_cr("Dumping core ...");
1363 #endif
1364     ::abort(); // dump core
1365   }
1366 
1367   ::exit(1);
1368 }
1369 
1370 // Die immediately, no exit hook, no abort hook, no cleanup.
1371 void os::die() {
1372   ::abort();
1373 }
1374 
1375 
1376 // This method is a copy of JDK's sysGetLastErrorString
1377 // from src/solaris/hpi/src/system_md.c
1378 
1379 size_t os::lasterror(char *buf, size_t len) {
1380   if (errno == 0)  return 0;
1381 
1382   const char *s = ::strerror(errno);
1383   size_t n = ::strlen(s);
1384   if (n >= len) {
1385     n = len - 1;
1386   }
1387   ::strncpy(buf, s, n);
1388   buf[n] = '\0';
1389   return n;
1390 }
1391 
1392 // thread_id is kernel thread id (similar to Solaris LWP id)
1393 intx os::current_thread_id() { return os::Linux::gettid(); }
1394 int os::current_process_id() {
1395   return ::getpid();
1396 }
1397 
1398 // DLL functions
1399 
1400 const char* os::dll_file_extension() { return ".so"; }
1401 
1402 // This must be hard coded because it's the system's temporary
1403 // directory not the java application's temp directory, ala java.io.tmpdir.
1404 const char* os::get_temp_directory() { return "/tmp"; }
1405 
1406 static bool file_exists(const char* filename) {
1407   struct stat statbuf;
1408   if (filename == NULL || strlen(filename) == 0) {
1409     return false;
1410   }
1411   return os::stat(filename, &statbuf) == 0;
1412 }
1413 
1414 bool os::dll_build_name(char* buffer, size_t buflen,
1415                         const char* pname, const char* fname) {
1416   bool retval = false;
1417   // Copied from libhpi
1418   const size_t pnamelen = pname ? strlen(pname) : 0;
1419 
1420   // Return error on buffer overflow.
1421   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1422     return retval;
1423   }
1424 
1425   if (pnamelen == 0) {
1426     snprintf(buffer, buflen, "lib%s.so", fname);
1427     retval = true;
1428   } else if (strchr(pname, *os::path_separator()) != NULL) {
1429     int n;
1430     char** pelements = split_path(pname, &n);
1431     if (pelements == NULL) {
1432       return false;
1433     }
1434     for (int i = 0; i < n; i++) {
1435       // Really shouldn't be NULL, but check can't hurt
1436       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1437         continue; // skip the empty path values
1438       }
1439       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1440       if (file_exists(buffer)) {
1441         retval = true;
1442         break;
1443       }
1444     }
1445     // release the storage
1446     for (int i = 0; i < n; i++) {
1447       if (pelements[i] != NULL) {
1448         FREE_C_HEAP_ARRAY(char, pelements[i]);
1449       }
1450     }
1451     if (pelements != NULL) {
1452       FREE_C_HEAP_ARRAY(char*, pelements);
1453     }
1454   } else {
1455     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1456     retval = true;
1457   }
1458   return retval;
1459 }
1460 
1461 // check if addr is inside libjvm.so
1462 bool os::address_is_in_vm(address addr) {
1463   static address libjvm_base_addr;
1464   Dl_info dlinfo;
1465 
1466   if (libjvm_base_addr == NULL) {
1467     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1468       libjvm_base_addr = (address)dlinfo.dli_fbase;
1469     }
1470     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1471   }
1472 
1473   if (dladdr((void *)addr, &dlinfo) != 0) {
1474     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1475   }
1476 
1477   return false;
1478 }
1479 
1480 bool os::dll_address_to_function_name(address addr, char *buf,
1481                                       int buflen, int *offset,
1482                                       bool demangle) {
1483   // buf is not optional, but offset is optional
1484   assert(buf != NULL, "sanity check");
1485 
1486   Dl_info dlinfo;
1487 
1488   if (dladdr((void*)addr, &dlinfo) != 0) {
1489     // see if we have a matching symbol
1490     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1491       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1492         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1493       }
1494       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1495       return true;
1496     }
1497     // no matching symbol so try for just file info
1498     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1499       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1500                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1501         return true;
1502       }
1503     }
1504   }
1505 
1506   buf[0] = '\0';
1507   if (offset != NULL) *offset = -1;
1508   return false;
1509 }
1510 
1511 struct _address_to_library_name {
1512   address addr;          // input : memory address
1513   size_t  buflen;        //         size of fname
1514   char*   fname;         // output: library name
1515   address base;          //         library base addr
1516 };
1517 
1518 static int address_to_library_name_callback(struct dl_phdr_info *info,
1519                                             size_t size, void *data) {
1520   int i;
1521   bool found = false;
1522   address libbase = NULL;
1523   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1524 
1525   // iterate through all loadable segments
1526   for (i = 0; i < info->dlpi_phnum; i++) {
1527     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1528     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1529       // base address of a library is the lowest address of its loaded
1530       // segments.
1531       if (libbase == NULL || libbase > segbase) {
1532         libbase = segbase;
1533       }
1534       // see if 'addr' is within current segment
1535       if (segbase <= d->addr &&
1536           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1537         found = true;
1538       }
1539     }
1540   }
1541 
1542   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1543   // so dll_address_to_library_name() can fall through to use dladdr() which
1544   // can figure out executable name from argv[0].
1545   if (found && info->dlpi_name && info->dlpi_name[0]) {
1546     d->base = libbase;
1547     if (d->fname) {
1548       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1549     }
1550     return 1;
1551   }
1552   return 0;
1553 }
1554 
1555 bool os::dll_address_to_library_name(address addr, char* buf,
1556                                      int buflen, int* offset) {
1557   // buf is not optional, but offset is optional
1558   assert(buf != NULL, "sanity check");
1559 
1560   Dl_info dlinfo;
1561   struct _address_to_library_name data;
1562 
1563   // There is a bug in old glibc dladdr() implementation that it could resolve
1564   // to wrong library name if the .so file has a base address != NULL. Here
1565   // we iterate through the program headers of all loaded libraries to find
1566   // out which library 'addr' really belongs to. This workaround can be
1567   // removed once the minimum requirement for glibc is moved to 2.3.x.
1568   data.addr = addr;
1569   data.fname = buf;
1570   data.buflen = buflen;
1571   data.base = NULL;
1572   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1573 
1574   if (rslt) {
1575     // buf already contains library name
1576     if (offset) *offset = addr - data.base;
1577     return true;
1578   }
1579   if (dladdr((void*)addr, &dlinfo) != 0) {
1580     if (dlinfo.dli_fname != NULL) {
1581       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1582     }
1583     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1584       *offset = addr - (address)dlinfo.dli_fbase;
1585     }
1586     return true;
1587   }
1588 
1589   buf[0] = '\0';
1590   if (offset) *offset = -1;
1591   return false;
1592 }
1593 
1594 // Loads .dll/.so and
1595 // in case of error it checks if .dll/.so was built for the
1596 // same architecture as Hotspot is running on
1597 
1598 
1599 // Remember the stack's state. The Linux dynamic linker will change
1600 // the stack to 'executable' at most once, so we must safepoint only once.
1601 bool os::Linux::_stack_is_executable = false;
1602 
1603 // VM operation that loads a library.  This is necessary if stack protection
1604 // of the Java stacks can be lost during loading the library.  If we
1605 // do not stop the Java threads, they can stack overflow before the stacks
1606 // are protected again.
1607 class VM_LinuxDllLoad: public VM_Operation {
1608  private:
1609   const char *_filename;
1610   char *_ebuf;
1611   int _ebuflen;
1612   void *_lib;
1613  public:
1614   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1615     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1616   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1617   void doit() {
1618     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1619     os::Linux::_stack_is_executable = true;
1620   }
1621   void* loaded_library() { return _lib; }
1622 };
1623 
1624 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1625   void * result = NULL;
1626   bool load_attempted = false;
1627 
1628   // Check whether the library to load might change execution rights
1629   // of the stack. If they are changed, the protection of the stack
1630   // guard pages will be lost. We need a safepoint to fix this.
1631   //
1632   // See Linux man page execstack(8) for more info.
1633   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1634     ElfFile ef(filename);
1635     if (!ef.specifies_noexecstack()) {
1636       if (!is_init_completed()) {
1637         os::Linux::_stack_is_executable = true;
1638         // This is OK - No Java threads have been created yet, and hence no
1639         // stack guard pages to fix.
1640         //
1641         // This should happen only when you are building JDK7 using a very
1642         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1643         //
1644         // Dynamic loader will make all stacks executable after
1645         // this function returns, and will not do that again.
1646         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1647       } else {
1648         warning("You have loaded library %s which might have disabled stack guard. "
1649                 "The VM will try to fix the stack guard now.\n"
1650                 "It's highly recommended that you fix the library with "
1651                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1652                 filename);
1653 
1654         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1655         JavaThread *jt = JavaThread::current();
1656         if (jt->thread_state() != _thread_in_native) {
1657           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1658           // that requires ExecStack. Cannot enter safe point. Let's give up.
1659           warning("Unable to fix stack guard. Giving up.");
1660         } else {
1661           if (!LoadExecStackDllInVMThread) {
1662             // This is for the case where the DLL has an static
1663             // constructor function that executes JNI code. We cannot
1664             // load such DLLs in the VMThread.
1665             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1666           }
1667 
1668           ThreadInVMfromNative tiv(jt);
1669           debug_only(VMNativeEntryWrapper vew;)
1670 
1671           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1672           VMThread::execute(&op);
1673           if (LoadExecStackDllInVMThread) {
1674             result = op.loaded_library();
1675           }
1676           load_attempted = true;
1677         }
1678       }
1679     }
1680   }
1681 
1682   if (!load_attempted) {
1683     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1684   }
1685 
1686   if (result != NULL) {
1687     // Successful loading
1688     return result;
1689   }
1690 
1691   Elf32_Ehdr elf_head;
1692   int diag_msg_max_length=ebuflen-strlen(ebuf);
1693   char* diag_msg_buf=ebuf+strlen(ebuf);
1694 
1695   if (diag_msg_max_length==0) {
1696     // No more space in ebuf for additional diagnostics message
1697     return NULL;
1698   }
1699 
1700 
1701   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1702 
1703   if (file_descriptor < 0) {
1704     // Can't open library, report dlerror() message
1705     return NULL;
1706   }
1707 
1708   bool failed_to_read_elf_head=
1709     (sizeof(elf_head)!=
1710      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1711 
1712   ::close(file_descriptor);
1713   if (failed_to_read_elf_head) {
1714     // file i/o error - report dlerror() msg
1715     return NULL;
1716   }
1717 
1718   typedef struct {
1719     Elf32_Half  code;         // Actual value as defined in elf.h
1720     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1721     char        elf_class;    // 32 or 64 bit
1722     char        endianess;    // MSB or LSB
1723     char*       name;         // String representation
1724   } arch_t;
1725 
1726 #ifndef EM_486
1727   #define EM_486          6               /* Intel 80486 */
1728 #endif
1729 #ifndef EM_AARCH64
1730   #define EM_AARCH64    183               /* ARM AARCH64 */
1731 #endif
1732 
1733   static const arch_t arch_array[]={
1734     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1735     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1736     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1737     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1738     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1739     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1740     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1741     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1742 #if defined(VM_LITTLE_ENDIAN)
1743     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1744 #else
1745     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1746 #endif
1747     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1748     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1749     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1750     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1751     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1752     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1753     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1754     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1755   };
1756 
1757 #if  (defined IA32)
1758   static  Elf32_Half running_arch_code=EM_386;
1759 #elif   (defined AMD64)
1760   static  Elf32_Half running_arch_code=EM_X86_64;
1761 #elif  (defined IA64)
1762   static  Elf32_Half running_arch_code=EM_IA_64;
1763 #elif  (defined __sparc) && (defined _LP64)
1764   static  Elf32_Half running_arch_code=EM_SPARCV9;
1765 #elif  (defined __sparc) && (!defined _LP64)
1766   static  Elf32_Half running_arch_code=EM_SPARC;
1767 #elif  (defined __powerpc64__)
1768   static  Elf32_Half running_arch_code=EM_PPC64;
1769 #elif  (defined __powerpc__)
1770   static  Elf32_Half running_arch_code=EM_PPC;
1771 #elif  (defined ARM)
1772   static  Elf32_Half running_arch_code=EM_ARM;
1773 #elif  (defined S390)
1774   static  Elf32_Half running_arch_code=EM_S390;
1775 #elif  (defined ALPHA)
1776   static  Elf32_Half running_arch_code=EM_ALPHA;
1777 #elif  (defined MIPSEL)
1778   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1779 #elif  (defined PARISC)
1780   static  Elf32_Half running_arch_code=EM_PARISC;
1781 #elif  (defined MIPS)
1782   static  Elf32_Half running_arch_code=EM_MIPS;
1783 #elif  (defined M68K)
1784   static  Elf32_Half running_arch_code=EM_68K;
1785 #elif  (defined AARCH64)
1786   static  Elf32_Half running_arch_code=EM_AARCH64;
1787 #else
1788     #error Method os::dll_load requires that one of following is defined:\
1789          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1790 #endif
1791 
1792   // Identify compatability class for VM's architecture and library's architecture
1793   // Obtain string descriptions for architectures
1794 
1795   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1796   int running_arch_index=-1;
1797 
1798   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1799     if (running_arch_code == arch_array[i].code) {
1800       running_arch_index    = i;
1801     }
1802     if (lib_arch.code == arch_array[i].code) {
1803       lib_arch.compat_class = arch_array[i].compat_class;
1804       lib_arch.name         = arch_array[i].name;
1805     }
1806   }
1807 
1808   assert(running_arch_index != -1,
1809          "Didn't find running architecture code (running_arch_code) in arch_array");
1810   if (running_arch_index == -1) {
1811     // Even though running architecture detection failed
1812     // we may still continue with reporting dlerror() message
1813     return NULL;
1814   }
1815 
1816   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1817     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1818     return NULL;
1819   }
1820 
1821 #ifndef S390
1822   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1823     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1824     return NULL;
1825   }
1826 #endif // !S390
1827 
1828   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1829     if (lib_arch.name!=NULL) {
1830       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1831                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1832                  lib_arch.name, arch_array[running_arch_index].name);
1833     } else {
1834       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1835                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1836                  lib_arch.code,
1837                  arch_array[running_arch_index].name);
1838     }
1839   }
1840 
1841   return NULL;
1842 }
1843 
1844 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1845                                 int ebuflen) {
1846   void * result = ::dlopen(filename, RTLD_LAZY);
1847   if (result == NULL) {
1848     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1849     ebuf[ebuflen-1] = '\0';
1850   }
1851   return result;
1852 }
1853 
1854 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1855                                        int ebuflen) {
1856   void * result = NULL;
1857   if (LoadExecStackDllInVMThread) {
1858     result = dlopen_helper(filename, ebuf, ebuflen);
1859   }
1860 
1861   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1862   // library that requires an executable stack, or which does not have this
1863   // stack attribute set, dlopen changes the stack attribute to executable. The
1864   // read protection of the guard pages gets lost.
1865   //
1866   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1867   // may have been queued at the same time.
1868 
1869   if (!_stack_is_executable) {
1870     JavaThread *jt = Threads::first();
1871 
1872     while (jt) {
1873       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1874           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1875         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1876           warning("Attempt to reguard stack yellow zone failed.");
1877         }
1878       }
1879       jt = jt->next();
1880     }
1881   }
1882 
1883   return result;
1884 }
1885 
1886 void* os::dll_lookup(void* handle, const char* name) {
1887   void* res = dlsym(handle, name);
1888   return res;
1889 }
1890 
1891 void* os::get_default_process_handle() {
1892   return (void*)::dlopen(NULL, RTLD_LAZY);
1893 }
1894 
1895 static bool _print_ascii_file(const char* filename, outputStream* st) {
1896   int fd = ::open(filename, O_RDONLY);
1897   if (fd == -1) {
1898     return false;
1899   }
1900 
1901   char buf[32];
1902   int bytes;
1903   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1904     st->print_raw(buf, bytes);
1905   }
1906 
1907   ::close(fd);
1908 
1909   return true;
1910 }
1911 
1912 void os::print_dll_info(outputStream *st) {
1913   st->print_cr("Dynamic libraries:");
1914 
1915   char fname[32];
1916   pid_t pid = os::Linux::gettid();
1917 
1918   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1919 
1920   if (!_print_ascii_file(fname, st)) {
1921     st->print("Can not get library information for pid = %d\n", pid);
1922   }
1923 }
1924 
1925 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1926   FILE *procmapsFile = NULL;
1927 
1928   // Open the procfs maps file for the current process
1929   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1930     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1931     char line[PATH_MAX + 100];
1932 
1933     // Read line by line from 'file'
1934     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1935       u8 base, top, offset, inode;
1936       char permissions[5];
1937       char device[6];
1938       char name[PATH_MAX + 1];
1939 
1940       // Parse fields from line
1941       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1942              &base, &top, permissions, &offset, device, &inode, name);
1943 
1944       // Filter by device id '00:00' so that we only get file system mapped files.
1945       if (strcmp(device, "00:00") != 0) {
1946 
1947         // Call callback with the fields of interest
1948         if(callback(name, (address)base, (address)top, param)) {
1949           // Oops abort, callback aborted
1950           fclose(procmapsFile);
1951           return 1;
1952         }
1953       }
1954     }
1955     fclose(procmapsFile);
1956   }
1957   return 0;
1958 }
1959 
1960 void os::print_os_info_brief(outputStream* st) {
1961   os::Linux::print_distro_info(st);
1962 
1963   os::Posix::print_uname_info(st);
1964 
1965   os::Linux::print_libversion_info(st);
1966 
1967 }
1968 
1969 void os::print_os_info(outputStream* st) {
1970   st->print("OS:");
1971 
1972   os::Linux::print_distro_info(st);
1973 
1974   os::Posix::print_uname_info(st);
1975 
1976   // Print warning if unsafe chroot environment detected
1977   if (unsafe_chroot_detected) {
1978     st->print("WARNING!! ");
1979     st->print_cr("%s", unstable_chroot_error);
1980   }
1981 
1982   os::Linux::print_libversion_info(st);
1983 
1984   os::Posix::print_rlimit_info(st);
1985 
1986   os::Posix::print_load_average(st);
1987 
1988   os::Linux::print_full_memory_info(st);
1989 }
1990 
1991 // Try to identify popular distros.
