1 /*
   2  * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "memory/allocation.inline.hpp"
  36 #include "memory/filemap.hpp"
  37 #include "mutex_linux.inline.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_linux.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "prims/jniFastGetField.hpp"
  42 #include "prims/jvm.h"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.inline.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.hpp"
  49 #include "runtime/init.hpp"
  50 #include "runtime/java.hpp"
  51 #include "runtime/javaCalls.hpp"
  52 #include "runtime/mutexLocker.hpp"
  53 #include "runtime/objectMonitor.hpp"
  54 #include "runtime/orderAccess.inline.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/perfMemory.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/statSampler.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "runtime/thread.inline.hpp"
  61 #include "runtime/threadCritical.hpp"
  62 #include "runtime/timer.hpp"
  63 #include "semaphore_posix.hpp"
  64 #include "services/attachListener.hpp"
  65 #include "services/memTracker.hpp"
  66 #include "services/runtimeService.hpp"
  67 #include "utilities/decoder.hpp"
  68 #include "utilities/defaultStream.hpp"
  69 #include "utilities/events.hpp"
  70 #include "utilities/elfFile.hpp"
  71 #include "utilities/growableArray.hpp"
  72 #include "utilities/macros.hpp"
  73 #include "utilities/vmError.hpp"
  74 
  75 // put OS-includes here
  76 # include <sys/types.h>
  77 # include <sys/mman.h>
  78 # include <sys/stat.h>
  79 # include <sys/select.h>
  80 # include <pthread.h>
  81 # include <signal.h>
  82 # include <errno.h>
  83 # include <dlfcn.h>
  84 # include <stdio.h>
  85 # include <unistd.h>
  86 # include <sys/resource.h>
  87 # include <pthread.h>
  88 # include <sys/stat.h>
  89 # include <sys/time.h>
  90 # include <sys/times.h>
  91 # include <sys/utsname.h>
  92 # include <sys/socket.h>
  93 # include <sys/wait.h>
  94 # include <pwd.h>
  95 # include <poll.h>
  96 # include <semaphore.h>
  97 # include <fcntl.h>
  98 # include <string.h>
  99 # include <syscall.h>
 100 # include <sys/sysinfo.h>
 101 # include <gnu/libc-version.h>
 102 # include <sys/ipc.h>
 103 # include <sys/shm.h>
 104 # include <link.h>
 105 # include <stdint.h>
 106 # include <inttypes.h>
 107 # include <sys/ioctl.h>
 108 
 109 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
 110 
 111 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 112 // getrusage() is prepared to handle the associated failure.
 113 #ifndef RUSAGE_THREAD
 114   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 115 #endif
 116 
 117 #define MAX_PATH    (2 * K)
 118 
 119 #define MAX_SECS 100000000
 120 
 121 // for timer info max values which include all bits
 122 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 123 
 124 #define LARGEPAGES_BIT (1 << 6)
 125 ////////////////////////////////////////////////////////////////////////////////
 126 // global variables
 127 julong os::Linux::_physical_memory = 0;
 128 
 129 address   os::Linux::_initial_thread_stack_bottom = NULL;
 130 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 131 
 132 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 133 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 134 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 135 Mutex* os::Linux::_createThread_lock = NULL;
 136 pthread_t os::Linux::_main_thread;
 137 int os::Linux::_page_size = -1;
 138 const int os::Linux::_vm_default_page_size = (8 * K);
 139 bool os::Linux::_supports_fast_thread_cpu_time = false;
 140 const char * os::Linux::_glibc_version = NULL;
 141 const char * os::Linux::_libpthread_version = NULL;
 142 pthread_condattr_t os::Linux::_condattr[1];
 143 
 144 static jlong initial_time_count=0;
 145 
 146 static int clock_tics_per_sec = 100;
 147 
 148 // For diagnostics to print a message once. see run_periodic_checks
 149 static sigset_t check_signal_done;
 150 static bool check_signals = true;
 151 
 152 // Signal number used to suspend/resume a thread
 153 
 154 // do not use any signal number less than SIGSEGV, see 4355769
 155 static int SR_signum = SIGUSR2;
 156 sigset_t SR_sigset;
 157 
 158 // Declarations
 159 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 160 
 161 // utility functions
 162 
 163 static int SR_initialize();
 164 
 165 julong os::available_memory() {
 166   return Linux::available_memory();
 167 }
 168 
 169 julong os::Linux::available_memory() {
 170   // values in struct sysinfo are "unsigned long"
 171   struct sysinfo si;
 172   sysinfo(&si);
 173 
 174   return (julong)si.freeram * si.mem_unit;
 175 }
 176 
 177 julong os::physical_memory() {
 178   return Linux::physical_memory();
 179 }
 180 
 181 // Return true if user is running as root.
 182 
 183 bool os::have_special_privileges() {
 184   static bool init = false;
 185   static bool privileges = false;
 186   if (!init) {
 187     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 188     init = true;
 189   }
 190   return privileges;
 191 }
 192 
 193 
 194 #ifndef SYS_gettid
 195 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 196   #ifdef __ia64__
 197     #define SYS_gettid 1105
 198   #else
 199     #ifdef __i386__
 200       #define SYS_gettid 224
 201     #else
 202       #ifdef __amd64__
 203         #define SYS_gettid 186
 204       #else
 205         #ifdef __sparc__
 206           #define SYS_gettid 143
 207         #else
 208           #error define gettid for the arch
 209         #endif
 210       #endif
 211     #endif
 212   #endif
 213 #endif
 214 
 215 // Cpu architecture string
 216 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 217 
 218 
 219 // pid_t gettid()
 220 //
 221 // Returns the kernel thread id of the currently running thread. Kernel
 222 // thread id is used to access /proc.
 223 pid_t os::Linux::gettid() {
 224   int rslt = syscall(SYS_gettid);
 225   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 226   return (pid_t)rslt;
 227 }
 228 
 229 // Most versions of linux have a bug where the number of processors are
 230 // determined by looking at the /proc file system.  In a chroot environment,
 231 // the system call returns 1.  This causes the VM to act as if it is
 232 // a single processor and elide locking (see is_MP() call).
 233 static bool unsafe_chroot_detected = false;
 234 static const char *unstable_chroot_error = "/proc file system not found.\n"
 235                      "Java may be unstable running multithreaded in a chroot "
 236                      "environment on Linux when /proc filesystem is not mounted.";
 237 
 238 void os::Linux::initialize_system_info() {
 239   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 240   if (processor_count() == 1) {
 241     pid_t pid = os::Linux::gettid();
 242     char fname[32];
 243     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 244     FILE *fp = fopen(fname, "r");
 245     if (fp == NULL) {
 246       unsafe_chroot_detected = true;
 247     } else {
 248       fclose(fp);
 249     }
 250   }
 251   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 252   assert(processor_count() > 0, "linux error");
 253 }
 254 
 255 void os::init_system_properties_values() {
 256   // The next steps are taken in the product version:
 257   //
 258   // Obtain the JAVA_HOME value from the location of libjvm.so.
 259   // This library should be located at:
 260   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 261   //
 262   // If "/jre/lib/" appears at the right place in the path, then we
 263   // assume libjvm.so is installed in a JDK and we use this path.
 264   //
 265   // Otherwise exit with message: "Could not create the Java virtual machine."
 266   //
 267   // The following extra steps are taken in the debugging version:
 268   //
 269   // If "/jre/lib/" does NOT appear at the right place in the path
 270   // instead of exit check for $JAVA_HOME environment variable.
 271   //
 272   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 273   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 274   // it looks like libjvm.so is installed there
 275   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 276   //
 277   // Otherwise exit.
 278   //
 279   // Important note: if the location of libjvm.so changes this
 280   // code needs to be changed accordingly.
 281 
 282   // See ld(1):
 283   //      The linker uses the following search paths to locate required
 284   //      shared libraries:
 285   //        1: ...
 286   //        ...
 287   //        7: The default directories, normally /lib and /usr/lib.
 288 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 289   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 290 #else
 291   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 292 #endif
 293 
 294 // Base path of extensions installed on the system.
 295 #define SYS_EXT_DIR     "/usr/java/packages"
 296 #define EXTENSIONS_DIR  "/lib/ext"
 297 
 298   // Buffer that fits several sprintfs.
 299   // Note that the space for the colon and the trailing null are provided
 300   // by the nulls included by the sizeof operator.
 301   const size_t bufsize =
 302     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 303          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 304   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 305 
 306   // sysclasspath, java_home, dll_dir
 307   {
 308     char *pslash;
 309     os::jvm_path(buf, bufsize);
 310 
 311     // Found the full path to libjvm.so.
 312     // Now cut the path to <java_home>/jre if we can.
 313     pslash = strrchr(buf, '/');
 314     if (pslash != NULL) {
 315       *pslash = '\0';            // Get rid of /libjvm.so.
 316     }
 317     pslash = strrchr(buf, '/');
 318     if (pslash != NULL) {
 319       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 320     }
 321     Arguments::set_dll_dir(buf);
 322 
 323     if (pslash != NULL) {
 324       pslash = strrchr(buf, '/');
 325       if (pslash != NULL) {
 326         *pslash = '\0';          // Get rid of /<arch>.
 327         pslash = strrchr(buf, '/');
 328         if (pslash != NULL) {
 329           *pslash = '\0';        // Get rid of /lib.
 330         }
 331       }
 332     }
 333     Arguments::set_java_home(buf);
 334     set_boot_path('/', ':');
 335   }
 336 
 337   // Where to look for native libraries.
 338   //
 339   // Note: Due to a legacy implementation, most of the library path
 340   // is set in the launcher. This was to accomodate linking restrictions
 341   // on legacy Linux implementations (which are no longer supported).
 342   // Eventually, all the library path setting will be done here.
 343   //
 344   // However, to prevent the proliferation of improperly built native
 345   // libraries, the new path component /usr/java/packages is added here.
 346   // Eventually, all the library path setting will be done here.
 347   {
 348     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 349     // should always exist (until the legacy problem cited above is
 350     // addressed).
 351     const char *v = ::getenv("LD_LIBRARY_PATH");
 352     const char *v_colon = ":";
 353     if (v == NULL) { v = ""; v_colon = ""; }
 354     // That's +1 for the colon and +1 for the trailing '\0'.
 355     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 356                                                      strlen(v) + 1 +
 357                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 358                                                      mtInternal);
 359     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 360     Arguments::set_library_path(ld_library_path);
 361     FREE_C_HEAP_ARRAY(char, ld_library_path);
 362   }
 363 
 364   // Extensions directories.
 365   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 366   Arguments::set_ext_dirs(buf);
 367 
 368   FREE_C_HEAP_ARRAY(char, buf);
 369 
 370 #undef DEFAULT_LIBPATH
 371 #undef SYS_EXT_DIR
 372 #undef EXTENSIONS_DIR
 373 }
 374 
 375 ////////////////////////////////////////////////////////////////////////////////
 376 // breakpoint support
 377 
 378 void os::breakpoint() {
 379   BREAKPOINT;
 380 }
 381 
 382 extern "C" void breakpoint() {
 383   // use debugger to set breakpoint here
 384 }
 385 
 386 ////////////////////////////////////////////////////////////////////////////////
 387 // signal support
 388 
 389 debug_only(static bool signal_sets_initialized = false);
 390 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 391 
 392 bool os::Linux::is_sig_ignored(int sig) {
 393   struct sigaction oact;
 394   sigaction(sig, (struct sigaction*)NULL, &oact);
 395   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 396                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 397   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 398     return true;
 399   } else {
 400     return false;
 401   }
 402 }
 403 
 404 void os::Linux::signal_sets_init() {
 405   // Should also have an assertion stating we are still single-threaded.
 406   assert(!signal_sets_initialized, "Already initialized");
 407   // Fill in signals that are necessarily unblocked for all threads in
 408   // the VM. Currently, we unblock the following signals:
 409   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 410   //                         by -Xrs (=ReduceSignalUsage));
 411   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 412   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 413   // the dispositions or masks wrt these signals.
 414   // Programs embedding the VM that want to use the above signals for their
 415   // own purposes must, at this time, use the "-Xrs" option to prevent
 416   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 417   // (See bug 4345157, and other related bugs).
 418   // In reality, though, unblocking these signals is really a nop, since
 419   // these signals are not blocked by default.
 420   sigemptyset(&unblocked_sigs);
 421   sigemptyset(&allowdebug_blocked_sigs);
 422   sigaddset(&unblocked_sigs, SIGILL);
 423   sigaddset(&unblocked_sigs, SIGSEGV);
 424   sigaddset(&unblocked_sigs, SIGBUS);
 425   sigaddset(&unblocked_sigs, SIGFPE);
 426 #if defined(PPC64)
 427   sigaddset(&unblocked_sigs, SIGTRAP);
 428 #endif
 429   sigaddset(&unblocked_sigs, SR_signum);
 430 
 431   if (!ReduceSignalUsage) {
 432     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 433       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 434       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 435     }
 436     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 437       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 438       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 439     }
 440     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 441       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 442       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 443     }
 444   }
 445   // Fill in signals that are blocked by all but the VM thread.
 446   sigemptyset(&vm_sigs);
 447   if (!ReduceSignalUsage) {
 448     sigaddset(&vm_sigs, BREAK_SIGNAL);
 449   }
 450   debug_only(signal_sets_initialized = true);
 451 
 452 }
 453 
 454 // These are signals that are unblocked while a thread is running Java.
 455 // (For some reason, they get blocked by default.)
 456 sigset_t* os::Linux::unblocked_signals() {
 457   assert(signal_sets_initialized, "Not initialized");
 458   return &unblocked_sigs;
 459 }
 460 
 461 // These are the signals that are blocked while a (non-VM) thread is
 462 // running Java. Only the VM thread handles these signals.
 463 sigset_t* os::Linux::vm_signals() {
 464   assert(signal_sets_initialized, "Not initialized");
 465   return &vm_sigs;
 466 }
 467 
 468 // These are signals that are blocked during cond_wait to allow debugger in
 469 sigset_t* os::Linux::allowdebug_blocked_signals() {
 470   assert(signal_sets_initialized, "Not initialized");
 471   return &allowdebug_blocked_sigs;
 472 }
 473 
 474 void os::Linux::hotspot_sigmask(Thread* thread) {
 475 
 476   //Save caller's signal mask before setting VM signal mask
 477   sigset_t caller_sigmask;
 478   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 479 
 480   OSThread* osthread = thread->osthread();
 481   osthread->set_caller_sigmask(caller_sigmask);
 482 
 483   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 484 
 485   if (!ReduceSignalUsage) {
 486     if (thread->is_VM_thread()) {
 487       // Only the VM thread handles BREAK_SIGNAL ...
 488       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 489     } else {
 490       // ... all other threads block BREAK_SIGNAL
 491       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 492     }
 493   }
 494 }
 495 
 496 //////////////////////////////////////////////////////////////////////////////
 497 // detecting pthread library
 498 
 499 void os::Linux::libpthread_init() {
 500   // Save glibc and pthread version strings.
 501 #if !defined(_CS_GNU_LIBC_VERSION) || \
 502     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 503   #error "glibc too old (< 2.3.2)"
 504 #endif
 505 
 506   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 507   assert(n > 0, "cannot retrieve glibc version");
 508   char *str = (char *)malloc(n, mtInternal);
 509   confstr(_CS_GNU_LIBC_VERSION, str, n);
 510   os::Linux::set_glibc_version(str);
 511 
 512   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 513   assert(n > 0, "cannot retrieve pthread version");
 514   str = (char *)malloc(n, mtInternal);
 515   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 516   os::Linux::set_libpthread_version(str);
 517 }
 518 
 519 /////////////////////////////////////////////////////////////////////////////
 520 // thread stack expansion
 521 
 522 // os::Linux::manually_expand_stack() takes care of expanding the thread
 523 // stack. Note that this is normally not needed: pthread stacks allocate
 524 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 525 // committed. Therefore it is not necessary to expand the stack manually.
 526 //
 527 // Manually expanding the stack was historically needed on LinuxThreads
 528 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 529 // it is kept to deal with very rare corner cases:
 530 //
 531 // For one, user may run the VM on an own implementation of threads
 532 // whose stacks are - like the old LinuxThreads - implemented using
 533 // mmap(MAP_GROWSDOWN).
 534 //
 535 // Also, this coding may be needed if the VM is running on the primordial
 536 // thread. Normally we avoid running on the primordial thread; however,
 537 // user may still invoke the VM on the primordial thread.
 538 //
 539 // The following historical comment describes the details about running
 540 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 541 
 542 
 543 // Force Linux kernel to expand current thread stack. If "bottom" is close
 544 // to the stack guard, caller should block all signals.
 545 //
 546 // MAP_GROWSDOWN:
 547 //   A special mmap() flag that is used to implement thread stacks. It tells
 548 //   kernel that the memory region should extend downwards when needed. This
 549 //   allows early versions of LinuxThreads to only mmap the first few pages
 550 //   when creating a new thread. Linux kernel will automatically expand thread
 551 //   stack as needed (on page faults).
 552 //
 553 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 554 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 555 //   region, it's hard to tell if the fault is due to a legitimate stack
 556 //   access or because of reading/writing non-exist memory (e.g. buffer
 557 //   overrun). As a rule, if the fault happens below current stack pointer,
 558 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 559 //   application (see Linux kernel fault.c).
 560 //
 561 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 562 //   stack overflow detection.
 563 //
 564 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 565 //   not use MAP_GROWSDOWN.
 566 //
 567 // To get around the problem and allow stack banging on Linux, we need to
 568 // manually expand thread stack after receiving the SIGSEGV.
 569 //
 570 // There are two ways to expand thread stack to address "bottom", we used
 571 // both of them in JVM before 1.5:
 572 //   1. adjust stack pointer first so that it is below "bottom", and then
 573 //      touch "bottom"
 574 //   2. mmap() the page in question
 575 //
 576 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 577 // if current sp is already near the lower end of page 101, and we need to
 578 // call mmap() to map page 100, it is possible that part of the mmap() frame
 579 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 580 // That will destroy the mmap() frame and cause VM to crash.
 581 //
 582 // The following code works by adjusting sp first, then accessing the "bottom"
 583 // page to force a page fault. Linux kernel will then automatically expand the
 584 // stack mapping.
 585 //
 586 // _expand_stack_to() assumes its frame size is less than page size, which
 587 // should always be true if the function is not inlined.
 588 
 589 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 590   #define NOINLINE
 591 #else
 592   #define NOINLINE __attribute__ ((noinline))
 593 #endif
 594 
 595 static void _expand_stack_to(address bottom) NOINLINE;
 596 
 597 static void _expand_stack_to(address bottom) {
 598   address sp;
 599   size_t size;
 600   volatile char *p;
 601 
 602   // Adjust bottom to point to the largest address within the same page, it
 603   // gives us a one-page buffer if alloca() allocates slightly more memory.
 604   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 605   bottom += os::Linux::page_size() - 1;
 606 
 607   // sp might be slightly above current stack pointer; if that's the case, we
 608   // will alloca() a little more space than necessary, which is OK. Don't use
 609   // os::current_stack_pointer(), as its result can be slightly below current
 610   // stack pointer, causing us to not alloca enough to reach "bottom".
 611   sp = (address)&sp;
 612 
 613   if (sp > bottom) {
 614     size = sp - bottom;
 615     p = (volatile char *)alloca(size);
 616     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 617     p[0] = '\0';
 618   }
 619 }
 620 
 621 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 622   assert(t!=NULL, "just checking");
 623   assert(t->osthread()->expanding_stack(), "expand should be set");
 624   assert(t->stack_base() != NULL, "stack_base was not initialized");
 625 
 626   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 627     sigset_t mask_all, old_sigset;
 628     sigfillset(&mask_all);
 629     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 630     _expand_stack_to(addr);
 631     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 632     return true;
 633   }
 634   return false;
 635 }
 636 
 637 //////////////////////////////////////////////////////////////////////////////
 638 // create new thread
 639 
 640 // Thread start routine for all newly created threads
 641 static void *java_start(Thread *thread) {
 642   // Try to randomize the cache line index of hot stack frames.
 643   // This helps when threads of the same stack traces evict each other's
 644   // cache lines. The threads can be either from the same JVM instance, or
 645   // from different JVM instances. The benefit is especially true for
 646   // processors with hyperthreading technology.
 647   static int counter = 0;
 648   int pid = os::current_process_id();
 649   alloca(((pid ^ counter++) & 7) * 128);
 650 
 651   ThreadLocalStorage::set_thread(thread);
 652 
 653   OSThread* osthread = thread->osthread();
 654   Monitor* sync = osthread->startThread_lock();
 655 
 656   osthread->set_thread_id(os::current_thread_id());
 657 
 658   if (UseNUMA) {
 659     int lgrp_id = os::numa_get_group_id();
 660     if (lgrp_id != -1) {
 661       thread->set_lgrp_id(lgrp_id);
 662     }
 663   }
 664   // initialize signal mask for this thread
 665   os::Linux::hotspot_sigmask(thread);
 666 
 667   // initialize floating point control register
 668   os::Linux::init_thread_fpu_state();
 669 
 670   // handshaking with parent thread
 671   {
 672     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 673 
 674     // notify parent thread
 675     osthread->set_state(INITIALIZED);
 676     sync->notify_all();
 677 
 678     // wait until os::start_thread()
 679     while (osthread->get_state() == INITIALIZED) {
 680       sync->wait(Mutex::_no_safepoint_check_flag);
 681     }
 682   }
 683 
 684   // call one more level start routine
 685   thread->run();
 686 
 687   return 0;
 688 }
 689 
 690 bool os::create_thread(Thread* thread, ThreadType thr_type,
 691                        size_t stack_size) {
 692   assert(thread->osthread() == NULL, "caller responsible");
 693 
 694   // Allocate the OSThread object
 695   OSThread* osthread = new OSThread(NULL, NULL);
 696   if (osthread == NULL) {
 697     return false;
 698   }
 699 
 700   // set the correct thread state
 701   osthread->set_thread_type(thr_type);
 702 
 703   // Initial state is ALLOCATED but not INITIALIZED
 704   osthread->set_state(ALLOCATED);
 705 
 706   thread->set_osthread(osthread);
 707 
 708   // init thread attributes
 709   pthread_attr_t attr;
 710   pthread_attr_init(&attr);
 711   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 712 
 713   // stack size
 714   // calculate stack size if it's not specified by caller
 715   if (stack_size == 0) {
 716     stack_size = os::Linux::default_stack_size(thr_type);
 717 
 718     switch (thr_type) {
 719     case os::java_thread:
 720       // Java threads use ThreadStackSize which default value can be
 721       // changed with the flag -Xss
 722       assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 723       stack_size = JavaThread::stack_size_at_create();
 724       break;
 725     case os::compiler_thread:
 726       if (CompilerThreadStackSize > 0) {
 727         stack_size = (size_t)(CompilerThreadStackSize * K);
 728         break;
 729       } // else fall through:
 730         // use VMThreadStackSize if CompilerThreadStackSize is not defined
 731     case os::vm_thread:
 732     case os::pgc_thread:
 733     case os::cgc_thread:
 734     case os::watcher_thread:
 735       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 736       break;
 737     }
 738   }
 739 
 740   stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 741   pthread_attr_setstacksize(&attr, stack_size);
 742 
 743   // glibc guard page
 744   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 745 
 746   ThreadState state;
 747 
 748   {
 749     pthread_t tid;
 750     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 751 
 752     pthread_attr_destroy(&attr);
 753 
 754     if (ret != 0) {
 755       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 756         perror("pthread_create()");
 757       }
 758       // Need to clean up stuff we've allocated so far
 759       thread->set_osthread(NULL);
 760       delete osthread;
 761       return false;
 762     }
 763 
 764     // Store pthread info into the OSThread
 765     osthread->set_pthread_id(tid);
 766 
 767     // Wait until child thread is either initialized or aborted
 768     {
 769       Monitor* sync_with_child = osthread->startThread_lock();
 770       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 771       while ((state = osthread->get_state()) == ALLOCATED) {
 772         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 773       }
 774     }
 775   }
 776 
 777   // Aborted due to thread limit being reached
 778   if (state == ZOMBIE) {
 779     thread->set_osthread(NULL);
 780     delete osthread;
 781     return false;
 782   }
 783 
 784   // The thread is returned suspended (in state INITIALIZED),
 785   // and is started higher up in the call chain
 786   assert(state == INITIALIZED, "race condition");
 787   return true;
 788 }
 789 
 790 /////////////////////////////////////////////////////////////////////////////
 791 // attach existing thread
 792 
 793 // bootstrap the main thread
 794 bool os::create_main_thread(JavaThread* thread) {
 795   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 796   return create_attached_thread(thread);
 797 }
 798 
 799 bool os::create_attached_thread(JavaThread* thread) {
 800 #ifdef ASSERT
 801   thread->verify_not_published();
 802 #endif
 803 
 804   // Allocate the OSThread object
 805   OSThread* osthread = new OSThread(NULL, NULL);
 806 
 807   if (osthread == NULL) {
 808     return false;
 809   }
 810 
 811   // Store pthread info into the OSThread
 812   osthread->set_thread_id(os::Linux::gettid());
 813   osthread->set_pthread_id(::pthread_self());
 814 
 815   // initialize floating point control register
 816   os::Linux::init_thread_fpu_state();
 817 
 818   // Initial thread state is RUNNABLE
 819   osthread->set_state(RUNNABLE);
 820 
 821   thread->set_osthread(osthread);
 822 
 823   if (UseNUMA) {
 824     int lgrp_id = os::numa_get_group_id();
 825     if (lgrp_id != -1) {
 826       thread->set_lgrp_id(lgrp_id);
 827     }
 828   }
 829 
 830   if (os::Linux::is_initial_thread()) {
 831     // If current thread is initial thread, its stack is mapped on demand,
 832     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 833     // the entire stack region to avoid SEGV in stack banging.