1992 // Most Linux distributions have a /etc/XXX-release file, which contains
1993 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1994 // file that also contains the OS version string. Some have more than one
1995 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1996 // /etc/redhat-release.), so the order is important.
1997 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1998 // their own specific XXX-release file as well as a redhat-release file.
1999 // Because of this the XXX-release file needs to be searched for before the
2000 // redhat-release file.
2001 // Since Red Hat has a lsb-release file that is not very descriptive the
2002 // search for redhat-release needs to be before lsb-release.
2003 // Since the lsb-release file is the new standard it needs to be searched
2004 // before the older style release files.
2005 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2006 // next to last resort.  The os-release file is a new standard that contains
2007 // distribution information and the system-release file seems to be an old
2008 // standard that has been replaced by the lsb-release and os-release files.
2009 // Searching for the debian_version file is the last resort.  It contains
2010 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2011 // "Debian " is printed before the contents of the debian_version file.
2012 
2013 const char* distro_files[] = {
2014   "/etc/oracle-release",
2015   "/etc/mandriva-release",
2016   "/etc/mandrake-release",
2017   "/etc/sun-release",
2018   "/etc/redhat-release",
2019   "/etc/lsb-release",
2020   "/etc/SuSE-release",
2021   "/etc/turbolinux-release",
2022   "/etc/gentoo-release",
2023   "/etc/ltib-release",
2024   "/etc/angstrom-version",
2025   "/etc/system-release",
2026   "/etc/os-release",
2027   NULL };
2028 
2029 void os::Linux::print_distro_info(outputStream* st) {
2030   for (int i = 0;; i++) {
2031     const char* file = distro_files[i];
2032     if (file == NULL) {
2033       break;  // done
2034     }
2035     // If file prints, we found it.
2036     if (_print_ascii_file(file, st)) {
2037       return;
2038     }
2039   }
2040 
2041   if (file_exists("/etc/debian_version")) {
2042     st->print("Debian ");
2043     _print_ascii_file("/etc/debian_version", st);
2044   } else {
2045     st->print("Linux");
2046   }
2047   st->cr();
2048 }
2049 
2050 static void parse_os_info(char* distro, size_t length, const char* file) {
2051   FILE* fp = fopen(file, "r");
2052   if (fp != NULL) {
2053     char buf[256];
2054     // get last line of the file.
2055     while (fgets(buf, sizeof(buf), fp)) { }
2056     // Edit out extra stuff in expected ubuntu format
2057     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2058       char* ptr = strstr(buf, "\"");  // the name is in quotes
2059       if (ptr != NULL) {
2060         ptr++; // go beyond first quote
2061         char* nl = strchr(ptr, '\"');
2062         if (nl != NULL) *nl = '\0';
2063         strncpy(distro, ptr, length);
2064       } else {
2065         ptr = strstr(buf, "=");
2066         ptr++; // go beyond equals then
2067         char* nl = strchr(ptr, '\n');
2068         if (nl != NULL) *nl = '\0';
2069         strncpy(distro, ptr, length);
2070       }
2071     } else {
2072       // if not in expected Ubuntu format, print out whole line minus \n
2073       char* nl = strchr(buf, '\n');
2074       if (nl != NULL) *nl = '\0';
2075       strncpy(distro, buf, length);
2076     }
2077     // close distro file
2078     fclose(fp);
2079   }
2080 }
2081 
2082 void os::get_summary_os_info(char* buf, size_t buflen) {
2083   for (int i = 0;; i++) {
2084     const char* file = distro_files[i];
2085     if (file == NULL) {
2086       break; // ran out of distro_files
2087     }
2088     if (file_exists(file)) {
2089       parse_os_info(buf, buflen, file);
2090       return;
2091     }
2092   }
2093   // special case for debian
2094   if (file_exists("/etc/debian_version")) {
2095     strncpy(buf, "Debian ", buflen);
2096     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2097   } else {
2098     strncpy(buf, "Linux", buflen);
2099   }
2100 }
2101 
2102 void os::Linux::print_libversion_info(outputStream* st) {
2103   // libc, pthread
2104   st->print("libc:");
2105   st->print("%s ", os::Linux::glibc_version());
2106   st->print("%s ", os::Linux::libpthread_version());
2107   st->cr();
2108 }
2109 
2110 void os::Linux::print_full_memory_info(outputStream* st) {
2111   st->print("\n/proc/meminfo:\n");
2112   _print_ascii_file("/proc/meminfo", st);
2113   st->cr();
2114 }
2115 
2116 void os::print_memory_info(outputStream* st) {
2117 
2118   st->print("Memory:");
2119   st->print(" %dk page", os::vm_page_size()>>10);
2120 
2121   // values in struct sysinfo are "unsigned long"
2122   struct sysinfo si;
2123   sysinfo(&si);
2124 
2125   st->print(", physical " UINT64_FORMAT "k",
2126             os::physical_memory() >> 10);
2127   st->print("(" UINT64_FORMAT "k free)",
2128             os::available_memory() >> 10);
2129   st->print(", swap " UINT64_FORMAT "k",
2130             ((jlong)si.totalswap * si.mem_unit) >> 10);
2131   st->print("(" UINT64_FORMAT "k free)",
2132             ((jlong)si.freeswap * si.mem_unit) >> 10);
2133   st->cr();
2134 }
2135 
2136 // Print the first "model name" line and the first "flags" line
2137 // that we find and nothing more. We assume "model name" comes
2138 // before "flags" so if we find a second "model name", then the
2139 // "flags" field is considered missing.
2140 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2141 #if defined(IA32) || defined(AMD64)
2142   // Other platforms have less repetitive cpuinfo files
2143   FILE *fp = fopen("/proc/cpuinfo", "r");
2144   if (fp) {
2145     while (!feof(fp)) {
2146       if (fgets(buf, buflen, fp)) {
2147         // Assume model name comes before flags
2148         bool model_name_printed = false;
2149         if (strstr(buf, "model name") != NULL) {
2150           if (!model_name_printed) {
2151             st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2152             st->print_raw(buf);
2153             model_name_printed = true;
2154           } else {
2155             // model name printed but not flags?  Odd, just return
2156             fclose(fp);
2157             return true;
2158           }
2159         }
2160         // print the flags line too
2161         if (strstr(buf, "flags") != NULL) {
2162           st->print_raw(buf);
2163           fclose(fp);
2164           return true;
2165         }
2166       }
2167     }
2168     fclose(fp);
2169   }
2170 #endif // x86 platforms
2171   return false;
2172 }
2173 
2174 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2175   // Only print the model name if the platform provides this as a summary
2176   if (!print_model_name_and_flags(st, buf, buflen)) {
2177     st->print("\n/proc/cpuinfo:\n");
2178     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2179       st->print_cr("  <Not Available>");
2180     }
2181   }
2182 }
2183 
2184 #if defined(AMD64) || defined(IA32) || defined(X32)
2185 const char* search_string = "model name";
2186 #elif defined(SPARC)
2187 const char* search_string = "cpu";
2188 #elif defined(PPC64)
2189 const char* search_string = "cpu";
2190 #else
2191 const char* search_string = "Processor";
2192 #endif
2193 
2194 // Parses the cpuinfo file for string representing the model name.
2195 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2196   FILE* fp = fopen("/proc/cpuinfo", "r");
2197   if (fp != NULL) {
2198     while (!feof(fp)) {
2199       char buf[256];
2200       if (fgets(buf, sizeof(buf), fp)) {
2201         char* start = strstr(buf, search_string);
2202         if (start != NULL) {
2203           char *ptr = start + strlen(search_string);
2204           char *end = buf + strlen(buf);
2205           while (ptr != end) {
2206              // skip whitespace and colon for the rest of the name.
2207              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2208                break;
2209              }
2210              ptr++;
2211           }
2212           if (ptr != end) {
2213             // reasonable string, get rid of newline and keep the rest
2214             char* nl = strchr(buf, '\n');
2215             if (nl != NULL) *nl = '\0';
2216             strncpy(cpuinfo, ptr, length);
2217             fclose(fp);
2218             return;
2219           }
2220         }
2221       }
2222     }
2223     fclose(fp);
2224   }
2225   // cpuinfo not found or parsing failed, just print generic string.  The entire
2226   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2227 #if defined(AMD64)
2228   strncpy(cpuinfo, "x86_64", length);
2229 #elif defined(IA32)
2230   strncpy(cpuinfo, "x86_32", length);
2231 #elif defined(IA64)
2232   strncpy(cpuinfo, "IA64", length);
2233 #elif defined(SPARC)
2234   strncpy(cpuinfo, "sparcv9", length);
2235 #elif defined(AARCH64)
2236   strncpy(cpuinfo, "AArch64", length);
2237 #elif defined(ARM)
2238   strncpy(cpuinfo, "ARM", length);
2239 #elif defined(PPC)
2240   strncpy(cpuinfo, "PPC64", length);
2241 #elif defined(ZERO_LIBARCH)
2242   strncpy(cpuinfo, ZERO_LIBARCH, length);
2243 #else
2244   strncpy(cpuinfo, "unknown", length);
2245 #endif
2246 }
2247 
2248 static void print_signal_handler(outputStream* st, int sig,
2249                                  char* buf, size_t buflen);
2250 
2251 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2252   st->print_cr("Signal Handlers:");
2253   print_signal_handler(st, SIGSEGV, buf, buflen);
2254   print_signal_handler(st, SIGBUS , buf, buflen);
2255   print_signal_handler(st, SIGFPE , buf, buflen);
2256   print_signal_handler(st, SIGPIPE, buf, buflen);
2257   print_signal_handler(st, SIGXFSZ, buf, buflen);
2258   print_signal_handler(st, SIGILL , buf, buflen);
2259   print_signal_handler(st, SR_signum, buf, buflen);
2260   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2261   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2262   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2263   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2264 #if defined(PPC64)
2265   print_signal_handler(st, SIGTRAP, buf, buflen);
2266 #endif
2267 }
2268 
2269 static char saved_jvm_path[MAXPATHLEN] = {0};
2270 
2271 // Find the full path to the current module, libjvm.so
2272 void os::jvm_path(char *buf, jint buflen) {
2273   // Error checking.
2274   if (buflen < MAXPATHLEN) {
2275     assert(false, "must use a large-enough buffer");
2276     buf[0] = '\0';
2277     return;
2278   }
2279   // Lazy resolve the path to current module.
2280   if (saved_jvm_path[0] != 0) {
2281     strcpy(buf, saved_jvm_path);
2282     return;
2283   }
2284 
2285   char dli_fname[MAXPATHLEN];
2286   bool ret = dll_address_to_library_name(
2287                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2288                                          dli_fname, sizeof(dli_fname), NULL);
2289   assert(ret, "cannot locate libjvm");
2290   char *rp = NULL;
2291   if (ret && dli_fname[0] != '\0') {
2292     rp = realpath(dli_fname, buf);
2293   }
2294   if (rp == NULL) {
2295     return;
2296   }
2297 
2298   if (Arguments::sun_java_launcher_is_altjvm()) {
2299     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2300     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2301     // If "/jre/lib/" appears at the right place in the string, then
2302     // assume we are installed in a JDK and we're done. Otherwise, check
2303     // for a JAVA_HOME environment variable and fix up the path so it
2304     // looks like libjvm.so is installed there (append a fake suffix
2305     // hotspot/libjvm.so).
2306     const char *p = buf + strlen(buf) - 1;
2307     for (int count = 0; p > buf && count < 5; ++count) {
2308       for (--p; p > buf && *p != '/'; --p)
2309         /* empty */ ;
2310     }
2311 
2312     if (strncmp(p, "/jre/lib/", 9) != 0) {
2313       // Look for JAVA_HOME in the environment.
2314       char* java_home_var = ::getenv("JAVA_HOME");
2315       if (java_home_var != NULL && java_home_var[0] != 0) {
2316         char* jrelib_p;
2317         int len;
2318 
2319         // Check the current module name "libjvm.so".
2320         p = strrchr(buf, '/');
2321         if (p == NULL) {
2322           return;
2323         }
2324         assert(strstr(p, "/libjvm") == p, "invalid library name");
2325 
2326         rp = realpath(java_home_var, buf);
2327         if (rp == NULL) {
2328           return;
2329         }
2330 
2331         // determine if this is a legacy image or modules image
2332         // modules image doesn't have "jre" subdirectory
2333         len = strlen(buf);
2334         assert(len < buflen, "Ran out of buffer room");
2335         jrelib_p = buf + len;
2336         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2337         if (0 != access(buf, F_OK)) {
2338           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2339         }
2340 
2341         if (0 == access(buf, F_OK)) {
2342           // Use current module name "libjvm.so"
2343           len = strlen(buf);
2344           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2345         } else {
2346           // Go back to path of .so
2347           rp = realpath(dli_fname, buf);
2348           if (rp == NULL) {
2349             return;
2350           }
2351         }
2352       }
2353     }
2354   }
2355 
2356   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2357   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2358 }
2359 
2360 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2361   // no prefix required, not even "_"
2362 }
2363 
2364 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2365   // no suffix required
2366 }
2367 
2368 ////////////////////////////////////////////////////////////////////////////////
2369 // sun.misc.Signal support
2370 
2371 static volatile jint sigint_count = 0;
2372 
2373 static void UserHandler(int sig, void *siginfo, void *context) {
2374   // 4511530 - sem_post is serialized and handled by the manager thread. When
2375   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2376   // don't want to flood the manager thread with sem_post requests.
2377   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2378     return;
2379   }
2380 
2381   // Ctrl-C is pressed during error reporting, likely because the error
2382   // handler fails to abort. Let VM die immediately.
2383   if (sig == SIGINT && is_error_reported()) {
2384     os::die();
2385   }
2386 
2387   os::signal_notify(sig);
2388 }
2389 
2390 void* os::user_handler() {
2391   return CAST_FROM_FN_PTR(void*, UserHandler);
2392 }
2393 
2394 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2395   struct timespec ts;
2396   // Semaphore's are always associated with CLOCK_REALTIME
2397   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2398   // see unpackTime for discussion on overflow checking
2399   if (sec >= MAX_SECS) {
2400     ts.tv_sec += MAX_SECS;
2401     ts.tv_nsec = 0;
2402   } else {
2403     ts.tv_sec += sec;
2404     ts.tv_nsec += nsec;
2405     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2406       ts.tv_nsec -= NANOSECS_PER_SEC;
2407       ++ts.tv_sec; // note: this must be <= max_secs
2408     }
2409   }
2410 
2411   return ts;
2412 }
2413 
2414 extern "C" {
2415   typedef void (*sa_handler_t)(int);
2416   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2417 }
2418 
2419 void* os::signal(int signal_number, void* handler) {
2420   struct sigaction sigAct, oldSigAct;
2421 
2422   sigfillset(&(sigAct.sa_mask));
2423   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2424   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2425 
2426   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2427     // -1 means registration failed
2428     return (void *)-1;
2429   }
2430 
2431   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2432 }
2433 
2434 void os::signal_raise(int signal_number) {
2435   ::raise(signal_number);
2436 }
2437 
2438 // The following code is moved from os.cpp for making this
2439 // code platform specific, which it is by its very nature.
2440 
2441 // Will be modified when max signal is changed to be dynamic
2442 int os::sigexitnum_pd() {
2443   return NSIG;
2444 }
2445 
2446 // a counter for each possible signal value
2447 static volatile jint pending_signals[NSIG+1] = { 0 };
2448 
2449 // Linux(POSIX) specific hand shaking semaphore.
2450 static sem_t sig_sem;
2451 static PosixSemaphore sr_semaphore;
2452 
2453 void os::signal_init_pd() {
2454   // Initialize signal structures
2455   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2456 
2457   // Initialize signal semaphore
2458   ::sem_init(&sig_sem, 0, 0);
2459 }
2460 
2461 void os::signal_notify(int sig) {
2462   Atomic::inc(&pending_signals[sig]);
2463   ::sem_post(&sig_sem);
2464 }
2465 
2466 static int check_pending_signals(bool wait) {
2467   Atomic::store(0, &sigint_count);
2468   for (;;) {
2469     for (int i = 0; i < NSIG + 1; i++) {
2470       jint n = pending_signals[i];
2471       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2472         return i;
2473       }
2474     }
2475     if (!wait) {
2476       return -1;
2477     }
2478     JavaThread *thread = JavaThread::current();
2479     ThreadBlockInVM tbivm(thread);
2480 
2481     bool threadIsSuspended;
2482     do {
2483       thread->set_suspend_equivalent();
2484       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2485       ::sem_wait(&sig_sem);
2486 
2487       // were we externally suspended while we were waiting?
2488       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2489       if (threadIsSuspended) {
2490         // The semaphore has been incremented, but while we were waiting
2491         // another thread suspended us. We don't want to continue running
2492         // while suspended because that would surprise the thread that
2493         // suspended us.
2494         ::sem_post(&sig_sem);
2495 
2496         thread->java_suspend_self();
2497       }
2498     } while (threadIsSuspended);
2499   }
2500 }
2501 
2502 int os::signal_lookup() {
2503   return check_pending_signals(false);
2504 }
2505 
2506 int os::signal_wait() {
2507   return check_pending_signals(true);
2508 }
2509 
2510 ////////////////////////////////////////////////////////////////////////////////
2511 // Virtual Memory
2512 
2513 int os::vm_page_size() {
2514   // Seems redundant as all get out
2515   assert(os::Linux::page_size() != -1, "must call os::init");
2516   return os::Linux::page_size();
2517 }
2518 
2519 // Solaris allocates memory by pages.
2520 int os::vm_allocation_granularity() {
2521   assert(os::Linux::page_size() != -1, "must call os::init");
2522   return os::Linux::page_size();
2523 }
2524 
2525 // Rationale behind this function:
2526 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2527 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2528 //  samples for JITted code. Here we create private executable mapping over the code cache
2529 //  and then we can use standard (well, almost, as mapping can change) way to provide
2530 //  info for the reporting script by storing timestamp and location of symbol
2531 void linux_wrap_code(char* base, size_t size) {
2532   static volatile jint cnt = 0;
2533 
2534   if (!UseOprofile) {
2535     return;
2536   }
2537 
2538   char buf[PATH_MAX+1];
2539   int num = Atomic::add(1, &cnt);
2540 
2541   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2542            os::get_temp_directory(), os::current_process_id(), num);
2543   unlink(buf);
2544 
2545   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2546 
2547   if (fd != -1) {
2548     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2549     if (rv != (off_t)-1) {
2550       if (::write(fd, "", 1) == 1) {
2551         mmap(base, size,
2552              PROT_READ|PROT_WRITE|PROT_EXEC,
2553              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2554       }
2555     }
2556     ::close(fd);
2557     unlink(buf);
2558   }
2559 }
2560 
2561 static bool recoverable_mmap_error(int err) {
2562   // See if the error is one we can let the caller handle. This
2563   // list of errno values comes from JBS-6843484. I can't find a
2564   // Linux man page that documents this specific set of errno
2565   // values so while this list currently matches Solaris, it may
2566   // change as we gain experience with this failure mode.