 834     // It is also useful to get around the heap-stack-gap problem on SuSE
 835     // kernel (see 4821821 for details). We first expand stack to the top
 836     // of yellow zone, then enable stack yellow zone (order is significant,
 837     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 838     // is no gap between the last two virtual memory regions.
 839 
 840     JavaThread *jt = (JavaThread *)thread;
 841     address addr = jt->stack_yellow_zone_base();
 842     assert(addr != NULL, "initialization problem?");
 843     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 844 
 845     osthread->set_expanding_stack();
 846     os::Linux::manually_expand_stack(jt, addr);
 847     osthread->clear_expanding_stack();
 848   }
 849 
 850   // initialize signal mask for this thread
 851   // and save the caller's signal mask
 852   os::Linux::hotspot_sigmask(thread);
 853 
 854   return true;
 855 }
 856 
 857 void os::pd_start_thread(Thread* thread) {
 858   OSThread * osthread = thread->osthread();
 859   assert(osthread->get_state() != INITIALIZED, "just checking");
 860   Monitor* sync_with_child = osthread->startThread_lock();
 861   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 862   sync_with_child->notify();
 863 }
 864 
 865 // Free Linux resources related to the OSThread
 866 void os::free_thread(OSThread* osthread) {
 867   assert(osthread != NULL, "osthread not set");
 868 
 869   if (Thread::current()->osthread() == osthread) {
 870     // Restore caller's signal mask
 871     sigset_t sigmask = osthread->caller_sigmask();
 872     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 873   }
 874 
 875   delete osthread;
 876 }
 877 
 878 //////////////////////////////////////////////////////////////////////////////
 879 // thread local storage
 880 
 881 // Restore the thread pointer if the destructor is called. This is in case
 882 // someone from JNI code sets up a destructor with pthread_key_create to run
 883 // detachCurrentThread on thread death. Unless we restore the thread pointer we
 884 // will hang or crash. When detachCurrentThread is called the key will be set
 885 // to null and we will not be called again. If detachCurrentThread is never
 886 // called we could loop forever depending on the pthread implementation.
 887 static void restore_thread_pointer(void* p) {
 888   Thread* thread = (Thread*) p;
 889   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
 890 }
 891 
 892 int os::allocate_thread_local_storage() {
 893   pthread_key_t key;
 894   int rslt = pthread_key_create(&key, restore_thread_pointer);
 895   assert(rslt == 0, "cannot allocate thread local storage");
 896   return (int)key;
 897 }
 898 
 899 // Note: This is currently not used by VM, as we don't destroy TLS key
 900 // on VM exit.
 901 void os::free_thread_local_storage(int index) {
 902   int rslt = pthread_key_delete((pthread_key_t)index);
 903   assert(rslt == 0, "invalid index");
 904 }
 905 
 906 void os::thread_local_storage_at_put(int index, void* value) {
 907   int rslt = pthread_setspecific((pthread_key_t)index, value);
 908   assert(rslt == 0, "pthread_setspecific failed");
 909 }
 910 
 911 extern "C" Thread* get_thread() {
 912   return ThreadLocalStorage::thread();
 913 }
 914 
 915 //////////////////////////////////////////////////////////////////////////////
 916 // initial thread
 917 
 918 // Check if current thread is the initial thread, similar to Solaris thr_main.
 919 bool os::Linux::is_initial_thread(void) {
 920   char dummy;
 921   // If called before init complete, thread stack bottom will be null.
 922   // Can be called if fatal error occurs before initialization.
 923   if (initial_thread_stack_bottom() == NULL) return false;
 924   assert(initial_thread_stack_bottom() != NULL &&
 925          initial_thread_stack_size()   != 0,
 926          "os::init did not locate initial thread's stack region");
 927   if ((address)&dummy >= initial_thread_stack_bottom() &&
 928       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 929     return true;
 930   } else {
 931     return false;
 932   }
 933 }
 934 
 935 // Find the virtual memory area that contains addr
 936 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 937   FILE *fp = fopen("/proc/self/maps", "r");
 938   if (fp) {
 939     address low, high;
 940     while (!feof(fp)) {
 941       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 942         if (low <= addr && addr < high) {
 943           if (vma_low)  *vma_low  = low;
 944           if (vma_high) *vma_high = high;
 945           fclose(fp);
 946           return true;
 947         }
 948       }
 949       for (;;) {
 950         int ch = fgetc(fp);
 951         if (ch == EOF || ch == (int)'\n') break;
 952       }
 953     }
 954     fclose(fp);
 955   }
 956   return false;
 957 }
 958 
 959 // Locate initial thread stack. This special handling of initial thread stack
 960 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 961 // bogus value for initial thread.
 962 void os::Linux::capture_initial_stack(size_t max_size) {
 963   // stack size is the easy part, get it from RLIMIT_STACK
 964   size_t stack_size;
 965   struct rlimit rlim;
 966   getrlimit(RLIMIT_STACK, &rlim);
 967   stack_size = rlim.rlim_cur;
 968 
 969   // 6308388: a bug in ld.so will relocate its own .data section to the
 970   //   lower end of primordial stack; reduce ulimit -s value a little bit
 971   //   so we won't install guard page on ld.so's data section.
 972   stack_size -= 2 * page_size();
 973 
 974   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
 975   //   7.1, in both cases we will get 2G in return value.
 976   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
 977   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
 978   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
 979   //   in case other parts in glibc still assumes 2M max stack size.
 980   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
 981   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
 982   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
 983     stack_size = 2 * K * K IA64_ONLY(*2);
 984   }
 985   // Try to figure out where the stack base (top) is. This is harder.
 986   //
 987   // When an application is started, glibc saves the initial stack pointer in
 988   // a global variable "__libc_stack_end", which is then used by system
 989   // libraries. __libc_stack_end should be pretty close to stack top. The
 990   // variable is available since the very early days. However, because it is
 991   // a private interface, it could disappear in the future.
 992   //
 993   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 994   // to __libc_stack_end, it is very close to stack top, but isn't the real
 995   // stack top. Note that /proc may not exist if VM is running as a chroot
 996   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 997   // /proc/<pid>/stat could change in the future (though unlikely).
 998   //
 999   // We try __libc_stack_end first. If that doesn't work, look for
1000   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1001   // as a hint, which should work well in most cases.
1002 
1003   uintptr_t stack_start;
1004 
1005   // try __libc_stack_end first
1006   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1007   if (p && *p) {
1008     stack_start = *p;
1009   } else {
1010     // see if we can get the start_stack field from /proc/self/stat
1011     FILE *fp;
1012     int pid;
1013     char state;
1014     int ppid;
1015     int pgrp;
1016     int session;
1017     int nr;
1018     int tpgrp;
1019     unsigned long flags;
1020     unsigned long minflt;
1021     unsigned long cminflt;
1022     unsigned long majflt;
1023     unsigned long cmajflt;
1024     unsigned long utime;
1025     unsigned long stime;
1026     long cutime;
1027     long cstime;
1028     long prio;
1029     long nice;
1030     long junk;
1031     long it_real;
1032     uintptr_t start;
1033     uintptr_t vsize;
1034     intptr_t rss;
1035     uintptr_t rsslim;
1036     uintptr_t scodes;
1037     uintptr_t ecode;
1038     int i;
1039 
1040     // Figure what the primordial thread stack base is. Code is inspired
1041     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1042     // followed by command name surrounded by parentheses, state, etc.
1043     char stat[2048];
1044     int statlen;
1045 
1046     fp = fopen("/proc/self/stat", "r");
1047     if (fp) {
1048       statlen = fread(stat, 1, 2047, fp);
1049       stat[statlen] = '\0';
1050       fclose(fp);
1051 
1052       // Skip pid and the command string. Note that we could be dealing with
1053       // weird command names, e.g. user could decide to rename java launcher
1054       // to "java 1.4.2 :)", then the stat file would look like
1055       //                1234 (java 1.4.2 :)) R ... ...
1056       // We don't really need to know the command string, just find the last
1057       // occurrence of ")" and then start parsing from there. See bug 4726580.
1058       char * s = strrchr(stat, ')');
1059 
1060       i = 0;
1061       if (s) {
1062         // Skip blank chars
1063         do { s++; } while (s && isspace(*s));
1064 
1065 #define _UFM UINTX_FORMAT
1066 #define _DFM INTX_FORMAT
1067 
1068         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1069         //              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
1070         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,
1071                    &state,          // 3  %c
1072                    &ppid,           // 4  %d
1073                    &pgrp,           // 5  %d
1074                    &session,        // 6  %d
1075                    &nr,             // 7  %d
1076                    &tpgrp,          // 8  %d
1077                    &flags,          // 9  %lu
1078                    &minflt,         // 10 %lu
1079                    &cminflt,        // 11 %lu
1080                    &majflt,         // 12 %lu
1081                    &cmajflt,        // 13 %lu
1082                    &utime,          // 14 %lu
1083                    &stime,          // 15 %lu
1084                    &cutime,         // 16 %ld
1085                    &cstime,         // 17 %ld
1086                    &prio,           // 18 %ld
1087                    &nice,           // 19 %ld
1088                    &junk,           // 20 %ld
1089                    &it_real,        // 21 %ld
1090                    &start,          // 22 UINTX_FORMAT
1091                    &vsize,          // 23 UINTX_FORMAT
1092                    &rss,            // 24 INTX_FORMAT
1093                    &rsslim,         // 25 UINTX_FORMAT
1094                    &scodes,         // 26 UINTX_FORMAT
1095                    &ecode,          // 27 UINTX_FORMAT
1096                    &stack_start);   // 28 UINTX_FORMAT
1097       }
1098 
1099 #undef _UFM
1100 #undef _DFM
1101 
1102       if (i != 28 - 2) {
1103         assert(false, "Bad conversion from /proc/self/stat");
1104         // product mode - assume we are the initial thread, good luck in the
1105         // embedded case.
1106         warning("Can't detect initial thread stack location - bad conversion");
1107         stack_start = (uintptr_t) &rlim;
1108       }
1109     } else {
1110       // For some reason we can't open /proc/self/stat (for example, running on
1111       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1112       // most cases, so don't abort:
1113       warning("Can't detect initial thread stack location - no /proc/self/stat");
1114       stack_start = (uintptr_t) &rlim;
1115     }
1116   }
1117 
1118   // Now we have a pointer (stack_start) very close to the stack top, the
1119   // next thing to do is to figure out the exact location of stack top. We
1120   // can find out the virtual memory area that contains stack_start by
1121   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1122   // and its upper limit is the real stack top. (again, this would fail if
1123   // running inside chroot, because /proc may not exist.)
1124 
1125   uintptr_t stack_top;
1126   address low, high;
1127   if (find_vma((address)stack_start, &low, &high)) {
1128     // success, "high" is the true stack top. (ignore "low", because initial
1129     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1130     stack_top = (uintptr_t)high;
1131   } else {
1132     // failed, likely because /proc/self/maps does not exist
1133     warning("Can't detect initial thread stack location - find_vma failed");
1134     // best effort: stack_start is normally within a few pages below the real
1135     // stack top, use it as stack top, and reduce stack size so we won't put
1136     // guard page outside stack.
1137     stack_top = stack_start;
1138     stack_size -= 16 * page_size();
1139   }
1140 
1141   // stack_top could be partially down the page so align it
1142   stack_top = align_size_up(stack_top, page_size());
1143 
1144   if (max_size && stack_size > max_size) {
1145     _initial_thread_stack_size = max_size;
1146   } else {
1147     _initial_thread_stack_size = stack_size;
1148   }
1149 
1150   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1151   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1152 }
1153 
1154 ////////////////////////////////////////////////////////////////////////////////
1155 // time support
1156 
1157 // Time since start-up in seconds to a fine granularity.
1158 // Used by VMSelfDestructTimer and the MemProfiler.
1159 double os::elapsedTime() {
1160 
1161   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1162 }
1163 
1164 jlong os::elapsed_counter() {
1165   return javaTimeNanos() - initial_time_count;
1166 }
1167 
1168 jlong os::elapsed_frequency() {
1169   return NANOSECS_PER_SEC; // nanosecond resolution
1170 }
1171 
1172 bool os::supports_vtime() { return true; }
1173 bool os::enable_vtime()   { return false; }
1174 bool os::vtime_enabled()  { return false; }
1175 
1176 double os::elapsedVTime() {
1177   struct rusage usage;
1178   int retval = getrusage(RUSAGE_THREAD, &usage);
1179   if (retval == 0) {
1180     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);
1181   } else {
1182     // better than nothing, but not much
1183     return elapsedTime();
1184   }
1185 }
1186 
1187 jlong os::javaTimeMillis() {
1188   timeval time;
1189   int status = gettimeofday(&time, NULL);
1190   assert(status != -1, "linux error");
1191   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1192 }
1193 
1194 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1195   timeval time;
1196   int status = gettimeofday(&time, NULL);
1197   assert(status != -1, "linux error");
1198   seconds = jlong(time.tv_sec);
1199   nanos = jlong(time.tv_usec) * 1000;
1200 }
1201 
1202 
1203 #ifndef CLOCK_MONOTONIC
1204   #define CLOCK_MONOTONIC (1)
1205 #endif
1206 
1207 void os::Linux::clock_init() {
1208   // we do dlopen's in this particular order due to bug in linux
1209   // dynamical loader (see 6348968) leading to crash on exit
1210   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1211   if (handle == NULL) {
1212     handle = dlopen("librt.so", RTLD_LAZY);
1213   }
1214 
1215   if (handle) {
1216     int (*clock_getres_func)(clockid_t, struct timespec*) =
1217            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1218     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1219            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1220     if (clock_getres_func && clock_gettime_func) {
1221       // See if monotonic clock is supported by the kernel. Note that some
1222       // early implementations simply return kernel jiffies (updated every
1223       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1224       // for nano time (though the monotonic property is still nice to have).
1225       // It's fixed in newer kernels, however clock_getres() still returns
1226       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1227       // resolution for now. Hopefully as people move to new kernels, this
1228       // won't be a problem.
1229       struct timespec res;
1230       struct timespec tp;
1231       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1232           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1233         // yes, monotonic clock is supported
1234         _clock_gettime = clock_gettime_func;
1235         return;
1236       } else {
1237         // close librt if there is no monotonic clock
1238         dlclose(handle);
1239       }
1240     }
1241   }
1242   warning("No monotonic clock was available - timed services may " \
1243           "be adversely affected if the time-of-day clock changes");
1244 }
1245 
1246 #ifndef SYS_clock_getres
1247   #if defined(IA32) || defined(AMD64)
1248     #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1249     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1250   #else
1251     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1252     #define sys_clock_getres(x,y)  -1
1253   #endif
1254 #else
1255   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1256 #endif
1257 
1258 void os::Linux::fast_thread_clock_init() {
1259   if (!UseLinuxPosixThreadCPUClocks) {
1260     return;
1261   }
1262   clockid_t clockid;
1263   struct timespec tp;
1264   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1265       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1266 
1267   // Switch to using fast clocks for thread cpu time if
1268   // the sys_clock_getres() returns 0 error code.
1269   // Note, that some kernels may support the current thread
1270   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1271   // returned by the pthread_getcpuclockid().
1272   // If the fast Posix clocks are supported then the sys_clock_getres()
1273   // must return at least tp.tv_sec == 0 which means a resolution
1274   // better than 1 sec. This is extra check for reliability.
1275 
1276   if (pthread_getcpuclockid_func &&
1277       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1278       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1279     _supports_fast_thread_cpu_time = true;
1280     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1281   }
1282 }
1283 
1284 jlong os::javaTimeNanos() {
1285   if (os::supports_monotonic_clock()) {
1286     struct timespec tp;
1287     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1288     assert(status == 0, "gettime error");
1289     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1290     return result;
1291   } else {
1292     timeval time;
1293     int status = gettimeofday(&time, NULL);
1294     assert(status != -1, "linux error");
1295     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1296     return 1000 * usecs;
1297   }
1298 }
1299 
1300 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1301   if (os::supports_monotonic_clock()) {
1302     info_ptr->max_value = ALL_64_BITS;
1303 
1304     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1305     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1306     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1307   } else {
1308     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1309     info_ptr->max_value = ALL_64_BITS;
1310 
1311     // gettimeofday is a real time clock so it skips
1312     info_ptr->may_skip_backward = true;
1313     info_ptr->may_skip_forward = true;
1314   }
1315 
1316   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1317 }
1318 
1319 // Return the real, user, and system times in seconds from an
1320 // arbitrary fixed point in the past.
1321 bool os::getTimesSecs(double* process_real_time,
1322                       double* process_user_time,
1323                       double* process_system_time) {
1324   struct tms ticks;
1325   clock_t real_ticks = times(&ticks);
1326 
1327   if (real_ticks == (clock_t) (-1)) {
1328     return false;
1329   } else {
1330     double ticks_per_second = (double) clock_tics_per_sec;
1331     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1332     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1333     *process_real_time = ((double) real_ticks) / ticks_per_second;
1334 
1335     return true;
1336   }
1337 }
1338 
1339 
1340 char * os::local_time_string(char *buf, size_t buflen) {
1341   struct tm t;
1342   time_t long_time;
1343   time(&long_time);
1344   localtime_r(&long_time, &t);
1345   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1346                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1347                t.tm_hour, t.tm_min, t.tm_sec);
1348   return buf;
1349 }
1350 
1351 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1352   return localtime_r(clock, res);
1353 }
1354 
1355 ////////////////////////////////////////////////////////////////////////////////
1356 // runtime exit support
1357 
1358 // Note: os::shutdown() might be called very early during initialization, or
1359 // called from signal handler. Before adding something to os::shutdown(), make
1360 // sure it is async-safe and can handle partially initialized VM.
1361 void os::shutdown() {
1362 
1363   // allow PerfMemory to attempt cleanup of any persistent resources
1364   perfMemory_exit();
1365 
1366   // needs to remove object in file system
1367   AttachListener::abort();
1368 
1369   // flush buffered output, finish log files
1370   ostream_abort();
1371 
1372   // Check for abort hook
1373   abort_hook_t abort_hook = Arguments::abort_hook();
1374   if (abort_hook != NULL) {
1375     abort_hook();
1376   }
1377 
1378 }
1379 
1380 // Note: os::abort() might be called very early during initialization, or
1381 // called from signal handler. Before adding something to os::abort(), make
1382 // sure it is async-safe and can handle partially initialized VM.
1383 void os::abort(bool dump_core, void* siginfo, void* context) {
1384   os::shutdown();
1385   if (dump_core) {
1386 #ifndef PRODUCT
1387     fdStream out(defaultStream::output_fd());
1388     out.print_raw("Current thread is ");
1389     char buf[16];
1390     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1391     out.print_raw_cr(buf);
1392     out.print_raw_cr("Dumping core ...");
1393 #endif
1394     ::abort(); // dump core
1395   }
1396 
1397   ::exit(1);
1398 }
1399 
1400 // Die immediately, no exit hook, no abort hook, no cleanup.
1401 void os::die() {
1402   ::abort();
1403 }
1404 
1405 
1406 // This method is a copy of JDK's sysGetLastErrorString
1407 // from src/solaris/hpi/src/system_md.c
1408 
1409 size_t os::lasterror(char *buf, size_t len) {
1410   if (errno == 0)  return 0;
1411 
1412   const char *s = ::strerror(errno);
1413   size_t n = ::strlen(s);
1414   if (n >= len) {
1415     n = len - 1;
1416   }
1417   ::strncpy(buf, s, n);
1418   buf[n] = '\0';
1419   return n;
1420 }
1421 
1422 // thread_id is kernel thread id (similar to Solaris LWP id)
1423 intx os::current_thread_id() { return os::Linux::gettid(); }
1424 int os::current_process_id() {
1425   return ::getpid();
1426 }
1427 
1428 // DLL functions
1429 
1430 const char* os::dll_file_extension() { return ".so"; }
1431 
1432 // This must be hard coded because it's the system's temporary
1433 // directory not the java application's temp directory, ala java.io.tmpdir.
1434 const char* os::get_temp_directory() { return "/tmp"; }
1435 
1436 static bool file_exists(const char* filename) {
1437   struct stat statbuf;
1438   if (filename == NULL || strlen(filename) == 0) {
1439     return false;
1440   }
1441   return os::stat(filename, &statbuf) == 0;
1442 }
1443 
1444 bool os::dll_build_name(char* buffer, size_t buflen,
1445                         const char* pname, const char* fname) {
1446   bool retval = false;
1447   // Copied from libhpi
1448   const size_t pnamelen = pname ? strlen(pname) : 0;
1449 
1450   // Return error on buffer overflow.