2567   switch (err) {
2568   case EBADF:
2569   case EINVAL:
2570   case ENOTSUP:
2571     // let the caller deal with these errors
2572     return true;
2573 
2574   default:
2575     // Any remaining errors on this OS can cause our reserved mapping
2576     // to be lost. That can cause confusion where different data
2577     // structures think they have the same memory mapped. The worst
2578     // scenario is if both the VM and a library think they have the
2579     // same memory mapped.
2580     return false;
2581   }
2582 }
2583 
2584 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2585                                     int err) {
2586   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2587           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2588           strerror(err), err);
2589 }
2590 
2591 static void warn_fail_commit_memory(char* addr, size_t size,
2592                                     size_t alignment_hint, bool exec,
2593                                     int err) {
2594   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2595           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2596           alignment_hint, exec, strerror(err), err);
2597 }
2598 
2599 // NOTE: Linux kernel does not really reserve the pages for us.
2600 //       All it does is to check if there are enough free pages
2601 //       left at the time of mmap(). This could be a potential
2602 //       problem.
2603 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2604   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2605   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2606                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2607   if (res != (uintptr_t) MAP_FAILED) {
2608     if (UseNUMAInterleaving) {
2609       numa_make_global(addr, size);
2610     }
2611     return 0;
2612   }
2613 
2614   int err = errno;  // save errno from mmap() call above
2615 
2616   if (!recoverable_mmap_error(err)) {
2617     warn_fail_commit_memory(addr, size, exec, err);
2618     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2619   }
2620 
2621   return err;
2622 }
2623 
2624 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2625   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2626 }
2627 
2628 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2629                                   const char* mesg) {
2630   assert(mesg != NULL, "mesg must be specified");
2631   int err = os::Linux::commit_memory_impl(addr, size, exec);
2632   if (err != 0) {
2633     // the caller wants all commit errors to exit with the specified mesg:
2634     warn_fail_commit_memory(addr, size, exec, err);
2635     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2636   }
2637 }
2638 
2639 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2640 #ifndef MAP_HUGETLB
2641   #define MAP_HUGETLB 0x40000
2642 #endif
2643 
2644 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2645 #ifndef MADV_HUGEPAGE
2646   #define MADV_HUGEPAGE 14
2647 #endif
2648 
2649 int os::Linux::commit_memory_impl(char* addr, size_t size,
2650                                   size_t alignment_hint, bool exec) {
2651   int err = os::Linux::commit_memory_impl(addr, size, exec);
2652   if (err == 0) {
2653     realign_memory(addr, size, alignment_hint);
2654   }
2655   return err;
2656 }
2657 
2658 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2659                           bool exec) {
2660   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2661 }
2662 
2663 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2664                                   size_t alignment_hint, bool exec,
2665                                   const char* mesg) {
2666   assert(mesg != NULL, "mesg must be specified");
2667   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2668   if (err != 0) {
2669     // the caller wants all commit errors to exit with the specified mesg:
2670     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2671     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2672   }
2673 }
2674 
2675 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2676   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2677     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2678     // be supported or the memory may already be backed by huge pages.
2679     ::madvise(addr, bytes, MADV_HUGEPAGE);
2680   }
2681 }
2682 
2683 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2684   // This method works by doing an mmap over an existing mmaping and effectively discarding
2685   // the existing pages. However it won't work for SHM-based large pages that cannot be
2686   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2687   // small pages on top of the SHM segment. This method always works for small pages, so we
2688   // allow that in any case.
2689   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2690     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2691   }
2692 }
2693 
2694 void os::numa_make_global(char *addr, size_t bytes) {
2695   Linux::numa_interleave_memory(addr, bytes);
2696 }
2697 
2698 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2699 // bind policy to MPOL_PREFERRED for the current thread.
2700 #define USE_MPOL_PREFERRED 0
2701 
2702 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2703   // To make NUMA and large pages more robust when both enabled, we need to ease
2704   // the requirements on where the memory should be allocated. MPOL_BIND is the
2705   // default policy and it will force memory to be allocated on the specified
2706   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2707   // the specified node, but will not force it. Using this policy will prevent
2708   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2709   // free large pages.
2710   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2711   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2712 }
2713 
2714 bool os::numa_topology_changed() { return false; }
2715 
2716 size_t os::numa_get_groups_num() {
2717   int max_node = Linux::numa_max_node();
2718   return max_node > 0 ? max_node + 1 : 1;
2719 }
2720 
2721 int os::numa_get_group_id() {
2722   int cpu_id = Linux::sched_getcpu();
2723   if (cpu_id != -1) {
2724     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2725     if (lgrp_id != -1) {
2726       return lgrp_id;
2727     }
2728   }
2729   return 0;
2730 }
2731 
2732 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2733   for (size_t i = 0; i < size; i++) {
2734     ids[i] = i;
2735   }
2736   return size;
2737 }
2738 
2739 bool os::get_page_info(char *start, page_info* info) {
2740   return false;
2741 }
2742 
2743 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2744                      page_info* page_found) {
2745   return end;
2746 }
2747 
2748 
2749 int os::Linux::sched_getcpu_syscall(void) {
2750   unsigned int cpu = 0;
2751   int retval = -1;
2752 
2753 #if defined(IA32)
2754   #ifndef SYS_getcpu
2755     #define SYS_getcpu 318
2756   #endif
2757   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2758 #elif defined(AMD64)
2759 // Unfortunately we have to bring all these macros here from vsyscall.h
2760 // to be able to compile on old linuxes.
2761   #define __NR_vgetcpu 2
2762   #define VSYSCALL_START (-10UL << 20)
2763   #define VSYSCALL_SIZE 1024
2764   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2765   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2766   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2767   retval = vgetcpu(&cpu, NULL, NULL);
2768 #endif
2769 
2770   return (retval == -1) ? retval : cpu;
2771 }
2772 
2773 // Something to do with the numa-aware allocator needs these symbols
2774 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2775 extern "C" JNIEXPORT void numa_error(char *where) { }
2776 
2777 
2778 // If we are running with libnuma version > 2, then we should
2779 // be trying to use symbols with versions 1.1
2780 // If we are running with earlier version, which did not have symbol versions,
2781 // we should use the base version.
2782 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2783   void *f = dlvsym(handle, name, "libnuma_1.1");
2784   if (f == NULL) {
2785     f = dlsym(handle, name);
2786   }
2787   return f;
2788 }
2789 
2790 bool os::Linux::libnuma_init() {
2791   // sched_getcpu() should be in libc.
2792   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2793                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2794 
2795   // If it's not, try a direct syscall.
2796   if (sched_getcpu() == -1) {
2797     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2798                                     (void*)&sched_getcpu_syscall));
2799   }
2800 
2801   if (sched_getcpu() != -1) { // Does it work?
2802     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2803     if (handle != NULL) {
2804       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2805                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2806       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2807                                        libnuma_dlsym(handle, "numa_max_node")));
2808       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2809                                         libnuma_dlsym(handle, "numa_available")));
2810       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2811                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2812       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2813                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2814       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2815                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2816 
2817 
2818       if (numa_available() != -1) {
2819         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2820         // Create a cpu -> node mapping
2821         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2822         rebuild_cpu_to_node_map();
2823         return true;
2824       }
2825     }
2826   }
2827   return false;
2828 }
2829 
2830 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2831 // The table is later used in get_node_by_cpu().
2832 void os::Linux::rebuild_cpu_to_node_map() {
2833   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2834                               // in libnuma (possible values are starting from 16,
2835                               // and continuing up with every other power of 2, but less
2836                               // than the maximum number of CPUs supported by kernel), and
2837                               // is a subject to change (in libnuma version 2 the requirements
2838                               // are more reasonable) we'll just hardcode the number they use
2839                               // in the library.
2840   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2841 
2842   size_t cpu_num = os::active_processor_count();
2843   size_t cpu_map_size = NCPUS / BitsPerCLong;
2844   size_t cpu_map_valid_size =
2845     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2846 
2847   cpu_to_node()->clear();
2848   cpu_to_node()->at_grow(cpu_num - 1);
2849   size_t node_num = numa_get_groups_num();
2850 
2851   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2852   for (size_t i = 0; i < node_num; i++) {
2853     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2854       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2855         if (cpu_map[j] != 0) {
2856           for (size_t k = 0; k < BitsPerCLong; k++) {
2857             if (cpu_map[j] & (1UL << k)) {
2858               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2859             }
2860           }
2861         }
2862       }
2863     }
2864   }
2865   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2866 }
2867 
2868 int os::Linux::get_node_by_cpu(int cpu_id) {
2869   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2870     return cpu_to_node()->at(cpu_id);
2871   }
2872   return -1;
2873 }
2874 
2875 GrowableArray<int>* os::Linux::_cpu_to_node;
2876 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2877 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2878 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2879 os::Linux::numa_available_func_t os::Linux::_numa_available;
2880 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2881 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2882 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2883 unsigned long* os::Linux::_numa_all_nodes;
2884 
2885 bool os::pd_uncommit_memory(char* addr, size_t size) {
2886   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2887                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2888   return res  != (uintptr_t) MAP_FAILED;
2889 }
2890 
2891 static address get_stack_commited_bottom(address bottom, size_t size) {
2892   address nbot = bottom;
2893   address ntop = bottom + size;
2894 
2895   size_t page_sz = os::vm_page_size();
2896   unsigned pages = size / page_sz;
2897 
2898   unsigned char vec[1];
2899   unsigned imin = 1, imax = pages + 1, imid;
2900   int mincore_return_value = 0;
2901 
2902   assert(imin <= imax, "Unexpected page size");
2903 
2904   while (imin < imax) {
2905     imid = (imax + imin) / 2;
2906     nbot = ntop - (imid * page_sz);
2907 
2908     // Use a trick with mincore to check whether the page is mapped or not.
2909     // mincore sets vec to 1 if page resides in memory and to 0 if page
2910     // is swapped output but if page we are asking for is unmapped
2911     // it returns -1,ENOMEM
2912     mincore_return_value = mincore(nbot, page_sz, vec);
2913 
2914     if (mincore_return_value == -1) {
2915       // Page is not mapped go up
2916       // to find first mapped page
2917       if (errno != EAGAIN) {
2918         assert(errno == ENOMEM, "Unexpected mincore errno");
2919         imax = imid;
2920       }
2921     } else {
2922       // Page is mapped go down
2923       // to find first not mapped page
2924       imin = imid + 1;
2925     }
2926   }
2927 
2928   nbot = nbot + page_sz;
2929 
2930   // Adjust stack bottom one page up if last checked page is not mapped
2931   if (mincore_return_value == -1) {
2932     nbot = nbot + page_sz;
2933   }
2934 
2935   return nbot;
2936 }
2937 
2938 
2939 // Linux uses a growable mapping for the stack, and if the mapping for
2940 // the stack guard pages is not removed when we detach a thread the
2941 // stack cannot grow beyond the pages where the stack guard was
2942 // mapped.  If at some point later in the process the stack expands to
2943 // that point, the Linux kernel cannot expand the stack any further
2944 // because the guard pages are in the way, and a segfault occurs.
2945 //
2946 // However, it's essential not to split the stack region by unmapping
2947 // a region (leaving a hole) that's already part of the stack mapping,
2948 // so if the stack mapping has already grown beyond the guard pages at
2949 // the time we create them, we have to truncate the stack mapping.
2950 // So, we need to know the extent of the stack mapping when
2951 // create_stack_guard_pages() is called.
2952 
2953 // We only need this for stacks that are growable: at the time of
2954 // writing thread stacks don't use growable mappings (i.e. those
2955 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2956 // only applies to the main thread.
2957 
2958 // If the (growable) stack mapping already extends beyond the point
2959 // where we're going to put our guard pages, truncate the mapping at
2960 // that point by munmap()ping it.  This ensures that when we later
2961 // munmap() the guard pages we don't leave a hole in the stack
2962 // mapping. This only affects the main/initial thread
2963 
2964 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2965   if (os::Linux::is_initial_thread()) {
2966     // As we manually grow stack up to bottom inside create_attached_thread(),
2967     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2968     // we don't need to do anything special.
2969     // Check it first, before calling heavy function.
2970     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2971     unsigned char vec[1];
2972 
2973     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2974       // Fallback to slow path on all errors, including EAGAIN
2975       stack_extent = (uintptr_t) get_stack_commited_bottom(
2976                                                            os::Linux::initial_thread_stack_bottom(),
2977                                                            (size_t)addr - stack_extent);
2978     }
2979 
2980     if (stack_extent < (uintptr_t)addr) {
2981       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2982     }
2983   }
2984 
2985   return os::commit_memory(addr, size, !ExecMem);
2986 }
2987 
2988 // If this is a growable mapping, remove the guard pages entirely by
2989 // munmap()ping them.  If not, just call uncommit_memory(). This only
2990 // affects the main/initial thread, but guard against future OS changes
2991 // It's safe to always unmap guard pages for initial thread because we
2992 // always place it right after end of the mapped region
2993 
2994 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2995   uintptr_t stack_extent, stack_base;
2996 
2997   if (os::Linux::is_initial_thread()) {
2998     return ::munmap(addr, size) == 0;
2999   }
3000 
3001   return os::uncommit_memory(addr, size);
3002 }
3003 
3004 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3005 // at 'requested_addr'. If there are existing memory mappings at the same
3006 // location, however, they will be overwritten. If 'fixed' is false,
3007 // 'requested_addr' is only treated as a hint, the return value may or
3008 // may not start from the requested address. Unlike Linux mmap(), this
3009 // function returns NULL to indicate failure.
3010 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3011   char * addr;
3012   int flags;
3013 
3014   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3015   if (fixed) {
3016     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3017     flags |= MAP_FIXED;
3018   }
3019 
3020   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3021   // touch an uncommitted page. Otherwise, the read/write might
3022   // succeed if we have enough swap space to back the physical page.
3023   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3024                        flags, -1, 0);
3025 
3026   return addr == MAP_FAILED ? NULL : addr;
3027 }
3028 
3029 static int anon_munmap(char * addr, size_t size) {
3030   return ::munmap(addr, size) == 0;
3031 }
3032 
3033 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3034                             size_t alignment_hint) {
3035   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3036 }
3037 
3038 bool os::pd_release_memory(char* addr, size_t size) {
3039   return anon_munmap(addr, size);
3040 }
3041 
3042 static bool linux_mprotect(char* addr, size_t size, int prot) {
3043   // Linux wants the mprotect address argument to be page aligned.
3044   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3045 
3046   // According to SUSv3, mprotect() should only be used with mappings
3047   // established by mmap(), and mmap() always maps whole pages. Unaligned
3048   // 'addr' likely indicates problem in the VM (e.g. trying to change
3049   // protection of malloc'ed or statically allocated memory). Check the
3050   // caller if you hit this assert.
3051   assert(addr == bottom, "sanity check");
3052 
3053   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3054   return ::mprotect(bottom, size, prot) == 0;
3055 }
3056 
3057 // Set protections specified
3058 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3059                         bool is_committed) {
3060   unsigned int p = 0;
3061   switch (prot) {
3062   case MEM_PROT_NONE: p = PROT_NONE; break;
3063   case MEM_PROT_READ: p = PROT_READ; break;
3064   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3065   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3066   default:
3067     ShouldNotReachHere();
3068   }
3069   // is_committed is unused.
3070   return linux_mprotect(addr, bytes, p);
3071 }
3072 
3073 bool os::guard_memory(char* addr, size_t size) {
3074   return linux_mprotect(addr, size, PROT_NONE);
3075 }
3076 
3077 bool os::unguard_memory(char* addr, size_t size) {
3078   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3079 }
3080 
3081 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3082                                                     size_t page_size) {
3083   bool result = false;
3084   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3085                  MAP_ANONYMOUS|MAP_PRIVATE,
3086                  -1, 0);
3087   if (p != MAP_FAILED) {
3088     void *aligned_p = align_ptr_up(p, page_size);
3089 
3090     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3091 
3092     munmap(p, page_size * 2);
3093   }
3094 
3095   if (warn && !result) {
3096     warning("TransparentHugePages is not supported by the operating system.");
3097   }
3098 
3099   return result;
3100 }
3101 
3102 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3103   bool result = false;
3104   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3105                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3106                  -1, 0);
3107 
3108   if (p != MAP_FAILED) {
3109     // We don't know if this really is a huge page or not.
3110     FILE *fp = fopen("/proc/self/maps", "r");
3111     if (fp) {
3112       while (!feof(fp)) {
3113         char chars[257];
3114         long x = 0;
3115         if (fgets(chars, sizeof(chars), fp)) {
3116           if (sscanf(chars, "%lx-%*x", &x) == 1
3117               && x == (long)p) {
3118             if (strstr (chars, "hugepage")) {
3119               result = true;
3120               break;
3121             }
3122           }
3123         }
3124       }
3125       fclose(fp);
3126     }
3127     munmap(p, page_size);
3128   }
3129 
3130   if (warn && !result) {
3131     warning("HugeTLBFS is not supported by the operating system.");
3132   }
3133 
3134   return result;
3135 }
3136 
3137 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3138 //
3139 // From the coredump_filter documentation:
3140 //
3141 // - (bit 0) anonymous private memory
3142 // - (bit 1) anonymous shared memory
3143 // - (bit 2) file-backed private memory
3144 // - (bit 3) file-backed shared memory
3145 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3146 //           effective only if the bit 2 is cleared)
3147 // - (bit 5) hugetlb private memory
3148 // - (bit 6) hugetlb shared memory
3149 //
3150 static void set_coredump_filter(void) {
3151   FILE *f;
3152   long cdm;
3153 
3154   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3155     return;
3156   }
3157 
3158   if (fscanf(f, "%lx", &cdm) != 1) {
3159     fclose(f);
3160     return;
3161   }
3162 
3163   rewind(f);
3164 
3165   if ((cdm & LARGEPAGES_BIT) == 0) {
3166     cdm |= LARGEPAGES_BIT;
3167     fprintf(f, "%#lx", cdm);
3168   }
3169 
3170   fclose(f);
3171 }
3172 
3173 // Large page support
3174 
3175 static size_t _large_page_size = 0;
3176 
3177 size_t os::Linux::find_large_page_size() {
3178   size_t large_page_size = 0;
3179 
3180   // large_page_size on Linux is used to round up heap size. x86 uses either
3181   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3182   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3183   // page as large as 256M.
3184   //
3185   // Here we try to figure out page size by parsing /proc/meminfo and looking
3186   // for a line with the following format:
3187   //    Hugepagesize:     2048 kB
3188   //
3189   // If we can't determine the value (e.g. /proc is not mounted, or the text
3190   // format has been changed), we'll use the largest page size supported by
3191   // the processor.
3192 
3193 #ifndef ZERO
3194   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3195                      ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3196 #endif // ZERO
3197 
3198   FILE *fp = fopen("/proc/meminfo", "r");
3199   if (fp) {
3200     while (!feof(fp)) {
3201       int x = 0;
3202       char buf[16];
3203       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3204         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3205           large_page_size = x * K;
3206           break;
3207         }
3208       } else {
3209         // skip to next line
3210         for (;;) {
3211           int ch = fgetc(fp);
3212           if (ch == EOF || ch == (int)'\n') break;
3213         }
3214       }
3215     }
3216     fclose(fp);
3217   }
3218 
3219   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3220     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3221             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3222             proper_unit_for_byte_size(large_page_size));
3223   }
3224 
3225   return large_page_size;
3226 }
3227 
3228 size_t os::Linux::setup_large_page_size() {
3229   _large_page_size = Linux::find_large_page_size();
3230   const size_t default_page_size = (size_t)Linux::page_size();
3231   if (_large_page_size > default_page_size) {
3232     _page_sizes[0] = _large_page_size;
3233     _page_sizes[1] = default_page_size;
3234     _page_sizes[2] = 0;
3235   }
3236 
3237   return _large_page_size;
3238 }
3239 
3240 bool os::Linux::setup_large_page_type(size_t page_size) {
3241   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3242       FLAG_IS_DEFAULT(UseSHM) &&
3243       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3244 
3245     // The type of large pages has not been specified by the user.
3246 
3247     // Try UseHugeTLBFS and then UseSHM.
3248     UseHugeTLBFS = UseSHM = true;
3249 
3250     // Don't try UseTransparentHugePages since there are known
3251     // performance issues with it turned on. This might change in the future.
3252     UseTransparentHugePages = false;
3253   }
3254 
3255   if (UseTransparentHugePages) {
3256     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3257     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3258       UseHugeTLBFS = false;
3259       UseSHM = false;
3260       return true;
3261     }
3262     UseTransparentHugePages = false;
3263   }
3264 
3265   if (UseHugeTLBFS) {
3266     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3267     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3268       UseSHM = false;
3269       return true;
3270     }
3271     UseHugeTLBFS = false;
3272   }
3273 
3274   return UseSHM;
3275 }
3276 
3277 void os::large_page_init() {
3278   if (!UseLargePages &&
3279       !UseTransparentHugePages &&
3280       !UseHugeTLBFS &&
3281       !UseSHM) {
3282     // Not using large pages.
3283     return;
3284   }
3285 
3286   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3287     // The user explicitly turned off large pages.
3288     // Ignore the rest of the large pages flags.
3289     UseTransparentHugePages = false;
3290     UseHugeTLBFS = false;
3291     UseSHM = false;
3292     return;
3293   }
3294 
3295   size_t large_page_size = Linux::setup_large_page_size();
3296   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3297 
3298   set_coredump_filter();
3299 }
3300 
3301 #ifndef SHM_HUGETLB
3302   #define SHM_HUGETLB 04000
3303 #endif
3304 
3305 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3306                                             char* req_addr, bool exec) {
3307   // "exec" is passed in but not used.  Creating the shared image for
3308   // the code cache doesn't have an SHM_X executable permission to check.
3309   assert(UseLargePages && UseSHM, "only for SHM large pages");
3310   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3311 
3312   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3313     return NULL; // Fallback to small pages.
3314   }
3315 
3316   key_t key = IPC_PRIVATE;
3317   char *addr;
3318 
3319   bool warn_on_failure = UseLargePages &&
3320                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3321                          !FLAG_IS_DEFAULT(UseSHM) ||
3322                          !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3323   char msg[128];
3324 
3325   // Create a large shared memory region to attach to based on size.
3326   // Currently, size is the total size of the heap
3327   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3328   if (shmid == -1) {
3329     // Possible reasons for shmget failure:
3330     // 1. shmmax is too small for Java heap.
3331     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3332     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3333     // 2. not enough large page memory.
3334     //    > check available large pages: cat /proc/meminfo
3335     //    > increase amount of large pages:
3336     //          echo new_value > /proc/sys/vm/nr_hugepages
3337     //      Note 1: different Linux may use different name for this property,
3338     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3339     //      Note 2: it's possible there's enough physical memory available but
3340     //            they are so fragmented after a long run that they can't
3341     //            coalesce into large pages. Try to reserve large pages when
3342     //            the system is still "fresh".
3343     if (warn_on_failure) {
3344       jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3345       warning("%s", msg);
3346     }
3347     return NULL;
3348   }
3349 
3350   // attach to the region
3351   addr = (char*)shmat(shmid, req_addr, 0);
3352   int err = errno;
3353 
3354   // Remove shmid. If shmat() is successful, the actual shared memory segment
3355   // will be deleted when it's detached by shmdt() or when the process
3356   // terminates. If shmat() is not successful this will remove the shared
3357   // segment immediately.
3358   shmctl(shmid, IPC_RMID, NULL);
3359 
3360   if ((intptr_t)addr == -1) {
3361     if (warn_on_failure) {
3362       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3363       warning("%s", msg);
3364     }
3365     return NULL;
3366   }
3367 
3368   return addr;
3369 }
3370 
3371 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3372                                         int error) {
3373   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3374 
3375   bool warn_on_failure = UseLargePages &&
3376       (!FLAG_IS_DEFAULT(UseLargePages) ||
3377        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3378        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3379 
3380   if (warn_on_failure) {
3381     char msg[128];
3382     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3383                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3384     warning("%s", msg);
3385   }
3386 }
3387 
3388 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3389                                                         char* req_addr,
3390                                                         bool exec) {
3391   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3392   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3393   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3394 
3395   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3396   char* addr = (char*)::mmap(req_addr, bytes, prot,
3397                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3398                              -1, 0);
3399 
3400   if (addr == MAP_FAILED) {
3401     warn_on_large_pages_failure(req_addr, bytes, errno);
3402     return NULL;
3403   }
3404 
3405   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3406 
3407   return addr;
3408 }
3409 
3410 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3411 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3412 //   (req_addr != NULL) or with a given alignment.
3413 //  - bytes shall be a multiple of alignment.
3414 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3415 //  - alignment sets the alignment at which memory shall be allocated.
3416 //     It must be a multiple of allocation granularity.
3417 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3418 //  req_addr or NULL.
3419 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3420 
3421   size_t extra_size = bytes;
3422   if (req_addr == NULL && alignment > 0) {
3423     extra_size += alignment;
3424   }
3425 
3426   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3427     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3428     -1, 0);
3429   if (start == MAP_FAILED) {
3430     start = NULL;
3431   } else {
3432     if (req_addr != NULL) {
3433       if (start != req_addr) {
3434         ::munmap(start, extra_size);
3435         start = NULL;
3436       }
3437     } else {
3438       char* const start_aligned = (char*) align_ptr_up(start, alignment);
3439       char* const end_aligned = start_aligned + bytes;
3440       char* const end = start + extra_size;
3441       if (start_aligned > start) {
3442         ::munmap(start, start_aligned - start);
3443       }
3444       if (end_aligned < end) {
3445         ::munmap(end_aligned, end - end_aligned);
3446       }
3447       start = start_aligned;
3448     }
3449   }
3450   return start;
3451 
3452 }
3453 
3454 // Reserve memory using mmap(MAP_HUGETLB).
3455 //  - bytes shall be a multiple of alignment.
3456 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3457 //  - alignment sets the alignment at which memory shall be allocated.
3458 //     It must be a multiple of allocation granularity.
3459 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3460 //  req_addr or NULL.
3461 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3462                                                          size_t alignment,
3463                                                          char* req_addr,
3464                                                          bool exec) {
3465   size_t large_page_size = os::large_page_size();
3466   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3467 
3468   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3469   assert(is_size_aligned(bytes, alignment), "Must be");
3470 
3471   // First reserve - but not commit - the address range in small pages.
3472   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3473 
3474   if (start == NULL) {
3475     return NULL;
3476   }
3477 
3478   assert(is_ptr_aligned(start, alignment), "Must be");
3479 
3480   char* end = start + bytes;
3481 
3482   // Find the regions of the allocated chunk that can be promoted to large pages.
3483   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3484   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3485 
3486   size_t lp_bytes = lp_end - lp_start;
3487 
3488   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3489 
3490   if (lp_bytes == 0) {
3491     // The mapped region doesn't even span the start and the end of a large page.
3492     // Fall back to allocate a non-special area.
3493     ::munmap(start, end - start);
3494     return NULL;
3495   }
3496 
3497   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3498 
3499   void* result;
3500 
3501   // Commit small-paged leading area.
3502   if (start != lp_start) {
3503     result = ::mmap(start, lp_start - start, prot,
3504                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3505                     -1, 0);
3506     if (result == MAP_FAILED) {
3507       ::munmap(lp_start, end - lp_start);
3508       return NULL;
3509     }
3510   }
3511 
3512   // Commit large-paged area.
3513   result = ::mmap(lp_start, lp_bytes, prot,
3514                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3515                   -1, 0);
3516   if (result == MAP_FAILED) {
3517     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3518     // If the mmap above fails, the large pages region will be unmapped and we
3519     // have regions before and after with small pages. Release these regions.
3520     //
3521     // |  mapped  |  unmapped  |  mapped  |
3522     // ^          ^            ^          ^
3523     // start      lp_start     lp_end     end
3524     //
3525     ::munmap(start, lp_start - start);
3526     ::munmap(lp_end, end - lp_end);
3527     return NULL;
3528   }
3529 
3530   // Commit small-paged trailing area.
3531   if (lp_end != end) {
3532     result = ::mmap(lp_end, end - lp_end, prot,
3533                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3534                     -1, 0);
3535     if (result == MAP_FAILED) {
3536       ::munmap(start, lp_end - start);
3537       return NULL;
3538     }
3539   }
3540 
3541   return start;
3542 }
3543 
3544 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3545                                                    size_t alignment,
3546                                                    char* req_addr,
3547                                                    bool exec) {
3548   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3549   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3550   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3551   assert(is_power_of_2(os::large_page_size()), "Must be");
3552   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3553 
3554   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3555     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3556   } else {
3557     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3558   }
3559 }
3560 
3561 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3562                                  char* req_addr, bool exec) {
3563   assert(UseLargePages, "only for large pages");
3564 
3565   char* addr;
3566   if (UseSHM) {
3567     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3568   } else {
3569     assert(UseHugeTLBFS, "must be");
3570     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3571   }
3572 
3573   if (addr != NULL) {
3574     if (UseNUMAInterleaving) {
3575       numa_make_global(addr, bytes);
3576     }
3577 
3578     // The memory is committed
3579     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3580   }
3581 
3582   return addr;
3583 }
3584 
3585 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3586   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3587   return shmdt(base) == 0;
3588 }
3589 
3590 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3591   return pd_release_memory(base, bytes);
3592 }
3593 
3594 bool os::release_memory_special(char* base, size_t bytes) {
3595   bool res;
3596   if (MemTracker::tracking_level() > NMT_minimal) {
3597     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3598     res = os::Linux::release_memory_special_impl(base, bytes);
3599     if (res) {
3600       tkr.record((address)base, bytes);
3601     }
3602 
3603   } else {
3604     res = os::Linux::release_memory_special_impl(base, bytes);
3605   }
3606   return res;
3607 }
3608 
3609 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3610   assert(UseLargePages, "only for large pages");
3611   bool res;
3612 
3613   if (UseSHM) {
3614     res = os::Linux::release_memory_special_shm(base, bytes);
3615   } else {
3616     assert(UseHugeTLBFS, "must be");
3617     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3618   }
3619   return res;
3620 }
3621 
3622 size_t os::large_page_size() {
3623   return _large_page_size;
3624 }
3625 
3626 // With SysV SHM the entire memory region must be allocated as shared
3627 // memory.
3628 // HugeTLBFS allows application to commit large page memory on demand.
3629 // However, when committing memory with HugeTLBFS fails, the region
3630 // that was supposed to be committed will lose the old reservation
3631 // and allow other threads to steal that memory region. Because of this
3632 // behavior we can't commit HugeTLBFS memory.
3633 bool os::can_commit_large_page_memory() {
3634   return UseTransparentHugePages;
3635 }
3636 
3637 bool os::can_execute_large_page_memory() {
3638   return UseTransparentHugePages || UseHugeTLBFS;
3639 }
3640 
3641 // Reserve memory at an arbitrary address, only if that area is
3642 // available (and not reserved for something else).
3643 
3644 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3645   const int max_tries = 10;
3646   char* base[max_tries];
3647   size_t size[max_tries];
3648   const size_t gap = 0x000000;
3649 
3650   // Assert only that the size is a multiple of the page size, since
3651   // that's all that mmap requires, and since that's all we really know
3652   // about at this low abstraction level.  If we need higher alignment,
3653   // we can either pass an alignment to this method or verify alignment
3654   // in one of the methods further up the call chain.  See bug 5044738.
3655   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3656 
3657   // Repeatedly allocate blocks until the block is allocated at the
3658   // right spot.
3659 
3660   // Linux mmap allows caller to pass an address as hint; give it a try first,
3661   // if kernel honors the hint then we can return immediately.
3662   char * addr = anon_mmap(requested_addr, bytes, false);
3663   if (addr == requested_addr) {
3664     return requested_addr;
3665   }
3666 
3667   if (addr != NULL) {
3668     // mmap() is successful but it fails to reserve at the requested address
3669     anon_munmap(addr, bytes);
3670   }
3671 
3672   int i;
3673   for (i = 0; i < max_tries; ++i) {
3674     base[i] = reserve_memory(bytes);
3675 
3676     if (base[i] != NULL) {
3677       // Is this the block we wanted?
3678       if (base[i] == requested_addr) {
3679         size[i] = bytes;
3680         break;
3681       }
3682 
3683       // Does this overlap the block we wanted? Give back the overlapped
3684       // parts and try again.
3685 
3686       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3687       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3688         unmap_memory(base[i], top_overlap);
3689         base[i] += top_overlap;
3690         size[i] = bytes - top_overlap;
3691       } else {
3692         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3693         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3694           unmap_memory(requested_addr, bottom_overlap);
3695           size[i] = bytes - bottom_overlap;
3696         } else {
3697           size[i] = bytes;
3698         }
3699       }
3700     }
3701   }
3702 
3703   // Give back the unused reserved pieces.
3704 
3705   for (int j = 0; j < i; ++j) {
3706     if (base[j] != NULL) {
3707       unmap_memory(base[j], size[j]);
3708     }
3709   }
3710 
3711   if (i < max_tries) {
3712     return requested_addr;
3713   } else {
3714     return NULL;
3715   }
3716 }
3717 
3718 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3719   return ::read(fd, buf, nBytes);
3720 }
3721 
3722 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3723   return ::pread(fd, buf, nBytes, offset);
3724 }
3725 
3726 // Short sleep, direct OS call.
3727 //
3728 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3729 // sched_yield(2) will actually give up the CPU:
3730 //
3731 //   * Alone on this pariticular CPU, keeps running.
3732 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3733 //     (pre 2.6.39).
3734 //
3735 // So calling this with 0 is an alternative.
3736 //
3737 void os::naked_short_sleep(jlong ms) {
3738   struct timespec req;
3739 
3740   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3741   req.tv_sec = 0;
3742   if (ms > 0) {
3743     req.tv_nsec = (ms % 1000) * 1000000;
3744   } else {
3745     req.tv_nsec = 1;
3746   }
3747 
3748   nanosleep(&req, NULL);
3749 
3750   return;
3751 }
3752 
3753 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3754 void os::infinite_sleep() {
3755   while (true) {    // sleep forever ...
3756     ::sleep(100);   // ... 100 seconds at a time
3757   }
3758 }
3759 
3760 // Used to convert frequent JVM_Yield() to nops
3761 bool os::dont_yield() {
3762   return DontYieldALot;
3763 }
3764 
3765 void os::naked_yield() {
3766   sched_yield();
3767 }
3768 
3769 ////////////////////////////////////////////////////////////////////////////////
3770 // thread priority support
3771 
3772 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3773 // only supports dynamic priority, static priority must be zero. For real-time
3774 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3775 // However, for large multi-threaded applications, SCHED_RR is not only slower
3776 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3777 // of 5 runs - Sep 2005).
3778 //
3779 // The following code actually changes the niceness of kernel-thread/LWP. It
3780 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3781 // not the entire user process, and user level threads are 1:1 mapped to kernel
3782 // threads. It has always been the case, but could change in the future. For
3783 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3784 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3785 
3786 int os::java_to_os_priority[CriticalPriority + 1] = {
3787   19,              // 0 Entry should never be used
3788 
3789    4,              // 1 MinPriority
3790    3,              // 2
3791    2,              // 3
3792 
3793    1,              // 4
3794    0,              // 5 NormPriority
3795   -1,              // 6
3796 
3797   -2,              // 7
3798   -3,              // 8
3799   -4,              // 9 NearMaxPriority
3800 
3801   -5,              // 10 MaxPriority
3802 
3803   -5               // 11 CriticalPriority
3804 };
3805 
3806 static int prio_init() {
3807   if (ThreadPriorityPolicy == 1) {
3808     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3809     // if effective uid is not root. Perhaps, a more elegant way of doing
3810     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3811     if (geteuid() != 0) {
3812       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3813         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3814       }
3815       ThreadPriorityPolicy = 0;
3816     }
3817   }
3818   if (UseCriticalJavaThreadPriority) {
3819     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3820   }
3821   return 0;
3822 }
3823 
3824 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3825   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3826 
3827   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3828   return (ret == 0) ? OS_OK : OS_ERR;
3829 }
3830 
3831 OSReturn os::get_native_priority(const Thread* const thread,
3832                                  int *priority_ptr) {
3833   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3834     *priority_ptr = java_to_os_priority[NormPriority];
3835     return OS_OK;
3836   }
3837 
3838   errno = 0;
3839   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3840   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3841 }
3842 
3843 // Hint to the underlying OS that a task switch would not be good.