1451   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1452     return retval;
1453   }
1454 
1455   if (pnamelen == 0) {
1456     snprintf(buffer, buflen, "lib%s.so", fname);
1457     retval = true;
1458   } else if (strchr(pname, *os::path_separator()) != NULL) {
1459     int n;
1460     char** pelements = split_path(pname, &n);
1461     if (pelements == NULL) {
1462       return false;
1463     }
1464     for (int i = 0; i < n; i++) {
1465       // Really shouldn't be NULL, but check can't hurt
1466       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1467         continue; // skip the empty path values
1468       }
1469       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1470       if (file_exists(buffer)) {
1471         retval = true;
1472         break;
1473       }
1474     }
1475     // release the storage
1476     for (int i = 0; i < n; i++) {
1477       if (pelements[i] != NULL) {
1478         FREE_C_HEAP_ARRAY(char, pelements[i]);
1479       }
1480     }
1481     if (pelements != NULL) {
1482       FREE_C_HEAP_ARRAY(char*, pelements);
1483     }
1484   } else {
1485     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1486     retval = true;
1487   }
1488   return retval;
1489 }
1490 
1491 // check if addr is inside libjvm.so
1492 bool os::address_is_in_vm(address addr) {
1493   static address libjvm_base_addr;
1494   Dl_info dlinfo;
1495 
1496   if (libjvm_base_addr == NULL) {
1497     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1498       libjvm_base_addr = (address)dlinfo.dli_fbase;
1499     }
1500     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1501   }
1502 
1503   if (dladdr((void *)addr, &dlinfo) != 0) {
1504     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1505   }
1506 
1507   return false;
1508 }
1509 
1510 bool os::dll_address_to_function_name(address addr, char *buf,
1511                                       int buflen, int *offset,
1512                                       bool demangle) {
1513   // buf is not optional, but offset is optional
1514   assert(buf != NULL, "sanity check");
1515 
1516   Dl_info dlinfo;
1517 
1518   if (dladdr((void*)addr, &dlinfo) != 0) {
1519     // see if we have a matching symbol
1520     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1521       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1522         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1523       }
1524       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1525       return true;
1526     }
1527     // no matching symbol so try for just file info
1528     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1529       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1530                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1531         return true;
1532       }
1533     }
1534   }
1535 
1536   buf[0] = '\0';
1537   if (offset != NULL) *offset = -1;
1538   return false;
1539 }
1540 
1541 struct _address_to_library_name {
1542   address addr;          // input : memory address
1543   size_t  buflen;        //         size of fname
1544   char*   fname;         // output: library name
1545   address base;          //         library base addr
1546 };
1547 
1548 static int address_to_library_name_callback(struct dl_phdr_info *info,
1549                                             size_t size, void *data) {
1550   int i;
1551   bool found = false;
1552   address libbase = NULL;
1553   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1554 
1555   // iterate through all loadable segments
1556   for (i = 0; i < info->dlpi_phnum; i++) {
1557     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1558     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1559       // base address of a library is the lowest address of its loaded
1560       // segments.
1561       if (libbase == NULL || libbase > segbase) {
1562         libbase = segbase;
1563       }
1564       // see if 'addr' is within current segment
1565       if (segbase <= d->addr &&
1566           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1567         found = true;
1568       }
1569     }
1570   }
1571 
1572   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1573   // so dll_address_to_library_name() can fall through to use dladdr() which
1574   // can figure out executable name from argv[0].
1575   if (found && info->dlpi_name && info->dlpi_name[0]) {
1576     d->base = libbase;
1577     if (d->fname) {
1578       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1579     }
1580     return 1;
1581   }
1582   return 0;
1583 }
1584 
1585 bool os::dll_address_to_library_name(address addr, char* buf,
1586                                      int buflen, int* offset) {
1587   // buf is not optional, but offset is optional
1588   assert(buf != NULL, "sanity check");
1589 
1590   Dl_info dlinfo;
1591   struct _address_to_library_name data;
1592 
1593   // There is a bug in old glibc dladdr() implementation that it could resolve
1594   // to wrong library name if the .so file has a base address != NULL. Here
1595   // we iterate through the program headers of all loaded libraries to find
1596   // out which library 'addr' really belongs to. This workaround can be
1597   // removed once the minimum requirement for glibc is moved to 2.3.x.
1598   data.addr = addr;
1599   data.fname = buf;
1600   data.buflen = buflen;
1601   data.base = NULL;
1602   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1603 
1604   if (rslt) {
1605     // buf already contains library name
1606     if (offset) *offset = addr - data.base;
1607     return true;
1608   }
1609   if (dladdr((void*)addr, &dlinfo) != 0) {
1610     if (dlinfo.dli_fname != NULL) {
1611       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1612     }
1613     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1614       *offset = addr - (address)dlinfo.dli_fbase;
1615     }
1616     return true;
1617   }
1618 
1619   buf[0] = '\0';
1620   if (offset) *offset = -1;
1621   return false;
1622 }
1623 
1624 // Loads .dll/.so and
1625 // in case of error it checks if .dll/.so was built for the
1626 // same architecture as Hotspot is running on
1627 
1628 
1629 // Remember the stack's state. The Linux dynamic linker will change
1630 // the stack to 'executable' at most once, so we must safepoint only once.
1631 bool os::Linux::_stack_is_executable = false;
1632 
1633 // VM operation that loads a library.  This is necessary if stack protection
1634 // of the Java stacks can be lost during loading the library.  If we
1635 // do not stop the Java threads, they can stack overflow before the stacks
1636 // are protected again.
1637 class VM_LinuxDllLoad: public VM_Operation {
1638  private:
1639   const char *_filename;
1640   char *_ebuf;
1641   int _ebuflen;
1642   void *_lib;
1643  public:
1644   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1645     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1646   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1647   void doit() {
1648     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1649     os::Linux::_stack_is_executable = true;
1650   }
1651   void* loaded_library() { return _lib; }
1652 };
1653 
1654 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1655   void * result = NULL;
1656   bool load_attempted = false;
1657 
1658   // Check whether the library to load might change execution rights
1659   // of the stack. If they are changed, the protection of the stack
1660   // guard pages will be lost. We need a safepoint to fix this.
1661   //
1662   // See Linux man page execstack(8) for more info.
1663   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1664     ElfFile ef(filename);
1665     if (!ef.specifies_noexecstack()) {
1666       if (!is_init_completed()) {
1667         os::Linux::_stack_is_executable = true;
1668         // This is OK - No Java threads have been created yet, and hence no
1669         // stack guard pages to fix.
1670         //
1671         // This should happen only when you are building JDK7 using a very
1672         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1673         //
1674         // Dynamic loader will make all stacks executable after
1675         // this function returns, and will not do that again.
1676         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1677       } else {
1678         warning("You have loaded library %s which might have disabled stack guard. "
1679                 "The VM will try to fix the stack guard now.\n"
1680                 "It's highly recommended that you fix the library with "
1681                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1682                 filename);
1683 
1684         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1685         JavaThread *jt = JavaThread::current();
1686         if (jt->thread_state() != _thread_in_native) {
1687           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1688           // that requires ExecStack. Cannot enter safe point. Let's give up.
1689           warning("Unable to fix stack guard. Giving up.");
1690         } else {
1691           if (!LoadExecStackDllInVMThread) {
1692             // This is for the case where the DLL has an static
1693             // constructor function that executes JNI code. We cannot
1694             // load such DLLs in the VMThread.
1695             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1696           }
1697 
1698           ThreadInVMfromNative tiv(jt);
1699           debug_only(VMNativeEntryWrapper vew;)
1700 
1701           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1702           VMThread::execute(&op);
1703           if (LoadExecStackDllInVMThread) {
1704             result = op.loaded_library();
1705           }
1706           load_attempted = true;
1707         }
1708       }
1709     }
1710   }
1711 
1712   if (!load_attempted) {
1713     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1714   }
1715 
1716   if (result != NULL) {
1717     // Successful loading
1718     return result;
1719   }
1720 
1721   Elf32_Ehdr elf_head;
1722   int diag_msg_max_length=ebuflen-strlen(ebuf);
1723   char* diag_msg_buf=ebuf+strlen(ebuf);
1724 
1725   if (diag_msg_max_length==0) {
1726     // No more space in ebuf for additional diagnostics message
1727     return NULL;
1728   }
1729 
1730 
1731   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1732 
1733   if (file_descriptor < 0) {
1734     // Can't open library, report dlerror() message
1735     return NULL;
1736   }
1737 
1738   bool failed_to_read_elf_head=
1739     (sizeof(elf_head)!=
1740      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1741 
1742   ::close(file_descriptor);
1743   if (failed_to_read_elf_head) {
1744     // file i/o error - report dlerror() msg
1745     return NULL;
1746   }
1747 
1748   typedef struct {
1749     Elf32_Half  code;         // Actual value as defined in elf.h
1750     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1751     char        elf_class;    // 32 or 64 bit
1752     char        endianess;    // MSB or LSB
1753     char*       name;         // String representation
1754   } arch_t;
1755 
1756 #ifndef EM_486
1757   #define EM_486          6               /* Intel 80486 */
1758 #endif
1759 #ifndef EM_AARCH64
1760   #define EM_AARCH64    183               /* ARM AARCH64 */
1761 #endif
1762 
1763   static const arch_t arch_array[]={
1764     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1765     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1766     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1767     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1768     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1769     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1770     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1771     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1772 #if defined(VM_LITTLE_ENDIAN)
1773     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1774 #else
1775     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1776 #endif
1777     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1778     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1779     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1780     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1781     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1782     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1783     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1784     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1785   };
1786 
1787 #if  (defined IA32)
1788   static  Elf32_Half running_arch_code=EM_386;
1789 #elif   (defined AMD64)
1790   static  Elf32_Half running_arch_code=EM_X86_64;
1791 #elif  (defined IA64)
1792   static  Elf32_Half running_arch_code=EM_IA_64;
1793 #elif  (defined __sparc) && (defined _LP64)
1794   static  Elf32_Half running_arch_code=EM_SPARCV9;
1795 #elif  (defined __sparc) && (!defined _LP64)
1796   static  Elf32_Half running_arch_code=EM_SPARC;
1797 #elif  (defined __powerpc64__)
1798   static  Elf32_Half running_arch_code=EM_PPC64;
1799 #elif  (defined __powerpc__)
1800   static  Elf32_Half running_arch_code=EM_PPC;
1801 #elif  (defined ARM)
1802   static  Elf32_Half running_arch_code=EM_ARM;
1803 #elif  (defined S390)
1804   static  Elf32_Half running_arch_code=EM_S390;
1805 #elif  (defined ALPHA)
1806   static  Elf32_Half running_arch_code=EM_ALPHA;
1807 #elif  (defined MIPSEL)
1808   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1809 #elif  (defined PARISC)
1810   static  Elf32_Half running_arch_code=EM_PARISC;
1811 #elif  (defined MIPS)
1812   static  Elf32_Half running_arch_code=EM_MIPS;
1813 #elif  (defined M68K)
1814   static  Elf32_Half running_arch_code=EM_68K;
1815 #elif  (defined AARCH64)
1816   static  Elf32_Half running_arch_code=EM_AARCH64;
1817 #else
1818     #error Method os::dll_load requires that one of following is defined:\
1819          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1820 #endif
1821 
1822   // Identify compatability class for VM's architecture and library's architecture
1823   // Obtain string descriptions for architectures
1824 
1825   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1826   int running_arch_index=-1;
1827 
1828   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1829     if (running_arch_code == arch_array[i].code) {
1830       running_arch_index    = i;
1831     }
1832     if (lib_arch.code == arch_array[i].code) {
1833       lib_arch.compat_class = arch_array[i].compat_class;
1834       lib_arch.name         = arch_array[i].name;
1835     }
1836   }
1837 
1838   assert(running_arch_index != -1,
1839          "Didn't find running architecture code (running_arch_code) in arch_array");
1840   if (running_arch_index == -1) {
1841     // Even though running architecture detection failed
1842     // we may still continue with reporting dlerror() message
1843     return NULL;
1844   }
1845 
1846   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1847     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1848     return NULL;
1849   }
1850 
1851 #ifndef S390
1852   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1853     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1854     return NULL;
1855   }
1856 #endif // !S390
1857 
1858   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1859     if (lib_arch.name!=NULL) {
1860       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1861                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1862                  lib_arch.name, arch_array[running_arch_index].name);
1863     } else {
1864       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1865                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1866                  lib_arch.code,
1867                  arch_array[running_arch_index].name);
1868     }
1869   }
1870 
1871   return NULL;
1872 }
1873 
1874 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1875                                 int ebuflen) {
1876   void * result = ::dlopen(filename, RTLD_LAZY);
1877   if (result == NULL) {
1878     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1879     ebuf[ebuflen-1] = '\0';
1880   }
1881   return result;
1882 }
1883 
1884 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1885                                        int ebuflen) {
1886   void * result = NULL;
1887   if (LoadExecStackDllInVMThread) {
1888     result = dlopen_helper(filename, ebuf, ebuflen);
1889   }
1890 
1891   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1892   // library that requires an executable stack, or which does not have this
1893   // stack attribute set, dlopen changes the stack attribute to executable. The
1894   // read protection of the guard pages gets lost.
1895   //
1896   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1897   // may have been queued at the same time.
1898 
1899   if (!_stack_is_executable) {
1900     JavaThread *jt = Threads::first();
1901 
1902     while (jt) {
1903       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
1904           jt->stack_yellow_zone_enabled()) {       // No pending stack overflow exceptions
1905         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
1906                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
1907           warning("Attempt to reguard stack yellow zone failed.");
1908         }
1909       }
1910       jt = jt->next();
1911     }
1912   }
1913 
1914   return result;
1915 }
1916 
1917 void* os::dll_lookup(void* handle, const char* name) {
1918   void* res = dlsym(handle, name);
1919   return res;
1920 }
1921 
1922 void* os::get_default_process_handle() {
1923   return (void*)::dlopen(NULL, RTLD_LAZY);
1924 }
1925 
1926 static bool _print_ascii_file(const char* filename, outputStream* st) {
1927   int fd = ::open(filename, O_RDONLY);
1928   if (fd == -1) {
1929     return false;
1930   }
1931 
1932   char buf[32];
1933   int bytes;
1934   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1935     st->print_raw(buf, bytes);
1936   }
1937 
1938   ::close(fd);
1939 
1940   return true;
1941 }
1942 
1943 void os::print_dll_info(outputStream *st) {
1944   st->print_cr("Dynamic libraries:");
1945 
1946   char fname[32];
1947   pid_t pid = os::Linux::gettid();
1948 
1949   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1950 
1951   if (!_print_ascii_file(fname, st)) {
1952     st->print("Can not get library information for pid = %d\n", pid);
1953   }
1954 }
1955 
1956 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1957   FILE *procmapsFile = NULL;
1958 
1959   // Open the procfs maps file for the current process
1960   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1961     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1962     char line[PATH_MAX + 100];
1963 
1964     // Read line by line from 'file'
1965     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1966       u8 base, top, offset, inode;
1967       char permissions[5];
1968       char device[6];
1969       char name[PATH_MAX + 1];
1970 
1971       // Parse fields from line
1972       sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
1973 
1974       // Filter by device id '00:00' so that we only get file system mapped files.
1975       if (strcmp(device, "00:00") != 0) {
1976 
1977         // Call callback with the fields of interest
1978         if(callback(name, (address)base, (address)top, param)) {
1979           // Oops abort, callback aborted
1980           fclose(procmapsFile);
1981           return 1;
1982         }
1983       }
1984     }
1985     fclose(procmapsFile);
1986   }
1987   return 0;
1988 }
1989 
1990 void os::print_os_info_brief(outputStream* st) {
1991   os::Linux::print_distro_info(st);
1992 
1993   os::Posix::print_uname_info(st);
1994 
1995   os::Linux::print_libversion_info(st);
1996 
1997 }
1998 
1999 void os::print_os_info(outputStream* st) {
2000   st->print("OS:");
2001 
2002   os::Linux::print_distro_info(st);
2003 
2004   os::Posix::print_uname_info(st);
2005 
2006   // Print warning if unsafe chroot environment detected
2007   if (unsafe_chroot_detected) {
2008     st->print("WARNING!! ");
2009     st->print_cr("%s", unstable_chroot_error);
2010   }
2011 
2012   os::Linux::print_libversion_info(st);
2013 
2014   os::Posix::print_rlimit_info(st);
2015 
2016   os::Posix::print_load_average(st);
2017 
2018   os::Linux::print_full_memory_info(st);
2019 }
2020 
2021 // Try to identify popular distros.
2022 // Most Linux distributions have a /etc/XXX-release file, which contains
2023 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2024 // file that also contains the OS version string. Some have more than one
2025 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2026 // /etc/redhat-release.), so the order is important.
2027 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2028 // their own specific XXX-release file as well as a redhat-release file.
2029 // Because of this the XXX-release file needs to be searched for before the
2030 // redhat-release file.
2031 // Since Red Hat has a lsb-release file that is not very descriptive the
2032 // search for redhat-release needs to be before lsb-release.
2033 // Since the lsb-release file is the new standard it needs to be searched
2034 // before the older style release files.
2035 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2036 // next to last resort.  The os-release file is a new standard that contains
2037 // distribution information and the system-release file seems to be an old
2038 // standard that has been replaced by the lsb-release and os-release files.
2039 // Searching for the debian_version file is the last resort.  It contains
2040 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2041 // "Debian " is printed before the contents of the debian_version file.
2042 
2043 const char* distro_files[] = {
2044   "/etc/oracle-release",
2045   "/etc/mandriva-release",
2046   "/etc/mandrake-release",
2047   "/etc/sun-release",
2048   "/etc/redhat-release",
2049   "/etc/lsb-release",
2050   "/etc/SuSE-release",
2051   "/etc/turbolinux-release",
2052   "/etc/gentoo-release",
2053   "/etc/ltib-release",
2054   "/etc/angstrom-version",
2055   "/etc/system-release",
2056   "/etc/os-release",
2057   NULL };
2058 
2059 void os::Linux::print_distro_info(outputStream* st) {
2060   for (int i = 0;; i++) {
2061     const char* file = distro_files[i];
2062     if (file == NULL) {
2063       break;  // done
2064     }
2065     // If file prints, we found it.
2066     if (_print_ascii_file(file, st)) {
2067       return;
2068     }
2069   }
2070 
2071   if (file_exists("/etc/debian_version")) {
2072     st->print("Debian ");
2073     _print_ascii_file("/etc/debian_version", st);
2074   } else {
2075     st->print("Linux");
2076   }
2077   st->cr();
2078 }
2079 
2080 static void parse_os_info(char* distro, size_t length, const char* file) {
2081   FILE* fp = fopen(file, "r");
2082   if (fp != NULL) {
2083     char buf[256];
2084     // get last line of the file.
2085     while (fgets(buf, sizeof(buf), fp)) { }
2086     // Edit out extra stuff in expected ubuntu format
2087     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2088       char* ptr = strstr(buf, "\"");  // the name is in quotes
2089       if (ptr != NULL) {
2090         ptr++; // go beyond first quote
2091         char* nl = strchr(ptr, '\"');
2092         if (nl != NULL) *nl = '\0';
2093         strncpy(distro, ptr, length);
2094       } else {
2095         ptr = strstr(buf, "=");
2096         ptr++; // go beyond equals then
2097         char* nl = strchr(ptr, '\n');
2098         if (nl != NULL) *nl = '\0';
2099         strncpy(distro, ptr, length);
2100       }
2101     } else {
2102       // if not in expected Ubuntu format, print out whole line minus \n
2103       char* nl = strchr(buf, '\n');
2104       if (nl != NULL) *nl = '\0';
2105       strncpy(distro, buf, length);
2106     }
2107     // close distro file
2108     fclose(fp);
2109   }
2110 }
2111 
2112 void os::get_summary_os_info(char* buf, size_t buflen) {
2113   for (int i = 0;; i++) {
2114     const char* file = distro_files[i];
2115     if (file == NULL) {
2116       break; // ran out of distro_files
2117     }
2118     if (file_exists(file)) {
2119       parse_os_info(buf, buflen, file);
2120       return;
2121     }
2122   }
2123   // special case for debian
2124   if (file_exists("/etc/debian_version")) {
2125     strncpy(buf, "Debian ", buflen);
2126     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2127   } else {
2128     strncpy(buf, "Linux", buflen);
2129   }
2130 }
2131 
2132 void os::Linux::print_libversion_info(outputStream* st) {
2133   // libc, pthread
2134   st->print("libc:");
2135   st->print("%s ", os::Linux::glibc_version());
2136   st->print("%s ", os::Linux::libpthread_version());
2137   st->cr();
2138 }
2139 
2140 void os::Linux::print_full_memory_info(outputStream* st) {
2141   st->print("\n/proc/meminfo:\n");
2142   _print_ascii_file("/proc/meminfo", st);
2143   st->cr();
2144 }
2145 
2146 void os::print_memory_info(outputStream* st) {
2147 
2148   st->print("Memory:");
2149   st->print(" %dk page", os::vm_page_size()>>10);
2150 
2151   // values in struct sysinfo are "unsigned long"
2152   struct sysinfo si;
2153   sysinfo(&si);
2154 
2155   st->print(", physical " UINT64_FORMAT "k",
2156             os::physical_memory() >> 10);
2157   st->print("(" UINT64_FORMAT "k free)",
2158             os::available_memory() >> 10);
2159   st->print(", swap " UINT64_FORMAT "k",
2160             ((jlong)si.totalswap * si.mem_unit) >> 10);
2161   st->print("(" UINT64_FORMAT "k free)",
2162             ((jlong)si.freeswap * si.mem_unit) >> 10);
2163   st->cr();
2164 }
2165 
2166 // Print the first "model name" line and the first "flags" line
2167 // that we find and nothing more. We assume "model name" comes
2168 // before "flags" so if we find a second "model name", then the
2169 // "flags" field is considered missing.
2170 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2171 #if defined(IA32) || defined(AMD64)
2172   // Other platforms have less repetitive cpuinfo files
2173   FILE *fp = fopen("/proc/cpuinfo", "r");
2174   if (fp) {
2175     while (!feof(fp)) {
2176       if (fgets(buf, buflen, fp)) {
2177         // Assume model name comes before flags
2178         bool model_name_printed = false;
2179         if (strstr(buf, "model name") != NULL) {
2180           if (!model_name_printed) {
2181             st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2182             st->print_raw(buf);
2183             model_name_printed = true;
2184           } else {
2185             // model name printed but not flags?  Odd, just return
2186             fclose(fp);
2187             return true;
2188           }
2189         }
2190         // print the flags line too
2191         if (strstr(buf, "flags") != NULL) {
2192           st->print_raw(buf);
2193           fclose(fp);
2194           return true;
2195         }
2196       }
2197     }
2198     fclose(fp);
2199   }
2200 #endif // x86 platforms
2201   return false;
2202 }
2203 
2204 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2205   // Only print the model name if the platform provides this as a summary
2206   if (!print_model_name_and_flags(st, buf, buflen)) {
2207     st->print("\n/proc/cpuinfo:\n");
2208     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2209       st->print_cr("  <Not Available>");
2210     }
2211   }
2212 }
2213 
2214 #if defined(AMD64) || defined(IA32) || defined(X32)
2215 const char* search_string = "model name";
2216 #elif defined(SPARC)
2217 const char* search_string = "cpu";
2218 #else
2219 const char* search_string = "Processor";
2220 #endif
2221 
2222 // Parses the cpuinfo file for string representing the model name.
2223 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2224   FILE* fp = fopen("/proc/cpuinfo", "r");
2225   if (fp != NULL) {
2226     while (!feof(fp)) {
2227       char buf[256];
2228       if (fgets(buf, sizeof(buf), fp)) {
2229         char* start = strstr(buf, search_string);
2230         if (start != NULL) {
2231           char *ptr = start + strlen(search_string);
2232           char *end = buf + strlen(buf);
2233           while (ptr != end) {
2234              // skip whitespace and colon for the rest of the name.