3844 // Void return because it's a hint and can fail.
3845 void os::hint_no_preempt() {}
3846 
3847 ////////////////////////////////////////////////////////////////////////////////
3848 // suspend/resume support
3849 
3850 //  the low-level signal-based suspend/resume support is a remnant from the
3851 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3852 //  within hotspot. Now there is a single use-case for this:
3853 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3854 //      that runs in the watcher thread.
3855 //  The remaining code is greatly simplified from the more general suspension
3856 //  code that used to be used.
3857 //
3858 //  The protocol is quite simple:
3859 //  - suspend:
3860 //      - sends a signal to the target thread
3861 //      - polls the suspend state of the osthread using a yield loop
3862 //      - target thread signal handler (SR_handler) sets suspend state
3863 //        and blocks in sigsuspend until continued
3864 //  - resume:
3865 //      - sets target osthread state to continue
3866 //      - sends signal to end the sigsuspend loop in the SR_handler
3867 //
3868 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3869 
3870 static void resume_clear_context(OSThread *osthread) {
3871   osthread->set_ucontext(NULL);
3872   osthread->set_siginfo(NULL);
3873 }
3874 
3875 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3876                                  ucontext_t* context) {
3877   osthread->set_ucontext(context);
3878   osthread->set_siginfo(siginfo);
3879 }
3880 
3881 // Handler function invoked when a thread's execution is suspended or
3882 // resumed. We have to be careful that only async-safe functions are
3883 // called here (Note: most pthread functions are not async safe and
3884 // should be avoided.)
3885 //
3886 // Note: sigwait() is a more natural fit than sigsuspend() from an
3887 // interface point of view, but sigwait() prevents the signal hander
3888 // from being run. libpthread would get very confused by not having
3889 // its signal handlers run and prevents sigwait()'s use with the
3890 // mutex granting granting signal.
3891 //
3892 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3893 //
3894 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3895   // Save and restore errno to avoid confusing native code with EINTR
3896   // after sigsuspend.
3897   int old_errno = errno;
3898 
3899   Thread* thread = Thread::current();
3900   OSThread* osthread = thread->osthread();
3901   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3902 
3903   os::SuspendResume::State current = osthread->sr.state();
3904   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3905     suspend_save_context(osthread, siginfo, context);
3906 
3907     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3908     os::SuspendResume::State state = osthread->sr.suspended();
3909     if (state == os::SuspendResume::SR_SUSPENDED) {
3910       sigset_t suspend_set;  // signals for sigsuspend()
3911 
3912       // get current set of blocked signals and unblock resume signal
3913       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3914       sigdelset(&suspend_set, SR_signum);
3915 
3916       sr_semaphore.signal();
3917       // wait here until we are resumed
3918       while (1) {
3919         sigsuspend(&suspend_set);
3920 
3921         os::SuspendResume::State result = osthread->sr.running();
3922         if (result == os::SuspendResume::SR_RUNNING) {
3923           sr_semaphore.signal();
3924           break;
3925         }
3926       }
3927 
3928     } else if (state == os::SuspendResume::SR_RUNNING) {
3929       // request was cancelled, continue
3930     } else {
3931       ShouldNotReachHere();
3932     }
3933 
3934     resume_clear_context(osthread);
3935   } else if (current == os::SuspendResume::SR_RUNNING) {
3936     // request was cancelled, continue
3937   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3938     // ignore
3939   } else {
3940     // ignore
3941   }
3942 
3943   errno = old_errno;
3944 }
3945 
3946 static int SR_initialize() {
3947   struct sigaction act;
3948   char *s;
3949 
3950   // Get signal number to use for suspend/resume
3951   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3952     int sig = ::strtol(s, 0, 10);
3953     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
3954         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
3955       SR_signum = sig;
3956     } else {
3957       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
3958               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
3959     }
3960   }
3961 
3962   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3963          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3964 
3965   sigemptyset(&SR_sigset);
3966   sigaddset(&SR_sigset, SR_signum);
3967 
3968   // Set up signal handler for suspend/resume
3969   act.sa_flags = SA_RESTART|SA_SIGINFO;
3970   act.sa_handler = (void (*)(int)) SR_handler;
3971 
3972   // SR_signum is blocked by default.
3973   // 4528190 - We also need to block pthread restart signal (32 on all
3974   // supported Linux platforms). Note that LinuxThreads need to block
3975   // this signal for all threads to work properly. So we don't have
3976   // to use hard-coded signal number when setting up the mask.
3977   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3978 
3979   if (sigaction(SR_signum, &act, 0) == -1) {
3980     return -1;
3981   }
3982 
3983   // Save signal flag
3984   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3985   return 0;
3986 }
3987 
3988 static int sr_notify(OSThread* osthread) {
3989   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3990   assert_status(status == 0, status, "pthread_kill");
3991   return status;
3992 }
3993 
3994 // "Randomly" selected value for how long we want to spin
3995 // before bailing out on suspending a thread, also how often
3996 // we send a signal to a thread we want to resume
3997 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3998 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3999 
4000 // returns true on success and false on error - really an error is fatal
4001 // but this seems the normal response to library errors
4002 static bool do_suspend(OSThread* osthread) {
4003   assert(osthread->sr.is_running(), "thread should be running");
4004   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4005 
4006   // mark as suspended and send signal
4007   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4008     // failed to switch, state wasn't running?
4009     ShouldNotReachHere();
4010     return false;
4011   }
4012 
4013   if (sr_notify(osthread) != 0) {
4014     ShouldNotReachHere();
4015   }
4016 
4017   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4018   while (true) {
4019     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4020       break;
4021     } else {
4022       // timeout
4023       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4024       if (cancelled == os::SuspendResume::SR_RUNNING) {
4025         return false;
4026       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4027         // make sure that we consume the signal on the semaphore as well
4028         sr_semaphore.wait();
4029         break;
4030       } else {
4031         ShouldNotReachHere();
4032         return false;
4033       }
4034     }
4035   }
4036 
4037   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4038   return true;
4039 }
4040 
4041 static void do_resume(OSThread* osthread) {
4042   assert(osthread->sr.is_suspended(), "thread should be suspended");
4043   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4044 
4045   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4046     // failed to switch to WAKEUP_REQUEST
4047     ShouldNotReachHere();
4048     return;
4049   }
4050 
4051   while (true) {
4052     if (sr_notify(osthread) == 0) {
4053       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4054         if (osthread->sr.is_running()) {
4055           return;
4056         }
4057       }
4058     } else {
4059       ShouldNotReachHere();
4060     }
4061   }
4062 
4063   guarantee(osthread->sr.is_running(), "Must be running!");
4064 }
4065 
4066 ///////////////////////////////////////////////////////////////////////////////////
4067 // signal handling (except suspend/resume)
4068 
4069 // This routine may be used by user applications as a "hook" to catch signals.
4070 // The user-defined signal handler must pass unrecognized signals to this
4071 // routine, and if it returns true (non-zero), then the signal handler must
4072 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4073 // routine will never retun false (zero), but instead will execute a VM panic
4074 // routine kill the process.
4075 //
4076 // If this routine returns false, it is OK to call it again.  This allows
4077 // the user-defined signal handler to perform checks either before or after
4078 // the VM performs its own checks.  Naturally, the user code would be making
4079 // a serious error if it tried to handle an exception (such as a null check
4080 // or breakpoint) that the VM was generating for its own correct operation.
4081 //
4082 // This routine may recognize any of the following kinds of signals:
4083 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4084 // It should be consulted by handlers for any of those signals.
4085 //
4086 // The caller of this routine must pass in the three arguments supplied
4087 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4088 // field of the structure passed to sigaction().  This routine assumes that
4089 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4090 //
4091 // Note that the VM will print warnings if it detects conflicting signal
4092 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4093 //
4094 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4095                                                  siginfo_t* siginfo,
4096                                                  void* ucontext,
4097                                                  int abort_if_unrecognized);
4098 
4099 void signalHandler(int sig, siginfo_t* info, void* uc) {
4100   assert(info != NULL && uc != NULL, "it must be old kernel");
4101   int orig_errno = errno;  // Preserve errno value over signal handler.
4102   JVM_handle_linux_signal(sig, info, uc, true);
4103   errno = orig_errno;
4104 }
4105 
4106 
4107 // This boolean allows users to forward their own non-matching signals
4108 // to JVM_handle_linux_signal, harmlessly.
4109 bool os::Linux::signal_handlers_are_installed = false;
4110 
4111 // For signal-chaining
4112 struct sigaction sigact[NSIG];
4113 uint64_t sigs = 0;
4114 #if (64 < NSIG-1)
4115 #error "Not all signals can be encoded in sigs. Adapt its type!"
4116 #endif
4117 bool os::Linux::libjsig_is_loaded = false;
4118 typedef struct sigaction *(*get_signal_t)(int);
4119 get_signal_t os::Linux::get_signal_action = NULL;
4120 
4121 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4122   struct sigaction *actp = NULL;
4123 
4124   if (libjsig_is_loaded) {
4125     // Retrieve the old signal handler from libjsig
4126     actp = (*get_signal_action)(sig);
4127   }
4128   if (actp == NULL) {
4129     // Retrieve the preinstalled signal handler from jvm
4130     actp = get_preinstalled_handler(sig);
4131   }
4132 
4133   return actp;
4134 }
4135 
4136 static bool call_chained_handler(struct sigaction *actp, int sig,
4137                                  siginfo_t *siginfo, void *context) {
4138   // Call the old signal handler
4139   if (actp->sa_handler == SIG_DFL) {
4140     // It's more reasonable to let jvm treat it as an unexpected exception
4141     // instead of taking the default action.
4142     return false;
4143   } else if (actp->sa_handler != SIG_IGN) {
4144     if ((actp->sa_flags & SA_NODEFER) == 0) {
4145       // automaticlly block the signal
4146       sigaddset(&(actp->sa_mask), sig);
4147     }
4148 
4149     sa_handler_t hand = NULL;
4150     sa_sigaction_t sa = NULL;
4151     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4152     // retrieve the chained handler
4153     if (siginfo_flag_set) {
4154       sa = actp->sa_sigaction;
4155     } else {
4156       hand = actp->sa_handler;
4157     }
4158 
4159     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4160       actp->sa_handler = SIG_DFL;
4161     }
4162 
4163     // try to honor the signal mask
4164     sigset_t oset;
4165     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4166 
4167     // call into the chained handler
4168     if (siginfo_flag_set) {
4169       (*sa)(sig, siginfo, context);
4170     } else {
4171       (*hand)(sig);
4172     }
4173 
4174     // restore the signal mask
4175     pthread_sigmask(SIG_SETMASK, &oset, 0);
4176   }
4177   // Tell jvm's signal handler the signal is taken care of.
4178   return true;
4179 }
4180 
4181 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4182   bool chained = false;
4183   // signal-chaining
4184   if (UseSignalChaining) {
4185     struct sigaction *actp = get_chained_signal_action(sig);
4186     if (actp != NULL) {
4187       chained = call_chained_handler(actp, sig, siginfo, context);
4188     }
4189   }
4190   return chained;
4191 }
4192 
4193 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4194   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4195     return &sigact[sig];
4196   }
4197   return NULL;
4198 }
4199 
4200 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4201   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4202   sigact[sig] = oldAct;
4203   sigs |= (uint64_t)1 << (sig-1);
4204 }
4205 
4206 // for diagnostic
4207 int sigflags[NSIG];
4208 
4209 int os::Linux::get_our_sigflags(int sig) {
4210   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4211   return sigflags[sig];
4212 }
4213 
4214 void os::Linux::set_our_sigflags(int sig, int flags) {
4215   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4216   if (sig > 0 && sig < NSIG) {
4217     sigflags[sig] = flags;
4218   }
4219 }
4220 
4221 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4222   // Check for overwrite.
4223   struct sigaction oldAct;
4224   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4225 
4226   void* oldhand = oldAct.sa_sigaction
4227                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4228                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4229   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4230       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4231       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4232     if (AllowUserSignalHandlers || !set_installed) {
4233       // Do not overwrite; user takes responsibility to forward to us.
4234       return;
4235     } else if (UseSignalChaining) {
4236       // save the old handler in jvm
4237       save_preinstalled_handler(sig, oldAct);
4238       // libjsig also interposes the sigaction() call below and saves the
4239       // old sigaction on it own.
4240     } else {
4241       fatal("Encountered unexpected pre-existing sigaction handler "
4242             "%#lx for signal %d.", (long)oldhand, sig);
4243     }
4244   }
4245 
4246   struct sigaction sigAct;
4247   sigfillset(&(sigAct.sa_mask));
4248   sigAct.sa_handler = SIG_DFL;
4249   if (!set_installed) {
4250     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4251   } else {
4252     sigAct.sa_sigaction = signalHandler;
4253     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4254   }
4255   // Save flags, which are set by ours
4256   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4257   sigflags[sig] = sigAct.sa_flags;
4258 
4259   int ret = sigaction(sig, &sigAct, &oldAct);
4260   assert(ret == 0, "check");
4261 
4262   void* oldhand2  = oldAct.sa_sigaction
4263                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4264                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4265   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4266 }
4267 
4268 // install signal handlers for signals that HotSpot needs to
4269 // handle in order to support Java-level exception handling.
4270 
4271 void os::Linux::install_signal_handlers() {
4272   if (!signal_handlers_are_installed) {
4273     signal_handlers_are_installed = true;
4274 
4275     // signal-chaining
4276     typedef void (*signal_setting_t)();
4277     signal_setting_t begin_signal_setting = NULL;
4278     signal_setting_t end_signal_setting = NULL;
4279     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4280                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4281     if (begin_signal_setting != NULL) {
4282       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4283                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4284       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4285                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4286       libjsig_is_loaded = true;
4287       assert(UseSignalChaining, "should enable signal-chaining");
4288     }
4289     if (libjsig_is_loaded) {
4290       // Tell libjsig jvm is setting signal handlers
4291       (*begin_signal_setting)();
4292     }
4293 
4294     set_signal_handler(SIGSEGV, true);
4295     set_signal_handler(SIGPIPE, true);
4296     set_signal_handler(SIGBUS, true);
4297     set_signal_handler(SIGILL, true);
4298     set_signal_handler(SIGFPE, true);
4299 #if defined(PPC64)
4300     set_signal_handler(SIGTRAP, true);
4301 #endif
4302     set_signal_handler(SIGXFSZ, true);
4303 
4304     if (libjsig_is_loaded) {
4305       // Tell libjsig jvm finishes setting signal handlers
4306       (*end_signal_setting)();
4307     }
4308 
4309     // We don't activate signal checker if libjsig is in place, we trust ourselves
4310     // and if UserSignalHandler is installed all bets are off.
4311     // Log that signal checking is off only if -verbose:jni is specified.
4312     if (CheckJNICalls) {
4313       if (libjsig_is_loaded) {
4314         if (PrintJNIResolving) {
4315           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4316         }
4317         check_signals = false;
4318       }
4319       if (AllowUserSignalHandlers) {
4320         if (PrintJNIResolving) {
4321           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4322         }
4323         check_signals = false;
4324       }
4325     }
4326   }
4327 }
4328 
4329 // This is the fastest way to get thread cpu time on Linux.
4330 // Returns cpu time (user+sys) for any thread, not only for current.
4331 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4332 // It might work on 2.6.10+ with a special kernel/glibc patch.
4333 // For reference, please, see IEEE Std 1003.1-2004:
4334 //   http://www.unix.org/single_unix_specification
4335 
4336 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4337   struct timespec tp;
4338   int rc = os::Linux::clock_gettime(clockid, &tp);
4339   assert(rc == 0, "clock_gettime is expected to return 0 code");
4340 
4341   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4342 }
4343 
4344 /////
4345 // glibc on Linux platform uses non-documented flag
4346 // to indicate, that some special sort of signal
4347 // trampoline is used.
4348 // We will never set this flag, and we should
4349 // ignore this flag in our diagnostic
4350 #ifdef SIGNIFICANT_SIGNAL_MASK
4351   #undef SIGNIFICANT_SIGNAL_MASK
4352 #endif
4353 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4354 
4355 static const char* get_signal_handler_name(address handler,
4356                                            char* buf, int buflen) {
4357   int offset = 0;
4358   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4359   if (found) {
4360     // skip directory names
4361     const char *p1, *p2;
4362     p1 = buf;
4363     size_t len = strlen(os::file_separator());
4364     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4365     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4366   } else {
4367     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4368   }
4369   return buf;
4370 }
4371 
4372 static void print_signal_handler(outputStream* st, int sig,
4373                                  char* buf, size_t buflen) {
4374   struct sigaction sa;
4375 
4376   sigaction(sig, NULL, &sa);
4377 
4378   // See comment for SIGNIFICANT_SIGNAL_MASK define
4379   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4380 
4381   st->print("%s: ", os::exception_name(sig, buf, buflen));
4382 
4383   address handler = (sa.sa_flags & SA_SIGINFO)
4384     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4385     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4386 
4387   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4388     st->print("SIG_DFL");
4389   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4390     st->print("SIG_IGN");
4391   } else {
4392     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4393   }
4394 
4395   st->print(", sa_mask[0]=");
4396   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4397 
4398   address rh = VMError::get_resetted_sighandler(sig);
4399   // May be, handler was resetted by VMError?
4400   if (rh != NULL) {
4401     handler = rh;
4402     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4403   }
4404 
4405   st->print(", sa_flags=");
4406   os::Posix::print_sa_flags(st, sa.sa_flags);
4407 
4408   // Check: is it our handler?
4409   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4410       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4411     // It is our signal handler
4412     // check for flags, reset system-used one!
4413     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4414       st->print(
4415                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4416                 os::Linux::get_our_sigflags(sig));
4417     }
4418   }
4419   st->cr();
4420 }
4421 
4422 
4423 #define DO_SIGNAL_CHECK(sig)                      \
4424   do {                                            \
4425     if (!sigismember(&check_signal_done, sig)) {  \
4426       os::Linux::check_signal_handler(sig);       \
4427     }                                             \
4428   } while (0)
4429 
4430 // This method is a periodic task to check for misbehaving JNI applications
4431 // under CheckJNI, we can add any periodic checks here
4432 
4433 void os::run_periodic_checks() {
4434   if (check_signals == false) return;
4435 
4436   // SEGV and BUS if overridden could potentially prevent
4437   // generation of hs*.log in the event of a crash, debugging
4438   // such a case can be very challenging, so we absolutely
4439   // check the following for a good measure:
4440   DO_SIGNAL_CHECK(SIGSEGV);
4441   DO_SIGNAL_CHECK(SIGILL);
4442   DO_SIGNAL_CHECK(SIGFPE);
4443   DO_SIGNAL_CHECK(SIGBUS);
4444   DO_SIGNAL_CHECK(SIGPIPE);
4445   DO_SIGNAL_CHECK(SIGXFSZ);
4446 #if defined(PPC64)
4447   DO_SIGNAL_CHECK(SIGTRAP);
4448 #endif
4449 
4450   // ReduceSignalUsage allows the user to override these handlers
4451   // see comments at the very top and jvm_solaris.h
4452   if (!ReduceSignalUsage) {
4453     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4454     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4455     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4456     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4457   }
4458 
4459   DO_SIGNAL_CHECK(SR_signum);
4460 }
4461 
4462 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4463 
4464 static os_sigaction_t os_sigaction = NULL;
4465 
4466 void os::Linux::check_signal_handler(int sig) {
4467   char buf[O_BUFLEN];
4468   address jvmHandler = NULL;
4469 
4470 
4471   struct sigaction act;
4472   if (os_sigaction == NULL) {
4473     // only trust the default sigaction, in case it has been interposed
4474     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4475     if (os_sigaction == NULL) return;
4476   }
4477 
4478   os_sigaction(sig, (struct sigaction*)NULL, &act);
4479 
4480 
4481   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4482 
4483   address thisHandler = (act.sa_flags & SA_SIGINFO)
4484     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4485     : CAST_FROM_FN_PTR(address, act.sa_handler);
4486 
4487 
4488   switch (sig) {
4489   case SIGSEGV:
4490   case SIGBUS:
4491   case SIGFPE:
4492   case SIGPIPE:
4493   case SIGILL:
4494   case SIGXFSZ:
4495     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4496     break;
4497 
4498   case SHUTDOWN1_SIGNAL:
4499   case SHUTDOWN2_SIGNAL:
4500   case SHUTDOWN3_SIGNAL:
4501   case BREAK_SIGNAL:
4502     jvmHandler = (address)user_handler();
4503     break;
4504 
4505   default:
4506     if (sig == SR_signum) {
4507       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4508     } else {
4509       return;
4510     }
4511     break;
4512   }
4513 
4514   if (thisHandler != jvmHandler) {
4515     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4516     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4517     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4518     // No need to check this sig any longer
4519     sigaddset(&check_signal_done, sig);
4520     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4521     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4522       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4523                     exception_name(sig, buf, O_BUFLEN));
4524     }
4525   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4526     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4527     tty->print("expected:");
4528     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4529     tty->cr();
4530     tty->print("  found:");
4531     os::Posix::print_sa_flags(tty, act.sa_flags);
4532     tty->cr();
4533     // No need to check this sig any longer
4534     sigaddset(&check_signal_done, sig);
4535   }
4536 
4537   // Dump all the signal
4538   if (sigismember(&check_signal_done, sig)) {
4539     print_signal_handlers(tty, buf, O_BUFLEN);
4540   }
4541 }
4542 
4543 extern void report_error(char* file_name, int line_no, char* title,
4544                          char* format, ...);
4545 
4546 // this is called _before_ the most of global arguments have been parsed
4547 void os::init(void) {
4548   char dummy;   // used to get a guess on initial stack address
4549 //  first_hrtime = gethrtime();
4550 
4551   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4552 
4553   init_random(1234567);
4554 
4555   ThreadCritical::initialize();
4556 
4557   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4558   if (Linux::page_size() == -1) {
4559     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4560           strerror(errno));
4561   }
4562   init_page_sizes((size_t) Linux::page_size());
4563 
4564   Linux::initialize_system_info();
4565 
4566   // main_thread points to the aboriginal thread
4567   Linux::_main_thread = pthread_self();
4568 
4569   Linux::clock_init();
4570   initial_time_count = javaTimeNanos();
4571 
4572   // pthread_condattr initialization for monotonic clock
4573   int status;
4574   pthread_condattr_t* _condattr = os::Linux::condAttr();
4575   if ((status = pthread_condattr_init(_condattr)) != 0) {
4576     fatal("pthread_condattr_init: %s", strerror(status));
4577   }
4578   // Only set the clock if CLOCK_MONOTONIC is available
4579   if (os::supports_monotonic_clock()) {
4580     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4581       if (status == EINVAL) {
4582         warning("Unable to use monotonic clock with relative timed-waits" \
4583                 " - changes to the time-of-day clock may have adverse affects");
4584       } else {
4585         fatal("pthread_condattr_setclock: %s", strerror(status));
4586       }
4587     }
4588   }
4589   // else it defaults to CLOCK_REALTIME
4590 
4591   // retrieve entry point for pthread_setname_np
4592   Linux::_pthread_setname_np =
4593     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4594 
4595 }
4596 
4597 // To install functions for atexit system call
4598 extern "C" {
4599   static void perfMemory_exit_helper() {
4600     perfMemory_exit();
4601   }
4602 }
4603 
4604 // this is called _after_ the global arguments have been parsed
4605 jint os::init_2(void) {
4606   Linux::fast_thread_clock_init();
4607 
4608   // Allocate a single page and mark it as readable for safepoint polling
4609   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4610   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4611 
4612   os::set_polling_page(polling_page);
4613 
4614 #ifndef PRODUCT
4615   if (Verbose && PrintMiscellaneous) {
4616     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4617                (intptr_t)polling_page);
4618   }
4619 #endif
4620 
4621   if (!UseMembar) {
4622     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4623     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4624     os::set_memory_serialize_page(mem_serialize_page);
4625 
4626 #ifndef PRODUCT
4627     if (Verbose && PrintMiscellaneous) {
4628       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n",
4629                  (intptr_t)mem_serialize_page);
4630     }
4631 #endif
4632   }
4633 
4634   // initialize suspend/resume support - must do this before signal_sets_init()
4635   if (SR_initialize() != 0) {
4636     perror("SR_initialize failed");
4637     return JNI_ERR;
4638   }
4639 
4640   Linux::signal_sets_init();
4641   Linux::install_signal_handlers();
4642 
4643   // Check minimum allowable stack size for thread creation and to initialize
4644   // the java system classes, including StackOverflowError - depends on page
4645   // size.  Add a page for compiler2 recursion in main thread.
4646   // Add in 2*BytesPerWord times page size to account for VM stack during
4647   // class initialization depending on 32 or 64 bit VM.
4648   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4649                                       JavaThread::stack_guard_zone_size() +
4650                                       JavaThread::stack_shadow_zone_size() +
4651                                       (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4652 
4653   size_t threadStackSizeInBytes = ThreadStackSize * K;
4654   if (threadStackSizeInBytes != 0 &&
4655       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4656     tty->print_cr("\nThe stack size specified is too small, "
4657                   "Specify at least " SIZE_FORMAT "k",
4658                   os::Linux::min_stack_allowed/ K);
4659     return JNI_ERR;
4660   }
4661 
4662   // Make the stack size a multiple of the page size so that
4663   // the yellow/red zones can be guarded.
4664   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4665                                                 vm_page_size()));
4666 
4667   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4668 
4669 #if defined(IA32)
4670   workaround_expand_exec_shield_cs_limit();
4671 #endif
4672 
4673   Linux::libpthread_init();
4674   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4675     tty->print_cr("[HotSpot is running with %s, %s]\n",
4676                   Linux::glibc_version(), Linux::libpthread_version());
4677   }
4678 
4679   if (UseNUMA) {
4680     if (!Linux::libnuma_init()) {
4681       UseNUMA = false;
4682     } else {
4683       if ((Linux::numa_max_node() < 1)) {
4684         // There's only one node(they start from 0), disable NUMA.
4685         UseNUMA = false;
4686       }
4687     }
4688     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4689     // we can make the adaptive lgrp chunk resizing work. If the user specified
4690     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4691     // disable adaptive resizing.
4692     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4693       if (FLAG_IS_DEFAULT(UseNUMA)) {
4694         UseNUMA = false;
4695       } else {
4696         if (FLAG_IS_DEFAULT(UseLargePages) &&
4697             FLAG_IS_DEFAULT(UseSHM) &&
4698             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4699           UseLargePages = false;
4700         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4701           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4702           UseAdaptiveSizePolicy = false;
4703           UseAdaptiveNUMAChunkSizing = false;
4704         }
4705       }
4706     }
4707     if (!UseNUMA && ForceNUMA) {
4708       UseNUMA = true;
4709     }
4710   }
4711 
4712   if (MaxFDLimit) {
4713     // set the number of file descriptors to max. print out error
4714     // if getrlimit/setrlimit fails but continue regardless.
4715     struct rlimit nbr_files;
4716     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4717     if (status != 0) {
4718       if (PrintMiscellaneous && (Verbose || WizardMode)) {
4719         perror("os::init_2 getrlimit failed");
4720       }
4721     } else {
4722       nbr_files.rlim_cur = nbr_files.rlim_max;
4723       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4724       if (status != 0) {
4725         if (PrintMiscellaneous && (Verbose || WizardMode)) {
4726           perror("os::init_2 setrlimit failed");
4727         }
4728       }
4729     }
4730   }
4731 
4732   // Initialize lock used to serialize thread creation (see os::create_thread)
4733   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4734 
4735   // at-exit methods are called in the reverse order of their registration.
4736   // atexit functions are called on return from main or as a result of a
4737   // call to exit(3C). There can be only 32 of these functions registered
4738   // and atexit() does not set errno.
4739 
4740   if (PerfAllowAtExitRegistration) {
4741     // only register atexit functions if PerfAllowAtExitRegistration is set.
4742     // atexit functions can be delayed until process exit time, which
4743     // can be problematic for embedded VM situations. Embedded VMs should
4744     // call DestroyJavaVM() to assure that VM resources are released.
4745 
4746     // note: perfMemory_exit_helper atexit function may be removed in
4747     // the future if the appropriate cleanup code can be added to the
4748     // VM_Exit VMOperation's doit method.
4749     if (atexit(perfMemory_exit_helper) != 0) {
4750       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4751     }
4752   }
4753 
4754   // initialize thread priority policy
4755   prio_init();
4756 
4757   return JNI_OK;
4758 }
4759 
4760 // Mark the polling page as unreadable
4761 void os::make_polling_page_unreadable(void) {
4762   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4763     fatal("Could not disable polling page");
4764   }
4765 }
4766 
4767 // Mark the polling page as readable
4768 void os::make_polling_page_readable(void) {
4769   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4770     fatal("Could not enable polling page");
4771   }
4772 }
4773 
4774 // Get the current number of available processors for this process.
4775 // This value can change at any time during a process's lifetime.
4776 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4777 // If it appears there may be more than 1024 processors then we do a
4778 // dynamic check - see 6515172 for details.
4779 // If anything goes wrong we fallback to returning the number of online
4780 // processors - which can be greater than the number available to the process.
4781 int os::active_processor_count() {
4782   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
4783   cpu_set_t* cpus_p = &cpus;
4784   int cpus_size = sizeof(cpu_set_t);
4785 
4786   int configured_cpus = processor_count();  // upper bound on available cpus
4787   int cpu_count = 0;
4788 
4789   // To enable easy testing of the dynamic path on different platforms we
4790   // introduce an experimental flag: UseCpuAllocPath
4791   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4792     // kernel may use a mask bigger than cpu_set_t
4793     log_trace(os)("active_processor_count: using dynamic path %s"
4794                   "- configured processors: %d",
4795                   UseCpuAllocPath ? "(forced) " : "",
4796                   configured_cpus);
4797     cpus_p = CPU_ALLOC(configured_cpus);
4798     if (cpus_p != NULL) {
4799       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4800       // zero it just to be safe
4801       CPU_ZERO_S(cpus_size, cpus_p);
4802     }
4803     else {
4804        // failed to allocate so fallback to online cpus
4805        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4806        log_trace(os)("active_processor_count: "
4807                      "CPU_ALLOC failed (%s) - using "
4808                      "online processor count: %d",
4809                      strerror(errno), online_cpus);
4810        return online_cpus;
4811     }
4812   }
4813   else {
4814     log_trace(os)("active_processor_count: using static path - configured processors: %d",
4815                   configured_cpus);
4816   }
4817        
4818   // pid 0 means the current thread - which we have to assume represents the process
4819   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4820     if (cpus_p != &cpus) {
4821       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4822     }
4823     else {
4824       cpu_count = CPU_COUNT(cpus_p);
4825     }
4826     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4827   }
4828   else {
4829     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4830     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4831             "which may exceed available processors", strerror(errno), cpu_count);
4832   }
4833 
4834   if (cpus_p != &cpus) {
4835     CPU_FREE(cpus_p);
4836   }  
4837 
4838   assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
4839   return cpu_count;
4840 }
4841 
4842 void os::set_native_thread_name(const char *name) {
4843   if (Linux::_pthread_setname_np) {
4844     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4845     snprintf(buf, sizeof(buf), "%s", name);
4846     buf[sizeof(buf) - 1] = '\0';
4847     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4848     // ERANGE should not happen; all other errors should just be ignored.
4849     assert(rc != ERANGE, "pthread_setname_np failed");
4850   }
4851 }
4852 
4853 bool os::distribute_processes(uint length, uint* distribution) {
4854   // Not yet implemented.
4855   return false;
4856 }
4857 
4858 bool os::bind_to_processor(uint processor_id) {
4859   // Not yet implemented.
4860   return false;
4861 }
4862 
4863 ///
4864 
4865 void os::SuspendedThreadTask::internal_do_task() {
4866   if (do_suspend(_thread->osthread())) {
4867     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4868     do_task(context);
4869     do_resume(_thread->osthread());
4870   }
4871 }
4872 
4873 class PcFetcher : public os::SuspendedThreadTask {
4874  public:
4875   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4876   ExtendedPC result();
4877  protected:
4878   void do_task(const os::SuspendedThreadTaskContext& context);
4879  private:
4880   ExtendedPC _epc;
4881 };
4882 
4883 ExtendedPC PcFetcher::result() {
4884   guarantee(is_done(), "task is not done yet.");
4885   return _epc;
4886 }
4887 
4888 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4889   Thread* thread = context.thread();
4890   OSThread* osthread = thread->osthread();
4891   if (osthread->ucontext() != NULL) {
4892     _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
4893   } else {
4894     // NULL context is unexpected, double-check this is the VMThread
4895     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4896   }
4897 }
4898 
4899 // Suspends the target using the signal mechanism and then grabs the PC before
4900 // resuming the target. Used by the flat-profiler only
4901 ExtendedPC os::get_thread_pc(Thread* thread) {
4902   // Make sure that it is called by the watcher for the VMThread
4903   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4904   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4905 
4906   PcFetcher fetcher(thread);
4907   fetcher.run();
4908   return fetcher.result();
4909 }
4910 
4911 ////////////////////////////////////////////////////////////////////////////////
4912 // debug support
4913 
4914 bool os::find(address addr, outputStream* st) {
4915   Dl_info dlinfo;
4916   memset(&dlinfo, 0, sizeof(dlinfo));
4917   if (dladdr(addr, &dlinfo) != 0) {
4918     st->print(PTR_FORMAT ": ", p2i(addr));
4919     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4920       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
4921                 p2i(addr) - p2i(dlinfo.dli_saddr));
4922     } else if (dlinfo.dli_fbase != NULL) {
4923       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
4924     } else {
4925       st->print("<absolute address>");
4926     }
4927     if (dlinfo.dli_fname != NULL) {
4928       st->print(" in %s", dlinfo.dli_fname);
4929     }
4930     if (dlinfo.dli_fbase != NULL) {
4931       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
4932     }
4933     st->cr();
4934 
4935     if (Verbose) {
4936       // decode some bytes around the PC
4937       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4938       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4939       address       lowest = (address) dlinfo.dli_sname;
4940       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4941       if (begin < lowest)  begin = lowest;
4942       Dl_info dlinfo2;
4943       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4944           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4945         end = (address) dlinfo2.dli_saddr;
4946       }
4947       Disassembler::decode(begin, end, st);
4948     }
4949     return true;
4950   }
4951   return false;
4952 }
4953 
4954 ////////////////////////////////////////////////////////////////////////////////
4955 // misc
4956 
4957 // This does not do anything on Linux. This is basically a hook for being
4958 // able to use structured exception handling (thread-local exception filters)
4959 // on, e.g., Win32.
4960 void
4961 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
4962                          JavaCallArguments* args, Thread* thread) {
4963   f(value, method, args, thread);
4964 }
4965 
4966 void os::print_statistics() {
4967 }
4968 
4969 bool os::message_box(const char* title, const char* message) {
4970   int i;
4971   fdStream err(defaultStream::error_fd());
4972   for (i = 0; i < 78; i++) err.print_raw("=");
4973   err.cr();
4974   err.print_raw_cr(title);
4975   for (i = 0; i < 78; i++) err.print_raw("-");
4976   err.cr();
4977   err.print_raw_cr(message);
4978   for (i = 0; i < 78; i++) err.print_raw("=");
4979   err.cr();
4980 
4981   char buf[16];
4982   // Prevent process from exiting upon "read error" without consuming all CPU
4983   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4984 
4985   return buf[0] == 'y' || buf[0] == 'Y';
4986 }
4987 
4988 int os::stat(const char *path, struct stat *sbuf) {
4989   char pathbuf[MAX_PATH];
4990   if (strlen(path) > MAX_PATH - 1) {
4991     errno = ENAMETOOLONG;
4992     return -1;
4993   }
4994   os::native_path(strcpy(pathbuf, path));
4995   return ::stat(pathbuf, sbuf);
4996 }
4997 
4998 bool os::check_heap(bool force) {
4999   return true;
5000 }
5001 
5002 // Is a (classpath) directory empty?