2235              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2236                break;
2237              }
2238              ptr++;
2239           }
2240           if (ptr != end) {
2241             // reasonable string, get rid of newline and keep the rest
2242             char* nl = strchr(buf, '\n');
2243             if (nl != NULL) *nl = '\0';
2244             strncpy(cpuinfo, ptr, length);
2245             fclose(fp);
2246             return;
2247           }
2248         }
2249       }
2250     }
2251     fclose(fp);
2252   }
2253   // cpuinfo not found or parsing failed, just print generic string.  The entire
2254   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2255 #if defined(AMD64)
2256   strncpy(cpuinfo, "x86_64", length);
2257 #elif defined(IA32)
2258   strncpy(cpuinfo, "x86_32", length);
2259 #elif defined(IA64)
2260   strncpy(cpuinfo, "IA64", length);
2261 #elif defined(SPARC)
2262   strncpy(cpuinfo, "sparcv9", length);
2263 #elif defined(AARCH64)
2264   strncpy(cpuinfo, "AArch64", length);
2265 #elif defined(ARM)
2266   strncpy(cpuinfo, "ARM", length);
2267 #elif defined(PPC)
2268   strncpy(cpuinfo, "PPC64", length);
2269 #elif defined(ZERO_LIBARCH)
2270   strncpy(cpuinfo, ZERO_LIBARCH, length);
2271 #else
2272   strncpy(cpuinfo, "unknown", length);
2273 #endif
2274 }
2275 
2276 void os::print_siginfo(outputStream* st, void* siginfo) {
2277   const siginfo_t* si = (const siginfo_t*)siginfo;
2278 
2279   os::Posix::print_siginfo_brief(st, si);
2280 #if INCLUDE_CDS
2281   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2282       UseSharedSpaces) {
2283     FileMapInfo* mapinfo = FileMapInfo::current_info();
2284     if (mapinfo->is_in_shared_space(si->si_addr)) {
2285       st->print("\n\nError accessing class data sharing archive."   \
2286                 " Mapped file inaccessible during execution, "      \
2287                 " possible disk/network problem.");
2288     }
2289   }
2290 #endif
2291   st->cr();
2292 }
2293 
2294 
2295 static void print_signal_handler(outputStream* st, int sig,
2296                                  char* buf, size_t buflen);
2297 
2298 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2299   st->print_cr("Signal Handlers:");
2300   print_signal_handler(st, SIGSEGV, buf, buflen);
2301   print_signal_handler(st, SIGBUS , buf, buflen);
2302   print_signal_handler(st, SIGFPE , buf, buflen);
2303   print_signal_handler(st, SIGPIPE, buf, buflen);
2304   print_signal_handler(st, SIGXFSZ, buf, buflen);
2305   print_signal_handler(st, SIGILL , buf, buflen);
2306   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2307   print_signal_handler(st, SR_signum, buf, buflen);
2308   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2309   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2310   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2311   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2312 #if defined(PPC64)
2313   print_signal_handler(st, SIGTRAP, buf, buflen);
2314 #endif
2315 }
2316 
2317 static char saved_jvm_path[MAXPATHLEN] = {0};
2318 
2319 // Find the full path to the current module, libjvm.so
2320 void os::jvm_path(char *buf, jint buflen) {
2321   // Error checking.
2322   if (buflen < MAXPATHLEN) {
2323     assert(false, "must use a large-enough buffer");
2324     buf[0] = '\0';
2325     return;
2326   }
2327   // Lazy resolve the path to current module.
2328   if (saved_jvm_path[0] != 0) {
2329     strcpy(buf, saved_jvm_path);
2330     return;
2331   }
2332 
2333   char dli_fname[MAXPATHLEN];
2334   bool ret = dll_address_to_library_name(
2335                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2336                                          dli_fname, sizeof(dli_fname), NULL);
2337   assert(ret, "cannot locate libjvm");
2338   char *rp = NULL;
2339   if (ret && dli_fname[0] != '\0') {
2340     rp = realpath(dli_fname, buf);
2341   }
2342   if (rp == NULL) {
2343     return;
2344   }
2345 
2346   if (Arguments::sun_java_launcher_is_altjvm()) {
2347     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2348     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2349     // If "/jre/lib/" appears at the right place in the string, then
2350     // assume we are installed in a JDK and we're done. Otherwise, check
2351     // for a JAVA_HOME environment variable and fix up the path so it
2352     // looks like libjvm.so is installed there (append a fake suffix
2353     // hotspot/libjvm.so).
2354     const char *p = buf + strlen(buf) - 1;
2355     for (int count = 0; p > buf && count < 5; ++count) {
2356       for (--p; p > buf && *p != '/'; --p)
2357         /* empty */ ;
2358     }
2359 
2360     if (strncmp(p, "/jre/lib/", 9) != 0) {
2361       // Look for JAVA_HOME in the environment.
2362       char* java_home_var = ::getenv("JAVA_HOME");
2363       if (java_home_var != NULL && java_home_var[0] != 0) {
2364         char* jrelib_p;
2365         int len;
2366 
2367         // Check the current module name "libjvm.so".
2368         p = strrchr(buf, '/');
2369         if (p == NULL) {
2370           return;
2371         }
2372         assert(strstr(p, "/libjvm") == p, "invalid library name");
2373 
2374         rp = realpath(java_home_var, buf);
2375         if (rp == NULL) {
2376           return;
2377         }
2378 
2379         // determine if this is a legacy image or modules image
2380         // modules image doesn't have "jre" subdirectory
2381         len = strlen(buf);
2382         assert(len < buflen, "Ran out of buffer room");
2383         jrelib_p = buf + len;
2384         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2385         if (0 != access(buf, F_OK)) {
2386           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2387         }
2388 
2389         if (0 == access(buf, F_OK)) {
2390           // Use current module name "libjvm.so"
2391           len = strlen(buf);
2392           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2393         } else {
2394           // Go back to path of .so
2395           rp = realpath(dli_fname, buf);
2396           if (rp == NULL) {
2397             return;
2398           }
2399         }
2400       }
2401     }
2402   }
2403 
2404   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2405   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2406 }
2407 
2408 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2409   // no prefix required, not even "_"
2410 }
2411 
2412 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2413   // no suffix required
2414 }
2415 
2416 ////////////////////////////////////////////////////////////////////////////////
2417 // sun.misc.Signal support
2418 
2419 static volatile jint sigint_count = 0;
2420 
2421 static void UserHandler(int sig, void *siginfo, void *context) {
2422   // 4511530 - sem_post is serialized and handled by the manager thread. When
2423   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2424   // don't want to flood the manager thread with sem_post requests.
2425   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2426     return;
2427   }
2428 
2429   // Ctrl-C is pressed during error reporting, likely because the error
2430   // handler fails to abort. Let VM die immediately.
2431   if (sig == SIGINT && is_error_reported()) {
2432     os::die();
2433   }
2434 
2435   os::signal_notify(sig);
2436 }
2437 
2438 void* os::user_handler() {
2439   return CAST_FROM_FN_PTR(void*, UserHandler);
2440 }
2441 
2442 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2443   struct timespec ts;
2444   // Semaphore's are always associated with CLOCK_REALTIME
2445   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2446   // see unpackTime for discussion on overflow checking
2447   if (sec >= MAX_SECS) {
2448     ts.tv_sec += MAX_SECS;
2449     ts.tv_nsec = 0;
2450   } else {
2451     ts.tv_sec += sec;
2452     ts.tv_nsec += nsec;
2453     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2454       ts.tv_nsec -= NANOSECS_PER_SEC;
2455       ++ts.tv_sec; // note: this must be <= max_secs
2456     }
2457   }
2458 
2459   return ts;
2460 }
2461 
2462 extern "C" {
2463   typedef void (*sa_handler_t)(int);
2464   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2465 }
2466 
2467 void* os::signal(int signal_number, void* handler) {
2468   struct sigaction sigAct, oldSigAct;
2469 
2470   sigfillset(&(sigAct.sa_mask));
2471   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2472   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2473 
2474   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2475     // -1 means registration failed
2476     return (void *)-1;
2477   }
2478 
2479   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2480 }
2481 
2482 void os::signal_raise(int signal_number) {
2483   ::raise(signal_number);
2484 }
2485 
2486 // The following code is moved from os.cpp for making this
2487 // code platform specific, which it is by its very nature.
2488 
2489 // Will be modified when max signal is changed to be dynamic
2490 int os::sigexitnum_pd() {
2491   return NSIG;
2492 }
2493 
2494 // a counter for each possible signal value
2495 static volatile jint pending_signals[NSIG+1] = { 0 };
2496 
2497 // Linux(POSIX) specific hand shaking semaphore.
2498 static sem_t sig_sem;
2499 static PosixSemaphore sr_semaphore;
2500 
2501 void os::signal_init_pd() {
2502   // Initialize signal structures
2503   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2504 
2505   // Initialize signal semaphore
2506   ::sem_init(&sig_sem, 0, 0);
2507 }
2508 
2509 void os::signal_notify(int sig) {
2510   Atomic::inc(&pending_signals[sig]);
2511   ::sem_post(&sig_sem);
2512 }
2513 
2514 static int check_pending_signals(bool wait) {
2515   Atomic::store(0, &sigint_count);
2516   for (;;) {
2517     for (int i = 0; i < NSIG + 1; i++) {
2518       jint n = pending_signals[i];
2519       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2520         return i;
2521       }
2522     }
2523     if (!wait) {
2524       return -1;
2525     }
2526     JavaThread *thread = JavaThread::current();
2527     ThreadBlockInVM tbivm(thread);
2528 
2529     bool threadIsSuspended;
2530     do {
2531       thread->set_suspend_equivalent();
2532       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2533       ::sem_wait(&sig_sem);
2534 
2535       // were we externally suspended while we were waiting?
2536       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2537       if (threadIsSuspended) {
2538         // The semaphore has been incremented, but while we were waiting
2539         // another thread suspended us. We don't want to continue running
2540         // while suspended because that would surprise the thread that
2541         // suspended us.
2542         ::sem_post(&sig_sem);
2543 
2544         thread->java_suspend_self();
2545       }
2546     } while (threadIsSuspended);
2547   }
2548 }
2549 
2550 int os::signal_lookup() {
2551   return check_pending_signals(false);
2552 }
2553 
2554 int os::signal_wait() {
2555   return check_pending_signals(true);
2556 }
2557 
2558 ////////////////////////////////////////////////////////////////////////////////
2559 // Virtual Memory
2560 
2561 int os::vm_page_size() {
2562   // Seems redundant as all get out
2563   assert(os::Linux::page_size() != -1, "must call os::init");
2564   return os::Linux::page_size();
2565 }
2566 
2567 // Solaris allocates memory by pages.
2568 int os::vm_allocation_granularity() {
2569   assert(os::Linux::page_size() != -1, "must call os::init");
2570   return os::Linux::page_size();
2571 }
2572 
2573 // Rationale behind this function:
2574 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2575 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2576 //  samples for JITted code. Here we create private executable mapping over the code cache
2577 //  and then we can use standard (well, almost, as mapping can change) way to provide
2578 //  info for the reporting script by storing timestamp and location of symbol
2579 void linux_wrap_code(char* base, size_t size) {
2580   static volatile jint cnt = 0;
2581 
2582   if (!UseOprofile) {
2583     return;
2584   }
2585 
2586   char buf[PATH_MAX+1];
2587   int num = Atomic::add(1, &cnt);
2588 
2589   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2590            os::get_temp_directory(), os::current_process_id(), num);
2591   unlink(buf);
2592 
2593   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2594 
2595   if (fd != -1) {
2596     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2597     if (rv != (off_t)-1) {
2598       if (::write(fd, "", 1) == 1) {
2599         mmap(base, size,
2600              PROT_READ|PROT_WRITE|PROT_EXEC,
2601              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2602       }
2603     }
2604     ::close(fd);
2605     unlink(buf);
2606   }
2607 }
2608 
2609 static bool recoverable_mmap_error(int err) {
2610   // See if the error is one we can let the caller handle. This
2611   // list of errno values comes from JBS-6843484. I can't find a
2612   // Linux man page that documents this specific set of errno
2613   // values so while this list currently matches Solaris, it may
2614   // change as we gain experience with this failure mode.
2615   switch (err) {
2616   case EBADF:
2617   case EINVAL:
2618   case ENOTSUP:
2619     // let the caller deal with these errors
2620     return true;
2621 
2622   default:
2623     // Any remaining errors on this OS can cause our reserved mapping
2624     // to be lost. That can cause confusion where different data
2625     // structures think they have the same memory mapped. The worst
2626     // scenario is if both the VM and a library think they have the
2627     // same memory mapped.
2628     return false;
2629   }
2630 }
2631 
2632 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2633                                     int err) {
2634   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2635           ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2636           strerror(err), err);
2637 }
2638 
2639 static void warn_fail_commit_memory(char* addr, size_t size,
2640                                     size_t alignment_hint, bool exec,
2641                                     int err) {
2642   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2643           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2644           alignment_hint, exec, strerror(err), err);
2645 }
2646 
2647 // NOTE: Linux kernel does not really reserve the pages for us.
2648 //       All it does is to check if there are enough free pages
2649 //       left at the time of mmap(). This could be a potential
2650 //       problem.
2651 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2652   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2653   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2654                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2655   if (res != (uintptr_t) MAP_FAILED) {
2656     if (UseNUMAInterleaving) {
2657       numa_make_global(addr, size);
2658     }
2659     return 0;
2660   }
2661 
2662   int err = errno;  // save errno from mmap() call above
2663 
2664   if (!recoverable_mmap_error(err)) {
2665     warn_fail_commit_memory(addr, size, exec, err);
2666     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2667   }
2668 
2669   return err;
2670 }
2671 
2672 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2673   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2674 }
2675 
2676 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2677                                   const char* mesg) {
2678   assert(mesg != NULL, "mesg must be specified");
2679   int err = os::Linux::commit_memory_impl(addr, size, exec);
2680   if (err != 0) {
2681     // the caller wants all commit errors to exit with the specified mesg:
2682     warn_fail_commit_memory(addr, size, exec, err);
2683     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2684   }
2685 }
2686 
2687 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2688 #ifndef MAP_HUGETLB
2689   #define MAP_HUGETLB 0x40000
2690 #endif
2691 
2692 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2693 #ifndef MADV_HUGEPAGE
2694   #define MADV_HUGEPAGE 14
2695 #endif
2696 
2697 int os::Linux::commit_memory_impl(char* addr, size_t size,
2698                                   size_t alignment_hint, bool exec) {
2699   int err = os::Linux::commit_memory_impl(addr, size, exec);
2700   if (err == 0) {
2701     realign_memory(addr, size, alignment_hint);
2702   }
2703   return err;
2704 }
2705 
2706 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2707                           bool exec) {
2708   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2709 }
2710 
2711 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2712                                   size_t alignment_hint, bool exec,
2713                                   const char* mesg) {
2714   assert(mesg != NULL, "mesg must be specified");
2715   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2716   if (err != 0) {
2717     // the caller wants all commit errors to exit with the specified mesg:
2718     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2719     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2720   }
2721 }
2722 
2723 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2724   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2725     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2726     // be supported or the memory may already be backed by huge pages.
2727     ::madvise(addr, bytes, MADV_HUGEPAGE);
2728   }
2729 }
2730 
2731 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2732   // This method works by doing an mmap over an existing mmaping and effectively discarding
2733   // the existing pages. However it won't work for SHM-based large pages that cannot be
2734   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2735   // small pages on top of the SHM segment. This method always works for small pages, so we
2736   // allow that in any case.
2737   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2738     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2739   }
2740 }
2741 
2742 void os::numa_make_global(char *addr, size_t bytes) {
2743   Linux::numa_interleave_memory(addr, bytes);
2744 }
2745 
2746 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2747 // bind policy to MPOL_PREFERRED for the current thread.
2748 #define USE_MPOL_PREFERRED 0
2749 
2750 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2751   // To make NUMA and large pages more robust when both enabled, we need to ease
2752   // the requirements on where the memory should be allocated. MPOL_BIND is the
2753   // default policy and it will force memory to be allocated on the specified
2754   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2755   // the specified node, but will not force it. Using this policy will prevent
2756   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2757   // free large pages.
2758   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2759   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2760 }
2761 
2762 bool os::numa_topology_changed() { return false; }
2763 
2764 size_t os::numa_get_groups_num() {
2765   int max_node = Linux::numa_max_node();
2766   return max_node > 0 ? max_node + 1 : 1;
2767 }
2768 
2769 int os::numa_get_group_id() {
2770   int cpu_id = Linux::sched_getcpu();
2771   if (cpu_id != -1) {
2772     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2773     if (lgrp_id != -1) {
2774       return lgrp_id;
2775     }
2776   }
2777   return 0;
2778 }
2779 
2780 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2781   for (size_t i = 0; i < size; i++) {
2782     ids[i] = i;
2783   }
2784   return size;
2785 }
2786 
2787 bool os::get_page_info(char *start, page_info* info) {
2788   return false;
2789 }
2790 
2791 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2792                      page_info* page_found) {
2793   return end;
2794 }
2795 
2796 
2797 int os::Linux::sched_getcpu_syscall(void) {
2798   unsigned int cpu;
2799   int retval = -1;
2800 
2801 #if defined(IA32)
2802   #ifndef SYS_getcpu
2803     #define SYS_getcpu 318
2804   #endif
2805   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2806 #elif defined(AMD64)
2807 // Unfortunately we have to bring all these macros here from vsyscall.h
2808 // to be able to compile on old linuxes.
2809   #define __NR_vgetcpu 2
2810   #define VSYSCALL_START (-10UL << 20)
2811   #define VSYSCALL_SIZE 1024
2812   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2813   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2814   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2815   retval = vgetcpu(&cpu, NULL, NULL);
2816 #endif
2817 
2818   return (retval == -1) ? retval : cpu;
2819 }
2820 
2821 // Something to do with the numa-aware allocator needs these symbols
2822 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2823 extern "C" JNIEXPORT void numa_error(char *where) { }
2824 extern "C" JNIEXPORT int fork1() { return fork(); }
2825 
2826 
2827 // If we are running with libnuma version > 2, then we should
2828 // be trying to use symbols with versions 1.1
2829 // If we are running with earlier version, which did not have symbol versions,
2830 // we should use the base version.
2831 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2832   void *f = dlvsym(handle, name, "libnuma_1.1");
2833   if (f == NULL) {
2834     f = dlsym(handle, name);
2835   }
2836   return f;
2837 }
2838 
2839 bool os::Linux::libnuma_init() {
2840   // sched_getcpu() should be in libc.
2841   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2842                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2843 
2844   // If it's not, try a direct syscall.
2845   if (sched_getcpu() == -1) {
2846     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2847                                     (void*)&sched_getcpu_syscall));
2848   }
2849 
2850   if (sched_getcpu() != -1) { // Does it work?
2851     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2852     if (handle != NULL) {
2853       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2854                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2855       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2856                                        libnuma_dlsym(handle, "numa_max_node")));
2857       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2858                                         libnuma_dlsym(handle, "numa_available")));
2859       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2860                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2861       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2862                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2863       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2864                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2865 
2866 
2867       if (numa_available() != -1) {
2868         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2869         // Create a cpu -> node mapping
2870         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2871         rebuild_cpu_to_node_map();
2872         return true;
2873       }
2874     }
2875   }
2876   return false;
2877 }
2878 
2879 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2880 // The table is later used in get_node_by_cpu().
2881 void os::Linux::rebuild_cpu_to_node_map() {
2882   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2883                               // in libnuma (possible values are starting from 16,
2884                               // and continuing up with every other power of 2, but less
2885                               // than the maximum number of CPUs supported by kernel), and
2886                               // is a subject to change (in libnuma version 2 the requirements
2887                               // are more reasonable) we'll just hardcode the number they use
2888                               // in the library.
2889   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2890 
2891   size_t cpu_num = os::active_processor_count();
2892   size_t cpu_map_size = NCPUS / BitsPerCLong;
2893   size_t cpu_map_valid_size =
2894     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2895 
2896   cpu_to_node()->clear();
2897   cpu_to_node()->at_grow(cpu_num - 1);
2898   size_t node_num = numa_get_groups_num();
2899 
2900   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2901   for (size_t i = 0; i < node_num; i++) {
2902     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2903       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2904         if (cpu_map[j] != 0) {
2905           for (size_t k = 0; k < BitsPerCLong; k++) {
2906             if (cpu_map[j] & (1UL << k)) {
2907               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2908             }
2909           }
2910         }
2911       }
2912     }
2913   }
2914   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2915 }
2916 
2917 int os::Linux::get_node_by_cpu(int cpu_id) {
2918   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2919     return cpu_to_node()->at(cpu_id);
2920   }
2921   return -1;
2922 }
2923 
2924 GrowableArray<int>* os::Linux::_cpu_to_node;
2925 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2926 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2927 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2928 os::Linux::numa_available_func_t os::Linux::_numa_available;
2929 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2930 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2931 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2932 unsigned long* os::Linux::_numa_all_nodes;
2933 
2934 bool os::pd_uncommit_memory(char* addr, size_t size) {
2935   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2936                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2937   return res  != (uintptr_t) MAP_FAILED;
2938 }
2939 
2940 static address get_stack_commited_bottom(address bottom, size_t size) {
2941   address nbot = bottom;
2942   address ntop = bottom + size;
2943 
2944   size_t page_sz = os::vm_page_size();
2945   unsigned pages = size / page_sz;
2946 
2947   unsigned char vec[1];
2948   unsigned imin = 1, imax = pages + 1, imid;
2949   int mincore_return_value = 0;
2950 
2951   assert(imin <= imax, "Unexpected page size");
2952 
2953   while (imin < imax) {
2954     imid = (imax + imin) / 2;
2955     nbot = ntop - (imid * page_sz);
2956 
2957     // Use a trick with mincore to check whether the page is mapped or not.
2958     // mincore sets vec to 1 if page resides in memory and to 0 if page
2959     // is swapped output but if page we are asking for is unmapped
2960     // it returns -1,ENOMEM
2961     mincore_return_value = mincore(nbot, page_sz, vec);
2962 
2963     if (mincore_return_value == -1) {
2964       // Page is not mapped go up
2965       // to find first mapped page
2966       if (errno != EAGAIN) {
2967         assert(errno == ENOMEM, "Unexpected mincore errno");
2968         imax = imid;
2969       }
2970     } else {
2971       // Page is mapped go down
2972       // to find first not mapped page
2973       imin = imid + 1;
2974     }
2975   }
2976 
2977   nbot = nbot + page_sz;
2978 
2979   // Adjust stack bottom one page up if last checked page is not mapped
2980   if (mincore_return_value == -1) {
2981     nbot = nbot + page_sz;
2982   }
2983 
2984   return nbot;
2985 }
2986 
2987 
2988 // Linux uses a growable mapping for the stack, and if the mapping for
2989 // the stack guard pages is not removed when we detach a thread the
2990 // stack cannot grow beyond the pages where the stack guard was
2991 // mapped.  If at some point later in the process the stack expands to
2992 // that point, the Linux kernel cannot expand the stack any further
2993 // because the guard pages are in the way, and a segfault occurs.
2994 //
2995 // However, it's essential not to split the stack region by unmapping
2996 // a region (leaving a hole) that's already part of the stack mapping,
2997 // so if the stack mapping has already grown beyond the guard pages at
2998 // the time we create them, we have to truncate the stack mapping.
2999 // So, we need to know the extent of the stack mapping when
3000 // create_stack_guard_pages() is called.
3001 
3002 // We only need this for stacks that are growable: at the time of
3003 // writing thread stacks don't use growable mappings (i.e. those
3004 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3005 // only applies to the main thread.
3006 
3007 // If the (growable) stack mapping already extends beyond the point
3008 // where we're going to put our guard pages, truncate the mapping at
3009 // that point by munmap()ping it.  This ensures that when we later
3010 // munmap() the guard pages we don't leave a hole in the stack
3011 // mapping. This only affects the main/initial thread
3012 
3013 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3014   if (os::Linux::is_initial_thread()) {
3015     // As we manually grow stack up to bottom inside create_attached_thread(),
3016     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3017     // we don't need to do anything special.
3018     // Check it first, before calling heavy function.