5003 bool os::dir_is_empty(const char* path) {
5004   DIR *dir = NULL;
5005   struct dirent *ptr;
5006 
5007   dir = opendir(path);
5008   if (dir == NULL) return true;
5009 
5010   // Scan the directory
5011   bool result = true;
5012   char buf[sizeof(struct dirent) + MAX_PATH];
5013   while (result && (ptr = ::readdir(dir)) != NULL) {
5014     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5015       result = false;
5016     }
5017   }
5018   closedir(dir);
5019   return result;
5020 }
5021 
5022 // This code originates from JDK's sysOpen and open64_w
5023 // from src/solaris/hpi/src/system_md.c
5024 
5025 int os::open(const char *path, int oflag, int mode) {
5026   if (strlen(path) > MAX_PATH - 1) {
5027     errno = ENAMETOOLONG;
5028     return -1;
5029   }
5030 
5031   // All file descriptors that are opened in the Java process and not
5032   // specifically destined for a subprocess should have the close-on-exec
5033   // flag set.  If we don't set it, then careless 3rd party native code
5034   // might fork and exec without closing all appropriate file descriptors
5035   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5036   // turn might:
5037   //
5038   // - cause end-of-file to fail to be detected on some file
5039   //   descriptors, resulting in mysterious hangs, or
5040   //
5041   // - might cause an fopen in the subprocess to fail on a system
5042   //   suffering from bug 1085341.
5043   //
5044   // (Yes, the default setting of the close-on-exec flag is a Unix
5045   // design flaw)
5046   //
5047   // See:
5048   // 1085341: 32-bit stdio routines should support file descriptors >255
5049   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5050   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5051   //
5052   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5053   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5054   // because it saves a system call and removes a small window where the flag
5055   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5056   // and we fall back to using FD_CLOEXEC (see below).
5057 #ifdef O_CLOEXEC
5058   oflag |= O_CLOEXEC;
5059 #endif
5060 
5061   int fd = ::open64(path, oflag, mode);
5062   if (fd == -1) return -1;
5063 
5064   //If the open succeeded, the file might still be a directory
5065   {
5066     struct stat64 buf64;
5067     int ret = ::fstat64(fd, &buf64);
5068     int st_mode = buf64.st_mode;
5069 
5070     if (ret != -1) {
5071       if ((st_mode & S_IFMT) == S_IFDIR) {
5072         errno = EISDIR;
5073         ::close(fd);
5074         return -1;
5075       }
5076     } else {
5077       ::close(fd);
5078       return -1;
5079     }
5080   }
5081 
5082 #ifdef FD_CLOEXEC
5083   // Validate that the use of the O_CLOEXEC flag on open above worked.
5084   // With recent kernels, we will perform this check exactly once.
5085   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5086   if (!O_CLOEXEC_is_known_to_work) {
5087     int flags = ::fcntl(fd, F_GETFD);
5088     if (flags != -1) {
5089       if ((flags & FD_CLOEXEC) != 0)
5090         O_CLOEXEC_is_known_to_work = 1;
5091       else
5092         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5093     }
5094   }
5095 #endif
5096 
5097   return fd;
5098 }
5099 
5100 
5101 // create binary file, rewriting existing file if required
5102 int os::create_binary_file(const char* path, bool rewrite_existing) {
5103   int oflags = O_WRONLY | O_CREAT;
5104   if (!rewrite_existing) {
5105     oflags |= O_EXCL;
5106   }
5107   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5108 }
5109 
5110 // return current position of file pointer
5111 jlong os::current_file_offset(int fd) {
5112   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5113 }
5114 
5115 // move file pointer to the specified offset
5116 jlong os::seek_to_file_offset(int fd, jlong offset) {
5117   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5118 }
5119 
5120 // This code originates from JDK's sysAvailable
5121 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5122 
5123 int os::available(int fd, jlong *bytes) {
5124   jlong cur, end;
5125   int mode;
5126   struct stat64 buf64;
5127 
5128   if (::fstat64(fd, &buf64) >= 0) {
5129     mode = buf64.st_mode;
5130     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5131       int n;
5132       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5133         *bytes = n;
5134         return 1;
5135       }
5136     }
5137   }
5138   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5139     return 0;
5140   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5141     return 0;
5142   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5143     return 0;
5144   }
5145   *bytes = end - cur;
5146   return 1;
5147 }
5148 
5149 // Map a block of memory.
5150 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5151                         char *addr, size_t bytes, bool read_only,
5152                         bool allow_exec) {
5153   int prot;
5154   int flags = MAP_PRIVATE;
5155 
5156   if (read_only) {
5157     prot = PROT_READ;
5158   } else {
5159     prot = PROT_READ | PROT_WRITE;
5160   }
5161 
5162   if (allow_exec) {
5163     prot |= PROT_EXEC;
5164   }
5165 
5166   if (addr != NULL) {
5167     flags |= MAP_FIXED;
5168   }
5169 
5170   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5171                                      fd, file_offset);
5172   if (mapped_address == MAP_FAILED) {
5173     return NULL;
5174   }
5175   return mapped_address;
5176 }
5177 
5178 
5179 // Remap a block of memory.
5180 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5181                           char *addr, size_t bytes, bool read_only,
5182                           bool allow_exec) {
5183   // same as map_memory() on this OS
5184   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5185                         allow_exec);
5186 }
5187 
5188 
5189 // Unmap a block of memory.
5190 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5191   return munmap(addr, bytes) == 0;
5192 }
5193 
5194 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5195 
5196 static clockid_t thread_cpu_clockid(Thread* thread) {
5197   pthread_t tid = thread->osthread()->pthread_id();
5198   clockid_t clockid;
5199 
5200   // Get thread clockid
5201   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5202   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5203   return clockid;
5204 }
5205 
5206 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5207 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5208 // of a thread.
5209 //
5210 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5211 // the fast estimate available on the platform.
5212 
5213 jlong os::current_thread_cpu_time() {
5214   if (os::Linux::supports_fast_thread_cpu_time()) {
5215     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5216   } else {
5217     // return user + sys since the cost is the same
5218     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5219   }
5220 }
5221 
5222 jlong os::thread_cpu_time(Thread* thread) {
5223   // consistent with what current_thread_cpu_time() returns
5224   if (os::Linux::supports_fast_thread_cpu_time()) {
5225     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5226   } else {
5227     return slow_thread_cpu_time(thread, true /* user + sys */);
5228   }
5229 }
5230 
5231 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5232   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5233     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5234   } else {
5235     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5236   }
5237 }
5238 
5239 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5240   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5241     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5242   } else {
5243     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5244   }
5245 }
5246 
5247 //  -1 on error.
5248 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5249   pid_t  tid = thread->osthread()->thread_id();
5250   char *s;
5251   char stat[2048];
5252   int statlen;
5253   char proc_name[64];
5254   int count;
5255   long sys_time, user_time;
5256   char cdummy;
5257   int idummy;
5258   long ldummy;
5259   FILE *fp;
5260 
5261   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5262   fp = fopen(proc_name, "r");
5263   if (fp == NULL) return -1;
5264   statlen = fread(stat, 1, 2047, fp);
5265   stat[statlen] = '\0';
5266   fclose(fp);
5267 
5268   // Skip pid and the command string. Note that we could be dealing with
5269   // weird command names, e.g. user could decide to rename java launcher
5270   // to "java 1.4.2 :)", then the stat file would look like
5271   //                1234 (java 1.4.2 :)) R ... ...
5272   // We don't really need to know the command string, just find the last
5273   // occurrence of ")" and then start parsing from there. See bug 4726580.
5274   s = strrchr(stat, ')');
5275   if (s == NULL) return -1;
5276 
5277   // Skip blank chars
5278   do { s++; } while (s && isspace(*s));
5279 
5280   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5281                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5282                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5283                  &user_time, &sys_time);
5284   if (count != 13) return -1;
5285   if (user_sys_cpu_time) {
5286     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5287   } else {
5288     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5289   }
5290 }
5291 
5292 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5293   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5294   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5295   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5296   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5297 }
5298 
5299 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5300   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5301   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5302   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5303   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5304 }
5305 
5306 bool os::is_thread_cpu_time_supported() {
5307   return true;
5308 }
5309 
5310 // System loadavg support.  Returns -1 if load average cannot be obtained.
5311 // Linux doesn't yet have a (official) notion of processor sets,
5312 // so just return the system wide load average.
5313 int os::loadavg(double loadavg[], int nelem) {
5314   return ::getloadavg(loadavg, nelem);
5315 }
5316 
5317 void os::pause() {
5318   char filename[MAX_PATH];
5319   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5320     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5321   } else {
5322     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5323   }
5324 
5325   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5326   if (fd != -1) {
5327     struct stat buf;
5328     ::close(fd);
5329     while (::stat(filename, &buf) == 0) {
5330       (void)::poll(NULL, 0, 100);
5331     }
5332   } else {
5333     jio_fprintf(stderr,
5334                 "Could not open pause file '%s', continuing immediately.\n", filename);
5335   }
5336 }
5337 
5338 
5339 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5340 // comment paragraphs are worth repeating here:
5341 //
5342 // Assumption:
5343 //    Only one parker can exist on an event, which is why we allocate
5344 //    them per-thread. Multiple unparkers can coexist.
5345 //
5346 // _Event serves as a restricted-range semaphore.
5347 //   -1 : thread is blocked, i.e. there is a waiter
5348 //    0 : neutral: thread is running or ready,
5349 //        could have been signaled after a wait started
5350 //    1 : signaled - thread is running or ready
5351 //
5352 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5353 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5354 // For specifics regarding the bug see GLIBC BUGID 261237 :
5355 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5356 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5357 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5358 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5359 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5360 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5361 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5362 // of libpthread avoids the problem, but isn't practical.
5363 //
5364 // Possible remedies:
5365 //
5366 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5367 //      This is palliative and probabilistic, however.  If the thread is preempted
5368 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5369 //      than the minimum period may have passed, and the abstime may be stale (in the
5370 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5371 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5372 //
5373 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5374 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5375 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5376 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5377 //      thread.
5378 //
5379 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5380 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5381 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5382 //      This also works well.  In fact it avoids kernel-level scalability impediments
5383 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5384 //      timers in a graceful fashion.
5385 //
5386 // 4.   When the abstime value is in the past it appears that control returns
5387 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5388 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5389 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5390 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5391 //      It may be possible to avoid reinitialization by checking the return
5392 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5393 //      condvar we must establish the invariant that cond_signal() is only called
5394 //      within critical sections protected by the adjunct mutex.  This prevents
5395 //      cond_signal() from "seeing" a condvar that's in the midst of being
5396 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5397 //      desirable signal-after-unlock optimization that avoids futile context switching.
5398 //
5399 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5400 //      structure when a condvar is used or initialized.  cond_destroy()  would
5401 //      release the helper structure.  Our reinitialize-after-timedwait fix
5402 //      put excessive stress on malloc/free and locks protecting the c-heap.
5403 //
5404 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5405 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5406 // and only enabling the work-around for vulnerable environments.
5407 
5408 // utility to compute the abstime argument to timedwait:
5409 // millis is the relative timeout time
5410 // abstime will be the absolute timeout time
5411 // TODO: replace compute_abstime() with unpackTime()
5412 
5413 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5414   if (millis < 0)  millis = 0;
5415 
5416   jlong seconds = millis / 1000;
5417   millis %= 1000;
5418   if (seconds > 50000000) { // see man cond_timedwait(3T)
5419     seconds = 50000000;
5420   }
5421 
5422   if (os::supports_monotonic_clock()) {
5423     struct timespec now;
5424     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5425     assert_status(status == 0, status, "clock_gettime");
5426     abstime->tv_sec = now.tv_sec  + seconds;
5427     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5428     if (nanos >= NANOSECS_PER_SEC) {
5429       abstime->tv_sec += 1;
5430       nanos -= NANOSECS_PER_SEC;
5431     }
5432     abstime->tv_nsec = nanos;
5433   } else {
5434     struct timeval now;
5435     int status = gettimeofday(&now, NULL);
5436     assert(status == 0, "gettimeofday");
5437     abstime->tv_sec = now.tv_sec  + seconds;
5438     long usec = now.tv_usec + millis * 1000;
5439     if (usec >= 1000000) {
5440       abstime->tv_sec += 1;
5441       usec -= 1000000;
5442     }
5443     abstime->tv_nsec = usec * 1000;
5444   }
5445   return abstime;
5446 }
5447 
5448 void os::PlatformEvent::park() {       // AKA "down()"
5449   // Transitions for _Event:
5450   //   -1 => -1 : illegal
5451   //    1 =>  0 : pass - return immediately
5452   //    0 => -1 : block; then set _Event to 0 before returning
5453 
5454   // Invariant: Only the thread associated with the Event/PlatformEvent
5455   // may call park().
5456   // TODO: assert that _Assoc != NULL or _Assoc == Self
5457   assert(_nParked == 0, "invariant");
5458 
5459   int v;
5460   for (;;) {
5461     v = _Event;
5462     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5463   }
5464   guarantee(v >= 0, "invariant");
5465   if (v == 0) {
5466     // Do this the hard way by blocking ...
5467     int status = pthread_mutex_lock(_mutex);
5468     assert_status(status == 0, status, "mutex_lock");
5469     guarantee(_nParked == 0, "invariant");
5470     ++_nParked;
5471     while (_Event < 0) {
5472       status = pthread_cond_wait(_cond, _mutex);
5473       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5474       // Treat this the same as if the wait was interrupted
5475       if (status == ETIME) { status = EINTR; }
5476       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5477     }
5478     --_nParked;
5479 
5480     _Event = 0;
5481     status = pthread_mutex_unlock(_mutex);
5482     assert_status(status == 0, status, "mutex_unlock");
5483     // Paranoia to ensure our locked and lock-free paths interact
5484     // correctly with each other.
5485     OrderAccess::fence();
5486   }
5487   guarantee(_Event >= 0, "invariant");
5488 }
5489 
5490 int os::PlatformEvent::park(jlong millis) {
5491   // Transitions for _Event:
5492   //   -1 => -1 : illegal
5493   //    1 =>  0 : pass - return immediately
5494   //    0 => -1 : block; then set _Event to 0 before returning
5495 
5496   guarantee(_nParked == 0, "invariant");
5497 
5498   int v;
5499   for (;;) {
5500     v = _Event;
5501     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5502   }
5503   guarantee(v >= 0, "invariant");
5504   if (v != 0) return OS_OK;
5505 
5506   // We do this the hard way, by blocking the thread.
5507   // Consider enforcing a minimum timeout value.
5508   struct timespec abst;
5509   compute_abstime(&abst, millis);
5510 
5511   int ret = OS_TIMEOUT;
5512   int status = pthread_mutex_lock(_mutex);
5513   assert_status(status == 0, status, "mutex_lock");
5514   guarantee(_nParked == 0, "invariant");
5515   ++_nParked;
5516 
5517   // Object.wait(timo) will return because of
5518   // (a) notification
5519   // (b) timeout
5520   // (c) thread.interrupt
5521   //
5522   // Thread.interrupt and object.notify{All} both call Event::set.
5523   // That is, we treat thread.interrupt as a special case of notification.
5524   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5525   // We assume all ETIME returns are valid.
5526   //
5527   // TODO: properly differentiate simultaneous notify+interrupt.
5528   // In that case, we should propagate the notify to another waiter.
5529 
5530   while (_Event < 0) {
5531     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5532     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5533       pthread_cond_destroy(_cond);
5534       pthread_cond_init(_cond, os::Linux::condAttr());
5535     }
5536     assert_status(status == 0 || status == EINTR ||
5537                   status == ETIME || status == ETIMEDOUT,
5538                   status, "cond_timedwait");
5539     if (!FilterSpuriousWakeups) break;                 // previous semantics
5540     if (status == ETIME || status == ETIMEDOUT) break;
5541     // We consume and ignore EINTR and spurious wakeups.
5542   }
5543   --_nParked;
5544   if (_Event >= 0) {
5545     ret = OS_OK;
5546   }
5547   _Event = 0;
5548   status = pthread_mutex_unlock(_mutex);
5549   assert_status(status == 0, status, "mutex_unlock");
5550   assert(_nParked == 0, "invariant");
5551   // Paranoia to ensure our locked and lock-free paths interact
5552   // correctly with each other.
5553   OrderAccess::fence();
5554   return ret;
5555 }
5556 
5557 void os::PlatformEvent::unpark() {
5558   // Transitions for _Event:
5559   //    0 => 1 : just return
5560   //    1 => 1 : just return
5561   //   -1 => either 0 or 1; must signal target thread
5562   //         That is, we can safely transition _Event from -1 to either
5563   //         0 or 1.
5564   // See also: "Semaphores in Plan 9" by Mullender & Cox
5565   //
5566   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5567   // that it will take two back-to-back park() calls for the owning
5568   // thread to block. This has the benefit of forcing a spurious return
5569   // from the first park() call after an unpark() call which will help
5570   // shake out uses of park() and unpark() without condition variables.
5571 
5572   if (Atomic::xchg(1, &_Event) >= 0) return;
5573 
5574   // Wait for the thread associated with the event to vacate
5575   int status = pthread_mutex_lock(_mutex);
5576   assert_status(status == 0, status, "mutex_lock");
5577   int AnyWaiters = _nParked;
5578   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5579   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5580     AnyWaiters = 0;
5581     pthread_cond_signal(_cond);
5582   }
5583   status = pthread_mutex_unlock(_mutex);
5584   assert_status(status == 0, status, "mutex_unlock");
5585   if (AnyWaiters != 0) {
5586     // Note that we signal() *after* dropping the lock for "immortal" Events.
5587     // This is safe and avoids a common class of  futile wakeups.  In rare
5588     // circumstances this can cause a thread to return prematurely from
5589     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5590     // will simply re-test the condition and re-park itself.
5591     // This provides particular benefit if the underlying platform does not
5592     // provide wait morphing.
5593     status = pthread_cond_signal(_cond);
5594     assert_status(status == 0, status, "cond_signal");
5595   }
5596 }
5597 
5598 
5599 // JSR166
5600 // -------------------------------------------------------
5601 
5602 // The solaris and linux implementations of park/unpark are fairly
5603 // conservative for now, but can be improved. They currently use a
5604 // mutex/condvar pair, plus a a count.
5605 // Park decrements count if > 0, else does a condvar wait.  Unpark
5606 // sets count to 1 and signals condvar.  Only one thread ever waits
5607 // on the condvar. Contention seen when trying to park implies that someone
5608 // is unparking you, so don't wait. And spurious returns are fine, so there
5609 // is no need to track notifications.
5610 
5611 // This code is common to linux and solaris and will be moved to a
5612 // common place in dolphin.
5613 //
5614 // The passed in time value is either a relative time in nanoseconds
5615 // or an absolute time in milliseconds. Either way it has to be unpacked
5616 // into suitable seconds and nanoseconds components and stored in the
5617 // given timespec structure.
5618 // Given time is a 64-bit value and the time_t used in the timespec is only
5619 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5620 // overflow if times way in the future are given. Further on Solaris versions
5621 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5622 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5623 // As it will be 28 years before "now + 100000000" will overflow we can
5624 // ignore overflow and just impose a hard-limit on seconds using the value
5625 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5626 // years from "now".
5627 
5628 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5629   assert(time > 0, "convertTime");
5630   time_t max_secs = 0;
5631 
5632   if (!os::supports_monotonic_clock() || isAbsolute) {
5633     struct timeval now;
5634     int status = gettimeofday(&now, NULL);
5635     assert(status == 0, "gettimeofday");
5636 
5637     max_secs = now.tv_sec + MAX_SECS;
5638 
5639     if (isAbsolute) {
5640       jlong secs = time / 1000;
5641       if (secs > max_secs) {
5642         absTime->tv_sec = max_secs;
5643       } else {
5644         absTime->tv_sec = secs;
5645       }
5646       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5647     } else {
5648       jlong secs = time / NANOSECS_PER_SEC;
5649       if (secs >= MAX_SECS) {
5650         absTime->tv_sec = max_secs;
5651         absTime->tv_nsec = 0;
5652       } else {
5653         absTime->tv_sec = now.tv_sec + secs;
5654         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5655         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5656           absTime->tv_nsec -= NANOSECS_PER_SEC;
5657           ++absTime->tv_sec; // note: this must be <= max_secs
5658         }
5659       }
5660     }
5661   } else {
5662     // must be relative using monotonic clock
5663     struct timespec now;
5664     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5665     assert_status(status == 0, status, "clock_gettime");
5666     max_secs = now.tv_sec + MAX_SECS;
5667     jlong secs = time / NANOSECS_PER_SEC;
5668     if (secs >= MAX_SECS) {
5669       absTime->tv_sec = max_secs;
5670       absTime->tv_nsec = 0;
5671     } else {
5672       absTime->tv_sec = now.tv_sec + secs;
5673       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5674       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5675         absTime->tv_nsec -= NANOSECS_PER_SEC;
5676         ++absTime->tv_sec; // note: this must be <= max_secs
5677       }
5678     }
5679   }
5680   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5681   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5682   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5683   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5684 }
5685 
5686 void Parker::park(bool isAbsolute, jlong time) {
5687   // Ideally we'd do something useful while spinning, such
5688   // as calling unpackTime().
5689 
5690   // Optional fast-path check:
5691   // Return immediately if a permit is available.
5692   // We depend on Atomic::xchg() having full barrier semantics
5693   // since we are doing a lock-free update to _counter.
5694   if (Atomic::xchg(0, &_counter) > 0) return;
5695 
5696   Thread* thread = Thread::current();
5697   assert(thread->is_Java_thread(), "Must be JavaThread");
5698   JavaThread *jt = (JavaThread *)thread;
5699 
5700   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5701   // Check interrupt before trying to wait
5702   if (Thread::is_interrupted(thread, false)) {
5703     return;
5704   }
5705 
5706   // Next, demultiplex/decode time arguments
5707   timespec absTime;
5708   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5709     return;
5710   }
5711   if (time > 0) {
5712     unpackTime(&absTime, isAbsolute, time);
5713   }
5714 
5715 
5716   // Enter safepoint region
5717   // Beware of deadlocks such as 6317397.
5718   // The per-thread Parker:: mutex is a classic leaf-lock.
5719   // In particular a thread must never block on the Threads_lock while
5720   // holding the Parker:: mutex.  If safepoints are pending both the
5721   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5722   ThreadBlockInVM tbivm(jt);
5723 
5724   // Don't wait if cannot get lock since interference arises from
5725   // unblocking.  Also. check interrupt before trying wait
5726   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5727     return;
5728   }
5729 
5730   int status;
5731   if (_counter > 0)  { // no wait needed
5732     _counter = 0;
5733     status = pthread_mutex_unlock(_mutex);
5734     assert(status == 0, "invariant");
5735     // Paranoia to ensure our locked and lock-free paths interact
5736     // correctly with each other and Java-level accesses.
5737     OrderAccess::fence();
5738     return;
5739   }
5740 
5741 #ifdef ASSERT
5742   // Don't catch signals while blocked; let the running threads have the signals.
5743   // (This allows a debugger to break into the running thread.)
5744   sigset_t oldsigs;
5745   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5746   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5747 #endif
5748 
5749   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5750   jt->set_suspend_equivalent();
5751   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5752 
5753   assert(_cur_index == -1, "invariant");
5754   if (time == 0) {
5755     _cur_index = REL_INDEX; // arbitrary choice when not timed
5756     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5757   } else {
5758     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5759     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5760     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5761       pthread_cond_destroy(&_cond[_cur_index]);
5762       pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5763     }
5764   }
5765   _cur_index = -1;
5766   assert_status(status == 0 || status == EINTR ||
5767                 status == ETIME || status == ETIMEDOUT,
5768                 status, "cond_timedwait");
5769 
5770 #ifdef ASSERT
5771   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5772 #endif
5773 
5774   _counter = 0;
5775   status = pthread_mutex_unlock(_mutex);
5776   assert_status(status == 0, status, "invariant");
5777   // Paranoia to ensure our locked and lock-free paths interact
5778   // correctly with each other and Java-level accesses.
5779   OrderAccess::fence();
5780 
5781   // If externally suspended while waiting, re-suspend
5782   if (jt->handle_special_suspend_equivalent_condition()) {
5783     jt->java_suspend_self();
5784   }
5785 }
5786 
5787 void Parker::unpark() {
5788   int status = pthread_mutex_lock(_mutex);
5789   assert(status == 0, "invariant");
5790   const int s = _counter;
5791   _counter = 1;
5792   if (s < 1) {
5793     // thread might be parked
5794     if (_cur_index != -1) {
5795       // thread is definitely parked
5796       if (WorkAroundNPTLTimedWaitHang) {
5797         status = pthread_cond_signal(&_cond[_cur_index]);
5798         assert(status == 0, "invariant");
5799         status = pthread_mutex_unlock(_mutex);
5800         assert(status == 0, "invariant");
5801       } else {
5802         // must capture correct index before unlocking
5803         int index = _cur_index;
5804         status = pthread_mutex_unlock(_mutex);
5805         assert(status == 0, "invariant");
5806         status = pthread_cond_signal(&_cond[index]);
5807         assert(status == 0, "invariant");
5808       }
5809     } else {
5810       pthread_mutex_unlock(_mutex);
5811       assert(status == 0, "invariant");
5812     }
5813   } else {
5814     pthread_mutex_unlock(_mutex);
5815     assert(status == 0, "invariant");
5816   }
5817 }
5818 
5819 
5820 extern char** environ;
5821 
5822 // Run the specified command in a separate process. Return its exit value,
5823 // or -1 on failure (e.g. can't fork a new process).
5824 // Unlike system(), this function can be called from signal handler. It
5825 // doesn't block SIGINT et al.
5826 int os::fork_and_exec(char* cmd) {
5827   const char * argv[4] = {"sh", "-c", cmd, NULL};
5828 
5829   pid_t pid = fork();
5830 
5831   if (pid < 0) {
5832     // fork failed
5833     return -1;
5834 
5835   } else if (pid == 0) {
5836     // child process
5837 
5838     execve("/bin/sh", (char* const*)argv, environ);
5839 
5840     // execve failed
5841     _exit(-1);
5842 
5843   } else  {
5844     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5845     // care about the actual exit code, for now.
5846 
5847     int status;
5848 
5849     // Wait for the child process to exit.  This returns immediately if
5850     // the child has already exited. */
5851     while (waitpid(pid, &status, 0) < 0) {
5852       switch (errno) {
5853       case ECHILD: return 0;
5854       case EINTR: break;
5855       default: return -1;
5856       }
5857     }
5858 
5859     if (WIFEXITED(status)) {
5860       // The child exited normally; get its exit code.
5861       return WEXITSTATUS(status);
5862     } else if (WIFSIGNALED(status)) {
5863       // The child exited because of a signal
5864       // The best value to return is 0x80 + signal number,
5865       // because that is what all Unix shells do, and because
5866       // it allows callers to distinguish between process exit and
5867       // process death by signal.
5868       return 0x80 + WTERMSIG(status);
5869     } else {
5870       // Unknown exit code; pass it through
5871       return status;
5872     }
5873   }
5874 }
5875 
5876 // is_headless_jre()
5877 //
5878 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5879 // in order to report if we are running in a headless jre
5880 //
5881 // Since JDK8 xawt/libmawt.so was moved into the same directory
5882 // as libawt.so, and renamed libawt_xawt.so
5883 //
5884 bool os::is_headless_jre() {
5885   struct stat statbuf;
5886   char buf[MAXPATHLEN];
5887   char libmawtpath[MAXPATHLEN];
5888   const char *xawtstr  = "/xawt/libmawt.so";
5889   const char *new_xawtstr = "/libawt_xawt.so";
5890   char *p;
5891 
5892   // Get path to libjvm.so
5893   os::jvm_path(buf, sizeof(buf));
5894 
5895   // Get rid of libjvm.so
5896   p = strrchr(buf, '/');
5897   if (p == NULL) {
5898     return false;
5899   } else {
5900     *p = '\0';
5901   }
5902 
5903   // Get rid of client or server
5904   p = strrchr(buf, '/');
5905   if (p == NULL) {
5906     return false;
5907   } else {
5908     *p = '\0';
5909   }
5910 
5911   // check xawt/libmawt.so
5912   strcpy(libmawtpath, buf);
5913   strcat(libmawtpath, xawtstr);
5914   if (::stat(libmawtpath, &statbuf) == 0) return false;
5915 
5916   // check libawt_xawt.so
5917   strcpy(libmawtpath, buf);
5918   strcat(libmawtpath, new_xawtstr);
5919   if (::stat(libmawtpath, &statbuf) == 0) return false;
5920 
5921   return true;
5922 }
5923 
5924 // Get the default path to the core file
5925 // Returns the length of the string
5926 int os::get_core_path(char* buffer, size_t bufferSize) {
5927   /*
5928    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5929    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5930    */
5931   const int core_pattern_len = 129;
5932   char core_pattern[core_pattern_len] = {0};
5933 
5934   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5935   if (core_pattern_file == -1) {
5936     return -1;
5937   }
5938 
5939   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5940   ::close(core_pattern_file);
5941   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5942     return -1;
5943   }
5944   if (core_pattern[ret-1] == '\n') {
5945     core_pattern[ret-1] = '\0';
5946   } else {
5947     core_pattern[ret] = '\0';
5948   }
5949 
5950   char *pid_pos = strstr(core_pattern, "%p");
5951   int written;
5952 
5953   if (core_pattern[0] == '/') {
5954     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5955   } else {
5956     char cwd[PATH_MAX];
5957 
5958     const char* p = get_current_directory(cwd, PATH_MAX);
5959     if (p == NULL) {
5960       return -1;
5961     }
5962 
5963     if (core_pattern[0] == '|') {
5964       written = jio_snprintf(buffer, bufferSize,
5965                         "\"%s\" (or dumping to %s/core.%d)",
5966                                      &core_pattern[1], p, current_process_id());
5967     } else {
5968       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5969     }
5970   }
5971 
5972   if (written < 0) {
5973     return -1;
5974   }
5975 
5976   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5977     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5978 
5979     if (core_uses_pid_file != -1) {
5980       char core_uses_pid = 0;
5981       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5982       ::close(core_uses_pid_file);
5983 
5984       if (core_uses_pid == '1') {
5985         jio_snprintf(buffer + written, bufferSize - written,
5986                                           ".%d", current_process_id());
5987       }
5988     }
5989   }
5990 
5991   return strlen(buffer);
5992 }
5993 
5994 bool os::start_debugging(char *buf, int buflen) {
5995   int len = (int)strlen(buf);
5996   char *p = &buf[len];
5997 
5998   jio_snprintf(p, buflen-len,
5999              "\n\n"
6000              "Do you want to debug the problem?\n\n"
6001              "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6002              "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6003              "Otherwise, press RETURN to abort...",
6004              os::current_process_id(), os::current_process_id(),
6005              os::current_thread_id(), os::current_thread_id());
6006 
6007   bool yes = os::message_box("Unexpected Error", buf);
6008 
6009   if (yes) {
6010     // yes, user asked VM to launch debugger
6011     jio_snprintf(buf, sizeof(buf), "gdb /proc/%d/exe %d",
6012                      os::current_process_id(), os::current_process_id());
6013 
6014     os::fork_and_exec(buf);
6015     yes = false;
6016   }
6017   return yes;
6018 }
6019 
6020 
6021 
6022 /////////////// Unit tests ///////////////
6023 
6024 #ifndef PRODUCT
6025 
6026 #define test_log(...)              \
6027   do {                             \
6028     if (VerboseInternalVMTests) {  \
6029       tty->print_cr(__VA_ARGS__);  \
6030       tty->flush();                \
6031     }                              \
6032   } while (false)
6033 
6034 class TestReserveMemorySpecial : AllStatic {
6035  public:
6036   static void small_page_write(void* addr, size_t size) {
6037     size_t page_size = os::vm_page_size();
6038 
6039     char* end = (char*)addr + size;
6040     for (char* p = (char*)addr; p < end; p += page_size) {
6041       *p = 1;
6042     }
6043   }
6044 
6045   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6046     if (!UseHugeTLBFS) {
6047       return;
6048     }
6049 
6050     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6051 
6052     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6053 
6054     if (addr != NULL) {
6055       small_page_write(addr, size);
6056 
6057       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6058     }
6059   }
6060 
6061   static void test_reserve_memory_special_huge_tlbfs_only() {
6062     if (!UseHugeTLBFS) {
6063       return;
6064     }
6065 
6066     size_t lp = os::large_page_size();
6067 
6068     for (size_t size = lp; size <= lp * 10; size += lp) {
6069       test_reserve_memory_special_huge_tlbfs_only(size);
6070     }
6071   }
6072 
6073   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6074     size_t lp = os::large_page_size();
6075     size_t ag = os::vm_allocation_granularity();
6076 
6077     // sizes to test
6078     const size_t sizes[] = {
6079       lp, lp + ag, lp + lp / 2, lp * 2,
6080       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6081       lp * 10, lp * 10 + lp / 2
6082     };
6083     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6084 
6085     // For each size/alignment combination, we test three scenarios:
6086     // 1) with req_addr == NULL
6087     // 2) with a non-null req_addr at which we expect to successfully allocate
6088     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6089     //    expect the allocation to either fail or to ignore req_addr
6090 
6091     // Pre-allocate two areas; they shall be as large as the largest allocation
6092     //  and aligned to the largest alignment we will be testing.
6093     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6094     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6095       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6096       -1, 0);
6097     assert(mapping1 != MAP_FAILED, "should work");
6098 
6099     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6100       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6101       -1, 0);
6102     assert(mapping2 != MAP_FAILED, "should work");
6103 
6104     // Unmap the first mapping, but leave the second mapping intact: the first
6105     // mapping will serve as a value for a "good" req_addr (case 2). The second
6106     // mapping, still intact, as "bad" req_addr (case 3).
6107     ::munmap(mapping1, mapping_size);
6108 
6109     // Case 1
6110     test_log("%s, req_addr NULL:", __FUNCTION__);
6111     test_log("size            align           result");
6112 
6113     for (int i = 0; i < num_sizes; i++) {
6114       const size_t size = sizes[i];
6115       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6116         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6117         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6118                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6119         if (p != NULL) {
6120           assert(is_ptr_aligned(p, alignment), "must be");
6121           small_page_write(p, size);
6122           os::Linux::release_memory_special_huge_tlbfs(p, size);
6123         }
6124       }
6125     }
6126 
6127     // Case 2
6128     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6129     test_log("size            align           req_addr         result");
6130 
6131     for (int i = 0; i < num_sizes; i++) {
6132       const size_t size = sizes[i];
6133       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6134         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6135         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6136         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6137                  size, alignment, p2i(req_addr), p2i(p),
6138                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6139         if (p != NULL) {
6140           assert(p == req_addr, "must be");
6141           small_page_write(p, size);
6142           os::Linux::release_memory_special_huge_tlbfs(p, size);
6143         }
6144       }
6145     }
6146 
6147     // Case 3
6148     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6149     test_log("size            align           req_addr         result");
6150 
6151     for (int i = 0; i < num_sizes; i++) {
6152       const size_t size = sizes[i];
6153       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6154         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6155         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6156         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6157                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6158         // as the area around req_addr contains already existing mappings, the API should always
6159         // return NULL (as per contract, it cannot return another address)
6160         assert(p == NULL, "must be");
6161       }
6162     }
6163 
6164     ::munmap(mapping2, mapping_size);
6165 
6166   }
6167 
6168   static void test_reserve_memory_special_huge_tlbfs() {
6169     if (!UseHugeTLBFS) {
6170       return;
6171     }
6172 
6173     test_reserve_memory_special_huge_tlbfs_only();
6174     test_reserve_memory_special_huge_tlbfs_mixed();
6175   }
6176 
6177   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6178     if (!UseSHM) {
6179       return;
6180     }
6181 
6182     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6183 
6184     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6185 
6186     if (addr != NULL) {
6187       assert(is_ptr_aligned(addr, alignment), "Check");
6188       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6189 
6190       small_page_write(addr, size);
6191 
6192       os::Linux::release_memory_special_shm(addr, size);
6193     }
6194   }
6195 
6196   static void test_reserve_memory_special_shm() {
6197     size_t lp = os::large_page_size();
6198     size_t ag = os::vm_allocation_granularity();
6199 
6200     for (size_t size = ag; size < lp * 3; size += ag) {
6201       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6202         test_reserve_memory_special_shm(size, alignment);
6203       }
6204     }
6205   }
6206 
6207   static void test() {
6208     test_reserve_memory_special_huge_tlbfs();
6209     test_reserve_memory_special_shm();
6210   }
6211 };
6212 
6213 void TestReserveMemorySpecial_test() {
6214   TestReserveMemorySpecial::test();
6215 }
6216 
6217 #endif