3019     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3020     unsigned char vec[1];
3021 
3022     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3023       // Fallback to slow path on all errors, including EAGAIN
3024       stack_extent = (uintptr_t) get_stack_commited_bottom(
3025                                                            os::Linux::initial_thread_stack_bottom(),
3026                                                            (size_t)addr - stack_extent);
3027     }
3028 
3029     if (stack_extent < (uintptr_t)addr) {
3030       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3031     }
3032   }
3033 
3034   return os::commit_memory(addr, size, !ExecMem);
3035 }
3036 
3037 // If this is a growable mapping, remove the guard pages entirely by
3038 // munmap()ping them.  If not, just call uncommit_memory(). This only
3039 // affects the main/initial thread, but guard against future OS changes
3040 // It's safe to always unmap guard pages for initial thread because we
3041 // always place it right after end of the mapped region
3042 
3043 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3044   uintptr_t stack_extent, stack_base;
3045 
3046   if (os::Linux::is_initial_thread()) {
3047     return ::munmap(addr, size) == 0;
3048   }
3049 
3050   return os::uncommit_memory(addr, size);
3051 }
3052 
3053 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3054 // at 'requested_addr'. If there are existing memory mappings at the same
3055 // location, however, they will be overwritten. If 'fixed' is false,
3056 // 'requested_addr' is only treated as a hint, the return value may or
3057 // may not start from the requested address. Unlike Linux mmap(), this
3058 // function returns NULL to indicate failure.
3059 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3060   char * addr;
3061   int flags;
3062 
3063   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3064   if (fixed) {
3065     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3066     flags |= MAP_FIXED;
3067   }
3068 
3069   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3070   // touch an uncommitted page. Otherwise, the read/write might
3071   // succeed if we have enough swap space to back the physical page.
3072   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3073                        flags, -1, 0);
3074 
3075   return addr == MAP_FAILED ? NULL : addr;
3076 }
3077 
3078 static int anon_munmap(char * addr, size_t size) {
3079   return ::munmap(addr, size) == 0;
3080 }
3081 
3082 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3083                             size_t alignment_hint) {
3084   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3085 }
3086 
3087 bool os::pd_release_memory(char* addr, size_t size) {
3088   return anon_munmap(addr, size);
3089 }
3090 
3091 static bool linux_mprotect(char* addr, size_t size, int prot) {
3092   // Linux wants the mprotect address argument to be page aligned.
3093   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3094 
3095   // According to SUSv3, mprotect() should only be used with mappings
3096   // established by mmap(), and mmap() always maps whole pages. Unaligned
3097   // 'addr' likely indicates problem in the VM (e.g. trying to change
3098   // protection of malloc'ed or statically allocated memory). Check the
3099   // caller if you hit this assert.
3100   assert(addr == bottom, "sanity check");
3101 
3102   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3103   return ::mprotect(bottom, size, prot) == 0;
3104 }
3105 
3106 // Set protections specified
3107 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3108                         bool is_committed) {
3109   unsigned int p = 0;
3110   switch (prot) {
3111   case MEM_PROT_NONE: p = PROT_NONE; break;
3112   case MEM_PROT_READ: p = PROT_READ; break;
3113   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3114   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3115   default:
3116     ShouldNotReachHere();
3117   }
3118   // is_committed is unused.
3119   return linux_mprotect(addr, bytes, p);
3120 }
3121 
3122 bool os::guard_memory(char* addr, size_t size) {
3123   return linux_mprotect(addr, size, PROT_NONE);
3124 }
3125 
3126 bool os::unguard_memory(char* addr, size_t size) {
3127   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3128 }
3129 
3130 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3131                                                     size_t page_size) {
3132   bool result = false;
3133   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3134                  MAP_ANONYMOUS|MAP_PRIVATE,
3135                  -1, 0);
3136   if (p != MAP_FAILED) {
3137     void *aligned_p = align_ptr_up(p, page_size);
3138 
3139     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3140 
3141     munmap(p, page_size * 2);
3142   }
3143 
3144   if (warn && !result) {
3145     warning("TransparentHugePages is not supported by the operating system.");
3146   }
3147 
3148   return result;
3149 }
3150 
3151 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3152   bool result = false;
3153   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3154                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3155                  -1, 0);
3156 
3157   if (p != MAP_FAILED) {
3158     // We don't know if this really is a huge page or not.
3159     FILE *fp = fopen("/proc/self/maps", "r");
3160     if (fp) {
3161       while (!feof(fp)) {
3162         char chars[257];
3163         long x = 0;
3164         if (fgets(chars, sizeof(chars), fp)) {
3165           if (sscanf(chars, "%lx-%*x", &x) == 1
3166               && x == (long)p) {
3167             if (strstr (chars, "hugepage")) {
3168               result = true;
3169               break;
3170             }
3171           }
3172         }
3173       }
3174       fclose(fp);
3175     }
3176     munmap(p, page_size);
3177   }
3178 
3179   if (warn && !result) {
3180     warning("HugeTLBFS is not supported by the operating system.");
3181   }
3182 
3183   return result;
3184 }
3185 
3186 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3187 //
3188 // From the coredump_filter documentation:
3189 //
3190 // - (bit 0) anonymous private memory
3191 // - (bit 1) anonymous shared memory
3192 // - (bit 2) file-backed private memory
3193 // - (bit 3) file-backed shared memory
3194 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3195 //           effective only if the bit 2 is cleared)
3196 // - (bit 5) hugetlb private memory
3197 // - (bit 6) hugetlb shared memory
3198 //
3199 static void set_coredump_filter(void) {
3200   FILE *f;
3201   long cdm;
3202 
3203   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3204     return;
3205   }
3206 
3207   if (fscanf(f, "%lx", &cdm) != 1) {
3208     fclose(f);
3209     return;
3210   }
3211 
3212   rewind(f);
3213 
3214   if ((cdm & LARGEPAGES_BIT) == 0) {
3215     cdm |= LARGEPAGES_BIT;
3216     fprintf(f, "%#lx", cdm);
3217   }
3218 
3219   fclose(f);
3220 }
3221 
3222 // Large page support
3223 
3224 static size_t _large_page_size = 0;
3225 
3226 size_t os::Linux::find_large_page_size() {
3227   size_t large_page_size = 0;
3228 
3229   // large_page_size on Linux is used to round up heap size. x86 uses either
3230   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3231   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3232   // page as large as 256M.
3233   //
3234   // Here we try to figure out page size by parsing /proc/meminfo and looking
3235   // for a line with the following format:
3236   //    Hugepagesize:     2048 kB
3237   //
3238   // If we can't determine the value (e.g. /proc is not mounted, or the text
3239   // format has been changed), we'll use the largest page size supported by
3240   // the processor.
3241 
3242 #ifndef ZERO
3243   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3244                      ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3245 #endif // ZERO
3246 
3247   FILE *fp = fopen("/proc/meminfo", "r");
3248   if (fp) {
3249     while (!feof(fp)) {
3250       int x = 0;
3251       char buf[16];
3252       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3253         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3254           large_page_size = x * K;
3255           break;
3256         }
3257       } else {
3258         // skip to next line
3259         for (;;) {
3260           int ch = fgetc(fp);
3261           if (ch == EOF || ch == (int)'\n') break;
3262         }
3263       }
3264     }
3265     fclose(fp);
3266   }
3267 
3268   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3269     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3270             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3271             proper_unit_for_byte_size(large_page_size));
3272   }
3273 
3274   return large_page_size;
3275 }
3276 
3277 size_t os::Linux::setup_large_page_size() {
3278   _large_page_size = Linux::find_large_page_size();
3279   const size_t default_page_size = (size_t)Linux::page_size();
3280   if (_large_page_size > default_page_size) {
3281     _page_sizes[0] = _large_page_size;
3282     _page_sizes[1] = default_page_size;
3283     _page_sizes[2] = 0;
3284   }
3285 
3286   return _large_page_size;
3287 }
3288 
3289 bool os::Linux::setup_large_page_type(size_t page_size) {
3290   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3291       FLAG_IS_DEFAULT(UseSHM) &&
3292       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3293 
3294     // The type of large pages has not been specified by the user.
3295 
3296     // Try UseHugeTLBFS and then UseSHM.
3297     UseHugeTLBFS = UseSHM = true;
3298 
3299     // Don't try UseTransparentHugePages since there are known
3300     // performance issues with it turned on. This might change in the future.
3301     UseTransparentHugePages = false;
3302   }
3303 
3304   if (UseTransparentHugePages) {
3305     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3306     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3307       UseHugeTLBFS = false;
3308       UseSHM = false;
3309       return true;
3310     }
3311     UseTransparentHugePages = false;
3312   }
3313 
3314   if (UseHugeTLBFS) {
3315     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3316     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3317       UseSHM = false;
3318       return true;
3319     }
3320     UseHugeTLBFS = false;
3321   }
3322 
3323   return UseSHM;
3324 }
3325 
3326 void os::large_page_init() {
3327   if (!UseLargePages &&
3328       !UseTransparentHugePages &&
3329       !UseHugeTLBFS &&
3330       !UseSHM) {
3331     // Not using large pages.
3332     return;
3333   }
3334 
3335   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3336     // The user explicitly turned off large pages.
3337     // Ignore the rest of the large pages flags.
3338     UseTransparentHugePages = false;
3339     UseHugeTLBFS = false;
3340     UseSHM = false;
3341     return;
3342   }
3343 
3344   size_t large_page_size = Linux::setup_large_page_size();
3345   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3346 
3347   set_coredump_filter();
3348 }
3349 
3350 #ifndef SHM_HUGETLB
3351   #define SHM_HUGETLB 04000
3352 #endif
3353 
3354 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3355                                             char* req_addr, bool exec) {
3356   // "exec" is passed in but not used.  Creating the shared image for
3357   // the code cache doesn't have an SHM_X executable permission to check.
3358   assert(UseLargePages && UseSHM, "only for SHM large pages");
3359   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3360 
3361   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3362     return NULL; // Fallback to small pages.
3363   }
3364 
3365   key_t key = IPC_PRIVATE;
3366   char *addr;
3367 
3368   bool warn_on_failure = UseLargePages &&
3369                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3370                          !FLAG_IS_DEFAULT(UseSHM) ||
3371                          !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3372   char msg[128];
3373 
3374   // Create a large shared memory region to attach to based on size.
3375   // Currently, size is the total size of the heap
3376   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3377   if (shmid == -1) {
3378     // Possible reasons for shmget failure:
3379     // 1. shmmax is too small for Java heap.
3380     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3381     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3382     // 2. not enough large page memory.
3383     //    > check available large pages: cat /proc/meminfo
3384     //    > increase amount of large pages:
3385     //          echo new_value > /proc/sys/vm/nr_hugepages
3386     //      Note 1: different Linux may use different name for this property,
3387     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3388     //      Note 2: it's possible there's enough physical memory available but
3389     //            they are so fragmented after a long run that they can't
3390     //            coalesce into large pages. Try to reserve large pages when
3391     //            the system is still "fresh".
3392     if (warn_on_failure) {
3393       jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3394       warning("%s", msg);
3395     }
3396     return NULL;
3397   }
3398 
3399   // attach to the region
3400   addr = (char*)shmat(shmid, req_addr, 0);
3401   int err = errno;
3402 
3403   // Remove shmid. If shmat() is successful, the actual shared memory segment
3404   // will be deleted when it's detached by shmdt() or when the process
3405   // terminates. If shmat() is not successful this will remove the shared
3406   // segment immediately.
3407   shmctl(shmid, IPC_RMID, NULL);
3408 
3409   if ((intptr_t)addr == -1) {
3410     if (warn_on_failure) {
3411       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3412       warning("%s", msg);
3413     }
3414     return NULL;
3415   }
3416 
3417   return addr;
3418 }
3419 
3420 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3421                                         int error) {
3422   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3423 
3424   bool warn_on_failure = UseLargePages &&
3425       (!FLAG_IS_DEFAULT(UseLargePages) ||
3426        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3427        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3428 
3429   if (warn_on_failure) {
3430     char msg[128];
3431     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3432                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3433     warning("%s", msg);
3434   }
3435 }
3436 
3437 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3438                                                         char* req_addr,
3439                                                         bool exec) {
3440   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3441   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3442   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3443 
3444   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3445   char* addr = (char*)::mmap(req_addr, bytes, prot,
3446                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3447                              -1, 0);
3448 
3449   if (addr == MAP_FAILED) {
3450     warn_on_large_pages_failure(req_addr, bytes, errno);
3451     return NULL;
3452   }
3453 
3454   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3455 
3456   return addr;
3457 }
3458 
3459 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3460 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3461 //   (req_addr != NULL) or with a given alignment.
3462 //  - bytes shall be a multiple of alignment.
3463 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3464 //  - alignment sets the alignment at which memory shall be allocated.
3465 //     It must be a multiple of allocation granularity.
3466 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3467 //  req_addr or NULL.
3468 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3469 
3470   size_t extra_size = bytes;
3471   if (req_addr == NULL && alignment > 0) {
3472     extra_size += alignment;
3473   }
3474 
3475   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3476     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3477     -1, 0);
3478   if (start == MAP_FAILED) {
3479     start = NULL;
3480   } else {
3481     if (req_addr != NULL) {
3482       if (start != req_addr) {
3483         ::munmap(start, extra_size);
3484         start = NULL;
3485       }
3486     } else {
3487       char* const start_aligned = (char*) align_ptr_up(start, alignment);
3488       char* const end_aligned = start_aligned + bytes;
3489       char* const end = start + extra_size;
3490       if (start_aligned > start) {
3491         ::munmap(start, start_aligned - start);
3492       }
3493       if (end_aligned < end) {
3494         ::munmap(end_aligned, end - end_aligned);
3495       }
3496       start = start_aligned;
3497     }
3498   }
3499   return start;
3500 
3501 }
3502 
3503 // Reserve memory using mmap(MAP_HUGETLB).
3504 //  - bytes shall be a multiple of alignment.
3505 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3506 //  - alignment sets the alignment at which memory shall be allocated.
3507 //     It must be a multiple of allocation granularity.
3508 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3509 //  req_addr or NULL.
3510 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3511                                                          size_t alignment,
3512                                                          char* req_addr,
3513                                                          bool exec) {
3514   size_t large_page_size = os::large_page_size();
3515   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3516 
3517   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3518   assert(is_size_aligned(bytes, alignment), "Must be");
3519 
3520   // First reserve - but not commit - the address range in small pages.
3521   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3522 
3523   if (start == NULL) {
3524     return NULL;
3525   }
3526 
3527   assert(is_ptr_aligned(start, alignment), "Must be");
3528 
3529   char* end = start + bytes;
3530 
3531   // Find the regions of the allocated chunk that can be promoted to large pages.
3532   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3533   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3534 
3535   size_t lp_bytes = lp_end - lp_start;
3536 
3537   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3538 
3539   if (lp_bytes == 0) {
3540     // The mapped region doesn't even span the start and the end of a large page.
3541     // Fall back to allocate a non-special area.
3542     ::munmap(start, end - start);
3543     return NULL;
3544   }
3545 
3546   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3547 
3548   void* result;
3549 
3550   // Commit small-paged leading area.
3551   if (start != lp_start) {
3552     result = ::mmap(start, lp_start - start, prot,
3553                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3554                     -1, 0);
3555     if (result == MAP_FAILED) {
3556       ::munmap(lp_start, end - lp_start);
3557       return NULL;
3558     }
3559   }
3560 
3561   // Commit large-paged area.
3562   result = ::mmap(lp_start, lp_bytes, prot,
3563                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3564                   -1, 0);
3565   if (result == MAP_FAILED) {
3566     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3567     // If the mmap above fails, the large pages region will be unmapped and we
3568     // have regions before and after with small pages. Release these regions.
3569     //
3570     // |  mapped  |  unmapped  |  mapped  |
3571     // ^          ^            ^          ^
3572     // start      lp_start     lp_end     end
3573     //
3574     ::munmap(start, lp_start - start);
3575     ::munmap(lp_end, end - lp_end);
3576     return NULL;
3577   }
3578 
3579   // Commit small-paged trailing area.
3580   if (lp_end != end) {
3581     result = ::mmap(lp_end, end - lp_end, prot,
3582                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3583                     -1, 0);
3584     if (result == MAP_FAILED) {
3585       ::munmap(start, lp_end - start);
3586       return NULL;
3587     }
3588   }
3589 
3590   return start;
3591 }
3592 
3593 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3594                                                    size_t alignment,
3595                                                    char* req_addr,
3596                                                    bool exec) {
3597   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3598   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3599   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3600   assert(is_power_of_2(os::large_page_size()), "Must be");
3601   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3602 
3603   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3604     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3605   } else {
3606     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3607   }
3608 }
3609 
3610 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3611                                  char* req_addr, bool exec) {
3612   assert(UseLargePages, "only for large pages");
3613 
3614   char* addr;
3615   if (UseSHM) {
3616     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3617   } else {
3618     assert(UseHugeTLBFS, "must be");
3619     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3620   }
3621 
3622   if (addr != NULL) {
3623     if (UseNUMAInterleaving) {
3624       numa_make_global(addr, bytes);
3625     }
3626 
3627     // The memory is committed
3628     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3629   }
3630 
3631   return addr;
3632 }
3633 
3634 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3635   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3636   return shmdt(base) == 0;
3637 }
3638 
3639 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3640   return pd_release_memory(base, bytes);
3641 }
3642 
3643 bool os::release_memory_special(char* base, size_t bytes) {
3644   bool res;
3645   if (MemTracker::tracking_level() > NMT_minimal) {
3646     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3647     res = os::Linux::release_memory_special_impl(base, bytes);
3648     if (res) {
3649       tkr.record((address)base, bytes);
3650     }
3651 
3652   } else {
3653     res = os::Linux::release_memory_special_impl(base, bytes);
3654   }
3655   return res;
3656 }
3657 
3658 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3659   assert(UseLargePages, "only for large pages");
3660   bool res;
3661 
3662   if (UseSHM) {
3663     res = os::Linux::release_memory_special_shm(base, bytes);
3664   } else {
3665     assert(UseHugeTLBFS, "must be");
3666     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3667   }
3668   return res;
3669 }
3670 
3671 size_t os::large_page_size() {
3672   return _large_page_size;
3673 }
3674 
3675 // With SysV SHM the entire memory region must be allocated as shared
3676 // memory.
3677 // HugeTLBFS allows application to commit large page memory on demand.
3678 // However, when committing memory with HugeTLBFS fails, the region
3679 // that was supposed to be committed will lose the old reservation
3680 // and allow other threads to steal that memory region. Because of this
3681 // behavior we can't commit HugeTLBFS memory.
3682 bool os::can_commit_large_page_memory() {
3683   return UseTransparentHugePages;
3684 }
3685 
3686 bool os::can_execute_large_page_memory() {
3687   return UseTransparentHugePages || UseHugeTLBFS;
3688 }
3689 
3690 // Reserve memory at an arbitrary address, only if that area is
3691 // available (and not reserved for something else).
3692 
3693 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3694   const int max_tries = 10;
3695   char* base[max_tries];
3696   size_t size[max_tries];
3697   const size_t gap = 0x000000;
3698 
3699   // Assert only that the size is a multiple of the page size, since
3700   // that's all that mmap requires, and since that's all we really know
3701   // about at this low abstraction level.  If we need higher alignment,
3702   // we can either pass an alignment to this method or verify alignment
3703   // in one of the methods further up the call chain.  See bug 5044738.
3704   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3705 
3706   // Repeatedly allocate blocks until the block is allocated at the
3707   // right spot.
3708 
3709   // Linux mmap allows caller to pass an address as hint; give it a try first,
3710   // if kernel honors the hint then we can return immediately.
3711   char * addr = anon_mmap(requested_addr, bytes, false);
3712   if (addr == requested_addr) {
3713     return requested_addr;
3714   }
3715 
3716   if (addr != NULL) {
3717     // mmap() is successful but it fails to reserve at the requested address
3718     anon_munmap(addr, bytes);
3719   }
3720 
3721   int i;
3722   for (i = 0; i < max_tries; ++i) {
3723     base[i] = reserve_memory(bytes);
3724 
3725     if (base[i] != NULL) {
3726       // Is this the block we wanted?
3727       if (base[i] == requested_addr) {
3728         size[i] = bytes;
3729         break;
3730       }
3731 
3732       // Does this overlap the block we wanted? Give back the overlapped
3733       // parts and try again.
3734 
3735       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3736       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3737         unmap_memory(base[i], top_overlap);
3738         base[i] += top_overlap;
3739         size[i] = bytes - top_overlap;
3740       } else {
3741         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3742         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3743           unmap_memory(requested_addr, bottom_overlap);
3744           size[i] = bytes - bottom_overlap;
3745         } else {
3746           size[i] = bytes;
3747         }
3748       }
3749     }
3750   }
3751 
3752   // Give back the unused reserved pieces.
3753 
3754   for (int j = 0; j < i; ++j) {
3755     if (base[j] != NULL) {
3756       unmap_memory(base[j], size[j]);
3757     }
3758   }
3759 
3760   if (i < max_tries) {
3761     return requested_addr;
3762   } else {
3763     return NULL;
3764   }
3765 }
3766 
3767 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3768   return ::read(fd, buf, nBytes);
3769 }
3770 
3771 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3772   return ::pread(fd, buf, nBytes, offset);
3773 }
3774 
3775 // Short sleep, direct OS call.
3776 //
3777 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3778 // sched_yield(2) will actually give up the CPU:
3779 //
3780 //   * Alone on this pariticular CPU, keeps running.
3781 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3782 //     (pre 2.6.39).
3783 //
3784 // So calling this with 0 is an alternative.
3785 //
3786 void os::naked_short_sleep(jlong ms) {
3787   struct timespec req;
3788 
3789   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3790   req.tv_sec = 0;
3791   if (ms > 0) {
3792     req.tv_nsec = (ms % 1000) * 1000000;
3793   } else {
3794     req.tv_nsec = 1;
3795   }
3796 
3797   nanosleep(&req, NULL);
3798 
3799   return;
3800 }
3801 
3802 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3803 void os::infinite_sleep() {
3804   while (true) {    // sleep forever ...
3805     ::sleep(100);   // ... 100 seconds at a time
3806   }
3807 }
3808 
3809 // Used to convert frequent JVM_Yield() to nops
3810 bool os::dont_yield() {
3811   return DontYieldALot;
3812 }
3813 
3814 void os::naked_yield() {
3815   sched_yield();
3816 }
3817 
3818 ////////////////////////////////////////////////////////////////////////////////
3819 // thread priority support
3820 
3821 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3822 // only supports dynamic priority, static priority must be zero. For real-time
3823 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3824 // However, for large multi-threaded applications, SCHED_RR is not only slower
3825 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3826 // of 5 runs - Sep 2005).
3827 //
3828 // The following code actually changes the niceness of kernel-thread/LWP. It
3829 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3830 // not the entire user process, and user level threads are 1:1 mapped to kernel
3831 // threads. It has always been the case, but could change in the future. For
3832 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3833 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3834 
3835 int os::java_to_os_priority[CriticalPriority + 1] = {
3836   19,              // 0 Entry should never be used
3837 
3838    4,              // 1 MinPriority
3839    3,              // 2
3840    2,              // 3
3841 
3842    1,              // 4
3843    0,              // 5 NormPriority
3844   -1,              // 6
3845 
3846   -2,              // 7
3847   -3,              // 8
3848   -4,              // 9 NearMaxPriority
3849 
3850   -5,              // 10 MaxPriority
3851 
3852   -5               // 11 CriticalPriority
3853 };
3854 
3855 static int prio_init() {
3856   if (ThreadPriorityPolicy == 1) {
3857     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3858     // if effective uid is not root. Perhaps, a more elegant way of doing
3859     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3860     if (geteuid() != 0) {
3861       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3862         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3863       }
3864       ThreadPriorityPolicy = 0;
3865     }
3866   }
3867   if (UseCriticalJavaThreadPriority) {
3868     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3869   }
3870   return 0;
3871 }
3872 
3873 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3874   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3875 
3876   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3877   return (ret == 0) ? OS_OK : OS_ERR;
3878 }
3879 
3880 OSReturn os::get_native_priority(const Thread* const thread,
3881                                  int *priority_ptr) {
3882   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3883     *priority_ptr = java_to_os_priority[NormPriority];
3884     return OS_OK;
3885   }
3886 
3887   errno = 0;
3888   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3889   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3890 }
3891 
3892 // Hint to the underlying OS that a task switch would not be good.
3893 // Void return because it's a hint and can fail.
3894 void os::hint_no_preempt() {}
3895 
3896 ////////////////////////////////////////////////////////////////////////////////
3897 // suspend/resume support
3898 
3899 //  the low-level signal-based suspend/resume support is a remnant from the
3900 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3901 //  within hotspot. Now there is a single use-case for this:
3902 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3903 //      that runs in the watcher thread.
3904 //  The remaining code is greatly simplified from the more general suspension
3905 //  code that used to be used.
3906 //
3907 //  The protocol is quite simple:
3908 //  - suspend:
3909 //      - sends a signal to the target thread
3910 //      - polls the suspend state of the osthread using a yield loop
3911 //      - target thread signal handler (SR_handler) sets suspend state
3912 //        and blocks in sigsuspend until continued
3913 //  - resume:
3914 //      - sets target osthread state to continue
3915 //      - sends signal to end the sigsuspend loop in the SR_handler
3916 //
3917 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3918 
3919 static void resume_clear_context(OSThread *osthread) {
3920   osthread->set_ucontext(NULL);
3921   osthread->set_siginfo(NULL);
3922 }
3923 
3924 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3925                                  ucontext_t* context) {
3926   osthread->set_ucontext(context);
3927   osthread->set_siginfo(siginfo);
3928 }
3929 
3930 // Handler function invoked when a thread's execution is suspended or
3931 // resumed. We have to be careful that only async-safe functions are
3932 // called here (Note: most pthread functions are not async safe and
3933 // should be avoided.)
3934 //
3935 // Note: sigwait() is a more natural fit than sigsuspend() from an
3936 // interface point of view, but sigwait() prevents the signal hander
3937 // from being run. libpthread would get very confused by not having
3938 // its signal handlers run and prevents sigwait()'s use with the
3939 // mutex granting granting signal.
3940 //
3941 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3942 //
3943 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3944   // Save and restore errno to avoid confusing native code with EINTR
3945   // after sigsuspend.
3946   int old_errno = errno;
3947 
3948   Thread* thread = Thread::current();
3949   OSThread* osthread = thread->osthread();
3950   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3951 
3952   os::SuspendResume::State current = osthread->sr.state();
3953   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3954     suspend_save_context(osthread, siginfo, context);
3955 
3956     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3957     os::SuspendResume::State state = osthread->sr.suspended();
3958     if (state == os::SuspendResume::SR_SUSPENDED) {
3959       sigset_t suspend_set;  // signals for sigsuspend()
3960 
3961       // get current set of blocked signals and unblock resume signal
3962       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3963       sigdelset(&suspend_set, SR_signum);
3964 
3965       sr_semaphore.signal();
3966       // wait here until we are resumed
3967       while (1) {
3968         sigsuspend(&suspend_set);
3969 
3970         os::SuspendResume::State result = osthread->sr.running();
3971         if (result == os::SuspendResume::SR_RUNNING) {
3972           sr_semaphore.signal();
3973           break;
3974         }
3975       }
3976 
3977     } else if (state == os::SuspendResume::SR_RUNNING) {
3978       // request was cancelled, continue
3979     } else {
3980       ShouldNotReachHere();
3981     }
3982 
3983     resume_clear_context(osthread);
3984   } else if (current == os::SuspendResume::SR_RUNNING) {
3985     // request was cancelled, continue
3986   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3987     // ignore
3988   } else {
3989     // ignore
3990   }
3991 
3992   errno = old_errno;
3993 }
3994 
3995 
3996 static int SR_initialize() {
3997   struct sigaction act;
3998   char *s;
3999   // Get signal number to use for suspend/resume
4000   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4001     int sig = ::strtol(s, 0, 10);
4002     if (sig > 0 || sig < _NSIG) {
4003       SR_signum = sig;
4004     }
4005   }
4006 
4007   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4008          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4009 
4010   sigemptyset(&SR_sigset);
4011   sigaddset(&SR_sigset, SR_signum);
4012 
4013   // Set up signal handler for suspend/resume
4014   act.sa_flags = SA_RESTART|SA_SIGINFO;
4015   act.sa_handler = (void (*)(int)) SR_handler;
4016 
4017   // SR_signum is blocked by default.
4018   // 4528190 - We also need to block pthread restart signal (32 on all
4019   // supported Linux platforms). Note that LinuxThreads need to block
4020   // this signal for all threads to work properly. So we don't have
4021   // to use hard-coded signal number when setting up the mask.
4022   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4023 
4024   if (sigaction(SR_signum, &act, 0) == -1) {
4025     return -1;
4026   }
4027 
4028   // Save signal flag
4029   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4030   return 0;
4031 }
4032 
4033 static int sr_notify(OSThread* osthread) {
4034   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4035   assert_status(status == 0, status, "pthread_kill");
4036   return status;
4037 }
4038 
4039 // "Randomly" selected value for how long we want to spin
4040 // before bailing out on suspending a thread, also how often
4041 // we send a signal to a thread we want to resume
4042 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4043 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4044 
4045 // returns true on success and false on error - really an error is fatal
4046 // but this seems the normal response to library errors
4047 static bool do_suspend(OSThread* osthread) {
4048   assert(osthread->sr.is_running(), "thread should be running");
4049   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4050 
4051   // mark as suspended and send signal
4052   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4053     // failed to switch, state wasn't running?
4054     ShouldNotReachHere();
4055     return false;
4056   }
4057 
4058   if (sr_notify(osthread) != 0) {
4059     ShouldNotReachHere();
4060   }
4061 
4062   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4063   while (true) {
4064     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4065       break;
4066     } else {
4067       // timeout
4068       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4069       if (cancelled == os::SuspendResume::SR_RUNNING) {
4070         return false;
4071       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4072         // make sure that we consume the signal on the semaphore as well
4073         sr_semaphore.wait();
4074         break;
4075       } else {
4076         ShouldNotReachHere();
4077         return false;
4078       }
4079     }
4080   }
4081 
4082   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4083   return true;
4084 }
4085 
4086 static void do_resume(OSThread* osthread) {
4087   assert(osthread->sr.is_suspended(), "thread should be suspended");
4088   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4089 
4090   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4091     // failed to switch to WAKEUP_REQUEST
4092     ShouldNotReachHere();
4093     return;
4094   }
4095 
4096   while (true) {
4097     if (sr_notify(osthread) == 0) {
4098       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4099         if (osthread->sr.is_running()) {
4100           return;
4101         }
4102       }
4103     } else {
4104       ShouldNotReachHere();
4105     }
4106   }
4107 
4108   guarantee(osthread->sr.is_running(), "Must be running!");
4109 }
4110 
4111 ///////////////////////////////////////////////////////////////////////////////////
4112 // signal handling (except suspend/resume)
4113 
4114 // This routine may be used by user applications as a "hook" to catch signals.
4115 // The user-defined signal handler must pass unrecognized signals to this
4116 // routine, and if it returns true (non-zero), then the signal handler must
4117 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4118 // routine will never retun false (zero), but instead will execute a VM panic
4119 // routine kill the process.
4120 //
4121 // If this routine returns false, it is OK to call it again.  This allows
4122 // the user-defined signal handler to perform checks either before or after
4123 // the VM performs its own checks.  Naturally, the user code would be making
4124 // a serious error if it tried to handle an exception (such as a null check
4125 // or breakpoint) that the VM was generating for its own correct operation.
4126 //
4127 // This routine may recognize any of the following kinds of signals:
4128 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4129 // It should be consulted by handlers for any of those signals.
4130 //
4131 // The caller of this routine must pass in the three arguments supplied
4132 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4133 // field of the structure passed to sigaction().  This routine assumes that
4134 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4135 //
4136 // Note that the VM will print warnings if it detects conflicting signal
4137 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4138 //
4139 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4140                                                  siginfo_t* siginfo,
4141                                                  void* ucontext,
4142                                                  int abort_if_unrecognized);
4143 
4144 void signalHandler(int sig, siginfo_t* info, void* uc) {
4145   assert(info != NULL && uc != NULL, "it must be old kernel");
4146   int orig_errno = errno;  // Preserve errno value over signal handler.
4147   JVM_handle_linux_signal(sig, info, uc, true);
4148   errno = orig_errno;
4149 }
4150 
4151 
4152 // This boolean allows users to forward their own non-matching signals
4153 // to JVM_handle_linux_signal, harmlessly.
4154 bool os::Linux::signal_handlers_are_installed = false;
4155 
4156 // For signal-chaining
4157 struct sigaction os::Linux::sigact[MAXSIGNUM];
4158 unsigned int os::Linux::sigs = 0;
4159 bool os::Linux::libjsig_is_loaded = false;
4160 typedef struct sigaction *(*get_signal_t)(int);
4161 get_signal_t os::Linux::get_signal_action = NULL;
4162 
4163 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4164   struct sigaction *actp = NULL;
4165 
4166   if (libjsig_is_loaded) {
4167     // Retrieve the old signal handler from libjsig
4168     actp = (*get_signal_action)(sig);
4169   }
4170   if (actp == NULL) {
4171     // Retrieve the preinstalled signal handler from jvm
4172     actp = get_preinstalled_handler(sig);
4173   }
4174 
4175   return actp;
4176 }
4177 
4178 static bool call_chained_handler(struct sigaction *actp, int sig,
4179                                  siginfo_t *siginfo, void *context) {
4180   // Call the old signal handler
4181   if (actp->sa_handler == SIG_DFL) {
4182     // It's more reasonable to let jvm treat it as an unexpected exception
4183     // instead of taking the default action.
4184     return false;
4185   } else if (actp->sa_handler != SIG_IGN) {
4186     if ((actp->sa_flags & SA_NODEFER) == 0) {
4187       // automaticlly block the signal
4188       sigaddset(&(actp->sa_mask), sig);
4189     }
4190 
4191     sa_handler_t hand;
4192     sa_sigaction_t sa;
4193     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4194     // retrieve the chained handler
4195     if (siginfo_flag_set) {
4196       sa = actp->sa_sigaction;
4197     } else {
4198       hand = actp->sa_handler;
4199     }
4200 
4201     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4202       actp->sa_handler = SIG_DFL;
4203     }
4204 
4205     // try to honor the signal mask
4206     sigset_t oset;
4207     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4208 
4209     // call into the chained handler
4210     if (siginfo_flag_set) {
4211       (*sa)(sig, siginfo, context);
4212     } else {
4213       (*hand)(sig);
4214     }
4215 
4216     // restore the signal mask
4217     pthread_sigmask(SIG_SETMASK, &oset, 0);
4218   }
4219   // Tell jvm's signal handler the signal is taken care of.
4220   return true;
4221 }
4222 
4223 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4224   bool chained = false;
4225   // signal-chaining
4226   if (UseSignalChaining) {
4227     struct sigaction *actp = get_chained_signal_action(sig);
4228     if (actp != NULL) {
4229       chained = call_chained_handler(actp, sig, siginfo, context);
4230     }
4231   }
4232   return chained;
4233 }
4234 
4235 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4236   if ((((unsigned int)1 << sig) & sigs) != 0) {
4237     return &sigact[sig];
4238   }
4239   return NULL;
4240 }
4241 
4242 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4243   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4244   sigact[sig] = oldAct;
4245   sigs |= (unsigned int)1 << sig;
4246 }
4247 
4248 // for diagnostic
4249 int os::Linux::sigflags[MAXSIGNUM];
4250 
4251 int os::Linux::get_our_sigflags(int sig) {
4252   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4253   return sigflags[sig];
4254 }
4255 
4256 void os::Linux::set_our_sigflags(int sig, int flags) {
4257   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4258   sigflags[sig] = flags;
4259 }
4260 
4261 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4262   // Check for overwrite.
4263   struct sigaction oldAct;
4264   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4265 
4266   void* oldhand = oldAct.sa_sigaction
4267                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4268                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4269   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4270       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4271       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4272     if (AllowUserSignalHandlers || !set_installed) {
4273       // Do not overwrite; user takes responsibility to forward to us.
4274       return;
4275     } else if (UseSignalChaining) {
4276       // save the old handler in jvm
4277       save_preinstalled_handler(sig, oldAct);
4278       // libjsig also interposes the sigaction() call below and saves the
4279       // old sigaction on it own.
4280     } else {
4281       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4282                     "%#lx for signal %d.", (long)oldhand, sig));
4283     }
4284   }
4285 
4286   struct sigaction sigAct;
4287   sigfillset(&(sigAct.sa_mask));
4288   sigAct.sa_handler = SIG_DFL;
4289   if (!set_installed) {
4290     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4291   } else {
4292     sigAct.sa_sigaction = signalHandler;
4293     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4294   }
4295   // Save flags, which are set by ours
4296   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4297   sigflags[sig] = sigAct.sa_flags;
4298 
4299   int ret = sigaction(sig, &sigAct, &oldAct);
4300   assert(ret == 0, "check");
4301 
4302   void* oldhand2  = oldAct.sa_sigaction
4303                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4304                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4305   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4306 }
4307 
4308 // install signal handlers for signals that HotSpot needs to
4309 // handle in order to support Java-level exception handling.
4310 
4311 void os::Linux::install_signal_handlers() {
4312   if (!signal_handlers_are_installed) {
4313     signal_handlers_are_installed = true;
4314 
4315     // signal-chaining
4316     typedef void (*signal_setting_t)();
4317     signal_setting_t begin_signal_setting = NULL;
4318     signal_setting_t end_signal_setting = NULL;
4319     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4320                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4321     if (begin_signal_setting != NULL) {
4322       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4323                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4324       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4325                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4326       libjsig_is_loaded = true;
4327       assert(UseSignalChaining, "should enable signal-chaining");
4328     }
4329     if (libjsig_is_loaded) {
4330       // Tell libjsig jvm is setting signal handlers
4331       (*begin_signal_setting)();
4332     }
4333 
4334     set_signal_handler(SIGSEGV, true);
4335     set_signal_handler(SIGPIPE, true);
4336     set_signal_handler(SIGBUS, true);
4337     set_signal_handler(SIGILL, true);
4338     set_signal_handler(SIGFPE, true);
4339 #if defined(PPC64)
4340     set_signal_handler(SIGTRAP, true);
4341 #endif
4342     set_signal_handler(SIGXFSZ, true);
4343 
4344     if (libjsig_is_loaded) {
4345       // Tell libjsig jvm finishes setting signal handlers
4346       (*end_signal_setting)();
4347     }
4348 
4349     // We don't activate signal checker if libjsig is in place, we trust ourselves
4350     // and if UserSignalHandler is installed all bets are off.
4351     // Log that signal checking is off only if -verbose:jni is specified.
4352     if (CheckJNICalls) {
4353       if (libjsig_is_loaded) {
4354         if (PrintJNIResolving) {
4355           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4356         }
4357         check_signals = false;
4358       }
4359       if (AllowUserSignalHandlers) {
4360         if (PrintJNIResolving) {
4361           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4362         }
4363         check_signals = false;
4364       }
4365     }
4366   }
4367 }
4368 
4369 // This is the fastest way to get thread cpu time on Linux.
4370 // Returns cpu time (user+sys) for any thread, not only for current.
4371 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4372 // It might work on 2.6.10+ with a special kernel/glibc patch.
4373 // For reference, please, see IEEE Std 1003.1-2004:
4374 //   http://www.unix.org/single_unix_specification
4375 
4376 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4377   struct timespec tp;
4378   int rc = os::Linux::clock_gettime(clockid, &tp);
4379   assert(rc == 0, "clock_gettime is expected to return 0 code");
4380 
4381   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4382 }
4383 
4384 /////
4385 // glibc on Linux platform uses non-documented flag
4386 // to indicate, that some special sort of signal
4387 // trampoline is used.
4388 // We will never set this flag, and we should
4389 // ignore this flag in our diagnostic
4390 #ifdef SIGNIFICANT_SIGNAL_MASK
4391   #undef SIGNIFICANT_SIGNAL_MASK
4392 #endif
4393 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4394 
4395 static const char* get_signal_handler_name(address handler,
4396                                            char* buf, int buflen) {
4397   int offset;
4398   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4399   if (found) {
4400     // skip directory names
4401     const char *p1, *p2;
4402     p1 = buf;
4403     size_t len = strlen(os::file_separator());
4404     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4405     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4406   } else {
4407     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4408   }
4409   return buf;
4410 }
4411 
4412 static void print_signal_handler(outputStream* st, int sig,
4413                                  char* buf, size_t buflen) {
4414   struct sigaction sa;
4415 
4416   sigaction(sig, NULL, &sa);
4417 
4418   // See comment for SIGNIFICANT_SIGNAL_MASK define
4419   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4420 
4421   st->print("%s: ", os::exception_name(sig, buf, buflen));
4422 
4423   address handler = (sa.sa_flags & SA_SIGINFO)
4424     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4425     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4426 
4427   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4428     st->print("SIG_DFL");
4429   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4430     st->print("SIG_IGN");
4431   } else {
4432     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4433   }
4434 
4435   st->print(", sa_mask[0]=");
4436   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4437 
4438   address rh = VMError::get_resetted_sighandler(sig);
4439   // May be, handler was resetted by VMError?
4440   if (rh != NULL) {
4441     handler = rh;
4442     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4443   }
4444 
4445   st->print(", sa_flags=");
4446   os::Posix::print_sa_flags(st, sa.sa_flags);
4447 
4448   // Check: is it our handler?
4449   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4450       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4451     // It is our signal handler
4452     // check for flags, reset system-used one!
4453     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4454       st->print(
4455                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4456                 os::Linux::get_our_sigflags(sig));
4457     }
4458   }
4459   st->cr();
4460 }
4461 
4462 
4463 #define DO_SIGNAL_CHECK(sig)                      \
4464   do {                                            \
4465     if (!sigismember(&check_signal_done, sig)) {  \
4466       os::Linux::check_signal_handler(sig);       \
4467     }                                             \
4468   } while (0)
4469 
4470 // This method is a periodic task to check for misbehaving JNI applications
4471 // under CheckJNI, we can add any periodic checks here
4472 
4473 void os::run_periodic_checks() {
4474   if (check_signals == false) return;
4475 
4476   // SEGV and BUS if overridden could potentially prevent
4477   // generation of hs*.log in the event of a crash, debugging
4478   // such a case can be very challenging, so we absolutely
4479   // check the following for a good measure:
4480   DO_SIGNAL_CHECK(SIGSEGV);
4481   DO_SIGNAL_CHECK(SIGILL);
4482   DO_SIGNAL_CHECK(SIGFPE);
4483   DO_SIGNAL_CHECK(SIGBUS);
4484   DO_SIGNAL_CHECK(SIGPIPE);
4485   DO_SIGNAL_CHECK(SIGXFSZ);
4486 #if defined(PPC64)
4487   DO_SIGNAL_CHECK(SIGTRAP);
4488 #endif
4489 
4490   // ReduceSignalUsage allows the user to override these handlers
4491   // see comments at the very top and jvm_solaris.h
4492   if (!ReduceSignalUsage) {
4493     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4494     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4495     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4496     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4497   }
4498 
4499   DO_SIGNAL_CHECK(SR_signum);
4500   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4501 }
4502 
4503 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4504 
4505 static os_sigaction_t os_sigaction = NULL;
4506 
4507 void os::Linux::check_signal_handler(int sig) {
4508   char buf[O_BUFLEN];
4509   address jvmHandler = NULL;
4510 
4511 
4512   struct sigaction act;
4513   if (os_sigaction == NULL) {
4514     // only trust the default sigaction, in case it has been interposed
4515     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4516     if (os_sigaction == NULL) return;
4517   }
4518 
4519   os_sigaction(sig, (struct sigaction*)NULL, &act);
4520 
4521 
4522   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4523 
4524   address thisHandler = (act.sa_flags & SA_SIGINFO)
4525     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4526     : CAST_FROM_FN_PTR(address, act.sa_handler);
4527 
4528 
4529   switch (sig) {
4530   case SIGSEGV:
4531   case SIGBUS:
4532   case SIGFPE:
4533   case SIGPIPE:
4534   case SIGILL:
4535   case SIGXFSZ:
4536     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4537     break;
4538 
4539   case SHUTDOWN1_SIGNAL:
4540   case SHUTDOWN2_SIGNAL:
4541   case SHUTDOWN3_SIGNAL:
4542   case BREAK_SIGNAL:
4543     jvmHandler = (address)user_handler();
4544     break;
4545 
4546   case INTERRUPT_SIGNAL:
4547     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4548     break;
4549 
4550   default:
4551     if (sig == SR_signum) {
4552       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4553     } else {
4554       return;
4555     }
4556     break;
4557   }
4558 
4559   if (thisHandler != jvmHandler) {
4560     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4561     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4562     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4563     // No need to check this sig any longer
4564     sigaddset(&check_signal_done, sig);
4565     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4566     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4567       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4568                     exception_name(sig, buf, O_BUFLEN));
4569     }
4570   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4571     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4572     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4573     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4574     // No need to check this sig any longer
4575     sigaddset(&check_signal_done, sig);
4576   }
4577 
4578   // Dump all the signal
4579   if (sigismember(&check_signal_done, sig)) {
4580     print_signal_handlers(tty, buf, O_BUFLEN);
4581   }
4582 }
4583 
4584 extern void report_error(char* file_name, int line_no, char* title,
4585                          char* format, ...);
4586 
4587 extern bool signal_name(int signo, char* buf, size_t len);
4588 
4589 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4590   if (0 < exception_code && exception_code <= SIGRTMAX) {
4591     // signal
4592     if (!signal_name(exception_code, buf, size)) {
4593       jio_snprintf(buf, size, "SIG%d", exception_code);
4594     }
4595     return buf;
4596   } else {
4597     return NULL;
4598   }
4599 }
4600 
4601 // this is called _before_ the most of global arguments have been parsed
4602 void os::init(void) {
4603   char dummy;   // used to get a guess on initial stack address
4604 //  first_hrtime = gethrtime();
4605 
4606   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4607 
4608   init_random(1234567);
4609 
4610   ThreadCritical::initialize();
4611 
4612   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4613   if (Linux::page_size() == -1) {
4614     fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4615                   strerror(errno)));
4616   }
4617   init_page_sizes((size_t) Linux::page_size());
4618 
4619   Linux::initialize_system_info();
4620 
4621   // main_thread points to the aboriginal thread
4622   Linux::_main_thread = pthread_self();
4623 
4624   Linux::clock_init();
4625   initial_time_count = javaTimeNanos();
4626 
4627   // pthread_condattr initialization for monotonic clock
4628   int status;
4629   pthread_condattr_t* _condattr = os::Linux::condAttr();
4630   if ((status = pthread_condattr_init(_condattr)) != 0) {
4631     fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4632   }
4633   // Only set the clock if CLOCK_MONOTONIC is available
4634   if (os::supports_monotonic_clock()) {
4635     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4636       if (status == EINVAL) {
4637         warning("Unable to use monotonic clock with relative timed-waits" \
4638                 " - changes to the time-of-day clock may have adverse affects");
4639       } else {
4640         fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4641       }
4642     }
4643   }
4644   // else it defaults to CLOCK_REALTIME
4645 
4646   // If the pagesize of the VM is greater than 8K determine the appropriate
4647   // number of initial guard pages.  The user can change this with the
4648   // command line arguments, if needed.
4649   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4650     StackYellowPages = 1;
4651     StackRedPages = 1;
4652     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4653   }
4654 
4655   // retrieve entry point for pthread_setname_np
4656   Linux::_pthread_setname_np =
4657     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4658 
4659 }
4660 
4661 // To install functions for atexit system call
4662 extern "C" {
4663   static void perfMemory_exit_helper() {
4664     perfMemory_exit();
4665   }
4666 }
4667 
4668 // this is called _after_ the global arguments have been parsed
4669 jint os::init_2(void) {
4670   Linux::fast_thread_clock_init();
4671 
4672   // Allocate a single page and mark it as readable for safepoint polling
4673   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4674   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4675 
4676   os::set_polling_page(polling_page);
4677 
4678 #ifndef PRODUCT
4679   if (Verbose && PrintMiscellaneous) {
4680     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4681                (intptr_t)polling_page);
4682   }
4683 #endif
4684 
4685   if (!UseMembar) {
4686     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4687     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4688     os::set_memory_serialize_page(mem_serialize_page);
4689 
4690 #ifndef PRODUCT
4691     if (Verbose && PrintMiscellaneous) {
4692       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n",
4693                  (intptr_t)mem_serialize_page);
4694     }
4695 #endif
4696   }
4697 
4698   // initialize suspend/resume support - must do this before signal_sets_init()
4699   if (SR_initialize() != 0) {
4700     perror("SR_initialize failed");
4701     return JNI_ERR;
4702   }
4703 
4704   Linux::signal_sets_init();
4705   Linux::install_signal_handlers();
4706 
4707   // Check minimum allowable stack size for thread creation and to initialize
4708   // the java system classes, including StackOverflowError - depends on page
4709   // size.  Add a page for compiler2 recursion in main thread.
4710   // Add in 2*BytesPerWord times page size to account for VM stack during
4711   // class initialization depending on 32 or 64 bit VM.
4712   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4713                                       (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4714                                       (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4715 
4716   size_t threadStackSizeInBytes = ThreadStackSize * K;
4717   if (threadStackSizeInBytes != 0 &&
4718       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4719     tty->print_cr("\nThe stack size specified is too small, "
4720                   "Specify at least %dk",
4721                   os::Linux::min_stack_allowed/ K);
4722     return JNI_ERR;
4723   }
4724 
4725   // Make the stack size a multiple of the page size so that
4726   // the yellow/red zones can be guarded.
4727   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4728                                                 vm_page_size()));
4729 
4730   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4731 
4732 #if defined(IA32)
4733   workaround_expand_exec_shield_cs_limit();
4734 #endif
4735 
4736   Linux::libpthread_init();
4737   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4738     tty->print_cr("[HotSpot is running with %s, %s]\n",
4739                   Linux::glibc_version(), Linux::libpthread_version());
4740   }
4741 
4742   if (UseNUMA) {
4743     if (!Linux::libnuma_init()) {
4744       UseNUMA = false;
4745     } else {
4746       if ((Linux::numa_max_node() < 1)) {
4747         // There's only one node(they start from 0), disable NUMA.
4748         UseNUMA = false;
4749       }
4750     }
4751     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4752     // we can make the adaptive lgrp chunk resizing work. If the user specified
4753     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4754     // disable adaptive resizing.
4755     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4756       if (FLAG_IS_DEFAULT(UseNUMA)) {
4757         UseNUMA = false;
4758       } else {
4759         if (FLAG_IS_DEFAULT(UseLargePages) &&
4760             FLAG_IS_DEFAULT(UseSHM) &&
4761             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4762           UseLargePages = false;
4763         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4764           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4765           UseAdaptiveSizePolicy = false;
4766           UseAdaptiveNUMAChunkSizing = false;
4767         }
4768       }
4769     }
4770     if (!UseNUMA && ForceNUMA) {
4771       UseNUMA = true;
4772     }
4773   }
4774 
4775   if (MaxFDLimit) {
4776     // set the number of file descriptors to max. print out error
4777     // if getrlimit/setrlimit fails but continue regardless.
4778     struct rlimit nbr_files;
4779     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4780     if (status != 0) {
4781       if (PrintMiscellaneous && (Verbose || WizardMode)) {
4782         perror("os::init_2 getrlimit failed");
4783       }
4784     } else {
4785       nbr_files.rlim_cur = nbr_files.rlim_max;
4786       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4787       if (status != 0) {
4788         if (PrintMiscellaneous && (Verbose || WizardMode)) {
4789           perror("os::init_2 setrlimit failed");
4790         }
4791       }
4792     }
4793   }
4794 
4795   // Initialize lock used to serialize thread creation (see os::create_thread)
4796   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4797 
4798   // at-exit methods are called in the reverse order of their registration.
4799   // atexit functions are called on return from main or as a result of a
4800   // call to exit(3C). There can be only 32 of these functions registered
4801   // and atexit() does not set errno.
4802 
4803   if (PerfAllowAtExitRegistration) {
4804     // only register atexit functions if PerfAllowAtExitRegistration is set.
4805     // atexit functions can be delayed until process exit time, which
4806     // can be problematic for embedded VM situations. Embedded VMs should
4807     // call DestroyJavaVM() to assure that VM resources are released.
4808 
4809     // note: perfMemory_exit_helper atexit function may be removed in
4810     // the future if the appropriate cleanup code can be added to the
4811     // VM_Exit VMOperation's doit method.
4812     if (atexit(perfMemory_exit_helper) != 0) {
4813       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4814     }
4815   }
4816 
4817   // initialize thread priority policy
4818   prio_init();
4819 
4820   return JNI_OK;
4821 }
4822 
4823 // Mark the polling page as unreadable
4824 void os::make_polling_page_unreadable(void) {
4825   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4826     fatal("Could not disable polling page");
4827   }
4828 }
4829 
4830 // Mark the polling page as readable
4831 void os::make_polling_page_readable(void) {
4832   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4833     fatal("Could not enable polling page");
4834   }
4835 }
4836 
4837 int os::active_processor_count() {
4838   // Linux doesn't yet have a (official) notion of processor sets,
4839   // so just return the number of online processors.
4840   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4841   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4842   return online_cpus;
4843 }
4844 
4845 void os::set_native_thread_name(const char *name) {
4846   if (Linux::_pthread_setname_np) {
4847     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4848     snprintf(buf, sizeof(buf), "%s", name);
4849     buf[sizeof(buf) - 1] = '\0';
4850     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4851     // ERANGE should not happen; all other errors should just be ignored.
4852     assert(rc != ERANGE, "pthread_setname_np failed");
4853   }
4854 }
4855 
4856 bool os::distribute_processes(uint length, uint* distribution) {
4857   // Not yet implemented.
4858   return false;
4859 }
4860 
4861 bool os::bind_to_processor(uint processor_id) {
4862   // Not yet implemented.
4863   return false;
4864 }
4865 
4866 ///
4867 
4868 void os::SuspendedThreadTask::internal_do_task() {
4869   if (do_suspend(_thread->osthread())) {
4870     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4871     do_task(context);
4872     do_resume(_thread->osthread());
4873   }
4874 }
4875 
4876 class PcFetcher : public os::SuspendedThreadTask {
4877  public:
4878   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4879   ExtendedPC result();
4880  protected:
4881   void do_task(const os::SuspendedThreadTaskContext& context);
4882  private:
4883   ExtendedPC _epc;
4884 };
4885 
4886 ExtendedPC PcFetcher::result() {
4887   guarantee(is_done(), "task is not done yet.");
4888   return _epc;
4889 }
4890 
4891 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4892   Thread* thread = context.thread();
4893   OSThread* osthread = thread->osthread();
4894   if (osthread->ucontext() != NULL) {
4895     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4896   } else {
4897     // NULL context is unexpected, double-check this is the VMThread
4898     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4899   }
4900 }
4901 
4902 // Suspends the target using the signal mechanism and then grabs the PC before
4903 // resuming the target. Used by the flat-profiler only
4904 ExtendedPC os::get_thread_pc(Thread* thread) {
4905   // Make sure that it is called by the watcher for the VMThread
4906   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4907   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4908 
4909   PcFetcher fetcher(thread);
4910   fetcher.run();
4911   return fetcher.result();
4912 }
4913 
4914 ////////////////////////////////////////////////////////////////////////////////
4915 // debug support
4916 
4917 bool os::find(address addr, outputStream* st) {
4918   Dl_info dlinfo;
4919   memset(&dlinfo, 0, sizeof(dlinfo));
4920   if (dladdr(addr, &dlinfo) != 0) {
4921     st->print(PTR_FORMAT ": ", addr);
4922     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4923       st->print("%s+%#x", dlinfo.dli_sname,
4924                 addr - (intptr_t)dlinfo.dli_saddr);
4925     } else if (dlinfo.dli_fbase != NULL) {
4926       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4927     } else {
4928       st->print("<absolute address>");
4929     }
4930     if (dlinfo.dli_fname != NULL) {
4931       st->print(" in %s", dlinfo.dli_fname);
4932     }
4933     if (dlinfo.dli_fbase != NULL) {
4934       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4935     }
4936     st->cr();
4937 
4938     if (Verbose) {
4939       // decode some bytes around the PC
4940       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4941       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4942       address       lowest = (address) dlinfo.dli_sname;
4943       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4944       if (begin < lowest)  begin = lowest;
4945       Dl_info dlinfo2;
4946       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4947           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4948         end = (address) dlinfo2.dli_saddr;
4949       }
4950       Disassembler::decode(begin, end, st);
4951     }
4952     return true;
4953   }
4954   return false;
4955 }
4956 
4957 ////////////////////////////////////////////////////////////////////////////////
4958 // misc
4959 
4960 // This does not do anything on Linux. This is basically a hook for being
4961 // able to use structured exception handling (thread-local exception filters)
4962 // on, e.g., Win32.
4963 void
4964 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4965                          JavaCallArguments* args, Thread* thread) {
4966   f(value, method, args, thread);
4967 }
4968 
4969 void os::print_statistics() {
4970 }
4971 
4972 int os::message_box(const char* title, const char* message) {
4973   int i;
4974   fdStream err(defaultStream::error_fd());
4975   for (i = 0; i < 78; i++) err.print_raw("=");
4976   err.cr();
4977   err.print_raw_cr(title);
4978   for (i = 0; i < 78; i++) err.print_raw("-");
4979   err.cr();
4980   err.print_raw_cr(message);
4981   for (i = 0; i < 78; i++) err.print_raw("=");
4982   err.cr();
4983 
4984   char buf[16];
4985   // Prevent process from exiting upon "read error" without consuming all CPU
4986   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4987 
4988   return buf[0] == 'y' || buf[0] == 'Y';
4989 }
4990 
4991 int os::stat(const char *path, struct stat *sbuf) {
4992   char pathbuf[MAX_PATH];
4993   if (strlen(path) > MAX_PATH - 1) {
4994     errno = ENAMETOOLONG;
4995     return -1;
4996   }
4997   os::native_path(strcpy(pathbuf, path));
4998   return ::stat(pathbuf, sbuf);
4999 }
5000 
5001 bool os::check_heap(bool force) {
5002   return true;
5003 }
5004 
5005 // Is a (classpath) directory empty?
5006 bool os::dir_is_empty(const char* path) {
5007   DIR *dir = NULL;
5008   struct dirent *ptr;
5009 
5010   dir = opendir(path);
5011   if (dir == NULL) return true;
5012 
5013   // Scan the directory
5014   bool result = true;
5015   char buf[sizeof(struct dirent) + MAX_PATH];
5016   while (result && (ptr = ::readdir(dir)) != NULL) {
5017     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5018       result = false;
5019     }
5020   }
5021   closedir(dir);
5022   return result;
5023 }
5024 
5025 // This code originates from JDK's sysOpen and open64_w
5026 // from src/solaris/hpi/src/system_md.c
5027 
5028 int os::open(const char *path, int oflag, int mode) {
5029   if (strlen(path) > MAX_PATH - 1) {
5030     errno = ENAMETOOLONG;
5031     return -1;
5032   }
5033 
5034   // All file descriptors that are opened in the Java process and not
5035   // specifically destined for a subprocess should have the close-on-exec
5036   // flag set.  If we don't set it, then careless 3rd party native code
5037   // might fork and exec without closing all appropriate file descriptors
5038   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5039   // turn might:
5040   //
5041   // - cause end-of-file to fail to be detected on some file
5042   //   descriptors, resulting in mysterious hangs, or
5043   //
5044   // - might cause an fopen in the subprocess to fail on a system
5045   //   suffering from bug 1085341.
5046   //
5047   // (Yes, the default setting of the close-on-exec flag is a Unix
5048   // design flaw)
5049   //
5050   // See:
5051   // 1085341: 32-bit stdio routines should support file descriptors >255
5052   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5053   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5054   //
5055   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5056   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5057   // because it saves a system call and removes a small window where the flag
5058   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5059   // and we fall back to using FD_CLOEXEC (see below).
5060 #ifdef O_CLOEXEC
5061   oflag |= O_CLOEXEC;
5062 #endif
5063 
5064   int fd = ::open64(path, oflag, mode);
5065   if (fd == -1) return -1;
5066 
5067   //If the open succeeded, the file might still be a directory
5068   {
5069     struct stat64 buf64;
5070     int ret = ::fstat64(fd, &buf64);
5071     int st_mode = buf64.st_mode;
5072 
5073     if (ret != -1) {
5074       if ((st_mode & S_IFMT) == S_IFDIR) {
5075         errno = EISDIR;
5076         ::close(fd);
5077         return -1;
5078       }
5079     } else {
5080       ::close(fd);
5081       return -1;
5082     }
5083   }
5084 
5085 #ifdef FD_CLOEXEC
5086   // Validate that the use of the O_CLOEXEC flag on open above worked.
5087   // With recent kernels, we will perform this check exactly once.
5088   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5089   if (!O_CLOEXEC_is_known_to_work) {
5090     int flags = ::fcntl(fd, F_GETFD);
5091     if (flags != -1) {
5092       if ((flags & FD_CLOEXEC) != 0)
5093         O_CLOEXEC_is_known_to_work = 1;
5094       else
5095         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5096     }
5097   }
5098 #endif
5099 
5100   return fd;
5101 }
5102 
5103 
5104 // create binary file, rewriting existing file if required
5105 int os::create_binary_file(const char* path, bool rewrite_existing) {
5106   int oflags = O_WRONLY | O_CREAT;
5107   if (!rewrite_existing) {
5108     oflags |= O_EXCL;
5109   }
5110   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5111 }
5112 
5113 // return current position of file pointer
5114 jlong os::current_file_offset(int fd) {
5115   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5116 }
5117 
5118 // move file pointer to the specified offset
5119 jlong os::seek_to_file_offset(int fd, jlong offset) {
5120   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5121 }
5122 
5123 // This code originates from JDK's sysAvailable
5124 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5125 
5126 int os::available(int fd, jlong *bytes) {
5127   jlong cur, end;
5128   int mode;
5129   struct stat64 buf64;
5130 
5131   if (::fstat64(fd, &buf64) >= 0) {
5132     mode = buf64.st_mode;
5133     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5134       // XXX: is the following call interruptible? If so, this might
5135       // need to go through the INTERRUPT_IO() wrapper as for other
5136       // blocking, interruptible calls in this file.
5137       int n;
5138       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5139         *bytes = n;
5140         return 1;
5141       }
5142     }
5143   }
5144   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5145     return 0;
5146   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5147     return 0;
5148   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5149     return 0;
5150   }
5151   *bytes = end - cur;
5152   return 1;
5153 }
5154 
5155 // Map a block of memory.
5156 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5157                         char *addr, size_t bytes, bool read_only,
5158                         bool allow_exec) {
5159   int prot;
5160   int flags = MAP_PRIVATE;
5161 
5162   if (read_only) {
5163     prot = PROT_READ;
5164   } else {
5165     prot = PROT_READ | PROT_WRITE;
5166   }
5167 
5168   if (allow_exec) {
5169     prot |= PROT_EXEC;
5170   }
5171 
5172   if (addr != NULL) {
5173     flags |= MAP_FIXED;
5174   }
5175 
5176   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5177                                      fd, file_offset);
5178   if (mapped_address == MAP_FAILED) {
5179     return NULL;
5180   }
5181   return mapped_address;
5182 }
5183 
5184 
5185 // Remap a block of memory.
5186 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5187                           char *addr, size_t bytes, bool read_only,
5188                           bool allow_exec) {
5189   // same as map_memory() on this OS
5190   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5191                         allow_exec);
5192 }
5193 
5194 
5195 // Unmap a block of memory.
5196 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5197   return munmap(addr, bytes) == 0;
5198 }
5199 
5200 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5201 
5202 static clockid_t thread_cpu_clockid(Thread* thread) {
5203   pthread_t tid = thread->osthread()->pthread_id();
5204   clockid_t clockid;
5205 
5206   // Get thread clockid
5207   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5208   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5209   return clockid;
5210 }
5211 
5212 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5213 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5214 // of a thread.
5215 //
5216 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5217 // the fast estimate available on the platform.
5218 
5219 jlong os::current_thread_cpu_time() {
5220   if (os::Linux::supports_fast_thread_cpu_time()) {
5221     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5222   } else {
5223     // return user + sys since the cost is the same
5224     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5225   }
5226 }
5227 
5228 jlong os::thread_cpu_time(Thread* thread) {
5229   // consistent with what current_thread_cpu_time() returns
5230   if (os::Linux::supports_fast_thread_cpu_time()) {
5231     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5232   } else {
5233     return slow_thread_cpu_time(thread, true /* user + sys */);
5234   }
5235 }
5236 
5237 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5238   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5239     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5240   } else {
5241     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5242   }
5243 }
5244 
5245 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5246   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5247     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5248   } else {
5249     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5250   }
5251 }
5252 
5253 //  -1 on error.
5254 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5255   pid_t  tid = thread->osthread()->thread_id();
5256   char *s;
5257   char stat[2048];
5258   int statlen;
5259   char proc_name[64];
5260   int count;
5261   long sys_time, user_time;
5262   char cdummy;
5263   int idummy;
5264   long ldummy;
5265   FILE *fp;
5266 
5267   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5268   fp = fopen(proc_name, "r");
5269   if (fp == NULL) return -1;
5270   statlen = fread(stat, 1, 2047, fp);
5271   stat[statlen] = '\0';
5272   fclose(fp);
5273 
5274   // Skip pid and the command string. Note that we could be dealing with
5275   // weird command names, e.g. user could decide to rename java launcher
5276   // to "java 1.4.2 :)", then the stat file would look like
5277   //                1234 (java 1.4.2 :)) R ... ...
5278   // We don't really need to know the command string, just find the last
5279   // occurrence of ")" and then start parsing from there. See bug 4726580.
5280   s = strrchr(stat, ')');
5281   if (s == NULL) return -1;
5282 
5283   // Skip blank chars
5284   do { s++; } while (s && isspace(*s));
5285 
5286   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5287                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5288                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5289                  &user_time, &sys_time);
5290   if (count != 13) return -1;
5291   if (user_sys_cpu_time) {
5292     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5293   } else {
5294     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5295   }
5296 }
5297 
5298 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5299   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5300   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5301   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5302   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5303 }
5304 
5305 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5306   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5307   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5308   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5309   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5310 }
5311 
5312 bool os::is_thread_cpu_time_supported() {
5313   return true;
5314 }
5315 
5316 // System loadavg support.  Returns -1 if load average cannot be obtained.
5317 // Linux doesn't yet have a (official) notion of processor sets,
5318 // so just return the system wide load average.
5319 int os::loadavg(double loadavg[], int nelem) {
5320   return ::getloadavg(loadavg, nelem);
5321 }
5322 
5323 void os::pause() {
5324   char filename[MAX_PATH];
5325   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5326     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5327   } else {
5328     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5329   }
5330 
5331   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5332   if (fd != -1) {
5333     struct stat buf;
5334     ::close(fd);
5335     while (::stat(filename, &buf) == 0) {
5336       (void)::poll(NULL, 0, 100);
5337     }
5338   } else {
5339     jio_fprintf(stderr,
5340                 "Could not open pause file '%s', continuing immediately.\n", filename);
5341   }
5342 }
5343 
5344 
5345 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5346 // comment paragraphs are worth repeating here:
5347 //
5348 // Assumption:
5349 //    Only one parker can exist on an event, which is why we allocate
5350 //    them per-thread. Multiple unparkers can coexist.
5351 //
5352 // _Event serves as a restricted-range semaphore.
5353 //   -1 : thread is blocked, i.e. there is a waiter
5354 //    0 : neutral: thread is running or ready,
5355 //        could have been signaled after a wait started
5356 //    1 : signaled - thread is running or ready
5357 //
5358 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5359 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5360 // For specifics regarding the bug see GLIBC BUGID 261237 :
5361 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5362 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5363 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5364 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5365 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5366 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5367 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5368 // of libpthread avoids the problem, but isn't practical.
5369 //
5370 // Possible remedies:
5371 //
5372 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5373 //      This is palliative and probabilistic, however.  If the thread is preempted
5374 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5375 //      than the minimum period may have passed, and the abstime may be stale (in the
5376 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5377 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5378 //
5379 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5380 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5381 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5382 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5383 //      thread.
5384 //
5385 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5386 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5387 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5388 //      This also works well.  In fact it avoids kernel-level scalability impediments
5389 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5390 //      timers in a graceful fashion.
5391 //
5392 // 4.   When the abstime value is in the past it appears that control returns
5393 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5394 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5395 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5396 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5397 //      It may be possible to avoid reinitialization by checking the return
5398 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5399 //      condvar we must establish the invariant that cond_signal() is only called
5400 //      within critical sections protected by the adjunct mutex.  This prevents
5401 //      cond_signal() from "seeing" a condvar that's in the midst of being
5402 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5403 //      desirable signal-after-unlock optimization that avoids futile context switching.
5404 //
5405 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5406 //      structure when a condvar is used or initialized.  cond_destroy()  would
5407 //      release the helper structure.  Our reinitialize-after-timedwait fix
5408 //      put excessive stress on malloc/free and locks protecting the c-heap.
5409 //
5410 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5411 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5412 // and only enabling the work-around for vulnerable environments.
5413 
5414 // utility to compute the abstime argument to timedwait:
5415 // millis is the relative timeout time
5416 // abstime will be the absolute timeout time
5417 // TODO: replace compute_abstime() with unpackTime()
5418 
5419 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5420   if (millis < 0)  millis = 0;
5421 
5422   jlong seconds = millis / 1000;
5423   millis %= 1000;
5424   if (seconds > 50000000) { // see man cond_timedwait(3T)
5425     seconds = 50000000;
5426   }
5427 
5428   if (os::supports_monotonic_clock()) {
5429     struct timespec now;
5430     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5431     assert_status(status == 0, status, "clock_gettime");
5432     abstime->tv_sec = now.tv_sec  + seconds;
5433     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5434     if (nanos >= NANOSECS_PER_SEC) {
5435       abstime->tv_sec += 1;
5436       nanos -= NANOSECS_PER_SEC;
5437     }
5438     abstime->tv_nsec = nanos;
5439   } else {
5440     struct timeval now;
5441     int status = gettimeofday(&now, NULL);
5442     assert(status == 0, "gettimeofday");
5443     abstime->tv_sec = now.tv_sec  + seconds;
5444     long usec = now.tv_usec + millis * 1000;
5445     if (usec >= 1000000) {
5446       abstime->tv_sec += 1;
5447       usec -= 1000000;
5448     }
5449     abstime->tv_nsec = usec * 1000;
5450   }
5451   return abstime;
5452 }
5453 
5454 void os::PlatformEvent::park() {       // AKA "down()"
5455   // Transitions for _Event:
5456   //   -1 => -1 : illegal
5457   //    1 =>  0 : pass - return immediately
5458   //    0 => -1 : block; then set _Event to 0 before returning
5459 
5460   // Invariant: Only the thread associated with the Event/PlatformEvent
5461   // may call park().
5462   // TODO: assert that _Assoc != NULL or _Assoc == Self
5463   assert(_nParked == 0, "invariant");
5464 
5465   int v;
5466   for (;;) {
5467     v = _Event;
5468     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5469   }
5470   guarantee(v >= 0, "invariant");
5471   if (v == 0) {
5472     // Do this the hard way by blocking ...
5473     int status = pthread_mutex_lock(_mutex);
5474     assert_status(status == 0, status, "mutex_lock");
5475     guarantee(_nParked == 0, "invariant");
5476     ++_nParked;
5477     while (_Event < 0) {
5478       status = pthread_cond_wait(_cond, _mutex);
5479       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5480       // Treat this the same as if the wait was interrupted
5481       if (status == ETIME) { status = EINTR; }
5482       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5483     }
5484     --_nParked;
5485 
5486     _Event = 0;
5487     status = pthread_mutex_unlock(_mutex);
5488     assert_status(status == 0, status, "mutex_unlock");
5489     // Paranoia to ensure our locked and lock-free paths interact
5490     // correctly with each other.
5491     OrderAccess::fence();
5492   }
5493   guarantee(_Event >= 0, "invariant");
5494 }
5495 
5496 int os::PlatformEvent::park(jlong millis) {
5497   // Transitions for _Event:
5498   //   -1 => -1 : illegal
5499   //    1 =>  0 : pass - return immediately
5500   //    0 => -1 : block; then set _Event to 0 before returning
5501 
5502   guarantee(_nParked == 0, "invariant");
5503 
5504   int v;
5505   for (;;) {
5506     v = _Event;
5507     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5508   }
5509   guarantee(v >= 0, "invariant");
5510   if (v != 0) return OS_OK;
5511 
5512   // We do this the hard way, by blocking the thread.
5513   // Consider enforcing a minimum timeout value.
5514   struct timespec abst;
5515   compute_abstime(&abst, millis);
5516 
5517   int ret = OS_TIMEOUT;
5518   int status = pthread_mutex_lock(_mutex);
5519   assert_status(status == 0, status, "mutex_lock");
5520   guarantee(_nParked == 0, "invariant");
5521   ++_nParked;
5522 
5523   // Object.wait(timo) will return because of
5524   // (a) notification
5525   // (b) timeout
5526   // (c) thread.interrupt
5527   //
5528   // Thread.interrupt and object.notify{All} both call Event::set.
5529   // That is, we treat thread.interrupt as a special case of notification.
5530   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5531   // We assume all ETIME returns are valid.
5532   //
5533   // TODO: properly differentiate simultaneous notify+interrupt.
5534   // In that case, we should propagate the notify to another waiter.
5535 
5536   while (_Event < 0) {
5537     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5538     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5539       pthread_cond_destroy(_cond);
5540       pthread_cond_init(_cond, os::Linux::condAttr());
5541     }
5542     assert_status(status == 0 || status == EINTR ||
5543                   status == ETIME || status == ETIMEDOUT,
5544                   status, "cond_timedwait");
5545     if (!FilterSpuriousWakeups) break;                 // previous semantics
5546     if (status == ETIME || status == ETIMEDOUT) break;
5547     // We consume and ignore EINTR and spurious wakeups.
5548   }
5549   --_nParked;
5550   if (_Event >= 0) {
5551     ret = OS_OK;
5552   }
5553   _Event = 0;
5554   status = pthread_mutex_unlock(_mutex);
5555   assert_status(status == 0, status, "mutex_unlock");
5556   assert(_nParked == 0, "invariant");
5557   // Paranoia to ensure our locked and lock-free paths interact
5558   // correctly with each other.
5559   OrderAccess::fence();
5560   return ret;
5561 }
5562 
5563 void os::PlatformEvent::unpark() {
5564   // Transitions for _Event:
5565   //    0 => 1 : just return
5566   //    1 => 1 : just return
5567   //   -1 => either 0 or 1; must signal target thread
5568   //         That is, we can safely transition _Event from -1 to either
5569   //         0 or 1.
5570   // See also: "Semaphores in Plan 9" by Mullender & Cox
5571   //
5572   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5573   // that it will take two back-to-back park() calls for the owning
5574   // thread to block. This has the benefit of forcing a spurious return
5575   // from the first park() call after an unpark() call which will help
5576   // shake out uses of park() and unpark() without condition variables.
5577 
5578   if (Atomic::xchg(1, &_Event) >= 0) return;
5579 
5580   // Wait for the thread associated with the event to vacate
5581   int status = pthread_mutex_lock(_mutex);
5582   assert_status(status == 0, status, "mutex_lock");
5583   int AnyWaiters = _nParked;
5584   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5585   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5586     AnyWaiters = 0;
5587     pthread_cond_signal(_cond);
5588   }
5589   status = pthread_mutex_unlock(_mutex);
5590   assert_status(status == 0, status, "mutex_unlock");
5591   if (AnyWaiters != 0) {
5592     // Note that we signal() *after* dropping the lock for "immortal" Events.
5593     // This is safe and avoids a common class of  futile wakeups.  In rare
5594     // circumstances this can cause a thread to return prematurely from
5595     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5596     // will simply re-test the condition and re-park itself.
5597     // This provides particular benefit if the underlying platform does not
5598     // provide wait morphing.
5599     status = pthread_cond_signal(_cond);
5600     assert_status(status == 0, status, "cond_signal");
5601   }
5602 }
5603 
5604 
5605 // JSR166
5606 // -------------------------------------------------------
5607 
5608 // The solaris and linux implementations of park/unpark are fairly
5609 // conservative for now, but can be improved. They currently use a
5610 // mutex/condvar pair, plus a a count.
5611 // Park decrements count if > 0, else does a condvar wait.  Unpark
5612 // sets count to 1 and signals condvar.  Only one thread ever waits
5613 // on the condvar. Contention seen when trying to park implies that someone
5614 // is unparking you, so don't wait. And spurious returns are fine, so there
5615 // is no need to track notifications.
5616 
5617 // This code is common to linux and solaris and will be moved to a
5618 // common place in dolphin.
5619 //
5620 // The passed in time value is either a relative time in nanoseconds
5621 // or an absolute time in milliseconds. Either way it has to be unpacked
5622 // into suitable seconds and nanoseconds components and stored in the
5623 // given timespec structure.
5624 // Given time is a 64-bit value and the time_t used in the timespec is only
5625 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5626 // overflow if times way in the future are given. Further on Solaris versions
5627 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5628 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5629 // As it will be 28 years before "now + 100000000" will overflow we can
5630 // ignore overflow and just impose a hard-limit on seconds using the value
5631 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5632 // years from "now".
5633 
5634 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5635   assert(time > 0, "convertTime");
5636   time_t max_secs = 0;
5637 
5638   if (!os::supports_monotonic_clock() || isAbsolute) {
5639     struct timeval now;
5640     int status = gettimeofday(&now, NULL);
5641     assert(status == 0, "gettimeofday");
5642 
5643     max_secs = now.tv_sec + MAX_SECS;
5644 
5645     if (isAbsolute) {
5646       jlong secs = time / 1000;
5647       if (secs > max_secs) {
5648         absTime->tv_sec = max_secs;
5649       } else {
5650         absTime->tv_sec = secs;
5651       }
5652       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5653     } else {
5654       jlong secs = time / NANOSECS_PER_SEC;
5655       if (secs >= MAX_SECS) {
5656         absTime->tv_sec = max_secs;
5657         absTime->tv_nsec = 0;
5658       } else {
5659         absTime->tv_sec = now.tv_sec + secs;
5660         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5661         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5662           absTime->tv_nsec -= NANOSECS_PER_SEC;
5663           ++absTime->tv_sec; // note: this must be <= max_secs
5664         }
5665       }
5666     }
5667   } else {
5668     // must be relative using monotonic clock
5669     struct timespec now;
5670     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5671     assert_status(status == 0, status, "clock_gettime");
5672     max_secs = now.tv_sec + MAX_SECS;
5673     jlong secs = time / NANOSECS_PER_SEC;
5674     if (secs >= MAX_SECS) {
5675       absTime->tv_sec = max_secs;
5676       absTime->tv_nsec = 0;
5677     } else {
5678       absTime->tv_sec = now.tv_sec + secs;
5679       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5680       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5681         absTime->tv_nsec -= NANOSECS_PER_SEC;
5682         ++absTime->tv_sec; // note: this must be <= max_secs
5683       }
5684     }
5685   }
5686   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5687   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5688   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5689   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5690 }
5691 
5692 void Parker::park(bool isAbsolute, jlong time) {
5693   // Ideally we'd do something useful while spinning, such
5694   // as calling unpackTime().
5695 
5696   // Optional fast-path check:
5697   // Return immediately if a permit is available.
5698   // We depend on Atomic::xchg() having full barrier semantics
5699   // since we are doing a lock-free update to _counter.
5700   if (Atomic::xchg(0, &_counter) > 0) return;
5701 
5702   Thread* thread = Thread::current();
5703   assert(thread->is_Java_thread(), "Must be JavaThread");
5704   JavaThread *jt = (JavaThread *)thread;
5705 
5706   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5707   // Check interrupt before trying to wait
5708   if (Thread::is_interrupted(thread, false)) {
5709     return;
5710   }
5711 
5712   // Next, demultiplex/decode time arguments
5713   timespec absTime;
5714   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5715     return;
5716   }
5717   if (time > 0) {
5718     unpackTime(&absTime, isAbsolute, time);
5719   }
5720 
5721 
5722   // Enter safepoint region
5723   // Beware of deadlocks such as 6317397.
5724   // The per-thread Parker:: mutex is a classic leaf-lock.
5725   // In particular a thread must never block on the Threads_lock while
5726   // holding the Parker:: mutex.  If safepoints are pending both the
5727   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5728   ThreadBlockInVM tbivm(jt);
5729 
5730   // Don't wait if cannot get lock since interference arises from
5731   // unblocking.  Also. check interrupt before trying wait
5732   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5733     return;
5734   }
5735 
5736   int status;
5737   if (_counter > 0)  { // no wait needed
5738     _counter = 0;
5739     status = pthread_mutex_unlock(_mutex);
5740     assert(status == 0, "invariant");
5741     // Paranoia to ensure our locked and lock-free paths interact
5742     // correctly with each other and Java-level accesses.
5743     OrderAccess::fence();
5744     return;
5745   }
5746 
5747 #ifdef ASSERT
5748   // Don't catch signals while blocked; let the running threads have the signals.
5749   // (This allows a debugger to break into the running thread.)
5750   sigset_t oldsigs;
5751   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5752   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5753 #endif
5754 
5755   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5756   jt->set_suspend_equivalent();
5757   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5758 
5759   assert(_cur_index == -1, "invariant");
5760   if (time == 0) {
5761     _cur_index = REL_INDEX; // arbitrary choice when not timed
5762     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5763   } else {
5764     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5765     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5766     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5767       pthread_cond_destroy(&_cond[_cur_index]);
5768       pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5769     }
5770   }
5771   _cur_index = -1;
5772   assert_status(status == 0 || status == EINTR ||
5773                 status == ETIME || status == ETIMEDOUT,
5774                 status, "cond_timedwait");
5775 
5776 #ifdef ASSERT
5777   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5778 #endif
5779 
5780   _counter = 0;
5781   status = pthread_mutex_unlock(_mutex);
5782   assert_status(status == 0, status, "invariant");
5783   // Paranoia to ensure our locked and lock-free paths interact
5784   // correctly with each other and Java-level accesses.
5785   OrderAccess::fence();
5786 
5787   // If externally suspended while waiting, re-suspend
5788   if (jt->handle_special_suspend_equivalent_condition()) {
5789     jt->java_suspend_self();
5790   }
5791 }
5792 
5793 void Parker::unpark() {
5794   int status = pthread_mutex_lock(_mutex);
5795   assert(status == 0, "invariant");
5796   const int s = _counter;
5797   _counter = 1;
5798   if (s < 1) {
5799     // thread might be parked
5800     if (_cur_index != -1) {
5801       // thread is definitely parked
5802       if (WorkAroundNPTLTimedWaitHang) {
5803         status = pthread_cond_signal(&_cond[_cur_index]);
5804         assert(status == 0, "invariant");
5805         status = pthread_mutex_unlock(_mutex);
5806         assert(status == 0, "invariant");
5807       } else {
5808         // must capture correct index before unlocking
5809         int index = _cur_index;
5810         status = pthread_mutex_unlock(_mutex);
5811         assert(status == 0, "invariant");
5812         status = pthread_cond_signal(&_cond[index]);
5813         assert(status == 0, "invariant");
5814       }
5815     } else {
5816       pthread_mutex_unlock(_mutex);
5817       assert(status == 0, "invariant");
5818     }
5819   } else {
5820     pthread_mutex_unlock(_mutex);
5821     assert(status == 0, "invariant");
5822   }
5823 }
5824 
5825 
5826 extern char** environ;
5827 
5828 // Run the specified command in a separate process. Return its exit value,
5829 // or -1 on failure (e.g. can't fork a new process).
5830 // Unlike system(), this function can be called from signal handler. It
5831 // doesn't block SIGINT et al.
5832 int os::fork_and_exec(char* cmd) {
5833   const char * argv[4] = {"sh", "-c", cmd, NULL};
5834 
5835   pid_t pid = fork();
5836 
5837   if (pid < 0) {
5838     // fork failed
5839     return -1;
5840 
5841   } else if (pid == 0) {
5842     // child process
5843 
5844     execve("/bin/sh", (char* const*)argv, environ);
5845 
5846     // execve failed
5847     _exit(-1);
5848 
5849   } else  {
5850     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5851     // care about the actual exit code, for now.
5852 
5853     int status;
5854 
5855     // Wait for the child process to exit.  This returns immediately if
5856     // the child has already exited. */
5857     while (waitpid(pid, &status, 0) < 0) {
5858       switch (errno) {
5859       case ECHILD: return 0;
5860       case EINTR: break;
5861       default: return -1;
5862       }
5863     }
5864 
5865     if (WIFEXITED(status)) {
5866       // The child exited normally; get its exit code.
5867       return WEXITSTATUS(status);
5868     } else if (WIFSIGNALED(status)) {
5869       // The child exited because of a signal
5870       // The best value to return is 0x80 + signal number,
5871       // because that is what all Unix shells do, and because
5872       // it allows callers to distinguish between process exit and
5873       // process death by signal.
5874       return 0x80 + WTERMSIG(status);
5875     } else {
5876       // Unknown exit code; pass it through
5877       return status;
5878     }
5879   }
5880 }
5881 
5882 // is_headless_jre()
5883 //
5884 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5885 // in order to report if we are running in a headless jre
5886 //
5887 // Since JDK8 xawt/libmawt.so was moved into the same directory
5888 // as libawt.so, and renamed libawt_xawt.so
5889 //
5890 bool os::is_headless_jre() {
5891   struct stat statbuf;
5892   char buf[MAXPATHLEN];
5893   char libmawtpath[MAXPATHLEN];
5894   const char *xawtstr  = "/xawt/libmawt.so";
5895   const char *new_xawtstr = "/libawt_xawt.so";
5896   char *p;
5897 
5898   // Get path to libjvm.so
5899   os::jvm_path(buf, sizeof(buf));
5900 
5901   // Get rid of libjvm.so
5902   p = strrchr(buf, '/');
5903   if (p == NULL) {
5904     return false;
5905   } else {
5906     *p = '\0';
5907   }
5908 
5909   // Get rid of client or server
5910   p = strrchr(buf, '/');
5911   if (p == NULL) {
5912     return false;
5913   } else {
5914     *p = '\0';
5915   }
5916 
5917   // check xawt/libmawt.so
5918   strcpy(libmawtpath, buf);
5919   strcat(libmawtpath, xawtstr);
5920   if (::stat(libmawtpath, &statbuf) == 0) return false;
5921 
5922   // check libawt_xawt.so
5923   strcpy(libmawtpath, buf);
5924   strcat(libmawtpath, new_xawtstr);
5925   if (::stat(libmawtpath, &statbuf) == 0) return false;
5926 
5927   return true;
5928 }
5929 
5930 // Get the default path to the core file
5931 // Returns the length of the string
5932 int os::get_core_path(char* buffer, size_t bufferSize) {
5933   /*
5934    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5935    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5936    */
5937   const int core_pattern_len = 129;
5938   char core_pattern[core_pattern_len] = {0};
5939 
5940   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5941   if (core_pattern_file != -1) {
5942     ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5943     ::close(core_pattern_file);
5944 
5945     if (ret > 0) {
5946       char *last_char = core_pattern + strlen(core_pattern) - 1;
5947 
5948       if (*last_char == '\n') {
5949         *last_char = '\0';
5950       }
5951     }
5952   }
5953 
5954   if (strlen(core_pattern) == 0) {
5955     return -1;
5956   }
5957 
5958   char *pid_pos = strstr(core_pattern, "%p");
5959   int written;
5960 
5961   if (core_pattern[0] == '/') {
5962     written = jio_snprintf(buffer, bufferSize, core_pattern);
5963   } else {
5964     char cwd[PATH_MAX];
5965 
5966     const char* p = get_current_directory(cwd, PATH_MAX);
5967     if (p == NULL) {
5968       return -1;
5969     }
5970 
5971     if (core_pattern[0] == '|') {
5972       written = jio_snprintf(buffer, bufferSize,
5973                         "\"%s\" (or dumping to %s/core.%d)",
5974                                      &core_pattern[1], p, current_process_id());
5975     } else {
5976       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5977     }
5978   }
5979 
5980   if (written < 0) {
5981     return -1;
5982   }
5983 
5984   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5985     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5986 
5987     if (core_uses_pid_file != -1) {
5988       char core_uses_pid = 0;
5989       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5990       ::close(core_uses_pid_file);
5991 
5992       if (core_uses_pid == '1') {
5993         jio_snprintf(buffer + written, bufferSize - written,
5994                                           ".%d", current_process_id());
5995       }
5996     }
5997   }
5998 
5999   return strlen(buffer);
6000 }
6001 
6002 /////////////// Unit tests ///////////////
6003 
6004 #ifndef PRODUCT
6005 
6006 #define test_log(...)              \
6007   do {                             \
6008     if (VerboseInternalVMTests) {  \
6009       tty->print_cr(__VA_ARGS__);  \
6010       tty->flush();                \
6011     }                              \
6012   } while (false)
6013 
6014 class TestReserveMemorySpecial : AllStatic {
6015  public:
6016   static void small_page_write(void* addr, size_t size) {
6017     size_t page_size = os::vm_page_size();
6018 
6019     char* end = (char*)addr + size;
6020     for (char* p = (char*)addr; p < end; p += page_size) {
6021       *p = 1;
6022     }
6023   }
6024 
6025   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6026     if (!UseHugeTLBFS) {
6027       return;
6028     }
6029 
6030     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6031 
6032     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6033 
6034     if (addr != NULL) {
6035       small_page_write(addr, size);
6036 
6037       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6038     }
6039   }
6040 
6041   static void test_reserve_memory_special_huge_tlbfs_only() {
6042     if (!UseHugeTLBFS) {
6043       return;
6044     }
6045 
6046     size_t lp = os::large_page_size();
6047 
6048     for (size_t size = lp; size <= lp * 10; size += lp) {
6049       test_reserve_memory_special_huge_tlbfs_only(size);
6050     }
6051   }
6052 
6053   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6054     size_t lp = os::large_page_size();
6055     size_t ag = os::vm_allocation_granularity();
6056 
6057     // sizes to test
6058     const size_t sizes[] = {
6059       lp, lp + ag, lp + lp / 2, lp * 2,
6060       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6061       lp * 10, lp * 10 + lp / 2
6062     };
6063     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6064 
6065     // For each size/alignment combination, we test three scenarios:
6066     // 1) with req_addr == NULL
6067     // 2) with a non-null req_addr at which we expect to successfully allocate
6068     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6069     //    expect the allocation to either fail or to ignore req_addr
6070 
6071     // Pre-allocate two areas; they shall be as large as the largest allocation
6072     //  and aligned to the largest alignment we will be testing.
6073     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6074     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6075       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6076       -1, 0);
6077     assert(mapping1 != MAP_FAILED, "should work");
6078 
6079     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6080       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6081       -1, 0);
6082     assert(mapping2 != MAP_FAILED, "should work");
6083 
6084     // Unmap the first mapping, but leave the second mapping intact: the first
6085     // mapping will serve as a value for a "good" req_addr (case 2). The second
6086     // mapping, still intact, as "bad" req_addr (case 3).
6087     ::munmap(mapping1, mapping_size);
6088 
6089     // Case 1
6090     test_log("%s, req_addr NULL:", __FUNCTION__);
6091     test_log("size            align           result");
6092 
6093     for (int i = 0; i < num_sizes; i++) {
6094       const size_t size = sizes[i];
6095       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6096         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6097         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6098             size, alignment, p, (p != NULL ? "" : "(failed)"));
6099         if (p != NULL) {
6100           assert(is_ptr_aligned(p, alignment), "must be");
6101           small_page_write(p, size);
6102           os::Linux::release_memory_special_huge_tlbfs(p, size);
6103         }
6104       }
6105     }
6106 
6107     // Case 2
6108     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6109     test_log("size            align           req_addr         result");
6110 
6111     for (int i = 0; i < num_sizes; i++) {
6112       const size_t size = sizes[i];
6113       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6114         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6115         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6116         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6117             size, alignment, req_addr, p,
6118             ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6119         if (p != NULL) {
6120           assert(p == req_addr, "must be");
6121           small_page_write(p, size);
6122           os::Linux::release_memory_special_huge_tlbfs(p, size);
6123         }
6124       }
6125     }
6126 
6127     // Case 3
6128     test_log("%s, req_addr non-NULL with preexisting mapping:", __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(mapping2, 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, req_addr, p,
6138             ((p != NULL ? "" : "(failed)")));
6139         // as the area around req_addr contains already existing mappings, the API should always
6140         // return NULL (as per contract, it cannot return another address)
6141         assert(p == NULL, "must be");
6142       }
6143     }
6144 
6145     ::munmap(mapping2, mapping_size);
6146 
6147   }
6148 
6149   static void test_reserve_memory_special_huge_tlbfs() {
6150     if (!UseHugeTLBFS) {
6151       return;
6152     }
6153 
6154     test_reserve_memory_special_huge_tlbfs_only();
6155     test_reserve_memory_special_huge_tlbfs_mixed();
6156   }
6157 
6158   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6159     if (!UseSHM) {
6160       return;
6161     }
6162 
6163     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6164 
6165     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6166 
6167     if (addr != NULL) {
6168       assert(is_ptr_aligned(addr, alignment), "Check");
6169       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6170 
6171       small_page_write(addr, size);
6172 
6173       os::Linux::release_memory_special_shm(addr, size);
6174     }
6175   }
6176 
6177   static void test_reserve_memory_special_shm() {
6178     size_t lp = os::large_page_size();
6179     size_t ag = os::vm_allocation_granularity();
6180 
6181     for (size_t size = ag; size < lp * 3; size += ag) {
6182       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6183         test_reserve_memory_special_shm(size, alignment);
6184       }
6185     }
6186   }
6187 
6188   static void test() {
6189     test_reserve_memory_special_huge_tlbfs();
6190     test_reserve_memory_special_shm();
6191   }
6192 };
6193 
6194 void TestReserveMemorySpecial_test() {
6195   TestReserveMemorySpecial::test();
6196 }
6197 
6198 #endif